Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

default · ~99.6k rows (showing the first 120k)

Split (2)

train · ~72.4k rows (showing the first 93.1k)

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1512.03391	2	1512	2015-12-30T20:21:19	Topology of Exceptional Orbit Hypersurfaces of Prehomogeneous Spaces	[ "math.AG", "math.AT" ]	We consider the topology for a class of hypersurfaces with highly nonisolated singularites which arise as exceptional orbit varieties of a special class of prehomogeneous vector spaces, which are representations of linear algebraic groups with open orbits. These hypersurface singularities include both determinantal hypersurfaces and linear free (and free*) divisors. Although these hypersurfaces have highly nonisolated singularities, we determine the topology of their Milnor fibers, complements and links. We do so by using the action of linear algebraic groups beginning with the complement, instead of using Morse type arguments on the Milnor fibers. This includes replacing the local Milnor fiber by a global Milnor fiber which has a complex geometry resulting from a transitive action of an appropriate algebraic group, yielding a compact model submanifold for the homotopy type of the Milnor fiber. The topology includes the (co)homology (in characteristic 0, and 2 torsion in one family) and homotopy groups, and we deduce the triviality of the monodromy transformations on rational (or complex) cohomology. The cohomology of the Milnor fibers and complements are isomorphic as algebras to exterior algebras or for one family, modules over exterior algebras; and cohomology of the link is, as a vector space, a truncated and shifted exterior algebra, for which the cohomology product structure is essentially trivial. We also deduce from Bott's periodicity theorem, the homotopy groups of the Milnor fibers for determinantal hypersurfaces in the stable range as the stable homotopy groups of the associated infinite dimensional symmetric spaces. Applying a Theorem of Oka we obtain a class of formal linear combinations of exceptional orbit hypersurfaces which have Milnor fibers which are homotopy equivalent to joins of the compact model submanifolds.	math.AG	math	TOPOLOGY OF EXCEPTIONAL ORBIT HYPERSURFACES OF PREHOMOGENEOUS SPACES JAMES DAMON1 Abstract. We consider the topology for a class of hypersurfaces with highly nonisolated singularites which arise as "exceptional orbit varieties"of a special class of prehomogeneous vector spaces, which are representations of linear alge- braic groups with open orbits. These hypersurface singularities include both determinantal hypersurfaces and linear free (and free) divisors. Although these hypersurfaces have highly nonisolated singularities, we determine the topology of their Milnor fibers, complements and links. We do so by using the action of linear algebraic groups beginning with the complement, instead of using Morse-type arguments on the Milnor fibers. This includes replac- ing the local Milnor fiber by a global Milnor fiber which has a "complex ge- ometry"resulting from a transitive action of an appropriate algebraic group, yielding a compact "model submanifold"for the homotopy type of the Mil- nor fiber. The topology includes the (co)homology (in characteristic 0, and 2-torsion in one family) and homotopy groups, and we deduce the triviality of the monodromy transformations on rational (or complex) cohomology. Unlike isolated singularities, the cohomology of the Milnor fibers and com- plements are isomorphic as algebras to exterior algebras or for one family, mod- ules over exterior algebras; and cohomology of the link is, as a vector space, a truncated and shifted exterior algebra, for which the cohomology product structure is essentially trivial. We also deduce from Bott's periodicity theo- rem, the homotopy groups of the Milnor fibers for determinantal hypersurfaces in the "stable range"as the stable homotopy groups of the associated infinite dimensional symmetric spaces. Lastly, we combine the preceding with a The- orem of Oka to obtain a class of "formal linear combinations"of exceptional orbit hypersurfaces which have Milnor fibers which are homotopy equivalent to joins of the compact model submanifolds. It follows that Milnor fibers for all of these hypersurfaces are essentially never homotopy equivalent to bouquets of spheres (even allowing differing dimensions). Introduction In this paper we investigate the topology of a class of highly nonisolated hy- persurface singularities E, each of which arises as the hypersurface of exceptional orbits, the exceptional orbit variety, for a rational representation of connected com- plex linear algebraic group ρ : G → GL(V ) with an open orbit U. Such a space with group action has been studied by Sato and Kimura [So], [SK] and is called by 1991 Mathematics Subject Classification. Primary: 11S90, 32S25, 55R80 Secondary: 57T15, 14M12, 20G05. Key words and phrases. linear algebraic groups, solvable, reductive, prehomogeneous spaces, determinantal hypersurfaces, classical symmetric spaces, exceptional orbit varieties, linear free divisors, global Milnor fibration, cohomological triviality of Milnor fibration, topology of Milnor fibers, complements, links, quivers, Hopf algebras, exterior algebras. (1) Partially supported by the Simons Foundation grant 230298 and the National Science Foundation grant DMS-1105470. 1 2 JAMES DAMON them a prehomogeneous vector space, which we will shorten in this paper to just prehomogeneous space. Both determinantal hypersurfaces and linear free divisors belong to this class. We consider how the topology of such singularities can be determined. For nonisolated singularities with small dimensional singular set, a body of work by Siersma [Si], [Si2], [Si3], Tibar [Ti], Nemethi [Ne], Zaharia [Z], etc. has used Morse-theoretic methods to extend the results for isolated singularities, and show that the Milnor fiber is still homotopy equivalent to a bouquet of spheres, except now the spheres may have different dimensions. One might ask to what extent such results apply to these hypersurfaces of exceptional orbits, which now have highly nonisolated singularities. Complex Morse theory has been applied to determine the vanishing topology of "nonlinear sections"of these hypersurfaces in terms of "singu- lar Milnor fibers"in results by Mond, Goryunov, Bruce, Pike, and this author, see e.g. [DM], [Br], [D1, D2], [GM], and [DP3]. However, these results provide little information about the topology of the hypersurface singularities themselves. We shall see that the structure of these singularities are quite different from those stud- ied in the above work. In fact, to determine their topology we take a very different approach which makes substantial use of their representations via prehomogeneous spaces. For the exceptional orbit varieties, we will be concerned with the topology of the Milnor fiber, the complement and the link; and we determine their homotopy types along with the (co)homology structure, homotopy groups, and the monodromy action. The changes in approach which make this possible are: i) reversing the usual approach from first determining the Milnor fiber and monodromy to then compute the topology of the link and complement by beginning instead with the complement and deducing the topology of the link and Milnor fiber. ii) replacing the local Milnor fiber by a global Milnor fiber, which is a smooth affine hypersurface that has a "model complex geometry"resulting from the transitive action of an associated linear algebraic group, yielding as a deformation retract a compact submanifold; iii) using the relation between the two algebraic group actions and the topology of maximal compact subgroups to deduce the cohomological triviality of an associated fibration of the groups; and iv) using the preceding to determine the topology and cohomology of the Mil- nor fiber. We will explicity compute the cohomology of the Milnor fiber, the complement and the link for two classes: determinantal hypersurfaces, which are varieties of singular matrices in the spaces of m×m matrices which may be symmetric, general, or skew-symmetric (with m even); and exceptional orbit varieties in the general equidimensional case. This second class will in particular apply to both linear free (and free) divisors, introduced for reductive groups by Buchweitz-Mond [BM] and determined for "block representations"of solvable groups in Damon-Pike [DP], [DP2]. We express the cohomology algebras of the Milnor fibers as either exterior al- gebras, or for the one case of symmetric m × m matrices with m even, modules over an exterior algebra on two generators, see Theorems 3.1 and 5.1. For determi- nantal hypersurfaces, we specifically show, Theorem 3.1, that the Milnor fibers are EXCEPTIONAL ORBIT HYPERSURFACES 3 homotopy equivalent to compact classical symmetric spaces of Cartan; and besides obtaining their cohomology we give their homotopy groups in a stable range using Bott periodicity. As well, certain of the Milnor fibers exhibit 2-torsion in their cohomology. We further show that for either class, excluding the symmetric m × m matrices with m even, the monodromy acts trivially on the (rational or) complex cohomology of the Milnor fiber. For the links in either case, we compute the co- homology vector space as a truncated and shifted exterior algebra and determine that its cohomology product structure is "almost trivial", Theorems 3.2 and 4.5. For the complex cohomology of the complement for reductive groups these results extend those obtained in Granger-Mond et al. [GMNS], although in that paper they also prove the cohomology is computed by the logarithmic complex, which we do not. The results here also extend the results obtained for the complement and Milnor fiber in the case of solvable algebraic groups in Damon-Pike [DP]. Lastly, we combine in §6 the results on exceptional orbit hypersurfaces we have described together with a Theorem of Mutsuo Oka [Ok] and the results of Siersma et al to compute the topology of a class of hypersurface singularities which are formed as "formal sums"of hypersurface singularities each of which is either an exceptional orbit hypersurface which we consider or is a weighted homogeneous nonisolated hypersurface singularity considered by the Siersma group. These yield hypersurface singularities whose Milnor fibers are homotopy equivalent to joins of compact manifolds, or more generally to a bouquet of suspensions of such joins of compact manifolds. We would like to thank both David Mond and Shrawan Kumar for their input from several valuable conversations on these questions, and the referee for a number of very useful suggestions. 1. Prehomogeneous Spaces and the Wang Sequence A Special Class of Prehomogeneous Spaces. We consider a special class of representations of a connected complex linear algebraic groups ρ : G → GL(V ) which have an open orbit U. We specifically consider the cases where the excep- tional orbit variety, which is the union of the orbits of positive codimension, is a hypersurface E. If H is the isotropy subgroup for a point v0 ∈ U, then H is a closed algebraic subgroup of G, and it is a basic fact, see e.g. Borel [Bo2], that both G and H have maximal compact subgroups K, resp. L, with L ⊂ K, which are strong deformation retracts of G, resp. H, and of the same ranks as G and H. For example this will include the cases where V = M is one of the spaces of complex matrices M = Symm or M = Skm (for m = 2k) acted on by GLm(C) by B · A = BABT , or , M = Mm,m and GLm(C) acts by left multiplication. Each of these representations have open orbits and the resulting prehomogeneous space has an exceptional orbit variety E which is a hypersurface. Definition 1.1. The determinantal hypersurface for the space of m × m symmetric or general matrices, denoted by M = Symm or M = Mm,m is the hypersurface of singular matrices defined by det : M → C and denoted by Dsy m for M = Symm, or Dm for M = Mm,m. For the space of m × m skew-symmetric matrices M = Skm (for m = 2k) the determinantal hypersurface of singular matrices is defined by the Pfaffian Pf : Skm → C, and is denoted by Dsk m . A second class of examples consists of "equidimensional representations"where dim CV = dim CG with an open orbit U. Then, necessarily E is a hypersurface and 4 JAMES DAMON the isotropy subgroup H of a point in the open orbit is finite. If an appropriate defining equation for E, obtained from the coefficient determinant of the associated vector fields for the action, is reduced then E is a "linear free divisor", introduced by Mond and Buchweitz [BM]; and otherwise it is a slightly weaker linear free* divisor with nonreduced defining equation. Partial results on the topology of the complement were obtained if G is reductive, see e.g. [GMNS]) and also for the complement and Milnor fiber for "block representations"for G solvable (see [DP], [DP2]). We shall determine the topology in the general case. For any of these cases, the action of G commutes with the usual C∗-action on V ; hence, the exceptional orbit variety E is also invariant, and hence has a homo- geneous reduced defining equation f of degree n. In the case of equidimensional representations there is a defining equation given by the coefficient determinant of the associated vector fields of degree N = dim C V = dim CG. If E is a linear free divisor, this is a reduced definng equation, which we may choose for f ; if not then deg f = n < N . δ ) ∩ D2N ε We consider the Milnor fibration of the hypersurface germ (E, 0) in the standard form as f −1(S1 the disk about 0 of sufficiently small radius ε, δ the boundary of the disk about 0 in C of radius δ, for 0 < δ << ε. Because of S1 the homogeneity of f we may adapt a standard argument for isolated singularities to obtain the global description of the Milnor fibration. ε with D2N Lemma 1.2. The Milnor fibration of (E, 0) is diffeomorphic to the fibration f E : E → S1, where E = f −1(S1), with fiber F = f −1(1). This is the restriction of the fibration f : V \E → C∗ to S1 ⊂ C∗, and the inclusion E ⊂ V \E is a homotopy equivalence. Proof. The proof is a slight modification of that for the case of isolated singularities. It uses the induced homogeneous R+-action to establish a diffeomorphism of fibra- tions between E = f −1(S1) and f −1(S1 δ ). If 0 < δ << ε are sufficiently small, it is proven that the restriction of the distance-squared function k · k2 to f −1(S1 δ )\D2N has no critical points, and by Morse theory we deduce that the Milnor fibration is diffeomorphic as a fibration to f −1(S1 δ ). For the case of equidimensional represen- tations of solvable linear algebraic groups the details are given in the proof of [DP, Thm 3.2]. The general case follows the same line of reasoning. (cid:3) ε We refer to the fibration F ֒→ E → S1 as the global Milnor fibration and F as the global Milnor fiber. We note that by e.g. Kato-Masumoto [KM], provided N ≥ 2, F is path-connected. Second, the cohomology of the complement V \E (and hence E), can be expressed in terms of the maximal compact subgroups. Lemma 1.3. If G and the isotropy subgroup H of v0 ∈ U have maximal compact subgroups K, resp. L, then V \E is homotopy equivalent to K/L. Hence, H ∗(V \E) ≃ H ∗(K/L) . Proof. First, the action of G on v0 ∈ U, gives the open orbit V \E = U ≃ G/H as a homogenenous space. It is sufficient to show the inclusion K/L ֒→ G/H is a homotopy equivalence. As pointed out by Shrawan Kumar, this is a simple consequence of the long exact homotopy sequence for a fibration. The groups G, resp. H, are homotopy equivalent to their maximal compact subgroups, K, resp.L. EXCEPTIONAL ORBIT HYPERSURFACES 5 We consider the long exact homotopy sequence of the fibrations (see e.g. [Sp, Chap 7, §2, Thm 10]) in the horizontal rows of (1.1). L −−−−→ K −−−−→ K/L H −−−−→ G −−−−→ G/H i0y i ′y (1.1) which has the form (1.2) i ′ ∗y iy i∗y −−−−→ πj(L, Id) −−−−→ πj(K, Id) −−−−→ πj (K/L, {L}) −−−−→ i0 ∗y −−−−→ πj(H, Id) −−−−→ πj(G, Id) −−−−→ πj(G/H, {H}) −−−−→ We note that when j = 0, both i∗ and i ′ ∗ are bijections of pointed sets so K is connected and H and L have the same finite number of connected components. The homotopy equivalence i ′ : L ֒→ H restricts to a homotopy equivalence i ′ : L0 ֒→ H0 of the connected components of the identity. Thus, i ′ ∗ : πj(L, Id) = πj(L0, Id) ≃ πj(H0, Id) = πj(H, Id) is an isomorphism for all j ≥ 0. Hence, we can apply the five Lemma to conclude first that πi(K/L, {L}) ≃ πi(G/H, {H}) for all i ≥ 0. As both are path-connected CW-complexes, it follows by Whitehead's theorem that K/L is homotopy equivalent to G/H, giving the result. (cid:3) Next, we consider the global Milnor fibration F ֒→ E → S1. First, from the long exact homotopy sequence for the fibration, we obtain πi(F ) ≃ πi(E) for i ≥ 2 and, as F is path-connected, the short exact sequence p 1 −−−−→ π1(F ) −−−−→ π1(E) −−−−→ π1(S1) −−−−→ 1 . (1.3) Hence, by the Hurewicz theorem and the universal coefficient theorem, p∗ : H1(E) → H1(S1) is surjective, and p∗ : H 1(S1; k) → H 1(E; k) is injective for any field k of characteristic 0, so the image of a generator of H 1(S1; k) is a nonzero class s1 ∈ H1(E). Furthermore, this class restricts to 0 in H 1(F ; k), so then does the ideal hs1i · H ∗(E; k). Example 1.4. For the case of complex cohomology, if γ is a generator of π1(S1), then there is a β ∈ π1(E) such that f∗(β) = γ. Hence, df z ) is a closed 1-form which satisfies f = f ∗( dz Zβ df f = Zf∗(β) dz z = 2πi . Hence ,ω1 = df f defines a nontrivial cohomology class s1 in H 1(E; C). Moreover, its restriction to F is 0 as df = 0 as f ≡ 1 on F . Thus, the ideal Chs1i · H ∗(E; C) restricts to 0 in H ∗(F ; C). Model Complex Geometry for the Milnor Fiber. We note that although by Lemma 1.2 the Milnor fiber and global Milnor fiber are diffeomorphic, they are not holomorphically diffeomorphic. The different complex structure on the global Milnor fiber allows us to introduce an alternate way to view the topology on the global Milnor fiber resulting from it having a form of "complex model geometry"in the sense of Thurston (see e.g. [Th, Chap. 3]). By this we mean a smooth affine submanifold X ⊂ CN together with an algebraic action of a connected linear alge- braic group G on X, which is transitive with H an isotropy subgroup of a point. 6 JAMES DAMON We denote it by the triple (X, G, H). It provides a method for beginning to under- stand the geometry of the global Milnor fiber. Actually, Thurston's definition of geometry over the reals required X be simply connected, which we refer to it as a simple model geometry. However, without this restriction, the following is true. Proposition 1.5. For a prehomogeneous space defined by the representation ρ : G → GL(V ) with exceptional orbit variety a hypersurface, there is a connected codi- mension one algebraic subgroup G′ 0 of G which acts transitively on the global Milnor fiber F , with isotropy subgroup H ′ so that (F, G′, H ′) defines a model complex ge- ometry on F . This model is simple in the case of determinantal hypersurfaces. The proof of this will follow from results in § 2. This model geometry type representation of the global Milnor fiber is the essential ingredient, in place of Morse theory, for the analysis of the topology. To carry out this analysis, we will make considerable use of the Wang cohomology sequence to relate the cohomology of E with that of F . p → S1. We let σ = σ1 be The Wang Sequence. We consider a fibration F ֒→ E the monodromy map where σt denotes the lift of the curve ϕ(t) = e2πit on S1. to the family of homeomorphisms σt : F → Ft = p−1(ϕ(t)). Definition 1.6. We say that the fibration F ֒→ E → S1 is (rationally) cohomo- logically trivial if σ∗ : H ∗(F ; Q) → H ∗(F ; Q) is the identity. Remark 1.7. By the universal coefficient theorem it likewise follows that for any field k of characteristic 0, σ∗ = id on H ∗(F ; k) and conversely if it holds for k of characteristic 0 then it also holds for Q. Also, if H ∗(F ; Z) is a free abelian group then it also follows that the Milnor fibration is cohomologically trivial for integer coefficients. We also might ask whether the monodromy is geometrically trivial in that it is homotopy equivalent to the identity. David Mond has been able to show di- rectly for several examples in the equidimensional case, including certain quiver representations, that the monodromy is geometrically trivial; however, for certain other equidimensional cases we discussed neither of us were unable to verify this. For now there is no general result for geometric triviality of monodromy, even for certain special classes such as quiver representations, so it is still an open ques- tion. However, we will see that being cohomologically trivial will suffice for many questions. To begin, there is the following result. Proposition 1.8. For the fibration F ֒→ E → S1, with F connected and k a field of characteristic 0, the following are equivalent. i) The fibration is cohomologically trivial. ii) there is an isomorphism of graded vector spaces. (1.4) iii) (1.5) H ∗(E; k) ≃ Λ∗khs1i ⊗ H ∗(F ; k) ≃ H ∗(F ; k) ⊕ khs1i ⊗ H ∗(F ; k) . dim kH ∗(E; k) = 2 dim kH ∗(F ; k) . EXCEPTIONAL ORBIT HYPERSURFACES 7 Moreover, if the preceding hold then (1.4) is an isomorphism of graded Λ∗khs1i- modules, where the exterior algebra Λ∗khs1i is on one generator s1. Proof. First, we immediately observe that ii) imples iii). Second, suppose i) holds. We consider the Wang sequence in cohomology (see e.g. [Sp, Chap. 8 §5 Cor. 6]) applied to our situation. (1.6) −−−−→ H q(F ; k) θ−−−−→ H q(F ; k) −−−−→ H q+1(E; k) i∗ −−−−→ H q+1(F ; k) θ−−−−→ H q+1(F ; k) −−−−→ where θ = id − σ∗ for σ the monodromy map. If the fibration is cohomologically trivial, then σ∗ = id, and (1.6) reduces to the short exact sequences for q ≥ −1, (1.7) 0 −−−−→ H q(F ; k) −−−−→ H q+1(E; k) i∗ −−−−→ H q+1(F ; k) −−−−→ 0 It then follows that the Betti numbers, which equal βq = dim kH q( · ; k), satisfy βq+1(E) = βq(F ) + βq+1(F ) for q ≥ −1 (where β−1 = 0). This says that there is a graded vector space isomorphism (1.4) and also that (1.5) holds. Third, if iii) holds we show that i) holds. For any q ≥ 0, let Kq = ker(I − σ∗) : H q(F ; k) → H q(F ; k) and Cq = coker (I − σ∗). Then, dim kKq = dim kCq and from (1.6) (1.8) βq+1(E) = dim kKq+1 + dim kCq = dim kKq+1 + dim kKq . Hence, if we choose an r > dim F so that H j(F ; k) = 0 for j ≥ r, then summing (1.8) over q yields dim kH ∗(E; k) = r−1Xq=−1 rXq=0 = 2 dim kKq+1 + rXq=0 dim kKq dim kKq (1.9) Hence, if the fibration is not cohomologically trivial then for some q, dim kKq < dim kH q(F ; k). It follows that the RHS of (1.9) is less than 2 dim kH ∗(F ; k), so iii) doesn't hold. Lastly, if the above hold, then (1.4) is an isomorphism of graded vector spaces, and (1.6) reduces to the short exact sequences (1.7). Thus, for each q, we can find a set of elements {ϕ(q) mq } in H q(E; k) which restrict to a basis for H q(F ; k). Then, we can apply the Leray-Hirsch Theorem, see e.g. [Sp, Chap. 5, §7,Thm 9], to conclude that H ∗(E; k) is a free H ∗(S1; k)-module on this set of generators. As H ∗(S1; k) ≃ Λ∗khs1i, we conclude that (1.4) is an isomorphism of graded Λ∗khs1i-modules. (cid:3) 1 , . . . , ϕ(q) Topology of the Complement, Milnor Fiber, and Link. We consider a rep- resentation ρ : G → GL(V ) which belongs to the special class of prehomogeneous spaces: with open orbit U, isotropy subgroup H of v0 ∈ U, with maximal com- pact subgroups K, resp. L, exceptional orbit variety E which is a hypersurface, and whose global Milnor fiber is cohomologically trivial. We apply the preceding 8 JAMES DAMON results to draw conclusions about the cohomology of the complement, the Milnor fiber and link of E using the preceding notation. We have already considered both the cohomology of the complement and Milnor fiber. Since E is invariant under the usual C∗-action, we may use the restricted R+-action to conclude that L(E) = E ∩ S2N −1 is diffeomorphic to the link of E, where S2N −1 denotes the unit sphere about the origin in V . To describe the cohomology of the link we use the following notation. For a compact connected orientable manifold X, we introduce the graded vector space ^H ∗(X; k) [r] which will denote the vector space H ∗(X; k), truncated at the top degree and shifted upward by degree r. Then we can summarize the topology of the exceptional orbit variety E by the following. Proposition 1.9. Suppose ρ : G → GL(V ) is a representation which belongs to the special class of prehomogeneous spaces. i) Cohomology of Milnor fiber F : If the global Milnor fibration is cohomologically trivial, then H ∗(F ; k) ≃ H ∗(E; k)/ (khs1i ⌣ H ∗(E; k)) where as graded vector spaces, H ∗(E; k) ≃ H ∗(F ; k) ⊕ (khs1i ⊗ H ∗(F ; k)) ; ii) Cohomology of the Complement V \E: H ∗(V \E; k) ≃ H ∗(K/L; k) ; iii) Cohomology of the Link L(E): If K/L is orientable, then Proof. We have already discussed i) in Proposition 1.8, and ii) in Lemma 1.3. eH ∗(L(E); k) ≃ ^H ∗(K/L; k) [2N − 2 − dim RK/L] ; For the link, as E is homogeneous (of degree n), the complement V \E is also invariant under the usual C∗ action. Thus, we may contract using the R+-action to conclude that E\{0} is homotopy equivalent to the link L(E) and V \E is homotopy equivalent to the complement of the link S2N −1\L(E). Thus, (1.10) H ∗(S2N −1\L(E); k) = H ∗(V \E; k) . Second, we apply Alexander duality for subspaces of spheres (see e.g. [Ma, Chap. XIV, Thm 6.6]) to conclude for all j (1.11) Hence, combined with (1.10), we obtain for all j eH j(L(E); k) ≃ eH2N −2−j(S2N −1\L(E); k) . (1.12) By Lemma 1.3 (1.13) eH j(L(E); k) ≃ eH2N −2−j(V \E; k) . H ∗(V \E; k) ≃ H ∗(K/L; k) If q = dim RK/L, then, as K/L is compact and orientable, by Poincar´e duality H q−r(K/L; k) ≃ H r(K/L; k) for all 0 ≤ r ≤ q. Thus, for j = 2N − 2 − (q − r), we obtain from (1.12) for 0 < r < q (1.14) eH 2N −2−q+r(L(E); k) ≃ eHq−r(K/L; k) ≃ eH q−r(K/L; k) ≃ eH r(K/L; k) . EXCEPTIONAL ORBIT HYPERSURFACES 9 If r = q we obtain This last equation also follows as dim RL(E) = 2N − 3. If j = 0 in (1.12) and N > 2 then as 2N − 2 > N = dim RK/L, eH 2N −2)(L(E); k) ≃ eH 0(K/L; k) = 0 . Combining (1.14) and the comment following it with (1.13), yields the result if N > 2. eH 0(L(E); k) ≃ eH 2N −2(K/L; k) = 0 . The only cases when N ≤ 2 are Sym1, M1,1, and Sk2, each of which is dimension 1 with GL1(C) = C ∗, K = S1, H = {1}, and E = {0}. In each case the Milnor fiber is a single point, the link is empty, and the complement is homotopy equivalent to S1. Then, iii) has to be understood as correct for eH j(L(E); k) for j ≥ 0 and then the result is true. (cid:3) Then, we will use these results to determine when the global Milnor fibrations arising from two classes of exceptional orbit hypersurfaces are cohomologically triv- ial, and show that the cohomology H ∗(K/L; k) is either an exterior algebra or a module over an exterior algebra. 2. Modeling Global Milnor Fibers for Exceptional Orbit Hypersurfaces Let ρ : G → GL(V ) be a representation of a connected linear algebraic group G defining a prehomogeneous space with exceptional orbit variety E a hypersurface. We will identify a subgroup of G which defines a model complex geometry on the global Milnor fiber F . Using this we relate the corresponding maximal connected compact subgroups via an exact sequence that gives a fibration which we show is always cohomologically trivial. For now we consider the general case with E a hypersurface with reduced ho- mogenous defining equation h = 0. We first define a representation χ of G on Chhi as follows. If g ∈ G, then we let g∗h(v) = h(g−1 · v) for v ∈ V . As the action of g−1 preserves the orbits of G, g∗h still vanishes on E. Thus, by the Nullstellensatz, g∗h is a multiple of h. Also, as the action of g is given by a linear transformation of V , g∗h is again a polynomial of the same degree as h; hence, g∗h = cg · h, for cg ∈ C. This defines a representation χ0 : G → GL(Chhi) ≃ C∗. We let G′ = ker(χ0) and G′ 0 denote the connected component of the identity of G′. We begin with a simple Lemma. Lemma 2.1. In the preceding situation, χ0 is non-trivial and induces by a lifting an exact sequence (2.1) χ −−−−→ C∗ −−−−→ 1 1 −−−−→ G′ 0 are linear algebraic groups with dim CG′ 0 −−−−→ G where G′ and G′ rank (G′ so is G′ 0 = dim CG − 1 and 0) = rank (G) − 1. Moreover, if G is reductive, respectively solvable, then 0 reductive, respectively solvable. Proof. First, if χ0 were trivial, then G acts trivially on h; hence, it will preserve the fibers h−1(w) for any w ∈ C. This says that the orbits of G are all of positive codimension, so there is no open orbit, a contradiction. Thus, χ0 is surjective and G/G′ ≃ C∗. 10 JAMES DAMON Since G′ is a closed subgroup of G, it is an algebraic subgroup, as is the subgroup 0 have the same dimension and rank. From G/G′ ≃ C∗ we see 0. Also, G′ and G′ G′ dim CG′ 0 = dim CG′ = dim CG − 1. As G′ is algebraic, it has only finitely many connected components. Thus, the quotient map G/G′ 0 ≃ C∗. Thus, we have the exact sequence given in (2.1), where χ denotes the lifting of χ0 for the covering map. 0 → G/G′ ≃ C∗ is a finite covering map of C∗ and hence G/G′ Next, if T denotes a maximal algebraic torus of G of dimension k = rank (G), then χ0T is still onto C∗, otherwise its image would be {1}. Since every element in G is conjugate to an element of T , χ would be trivial on G, which as we just saw is impossible. Thus, ker(χ0T ) is a torus of dimension k − 1. Thus, rank (G′) ≥ k − 1. If rank (G′) = k, then there is a maximal algebraic torus T ′ ⊂ G′ of rank k. Then, T ′ is also a maximal algebraic torus of G and χ0T ′ is trivial, a contradiction. Thus, rank (G′) = k − 1. If G is solvable, then so is its subgroup G′ 0. If G is reductive, and K is the maximal compact subgroup with Lie algebra k, then g = k + ik. If χ : g → C is the Lie algebra homomorphism associated to χ, then as it is C linear, it is the complex- ification of χk. Hence, the Lie algebra of G′, g′ = ker( χ) is the complexification of ker( χ) = k′, the Lie algebra of K ′. Thus, if G′ 0 denote the corresponding connected components of the identity, then as K ′ 0 is compact, and k′ a real form for g′, then by [GW, Thm 2.4.7, Prop. 2.4.2], G′ 0 and then G′ are reductive. 0 and K ′ This completes the proof. (cid:3) Second, we consider the associated exact sequence of maximal compact sub- groups. For the maximal compact subgroup K ⊂ G, χ(K) is a connected compact subgroup of C∗, and hence is either {1} or S1. Again if χ(K) = {1}, let T0 be a maximal torus of K contained in a maximal algebraic torus T of G and homotopy equivalent to it. Since χ0T0 is trivial, it follows that χT is trivial, which is a contradiction. We let χ′ = χK and K ′ = ker(χ′), G′ ∩ K. We also let K ′ 0 be the connected component of the identity of K ′, so K ′ 0 ⊂ G′ 0. Then, we have the following diagram with rows exact, and the vertical arrows are inclusions. (2.2) 1 −−−−→ G′ 0 −−−−→ G χ −−−−→ C∗ −−−−→ 1 i ′x ix i0x 1 −−−−→ K ′ 0 −−−−→ K χ′ −−−−→ S1 −−−−→ 1 We remark that the bottom row is exact, because the covering K/K ′ is the restriction of the covering for G/G′ 0 → G/G′. 0 → K/K ′ ≃ S1 There are three things that we want to show in this situation: i) i ′ is a homotopy equivalence; ii) G′ iii) the fibration K ′ 0 acts transitively on the global Milnor fiber F ; and 0 ֒→ K → S1 is cohomologically trivial. Lemma 2.2. In the above situation in the diagram (2.2), i ′ is a homotopy equiv- alence. EXCEPTIONAL ORBIT HYPERSURFACES 11 Proof. Each of the rows of (2.2) gives fibrations: G′ 0 ֒→ K → S1. Since K is the maximal compact subgroup of G, i : K ֒→ G is a homotopy equivalence, as is the inclusion i0 : S1 ֒→ C∗. Then, we may consider the induced maps for the long exact sequences in homotopy for the two fibrations, and again by the 5-lemma, since both i and i0 are homotopy equivalences, it follows that i ′ ∗ : πj(K ′ 0) for all j ≥ 0. Since they are both path-connected CW- complexes, it follows by Whitehead's theorem that i ′ is a homotopy equivalence. (cid:3) 0 ֒→ G → C∗ and K ′ 0) ≃ πj(G′ Next, as G′ 0 acts trivially on h, it acts on the fibers of h, and in particular on 0 for a point v0 in the global Milnor fiber F . Let H ′ be the isotropy subgroup of G′ F , and let L′ ⊂ K ′ 0 be the maximal compact subgroup of H ′. Lemma 2.3. In the preceding situation, G′ fiber and K ′ finite and hence equals L′ and G′ of covering transformations H ′. 0 acts transitively on the global Milnor 0/L′ ⊂ F is a deformation retract. In the equidimensional case, H ′ is 0 is a finite regular covering space of F with group Proof. By Lemma 2.1, dim CG′ 0 = dim CG − 1 and it acts on the global Milnor fiber F of dim CF = N − 1, where dim CV = N . If v0 ∈ F , then as v0 ∈ U, the orbit map Φ : G → V sending g 7→ g · v0 is a local submersion at Id. Hence, if dΦ(Id)TIdG′ 0 has rank < N − 1, then dΦ(Id) has rank < N , contradicting it being a submersion. Thus, dΦ(Id)TIdG′ 0 preserves the fibers of h, Φ(G′ 0) ⊂ F , so its image is a neighborhood of v0. As this is true at each point v0 ∈ F , The orbit of v0 under G′ 0 is open in F . Thus, any orbit of G′ in F is open in F . As F is connected (by e.g. Kato-Matsumoto [KM]), there can only be one orbit and G′ 0 of v0, F is diffeomorphic to G′ 0/H ′. Also, by the same argument given in Lemma 1.3, K ′ 0 acts transitively on F . Hence, if H ′ is the isotropy subgroup in G′ 0/H ′ is a deformation retract, establishing the first result. 0 has rank N − 1, and as G′ 0/L′ ⊂ G′ In the equidimensional case, dim CG′ 0 = dim CF = N − 1. As they have the same dimension, the isotropy subgroup of a point is a zero dimensional algebraic subgroup, and hence finite. As G′ 0 is a covering space as stated. 0 is connected and F ≃ G′ 0/H ′, G′ (cid:3) Remark 2.4. We see now that Proposition 1.5 follows from Lemma 2.3, since 0, H ′) gives the model complex geometry for F . For the determinantal hyper- (F, G′ surfaces, the singular sets for Dsy m (m even) have codimension ≥ 2, so by Kato-Matsumoto [KM] the Milnor fibers are simply connected, so the model is simple. m , Dm, and Dsk We also remark that Lemma 2.3 gives an alternate approach to the topological structure of the Milnor fiber. Rather than trying to represent it as being homotopy equivalent to a bouquet of spheres, it gives a compact submanifold which is a deformation retract. We complete this section by establishing for a fibration of connected compact Lie groups that cohomological triviality always hold. Lemma 2.5. For an exact sequence of compact connected Lie groups (2.3) 1 −−−−→ K ′ The associated fibration K ′ χ′ 0 −−−−→ K −−−−→ S1 −−−−→ 1 0 ֒→ K → S1 is cohomologically trivial. 12 JAMES DAMON Proof. First, by the same arguments as in Lemma 2.1 but for compact groups, dim RK ′ 0 = dim RK − 1 and rank K ′ 0 = rank K − 1 = k − 1. Thus, by the basic Hopf structure theorem, the cohomology with coefficients in a field k of characteristic 0 of a compact connected Lie group of rank k is a Hopf algebra which is isomorphic to an exterior algebra on k generators of odd degrees (see e.g. Borel [Bo] which treats the more general Leray Structure theorem for Hopf algebras, where k may be a perfect field). Thus, H ∗(K; k) is isomorphic to an exterior algebra on k generators and H ∗(K ′ 0; k) is isomorphic to an exterior algebra 0; k) = 2k−1. on k − 1 generators. Thus, dim kH ∗(K; k) = 2k, and dim kH ∗(K ′ Hence, dim kH ∗(K; k) = 2 dim kH ∗(K ′ 0; k), so by Proposition 1.8 the fibration is cohomologically trivial. (cid:3) This last result by itself will not imply that the global Milnor fibration is coho- mologically trivial. In fact, we see in the next section that it fails for a family of determinantal hypersurfaces. However, the preceding will provide the crucial pieces in §5 to prove that for any prehomogeneous space defined from an equidimensional representation ρ : G → GL(V ) of a connected linear algebraic group G, the global Milnor fibration is cohomologically trivial. 3. Topology of Determinantal Hypersurfaces We first consider the determinantal hypersurfaces which arise as the exceptional orbit varieties of the representations of GLm(C) on M = Symm, Mm,m, or Skm (for m = 2k), given before Definition 1.1. We separately state the results for the topology of the Milnor fibers, and the complements and links. Theorem 3.1. The Milnor fibers of the determinantal hypersurfaces are homoge- neous spaces homotopy equivalent to compact symmetric spaces of "classical type" (classified by Cartan) given in Table 1. The cohomology with coefficients in a field k of characteristic 0 of the Milnor fiber is either an exterior algebra, or a free module on two genersators over an exterior algebra, as given in Table 1. Also, the Milnor fibration is cohomologically trivial for all cases except the case of m × m symmetric matrices with m even. Moreover, for the regular and skew-symmetric cases the formulas remain true when k is replaced by Z, while for the symmetric case there is 2-torsion in the Milnor fiber and the mod-2 cohomology of the Milnor fiber is given by the Z2-exterior algebra H ∗(SUm/SOm(R); Z2) = Λ∗ Z2hs2, s3, ..., smi with sj of degree j. Along with the topology of the Milnor fiber, we can also give the topology of the complement and the (co)homology of the link in the next theorem. Theorem 3.2. The complements M \D of the determinantal hypersurfaces D are homogeneous spaces G/H which are homotopy equivalent to quotients of maximal compact subgroups K/L, given in column 2 of Table 2. The cohomology with coef- ficients in a field k of characteristic 0 of the complement is an exterior algebra as given in column 3 of Table 2. Moreover, for the regular and skew-symmetric cases the formulas remain true when C is replaced by Z. EXCEPTIONAL ORBIT HYPERSURFACES 13 Milnor Fiber Symmetric H ∗(F, k) F Space SLm(C)/SOm(C) SUm/SOm Λ∗khe5, e9, . . . , e2m−1i SLm(C)/SOm(C) SUm/SOm Λ∗khe5, e9, . . . , e2m−3i · {1, em} SUm SLm(C) SL2k(C)/Spk(C) Λ∗khe3, e5, . . . , e2m−1i SU2k/Spk Λ∗khe5, e9, . . . , e2m−3i Table 1. The generators of the cohomology ek are in degree k; and the structure is an exterior algebra for all cases except for Dsy 2k, for which the structure is a free module on 1 and em over the exterior algebra. Determinantal Hypersurface Dsy m (m = 2k+1) Dsy m (m = 2k) Dm Dsk m , m = 2k Determinantal Hypersurface Dsy m (m = 2k+1) Dsy m (m = 2k) Dm Dsk m (m = 2k) Complement M \D H ∗(M \D, k) ≃ H ∗(K/L, k) GLm(C)/Om(C) Λ∗khe1, e5, . . . , e2m−1i ∼ Um/Om(R) GLm(C)/Om(C) Λ∗khe1, e5, . . . , e2m−3i GLm(C) ∼ Um GL2k(C)/Spk(C) Λ∗khe1, e5, . . . , e2m−3i Λ∗khe1, e3, . . . , e2m−1i m2 − 2 ∼ U2k/Spk Shift (cid:0)m+1 2 (cid:1) − 2 (cid:0)m+1 2 (cid:1) + m − 2 (cid:0)m 2(cid:1) − 2 Table 2. The cohomology of the complements M \D and links L(D) for each determinantal hypersurface D. The complements, are homotopy equivalent to the quotients of maximal compact sub- groups K/L with cohomology given in the third column, where the generators of the cohomology ek are in degree k; and the structure is an exterior algebra. For the links L(D), the cohomology is iso- morphic as a vector space to the cohomology of the complement truncated in the top degree and shifted by the degree indicated in the last column. The cohomology of the link L(D) is isomorphic as a graded vector space to the cohomology of the complement truncated in the top degree and shifted in degree. The shift is N − 2 where N = dim CM for all cases except the symmetric case with m even; and in that case the shift is N − 2 + m. Then the products in the reduced 1 ⌣ 1∗, where 1 and 1∗ are the images in degrees N − 1, resp. N − 2, of e1, 1 ∈ H ∗(M \D; C) cohomology eH ∗(L(D); C) are 0 except possibly for the single product e∗ the e∗ in all cases except for the symmetric case with m even. Example 3.3 (Simplest Determinantal Hypersurfaces which are Morse Singulari- ties). We consider the complex cohomology for the simplest examples in Table 1 of dimension > 1 and clarify the notation. First, for Dsy 2 we have m = 2 with k = 1. As 4k − 3 = 1 < 5, the exterior algebra is on 0 generators. By this we mean it is just C · 1, with 1 the identity in the cohomology algebra. The Milnor fiber then has complex cohomology Ch1, e2i. As Dsy is defined by an equation of the form 2 14 JAMES DAMON xz −y2 = 0, we see it is a Morse singularity with cohomology as stated. In addition, in Table 2, the complement has cohomology Ch1, e1i; and after truncating e1 and shifting by 3, we obtain that the reduced complex cohomology of the link L(Dsy 2 ), which is RP 3, is nonzero only in degree 3, where it is C. Next D2 is defined by an equation of the form xw − yz = 0. Again it is a Morse singularity and the table with m = 2 gives for the complex cohomol- ogy of the Milnor fiber Λ∗Che3i = Ch1, e3i. The complement has cohomology Λ∗Che1, e3i = Ch1, e1, e3, e1e3i, which when truncated and shifted gives the re- duced complex cohomology of the link L(D2) as being C in degrees 2, 3 and 5 and zero otherwise. Similarly, Dsy 4 is defined by the Pfaffian of the form xw−yv+zu = 0, which is again a Morse singularity and the Table with m = 2 gives as the complex cohomology of the Milnor fiber Λ∗Che5i = Ch1, e5i; and the reduced complex coho- 2 ) is C in degrees 4, 5 and 9 and zero otherwise. Because mology of the link L(Dsk in the later two cases the link is a compact oriented manifold, we see that the one possible nonzero product e∗ 1 ⌣ 1∗ is indeed nontrivial. Example 3.4. Beyond the cases in Example 3.3, the classical approach of rep- resenting the Milnor fiber as a bouquet of spheres no longer applies as the coho- mology algebra would be trivial with no torsion. For example, for Dsy 3 , the Table gives the complex cohomology as Λ∗Che5i = Ch1, e5i; however, we also have for Z2-cohomology, Λ∗Z2hs2, s3i = Z2h1, s2, s3, s2s3i, which gives 2-torsion in degrees 2 and 3 and a non-trivial product structure. Likewise, other higher determinantal hypersurfaces have Milnor fibers not homotopy equivalent to a bouquet of spheres. The cohomology of the complement Sym3\Dsy 3 is the exterior algebra Λ∗Che1, e5i = Ch1, e1, e5, e1e5i. Also, N − 2 = 6 − 2 = 4, so the link L(Dsy cohomology Ch1∗, e∗ to be determined. 5i with degrees 4, 5, 9. where the product e∗ 1, e∗ 3 ) has reduced complex 1 ⌣ 1∗ still has Stable Homotopy Groups of the Milnor Fibers. Third, using the periodicity Theorems of Bott, we can also give the stable homotopy groups of the Milnor fibers up to the appropriate stable range. Again this is because we have already identified them as being homotopy equivalent to the classical symmetric spaces, so the results of Bott, as applied to these spaces, yield the calculations, with it only being necessary to observe where the stable range begins in each case. For the three cases, we the corresponding infinite dimensional symmetric spaces are given by: SU = ∪∞ SU/SO = ∪∞ n=1SUn, n=1SUn/SOn, n=1SU2n/Spn . We also let the Milnor fibers of the determinantal hypersurfaces be denoted by: Fm, F sy m . Then, the homotopy groups of the Milnor fibers can be given up to the end of the stable range in terms of the stable homotopy groups of the infinite dimensional symmetric spaces as follows. m , and F sk and SU/Sp = ∪∞ Theorem 3.5. The homotopy groups of the Milnor fibers up to the end of the stable range are as follows. i) ii) πj (Fm) ≃ πj(SUm) ≃ πj (SU) for j < 2m πj(F sy m ) ≃ πj(SUm/SOm) ≃ πj(SU/SO) for j < m − 1 EXCEPTIONAL ORBIT HYPERSURFACES 15 ii) for m = 2k πj(F sk m ) ≃ πj(SU2k/Spk) ≃ πj (SU/Sp) for j < 4k − 2 where the stable homotopy groups are given in Table 3. πj(G/H) j = SU SU/SO SU/Sp 0 0 0 0 4 5 2 0 1 0 0 Z2 Z2 0 0 6 3 7 8 9 Z 0 Z Z 0 Z 0 0 Z 0 0 Z 0 0 Z 0 Z Z2 Z2 0 Table 3. The stable homotopy groups of the infinite dimensional symmetric spaces G/H. They are periodic of period dividing 8, with the periodicity beginning at j = 2. We note that for the Milnor fibers Fm, m ≥ 6, F sy m , m ≥ 11, and F sk m , m ≥ 6, exhibit all of the homotopy groups appearing in Table 3 in their stable ranges. Proofs of the Theorems. Proof of Theorem 3.1. We first observe that in all three cases the determinantal hypersurface is homogeneous defined by either det−1(0) for M = Symm or M = Mm,m or Pf −1(0) for M = Skm (for m = 2k). By Lemma 1.2, we may consider the Milnor fibers of the global Milnor fibration. The simplest case is for the Milnor fiber for the determinantal hypersurface Dm for m × m matrices Mm,m, which is just F = det−1(1) = SLm(C). It is homotopy equivalent to its maximal compact subgroup SUm, and so, for example, by [MT, Chap. 3, Thm 6.5],its cohomology is a Hopf algebra given by the exterior algebra H ∗(SUm; Z) ≃ Λ∗Zhe3, . . . , e2m−1i . where on the ei are of degree i (which correspond by transgression homomorphisms to the Chern classes). Furthermore, by replacing Z by k, a field of characteristic 0, we obtain the corresponding result for cohomology with coefficients in k. The second case Dsy m is for the m × m symmetric case M = Symm. We claim the Milnor fiber F = det−1(1), is diffeomorphic to SLm(C)/SOm(C). First, the action of SLm(C) on F by A 7→ BABT is transitive. We know by the classification of symmetric matrices under the equivalence A 7→ BABT , that if det(A) 6= 0, then there is a B ∈ GLm(C) such that BABT = Im, the m × m identity matrix. If det(B) = b 6= 0 and det(A) = 1, then det(B) det(A) det(BT ) = b2 = 1 = det(Im). Hence, if we replace B by B′ = b−1B, then B′AB′ T = b−2 ·Im = Im, and det(B′) = 1. Thus, the orbit of Im under SLm(C) is F . Moreover, the isotropy subgroup of Im under the action of SLm(C) is {B ∈ SLm(C) : B Im BT = Im}, which is SOm(C). Hence, F ≃ SLm(C)/SOm(C). Lastly, the groups SLm(C), resp. SOm(C), are homotopy equivalent to their max- imal compact subgroups, which are SUm, resp. SOm(R). Hence, by the argument given in Lemma 1.3, SLm(C)/SOm(C) is homotopy equivalent to SUm/SOm(R), which is one of the classical symmetric spaces (see [MT, Chap. 3, §6]). Thus, H ∗(SLm(C)/SOm(C)) ≃ H ∗(SUm/SOm(R)). The calculation of H ∗(SUm/SOm(R)) 16 JAMES DAMON is given in e.g. [MT, Chap. 3, Thm 6.7] depends on whether m is even or odd. If m is odd = 2k + 1, then for a field k of characteristic 0, it is an exterior algebra H ∗(SUm/SOm(R); k) = Λ∗khe5, e9, ..., e2m−1i . If m is even = 2k, then it is a free module of rank 2 over an exterior algebra H ∗(SUm/SOm(R); k) = Λ∗khe5, e9, ..., e2m−3i{1, em} In the last case e2 odd degree generators.. m 6= 0, but instead equals an expression involving products of the The remaining case is for the m × m skew-symmetric case M = Skm with In this case the global Milnor fiber F = Pf −1(1) is diffeomorphic to m = 2k. SLm(C)/Spm(C). First, the action of SLm(C) on F by A 7→ BABT is transitive. We know by the classification of skew-symmetric matrices under the equivalence A 7→ BABT , that if det(A) 6= 0, then there is a B ∈ GLm(C) such that BABT = Jm, where Jm is the 2 × 2 block diagonal matrix with 2 × 2 diagonal blocks (cid:18) 0 −1 1 0(cid:19). Again if det(B) = b 6= 0 and Pf(A) = 1, then Pf(BABT ) = det(B)Pf(A) = b = 1 = Pf(Jm). Hence, B ∈ SLm(C); thus, the orbit of Jm under SLm(C) is F . The isotropy group of Jm in SLm(C) is {B ∈ SLm(C) : BJmBT = Jm} = Spk(C). Thus, as before, F ≃ SL2k(C)/Spk(C). Lastly, the groups SL2k(C), resp. Spk(C), are homotopy equivalent to their max- imal compact subgroups, which are SU2k(C), resp. Spk. We conclude SL2k(C)/Spk(C) is homotopy equivalent to SU2k(C)/Spk, which is again one of the classical sym- metric spaces of compact type (see [MT, Chap. 3, §6]). Then, by [MT, Chap. 3, Thm 6.7] H ∗(SU2k(C)/Spk; Z) = Λ∗Zhe5, e9, ..., e2m−3i where again ei has degree i and m = 2k. We obtain the corresponding result with Z replaced by k. For the symmetric case, we again apply [MT, Chap. 3, Thm 6.7] to conclude that the mod-2 cohomology is as claimed. The only remaining claim is that the Milnor fibrations are cohomologically trivial except in the case of m × m symmetric matrices with m even. This will follow from the results in Theorem 3.2 together with Proposition 1.8. (cid:3) Proof of Theorem 3.2. The proof will follow in each case from Proposition 1.9. In the case of Mm,m, the complement is GLm(C), which has maximal compact subgroup Um with cohomology a Hopf algebra given by the exterior algebra H ∗(Um; Z) ≃ Λ∗Zhe1, e3, . . . , e2m−1i , and replacing Z by k we obtain the corresponding result for cohomology with k coefficients. Second, in the skew-symmetric case M = Skm with m = 2k, the isotropy sub- group of GLm(C) for Jm is {B ∈ GLm(C) : BJmBT = Jm} = Spm(C). They have maximal compact subgroups Um, resp. Spk(R). Second by [MT, Chap. 3, Thm 6.7], H ∗(U2k(C)/Spk; Z) = Λ∗Zhe1, e5, e9, ..., e2m−3i , and we can again replace Z by k to obtain the corresponding result for cohomology with coefficients in k. EXCEPTIONAL ORBIT HYPERSURFACES 17 Third, the symmetric case is the most subtle. We will verify it for complex cohomology and then the result will follow for any field k of characteristic 0. For M = Symm, the isotropy subgroup of GLm(C) for Im is {B ∈ GLm(C) : BImBT = Im} = Om(C). They have maximal compact subgroups Um, resp. Om(R); and Om(R) has two connected components with SOm(R) being the con- nected component of the identity. Thus, first by Proposition 1.9, H ∗(M \E; C) ≃ H ∗(Um/Om(R); C). Second by [MT, Chap. 3, Thm 6.7], we may compute the cohomology H ∗(Um/SOm(R); C), which depends on whether m is even or odd. If m is odd = 2k + 1, then it is an exterior algebra (3.1) H ∗(Um/SOm(R); C) = Λ∗Che1, e5, e9, ..., e2m−1i . If m is even = 2k, then it is a free module of rank 2 over an exterior algebra H ∗(Um/SOm(R); C) = Λ∗Che1, e5, e9, ..., e2m−3i{1, em} . (3.2) To obtain the results for H ∗(Um/Om(R); C) from H ∗(Um/SOm(R); C) we use that p : Um/SOm(R) → Um/Om(R) is a double covering obtained from Um/SOm(R) as a quotient of the action of Om(R)/SOm(R) ≃ Z2. This action is given by a covering transformation τ of Um/SOm(R) defined as follows. We let Cm denote the diagonal matrix which equals 1 except for the last entry, which is −1. Then, τ (A · SOm(R)) = A · Cm · SOm(R). This is well-defined as SOm(R) is normal in Om(R) so if A′ = A · D with D ∈ SOm(R), then A′ · Cm = A · D · Cm = A · Cm · D′ with D′ = ·C −1 m · D · Cm ∈ SOm(R). Then, Um/Om(R) is the resulting quotient. We obtain by a standard result, see Lemma 4.2, (3.3) H ∗(Um/Om(R); C) ≃ H ∗(Um/SOm(R); C)Z2 . The subtle point of the calculation is to compute the RHS of (3.3). Then, the result for the symmetric case follows from the following Lemma. Lemma 3.6. The action of τ ∗ on H ∗(Um/Om(R); C) is given by τ ∗(e4i+1) = e4i+1 for all i, and if m is even, τ ∗(em) = −em It follows from the Lemma that in the odd dimensional case m = 2k + 1 the entire cohomology is invariant under τ ∗. By contrast, in the even dimensional case m = 2k, the exterior algebra generated by all of the odd degree generators e4i+1 is invariant under τ ∗; however the elements in the module on em over the exterior algebra are all sent to their negatives. Thus, the invariant cohomology under τ ∗ is exactly the exterior algebra as stated. Before proving the Lemma, which gives the result for the symmetric case, we give the remaining argument for the link. By Proposition 1.9 we obtain the cohomology as the truncated and shifted co- homology of the complement. If K/L is orientable then the degree is shifted by 2N − 2 − dim RK/L. With the sole exception of the m × m symmetric matrices with m even, for the determinantal hypersurfaces, K/L is orientable. For example, we can see this because the cohomology computed in Table 2 is nonzero in the degree = dim RK/L. However, for m × m symmetric matrices with m even, this is not true, so K/L is not orientable so we will consider this case separately. Then, for the other cases, the complement is a quotient of reductive groups G/H with N = dim C(G/H) = dim RK/L, so the shift is given by N − 2. For the symmetric case with m even, the cohomology is still an exterior algebra, but top nonzero cohomology occurs in degree m below the top degree (N = dim RK/L), if we follow the argument in Proposition 1.9, we see that the shift must be altered to N − 2 + m, giving the result as claimed. 18 JAMES DAMON The properties of the cohomology product follow by considering degrees. We have dim RL(D) = 2N − 3, H 2N −4(L(D); C) = 0 (as H 2(K/L; C) = 0) in each case. Excluding the symmetric case with m even, the lowest positive degree classes have degrees N − 2 for 1∗ and N − 1 for e∗ 1, then the lowest dimensional products in the reduced cohomology are given by: 1∗ ∪ 1∗ of degree 2N − 4 and hence 0, and e∗ 1 ∪ 1∗ of degree 2N − 3. All other products have higher degree and hence are 0. For the symmetric case with m even, the lowest degree term 1∗ has degree N − 2 + m, so the lowest degree product will have degree 2(N − 2 + m) = 2N − 3 + (2m − 1) > 2N − 3 so all products are zero in this case. Lastly, we return to the question of the Milnor fibration being cohomologically trivial. Since H ∗(E; C) ≃ H ∗(M \E; C), the calculations of the cohomology given by Theorems 3.1 and 3.2 (and the universal coefficient theorem) imply in each case except for the symmetric case with m even, that (1.5) is satisfied, so by Proposition 1.8, the Milnor fibration is cohomologically trivial. In the symmetric case with m = 2k even, from Table 1 we see the cohomology of the Milnor fiber is a free module on two generators over an exterior algebra on k−1 generators, and has total dimension 2k; while from Table 2 the cohomology of the total space is an exterior algebra on k generators and has the same total dimension 2k. Thus, (1.5) does not hold, so the Milnor fibration for these cases is not cohomologically trivial. (cid:3) Proof of Lemma 3.6. We let τ ′ denote the diffeomorphism of Un defined by τ ′(A) = A · Cm. Then, τ ′ descends in the quotient map p : Un → Un/On(R) to yield τ . Furthermore, by [MT, Chap. 3, §6, Thm 6.7], the map p∗ in complex cohomology uniquely sends p∗(e4j+1) = e4j+1 in H ∗(Un; C) for all j with e4j+1 nonzero in H ∗(Un/On(R); C). Here we use e2j+1 to denote the exterior algebra generators of H ∗(Un; C). Since Un is connected, we can choose a path from Cm to Id, and this defines a homotopy between τ ′ and the identity map on Un. Thus, τ ′ ∗(e4j+1) = e4j+1 for all j. Then, p∗(τ ∗(e4j+1)) = τ ′ ∗(p∗(e4j+1)) = τ ′ ∗ (e4j+1) = e4j+1 = p∗(e4j−3) . By the uniqueness of p∗(e4j+1), we conclude τ ∗(e4j+1) = e4j+1 as claimed. It remains to consider in the case m = 2k, τ ∗(em) which is the Euler class e(E) of the m dimensional bundle E over Um/SOm(R) defined from the stan- dard representation of SOm(R) on Rm. We consider the fibration p′ : Um/T k → Um/SOm(R) where T k is the k-torus SO2(R)×· · ·×SO2(R) (with k factors). Then p′ ∗ : H ∗(Um/SOm(R); C) → H ∗(Um/T k; C) is injective. Also, the pull-back p′ ∗(E) splits into oriented real 2-plane bundles (or using SO2(R) ≃ U1, complex line bun- dles) L1 ⊕ L2 ⊕ · · · ⊕ Lk. Then τ ∗(e(E)) = e(τ ∗E). Since Cm is in the normalizer of T k, we can define a diffeomorphism τ ′′ of Um/T k by τ ′′(A · T k) = (A · Cm · T k). This is seen to be well-define just as was τ ′. Then, p′ ∗(τ ∗(em)) = p′ ∗(τ ∗(e(E))) = τ ′′ ∗(p′ ∗(e(E))). Also, by the splitting, j=1 τ ′′ ∗e(Lj). Now, the effect of τ ′′ on the first k − 1 factors of T k is as the identity so τ ′′ ∗e(Lj) = e(Lj) for j < k. p′ ∗(e(E)) = Qk On the last factor τ ′′ acts by multiplication by the reflection matrix C2 =(cid:18)1 j=1 e(Lj). Thus, τ ′′ ∗(p′ ∗(e(E))) = Qk 0 −1(cid:19). 0 This changes the orientation of Lk and hence changes the sign of e(Lk). EXCEPTIONAL ORBIT HYPERSURFACES 19 Thus, τ ′′ ∗e(Lk) = −e(Lk). Hence, τ ′′ ∗(p′ ∗(e(E))) = −e(E). Finally, from the above we conclude p′ ∗(τ ∗(e(E))) = τ ′′ ∗(p′ ∗(e(E))) = −p′ ∗(e(E)) . Since p′ ∗ is injective, we finally conclude τ ∗(em) = τ ∗(e(E)) = −e(E) = −em, as claimed. (cid:3) Proof of Theorem 3.5. Lastly, for Theorem 3.5, we begin by noting that the groups in Table 3 are obtained from Bott periodicity, see [MT, Chap. 4, §6,Table 4.1]. Then, as we already know that the Milnor fibers are homotopy equivalent to the specific classical symmetric spaces, it is only necessary to see that the stable ranges are as stated. This follows by standard type arguments applying the homotopy long exact sequence to the fibrations SUn ֒→ SUn+1 → S2n+1, resp. SO(n)(R) ֒→ SOn+1(R) → Sn, and Spn ֒→ Spn+1 → S4n−1, together with SOn(R) ֒→ SUn → SUn/SOn(R), and Spn ֒→ SU2n → SU2n/Spn(R) to obtain i) ii) πj(SUn) ≃ πj(SUn+1) for j ≤ 2n − 1 ; πj (SUn/SOn(R)) ≃ πj(SUn+1/SOn+1(R)) for j ≤ n − 1 ; and iii) πj(SU2n/Spn) ≃ πj(SU2(n+1)/Spn+1) for j ≤ 4n − 2 . This gives the stable range. (cid:3) 4. Cohomology of the Complement and Link for the Equidimensional Case We next return to the class of special prehomogeneous spaces defined by equidi- mensional representations ρ : G → GL(V ) of a connected linear algebraic group. In this section, we compute the topology of the complement and link of the exceptional orbit variety E. Topology of the Complement. Theorem 4.1. Consider a prehomogeneous space defined by an equidimensional representation of a connected linear algebraic group ρ : G → GL(V ) with excep- tional orbit variety E and maximal compact subgroup K. Then, for a field k of characteristic 0, (4.1) H ∗(V \E; k) ≃ H ∗(K; k) = Λ∗khs1, s2, . . . , ski . where sj are classes of odd degree qj. In addition, πi(V \E) ≃ πi(K) for i > 1; and there is a short exact sequence (4.2) 1 −−−−→ π1(K) −−−−→ π1(V \E) −−−−→ H −−−−→ 1 where H is the isotropy subgroup of G at a point v0 ∈ V \E. Proof. First, for (4.1) we may again apply the Hopf structure theorem for a compact connected Lie group K and a field k of characteristic 0, H ∗(K; k), to conclude it is a Hopf algebra which is isomorphic to an exterior algebra on classes of odd degree. This gives the structure for H ∗(K; k) in (4.1) 20 JAMES DAMON As K is homotopy equivalent to G we conclude that H ∗(G; k) is also given by the RHS of (4.1). However, let v0 ∈ U, the open orbit of G, and let H be the isotropy subgroup of v0, which is finite as dim CG = dim CV . Then, G/H ≃ U = V \E. Thus, we may replace V \E by G/H. Lastly, it is sufficient to show that (4.3) H ∗(G/H; k) = H ∗(G; k) . This follows in two steps. First, there is the standard Lemma (see e.g. [Bn, Chap III, Thm2.4]). Lemma 4.2. If the finite group H acts freely on a manifold M , then for a field k of characteristic 0 or relatively prime to H, H ∗(M ; k)H = H ∗(M/H; k) where H ∗(M ; k)H denotes the subspace invariant under the induced action of H on cohomology Hence, (4.4) H ∗(G; k)H = H ∗(G/H; k) Second, we have the averaging map (4.5) avgH : H ∗(G; k) → H ∗(G; k)H [τ ] 7→ 1 H Xσ∈H σ∗([τ ]) which is an isomorphism. This follows since as G is connected, for any cocycle τ on G, the cohomology classes [σ∗(τ )] = [τ ] for all σ ∈ G; so [avgH (τ )] = [τ ]. For the second part, we use the long exact sequence in homotopy for the fibration H ֒→ G → G/H (just as in [DP, §3]) to obtain the desired isomorphisms and the exact sequence. (cid:3) We state an immediate consequence for the case of an equidimensional represen- tation ρ : G → GL(V ) of a connected solvable linear algebraic group with an open orbit. As G has a maximal compact subgroup T k, where k = rank (G), we obtain from Theorem 4.1. Corollary 4.3. Suppose ρ : G → GL(V ) is an equidimensional representation of a connected solvable linear algebraic group which defines a prehomogeneous space with exceptional orbit variety E. If rank (G) = k, then V \E is a K(π, 1) with π a finite extension of Zk by the finite isotropy group H of a point in U; and H ∗(V \E; k) = Λ∗khs1, s2, . . . , ski . where each sj is of degree one. Remark 4.4. Since an equidimensional representation of a connected solvable linear algebraic group with open orbit U is a (possibly nonreduced) block represen- tation, the first part of Corollary 4.3 was given in [DP]. The second part extends for cohomology the results in [DP] which for the special cases of representations corresponding to (modified) Cholesky-type factorizations furthermore showed that the complement and Milnor fibers were homotopy tori. EXCEPTIONAL ORBIT HYPERSURFACES 21 Topology of the Link. We can immediately deduce the cohomology of the link of the exceptional orbit variety E. For an exterior algebra A = Λ∗khs1, s2, . . . , ski, with generators si of odd degrees qi,we let denote the algebra obtained from A by removing the top degree term, and shifting degrees upward by degree r. Then, there is the following result for the link L(E) of E. gΛ∗khs1, s2, . . . , ski [r] Theorem 4.5. Let ρ : G → GL(V ) be an equidimensional representation of a linear algebraic group defining a prehomogeneous space with exceptional orbit variety E, and a maximal compact subgroup K. Then, for a field k of characteristic 0, there is an isomorphism of graded vector spaces eH∗(L(E); k) ≃ eH ∗(L(E); k) ≃ gΛ∗khs1, s2, . . . , ski [2N − 2 − dim RK] where N = dim CV = dim CG and k = rank (G). Proof. This result follows from Proposition 1.9 using Theorem 4.1. (cid:3) Corollary 4.6. In the situation of Theorem 4.5, i) If G is solvable and 6≃ (C∗)N , then the degree shift > N − 2; hence products in the reduced cohomology eH ∗(L(D); k) are 0. cohomology eH ∗(L(D); C) are 0 except possibly for the single product e∗ ii) When G is reductive, the degree shift equals N −2, so products in the reduced 1 ∪1∗, 1 and 1∗ are the images in degrees N − 1, resp. N − 2, of where the e∗ e1, 1 ∈ H ∗(M \D; k). Proof of Corollary 4.6. If G is solvable and 6≃ (C∗)N then dim RK < dim CG = N ; hence, 2N − 2 − dim RK > N − 2. Thus the product of two classes in eH ∗(L(D); k) have degree ≥ 2N − 2 and hence is 0. For the equidimensional case when G is reductive, dim CG = dim RK and H = L is finite, so N = dim CG = dim RK/L. Hence, the degree shift is simply N − 2. Then, the argument follows as for the analogous property in Theorem 3.2. (cid:3) As a second corollary we consider the number of irreducible components of E. Each component Wi such will contribute a generator from the fundamental class (cid:2)Wi ∩ S2N −1(cid:3) to H2N −3(L(E); k). Applying Theorem 4.5 we obtain as a corollary Corollary 4.7. For a representation ρ : G → GL(V ) as in Theorem 4.5, the number of irreducible components of E equals dim H1(K; k). In particular, when G is solvable, the number equals rank (G). Proof. The generators of H2N −3(L(E); k) ≃ H 2N −3(L(E); k) . are given by the fundamental classes (cid:2)Wi ∩ S2N −1(cid:3) for the components Wi of E. Thus, by Theorem 4.5, the number of components equals dim kH1(K; k). If G is solvable of rank k, then K = T k so dim kH1(T k; k) = k (cid:3) Example 4.8. We compare the topology of the links for the exceptional orbit hypersurfaces in M2,2 defined for 2 × 2 matrices (cid:18)x y z w(cid:19): D2, the determinantal hypersurface of singular matrices (defined by xw − yz), for the representation of left 22 JAMES DAMON multiplication by GL2(C); the linear free* divisor E ′ 2 (defined by x(xw −yz)) for the representation for Cholesky factorization using the solvable group B2 × N T 2 (with B2 the Borel subgroup of lower triangular matrices and N T 2 the upper triangular unipotent matrices); and the linear free divisor E2 (defined by xy(xw − yz)) for the representation for modified Cholesky factorization using the solvable group 2 ≃ C∗ B2 × C T the invertible diagonal matrices with top entry 1) (also see [DP2] for the second and third). Table 4 gives the maximal compact subgroups K, compact model for the Milnor fiber, and the reduced cohomology of the links in the nonvanishing dimensions. These exhibit the increased complexity of the cohomology of the link, and changes in the topology of the Milnor fiber resulting from successively adding two hyperplanes to obtain the linear free divisor E2. 2 (with B2 the Borel subgroup of lower triangular matrices and C T E D2 E ′ 2 E2 h K F model xw − yz U2 SU2 ≃ S3 T 2 x(xw − yz) xy(xw − yz) T 3 S1 T 2 eH J (L(E), k) j = 2 k 0 0 4 0 k 5 3 k k k2 0 k k3 k3 Table 4. Three exceptional orbit hypersurfaces arising from equidi- mensional representations in Example 4.8, with defining equation h = 0, together with the maximal compact subgroup K of G, the compact model for the global Milnor fiber F , and the nonzero reduced cohomol- ogy groups eH J (L(E ), k). 5. Topology of the Milnor Fiber for Equidimensional Representations We next apply the preceding results on the cohomology of the complement to- gether with properties of the the Wang sequence in §1 and the results for equidi- mensional representations in §2 to compute the topology of the Milnor fiber of E. We use the notation of §§1 and 2. Theorem 5.1. Consider a prehomogeneous space defined by an equidimensional representation ρ : G → GL(V ) of a connected linear algebraic group G of rank k with maximal compact subgroup K and exceptional orbit variety E. Then, the global Milnor fibration of E is cohomologically trivial. Furthermore, for a field k of characteristic 0, H ∗(F ; k) ≃ Λ∗k < e2, . . . , ek > . Here ej = i∗(sj), where i : F ֒→ V \E is the inclusion and H ∗(V \E; k) ≃ H ∗(K; k) ≃ Λ∗k < s1, s2, . . . , sk > with deg s1 = 1. Moreover, the homotopy groups of F are given by πj (F ) ≃ πj(G) for j ≥ 2; and there is the exact sequence (5.1) 1 −−−−→ π1(G′ 0) −−−−→ π1(F ) −−−−→ H −−−−→ 1 EXCEPTIONAL ORBIT HYPERSURFACES 23 where H is the isotropy group of G′ sequence 0 for a point in F and π1(G′ 0) is in the exact (5.2) 1 −−−−→ π1(G′ 0) −−−−→ π1(G) −−−−→ Z −−−−→ 1 Proof. We can apply apply Proposition 1.9 and Theorem 4.1, to compute for the global Milnor fibration, the cohomology (5.3) where (5.4) with k = rank K. H ∗(E; k) ≃ H ∗(V \E; k) ≃ H ∗(K; k) , H ∗(K; k) ≃ Λ∗khs1, . . . , ski Second, by Lemma 2.3, G′ 0 is a finite covering space of F , so by Lemma 4.2 and then Lemma 2.2, (5.5) H ∗(F ; k) ≃ H ∗(G′ 0; k) ≃ H ∗(K ′ 0; k) Finally by Lemma 2.5, since K ′ Proposition 1.8 that dim kH ∗(K; k) = 2 dim kH ∗(K ′ we obtain 0 ֒→ K → S1 is cohomologically trivial, we have by 0; k). Combining the preceding (5.6) dim kH ∗(E; k) = 2 dim kH ∗(F ; k) . Thus, by Proposition 1.8 the global Milnor fibration is cohomologically trivial. Also, by the discussion for (1.3) in §1, there is a nontrivial class s1 ∈ H 1(E; k), and we may choose s1 to be one of the generators of the exterior algebra. Thus, combining (5.3) and (5.4) we have an isomorphism of algebras (5.7) H ∗(E; k) ≃ Λ∗khs2, . . . , ski ⊕ (hs1i · Λ∗khs2, . . . , ski) . where the product in the second summand is cup prouct. As the cohomology of the global Milnor fibration is cohomologically trivial, the second summand on the RHS of (5.7) maps by i∗ : H ∗(E; k) → H ∗(F ; k) to be 0. It follows by consideration of dimensions and Proposition 1.8 that the first summand maps by i∗ isomorphically to H ∗(F ; C). This gives the desired conclusion. For the homotopy groups πj (F ), we use the finite covering space G′ 0 → F with covering group H ′. The exact sequence (5.1) follows from the basic relation between fundamental groups for a regular covering, and the higher homotopy groups follow from e.g. the long exact homotopy sequence for a fibration. Lastly the exact sequence (5.2) follows from the long exact homotopy sequence of the fibration G′ 0 ֒→ G → C∗ since both G′ (cid:3) 0 and G are connected. Linear Free and Free* Divisors for Solvable Linear Algebraic Groups. We consider the case of an equidimensional representation ρ : G → GL(V ) defining a prehomogeneous space when G is solvable, with exceptional orbit variety E. As G has a maximal compact subgroup T k, where k = rank (G), we already know by Theorems 4.5 and 5.1 that the cohomology of both the complement V \E and Milnor fiber F are an exterior algebras on k, resp. k − 1 generators, all of which are of degree 1. As in Theorem 4.1 of [DP], we can explicitly construct the generators from the basic relative invariants pi of the representation. ρ. By Corollary 4.7, there are k = rank G irreducible components Wi of E. By [SK], the homogeneous defin- ing equations pi for Wi are basic relative invariants and are independent. We let 24 JAMES DAMON ωi = dpi pi = p∗ i ( dz homogeneous defining equation for E is given by f =Qk z ), which provide k closed one-forms on U = V \E. The reduced f defines a cohomology class in H 1(V \E; C) whose restriction ωF = 0 as df F = 0 (since f ≡ 1 on F ). i=1 pi. Then, ω = df We deduce an extension of the result of Damon-Pike [DP]. Theorem 5.2. Let ρ : G → GL(V ) be an equidimensional representation of a solvable linear algebraic group G of rank k, defining a prehomogeneous space with exceptional orbit variety E. Then, i) H 1(V \E, C) is the exterior algebra on the set of generators ωi for i = 1, . . . k. ii) H 1(F, C) is generated by the {ωi, i = 1, . . . , k} with a single relationPk 0. Hence, H ∗(F, C) is the exterior algebra on any subset of k − 1 of the ωi. i=1 ωi = Proof. We know by Theorem 4.1 that H ∗(V \E; C) is an exterior algebra on k gen- erators of degree 1. Then, we can apply same the methods in [DP, §4] to show that the k closed 1-forms ωi defined from the basic relative invariants pi are a set of gen- erators for H 1(V \E; C); and hence, are exterior algebra generators for H ∗(V \E; C). By Theorem 5.1, the complex cohomology of the global Milnor fiber is an exterior algebra on the pull-backs of k − 1 of the generators {ωi} for H 1(V \E; C). Since the i=1 ωi = ω = 0, and k − 1 of the generators will suffice. pullback of Pk (cid:3) Equidimensional Representations of Reductive Linear Algebraic Groups. In the case where G is a connected reductive linear algebraic group with maximal compact subgroup K, and ρ : G → GL(V ) defines a prehomogeneous space, then the results apply even if the exceptional orbit variety E is not a linear free divisor. However, we give one consequence for an important class of such representations for quivers of finite type considered by Buchweitz-Mond which do give linear free divisors. Quivers of Finite Representation Type. A quiver is a connected finite directed graph Γ with edges e(Γ) = {ℓj}, vertices v(Γ) = {vi}, where we denote the initial vertex for ℓj by i(ℓj) and end point by e(ℓj). To define a representation of the quiver Γ, we associate to each vertex vi a finite dimensional complex vector space Vi of dimension di and for each edge ℓj a linear transformation ϕj : Vi(ℓj ) → Ve(ℓj ). Then, the tuple {ϕj}ℓj ∈e(Γ) of linear transformations ϕj : Vi(ℓj) → Ve(ℓj ) is a quiver representation. With an ordering on the vertices {vi ∈ v(Γ)}, we let d = (di) denote the dimension vector for the quiver representation. Together a set of such transformations {ϕj} forms a quiver representation space V ≃Qℓj ∈L Hom(Vi(ℓj ), Ve(ℓj )). The group G =Qvi∈v(Γ) GL(Vi) acts on V by {ψi} · {ϕj} = {ψe(ℓj ) ◦ ϕj ◦ ψ−1 i(ℓj )} f or {ψi} ∈ G, {ϕj} ∈ V The group C∗ embeds in each GL(Vi) as the standard C∗-action. The diagonal embedding of C∗ in G defines a subgroup of G which acts trivially on V . We let G = G/C∗. Then, the quiver is said to be of finite representation type if G has only a finite number of isomorphism classes of indecomposable quiver representations. The classification of quivers of finite representation type was done by Gabriel [G, G2] and they correspond to the Dynkin diagrams of type A, D, or E with the EXCEPTIONAL ORBIT HYPERSURFACES 25 indecomposable quiver representations for appropriate dimension vectors d which are the positive Schur roots corresponding to the Dynkin diagram. The quiver arrows can go in either direction. Then, it is a fact that for these dimension vectors the representation of G on V is an equidimensional representation with an open orbit, which hence defines a prehomogeneous space whose exceptional orbit variety, denoted D(Γ,d), is called the discriminant of the quiver. Buchweitz-Mond prove in [BM] that for quivers of finite representation type the discriminant D(Γ,d) is a linear free divisor. As a result of Theorems 5.1 and 4.5, we can compute the cohomology in charac- teristic 0 of both the link and Milnor fiber of the quiver discriminant. To do so we need a simple lemma. Let G be a connected linear algebraic group with j : C∗ ֒→ G a subgroup, such that for S1 ⊂ C∗, j ′ : S1 ⊂ K for K a maximal compact subgroup of G, of rank k. By Hopf's structure theorem, (5.8) H ∗(K; k) ≃ Λ∗k < s1, . . . , sk > . with each si of odd degree qi and s1 of degree 1. Lemma 5.3. In the preceding situation, suppose that j ′ ∗(s1) generates H 1(S1; k). Then, (5.9) H ∗(G/C∗; k) ≃ Λ∗k < s2, . . . , sk > . Proof. By the same argument given in Lemma 1.3, K/S1 ֒→ G/C∗ is a homotopy equivalence. Hence it is sufficient to prove the result for K/S1. Then, {1, s1} restrict to a basis for H ∗(S1; k). Hence, the Leray-Hirsch theorem applied to the p fibration S1 ֒→ K → K/S1 yields that H ∗(K; k) is a free H ∗(K/S1; k)-module on {1, s1}, where the module structure is via p∗ and p∗ is injective. Also by (5.8), H ∗(K; k) is a free Λ∗k < s2, . . . , sk >-module on {1, s1}. Thus, H ∗(K; k)/(s1 ⌣ H ∗(K; k)) ≃ Λ∗k < s2, . . . , sk > ; and also by the Leray-Hirsch theorem, H ∗(K; k)/(s1 ⌣ H ∗(K; k)) ≃ p∗H ∗(K/S1; k) · {1} . Thus, if we compose p∗ with projection onto the quotient by cup product with s1. We obtain an isomorphism H ∗(K/S1; k) ≃ Λ∗k < s2, . . . , sk > . (cid:3) Now in the case of quivers for a graph Γ and quiver representation space V with dimension vector d, we have the subgroup C∗ ֒→ G = Q GL(Vi). The max- imal compact subgroup of GL(Vi) ≃ GLdi(C) is the unitary group Udi, which has cohomology H ∗(Udi; k) ≃ Λ∗khs(i) 1 , . . . , s(i) di i . with s(i) subgroup of G and its cohomology is given by of degree 2j − 1. Then, K = Qvi ∈v(Γ) ∈ Udi is the maximal compact j (5.10) H ∗( K; k) ≃ ⊗vi ∈v(Γ)Λ∗khs(i) 1 , . . . , s(i) di i which is again an exterior algebra. We denote the RHS of (5.10) by Λ∗(Γ, d). We let K = K/S1 which is the maximal compact subgroup of G. 26 JAMES DAMON To compute the cohomology of K we use Lemma 5.3. First, the degree 1 gener- ator s(i) 1 arises at the pull-back of the generator of H 1(S1; k) via det : Udi → S1. The composition of S1 ֒→ Udi with the determinant map sends z → zdi, which is a covering map but still induces an isomorphism on cohomology with coefficients in k. Thus, the pull-back of s(i) 1 to H 1(S1; k) via the inclusion is a generator. Hence, the pull-back of any degree 1 generatorPvi ∈v(Γ) ais(i) 1 not belonging to a certain codimension one subspace of H ∗(Udi; k) will generate H 1(S1; k). We choose such a generator s1, which can be chosen to be one of the generators of the exterior algebra. Then, we have by Lemma 5.3 (5.11) H ∗(K; k) ≃ Λ∗(Γ, d)/(s1 · Λ∗(Γ, d)) which is an exterior algebra obtained by removing a generator of degree 1 from (5.10). Second, by the discussion for (1.3) in §1 and Lemma 2.5 in §2, there is a nonzero class s2 ∈ H 1(K; k) obtained as the pull-back of the generator of H 1(S1; k) via the composition of the map χ′ : K → S1 with the projection K → K/S1 = K. This will be different from s1. Then, we have the following structure theorem for the cohomology of the Milnor fiber and link of the quiver discriminant. Theorem 5.4. For the quiver representation space V for a quiver Γ of finite type with dimension vector d, having an open orbit of indecomposable quiver represen- tations, let F(Γ,d) denote the Milnor fiber of the discriminant D(Γ,d), and L(D(Γ,d)) the link. Then, (5.12) H ∗(F(Γ,d); k) ≃ Λ∗(Γ, d)/(s1, s2) · Λ∗(Γ, d) which is the exterior algebra on the generators of (5.10) but with two degree 1 generators removed. Also, (5.13) eH ∗(L(D(Γ,d)); k) ≃ ^Λ∗(Γ, d)/(s1 · ^Λ∗(Γ, d))[dim CV − 2] which is the exterior algebra on the generators of (5.10) with one degree 1 generator removed, then truncated in the top degree, and then shifted by degree dim CV − 2. Remark 5.5. For an odd integer k > 1, the number of generators in degree k of the exterior algebras in (5.12) and (5.13) is given by {vj ∈ v(Γ) : dj ≥ k+1 2 }. While the number of degree 1 generators is either 2, resp. 1, less than v(Γ). Proof. This follows from from the preceding discussion, Lemma 5.3, together with Theorems 5.1 and 4.5. (cid:3) Example 5.6. We consider the quiver representation corresponding to D4 which has 4 vertices with a central one v1 connected to the other three by edges, with the direction toward the central vertex. The dimension vector d has dimensions dim CV1 = 2, and for the other vertices dim CVi = 1. By Mond and Buchweitz, we can view the linear transformations as determined by vectors (xi, yi) ∈ C2, for i = 1, 2, 3. The quiver discriminant consists of the triple of vectors for which at least two of them lie in a common line. This is defined by the equation (x1y2 − x2y1)(x1y3 − x3y1)(x2y3 − x3y2) = 0 in C6. The group G =(cid:0)GL(C2) × (C∗)3(cid:1) /C∗ has maximal compact subgroup K = (U2 × T 3)/S1. Then, by Theorem 5.4 we have the cohomology of the Milnor fiber (5.14) H ∗(F(D4,d); k) ≃ Λ∗khs1, s2, s3i EXCEPTIONAL ORBIT HYPERSURFACES 27 and (5.15) eH ∗(L(D(D4,d)); k) ≃ gΛ∗khs1, s2, s3, s4i[4] where we let s1 have degree 3 and the other si have degree 1. Thus, H j(FD4 ; k) is nonzero in degrees 0 ≤ j ≤ 5 and has dimensions 1, 2, 1, 1, 2, 1; while eH j(L(D(D4,d)); k) is nonzero in degrees 4 ≤ j ≤ 9 and has dimensions 1, 3, 3, 2, 3, 3. The top degree 9 of dimension 3 corresponds to the 3 irreducible components of D(D4,d). Remark 5.7. Mond and Buchweitz have examined the effect of reversing various quiver arrows in a quiver representation of finite type. David Mond has indicated they have found that for certain quiver representations, the discriminant changes when the directions of certain quiver arrows are changed. However, it follows from the results obtained here that the rational cohomology of the Milnor fiber, the complement, and the cohomology of the link (as a graded vector space) will not change. A natural question is which topological invariants will detect this change. 6. Topology of Formal Sums of Exceptional Orbit Hypersurfaces We conclude by combining the preceding results with a result of Mutsuo Oka [Ok], together with the results from Siersma et al to exhibit a large collection of highly nonisolated hypersurface singularities whose Milnor fibers are either joins of compact manifolds or bouquets of suspensions of such a join, and whose topology we can explicitly compute. Topology of Formal Sums of Exceptional Orbit Hypersurfaces. We use the notion of a formal linear combination of hypersurface singularities. We consider fi : Cni , 0 → C, 0 for i = 1, . . . , r, defining hypersurfaces Xi, 0 ⊂ Cni, 0. We i=1 Cni → Cni denote projection i=1 aifi ◦ πi. Then, f defines a i=1 Cni which we will refer to as the formal linear combination regard the Cni as distinct spaces and let πi : Qr on the i-th factor. Then, for ai ∈ C∗ we let f =Pr hypersurface in Qr of the hypersurfaces defined by the fi. We will denote it by f = ⊕r i=1aifi. We are interested in the special case where the fi define exceptional orbit hy- persurfaces Ei of the types we considered earlier. Then, we will first combine the earlier results we obtained for them with the following result of Mutsuo Oka [Ok, Thm 1] which considerably extends a classical result of Thom-S´ebastiani. Theorem 6.1 (Oka). Let g : Cn, 0 → C, 0 and h : Cm, 0 → C, 0 be weighted homogeneous germs with Milnor fibers Y , resp. Z, then f = g ⊕ h : Cn+m, 0 → C, 0 has Milnor fiber X homotopy equivalent to the join Y ∗Z. Moreover, the monodromy of f is the join of the monodromies of g and h. To use this result we recall the consequences for cohomology with coefficients in a field k of characteristic 0 (see e.g. [Ok] or [CF, Chap. 5]). Hence, we have the following isomorphism of cohomology viewed as graded vector spaces (6.2) H ∗(Y ∗ Z; k) ≃ ( H ∗(Y ; k) ⊗ H ∗(Z; k))[1] where as earlier "[1]"will denote shift increasing degrees by 1. Then, if the mon- odromy on H ∗(Y ; k) is denoted by σg and that on H ∗(Z; k), by σh, then the join (6.1) H ℓ(Y ∗ Z; k) ≃ ⊕ℓ−1 i=0 (cid:16) H i(Y ; k) ⊗ H ℓ−i−1(Z; k)(cid:17) 28 JAMES DAMON of the monodromies, denoted σg ∗ σh, which by Theorem 6.1 gives the monodromy σf , is given by the tensor product σg ⊗ σh on each summand. We may combine Oka's theorem with our earlier results. For i = 1, . . . , r, let fi : Cni, 0 → C, 0 define an exceptional orbit hypersurface Ei, 0 with global Milnor fiber Fi with model compact submanifold Mi, global Milnor fibration pi : Ei → S1, and monodromy σi on H ∗(Fi; k). Theorem 6.2. Let f : Cn, 0 → C, 0 be given by the formal linear combination i=1aifi with ai ∈ C∗, and fi as above. It defines a hypersurface E, with global ⊕r Milnor fiber F , global Milnor fibration p : E → S1, and monodromy σ. Then i) F is homotopy equivalent to the join of compact manifolds M1∗M2∗· · ·∗Mr; ii) hence, (6.3) H ∗(F ; k) ≃ (cid:16)⊗r i=1 H ∗(Mi; k)(cid:17) [r − 1]. iii) Also, the monodromy σ = σ1 ∗ σ2 ∗ · · · ∗ σr; so iv) if the Milnor fibration of each fi is cohomologically trivial, then the Milnor fibration of f is cohomologically trivial. v) In the case of iv), (6.4) H ∗(Cn\E; k) ≃ H ∗(E; k) ≃ Λ∗k{s1} ⊗(cid:16)k{1} ⊕ (⊗r i=1 H ∗(Mi; k))[r − 1](cid:17) where k{1} has degree 0 and deg(s1) = 1. Proof. For i) ii) and and iii), we use induction on the number r of terms in the formal linear combination f = ⊕r i=1aifi. If the result is true when m < r then f = g ⊕ h where g = ⊕r−1 i=1 aifi and h = arfr. As each fi is homogeneous on different coordinates, both g and h are weighted homogeneous for a common set of weights on Cn, so Oka's theorem applies and the Milnor fiber X = F of f is homotopy equivalent to Y ∗ Z with Y and Z denoting the global Milnor fibers of g, resp. h (and the Milnor fiber of h is diffeomorphic to that of fr, i.e. Fr). Also, the monodromy σf = σg ∗ σh. By the inductive assumption Y is homotopy equivalent to F1 ∗ F2 ∗ · · · ∗ Fr−1, with monodromy σg = σ1 ∗ · · · ∗ σr−1. so by Oka's theorem, X is homotopy equivalent to (F1 ∗ F2 ∗ · · · ∗ Fr−1) ∗ Fr, with monodromy σf = σ1 ∗ · · · ∗ σr−1 ∗ σr. Hence, by repeated application of (6.2) we see that H ∗(F ; k) has the indicated form as a graded vector space. Furthermore, for iv) if each global Milnor fibration pi : Ei → S1 is cohomo- logically trivial, then σi ≡ id so by iii) σf ≡ id and the Milnor fiber of f is cohomologically trivial. Thus, by Proposition 1.8 H ∗(E; k) is given by (6.4). As f is weighted homogeneous, by an analogous argument as in Lemma 1.2 it follows that Cn\E has E as a deformation retract; thus, the remainder of (6.4) follows. (cid:3) We next separately give the form of the (co)homology of the link L(E). To do so we will introduce some notation. Given a (finite dimensional) graded vector space W = ⊕Wj over the field k, we let k[m] denote k with a grading of degree m; and apply a grading to hom(W, k[m]) ≃ ⊕ hom(Wj , k[m]) so that if Wj has graded degree ℓj, then hom(Wj , k[m]) has graded degree m − ℓj. EXCEPTIONAL ORBIT HYPERSURFACES 29 Corollary 6.3. If f is given as a formal sum as above and the Milnor fiber of each fi is cohomologically trivial, then (6.5) eH ∗(L(E); k) ≃ k[2n − 3] ⊕ hom(⊗r i=1 H ∗(Mi; k), k[2n − 2])[−(r − 1)] ⊕ hom(⊗r i=1 H ∗(Mi; k), k[2n − 2])[−r] . Proof. By v) of Theorem 6.2, H ∗(Cn\E; k) is given by (6.4). Then, by the same argument given in Proposition 1.9, S2n−1\L(E) is homotopy equivalent to Cn\E. Hence, we may again apply Alexander duality as (1.11) in the proof of Proposition 1.9 to obtain the isomorphism of graded vector spaces (6.6) (6.7) Summing (6.6) over j yields an isomorphism of graded vector spaces eH 2n−2−j(L(E); k) ≃ hom(eHj(S2n−1\L(E); k), k[2n − 2]) . eH ∗(L(E); k) ≃ hom(eH∗(S2n−1\L(E); k), k[2n − 2]) By (6.4), we may decompose the RHS of (6.7) as the direct sum of graded vector spaces (6.8) hom(k{s1}, k[2n − 2]) ⊕ hom(cid:16)(⊗r i=1 ⊕ hom(cid:16)k{s1} ⊗ (⊗r H ∗(Mi; k))[(r − 1)], k[2n − 2](cid:17) H ∗(Mi; k))[r − 1], k[2n − 2](cid:17) . i=1 As graded vector spaces, the first summand is k[2n − 3], and k{s1} ⊗ (⊗r i=1 H ∗(Mi; k))[r − 1] ≃ (⊗r i=1 H ∗(Mi; k))[r] . Then, for the second and third summands in (6.8), the positive shift inside "hom" can be moved to a negative shift outside. Hence, when we add k{1} to both sides of (6.7), we obtain from (6.6) and (6.7) the isomorphism of graded vector spaces (6.5). (cid:3) Unfortunately H ∗(L(E); k) in this more general case cannot be given in such a simple form as a shifted and upper truncated cohomology of a compact orientable manifold or an exterior algebra as in iii) of Proposition 1.9 and Theorems 3.2 and 4.5. However, we can still view the cohomology as a direct sum of the shifted upper truncated exterior algebras which appear in the theorems. Example 6.4. As a simple example we consider the formal sum f = a1 det  x1 x2 x3 x4 x5 x6 x7 x8 x9   + a2y1y2 det(cid:18)y1 y3 y2 y4(cid:19) which defines a hypersurface singularity in C13. We denote the first term by f1(x) and the second by f2(y). Then, by Theorem 3.1, the Milnor fiber of f1 is ho- motopy equivalent to SU3 and has cohomology isomorphic to an exterior algebra Λ∗k(e3, e5). Also, by [DP2] (see also Example 4.8), f2 defines a linear free divisor resulting from the action of a solvable linear algebraic group and by [DP] it has Milnor fiber homotopy equivalent to torus, with cohomology an exterior algebra Λ∗k(e1, e′ 1). Thus, by Theorem 6.2 the Milnor fiber of f is homotopy equivalent to SU3 ∗ T 2 with reduced cohomology (cid:16)eΛ∗k(e3, e5) ⊗eΛ∗k(e1, e′ 1)(cid:17) [1], which is 0 from the dimensions that it is a direct sum of three shifted copies of eΛ∗k(e1, e′ in degrees < 5 and in degrees 5 ≤ ℓ ≤ 11 has dimensions 2, 1, 2, 1, 0, 2, 1. We see 1) 30 JAMES DAMON with nonzero dimensions 2, 1 or three shifted copies of eΛ∗k(e3, e5) which has its dimensions 1, 0, 1, 0, 0, 1 between degrees 3 and 8. Also, both Milnor fibrations are cohomologically trivial by Theorems 3.1 and 5.1; thus, by iii) of Theorem 6.2, the Milnor fibration of f is cohomologically trivial. Hence, by Corollary 6.3, the link L(E) has reduced cohomology which is 0 in degrees < 12 and in degrees 12 ≤ ℓ ≤ 23 has dimensions 1, 3, 2, 1, 3, 3, 3, 2, 0, 0, 0, 1. Here we see the group of lower nonzero dimensions obtained from those for the Milnor fiber written in reverse order 1, 2, 0, 1, 2, 1, 2 and added to another copy shifted by 1. Milnor Fibers Homotopy Equivalent to a Bouquet of Suspensions of Joins of Compact Manifolds. As a last step, we consider a formal sum h = f ⊕ g, where f is a formal sum of fi which define exceptional orbit hypersurfaces as considered earlier, and g is weighted homogeneous and has a Milnor fiber which is homotopy equivalent to a bouquet of spheres ∨k i=1Sni . Then, we may again apply Oka's Theorem to conclude Proposition 6.5. The Milnor fiber of h = f ⊕ g as above has the homotopy type of a bouquet of spaces, each of which is an iterated ni + 1 suspension Sni+1(∗r j=1Mj) for i = 1, . . . , k. Remark 6.6. Thus, within the class of nonisolated hypersurface singularities ob- tained from formal sums, the Milnor fibers which are a bouquet of spheres are replaced more generally by bouquets of spaces each of which are suspensions of joins of compact manifolds. Proof. Let X = ∗r j=1Mj,which is homotopy equivalent to the Milnor fiber of f . By Oka's Theorem, the Milnor fiber of h is homotopy equivalent to the join X ∗ i=1Sni). Let p denote the common point of the bouquet. Then, this join is (∨k the union of the joins X ∗ Sni , and any two intersect in the common subspace X ∗ {p}. This is a cone on X, and is hence contractible. Moreover, let Ui ⊂ Sni be a contractible neighborhood of p, with ϕi : Ui → {p} a strong deformation retraction. Then, the join of ϕi with the identity on X gives a strong deformation retraction ϕi of X ∗ Ui to X ∗ {p}. Thus, together these give a strong deformation retraction of ∪k i=1X ∗ Ui to X ∗ {p}. As X ∗ {p} has {p} as a strong deformation retract, we conclude that ∪k i=1X ∗Ui has {p} as a strong deformation retract. Hence, we may collapse ∪k i=1X ∗ Ui to {p} and retain the same homotopy type. It follows that X ∗ (∨k i=1(X ∗ Sni ), and each X ∗ Sni is homeomorphic to the ni + 1 iterated suspension Sni+1(X). Together these give the result. (cid:3) i=1Sni) is homotopy equivalent to ∨k References [Bo] [Bo2] [Bn] [Br] Borel, A. Sur les cohomologie des espaces fibr´es principaux et des espaces homogenes de groupes de Lie compacts Annals Math. 57, (1953) 115 -- 207. Sous-Groupes Compacts Maximaux des Groups de Lie, Seminaire Bourbaki, De- cember 1950 Bredon, G. E. Introduction to Compact Transformation Groups, Academic Press, 1972 Bruce, J.W. Families of Symmetric Matrices Moscow Math. Jour 3 no. 2, Special Issue in honor of V. I. Arnol'd, (2003) 335 -- 360. [BM] Buchweitz, R. O. and Mond, D. Linear Free Divisors and Quiver Representations in Singularities and Computer Algebra London math. Soc. Lect. Ser. vol 324 Cambridge Univ. Press, 2006, 41 -- 77. EXCEPTIONAL ORBIT HYPERSURFACES 31 [CF] [D1] [D2] Cooke, G. and Finney, R. Homology of Cell Complexes, based on lectures by N. E. Steen- rod, Princeton Mathematical Notes, Princeton Univ. Press (1967) Damon, J. Higher Multiplicities and Almost Free Divisors and Complete Intersections Memoirs Amer. Math. Soc. 123 no 589 (1996) Damon, J. Nonlinear Sections of Non-isolated Complete Intersections New Developments in Singularity Theory, Eds. D. Siersma, C. T. C. Wall, V. Zakalyukin, Nato Science Series 21, Kluwer Acad. Publ. (2001), 405 -- 445. [DM] Damon, J. and Mond, D. A -- Codimension and the Vanishing Topology of Discriminants Invent. Math. 106 (1991), 217 -- 242. [DP] Damon, J. and Pike, B. Solvable Group Representations and Free Divisors whose Com- plements are K(π, 1)'s, Proc. Conference on Singularities and Generic Geometry, in Top. and its Appl. 159 (2012) 437 -- 449. [DP2] Damon, J. and Pike , B. Solvable Groups, Free Divisors and Nonisolated Matrix Singu- larities I: Towers of Free Divisors, Annales de l'Inst. Fourier 65 no. 3 (2015) 1251 -- 1300. [DP3] Damon, J. and Pike, B. Solvable Groups, Free Divisors and Nonisolated Matrix Singular- ities II: Vanishing Topology, Geom. and Top. 18 no. 2 (2014) 911 -- 962. Gabriel, P. Unzerlegbare Darstellungen I Manuscripta Math. 6 (1972) 71 -- 103. R´epresentations Ind´ecomposables S´eminaire Bourbaki Expos´e 444 (1974) in [G] [G2] Springer Lect Notes. 431 (1975) 143 -- 169. [GW] Goodman, R. and Wallach, N. Representations and Invariants of the Classical Groups Cambridge Univ. Press. 1998 [GM] Goryunov, V. and Mond, D. Tjurina and Milnor Numbers of Matrix Singularities J. London Math. Soc. 72 (2) (2005), 205 -- 224. [GMNS] Granger, M., Mond, D., Nieto-Reyes, A., and Schulze, M. Linear Free Divisors and the Global Logarithmic Comparison Theorem, Annales de l'Inst. Fourier 59 (2009) 811 -- 850. [KM] Kato, M. and Matsumoto, Y. On the Connectivity of the Milnor Fiber of a Holomorphic Function at a Critical Point, Manifolds-Tokyo 1973 (Proc. Int'l. Conf., Tokyo, 1973), Univ. Tokyo Press (1975) 131 -- 136. [MT] Mimura, M. and Toda, H. Topology of Lie Groups I and II, Translations of Math. Mono- [Ne] graphs 91, Amer. Math. Soc. (1991) Nemethi, A. The Milnor fiber and the Zeta Function of Singularities of Type f = P(h, g) Compositio Math. 79 (1991), 63 -- 97. [Ma] Massey, W. A basic Course in Algebraic Topology Springer Graduate texts 127, 1991 [Ok] Oka, M. On the Homotopy Type of Hypersurfaces Defined by Weighted Homogeneous [Sa] [So] [SK] [Si] [Si1] [Si2] [Si3] [Sp] [Th] [Ti] [Z] Polynomials Topology 12 (1973), 19 -- 32. Saito, K. Theory of Logarithmic Differential Forms and Logarithmic Vector Fields J. Fac. Sci. Univ. Tokyo Sect. Math. 27 (1980), 265 -- 291. Sato, M. Theory of Prehomogeneous Vector Spaces (algebraic part) -- English translation of Sato's lecture from notes by T. Shintani, Nagoya Math. J. 120 (1990), 1 -- 34. Sato, M. and Kimura, T. A Classification of Irreducible Prehomogeneous Vector Spaces and Their Relative Invariants, Nagoya Math. Jorn. 65, (1977) 1 -- 155. Siersma, D. The Vanishing Topology of Non-isolated Singularities New Developments in Singularity Theory, Eds. D. Siersma, C. T. C. Wall, V. Zakalyukin, Nato Science Series 21, Kluwer Acad. Publ. (2001), 447- 472. Isolated Line Singularities Proc. Sym. Pure Math. 40 Part 2 (1983), 485 -- 496. Singularities with Critical Locus a One -- Dimensional Complete Intersection and Transversal Type A1 Topology and Appl. 27 (1987), 51 -- 73. A Bouquet Theorem for the Milnor Fiber Jour. Alg. Geom. 4 (1995), 51 -- 66. Spanier, E. H. Algebraic Topology, McGraw- Hill Publ. (1966) Thurston, W. Three Dimensional Geometry and Topology Princeton Univ. Press, 1997. Tibar, M. Bouquet Decomposition for the Milnor Fiber Topology 35 (1996), 227 -- 241. Zaharia, Topological Properties of Certain Singularities with Critical Locus a 2- dimensional Complete Intersection Top. and its Appl. 60 (1994) 153 -- 171. Department of Mathematics, University of North Carolina, Chapel Hill, NC 27599- 3250, USA
1610.04368	1	1610	2016-10-14T08:47:15	Teleman's classification of 2D semisimple cohomological field theories	[ "math.AG", "math.AT" ]	In his 2011 paper, Teleman proved that a cohomological field theory on the moduli space $\overline{\mathcal{M}}_{g,n}$ of stable complex curves is uniquely determined by its restriction to the smooth part $\mathcal{M}_{g,n}$, provided that the underlying Frobenius algebra is semisimple. This leads to a classification of all semisimple cohomological field theories. The present paper, the outcome of the author's master's thesis, presents Teleman's proof following his original paper. The author claims no originality: the main motivation has been to keep the exposition as complete and self-contained as possible.	math.AG	math	Teleman's classification of semisimple cohomological field theories Simone Melchiorre Chiarello∗ 6 1 0 2 t c O 4 1 ] Abstract In his paper [13] Teleman proved that a cohomological field theory on the moduli space Mg,n of stable complex curves is uniquely determined by its re- striction to the smooth part Mg,n, provided that the underlying Frobenius algebra is semisimple. This leads to a classification of all semisimple cohomo- logical field theories. The present paper, the outcome of the author's master's thesis, presents Teleman's proof following the above-mentioned paper. The au- thor claims no originality: the main motivation has been to keep the exposition as complete and self-contained as possible. . G A h t a m [ 1 v 8 6 3 4 0 . 0 1 6 1 : v i X r a Introduction 1 In this work, we will classify the semisimple cohomological field theories (CohFT), following Teleman's 2011 article [13]. A cohomological field theory is the datum of a finite dimensional C-vector space A with a nondegenerate bilinear form η on it, a nonzero vector 1 ∈ A, and for every g and n with 2g − 2 + n > 0, a C-linear homomorphism Ωg,n : A⊗n → H•(Mg,n) satisfying certain axioms related to the geometrical structure of Mg,n (the moduli space of stable complex projective curves with n marked points), namely the presence of a natural Sn-action, of sewing maps and of the forgetful map Mg,n → Mg,n−1. More details will be given in Section 3. We define a Frobenius algebra structure on A by imposing Ω0,3(v1 ⊗ v2 ⊗ v3) = η(v1·v2, v3), so we get 1 as the neutral element for the multiplication. This algebra is said to be semisimple if there exists an orthonormal basis Pi for η made of projectors, that is elements such that Pi·Pi = θ−1 and Pi·Pj = 0 for i (cid:54)= j; in this case the CohFT is said to be semisimple. The classification will be restricted to this type of theories. One property of semisimple Frobenius i θ−1 i Pi, called The classification will be carried out in three steps. First of all, we will consider algebras, that will be used in a crucial way, is that the element α =(cid:80) the space (cid:102)Mg,n which is the n-torus bundle over Mg,n corresponding to assigning a i Pi for some nonzero complex numbers θ−1 the Euler class of the algebra, is invertible. i tangent direction to the curve at each marked point, and we define (cid:101)Ωg,n : A⊗n → H•((cid:102)Mg,n) ∗[email protected], Université de Genève, Villa Battelle, route de Drize 9, 1227 Carouge (Switzerland). Partially supported by the ENS grant "Sélection Internationale". 1 and such that R(ψ)∗ = R(−ψ)−1 log(cid:101)Ω+(v) = −η(βlog(R(ψ)−11), v). with the same axioms of a cohomological field theory, properly modified with a de- formation of the sewing maps so to always have smooth curves. These theories are called fixed boundary theories and can be easily classified thanks to several funda- mental results concerning the cohomology of the smooth part Mg,n, namely Harer's stability theorem for Mg, its extension to Mg,n by Looijenga, and Madsen-Weiss's see that the answer for the first step is given by a homomorphism(cid:101)Ω+ : A → C[κj]j≥1 theorem (former Mumford's conjecture) about the stable cohomology of Mg. We will of the form (cid:101)Ω+ = exp((cid:80) j>0 φjκj) where the φj ∈ A∗ are freely chosen co-vectors, have a theory by setting (cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) =(cid:101)Ω+(αg · v1 · . . . · vn)(cid:102)Mg,n and the κj's are the κ-classes on the moduli space. For every choice of the φj's we The second step deals with CohFTs on Mg,n, again with axioms slightly modified case we get the classification by pulling back to (cid:102)Mg,n (this procedure yields indeed to always have smooth curves. These are called free boundary theories and in this a fixed boundary theory) and using the result of the first step. The classification in this case is given by a fixed boundary theory and a ψ-valued endomorphism R(ψ) ∈ End(A)[[ψ]] with R(0) = Id, satisfying the symplectic condition . The quantities involved are defined in Section 6.2. Under these conditions, the theory is given by Ωg,n(v1 ⊗ . . . ⊗ vn) =(cid:101)Ω+(αg · R(ψ1)−1(v1) · . . . · R(ψn)−1(vn))Mg,n where the ψi are the ψ-classes at the marked points. The last step is to get the full theories on Mg,n, also called Deligne-Mumford theories or nodal theories. The fact that for every smooth theory Ωg,n there is a nodal theory Mg,n that restricts to Ωg,n on the smooth part Mg,n will be an easy consequence of Givental's group action. However, the fact that this nodal theory is unique is not trivial at all. To show it, a new stratification of Mg,n is introduced. Acknowledgments This work has been written on spring 2015 as a master's thesis at the Université Pierre et Marie Curie (Paris VI) under the supervision of Dimitri Zvonkine: I would like to thank him for his infinite patience and for having taught me all I know about this subject and about how to write a paper. I also thank the jury members Alessandro Chiodo and Julien Marché. I thank Alessandro Chiodo for very useful discussions and support about my future as a PhD student. I would also like to thank the ENS for having provided the funds and the resources that have allowed me to complete my career as a master student. 2 Basic notions We collect in this section all the basic notions that will be assumed throughout the paper. This is just an expository section and contains no proofs. A good introduction can be found, for example, in [3] or in [15]. For the founding principles of moduli spaces, see Mumford's book [11]. 2 2.1 Moduli spaces of stable curves A a Riemann surface C with n distinct marked points x1, . . . , xn is stable if there is only a finite number of automorphisms of C that leave all the points xi fixed. It can be shown that a Riemann surface of genus g with n marked point is stable if and only if 2g − 2 + n > 0. We will use the notation Mg,n to denote the coarse moduli space of stable Riemann surfaces of genus g and n marked points; roughly speaking, it is the class of all stable Riemann surfaces of genus g with n marked points, modulo isomorphism. For example, M0,3 consists of just one point: any Riemann surface of genus 0 is isomorphic to P1(C), and for any two sets consisting each of three distinct points, there is an automorphism of P1(C) that sends one set into the other one. We will denote by (C, x1, . . . , xn) ∈ Mg,n a stable curve, where C is a Riemann surface of genus g and x1, . . . , xn ∈ C. The space Mg,n exists only in the case that 2g − 2 + n > 0; in this case, its complex dimension is 3g − 3 + n. We will write Mg,n for the the Deligne-Mumford compactification of Mg,n. It is a compact orbifold that has Mg,n as an open dense suborbifold. The elements of Mg,n are called stable curves of genus g with n marked points. It can be shown that any stable curve is a connected nodal curve, with only simple nodes. k i and y(cid:48)(cid:48) i i For (C, x1, . . . , xn) a nodal curve of geometric genus g with only simple nodes and n marked points distinct from the nodes, let ν : C(cid:48) → C be its normalization. on C(cid:48), while each marked point xi Then each node yi has two pre-images y(cid:48) i ∈ C(cid:48). If C = C1 ∪ . . . ∪ Ck where the Ci's are the irreducible has one pre-image x(cid:48) components of C, we have a corresponding decomposition C(cid:48) = C(cid:48) into connected smooth components, and for each i we have νCi : C(cid:48) Let (C, x1, . . . , xn) be a connected nodal curve with only simple nodes which are distinct from the marked points, let N ⊆ C be the set of its nodes, and M = {x1, . . . xn} be the set of its marked points. Each component C(cid:48) has natural markings, i = ν−1(M ) ∩ namely the k(cid:48) points of N(cid:48) . Then (C, x1, . . . , xn) is stable (that is, it belongs to Mg,n) if and only if every C(cid:48) (Ci, N(cid:48) for some gi and ni such that 2gi − 2 + ni > 0. A point which is either a node or a marked point is called a special point. and the n(cid:48) points of M(cid:48) i ) is a stable Riemann surface, that is, it belongs to Mgi,ni 1 (cid:116) . . . (cid:116) C(cid:48) i = ν−1(N ) ∩ C(cid:48) i If (C1, x1, . . . , xn1+1) ∈ Mg1,n1+1 and (C2, y1, . . . , yn2+1) ∈ Mg2,n2+1, let (cid:16) Ci. i i i (cid:116) M(cid:48) s(C1, C2) = (C1 (cid:116) C2)/ ∼ where ∼ is the equivalence relation in which xn1+1 ∼ yn2+1 and all the other points are identified with nothing but themselves. Then s(C1, C2) is a nodal curve and it can be easily shown that (s(C1, C2), x1, . . . , xn1, y1, . . . yn2) is in fact a stable curve of genus g = g1 + g2 with n = n1 + n2 marked points. Therefore, s induces a continuous map s : Mg1,n1+1 × Mg2,n2+1 → Mg,n which we will call the separating sewing map, and the operation of taking s will be also described as "sewing the curves in Mg1,n1+1 with those in Mg2,n2+1 along the points marked n1 + 1 and n2 + 1 respectively". Notice that s is an injective map. Analogously, if (C, x1, . . . , xn+2) ∈ Mg−1,n+2, let q(C) = C/ ∼ 3 where ∼ is the equivalence relation in which xn+1 ∼ xn+2 and all the other points are identified with nothing but themselves. Then q(C) is a nodal curve and it can be easily shown that (q(C), x1, . . . , xn) is in fact a stable curve of genus g with n marked points. Therefore, q induces a continuous map q : Mg−1,n+1 → Mg,n which we will call the non-separating sewing map. Notice that q is a 2-fold covering map on its image. Let us suppose that 2g − 2 + n > 0 and let (C, x1, . . . , xn+1) ∈ Mg,n+1. If the curve (C, x1, . . . , xn) obtained by cancelling the last marked point is stable, we define it to be p(C, x1, . . . , xn+1). However, (C, x1, . . . , xn) can also be not stable. This happens if the last marked point lies on an irreducible component that has genus zero and only three special points. If this is the case, we call the stabilization of (C, x1, . . . , xn) the curve obtained by contracting the whole non-stable component to a point. We define this stabilization to be p(C, x1, . . . , xn, xn+1). Therefore, p induces a continuous map p : Mg,n+1 → Mg,n called the forgetful map, and the operation of taking p will be also described as "forgetting the last marked point". Notice that p is a proper map and it is surjective. 2.1.1 Cohomology classes on Mg,n We denote by H•(Mg,n) the cohomology with complex coefficients of Mg,n consid- ered just as a topological space. We use a similar notation for Mg,n. For each (C, x1, . . . , xn) ∈ Mg,n, let (TxiC)∗ be the cotangent space to C at xi. Assigning this object to each curve of Mg,n defines a line bundle Li → Mg,n which we will call, by abuse of language, the "cotangent bundle at the i-th marked point". Its Chern class is denoted by ψi := c(Li) ∈ H 2(Mg,n) and we will call it the ψ-class at the i-th marked point. the push-forward p∗ : H•(Mg,n+1) → H•−2(Mg,n) is well-defined. We define Let us consider the forgetful map p : Mg,n+1 → Mg,n. Since it is a proper map, κm = p∗(ψm+1 n+1 ) ∈ H 2m(Mg,n) and we will call it the m-th κ-class. 2.1.2 Boundary strata Let us fix an element C ∈ Mg,n, and let SC ⊆ Mg,n be the set of curves which are diffeomorphic to C. Then SC is, by definition, the open boundary stratum cor- responding to the topological type of C. The topological closure SC of an open boudary stratum is called a closed boundary stratum. 4 If C is a smooth curve, we get SC = Mg,n, therefore Mg,n is an open boundary stratum, and we will call it the smooth stratum of Mg,n. By contrast, the com- plementary Mg,n \ Mg,n = ∂Mg,n is called the boundary of Mg,n. Except for the smooth stratum, every open or closed boundary stratum is a subset of the boudary ∂Mg,n (this justifies the term "boundary stratum"). Notice that, while a closed boundary stratum is obviously a closed subset of Mg,n, an open boundary stratum which is not the smooth stratum is never an open subset of Mg,n. It can be shown that open boundary strata are smooth suborbifolds of Mg,n, meaning that they have no self intersections. 1 ∪ . . . ∪ C(cid:48) k the connected components, and let C(cid:48) . Then it is clear (one can prove it by induction on k) that ν is simply a sequence of s and q maps that can be taken, without loss of generality, in order: Let C ∈ Mg,n and let ν : C(cid:48) → C be its normalization, with C(cid:48) = C(cid:48) i ∈ Mgi,ni (1) Since each si is injective and each qj is a covering map, we see that the sequence writ- to SC; let F be the monodromy ν = q1 ◦ . . . ◦ ql ◦ s1 ◦ . . . ◦ sm. ten above induces a covering map from(cid:81)k k(cid:89) group of this covering. Then we have i=1 Mgi,ni Therefore we also get SC (cid:39) ((cid:81)k i=1 Mgi,ni)/F . From this description, it is clear that the correspondence SC (cid:55)→ SC is a bijection between the open and the closed boundary strata. Let N be the normal bundle of SC in Mg,n, and let (x(cid:48) i ) for i = 1, . . . , l + m be the pairs of points that are sewed together by the maps in the sequence (1). Then it can be shown that i, x(cid:48)(cid:48) are, respectively, the tangent bundles at the marked points x(cid:48) and i . Therefore, by definition of ψ-classes, its Chern class is i where Tx(cid:48) x(cid:48)(cid:48) i and Tx(cid:48)(cid:48) i SC (cid:39) ( Mgi,ni)/F. i=1 N (cid:39) l+m(cid:77) i=1 ⊗ Tx(cid:48)(cid:48) i Tx(cid:48) i c(N ) = − l+m(cid:88) (ψ(cid:48) i + ψ(cid:48)(cid:48) i ). i=1 2.1.3 The stable range The stable range of the cohomology of Mg,n is the part of H•(Mg,n) of degree at most g/3 (unless when the converse is clear, all degrees, as long as dimensions and codimensions, are intended as complex). The reason for this terminology will be more clear when we will state Harer's stability theorem in the next section. In the stable range we have the following crucial theorems. Theorem 1 (Looijenga). Let g ≥ 2 and Mg = Mg,0. Then if d ≤ g/3 we have H d(Mg,n) ∼= (H•(Mg)[ψ1, . . . , ψn])d where the subscript d at the right-hand side indicates the component of degree d. 5 Proof. See, for example, Corollary 2.18 of [13]. For the improvement of the bound on the degree, see [14]. The next theorem has been known as "Mumford's conjecture" for a long time. Theorem 2 (Madsen-Weiss, 2005). Let g ≥ 2 and let Mg = Mg,0. Then if d ≤ g/3 we have H d(Mg) ∼= (C[κj]j≥1)d where the subscript d at the right-hand side indicates the component of degree d. Proof. See [10]. 2.2 The tubular neighbourhood theorem Let M be a smooth manifold and X ⊆ M a closed submanifold of codimension k. By tubular neighbourhood of X in M we mean an open neighbourhood T of X in M, endowed with a smooth map π : T → X with fibers homeomorphic to Bk, the open real ball of dimension k. We state the following technical theorem, which will be crucial in the sequel. Theorem 3 (Tubular neighbourhood theorem). Let M be a smooth manifold and X ⊆ M a closed submanifold. Let ν : N → X be the normal bundle of X in M. Then there exists a tubular neighbourhood T ⊆ M of X and an homeomorphism φ : N Proof. See Theorem 6.5 in [4]. ∼−→ T such that π ◦ φ = ν. Both T and φ are unique, up to homotopy. 3 Cohomological field theories (Our definition and treatment of cohomological field theories will be in Kontsevich- Manin's style, as in [9]. We will not use universal classes as in [13], but it can be seen that the two definitions are equivalent.) Let q : Mg−1,n+2 → Mg,n be the map that sews the points labeled n + 1 and n + 2, and let s : Mg1,n1+1 × Mg2,n2+1 → Mg,n be the map that sews together two surfaces at the points marked n1 + 1 and n2 + 1 respectively; here g = g1 + g2 and n = n1 + n2. Finally let p : Mg,n+1 → Mg,n be the map that forgets the marked point labeled n + 1 and then stabilizes the resulting curve. 6 3.1 Nodal theories We start from nodal theories since they have the simplest sewing axioms. Definition 1. Let A be a finite-dimensional C-vector space, 1 ∈ A a non-zero vector, and η = ηµνeµ ⊗ eν a nondegenerate symmetric bilinear form on A with inverse bi- vector ηµνeµ ⊗ eν. A nodal cohomological field theory (nodal CohFT) with base (A, η, 1) is a set of linear maps for every nonnegative integer numbers n, g such that 2g − 2 + n > 0, that satisfy the following conditions: Ωg,n : A⊗n → H•(Mg,n) 1. Ωg,n is Sn-equivariant for every g, n; 2. Ω0,3(1 ⊗ u ⊗ v) = η(u, v), for every u, v ∈ A; 3. q∗Ωg,n(v1⊗. . .⊗vn) = ηµνΩg−1,n+2(v1⊗. . .⊗vn⊗eµ⊗eν), for any v1, . . . , vn ∈ A; 4. s∗Ωg,n(v1 ⊗ . . .⊗ vn) = ηµνΩg1,n1+1(v1 ⊗ . . .⊗ vn1 ⊗ eµ)× Ωg2,n2+1(vn1+1 ⊗ . . .⊗ vn ⊗ eν), for any v1, . . . , vn ∈ A; 5. p∗Ωg,n(v1 ⊗ . . . ⊗ vn) = Ωg,n+1(v1 ⊗ . . . ⊗ vn ⊗ 1), for any v1, . . . , vn ∈ A. The summation over µ and ν is always assumed. Remark 1. We explain better the meaning of Sn-equivariance. Let ρ ∈ Sn be a permutation of the index set {1, . . . , n}. Since we have an action of Sn on Mg,n by permutation of the marked points, we have an induced cohomology map ρ∗ : H•(Mg,n) → H•(Mg,n). The first axiom then means that for every ρ ∈ Sn, we must have Ωg,n(vρ(1) ⊗ . . . ⊗ vρ(n)) = ρ∗Ωg,n(v1 ⊗ . . . ⊗ vn) 3.2 Fixed boundary theories We now define fixed boundary theories. They are smooth theories, meaning that they deal with the moduli space of smooth Riemann surfaces. Notation 1. For every g and n such that 2g − 2 + n > 0, let I ⊆ {1, . . . , n} and let πI : (cid:103)MI g,n → Mg,n be the torus bundle defined as follows. For a complex vector space V , we call S(V ) its spherization, defined by i∈I S(TxiC) S(V ) = (V \ {0})/R+. where each TxiC is the complex tangent line to C at the point xi. If C ∈ Mg,n with marked points x1, . . . , xn, then the fiber of πI at C is(cid:81) If I = {1, . . . , n} we will simply write (cid:102)Mg,n for (cid:103)MI We also write πI : (cid:102)Mg,n → (cid:103)MI I (C) is(cid:81) We write (cid:102)MI for the restriction of πI to Mg,n. g,n of π−1 i /∈I S(TxiC). g,n and π for πI. g,n for the Tn−I-bundle whose fiber at each point 7 Remark 2. Each S(TxiC) is a circle bundle endowed with a natural action of S1, the set of complex numbers with norm 1, defined by z · [v] = [zv] for z ∈ S1 and [v] ∈ S(TxiC). Let I = {i1, . . . , iI}, let (C, x1, . . . xn) ∈ Mg,n and [vij ] ∈ S(Txij C). We have an action of TI = (S1)I on (cid:102)MI defined by (z1, . . . , zI) · (C, [vi1], . . . , [viI]) = (C, [z1vi1], . . . , [zIviI]). g,n Thus πI is a principal TI-bundle. Remark 3. It is clear that this bundle can be thought of as a way of "attaching a tangent direction to each marked point", thus (cid:102)Mg,n is the moduli space of Riemann Definition 2. If I ⊆ {1, . . . , n} and i ∈ I, the point xi is a framed point of (cid:102)MI surfaces with marked points and tangent directions for each of them. ; if g,n i /∈ I, the point xi is a free point. Let I1 = {1, 2, . . . , n1} and I2 = {1, 2, . . . , n2} and let s : (cid:102)MI1 g1,n1+1 × (cid:102)MI2 g2,n2+1 → (cid:102)Mg,n be the map that sews the points marked n1 + 1 and n2 + 1 respectively on the first and second curve. Note that these points are free, so that we do not have to worry about what it means sewing two framed points yet. Let (cid:101)S be the image of s. Remark 4. This image (cid:101)S ⊆ (cid:102)M is a boundary stratum of codimension 1. In fact, if s is the usual sewing map, we can draw the following commutative diagram: (cid:102)MI1 g1,n1+1 × (cid:102)MI2 (cid:121) πI1×πI2 g2,n2+1 s−−−−→ (cid:102)Mg,n (cid:121)π s Mg,n Mg1,n1+1 × Mg2,n2+1 −−−−→ tubular neighbourhood theorem, there exists a tubular neighbourhood (unique up boundary stratum of codimension 1, and taking the pre-image along a fiber bundle does not change the codimension. thus (cid:101)S is nothing but the pre-image π−1(S) of the image S of s, which is indeed a Notation 2. We denote by ν : (cid:101)N → (cid:101)S the normal bundle to (cid:101)S = Im(s). By the to isotopy) Y of (cid:101)S and a homeomorphism φ : Y → (cid:101)N that commutes with the respective inclusions. Thus we get a map ν ◦ φ : Y → (cid:101)S. As a consequence of the the identification φ, ∂Y corresponds to a circle subbundle (cid:103)∂N of (cid:101)N. In the sequel, will write (cid:101)N to denote both the normal bundle to (cid:101)S and the associated tubular the map that we have previously written ν ◦ φ. With (cid:103)∂N we will mean both the circle subbundle of (cid:101)N and the circular neighbourhood of (cid:101)S that we have previously tubular neighbourhood theorem, the boundary ∂Y is homeomorphic to a subspace of Y by means of a homeomorphism ψ such that ν ◦ φ ◦ ψ = ν ◦ φ. Therefore under we suppose to have fixed Y and φ, and we forget about the homeomorphism: we neighbourhood Y . We therefore write ν to denote both the normal bundle map, and denoted by ∂Y . 8 Lemma 1. Let S1 act on (cid:102)Mg1,n1+1 at the (n1 + 1)-th marked point and on (cid:102)Mg2,n2+1 at the (n2 + 1)-th marked point respectively, as in Remark 2. Let ν : (cid:103)∂N → (cid:101)S be the circle subbundle of (cid:101)N as in Notation 2, and let (cid:102)Mg1,n1+1 ×S1 (cid:102)Mg2,n2+1 be the quotient of (cid:102)Mg1,n1+1 × (cid:102)Mg1,n2+1 by the identification (C(cid:48)([v1], . . . , [vn1], [zv(cid:48)]), C(cid:48)(cid:48)([vn1+1, . . . , [vn], [v(cid:48)(cid:48)])) ∼ ∼ (C(cid:48)([v1], . . . [vn1], [v(cid:48)], C(cid:48)(cid:48)([vn1+1, . . . , [vn], [zv(cid:48)(cid:48)])). f : (cid:102)Mg1,n1+1 ×S1 (cid:102)Mg2,n2+1 → (cid:101)S Then the map that forgets the tangent directions and sews the points marked n1 + 1 and n2 + 1 respectively, is a bundle map isomorphic to ν. Proof. Since by Remark 4, (cid:101)N = π∗N where N is the normal bundle of S in Mg,n, we see that if C = s(C(cid:48), C(cid:48)(cid:48)) ∈ (cid:101)S then the fiber of (cid:101)N at C is Txn1+1C(cid:48) ⊗C Txn2+1C(cid:48)(cid:48). We define a map (cid:101)N → (cid:102)Mg1,n1+1 ×S1 (cid:102)Mg2,n2+1 by writing (s(C(cid:48), C(cid:48)(cid:48)), [v1], . . . , [vn], v(cid:48)⊗v(cid:48)(cid:48)) (cid:55)→ ((C(cid:48), [v1], . . . , [vn1], [v(cid:48)]), (C(cid:48)(cid:48), [vn1+1], . . . , [vn], [v(cid:48)(cid:48)])) This maps is surjective and becomes bijective when restricted to(cid:103)∂N. The fact that this gives an isomorphism between the two bundles is clear by the definition of this map. Proposition 1. There exists a sewing-smoothing map σ : (cid:102)Mg1,n1+1 × (cid:102)Mg2,n2+1 → (cid:102)Mg,n and a circular neighbourhood ν :(cid:103)∂N → (cid:101)S as in Notation 2, such that 1. Im(σ) ⊆(cid:103)∂N, 2. ν ◦ σ = s ◦ (πI1 × πI2). Proof. The map σ is obtained by composing the quotient map (cid:102)Mg1,n1+1 × (cid:102)Mg2,n2+1 → (cid:102)Mg1,n1+1 ×S1 (cid:102)Mg2,n2+1 with the isomorphism of Lemma 1(cid:102)Mg1,n1+1 ×S1 (cid:102)Mg2,n2+1 →(cid:103)∂N The verification of properties 1 and 2 is immediate. To define a cohomological field theory, we must be able to deal with self-sewing maps as well. Thus, let J = {1, . . . , n} and let be the map that sews together the points marked n + 1 and n + 2. We stress again the fact that they are free points. Let (cid:101)T be the image of q. Then, as in Remark 4, (cid:101)T is a boundary stratum of (cid:102)Mg,n of codimension 1, and the proof is exactly the same. q : (cid:102)MJ g−1,n+2 → (cid:102)Mg,n 9 identifications can be found in Notation 2. Notation 3. We call νq : (cid:101)P → (cid:101)T the normal bundle of (cid:101)T and the relative tubu- lar neighbourhood given by the tubular neighbourhood theorem. We call (cid:102)∂P be the boundary of (cid:101)P and the circular subbundle of νq. The justification for these on (cid:102)Mg−1,n+2. We have a Z2-action on (cid:101)P (cid:48) generated by bundle (cid:101)P (cid:48) = Txn+1 ⊗ Txn+2 Then the normal bundle (cid:101)P to (cid:101)T is, by definition, the quotient of (cid:101)P (cid:48) by this action. Its fiber over a curve C ∈ (cid:102)Mg,n is Txn+1C(cid:48) ⊗ Txn+2C(cid:48) where C(cid:48) ∈ (cid:102)Mg−1,n+2 is any ((C, x1, . . . , xn, xn+1, xn+2), (u ⊗ v)) (cid:55)→ ((C, x1, . . . , xn, xn+2, xn+1), (v ⊗ u)). Remark 5. Recall that we are talking about orbifolds: let us consider the vector of the two normalizations of C. The tubular neighbourhood theorem still holds for this orbifold version of the normal bundle. This is perhaps the best place to state, without proof, the following fundamental theorem. Theorem 4 (Harer's stability). Let s0 : (cid:102)Mg,n → (cid:102)Mg,n+1 be the map that sews together the curves of(cid:102)Mg,n and (cid:102)M0,3 (which is a space consisting of just one point) along the points marked n and 3 respectively. Let p : (cid:102)Mg,n+1 → (cid:102)Mg,n be the forgetful map. Then, if k ≤ g/3, the induced map in cohomology 0 : H k((cid:102)Mg,n+1) → H k((cid:102)Mg,n) s∗ is an isomorphism, with inverse p∗. Proof. See [8]. By this theorem, we call the range of degree less than g/3 the stable range of the cohomology of Mg,n. The proofs of the following lemma, corollary and proposition are very similar to those of Lemma 1 and Proposition 1 respectively, therefore they will be omitted. Lemma 2. Let ((cid:102)Mg−1,n+2)S1 be the quotient space of (cid:102)Mg−1,n+2 obtained by identi- fying (C, [v1], . . . , [z · vn+1], [vn+2]) with (C, [v1], . . . , [vn+1], [z · vn+2]) for z ∈ S1 that acts by multiplication on S(Txi). Then the map g : ((cid:102)Mg−1,n+2)S1 → (cid:101)T that forgets the tangent directions and then applies q is isomorphic to the circle subbundle νq : (cid:102)∂P → (cid:101)T . Proposition 2. There exists a sewing-smoothing map τ : (cid:102)Mg−1,n+2 → (cid:102)Mg,n and a circular neighbourhood νq : (cid:102)∂P → (cid:101)T as in Notation 3, such that 1. Im(τ ) ⊆ (cid:102)∂P , 2. νq ◦ τ = q ◦ πJ 10 We are now ready to define fixed boundary cohomological field theories. Definition 3. Let A be a C-vector space of finite dimension, 1 ∈ A a non-zero vector, and η = ηµνeµ ⊗ eν a nondegenerate symmetric bilinear form on A with inverse co- form ηµνeµ⊗eν. A fixed boundary cohomological field theory (fixed boundary CohFT) with base (A, η, 1) is a set of linear maps (cid:101)Ωg,n : A⊗n → H•((cid:102)Mg,n) for every nonnegative integer numbers n, g such that 2g − 2 + n > 0, that satisfy the following conditions: 1. (cid:101)Ωg,n is Sn-equivariant for every g, n (compare with Remark 1); 2. (cid:101)Ω0,3(1 ⊗ u ⊗ v) = η(u, v) ∈ H 0((cid:102)M0,3), for any u, v ∈ A; 3. τ∗(cid:101)Ωg,n(v1⊗. . .⊗vn) = ηµν(cid:101)Ωg−1,n+2(v1⊗. . .⊗vn⊗eµ⊗eν), for any v1, . . . , vn ∈ A, 4. σ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = ηµν(cid:101)Ωg1,n1+1(v1 ⊗ . . . ⊗ vn1 ⊗ eµ) ×(cid:101)Ωg2,n2+1(vn1+1 ⊗ 5. p∗(cid:101)Ωg,n(v1 ⊗ . . .⊗ vn) =(cid:101)Ωg,n+1(v1 ⊗ . . .⊗ vn ⊗ 1), for any v1, . . . , vn ∈ A, where p : (cid:102)Mg,n+1 → (cid:102)Mg,n is the map that forgets the last marked points, together . . . ⊗ vn ⊗ eν), for any v1, . . . , vn ∈ A, where σ is the map of Proposition 1; where τ is the map of Proposition 2; with the tangent direction attached to it. The summation over µ and ν is always assumed. 3.3 Free boundary theories Now we consider free boundary theories. This means that we don't assign any tangent vector to the marked points and we look at homomorphisms Ωg,n : A⊗n → H•(Mg,n) Notice that we are still talking about smooth theories, that is, theories in which only the cohomology ring of the smooth part of the moduli space of curves is involved. Let s : Mg1,n1+1 × Mg2,n2+1 → Mg,n be the map that sews together the points marked n1 + 1 and n2 + 1 respectively on the two curves, and let S be the image of s. Notation 4. We call νs : Ns → S both the normal bundle to S and a fixed tubular neighbourhood of S given by the tubular neighbourhood theorem. We call νs : ∂Ns → S the boundary of the tubular neighbourhood Ns and the circle subbundle of νs given by the tubular neighbourhood theorem. Let q : Mg−1,n+2 → Mg,n be the self-sewing map, and let T be its image. We call νq : Nq → T 11 both its normal bundle and its respective tubular neighbourhood given by the tubular neighbourhood theorem. We call νq : ∂Nq → T both the circular neighbourhood boundary of Nq and the restriction of the normal bundle to the circle subbundle given by the tubular neighbourhood theorem. Remark 6. Clearly, S (cid:39) Mg1,n1+1 × Mg2,n2+1. However, the self-sewing map q : Mg−1,n+2 → T is a 2-sheeted covering, where two marked curves (C1, x1, . . . , xn+2), (C2, y1, . . . , yn+2) ∈ Mg−1,n+2 belong to the same fiber of q if and only if C1 = C2, xi = yi for 1 ≤ i ≤ n and xn+1 = yn+2, xn+2 = yn+1. Therefore the automorphism ρ of Mg−1,n+2 that switches the two last marked points is the only nontrivial automorphism of the covering q. By the general theory of covering maps, a cohomology class α on Mg−1,n+2 is ρ-invariant if and only if it is the pull-back of a cohomology class on T . Indeed, under the hypothesis of ρ-invariance, we have α = q∗(cid:18) 1 (cid:19) q∗α 2 where q∗ is the push-forward map in cohomology, which is well-defined since q is proper. Let us now define the free boundary cohomological field theories. Definition 4. Let A be a C-vector space of finite dimension, 1 ∈ A a non-zero vector, and η = ηµνeµ ⊗ eν a nondegenerate symmetric bilinear form on A with inverse co- form ηµνeµ ⊗ eν. A free boundary cohomological field theory (free boundary CohFT) with base (A, η, 1) is a set of linear maps Ωg,n : A⊗n → H•(Mg,n) for every nonnegative integer numbers n, g such that 2g − 2 + n > 0, that satisfy the following conditions: 1. Ωg,n is Sn-equivariant for every g, n; 2. Ω0,3(1 ⊗ u ⊗ v) = η(u, v), for every u, v ∈ A; 3. Ωg,n(v1 ⊗ . . . ⊗ vn)∂Nq = ν∗ 2 q∗ηµνΩg−1,n+2(v1 ⊗ . . . ⊗ vn ⊗ eµ ⊗ eν)), for any q ( 1 v1, . . . , vn ∈ A (see Remark 6 for the reason of the map 1 2 q∗); 4. Ωg,n(v1 ⊗ . . .⊗ vn)∂Ns = ν∗ . . . ⊗ vn ⊗ eν), for any v1, . . . vn ∈ A; s ηµνΩg1,n1+1(v1 ⊗ . . .⊗ vn1 ⊗ eµ)× Ωg2,n2+1(vn1+1 ⊗ 5. p∗Ωg,n(v1 ⊗ . . .⊗ vn) = Ωg,n+1(v1 ⊗ . . .⊗ vn ⊗ 1), for any v1, . . . , vn ∈ A, where p is the map that forgets the last marked point. Remark 7. In axiom 4, we are identifying S with Mg1,n1+1 × Mg2,n2+1 by means of the map s, which is an homeomorphism with its image. 12 The next proposition allows us to lift free boundary CohFTs to fixed boundary ones along the torus bundle. Proposition 3. Let (Ωg,n)g,n be a free boundary CohFT , and let π : (cid:102)Mg,n → Mg,n be the map forgetting the tangent directions. Then (π∗Ωg,n)g,n define a fixed boundary CohFT . Proof. We have to show that the classes (π∗Ωg,n)g,n satisfy the axioms of a cohomo- logical field theory with fixed boundaries. The bilinear form η and the unity vector 1 stay the same in the two cases and the axiom involving the forgetful map is trivially satisfied, thus we only have to deal with the sewing axiom. We have to show that σ∗π∗Ωg,n = η−1π∗Ωg1,n1+1 × π∗Ωg2,n2+1 where σ is as in Proposition 1 and η−1 = ηµνeµ ⊗ eν is inserted at the entries corresponding to the points to be sewed. If Ns is the normal bundle to S, then (cid:101)N = π∗Ns where ν : (cid:101)N → (cid:101)S is the normal bundle to (cid:101)S (notations as in Lemma 1). Thus we can suppose to have chosen a circular neighbourhoods such that ∂Ns = π((cid:103)∂N ), therefore Im(π ◦ σ) ⊆ ∂Ns by Proposition 1. Therefore we have, by the property of the Ωg,n's of satisfying the sewing axiom σ∗π∗Ωg,n = σ∗π∗Ωg,n∂Ns = σ∗π∗ν∗ s (η−1Ωg1,n1+1 × Ωg2,n2+1) but νs ◦ π = π ◦ ν, and by Proposition 1, ν ◦ σ = s ◦ (πI1 × πI2), and in turn π ◦ s = s ◦ (πI1 × πI2), and trivially (πI1 × πI2) ◦ (πI1 × πI2) = π × π, thus altogether we have νs ◦ π ◦ σ = s ◦ (π × π). Because of Remark 7, the map s∗ accounts for the identification of S with Mg1,n1+1 × Mg2,n2+1, therefore can omit it in the sewing axiom and we can rewrite the right-hand side of the equality above as (π × π)∗(η−1Ωg1,n1+1 × Ωg2,n2+1) = η−1(π∗Ωg1,n1+1 × π∗Ωg2,n2+1) that is, the classes (π∗Ωg,n)g,n satisfy the separating sewing axiom. Let now q : Mg−1,n+2 → T be the self-sewing map and τ : (cid:102)Mg−1,n+2 → (cid:102)Mg,n as in Proposition 2. We have to show that τ∗π∗Ωg,n = η−1π∗Ωg−1,n+2 with η−1 inserted at the entries corresponding to the points to be sewed. Just like before, we can suppose to have chosen circular neighbourhoods (cid:102)∂P of (cid:101)T (see Notation 3) and ∂Nq of T such that π((cid:102)∂P ) ⊆ ∂Nq, therefore Im(π ◦ τ ) ⊆ ∂Nq by Proposition 2. Therefore we have, by the property of the Ωg,n's of satisfying the sewing axiom, (cid:19) q∗η−1Ωg−1,n+2 (cid:18) 1 2 By the same reasoning as in the separating case, we have νq ◦ π◦ τ = q◦ π; moreover, since by hypothesis Ωg−1,n+2 is Sn+2-invariant, we have by Remark 6 that τ∗π∗Ωg,n = τ∗π∗Ωg,n∂Nq = τ∗π∗ν∗ (cid:19) q q∗(cid:18) 1 2 q∗Ωg−1,n+2 = Ωg−1,n+2 which finally implies that (π∗Ωg,n)g,n satisfy the non-separating sewing axiom as well. This completes the proof. 13 The following proposition implies, with the previous one, that these axioms agree with those of a nodal CohFT. Proposition 4. Let Ωg,n : A⊗n → H•(Mg,n) for 2g − 2 + n > 0 be homomorphisms defining a nodal CohFT. Then the restrictions of Ωg,n to Mg,n define a smooth CohFT with free boundaries. Proof. Let ι : Mg,n → Mg,n be the inclusion, we have to show that ι∗Ωg,n satisfy the axioms of a free boundary CohFT. The only nontrivial verification is the one about the sewing axiom. Let us consider a sewing map s : Mg1,n1+1 × Mg2,n2+1 → Mg,n and let S be its image. Then, since ∂Ns ⊆ Mg,n, we have (again identifying the circular neighbourhood ∂Ns with the respective spherized bundle) ι∗Ωg,n(v1 ⊗ . . . ⊗ vn)∂Ns = Ωg,n(v1 ⊗ . . . ⊗ vn)∂Ns = ν∗ s Ωg,n(v1 ⊗ . . . ⊗ vn)S ∼= S and of course we can write now we have the isomorphism s : Mg1,n1+1×Mg2,n2+1 s = s ◦ (ι × ι) where s is the sewing map on the Deligne-Mumford compactifications (this map is involved in the separating sewing axiom of a nodal CohFT). Under the isomorphism s we have Ωg,n(v1⊗. . .⊗vn)S = ηµνι∗Ωg1,n1+1(v1⊗. . .⊗vn1⊗eµ)×ι∗Ωg2,n2+1(vn1+1⊗. . .⊗vn⊗eν) by the separating sewing axiom that the Ωg,n's satisfy being a nodal CohFT, and pulling back by νs we have exactly the separating sewing axiom for smooth fixed boundary theories. For the non-separating sewing map q : Mg−1,n+2 → Mg,n, with the notation of the preceding part, we have ι∗Ωg,n(v1 ⊗ . . . ⊗ vn)∂Nq = Ωg,n(v1 ⊗ . . . ⊗ vn)∂Nq = ν∗ q Ωg,n(v1 ⊗ . . . ⊗ vn)T Obviously we have q = q◦ι where q is the non-separating sewing map on the Deligne- Mumford compactification (this map is involved in the non-separating sewing axiom of a nodal CohFT). Since q : Mg−1,n+2 → T is a 2-sheeted covering, and since every Ωg,n is Sn-invariant, we have by Remark 6 Ωg,n(v1⊗. . .⊗vn)T = q∗(q∗Ωg,n(v1⊗. . .⊗vn)T ) = 1 2 q∗(ηµνι∗Ωg−1,n+2(v1,⊗ . . .⊗vn⊗eµ⊗eν)) 1 2 by the non-separating sewing axiom that the Ωg,n's satisfy being a nodal CohFT. Pulling back by νs we have exactly the non-separating sewing axiom for smooth fixed boundary theories, and the proof is complete. 4 Givental's group action In this section, we introduce a particular group action on the category of all nodal CohFTs that allows one to build new CohFTs from old ones. To do so, we must investigate more deeply the structure of the boundary strata of Mg,n, by introducing the concept of dual graph. 14 4.1 Dual graphs Let S ⊆ Mg,n be an open boundary stratum. Let C ∈ S be a stable curve and let ν : C(cid:48) → C be its normalization. We associate the following objects to C: 1. the set V of the components of C(cid:48) and, for each component v ∈ V , its genus g(v); 2. the set H of the special points of C(cid:48) (recall that they are the points in ν−1(p) where p ∈ C is either a marked point or a node); 3. a function v : H → V that sends every special point to the component of C(cid:48) it belongs to; 4. an involution ι : H → H that sends every special point to its conjugate (that is, ι(h) = h(cid:48) if {h, h(cid:48)} = ν−1(p) where p ∈ C is a node, and ι(h) = h if {h} = ν−1(p) where p ∈ C is a marked point); let then E = {{h, ι(h)} h ∈ H and ι(h) (cid:54)= h} and L = {h ∈ H ι(h) = h}; Definition 5. Let C ∈ Mg,n be a stable curve. The dual graph of C is the graph ΓC with vertex set V , edge set E and leg set L, such that • two vertices v1 and v2 are linked by an edge e ∈ E if and only if v(e) = {v1, v2}; • a leg l ∈ L belongs to the vertex v if and only if v(l) = v. An automorphism of ΓC is a permutation of the sets V and H which leaves invariant g, v and ι. The group of automorphisms of ΓC is denoted by Aut(ΓC). Remark 8. If C ∈ Mg,n is a stable curve with normalization ν : C(cid:48) → C, then its graph ΓC is connected because C is connected, that is, we can reach any component of its from any other one, by passing through its nodes. Dually, this means that we can reach any vertex in ΓC from any other one, by passing through its edges. Remark 9. Let C ∈ Mg,n be a stable curve and for each vertex v of ΓC, let n(v) = v−1(v) be its valence. Then the stability condition on C says that, for each vertex v of ΓC, we must have 2g(v) − 2 + n(v) > 0 Remark 10. All the curves in an open boundary stratum S ⊆ Mg,n have the same dual graph. Indeed, if C ∈ S is any curve in the stratum with dual graph ΓC, let n(v) be its valence as in remark 9 for each v ∈ V . Then S is the image of a map (cid:89) v∈V Mg(v),n(v) → Mg,n ξΓC : defined in the following way: • If there are k edges e1, . . . , ek with ei = {hi, ι(hi)} between two distinct vertices v1 and v2, then we sew together Mg(v1),n(v1) and Mg(v2),n(v2) along the special points hi and ι(hi) respectively, for 1 ≤ i ≤ k. 15 • If there are k edges e1, . . . , ek with ei = {hi, ι(hi)} that link v to itself (this means that v(ei) = {v} for every i), then we apply a non-separating sewing map on Mg(v),n(v) by sewing together the special points hi and ι(hi) for 1 ≤ i ≤ k. give rise to a graph which It is then evident that all the elements of the image of ξΓC is the same as ΓC. Thus this remark justifies the following definition. Definition 6. Let S ⊆ Mg,n be an open boundary stratum. Then its dual graph ΓS is the dual graph ΓC of any curve C ∈ S. Let T ⊆ Mg,n be a closed boundary stratum. Then its dual graph ΓT is the dual graph ΓS of the open boundary stratum S such that T = S. Since the correspondence S (cid:55)→ S is a bijection, in the sequel we forget the distinction between open and closed strata, when we talk about the dual graphs associated to them. Remark 11. If S is a closed boundary stratum, with dual graph ΓS, then we can define a map (cid:89) v∈V ξΓS : Mg(v),n(v) → Mg,n such that with image S, in the same way as in Remark 10. Remark 12. The genus of a curve C ∈ Mg,n is the sum of the genera of its compo- nents, plus the maximal length k of a sequence of self-sewing maps qi : Mgi−1,ni+2 → Mgi,ni (2) for some C(cid:48) ∈ Mg1,n1 If C = q1(C(cid:48)), then ΓC is the graph obtained by ΓC(cid:48) by attaching a loop to the vertex corresponding to the components sewn by q1. Therefore h1(ΓC) = h1(ΓC(cid:48))+1, and similarly if C = qk◦. . .◦q1(C(cid:48)) then h1(ΓC) = h1(ΓC(cid:48))+k. Moreover there exists C(cid:48) such that q(C(cid:48)) = C if and only if ΓC is not simply connected (it suffices to take as C(cid:48) the normalization of C at a node corresponding of an edge that borders a loop in ΓC), therefore if k is the maximal length of a sequence like (2), then h1(ΓC) = k. C = qk ◦ . . . ◦ q1(C(cid:48)) . Putting all together, we have the formula (cid:88) v∈V g = g(v) + h1(ΓC) the first term accounts for the sum of the genera of the components of C, and the second one for the non-separating nodes. 4.2 R-matrix action This section follows closely [12], where all the missing proofs can be found, along with a much deeper treatment of the subject. Let A be a finite-dimensional vector space over C and η a symmetric non- degenerate bilinear form on A. For a variable z, let R(z) = Id + R1z + R2z2 + . . . ∈ End(A)[[z]] be an End(A)[[z]]-valued power series. Let R(z)∗ be the adjoint of R(z) with respect to η, that is, R(z)∗ = η−1R(z)tη where R(z)t = Id + Rt 1z + Rt 2z2 + . . . ∈ End(A)[[z]] 16 is the transpose of R(z). We say that R(z) satisfies the symplectic condition if R(z)R(−z)∗ = Id. If R(z) satisfies the symplectic condition, then the quantity η−1 − R(z)η−1R(w)t z + w is a power series in the variables z and w. symplectic condition as well. Therefore the quantity Note that R−1(z) = 1 R(z) = R(−z)∗ is still a power series, and it satisfies the η−1 − R−1(z)η−1R−1(w)t z + w is a power series. If v ∈ A is a vector, we denote by R(z)v the A-valued power series R(z)v = v + R1(v)z + R2(v)z2 + . . . ∈ A[[z]]. If φ : A → B is a homomorphism of vector spaces and v ∈ A, we denote by φ(R(z)v) the B-valued power series φ(R(z)v) = φ(v) + φ(R1(v))z + φ(R2(v))z2 + . . . ∈ B[[z]]. Definition 7. Let (Ωg,n)g,n be a nodal cohomological field theory with base (A, 1, η), and let T (z) = z · [1 − R(z)−11] ∈ z2A[[z]]. Let S ⊆ Mg,n be a closed boundary stratum and let ΓS be its dual graph, with vertex set V . We define the contribution of S ContΓS : A⊗n → H•(Mg,n) in the following way: 1. At every vertex v of ΓS, we place (cid:88) m≥0 (pm)∗Ωg(v),n(v)+m(· ⊗ T (ψn(v)+1) ⊗ . . . ⊗ T (ψn(v)+m)) 1 m! where pm : Mg,n+m → Mg,n is the map that forgets the last m marked points (in the sum, the term corresponding to m = 0 is simply Ωg(v),n(v)). 2. At every leg l of ΓS, we place R−1(ψl) (recall that a leg corresponds to a marked point, that has a respective ψ-class); 3. At every edge e of ΓS, we place η−1 − R−1(ψ(cid:48) e)η−1R−1(ψ(cid:48)(cid:48) e )t ψ(cid:48) e + ψ(cid:48)(cid:48) e (recall that an edge corresponds to a node, that has two ψ-classes attached to the two special points of the normalization at that node). 17 4. In this way we have a map A⊗n → (cid:81) map along the map ξΓS of Remark 11, which is proper, to obtain ContΓS . Definition 8. Let Ωg,n be a nodal CohFT with base (A, 1, η). We define (RΩ)g,n : A⊗n → H•(Mg,n) with the following formula v∈V Mg(v),n(v). We push forward this (cid:88) S⊆Mg,n (RΩ)g,n = 1 Aut(ΓS)ContΓS where S runs among the closed boundary strata of Mg,n. We say that RΩ is obtained from Ω by an R-matrix action. Example 1. Let us consider a graph Γ with a vertex v of genus g, and a leg l attached to it. It corresponds to the open stratum Mg,1 or to the closed stratum S = Mg,1. Then, for every a ∈ A, we have ContΓ(a) = (pm)∗Ωg,1+m(R−1(ψl)a ⊗ T (ψ2) ⊗ . . . ⊗ T (ψ1+m)). 1 m! (cid:88) m≥0 Example 2. Let us consider a graph Γ with a vertex v of genus g, no legs, and an edge e that joins v to itself. The valence of the only vertex of Γ is 2. Then with the convention A⊗0 = C, we have naturally ContΓ ∈ H•(Mg) and it is precisely (cid:88) m≥0 ContΓ = 1 m! (pm)∗ηµνΩg,2+m (cid:18) eµ ⊗ eν − R−1(ψ(cid:48) e)eµ ⊗ R−1(ψ(cid:48)(cid:48) ψ(cid:48) e + ψ(cid:48)(cid:48) e e )teν ⊗ T (ψ3) ⊗ . . . ⊗ T (ψ2+m) (cid:19) . Example 3. Let Γ be a graph with two vertices v1 and v2 of genera g1 and g2 respectively, no legs, and an edge e that joins v1 to v2. In this case the valence of each vertex is 1, and again ContΓ ∈ H•(Mg). If we set Cont(cid:48) m(eµ) = 1 m! (pm)∗Ωg1,1+m((eµ − R−1(ψ(cid:48) is the same but with g2 and R−1(ψ(cid:48)(cid:48) e)eµ) ⊗ T (ψ2) ⊗ . . . ⊗ T (ψ1+m)) e )t instead of g1 and R−1(ψ(cid:48) e ), we and Cont(cid:48)(cid:48) have m ContΓ = ηµν ψ(cid:48) e + ψ(cid:48)(cid:48) e (cid:88) m≥0 m(eµ) ×(cid:88) m≥0 Cont(cid:48) Cont(cid:48)(cid:48) m(eν). Example 4. Let us consider the case in which R(z) ≡ Id. Then the power series of rule 2 of Definition 7 is η−1 − R−1(z)η−1R−1(w)t = 0, and T (z) = 0. Therefore all the contributions are zero, except the one corresponding to the stratum S ⊆ Mg,n whose graph ΓS has no edges. By Remark 8 it must have only one vertex, with n legs attached to it, and it must clearly correspond to the smooth stratum S = Mg,n ⊆ Mg,n, or to the closed stratum S = Mg,n. In this case it is easily seen that the contribution is simply ContΓS = Ωg,n. Therefore we have IdΩ = Ω. Proposition 5. If Ω is a nodal CohFT, then RΩ is a nodal CohFT. The R-matrix action on CohFTs is a left group action. 18 Remark 13. We could also have defined, in Definition 7, contributions associated to a free boundary cohomological field theory Ωg,n, and a matrix action RΩ in Definition 8 as well, using the same formulas as in the nodal case. It turns out that Proposition 5 stays valid, therefore RΩ will be a (nodal!) cohomological field theory. We will now state one lemma that will be needed in the sequel. Lemma 3. Let A be a C-algebra and let M (z) = z(1 − exp(−a1z − a2z2 − . . .)) ∈ A[[z]]. Fix g, n integers such that 2g − 2 + n > 0, and for each m ≥ 0, let pm : Mg,n+m → Mg,n be the map that forgets the last m marked points. Then in Mg,n we have the equality exp(a1κ1 + a2κ2 + . . .) = (pm)∗(M (ψn+1)··· M (ψn+m)) 1 m! Proof. We need a technical fact about κ-classes. For each permutation σ ∈ Sm, let σ = ρ1 ··· ρl be its decomposition in cycles, where also 1-cycles have been written. n+m ), then For each cycle ρi, define kρi =(cid:80) kj. If κk1,...,km = (pm)∗(ψk1+1 n+1 ··· ψkm+1 j∈ρi ∞(cid:88) m=0 κk1,...,km = κkρi . (3) (cid:88) l(cid:89) σ∈Sm i=1 l(cid:89) For example, κ1,1 = κ2 Let B(z) = M (z) 1 + κ2 and κ1,2,3 = κ1κ2κ3 + κ1κ5 + κ2κ4 + κ2 3 + 2κ6. z ∈ A[[z]]. For each set of m strictly positive integers {k1, . . . , km}, let bk1,...,km = ak1 ··· akmκk1,...,kmzk1+...+km. If σ ∈ Sm is a permutation whose decomposition in cycles is σ = ρ1 ··· ρl, let )zk1+...+km bk1,...,km,σ. Let i=1 κkρi bk1,...,km,σ = (ak1 ··· akm then by Equation 3 we have bk1,...,km =(cid:80) (cid:88) and let F =(cid:80) th number weighted by akizki, along with a weighting(cid:81)l k1,...,km≥1 bk1,...,km Nm = σ∈Sm Nm m! m≥0 . We see that F is the generating function, evaluated at 1, of some combinatorial objects, namely the sets of m numbers {k1, . . . , km}, with the i- where ρ1 ··· ρl ∈ Sm is the decomposition in cycles of a permutation in Sm. By the theory of generating functions, F = eG where G is the generating function, evaluated at 1, of the sets of n numbers {k1, . . . , kn}, with the i-th number weighted by akizki, along with a where ρ is an n-cycle in Sn. Let φ : A[[z]] → A[[z, κi]]i≥1 be the weighting κkρ A-linear map that sends every zi to ziκi. Since kρ = k1 + . . . + kn does not depend on ρ, and since there are exactly (n − 1)! n-cycles in Sn, we see that i=1 κkρi (cid:88) n≥0 G = φ (cid:88) (n − 1)! n! k1,...,kn≥1 19  = ak1 ··· aknzk1+...+kn (cid:88) n≥0 = φ B(z)n n  = −φ(log(1 − B(z))) = a1κ1z + a2κ2z2 + . . . Setting z = 1 in the equality F = eG we get precisely the formula we need to prove. 5 Frobenius algebras Definition 9. Let (A,·, 1) be a finite-dimensional associative algebra with unit ele- ment 1 over the field C (this will be the only case of real interest in our discussion). This algebra is a Frobenius algebra if it is equipped with a non-degenerete bilinear form η : A ⊗ A → C such that η(a · b, c) = η(a, b · c) for every a, b, c ∈ A. If η is symmetric, then A is called a symmetric Frobenius algebra. ∼−→ A∗ between A and the dual space A∗, Remark 14. We have an isomorphism i : A given by the nondegenerate form η, namely i(v) = η(v,·). We can define an algebra structure on A∗ by imposing i to be an isomorphism of algebras. Also, on the dual space A∗ we have (ηt)−1 the inverse transpose form, which is still non-degenerate and defines a structure of Frobenius algebra on the dual space A∗ (the verification is immediate). ∼−→ A∗ as in the previous remark, the functional Definition 10. By letting i : A θ = i(1) is called the Frobenius trace of the algebra. In other words, the Frobenius trace is defined as θ(a) = η(a, 1) for every a ∈ A. Definition 11. Let (A,·, 1, η) be a Frobenius algebra. The vector α = ηµνeµ · eν µeρ µAσ ρ )ρ,σ, then fµ = Aρ is called the Euler class of A. Remark 15. If Aeµ = fµ for some invertible matrix A = (Aσ and ν (A−1eρ) · (A−1eσ) = (AtηA)ρσfρ · fσ ηµνeµ · eν = ηµν(A−1fµ) · (A−1fν) = ηµνAρ Where At is the transpose matrix of A. Since the matrix η, in the basis (fρ)ρ, turns into the matrix AtηA, we see that the Euler class does not depend on the choice of a basis of A, it is therefore determined by the Frobenius algebra structure. Remark 16. The Euler class α does not lie in the kernel of the Frobenius trace of A; in particular, it is never zero. Indeed, let θ be the Frobenius trace of A. Then we have θ(α) = ηµνη(eµ, eν) = ηµνηµν = dimA, which is not zero. Definition 12. Let (A,·, 1, η) be a Frobenius algebra of complex dimension k. A is said to be semisimple if there exist an orthonormal (with respect to η) basis (eµ)1≤µ≤k and non-zero complex numbers θµ, such that for every i and j, we have (4) eµ · eν = δµνθ−1 µ eµ A basis that satisfies the conditions (4) is called a semisimple basis; the moduli θµ are called the weights of the algebra. 20 Definition 10. Therefore we have 1 =(cid:80) It is clear that in this case θµ = η(eµ, 1) and also θµ = θ(eµ) in the notation of µ θµeµ for a semisimple basis (eµ)µ. Remark 17. The unordered set of weights is well-defined. Indeed, if fµ is another semisimple basis, then there exists an η-orthogonal matrix P (i.e. P tηP = η) such that fµ = P eµ. Let us expand the product for general µ and ν P eµ · P eν = P ρ ν eρ · eσ = P ρ µ P σ µ θ−1 ρ eρ µ P ρ µ P σ For µ (cid:54)= ν, we must have P ρ ν = 0 for every ρ, therefore the matrix P has at most one (and then exactly one since it must be non-singular) non-zero entry for each row. Thus, eventually conjugating by a permutation matrix (which is orthogonal), we can suppose that P is diagonal. Since P is η-orthogonal, the entries in the diagonal must be 1 or −1, therefore the weights remain unchanged since the θµ can only change sign. In fact we have shown that the semisimple basis is uniquely determined, up to permutation and to sign switch of the vectors. Proposition 6. Let (A,·, 1, η) be a Frobenius algebra. Then A is semisimple if and only if its Euler class α is invertible. Proof. We will prove just the easy part: if the Frobenius algebra is semisimple, then α is invertible. The other part is proved in [2] and will not be needed in the sequel. Let A be semisimple and let (eµ)µ be a semisimple basis. Then by Definition 11 µ eµ in this basis. It is then easy to show and Remark 15 we can write α = (cid:80) that α−1 =(cid:80) µeµ is an element such that α · α−1 = 1, that is, α is invertible. µ θ−1 µ θ3 Definition 13. Let (Ωg,n)g,n be a CohFT of any type with base (A, η, 1) and let i : A → A∗ be the isomorphism of Remark 14. We define a product · on A by writing, for every v1, v2 ∈ A, v1 · v2 = i−1Ω0,3(v1 ⊗ v2 ⊗ ·) or in other words, η(v1 · v2, v3) = Ω0,3(v1 ⊗ v2 ⊗ v3). Remark 18. We notice that Ω0,3 and Ω0,3 both have values in the cohomology ring of M0,3 = M0,3, which is just C, and by Proposition 4 they must be the same homomorphism. The same is true for (cid:101)Ω0,3, by Proposition 3; thus we can avoid specifying the type of the CohFT, because every type leads to the same algebra structure. Proposition 7. Let (Ωg,n)g,n be a CohFT of any type with base (A, η, 1). Then there exists a unique Frobenius algebra structure on A such that, for every vi ∈ A, we have Ω0,3(v1 ⊗ v2 ⊗ v3) = η(v1 · v2, v3) This Frobenius algebra is symmetric and commutative, and has 1 as neutral element. We will need the following lemma. Lemma 4. Let s : Mg,n × M0,3 → Mg,n+1 be the map that sews together the points marked n and 3 respectively. Then we have s∗Ωg,n+1(v1 ⊗ v2 ⊗ . . . ⊗ vn+1) = Ωg,n(v1 ⊗ . . . ⊗ vn−1 ⊗ (vn · vn+1)) 21 Proof. By the sewing axiom of Definition 1, we have s∗Ωg,n+1(v1 ⊗ v2 ⊗ . . .⊗ vn+1) = ηµνΩg,n(v1 ⊗ . . .⊗ vn−1 ⊗ eν)× Ω0,3(vn ⊗ vn+1 ⊗ eµ) and by Definition 13, the last quantity is ηµνη(vn·vn+1, eµ)Ωg,n(v1⊗. . .⊗vn−1⊗eν) = Ωg,n(v1⊗. . .⊗vn−1⊗ηµνη(vn·vn+1, eµ)eν) since ηµνη(vn · vn+1, eµ)eν = vn · vn+1, we get the required formula. Remark 19. The same lemma holds for smooth theories as well, provided we use the right sewing axioms. This means that if σ : (cid:102)Mg,n × (cid:102)M0,3 → (cid:102)Mg,n+1 is the sewing-smoothing map of Proposition 1, then we have σ∗(cid:101)Ωg,n+1(v1 ⊗ v2 ⊗ . . . ⊗ vn+1) =(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn−1 ⊗ (vn · vn+1)) Analogously, if νs : ∂Ns → S are as in Remark 4, then we have Ωg,n+1(v1 ⊗ v2 ⊗ . . . ⊗ vn+1)∂Ns = ν∗ s Ωg,n(v1 ⊗ . . . ⊗ vn−1 ⊗ (vn · vn+1)) The proof is literally the same as in the previous lemma, since the right-hand side is the same in the three sewing axioms. Proof of Proposition 7. First of all, we see that for every v ∈ A, v · 1 = v. Indeed, we have v · 1 = i−1Ω0,3(v ⊗ 1 ⊗ ·) = i−1Ω0,3(1 ⊗ v ⊗ ·) = i−1η(v,·) = v where we have used the S3-equivariance and the second axiom of CohFTs. Now associativity, which is equivalent to Ω0,3((v1 · v2) ⊗ v3 ⊗ v4) = Ω0,3(v1 ⊗ (v2 · v3) ⊗ v4) for every vi ∈ A. By Lemma 4 and by S3-equivariance, these two quantities are s∗Ω0,4(v3 ⊗ v4 ⊗ v1 ⊗ v2) and s∗Ω0,4(v1 ⊗ v4 ⊗ v2 ⊗ v3) respectively, where s : M0,3 × M0,3 → M0,4 is the sewing map as in Lemma 4. If ρ = (1, 2, 3) ∈ S4 then by S4-equivariance we have ρ∗Ω0,4(v1⊗v2⊗v3⊗v4) = Ω0,4(v2⊗v3⊗v1⊗v4), but ρ∗ is the identity in degree 0, and s∗ kills every positive-degree cohomology class, since M0,3 has dimension 0. Therefore we have s∗Ω0,4(v3 ⊗ v4 ⊗ v1 ⊗ v2) = s∗Ω0,4(v1 ⊗ v4 ⊗ v2 ⊗ v3) and by Lemma 4 we have associativity. Finally, we have η(v1· v2, v3) = η(v1, v2· v3) for every vi ∈ A because this equality means Ω0,3((v1 · v2) ⊗ v3 ⊗ 1) = Ω0,3(v1 ⊗ (v2 · v3) ⊗ 1) which is true thanks to associativity. 22 Lemma 5. Let (Ωg,n)g,n be a CohFT of any type with base (A, η, 1), let α be its Euler class, and let q : Mg−1,n+2 → Mg,n be the self-sewing map. Then q∗Ω1,1(v) = η(v, α) In particular, the degree zero part of Ω1,1(v) is η(v, α). Proof. By the self-sewing axiom, we have q∗Ω1,1(v) = ηµνΩ0,3(v ⊗ eµ ⊗ eν) = ηµνη(v, eµ · eν) = η(v, α) which is the statement. Lemma 6. Let (Ωg,n)g,n be a semisimple CohFT (of any type) with Euler class α. Then the degree-zero part of Ω1,2(v ⊗ w) is Ω1,2(v ⊗ w)0 = η(v · w, α) Proof. We do the proof for a fixed boundary theory ((cid:101)Ωg,n)g,n, but the proof is the same for other theories if we take care of using the right sewing axioms. We consider the sewing-smoothing map By Remark 19 we get (cid:102)M0,3 × (cid:102)M1,1 σ−→ (cid:102)M1,2 σ∗(cid:101)Ω1,2(v ⊗ w) =(cid:101)Ω1,1(v · w) By the previous lemma, the degree zero part of the last quantity is η(v · w, α) and the formula is proved. 5.1 Sewing smooth surfaces The next three propositions show the effect of sewing a fixed smooth g(cid:48)(cid:48)-genus surface Σ to the curves in Mg,n. Using fixed boundaries, free boundaries, or nodal theories changes the way this result is stated, even though the proofs are just the same in the three cases. For the sake of clarity and completeness, we state three different propositions for the three types of theories we have discussed. Proposition 8. Let g = g(cid:48) + g(cid:48)(cid:48) and Σ ∈ (cid:102)Mg(cid:48)(cid:48),2 be a smooth genus g(cid:48)(cid:48) surface with two framed points, and let σΣ : (cid:102)Mg(cid:48),n × Σ → (cid:102)Mg,n be the restriction of the sewing-smoothing map σ of Proposition 1 that sews together the points marked n and 2 respectively. Then for any v1, . . . , vn ∈ A we have ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) =(cid:101)Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ (αg(cid:48)(cid:48) · vn)) σΣ 23 Proof. Let us first consider the case g(cid:48)(cid:48) = 1, so Σ is a torus with two fixed marked points. By the sewing axiom, we have σ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = ηµν(cid:101)Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ eµ) ×(cid:101)Ω1,2(vn ⊗ eν) Now we must take the restriction of this class (which is in H•((cid:102)Mg(cid:48),n × (cid:102)M1,1)) to the closed subset (cid:102)Mg(cid:48),n×Σ, and the result of this operation is just taking the degree-zero part in each right factor and summing up; thus the class we are looking for is ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = ηµν(cid:101)Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ eµ)(cid:101)Ω1,2(vn ⊗ eν)0 where the subscript 0 stands for the degree-zero part. If we put w = eν in Lemma 6, we get (cid:101)Ω1,2(vn ⊗ eν)0 = η(vn · eν, α), thus finally (cid:101)Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ ηµνη(α · vn, eµ)eν) =(cid:101)Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ (α · vn)) the quantity we are looking for is σΣ The general case comes simply by induction noticing that sewing on a genus g(cid:48)(cid:48) + 1 surface is the same as sewing in a row a genus g(cid:48)(cid:48) surface and a torus. Proposition 9. Let g = g(cid:48) + g(cid:48)(cid:48) and Σ ∈ Mg(cid:48)(cid:48),2 be a smooth genus g(cid:48)(cid:48) surface with two free points, and let sΣ : Mg(cid:48),n × Σ → Mg,n be the restriction of the sewing map s that sews together the points marked n and 2 respectively, and let SΣ be its image. With notations as in Notation 2, let ∂NΣ = ν−1 s (SΣ) and let be the restriction of νs. Then for every vi ∈ A, we have νΣ : ∂NΣ → SΣ Ωg,n(v1 ⊗ . . . ⊗ vn)∂NΣ = ν∗ ΣΩg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ (αg(cid:48)(cid:48) · vn)) Proof. Since the term on the right of the sewing axiom in Definition 4 is the same as the respective one in Definition 3 (except for a ν∗ that appears), the same calculation s )H•(SΣ) = of the proof of the previous proposition, together with the obvious fact (ν∗ ν∗ Proposition 10. Let g = g(cid:48) + g(cid:48)(cid:48) and Σ ∈ Mg(cid:48)(cid:48),2 be a smooth genus g(cid:48)(cid:48) surface with two free points, and let , leads to the desired result. Σ s sΣ : Mg(cid:48),n × Σ → Mg,n be the restriction of the sewing map s that sews together the points marked n and 2 respectively. Then for every vi ∈ A, we have ∗Ωg,n(v1 ⊗ . . . ⊗ vn) = Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn−1 ⊗ (αg(cid:48)(cid:48) · vn)) sΣ Proof. The proof is the same as that of Proposition 9, thanks to the fact that the term in the right-hand side of the sewing axiom in Definition 1 is the same as in Definition 3. 24 Remark 20. Let us suppose that we want to construct, say, a nodal CohFT (Ωg,n)g,n, and let us suppose we know the theory on Mg(cid:48),n × Mg(cid:48)(cid:48),2. If the base (A,·, 1, η) is semisimple, then we can recover the theory on Mg(cid:48),n. In fact, thanks to Proposition 10 and to Proposition 6 that assures that α is invertible in a semisimple theory, we have Ωg(cid:48),n(v1 ⊗ . . . ⊗ vn) = s∗ ΣΩg,n(v1 ⊗ . . . ⊗ vn−1 ⊗ (α−g(cid:48)(cid:48) · vn)) It is important to notice that to calculate the right-hand side of this equality we don't need to know the whole homomorphism Ωg,n : A⊗n → H•(Mg,n), but just its restriction to Mg(cid:48),n × Mg(cid:48)(cid:48),2, because the image of sΣ is contained in this subspace. This will play an important role in the classification of nodal cohomological field theories. 6 Classification From now on, all the theories are supposed to be semisimple. 6.1 Fixed boundary theories Let(cid:101)Ω be a fixed boundary CohFT with semisimple base (A, 1, η) and with semisimple basis (eµ)µ. We consider (cid:102)Mg,1 embedded in (cid:102)Mg+1,1 by the map where (T, [v1], [v2]) ∈ (cid:102)M1,2 is any fixed smooth genus 1 curve with two marked smoothing map of Proposition 1. We call σg,T : (cid:102)Mg,1 → (cid:102)Mg+1,1 this embedding. Therefore, we have a projective system (H•((cid:102)Mg,1))g≥1 with projection maps points, [v] and [vi] are tangent directions at the marked points, and σ is the sewing- (C, [v]) (cid:55)→ σ((C, [v]), (T, [v1], [v2])) . The limit is naturally isomorphic to C[κj]j≥1 by Madsen-Weiss theorem: σ∗ g,T σ∗ But σ∗ Proposition 11. The limit (cid:101)Ω+ = lim←− (cid:101)Ωg,1(α−g·) is a well-defined element of A∗ ⊗ C[κj]j≥1. Proof. Letting Σ = T in Proposition 8 we find that g,T(cid:101)Ωg+1,1(α−g−1 · v) =(cid:101)Ωg,1(α−g−1 · α · v) =(cid:101)Ωg,1(α−g · v). are the projections in the system (H•((cid:102)Mg,1))g≥1, thus the homomorphisms (cid:101)Ωg,1(α−g·) are compatible with the projective system. Thus, they pass to the limit to get a well-defined homomorphism (cid:101)Ω+ : A → lim←−H•((cid:102)Mg,1) which is an element of A∗ ⊗ C[κj]j≥1 by Madsen-Weiss theorem. Remark 21. Let i : A → A∗ be the isomorphism of remark 14. Since (eµ)µ is an orthonormal basis for η, we have i(eν) = eν for each ν, where (eµ)µ is the dual basis to (eµ)µ. By imposing i to be an isomorphism of algebras, we immediately see that (A∗, η−1, i(1)) is a semisimple Frobenius algebra, with semisimple basis (eµ)µ and same weights as A: g,T lim←−H•((cid:102)Mg,1) ∼= C[κj]j≥1. eµ · eν = δµ,νθ−1 µ eµ 25 Thus we can write an explicit formula for the effect of applying a vector tρeρ ∈ A to the product of the covectors v = vµeµ and w = wνeν: ((vµeµ) · (wνeν))(tρeρ) = vµwµθ−1 µ tµ = θ−1 µ tµv(eµ)w(eµ) ture in A∗ and the cross product in cohomology to define the product of two homo- Remark 22. Since (cid:101)Ωg,1 ∈ A∗ ⊗ H•((cid:102)Mg,1), we can use the Frobenius algebra struc- morphisms (cid:101)Ωg,1 ·(cid:101)Ωh,1. More precisely, thanks to Remark 21, we have (cid:101)Ωg,1 ·(cid:101)Ωh,1(vµeµ) = θ−1 where × is the usual cross product in the cohomology of the product (cid:102)Mg,1 × (cid:102)Mh,1. Proposition 12. Let (cid:101)C ∈ (cid:102)M0,3 be any curve with framed points, and let mg,h : (cid:102)Mg,1 × (cid:102)Mh,1 → (cid:102)Mg+h,1 be the composition ∼−→ (cid:102)Mg,1 × {(cid:101)C} × (cid:102)Mh,1 µ vµ(cid:101)Ωg,1(eµ) ×(cid:101)Ωh,1(eµ) σ1×Id−−−−→ (cid:102)Mg,2 × (cid:102)Mh,1 (cid:102)Mg,1 × (cid:102)Mh,1 σ2−→ (cid:102)Mg+h,1 where σ1 and σ2 are the obvious sewing-smoothing maps. Then, in the notation of Remark 22, we have Proof. By applying the sewing axiom for fixed boundary theories we get m∗ g,h(cid:101)Ωg+h,1 =(cid:101)Ωg,1 ·(cid:101)Ωh,1 (cid:88) (cid:101)Ωg,2(v ⊗ eµ) ×(cid:101)Ωh,1(eµ) σ∗ 2(cid:101)Ωg+h,1(v) = (cid:88) and 1(cid:101)Ωg,2(v ⊗ eµ) = σ∗ µ η(v · eµ, eν)(cid:101)Ωg,1(eν) = η(v, eµ · eµ)(cid:101)Ωg,1(eµ) This formula is correct since (cid:101)Ω0,3, by Definition 3, has values in the degree zero part of H•((cid:102)M0,3), therefore in this case we have σ∗ 1 = (σ1(cid:102)Mg,1×{C})∗. Using the fact that ν µ eµ, we get the result thanks to Remark 21. eµ · eµ = θ−1 Definition 14. We write ig,n : C[κj]j≥1 → H•((cid:102)Mg,n) ∼= lim←−H•((cid:102)Mg,1). C[κj]j≥1 Remark 23. The maps for the C-algebra homomorphism that sends each κj to the respective κ-class in (cid:102)Mg,n. Notice that ig,1 is the projection to the g-th factor under the identification induced in cohomology by the maps mg,h of Proposition 12 pass to the projective limit giving a map m∗ g,h : H•((cid:102)Mg+h,1) → H•((cid:102)Mg,1) ⊗ H•((cid:102)Mh,1) m∗ : lim←−H•((cid:102)Mg,1) → (lim←−H•((cid:102)Mg,1)) ⊗ (lim←−H•((cid:102)Mg,1)) that is, by Madsen-Weiss's theorem, a map m∗ : C[κj]j≥1 → (C[κj]j≥1) ⊗ (C[κj]j≥1) From now on, by m∗ we will mean this map if not otherwise specified. g,h ◦ ig+h,1 = (ig ⊗ ih) ◦ m∗ by Definition 14. Of course we have m∗ 26 Definition 15. Let X, Y ∈ A∗ ⊗ C[κj]j≥1; we define X · Y (vµeµ) = θ−1 µ vµX(eµ) ⊗ Y (eµ) which is an element of A∗ ⊗ (C[κj]j≥1 ⊗ C[κj]j≥1). homomorphisms Xg : A → H•((cid:102)Mg,1), then With this definition, if X = (Xg)g≥1 and Y = (Yh)h≥1 are projective systems of (ig ⊗ ih)X · Y = Xg · Yh where at the right-hand side we have used the product of Remark 22. With the identification C ⊗ C[κj]j≥1 ∼= C[κj]j≥1, we see by an easy computation that the neutral element of the product is θ, the Frobenius trace of A as in Definition 10, which is indeed a homomorphism of A into C ⊆ C[κj]j≥1. Corollary 1. With the notations of Remark 23 and Definition 15, the limit homo- morphism (cid:101)Ω+ satisfies m∗(cid:101)Ω+ =(cid:101)Ω+ ·(cid:101)Ω+ Proof. By Proposition 12, by Remark 21 and by the identity αk =(cid:80) vµ(cid:101)Ωg,1(eµ)×(cid:101)Ωh,1(eµ) = ((cid:101)Ωg,1(α−g·)·(cid:101)Ωh,1(α−h·))(v) m∗(cid:101)Ωg+h,1(α−g−h · v) = θ2(g+h)−1 Recalling that(cid:101)Ω+ = lim←−(cid:101)Ωg,1(α−g·) we get the result by Remark 23 and Definition µ θ−2k+1 eµ, we have µ µ 15. 6.1.1 Properties of C[κj]j≥1 We want to study more deeply the properties of C[κj]j≥1, in particular the map m∗ and the product structure of Definition 15. To this purpose, we give a definition and prove a series of lemmas which will be useful in the sequel. Lemma 7. With notation as in Remark 23, we have m∗κj = κj ⊗ 1 + 1 ⊗ κj for each j ≥ 1. Proof. Let g and h be integers and let us consider the following commutative diagram. (cid:102)Mg,2 × (cid:102)Mh,1 (cid:116) (cid:102)Mg,1 × (cid:102)Mh,2 p(cid:48)(cid:121) (cid:102)Mg,1 × (cid:102)Mh,1 m(cid:48) g,h−−−−→ (cid:102)Mg+h,2 (cid:121) mg,h−−−−→ (cid:102)Mg+h,1 p where p(cid:48) is the map that forgets the second marked point at each curve, and m(cid:48) the map that sews a fixed element of (cid:102)M0,3 to the first marked points of each factor. and ψ2 ⊗ ψ2 = 0 ∈ H•((cid:102)Mg,2 × (cid:102)Mh,1 (cid:116) (cid:102)Mg,1 × (cid:102)Mh,2). Therefore m(cid:48)(ψj+1 Then κj = p∗(ψj+1 2 ⊗ 1 + 1 ⊗ ψj+1 ψj+1 h we get the statement, thanks to Remark 23. ) But m(cid:48)∗(ψ2) = ψ2 ⊗ 1 + 1 ⊗ ψ2 ) = p(cid:48)∗m(cid:48)∗(ψj+1 ) = ) = κj ⊗ 1 + 1 ⊗ κj. Taking the limit in g and ) and m∗p∗(ψj+1 and p(cid:48)∗m(cid:48)∗(ψj+1 is g,h 2 2 2 2 2 2 27 Remark 24. The last lemma says that the polynomial ring C[κj]j≥1 is a Hopf algebra, the co-product being m∗, and the antipode map being S : κj (cid:55)→ −κj for each j and extended to a C-linear homomorphism C[κj]j≥1 → C[κj]j≥1. If A is a Frobenius algebra, it is in fact a bialgebra (but not necessarily a Hopf algebra) via the co-product ∆(v) = η(v, eµ · eν)eµ ⊗ eν which is such that, for every w, t ∈ A, η ⊗ η(∆(v), eµ ⊗ eν) = η(v, eµ · eµ). Notice that, in the notation of Definition 15, we have X · Y = (X ⊗ Y ) ◦ ∆. µ vµeµ ⊗ eµ. Let us In the semisimple case, the co-product simplifies to ∆(v) = θ−1 µeµ, we see that ∆ has a left inverse stick to the semisimple case. Taking P (v) = vµθ2 given by · ◦ (Id ⊗ P ), where we have denoted by · : A ⊗ A → A the product of the Frobenius algebra A. The endomorphism P of A defines a co-product ∆∗ on A∗ that makes it a bialgebra via the formula ∆∗(φ)(v ⊗ w) = φ(P (v) · P (w)). Notice that with this definition the isomorphism i : A → A∗ of Remark 14 is not an isomorphism of bialgebras. Thus, A∗ ⊗ C[κj]j≥1 is a bialgebra: the co-product (cid:101)∆ being the tensor prod- ∼= uct of the two co-products on each factor. By the identification A∗ ⊗ C[κj]j≥1 HomC(A, C[κj]j≥1), we see that the following diagram commutes. A ∆ f−−−−→ (cid:121) A ⊗ A −−−−→(cid:101)∆f C[κj]j≥1 m∗(cid:121) C[κj]j≥1 ⊗ C[κj]j≥1 Definition 16. Let y ∈ C[κj]j≥1. Then y is primitive if m∗y = y ⊗ 1 + 1 ⊗ y with the notation of Remark 23. Lemma 8. A class y ∈ C[κj]j≥1 is primitive if and only if it is a linear combination of κ classes. Proof. Each κj is primitive by Lemma 7. Since the primitive classes clearly form a C-vector space, we deduce that every linear combination of κ classes is primitive. Conversely, Milnor-Moore's theorem says that a Hopf algebra is a free algebra on its primitive classes. Since C[κj]j≥1 is a free algebra on the κj's, and since they are primitive, there cannot be any other primitive class. This completes the proof. 28 Definition 17. Let X : A → C[κj]j≥1 be a homomorphism. X is group-like if X0 = θ, the Frobenius trace of A, and it satisfies the equality X is primitive if it satisfies the equality m∗X = X · X m∗X = X ⊗ 1 + 1 ⊗ X Remark 25. Lemma 8 says that a homomorphism x : A → C[κj]j≥1 is primitive if and only if there are covectors φj ∈ A∗ such that x =(cid:80) j φjκj. Lemma 9. Let x : A → C[κj]j≥1 be a homomorphism, and let exp(x) be the homo- morphism defined by the infinite series ∞(cid:88) n=0 x·n n! exp(x) = where x·n : A → C[κj]j≥1 is the n-th power for the product of Definition 15 composed with the product of polynomials C[κj]⊗n j≥1 → C[κj]j≥1, with the convention x0 = θ the Frobenius trace of A as in Definition 10. Then we have Proof. By linearity we have m∗exp(x) = exp(m∗x) m∗exp(x) = (cid:88) n m∗(x·n) n! therefore we have to calculate m∗(x·n). By applying Definition 15 with X = x, Y = xn−1, we see that vµ(x(eµ))n Where now (x(eµ))n is the usual power of polynomials. x·n(vµeµ) = θ−n+1 µ Now we have m∗[x(eµ)n] = [m∗(x(eµ))]n, thus m∗(x·n) = (m∗x)·n which immediately yields our result. Lemma 10. Let X : A → C[κj]j≥1 be a homomorphism. Then X is group-like if and only if there exists a primitive homomorphism x such that X = exp(x). Proof. Let x be primitive, and let X = exp(x). Then by Lemma 9 we have m∗X = exp(m∗x) = exp(x ⊗ 1 + 1 ⊗ x). Moreover, expanding out the powers, we get ∞(cid:88) (cid:88) n=0 k1+k2=n exp(x⊗ 1 + 1⊗ x) = 1 k1!k2! (x⊗ 1)·k1 · (1⊗ x)·k2 = exp(x⊗ 1)· exp(1⊗ x) Now we clearly have (x⊗1)·n = x·n⊗1, therefore the quantity in the above expression is (X ⊗ 1) · (1 ⊗ X), and this is in turn X · X since by Remark 15 we have (X ⊗ 1)· (1⊗ X)(v) = θ−1 µ vµ(X(eµ)⊗ 1)(1⊗ X(eµ)) = θ−1 µ vµX(eµ)X(eµ) = X · X(v) 29 Thus we see that X is group-like. Conversely, let X be group-like and let us define x = log(X) := (−1)n−1 (X − θ)·n n ∞(cid:88) n=1 then x is well-defined since the sum at the right-hand side is finite for each degree, and formally exp(x) = X. Moreover, the technique used in the proof of Lemma 9 yields m∗x = log(m∗X), therefore using the fact that X is group-like by hypothesis we get m∗X = log(X · X) Writing X·X = (X⊗1)·(1⊗X) and (X−θ)·n⊗1 = ((X−θ)⊗1)·n in the expression for log(X), we get the expression log(X · X) = log(X) ⊗ 1 + 1 ⊗ log(X) from which we see that x is primitive. This completes the proof. j>0 φjκj). C[κj]j≥1, we have have the first classification theorem. 6.1.2 Classification of smooth theories We are now ready to classify fixed boundary theories. this x is an A∗-linear combination of κj classes and this yields our result. Proposition 13. There exist covectors φj ∈ A∗ such that (cid:101)Ω+ = exp((cid:80) Proof. By Corollary 1 we know that (cid:101)Ω+ is group-like. By Lemma 10, there exists a primitive homomorphism x : A → C[κj]j≥1 such that (cid:101)Ω+ = exp(x). By Remark 25 Let us now see how to recover the homomorphisms (cid:101)Ωg,n from (cid:101)Ω+. For this, we Theorem 5. Let (cid:101)Ωg,n : A⊗n → H•((cid:102)Mg,n) be a fixed boundary CohFT. Then there exist covectors φj ∈ A∗ for j ≥ 1 such that, setting (cid:101)Ω+ = exp((cid:80) j≥1 φjκj) : A → (cid:101)Ωg,n(v1 ⊗ ··· ⊗ vn) = ig,n(cid:101)Ω+(αg · v1 ··· vn) (5) for g ≥ 1. Conversely, if a semisimple, symmetric, commutative Frobenius algebra structure is defined on (A,·, 1, η), then for every choice of covectors φj ∈ A∗ for j ≥ 1, we get a fixed boundary CohFT using the above formulas for g ≥ 1, and (cid:101)Ω0,n(v1 ⊗ . . . ⊗ vn) = s∗(cid:101)Ω1,n(v1 ⊗ . . . ⊗ (α−1 · vn)) where s : (cid:102)M0,n × Σ → (cid:102)M1,n for any Σ ∈ (cid:102)M1,2. Proof. From the definition (cid:101)Ω+ = lim←−(cid:101)Ωg,1(α−g·) we immediately find (cid:101)Ωg,1(v) = ig,1(cid:101)Ω+(αg · v). Let ϕg,n : (cid:102)Mg,1 → (cid:102)Mg,n for n ≥ 2 be the sewing-smoothing map applied to a fixed element of (cid:102)M0,n+2, and let sg,n : (cid:102)Mg,n → (cid:102)Mg+g(cid:48),n be the sewing- smoothing map applied to a fixed element of (cid:102)Mg(cid:48),2 sewed to (cid:102)Mg,n. It is easy to see by induction on n that g,n(cid:101)Ωg,n(v1 ⊗ ··· ⊗ vn) =(cid:101)Ωg,1(v1 ··· vn) = ig,1(cid:101)Ω+(αg · v1 ··· vn) ϕ∗ (6) 30 and by Proposition 8 we have (cid:101)Ωg,n(v1 ⊗ ··· ⊗ vn) = s∗ g,n(cid:101)Ωg+g(cid:48),n(v1 ⊗ ··· ⊗ (α−g(cid:48) · vn)) (7) Harer's stability theorem says that the map ϕg+g(cid:48),n is a homology equivalence in degree less than (g + g(cid:48))/3, thus if this quantity is greater than 3g− 3 + n, we deduce by (6) and (7) the following formula (cid:101)Ωg,n(v1 ⊗ ··· ⊗ vn) = s∗ g+g(cid:48),n)−1ig+g(cid:48),1(cid:101)Ω+(αg · v1 ··· vn) g,n(ϕ∗ (8) which is well-defined since the degree of the left-hand side is in the range in which ϕg+g(cid:48),n is an isomorphism. Now, in the stable range, we know by Harer's theorem that the inverse of ϕg+g(cid:48),n is exactly p, the forgetful map. Moreover it is clear that p ◦ sg,n = sg,1 ◦ p and that s∗ g,1ig+g(cid:48),1 = ig,1. Substituting in (8) we get the desired formula. Thus we have shown the first part. For the second part, we need to verify the axioms. Let then (eµ)µ be a semisimple Since (A,·) is commutative, (cid:101)Ωg,n is Sn-invariant, so the first axiom is verified. basis for the Frobenius algebra (A,·, 1, η). 1(cid:101)Ω+(v1 · v2) For the second axiom, we have (cid:101)Ω0,3(v1 ⊗ v2 ⊗ 1) = s∗(cid:101)Ω1,3(v1 ⊗ v2 ⊗ (α−1 · 1)) = s∗p∗i∗ Now from the definition of (cid:101)Ω+ we see that i∗ H•((cid:102)M1,1) to which i∗ 1(cid:101)Ω+(v1 · v2) = θ(v1 · v2) + φ1(v1 · v2)κ1, 1(cid:101)Ω+(v1 · v2) belongs); moreover, we have p∗κ1 = κ1 − ψ2 − ψ3, where θ is the Frobenius trace of A (the classes κj for j ≥ 2 are zero in the ring thus s∗p∗κ1 = 0 and finally s∗p∗i∗ 1(cid:101)Ω+(v1 · v2) = θ(v1 · v2) = η(v1, v2) and the axiom is verified. For the third one, let τ : (cid:102)Mg−1,n+2 → (cid:102)Mg,n be the non-separating sewing- smoothing map. Then g(cid:101)Ω+(αg · v1 ··· vn) forgetting all the points except the ones marked 1, n + 1 and n + 2, therefore we must compute τ∗i∗ every i, therefore τ∗κi = κi for every i and we simply have τ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = τ∗p∗i∗ Now p ◦ τ = τ ◦ p1,n+1,n+2 where p1,n+1,n+2 : (cid:102)Mg−1,n+2 → (cid:102)Mg−1,3 is the map g(cid:101)Ω+(αg · v1 ··· vn) where τ : (cid:102)Mg−1,3 → (cid:102)Mg,1. Now τ∗ψi = ψi for g(cid:101)Ω+(αg · v1 ··· vn) = p∗ g−1(cid:101)Ω+(αg · v1 ··· vn) 2,3i∗ g−1(cid:101)Ω+(αg · v1 ··· vn) = τ∗(cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = p∗i∗ (cid:88) (cid:101)Ω+(αg−1 · v1 ··· · vn · eµ · eµ) = (cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn ⊗ eµ ⊗ eµ) Putting all this together, we finally have = p∗i∗ g−1 (cid:88) τ∗ι∗ µ µ taking into account that eµ is a semisimple basis, we have verified the third axiom. The fourth axiom is proved similarly, while the fifth axiom is obvious. 31 6.2 Free boundary theories Now that we have proved the classification theorem for fixed boundary theories, we proceed to study free boundary theories. Thus, let Ω be a semisimple free boundary CohFT with semisimple base (A,·, 1, η) with semisimple basis (eµ)µ. We recall that if π : (cid:102)Mg,n → Mg,n is the map that forgets the tangent directions, then Proposition 3 says that (cid:101)Ω = π∗Ω is a fixed boundary CohFT, with the same base. Therefore, by Theorem 5, there exists a homomorphism (cid:101)Ω+ : A → C[κj]j≥1 such that Let π{2} : (cid:102)M{2} π∗Ωg,n(v1 ⊗ . . . ⊗ vn) = ig,n(cid:101)Ω+(αg · v1 ··· vn) g,2 → (cid:102)Mg,2 be the bundle map like in Notation 1, let s{2} Σ = {2} g,2 = π{2}∗Ωg,2. Then, by π{2}∗sΣ with notation as in Proposition 8, and let Ω Proposition 8, we have therefore we have a projective system Ω s{2}∗ Σ Ω {2} g(cid:48),2(v1 ⊗ (αg(cid:48) · v2)) {2} g,2 (v1 ⊗ v2) = Ω {2} g,2 (· ⊗ (α−g·)). We denote the limit object Ω+ = lim←−Ω {2} g,2 (· ⊗ (α−g·)) : A ⊗ A → C[ψ, κj]j≥1 where ψ is the ψ-class of the only free point of (cid:102)M{2} . We can see Ω+ as an element of (A∗ ⊗ A∗)[κ, ψj]j≥1, therefore composing with the isomorphism i−1 : A∗ → A of Remark 14 we get an element Z(κ, ψj)j≥1 ∈ (A∗ ⊗ A)[[κ, ψj]]j≥1. Recalling that there is a natural isomorphism A∗ ⊗ A (cid:39) End(A), we define R(ψ) = Z(0,−ψ)∗ ∈ End(A)[[ψ]] g,2 where the star indicates the adjoint endomorphism with respect to η. Remark 26. This definition simply means that for every v, w ∈ A, we have η(Z(κ, ψ)v, w) = Ω+(v ⊗ w)(κ, ψ) coefficient-wise, that is, η(((cid:80) (9) where Ω+(v ⊗ w)(κ, ψj)j≥1 ∈ C[κ, ψj]j≥1 and η at the left-hand side is applied i η(fi(v), w)ψi. By Lemma 6, the {2} 1,2 (v ⊗ α−1 · w)0 is η(v · α−1 · w, α) = η(v, w); therefore it is the i fiψi)v, w) = (cid:80) degree-zero part Ω same for Ω {2} g,2 (v ⊗ α−g · v)0 for every g ≥ 1, thus we see η(v, R(−ψ)0w) = Ω+(v ⊗ w)(ψ, 0)0 = η(v, w) for every v, w ∈ A, and we deduce R(ψ)0 = Id. From now on, we will write R(ψ) = Id + R1ψ + R2ψ2 + . . . for Ri ∈ End(A). In particular, R(ψ) is invertible. Lemma 11. We have (cid:101)Ω+(v1 · v2)(κ) = η(Z(κ, ψ)v1, R(−ψ)−1v2) = Ω+(v1 ⊗ R(−ψ)−1v2) 32 g1,g2 where Proof. Throughout this proof, we refer to the notation used in chapter 3, in particular to Notation 4. The sewing axiom for free boundary CohFTs says that Ωg,2(α−g1 · v1 ⊗ α−g2 · v2)∂Ns = ν∗ νg1,g2 : ∂Ns (cid:39) (cid:102)M{2} (cid:88) g1,2 ×S1 (cid:102)M{2} Ωg1,2(eµ ⊗ α−g1 · v1) × Ωg2,2(eµ ⊗ α−g2 · v2) g2,2 → Mg1,2 × Mg2,2 (just for this time, we have sewed the points marked 1 on each curve, the sewing axiom is therefore slightly modified). forgetting the tangent directions, we get If π : (cid:102)Mg,n → Mg,n is the bundle map (cid:88) {2} g2,2(eµ⊗α−g2·v2) {2} g1,2(eµ⊗α−g1·v1)×Ω Ω µ (cid:101)Ωg,2(α−g1·v1⊗α−g2·v2)π−1(∂Ns) = ν∗ (cid:101)Ω+(v1 · v2)(κ) = ν∗(cid:88) By passing to the limit in g1 and g2 we get g1,g2 µ Ω+(eµ ⊗ v1)(κ(cid:48), ψ(cid:48))Ω+(eµ ⊗ v2)(κ(cid:48)(cid:48), ψ(cid:48)(cid:48)) µ respective classes in (cid:102)M{2} where the left-hand side is computed with Theorem 5. Here κ(cid:48) and ψ(cid:48) refer to the 's are algebraically independent. We set κ = κ(cid:48) + κ(cid:48)(cid:48) and ν∗ to be the limit of the maps ν∗ sends the other classes to the respective classes on Mg,2, we can write . Then the κ(cid:48) g1,g2ψ(cid:48)(cid:48), and ν∗ . Since for every g1, g2 we have ν∗ , κ(cid:48)(cid:48) and ψ(cid:48)(cid:48) in (cid:102)M{2} g1,g2ψ(cid:48) = −ν∗ (cid:88) 's and the κ(cid:48)(cid:48) Ω+(eµ ⊗ v1)(κ(cid:48),−ψ)Ω+(eµ ⊗ v2)(κ(cid:48)(cid:48), ψ) (cid:101)Ω+(v1 · v2)(κ) = (10) g1,g2 g1,g2 g2,2 g1,2 i i µ therefore, setting κ(cid:48) = 0, κ(cid:48)(cid:48) = κ in (10) and using Equation (9), we get (cid:101)Ω+(v1 · v2)(κ) = (cid:88) η(R(ψ)v1, eµ)η(v2, Z(κ, ψ)eµ) = η(R(ψ)v1, Z(κ, ψ)∗v2). (11) µ Lemma 12 below implies that Z(κ, ψ) commutes with its adjoint, therefore the last term in Equation (11) is η(Z(κ, ψ)v1, R(ψ)∗v2). This is the first equality again by Lemma 12 below, and the second one is simply Equation (9). Lemma 12. The endomorphism R(ψ) satisfies the symplectic condition, that is R(ψ)∗ = R(−ψ)−1 Where R(ψ)∗ is the adjoint of R(ψ) with respect to the bilinear form η. Proof. Setting κ = 0 in Equation (11) we have η(R(ψ)v1, R(−ψ)v2) = η(v1, v2) that is, R(ψ)∗ = R(−ψ)−1 since R(−ψ) is invertible by Remark 26. Corollary 2. We have (cid:101)Ω+(v1 · v2) = Ω+(R(ψ)v1 ⊗ v2) 33 Proof. By Lemma 11 we have (cid:101)Ω+(v1 · v2)(κ) = η(R(ψ)Z(κ, ψ)v1, v2) we now get the corollary by the fact that Z(κ, ψ) and R(ψ) commute, and by Equa- tion 9. Similarly to the previous subsection, we want to recover Ωg,n from the limit object Ω+. Theorem 6. Let ig,n : C[κj]j≥1 → H•(Mg,n) be the C-linear map that sends each κ-class to the respective class in Mg,n. Then we have Ωg,n(v1 ⊗ . . . ⊗ vn) = ig,n(cid:101)Ω+(αg · R(ψ1)−1v1 · . . . · R(ψn)−1vn)(κ) Proof. Let us consider a multi-sewing map gn,2) → (cid:102)Mg+G,n s : Mg,n × ((cid:102)M{2} g1,2 × ··· × (cid:102)M{2} g+g1+...+gi−1,n ×Mgi,2 → (cid:102)M{1,...,i} that sews the k-th marked point of Mg,n to the first marked point of M{2} for each of the maps si : (cid:102)M{1,...,i−1} k. Here G = g1 + . . . + gn. This map is nothing but the composition, for 1 ≤ i ≤ k, that sew the i-th marked point of the first curve to the first point of the second curve. Now let us apply the sewing axiom for free boundary CohFTs n times, and multiply each i-th entry by α−gi; we get g+g1+...+gi,2 gk,2 Ωg+G,n((α−g1 · v1) ⊗ . . . ⊗ (α−gn · vn))∂Ns = {2} {2} g1,2(eµ1⊗(α−g1·v1))×···×Ω gn,2(eµn⊗(α−gn·vn)) Ωg,n(eµ1⊗. . .⊗eµn)×Ω = ν∗ (cid:88) µ1,...,µn where now ∂Ns is the circular neighbourhood, given by the circular neighbourhood theorem, of the image S of s, and ν : ∂Ns → S is the respective circle bundle. Now let g1, . . . , gn go to infinity, and use Lemma 11 to compute the factors at the right-hand side of the equation. Then we get Ωg,n(eµ1⊗. . .⊗eµn)η(Z(κ,−ψ1)eµ1, v1)··· η(Z(κ,−ψn)eµn, vn) (cid:101)Ω+(αg·v1 ··· vn) = (cid:88) is attached to the free point of (cid:102)M{2} µ1,...,µn where the ψi's are the ψ-classes attached to the marked points of Mg,n (therefore ν∗ψ(cid:48) ). Let κ(g,n) denote the κ-classes on Mg,n, then we have i = −ψi if ψ(cid:48) (cid:101)Ω+(αg · v1 ··· vn)(κ(g,n) + κ) = Ωg,n(Z(κ(1),−ψ1)∗v1 ⊗ . . . ⊗ Z(κ(n),−ψn)∗vn) g1,2 i where κ = κ(1) + . . . + κ(n). Setting one by one each κ(i) = 0 we get Ωg,n(v1 ⊗ . . . ⊗ vn) =(cid:101)Ω+(αg · R(ψ1)−1v1 · . . . · R(ψn)−1vn)(κ(g,n)) = = ig,n(cid:101)Ω+(αg · R(ψ1)−1v1 · . . . · R(ψn)−1vn)(κ) which is the statement of the theorem. 34 for every v ∈ A. Using the fact that p∗κj = κj − ψj for every j, we get p∗ig,1(cid:101)Ω+(v) = ig,2(cid:101)Ω+(v · R(ψ2)−11) (cid:88) φjκj) · exp(−(cid:88) ((cid:101)Ω+ · exp(−(cid:88) 2))(v) =(cid:101)Ω+(v · R(ψ2)−11) φjκj) = exp( (cid:88) φjψj 2 j j j p∗exp( φjψj 2) thus for every v ∈ A, we must have This theorem, in turn, allows us to prove a relation that must exist between (cid:101)Ω+ (cid:101)Ω+ : A → C[κj]j≥1 be group-like, and let R(ψ) ∈ End(A)[[ψ]] be such that R(0) = Id and R(ψ). Proposition 14. Let A be a symmetric and commutative Frobenius algebra. Let and R(ψ)∗ = R(−ψ)−1. Then the formulas of Theorem 6 define a free boundary CohFT if and only if for every v ∈ A, log(cid:101)Ω+(v) = −η(βlog(R(ψ)−11), v) where the logarithm is defined as in Proposition 10 and β : A[[ψ]] → A[[κj]]j≥1 is the A-linear map that sends each ψj to κj. Proof. Let us first suppose that Ωg,n is a free boundary CohFT. Since Ωg,n satisfies the axiom involving the forgetful map, we must have in particular, for every g ≥ 1, which immediately implies R(ψ2)−11 = (exp(−(cid:80) j the dual with respect to η. This is precisely the stated formula. Let us define, for the rest of the proof, j φjψj 2))∗ where the star indicates ωg,n(v1 ⊗ . . . ⊗ vn) = θ(αg · v1 · . . . · vn) for every v1, . . . , vn ∈ A, where θ is the Frobenius trace of A. It is easily proved that ωg,n is a nodal CohFT. Notice that it is nothing but the degree-zero part of any cohomological field theory, of any type, with base A. Let us now suppose that the formula in the statement of the proposition is satis- fied. This formula simply says the following: if(cid:101)Ω+ = exp(φ1κ1+φ2κ2+. . .) (compare with Proposition 13), then R(ψ)−11 = exp(−a1ψ − a2ψ2 − . . .) where the ai are de- fined in such a way that, for every w ∈ A and every i, η(ai, w) = θ(ai · w) = φi(w). Then setting T (z) = z(1 − R(ψ)−11) ∈ z2A[[z]], Lemma 3 says that for every v1, . . . , vn ∈ A, we have θ(αg · R(ψ1)−1v1 · . . . · R(ψn)−1vn ·(cid:88) (pm)∗(T (ψn+1)··· T (ψn+m))) = 1 m! = ig,n(cid:101)Ω+(αg · R(ψ1)−1vn · . . . · R(ψn)−1vn). m≥0 The left-hand side of the last equality is the restriction of Rω of Definition 8 to the smooth parts Mg,n of the moduli spaces. Now, Rω is a nodal CohFT by Proposition 5 and its restriction to Mg,n is a free boundary CohFT by Proposition 4. Thus, the formulas of Theorem 6 define a free boundary CohFT. 35 Looking carefully at the last proof, we see that we do not need semisimplicity: the formulas of Theorem 6 will give a free boundary CohFT even if A is not semisimple. However, it is not true anymore that every free boundary CohFT is given by those formulas, since the very construction of (cid:101)Ω+ requires the invertibility of α, which is equivalent to semisimplicity by Proposition 6. 7 Classification of nodal theories 7.1 Existence of nodal theories Up to now, we have classified smooth theories. In particular, fixing a symmetric, commutative, semisimple Frobenius algebra structure (A,·, 1, η), we have found what follows. • A fixed boundaries theory ((cid:101)Ωg,n)g,n is uniquely determined by a sequence (φj)j≥1 of elements of A∗. Setting (cid:101)Ω+ = exp((cid:80) j≥1 φjκj) : A → C[κj]j≥1, (cid:101)Ωg,n(v1 ⊗ . . . ⊗ vn) = p∗i∗ g(cid:101)Ω+(αg · v1 · . . . · vn) we have with notation as in Theorem 5, for g ≥ 1, while the part g = 0 can be obtained from the part g = 1. theory ((cid:101)Ωg,n)g,n, which is its pull-back under the bundle map π : (cid:102)Mg,n → • A free boundaries theory (Ωg,n)g,n is uniquely determined by a fixed boundaries Mg,n, and by and End(A)-valued power series R(z) that satisfies the symplectic condition and R(0) = Id. In this way we have Ωg,n(v1 ⊗ . . . ⊗ vn) =(cid:101)Ωg,n(R(ψ1)−1v1 ⊗ . . . ⊗ R(ψn)−1vn). The endomorphism R(z) cannot be freely chosen: it must satisfy the condition log(cid:101)Ω+(v) = −η(βlog(R(ψ)−11), v) for every v ∈ A, as stated in Proposition 14. We will now deal with nodal theories. When talking about dual graphs, contri- butions and so on, we refer to the notations of Section 4. Proposition 15. Let (Ωg,n)g,n be a free boundaries theory with base (A,·, 1, η). Then there exists a nodal theory (Ωg,n)g,n whose restriction to the smooth part Mg,n of the moduli space Mg,n is (Ωg,n)g,n. Proof. Let R(z) be the End(A)-valued power series determined by the theory Ω. By Corollary 12 and Remark 26, R(z) satisfies the symplectic condition and R(0) = Id; it also satisfy the compatibility condition of Proposition 14. Let ωg,n : A⊗n → H•(Mg,n) be defined as ωg,n(v1 ⊗ . . . ⊗ vn) = θ(αg · v1 · . . . · vn) where θ is the Frobenius trace of A. Then it is easy to see that ω is a CohFT. Notice that this is simply the degree zero part of Ω. We define, as in the proof of Proposition 14, Ω = Rω 36 using the R-matrix action of Definition 8. By Proposition 5, Ω is a nodal field theory, and the fact that its restriction to Mg,n is precisely Ω is proven in Proposition 14. 7.2 Uniqueness for nodal theories Now that we have seen how to construct a nodal theory from a smooth one, we pro- ceed to show uniqueness. Notice that we did not mention semisimplicity in Proposi- tion 15; however, in the following theorem, semisimplicity is essential. Theorem 7. Let Ω be a free boundaries CohFT with semisimple base (A,·, 1, η). Then there exists a unique nodal CohFT Ω that has Ω as its restriction to the smooth stratum Mg,n. We first prove a general lemma which explains how an ambiguity can arise when patching two cohomology classes. Lemma 13. Let M be a smooth complex manifold and let j : S → M be the immer- sion of a closed smooth submanifold of complex codimension c. Let νS : N → S be its normal bundle with N the tubular neighbourhood of S given by the tubular neighbour- hood theorem. Let [α1], [α2] ∈ H k(M ) such that [α1]S = [α2]S and [α1]Sc = [α2]Sc. Then [α1] − [α2] ∈ j∗Ann H k−2c(S)(c(νS)) where c(νS) is the Chern class of the vector bundle νS. ∼−→ Proof. The Thom isomorphism theorem says that, for every integer k, j∗ : H k−2c(S) H k(N, N \ S), and the excision property of cohomology implies that H k(N, N \ S) (cid:39) H k(M, M \ S). Notice that excision can be applied as long as S is a closed subman- ifold. This allows us to write the cohomology long exact seqence of the immersion M \ S ⊆ M as ··· → H k−2c(S) → H k(M ) → H k(M \ S) → H k−2c+1(S) → ··· Now, [α1] − [α2] = j∗(b) for some b ∈ H k−2c(S) since [α1] and [α2] go to same class when pushed to H k(M \ S). Similarly, ([α1] − [α2])S = 0. But j∗(b)S = eul(νS) · b, therefore we get our result. Thus to prove the uniqueness of patchings we have to show that we can always reduce ourselves to the case in which c(νS) is not a zero divisor. To do so, we will construct a special stratification of the moduli space Mg,n. 7.2.1 Stratification of Mg,n Definition 18. Suppose n > 0 and let C be a curve in Mg,n. The special component of C is its irreducible component containing the n-th marked point. The connected components of the curve obtained from C by cancelling out the special component are called non-special. Fixing a curve C, the datum of • the topological type of the special component of C, • the number of marked points of the special component C, 37 • the number of nodes that link the special component of C with the non-special ones, is called the special type of C. In other words, two curves in Mg,n are said to be of the same special type if their special components are homeomorphic, have the same number of marked points, and the same number of nodes that link them to the non-special components. We denote by Mg,n τ ⊆ Mg,n the set of curves with special type τ. τ C C Remark 27. Let C ∈ Mg,n be a curve and let ΓC be its dual graph. Let vsp ∈ ΓC be the vertex to which the leg labeled n is attached. Then the special type of C is determined by the subgraph Γ(cid:48) obtained from C by cancelling out all the vertices except vsp, and all the edges except those that have vsp at one end. Thus, if C ∈ S where S is an open boundary stratum of Mg,n, and if τ is the special type determined by Γ(cid:48) . Indeed, all the curves of an open boundary stratum have the same dual graph. This shows that each Mg,n , then S ⊆ Mg,n is a union of open boundary strata. Since two different σ = ∅. τ ∩Mg,n open boundary strata are disjoint, we see at once that if τ (cid:54)= σ, then Mg,n This is just another way to say that the special type of a curve is well defined. τ ⊆ Mg,n with τ a special type represented by a smooth Remark 28. Let Mg,n special component Csp of genus γ with ν + k marked points (the nodes linking Csp to the non-special components are counted as marked points as well, and they account for the k term in the previous expression). Let C ∈ Mg,n and let C(cid:48) be the curve obtained from C by eliminating the special component. Let C(cid:48) = C(cid:48) be its decomposition in connected components, with C(cid:48) for each i, µi being the number of nodes that C(cid:48) shares with Csp. Then we have the relations i ∈ Mgi,ni+µi 1 (cid:116) . . .(cid:116) C(cid:48) τ τ l i γ + (gi + µi − 1) = g , ν + ni = n and µi = k (12) i=1 n(cid:48) + µ(cid:48) > 0, g(cid:48) ≤ g, n(cid:48) ≤ n and µ(cid:48) ≤ g + 1, let us write MA =(cid:81) For A a finite set made of triples (g(cid:48), n(cid:48), µ(cid:48)) of non-negative integers such that 2g(cid:48)−2+ (gi,ni,µi)∈A Mgi,ni+µi . Let X be the family of the A's whose elements satisfy Relations (12). Then we have i=1 i=1 l(cid:88) l(cid:88) l(cid:88) (cid:71) A∈X Mg,n τ = (MA × Mγ,ν+k)/FA where FA is the (finite) group of automorphisms of some dual graph that depends on A and that is not of great interest here. Notice that the sets MA are open strata which form, taking the union, a bunch of spaces M i, i ∈ I, that are products of closed moduli spaces. The A's that contribute to form a certain M i must give the same automorphism group FA = Fi, since this group is determined by the nodes that the special component shares with the non-special ones, and it does not depend on the specific topological type of the non-special components. Thus we can rewrite the previous decomposition as Mg,n (M i × Mγ,ν+k)/Fi (cid:71) (13) τ = i∈I 38 τ in connected components, all with the same dimension. Using Relations (12) we see that the codimension in Mg,n of each component (M i × Mγ,ν+k)/FA is k, that depends only on τ. Therefore, in (13) we have decomposed Mg,n Let τ be a special type represented by a non-smooth special component Csp ∈ Mγ(cid:48),ν(cid:48)+k, which must therefore have only non-separating nodes. If µ is the number of non-separating nodes of Csp, if γ = γ(cid:48) − µ and ν = ν(cid:48) + 2µ, then the normalization of the special component lies in Mγ,ν+k. The decomposition (13) is still valid, with γ(cid:48), ν(cid:48) instead of γ, ν in Relations (12) (but not in decomposition (13)!). The decomposition is still equidimensional, and the codimension of Mg,n in Mg,n is now µ + k. τ 7.2.2 Order of the strata In all this section we will fix, once and for all, an ambient space Mg,n where all the strata lie. We will omit the indices g and n when it causes no confusion. Definition 19. We say that two special types are such that τ > τ(cid:48) (the special type τ is greater then τ(cid:48)) if τ (cid:54)= τ(cid:48) and at least one point of Mτ(cid:48) lies in the closure of Mτ , that is, Mτ ∩ Mτ(cid:48) (cid:54)= ∅. This relation is transitive and therefore determines a partial order on the special types. Notation 5. Denote by Uτ the union(cid:83) τ(cid:48)≥τ Mτ(cid:48). Proposition 16. The set Uτ is open in Mg,n. Proof. This follows from the definition of the order: for any x ∈ Mτ(cid:48) and y close enough to x, we have y ∈ Mτ(cid:48)(cid:48) for some τ(cid:48)(cid:48) ≥ τ(cid:48). Example 5. Here is a graphic representation of the stratification we have defined in the case of M1,2. The special strata are four, labeled in the following figure by the letters A, B, C and D. Notice that they are disjoint (see Remark 27). 39 12121212ABCDB = XCA =CC = D = 12X = The partial order of these strata is shown in the diagram, the smaller strata being on the left. Note that the order is not compatible with the dimensions of the strata: for instance, we have B < C even though these cells have the same dimension. This is due to the fact that B does not lie in the closure of C entirely, but contains one point X that does. In the example above we can observe that A, A ∪ C, A ∪ C ∪ B, A ∪ C ∪ D, and A ∪ B ∪ C ∪ D are open. Moreover • C is closed and smooth in C ∪ A with a normal bundle of rank 1; • B is closed and smooth in B ∪ C ∪ A with a normal bundle of rank 1; • D is closed and smooth in D ∪ C ∪ A with a normal bundle of rank 2. We will now see that these properties hold in general. Lemma 14. Every Mτ is a smooth closed sub-orbifold of Uτ . Proof. Suppose that Mτ has self-intersections, let Σ be a point in the self-intersection, and let σ ∈ Aut(Σ) be an automorphism that switches two intersecting branches of Mτ . Now, let V ∈ TΣMτ be tangent to one of the intersecting branches (therefore it is not tangent to the other branch, since all the strata have normal crossings). Then V represents a deformation of Σ which does not change the special component, and a deformation of σ(Σ) which changes the special component. Since however any automorphism of Σ must preserve its special component, we have a contradiction. Let x ∈ Uτ \ Mτ , so that x ∈ Mτ(cid:48) for some τ(cid:48) > τ. If there were no neighbour- hoods V of x with V ∩Mτ = ∅, then x ∈ Mτ , so that Mτ(cid:48) ∩Mτ (cid:54)= ∅ and τ > τ(cid:48) by definition of the order. This contradiction shows that Mτ ⊆ Uτ is closed. Lemma 15. The normal bundle of Mτ ⊆ Uτ is the direct sum of the line bundles L(cid:48) ⊗ L(cid:48)(cid:48) for (L(cid:48), L(cid:48)(cid:48)) tangent pairs at the nodes that link the special component to the non-special ones, and at the non-separating nodes of the special component. Proof. Let us keep the notations of Remark 28. First, let τ be a special type repre- sented by a smooth special component. From decomposition (13) of Remark 28, it suffices to show the statement for each (MA × Mγ,ν+k)/FA. This last object is a semi-open boundary stratum, which is the image of a sewing map s : MA×Mγ,ν+k → Mg,n. Thus its normal bundle is the sum of the line bundles L(cid:48)⊗L(cid:48)(cid:48) for tangent pairs (L(cid:48), L(cid:48)(cid:48)) at the points sewed together by s. Since the special component is smooth, the lemma is proved for this case. Let now τ be a special type represented by a non-smooth special component Csp ∈ Mγ(cid:48),ν(cid:48)+k whose normalization lies in Mγ,ν+k. Then by Remark 28, each com- ponent of the decomposition (13) is the image of a sewing map s : MA ×Mγ,ν+k → Mg−µ,n+2µ followed by a non-separating sewing map q : Mg−µ,n+2µ → Mg,n. The normal bundle to the image of q is the sum of the line bundles L(cid:48) ⊗ L(cid:48)(cid:48) for tangent 40 DCBApairs (L(cid:48), L(cid:48)(cid:48)) at the points sewed together by q. These account for the non-separating nodes of the special components, while the other nodes have been dealt with in the smooth case, and the lemma is proved. Lemma 16. Let us consider the decomposition (13) of Remark 28: (cid:71) A∈X Mτ = (MA × Mγ,ν+k)/FA. Then, below degree γ/3, the Chern class of the normal bundle ντ of Mτ in Uτ is not a zero divisor. In other words, every element of Ann H•−2k(Mτ )(c(ντ )) has degree strictly higher than γ/3. Proof. Let us write (cid:71) A∈X Mτ = (MA × Mγ,ν+k)/FA := (cid:71) A∈X TA with notation as in Remark 28. Looijenga's theorem says that, below degree γ/3 (the so-called "stable range "), H•(Mγ,ν+k) (cid:39) H•(Mγ)[ψ1, . . . , ψν+k]. In particular the ψ-classes are free generators of the stable range of the cohomology of Mγ,ν+k. Now, for each A ∈ X, we have H•(TA) ⊆ H•(MA) ⊗ H•(Mγ,ν+k), and c(νTA) = − k(cid:88) (ψ(cid:48) i ⊗ 1 + 1 ⊗ ψν+i) ∈ H•(TA) i=1 i γ/3. By Lemma 15, we have c(ντ ) =(cid:80) is the ψ-class on MA at the point sewed to the point marked where for each i, ψ(cid:48) ν + i on Mγ,ν+k. This implies that each c(νTA) is not a zero-divisor below degree c(νTA), and this is in fact a direct sum. Therefore we conclude that c(ντ ) is not a zero-divisor below degree γ/3, which is what we had to prove. A∈X We will need one more technical lemma. Lemma 17. Let τ0 be a special stratum and suppose that the restriction of Ωg,n on each stratum Mτ ⊆ Mg,n with τ ≥ τ0 is known. Then the restriction of Ωg,n to each Uτ with τ ≥ τ0 is known below degree γ/3, where γ is the genus of the special component corresponding to the stratum τ. Proof. We reason by descending induction using the order of Definition 19 on the special strata. If τ is the biggest special type, then Uτ = Mτ , where Ωg,n is deter- mined by hypothesis. Now if we have determined the theory on each Uτ for τ in some union of intervals I containing the biggest special type, choose a maximal τ(cid:48) /∈ I and consider Uτ(cid:48). Then, by induction, the theory is determined on Uτ(cid:48) \Mτ(cid:48) up to degree γ(cid:48)/3 (notice that if τ(cid:48) ≤ τ then γ(cid:48) ≤ γ). By Lemma 14, Mτ(cid:48) is a smooth closed sub-orbifold of Uτ(cid:48). Then we can patch, uniquely below degree γ(cid:48)/3, the theories on Mτ(cid:48) and on Uτ(cid:48) \ Mτ(cid:48) thanks to Lemmas 13 and 16. 41 Now we are ready to prove the "uniqueness" part of the classification of nodal theories. We keep the notation of the beginning of this section. Theorem 8. A nodal CohFT Ω is uniquely determined by its restriction Ω to the smooth part Mg,n. Proof. By definition, Ω0,3(v1 ⊗ v2 ⊗ v3) = η(v1 · v2, v3) = Ω0,3(v1 ⊗ v2 ⊗ v3) thus Ω0,3 is uniquely determined by Ω0,3. and let 3g − 3 + n = d. We have to show that we can determine Ωg,n. Let d ≥ 1, let us suppose we know Ωg(cid:48),n(cid:48) for every (g(cid:48), n(cid:48)) such that 3g(cid:48)−3+n(cid:48) < d, Let G > 9g − 9 + 3n, and let us consider the sewing map s : Mg,n × MG,2 → Mg+G,n that identifies the points labeled n and 2 respectively (notice the lack of the overline in the second space: we only want to consider sewings of smooth surfaces). Therefore, the point marked 1 on MG,2 goes to the point marked n on Mg+G,n. Let N be the usual tubular neighbourhood of the boundary stratum S in which the image of this map lies. Let I be the set of the τ's such that Mg+G,n ∩ ∂N (cid:54)= ∅, and let τ (cid:91) τ∈I U = Mg+G,n τ This accounts for a finite stratification of U in which every stratum has special component of genus G or higher. Here is a graphic representation of the strata of U in the case g = 1, n = 2. See also Example 5 for more comments. Now every special stratum Mg+G,n with τ ∈ I is a product of a smooth moduli space and some Mg(cid:48),n(cid:48)'s with 3g(cid:48) − 3 + n(cid:48) < d (see Remark 28). Thus our inductive τ 42 1212G12GABCDB = XCA =CC = D = 1GG+1G122X = τ is uniquely determined. hypothesis together with the sewing axiom implies that the restriction of Ωg+G,n to each Mg+G,n Let ΩU be the restriction of Ωg+G,n to U. Notice that U is contained in Uτ0 where τ0 is the special stratum (in Mg+G,n) corresponding to curves whose special component is a smooth curve of genus G with one marked point and one node attaching it to the other components. By Lemma 17, ΩU is uniquely determined below degree G/3. Since ∂N ⊆ U, we see that the restriction of Ωg+G,n to ∂N is uniquely determined below degree G/3. Now the sewing axiom yields Ωg+G,n(v1 ⊗ . . . ⊗ vn)∂N = ν∗ηµνΩg,n(v1 ⊗ . . . ⊗ eµ) × ΩG,2(vn, eν) and the Gysin sequence for the circle bundle ∂N → S implies that ν∗ induces an isomorphism below degree G/3 H•(∂N ) (cid:39) H•(Mg,n) ⊗ H•(MG)[ψ(cid:48) 1] is the ψ-class attached to the first marked point in MG,2. We can conclude where ψ(cid:48) that 1 ηµνΩg,n(v1 ⊗ . . . ⊗ eµ) × ΩG,2(vn, eν) is known below degree G/3. Restricting to any fixed smooth surface Σ ∈ MG,2 gives Ωg,n(v1 ⊗ . . . ⊗ (αG · vn)), (see the proofs of Propositions 8-10) and since in a semisimple theory α is invertible, we conclude that Ωg,n is uniquely determined below degree G/3 > 3g − 3 + n = dimMg,n. Thus Ωg,n is totally determined: this completes the inductive step and the proof. References [1] Abrams, L. Two-dimensional topological quantum field theories and Frobenius algebras. J. Knot Theory Ramifications 5, 569-587 (1996) [2] Abrams, L. The quantum Euler class and the quantum cohomology of the Grass- mannians. Israel J. Math. 117, 335-352 (2000) [3] Arbarello, E.; Cornalba, M.; Griffiths, P. A.; Harris, J. D. Geometry of algebraic curves, vol. I et II. Springer "A Series of Comprehensive Studies in Mathematics", vol. 268 (1985-2011). [4] Cannas da Silva, A. Lectures on Symplectic Geometry. Springer Lecture Notes in Mathematics, 1764 (2006) [5] Dubrovin, B. Geometry and analytic arXiv:math/9807034 (1998) theory of Frobenius manifolds. [6] Fulton, W.; Pandharipande, R. Notes on stable maps and quantum cohomology. arXiv:alg-geom/9608011v2 (1997) 43 [7] Givental, A. Semi-simple Frobenius structures in higher genus. Internat. Math. Res. Notices 23, 1265-1286 (2001) [8] Harer, J. L. Stability of the homology of the mapping class groups of orientable surfaces. Ann. of Math. (2) 121, 215-249 (1985) [9] Kontsevich, M.; Manin, Y. Gromov-Witten classes, quantum cohomology, and enumerative geometry. Comm.Math.Phys.,164:3 (1994), 525-562, et Mirror Sym- metry II, B.Greene and S.-T.Yau (editors), (1997). AMS and International Press, 607-654. [10] Madsen, I.; Weiss, M. The stable mapping class group and stable homotopy theory. European Congress of Mathematics, 283-307, Eur. Math. Soc., Zürich (2005) [11] Mumford, D.; Fogarty, J.; Kirwan, F. Geometric invariant theory, 3rd edition. Springer "A Series of Modern Surveys in Mathematics", vol. 34 (1994) [12] Pandharipande, R.; Pixton, A.; Zvonkine, D. Relations on Mg,n via 3-spin structures. arXiv:1303.1043v2 (2013) [13] Teleman, C. The Structure of 2D Semi-simple Field Theories. arXiv:0712.0160v3 (2011) [14] Wahl, N. Homological stability for mapping class groups of surfaces. Handbook of Moduli, vol. III, 547-583. Advanced Lectures in Mathematics 26 (2013) [15] Zvonkine, D. An introduction to moduli spaces of curves and their intersection theory. Leçons des journées mathématiques de Glanon (2014) 44
0910.2676	3	0910	2010-01-20T21:30:50	Hodge groups of certain superelliptic jacobians	[ "math.AG", "math.NT" ]	Suppose that $K$ is a field of characteristic 0, $p$ is an odd prime, $r$ a positive integer, $q=p^r$ a prime power. Suppose that $f(x)$ is a polynomial of degree $n > 4$ with coefficients in $K$ and without multiple roots. Let us consider the superelliptic curve $C: y^q=f(x)$ and its jacobian $J(C)$. Assuming that $K$ is a subfield of the field of complex numbers, we study the (connected reductive algebraic) Hodge group $Hdg$ of the corresponding complex abelian variety $J(C)$. In our previous paper (arXiv:0907.1563 [math.AG]) we studied the center of $Hdg. In this paper we study the semisimple part (commutator subgroup) of $Hdg$. Assuming that $p$ does not divide $n$ and $n-1$ is not divisible by $q$, the Galois group of $f(x)$ over $K$ is either the full symmetric group $S_n$ or the alternating group $A_n$, we prove that the semisimple part of $Hdg$ is "as large as possible".	math.AG	math	HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS JIANGWEI XUE AND YURI G. ZARHIN Throughout this paper K is a field of characteristic zero, ¯K its algebraic closure and Gal(K) = Aut( ¯K/K) the absolute Galois group of K. If X is an abelian variety over ¯K then we write End(X) for the ring of all its ¯K-endomorphisms and End0(X) for the corresponding Q-algebra End(X) ⊗ Q; the notation 1X stands for the identity automorphism of X. Let f (x) ∈ K[x] be a polynomial of degree n ≥ 3 with coefficients in K and without multiple roots, Rf ⊂ ¯K the (n-element) set of roots of f and K(Rf ) ⊂ ¯K the splitting field of f . We write Gal(f ) = Gal(f /K) for the Galois group Gal(K(Rf )/K) of f ; it permutes roots of f and may be viewed as a certain per- mutation group of Rf , i.e., as as a subgroup of the group Perm(Rf ) ∼= Sn of permutation of Rf . (Gal(f ) is transitive if and only if f is irreducible.) Suppose that p is a prime that does not divide n and a positive integer q = pr is a power of p. We write Cf,q for the superelliptic K-curve yq = f (x) and J(Cf,q) for its jacobian. Clearly, J(Cf,q) is an abelian variety that is defined over K and dim(J(Cf,q)) = (n − 1)(q − 1) 2 . (canonically) isomorphic to a product Qr In a series of papers Assume that K contains a primitive qth root of unity ζq. [12, 14, 15, 16], one of the authors (Y.Z.) discussed the structure of End0(J(Cf,q)), assuming that n ≥ 5 and the Galois group Gal(f ) of f (x) over K is, at least, doubly transitive. In particular, he proved that if n ≥ 5 and Gal(f ) coincides either with full symmetric group Sn or with alternating group An then End0(J(Cf,q)) is i=1 Q(ζpi ) of cyclotomic fields. (If q = p then we proved that End(J(Cf,p)) = Z[ζp].) More precisely, if q 6= p then the map (x, y) → (x, yp) defines the map of curves Cf,q → Cf,q/p, which induces (by Albanese functoriality) the surjective homomorphism J(Cf,q) → J(Cf,q/p) of abelian varieties over K; we write J (f,q) for the identity component of its kernel. (If q = p then we put J (f,q) = J(Cf,p).) One may check [16] that J(Cf,q) is K- i=1 J (f,pi) and the automorphism δq : (x, y) 7→ (x, ζqy) of Cf,q gives rise to an embedding Z[ζq] ֒→ End(J (f,q)), ζq 7→ δq. What was actually proved in [16, 18] is that isogenous to the productQr Z[ζq] ∼= Z[δq] = End(J (f,q)) if Gal(f ) = Sn, p ≥ 3 and n ≥ 4 or Gal(f ) = An and n ≥ 5. Let us assume that K ⊂ C and let the field ¯K be the algebraic closure of K in C. This allows us to consider J(Cf,q) and J (f,q) as complex abelian varieties. Our goal is to study the (reductive Q-algebraic connected) Hodge group Hdg = Hdg(J (f,q)) of J (f,q). Notice that when q = 2 (i.e., in the hyperelliptic case) this group was completely determined in [13] (when f (x) has "large" Galois group). When q > 2 1 2 JIANGWEI XUE AND YURI G. ZARHIN we determined in our previous paper [9] the center of Hdg(J (f,q)), also assuming that the Galois group of f (x) is "large". Let us assume that q > 2. In order to describe our results let us recall that the jacobian J(Cf,q) carries the canonical principal polarization that is invariant under all automorphisms (induced by automorphisms) of Cf,q. This implies that the in- duced polarization on the abelian subvariety J (f,q) is δq-invariant. This polarization gives rise to the δq-invariant nondegenerate alternating Q-bilinear form ψq : H1(J (f,q), Q) × H1(J (f,q), Q) → Q on the first rational homology group of the complex abelian variety J (f,q). On the other hand, H1(J (f,q), Q) carries the natural structure of Q[δq] ∼= Q(ζq)-vector space. The δq-invariance of ψq implies that ψq(ex, y) = ψq(x, ¯ey) ∀e ∈ Q(ζq); x, y ∈ H1(J (f,q), Q). Here e 7→ ¯e stands for the complex conjugation map. Let Q(ζq)+ = {e ∈ Q(ζq) ¯e = e} be the maximal totally real subfield of the cyclotomic CM field Q(ζq) and let Q(ζq)− = {e ∈ Q(ζq) ¯e = −e}. Pick a non-zero element α ∈ Q(ζq)−. Now the standard construction (see, for instance, [5, p. 531]) allows us to define the non-degenerate Q(ζq)-sesquilinear Hermitian form φq : H1(J (f,q), Q) × H1(J (f,q), Q) → Q(ζq) such that ψq(x, y) = TrQ(ζq )/Q(αφq(x, y)) ∀x, y ∈ H1(J (f,q), Q). We write U(H1(J (f,q), Q), φq) for the unitary group of φq of the Q(ζq)-vector space H1(J (f,q), Q), viewed as an algebraic Q-subgroup of GL(H1(J (f,q), Q)) (via Weil's restriction of scalars from Q(ζq)+ to Q (ibid). Since the Hodge group respects the polarization and commutes with endomorphisms of J (f,q), Hdg(J (f,q)) ⊂ U(H1(J (f,q), Q), φq). Our main result is the following statement. Theorem 0.1. Suppose that n ≥ 4 and p is a prime that does not divide n. Let f (x) ∈ C[x] be a degree n polynomial without multiple roots. Let r be a positive integer and q = pr. Suppose that there exists a subfield K of C that contains all the coefficients of f (x). Let us assume that f (x) is irreducible over K and the Galois group Gal(f ) of f (x) over K is either Sn or An. Assume additionally that either n ≥ 5 or n = 4 and Gal(f ) = S4. Suppose that n > q and one of the following three conditions holds: (A) q < n < 2q; (B) p is odd and n 6≡ 1 mod q; (C) p = 2, n 6≡ 1 mod q and n 6≡ q − 1 mod 2q. Then Hdg(J (f,q)) = U(H1(J (f,q), Q), φq). Remark 0.2. The case of q = p = 3 was earlier treated in [14]. HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 3 Lefschetz's theorem about algebraicity of 2-dimensional Hodge classes and clas- sical invariant theory for the unitary groups [5, Theorem 0 on p. 524; see also pp. 531 -- 532 ] imply the following corollary to Theorem 0.1. Corollary 0.3. Let (n, p, q, f (x)) satisfy the conditions of Theorem 0.1. Then every Hodge class on each self-product (J (f,q))m of J (f,q) can be presented as a linear combination with rational coefficients of products of divisor classes. In particular, the Hodge conjecture holds true for (J (f,q))m. The paper is organized as follows. In Section 1 we discuss Lie algebras of Hodge groups of complex abelian varieties. Its main result, Theorem 1.1 (that may be of independent interest) asserts that under certain conditions the semisimple part of the Hodge group (and its Lie algebra) is "as large as possible". We deduce Theorem 0.1 from Theorem 1.1, using auxiliary results from Section 2 concerning divisibility properties of certain arithmetic functions. In Section 3 we discuss linear reductive Lie algebras. The last Section contains the proof of Theorem 1.1. 1. Complex abelian varieties Let Z be a complex abelian variety of positive dimension and let Ω1(Z) be the dim(Z)-dimensional complex vector space of regular differential 1-forms on Z. We write CZ for the center of the semisimple finite-dimensional Q-algebra End0(Z). We write H1(Z, Q) for its first rational homology group. It is well known that H1(Z, Q) is a 2dim(Z)-dimensional Q-vector space. The Q-algebra End0(Z) acts faithfully on H1(Z, Q). In particular, if E is a subfield of End0(Z) that contains the identity map then H1(Z, Q)carries the natural structure of E-vector space of dimension d(Z, E) = 2dim(Z) [E : Q] . Let ΣE be the set of all field embeddings σ : E ֒→ C. It is well-known that Cσ := E ⊗E,σ C = C, EC = E ⊗Q C = Yσ∈ΣE E ⊗E,σ C = Yσ∈ΣE Cσ. If σ ∈ ΣE then we write ¯σ for the complex-conjugate of σ. We write XE for the Q-vector space of functions φ : ΣE → Q with φ(¯σ) + φ(σ) = 0 ∀σ ∈ ΣE. If E/Q is Galois then XE carries the natural structure of Gal(E/Q)-module. Let Lie(Z) be the tangent space to the origin of Z; it is a dim(Z)-dimensional C- vector space. By functoriality, End0(Z) and therefore E act on Lie(Z) and therefore provide Lie(Z) with a natural structure of E ⊗Q C-module. Clearly, Lie(Z) = Mσ∈ΣE CσLie(Z) = ⊕σ∈ΣE Lie(Z)σ where Lie(Z)σ := CσLie(Z) = {x ∈ Lie(Z) ex = σ(e)x ∀e ∈ E}. Let us put nσ = nσ(Z, E) = dimCσ Lie(Z)σ = dimCLie(Z)σ. It is well-known that the natural map Ω1(Z) → HomC(Lie(Z), C) is an isomorphism. This allows us to define via duality the natural homomorphism E → EndC(HomC(Lie(Z), C)) = 4 JIANGWEI XUE AND YURI G. ZARHIN EndC(Ω1(Z)). This provides Ω1(Z) with a natural structure of E ⊗Q C-module in such a way that Ω1(Z)σ := CσΩ1(Z) ∼= HomC(Lie(Z)σ, C). In particular, nσ = dimC(Lie(Z)σ) = dimC(Ω1(Z)σ) (1). Let us consider the first complex homology group of Z H1(Z, C) = H1(Z, Q) ⊗Q C, which is a 2dim(Z)-dimensional complex vector space. If E is as above then H1(Z, C) carries the natural structure of a free EC := E ⊗Q C-module of rank d(Z, E). We have H1(Z, C) = Mσ∈ΣE CσH1(Z, C) = ⊕σ∈ΣE H1(Z, Q)σ where H1(Z, Q)σ := CσH1(Z, C) = {x ∈ H1(Z, C) ex = σ(e)x ∀e ∈ E} = H1(Z, Q)⊗E,σC. Clearly, every H1(Z, Q)σ is a d(Z, E)-dimensional C-vector subspace that is also a EC-submodule of H1(Z, C). There is a canonical Hodge decomposition ([3, chapter 1], [1, pp. 52 -- 53]) H1(Z, C) = H −1,0 ⊕ H 0,−1 where H −1,0 = H −1,0(Z) and H 0,−1 = H 0,−1(Z) are mutually "complex conjugate" dim(Z)-dimensional complex vector spaces. This splitting is End0(Z)-invariant and the End0(Z)-module H −1,0 is canonically isomorphic to the commutative Lie algebra Lie(Z) of Z. This implies that H −1,0 and Lie(Z) are isomorphic as E- modules and even as EC-modules. Let be the C-linear operator in H1(Z, C) defined as follows. f0 H = f0 H,Z : H1(Z, C) → H1(Z, C) fH (x) = − 1 2 x ∀ x ∈ H −1,0; f0 H (x) = 1 2 x ∀ x ∈ H 0,−1. H commutes with End0(Z). In particular, every H1(Z, Q)σ is f0 Clearly, f0 H -invariant. More precisely, the linear operator f0 H : H1(Z, Q)σ → H1(Z, Q)σ is semisimple and its spectrum lies in the two-element set {1/2, −1/2}. Taking into account that the EC-modules H −1,0 and Lie(Z) are isomorphic, we conclude that the multiplicity of eigenvalue −1/2 is nσ = nσ(Z, E) while the multiplicity of eigenvalue 1/2 is d(Z, E) − nσ(Z, E). Let ¯σ : E ֒→ C be the composition of σ : E ֒→ C and the complex conjugation C → C. It is known ([1], [2]) that nσ + n¯σ = d(Z, E). This implies that the multiplicity of eigenvalue 1/2 is n¯σ. We refer to [5], [10, Sect. 6.6.1 and 6.6.2] for the definition and basic proper- ties of the Hodge group (aka special Mumford -- Tate group) Hdg = HdgZ of (the rational Hodge structure H1(Z, Q) and of) Z. Recall that Hdg is a connected reduc- tive algebraic Q-subgroup of GL(H1(Z, Q)), whose centralizer in EndQ(H1(Z, Q)) coincides with End0(Z). Let hdg = hdgZ ⊂ EndQ(H1(Z, Q)) HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 5 be the Q-Lie algebra of Hdg; it is a reductive Q-Lie subalgebra of EndQ(H1(Z, Q)), its natural representation in H1(Z, Q) is completely reducible and its centralizer there coincides with End0(Z). Notice also that its complexification hdgC = hdg ⊗Q C ⊂ EndQ(H1(Z, Q)) ⊗Q C = EndC(H1(Z, C)) H [9, Sect. 3.4]. contains f0 Suppose that E = End0(Z) is a CM field. Choose a polarization on Z. The corresponding Rosati involution on End0(Z) coincides with the complex conjugation e 7→ ¯e on E. The polarization gives rise to the nondegenerate alternating Q-bilinear form ψ : H1(Z, Q) × H1(Z, Q) → Q ψ(ex, y) = ψ(x, ¯ey) ∀x, y ∈ H1(Z, Q); e ∈ E. such that Let be the maximal totally real subfield of the CM field E and let E+ = {e ∈ E ¯e = e} E− = {e ∈ E ¯e = −e}. Pick a non-zero element α ∈ Q(ζq)−. Now the standard construction (see, for in- stance, [5, p. 531]) allows us to define the non-degenerate E-sesquilinear Hermitian form φ : H1(Z, Q) × H1(Z, Q) → E such that ψ(x, y) = TrE/Q(αφ(x, y)) ∀x, y ∈ H1(Z, Q). We write U(H1(Z, Q), φ) for the unitary group of φ of the E-vector space H1(Z, Q), viewed as an algebraic Q-subgroup of GL(H1(Z, Q)) (via Weil's restriction of scalars from E+ to Q (ibid). It is well-known that U(H1(Z, Q), φ) is reductive and its Q- dimension is [E+ : Q]d(Z, E)2 = [E : Q]d(Z, E)2. 1 2 Let u(H1(Z, Q), φ) be the Q-Lie algebra of U(H1(Z, Q), φ): it is a reductive Q-Lie subalgebra of EndQ(H1(Z, Q)). Explicitly, u(H1(Z, Q), φ) = {u ∈ EndE(H1(Z, Q)) φ(ux, y)+φ(x, uy) = 0 ∀x, y ∈ H1(Z, Q)}. The reductive Q-Lie algebra u(H1(Z, Q), φ) splits into a direct sum u(H1(Z, Q), φ) = E− ⊕ su(H1(Z, Q), φ) of its center E− and the semisimple Q-Lie algebra su(H1(Z, Q), φ) = {u ∈ u(H1(Z, Q), φ) TrE(u) = 0.} Here TrE : EndE(H1(Z, Q)) → E is the trace map. One may easily check that 1 2 dimQ(su(H1(Z, Q), φ)) = [E : Q]{d(Z, E)2 − 1}. Since the Hodge group respects the polarization and commutes with endomor- phisms of Z, Hdg(Z) ⊂ U(H1(Z, Q), φ) 6 JIANGWEI XUE AND YURI G. ZARHIN (ibid). This implies that hdg ⊂ u(H1(Z, Q), φ) ⊂ EndQ(H1(Z, Q)). This implies that the semisimple part hdgss = [hdg, hdg] of hdg lies in su(H1(Z, Q), φ). In particular, dimQ(hdgss) ≤ 1 2 [E : Q]{d(Z, E)2 − 1}; the equality holds if and only if hdgss = su(H1(Z, Q), φ). The following statement may be viewed as a partial generalization of Theorem 3 in [5, p. 526]. Theorem 1.1. Suppose that E = End0(Z) is a CM field. Assume that all nσ(Z, E) are distinct positive integers. Assume additionally that there exists a field embedding σ : E ֒→ C such that nσ(Z, E) and d(Z, E) are relatively prime. Then hdgss = su(H1(Z, Q), φ). Remark 1.2. Clearly, in the course of the proof of Theorem 1.1, it suffices to check that dimQ(hdgss) ≥ [E : Q]{d(Z, E)2 − 1}. 1 2 We prove Theorem 1.1 in Section 4. Proof of Theorem 0.1. Let us put Z = J (f,q) and E = Q(ζq). Clearly, d(Z, E) = n − 1. We know that End0(Z) = E = Q(ζq) [16, 18] and if σ = σi : Q(ζq) ֒→ C is a field embedding that sends ζq to ζ −i q with 1 ≤ i < q, (i, p) = 1 then nσ = [ni/q] [16, 17]. (Clearly, every field embedding Q(ζq) ֒→ C is of the form σi.) Since n > q, all integers [ni/q] are positive and distinct. Propositions 2.1 and 2.2 below imply that under our assumptions on (p, q, n) there exists a positive integer i < q such that (i, p) = 1 and [ni/q] and n − 1 are relatively prime. This allows us to apply Theorem 1.1 and conclude that hdgss = su(H1(Z, Q), φ). On the other hand, by a result from [9], the center of hdg coincides with E−. This implies that the reductive Q-Lie algebra hdg coincides with the direct sum E− ⊕ su(H1(Z, Q), φ) = u(H1(Z, Q), φ). Now the connectedness of the Hodge group and the unitary group implies that Hdg(Z) = U(H1(Z, Q), φ), i.e., Hdg(J (f,q)) = U(H1(J (f,q), Q), φq). (cid:3) 2. Divisibility properties of integral parts Proposition 2.1. Suppose that p is a prime, r a positive integer, q = pr and n is a positive integer that is not divisible by p. Suppose that one of the following conditions holds: (i) q < n < 2q; (ii) The prime p is odd. In addition, either p ∤ (n − 1) or n < 2q. HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 7 Then there exists an integer i such that and integers [ni/p] and n − 1 are relatively prime. 1 ≤ i ≤ q − 1, (i, p) = 1 Proof. If q < n < 2q then and we may take i = 1. [n · 1/q] = [n/q] = 1 So, further we assume that p is odd and either n < q or n − 1 is not divisible by p. If q > n > q/2 then [2n/q] = 1 and we may take i = 2. If 0 < n < q/2 then there exists a positive integer µ such that Since q is a power of p and n is not divisible by p, q ≤ µn < (µ + 1)n < 2q. Clearly, q < µn < (µ + 1)n < 2q. 1 = [nµ/q] = [n(µ + 1)/q]. So, we take as i either µ or µ + 1, depending on which one is not divisible by p. Now let us assume that n − 1 is not divisible by q. Let us put k = [n/q], c = n − kq, d = c − 1. We have c = d + 1, n = qk + c, n − 1 = qk + d; 2 ≤ c ≤ q − 1, 1 ≤ d ≤ q − 2, (c, p) = 1, (d, p) = 1. Let i be an integer such that 1 ≤ i ≤ q − 1 and (i, p) = 1. Put j = [ci/q]. Clearly, j is a nonnegative integer such that qj < ic < (q + 1)j. (The first inequality holds, because neither c nor i are divisible by p.) In other words, 0 < ic − qj < q. In addition, [ni/q] = [(kq + c)i/q] = ik + [ci/q] = ik + j. Suppose that n− 1 and [ni/q] are not relatively prime. Then there exists a prime ℓ that divides both n − 1 and [ni/q]. This implies that q · k + d · 1 = 0 in Z/ℓZ and i · k + j · 1 = 0 in Z/ℓZ. So, we get the homogeneous system of two linear equations over the field Z/ℓZ that admits the nontrivial solution (k, 1) 6= (0, 0). By Cramer's rule, the determinant id − qj is zero in Z/ℓZ, i.e., the integer id − qj is divisible by ℓ. In particular, id − qj 6= ±1. So, we prove the Proposition if we find such i that id − qj is either 1 or −1. In order to do that, notice that id − qj = i(c − 1) − qj = (ic − qj) − i. This implies that Since 1 ≤ i ≤ q − 1, −i < id − qj < q − i. 1 − q < id − qj < q − 1. 8 JIANGWEI XUE AND YURI G. ZARHIN Now if we choose i in such a way that 1 ≤ i ≤ q − 1 and id is congruent to 1 modulo q (such choice is possible, because d is not divisible by p) then id − qj is congruent to 1 modulo q and therefore id − qj = 1. In addition, the latter equality implies that i is not divisible by p. (cid:3) Proposition 2.1 admits the following (partial) generalization. Proposition 2.2. Suppose that p is a prime, r a positive integer, q = pr and n is a positive integer that is not divisible by p. Suppose that n − 1 is not divisible by q. If p = 2, assume additionally that q = 2r > 2 and n 6≡ q − 1 mod 2q. Then there exists an integer i such that and integers [ni/p] and n − 1 are relatively prime. 1 ≤ i ≤ q − 1, (i, p) = 1 Remark 2.3. If q = p = 2 then every odd n is congruent to 1 modulo 2. If p = 2, q = 4 then (n, q) satisfy the conditions of Proposition 2.2 if and only if n − 7 is divisible by 8. Proof of Proposition 2.2. As in the proof of Proposition 2.1, let us put k = [n/q], c = n − kq, d = c − 1. We have (1) (c, p) = 1, 2 ≤ c ≤ q − 1, 1 ≤ d ≤ q − 2. We are given that q does not divide d. However, in the light of Proposition 2.1, we may and will assume that p divides d; in particular, q > p and d ≥ p ≥ 2 > 1. Let Then t = (d, q), d′ = d/t, q′ = q/t. and both t and q′ are powers of p. This implies that q′ > 1, t > 1, (d′, q′) = 1 (d′, p) = 1, t ≥ p ≥ 2, q′ ≥ p ≥ 2 > 1. Since q′ > 1, d′ ≥ 1 and (d′, q′) = 1, there exists a unique pair of integers (i, j) such that d′i − q′j = 1, 0 < i ≤ q′ − 1, j ≥ 0. Clearly, (i, p) = 1 and therefore (i + q′, p) = 1, because q′ is a power of p. Since t ≥ 2, we have i + q′ < 2q′ ≤ tq′ = q. We will treat the case p = 2, n ≡ −1 mod q and n 6≡ q − 1 mod 2q separately at the end. So we further assume that if p = 2, then n + 1 is not divisible by q. We will show that either i or i + q′ is the desired integer, i.e., either [ni/q] and n − 1 are relatively prime or [n(i + q′)/q] and n − 1 are relatively prime. It is convenient to consider both integers [ni/q] and [n(i + q′)/q] as [n(i + ǫq′)/q] with ǫ = 0, 1. For every nonnegative integer ǫ we have (2) d′(i + ǫq′) − q′(j + ǫd′) = 1. Multiplying both side of (2) by t, we get (3) d(i + ǫq′) − q(j + ǫd′) = t. HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 9 Since c = d + 1, it follows that c(i + ǫq′) q = (d + 1)(i + ǫq′) q = j + ǫd′ + t + i + ǫq′ q and therefore (4) (cid:20) c(i + ǫq′) q (cid:21) − (j + ǫd′) = (cid:20) t + i + ǫq′ q (cid:21) . The following Lemma will be proven at the end of this Section. Lemma 2.4. We keep the assumptions of Proposition 2.2. If p = 2, we assume additionally that q ∤ n + 1. Then [(t + i + q′)/q] = 0 and therefore [(t + i)/q] = 0. Let us assume that either p is odd or p = 2 and n + 1 is not divisible by q. Combining Lemma 2.4 with (4), we conclude that (5) (cid:20) c(i + ǫq′) q (cid:21) = (j + ǫd′) if ǫ = 0 or 1. It follows that if ǫ = 0 or 1 then [n(i + ǫq′)/q] = [(kq + c)(i + ǫq′)/q] = k(i + ǫq′)+ [c(i + ǫq′)/q] = k(i + ǫq′)+ (j + ǫd′). Now suppose that n − 1 and [n(i + ǫq′)]/q] are not relatively prime for some ǫ = 0 or 1. Then there exists a prime ℓ that divides both n − 1 and [n(i + ǫq′)/q]. This implies that q · k + d · 1 = 0 in Z/ℓZ and (i + ǫq′) · k + (j + ǫd′) · 1 = 0 in Z/ℓZ. So we get the homogeneous system of linear equations over the field Z/ℓZ with determinant d(i + ǫq′) − q(j + ǫd′) = t (by (3)), which admits a non-trivial solution (k, 1) 6= (0, 0). By Cramer's rule the determinant t is zero in Z/ℓZ, i.e., ℓ t. Since t is a power of p, we conclude that ℓ = p. Since (i+ǫq′)·k+(j+ǫd′)·1 = 0 in Z/ℓZ and ℓ = p, the integer (i+ǫq′)·k+(j+ǫd′) is divisible by p. Now suppose that n − 1 and [n(i + ǫq′)/q] are not relatively prime for both ǫ = 0 and 1. This implies that both integers i · k + j and (i + q′) · k + (j + d′) are divisible by p. Therefore their difference q′k + d′ is also divisible by p, which is not the case, because q′ is a power of p while (d′, p) = 1. The obtained contradiction proves Proposition when either p is odd or p = 2 and n + 1 is not divisible by q. At last, let us treat the remaining case when p = 2, q = 2r with r ≥ 2, the integer n + 1 is divisible by q but n 6≡ q − 1 mod 2q. Then q divides n + 1 and the ratio k := (n + 1)/q is an even integer. We have n = 2rk − 1. Let us put i = 2r−1 − 1. Since r ≥ 2, the integer i is odd. We have q (cid:21) = (cid:20) (2rk − 1)(2r−1 − 1) (cid:20) ni 2r (cid:21) = (2r−1 − 1)k − 1. It follows that In particular, [ni/q] is odd, since k is even. Notice that [ni/q] ≡ −1 mod k. (n − 1)/2 = [ni/q] + k. 10 JIANGWEI XUE AND YURI G. ZARHIN Combining all those assertions, we get (n − 1, [ni/q]) = ((n − 1)/2, [ni/q]) = ([ni/q] + k, [ni/q]) = (k, [ni/q]) = (k, −1) = 1. Proof of Lemma 2.4. Recall that q′ and t are powers of p. This implies that t ≥ p ≥ 3 if p is odd; if p = 2 then either t = 2 or t ≥ 4. First, let us assume that p is odd and therefore t ≥ p ≥ 3 and q′ ≥ p ≥ 3. It (cid:3) follows that Since i ≤ q′ − 1, q′(t − 2) ≥ 3(t − 2) = 3t − 6 ≥ t. q = tq′ ≥ t + 2q′ > t + i + q′. This implies that [(t + i + q′)/q] = 0. Second, assume that p = 2 and t ≥ 4. Then q′ ≥ 2 and therefore q′(t − 2) ≥ 2(t − 2) = 2t − 4 ≥ t. As above, i ≤ q′ − 1, q = tq′ = (t − 2)q′ + 2q′ ≥ t + 2q′ > t + i + q′ and therefore [(t + i + q′)/q] = 0. Third, assume that p = 2 and t = 2. Then t = 2, d = 2d′, q′ = q 2 = 2r−1. Recall that (i, p) = 1, i. e., i is odd. Since i ≤ q′ − 1, the sum t + i + q′ = 2 + i + 2r−1 is greater or equal than q = 2r only if i = q′ − 1 = 2r−1 − 1. By (2), d′ · i ≡ 1 mod q′. If i = q′ − 1, then d′ = q′ − 1, since 1 ≤ d′ < q′. This implies that c = d + 1 = 2d′ + 1 = 2q′ − 1 = q − 1, which contradicts the assumption that is not divisible by q. So, this case does not occur. n + 1 = (kq + c) + 1 = kq + (c + 1) (cid:3) 3. Linear reductive Lie algebras Throughout this Section, Q is a field of characteristic zero, C is an algebraically If W is a Q-vector space (resp. Q-algebra or Q-Lie closed field containing Q. algebra) then we write WC for W ⊗Q C provided with the natural structure of a C-vector space (resp. C-algebra or C-Lie algebra). Let W be a nonzero finite-dimensional Q-vector space. Let E ⊂ EndQ(W ) be a subfield that contains the scalars Q · IdW . Then E/Q is a finite algebraic extension and W carries the natural structure of E-vector space; in addition, dimQ(W ) = [E : Q] · dimE(W ). We write Σ for the set of all Q-linear field embedding σ : E ֒→ C. If σ ∈ Σ then we write Wσ for the C-vector space W ⊗E,σ C; clearly, dimC (Wσ) = dimE(W ); HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 11 there are natural surjective homomorphisms WC ։ Wσ. Their "direct sum" WC → ⊕σ∈ΣWσ is an isomorphism of C-vector spaces, so one may view every Wσ as a C-vector subspace (direct summand) of WC . Remark 3.1. Let S be a Q-vector subspace of W . Let ES ⊂ W be the E-vector subspace of W generated by S, i.e., ES is the set of all linear combinations of elements of S with coefficients in E. Clearly, ES = W if and only if S contains a basis of the E-vector space W . It is also clear that the image of the composition SC ⊂ WC ։ Wσ coincides with {ES}σ = ES ⊗E,σ C ⊂ W ⊗E,σ C = Wσ; in particular, the C-dimension of this image coincides with dimE(ES) and does not depend on the choice of σ. Remark 3.2. Let k ⊂ EndQ(W ) be a reductive Q-Lie subalgebra such that the natural representation of k is completely reducible and the centralizer Endk(W ) of k in EndQ(W ) coincides with E; in particular, the k-module W is simple and k ⊂ EndE(W ). Clearly, Ek is a reductive E-Lie subalgebra of EndE(W ) and the centralizer of Ek in EndE(W ) coincides with E. In other words, the E-vector space W is an absolutely simple Ek-module.This implies that the C-vector space Wσ is an absolutely simple {Ek}σ-module. However, applying Remark 3.1 to the E-vector space EndE(W ) (instead of W ) and its Q-vector subspace S = k, we conclude that the C-Lie subalgebra {Ek}σ ⊂ EndC (Wσ) is the image of kC → EndE(W ) ⊗E,σ C = EndC (Wσ). This implies that the kC -module Wσ is also absolutely simple. On the other hand, the reductiveness of k and the equality Endk(W ) = E imply that k splits into the direct sum k = kss ⊕ c of its center c and the semisimple Q-Lie algebra kss = [k, k]; in addition, c ⊂ E. This implies that Ek = Ekss ⊕ Ec where Ec is either E or {0}. It follows easily that in both cases the kss C -module Wσ remains absolutely simple and the image of kss C in EndC (Wσ) coincides with {Ekss}σ = Ekss ⊗E,σ C = [{Ek}σ, {Ek}σ]. The following statement is a variant of [6, Sect. 4, Prop. 5] Lemma 3.3. Let W be a finite-dimensional C-vector space of positive dimension, let sl(W ) be the Lie algebra of traceless linear operators in W . Let g ⊂ EndC (W ) be an irreducible semisimple linear C-Lie subalgebra. Assume that there exists a diagonalizable operator f ∈ g ⊂ EndC (W ) that enjoys the following property: f acts in W as a linear operator with exactly two eigenvalues, whose multiplicities are relatively prime. Then g = sl(W ). 12 JIANGWEI XUE AND YURI G. ZARHIN Proof of Lemma 3.3. Clearly, f 6= 0 (otherwise, it would have only one eigenvalue). Let α1 and α2 be the eigenvalues of f and let W1 and W2 are the corresponding eigenspaces. Since f 6= 0, either α1 6= 0 or α2 6= 0. We claim that g is simple. Indeed, let us split the semisimple g into a direct sum g = ⊕r i=1gi of simple C-Lie algebras gi. Then the simple g-module W splits into a tensor product W = ⊗r i=1Ui of simple gi-modules Ui. Since W is a faithful g-module, all dimC(Ui) > 1. We i=1 fi with fi ∈ gi. Clearly, if some fi = 0 then the multiplicities of all eigenvalues of f are divisible by dimC (Ui). However, the spectrum of f consists of (only) two eigenvalues, whose multiplicities are relatively prime. This implies that all fi 6= 0, i.e., f does not belong to a proper ideal of g. Now the simplicity of g follows from [11, Th. 1.5 on p. 286] (with k = C and V = W ). have f = Pr The semisimplicity of f means that Let us put W = W1 ⊕ W2. n = dimC(W1), m = dimC (W2). By assumption, n and m are positive integers that are relatively prime. Let us put h := n + m = dimC (W ). Clearly, h and m are relatively prime. The semisimplicity of g implies that g ⊂ sl(W ); in particular, the trace of f is zero. This means that nα1 + mα2 = 0. It follows that both α1 and α2 do not vanish. Replacing f by (m/α1)f , we may and will assume that the spectrum of f consists of eigenvalue m of multiplicity n and eigenvalue −n of multiplicity m. In addition, W1 is the eigenspace attached to eigenvalue m and W2 is the eigenspace attached to eigenvalue −n. Let Θ ⊂ GL(W ) be the linear algebraic subgroup that consists of all linear operators A(u, v) that act on W1 as multiplication by u and on W2 as multiplication by v where (u, v) ∈ C ∗ × C ∗ satisfy unvm = 1. Let us consider the multiplicative algebraic group Gm over C. The morphism of algebraic C-groups ρ : Gm → Θ, z 7→ A(zm, z−n) is an isomorphism: in order to construct the inverse map, pick integers a and b with ma − nb = 1 (here we use the assumption that n and m are relatively prime). Then the morphism of algebraic group Θ → Gm, A(u, v) 7→ uavb is the inverse of ρ. So, Θ is isomorphic to Gm; in particular, it is a one-dimensional connected linear algebraic group and its Lie algebra is a one-dimensional C-Lie subalgebra of EndC (W ). Considering the tangent map to the composition Gm → Θ ⊂ GL(W ), one may easily find that the Lie algebra Lie(Θ) of Θ coincides with C·f ⊂ EndC (W ). On the other hand, let µh ⊂ C ∗ be the order h cyclic group of hth root of unity. If HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 13 z ∈ µh then zm = z−n (since n + m = h) and therefore A(zm, z−n) coincides with multiplication by zm. This implies that µh · Id ⊂ Θ ⊂ GL(W ) where Id is the identity map in W . Since every (linear) simple Lie algebra is algebraic, there exists a connected simple linear algebraic subgroup G ⊂ GL(W ), whose Lie algebra coincides with g. Since f ∈ g, we have C · f ⊂ g. In other words, the Lie algebra of connected Θ lies in the Lie algebra of G. This implies that Θ ⊂ G and therefore µh · Id ⊂ Θ ⊂ G ⊂ GL(W ); in particular, the order of the center of simple G is divisible by h = dimC(W ). It follows from Lemma 2 of Sect. 4 in [6] that G = SL(W ). Since the Lie algebra of SL(W ) coincides with sl(W ), we conclude that g = sl(W ). (cid:3) Theorem 3.4. Let V be a Q-vector space of positive finite dimension and let k ⊂ EndQ(V ) a reductive Lie algebra, whose natural representation in V is completely reducible. Suppose that the centralizer is a field. Let us put Let E := Endk(V ) h = dimE(V ). f ∈ kC ⊂ EndQ(V ) ⊗Q C = EndC (VC ) be a non-zero semisimple element that enjoys the following properties. (i) The spectrum of the C-linear operator f : VC → VC consists of two eigen- values, λ and µ; (ii) For every σ ∈ Σ we write nσ and mσ for the multiplicities of of eigenvalues λ and µ of the C-linear operator f : Vσ → Vσ . Then: (1) all the numbers {nσ}σ∈Σ are distinct positive integers. Assume also that all mσ are positive integers. (2) There exists τ ∈ Σ such that nτ and mτ are relatively prime. Then the semisimple part kss C = [kC , kC ] of kC contains an ideal that is isomorphic to a direct sum of r copies of the C-Lie algebra sl(h, C) of traceless matrices of size h with r ≥ [E : Q]/2. In particular, if kss = [k, k] is the semisimple part of k then dimQ(kss) = dimC(kss C ) ≥ [E : Q](h2 − 1). 1 2 In order to prove Theorem 3.4, we need the following two statements. Lemma 3.5. Let g1 and g2 be non-zero finite-dimensional simple Lie algebras over a field of characteristic zero. Let g ⊂ g1 ⊕ g2 be a semisimple Lie subalgebra such that the both projection maps g → g1, g → g2 are surjective. Then either g = g1 ⊕ g2 or there exists a Lie algebra isomorphism φ : g1 ∼= g2 such that g coincides with the graph Γ(φ) of φ in g1 × g2 = g1 ⊕ g2. In particular, if g1 and g2 are not isomorphic then g = g1 ⊕ g2. 14 JIANGWEI XUE AND YURI G. ZARHIN Lemma 3.6. Let n ≥ 2 and d be positive integers. Let a1, . . . , ad be d distinct positive integers such that 1 ≤ ai < n ∀i; ai 6= n − aj ∀i, j. Let C be an algebraically closed field of characteristic zero and W1, . . . , Wd be n- dimensional C-vector spaces. Let us put and let W = ⊕d i=1Wi i=1EndC (Wi) ⊂ EndC (W ) be a semisimple C-Lie algebra that enjoys the following properties: i=1sl(Wi) ⊂ ⊕d kss ⊂ ⊕d (i) The projection map kss → sl(Wi) is surjective for all i. (ii) There exists a semisimple element f ∈ kss ⊂ ⊕d i=1EndC (Wi) such that for all i the element f acts in Wi as a linear operator with two eigenvalues of multiplicity ai and n − ai respectively. Then kss = ⊕d i=1sl(Wi). Proof of Lemma 3.5. Let g0 be the kernel of the first projection map g ։ g1. By definition, g0 ⊂ {0} ⊕ g2 ⊂ g1 ⊕ g2. The surjectivity of the second projection g → g2 implies that g0 is an ideal in {0} ⊕ g2 ∼= g2. The simplicity of g2 implies that either g0 = {0} or g0 = {0} ⊕ g2. In the latter case g = g1 ⊕ g2. So, let us assume that g0 = {0}, i.e., the first projection map is an isomorphism. This means that there is a Lie algebra homomorphism φ : g1 → g2 such that g coincides with the graph Γ(φ) of φ. Now the surjectiveness of the second projection map means that φ is surjective. Since g1 is simple, φ is injective and therefore is an isomorphism. (cid:3) Proof of Lemma 3.6. Let us denote by fi : Wi → Wi the linear operator in Wi induced by f (for all i). If d = 1 then the result follows from the property (i). Assume now that d = 2. If kss 6= sl(W1) ⊕ sl(W2) then it follows from Lemma 3.5 that kss coincides with the graph Γ(φ) of a certain Lie algebra isomorphism φ : sl(W1) ∼= sl(W2). It is well known that such a Lie algebra isomorphism is induced either by an isomorphism of vector spaces either between W1 or W2 or between W ∗ 1 = HomC (W1, C) and W2. In the former case the spectra of f1 and f2 coincide (including the multiplicities). In the latter case the spectra of −f1 and f2 coincide (including the multiplicities). This implies that either a1 = a2 or a1 = n − a2. This contradicts our assumptions and proves the case of d = 2. In the case of arbitrary d ≥ 2, let us apply Lemma 3.5 to the image of kss in sl(Wi) ⊕ sl(Wj): we obtain that for every pair of distinct indices i, j ≤ d the projection map kss → sl(Wi) ⊕ sl(Wj ) is surjective. Now the case of arbitrary d follows from Lemma on pp. 790 -- 791 of [4]. (cid:3) Proof of Theorem 3.4. Let us split the reductive Q-Lie algebra kC Lie into a direct sum of its center c ⊂ E and the semisimple Q-Lie algebra kss = [k, k] = [k, k] kC = kss ⊕ c HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 15 Then f = f0 + fc with f0 ∈ kss C , fc ∈ cC . By Remark 3.2, the kC -module Vσ is absolutely simple for all σ ∈ Σ. By Schur's Lemma there exists cσ ∈ C such that fc acts in Vσ as multiplication by cσ. It follows that f0 acts in Vσ as a diagonalizable operator with eigenvalues λ − cσ of positive multiplicity nσ and µ − cσ of positive multiplicity mσ. Let us denote by fσ the linear operator in Vσ induced by f0. It follows from Remark 3.2 that fσ ∈ {Ekss}σ ⊂ EndC (Vσ) and {Ekss}σ is an irreducible linear semisimple Lie C-(sub)algebra. It is also clear that fσ is a diagonalizable operator with (exactly two) eigenvalues λ − cσ of mul- tiplicity nσ and µ − cσ of multiplicity mσ. Taking σ = τ and applying Lemma 3.3 to W = Wτ , f = fτ and g = {Ekss}σ, we conclude that {Ekss}τ = sl(Wτ ). In other words, the image of kss C → EndC (Vτ ) coincides with sl(Vτ ); in particular, the C-dimension of the image is h2 − 1. By Remark 3.1, for all σ ∈ C the image of kss C → EndC (Vσ) also has C-dimension h2−1. The semisimplicity of kss C implies the semisimplity of the image and therefore this image must lie in sl(Vσ); in particular, its C-dimension does not exceed h2 − 1 (recall that h = dimC (Vσ) ). It follows that the image of kss C → EndC (Vσ) coincides with sl(Vσ) for all σ ∈ Σ. Now let us choose a maximal subset Π ⊂ Σ with respect to the following property: nσ 6= mκ ∀σ, κ ∈ Π. We claim that #(Π) ≥ #(Σ)/2. Indeed, if #(Π) < #(Σ)/2 then the cardinality of the set does not exceed NΠ = {nσ σ ∈ Π} ∪ {mσ σ ∈ Π} 2 · #(Π) < 2 · #(Σ) 2 = #(Σ). This implies that #(NΠ) < #(Σ) and therefore there exists κ ∈ Σ such that nκ does not belong to NΠ; in particular, κ does not belong to Π. It follows from the very definition of NΠ that mκ also does not belong to NΠ. This implies that Π is not maximal, because we could replace it by Π ∪ {κ}. The obtained contradiction proves the desired inequality. Since #(Σ) = [E : Q], we have #(Π) ≥ [E : Q]/2. Now let us put and denote by kΠ the image of kss C in W := ⊕σ∈ΠVσ ⊕σ∈ΠEndC (Vσ) ⊂ EndC (⊕σ∈ΠVσ) = EndC (W ). Clearly, the linear operator fΠ : W → W induced by f is a semisimple element of kΠ. Now the result follows from Lemma 3.6 applied to W , the semisimple C-Lie algebra kΠ ⊂ EndC (W ) and fΠ. (cid:3) The following statement will be used in Section 4. Lemma 3.7. Let r ≥ 2 be a positive integer and let g1, . . . , gr be mutually noniso- morphic finite-dimensional simple Q-Lie algebras. Let g ⊂ ⊕r i=1gi be a semisim- ple Q-Lie subalgebra such that every projection map g → gi is surjective. Then g = ⊕r i=1gi 16 JIANGWEI XUE AND YURI G. ZARHIN Proof. Let i, j ≤ r be two distinct positive integers. Applying Lemma 3.5 to the image of g in gi⊕gj (with respect to the corresponding projection map), we conclude that the projection map g → gi ⊕ gj is surjective. Now the result follows from Lemma on pp. 790 -- 791 of [4]. 4. Semisimple components of Hodge groups and their Lie algebras Proof of Theorem 1.1. Let us apply Theorem 3.4 to Q = Q, C = C, V = H1(Z, Q), h = d(Z, E), k = hdg, f = f0 H , λ = − 1 2 , µ = 1 2 , nσ = nσ(Z, E), mσ = d(Z, E) − nσ(Z, E) = n¯σ(Z, E). We conclude that dimQ(hdgss) ≥ 1 2 [E : Q]{d(Z, E)2 − 1}. By Remark 1.2, this implies that hdgss = su(H1(Z, Q), φ) and we are done. (cid:3) (cid:3) 4.1. Let us assume that p is odd and q = pr and consider the abelian variety Z = Qr i=1H1(J (f,pi), Q). i=1 J (f,pi) and its first rational homology group H1(Z, Q) = ⊕r Then every subspace H1(J (f,pi), Q) is Hdg(Z)-invariant and the image of Hdg(Z) in GLQ(H1(J (f,pi), Q)) coincides with Hdg(J (f,pi)). (This assertion follows easily from the minimality property of Hodge groups.) It follows that every H1(J (f,pi), Q) is hdgZ-invariant and the image of hdgZ in EndQ(H1(J (f,pi), Q)) coincides with the Lie algebra of Hdg(J (f,pi)). This implies that the image of the semisimple Q-Lie Z in EndQ(H1(J (f,pi), Q)) coincides with the semisimple part of the algebra hdgss Lie algebra of Hdg(J (f,pi)). Theorem 4.2. Suppose that p is an odd prime, n ≥ 4 is a positive integer such that p does not divide n(n − 1). Suppose that f (x) ∈ C[x] is a degree n polynomial without multiple roots. Suppose that there exists a subfield K of C that contains all the coefficients of f (x). Let us assume that f (x) is irreducible over K and the Galois group Gal(f ) of f (x) over K is either Sn or An. Assume additionally that either n ≥ 5 or n = 4 and Gal(f ) = S4. Let r be a positive integer and q = pr. Let us put Z = Qr i=1 J (f,pi). If n > q then hdgZ ss = ⊕r i=1su(H1(J (f,pi), Q), φpi ). Proof. It follows from Theorem 0.1 that the semisimple part of the Lie algebra of Hdg(J (f,pi)) coincides with su(H1(J (f,pi), Q), φpi ) for all i ≤ r; notice that all su(H1(J (f,pi), Q), φpi )'s are mutually nonisomorphic simple Q-Lie algebras. Now the result follows from arguments of Subsect. 4.1 combined with Lemma 3.7. (cid:3) HODGE GROUPS OF CERTAIN SUPERELLIPTIC JACOBIANS 17 We keep the notation and assumptions of Theorem 4.2. For every positive integer i let us put Q(ζpi )− := {e ∈ Q(ζpi ) ¯e = −e} ⊂ Q(ζpi ) = End0(J (f,pi)) ⊂ EndQ(H1(J (f,pi), Q)). Let us put Tri = TrQ(ζpi+1 )/Q(ζpi ) : Q(ζpi+1) → Q(ζpi ), − := {(ei)r Ep,r i=1Q(ζpi )− ⊂ ⊕r ⊕r i=1 ∈ ⊕r i=1Q(ζpi )− Tri(ei+1) = ei ∀i < r} ⊂ i=1EndQ(H1(J (f,pi), Q)) ⊂ EndQ(H1(Z, Q)). Recall [9] that (under our assumptions) the center of the Q-Lie algebra hdgZ coin- cides with Ep,r − ⊂ EndQ(H1(Z, Q)). Then the reductiveness of hdgZ combined with Theorem 4.2 implies the following statement. Theorem 4.3. Suppose that p is an odd prime, n ≥ 4 is a positive integer such that p does not divide n(n − 1). Suppose that f (x) ∈ C[x] is a degree n polynomial without multiple roots. Suppose that there exists a subfield K of C that contains all the coefficients of f (x). Let us assume that f (x) is irreducible over K and the Galois group Gal(f ) of f (x) over K is either Sn or An. Assume additionally that either n ≥ 5 or n = 4 and Gal(f ) = S4. Let r be a positive integer and q = pr. Let us put Z = Qr i=1 J (f,pi). If n > q then − ⊕ [⊕r hdgZ = Ep,r i=1su(H1(J (f,pi), Q), φpi )]. Remark 4.4. We keep the assumptions of Theorem 4.3. Let us fix an isogeny α : J(Cf,pr ) → Qr i=1 J (f,pi) = Z. Then α induces an isomorphism of Q-vector spaces α : H1(J(Cf,pr ), Q) ∼= H1(Z, Q). Clearly, the Hodge group of Hdg(J(Cf,pr )) coincides with α−1Hdg(Z)α. This implies that the Q-Lie algebra of Hdg(J(Cf,pr ))) coincides with α{Ep,r − ⊕ [⊕r i=1su(H1(J (f,pi), Q), φpi )]}α−1. References [1] P. Deligne, Hodge cycles on abelian varieties (notes by J.S. Milne). Lecture Notes in Math., vol. 900 (Springer-Verlag, 1982), pp. 9 -- 100. [2] B. Moonen, Yu. G. Zarhin, Weil classes on abelian varieties. J. reine angew. Math. 496 (1998), 83-92. [3] Mumford, D.: Abelian varieties, Second edition. Oxford University Press, London, 1974. [4] K. Ribet, Galois action on division points of Abelian varieties with real multiplications. Amer. J. Math. 98, 751 -- 804 (1976). [5] K. Ribet, Hodge classes on certain abelian varieties. Amer. J. Math. 105 (1983), 523 -- 538. [6] Serre, J.-P. Sur les groupes de Galois attach´es aux groupes p-divisibles. Proc. Conf. Local Fields (Driebergen, 1966), pp. 118 -- 131, Springer, Berlin, 1967. [7] Serre, J.-P.: Repr´esentations lin´eares des groupes finis, Troisi´eme ´edition. Hermann, Paris, 1978. [8] G. Shimura, Abelian varieties with complex multiplication and modular functions. Princeton University Press, Princeton, 1997. [9] J. Xue, Yu.G. Zarhin, Centers of Hodge groups of superelliptic jacobians. Transformation Groups, to appear; arXiv:0907.1563 [math.AG]. [10] Yu.G. Zarhin, Weights of simple Lie algebras in the cohomology of algebraic varieties. Izv. Akad. Nauk SSSR Ser. Mat. 48 (1984), 264 -- 304; Math. USSR Izv. 24 (1985), 245 - 281. 18 JIANGWEI XUE AND YURI G. ZARHIN [11] Yu.G. Zarhin, Linear irreducible Lie algebras and Hodge structures. Algebraic geometry (Chicago, IL, 1989), 281 -- 297, Lecture Notes in Math. 1479, Springer, Berlin, 1991. [12] Yu.G. Zarhin, Hyperelliptic jacobians without complex multiplication. Math. Res. Letters 7 (2000), 123 -- 132. [13] Yu.G. Zarhin, Very simple 2-adic representations and hyperelliptic jacobians. Moscow Math. J. 2 (2002), issue 2, 403-431. [14] Yu.G. Zarhin, Cyclic covers, their Jacobians and endomorphisms. J. reine angew. Math. 544 (2002), 91 -- 110 . [15] Yu.G. Zarhin, The endomorphism rings of Jacobians of cyclic covers of the projective line. Math. Proc. Cambridge Philos. Soc. 136 (2004), 257 -- 267. [16] Yu.G. Zarhin, Endomorphism algebras of superelliptic Jacobians. In: F. Bogomolov, Yu. Tschinkel (eds.) Geometric methods in Algebra and Number Theory, Progress in Math. 235, 339 -- 362, Birkhauser, Boston Basel Berlin, 2005. [17] Yu. G. Zarhin, Superelliptic jacobians. In: "Diophantine Geometry" Proceedings (U. Zannier, ed.), Edizioni Della Normali, Pisa 2007, pp. 363-390. [18] Yu.G. Zarhin, Endomorphisms of superelliptic jacobians. Math. Z., 261 (2009), 691 -- 707, 709. Department of Mathematics, Pennsylvania State University, University Park, PA 16802, USA E-mail address: xue [email protected] Department of Mathematics, Pennsylvania State University, University Park, PA 16802, USA E-mail address: [email protected]
1304.5558	1	1304	2013-04-19T22:39:04	A probabilistic symbolic algorithm to find the minimum of a polynomial function on a basic closed semialgebraic set	[ "math.AG", "math.OC" ]	We consider the problem of computing the minimum of a polynomial function g on a basic closed semialgebraic set E in R^n. We present a probabilistic symbolic algorithm to find a finite set of sample points of the subset E^{min} of E where the minimum of g is attained, provided that E^{min} is non-empty and has at least one compact connected component.	math.AG	math	A probabilistic symbolic algorithm to find the minimum of a polynomial function on a basic closed semialgebraic set Gabriela Jeronimo♦,♮,∗, Daniel Perrucci♦,♭,∗, ♦ Departamento de Matem´atica, FCEN, Universidad de Buenos Aires, Argentina ♮ IMAS, CONICET -- UBA, Argentina ♭ CONICET, Argentina June 27, 2018 Abstract We consider the problem of computing the minimum of a polynomial function g on a basic closed semialgebraic set E ⊂ Rn. We present a probabilistic symbolic algorithm to find a finite set of sample points of the subset Emin of E where the minimum of g is attained, provided that Emin is non-empty and has at least one compact connected component. 1 Introduction The minimization of polynomial functions over Rn, unrestricted or subject to polynomial con- straints, is a classical problem with a variety of applications. In the last years, it has been exten- sively studied in the algorithmic framework through numerical or symbolic-numerical methods based on certificates of positivity (see, for instance, [20], [22], [23], [21], [27], [11], [9]). From the symbolic computation perspective, a possible way to tackle the problem is to restate it as a quantifier elimination problem over the reals and to apply a symbolic algorithm to solve this more general task. However, even the most efficient algorithms for quantifier elimination (see [1, Chapter 14]) lead to high complexities since they are not particulary designed for this specific problem. In [25], a probabilistic algorithm for global optimization of polynomial functions over Rn with good complexity estimates is presented. The better complexity is due to an alternative strategy relying on the computation of generalized critical values. Also, the problem of deciding algorithmically whether the global infimum of a polynomial function is attained is considered in [10]. Recently, in [18] (see also [6], [15]), another approach based on computer algebra techniques led to a lower bound for the minimum of a polynomial function on a basic closed semialgebraic set, provided that the set where this minimum is attained is compact. Here, we adapt the techniques underlying these theoretical result to the algorithmic framework. More precisely, we consider the following problem. Let K ⊂ R be an effective field. Let f1, . . . , fm, g ∈ K[x1, . . . , xn], with n ≥ 2, and E = {x ∈ Rn f1(x) = · · · = fl(x) = 0, fl+1(x) ≥ 0, . . . , fm(x) ≥ 0}, ∗Partially supported by the following grants: PIP 099/11 CONICET and UBACYT 20020090100069 (2010/2012). 1 and suppose that g attains a minimum value gmin at E. We look for a symbolic algorithm to compute at least one point in Emin = {x ∈ E g(x) = gmin}. In this paper we assume that Emin has at least one compact connected component. This is the case, for instance, in many families of instances of known optimization problems (see [18, Section 4]). Our approach consists in finding a finite set of points containing at least one point in each compact connected component of Emin. The natural tool to use when solving this problem is the Lagrange Multiplier's Theorem; nevertheless, a direct application of this result may lead to a degenerate system or a system with infinitely many solutions. In order to overcome this difficulty, we apply deformation techniques as in [17] (see also [1, Chapter 13]), which enable us to deal with "nice" systems that, in the limit, define finite sets containing the required minimizing points. These sets are described by geometric resolutions, which are parametric representations where the parameter ranges over the set of roots of a univariate polynomial. Finally, we compare the values that the given polynomial function g takes at the computed points and obtain the Thom encodings characterizing the minimizers. The main result of the paper is the following: Theorem 1 Let E = {x ∈ Rn f1(x) = · · · = fl(x) = 0, fl+1(x) ≥ 0, . . . , fm(x) ≥ 0} be defined by polynomials f1, . . . , fm ∈ K[x1, . . . , xn] with n ≥ 2 and degrees bounded by an even integer d. Let g ∈ K[x1, . . . , xn] be a polynomial of degree at most d that attains a minimum value gmin at E in a non-empty set Emin with at least one compact connected component. Algorithm FindingMinimum (see Section 5) is a probabilistic procedure that, taking as input the integer d and the polynomials f1, . . . , fm, g encoded by a straight-line program of length L, computes a family (cid:8)(cid:0)(pi, vi,1, . . . , vi,n), τi(cid:1)(cid:9)i∈I where I is a finite set and for every i ∈ I, (pi, vi,1, . . . , vi,n) is a geometric resolution in K[u] and τi ∈ {−1, 0, 1}deg pi is the Thom encoding of a real root ξi of pi such that the set {(vi,1(ξi), . . . , vi,n(ξi))}i∈I is included in Emin and intersects all its compact connected components. The complexity of the algorithm is where • D = • Υ = max O(cid:16)(n3(L + dn + nΩ−1)D2 log2(D) log log2(D) + (m + D)D2 log3(D))Υ(cid:17), 0≤s≤min{n,m}(cid:0)n P0≤s≤min{n,m} s(cid:1)ds(d − 1)n−s, 0≤s1≤l, 0≤s2≤m−l(cid:0) l Ps1+s2=s P0≤s≤min{n,m}(cid:0)m s(cid:1)2s. s1(cid:1)(cid:0)m−l s2 (cid:1)2s1 ≤ In the above statement, Ω denotes a positive real number such that for any ring R, addition, multiplication and the computation of determinant and adjoint of matrices in Rk×k can be performed within O(kΩ) operations in R. We may assume Ω ≤ 4 (see [3]) and, in order to simplify complexity estimations, we also assume that Ω ≥ 3. 2 An extended abstract of this work containing Theorem 1 has been accepted for presentation at the conference Effective Methods in Algebraic Geometry (MEGA) 2013 ([16]). The paper is organized as follows. In Section 2, we introduce the basic notation and state some previous results we use throughout the paper. In Section 3, we present the deformation we apply and we prove some of its geometric properties. Section 4 is devoted to showing how the deformation leads to a geometric resolution of the finite set we look for. Finally, in Section 5 we show the algorithmic counterparts of the previous theoretical results, proving Theorem 1. 2 Notation and preliminaries Throughout the paper, we denote N the set of positive integers. For a field K, we write K for an algebraic closure of K, K(t) for the field of rational functions in a single variable t and K[[t]] for the set of formal power series in t. For a given n ∈ N, the n-dimensional affine and projective spaces over an algebraically closed K respectively. In the case when K = C, we write simply An K and Pn field K are denoted by An and Pn. 2.1 Algorithms and complexity The algorithms we consider in this paper are described over an effective field K ⊂ R. The notion of complexity of an algorithm we consider is the number of operations and comparisons over K that the execution of the algorithm requires. In this definition of complexity, accessing, reading and writing pre-computed objects is cost free. The objects we deal with are polynomials with coefficients in K. In our algorithms we represent each polynomial either as the array of all its coefficients in a pre-fixed order of its monomials (dense form) or by a straight-line program. Roughly speaking, a straight-line program (or slp, for short) over K encoding a list of polynomials in K[x1, . . . , xn] is a program without branches (an arithmetic circuit) which enables us to evaluate these polynomials at any given point in Kn. The number of instructions in the program is called the length of the slp (for a precise definition we refer to [4, Definition 4.2]; see also [13]). From the dense encoding of a family of m polynomials in K[x1, . . . , xn] of degrees bounded by d, we can obtain an slp of length O(cid:0)m(cid:0)d+n n (cid:1)(cid:1) encoding them. Unless otherwise stated, throughout the paper, univariate polynomials will be encoded in dense form. To estimate complexities we will use the following results. Operations between univariate polynomials with coefficients in a field K of degree bounded by d in dense form can be done using O(d log(d) log log(d)) operations in K (see [7, Chapters 8 and 9]) and gcd or resultant computations by means of the Extended Euclidean Algorithm (see [7, Chapter 11]) can be performed within O(d log2(d) log log(d)) operations in K. Given an slp of length L encoding a univariate polynomial of degree at most d, we can obtain its dense form within O(dL + d log2(d) log log(d)) operations in K (see [7, Corollary 10.12]). From an slp of length L encoding a polynomial f ∈ K[x1, . . . , xn], we can compute an slp of length O(L) encoding f and all its first order partial derivatives (see [2]). 3 2.2 Geometric resolutions A way of representing zero-dimensional affine varieties which is widely used in computer algebra nowadays is a geometric resolution (see, for instance, [8]). The precise definition we are going to use is the following: Let K be a field of characteristic 0 and V = {z1, . . . , zD} ⊂ An K be a zero-dimensional variety defined by polynomials in K[x1, . . . , xn]. Given a separating linear form ℓ = α1x1 + · · · + αnxn ∈ K[x1, . . . , xn] for V (that is, a linear form ℓ such that ℓ(zi) 6= ℓ(zj) if i 6= j), the following polynomials completely characterize the variety V : • the minimal polynomial p :=Q1≤i≤D(u − ℓ(zi)) ∈ K[u] of ℓ over the variety V (where u is a new variable), • polynomials v1, . . . , vn ∈ K[u] with deg(vj) < D for every 1 ≤ j ≤ n satisfying V =(cid:8)(cid:0)v1(ξ), . . . , vn(ξ)(cid:1) ∈ K n ξ ∈ K, p(ξ) = 0(cid:9). The family of univariate polynomials (p, v1, . . . , vn) is called a geometric resolution of V (asso- ciated with the linear form ℓ). Note that if K is a subfield of R, the real roots of p correspond to the real points of the variety V . Given geometric resolutions (p1, v11, . . . , v1n) and (p2, v21, . . . , v2n) of two zero-dimensional varieties V1 and V2 in An consisting of D1 and D2 points respectively, associated with the same linear form ℓ which separates the points in V1 ∪ V2, we can obtain a geometric resolution of V1 ∪ V2 within complexity O(nD log2(D) log log(D)), where D = max{D1, D2}, by means of the Chinese Remainder Theorem using the Extended Euclidean Algorithm. 2.3 Thom encoding of real algebraic numbers The Thom encoding of real algebraic numbers provides an algebraic approach to distinguish the different real roots of a real univariate polynomial. We recall here its definition and main properties (see [1, Chapter 2]). Given p ∈ K[u] and a real root ξ of p, the Thom encoding of ξ as a root of p is the sequence (sign(p′(ξ)), . . . , sign(p(deg p)(ξ))), where we represent the sign with an element of the set {−1, 0, 1}. If the sign of the leading coefficient of p is known, the Thom encoding can be shortened to (sign(p′(ξ)), . . . , sign(p(deg p−1)(ξ))). Two different real roots of p have different Thom encodings. In addition, given the Thom encodings (σ1,1, . . . , σ1,deg p) and (σ2,1, . . . , σ2,deg p) of two different real roots ξ1 and ξ2 of p, it is possible to decide which is the smallest between ξ1 and ξ2 as follows: Consider the largest value of k such that σ1,k 6= σ2,k; then k < deg p, since p(deg p) is a constant. Also σ1,k+1 = σ2,k+1 6= 0, since otherwise ξ1 and ξ2 would have the same Thom encoding with respect to the polynomial p(k+1) and therefore ξ1 = ξ2. Then, • if σ1,k+1 = σ2,k+1 = 1, we have that ξ1 < ξ2 if and only if σ1,k < σ2,k, • if σ1,k+1 = σ2,k+1 = −1, we have that ξ1 < ξ2 if and only if σ1,k > σ2,k. 4 3 The deformation 3.1 Defining the deformation Here we introduce the deformation we use. We denote: • q0 := n + 1 and q1 < · · · < qm the first m prime numbers greater than n + 1. Let A ∈ Q(m+1)×(n+1), A = (aij)0≤i≤m, 0≤j≤n be the Cauchy matrix defined by aij = 1 qi−j (note that each submatrix of A has maximal rank and aij > 0 for every i, j). • For e ∈ N, Te the Tchebychev polynomial of degree e (see [19, Section 6.1]). • g(x) =P1≤j≤n a0jTd(xj) and, for every 1 ≤ i ≤ m, fi(x) = ai0 +P1≤j≤n aij(cid:0)Td(xj) + 1(cid:1). • G(t, x) = tg(x) + (1 − t)g(x) and, for every 1 ≤ i ≤ m, F + i (t, x) = tfi(x) + (1 − t) fi(x) and i (t, x) = tfi(x) − (1 − t) fi(x). F − • For every S ⊂ {1, . . . , m} and σ ∈ {+, −}S , VS,σ = { (t, x, λ) ∈ A × An × PS F σi i (t, x) = 0 for every i ∈ S, λ0∇xG(t, x) =Xi∈S λi∇xF σi i (t, x) }. We consider the decomposition of VS,σ as VS,σ = V (t) S,σ ∪ VS,σ, where -- V (t) S,σ is the union of the irreducible components of VS,σ included in t = t0 for some t0 ∈ C, -- VS,σ is the union of the remaining irreducible components of VS,σ. • For a group of variables y, Πy the projection to the coordinates y. 3.2 Geometric properties Let C be a compact connected component of Emin. For δ > 0, we write • C=δ = {x ∈ Rn dist(x, C) = δ}, • C≤δ = {x ∈ Rn dist(x, C) ≤ δ}, • C<δ = {x ∈ Rn dist(x, C) < δ}. Let µ > 0 such that C≤µ and Emin \ C do not intersect. We consider E = {(t, x) ∈ R × Rn F + 1 (t, x) ≥ 0, . . . , F + F − 1 (t, x) ≤ 0, . . . , F − l (t, x) ≥ 0, F + l (t, x) ≤ 0} l+1(t, x) ≥ 0, . . . , F + m (t, x) ≥ 0, and, for every 0 ≤ t ≤ 1, Note that E0 = Rn, E1 = E and for 0 ≤ t1 ≤ t2 ≤ 1, Et2 ⊂ Et1. Et = Πx(cid:0) E ∩(cid:0){t} × Rn(cid:1)(cid:1) ⊂ Rn. 5 Lemma 2 There exists 0 < ε < 1 such that for every 1 − ε ≤ t ≤ 1, the minimum value that G(t, ·) takes on Et ∩ C≤µ is not attained at any point in Et ∩ C=µ. Proof. The minimum value of g(·) = G(1, ·) at E1 ∩ C≤µ is gmin, which over this set is only attained at points in C and, therefore, not in E1 ∩ C=µ. Assume the claim does not hold. Then there exists a strictly increasing sequence of positive numbers (tk)k∈N converging to 1 such that the minimum value of G(tk, ·) at Etk ∩ C≤µ is attained at a point zk ∈ Etk ∩ C=µ. The sequence (zk)k∈N is contained in the compact set C=µ; therefore, without loss of gener- ality, we may assume that it converges to a point z ∈ C=µ. On the other hand, the sequence (tk, zk)k∈N is contained in E and converges to (1, z), then (1, z) ∈ E and z ∈ E1. We will prove that g(z) = gmin. Take z′ ∈ E1 ∩ C≤µ, then z′ ∈ Etk ∩ C≤µ for every k ∈ N and g(z′) = G(1, z′) = lim k→∞ G(tk, z′) ≥ lim k→∞ G(tk, zk) = G(1, z) = g(z). Then g attains its minimum on E1 ∩ C≤µ at z, which is gmin. This leads to a contradiction since this value is not attained at any point in E1 ∩ C=µ. (cid:3) The following proposition shows that in order to obtain minimizers for the polynomial func- tion g on the compact connected component C, it is enough to consider at most as many of the equations and inequations defining E as the number of variables. We define the set S = {(S, σ) S ⊂ {1, . . . , m} with 0 ≤ S ≤ n and σ ∈ {+, −}S with σi = + for l + 1 ≤ i ≤ m}. Proposition 3 Let C be a compact connected component of Emin. There exist z ∈ C and (S, σ) ∈ S such that z ∈ Πx(VS,σ ∩ {t = 1}). We point out that if E (resp. Emin) is compact, we can easily prove Proposition 3 adapting [18, Theorem 14]). Now we prove it the arguments in the proof of [18, Proposition 7] (resp. under our weaker assumptions. Proof of Proposition 3. Consider 0 < ε < 1 such that • for every 1 − ε ≤ t ≤ 1, the minimum value that G(t, ·) takes on Et ∩ C≤µ is not attained at any point in Et ∩ C=µ, • for every S ⊂ {1, . . . , m} and σ ∈ {+, −}S , Πt(V (t) S,σ) ∩ (1 − ε, 1) = ∅, • for every 1 − ε ≤ t ≤ 1, S ⊂ {1, . . . , m} with S > n and σ ∈ {+, −}S , the set {x ∈ An F σi i (t, x) = 0 for every i ∈ S} is empty. The existence of such an ε follows from Lemma 2, the finitness of Πt(V (t) σ, and an adaptation of the arguments in [17, Lemma 21] or [18, Lemma 4]. S,σ) for every S and Let (tk)k∈N be an increasing sequence converging to 1 with t1 > 1 − ε and let zk ∈ Etk ∩ C<µ be a point such that G(tk, ·) attains its minimum value on the set Etk ∩ C≤µ at zk. Without loss of generality, we may assume that the sequence (zk)k∈N is convergent to a point z ∈ E1 ∩ C≤µ, and proceeding as in the proof of Lemma 2, we have that z ∈ C. 6 Now, for every k ∈ N and every x ∈ Rn, at most one F + i (tk, x) and F − i (tk, x) may vanish. Let Sk = {i ∈ {1, . . . , l} F + i (tk, zk) = 0 or F − i (tk, zk) = 0} ∪ {i ∈ {l + 1, . . . , m} F + i (tk, zk) = 0}; then 0 ≤ Sk ≤ n. Without loss of generality, we may assume that Sk is the same set S for every k ∈ N; moreover, we may assume that, for each i ∈ S ∩ {1, . . . , l}, it is always the same, F + i (tk, zk) or F − i (tk, zk), the one which vanishes, thus defining a function σ ∈ {+, −}S with σi = + for l + 1 ≤ i ≤ m. Then, (S, σ) ∈ S. For k ∈ N, if the set {∇xF σi i (tk, zk), i ∈ S} is linearly independent, since the function G(tk, ·) attains a local minimum at the point zk when restricted to the set Etk ∩ C<µ, by the Lagrange Multiplier's Theorem, there exists (λi,k)i∈S such that ∇xG(tk, zk) =Xi∈S λi,k∇xF σi i (tk, zk). We take λ0,k = 1 and we have that (tk, zk, (λ0,k, (λi,k)i∈S)) ∈ VS,σ; but since tk 6∈ Πt(V (t) have that (tk, zk, (λ0,k, (λi,k)i∈S)) ∈ VS,σ. On the other hand, ifPi∈S λi,k∇xF σi (λi,k)i∈S 6= 0, we take λ0,k = 0 and, as in the previous case, (tk, zk, (λ0,k, (λi,k)i∈S)) ∈ VS,σ. Without loss of generality, we may assume that (λ0,k, (λi,k)i∈S )k∈N converges to a point (λ0, (λi,0)i∈S)) ∈ PS; then (1, z, (λ0, (λi,0)i∈S)) ∈ VS,σ and, therefore, z ∈ Πx(VS,σ ∩ {t = 1}) as we wanted to prove. (cid:3) S,σ), we i (tk, zk) = 0 with Proposition 4 For every (S, σ) ∈ S, we have that Πx(VS,σ ∩ {t = 1}) is a finite set. The proof of Proposition 4 will follow from arguments in the next section. From Propositions 3 and 4 we deduce the following: Corollary 5 Under the assumptions of Theorem 1, the set Πx(VS,σ ∩ {t = 1}) [(S,σ)∈S is finite and contains a point in every compact connected component of Emin. Due to Corollary 5, to solve the problem we are considering, in the next section we will focus on describing the set Πx(VS,σ ∩ {t = 1}) for every (S, σ) ∈ S. 4 A geometric resolution Throughout this section, we consider fixed (S, σ) ∈ S and denote V = VS,σ and V = VS,σ; moreover, for simplicity, we suppose S = {1, . . . , s} and σ = {+}S. For 1 ≤ j ≤ n, let gj(x, λ) = λ0 ∂g ∂xj − X1≤i≤s λi ∂fi ∂xj ∈ K[x1, . . . , xn, λ0, . . . , λs], 7 gj(x, λ) = λ0 ∂g ∂xj − X1≤i≤s λi ∂ fi ∂xj ∈ K[x1, . . . , xn, λ0, . . . , λs], Gj(t, x, λ) = tgj(x, λ) + (1 − t)gj(x, λ) ∈ K[t, x1, . . . , xn, λ0, . . . , λs]. These polynomials are homogeneous of degree 1 in the variables λ; therefore, by the multi- homogeneous B´ezout theorem (see, for instance, [26, Chapter 4, Section 2.1]), the degree of the varieties V ∩ {t = t0} for t0 ∈ C is bounded by Ds :=(cid:18)n s(cid:19)ds(d − 1)n−s. The next lemma shows the key properties of the initial system in the deformation (c.f. [17, Lemma 20]). Lemma 6 The polynomials f1, . . . , fs, g1, . . . , gn define a 0-dimensional variety in An × {λ0 6= 0} ⊂ An × Ps with Ds distinct points w1, . . . , wDs satisfying Πx(wi) 6= Πx(wj) for i 6= j, and the Jacobian determinant of f1, . . . , fs and the polynomials obtained from g1, . . . , gn dehomogenizing with λ0 = 1 does not vanish at any of these points. Proof. Recalling that g(x) =P1≤j≤n a0jTd(xj) and fi = ai0 +P1≤j≤n aij(Td(xj) + 1), we have that, for every 1 ≤ j ≤ n, gj(x, λ) = T ′ d(xj)(cid:16)a0jλ0 − X1≤i≤s aijλi(cid:17). Therefore, the solution set of the system f1, . . . , fs, g1, . . . , gn can be decomposed as {T ′ d(xj) = 0 ∀j ∈ B, f1(x) = 0, . . . , fs(x) = 0} × {a0j λ0 − X1≤i≤s [B⊂{1,...,n} By our assumption on the matrix A, if B = n − s, the linear system a0jλ0 −P1≤i≤s aijλi = 0 ∀j /∈ B has a unique solution ΛB ∈ Ps; moreover, this solution lies in {λ0 6= 0}. For a fixed B ⊂ {1, . . . , n} with B = n − s, taking into account that T ′ aijλi = 0 ∀j /∈ B}. d has d − 1 real roots and Td takes the value 1 or −1 at each of these roots, we have that SB := {T ′ d(xj) = 0 ∀j ∈ B, f1(x) = 0, . . . , fs(x) = 0} decomposes as the union of the sets SB,e := {T ′ d(xj) = 0, Td(xj) = e(j) ∀j ∈ B, f B,e 1 = 0, . . . , f B,e s = 0} for all e : B → {1, −1}, where f B,e replacing Td(xj) = e(j) for every j ∈ B. i ∈ K[xj; j /∈ B] denotes the polynomial obtained from fi by Without loss of generality, in order to simplify notation, assume B = {s + 1, . . . , n}. Then, for e : B → {1, −1}, the system f B,e 1 = 0, . . . , f B,e s = 0 can be written in the form AB  Td(x1) + 1 ... Td(xs) + 1   =  αB,e 1 ... αB,e s   8 where AB := (aij)1≤i,j≤s and αB,e AB is invertible, we can solve the underlying linear system for Td(x1) + 1, . . . , Td(xs) + 1. By applying Cramer's rule, it can be seen that the coordinates of the solution to this linear system are rational numbers where the denominators are a multiple of the prime number qs, whereas the numerators are relatively prime with qs; therefore, no coordinate of a solution is an integer number. We deduce that the above system is equivalent to a system of the form i = −ai0 −Ps+1≤j≤n aij(e(j) + 1) for 1 ≤ i ≤ s. Since Td(x1) = cB,e 1 , . . . , Td(xs) = cB,e s where cB,e none of which equals a root of T ′ i 6= ±1 for every 1 ≤ i ≤ s. It follows that each of the equations has d distict roots, Moreover, the Jacobian matrix of f1, . . . , fs and the polynomials obtained from g1, . . . , gn d (thus, the sets SB,e are mutually disjoint). dehomogenizing with λ0 = 1 evaluated at any of its solutions is of the form s s { { n − s {   C1 ∗ ∗ 0 0 C3 0 C2 ∗   . {z}s {z}n−s {z}s It is easy to see that C1, C2 and C3 are invertible matrices and so, the Jacobian determinant does not vanish. We conclude that SB consists of (d−1)n−sds distinct points in An for every B with B = n−s. Hence, the system f1, . . . , fs, g1, . . . , gn has Ds isolated solutions in An × Ps whose projections to An are all distinct. Since Ds is an upper bound for the degree of the variety the system defines, it follows that these are all its solutions. (cid:3) As a consequence of Lemma 6, it follows that all the irreducible components of V intersect the set {t = 0} and have dimension 1. Proposition 4 is immediate from this fact. Moreover, the following further properties of the induced deformation hold. Lemma 7 The variety defined in An by F1, . . . , Fs, G1, . . . , Gn is 0-dimensional and has Ds distinct points W1, . . . , WDs in {λ0 6= 0} such that Πx(Wi) 6= Πx(Wj) for i 6= j. More- over, these points can be considered as elements in K[[t]]n+s. × Ps K(t) K(t) Proof: The multihomogeneous B´ezout Theorem states that the degree of the variety is bounded by Ds. If w1, . . . , wDs are the common zeros of f1, . . . , fs, g1, . . . , gn, the Jacobian with respect to x1, . . . , xn, λ1, . . . , λs of F1, . . . , Fs, and the polynomials obtained from G1, . . . , Gn dehomog- enizing with λ0 = 1 at t = 0 and (x, λ) = wi is nonzero. The result follows applying the Newton-Hensel lifting (see for example [12, Lemma 3]). (cid:3) Consider now new variables y1, . . . , yn and define ℓ(x, λ, y) = ℓ(x, y) = P1≤j≤n yjxj. For α1, . . . , αn ∈ C, let ℓα(x, λ) = ℓα(x) =P1≤j≤n αjxj. Let P (t, u, y) = Y1≤i≤Ds(cid:0)u − ℓ(Wi, y)(cid:1) = P0≤h≤Ds ph(t, y)uh ∈ K(t)[u, y], P (t, u, y) q(t) q(t) = 9 with P (t, u, y) ∈ K[t, u, y] with no factors in K[t]. Let Q(u, y) = gcd( P (1, u, y), ∂ P /∂u(1, u, y)). Then, for generic α ∈ Cn, if p(u) := P (1, u, α) Q(u, α) and, for every 1 ≤ j ≤ n, vj(u) := − ∂ P ∂yj (1, u, α) Q(u, α) ∂ P ∂u (1, u, α) Q(u, α) !−1 mod p(u), we have that (cid:0)p(u), v1(u), . . . , vn(u)(cid:1) is the geometric resolution associated to ℓα of a finite set P containing Πx(V ∩ {t = 1}) (see, for instance, [8, Algorithm 9] or [17, Proposition 8]). The computation of P will be done by means of a Newton-Hensel based approximation. The P (t, u, y) ≤ nDs, which can be proved required precision is obtained from the degree bound degt as in [17, Lemma 9]. 5 The algorithm In this section we present our algorithms and prove the main result of the paper. From Corollary 5, we know that the finite set Πx(VS,σ ∩ {t = 1}) [(S,σ)∈S contains a point in every compact connected component of Emin; nevertheless, for a fixed (S, σ) ∈ S, the set Πx(VS,σ ∩ {t = 1}) is not necessarily contained in Emin, or may even have an empty intersection with E. The idea of our main algorithm is to compute first finite sets PS,σ containing Πx(VS,σ ∩ {t = 1}); then, look for for the points of each PS,σ that lie in E and finally, compare the values that the function g takes at these points. First, we introduce three auxiliary subroutines we use to construct our main procedure. Our first subroutine is an algorithm to compute the geometric resolution of the finite set PS,σ containing Πx(VS,σ ∩ {t = 1}) introduced in the previous section. This algorithm relies on the global Newton lifting from [8] and it is essentially the procedure underlying [17, Proposition 13]; we include it here for the sake of completeness. In order to simplify notation, we assume that S = {1, . . . , s} and σ = {+}S. Algorithm GeometricResolution INPUT: Polynomials f1, . . . , fs, g ∈ K[x1, . . . , xn] encoded by an slp of length L, an even integer d ≥ deg(fi), deg(g), and a linear form ℓα ∈ K[x1, . . . , xn]. OUTPUT: The geometric resolution (p, v1, . . . , vn) associated to ℓα of a finite set P containing Πx(V ∩ {t = 1}). 1. Compute the geometric resolution associated to ℓα(x) = α1x1 + · · · + αnxn of the variety defined in An+s by the (dehomogenized) system f1, . . . , fs, g1, . . . , gn as follows: 10 (a) For every B ⊂ {1, . . . , n} and e : B → {−1, 1}, compute the geometric resolution associated to ℓα(x) of the variety SB,e. (b) Compute the geometric resolution associated to ℓα(x) of the variety SB,e SB,e. 2. Compute the geometric resolution associated to ℓ(x, y) = y1x1 + · · · + ynxn of the variety defined by the (dehomogenized) system f1, . . . , fs, g1, . . . , gn over K(y), modulo the ideal (y1 − α1, . . . , yn − αn)2. 3. Compute P (t, u, y) mod ((t)2nDs+1 + (y1 − α1, . . . , yn − αn)2)K[[t]][u, y]. 4. Compute P (1, u, α) = P0≤h≤Ds ph(1, α)uh and ∂ P follows: ∂yj (1, u, α) = P0≤h≤Ds ∂ph ∂yj (1, α)uh as (a) Compute ph(t, α) and ∂ph ∂yj (t, α) (1 ≤ j ≤ n, 0 ≤ h ≤ Ds). (b) Evaluate t = 1. 5. Compute Q(u, α) = gcd( P (1, u, α), ∂ P ∂u (1, u, α)), p(u) = P (1,u,α) Q(u,α) and, for every 1 ≤ j ≤ n, vj(u) := − ∂ P ∂yj (1,u,α) Q(u,α) (cid:18) ∂ P ∂u (1,u,α) Q(u,α) (cid:19)−1 mod p(u). Proposition 8 Given a generic α ∈ Kn and polynomials f1, . . . , fm ∈ K[x1, . . . , xn] of degree bounded by an even integer d and encoded by an slp of length L, Algorithm GeometricResolution computes the geometric resolution associated to the linear form ℓα(x) =P1≤j≤n αjxj of a finite set PS,σ containing Πx(VS,σ∩{t = 1}) within complexity O(cid:0)n3(L+dn+nΩ−1)D2 s log2(Ds) log log2(Ds)(cid:1). Proof: Step 1(a). The variety SB,e is defined by a square polynomial system in separated vari- ables; then, the required computation can be achieved as in [14, Section 5.2.1] within complexity O(D2 Step 1(b). This step can be done within complexity O(nDs log3(Ds) log log(Ds)) following the procedure in Section 2.2 and the strategy described in [7, Algorithm 10.3]. B,e log2(DB,e) log log(DB,e)), where DB,e is the cardinality of SB,e. Step 2. This step can be done applying [8, Algorithm 1] within complexity O((dn3 + nΩ+1) Ds log(Ds) log log(Ds)). Step 3. Since F1, . . . , Fs, G1, . . . , Gn can be encoded by an slp of length O(L + (d + s)n), a geometric resolution of the variety they define associated with the linear form ℓ(x, y) mod- ulo the ideal (t)2nDs+1 + (y1 − α1, . . . , yn − αn)2 can be obtained from the previously com- puted geometric resolution by applying [8, Algorithm 1] within complexity O(n3(L + dn + nΩ−1)D2 s log2(Ds) log log2(Ds)). Step 4(a). By expanding P (t, u, y) = P0≤h≤Ds ph(t,y) q(t) uh ∈ K[[t]][u, y] into powers of u, (y1 − α1), . . . , (yn − αn), we have that the coefficients corresponding to uh and uh(yj − αj) (1 ≤ j ≤ n, 0 ≤ h ≤ Ds) are ph(t, α)/q(t) and ∂ph (t, α)/q(t) respectively. As the degrees ∂yj of the polynomials involved in these fractions are bounded by nDs, they are uniquely deter- mined by their power series expansions modulo (t)2nDs+1K[[t]] (see [7, Corollary 5.21]) that were computed at Step 3. By using [7, Corollary 5.24 and Algorithm 11.4] and converting 11 all rational fractions to a common denominator, the computation is done within complexity O(n2D2 s log2(Ds) log log(Ds)). Step 5. This step is achieved by means of the Extended Euclidean algorithm and polynomial divisions with remainder within complexity O(nDs log2(Ds) log log(Ds)). (cid:3) The second subroutine presented here tells us, for a finite set PS,σ given by a geometric resolution, if the set PS,σ ∩ E is empty, and, if not, it computes the Thom encoding of all the real roots of the minimal polynomial pS,σ corresponding to the points where the minimum value of g on PS,σ ∩ E is attained. Algorithm MinimumInGeometricResolution INPUT: A geometric resolution (pS,σ, vS,σ,1, . . . , vS,σ,n) in K[u] of a finite set PS,σ, polynomials f1, . . . , fm, g ∈ K[x1, . . . , xn] encoded by an slp of length L, an integer 0 ≤ l ≤ m and an integer d ≥ deg(fi), deg(g). OUTPUT: A boolean variable "Empty" with the truth value of the statement "The set PS,σ ∩ E is empty" and a list of elements τ1, . . . , τk ∈ {−1, 0, 1}deg pS,σ−1 with k = 0 if Empty = True, representing the Thom encodings of all the real roots of pS,σ corresponding to the points where the minimum value of g on PS,σ ∩ E is attained. 1. Compute the list of realizable sign conditions for f1(vS,σ(u)), . . . , fm(vS,σ(u)) over the real zeros of pS,σ(u). 2. Determine Empty going through the obtained list of realizable sign conditions. 3. If Empty = False: (a) Compute h(u) := Resu(pS,σ(u), u − g(vS,σ(u))). (b) Compute the list of realizable sign conditions for f1(vS,σ(u)), . . . , fm(vS,σ(u)), p′ S,σ(u), (u), h′(g(vS,σ(u))), . . . , h(deg pS,σ−1)(g(vS,σ(u))) over the real zeros of (deg pS,σ−1) . . . , p S,σ pS,σ(u). (c) Determine τ1, . . . , τk going through the obtained list of realizable sign conditions. Proposition 9 Given a geoemetric resolution (pS,σ, vS,σ,1, . . . , vS,σ,n) of the set PS,σ and the polynomials f1, . . . , fm, g ∈ K[x1, . . . , xn] encoded by an slp of length L, Algorithm MinimumInGeometricResolution decides whether PS,σ ∩ E is empty or not and computes the Thom encodings of the real roots of pS,σ corresponding to the points where the minimum value of g on PS,σ ∩ E is attained within complexity O(nD2 s + LDs + (m + Ds)D2 s log3(Ds))). Proof. Step 1. First we compute the dense encoding of the polynomials f1(vS,σ(u)), . . . , fm(vS,σ(u)), within complexity O((nDs +L)Ds +mDs log2(Ds) log log(Ds)). Then we apply the sign determi- s log3(Ds)). nation algorithm from [1, Section 10.3] and [5], following [24], within complexity O(mD2 Step 2. This step can be done within complexity O(mDs). Step 3(a). At this step the algorithm computes the monic polynomial h whose roots are the values of g at the points in PS,σ. First we compute the dense encoding of the polynomial 12 (deg pS,σ−1) S,σ s log2(Ds) log log(Ds)). g(vS,σ) and then, the resultant polynomial by multi-point evaluation and interpolation within complexity O((nDs + L)Ds + D2 Step 3(b). We continue the sign determination algorithm adding the polynomials p′ S,σ(u), (u), h′(g(vS,σ(u))), . . . , h(deg pS,σ−1)(g(vS,σ(u))) to what we have already com- . . . , p puted at Step 1. To do this, we first obtain the dense encoding of the polynomials involved. This step can be done within complexity O((m + Ds)D2 Step 3(c). The list of sign conditions computed at Step 3(b) enables us to know the Thom encoding of every real root ξ of pS,σ, and to relate ξ with the Thom encoding of g(vS,σ(ξ)) as a root of h. This information is enough to compare the different values of g(vS,σ(ξ)) (see Section 2.3) and, so, we can give the Thom encoding as roots of pS,σ of the roots giving the points where the minimum value of g is attained. This step is done within complexity O((m + Ds)Ds). (cid:3) s log3(Ds)). Since we know that the minimum value of g on E is also the minimum value of g on Πx(VS,σ ∩ {t = 1}) ∩ E, [(S,σ)∈S the set of sample minimizing points will be obtained by comparing the minimum values that g takes on PS,σ ∩ E for all (S, σ) ∈ S. This task is achieved by the following subroutine. Algorithm ComparingMinimums INPUT: Geometric resolutions (pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n) and (pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n) of finite sets PS1,σ1 and PS2,σ2 associated with a linear form ℓα, g ∈ K[x1, . . . , xn] encoded by an slp of length L, an integer d ≥ deg(g), and Thom encodings τ1 ∈ {−1, 0, 1}deg pS1,σ1 −1 and τ2 ∈ {−1, 0, 1}deg pS2,σ2 −1 of real roots of pS1,σ1 and pS2,σ2 corresponding to points where the minimum of g on PS1,σ1 ∩ E and PS2,σ2 ∩ E is attained. OUTPUT: An integer "Sign" from the set {−1, 0, 1} representing the sign of the minimum value that g takes on PS1,σ1 ∩ E minus the minimum value that g takes on PS2,σ2 ∩ E. 1. Compute a geometric resolution (p, v1, . . . , vn) of the union of the sets described by (pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n) and (pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n). 2. Compute h(u) := Resu(p(u), u − g(v(u))). 3. Compute the list of realizable sign conditions for pS1,σ1(u), p′ pS2,σ2(u), p′ ros of p(u). S2,σ2 (u), . . . , p (deg pS2,σ2 −1) S2,σ2 (u), (u), h′(g(v(u))), . . . , h(deg p)(g(v(u))) over the real ze- (u), . . . , p S1,σ1 (deg pS1,σ1 −1) S1,σ1 4. Determine Sign going through the obtained list of realizable sign conditions. (pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n) of PS1,σ1 and Proposition 10 Given geometric resolutions (pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n) of PS2,σ2, and the Thom encodings of a real root of pS1,σ1 and pS2,σ2 corresponding to points where the minimum of g on PS1,σ1 ∩ E and PS2,σ2 ∩ E is attained, Algo- rithm ComparingMinimums compares these minimums within complexity O(nD2 s1,s2 + LDs1,s2 + s1,s2 log3(Ds1,s2)) where Ds1,s2 := max{Ds1, Ds2 } with s1 = S1 and s2 = S2. D3 13 Proof. Step 1. This step is achieved within complexity O(nDs1,s2 log2(Ds1,s2) log log(Ds1,s2)) as explained in Section 2. Steps 2, 3 and 4. Similar to Algorithm MinimumInGeometricResolution Step 3(a), (b) and (c). The overall complexity of these steps is O(nD2 (cid:3) s1,s2 + LDs1,s2 + D3 s1,s2 log3(Ds1,s2)). We give now the main algorithm of the paper. Algorithm FindingMinimum INPUT: Polynomials f1, . . . , fm, g ∈ K[x1, . . . , xn] encoded by an slp of length L, an integer 0 ≤ l ≤ m and an even integer d ≥ deg(fi), deg(g). OUTPUT: A family (cid:8)(cid:0)(pi, vi,1, . . . , vi,n), τi(cid:1)(cid:9)i∈I where I is a finite set and for every i ∈ I, (pi, vi,1, . . . , vi,n) is a geometric resolution in K[u] and τi ∈ {−1, 0, 1}deg pi is the Thom encoding of a real root ξ of pi. 1. Take α = (α1, . . . , αn) ∈ Kn at random and set ℓα := α1x1 + · · · + αnxn. 2. S = {(S, σ) S ⊂ {1, . . . , m} with 0 ≤ S ≤ n and σ ∈ {+, −}S with σi = + for l + 1 ≤ i ≤ m}. 3. Take (S1, σ1) ∈ S and remove it from S. 4. (pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n) = GeometricResolution(fi(i ∈ S1), g, σ1, d, ℓα). 5. (Empty, τS1,σ1,1, . . . , τS1,σ1,k) = MinimumInGeometricResolution(pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n, f1, . . . , fm, g, l, d). 6. While Empty = True: (a) Discard (S1, σ1), take a new (S1, σ1) and remove it from S. (b) (pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n) = GeometricResolution(fi(i ∈ S1), g, σ1, d, ℓα). (c) (Empty, τS1,σ1,1, . . . , τS1,σ1,k) = MinimumInGeometricResolution(pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n, f1, . . . , fm, g, l, d). 7. I =(cid:8)(cid:0)(pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n), τS1,σ1,1(cid:1), . . . ,(cid:0)(pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n), τS1,σ1,k(cid:1)(cid:9). 8. While S 6= ∅: (a) Take (S2, σ2) ∈ S and remove it from S. (b) (pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n) = GeometricResolution(fi(i ∈ S2), g, σ2, d, ℓα). (c) (Empty, τS2,σ2,1, . . . , τS2,σ2,k) = MinimumInGeometricResolution(pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n, f1, . . . , fm, g, l, d). (d) If Empty = False : i. Sign = ComparingMinimums(pS1,σ1, vS1,σ1,1, . . . , vS1,σ1,n, pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n, g, d τS1,σ1,1, τS2,σ2,1). ii. If Sign = 0 then I = I∪(cid:8)(cid:0)(pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n), τS2,σ2,1(cid:1), . . . ,(cid:0)(pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n), τS2,σ2,k′(cid:1)(cid:9). 14 iii. If Sign = 1 A. I =(cid:8)(cid:0)(pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n), τS2,σ2,1(cid:1), . . . ,(cid:0)(pS2,σ2, vS2,σ2,1, . . . , vS2,σ2,n), τS2,σ2,k′(cid:1)(cid:9). B. (S1, σ1) = (S2, σ2). Now we prove Theorem 1. Proof of Theorem 1. The correctness of the algorithm follows from the results in Sections 3 and 4. The complexity upper bound is obtained by adding the complexity bounds for the subroutines GeometricResolution and MinimumInGeometricResolution applied to every element in S, and ComparingMinimums applied successively to pairs of elements from S. (cid:3) References [1] S. Basu, R. Pollack, and M.-F. Roy, Algorithms in real algebraic geometry. Second edition. Algorithms and Computation in Mathematics 10. Springer-Verlag, Berlin, 2006. [2] W. Baur and V. Strassen, The complexity of partial derivatives. Theoret. Comput. Sci. 22 (1983), no. 3, 317 -- 330. [3] S. Berkowitz, On computing the determinant in small parallel time using a small number of processors. Inform. Process. Lett. 18 (1984), no. 3, 147 -- 150. [4] P. Burgisser, M. Clausen, and M. A. Shokrollahi, Algebraic complexity theory, Grundlehren der Mathematischen Wissenschaften 315. Springer-Verlag, Berlin, 1997. [5] J. Canny, Improved algorithms for sign determination and existential quantifier elimination. Comput. J. 36 (1993), no. 5, 409 -- 418. [6] I. Emiris, B. Mourrain, and E. Tsigaridas, The DMM bound: Multivariate (aggregate) sepa- ration bounds, ISSAC 2010 -- Proceedings of the 2010 International Symposium on Symbolic and Algebraic Computation, 243-250, ACM, New York, 2010. [7] J. von zur Gathen and J. Gerhard, Modern computer algebra. Cambridge University Press, New York, 1999. [8] M. Giusti, G. Lecerf, and B. Salvy, A Grobner free alternative for polynomial system solving. J. Complexity 17 (2001), no. 1, 154 -- 211. [9] A. Greuet, F. Guo, M. Safey El Din, and L. Zhi, Global optimization of polynomials restricted to a smooth variety using sums of squares. J. Symbolic Comput. 47 (2012), no. 5, 503 -- 518. [10] A. Greuet and M. Safey El Din, Deciding reachability of the infimum of a multivariate polynomial. ISSAC 2011 -- Proceedings of the 36th International Symposium on Symbolic and Algebraic Computation, 131 -- 138, ACM, New York, 2011. [11] F. Guo, M. Safey El Din, and L. Zhi, Global optimization of polynomials using generalized critical values and sums of squares. ISSAC 2010 -- Proceedings of the 2010 International Symposium on Symbolic and Algebraic Computation, 107 -- 114, ACM, New York, 2010. 15 [12] J. Heintz, T. Krick, S. Puddu, J. Sabia, and A. Waissbein, Deformation techniques for efficient polynomial equation solving. J. Complexity 16 (2000), no. 1, 70 -- 109. [13] J. Heintz and C.-P. Schnorr, Testing polynomials which are easy to compute. In Logic and algorithmic (Zurich, 1980), Monograph. Enseign. Math. 30, 237 -- 254. Univ. Gen`eve, Geneva, 1982. [14] G. Jeronimo, G. Matera, P. Solern´o, and A. Waissbein, Deformation techniques for sparse systems. Found. Comput. Math. 9 (2009), no. 1, 1 -- 50. [15] G. Jeronimo and D. Perrucci, On the minimum of a positive polynomial over the standard smplex. J. Symbolic Comput. 45 (2010), no. 4, 434 -- 442. [16] G. Jeronimo and D. Perrucci, A probabilistic symbolic algorithm to find the minimum of a polynomial function on a basic closed semialgebraic set (extended abstract). Accepted for presentation at the conference Effective Methods in Algebraic Geometry (MEGA) 2013. [17] G. Jeronimo, D. Perrucci, and J. Sabia, On sign conditions over real multivariate polyno- mials. Discrete Comput. Geom. 44 (2010), no. 1, 195 -- 222. [18] G. Jeronimo, D. Perrucci, and E. Tsigaridas, On the minimum of a polynomial function on a basic closed semialgebraic set and applications. SIAM J. Optim. 23 (2013), no. 1, 241 -- 255. [19] D. Kincaid and W. Cheney, Numerical analysis. Mathematics of scientific computing. Brooks/Cole Publishing Co., Pacific Grove, CA, 1991. [20] J. B. Lasserre, Global optimization with polynomials and the problem of moments. SIAM J. Optim. 11 (2001), no. 3, 796 -- 817. [21] J. Nie, J. Demmel, and B. Sturmfels, Minimizing polynomials via sum of squares over the gradient ideal. Mathematical Programming 106 (2006), no. 3, 587 -- 606. [22] P.A. Parrilo. Semi-definite relaxations for semi-algebraic problems. Mathematical Program- ming 92 (2003), no.2, 293 -- 320. [23] P.A. Parrilo and B. Sturmfels, Minimizing polynomial functions. In Algorithmic and quan- titative real algebraic geometry, DIMACS Series in Discrete Mathematics and Theoretical Computer Science, Vol. 60, 83 -- 99, AMS. [24] D. Perrucci, Linear solving for sign determination. Theoret. Comput. Sci. 412 (2011), no. 35, 4715 -- 4720. [25] M. Safey El Din. Computing the global optimum of a multivariate polynomial over the reals. ISSAC 2008, 71-78, ACM, New York, 2008. [26] I. Shafarevich, Basic algebraic geometry. Springer-Verlag, Berlin, study edition, 1977. [27] M. Schweighofer, Global optimization of polynomials using gradient tentacles and sums of squares. SIAM J. Optim. 17 (2006), no. 3, 920 -- 942. 16
1011.0191	2	1011	2010-12-02T15:06:45	On combinatorial invariance of the cohomology of Milnor fiber of arrangements and Catalan equation over function field	[ "math.AG" ]	We discuss combinatorial invariance of the betti numbers of the Milnor fiber for arrangements of lines with points of multiplicity at most three and describe a link between this problem and enumeration of solutions of the Catalan equation over function field in the case when its coefficients are products of linear forms and the equation defines an elliptic curve.	math.AG	math	ON COMBINATORIAL INVARIANCE OF THE COHOMOLOGY OF MILNOR FIBER OF ARRANGEMENTS AND CATALAN EQUATION OVER FUNCTION FIELDS. A.LIBGOBER Abstract. We discuss combinatorial invariance of the betti numbers of the Milnor fiber for arrangements of lines with points of multiplicity two and three and describe a link between this problem and enumeration of solutions of the Catalan equation over function field in the case when its coefficients are products of linear forms and the equation defines an elliptic curve. 1. Introduction An interesting problem in the theory of arrangements of hyperplanes over the field of complex numbers is to determine which topological invariants of the comple- ment depend only on the combinatorics of the arrangements i.e. the poset formed by intersections of the hyperplanes of the arrangement. A theorem, going back to Arnold, Orlik and Solomon (cf. [17]), asserts that the cohomology of the comple- ment is a combinatorial invariant. On the other hand, it is known that the homotopy type (cf. [18]) and even the fundamental group (cf. [3]) cannot be determined from the combinatorics of the arrangement since there exist pairs of arrangements with the same combinatorics but different such invariants. Given an arrangement A of hyperplanes in Pn+1 in which the equations of hyper- planes are Li = 0 where Li are linear forms, one can consider the affine hypersurface ΠLi = 1 which is called the Milnor fiber of arrangement. It is known that for the Milnor fiber MA the homology group H1(MA, Z) depends only on the fundamental group of the complement to A and more precisely it is the abelianization of the subgroup of π1(P2 − A) which the kernel of the surjection lk : π1(P2 − A) → ZCardA given by the total linking number with A cf. ([14]). One can ask about the combi- natorial invariance of the cohomology of MA and in particular rkH 1(MA, Q) called (the first) Milnor number of the arrangement (for arrangements of lines, this is equivalent to the combinatorial invariance of remaining cohomology groups). To this end we show the following: Theorem 1.1. If two arrangements of lines A1 and A2 with the multiple points of multiplicity three and two only are combinatorially equivalent and if the monodromy on the Milnor fiber of A1 has an eigenvalue different from one then so does the monodromy on the Milnor fiber the arrangement A2. This theorem will be derived from the following more precise statement relating the cohomology of Milnor fiber and other data of arrangement. We say that an arrangement A is composed of a pencil L of plane curves (cf.[15],[8]) if L = {λF + µGdegF = degG} contains three curves λF +µG with equations which are products The author was partially supported by NSF grant. 1 2 A.LIBGOBER of linear forms such that A is the zero set of the forms defining these curves. A pencil is reduced if equations of the above reducible curved are reduced. (see section 2 for other definitions needed for this theorem). Theorem 1.2. Let A be an arrangement of lines with double and triple points only. If the monodromy acting on one-dimensional cohomology of the Milnor number of A has an eigenvalue different from one then the arrangement A is composed of a reduced pencil. The characteristic variety of the arrangement A has a component containing the identity Charπ1(P2 − A). Vice versa, if A is composed of a reduced pencil then the monodromy has eigenvalue exp( 2πi 3 ). The argument for the proof of above results relies on the relation between the Alexander polynomials of plane curves and Mordell-Weil groups considered recently in [4] and study of the Catalan equation over function field. Catalan equation, i.e. X p − Y q = 1 or more generally (1) A1X p + A2Y q + A3Z r = 0 has a long history and its properties of course depend on the ground field. Over function field, i.e. the case when Ai ∈ C(X) where X is a projective manifold the equation (1) was considered by J.Silverman. A finiteness results was shown in [20] in the case the equation (1) defines over the function field a curve of genus greater than 1. Elementary treatment of Fermat equation (i.e. Ai ∈ C, p = q = r) is given in [10]. Finding solution in the case of genus zero is trivial. Present paper considers the case p = q = r = 3 and Ai are product of linear forms in 3 variables. The answer to question about enumeration of solutions depends on the geometry of the curve A1 · A2 · A3 = 0 in P2 and we describe the cases when the number of solutions is infinite (cf. theorems 3.1 and 3.2). The paper organized as follows. In the next section we recall terminology from the theory of characteristic varieties of arrangements and related background infor- mation. In section 3 we prove results on existence and enumeration of solutions to Catalan equation. Proof of theorems presented in the last section. The paper ends with remark on properties of Mordell Weil groups which would imply combinatorial invariance of the cohomology of Milnor fiber. The author wants to thank S.Kaliman and L.Katzarkov for hospitality during visit to University of Miami where this work was written as well as S.Yuzvinsky for a useful comments. Finally, I grateful to A.Dimca for pointing out example 4.1 which helped to correct error in original statement of theorem 1.2 and to J.I.Cogolludo-Agustin for numerous examples of Catalan equations which helped to correct theorem 3.1. 2. Preliminaries As was already mentioned, Milnor fiber of an arrangement A of hyperplanes in Pn+1 is affine hypersurface MA in Cn+2 given by equation ΠLi = 1 where Li are linear forms defining the hyperplanes of the arrangement. The standard map Cn+2 − O → Pn+1 (Ø is the origin) restricted on MA is a cyclic cover (2) MA → Pn+1 − A of degree equal to the number of hyperplanes r = CardA in A. This cyclic cover corresponds to the homomorphism π1(Pn+1 − A) → Zr onto the cyclic group of order r = CardA given by the total modulo r linking number with the union of hyperplanes in A. We are interested in dimH1(MA, C) and also in the eigenspaces ON COMBINATORIAL INVARIANCE AND CATALAN EQUATION 3 of the generator of the group Zr of deck transformations acting on H1(MA, C). By Lefschetz theorem applied to a generic plane H in Pn+1 this group is isomorphic to the corresponding group of line arrangement formed by intersections of H with the hyperplanes of A. So from now on we shall assume that n = 1. Since the main results in this note deal with the case of arrangements of lines with points of multiplicity 2 and 3 from now on we shall restrict ourselves to the case of such multiplicities as well. By [12], [13] (cf. also [14]) the rank of the cyclic cover (2) can be found in terms the Alexander polynomial of P2 − A. More precisely, the Alexander polynomial of P2 − A is equal to (t − 1)CardA−2(t2 + t + 1)s (3) for s = dimH 1(P2, JS( 2CardA 3 − 3)) where JS ⊂ ØP2 is the ideal sheaf having stalk different from ØP iff P is a triple point of the arrangement and equal to maximal ideal at those P . Moreover if r is divisible by 3 then (4) rkH1(MA)ω = rkH1(MA)ω2 = s, rkH1(MA)1 = r − 2 where H1(MA)ζ is the eigenspace of the deck transformation of the covering (2) with eigenvalue ζ and ω = exp( 2π√−1 ). If r is not divisible by 3 then the dimensions of eigenspaces with eigenvalue ω, ω2 are equal to zero. Hence from now on we shall assume that r is divisible by 3. 3 Together with unbranched cover MA of degree r one can consider a non-singular model of 3-fold branched cyclic covering ¯MA of P2 which is a compactification of the 3 ⊂ Zr. The integer rkH1( ¯MA, C) quotient of MA by the action of the subgroup Z r is birational invariant and hence depends only on MA. Using results of [12], one has: (5) 1 2 rkH1( ¯MA) = rkH1( ¯MA)ω = rkH1( ¯MA)ω2 = s Recall the following definition from [4] adapting it to the case of arrangements (rather than arbitrary plane curves). Definition 2.1. A quasi-toric relation of orbifold type (3, 3, 3) is the identity F1f 3 + F2g3 + F3h3 = 0 where Fi, i = 1, 2, 3 is either a constant or a reduced curve such that Fi = 0 are without common components and F1F2F3 has the arrangement A as its zero set. (A'priori, of course for a given arrangement there are several possibilities for Fi's, including Fi = 1). Let E be elliptic curve over C with j-invariant zero. As a model for E we shall use the 3-fold ramified covering of C branched at 0, 1 (and infinity) which is a closure of the affine variety in C4 given by the equations: (6) (the cyclic covering map is given by (a, b, c, s) → s). As a model of ¯MA for calcu- lation of H 1( ¯MA) we shall use the complete intersection in C5 (with coordinates (x, y, A, B, C)) given by: c3 = ab, a = s, b = 1 − s (7) A = F1(x, y) F3(x, y) , B = F2(x, y) F3(x, y) , C3 = AB 4 A.LIBGOBER (here Fi are dehomogenized forms used in definition 2.1. The surface (7) is the threefold cyclic cover of C2 branched over the curve F1F2F3 = 0 corresponding to the homomorphism of the fundamental group sending the classes of loops corre- sponding to the curves Fi = 0 to the same element of Z3. There is a correspondence between the Z3-equivariant maps φ : ¯MA → E and quasi-toric relations given as follows. Note that such a map φ takes a fixed point of Z3 to one of three ramification points of E → P1. Denoting the polynomials whose zero sets are the remaining components of preimages of ramification points as f, g, h respectively, one can write the morphism φ as a compactification of the restriction on affine portion of ¯MA of Φ : C5 → C4 given by: (8) a = A( f h )3, b = B( g h )3, c = C f g h2 , s = A( f h )3 Let us define the equivalence between quasi-toric relations 2.1 involving Fi, f, g, h and F ′i , f′, g′, h′ saying that two such relations are equivalent if for some λ, λf , λg, λh ∈ C, one has Fi = λF ′i , λf ∈ µ3 and λf λg = λ2 h. With this we have the following: ∈ µ3, λg λh λh Proposition 2.2. There is one to one correspondence between quasi-toric relations up to above equivalence and non constant equivariant maps ¯MA → E. Proof. Indeed it follows from formulas (6), (7) that (f, g, h) appearing in (8) satisfy the quasi-toric relation. Vice versa, if f, g, h satisfy a quasi-toric relation, the formulas (8) yield a map which when restricted on affine portion of ¯MA provide a dominant map φ : ¯MA → E. (cid:3) As in [4] we shall view the maps ¯MA → E as C( ¯MA)-points of the elliptic curve E. As such they form a group with the quotient by the subgroup of constant maps (the Mordell-Weil group) being finitely generated (Mordell-Weil theorem cf. [11]). The following is a special case of result shown in [4]. Theorem 2.3. The rank of the Mordell-Weil group M W (A) of C( ¯MA)-points of E is equal to 2s where s is the degree of the factor (t2 + t + 1) is (3). Finally we shall need few results on characteristic varieties of arrangements. Recall (cf. [14]) that one can view them the components of affine subvariety V of the torus Charπ1(P2 − A) consisting of characters χ such that (9) dimH1(P2 − A, χ) > 0 These affine subvarieties are finite unions of translated subgroups of Charπ1(P2−A) by the points of finite order in Charπ1(P2 − A) (cf. [1],[16]). The following results is contained in [14] and [15] respectively (cf. also [5]): Theorem 2.4. 1) The dimension of the ω-eigenspace of the monodromy acting on H 1(MA) of the Milnor number MA of an arrangement A is equal to dimH 1(P2 − A, Lω) where Lω is the local system π1(P2 − A) → C∗ sending each loop, having the linking number equal to one with a hyperplane in A and zero with all the others, to the a root of unity ω. 2)The dimension of H 1(P2 − A, Lχ) of the local system is not smaller than d − 1 where d is the dimension of the component of characteristic variety to which it belongs. ON COMBINATORIAL INVARIANCE AND CATALAN EQUATION 5 More specifically, 2) follows from semi-continuity of ranks of cohomology of local systems (as in [15]) and the fact that generic local system in a component of positive dimension is a twist of the pull back of a local system on P1 minus a collection of points. The latter does not change generically the rank of H 1 (cf. [1], Prop. 1.7) and relation in 2) is obvious for the local systems on complements to a finite union of point in P1. 3. Catalan Equation A1f 3 + A2g3 + A3h3 = 0 with coefficients being products of linear forms. The purpose of this section is to prove the following: Theorem 3.1. Let Ai = Πj≤ri Li,j, i = 1, 2, 3, be products of linear forms Li,j ∈ C[x1, x2, x3] (ri are non negative integers). Assume that Li,j is not a multiple of Li′,j ′ for any pair two pairs (i, j), (i′, j′) and that lines Li,j = 0 form an arrangement of lines with points of multiplicity 2 and 3 only. 1) The equation A1f 3 + A2g3 + A3h3 = 0 (10) has solution only if A1A2A3 = A′1A′2A′3 where A′1, A′2, A′3 are linear dependent over C. In this case the solutions of (10) with linearly dependent Ai, are the pullbacks of solutions of Catalan equation over C(t). 2) In the case Ai are linearly dependent over C the set of solutions to (10) is infinite. Proof. First we show that if the equation (10) with A1 = A2 = 1 has a solution then there exist factorization A3 = A′1A′2A′3 such that at least two factors are non-constant and the Catalan equation (10) with these factors as coefficients has a solution. Assume that A1 = A2 = 1, select f, g, h which form a solution with smallest possible degree of h and such that f, g, h have no common factors. Then i = −1). Let h be an irreducible factor of h. There is −A3h3 = Π(f + ωig) (ω3 only one factor f + ωig which is divisible by h, since if there are two, then both f and g are divisible by h contradicting the assumption of absence of common factors for f, g, h. If each irreducible factor of h divides single factor f + ωig of f 3 + g3, then f + g = ah′3A′3, f + ω2g = b(h′′)3A′1, f + ω3g = c(h′′′)3A′2. So 3degf = (dega + degb + degc) + 3(degh′ + degh′′ + 3degh′′′) + degA3 i.e. a = b = c are constants. Hence A′1h′3, A′2h′′3, A′3h′′′3 belong to a pencil i.e. h′, h′′, h′′′ are solutions to the Catalan equation with coefficients A′1, A′2, A′3. If A′1, A′2 are constant then the degree of h′′′ is strictly smaller than the degree of h. Hence such descent should lead either to Catalan equation with solutions having at least two non- constant coefficients or the Catalan equation f 3 +g3+A3 = 0 would have a solution. In the latter case, by unique factorization, each f + ωig is a product of factors of A3 and the Catalan equation with coefficients A′′i = f + ωig has solution in C i.e. satisfies conclusion of 1) since A′′i are in the pencil generated by f, g. Assume now that factors A1, A2 in (10) are not constant. Let us consider the restriction of relation (10) on one of the lines Li,j, say L = L1,j which equation is a factor of A1. Denoting the form obtained by restriction on L of a form M as LM we can write the resulting relation as (11) LA2(Lg)3 +L A3(Lh)3 = 0 6 A.LIBGOBER ai,j Let LAi = Πl i,j . If one ai,j > 2 then the zero of li,j on L is the point with belong to more than three lines Li,j = 0 forming the zero set of A1A2A3 = 0 contradicting the assumption multiplicities not exceeding 3. Hence the exponent of a factor of LAi is ai,j = 1 or ai,j = 2. Due to unique factorization in C[t] we obtain that none of roots of LA2 and LA3 is a root of g or h. Moreover LA2 and LA3 have the same Lg Lh ∈ C. Hence any intersection sets of zeros on L with the same multiplicities and point of L and Ai also belongs to Ai′ for i = 2, 3 and i′ = 2, 3. This implies that the same is true for any point in A1 = 0, that degA2 = degA3 and hence the conditions of Noether AF + BG theorem (cf. [9]) are satisfied (the multiplicities of Ai are equal to one). Hence (12) A1 = λ2A2 + λ3A3 where degλi = 0 Next consider rational map P2 → P1 given by P → (A2(P ), A3(P )). Its fibers consist of points P such that (13) A2 A3 = t ∈ C The solution (10) for Ai is also solution to (λ2t+ λ3)f 3 + tg3 + h3 = 0. Since Fermat equation over the function field of the curve (13) has only constant solutions (cf. [10]) it follows that f, g, h depend only on t. The same argument shows that if ai,j = 2 then f, g, h are pullbacks as well. Part 2) is hence proven. (cid:3) Next we shall consider solutions to equation (10) over C(t). Theorem 3.2. The solutions to (10) in case Ai ∈ C(t) form a group isomorphic to Hom(J ac(C), E) where C is the 3-fold branched cover of P1 ramified at the set S which is the union of zeros of Ai (1 ≤ i ≤ 3), J ac(C) is its Jacobian and E is the elliptic curve with an automorphism of order 3. In particular if S contains only three elements then the group of solutions is isomorphic to End(E) = Z2. Proof. Non-constant solutions to (10) correspond to non-constant sections of the elliptic surface (7). This elliptic surface is iso-trivial and is trivialized over the 3-fold cyclic cover C of P1 containing t-line as an open set totally ramified at the union of zeros of Ai and possibly the infinity. Indeed the holonomy around each zero is trivial on such 3-fold cover. For trivial elliptic surface C × E over C non constant sections correspond to the maps C → E by Torelli theorem which correspond to the maps of the Jacobian of C to E. The last claim in the statement 2) and statement 3) follows since the Jacobian of threefold cover of P1 branched at three points is E. (cid:3) Example 3.3. 1) Let ω = exp( 2π√−1 non constant solution to Catalan equation: ) be a primitive root of unity of degree 3. A 3 (t − ω)2(t − ω2)2f 3 = (1 + ω2)(t − 1)2(t − ω2)2g3 + (−ω2)(t − ω)2(t − 1)2h3 is f = t − 1, g = t − ω, h = t − ω2 Indeed, after substitution we get (1 + ω2 = −ω): (t3−1)2[(t−1)−(1+ω2)(t−ω)+ω2(t−ω2)] = (t3−1)[(t−1)+ω(t−ω)+ω2(t−ω2)] = 0 2) Suppose that F3 = F1 + F2 i.e. the arrangement is composed of a pencil. This yields the solution f = g = h = 1 to Catalan equation. Elliptic curve c3 = s2 − s ON COMBINATORIAL INVARIANCE AND CATALAN EQUATION 7 which is isomorphic to a = s, b = 1 − s, c3 = ab is equivalent to y2 = x3 + 1 where s − 1 3 c. Duplication formulas from [21] p.59, are in this case: 1 2 = y, x = 4 x1 := x4 − 8x 4x3 + 4 , y1 = −y − 4x3+4 − x) 3x2( x4−8x 2y (here (x1, y1) denotes 2(x, y) in the sense of addition on elliptic curve). From this we infer that if we take F3 = 1 (which is always the case after replacement F1 → F1 F3 ) start with map a = F1, b = F2 then we get for doubling: , F2 → F2 F3 i.e. a := −F1(F1 − 2)3 (2F1 − 1)3 b = (F1 − 1)(F1 + 1)3 (2F1 − 1)3 f h = −F2 − 1 2F1 − 1 g h = F1 + 1 2F1 − 1 are solutions to Catalan equation (indeed, one easily sees that a + b = 1). 4. On Combinatorial invariance Proof. (Of theorems 1.1 and 1.2) Let A1 and A2 be two arrangements as in the statement of the theorem. Recall that the characteristic polynomial of the mon- odromy of Milnor fiber for each arrangement is (t − 1)r(Ai)−2(t2 + t + 1)s(Ai) where r(Ai) = Card(Ai) and s(Ai) is the superabundance of curves of degree 2d−9 pass- ing through the triple points of the arrangement Ai. Hence we need to show that s(A1) > 0 implies s(A2) > 0 as well. For the rest recall the results from [4]. Let C be a possibly reducible plane curve for which ω = exp( 2π√−1 ) is a root of the Alexander polynomial. Let Πi∈RFi = 0 where R is the set of irreducible components of C be the reduced equation of C. Then the equation of C admits a quasi-toric relation corresponding to ordinary triple point (cf. [4]) i.e. there exist an equality (cf. (2.1)): 3 3 (14) with distinct Fi and R = Si=1,2,3 Ri, Ri ∩ Rj = ∅, i 6= j is a partition of the set of components R. Πi∈R1 Fif 3 + Πi∈R2 Fig3 + Πi∈R3 Fih3 = 0 Since s(A1) > 0, there exist non trivial quasi-toric decomposition corresponding to the curve which is a union of lines in A1. This means that equations of lines of arrangement can be split into 3 groups which can be used to construct Catalan equation which has solutions. By theorem 3.1, A1 is composed of reduced pencil. On the other hand, to each such pencil corresponds the 2-dimensional component of characteristic variety of A1 containing the identity of Char(π1(P2 − A1)) (it consists of the pullbacks of local systems from the complement in P1 to a triple of points). Each component of characteristic variety of A1 containing the identity of Char(π1(P2 − A1)) yields a component of the resonance variety of A1 (cf. [14], [15]). Recall that the resonance variety is the collection of a ∈ H 1(P2 − A1) such that the Aomoto complex (15) H 0(P2 − A1) ∧a−→ H 1(P2 − A1) ∧a−→ H 2(P2 − A1) has non-zero cohomology at the middle term. Since such cohomology is determined by the cohomology algebra H∗(P2 − A1) we obtain that the resonance variety of A2 has the same number of components as A1. To each component of the resonance 8 A.LIBGOBER variety of A2 corresponds pencil which is reduced since the corresponding compo- nent of resonance variety of A1 comes from reduced pencil. For a map of P2 − A2 onto the complement in P1 to a triple of points P, the pull back of the local system on the latter assigning to a standard generator of π1(P1 − P) value ω3 = exp( 2πi 3 ), is the local system on P2 − A2 which cohomology is isomorphic to the eigenspace with eigenvalue ω3 on H 1(MA2 ). Hence the claim follows. (cid:3) Remark 4.1. The cases when s > 1 correspond to arrangements which are composed of several pencils. One such example is the following. Let C be an elliptic curve and A an arrangement of lines in plane dual to the plane containing C consisting of lines dual to nine inflection points of C. Then this arrangement has no double points, 12 triple points and the superabundance of curves of degree 3 containing these 12 points is s = 2 (cf. [6],[2]). Indeed, the cohomology sequence corresponding to (16) 0 → JS(3) → ØP2(3) → ØS → 0 is the following: (17) 0 → 0 → C10 → C12 → H 1(P2, JS(3)) → 0 This arrangement is composed of 4 pencils (cf. [14] section 3.3 example 3). For example one can use as explicit equations for this arrangement union of three cubics x3 − y3, x3 − z3, y3 − z3 or [(y−z)(x−z)(x−y)], [(y−ω2z)(z−ω2y)(x−ωz)], [(y−ωz)(x−ω2z)(x−ωy)] (ω3 = 1) etc. Remark 4.2. The combinatorial invariance of s(A) would follows from the follow- ing refinement of the above argument describing the relation between s(A) and the number l(A) of pencils of which A is composed. This is the number of essential components of the resonance variety of A. The number of Catalan equations cor- responding to an arrangement A which have solutions by 3.1 is also l(A). To each such pencil correspond collection the map ¯MA → E (cf. [4]) of 3-fold cyclic covers of P2 and P1 respectively defined up to automorphism of E i.e. i.e. six elements of the Mordell Weil group M W (A1) differ by an automorphism of E. Recall that M W (A) is endowed with the canonical height pairing (cf. Identification of the number of elliptic pencils on ¯MA with the number of elements of minimal height in M W (A1) and showing that the number of elements of minimal height in M W (A) = Z[ω3]s(A) is determined by s(A) would imply that the number of pencils which composed A is determined by s(A) and vice versa. Since, as follows from the above argument, the number of pencils composing the arrangement is a combinatorial invariant this would imply the combinatorial invariance of s(A) and hence of the Milnor number. [19]). References [1] D.Arapura, Geometry of cohomology of support loci for local systems I, Alg. Geom., vol. 6, p. 563, (1997). [2] N.Budur, A.Dimca, M.Saito, First Milnor cohomology of hyperplane arrangements, arXiv:0905.1284 ON COMBINATORIAL INVARIANCE AND CATALAN EQUATION 9 [3] E.Artal Bartolo, R.Carmona, J.I.Cogolludo Agustin, M.Marco Buzunariz, Invariants of com- binatorial line arrangements and Rybnikov's example. Singularity theory and its applications, Adv. Stud. Pure Math., 43, Math. Soc. Japan, Tokyo, 2006. [4] J.I.Cogolludo-Agustin, A.Libgober, Mordell-Weil groups of elliptic threefolds and the Alexan- der module of plane curves, arXiv:1008.2018. [5] A.Dimca, Pencils of plane curves and characteristic varieties, arXiv:math/0606442. [6] A.Dimca, G.Lehrer, Hodge-Deligne equivariant polynomials and monodromy of hyperplane arrangements arXiv:1006.3462. [7] M.Falk, Arrangements and cohomology. Ann. Comb. 1 (1997), no. 2, 135-157. [8] M.Falk,S.Yuzvinsky, Multinets, resonance varieties, and pencils of plane curves. Compos. Math. 143 (2007), no. 4, 1069-1088, [9] W.Fulton, Adjoints and Max Noether's Fundamentalsatz. Algebra, arithmetic and geometry with applications (West Lafayette, IN, 2000), 301-313, Springer, Berlin, 2004. [10] N.Greenleaf, On Fermat's equation in C(t). Amer. Math. Monthly 76 1969 808-809. [11] S.Lang, A.Neron, Rational points of abelian varieties over function fields. Amer. J. Math. 81 1959 95-118. [12] A.Libgober, Alexander polynomial of plane algebraic curves and cyclic multiple planes. Duke Math. J. 49 (1982), no. 4, 833-851. [13] A.Libgober, Alexander invariants of plane algebraic curves. Singularities, Part 2 (Arcata, Calif., 1981), 135-143, Proc. Sympos. Pure Math., 40, Amer. Math. Soc., Providence, RI, 1983. [14] A.Libgober Characteristic varieties of algebraic curves. Applications of algebraic geometry to coding theory, physics and computation (Eilat, 2001), 215-254, NATO Sci. Ser. II Math. Phys. Chem., 36, Kluwer Acad. Publ., Dordrecht, 2001. [15] A.Libgober, S.Yuzvinsky, Cohomology of the Orlik-Solomon algebras and local systems. Com- positio Math. 121 (2000), no. 3, 337-361. [16] A.Libgober, Non vanishing loci of Hodge numbers of local systems. Manuscripta Math. 128 (2009), no. 1, 1-31. [17] P.Orlik, L.Solomon, Combinatorics and topology of complements of hyperplanes. Invent. Math. 56 (1980), no. 2, 167-189. [18] G.Rybnikov, On the fundamental group of the complement of a complex hyperplane ar- rangement", Tech. Rep. 94-13, DIMACS, Rutgers Univ., Piscataway, NJ, 1994; preprint, arxiv.org/abs/math/9805056] [19] J.P.Serre, Lectures on the Mordell-Weil theorem, Aspects of Mathematics, Friedr. Vieweg and Sohn, 1989. [20] J.Silverman, The Catalan equation over function fields. Trans. Amer. Math. Soc. 273 (1982), no. 1, 201-205. [21] J.Silverman, The arithmetic of elliptic curves. Graduate Texts in Mathematics, 106. Springer- Verlag, New York, 1986. Department of Mathematics, University of Illinois, Chicago, IL 60607 and Depart- ment of Mathematics, University of Miami, Coral Gables, Fl. E-mail address: [email protected]
1601.07074	1	1601	2016-01-26T15:50:49	On stable rationality of Fano threefolds and del Pezzo fibrations	[ "math.AG" ]	We prove that very general non-rational Fano threefolds which are not birational to cubic threefolds are not stably rational.	math.AG	math	ON STABLE RATIONALITY OF FANO THREEFOLDS AND DEL PEZZO FIBRATIONS BRENDAN HASSETT AND YURI TSCHINKEL 1. Introduction Recent breakthroughs of Voisin [Voi15], developed and amplified by Colliot-Th´el`ene–Pirutka [CTP14, CTP15], Beauville [Bea14], and To- taro [Tot15], have reshaped the classical study of rationality questions for higher-dimensional varieties. Failure of stable rationality is now known for large classes of rationally-connected threefolds. The key tool is (Chow-theoretic) integral decompositions of the diagonal, which nec- essarily exist for stably rational varieties. Integral decompositions of the diagonal specialize well, even to mildly singular varieties, connect- ing logically the stable rationality of various classes of varieties. This puts a premium on discovering appropriate degenerations linking dif- ferent classes of rationally connected varieties. Using these techniques, we prove: Theorem 1. Let X be a very general smooth non-rational Fano three- fold over C. Assume that X is not birational to a cubic threefold. Then X is not stably rational. While smooth cubic threefolds are all known to be non-rational, de- termining whether or not they are stably rational remains an open problem. No smooth cubic threefolds are known to be stably rational. However, Voisin [Voi14] has shown that the cubic threefolds where her techniques fail to apply, i.e., those admitting an integral decomposition of the diagonal, are dense in moduli. Several common geometric threads, developed in collaboration with Kresch, unify our approach to Theorem 1. In [HKT15b], we showed that very general conic bundles over rational surfaces with sufficiently large discriminant fail to be stably rational. The conic bundle struc- tures on cubic threefolds arising from projection from a line have quin- tic plane curves as their discriminants—too small for our techniques Date: January 25, 2016. 1 2 BRENDAN HASSETT AND YURI TSCHINKEL to apply. Nevertheless, conic bundles are a useful tool for analyzing stable rationality of Fano threefolds. Second, in [HKT15a] we classi- fied quartic del Pezzo surfaces with mild singular fibers and maximal monodromy; previously [HT14] we showed that a number of these arise as specializations of Fano threefolds of index one. Together, these fa- cilitate a streamlined approach to most families of non-rational Fano threefolds. Acknowledgments: The first author was supported by NSF grant 1551514. We are grateful to Andrew Kresch for his foundational con- tributions that made this research possible. We also benefitted from conversations with Alena Pirutka. Kresch and Pirutka also offered helpful feedback on early drafts of this paper. 2. Organization of the cases The tables in [IP99] enumerate non-rational Fano threefolds; see also the summary in [Bea15, Section 2.3], which includes references to meth- ods used to establish non-rationality. In the results that follow, ‘very general’ refers to the complement to a countable union of Zariski-closed proper subsets of the families enumerated in this section. Let V be a Fano threefold of Picard rank one. Write Pic(V ) = Zh, for some ample h, and express the anti-canonical class −KV = rh. Let h1,2 = dim H 1(V, Ω2 V ), which equals the dimension of the intermedi- ate Jacobian IJ(V ). We enumerate non-rational Fano threefolds V of Picard rank one, using the invariants (r, −K 3 V , h1,2). • (1, 2, 52): double cover of P3 ramified in a surface of degree 6, unirationality is unknown, very general V are not stably ratio- nal [Bea14] • (1, 4, 30): quartic in P4, unirationality is unknown, very general V are not stably rational [CTP14] • (1, 6, 20): intersection of a quadric and a cubic, unirational • (1, 8, 14): intersection of three quadrics in P6, unirational • (1, 10, 10): section of Gr(2, 5) by a subspace of codimension 2 and a quadric, general such V are non-rational, all are unira- tional • (1, 14, 5): section of Gr(2, 5) by a subspace of codimension 5, unirational • (2, 8, 21): V1, unirationality is unknown • (2, 8 · 2, 10): V2, double cover of P3, ramified in a smooth quar- tic, unirational, very general V2 are not stably rational [Voi15, Bea15] ON STABLE RATIONALITY 3 • (2, 8 · 3, 5): V3, cubic in P4, unirational We next list non-rational minimal Fano threefolds of Picard rank V , h1,2). All of these admit conic bundle ≥ 2, using the invariants (−K 3 structures, induced by projection onto a rational surface. • (6, 20): double cover of P1 × P2 branched in a divisor of bidegree (2, 4), unirational • (12, 9): divisor in P2 × P2 of bidegree (2, 2), or a double cover of F(1, 2) ⊂ P2 × P2 of bidegree (1, 1), ramified in B ∈ − KF(1,2), unirational • (14, 9): double cover of V7 ≃ Blp(P3), branched in B ∈ − KV7, unirational • (12, 8): double cover of P1 × P1 × P1 branched in a divisor of degree (2, 2, 2), unirational 3. Conic bundles over rational surfaces We recall the set-up for the results of [HKT15b]: Let S be a smooth projective rational surface over C. Fix a linear system L of effective divisors on S such that the generic member is smooth and irreducible. Consider the space of pairs (cid:8) D ∈ L nodal and reduced, D′ → D ´etale of degree two (cid:9) → L and let M be one of its irreducible components. Assume it contains a point {D′ → D} with the following properties: • the nodes of D are disjoint from the base locus of L; • D is reducible and for each irreducible component D1 ⊂ D the induced cover D′ ×D D1 → D1 is non-trivial. Results of Artin and Mumford [AM72] and Sarkisov [Sar82] allow us to assign to each point of M a conic bundle X → S, unique up to birational equivalences over S. Essentially, M parametrizes rami- fication data for the associated Brauer elements in the function field of S, which determine them as S is rational. The condition on the distinguished point implies that the corresponding conic bundle has non-trivial Brauer group. Using Voisin’s decomposition of the diagonal technique, we proved in [HKT15b] that a very general point [X] ∈ M parametrizes a threefold that fails to be stably rational. We first observe an obvious strengthening of the main theorem of [HKT15b]: M need not dominate the linear series L but can be any smooth irreducible parameter space of reduced nodal curves D ∈ L 4 BRENDAN HASSETT AND YURI TSCHINKEL with ´etale double covers D′ → D. Let K denote the image of M in L, so we have M ϕ → K ⊂ L where ϕ is ´etale and a covering space over the open subset parametriz- ing smooth curves. We still insist that there is a reducible member whose nodes are disjoint from the base locus of K, such that the cover over each component is non-trivial. Second, our result is easiest to apply in cases where the monodromy action is large, e.g., when M parametrizes all non-trivial double covers of the generic point [D] ∈ K, or equivalently, when the monodromy representation on H 1(D, Z/2Z)\{0} is transitive. This is the case when S = P2 and L parametrizes plane curves of even degree; in odd degree there are two such orbits [Bea86]. Large monodromy actions make it easier to decide which component contains a given distinguished point {D′ → D}. 4. Classification of quartic del Pezzo fibrations and stable rationality Consider quartic del Pezzo surface fibrations π : X → P1 satisfying two non-degeneracy conditions: • the discriminant is square-free, i.e., X is regular and the degen- erate fibers are complete intersections of two quadrics with at most one ordinary singularity; • the monodromy action on the Picard groups of the fibers is the full Weyl group W (D5). The fundamental invariant of such fibrations is the height h(X ) = deg(c1(ωπ)3) = −2 deg(π∗ω−1 π ), an even integer (see [HKT15a, HT14] for more background). The prin- cipal results we require are [HKT15a, Th. 10.2]: • under the non-degeneracy conditions we have h(X ) ≥ 8; • when h(X ) = 8 or 10, the moduli space of these fibrations has two irreducible components; • when h(X ) ≥ 12 the moduli space is irreducible. When h(X ) = 8, 10, 12 the total space X is either rational or bira- tional to a cubic threefold; see [HKT15a, §11] and [HT14, §8-10] for details. Thus we will focus on fibrations with heights at least fourteen. Note that Alexeev [Ale87] established non-rationality in these cases by ON STABLE RATIONALITY 5 relating the del Pezzo fibrations to conic bundles. (We will review this below.) Theorem 2. Let X → P1 be a fibration in quartic del Pezzo surfaces satisfying our non-degeneracy conditions, with h(X ) ≥ 14, and very general in moduli. Then X fails to admit an integral decomposition of the diagonal and thus is not stably rational. Proof. We first reduce to the conic bundle case, following Alexeev. Choose a section σ : P1 → X , which we may assume is not contained in any line of the generic fiber. Blowing up this section gives a cubic surface fibration with a distinguished line and projecting from this line gives a conic fibration: L ֒→ eX π→ S ց ↓ ւ P1 Here S → P1 is a rational ruled surface. The conic bundle structure over S yields a discriminant curve D ⊂ S and an ´etale double cover D′ → D. Note that D′ → D coincides with the spectral data introduced in [HKT15a, §§2,8] and S is the natural ruled surface containing D described in [HKT15a, §10]. Using [HKT15a, §6] we pin down the numerical invariants: Suppose first that h(X ) = 4n + 2 for n ≥ 3. Here the surface S ≃ F1, the Hirzebruch surface. Let ξ denote the (−1)-curve and f the class of a fiber. Then [D] = 5ξ +(n+3)f which has genus h(X )−4. If h(X ) = 4n for n ≥ 4 then S ≃ F0 ≃ P1 × P1. Here D has bidegree (n, 5), also of genus h(X ) − 4. The fundamental dictionary between del Pezzo fibrations and spec- tral data [HKT15a, Th. 10.1] implies that the D ⊂ S arising from del Pezzo fibrations are generic in the linear series L = D. The analy- sis of [HKT15a, §3] shows that the monodromy acts on H 1(D, Z/2Z) via the full symplectic group, hence transitively on the non-trivial el- ements. To apply the main result of Section 3, it suffices to exhibit a distinguished point in D, i.e., a reducible curve D = D1 ∪ D2 with D1 and D2 smooth of positive genus, intersecting transversally. Note that any ´etale double cover on D1∪D2 deforms to the parameter space M. Indeed, such double covers correspond to homomorphisms H1(D1 ∪ D2, Z) → Z/2Z; 6 BRENDAN HASSETT AND YURI TSCHINKEL a specialization D D1 ∪ D2 induces a homomorphism H1(D, Z) → H1(D1 ∪ D2, Z) collapsing vanishing cycles, thus a double cover of D via composition. Producing the reducible curve is simple: For S = F1 take D1 ∈ 2ξ + 3f , the proper transform of a cubic plane curve, and D2 ∈ 3ξ + nf , a smooth curve of genus 2n − 5 ≥ 1. For S = F0 ≃ P1 × P1 take D1 of bidegree (2, 2), an elliptic curve, and D2 of bidegree (n − 2, 3), of genus 2n − 6 ≥ 2. (cid:3) 5. Quartic del Pezzo surfaces of height 22 A quartic del Pezzo surface of height 22 may be constructed as follows [HT14, §4, Case 5]: Let V = OP1 ⊕ OP1(1)⊕4 and consider the injection V ∗ ֒→ O⊕9 P1 associated with the global sections of V . Then we have morphisms P(V ∗) ֒→ P1 × P8 π2−→ P8, where the composition collapses the distinguished section σ : P1 → P(V ∗) arising from the OP1 summand. We use π = π1 for the fibration over P1. Let ξ = c1(OP(V ∗)(1)) and h = π∗(c1(OP1(1))) so that ξ5 = 4ξ4h. A generic height 22 quartic del Pezzo X → P1 admits an embedding X ֒→ P(V ∗) ց ↓ P1 as a complete intersection of divisors of degrees 2ξ − h and 2ξ. Let Q → P1 denote the former divisor, which is canonically determined. It necessarily contains the section σ. The second divisor Q′ is a pull-back of a quadric hypersurface via π2; it is typically disjoint from σ. Consider projection from the section σ: : P(V ∗) → P(OP1(−1)⊕4) ≃ P1 × P3 inducing a birational map ∼ 99K P1 × P3. Q Restricting to X yields a generically finite morphism φ : X → P3. ON STABLE RATIONALITY 7 We compute its invariants via intersections in P(V ∗). The pullback of the hyperplane class on P3 via φ is ξ − h. First, we have deg(φ) = (ξ − h)3(2ξ)(2ξ − h) = 2 which means φ is a double cover. Its ramification divisor R = KX − φ∗KP4 = −ξ + h + 4(ξ − h) = 3(ξ − h) maps to the branch surface B ⊂ P3 of degree six. We interpret when φ fails to be finite. Points p ∈ P3 correspond to line subbundles σ(P1) ֒→ L(p) ≃ P(OP1 ⊕ OP1(−1)) ≃ F1 ֒→ P(V ∗) where F1 is the blowup of the projective plane at a point. Thus Q∩L(p) is the union of the (−1)-curve and the proper transform of a line ℓ and Q′ ∩ L(p) is the proper transform of a conic disjoint from the (−1)- curve. These typically meet at two points but the conic might contain the line ℓ, i.e., φ−1(p) = ℓ; this is a codimension-three condition on p and corresponds to singular points of B. How many singularities do we expect on B? For the moment, as- sume these are ordinary double points for generic X . We have [HT14, Cor. 3.5]: h1(Ω2 X ) = 22 − 5 = 17 but for a generic sextic double solid V we have h1(Ω2 V) = 52 [IP99]. If V admits n ordinary singularities and rank r class group then we have 52 = n − r + 1 + h1(Ω2 eV), comparing the topological Euler characteristics and class groups of V where eV → V is the blowup of the singularities. This can be seen by and eV. We must have r = 2 if V is a contraction of a del Pezzo fibration X → P1. Thus we expect n = 36. Proposition 3. Let y0, y1, y2, y3 denote coordinates on P3. • The equation for B takes the form det(M) = 0 where with L linear in the yi and the remaining entries quadratic. • The generic such matrix arises from a height 22 fibration in quartic del Pezzo surfaces. M =  L2 Q0 Q1 Q0 Q′ 00 Q′ 01 01 Q′ Q0 Q′ 11   , 8 BRENDAN HASSETT AND YURI TSCHINKEL • The singularities of B are of two types. The first type corre- sponds to the vanishing of the 2 × 2 minors of M ; there are 32 such singularities. The second type corresponds to the locus L = Q0 = Q1 = 0; there are four such singularities. Proof. Let x0 and x1 denote homogeneous coordinates on P1 and their pullbacks to P(V ∗). Designate generating global sections y0, y1, y2, y3 ∈ Γ(OP(V ∗)(ξ − h)) ≃ Γ(OP3(1)) and z, x0y0, x1y0, . . . , x0y3, x1y3 ∈ Γ(OP(V ∗)(ξ)). After completing the square to eliminate the term linear in z, the defin- ing equation Q′ may be written in the form z2 = Q′ 00x2 0 + 2Q′ 01x0x1 + Q′ 11x2 1, where the Q′ takes the form ij are quadratic in the yi. The defining equation for Q −zL(y0, y1, y2, y3) + Q0x0 + Q1x1 = 0, where L is linear and Q0 and Q1 are quadratic in the yi. Eliminating z we obtain x2 0(Q′ 00L2 − Q2 0) + 2x0x1(Q′ 01L2 − Q0Q1) + x2 1(Q′ 11L2 − Q2 1) = 0, which is the defining equation for the image of X in P1 × P3. The discriminant of this polynomial–regarded as a binary quadratic form in x0 and x1–can be written as L4((Q′ 01)2 − Q′ 00Q′ 11) + L2(−2Q′ 01Q0Q1 + Q′ 00Q2 1 + Q′ 11Q2 0). After dividing out by −L2 we obtain det(M). This proves the first assertion. Reversing the algebra gives the second assertion. We analyze the singularities of the hypersurface det(M) = 0. In general, the singularities of the determinant of a symmetric 3×3 matrix of forms is given by the vanishing of the 2 × 2 minors. In geometric terms, this is the Veronese surface Ver ֒→ P5 which has degree four. If the entries are quadratic forms in y0, . . . , y3 then the image of the associated morphism P3 → P5 has degree eight. Bezout’s Theorem gives 32 points of intersection. However, we also have to take into account singularities of the entries. Given the form of the upper-left entry of M, these occur precisely when L = 0. (The other entries are generic.) The determinantal hypersurface ON STABLE RATIONALITY 9 thus has additional singularities along the locus L = Q0 = Q1 = 0. Our genericity assumption implies this is a complete intersection, thus we obtain four additional ordinary double points. (cid:3) We have the following corollary: Corollary 4. The generic sextic double solid arises as a deformation of a nodal birational model of a generic height 22 fibration X → P1 in quartic del Pezzo surfaces. 6. Index one Fano threefolds Let V be a smooth Fano threefold with Pic(V ) = ZKV , i.e., with V takes the following rank one and index one. Its degree d(V ) = −K 3 values [IP99]: d(V ) = 2, 4, 6, 8, 10, 12, 14, 16, 18, 22. For each d(V ) there is an irreducible parameter space for the corre- sponding Fano threefolds. The cases d(V ) = 12, 16, 18, and 22 are rational. When d(V ) = 14 the generic X ⊂ P9 arises as a linear section of the Grassmannian Gr(2, 6). Projective duality gives a codimension ten section of the Pfaffian cubic hypersurface in P14, a cubic threefold V ′. There is a birational map V 99K V ′; see [IM00, §1], for example, for additional details. This example is related to quartic del Pezzo fibrations: One of the two species of quartic del Pezzo fibrations of height ten X → P1 admits a natural morphism [HKT15a, §11] X → V ⊂ P9; the image is a nodal Fano threefold of degree 14. However, stable rationality of cubic threefolds (and birationally equivalent varieties) remains an open problem. 6.1. d(V ) = 2: Sextic double solids. Failure of stable rationality in this case has been established by Beauville [Bea14] and by Colliot- Th´el`ene–Pirutka [CTP15]. It also follows naturally from our formalism: Apply Corollary 4 to reduce to the corresponding del Pezzo fibration of height 22. This realizes a generic height 22 fibration in quartic del Pezzo surfaces as a nodal sextic double solid. If a smooth threefold admits an integral decomposition of the diagonal the same holds true for a specialization with nodes [Voi15, Th. 1.1]. Since a very general del Pezzo fibration of height 22 lacks such a decomposition (by Theorem 2), the same holds for a very general sextic double solid. 10 BRENDAN HASSETT AND YURI TSCHINKEL 6.2. d(V ) = 4: Quartic threefolds. Failure of stable rationality in this case has been established by Colliot-Th´el`ene and Pirutka [CTP14]. To see this from the perspective of fibrations in quartic del Pezzo sur- faces, it suffices to recall that a generic such fibration of height 20 admits a birational model as a determinantal quartic threefold with sixteen nodes [HT14, §11 ] (cf. [Che06, Th. 11]). This lacks an integral decomposition of the diagonal (Theorem 2), so very general quartic threefolds have the same property, hence fail to be stably rational. 6.3. d(V ) = 6: Complete intersections of a quadric and a cubic in P5. We proceed as before, using the fact that a generic quartic del Pezzo fibration of height 18 admits a birational model as a complete intersection Y ⊂ P5 with eight nodes. Indeed, realize X ⊂ P(O⊕4 P1 ⊕ OP1(−1)) ⊂ P1 × P5 as a complete intersection of forms of bidegree (1, 1), (0, 2), and (1, 2), as in Case 3 of [HT14, §4]. (Here we are using the irreducibility of the moduli space of quartic del Pezzo fibrations of height 18.) Let Y ⊂ P5 denote the image of projection onto the second factor. Consider first the image in P5 [x0,...,x5] of the locus cut out by the forms of bidegree (1, 1) and (1, 2): sL0 + tL1 = sQ0 + tQ1 = 0, with L0, L1 ∈ C[x0, . . . , x5]1, Q0, Q1 ∈ C[x0, . . . , x5]2. Its equation is obtained by eliminating s and t, which yields L1Q0 − L0Q1 = det(cid:18)L0 L1 Q0 Q1(cid:19) = 0. This is a cubic fourfold W singular along the elliptic quartic curve C = {L0 = L1 = Q0 = Q1 = 0}. Let Q ∈ C[x0, . . . , x5]2 be the form of bidegree (0, 2), so that Y = W ∩ {Q = 0}. This is a complete intersection of Q with W, having eight nodes at the intersection C ∩ {Q = 0} = {p1, . . . , p8}. The preimages of these nodes in X are distinguished sections of X → P1. We establish failure of stable rationality for very general complete intersections as in the previous cases: Use Theorem 2 to deduce the failure of integral decomposition of the diagonal for very general quartic ON STABLE RATIONALITY 11 del Pezzo fibrations of height eighteen. It follows that very general complete intersections V ⊂ P5 of a quadric and a cubic also lack such decompositions, so stable rationality fails. 6.4. d(V ) = 8: Complete intersections of three quadrics in P6. Let V ⊂ P3 denote a complete intersection of three quadrics. Beauville [Bea77, §6.4,§6.23] has shown that V is birational to a conic fibration X → P2, with discriminant D ⊂ P2 of degree seven, and a generic plane curve of degree seven arises in this way. Thus the results in §3 apply: a very general such V fails to be stably rational. For unity of exposition, we offer a second proof using the fact that a generic quartic del Pezzo fibration of height 16 admits a birational model as a complete intersection Y ⊂ P6 with four nodes. Express X ⊂ P(O⊕3 P1 ⊕ OP1(−1)⊕2) ⊂ P1 × P6 as a complete intersection of two forms of bidegree (1, 1), and two quadratic forms from P6 [x0,...,x6] (see Case 3 of [HT14, §4]). Again, Y is the projection of X onto the second factor. Write the forms of bidegree (1, 1) as sL0 + tL1 = sM0 + tM1 = 0, L0, L1, M0, M1 ∈ C[x0, . . . , x6]1. Eliminating s and t gives a quadratic equation L1M0 − L0M1 = det(cid:18) L0 L1 M0 M1(cid:19) = 0. Let W ⊂ P6 denote the resulting quadric hypersurface; it is singular along the plane P = {L0 = L1 = M0 = M1 = 0} ⊂ P6. Then Y arises as the intersection of W with the zeros of two arbitrary Q0, Q1 ∈ C[x0, . . . , x6]2. It is a complete intersection of three quadrics with singular locus P ∩ {Q0 = Q1 = 0} = {p1, . . . , p4}. The preimage of the singularities are distinguished sections of X → P1. The argument for failure of stable rational for a very general complete intersection of three quadrics proceeds as before. Alternatively, we may observe that such an intersection is birational to a conic bundle over P2, ramified in a curve D of degree 7, and apply results in §3 directly. 12 BRENDAN HASSETT AND YURI TSCHINKEL 6.5. d(V ) = 10: Complete intersections in Gr(2, 5). Fano three- folds V of this type are obtained by intersecting the Grassmannian Gr(2, 5) ⊂ P9 with two linear forms and one quadratic form. In [HT14, §11] we showed that generic quartic del Pezzo fibrations of height fourteen are birational to Y ⊂ P7, where Y is a specialization of V with two nodes. Repeating the arguments above, we conclude that the very general such V fails to be stably rational. 7. Fano threefolds of index two In this section we consider Fano threefolds V with Pic(V ) = Z KV 2 , i.e., those of rank one and index two. Here the degree d(V ) = −K 3 V = 8 · δ(V ) where δ(V ) ∈ N. The possible values are δ(V ) = 1, 2, 3, 4, 5; if δ(V ) = 4 or 5 then V is rational, and when δ(V ) = 3 then V is a cubic threefold. 7.1. δ(V ) = 1: Double cover of Veronese cone. Let P := P(1, 1, 1, 2) ⊂ P6 denote the cone over the Veronese surface Ver ⊂ P5; the vertex p = [0, 0, 0, 1] is a terminal singularity of P of index 2. Let B ⊂ P denote the restriction of a generic cubic hypersurface in P6, which has degree six in the natural grading on P. Consider the double cover φ : V → P branched over B. It is also ramified over p; its preimage v0 ∈ V is smooth. We may regard V as a hypersurface in P(1, 1, 1, 2, 3) of degree six, which is clearly Fano of index two. Note that h1(Ω2 V ) = 21 [IP99, §12] and that V depends on 34 parameters. We specialize B so it contains v0, analyzing the resulting double (This imposes one condition so the construction cover φ : V → P. depends on 33 parameters.) Blowing up p gives a resolution β : P(OP2 ⊕ OP2(−2)) ≃ Blp(P) → P. Let ξ and h generate the Picard group of the projective bundle, where h is the pullback of the hyperplane class on P2 and ξ = c1(OP(O⊕O(−2))). Let E ≃ P2 ⊂ P(OP2 ⊕ OP2(−2)) denote the exceptional divisor; note that [E] = ξ−2h. Let B denote the proper transform of B with [ B] = 2ξ + 2h, V → P(OP1 ⊕ OP1(−2)) the double cover branched along B, and E ⊂ V the preimage of E. Note ON STABLE RATIONALITY 13 that E ≃ P1 × P1 as B meets E ≃ P2 in a plane conic C. Moreover, applying the Hurwitz formula and adjunction yields N E/ V ≃ OP1×P1(−2, −2). The induced birational morphism V → V resolves v0 with exceptional divisor E ≃ P1 × P1. In particular, V → V is universally CH0-trivial (see [CTP14, Prop. 1.8]). For the equivalence between universal CH0- triviality and the existence of integral decompositions of the diagonal, see [ACTP13, Lemma 1.3] and [Voi15, §1]. We compute the invariants of V : The bundle structure : P(OP2 ⊕ OP2(−2)) → P2 induces a morphism ψ : V → P2. Since B is a bisection of , ψ endows V with the structure of a conic bundle. Let D ⊂ P2 denote its discriminant curve, which coincides with the branch locus of : B → P2. An adjunction computation implies K B = h B so D is a plane octic curve, generically smooth. Now D and C are tangent at each point of their intersection, i.e., D ∩ C = 2(z1 + . . . + z8) = 2Z with IZ ⊂ OD the ideal sheaf. Thus D depends on 44 − 8 − 3 = 33 parameters; moreover, the parameter space of smooth octic plane curves eight-tangent to C is birational to a projective bundle over C [8], thus irreducible. Furthermore, η := OD(1)⊗IZ is a two-torsion element of the Jacobian of D. The double cover D′ → D associated with V → P2 is classified by η. From it, we read off the cohomology of V : IJ( V ) = Prym(D′ → D). The curve D has genus 21 so h1(Ω2 reduces this Hodge number by one. V ) = 20. Thus the singularity v0 Lemma 5. There exists a specialization D D1 ∪ D2 of octic curves eight-tangent to C, such that D1 and D2 are transverse plane quartics each four-tangent to C. This satisfies the requirements of §3. Proof. Consider the space of pairs (D1, D2) where D1 and D2 are plane quartics four-tangent to C, with D1 and D2 meeting transversally. Re- peating the argument above, the plane quartics four-tangent to C are 14 BRENDAN HASSETT AND YURI TSCHINKEL birational to a P6 bundle over C [4], an irreducible rational variety of dimension ten. It is easy to check that a generic pair of such curves meets transversally, yielding a rational parameter space of dimension twenty. Write D1 ∩ C = 2(z1 + z2 + z3 + z4) D2 ∩ C = 2(z5 + z6 + z7 + z8) and IZ ⊂ OD1∪D2 for the ideal sheaf of Z = {z1, . . . , z8}. Thus η0 := IZ(1) is two-torsion in the Picard group of D1 ∪ D2 and restricts to non-trivial two-torsion elements on D1 and D2 because z1 + z2 + z3 + z4 6≡ [OD1(1)], z5 + z6 + z7 + z8 6≡ [OD2(1)]. Otherwise, these four-tuples of points would be collinear. Linear algebra shows that we can smooth D1 ∪ D2 to a smooth plane octic D tangent to C at z1, . . . , z8. As we saw in the proof of Theorem 2, this gives rise to a cover D′ → D classified by the divisor η. As D D1 ∪ D2 we have η η0. (cid:3) Thus the results of §3 imply that V fails to admit an integral de- composition of the diagonal. An application of the results of [CTP14, §1] implies that a very general V ⊂ P(1, 1, 1, 2, 3) also fails to admit an integral decomposition of the diagonal, and thus is not stably rational. 7.2. δ(V ) = 2: Quartic double solids. Let V be a quartic double solid φ : V → P3 with branch locus a degree four K3 surface B. When V is smooth we have h1(Ω2 V ) = 10. Voisin [Voi15] and Colliot-Th´el`ene–Pirutka [CTP15] established the failure of stable rationality for very general varieties in this class. Here we discuss how to approach this through conic bundle fibrations. Now suppose V (or equivalently B) has a node p and write V = Blp(V ). Projection from p gives a conic bundle structure π : V → P2 branched along a sextic plane curve D. The plane curve is typically smooth but admits a six-tangent conic curve C corresponding to the exceptional divisor of the induced resolution of B. Write D ∩ C = 2Z, Z = z1 + · · · + z6 so that η := OD(1) ⊗ IZ is two-torsion on D. Here IZ is the ideal sheaf of Z. ON STABLE RATIONALITY 15 As we saw in the previous case, the parameter space of sextic plane curves six-tangent to a prescribed conic is irreducible, being a projective bundle over the Hilbert scheme C [6]. We can specialize D D1 ∪ D2, where D1 and D2 are smooth plane cubics meeting transversally, each three-tangent to C. Thus η specializes to a two-torsion divisor on D1 ∪ D2 that is non-trivial on each component. An application of the results of §3 implies that the very general quar- tic double solid fails to have an integral decomposition of the diagonal, thus is not stably rational. 8. Fano threefolds of higher Picard rank As before we write d(V ) = −K 3 V . 8.1. d(V ) = 6, h1,2(V ) = 20. The first case is double covers V → P1 × P2 branched over a divisor of bidegree (2, 4). These depend on 3 × 15 − (1 + 3 + 8) = 33 parameters. Projection onto the second factor gives a conic bundle structure V → P2 with octic discriminant curve D ⊂ P2. The equation of D is given by the vanishing of the determinant of a 2 × 2 symmetric matrix of quartic forms in three variables. In particular, D is not gen- eral in its linear series and each symmetric determinantal octic comes with a distinguished non-trivial two-torsion class, i.e., the one associ- ated with the ramification data of V → P2. This makes it hard to apply the methods of §3 directly. However, there is a natural degeneration of such Fano threefolds to another class of rationally connected varieties: Fix distinct points p, q ∈ P2 and consider divisors B0 ⊂ P1 × P2 of bidegree (2, 4) whose fibers over P1 admit nodes at p and q. (Equivalently, these are singular along P1 × {p, q}.) Consider the birational map P2 99K P1 × P1 blowing up p and q and blowing down the line joining them. This takes quartic plane curves singular at p and q to bidegree (2, 2) curves in P1 × P1. Using the induced birational map P1 × P2 99K P1 × P1 × P1, 16 BRENDAN HASSETT AND YURI TSCHINKEL we see that B0 is mapped to (2, 2, 2) divisor in the image. Conversely, (2, 2, 2) divisors in the image all arise from this construction. Lemma 6. Let V0 denote the double cover of P1 × P2 branched over a very general such B0. Let eV0 → V0 denote the blowup along P1 × {p, q}. If eV0 admits no integral decomposition of the diagonal then the same holds for the very general Fano variety V arising as a double cover of P1 × P2 branched over a divisor of bidegree (2, 4). Proof. The singularities of V0 are along the lines ℓp := P1 × {p} and ℓq := P1 ×{q}. The singularities of B0 are ordinary double points along the general points of these lines and cusps (analytically isomorphic to x2 + y3 = 0) at a finite number of special points. For special r ∈ P1 the local singularity type of V0 at (r, p) is of the form w2 = x2 + ty2 + y3, where {x, y} are local coordinates of P2 centered at p, t is a local coordinate of P1 centered at r, and w is used to realize the double cover over P1 × P2. Thus the singularities of V0 are resolved by blowing up the lines eV0 = Blℓp∪ℓq(V0) → V0. The exceptional fibers over the generic points of ℓp and ℓq are smooth conics; the fibers over special points are reducible conics. This com- putation is similar to, but simpler than, the singularity analysis of [CTP14, App.]. The key is the exceptional fibers are universally CH0-trivial, in the sense of [CTP14, D´ef. 1.2]. Applying the result on universal CH0- triviality in [CTP14, §1], we conclude that V fails to be universally CH0-trivial if the same holds for eV0. (See [ACTP13, Lemma 1.3] and [Voi15, §1] for the equivalence with integral decompositions of the di- agonal.) (cid:3) In §8.4 we will show that very general double covers of P1 × P1 × P1 branched over a (2, 2, 2) divisor do not admit integral decompositions of the diagonal. 8.2. d(V ) = 12, h1,2(V ) = 9. The first part of the second case is re- alized as a divisor in P2 × P2 of bidegree (2, 2), depending on 19 pa- rameters. Using either projection, we obtain a conic bundle over P2 with sextic discriminant. It is well known that the plane sextic can be chosen generically [vG05, HVA13, §9]. The main result of §3 implies that very general conic bundles over P2 with sextic discriminant fail to ON STABLE RATIONALITY 17 be stably rational. That is, for a very general pair (D, D′ → D), where D is a plane sextic and D′ → D is a non-trivial ´etale double cover, the corresponding conic bundle X → P2 fails to be stably rational. It follows that for very general D, every double cover D′ → D is associ- ated with a conic bundle that is not stably rational. In particular, this applies to the very general divisor of bidegree (2, 2) in P2 × P2. The second part of the second case is a double cover V of a hypersur- face F(1, 2) ⊂ P2 × P2 of bidegree (1, 1) branched over an anticanonical divisor B of bidegree (2, 2). This depends on 18 parameters. Again, either projection induces a conic bundle V → P2 with sextic discrim- inant curve D. The Mori-Mukai classification [MM82] shows this is a specialization of the first part. The surface B is a lattice polarized K3 surface of type Φ := f1 f2 f1 f2 4 2 4 2 and the generic such surface arises as a complete intersection of forms of bidegree (1, 1) and (2, 2) in P2 × P2. The branch curve of πiB : B → P2 coincides with the locus where πiV : V → P2 fails to be smooth, i.e., the discriminant curve D. This is a sextic plane curve that is not of general moduli—the associated K3 double cover has Picard rank two. The technique of §3 does not immediately apply in this case. It is easy to use degeneration techniques to reduce this to cases where there is no integral decomposition of the diagonal. Consider a quartic surface B0 ⊂ P3 with nodes n1 and n2 and minimal resolution fB0. Its Picard group takes the form h R1 R2 0 4 0 0 −2 h 0 R1 0 −2 R2 0 where h is the pullback of the hyperplane class and the Ri are the exceptional divisors. The lattice Φ embeds into this lattice f1 = h − R1, f2 = h − R2. Projection from the nodes n1 and n2 gives a morphism fB0 → P2 × P2 18 BRENDAN HASSETT AND YURI TSCHINKEL with image B◦ a complete intersection of hypersurfaces of degrees (1, 1) and (2, 2). This extends to a birational map P3 99K F(1, 2) := P(Ω1 P2(1)) ⊂ P2 × P2 onto the divisor of bidegree (1, 1). To summarize: Lemma 7. A double cover of P3 branched over a generic quartic sur- face with two nodes is birational to a double cover of the complete flag variety F(1, 2) along a special anticanonical divisor. By [Voi15, Th. 1.1] a very general such quartic double solid fails to admit an integral decomposition of the diagonal. Thus the same holds for very general Fano threefolds V → F(1, 2) and so these fail to be stably rational. 8.3. d(V ) = 14, h1,2(V ) = 9. The third case is the double cover of P3 blown up at a point, with anticanonical branch locus B meeting the exceptional divisor transversally. These conic bundles were addressed in §7.2 as degenerate quartic double solids. 8.4. d(V ) = 12, h1,2(V ) = 8. The fourth case is a double cover V → P1 × P1 × P1 branched over a divisor of degree (2, 2, 2). These depend on 27−1−9 = 17 parameters. For each projection onto P1 × P1 we obtain a conic bundle, with discriminant D of bidegree (4, 4). Note that this is not generic; it has equation D = {det(M) = 0}, M =(cid:18)M11 M12 M12 M22(cid:19) , with M11, M12, M22 ∈ Γ(OP1×P1(2, 2)). This may be interpreted geometrically: the K3 double cover has Picard group B → P1 × P1 Π := E1 E2 E3 0 2 2 2 2 0 2 0 2 E1 E2 E3 In this basis, D ≡ 4(E1 + E2). The curves E3 and 2(E1 + E2) − E3 are conjugate under the involution associated with the first two factors, ON STABLE RATIONALITY 19 which fixes D. Thus E1 + E2 − E3 restricts to a two-torsion divisor η on D, which classifies the double cover D′ → D. Given that D is not generic in its linear series, the techniques of §3 do not apply directly. Clearly the monodromy cannot act transitively on the non-trivial two-torsion of D, as there is a distinguished element η. Keeping track of what happens to η as D D1 ∪D2 can be delicate. The quickest proof that the very general V → P1 × P1 × P1 fails to admit a decomposition of the diagonal is via degeneration of the branch locus. Let B0 ⊂ P3 denote a quartic surface with nodes n1, n2, n3 and minimal resolution fB0. Let h denote the pullback of the polarization and R1, R2, R3 the exceptional divisors over n1, n2, n3: h R1 R2 R3 0 4 0 0 0 −2 0 0 0 −2 0 h 0 R1 0 −2 R2 0 R3 0 Note that R0 = h − R1 − R2 − R3 is also a smooth rational curve in fB0. We can embed Π naturally into this lattice: E1 = h − R2 − R3, E2 = h − R1 − R3, E3 = h − R1 − R2. These reflect elliptic fibrations induced by pencils of planes in P3 pass- ing through two of the three nodes. Note that Ei = R0 + Ri, i = 1, 2, 3, which means that each elliptic fibration admits a fiber of Kodaira type I2 containing R0 as a component. The connection between B0 ⊂ P3 and (2, 2, 2) K3 surfaces in P1 × P1 × P1 goes further. There is a birational map P3 99K P1 × P1 × P1, where the map onto each factors is given by the pencil of planes through a pair of nodes of B0. This maps B0 birationally onto a (2, 2, 2) nodal K3 surface B◦, as R0 is in the fiber of each of the elliptic fibrations. Conversely, generic nodal (2, 2, 2) surfaces B◦ ⊂ P1 × P1 × P1 yield quartic surfaces with three nodes. To summarize: Lemma 8. Generic double solids V0 → P3 branched over a quartic surface with three nodes yield double covers V◦ → P1 × P1 × P1 branched over a nodal (2, 2, 2) surface, and vice versa. 20 BRENDAN HASSETT AND YURI TSCHINKEL Voisin [Voi15, Th. 1.1] has shown that a double solid branched over a very general quartic surface with r ≤ 7 nodes fails to admit an integral decomposition of the diagonal. Thus the same holds for a double cover of P1 × P1 × P1 branched over a very general (2, 2, 2) surface. We conclude that such threefolds fail to be stably rational. References [ACTP13] Asher Auel, Jean-Louis Colliot-Th´el`ene, and Raman Parimala. Univer- sal unramified cohomology of cubic fourfolds containing a plane. In Brauer Groups and Obstruction Problems: Moduli Spaces and Arith- metic. 2013. Proceedings of a workshop at the American Institute of Mathematics. V. A. Alekseev. On conditions for the rationality of three-folds with a pencil of del Pezzo surfaces of degree 4. Mat. Zametki, 41(5):724–730, 766, 1987. [Ale87] [Bea77] [Bea86] [Bea14] [AM72] M. Artin and D. Mumford. Some elementary examples of unirational varieties which are not rational. Proc. London Math. Soc. (3), 25:75–95, 1972. Arnaud Beauville. Vari´et´es de Prym et jacobiennes interm´ediaires. Ann. Sci. ´Ecole Norm. Sup. (4), 10(3):309–391, 1977. Arnaud Beauville. Le groupe de monodromie des familles universelles d’hypersurfaces et d’intersections compl`etes. In Complex analysis and algebraic geometry (Gottingen, 1985), volume 1194 of Lecture Notes in Math., pages 8–18. Springer, Berlin, 1986. Arnaud Beauville. A very general sextic double solid is not stably ratio- nal. Bull. London Math. Soc., to appear, 2014. arXiv:1411.7484. Arnaud Beauville. The Luroth problem, 2015. arXiv:1507.02476. Ivan Cheltsov. Nonrational nodal quartic threefolds. Pacific J. Math., 226(1):65–81, 2006. Jean-Louis Colliot-Th´el`ene and Alena Pirutka. Hypersurfaces quar- tiques de dimension 3 : non rationalit´e stable. Ann. Sci. ENS, to appear, 2014. arXiv:1402.4153. Jean-Louis Colliot-Th´el`ene and Alena Pirutka. Cyclic covers that are not stably rational. Izvestiya RAN, Ser. Math., to appear, 2015. arXiv:1506.0042v2. [Bea15] [Che06] [CTP14] [CTP15] [HKT15a] Brendan Hassett, Andrew Kresch, and Yuri Tschinkel. On the moduli of degree 4 Del Pezzo surfaces, 2015. To appear in the proceedings of the ‘Development of Moduli Theory’ conference in honor of S. Mukai’s 60th birthday, arXiv:1312.6734. [HKT15b] Brendan Hassett, Andrew Kresch, and Yuri Tschinkel. Stable rationality and conic bundles. Math. Annalen, to appear, 2015. arXiv:1503.08497. Brendan Hassett and Yuri Tschinkel. Quartic del Pezzo surfaces over function fields of curves. Cent. Eur. J. Math., 12(3):395–420, 2014. [HT14] ON STABLE RATIONALITY 21 [IM00] [IP99] [MM82] [HVA13] Brendan Hassett and Anthony V´arilly-Alvarado. Failure of the Hasse principle on general K3 surfaces. J. Inst. Math. Jussieu, 12(4):853–877, 2013. A. Iliev and D. Markushevich. The Abel-Jacobi map for a cubic three- fold and periods of Fano threefolds of degree 14. Doc. Math., 5:23–47 (electronic), 2000. V. A. Iskovskikh and Yu. G. Prokhorov. Fano varieties. In Algebraic geometry, V, volume 47 of Encyclopaedia Math. Sci., pages 1–247. Springer, Berlin, 1999. Shigefumi Mori and Shigeru Mukai. Classification of Fano 3-folds with B2 ≥ 2. Manuscripta Math., 36(2):147–162, 1981/82. V. G. Sarkisov. On conic bundle structures. Izv. Akad. Nauk SSSR Ser. Mat., 46(2):371–408, 432, 1982. Burt Totaro. Hypersurfaces that are not stably rational. Journal of the American Mathematical Society, to appear, 2015. arXiv:1502.04040. Bert van Geemen. Some remarks on Brauer groups of K3 surfaces. Adv. Math., 197(1):222–247, 2005. Claire Voisin. On the universal CH0 group of cubic hypersurfaces. J. Eur. Math. Soc. (JEMS), to appear, 2014. arXiv:1407.7261v2. Claire Voisin. Unirational threefolds with no universal codimension 2 cycle. Invent. Math., 201(1):207–237, 2015. [Tot15] [Sar82] [vG05] [Voi14] [Voi15] Department of Mathematics, Brown University, Box 1917 151 Thayer Street Providence, RI 02912, USA E-mail address: [email protected] Courant Institute, New York University, New York, NY 10012, USA E-mail address: [email protected] Simons Foundation, 160 Fifth Avenue, New York, NY 10010, USA
1805.07121	2	1805	2019-11-04T16:56:18	Motivic periods and Grothendieck arithmetic invariants	[ "math.AG", "math.KT", "math.NT" ]	We construct a period regulator for motivic cohomology of an algebraic scheme over a subfield of the complex numbers. For the field of algebraic numbers we formulate a period conjecture for motivic cohomology by saying that this period regulator is surjective. Showing that a suitable Betti--de Rham realization of 1-motives is fully faithful we can verify this period conjecture in several cases. The divisibility properties of motivic cohomology imply that our conjecture is a neat generalization of the classical Grothendieck period conjecture for algebraic cycles on smooth and proper schemes. These divisibility properties are treated in an appendix by B. Kahn (extending previous work of Bloch and Colliot-Th\'el\`ene--Raskind).	math.AG	math	MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Abstract. We construct a period regulator for motivic cohomology of an algebraic scheme over a subfield of the complex numbers. For the field of algebraic numbers we formulate a period conjecture for motivic cohomology by saying that this period regula- tor is surjective. Showing that a suitable Betti -- de Rham realization of 1-motives is fully faithful we can verify this period conjecture in several cases. The divisibility properties of motivic cohomology imply that our conjecture is a neat generalization of the classical Grothendieck period conjecture for algebraic cycles on smooth and proper schemes. These divisibility properties are treated in an appendix by B. Kahn (extending previous work of Bloch and Colliot-Thélène -- Raskind). Contents Introduction 1. Periods: constructions and conjectures 2. Periods of 1-motives: fullness of Betti-de Rham realizations 3. Some evidence: description of some Grothendieck arithmetic invariants Appendix A. Divisibility properties of motivic cohomology (by B. Kahn) References 1 4 11 24 35 40 Introduction Let X be a scheme which is separated and of finite type over a subfield K of the complex numbers. Consider the q-twisted singular cohomology H p(Xan, Zan(q)) of the analytic space Xan associated to the base change of X to C and the pth de Rham cohomology H p dR(X), which is an algebraically defined K-vector space. We have the following natural C-linear isomorphism p,q X : H p(Xan, Zan(q)) ⊗Z C ∼= H p dR(X) ⊗K C providing a comparison between these cohomology theories. As Grothendieck originally remarked, for X defined over the field of algebraic numbers K = Q or a number field, the position of the whole H p X «yields an interesting arithmetic invariant, generalizing the "periods" of regular differential forms» (see [30, p. 101 & footnotes (9) and (10)], cf. [3, §7.5 & Chap. 23], [16], [17] and [33, Chap. dR(X) with respect to H p(Xan, Zan(q)) under p,q 2000 Mathematics Subject Classification. 14F42, 14F40, 19E15, 14C30, 14L15. Key words and phrases. Motives, Periods, Motivic and de Rham Cohomology. 1 2 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) 5 & 13]). For the comparison of several notions of "periods" and versions of the period conjecture we refer to Huber survey [32] (see also [17, §2.2.2]). The main goal of this paper is to describe this arithmetic invariant, at least for p = 1 X (in and all twists, notably, q = 1 and q = 0. Definition 1.2.4) by making use of Ayoub's period isomorphism (see Lemma 1.2.2) in Vo- evodsky's triangulated category DMeff ´et of motivic complexes for the étale topology. Denote by H p,q (X) the named arithmetic invariant, i.e., the subgroup of those cohomology classes in H p(Xan, Zan(q)) which are landing in H p dR(X) via p,q X . We then show the existence of a regulator map (see Corollary 1.2.6 and Definition 1.2.7) In more details, we first reconstruct p,q rp,q : H p,q(X) → H p,q (X) from étale motivic cohomology groups H p,q(X). Here we regard motivic cohomology canon- ically identified with H p ´eh(X, Z(q)) where Z(q) is the Suslin-Voevodsky motivic complex (see [47, Def. 3.1]), as a complex of sheaves for the ´eh-topology (introduced in [12, §10.2]). We are mostly interested in the case of q = 0, 1 so that Z(0) ∼= Z[0] and Z(1) ∼= Gm[−1] by a theorem of Voevodsky (see [47, Thm. 4.1]). Following Grothendieck's idea, we conjecture that the period regulator rp,q is surjective over Q and we actually show some evidence. We easily see that H 0,q (X) = 0 for q 6= 0 and r0,0 is an isomorphism: therefore, the first non-trivial case is for p = 1. Moreover, by making use of Suslin-Voevodsky rigidity theorem we can show that rp,q is surjective on torsion (see if the vanishing H p,q(X) ⊗ Q/Z = 0 holds true then Lemma 1.4.2). We can also show: the surjectivity of rp,q dR(X) ∩ H p(Xan, Qan(q)) = 0. The divisibility properties of motivic cohomology (see Appendix A) imply that our conjecture is a neat generalization of the classical period conjecture for algebraic cycles on smooth and proper schemes (see Proposition 1.4.4). is equivalent to the vanishing H p In order to study the case p = 1 we can make use of the description of H 1 via the Albanese 1-motive L1Alb(X). Recall the existence of the homological motivic Albanese complex LAlb(X), a complex of 1-motives whose pth homology LpAlb(X) is a 1-motive with cotorsion (see [12, §8.2] for details). We can regard complexes of 1-motives as objects of DMeff ´et and by the adjunction properties of LAlb (proven in [12, Thm. 6.2.1]) we have a natural map Extp(LAlb(X), Z(1)) → H p,1(X) ∼= H p−1 ´eh (X, Gm) which is an isomorphism, rationally, for all p (see the motivic Albanese map displayed in (3.2) and (3.3) below). We can also describe periods for 1-motives (see Definition 2.2.1) in such a way that we obtain suitable Betti-de Rham realizations in period categories (see Definitions 2.5.4 and 2.5.1): a key point is that these realizations are fully faithful over Q (see Theorem 2.7.1). The main ingredient in the proof of fullness is a theorem due to Waldschmidt [54, Thm. 5.2.1] in transcendence theory, generalizing the classical Schneider- Lang theorem (see also [16, Thm. 4.2]). An alternative proof can be given using a theorem of Wüstholz [56] (see our second proof of Theorem 2.7.1). A version of Baker's theorem and instances of Kontsevich period conjecture for 1-motives are further explored in a recent work of Huber and Wüstholz [34]. Note that Kontsevich's period conjecture for 1-motives was formulated in [57] (see also [3, §23.3.3]). MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 3 Actually, we show that the regulator rp,1 can be revisited by making use of 1-motives (see Lemmas 3.2.1 to 3.2.3 and Proposition 3.2.7). As a byproduct, all this promptly applies to show the surjectivity of r1,1 (X) via Ext(LAlb(X), Z(1)) verifying the conjecture for p = 1 and q = 1 (see Theorem 3.2.4). In fact, we can use the motivic Picard complex RPic(X) (see [12, §8.3]) so that Extp(Z(0), RPic(X)) = Extp(LAlb(X), Z(1)) by Cartier duality showing that Ext(LAlb(X), Z(1)) is an extension of the finitely generated group Hom(Z, RpPic(X)) by a divisible group; we thus conclude that rp,1 ´eh(X, Gm) → H 1,1 induces a map : H 0 θp : Hom(Z, RpPic(X)) → H p,1 (X) which in turn can be described making use of the mentioned Betti-de Rham realization. The surjectivity of rp,1 and the fullness of the Betti-de Rham realization. For p = 1, considering the 1-motive R1Pic(X) = [L∗ 1] which is the Cartier 1 u1→ G1], we get a canonical isomorphism dual of L1Alb(X) = [L1 can be translated via θp u∗ 1→ G∗ Ker u∗ 1 ∼= Hom(Z, R1Pic(X)) ∼= H 1 dR(X) ∩ H 1(Xan, Zan(1)) = H 1,1 (X). In particular, we obtain that H 1 for smooth projective varieties was previously obtained by Bost-Charles [17, Thm. 4.2]. dR(X) ∩ H 1(Xan, Z(1)) = 0 if X is proper. This vanishing With some more efforts, making now use of the motivic complex Lπ0(X) along with its adjunction property (as stated in [12, §5.4]), we get a map Extp(Lπ0(X), Z(0)) → H p,0(X) ∼= H p ´eh(X, Z). Analysing the composition of this map for p = 1 with r1,0 we see that r1,0 : H 1 ´et(X, Z) ∼= H 1,0 (X) = H 1 dR(X) ∩ H 1(Xan, Zan) is an isomorphism (see Theorem 3.3.1), which yields the case p = 1 and q = 0 of our conjecture. In particular, H 1 dR(X) ∩ H 1(Xan, Zan) = 0 for X normal. This vanishing for smooth quasi-projective varieties was previously obtained by Bost-Charles [17, Thm. 4.1]. (X) = 0 (see Corollary 3.4.2) so that the period For p = 1 and q 6= 0, 1 we have that H 1,q conjecture for motivic cohomology is trivially verified. Remarkably, the description of the Grothendieck arithmetic invariants H p,q (X) appears strongly related to the geometric properties encoded by motivic cohomology. These prop- erties are almost hidden for smooth schemes, since the divisibility properties of motivic cohomology of X smooth yields that for p /∈ [q, 2q] the surjectivity of rp,q is equivalent to the vanishing H p dR(X) ∩ H p(Xan, Qan(q)) = 0. However, for X smooth with a smooth compactification X and normal crossing boundary Y , we have that Ker (Div0 Y (X) u∗ 1−→ Pic0 X/Q) ∼= H 1 dR(X) ∩ H 1(Xan, Zan(1)) 1 is the canonical mapping sending a divisor D supported on Y to OX (D). where u∗ In fact, here R1Pic(X) is Cartier dual of L1Alb(X) = [0 → A0 ], the Serre-Albanese semi- abelian variety (see [12, Chap. 9]). Therefore, there exist smooth schemes X such that H 1,1 (X) is non-zero and the vanishings in [17, Thm. 4.1 & 4.2] are particular instances of our descriptions. X/Q 4 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) (see [12, Def. 8.7.1]) to describe the compactly supported variant H 1,q With similar techniques one can make use of the Borel-Moore Albanese complex LAlbc(X) c,(X), for any twist q. Finally, the cohomological Albanese complex LAlb∗(X) (see [12, Def. 8.6.2]) shall be providing a description of H 2d−j,q (X) for d = dim(X), at least for j = 0, 1 and q an arbitrary twist. A homological version of period regulators is also feasible and will be discussed in a future work. Aknowledgements. We would like to thank Y. André and J. Ayoub for some useful discus- sions on the matters treated in this paper. The first author was partially supported by the Cariplo project n. 2017-1570 and all Italian authors acknowledge the support of the Min- istero dell'Istruzione, dell'Università e della Ricerca (MIUR) through the Research Project (PRIN 2010-11) "Arithmetic Algebraic Geometry and Number Theory". We thank the ref- eree for her/his many comments that helped us to considerably improve the exposition (starting with the suggestion to change the title to something reflecting more accurately the contents of the paper). 1. Periods: constructions and conjectures Let DMeff τ be the effective (unbounded) triangulated category of Voevodsky motivic com- plexes of τ -sheaves over a field K of zero characteristic, i.e., the full triangulated subcategory of D(Shvtr τ (SmK)) given by A1-local complexes (e.g. see [6, §4.1] and, for complexes bounded above, see also [47, Lect. 14]). We here generically denote by τ either the Nisnevich or étale Grothendieck topology on SmK , the category of smooth schemes which are of finite type over the field K. Let Z(q) for q ≥ 0 be the Suslin-Voevodsky motivic complex regarded as a complex of étale sheaves with transfers. More precisely we consider a change of topology tensor functor α : DMeff Nis → DMeff ´et and Z(q) = αZNis(q) (see [12, Cor. 1.8.5 & Def. 1.8.6]) where ZNis(q) is the usual complex for the Nisnevich topology (see also [50, Def. 3.1]). We have the following canonical iso- morphisms Z(0) ∼= Z[0], Z(1) ∼= Gm[−1] and Z(q) ⊗ Z(q′) ∼= Z(q + q′) for any q, q′ ≥ 0 (see [50, Lemma 3.2]). For any object M ∈ DMeff ´et we here denote M (q) := M ⊗ Z(q). Recall that by inverting the Tate twist M M (1) we obtain DMτ (where every compact object is isomorphic to M (−n) for some n ≥ 0 and M compact and effective). For M ∈ DMeff ´et we shall define its motivic cohomology as H p,q(M ) := HomDMeff ´et (M, Z(q)[p]). For any algebraic scheme X we have the Voevodsky étale motive M (X) = αC•Ztr(X) ∈ DMeff ´et where C• is the Suslin complex and Ztr(X) is the representable Nisnevich sheaf with transfers (see [47, Def. 2.8, 2.14 & Properties 14.5] and compare with [12, Lemma 1.8.7 & Sect. 8.1]). We then write H p,q(X) := H p,q(M (X)) and we refer to it as the étale motivic cohomology of X. We have an isomorphism H p,q(X) ∼= H p ´eh(X, Z(q)) MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 5 [6] and [50, Prop. 1.8 & Def. 3.1]). where the last cohomology is, in general, computed by the ´eh-topology (see [12, §10.2] and ´eh(X, Z(q)) ∼= cf. H p ´et(X, Z(q)). Note that we also have the triangulated category of motivic complexes without transfers ´et and if we are interested in rational coefficients we may forget transfers or keep the In particular, if X is smooth H p DAeff Nisnevich topology as we have equivalences DAeff ´et,Q ∼= DMeff ´et,Q ∼= DMeff Nis,Q (see [6], [7, Cor. B.14] and [47, Thm. 14.30]). If we work with rational coefficients, we then have that motivic cohomology H p,q(X)Q is computed by the cdh-topology, i.e., H p Zar(X, Q(q)) if X is smooth. cdh(X, Q(q)), and H p cdh(X, Q(q)) ∼= H p ´eh(X, Q(q)) ∼= H p 1.1. de Rham regulator. Denote by Ω the object of DMeff ´et which represents de Rham cohomology. More precisely we here denote Ω := αΩNis where ΩNis is the corresponding object for the Nisnevich topology (see [45, §2.1] and cf. [7, §2.3] without transfers). This latter ΩNis is given by the complex of presheaves with transfers that associates to X ∈ SmK the global sections Γ(X, Ω• X/K ) of the usual algebraic de Rham complex. For M ∈ DMeff ´et we shall denote (cf. [44, §6] and [45, Def. 2.1.1 & Lemma 2.1.2]) H p dR(M ) := HomDMeff ´et (M, Ω[p]). For any algebraic scheme X and M = M (X) we here may also consider the sheafification of Ω for the ´eh-topology. Actually, we set dR(X) := H p dR(M (X)) ∼= H p ´eh(X, Ω) H p (see [12, Prop. 10.2.3]). Remark that this definition is equivalent to the definition of the algebraic de Rham cohomology in [33, Chap. 3] via the h-topology (as one can easily see via blow-up induction [12, Lemma 10.3.1 b)] after [33, Prop. 3.2.4] and [33, Lemma 3.1.14]). Note that for q = 0 we have a canonical map r0 : Z(0) → Ω yielding a map H p,0(X) ∼= H p ´eh(X, Z) → H p For q = 1 we have r1 := d log : Z(1) → Ω in DMeff topology and apply α) yielding a map H p,1(X) ∼= H p−1 dR(X) ∼= H p Following [45, (2.1.5)] an internal de Rham regulator rq in DMeff as the composition of ´eh (X, Gm) → H p rq : Z(q) ∼= Z(1)⊗q d log⊗q −→ Ω⊗q → Ω. (1.1) For M ∈ DMeff regulator map dR(X) ∼= H p ´et (see [45, Lemme 2.1.3] for the Nisnevich ´eh(X, Ω). ´eh(X, Ω). ´et for q ≥ 2 is then obtained ´et , composing a map M → Z(q)[p] with rq[p] we get an external de Rham (1.2) rp,q dR : H p,q(M ) → H p dR(M ) and in particular for M = M (X) we get rp,q dR : H p,q(X) ∼= H p ´eh(X, Z(q)) → H p dR(X) ∼= H p ´eh(X, Ω). 6 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Note that if X is smooth then H p classical algebraic de Rham cohomology (again, see [12, Prop. 10.2.3] and cf. 3.2.4]) and we thus obtain rp,q ´eh(X, Ω) ∼= H p ´et(X, Ω) ∼= H p Zar(X, Ω• ´et(X, Z(q)) → H p X ) in this case. dR : H p Zar(X, Ω• X ) coincides with the [33, Prop. 1.2. Periods. As soon as we have an embedding σ : K ֒→ C we may consider a Betti realization (e.g. see [45, §3.3] or [5, Def. 2.1]) in the derived category of abelian groups D(Z) as a triangulated functor βσ : DMeff (1.3) such that βσ(Z(q)) ∼= Zan(q) := (2πi)q Z[0]. Actually, following Ayoub (see also [7, §2.1.2] and [8, §1.1.2]) if we consider the analogue of the Voevodsky motivic category DMeff an obtained as the full subcategory of D(Shvtr an-local complexes, where we here replace smooth schemes SmK by the category AnC of complex analytic manifolds, we get an equivalence an(AnC)) given by A1 ´et → D(Z) β : DMeff an ≃ −→ D(Z) such that Man(X) Sing∗(X) is sent to the singular chain complex of X ∈ AnC. Moreover there is a natural triangulated functor σ : DMeff ´et → DMeff an such that M (X) Man(Xan) where the analytic space Xan is given by the C-points of the base change XC of any algebraic scheme X. We then set βσ := β ◦ σ. Thus it is clear that βσ(Z[0]) = βσ(M (Spec(K)) = Z[0]. Since a K-rational point of X yields M (X) = Z ⊕fM (X) we also see that βσ(Z(1)[1]) ∼= βσ(fM (Gm)) ∼= β(fMan(C∗)) ∼= Zan(1)[1] and then, βσ(Z(q)) ∼= Zan(q) in general, as it follows from the compatibility of βσ with the tensor structures, i.e., we here use the fact that βσ is unital and monoidal. For M ∈ DMeff ´et , we denote H p,q an (M ) := HomD(Z)(βσM, Zan(q)[p]) and we have a Betti regulator map rp,q an : H p,q(M ) → H p,q (1.4) an (M ) induced by βσ. In particular, for M = M (X), we obtain from Ayoub's construction (see also [45, Prop. 4.2.7]): 1.2.1. Lemma. For any algebraic K-scheme X and any field homomorphism σ : K ֒→ C we have H p,q an (X) := HomD(Z)(βσM (X), Zan(q)[p]) ∼= H p(Xan, Zan(q)) and a Betti regulator map rp,q an : H p,q(X) → H p(Xan, Zan(q)). Recall that the functor βσ admits a right adjoint βσ : D(Z) → DMeff ´et (see [8, Def. 1.7]). Note that the Betti regulator (1.4) is just given by composition with the unit (1.5) rq σ : Z(q) → βσβσ(Z(q)) of the adjunction. Actually, by making use of the classical Poincaré Lemma and Grothendieck comparison theorem ([30, Thm. 1′]) we get: MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 7 1.2.2. Lemma (Ayoub). There is a canonical quasi-isomorphism q : βσβσ(Z(q)) ⊗Z C q.i. −→ Ω ⊗K C whose composition with rq σ in (1.5) is the regulator rq in (1.1) after tensoring with C. Proof. See [7, Cor. 2.89 & Prop. 2.92] and also [9, §3.5]. 1.2.3. Remark. Note that applying βσ to q we obtain a quasi-isomorphism βσ(q) such that (cid:3) C ∼= βσ(Z(q)) ⊗Z C βσ(rq σ)C / βσβσβσ(Z(q)) ⊗Z C βσ(q) / βσ(Ω) ⊗K C where βσ(rq)C is a split injection but it is not a quasi-isomorphism (cf. [45, §4.1]). For M ∈ DMeff ´et , by composition with q we get a period isomorphism βσ(rq)C p,q M : H p,q an (M ) ⊗Z C ≃ −→ H p dR(M ) ⊗K C. 1.2.4. Definition. For any scheme X we shall call period isomorphism the C-isomorphism obtained by setting p,q the period isomorphism. p,q X : H p(Xan, Zan(q)) ⊗Z C X := p,q M (X) as above. We shall denote ηp,q −→ H p dR(X) ⊗K C X := (p,q ≃ X )−1 the inverse of We also get the following compatibility. 1.2.5. Proposition. For M ∈ DMeff of the period isomorphism p,q ´et along with a fixed embedding σ : K ֒→ C the inverse M above induces a commutative diagram H p,q(M ) rp,q dR H p dR(M ) rp,q an H p,q an (M ) ιp,q an (◗◗◗◗◗◗◗◗◗◗◗◗◗ ≃ H p dR(M ) ⊗K C / H p,q an (M ) ⊗Z C ιp,q dR where ιp,q dR and ιp,q an are the canonical mappings given by tensoring with C. Proof. This easily follows from Lemma 1.2.2. In fact, by construction, the claimed commu- tative diagram can be translated into the following commutative square: HomDMeff (M, Z(q)[p]) rp,q dR := rq[p]◦− HomDMeff (M, Ω[p]) ιp,q an ◦ rp,q an ιp,q dR −⊗K C / HomDMeff (M, βσβσZ(q)[p])C q[p]◦− / HomDMeff (M, Ω[p])C. (cid:3) 2 2 / s s / / / ( / / 3 3 / / 1 1 / 8 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) 1.2.6. Corollary. Let X be an algebraic K-scheme along with a fixed embedding σ : K ֒→ C. The period isomorphism p,q X above induces a commutative square H p,q(X) rp,q dR H p dR(X) rp,q an ιp,q dR H p(Xan, Zan(q)) ιp,q an / H p(Xan, C). Note that from Corollary 1.2.6 we get a refinement of the Betti regulator. 1.2.7. Definition. Define the algebraic singular cohomology classes as the elements of the subgroup H p,q an ⊆ H p(Xan, Zan(q)) given by the image of the motivic cohomol- ogy under the Betti regulator rp,q an . alg (X) := Im rp,q Define the -algebraic singular cohomology classes by the subgroup H p,q (X) := H p dR(X) ∩ H p(Xan, Zan(q)) ⊆ H p(Xan, Zan(q)) where ∩ means that we take elements in H p(Xan, Zan(q)) which are given by the inverse image (under ιp,q dR) inside H p(Xan, C) via the isomorphism p,q The groups H p,q an ) of elements in H p X above. (X) shall be called period cohomology groups and dR(X) regarded (under ιp,q rp,q : H p,q(X) → H p,q (X) induced by rp,q dR and rp,q an shall be called the period regulator. We get that: 1.2.8. Corollary. H p,q alg (X) ⊆ H p,q (X). For example, all torsion cohomology classes are -algebraic: we shall see in Lemma 1.4.2 that they are also algebraic. In particular, if H p(Xan, Zan(q)) is all algebraic, i.e., the Betti regulator rp,q an is surjective, then the canonical embedding ιp,q an of singular cohomology H p(Xan, Qan(q)) in the C-vector space H p(Xan, C) factors through an embedding of H p(Xan, Qan(q)) into the K-vector space H p dR(X). If K = Q this rarely happens. For example, if p = 0 it happens only if q = 0 and in this case r0,q is always surjective (as H 0,q (X) = 0 for q 6= 0). 1.3. Period conjecture for motivic cohomology. Over K = Q it seems reasonable to make the conjecture that all -algebraic classes are algebraic, i.e., to conjecture that the period regulator rp,q is surjective. In other words we may say that the period conjecture for motivic cohomology holds for X, in degree p and twist q, if (1.6) H p,q alg (X) = H p,q (X). Over a number field we may expect that this holds rationally. If (1.6) holds we also have that H p(Xan, Zan(q)) modulo torsion embeds into H p dR(X) if and only if H p(Xan, Zan(q)) is all algebraic. Note that using Proposition 1.2.5 we can define H p,q (M ) providing a version of the period conjecture for any object M ∈ DMeff ´et . / / / MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 9 1.3.1. Proposition. For any q ≥ 0 the period conjecture (1.6) holds true for X, in degree p and twist r, if and only if it holds true for M (X)(q), in degree p and twist q + r. Proof. By Voevodsky cancellation theorem [51] we have that twisting by q in motivic co- ≃ −→ H p,q+r(M (X)(q)) is an isomorphism of groups. If M = M (X)(q) homology H p,r(X) with q ≥ 0 we also get H p,r (M (X)(q)) canonically by twisting. In fact, we have a diagram induced by twisting −→ H p,q+r (X) ≃ H p(Xan, Zan(r)) ⊗ C X eωp,r H p dR(X) ⊗K C qC e qdR e H p,q+r an (M (X)(q)) ⊗Z C p,q+r M (X)(q) e / H p dR(M (X)(q)) ⊗K C ≃ where qC := q ⊗ C is the C-isomorphism given by the canonical integrally defined map- ping q : H p(Xan, Zan(r)) (M (X)(q)) which is sending a p-th cohomology class re- garded as a map βσM (X) = Sing∗(Xan) → Zan(r)[p] in D(Z) to the q-twist βσM (X)(q) = Sing∗(Xan)(q) → Zan(q + r)[p]. Similarly, the C-isomorphism qdR is induced by twisting, since Ω(−q) (cid:3) q.i. −→ Ω is a canonical isomorphism in DMeff → H p,q+r ´et and the claim follows. an For X smooth we have that H p,q(X) ∼= H p ´et(X, Z(q)) and with rational coefficients we ´et(X, Q(q)) ∼= CH q(X, 2q − p)Q. In particular, if X is smooth and p = 2q we get is the modern refinement of the classical cycle class map with rational coefficients have that H p that r2q,q (1.7) r2q,q = cℓq : CH q(X)Q → H 2q,q (X)Q for codimension q cycles on X considered in [17]. In this case, the period conjecture (1.6) with rational coefficients coincides with the classical Grothendieck period conjecture for algebraic cycles: see [17, §1.1.3] and [17, Prop. 2.13-14] comparing it with the conjecture on torsors of periods. 1.3.2. Remark. For K = C we may also think to refine the Hodge conjecture as previously hinted by Beilinson, conjecturing the surjectivity of rp,q Hodge : H p,q(X)Q → HomMHS(Q(0), H p(X)(q)). However, such a generalization doesn't hold, in general, e.g. see [20]. n ) ∼= H p ´et(X, µ⊗q ´eh(X, Z/n(q)) ∼= H p 1.4. Torsion cohomology classes are algebraic. Consider Z/n(q) := Z(q) ⊗ Z/n. By Suslin-Voevodsky rigidity we have a quasi-isomorphism of complexes of étale sheaves µ⊗q n → Z/n(q) yielding H p n ). For a proof of this key result see [47, Thm. 10.2 & Prop. 10.7] for X smooth and make use of [12, Prop. 12.1.1] to get it in general. 1.4.1. Lemma. For any algebraic scheme X over K = K ֒→ C we have H p H p(Xan, Z/n). Proof. As étale cohomology of µ⊗q is invariant under the extension σ : K ֒→ C of al- n gebraically closed fields we obtain the claimed comparison from the classical comparison result after choosing a root of unity. (cid:3) ´eh(X, Z/n(q)) ∼= ´eh(X, µ⊗q / / / 10 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) We then have (cf. [49, Prop. 3.1]): 1.4.2. Lemma. The regulator rp,q rp,q ⊗ Q/Z : H p,q(X) ⊗ Q/Z ֒→ H p,q (X) ⊗ Q/Z is injective. tor: H p,q(X)tor→→H p,q (X)tor is surjective on torsion and Proof. By construction, for any positive integer n, comparing the usual universal coefficient exact sequences, we have the following commutative diagram with exact rows 0 → H p ´eh(X, Z(q))/n H p ´eh(X, Z/n(q)) nH p+1 ´eh (X, Z(q)) → 0 rp,q an /n 1.4.1 nrp+1,q an 0 → H p(Xan, Zan)/n / H p(Xan, Z/n) / nH p+1(Xan, Zan) → 0. Passing to the direct limit on n we easily get the claim. In fact, nH p,q and rp,q an /n factors through rp,q /n. (X) = nH p(Xan, Zan) (cid:3) 1.4.3. Lemma. We have that rp,q if this is the case rp,q ⊗ Q/Z is an isomorphism. ⊗ Q is surjective if and only if rp,q is surjective; moreover, Proof. This follows from a simple diagram chase. (cid:3) In the situation that H p,q(X) ⊗ Q/Z = 0 the period conjecture for motivic cohomology (1.6) is then equivalent to (1.8) In particular: H p dR(X) ∩ H p(Xan, Qan(q)) = 0. 1.4.4. Proposition. If X is smooth then (1.6) for p /∈ [q, 2q] is equivalent to (1.8). If X is smooth and proper then (1.6) is equivalent to the surjectivity of cℓq in (1.7) for p = 2q and to the vanishing (1.8) for p 6= 2q. Proof. In fact, by the Appendix A, Theorem A.1.3, we have that for p /∈ [q, 2q] the group H p,q(X) is an extension of torsion by divisible groups so that H p,q(X) ⊗ Q/Z = 0. If X is proper the latter vanishing holds true for all p 6= 2q. (cid:3) Proposition 1.4.4 explains some weight properties related to the Grothendieck period conjecture, weight arguments which are also considered in [17]. 1.4.5. Remark. For K = C we have that rp,q tor: H p,q(X)tor→→H p(Xan, Zan(q))tor is surjec- β tive (as also remarked in [49] for X smooth projective): torsion motivic cohomology classes supply the defect of algebraic cycles providing the missing torsion algebraic cycles. In fact, from the well known Atiyah-Hirzebruch-Totaro counterexamples to the integral Hodge con- jecture we know that cℓp : CH p(X) → H 2p(Xan, Zan(p)) cannot be surjective on torsion for p ≥ 2 in general. / / / / / / MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 11 2. Periods of 1-motives: fullness of Betti-de Rham realizations Let tM1(K) be the abelian category of 1-motives with torsion over K (see [12, App. C]). We shall drop the reference to K if it is clear from the context. We shall denote MK = [uK : LK → GK] ∈ tM1(K) a 1-motive with torsion with LK in degree 0 and GK in degree 1; for brevity, we shall write MK = LK [0] if GK = 0 and MK = GK[−1] if LK = 0 and we omit the reference to K if unnecessary. Let Mtor := [Ltor ∩ Ker(u) → 0] be the torsion part of MK, let Mfr := [L/Ltor → G/u(Ltor)] be the free part of MK, and let Mtf := [L/Ltor ∩ Ker(u) → G] be the torsion free part of MK. There are short exact sequences of complexes (2.1) and (2.2) 0 → Mtor → MK → Mtf → 0 0 → [F = F ] → Mtf → Mfr → 0, where F = Ltor/Ltor ∩ Ker(u). Let Mab denote the 1-motive with torsion [L → G/T] where T is the maximal subtorus of G. Recall (see [12, Prop. C.7.1]) that the canonical functor M1 → tM1 from Deligne 1-motives admits a left adjoint/left inverse given by M Mfr. Any 1-motive M = [L → G] is canonically endowed with an increasing filtration of sub-1- motives, the weight filtration, defined as follows: with T the maximal subtorus of G. We have that Db(tM1) ∼= Db(M1) (see [12, Thm. 1.11.1]) and that there is a canonical embedding (see [12, Def 2.7.1]) (2.4) Tot : Db(M1) ֒→ DMeff ´et so that we can also regard 1-motives as motivic complexes of étale sheaves. The restriction of the Betti realization βσ in (1.3) can be described explicitly for 1-motives via Deligne's Hodge realization (see [12, Thm. 15.4.1]). Similarly, the restriction of the de Rham realization in [45] can be described via Deligne's de Rham realization as follows. 2.1. de Rham realization. Let K be a field of characteristic zero and let MK = [uK : LK → K : LK → G♮ GK] ∈ tM1(K) be a 1-motive with torsion over K. Note that for M♮ K] the universal Ga-extension of MK we have K := [u♮ 0 → V(M) → M♮ K ρM −→ MK → 0 where V(M) := Ext(MK , Ga)∨. The existence of universal extensions is well-known when LK is torsion-free; for the general case see [11, Proposition 2.2.1]. Recall (see [23, §10.1.7]) the following (2.3) Wi(M) = G[−1] T[−1] M if if if if 0 i ≥ 0 i = −1 i = −2 i ≤ −3 12 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) 2.1.1. Definition. The de Rham realization of MK is TdR(MK) := Lie(G♮ K) as a K-vector space. 2.1.2. Remark. Note that ρM = (idL, ρG) where ρG : G♮ V(M) so that GK is the semiabelian quotient of G♮ uK = ρG ◦ u♮ K and u♮ K→→GK is a quotient and Ker ρG = K is a canonical lifting of uK, i.e., K. Further V(M) ⊆ TdR(MK) is also the kernel of the morphism dρG : Lie(G♮ K ) → Lie(GK) induced by ρG, so that TdR(MK) together with the K-subspace V(M) can be regarded as a filtered K-vector space. This datum is called the Hodge filtration of TdR(MK). The algebraic K-group G♮ K fits in the following diagram [14, (2.15)] 0 0 / V(A) A♮ ×A G i / V(M) ρG G♮ G G / 0 / 0 V(L) L ⊗ Ga where we have omitted subscripts K and written V(A)) for V(A[−1]) = V(G[−1]). 2.1.3. Lemma. For K ⊂ K ′ we have a natural isomorphism (M♮ K )K ′ ∼= (MK ′)♮. 2.2. Base change to C and periods. Consider K a subfield of C and let MC = [uC : LC → GC] be the base change of MK to C. Let TZ(MC) be the finitely generated abelian group in the usual Deligne-Hodge realization of MC (see [23, 10.1.3] and [13, §1]) given by the pull-back 0 0 / H1(GC) / Lie(GC) exp / GC euC / TZ(GC) / TZ(MC) gexp uC LC / 0 0 where for brevity TZ(GC) denotes TZ(GC[−1]) which by definition is H1(GC). After base change to C and Lemma 2.1.3 we then get (M♮ K )C ∼= (MC)♮ hence an isomorphism (2.5) ι : TdR(MC) ≃ −→ TdR(MK) ⊗K C / / / _ / / _ / / / / / / / / / / / / / / / O O / / O O MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 13 and a commutative diagram (2.6) 0 0 0 / H1(GC) / Lie(GC) exp / GC / 0 dρG ρG / H1(G♮ C) / Lie(G♮ C) exp G♮ C euC / TZ(GC) / TZ(MC) u♮ C LC gexp uC 0 0 where the dotted arrow exists by definition of TZ(MC) and the fact that the upper right-hand square is cartesian. Hence also the lower right-hand square is cartesian and the sequence on the bottom is equivalently obtained by pull-back of the upper sequence via uC or of the sequence in the middle via u♮ C. Further TZ(GC) is identified with the kernel of both exponential maps and the dotted arrow gives a homomorphism TZ(MC) → TdR(MC) := Lie(G♮ C). Finally note that the weight filtration in (2.3) gives a filtration on TZ(MC): the immersion TC → GC gives an inclusion TZ(TC) = H1(TC) ⊆ H1(GC) = TZ(GC) while TZ(GC) ⊆ TZ(MC) comes from the previous diagram. 2.2.1. Definition. The homomorphism of periods is the unique homomorphism M,Z : TZ(MC) → TdR(MK) ⊗K C isomorphism ι in (2.5). that yields dρG ◦ M,Z =euC and exp ◦M,Z = u♮ C ◦gexp under the identification given by the Note thateuC is the pull-back of uC along exp and for x ∈ LC we may pick flog(x) ∈ TZ(MC), i.e., such that gexp(flog(x)) = x. We then get uC(x) = exp(euC(flog(x))) = exp(dρG(M,Z(flog(x)))). 2.2.2. Theorem. The induced C-linear mapping (2.7) M,C : TC(MC) := TZ(MC) ⊗Z C → TdR(MK ) ⊗K C is an isomorphism. Proof. Making use of the identification in (2.5) we are left to see that it holds true for K = C. The case of L without torsion is treated by Deligne [23, 10.1.8]. Actually, an easy proof can be given by dévissage to the case of lattices, tori and abelian varieties. For the general case note that M,C = Mtf ,C by (2.1). Indeed TdR(Mtor) = 0 and the kernel of the canonical morphism TZ(MC) → TZ(Mtf,C) is torsion. Further by (2.2) the map TZ(Mtf,C) → TZ(Mfr,C) is an isomorphism and we have an exact sequence 0 → [F = F ] → M♮ tf → M♮ fr → 0 so that the canonical morphism TdR(Mtf ) → TdR(Mfr) is an isomorphism too. Hence Mtf ,C = Mfr,C. We conclude that M,C = Mfr,C and the latter is an isomorphism since Mfr is a Deligne 1-motive. (cid:3) / / / / / / / / O O / / O O / / < < / / O O / / O O ] ] 14 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) 2.2.3. Examples. • If MK = [0 → Gm], then TZ(MC) = Z and the first and second rows in (2.6) are given by 0 → Z M,C : C → C, z 7→ 2πiz. → C∗ → 0. Hence M,Z =euC : Z → C, x 7→ 2πix and • If MK = [LK → 0], then TZ(MC) = LC and TdR(MC) = LC ⊗Z C and the map M,Z is 2πi → C exp the homomorphism LC → LC ⊗Z C, x 7→ x ⊗ 1 and M,C is the identity map. • If MK = [u : Z → Gm] with u(1) = a ∈ K ∗, then G♮ = Gm × Ga. Once fixed a complex logarithm log a of a we can construct an isomorphism Z × Z → TZ(MC) = {(z, y) ∈ C × Z exp(z) = ay} that maps the pair (k, y) to (y log a + 2πik, y). The homomorphism of periods becomes then the map M,Z : Z × Z → C × C that sends (k, y) to (y log a + 2πik, y). Note that over C the Hodge filtration V(M) ⊆ TdR(MK) of Remark 2.1.2 is obtained from the Hodge filtration of TC(MC) via M,C. 2.3. Periods and transcendence. The proof of Theorem 2.7.1, which is the main outcome of the second section of this paper, makes use of deep results of transcendence theory that we recall below. First consider [43, Theorem 2]. 2.3.1. Theorem. Let AK be an abelian variety of dimension d over K = Q. Let Θ : Cd → Aan be the homomorphism given by the theta functions, inducing an isomorphism of the complex torus onto Aan. Assume that the derivations ∂/∂zi, (i = 1, . . . , d) are defined over If α = (αi) ∈ Cd is a complex vector 6= 0 such that all αi lie in K, then Θ(α) is K. transcendental over K. In particular, the periods are transcendental. It can be generalized to semiabelian varieties as follows. 2.3.2. Theorem. Let GK be a semiabelian variety over K = Q. If 0 6= x ∈ Lie(GK), then exp(x) ∈ GC(C) is transcendental over K. In particular, Lie(GK) ∩ Ker(exp) = {0} in Lie(GC). Proof. The assertion is known if GK = Gd m,K due to fundamental work of Hermite and Lindemann on the transcendence of eβ for β a non zero algebraic number. The case GK an abelian variety is Theorem 2.3.1. The general case follows then by dévissage. (cid:3) This type of results has been further generalised by Waldschmidt ([54, Thm. 5.2.1]): 2.3.3. Theorem. Let G be a commutative connected algebraic group over K = Q. Let ϕ : Cn → Gan be an analytic map such that the induced morphism on Lie algebras arises from a homomorphism of K-vector spaces K n → Lie G. Let Γ ⊂ Cn be a subgroup containing n elements which are C-linearly independent and such that ϕ(Γ) ⊂ G(K). Then the algebraic dimension of ϕ, i.e. the dimension of the Zariski closure of the image of ϕ, is ≤ n. This implies the following result [17, Thm. 3.1], which is one of the technical inputs for the proof of our Theorem 2.7.1 2.3.4. Theorem. Let GK, HK be two connected commutative algebraic groups over K = Q. If the group H1(GC) generates Lie(GC) as a complex vector space, then the map Lie : HomK-gr(GK , HK ) → {ψ ∈ HomK(Lie GK, Lie HK )ψC(H1(GC)) ⊆ H1(HC)} is an isomorphism of Z-modules. MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 15 The condition on the group H1(GC) is satisfied whenever GC is a semiabelian variety (or its universal vectorial extension) since TZ(GC) = H1(GC) generates TdR(GC) = Lie(G♮ C) and hence Lie(GC). Note that Theorem 2.2.2 is a generalization of this fact. For completeness we also cite the following result [56, Thm 1], which is a consequence of the celebrated analytic subgroup theorem of Wüstholz. It implies the result of Waldschmidt is a special case. We will use it to give an alternative proof of Theorem 2.7.1: 2.3.5. Theorem. Let W be a commutative connected algebraic group over Q. Let S be a subset of exp−1(cid:0)W (Q)(cid:1) and let V ⊂ Lie W be the smallest Q-vector subspace whose C-span contains S. Then, there exists a connected algebraic subgroup Z ⊂ W such that Lie Z = V . 2.4. Period categories. For a fixed σ : K ֒→ C we consider a homological category for Betti-de Rham realizations as follows. Let ModZ,K be the following category: (i) objects are triples (HZ, HK, ω) where HZ is a finitely generated abelian group, HK is a finite dimensional K-vector space, and ω : HZ → HK ⊗K C is a homomorphism of groups; (ii) morphisms ϕ : (HZ, HK, ω) → (H ′ Z is a group homomorphism, ϕK : HK → H ′ K is a K-linear homomorphism and ϕ is compatible with ω and ω′, i.e., the following square K , ω′) are pairs ϕ := (ϕZ, ϕK ) where ϕZ : HZ → H ′ Z, H ′ (2.8) HZ ϕZ H ′ Z ω ω′ HK ⊗K C ϕK ⊗1C / H ′ K ⊗K C commutes. For H = (HZ, HK, ω) in ModZ,K let (2.9) ωC : HZ ⊗Z C → HK ⊗K C be the induced C-linear mapping and denote Mod by those objects such that ωC is a C-isomorphism. ∼= Z,K the full subcategory of ModZ,K given There is a Q-linear variant ModQ,K of this category where objects are (HQ, HK, ω) as ∼= ModZ,K ⊗ Q is above but HQ is a finite dimensional Q-vector space. Note that ModQ,K the category ModZ,K modulo torsion objects (see [12, B.3] for this notion). 2.4.1. Definition. We shall call Mod category) of homological periods. ∼= Z,K (resp. Mod ∼= Q,K) the category (resp. Q-linear Let Mod ∼=,fr Z,K (resp. Mod ∼=,tor Z,K ) be the full subcategory of Mod ∼= Z,K given by those objects H such that HZ is free (resp. is torsion). For any r ∈ Z we shall denote For H = (HZ, HK, ω) and H ′ = (H ′ (2.10) H ⊗ H ′ := (HZ ⊗Z H ′ Z(r) := (Z, K, (2πi)r ) ∈ Mod K , ω′) we can define Z, H ′ ∼=,fr Z,K . and set H(r) := H ⊗ Z(r) the Tate twist. For H ∈ Mod HZ free, we have duals H ∨ ∈ Mod ∼=,fr Z,K given by (2.11) (HZ, HK , ω)∨ := (H ∨ Z , H ∨ K, ω∨) K, ω ⊗ ω′) Z, HK ⊗K H ′ ∼=,fr Z,K , say that H = (HZ, HK, ω) with / / / 16 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) where H ∨ vector space, and Z = Hom(HZ, Z) is the dual abelian group, H ∨ K = Hom(HK , K) is the dual K- ω∨ : H ∨ ≃ −→ H ∨ Z → H ∨ is the composition of the canonical mapping H ∨ H ∨ Z ⊗Z C idC) ◦ ω−1 We clearly get that (H ∨)∨ = H and ( )∨ : Mod that Z(r)∨ = Z(−r) so that H(r)∨ = H ∨(−r) for r ∈ Z. Z ⊗Z C with the C-isomorphism K ⊗K C given by the inverse of the C-dual of ωC in (2.9), i.e., ω∨(f ) = (f ⊗Z K ⊗K C ≃ (HK ⊗K C)∨. ∼=,fr Z,K is a dualizing functor. Note C for any f : HZ → Z, up to the canonical isomorphism H ∨ K ⊗K C Z → H ∨ ∼=,fr Z,K → Mod Similar constructions can be done for the Q-linear variant Mod ∼= Q,K ∼=,fr (resp. Mod Z,K ) admits an internal Hom defined via the internal Hom of the category of finite dimensional Q-vector spaces (resp. lattices). Furthermore these categories do have an ∼= Q,K, respectively. For any object identity object: 1 = Z(0) ∈ Mod ∼= Q,K are H of Mod ∼= Q,K we have H ∨ = Hom(H, 1) and End(1) = Q. Hence all objects of Mod ∼=,fr Z,K and 1 = Q(0) ∈ Mod ∼= Q,K. Note that Mod reflexive. Similarly, for Mod ∼=,fr Z,K . 2.4.2. Lemma. The categories ModZ,K and Mod category Mod Q-vector spaces. ∼= Z,K are abelian tensor categories. The ∼= Q,K is a neutral Tannakian category with fibre functor the forgetful functor to Note that there is a cohomological version of Mod ∼= Z,K and Mod ∼= Q,K, which is called the de Rham -- Betti category in the existing literature (cf. [3, 7.5]). 2.4.3. Definition. Let Mod HK is a finite dimensional K-vector space, HZ is a finitely generated abelian group and ∼= K,Z be the category whose objects are triples (HK, HZ, η) where is an isomorphism of C-vector spaces. We shall call Mod the categories of cohomological periods. ∼= K,Z and its Q-linear variant Mod ∼= K,Q η : HK ⊗K C ≃ −→ HZ ⊗Z C The category Mod ∼=,fr K,Z is denoted CdRB in [17, §2.1] and in [16, §5.3]. The Q-linear variant ∼= K,Q is denoted (K, Q)-Vect in [33, Chap. 5]. For these categories we have an analogue Mod of Lemma 2.4.2; in particular, a dualizing functor exists. 2.4.4. Lemma. There is canonical equivalence given by the functor ς : Mod ∼= Z,K → Mod ∼= K,Z ς(HZ, HK , ω) := (HK, HZ, ω−1 C ) which induces an equivalence between the tensor subcategories Mod ∼=,fr Z,K and Mod ∼=,fr K,Z. We set Note that, for H ∈ Mod Z(r) := ς(Z(r)) ∈ Mod ∼=,fr Z,K we may consider H ◦ ∈ Mod ∼=,fr K,Z. ∼=,fr K,Z setting (2.12) (HZ, HK, ω)◦ := (H ∨ K, H ∨ Z , ω◦) = ς(H ∨) = ς(H)∨ MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 17 where ω◦ : H ∨ Z(r)◦ = Z(−r) ∈ Mod K ⊗K C Z ⊗Z C is just given by the C-dual of ωC in (2.9). We then have ≃ −→ H ∨ ∼=,fr K,Z so that H(r)◦ = H ◦(−r) for all r ∈ Z. The functor ( )◦ is an anti-equivalence and there is an induced equivalence Mod (Mod ∼= K,Q)op of neutral Tannakian categories. ∼= Q,K ∼= 2.5. Betti -- de Rham realization and Cartier duality. Now recall the period mapping M,Z : TZ(MC) → TdR(MK) ⊗K C provided by Definition 2.2.1. According to Theorem 2.2.2 we have that M,C is a C-isomorphism. 2.5.1. Definition. For K a subfield of C, MK ∈ tM1(K) and M,Z we set TBdR(MK) := (TZ(MC), TdR(MK), M,Z) ∈ Mod ∼= Z,K and the Q-linear variant T Q BdR(MK) := (TQ(MC), TdR(MK), M,Q) ∈ Mod ∼= Q,K where TQ(MC) := TZ(MC) ⊗Z Q. Call these realizations the Betti -- de Rham realizations. Since the period mapping M,Z in TBdR(MK) is covariantly functorial, by the construc- tions in (2.6) and (2.5), the Betti -- de Rham realization yields a functor (2.13) TBdR : tM1(K) → Mod ∼= Z,K. Similarly, with rational coefficients, we get a functor in the homological category Mod ∼= 1 (K) ∼= tMQ from 1-motives up to isogenies MQ Q,K. By Examples 2.2.3 we have TBdR(Z[0]) = Z(0) and TBdR(Gm[−1]) = Z(1). 1 (K) to Mod ∼= Z,K 2.5.2. Definition. For H = (HZ, HK , ω) ∈ Mod ∼=,fr Z,K define the Cartier dual H ∗ := (H ∨ Z , H ∨ K, 2πiω∨) = H ∨(1) = H(−1)∨ = Hom(H, Z(1)) ∈ Mod ∼=,fr Z,K. Note that this construction is reflexive. 2.5.3. Theorem. For MK ∈ tM1(K) free with Cartier dual M∗ K we have that TBdR(MK)∗ ∼= TBdR(M∗ K) Proof. It suffices to prove that the Poincaré biextension of MK provides a natural morphism T (MK) ⊗ T (M∗ K) → Z(1) which induces the usual dualities h , iZ on TZ's and h , idR on TdR's constructed in [23, §10.2.3 & §10.2.7]. This is proved in [23, Prop. 10.2.8]. (cid:3) Note that we also have a de Rham -- Betti contravariant realization in the cohomological ∼= K,Z. Recall from [12, §1.13] that we also have the category of 1-motives with category Mod cotorsion tM1. Cartier duality (2.14) ( )∗ : tM1 ≃ −→ tM1 is an anti-equivalence of abelian categories. 18 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) 2.5.4. Definition. For M ∈ tM1 denote TdRB(M) := ς(TBdR(M∗)) = (TdR(M∗), TZ(M∗), ηM∗ ) ∈ Mod ∼= K,Z where ηM∗ := −1 M∗,C is the inverse of the C-linear period isomorphism M∗,C of the Cartier dual M∗ ∈ tM1 (see Theorem 2.2.2). Call this realization (and its Q-linear variant) the de Rham -- Betti realization. With this definition we get a functor (2.15) TdRB : tMop 1 → Mod ∼= K,Z. Now we have TdRB(Z[0]) = Z(1) and TdRB(Gm[−1]) = Z(0). With the notation adopted in (2.12), we also have TBdR(M)◦(1) = (TdR(M)∨, TZ(M)∨, (2πi)−1◦ M). 2.5.5. Lemma. We have a natural isomorphism of functors TdRB( ) ∼= TBdR( )◦(1). Proof. For M ∈ M1 and its Cartier dual M∗ we have that TBdR(M)∗ ∼= TBdR(M∗) ∈ Mod by Theorem 2.5.3. Thus the period isomorphism of the Cartier dual M∗,Z = 2πi∨ its C-inverse −1 ∼=,fr Z,K M,Z and (cid:3) M∗,C = (2πi)−1◦ M. 2.6. Weight and Hodge filtrations. Consider the category FModZ,K given by objects in ModZ,K endowed with finite and exhaustive filtrations and morphisms that respect the filtrations. More precisely, an object of FModZ,K is an abelian group HZ endowed with a (weight) filtration W•HZ and a K-vector space HK endowed with two filtrations W•HK, F•HK, along with the corresponding compatibilities of the ω's on weight filtrations. Let MK = [LK → GK] be a 1-motive over K and, as usual, let TK denote the maximal subtorus of GK. Since the Betti-de Rham realization (2.13) is functorial and compatible with the canonical weight filtration (2.3) on TZ(MC) and TdR(MK) is filtered by V(M), the Hodge filtration as in Remark 2.1.2, we also get a realization functor (2.16) We have F TBdR : tM1(K) → FModZ,K. (TZ(MC), TdR(MK ), M,Z) ⊇ (TZ(GC), TdR(GK), G,Z) ⊇ (TZ(TC), TdR(TK ), T,Z). Note that: 2.6.1. Lemma. Let K = Q. Let MK and NK be two free 1-motives over K. Then any morphism ϕ : TBdR(MK ) → TBdR(NK ) in Mod ∼=,fr Z,K preserves the weight filtrations. Proof. Let ϕ = (ϕZ, ϕK ) : (TZ(MC), TdR(MK), M,Z) → (TZ(NC), TdR(NK ), N,Z) for MK and NK one of the following pure 1-motives: [ZK → 0], [0 → Gm,K] and [0 → AK], where AK is an abelian variety. We show that ϕ = 0 for different weights, in all cases. As K is algebraically closed this implies that ϕ = 0 for all pure 1-motives of different weights and this easily yields the claimed compatibility. For MK = [ZK → 0] and NK = [0 → Gm,K ] (respectively MK = [0 → Gm,K] and NK = [ZK → 0]) we have TBdR([ZK → 0]) = Z(0), TBdR([0 → Gm,K]) = Z(1) and ϕ = 0 as MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 19 ϕK : K → K is given by the multiplication by an algebraic number but the compatibility (2.8) forces such algebraic number to be n2πi (respectively n/2πi) for some n ∈ Z. Similarly, for MK = [ZK → 0] and NK = [0 → AK] we have that Z,N ◦ ϕZ(1) = ϕK (1) if, and only if, ϕ = 0. Indeed, the preceding equality implies that dρA ◦ Z,N ◦ ϕZ(1) = dρA ◦ ϕK (1). Now, the right-hand term is in Lie(AK) while by Remark 2.3.2 the left-hand term would give a transcendental point of Lie(AC) if ϕZ(1) 6= 0. Dually, for MK = [0 → AK] and NK = [0 → Gm,K] by making use of Theorem 2.5.3 we then get ϕ∗ = 0 thus ϕ = 0. Finally, for MK = [0 → Gm,K] and NK = [0 → AK] we can apply Theorem 2.3.4 to the pair (ϕK , ϕZ) so that, dually, making use of Theorem 2.5.3, the same holds for MK = [0 → AK] and NK = [ZK → 0]. (cid:3) Let MK = [uK : LK → GK] and NK = [vK : FK → HK] be free and let ϕ : TBdR(MK) → ∼= Z,K . Then we have a K-linear mapping ϕK : TdR(MK) → TBdR(NK ) be a morphism in Mod TdR(NK ) and a homomorphism ϕZ : TZ(MC) → TZ(NC) which is compatible with the weight filtrations, by Lemma 2.6.1. Moreover, ϕZ and ϕK are compatible with the 's as in (2.8). We have that ϕZ restricts to a homomorphism (2.17) W−1ϕZ : W−1TZ(MC) := TZ(GC) ∼= H1(G♮ and we get an induced map on grW C) → W−1TZ(NC) := TZ(HC) ∼= H1(H♮ C) 0 as follows (2.18) ϕZ,0 : grW 0 TZ(MC) = TZ(MC)/TZ(GC) = LC → grW 0 TZ(NC) = TZ(NC)/TZ(HC) = FC. Note that ϕZ,0 is indeed defined over Q. 2.6.2. Lemma. Let K = Q. Let MK and NK be two free 1-motives over K. Then any morphism ϕ : TBdR(MK ) → TBdR(NK ) in Mod ∼=,fr Z,K preserves the Hodge filtrations. Proof. Let MK = [uK : LK → GK] and NK = [vK : FK → HK ] be free and let ϕ : TBdR(MK ) → ∼= Z,K. We have to show that ϕK (V(M)) ⊆ V(N) where V(M) TBdR(NK ) be a morphism in Mod is the additive part of G♮ K ; see Remark 2.1.2. Recall the commutative diagram K and V(N) is that of H♮ (2.19) TZ(MC) ϕZ / TZ(NC) M,Z N,Z TdR(MC) ϕK ⊗idC / TdR(NC). By definition of M,Z and (2.17) there is then a commutative diagram H1(G♮ C) / H1(H♮ C) W−1ϕZ Lie(G♮ C) / Lie(H♮ C) ϕK ⊗idC / / / / 20 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) where the vertical arrows are those in the horizontal sequence in the middle of diagram (2.6) for M and N respectively. Hence there exists an analytic morphism hC : G♮ C → H♮ C with dhC = ϕK ⊗ idC. It is sufficient to prove that hC is algebraic and defined over K to conclude by the structure theorem of algebraic K-groups that hK(V(M)) ⊆ V(N) and hence that ϕK = dhK preserves the Hodge filtrations. If LK = 0, by Lemma 2.6.1, ϕ factors through W−1TBdR(NK ) = TBdR(HK ). Hence we may assume FK = 0 as well. It follows then from Theorem 2.3.4 applied to G♮ and H♮ that the above morphism hC is indeed algebraic and defined over K. Hence ϕK preserves the Hodge filtrations. Now let LK 6= 0 and set L♮ K = LK ⊗Ga,K. Since ϕ preserves the weights by Lemma 2.6.1 we get W−1ϕ : TBdR(GK) → TBdR(HK ). By the previous step W−1ϕK(V(G)) ⊆ V(H) ⊆ V(N). We thus obtain the following commutative diagram M,Z TZ(MC) / TdR(M)/V(G) ⊗ C exp G♮ C/V(G)C GC ⊕ L♮ C ϕZ γ N,Z TZ(NC) / TdR(N)/V(N) ⊗ C exp / H♮ C/V(N)C δ HC where the mapping γ is induced by ϕK , we have the canonical identification of G♮ GK ⊕ L♮ are left to show that β : L♮ previous diagram maps TZ(GC) to 0 ⊕ 0, we obtain a commutative square K/V(G) = K and δ = gC + β with gC = gK ⊗ idC and gK : GK → HK induced by W−1ϕK . We C → HC is zero. Since the composition of the upper arrows in the LC (u,1) / GC ⊕ L♮ C ϕZ,0 v FC δ / HC where ϕZ,0 is the induced map as in (2.18). In particular, for x ∈ LK (K) we have β(x ⊗ 1) = v(ϕZ,0(x)) − gK (u(x)) = γ − dgK ⊗ idC is in HK(K). On the other hand β(x ⊗ 1) = exp dβ(x ⊗ 1). Since dβ = dδ − dgK ⊗ idC we have that dβ(x ⊗ 1) belongs to Lie(HK ) regarded as a K-linear subspace of Lie(HC). By Remark 2.3.2 we get that β(x ⊗ 1) = 0 and therefore that β = 0. (cid:3) 2.7. Full faithfulness. We are now ready to show that our previous Lemmas 2.6.1 and 2.6.2 yield the full faithfulness of Betti -- de Rham and de Rham -- Betti realizations. 2.7.1. Theorem. The functors TBdR in (2.13) and TdRB in (2.15) restricted to M1(K) are fully faithful over K = Q. Proof. Clearly, the functor TBdR (resp. TdRB) is faithful (cf. [4, proof of Lemma 3.3.2]) and we are left to show the fullness. Making use of Lemma 2.5.5 we are left to check the fullness for TBdR. Let MK = [uK : LK → GK] and NK = [vK : FK → HK ] be free and let ϕ : TBdR(MK) → TBdR(NK ) be a morphism in Mod ∼= Z,K. / / / / / / / MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 21 For 0-motives, i.e., if GK = HK = 0, we have LK ∼= Zr K and FK ∼= Zs K, ϕZ : TZ(MC) ∼= Zr → TZ(NC) ∼= Zs K provides a morphism f : MK → NK such that TBdR(f ) = ϕ. If the weight −1 parts are non-zero, by Lemma 2.6.1 ϕZ restricts to a homomorphism W−1ϕZ as in (2.17) and it yields a morphism ϕZ,0 as in (2.18) i.e., ϕZ,0 is the map induced by ϕZ on grW 0 . If we set fC := ϕZ,0 the homomorphism fC : LC → FC trivially descends to a homomorphism fK : LK → FK over K = Q. Let's now consider ϕC := ϕK ⊗K idC and translate (2.17) and (2.19), as in the proof of Lemma 2.6.2, in the following commutative diagram with exact rows (2.20) 0 0 / H1(G♮ C) Lie(G♮ C) exp W−1ϕZ ϕC G♮ C ψ / H1(H♮ C) exp / H♮ C / Lie(H♮ C) C → H♮ 0 / 0 yielding a morphism of analytic groups ψ : G♮ C on the quotients via the exponential mapping exp, as indicated above. Now, since by Lemma 2.6.2, we have ϕK (V(M)) ⊆ V(N), ψ(V(MC)) ⊆ V(NC), the diagram (2.20) induces a commutative diagram 0 0 / H1(GC) exp Lie(GC) W−1ϕZ ϕ′ C 0 GC ψ′ / H1(HC) / Lie(HC) exp / HC / 0. As ϕ′ C is the base change of the K-linear map Lie(GK) → Lie(HK ) induced by ϕK , it follows from Theorem 2.3.4 that ψ′ = gC is the base change of the morphism gK : GK → HK over K = Q induced by W−1ϕK (see the proof of Lemma 2.6.2). We are left to check that h := (fK , gK ) gives a morphism h : MK → NK , i.e., that gK ◦uK = vK ◦ fK, and to see that TBdR(h) = ϕ. To show that h is a morphism of 1-motives we may work after base change to C and, using (2.6), it suffices to prove that ψ ◦ u♮ C ◦ fC. Consider the following diagram C = v♮ TZ(MC) M,Z gexp &▼▼▼▼▼▼▼▼▼▼▼▼ ϕZ LC fC TZ(NC) N,Z gexp '◆◆◆◆◆◆◆◆◆◆◆◆ / FC u♮ C v♮ C TdR(MK ) ⊗K C ϕC &▲▲▲▲▲▲▲▲▲▲▲▲ exp TdR(NK ) ⊗K C &▼▼▼▼▼▼▼▼▼▼▼ exp / H♮ C . G♮ C ψ All squares are commutative. Indeed, exp ◦M,Z = u♮ C ◦gexp by (2.6), fC ◦gexp =gexp ◦ ϕZ by definition of fC, ϕC ◦ M,Z = N,Z ◦ ϕZ by the compatibility C ◦gexp and exp ◦N,Z = v♮ / / / / / / / / / / / / / / / / / / / / / / / / & ' / / / & & / 22 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) of ϕZ with ϕK as in (2.8), and finally ψ ◦ exp = exp ◦ϕC by (2.20). One concludes by the surjectivity of the map gexp that also ψ ◦ u♮ Now consider the morphism α := TBdR(h) − ϕ : TBdR(MK) → TBdR(NK ) in Mod ∼= Z,K. By construction W−1ϕZ = W−1TZ(h) so that α is vanishing on TBdR(GK). Moreover we have ∼= that grW Z,K from TBdR(LK ) to TBdR(HK ) which is trivial by Lemma 2.6.1. (cid:3) 0 TZ(h) = ϕZ,0 so that α induces a morphism in Mod C = v♮ C ◦ fC. 2.7.2. Remark. In the proof of Theorem 2.7.1, in order to show that (fK , gK ) : MK → NK is a morphism we are left to check that gK ◦ uK = vK ◦ fK. Remark that this also follows mappings 's and (ii) the mappings ϕZ and ϕK are compatible with the 's. from two key facts: (i) the pullbacks euC and evC of uC and vC factor through the period In fact, according to the above notation, for x ∈ LC pick flog(x) ∈ TZ(MC) and note that ϕZ(flog(x)) = flog(f (x)). Making use of (2.7) we obtain gC(uC(x)) = gC exp dρGM,Z(flog(x)) = exp dρHdg♮ by the functoriality of exp. Now ψ = ϕK ⊗K 1C and we are assuming the compatibility (ϕK ⊗K 1C) ◦ M,Z = N,Z ◦ ϕZ so that CM,Z(flog(x)) gC(uC(x)) = exp dρHN,ZϕZ(flog(x)) = exp dρHN,Z(flog(f (x))) = vC(f (x)) using (2.7) again, as claimed. We notice that an alternative proof of Theorem 2.7.1 can be given using Wüstholz's an- alytic subgroup Theorem 2.3.5 as follows: Alternative Proof of Theorem 2.7.1. Let MK = [uK : LK → GK] and NK = [vK : FK → HK ] be free 1-motives and let ϕ = (ϕZ, ϕK ) : TBdR(MK) → TBdR(NK ) be a morphism in Mod K and note that we have commutative squares ∼= Z,K. Let W = G♮ K × H♮ T := TZ(MC) (id,ϕZ) TZ(MC) × TZ(NC) M,Z×N,Z LC (id,ϕZ,0) / LC × FC u♮×v♮ / Lie WC = Lie G♮ C ⊕ Lie H♮ C exp / WC := G♮ C × H♮ C where the horizontal arrows are injective. Let S denote the image of T in Lie WC; it is contained in exp−1(cid:0)W (Q)(cid:1) since the image of LC × FC via u♮ × v♮ is contained in W (Q). Let V denote the image of Lie G♮ K in Lie W via the map id ⊕ ϕK . By the compatibility of ϕZ and ϕK over C via the homomorphisms of periods, VC coincides with the C-span of S. It then follows from Theorem 2.3.5 that there exists an algebraic subgroup Z ⊂ W whose Lie algebra is V . Now, the composition of the inclusion Z → W with the projection W → G♮ K is an isogeny, since it is an isomorphism on Lie algebras. In fact, it is an isomorphism; indeed the injective map T → VC = Lie ZC ⊂ Lie WC maps H1(G♮ C) ⊂ T into H1(ZC) ⊂ H1(WC) and hence the isomorphism Lie ZC C). Let γ : G♮ K be the homomorphism of algebraic K-groups defined by composing the C restricts to an isomorphism H1(ZC) −→ H1(G♮ −→ Lie G♮ K → H♮ ∼ ∼ / / / / / MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 23 inverse of the isomorphism Z → G♮ W → H♮ K . By construction Lie γ = ϕK . K with the inclusion Z → W and the second projection In order to see that f := (ϕZ,0, γ) is a morphism of 1-motives with TBdR(f ) = ϕ it suffices to check that γC ◦ u♮ C. The latter fact is equivalent to the equality (idG♮, γ) ◦ u♮ = (u♮, v♮ ◦ ϕZ,0) as morphisms LC → WC, and, by the above diagram, this is satisfied whenever (idLie G♮, Lie γC) ◦ M,Z = (M,Z, N,Z ◦ ϕZ) : T → Lie WC. Then we conclude by the commutativity of diagram (2.8) since Lie γC = ϕK ⊗ idC. (cid:3) C ◦ϕZ,0 as morphisms LC → H ♮ C = v♮ 2.8. Descent to number fields. Let K ′/K be a field extension with K ′ ⊆ Q. Note the following commutative diagram of functors (2.21) M1(K) TBdR / Mod ∼=,fr Z,K M1(K ′) TBdR / / Mod ∼=,fr Z,K ′ where the functor on the left is the usual base-change and the vertical functor on the right maps (HZ, HK, ω) to (HZ, HK ⊗K K ′, ω) using the canonical isomorphism (HK ⊗K K ′) ⊗K ′ C ≃ HK ⊗K C. 2.8.1. Proposition. Let K be a subfield of Q. The functor TBdR : M1(K) → Mod fully faithful. ∼=,fr Z,K is Proof. The functor TBdR is fully faithful over Q by Theorem 2.7.1; hence it is faithful over K, since the left-hand vertical functor in (2.21) is faithful. Assume now MK = [u : LK → GK], NK = [v : FK → HK ] are 1-motives over K and let (ϕZ, ϕK ) : TBdR(MK ) → TBdR(NK ) be a morphism in ModZ,K. By Theorem 2.7.1 there exists a morphism ψ : M Q such that TBdR(ψ) = (ϕZ, ϕK ⊗K idQ). Note that there exists a subfield K ′ ⊂ Q with K ′/K finite Galois and ψ = (f, g) is defined over K ′. We may further assume that LK ′, FK ′ are constant free. Hence grW 0 ϕZ descends over K ′ and we have a commutative square Q → N (2.22) LK ′ f =grW 0 (ϕZ) FK ′ LK ′ ⊗ Ga,K ′ grW 0 (ϕK′ ) / FK ′ ⊗ Ga,K ′ 0 (N) respectively). By diagram (2.22) f descends over K since grW where the vertical morphisms map x to x⊗1 (and descend the homomorphism of periods for grW 0 (M) and grW 0 (ϕK ′) = grW 0 (ϕK ) ⊗ idK ′ and the vertical morphisms are injective on points. In order to check that ψ descends over K, we may then reduce to the case LK = FK = 0. By Cartier duality, we may further reduce to the case where LK = FK = 0 and GK = AK, HK = BK are abelian varieties. For any τ ∈ Gal(K ′/K) let τ also denote the corresponding K-automorphism of Spec K ′. K′ : AK ′ → AK ′ be equal Further let τ ∗AK ′ denote the base change of AK ′ along τ and let τA / / / / 24 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) to idAK ⊗ τ . It is not a morphism of K ′-schemes in general. Finally let ιA,τ : AK ′ → τ ∗AK ′ be the canonical morphism of K ′-schemes such that τA K′ is the composition of ιA,τ with the projection τ ∗AK ′ → AK ′. In order to prove that the morphism ψ : AK ′ → BK ′ is defined over K we have to check that for any τ ∈ Gal(K ′/K) it is τB K′ . In fact it is sufficient to check that ιB,τ ◦ψ = (τ ∗ψ)◦ιA,τ as morphisms of K ′-schemes AK ′ → τ ∗BK ′, where τ ∗ψ : τ ∗AK ′ → τ ∗BK ′ is the obvious base change of ψ. By faithfulness of TBdR, it is sufficient to check that K′ ◦ ψ = ψ ◦ τA (2.23) TBdR(ιB,τ ) ◦ TBdR(ψ) = TBdR(τ ∗ψ) ◦ TBdR(ιA,τ ). Note that since AK is a K-form of AK ′, we may identify AK ′ with τ ∗AK ′ so that ιA,τ becomes the identity map. Further TdR(τ ∗ψ) = τ ∗(ϕK ⊗ idK ′) may be identified with ϕK ⊗ idK ′ and TZ(τ ∗ψ) with ϕZ. We conclude that TBdR(τ ∗ψ) may be identified with (ϕZ, ϕK ⊗ idK ′) and hence (2.23) is clear. (cid:3) 3. Some evidence: description of some Grothendieck arithmetic invariants Throughout this section we assume that K = Q and by scheme we mean a separated scheme of finite type over K. In order to show the period conjecture for motivic cohomology (1.6) we are left to deal with rational coefficients. However, we prefer to keep the arguments integral when possible. In general, for any algebraic scheme X over K, by making use of the period isomorphism p,q X in Definition 1.2.4 we set X and its inverse ηp,q H p,q BdR(X) := (H p(Xan, Zan(q)), H p dR(X), p,q X ) ∈ Mod ∼= Z,K and H p,q dRB(X) := (H p dR(X), H p(Xan, Zan(q)), ηp,q X ) ∈ Mod X = (2πi)qp,0 ∼= K,Z. BdR(X)) = H p,q dR(X) ⊗K C X = (2πi)−qηp,0 dRB(X). We have that p,q Note that ς(H p,q X where ηp,0 X : H p −→ H p(Xan, C) is the usual de Rham -- Betti comparison isomor- phism (up to a sign cf. [33, Def. 5.3.1] and [45, Lemma 4.1.1 & Prop. 4.1.2] for the Nisnevich topology). In particular we have that H p,q X and ηp,q ≃ dRB(X) = H p,0 dRB(X)(q). 3.1. Period cohomology revisited. For H ∈ ModZ,K we set H := Hom(Z(0), H) where the Hom-group is taken in ModZ,K. This yields a functor ( ) : ModZ,K → ModZ to the category of finitely generated abelian groups. Similarly, let H := Hom(Z(0), H) ∼= ∼= K,Z. By Lemma 2.4.4, for K,Z where now the Hom-group is taken in Mod for H ∈ Mod H ∈ Mod ∼= Z,K we clearly have that H = ς(H). Moreover, for H ∈ Mod ∼=,fr Z,K we have H ∗ = H(−1)∨ ∈ Mod = Hom(Z(0), H(−1)∨) = Hom(H, Z(1)). H ∗ ∼=,fr Z,K so that MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 25 Note that for H ∈ Mod shall denote ∼=,fr Z,K we also have H ◦(1) = H(−1)◦ ∈ Mod ∼=,fr K,Z (see (2.12)) and we H := HomMod ∼=,fr K,Z (Z(0), H(−1)◦) = HomMod ∼=,fr Z,K (H, Z(1)) = H ∗ . With rational coefficients, for H ∈ ModZ,K ⊗ Q and the corresponding H Q ∈ ModQ,K we then have H ⊗Z Q ∼= H Q Q . We have (cf. [17, Def. 2.1] and [16, (5.15)]): := Hom(Q(0), H Q) and, similarly, H ⊗Z Q ∼= H 3.1.1. Lemma. For H = (HZ, HK , ω) ∈ ModZ,K we have H ∼= HZ ∩ HK where ∩ is the ∼= K,Z, inverse image of HK under ω : HZ → HK ⊗K C. Moreover, for H = (HK, HZ, η) ∈ Mod ∼= HK ∩ HZ where ∩ is the inverse image of HK under the composition of we have that H η−1 −→ HK ⊗K C. HZ → HZ ⊗Z C Proof. The identifications are given by mapping ϕ ∈ Hom(Z(0), H) to ϕ(1) ∈ HZ ∩ HK. (cid:3) Similarly, for H = (HK, HQ, ω) ∈ Mod obtain: ∼= K,Q we have that H Q ∼= HK ∩ HQ. We then clearly 3.1.2. Corollary. For H = H p,q cohomology of Definition 1.2.7. dRB(X) we have that H ∼= H p,q (X) coincides with the period Moreover, composing the functor H H with the Betti -- de Rham realization of 1- motives TBdR in (2.13) we obtain a functor (3.1) T := ( ) ◦ TBdR : tM1(K) → ModZ. For a 1-motive M ∈ tM1(K) we also have TdRB(M) ∈ Mod the de Rham -- Betti realization TdRB in (2.15) now yields a functor T := ( ) ◦ TdRB : tM1(K)op → ModZ. ∼= K,Z. Composing H H with We also note that Lemma 2.5.5 yields: 3.1.3. Corollary. For M ∈ M1(K) we have that T (M) := TdRB(M) ∼= TBdR(M). Working with rational coefficients we have T Q and we then get a functor T Q 1-motives up to isogenies MQ dimensional Q-vector spaces. Moreover, applying our Theorem 2.7.1 we have: dRB := TdRB ⊗ Q) (resp. a contravariant functor T Q ) from the category of 1 := M1 ⊗ Q ∼= tM1 ⊗ Q ∼= tM1 ⊗ Q to the category of finite BdR := TBdR ⊗ Q (resp. T Q 3.1.4. Corollary. For M = [u : L → G] ∈ M1(Q) with Cartier dual M∗ = [u∗ : L∗ → G∗] ∈ M1(Q) we have that T(M) ∼= TZ(MC) ∩ TdR(MK ) ∼= Ker u and T (M) ∼= TdR(M∗ K) ∩ TZ(M∗ C) ∼= Ker u∗. 26 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Proof. Note that Z(0) = TBdR(Z[0]) and for TBdR(M) = (TZ(MC), TdR(MK), M,Z) we have Ker u ∼= HomM1(K)(Z[0], M) TBdR−→ Hom(TBdR(Z[0]), TBdR(M)) = T(M) which is an isomorphism over K = Q as proven in Theorem 2.7.1. We just apply Lemma 3.1.1. Moreover, TdRB(M) = (TdR(M∗ C), ηM∗), Z(0) = TdRB(Gm[−1]) and we have an isomorphism K ), TZ(M∗ HomM1(K)(M, Gm[−1]) TdRB−→ Hom(TdRB(Gm[−1]), TdRB(M)) = T (M) and HomM1(K)(M, Gm[−1]) ∼= HomM1(K)(Z[0], M∗) = Ker u∗ showing the claim. 3.2. Period conjecture for q = 1. Recall that Z(1) ∈ DMeff Tot([0 → Gm]) = Gm[−1] (see [12, Lemma 1.8.7]). We then have ´et is canonically identified with (cid:3) H p,1(X) ∼= H p−1 ´eh (X, Gm) for all p ∈ Z. Recall the motivic Albanese triangulated functor LAlb : DMeff gm → Db(M1) gm ⊂ DMeff where DMeff Nis is the subcategory of compact objects, i.e., the category of geometric motives, which has been constructed in [12, Def. 5.2.1] (see also [10, Thm. 2.4.1]). This functor is integrally defined. Rationally, LAlb yields a left adjoint to the inclusion functor given by Tot in (2.4) (see [12, Thm. 6.2.1]). Applying LAlb to the motive of any algebraic scheme X we get LAlb(X) ∈ Db(M1), a complex of 1-motives whose p-th homology LpAlb(X) ∈ tM1 is a 1-motive (with cotorsion, see [12, Def. 8.2.1]). Dually, we have RPic(X) ∈ Db(tM1) (see [12, §8.3]). Taking the Cartier dual of LpAlb(X) we get RpPic(X) ∈ tM1 and conversely via (2.14). Now, the motivic Albanese map in DMeff ´et (see [12, §8.2.7]) yields an integrally defined map M (X) → Tot LAlb(X) (3.2) HomDb(tM1)(LAlb(X), [0 → Gm][p]) → HomDMeff Rationally (by adjunction), this map becomes a Q-linear isomorphism (3.3) H p−1 ´eh (X, Gm)Q ∼= HomDMeff (M (X), Z(1)[p]) ←− HomDb(MQ ≃ ´et,Q (M (X), Z(1)[p]) ∼= H p−1 ´eh (X, Gm). ´et 1 )(LAlb(X), Gm[−1][p]). Using (2.14) we set Extp(Z, RPic(X)) := HomDb(tM1)(Z, RPic(X)[p]) ∼= HomDb(tM1)(LAlb(X), Gm[−1][p]) for all p ∈ Z and we also have (cf. [12, Lemma 10.5.1]): 3.2.1. Lemma. For any X over K = Q and p ∈ Z there is an extension 0 → Ext(Z, Rp−1Pic(X)) → Extp(Z, RPic(X)) π−→ Hom(Z, RpPic(X)) → 0 where the Hom and Ext are here taken in the category tM1 of 1-motives with torsion. The composition of (3.2) with the period regulator rp,1 (X) induces a mapping ´eh (X, Gm) → H p,1 : H p−1 θp : Hom(Z, RpPic(X)) → H p,1 (X). MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 27 Proof. In fact, the canonical spectral sequence Ep,q 2 = Extp(Z, RqPic(X)) ⇒ Extp+q(Z, RPic(X)) yields the claimed extension since the abelian category of 1-motives with torsion tM1(K) is of homological dimension 1 over the algebraically closed field K = Q. Moreover, for any 1-motive M = RpPic(X) ∈ tM1(K) the group Ext(Z, M) is divisible and the group Hom(Z, M) is finitely generated (as it follows easily by making use of [12, §C.8]). The horizontal mapping in the following commutative diagram Ext(Z, Rp−1Pic(X)) zero Extp(Z, RPic(X)) H p−1 ´eh (X, Gm) rp,1 / H p,1 (X) π Hom(Z, RpPic(X)) θp obtained by the composition of (3.2) with the period regulator rp,1 Ext(Z, Rp−1Pic(X)) to zero, since H p,1 mapping θp , is therefore sending (X) is finitely generated. We then get the induced (cid:3) as indicated in the diagram. Also for the Betti realization, there is an integrally defined group homomorphism θp Z : TZ(RpPic(X)C)fr → H p(Xan, Zan(1))fr induced via Cartier duality, by applying the Betti realization βσ in (1.3) to the motivic Albanese (3.2) in a canonical way. This is justified after the natural identification of Deligne's TZ with the Betti realization βσ on 1-motives (see [12, Thm. 15.4.1] and [53] for an explicit construction of the natural isomorphism TZ ∼= βσ Tot). Rationally, it yields an injection θp Q : TQ(RpPic(X)C) ∼= H p (1)(Xan, Qan(1)) ⊂ H p(Xan, Qan(1)) where the notation H p (1) is taken to indicate the largest 1-motivic part of H p(Xan, Qan(1)) (more precisely, this is given by the underlying Q-vector space associated to the mixed Hodge structure, see [12, Cor. 15.3.1]). For the de Rham realization, similarly, we have a K-linear mapping θp dR : TdR(RpPic(X)) → H p dR(X). Actually, for M = LpAlb(X) and M∗ = RpPic(X), we have ηM∗ the C-inverse of the period isomorphism M∗,C in Theorem 2.2.2 and ηp,1 X which is the inverse of the period isomorphism in Definition 1.2.4. Together with θp Z and θp dR, we obtain a diagram θp Z ⊗C 2πi TZ(RpPic(X)C)C / H p(Xan, Zan(1))C / H p(Xan, C) ηM∗ TdR(RpPic(X))C θp dR ⊗C ηp,1 X 6♠♠♠♠♠♠♠♠♠♠♠♠♠ ηp,0 X / H p dR(X) ⊗K C. _ & & / / / 8 8 / / O O / O O 6 28 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) We have that this diagram commutes, in fact: 3.2.2. Lemma. Let X be over the field K = Q and p ∈ Z. There is a morphism dR, θp θp dRB := (θp ∼=,fr K,Z. Rationally θp in the category Mod are integrally defined isomorphisms. Z) : TdRB(LpAlb(X))fr → H p,1 dRB(X)fr dRB ⊗ Q becomes injective. Moreover, θ0 dRB and θ1 dRB Proof. This is a consequence of [12, Cor. 16.3.2]. For p = 0, 1 it is straightforward that they are isomorphisms. (cid:3) 3.2.3. Lemma. The map θp realization via the Cartier duals (2.14), i.e., we have the following factorization defined in Lemma 3.2.1 factors through the de Rham-Betti θp Hom(Z, RpPic(X)) Hom(LpAlb(X), Gm[−1]) TdRB/ / T (LpAlb(X)) ι / H p,1 (X) such that ι is given by θp dRB in Lemma 3.2.2, using Corollary 3.1.2, as follows T (LpAlb(X)) = Hom(Z(0), TdRB(LpAlb(X))) → Hom(Z(0), H p,1 dRB(X)) ∼= H p,1 (X) and the latter Hom is here taken in Mod ∼= K,Z. is induced by rp,1 Proof. By construction θp applying Betti and de Rham realizations so that the claimed factorization is clear. on a quotient via the motivic Albanese (3.2) (cid:3) Thus, showing the period conjecture (1.6) for q = 1 is equivalent to seeing that θp is surjective, rationally. Recall (see [12, Prop. 10.4.2]) that for any X of dimension d = dim(X) the 1-motive Ld+1Alb(X) is a group of multiplicative type and where G1 is connected, so that LpAlb(X) ∈ M1 is free for p = 0, 1 (see [12, Prop. 12.6.3 c)]). Thus R0Pic(X) = [Z[π0(X)] → 0]∗ = [0 → Z[π0(X)]∨ ⊗ Gm] is a torus and we have that Ext(Z, R0Pic(X)) = HomK (Z, R0Pic(X)) = K ∗ ⊗Z Z[π0(X)]∨ (see [12, Prop. C.8.3 (b)]). 3.2.4. Theorem. For any X over K = Q we have that (1.6) holds true for p = q = 1, i.e., the period regulator r1,1 (X) is surjective. Moreover, considering the 1-motive R1Pic(X) = [L∗ 1 isomorphism 1] which is the Cartier dual of L1Alb(X) we have a canonical : H 0 u∗ 1→ G∗ ´eh(X, Gm)→→H 1,1 Ker u∗ 1 dR(X) ∩ H 1(Xan, Zan(1)) = H 1,1 ∼= H 1 ´eh(X, Gm) ∼= K ∗⊗ZZ[π0(X)]∨ and H 1 (X). dR(X)∩H 1(Xan, Z(1)) = In particular, if X is proper H 0 0. LpAlb(X) = 0 [Z[π0(X)] → 0] [L1 0 u1→ G1] if p < 0 if p = 0 if p = 1 if p > max(2, d + 1) * * / MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 29 Proof. In fact, R1Pic(X) is free and therefore Hom(L1Alb(X), Gm[−1]) ∼= Hom(Z, R1Pic(X)) ∼= Ker u∗ 1. Thus the extension in Lemma 3.2.1 is 0 → K ∗ ⊗Z Z[π0(X)]∨ → Ext1(Z, RPic(X)) → Ker u∗ 1 → 0. Moreover θ1 implies that r1,1 : Hom(Z, R1Pic(X)) ∼= Ker u∗ 1 ≃ → H 1,1 (X) is an isomorphism, which in turn is a surjection. Actually, see Lemma 3.2.3, θ1 factors as follows Hom(Z, R1Pic(X)) (a) ∼= T (L1Alb(X)) (b) ∼= Hom(Z(0), H 1,1 dRB(X)) (c) ∼= H 1,1 (X) where: (a) is the isomorphism obtained applying Corollary 3.1.4 to M = L1Alb(X); (b) is the Hom(Z(0), −) of the isomorphism θ1 dRB(X) given by p = 1 in Lemma 3.2.2; (c) is the isomorphism in Corollary 3.1.2. If X is proper then L∗ 1 = 0, i.e., u1→ G1] with G1 an abelian variety (see [12, Cor. 12.6.6]) in such a way L1Alb(X) = [L1 that R1Pic(X) = [0 → G∗ ´eh(X, Gm) ∼= Gm(π0(X)) (see [12, Lemma 12.4.1]). dRB : TdRB(L1Alb(X)) ∼= H 1,1 1], and H 0 (cid:3) 3.2.5. Remark. We may actually compute R1Pic(X) by using descent. For example, if X is normal let X be a normal compactification of X, p : X• → X a smooth hypercovering and X • a smooth compactification with normal crossing boundary Y• such that p : X • → X is a hypercovering. Then p∗ : Pic0 is an abelian variety and → Pic0 ≃ X/K X •/K R1Pic(X) = [Div0 Y•(X •) u∗ 1→ Pic0 X/K] where Div0 Y•(X •) := Ker(Div0 Y0(X 0) → Div0 Y1(X 1)) (see [12, Prop. 12.7.2]). For X smooth we have that (see [12, Cor. 9.2.3]) LpAlb(X) = if p = 2 otherwise, X/K is the Serre-Albanese semi-abelian variety and NS∗ where A0 multiplicative type dual to the Néron-Severi group NSX/K . In this case, we then have X/K denotes the group of if p = 0 if p = 1 [Z[π0(X)] → 0] [0 → A0 [0 → NS∗ 0 X/K ] X/K ]   RpPic(X) = [0 → Z[π0(X)]∗] [Div0 u∗ 1→ Pic0 Y (X) X/K] [NSX/K → 0] 0 if p = 0 if p = 1 if p = 2 otherwise, for a smooth compactification X with normal crossing boundary Y . Note that, reducing to the smooth case by blow-up induction we can see that the map (3.2) is an isomorphism for p = 0, 1 (cf. [12, Lemma 12.6.4 b)]). We deduce the following: 3.2.6. Corollary. For any scheme X over K = Q we have a short exact sequence 0 → K ∗ ⊗Z Z[π0(X)]∨ → H 0 ´eh(X, Gm) r1,1 −→ H 1,1 (X) → 0. 30 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) In general, we also have: admits a factorization ´eh (X, Gm)Q→→T 3.2.7. Proposition. For K = Q the period regulator rp,1 Q (LpAlb(X)) ֒→ H p H p,1(X)Q ∼= H p−1 where the projection is given by Lemma 3.2.1 via T Q Q in Lemma 3.2.2. Therefore, the conjecture (1.6) is equivalent to T H p,1 Proof. In fact, using the adjunction (3.3), the Cartier dual π∗ of π in Lemma 3.2.1, the factorization of Lemma 3.2.3 and Theorem 2.7.1 we have the following commutative diagram dR(X) ∩ H p(Xan, Qan(1)) = H p,1 dRB and the inclusion is given by θp dRB ⊗ Q (LpAlb(X)) ∼= (X)Q (X)Q. HomDb(MQ 1 )(LAlb(X), Gm[−1][p]) π∗⊗Q ≃ H p−1 ´eh (X, Gm)Q 2❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡❡ 7♦♦♦♦♦♦♦♦♦♦♦♦♦ rp,1 ,Q / H p,1 (X)Q Hom(LpAlb(X), Gm[−1])Q For X smooth we further have that ≃ T Q dRB / T Q (LpAlb(X)). H p,1(X) ∼= H p−1 ´eh (X, Gm) ∼= H p−1 ´et (cid:3) (X, Gm) and this latter is vanishing after tensoring with Q for all p 6= 1, 2 (see [29, Prop. 1.4]). Accordingly, the period conjecture (1.6) for X smooth and p 6= 1, 2 is in fact equivalent to (1.8), i.e., H p,1 (X) = H p (3.4) For p = 2 and X smooth we have that H 2,1(X) ∼= Pic(X), r2,1 cycle class map and T Q (L2Alb(X)) = NS(X)Q. dR(X) ∩ H p(Xan, Qan(1)) = 0 p 6= 1, 2. = cℓ is induced by the usual We here recover the results of Bost-Charles (see [16, Thm. 5.1] and [17, Cor. 3.9-3.10]) as follows. We refer to [12, Chap. 4] for the notion of biextension of 1-motives. The following is a generalization of [17, Thm. 3.8 2)] and of the discussion of the sign issue in [17, §3.4]: 3.2.8. Lemma. For N, M ∈ M1(Q) we have that Biext(N, M; Gm) ∼= (TBdR(N)∨ ⊗ TBdR(M)∨ ⊗ Z(1)) and, when N = M, the subgroup of symmetric biextensions corresponds to alternating ele- ments. Proof. Recall that Biext(−, M; Gm) is representable by the Cartier dual M∗ for M ∈ M1(K) (see [12, Prop. 4.1.1]). Thus Biext(N, M; Gm) = Hom(N, M∗) ∼= Hom(TBdR(N), TBdR(M)∗) where we here use Theorem 2.5.3 and Theorem 2.7.1. Now TBdR(M)∗ = TBdR(M)∨(1) in such a way that Hom(TBdR(N), TBdR(M)∗) = Hom(Z(0), TBdR(N)∨ ⊗ TBdR(M)∨ ⊗ Z(1)) making use of the tensor structure of the category Mod ∼=,fr Z,K by Lemma 2.4.2. Assume N = M. Since Biext(M, M; Gm) ∼= Hom(TBdR(M), TBdR(M)∗), any biextension P corresponds to a pairing TBdR(M) ⊗ TBdR(M) → Z(1) which induces the pairing [23, 10.2.3] / / / $ 2 / * 7 MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 31 on Deligne-Hodge realizations and the pairing [23, 10.2.7] on de Rham realizations; if P is symmetric, the pairing is alternating by [23, 10.2.5 & 10.2.8]. (cid:3) 3.2.9. Corollary. For X over K = Q we have that Biext(L1Alb(X), L1Alb(X); Gm)sym ∼= (H 1,0 dRB(X) ⊗ H 1,0 dRB(X) ⊗ Z(1))alt . Proof. Applying Lemma 3.2.8 to the free 1-motive L1Alb(X) we obtain the claimed formula. In fact, recall that TZ(L1Alb(X)) ∼= H1(Xan, Zan)fr and observe that H 1,0 dRB(X) is identified with TBdR(L1Alb(X))∨ up to inverting the period isomorphism by the same argument of Lemma 3.2.2. (cid:3) This implies that the period conjecture for p = 2 holds true in several cases, e.g. for abelian varieties, as previously proved by Bost (see [16, Thm. 5.1]). 3.3. The case q = 0. Consider the case of Z(0) which is canonically identified with Tot([Z → 0]) = Z[0]. Note that H p,0(X) ∼= H p ´eh(X, Z). Let M0(K) ⊂ M1(K) be the full subcategory of 0-motives or Artin motives over K. Recall that the motivic π0 (see [12, §5.4] and [10, Cor. 2.3.4]) is a triangulated functor Lπ0 : DMeff gm → Db(M0) whence Lπ0(X) ∈ Db(M0), a complex in the derived category of Artin motives, associated to the motive of X. We have that M (X) → Tot Lπ0(X) ∈ DMeff ´et (see (2.4) for Tot) induces HomDb(M0)(Lπ0(X), Z[p]) → HomDMeff ´et (M (X), Z(0)[p]) ∼= H p ´eh(X, Z). [12, Lemma 12.6.4 b)]) and it This map is an isomorphism, integrally, for p = 0, 1 (cf. becomes, by adjunction, a Q-linear isomorphism, for all p. Recall that for any M ∈ DMeff gm we have (see [12, Prop. 8.2.3]) LAlb(M (q)) ∼=(Lπ0(M )(1) 0 if q = 1 for q ≥ 2 where an Artin motive twisted by one is a 1-motive of weight −2, i.e., the twist by one functor (−)(1) : Db(M0) → Db(M1) is induced by L [0 → L ⊗ Gm]. Note that as soon as K = Q Artin motives are of homological dimension 0 and we have that HomDb(M0)(Lπ0(X), Z[p]) = HomM0(Lpπ0(X), Z). Moreover, we have that H p ´eh(X, Z) ∼= HomDMeff ´et (M (X)(1), Z(1)[p]) by Voevodsky's cancellation theorem [51]. 3.3.1. Theorem. For any X over K = Q we have that (1.6) holds true for p = 1 and q = 0. Moreover, we have H 1 ´et(X, Z) ∼= H 1 dR(X) ∩ H 1(Xan, Zan) ∼= H 1,0 (X) which is vanishing if X is normal. 32 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Proof. Making use of Proposition 1.3.1 we are left to show the period conjecture for M (X)(1) in degree 1 and twist 1. We have that HomDMeff ´et (M (X)(1), Z(1)[1]) ∼= HomD(M1)(LAlb(M (X)(1)), Z(1)[1]). We have L0Alb(M (X)(1)) ∼= L0π0(X)(1) ∼= [0 → Z[π0(X)] ⊗ Gm] in such a way that and (cf. (3.3) for M (X)(1)) we obtain ExttM1(L0Alb(M (X)(1)), Gm[−1]) = 0 H 1 ´eh(X, Z) ∼= HomtM1(L1Alb(M (X)(1)), Gm[−1]). ´et(X, Z) ∼= H 1 (M (X)(1)) ∼= H 1,0 ´eh(X, Z) for any scheme X and H 1 Now T (L1Alb(M (X)(1))) ∼= H 1,1 (X) by Lemma 3.2.2 twisted by (−1) and the same argument in the proof of Theorem 3.2.4 applies here. Finally, recall that H 1 ´eh(X, Z) = 0 if X is normal (see [12, Lemma 12.3.2 & Prop. 12.3.4]). (cid:3) 3.3.2. Remark. For X not normal (e.g. for the nodal curve) the group H 1 ´et(X, Z) can be ´et(X, Z) ∼= L Pic(X) ֒→ non-zero. Moreover, for any X we have a geometric interpretation H 1 Pic(X[t, t−1]) by a theorem of Weibel [55, Thm. 7.6]. Note that this L Pic(X) is also a sub- quotient of the negative K-theory group K−1(X) (see [55, Thm. 8.5]). For X smooth we have a quasi-isomorphism Lπ0(X) ∼= Z[π0(X)][0] (see [12, Prop. 5.4.1]) which means that H p,0(X)Q = 0 for p 6= 0. This yields (as it also does Proposition 1.4.4 for X smooth) that the period conjecture (1.6) is equivalent to (3.5) H p dR(X) ∩ H p(Xan, Qan) = 0 p 6= 0. 3.3.3. Remark. The period conjecture (1.6) for q = 0 and X smooth is also equivalent to the surjectivity of f p (X)Q induced by the canonical morphism f : X → π0(X), for all p ≥ 0. In fact, the morphism f induces a map M (X) → M (π0(X)) and a commutative square by functoriality (π0(X))Q → H p,0 : H p,0 H p,0(X)Q rp,0 H p,0 (X)Q f p f p H p,0(π0(X))Q ∼= / H p,0 (π0(X))Q (M (π0(X)), Z[p])Q → HomDMeff where f p : HomDMeff X smooth; since dim π0(X) = 0 then rp,0 group H 0(Xan, Zan(0)) has rank equal to the rank of Z[π0(X)] and f 0 for p 6= 0 the surjectivity of f p is equivalent to the vanishing of all groups. is clearly an isomorphism for π0(X). For p = 0 the is an isomorphism; (M (X), Z[p])Q is an isomorphism for ´et ´et 3.4. Arbitrary twists. We now apply Waldschmidt's Theorem 2.3.3 to arbitrary twists. 3.4.1. Proposition. For M = [L → G] a free 1-motive over K = Q and q ∈ Z an integer we have that 1) the group Hom(cid:0)Z(q), TBdR(M)(cid:1) of homomorphisms in Mod for q 6= 0, 1; ∼= ∼= Q,K is trivial Z,K or Mod / / / O O MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 33 2) the group Hom(cid:0)Z(q), TdRB(M)(cid:1) of homomorphisms in Mod for q 6= 0, 1. ∼= K,Z or Mod ∼= K,Q is trivial Proof. 1) We work in Mod ∼= Z,K and leave the other case to the reader. We suppose first C) we can identify Γ with a subgroup of Lie(G♮ TZ(Z(q)) = Z ⊂ TdR(Z(q))C = C. Via the non trivial map ϕK ⊗ C : TdR(Z(q))C → TdR(G)C = Lie(G♮ C). This subgroup is con- that L = 0. Consider a non trivial ϕ ∈ HomBdR(cid:0)Z(q), TBdR(M)(cid:1) and the subgroup Γ = tained in VC with V ⊂ Lie(G♮) defined by the image ϕK(cid:0)TdR(Z(q))(cid:1). Via the exponential ∼= map Lie(G♮ Z,K ∼= Q,K). We deduce from Waldschmidt's Theorem 2.3.3 that V ⊂ Lie(G♮) is (respectively Mod the Lie algebra of a 1-dimensional algebraic subgroup H of G♮. There are only two possi- bilities H = Ga and H = Gm. In both cases the period morphism for Z(q) identifies Γ with the subgroup (2πi)q Z ⊂ Lie(HC) that goes to 0 via expHC. For H = Ga the map expHC is the identity, leading to a contradiction. For H = Gm the kernel of expHC is (2πi)Z forcing q = 1. C) → G♮(C) the image of Γ is 0 ∈ G♮(K) as ϕ is a map in the category Mod Secondly we suppose that G = 0. Consider a non trivial ϕ ∈ HomBdR(cid:0)Z(q), TBdR(M)(cid:1). Recall that TdR(M) = L ⊗ K and the period map is induced by the inclusion L ⊂ L ⊗ K. Let e = ϕK (1) ∈ L ⊗ K. It is a non-zero element. Using that TZ(Z(q)) is identified via the period morphism for Z(q) with (2πi)qZ, we deduce that ϕZ(1) = (2πi)q · e should lie in L ⊂ L ⊗ K. As π is transcendental, this forces q = 0. For general M = [L → G] we reduce to G and L to conclude the statement. 2) We prove the statement for Mod ∼= K,Q follows similarly. Given a 1-motive M and its Cartier dual M∗ we have a natural identification ∼= K,Z using Lemma 2.4.4. The analogue for Mod ς : Hom(cid:0)Z(q), TBdR(M∗)(cid:1) ∼= Hom(cid:0)ς(Z(q)), ς(TBdR(M∗)) = Hom(cid:0)Z(q), TdRB(M)(cid:1)(cid:1). The statement follows then from 1). (cid:3) Denote H p,q dRB,(1)(X)fr ⊂ H p,q dRB(X)fr the image of TdRB(LpAlb(X))fr(q −1) under θp dRB(q − 1) of Lemma 3.2.2 twisted by q − 1. We have: ,(1)(X)fr = 0 if q 6= 0, 1. For p = 1 we have H 1,q ,(1)(X) = ´et(X, Z) (see Theorem 3.3.1) 1 (see Theorem 3.2.4) if q = 0 if q = 1 q 6= 0, 1. 3.4.2. Corollary. We get that H p,q H 1,q (X) and H 1 Ker u∗ 0 H 1,q (X) = ,(1) (X)fr = HomdRB(cid:0)Z(0), H p,1 Proof. We apply Proposition 3.4.1 2) to M = LpAlb(X) to deduce that H p,1−q dRB,(1)(X)(−q))(cid:1) = HomdRB(cid:0)Z(q), TdRB(LpAlb(X))fr(cid:1) = 0 (cid:3) if q 6= 0, 1. Thus, for the period conjecture in degree p = 1, the previous computations for the twists q = 0, 1 are the only relevant. 34 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Z 3.5. Higher odd degrees. Next, let X be a smooth and projective variety over K = Q. Denote J 2k+1(X) the intermediate Jacobian: as a real analytic manifold, it is defined as the quotient of the image H 2k+1 (X) of H 2k+1(cid:0)Xan, Zan(k)(cid:1) in H 2k+1(cid:0)Xan, R(k)(cid:1). This defines a full lattice of H 2k+1(cid:0)Xan, R(k)(cid:1) so that J 2k+1(X) is compact. It has also a natural complex analytic structure induced by the identification J 2k+1(X) := H 2k+1(Xan, C)/(cid:0)F k+1H 2k+1(Xan, C) + (2k+1,k For integers n define N nH 2k+1(cid:0)Xan, Qan(k)(cid:1) ⊂ H 2k+1(cid:0)Xan, Qan(k)(cid:1), the n-th step of the )−1(cid:0)H 2k+1 geometric coniveau filtration, as the kernel of Thus J 2k+1(X) is a complex torus. (X)(cid:1)(cid:1). X Z H 2k+1(cid:0)Xan, Qan(k)(cid:1) −→ MZ⊂X H 2k+1(cid:0)Xan\Zan, Qan(k)(cid:1) for Z ⊂ X varying among the codimension ≥ n closed subschemes. 3.5.1. Lemma. Assume that H 2k+1(cid:0)Xan, Qan(k)(cid:1) has geometric coniveau k, i.e., that we have N kH 2k+1(cid:0)Xan, C(cid:1) = H 2k+1(cid:0)Xan, C(cid:1). Then J 2k+1(X) is an abelian variety, which descends to an abelian variety J 2k+1(X)K over K with TdRB(cid:0)J 2k+1(X)K(cid:1) =(cid:0)H 2k+1 dR (X), H 2k+1 Z (X), η2k+1,k X (cid:1). Z Proof. Under the assumption, H 2k+1 (X) is a polarized Hodge structure of type (1, 0) and (0, 1) so that J 2k+1(X) is polarizable and, hence, an abelian variety. The second statement follows from [1, Thm. A] where it is proven that there exists an abelian variety J over K and a correspondence Γ ∈ CHh(J ×K X) over K, for h = k + dim J 2k+1(X), inducing an compatibly with the period morphisms). Then set J 2k+1(X)K := J. The period conjecture (1.6) in odd degrees for X predicts that H 2k+1,q isomorphism Γ∗ : H 1(Jan, Qan) ∼= H 2k+1(cid:0)Xan, Qan(k)(cid:1) (and hence in de Rham cohomology, H 2k+1(cid:0)Xan, Zan(q)(cid:1)fr = 0 for every k ∈ N and every q ∈ Z. smooth and projective holds true if H 2k+1(cid:0)Xan, Qan(k)(cid:1) has geometric coniveau k. 3.5.2. Proposition. The period conjecture (1.6) in degree p = 2k + 1 and any twist q for X Proof. Thanks to Lemma 3.5.1 we have that H 2k+1,q (X) = Hom(cid:0)Z(0), TdRB(cid:0)J 2k+1(X)K(cid:1)(q − k)(cid:1) = Hom(cid:0)Z(k − q), TdRB(cid:0)J 2k+1(X)K(cid:1)(cid:1). This is trivial for k − q 6= 0, 1 by Proposition 3.4.1. Now use Theorem 2.7.1. For k − q = 0 we get that this coincides with the homomorphisms of 1-motives from [Z → 0] to [0 → J 2k+1(X)K ], which is 0. For k − q = 1 this coincides with the homomorphisms of 1-motives from [0 → Gm] to [0 → J 2k+1(X)K ], which is also 0. (cid:3) (X) = H 2k+1 dR (X)∩ (cid:3) 3.5.3. Remark. Lemma 3.5.1 is proven more generally in [1] for the Hodge structure N kH 2k+1(cid:0)Xan, Qan(k)(cid:1) ⊂ H 2k+1(cid:0)Xan, Qan(k)(cid:1) defined by the k-th step of the coniveau (cid:1). The filtration. Namely, if X is defined over a number filed L ⊂ K, there is an abelian vari- ety J 2k+1 (X) over L with TdRB(cid:0)J 2k+1 (X)(cid:1) =(cid:0)N kH 2k+1 dR (X), N kH 2k+1 (X), η2k+1,k X Z a a MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 35 (X) gives the following weak version of the period proof of Proposition 3.5.2 using J 2k+1 conjecture: a (3.6) N kH 2k+1 projective threefolds; see [2]. The assumption implies, and under the generalized Hodge conjecture is equivalent to, the dR (X) ∩ N kH 2k+1(cid:0)Xan, Zan(q)(cid:1)fr = 0 for every k ∈ N and every q ∈ Z. The assumption in Lemma 3.5.1 amounts to saying that N kH 2k+1(cid:0)Xan, Qan(k)(cid:1) is equal to H 2k+1(cid:0)Xan, Qan(k)(cid:1). This equality holds, for example, for k = 1 for uniruled smooth fact that the Hodge structure H 2k+1(cid:0)Xan, Q(cid:1) has Hodge coniveau k, i.e., H 2k+1(cid:0)Xan, C(cid:1) weaker condition on the Hodge coniveau one can still prove that H 2k+1(cid:0)Xan, Qan(k)(cid:1) is the Hodge structure associated to the abelian variety J 2k+1 (X) over C, called the algebraic intermediate Jacobian in J 2k+1(X). Unfortunately one lacks the descent to K. See the discussion in [2]. is the sum of the (k + 1, k) and (k, k + 1) pieces of the Hodge decomposition. Under this alg Appendix A. Divisibility properties of motivic cohomology (by B. Kahn) In this appendix, some results of Colliot-Thélène and Raskind on the K2-cohomology of smooth projective varieties over a separably closed field k are extended to the étale motivic cohomology of smooth, not necessarily projective, varieties over k. Some consequences are drawn, such as the degeneration of the Bloch-Lichtenbaum spectral sequence for any field containing k. Recall that in [19], Colliot-Thélène and Raskind study the structure of the K2-cohomology groups of a smooth projective variety X over a separably closed field. Following arguments of Bloch [15], their proofs use the Weil conjecture proven by Deligne [22] and the Merkurjev- Suslin theorem [48]. These results and proofs can be reformulated in terms of motivic cohomology, since or even in terms of étale motivic cohomology, since H i Zar(X, K2) ≃ H i+2(X, Z(2)) H j(X, Z(2)) ∼ −→ H j ´et(X, Z(2)) for j ≤ 3 as follows again from the Merkurjev-Suslin theorem. If we work in terms of étale motivic cohomology, the recourse to the latter theorem is irrelevant and only the results of [22] are needed; in this form, the results of [19] and their proofs readily extend to étale motivic cohomology of higher weights, as in [41, Prop. 4.17] and [42, Prop. 1] (see also [49, Prop. 1.3]). Here we generalise these results to the étale motivic cohomology of smooth varieties over a separably closed field: see Theorem A.1.3. This could be reduced by a dévissage to the smooth projective case, using de Jong's alteration theorem in the style of [40], but it is simpler to reason directly by using cohomology with compact supports, and Weil II [24] rather than Weil I [22]. (I thank Hélène Esnault and Eckart Viehweg for suggesting to use this approach). This descends somewhat to the case where the base field k is not separably closed, yielding information on the Hochschild-Serre filtration on étale motivic cohomology (Theorem A.2.1). The rest of the appendix is concerned with implications on 36 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) motivic cohomology of a field K containing a separably closed field: the main result, which uses the norm residue isomorphism theorem of Voevodsky, Rost et al. ([52], see also [31]) is that H i(K, Z(n)) is divisible for i 6= n (Theorem A.3.1). As an immediate consequence, the "Bloch-Lichtenbaum" spectral sequence of K from motivic cohomology to algebraic K- theory degenerates (Theorem A.4.1). We also show that the cokernel of the cup-product map H i−1(K, Z(n − 1)) ⊗ K ∗ → H i(K, Z(n)) is uniquely divisible for i < n (Theorem A.5.1). Throughout, motivic cohomology is understood in the sense of Suslin and Voevodsky (hypercohomology of the Suslin-Voevodsly complexes [50]). A.1. A weight and coniveau argument. Let X be a separated scheme of finite type over a finitely generated field k. A.1.1. Proposition. Let n ∈ Z, ks a separable closure of k and G = Gal(ks/k). Let X = X ⊗k ks. Then H j c (X, Zl(m))G are finite for j /∈ [2m, m + d] and any prime number l invertible in k, where d = dim X. c (X, Zl(m))G and H j Proof. Suppose first that k = Fq is finite. By [21, Cor. 5.5.3 p. 394], the eigenvalues of Frobenius acting on H j c (X, Ql) are algebraic integers which are divisible by qj−d if j ≥ d. This yields the necessary bound m ≥ j − d for an eigenvalue 1. On the other hand, by [24], these eigenvalues have archimedean absolute values ≤ qj/2: this gives the necessary bound m ≤ j/2 for an eigenvalue 1. The conclusion follows. In general, we may choose a regular model S of k, of finite type over Spec Z, such that X extends to a compactifiable separated morphism of finite type f : X → S. By [35, lemma 2.2.2 p. 274 and 2.2.3 p. 277], Rjf!Zl is a constructible Zl-sheaf on S and its formation commutes with any base change. Shrinking S, we may assume that it is locally constant and that l is invertible on S. For a closed point s ∈ S, this gives an isomorphism H j c (X, Zl) ≃ H j c (X s, Zl) compatible with Galois action, and the result follows from the first case. (cid:3) A.1.2. Corollary. If X is smooth in Proposition A.1.1, then H i(X, Zl(n))(G) is finite for i /∈ [n, 2n], where the superscript (G) denotes the subset of elements invariant under some open subgroup of G. If X is smooth projective, then H i(X, Zl(n))(G) is finite for i 6= 2n and 0 for almost all l. Proof. By Poincaré duality and Proposition A.1.1, H i(X, Zl(n))G is finite for i /∈ [n, 2n]; the claim follows since H i(X, Zl(n))) is a finitely generated Zl-module. In the projec- tive case, the Weil conjecture [22] actually gives the finiteness of H i(X, Zl(n))G, hence of H i(X, Zl(n))(G), for all i 6= 2n. But Gabber's theorem [25] says that H i(X, Zl(n)) is torsion-free for almost all l, hence the conclusion. (cid:3) A.1.3. Theorem. Let X be a smooth variety over a separably closed field k of exponential characteristic p. Then, for i /∈ [n, 2n], the group H i ´et(X, Z(n))[1/p] is an extension of a direct sum T of finite l-groups by a divisible group. If X is projective, this is true for all i 6= 2n, MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 37 and T is finite. If p > 1, H i H i ´et(X, Z(n)) ⊗ Q/Z = 0 for i < n. For i ≤ 1, H i ´et(X, Z(n)) is uniquely p-divisible for i < n. In particular, ´et(X, Z(n)) is divisible. The sequence 0 → H i−1 ´et (X, Q/Z(n)) → H i ´et(X, Z(n)) → H i ´et(X, Z(n)) ⊗ Q → 0 is exact for i < n. Proof. Away from p, it is identical to [42, proof of Prop. 1] (which is the projective case) in view of Corollary A.1.2. The unique p-divisibility of H i ´et(X, Z(n)) for i < n follows from [26, Th. 8.4] and requires no hypothesis on k. (cid:3) A.1.4. Corollary. Let K be a field containing a separably closed field k. Then, for i < n, the sequence ´et(K, Z(n)) → H i is exact and the left group has no p-torsion if p = char K. ´et (K, Q/Z(n)) → H i 0 → H i−1 ´et(K, Z(n)) ⊗ Q → 0 Proof. We may assume K/k finitely generated. By Theorem A.1.3, this is true for any smooth model of K over k, and we pass to the limit (see [37, Prop. 2.1 b)], or rather its proof, for the commutation of étale motivic cohomology with limits). (cid:3) A.1.5. Remarks. 1) At least away from p, the range of "bad" i's in Corollary A.1.2 and Theorem A.1.3 is [n, 2n] in general but shrinks to 2n when X is projective. If we remove a smooth closed subset, this range becomes [2n − 1, 2n]. As the proof of Proposition A.1.1 shows, it depends on the length of the weight filtration on H ∗(X, Ql). If X = Y − D, where Y is smooth projective and D is a simple normal crossing divisor with r irreducible components, the range is [2n − r, 2n]. It would be interesting to understand the optimal range in general, purely in terms of the geometry of X. 2) Using Proposition A.1.1 or more precisely its proof, one may recover the l-local version of [40, Th. 3] without a recourse to de Jong's alteration theorem. I don't see how to get the global finiteness of loc. cit. with the present method, because one does not know whether the torsion of H j c (X, Zl) vanishes for l large when X is not smooth projective. 3) Using a cycle class map to Borel-Moore l-adic homology, one could use Proposition A.1.1 to extend Theorem A.1.3 to higher Chow groups of arbitrary separated k-schemes of finite type. Such a cycle class map was constructed in [36, §1.3]. Note that Borel-Moore l-adic cohomology is dual to l-adic cohomology with compact supports, so the bounds for finiteness are obtained from those of Proposition A.1.1 by changing signs. A.2. Descent. A.2.1. Theorem. Let X be a smooth variety over a field k; write ks for a separable clo- sure of k, Xs for X ⊗k ks and G for Gal(ks/k). For a complex of sheaves C over X´et, write F rH i ´et(X, C) induced by the Hochschild-Serre spectral sequence ´et(X, C) for the filtration on H i Er,s 2 (C) = H r(G, H s ´et(Xs, C)) ⇒ H r+s ´et (X, C). 38 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Then, for i < n, the homomorphism F rH i−1 ´et (X, Q/Z(n)) → F rH i ´et(X, Z(n)) induced by the Bockstein homomorphism β is bijective for r ≥ 3 and surjective for r = 1, 2. Proof. By the functoriality of Er,s morphism of spectral sequences δr,s m : Er,s−1 m (C) with respect to morphisms of complexes, we have a m (Q/Z(n)) → Er,s m (Z(n)) converging to the Bockstein homomorphisms. By Theorem A.1.3, δr,i−r and surjective for r = 1. It follows that, for m ≥ 3, δr,i−r for r = 1, 2. The conclusion follows. is bijective for r ≥ 2 is bijective for r ≥ 3 and surjective (cid:3) m 2 A.2.2. Remarks. 1) Of course, F rH i ´et(X, Z(n)) is torsion for r > 0 by a transfer argument, hence is contained in βH i−1(X, Q/Z(n)). The information of Theorem A.2.1 is that it equals βF rH i−1(X, Q/Z(n)). 2) For i ≥ n, we have a similar conclusion for higher values of r, with the same proof: this is left to the reader. A.3. Getting the norm residue isomorphism theorem into play. Recall that for any field K and any i ≤ n, we have an isomorphism (A.1) H i(K, Z(n)) ∼ −→ H i ´et(K, Z(n)). Indeed, this is seen after localising at l for all prime numbers l. For l 6= char K, this follows from [50, 27] and the norm residue isomorphism theorem [52], while for l = char K it follows from [26]. Finally, H i(K, Z(n)) = 0 for i > n. This yields: A.3.1. Theorem. Let K be as in Corollary A.1.4. Then, for i 6= n, the group in (A.1) is divisible. Proof. Again it suffices to prove this statement after tensoring with Z(l) for all prime numbers l. This is an immediate consequence of Corollary A.1.4 since, by [52], one has an isomorphism for l 6= char k H i−1 ´et (K, Ql/Zl(n)) ≃ K M i−1(K) ⊗ Ql/Zl(n − i + 1) and the right hand side is divisible. (cid:3) A.4. Application: degeneration of the Bloch-Lichtenbaum spectral sequence. A.4.1. Theorem. Let K be as in Corollary A.1.4. Then the Bloch-Lichtenbaum spectral sequence [46, (1.8)] Ep,q 2 = H p−q(K, Z(−q)) ⇒ K−p−q(K) degenerates. For any n > 0, the map K M n (K) → Kn(K) is injective with divisible cokernel. Proof. By the Adams operations, the differentials are torsion [46, Th. 11.7]. By Theorem A.3.1, they vanish on the divisible groups Ep,q for p < 0. But H i(K, Z(n)) = 0 for i > n, 2 so Ep,q 2 = 0 for p > 0. The last statement follows from the degeneration plus Theorem A.3.1. (cid:3) MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 39 A.4.2. Remarks. 1) Again by the Adams operations, the filtration on Kn(K) induced by the Bloch-Lichtenbaum spectral sequence splits after inverting (n − 1)! for any field K. On the other hand, we constructed in [39] a canonical splitting of the corresponding spectral sequence with finite coefficients, including the abutment; the hypothesis that K contains a separably closed field is not required there. This implies in particular that the map K M n (K)/lν → Kn(K)/lν is split injective [39, Th. 1 (c)], hence bijective if K contains a separably closed subfield by Theorem A.4.1. Could it be that the mod lν splittings of [39] also exist integrally? 2) As in [39], Theorems A.3.1 and A.4.1 extend to regular semi-local rings of geometric origin containing a separably closed field; the point is that, for such rings R, the groups H i−1 ´et (R, Ql/Zl(n)) are divisible by the universal exactness of the Gersten complexes ([28], [18, Th. 6.2.1]). A.5. The map H i−1(K, Z(n − 1)) ⊗ K ∗ → H i(K, Z(n)). A.5.1. Theorem. Let K be as in Corollary A.1.4. Then, for i < n, (i) The cokernel of the cup-product map H 1(K, Z(n − i + 1)) ⊗ H i−1(K, Z(i − 1)) γi,n → H i(K, Z(n)) is uniquely divisible. (ii) The cokernel of the cup-product map H i−1(K, Z(n − 1)) ⊗ K ∗ δi,n → H i(K, Z(n)) is uniquely divisible. Proof. By (A.1), we may use the étale version of these groups. (i) Since H i ´et(K, Z(n)) is divisible by Theorem A.3.1, so is Coker γi,n. Let ν ≥ 1 and l prime 6= char K. The diagram H 1 ´et(K, Z(n − i + 1)) ⊗ H i−1 ´et (K, Z(i − 1)) γi,n −−−−→ H i ´et(K, Z(n)) H 0 ´et(K, Z/lν (n − i + 1)) ⊗ H i−1 ´et (K, Z(i − 1)) −−−−→ H i−1 ´et (K, Z/lν (n)) ∪ commutes, where β denotes Bockstein. The bottom horizontal map is surjective (even bijective) by the norm residue isomorphism theorem (resp. by [26]). By Theorem A.3.1 again, H 1 ´et (K, Z(i − 1)), and Coker γi,n is also l-torsion free by an easy diagram chase. ´et(K, Z(n − i + 1)) is l-divisible, hence so is H 1 ´et(K, Z(n − i + 1)) ⊗ H i−1 (ii) Consider the commutative diagram H 1 ´et(K, Z(n − i + 1)) ⊗ H i−2 ´et (K, Z(i − 2)) ⊗ K ∗ γi−1,n−1⊗1 −−−−−−−→ H i−1 ´et (K, Z(n − 1)) ⊗ K ∗ H 1 ´et(K, Z(n − i + 1)) ⊗ H i−1 ´et (K, Z(i − 1)) γi,n −−−−→ H i ´et(K, Z(n)). β⊗1x 1⊗∪y βx δi,ny 40 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) Since the left vertical map is surjective, we see that Coker δi,n is the quotient of Coker γi,n ´et (K, Z(n − 1)) ⊗ K ∗ (Theorem A.3.1), hence the by the image of the divisible group H i−1 claim follows from (i). (cid:3) It would be very interesting to describe Ker δi,n, but this seems out of range. References [1] J. D. Achter, S. Casalaina-Martin, and C. Vial. Distinguished models of intermediate jacobians. To appear in J. Inst. Math. Jussieu. DOI: https://doi.org/10.1017/S1474748018000245,. [2] J. D. Achter, S. Casalaina-Martin, and C. Vial. On descending cohomology geometrically. Compos. Math., 153(7):1446 -- 1478, 2017. [3] Y. André. Une introduction aux motifs (motifs purs, motifs mixtes, périodes), volume 17 of Panoramas et Synthèses. Société Mathématique de France, Paris, 2004. [4] F. Andreatta, L. Barbieri-Viale, and A. Bertapelle. Ogus realization of 1-motives. J. Algebra, 487:294 -- 316, 2017. [5] J. Ayoub. Note sur les opérations de Grothendieck et la réalisation de Betti. J. Inst. Math. Jussieu, 9(2):225 -- 263, 2010. [6] J. Ayoub. A guide to (étale) motivic sheaves. In Proceedings of the International Congress of Mathematicians -- Seoul 2014. Vol. II, pages 1101 -- 1124. Kyung Moon Sa, Seoul, 2014. [7] J. Ayoub. L'algèbre de Hopf et le groupe de Galois motiviques d'un corps de caractéristique nulle, I. J. Reine Angew. Math., 693:1 -- 149, 2014. [8] J. Ayoub. L'algèbre de Hopf et le groupe de Galois motiviques d'un corps de caractéristique nulle, II. J. Reine Angew. Math., 693:151 -- 226, 2014. [9] J. Ayoub. Une version relative de la conjecture des périodes de Kontsevich-Zagier. Ann. of Math. (2), 181(3):905 -- 992, 2015. [10] J. Ayoub and L. Barbieri-Viale. 1-motivic sheaves and the Albanese functor. J. Pure Appl. Algebra, 213(5):809 -- 839, 2009. [11] L. Barbieri-Viale and A. Bertapelle. Sharp de Rham realization. Adv. Math., 222(4):1308 -- 1338, 2009. [12] L. Barbieri-Viale and B. Kahn. On the derived category of 1-motives. Astérisque, 381:xi+254, 2016. [13] L. Barbieri-Viale, A. Rosenschon, and M. Saito. Deligne's conjecture on 1-motives. Ann. of Math. (2), 158(2):593 -- 633, 2003. [14] A. Bertapelle. Deligne's duality for de Rham realizations of 1-motives. Math. Nachr., 282(12):1637 -- 1655, 2009. [15] S. Bloch. Torsion algebraic cycles and a theorem of Roitman. Compositio Math., 39(1):107 -- 127, 1979. [16] J.-B. Bost. Algebraization, transcendence, and D-group schemes. Notre Dame J. Form. Log., 54(3- 4):377 -- 434, 2013. [17] J.-B. Bost and F. Charles. Some remarks concerning the Grothendieck period conjecture. J. Reine Angew. Math., 714:175 -- 208, 2016. [18] J.-L. Colliot-Thélène, R. T. Hoobler, and B. Kahn. The Bloch-Ogus-Gabber theorem. In Algebraic K-theory (Toronto, ON, 1996), volume 16 of Fields Inst. Commun., pages 31 -- 94. Amer. Math. Soc., Providence, RI, 1997. [19] J.-L. Colliot-Thélène and W. Raskind. K2-cohomology and the second Chow group. Math. Ann., 270(2):165 -- 199, 1985. [20] R. de Jeu and J. D. Lewis. Beilinson's Hodge conjecture for smooth varieties. J. K-Theory, 11(2):243 -- 282, 2013. With an appendix by Masanori Asakura. [21] P. Deligne. Théorèmes d'intégralité. In Groupes de monodromie en géométrie algébrique (SGA7), Vol. II, volume 340 of Lecture Notes in Mathematics, pages 363 -- 400. Springer-Verlag, Berlin-Heidelberg, 1973. Appendix to N. Katz Le niveau de la cohomologie des intersections complètes, exposé XXI. [22] P. Deligne. La conjecture de Weil. I. Publications Mathématiques de l'IHÉS, 43:273 -- 307, 1974. [23] P. Deligne. Théorie de Hodge. III. Publications Mathématiques de l'IHÉS, 44:5 -- 77, 1974. [24] P. Deligne. La conjecture de Weil. II. Publications Mathématiques de l'IHÉS, 52:137 -- 252, 1980. MOTIVIC PERIODS AND GROTHENDIECK ARITHMETIC INVARIANTS 41 [25] O. Gabber. Sur la torsion dans la cohomologie l-adique d'une variété. C. R. Acad. Sci. Paris Sér. I Math., 297(3):179 -- 182, 1983. [26] T. Geisser and M. Levine. The K-theory of fields in characteristic p. Invent. Math., 139(3):459 -- 493, 2000. [27] T. Geisser and M. Levine. The Bloch-Kato conjecture and a theorem of Suslin-Voevodsky. J. Reine Angew. Math., 530:55 -- 103, 2001. [28] D. R. Grayson. Universal exactness in algebraic K-theory. J. Pure Appl. Algebra, 36(2):139 -- 141, 1985. [29] A. Grothendieck. Le groupe de Brauer. II. Théorie cohomologique. In Séminaire Bourbaki : années 1964/65 1965/66, exposés 277-312, volume 9, pages 287 -- 307. Société mathématique de France, 1966. Exposé 297. [30] A. Grothendieck. On the de Rham cohomology of algebraic varieties. Publications Mathématiques de l'IHÉS, 29:95 -- 103, 1966. [31] C. Haesemeyer and C. A. Weibel. The norm residue theorem in motivic cohomology, volume 200 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, 2019. [32] A. Huber. Galois theory of periods, 2018. arXiv:1811.06268[math.NT]. [33] A. Huber and S. Müller-Stach. Periods and Nori motives, volume 65 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A, Series of Modern Surveys in Mathematics. Springer, Cham, 2017. With contributions by Benjamin Friedrich and Jonas von Wangenheim. linear and G. Wüstholz. Transcendence [34] A. Huber 2019. and relations of 1-periods, arXiv:1805.10104v4[math.NT]. [35] J.-P. Jouanolou. Cohomologie l-adique. In Cohomologie l-adique et Fonctions L, Séminaire de Géome- trie Algébrique du Bois-Marie 1965 -- 1966 (SGA 5), volume 589 of Lecture Notes in Mathematics, pages Exp. VI, 251 -- 281. Springer-Verlag, Berlin-New York, 1977. [36] B. Kahn. The Geisser-Levine method revisited. Preliminary version to [38], K-theory preprint archives 438, http://www.math.uiuc.edu/K-theory/0438/. [37] B. Kahn. La conjecture de Milnor (d'après V. Voevodsky). In Séminaire Bourbaki : volume 1996/97, exposés 820-834, number 245 in Astérisque, pages 379 -- 418. Société mathématique de France, 1997. Exposé 834. [38] B. Kahn. The Geisser-Levine method revisited and algebraic cycles over a finite field. Math. Ann., 324(3):581 -- 617, 2002. [39] B. Kahn. K-theory of semi-local rings with finite coefficients and étale cohomology. K-Theory, 25(2):99 -- 138, 2002. [40] B. Kahn. Some finiteness results for étale cohomology. J. Number Theory, 99(1):57 -- 73, 2003. [41] B. Kahn. Classes de cycles motiviques étales. Algebra Number Theory, 6(7):1369 -- 1407, 2012. [42] B. Kahn. The Brauer group and indecomposable (2, 1)-cycles. Compos. Math., 152(5):1041 -- 1051, 2016. [43] S. Lang. Transcendental points on group varieties. Topology, 1:313 -- 318, 1962. [44] F. Lecomte and N. Wach. Le complexe motivique de Rham. Manuscripta Math., 129(1):75 -- 90, 2009. [45] F. Lecomte and N. Wach. Réalisation de Hodge des motifs de Voevodsky. Manuscripta Math., 141(3- 4):663 -- 697, 2013. [46] M. Levine. K-theory and motivic cohomology of schemes, 1999. Preprint, https://faculty.math.illinois.edu/K-theory/0336/mot.pdf. [47] C. Mazza, V. Voevodsky, and C. Weibel. Lecture notes on motivic cohomology, volume 2 of Clay Math- ematics Monographs. American Mathematical Society, Providence, RI; Clay Mathematics Institute, Cambridge, MA, 2006. [48] A.S. Merkurjev and A.A. Suslin. K-cohomology of Severi-Brauer varieties and the norm residue homo- morphism. Izv. Akad. Nauk SSSR Ser. Mat., 46(5):1011 -- 1046, 1135 -- 1136, 1982. [49] A. Rosenschon and V. Srinivas. Étale motivic cohomology and algebraic cycles. J. Inst. Math. Jussieu, 15(3):511 -- 537, 2016. [50] A. Suslin and V. Voevodsky. Bloch-Kato conjecture and motivic cohomology with finite coefficients. In The arithmetic and geometry of algebraic cycles (Banff, AB, 1998), volume 548 of NATO Sci. Ser. C Math. Phys. Sci., pages 117 -- 189. Kluwer Acad. Publ., Dordrecht, 2000. 42 F. ANDREATTA, L. BARBIERI-VIALE AND A. BERTAPELLE (APPENDIX BY B. KAHN) [51] V. Voevodsky. Cancellation theorem. Doc. Math., Extra vol.: Andrei A. Suslin sixtieth birthday:671 -- 685, 2010. [52] V. Voevodsky. On motivic cohomology with Z/l-coefficients. Ann. of Math. (2), 174(1):401 -- 438, 2011. [53] V. Vologodsky. Hodge realizations of 1-motives and the derived Albanese. J. K-Theory, 10(2):371 -- 412, 2012. [54] M. Waldschmidt. Nombres transcendants et groupes algébriques, volume 69 of Astérisque. Société Math- ématique de France, Paris, 1979. [55] C. A. Weibel. Pic is a contracted functor. Invent. Math., 103(1):351 -- 377, 1991. [56] G. Wüstholz. Some remarks on a conjecture of Waldschmidt. In Diophantine approximations and tran- scendental numbers (Luminy, 1982), volume 31 of Progr. Math., pages 329 -- 336. Birkhäuser Boston, Boston, MA, 1983. [57] G. Wüstholz. Leibniz' conjecture, periods & motives. In Colloquium De Giorgi 2009, volume 3 of Colloquia, pages 33 -- 42. Ed. Norm., Pisa, 2012. Dipartimento di Matematica "F. Enriques", Università degli Studi di Milano, via C. Saldini, 50, Milano I-20133, Italy E-mail address: [email protected] E-mail address: [email protected] Dipartimento di Matematica "T. Levi-Civita", Università degli Studi di Padova, via Trieste, 63, Padova I-35121, Italy E-mail address: [email protected] IMJ-PRG, Case 247, 4 place Jussieu, 75252 Paris Cedex 05, France E-mail address: [email protected]
1612.09156	1	1612	2016-12-29T14:30:19	Mather-Jacobian multiplier ideals on toric varieties	[ "math.AG" ]	This paper provides a formula for the Mather-Jacobian multiplier ideals of torus invariant ideals on (not necessarily normal) toric varieties that generalizes Howald's formula for the multiplier ideal of monomial ideals in a polynomial ring.	math.AG	math	MATHER-JACOBIAN MULTIPLIER IDEALS ON TORIC VARIETIES HOWARD M THOMPSON Abstract. This paper provides a formula for the Mather-Jacobian multiplier ideals of torus invariant ideals on (not necessarily normal) toric varieties that generalizes HowaldâĂŹs formula for the multiplier ideal of monomial ideals in a polynomial ring. 1. Introduction In Section 2, we recall the definition of Nash blowup and the relationship between Nash blowups and Jacobian ideals. In Section 3 on the next page, we recall the definition of log- arithmic Jacobian ideals and the relationships between Nash blowups, logarithmic Jacobian ideals, and Jacobian ideals. In Section 4 on page 3, we recall the definition of Mather- Jacobian multiplier ideals and present a formula for the Mather-Jacobian multiplier ideals of torus invariant ideals on (not necessarily normal) toric varieties. 2. Nash Blowup and the Jacobian In this section we loosely follow de Fernex & Docampo [2, Section 2]. Let X be a separated reduced scheme of finite type over an algebraically closed field k of characteristic zero. The Jacobian ideal sheaf of X, JacX ⊆ OX , is the smallest non-zero Fitting ideal of the cotangent sheaf ΩX . If X is of pure dimension d, then JacX = Fittd(ΩX ) is the d-th Fitting ideal of ΩX . If Y is a smooth scheme and f : Y → X is a morphism of schemes, the Jacobian ideal of f is the 0-th Fitting ideal Jacf = Fitt0(ΩY /X ) of ΩY /X . If Y is equidimensional of dimension d, then Ωd Y is an invertible sheaf and the image of the map f ∗Ωd Y is Jacf · Ωd Y . X → Ωd Let X be a separated reduced scheme of finite type over k of pure dimesion d. The Nash blowup of X is the birational map π : bX → X satisfying the universal property: if µ : Y → X is a proper birational map such that µ∗Ω1 X /(torsion) is a locally free OY -module, then µ factors uniquely through π. Following Ein, Ishii & Mustaţă [3, Remark 2.3] and noticing that X need not be a variety to apply Lipman [6, Lemma 1], we see every log resolution of JacX factors through the Nash blowup of X. In fact, by Oneto & Zatini [7, Corollary 5.2], Date: November 21, 2018. 2010 Mathematics Subject Classification. Primary: 14F18; Secondary: 14M25. Key words and phrases. Multiplier ideals, toric varieties. 1 2 HOWARD M THOMPSON if JacY X is the restriction to X of the Jacobian of a locally complete intersection Y of dimension d containing X, then the Nash blowup is the blowup of JacY X. 3. The Jacobian and the Logarithmic Jacobian let S be an affine semigroup (finitely generated torsion-free cancellative From now on, monoid) and let M = ZS ∼= Zd be its quotient group. So, X = Spec k[S] is a (not nec- essarily normal) affine toric variety. We write m for an element of MR = M ⊗Z R and we write χm for the element of k[M] corresponding to m ∈ M. We have a k[S]-module homomorphism ϕ : Ω1 k[S]/k → k[S] ⊗Z M; ϕ(dχm) = χm ⊗ m. Definition 1. The logarithmic Jacobian ideal Jaclog X of X is the ideal such that the image of ∧dϕ is Jaclog X ⊗Z ∧dM. This ideal is generated by elements χm1+m2+···+md such that m1 ∧ m2 ∧ · · · ∧ md 6= 0. Evidently, it suffices to take the mi from some fixed generating set for S. According to González Pérez & Teissier [4, Proposition 60], the Nash blowup π : bX → X of X is the blowup of Jaclog X . Examining the proof of this proposition provides a convenient formula for the Jacobian of X. Fix a surjective monoid homomorphism φ : Nr → S. This surjection induces a short exact sequence of Abelian groups 0 / L / Zr φgp / M / 0 . For each element u ∈ Zr, write u = u+ − u− where u+ and u− are elements of Nr with disjoint support. The monoid homomorphism φ induces a k-algebra homomorphism φalg : k[x1, x2, . . . , xr] → k[S] with kernel IL = (cid:16)xu+ − xu− u ∈ L(cid:17). Set c = rank L = r − d, let e1, e2, . . . er be the standard basis in Nr, let 1 = Pr generating set nf1 = xu+ Let J be the r × n matrix h ∂fj ∂xii, let U be the r × n matrix with columns u1, u2, . . . , un, let K = {k1, k2, . . . , kc} ⊆ {1, 2, . . . , r}, and let L = {j1, j2, . . . , jc} ⊆ {1, 2, . . . , n}. González Pérez & Teissier prove following relationship between the c × c minors JKL and UKL. ei be the all ones vector, and fix a no for IL. 1 − xu− 1 , f2 = xu+ 2 − xu− 2 , . . . , fn = xu+ n − xu− i=1 (1) xek1 +ek2 +···ekc · JKL ≡ x u+ j1 +u+ j2 +···+u+ jc UKL mod IL. Now, fix 1 ≤ j1 < j2 < · · · < jc ≤ n such that 0 6= uj1 ∧ uj2 ∧ · · · ∧ ujc ∈ ∧cL. Let u+ j1 − x − u j1 , x u+ j2 − x u − j2 , . . . , xu+ jc − xu − jc(cid:17) ⊆ k[x] Then, Y = Spec (k[x]/I) is a reduced I =(cid:16)x complete intersection of dimension d containing X. After multiplying both sides of the congruence Eq. (1) by x1−(ek1 +ek2 +···ekc ), varying K, and considering the restriction to X, one obtains χφgp(1) · JacY X = χφgp(u+ j1 +u+ j2 +···+u+ jc ) · Jaclog X . / / / / MATHER-JACOBIAN MULTIPLIER IDEALS ON TORIC VARIETIES 3 Lemma 2. JacX is generated by (cid:26)χφgp(ei1 +ei2 +···+eid +u+ j1 +u+ j2 0 6= φgp(ei1) ∧ φgp(ei2) ∧ · · · ∧ φgp(eid) ∈ ∧dM; 0 6= uj1 ∧ uj2 ∧ · · · ∧ ujc ∈ ∧cL (cid:27) . +···+u+ jc −1) (cid:12)(cid:12)(cid:12)(cid:12) X ·(cid:16)χφgp(u+ j1 Proof. Continuing as above, allow L to vary as well. We deduce χφgp(1) · JacX = Jaclog +u+ j2 +···+u+ jc )(cid:12)(cid:12)(cid:12) 0 6= uj1 ∧ uj2 ∧ · · · ∧ ujc ∈ ∧cL(cid:17) . (cid:3) 4. Mather-Jacobian Multiplier Ideals Definition 3. Let a ⊆ OX be a nonzero ideal sheaf on a variety X, let µ : Y → X be a log resolution of a · JacX , let OY (cid:16)−bKY /X(cid:17) = Jacµ, let OY (−JY /X ) = JacX · OY , let OY (−ZY /X ) = a · OY , and let λ ∈ R≥0. The Mather-Jacobian multiplier ideal of a with exponent λ is bJ(X, a λ) = µ∗OY (cid:16)bKY /X − JY /X − ⌊λZY /X⌋(cid:17) . Now, we are ready to develop a formula for the Mather-Jacobian multiplier ideals of torus invariant ideals on (not necessarily normal) toric varieties. it suffice to consider the affine case. Let X = Spec k[S] be a (not necessarily normal) affine toric variety, let a ⊆ k[S] be a nonzero monomial ideal, let µ : XΣ → X be a toric log resolution of a · JacX . Lemma 4. If a ⊆ OX is a nonzero invariant ideal sheaf on a toric variety X and µ : XΣ → X is a toric log resolution of a · JacX , then Jacµ · Jaclog XΣ = µ∗Jaclog X and OXΣ(cid:16)bKXΣ/X(cid:17) = HomOXΣ(cid:16)µ∗Jaclog X , Jaclog XΣ(cid:17) . Proof. Consider the map µ∗Ωd X / Ωd XΣ µ∗(Jaclog X ⊗Z ∧dM) / Jaclog XΣ ⊗Z ∧dM with image Jacµ · Ωd bundles on XΣ, the result follows. XΣ = Jacµ · Jaclog XΣ ⊗Z ∧dM. Since µ∗Jaclog X , Jaclog XΣ and Jacµ are line (cid:3) / / 4 HOWARD M THOMPSON Recall the notation of Section 3, let a ⊆ k[S] be a nonzero monomial ideal, and consider the following subsets of MR. σ∨ = Conv(S) Jlog = {φgp(ei1 + ei2 + · · · + eid) φgp(ei1) ∧ φgp(ei2) ∧ · · · ∧ φgp(eid) 6= 0} Newt(cid:16)Jaclog X (cid:17) = Conv(Jlog + S) J′ = {φgp(u+ J = Jlog + J′ j1 + u+ j2 + · · · + u+ jc − 1) uj1 ∧ uj2 ∧ · · · ∧ ujc 6= 0} Newt (JacX ) = Conv(J + S) P = Newt(a) = Conv({m χm ∈ a}) Q = Conv(J′ + S) =nm m + Newt(cid:16)Jaclog X (cid:17) ⊆ Newt (JacX )o Let Sσ = σ∨ ∩ M. Thus, Xσ = Spec k[Sσ] is a normal toric variety and the canonical map Xσ → X induced by the inclusion S ⊆ Sσ is the normalization map. We say a map µ : XΣ → X from a normal toric variety is toric if if factors as a toric map XΣ → Xσ followed by the canonical map Xσ → X. Theorem 5. If a ⊆ OX is a nonzero invariant ideal sheaf on a toric variety X and µ : XΣ → X is a toric log resolution of a · JacX , then where int (Q + λP ) is the interior of the polyhedron Q + λP . λ) ⇔ m ∈ int (Q + λP ) , χm ∈bJ(X, a Proof. We will mimic the proof of Howald's Theorem [5] as it appears in Cox, Little & Schenck [1, Theorem 11.3.12]. Let Σ(1) be the set of rays of the fan Σ,let nρ be the primitive vector on the ray ρ for each ρ ∈ Σ(1), let Dρ be the prime divisor on XΣ for each ρ ∈ Σ(1), X = and let ZY /X = Pρ∈Σ(1) aρDρ. Exploiting the facts that Jaclog XΣ = Pρ∈Σ(1) Dρ, µ∗Jaclog Pρ∈Σ(1) minm∈Jloghm, nρiDρ, and JY /X =Pρ∈Σ(1) minm∈Jhm, nρiDρ along with Lemma 4 on hm, nρi − ⌊λaρ⌋(cid:19) Dρ. (cid:18)−1 + min the preceding page, we see hm, nρi − min m∈J m∈Jlog Thus, bKY /X − JY /X − ⌊λZY /X⌋ = Xρ∈Σ(1) χm ∈bJ(X, a λ) ⇔ hm, nρi ≥ 1 − min m∈Jlog ⇔ hm, nρi > − min m∈Jlog ⇔ hm, nρi > min m∈J′ hm, nρi + λaρ hm, nρi + min m∈J hm, nρi + ⌊λaρ⌋ for all ρ ∈ Σ(1) hm, nρi + min m∈J hm, nρi + λaρ for all ρ ∈ Σ(1) for all ρ ∈ Σ(1) Since the resolution µ : XΣ → X principalizes both Jaclog X and JacX , it also principalizes the fractional ideal (χm m ∈ J′). Therefore, every facet of Q+ λP is supported on a hyperplane of the form {m hm, nρi = bρ} for some ρ ∈ Σ(1) and bρ ∈ R. We conclude χm ∈bJ(X, a λ) ⇔ m ∈ int (Q + λP ) . MATHER-JACOBIAN MULTIPLIER IDEALS ON TORIC VARIETIES 5 (cid:3) References [1] D. A. Cox, J. B. Little, and H. K. Schenck, Toric varieties, Graduate Studies in Mathematics, vol. 124, American Mathematical Society, Providence, RI, 2011. MR2810322 ↑4 [2] T. de Fernex and R. Docampo, Jacobian discrepancies and rational singularities, J. Eur. Math. Soc. (JEMS) 16 (2014), no. 1, 165 -- 199, DOI 10.4171/JEMS/430. MR3141731 ↑1 [3] L. Ein, S. Ishii, and M. Mustaţă, Multiplier ideals via Mather discrepancy (2011), 17 pp., arXiv:1107.2192 [math.AG]. ↑1 [4] P. D. González Pérez and B. Teissier, Toric geometry and the Semple-Nash modification, Rev. R. Acad. Cienc. Exactas Fís. Nat. Ser. A Math. RACSAM 108 (2014), no. 1, 1 -- 48, DOI 10.1007/s13398-012-0096-0. MR3183106 ↑2 [5] J. A. Howald, Multiplier ideals of monomial ideals, Trans. Amer. Math. Soc. 353 (2001), no. 7, 2665 -- 2671 (electronic), DOI 10.1090/S0002-9947-01-02720-9. MR1828466 ↑4 [6] J. Lipman, On the Jacobian ideal of the module of differentials, Proc. Amer. Math. Soc. 21 (1969), 422 -- 426. MR0237511 ↑1 [7] A. Oneto and E. Zatini, Remarks on Nash blowing-up, Rend. Sem. Mat. Univ. Politec. Torino 49 (1991), no. 1, 71 -- 82 (1993). Commutative algebra and algebraic geometry, II (Italian) (Turin, 1990). MR1218672 ↑1 Mathematics Department, 402 Murchie Science Building, 303 East Kearsley Street, Flint, MI 48502-1950 E-mail address: [email protected] URL: http://homepages.umflint.edu/~hmthomps/
1908.04775	1	1908	2019-08-13T17:41:17	$p$-adic Integral Geometry	[ "math.AG", "math.MG", "math.NT", "math.PR" ]	We prove a $p$-adic version of the Integral Geometry Formula for averaging the intersection of two $p$-adic projective algebraic sets. We apply this result to give bounds on the number of points in the modulo $p^m$ reduction of a projective set (reproving a result by Oesterl\'e) and to the study of random $p$-adic polynomial systems of equations.	math.AG	math	P -- ADIC INTEGRAL GEOMETRY AVINASH KULKARNI AND ANTONIO LERARIO Abstract. We prove a p-adic version of the Integral Geometry Formula for averaging the intersection of two p-adic projective algebraic sets. We apply this result to give bounds on m reduction of a projective set (reproving a result by the number of points in the modulo p Oesterlé) and to the study of random p-adic polynomial systems of equations. 1. Introduction Integral Geometry deals with "averaging" metric properties (e.g. the volume) of the intersec- tion of two submanifolds of a homogeneous space under the action of a Lie group. This classical subject has a vast range of applications and connections to different areas of mathematics, in- cluding differential geometry [How93, San04, Ale03], representation theory [BF11, How93], convex geometry [SW08, SW93, Wei79], numerical anlysis [AB15, BC13], and random ge- ometry [EK95, BKL18]. The main goal of this paper is the adaptation of the ideas coming from this subject to the p-adic world. Before moving to the details, let us spend a few words discussing the classical framework. 1.1. Classical Integral Geometry. The main reference we have in mind for this section is the monograph [How93], where the spherical case is discussed; here we prefer to discuss the projective case as it offers a closer analogy with the results of this article. Let A, B ⊂ RPn be two compact submanifolds of dimension dim(A) = a and dim(B) = b. We endow the projective space with the quotient metric from the unit sphere Sn ⊂ Rn+1 and, restricting this metric to a submanifold Y of dimension dim(Y ) = k, we get a natural notion of Riemannian k-dimensional volume of Y , which we denote by volk(Y ). More generally, if Y is a codimension m− k submanifold of a Riemannian manifold M whose volume density is vol (1.1) and U (Y, ǫ) =Sx∈Y B(x, ǫ) denotes the ǫ-neighborhood of Y in M , then The volume of the projective space RPk ⊆ RPn with the quotient metric is volk(Y ) := lim ǫ→0 vol(BRm−k (0, ǫ)) vol(U (Y, ǫ)) . volk(RPk) = k+1 2 π 2 (cid:1) . Γ(cid:0) k+1 The group G = O(n+1) acts on the projective space by isometries, and by a standard transver- sality argument, the intersection A ∩ gB is transversal and of codimension (n − a) + (n − b), i.e. of dimension k = n − (n − a) − (n − b), for almost every g ∈ G. The Integral Geometry Formula [How93, Corollary 3.9] then tells that: (1.2) ZG volk(A ∩ gB) volk(RPk) dg = vola(A) vola(RPa) · volb(B) volb(RPb) , 2010 Mathematics Subject Classification. 53C65 (primary), 11S80 (secondary). Key words and phrases. Integral Geometry Formula, p-adic volume, zeros of random polynomials. 1 2 AVINASH KULKARNI AND ANTONIO LERARIO where the integral is with respect to the normalized Haar measureRG dg = 1, and the integrand is finite for almost every g ∈ G (by transversality). The identity (1.2) is very remarkable and has many interesting applications. The spirit of this paper is much influenced by one of these applications in particular, due to [EK95], where the idea of combining (1.2) with algebraic geometry and Example 1 first appeared. Example 1 (How many zeros of a random polynomial are real?). Let f ∈ R[x0, x1](d) be the random polynomial given by: (1.3) f (x0, x1) = ξ0xd 0 + ··· + ξks(cid:18)d k(cid:19)xd−k 0 xk 1 + ··· + ξdxd 1, k=0 is a family of i.i.d. where {ξk}d In [EK95] the authors use a nice argument from integral geometry to show that the expectation of the number of zeroes on RP1 of the random polynomial (1.3) is √d. The argument goes as follows. One first considers the Veronese embedding ν : RP1 → RPd given by: standard Gaussians. ν : [x0, x1] 7→"xd 0, . . . ,s(cid:18)d k(cid:19)xd−k 0 xk 1, . . . , xd 1# . Using the properties of the Gaussian distribution, one then proves that the average number of zeroes of (1.3) equals the average number of points of intersection of ν(RP1) with a randomly sampled hyperplane gRPn−1 ⊂ RPn, with g ∈ O(n + 1) drawn from a uniform distribution. Finally, by (1.2), this expectation equals (1.4) E#{f = 0} = vol1(ν(RP1)) vol1(RP1) = √d. In other words, by using the integral geometry formula, we see that the expectation of the number of zeros of a random polynomial with respect to a particular distribution is determined by the length of the corresponding Veronese curve. We observe, however, that computing the length of a Veronese curve is in general not trivial. In Theorem 36 we will prove the p-adic version of the results discussed in this example. 1.2. p-adic Integral Geometry. Keeping Section 1.1 in mind, let us discuss now the content of the current paper. We begin with the geometry of the ambient space. We denote by Qp the set field p-adic numbers, endowed with the standard p-adic norm, and by Zp the ring of p-adic integers. The vector space Qn+1 is endowed with the norm: p k(a0, . . . , an)kp = sup i aip, (a0, . . . , an) ∈ Qn+1 p . Together with the metric structure, we also consider the standard p-adic measure µ, defined on the Borel sets of Qn+1 and normalized by µ(Zn+1 ) = 1, see Section 2.4. p p p We also turn the p-adic projective space Pn into a metric measure space as follows. First we consider the Hopf fibration ϕ : Sn → Pn, where Sn is the open and compact subset of Qn+1 such that kakp = 1 and ϕ is the natural projection map. Unlike the real case, the p-adic unit sphere as a p-adic analytic manifold is not of dimension n, but of dimension n + 1 (see Figure 1). Nevertheless, we have decided to retain the same notation due to the strong parallels with the real case. consisting of points a = (a0, . . . , an) ∈ Qn+1 p The metric structure on Pn is then defined by: d(x, y) = kx ∧ ykp , (1.5) x, y ∈ Pn, P -- ADIC INTEGRAL GEOMETRY 3 where x ∈ ϕ−1(x) and y ∈ ϕ−1(y) are any two preimages and x ∧ y ∈ Λ2(Qn+1 measure, we take the normalized pushforward ϕ∗µ will still denote this measure by µ. ). For the 1+p−1 . When there is no risk of ambiguity we p The main ingredients to formulate an identity in the style of (1.2) are (1) the action on a topological space X of a topological compact group G endowed with a measure and (2) a notion of "volume" and "dimension" for subspaces of X. In this paper the group will be G = GLn+1(Zp), which acts by isometries and measure preserving transformations on X = Pn. For the sake of this introductory section we will take X = Pn, but the results we will present are also valid for the p-adic sphere Sn ⊂ Qn+1 (see Section 2.2). We endow GLn+1(Zp) with the normalized Haar measure: ZGLn+1(Zp) dg = 1. The class of submanifolds of Pn that we will consider consists of algebraic sets (but this condition can also be weakened to consider stratified Qp-analytic sets, see Section 2.2). In this case the dimension of an algebraic set A ⊆ Pn is the one coming from algebraic geometry; the notion of volume is instead more interesting. Denoting by B(x, r) the ball of radius r centered at x ∈ Pn (with respect to the metric (1.5)), for an algebraic set Y ⊆ Pn of dimension dim(Y ) = k we define its volume, based on the work of [Ser81] and [Oes82], by: p (1.6) volk(Y ) := lim m→∞ pm(n−k) · µ [x∈Y B(x, p−m)! . Notice that this definition mimics the classical definition given in (1.1). The volume of the p-adic projective space equals (via Section 2.4): volk(Pk) = 1 − p−(k+1) 1 − p−1 . The main theorem around which this paper revolves is Theorem 35, restated below. Theorem 2 (p-adic Integral Geometry Formula). Let A, B ⊂ Pn be algebraic sets of dimen- sions dim(A) = a, dim(B) = b. Then for almost all g ∈ GLn+1(Zp) the intersection A ∩ gB is transversal, of dimension k = n − (n − a) − (n − b), and: (1.7) volk (A ∩ gB) dg = . vola(A) vola(Pa) · volb(B) volb(Pb) volk (Pk) ZGLn+1(Zp) Notice the exact analogy between (1.2) and (1.7). Before presenting a couple of applications, let us commentate on the notion of volume (1.6) that we use in Theorem 2. First, since Y sits inside the metric space Pn, it is possible to define its Hausdorff dimension, which coincides with its dimension k = dim(Y ) as an algebraic set, and its k-dimensional Hausdorff measure turns out to be equal to its k-dimensional volume. Second, a result of Oesterlé [Oes82, Theorem 2] connects the number of congruence classes of points mod pm in Y and the volume of Y . Specifically, denoting by Nm(Y ) the number of points of Y in its modulo-pm reduction (see Definition 14), we prove that: volk(Y ) = lim m→∞ Nm(Y ) pmk . In fact, if Y is smooth, then the limit above "stabilizes" for large enough m. The aforementioned result, as well as the finiteness of the limit above, were both proven by Serre in [Ser81]. We give a simplified proof of Oesterlé's result, which we state as Corollary 25 and Corollary 28. 4 AVINASH KULKARNI AND ANTONIO LERARIO (−1, 1) + (3Z3)2 (0, 1) + (3Z3)2 (1, 1) + (3Z3)2 (−1, 0) + (3Z3)2 (0, 0) + (3Z3)2 (0, 0) + (Z3)2 (1, 0) + (3Z3)2 (−1, −1) + (3Z3)2 (0, −1) + (3Z3)2 (1, −1) + (3Z3)2 Figure 1. A depiction of the 3-adic unit circle S1 in Z2 an open ball of radius 1 unshaded cells is S1. The unit −1 ∈ Z× the origin. Note that S1 has non-zero measure inside Z2 ϕ−1(U ) of any non-empty open subset of P1. 3. Each cell represents 3 centered at the indicated point. The union of all the 3 acts by reflecting the diagram through 3, as well as the pullback Third, all this discussion on volumes and the p-adic Integral Geometry Formula is valid for open compact subsets of Y as well as for the subset of smooth points of Y . The volume of the smooth sublocus of Y and the volume of Y are equal, see Corollary 25 and Corollary 26. Fourth, by taking B ⊂ Pn to be a subspace of complementary dimension B = Pn−k, we see by the p-adic Integral Geometry formula that the volume of Y satisfies the identity: volk(Y ) = volk(Pk) ·ZGLn+1(Zp) #(cid:16)Y ∩ gPn−k(cid:17) dg. This last observation allows us to prove a slightly weaker version of a result of Oesterlé [Oes82, Theorem 1], though with a substantially simpler proof. Oesterlé's result provides an estimate for the number of points in the modulo-pm reduction of a Zp-algebraic set. Here again we state the projective version of this result. Corollary 3. Let Y ⊆ Pn be an algebraic set of dimension k and degree d. Then volk(Y ) ≤ d. Moreover, if volk(Y ) < d, then for m large enough: Nm(Y ) ≤ d · pmkvolk(Pk). Proof. Note that the integrand of 1.2 is bounded by the degree for almost all g ∈ GLn+1(Zp). If volk(Y ) < d, then the result follows immediately. (cid:3) Remark 4. The second conclusion of the result above holds under the hypothesis that Y is equidimensional. If Y is not the union of deg Y linear subspaces, then there is a smooth point y ∈ Y (Qp) such that the tangent space H := TyY intersects Y with multiplicity exactly 2 at y. By perturbing H, we can construct a linear subspace H ′ such that #(H ′ ∩ Y ) < d − 2. More precisely, such that at least two points of H ′ ∩ Y are defined over a ramified extension (1.8) g(t) := ζ0 + ζ1(cid:18)t 1(cid:19) + . . . + ζd(cid:18)t d(cid:19), P -- ADIC INTEGRAL GEOMETRY 5 of Qp. In particular, the interior of the subset is non-empty, and thus we have that volk(Y ) < d. The claim then follows from Corollary 3. U := {g ∈ GLn+1(Zp) : #(Y ∩ gPn−k) < d}. Conversely, the hypothesis that Y is equidimensional is necessary. For instance, for Y the union of d lines and a points in P2, then Nm(Y ) over-estimates the volume for almost all m. 1.3. Applications. We conclude this introductory section with an application of the p-adic Integral Geometry Formula to random p-adic polynomials, in a spirit similar to Example 1 above. In the paper [Eva06], Evans considers the problem of studying the expectation of the number of zeros in Zn p of a system of random p-adic polynomials. The distribution that Evans considers comes from picking Gaussian p-adic coefficients in front of the Mahler basis (see [Eva06] for more details on the notion of p-adic gaussian variable). When n = 1, this means considering the following random polynomial: where {ζk}d k=0 is a family of i.i.d. uniform variables in Zp. Evans computed the expected number of zeroes of g in Zp; using our techniques we recover and extend this result, computing the expectation of the number of zeroes on the whole p-adic line. More precisely, in Section 5.2 we prove the following result. Theorem 5. Let g(t) be the random p-adic polynomial (1.8). (1) The expected number of zeroes in Zp of g equals p⌊logp d⌋ (2) For every m ≥ 1 the expected number of zeroes of g in the annulus 1 1+p−1 (this is Evans result). pm Zp\ 1 pm−1 Zp equals dp pm 1−p−1 1+p−1 and in particular: E#{g = 0} = p⌊logp d⌋ 1 + p−1 +Xm≥1 dp pm 1 − p−1 1 + p−1 = p⌊logp d⌋ + dpp−1 1 + p−1 . In this paper we also consider a different model that seems more natural over the projecive space, due to its invariance under the group GLn+1(Zp). In the case n = 1, we define the random p-adic polynomial: f (x0, x1) = ζ0xd 0 + ··· + ζkxd−k (1.9) where {ζk}d k=0 is a family of i.i.d. uniform variables in Zp. This distribution is invariant under GL2(Zp)-change of variables -- there are no preferred points or directions in the projective line P1. For the random polynomial (1.9) we prove the following (surprising) result. Proposition 6. The expected number of zeroes of the polynomial (1.9) in P1 is 1. Moreover the density of zeroes on P1 is uniform. 1 + ··· + ζdxd 1, 0 xk In fact, we prove a much more general theorem which relates the expectation of the number of zeroes of a system of random p-adic equations with the volume of the image of Pn under an appropriate "Veronese" map (as in [EK95]). In the case of the polynomial (1.9) the Veronese map to consider is ν : P1 → Pd given by [x0, x1] 7→ [xd 1] and, in analogy with (1.4), we have: vol1(ν(P1)) 0,··· , xd−k 1, . . . , xd 0 xk E#{f = 0} = vol1(P1) = 1. Once again, the computation of the length of the Veronese curve is an essential step to obtain the result. For us, this is given in Proposition 38. 6 AVINASH KULKARNI AND ANTONIO LERARIO Acknowledgements. The authors want to thank the Max Planck Institute for Mathemat- ics in the Sciences (MiS) in Leipzig, for the stimulating atmosphere where this work was done. Special thanks also to Paul Breiding, Peter Bürgisser, and Bernd Sturmfels for helpful comments and valuable discussions. 2.1. Notation. 2. Preliminaries ·p -- The p-adic absolute value, with normalization pp = p−1. k·kp -- The norm of a vector in Qn+1 Rm -- The ring Z/pmZ of integers modulo pm. Pn -- Projective space over Qp p B(a; p−m) -- The ball in a metric space centered at a of radius p−m Nm(U ) -- The minumum number of balls of radius p−m needed to cover U (alternatively: the number of points in the modulo pm reduction of U ) Sn -- The unit sphere in Qn+1 p Jf (a)p -- The absolute value of the determinant of the Jacobian matrix for the polynomials f . (In the projective case, normalized by the norm of a.) ϕ -- The Hopf fibration ϕ : Sn → Pn 2.2. Affine and projective sets and their stratification. A Qp-analytic manifold is de- fined analogously to a real or complex analytic manifold. i.e, in terms of charts and gluing maps. For the reader interested in the technical details, we refer to [Sch11, Part II]. Similarly, if Qp is an algebraic closure of Qp, there is the corresponding notion of Qp-analytic manifold. For a smooth projective algebraic variety X defined over Qp, the set of Zp-points of X has the structure of a Qp-analytic manifold. It is necessary for us to consider algebro-geometric objects whose set of Zp-points does not have the structure of a Qp-analytic manifold. Definition 7. A stratified Qp-analytic set is a set X which admits a decomposition into finitely many subsets (called strata) X := Xi kai=1 such that each Xi is a Qp-analytic manifold. The dimension of X is the maximal dimension of the strata. If X is empty we consider the empty coproduct and we set dim(X) = −∞. The previous notion of stratified set is weaker than the usual notion coming from semialge- braic geometry, singularity theory or algebraic geometry, but all we need in this paper is to be able to decompose the sets we are working with into finitely many disjoint pieces (the strata) each one of which is a Qp-analytic manifold. Note that for a stratified Qp-analytic set, the pieces need not be open or closed submanifolds of X, nor is the decomposition unique. Our central category of interest is the category of (analytically) open compact subsets of algebraic sets. At this point, it is conveinent to define some notation we will use throught the paper. Definition 8. Let X ⊆ Qn point. Let f := (f1, . . . , fr) be local equations for X at a. Then we define p be an algebraic set of codimension r, and let a ∈ X be a smooth Jf (a)p := sup I nMf ,I (a)p : Mf ,I is an r × r minor of Jf (a)o . P -- ADIC INTEGRAL GEOMETRY 7 Let X ⊆ Pn be a projective algebraic set of codimension r, let a ∈ X is a smooth point, and let f := (f1, . . . , fr) be local equations for X at a. With δ :=Pr i=1(deg(fi) − 1), we define : Mf ,I is an r × r minor of Jf (a)) . Jf (a)p := sup I (Mf ,I (a)p kakδ p Note in the projective definition that Jf (a)p is independent of the choice of homogeneous coordinates of a. However, both definitions are sensitive to the choice of defining equations. Remark 9. If U ⊆ X is an open compact subset of the smooth subset of X, then there is a finite set of points a1, . . . , ak and open subsets U1, . . . , Uk of U such that each X ∩ Ui is defined by the local equations at ai and such that U =Sk i=1 Ui. In particular, the function a 7→ sup1≤i≤k Jfi(a)p is bounded away from 0 on U . Proposition 10. Let X ⊆ Qn Proof. Write X := Z(f1, . . . , fr). Let Qp denote an algebraic closure of Qp, and let X(Qp) be the common zero locus of f1, . . . , fr over Qp. We may assume that f1, . . . , fr generate a radical ideal. The singular sublocus Y (Qp)) = Sing(X(Qp)) of X(Qp) is defined by the vanishing of the maximal minors of the Jacobian matrix, and so has positive codimension in X(Qp). We may write X(Qp) := (X(Qp)\Y (Qp))∪ Y (Qp), and (X(Qp)\Y (Qp)) is a Qp-analytic manifold. Iterating, we obtain a decomposition p be an algebraic set. Then X is a stratified Qp-analytic set. X(Qp) := Xi(Qp). Note that a point in X(Qp) is smooth in the algebraic sense if and only if it is also a smooth point of X(Qp). As algebraic smoothness implies analytic smoothness, the result follows by considering the decomposition (with some strata possibly empty over Qp): X = X(Qp) = Xi(Qp) (cid:3) kai=1 kai=1 nar=0 Qr p Remark 11. Note that Pn can be written as Pn := and that this is a stratification. In particular, if X is any projective algebraic set then each p ∩ X) is a stratified Qp-analytic set. Alternatively, by restricting to affine local patches, we see the Qp-points of any separated scheme of finite type over Qp is also a stratified Qp-analytic set, and the assignment X 7→ X(Qp) is functorial. However, r=0(Rm)r when m > 1, since Rm has a nontrivial ideal. component of X =`r=0(Qr Pn(Rm) is not equal to`n For any projective variety X over Zp, we have that X(Zp) = X(Qp). There is a natural map defined by ϕ : Sn (x0, . . . , xn) → Pn(Zp) 7→ [x0, . . . , xn] which is well-defined since (0, . . . , 0) 6∈ Sn. Plainly, we have that ϕ is a surjective, Qp-analytic, and hence measurable map. We have ϕ−1{[x0, . . . , xn]} = {(λx0, . . . , λxn) : λ ∈ Z× p }. Inspired by the real and complex settings, we give this map the name: Definition 12. We call the map ϕ the Hopf fibration. 8 AVINASH KULKARNI AND ANTONIO LERARIO 2.3. Modulo pm reduction. We now discuss the relations of our models and their reductions modulo pm. Denote Sn(Rm) := {(x0, . . . , xn) : xi ∈ Rm and at least one xi ∈ R× m}. As the next result indicates, reduction modulo pm commutes with projectivizing coordinates. Proposition 13. Let m > 0 be an integer, let U ⊆ Pn be any subset. The following diagram (of continuous functions) commutes: ϕ ϕm U πm πm(U ) ϕ−1(U ) bπm bπm(ϕ−1(U )) The map ϕm is defined by restricting the natural map ϕm : Sn(Rm) → Pn(Rm) 7→ [x0, . . . , xn]. (x0, . . . , xn) Moreover, for any x ∈ Pn(Rm), we have (2.1) #ϕ−1 m (x) = #R× m = pm(cid:18)1 − 1 p(cid:19). Proof. By definition, any x ∈ U has at least one coordinate not in pZp. Since πm(Z× for all m, the result follows. Definition 14 (Number of points modulo pm). If U ⊆ A ⊆ Zn set, we define Nm(U ) = #πm(U ). we define Nm(U ) = #bπm(U ). In the projective case, if U ⊆ A ⊆ Pn is a subset of an algebraic Example 15. Proposition 13 allows us to determine Nm(Pn) easily. We have that p is a subset of an algebraic set, p ) = R× m (cid:3) #Sn(Rm) = (#Rm)n+1 − (#pRm)n+1 = pm(n+1) − p(m−1)(n+1), as the maximal ideal of Rm is generated by p. In particular, using (2.1), we get: #Pn(Rm) = pm(n+1) − p(m−1)(n+1) demonstrates why P(Rm) is generally not equal to`n pm − pm−1 pmn − pm(n−1)−1 . = 1 − p−1 r=0(Rm)r. Notice that this number is not equal to #(Rm)n + . . . + #Rm + 1 in general. This explicitly 2.4. Measures and metrics. On projective space, we obtain a naturally defined quotient measure and quotient metric from the unit sphere (see for instance [Cho00]). Definition 16. We define the (normalized) pushforward measure on projective space by µ(U ) := 1 p ) · µ(ϕ−1(U )). µ(Z× Definition 17. Define the metric d(x, y) for x, y ∈ Pn by d(x, y) := kx ∧ ykp (2.2) for any x ∈ ϕ−1(x) and y ∈ ϕ−1(y). P -- ADIC INTEGRAL GEOMETRY 9 The metric in Definition 17 is the natural quotient metric, and any subset A ⊂ Pn can be turned into a metric space by restricting the metric to A. Our normalization of the pushforward measure ensures that for any x ∈ Pn and m > 0, the ball B(x; p−m) ⊆ Pn has measure p−mn. When m = 0, we have B(x; 1) = Pn for any x ∈ Pn, so µ(Pn) = (1 − p−1)−1µ(Sn) = . 1 − p−(n+1) 1 − p−1 2.5. Hensel's lemma. One of the main ideas in our arguments of Section 4 is to locally approximate the smooth sublocus of a stratified p-adic algebraic set by its tangent cone. In this section, we review the technical results needed to implement this idea. Lemma 18 (Hensel's lemma). Let f = (f1, . . . , fn) ∈ Zp[x1, . . . , xn]n, let m be a positive integer, let a ∈ Zn p, then there is a unique α ∈ Zn Proof. See Keith Conrad's notes [Con] (unpublished) or [Fis97, Theorem 1] (published). (cid:3) p , and let Jf (a) be the Jacobian matrix of f at a. If kf (a)kp < Jf (a)2 p such that f (α) = 0 and α ≡ a (mod p)m. The p-adic analytic implicit function theorem is the second critical tool we need. The formulation below is due to Igusa [Igu00, Theorem 2.2.1]. Theorem 19 (Implicit Function Theorem). Denote ZpJx, yK = ZpJx1, . . . , xr, y1, . . . , yn−rK the ring of formal power series in the variables x1, . . . , yn−r with coefficients in Zp. Then: (1) If Fi(x, y) is in ZpJx, yK and Fi(0, 0) = 0 for all i and further ∂(F1, . . . , Fn−r) ∂(y1, . . . , yn−r) (0, 0) 6≡ 0 mod p, then every fi(x) in the unique solution f = (f1, . . . , fn−r) of F (x, f (x)) = 0 satisfying fi(0) = 0 is in ZpJxK. (2) If a ∈ B(0; p−1), then f (a) ∈ Zn−r p and F (a, f (a)) = 0. Furthermore, if (a, b) ∈ pZr p × pZn−r p satisfies F (a, b) = 0, then b = f (a). The statement of part (2) is slightly modified from Igusa [Igu00, Theorem 2.2.1], but is equivalent (see [Igu00, p22]). In our study of embedded p-adic algebraic sets, we can give an explicit quantitative version of the Implicit Function Theorem. Proposition 20. Let X ⊆ Qn p be an algebraic set and let a ∈ X be a smooth point. Let f be a set of r local generators for I(X) ∩ Zp[x1, . . . , xn] at a. Let m ∈ N be such that Jf (a)2 p 6≡ 0 (mod pm). Then there exists a bianalytic isometry π : (X ∩ B(a; p−m)) → (TxX ∩ B(a; p−m)), which in a suitable change of coordinates is given by a projection map. Proof. We may assume up to translation that a is the origin. Note that we may also act on the equations by taking GLr(Zp)-linear combinations and act by GLn(Zp)-changes of coordinates on X. Thus, by using the Smith normal form of a matrix over Zp, we may assume that Jf (0) = σ1 . . . . . . 0 . . . σr  with σ1p ≥ . . . ≥ σrp. We consider the new system of equations given by Fj(x) := σ−1 j σ−1fj(y) with y = (σx1, . . . , σxn) and σ := σ1 . . . σr. Observe: 10 AVINASH KULKARNI AND ANTONIO LERARIO -- The constant term of each Fj is zero. -- The linear term of Fj is xj, as the factors of σj, σ in the linear term cancel. -- The higher order terms are integral, as σjσp ≥ σ2 p for all j. Thus, each Fj ∈ Zp[x], and the Jacobian matrix for the equations {F1, . . . , Fr} is(cid:2) I We now apply the Implicit Function Theorem; denoting xtail := (xr+1, . . . , xn), we obtain 0 (cid:3). power series g1, . . . , gr ∈ ZpJxr+1, . . . , xnK such that formally Fj(g1(xtail), . . . , gr(xtail), xr+1, . . . , xn) = 0. Furthermore, on the ball B(0; p−1) the gj are convergent and we have for all b ∈ B(0; p−1) that F (a, b) = 0 implies a = g(b). Finally, since the last n − r coordinates are identified with the tangent space of X at 0, we have that each gj is at least quadratic in the xi. Thus, each gj is Lipschitz on B(0; p−1) with Lipschitz constant 1. In particular, the map γF : B(0, p−1) → x Z(F1, . . . , Fr)(Zp) 7→ (g1(xtail), . . . , gr(xtail), xr+1, . . . , xn) is an isometry, whose inverse is the projection. Since each gj is at least quadratic, we have that each σ−1gj(xtail) is Lipschitz on B(0;σp) with Lipschitz constant 1. Denoting g♭ j(x) := σ−1g(x) we have the isometry γf : B(0,σp) → x Z(f1, . . . , fr)(Zp) 7→ (g♭ 1(xtail), . . . , g♭ r(xtail), xr+1, . . . , xn) whose inverse is the projection. Finally, since p−m ≤ Jf (0)2 p ≤ σp, we are done. (cid:3) Note that Proposition 20 holds for any choice of local equations for the algebraic set X. It is useful to fix some global set of defining equations with nice properties. For instance, we use the following elementary lemma. Lemma 21. Let X ⊆ Pn be an algebraic set, and let I(X) ⊆ Qp[x0, . . . , xn] be the homogeneous ideal of equations vanishing on X. Then I(X) ∩ Zp[x0, . . . , xn] is a homogeneous ideal of Zp[x0, . . . , xn] saturated with respect to hpi. Proof. If f ∈ Zp[x0, . . . , xn] such that pmf ∈ I(X) for some m ≥ 0, then f vanishes at every x ∈ X(Qp), so by definition f ∈ I(X). (cid:3) 3. Volumes 3.1. The volume of an algebraic set. Definition 22 (The volume of an algebraic set). Let U ⊂ A ⊆ Zn subset of an algebraic set A of dimension a. We define the a-dimensional volume of U as: B(x, p−m)! . vola(U ) = lim m→∞ pm(n−a) · µn [x∈U (3.1) p be an open and compact In a similar way, if U ⊆ A ⊆ Pn is an open and compact subset of an algebraic set A of dimension dim(A) = a, we define its a-dimensional projective volume by (3.1), replacing µn with the pushforward measure on Pn and taking balls with respect to the projective metric. Moreover, if A ⊆ Sn−1 ⊂ Qn p , again we define the volume of U ⊆ A by (3.1). P -- ADIC INTEGRAL GEOMETRY 11 For (open compact subsets of) algebraic sets, Serre [Ser81, pp148-149] observed that the limit in Definition 22 exists and is finite. He moreover observed, via an inductive argument based on resolution of singularities, that the volume of an algebraic set is a rational number. The result also appears in an article by Oesterlé [Oes82, Theorem 2], though Oesterlé's proves that the limit exists by more elementary means. Oesterlé also proves the following: Proposition 23. The volume of an open and compact subset U ⊆ A of an algebraic set (both affine and projective) coincides with its dim(A)-Hausdorff measure. When U ⊆ Zn p is an open set of full dimension we have voln(U ) = µn(U ). Additionally, the volume of U ⊆ A ⊆ Pn and the volume of the (dim(A) + 1)-dimensional set ϕ−1(U ) ⊆ ϕ−1(A) ⊆ Sn are related by: (3.2) vola+1(ϕ−1(U )) vola+1(ϕ−1(U )) vola(U ) = = . vol1(Z∗ p) 1 − 1 p 3.2. Volume and number of points in the modulo pm reduction. Lemma 24. We have the following two identities: (1) Let U ⊆ A ⊆ Zn p be an open and compact subset of an algebraic set A of dimension dim(A) = a. Then Nm(U ) is the minimum number of affine balls of radius p−m that we need to cover U . (2) Let U ⊆ A ⊆ Pn be an open and compact subset of an algebraic set A of dimension dim(A) = a. Then Nm(U ) is the minimum number of projective balls of radius p−m that we need to cover U . Moreover, we have that Nm(U ) = Nm(ϕ−1(U )) p(cid:17) pm(cid:16)1 − 1 Proof. Both results follow from the definition of Nm(U ) and Proposition 13. (cid:3) Corollary 25. We have the following two identities. (1) Let U ⊆ A ⊆ Zn dim(A) = a. Then: p be an open and compact subset of an algebraic set A of dimension vol(U ) = lim m→∞ Nm(U ) pma . (2) Let U ⊆ A ⊆ Pn be an open and compact subset of an algebraic set A of dimension dim(A) = a. Then: vol(U ) = lim m→∞ Nm(U ) pma . Proof. In the affine case, using Lemma 24 we cover U with Nm(U ) disjoint balls of radius p−m and centered at points x1, . . . , xNm(U ) ∈ U . We have: vol(U ) = lim m→∞ = lim m→∞ pm(n−a) · µ [x∈U Nm(U ) pma . B(x, p−m)! = lim m→∞ pm(n−a) Nm(U )Xj=1 µ(cid:0)B(xj, p−m)(cid:1) 12 AVINASH KULKARNI AND ANTONIO LERARIO In the projective case, we use the result in the affine case to see that vol(U ) = vol(ϕ−1(U )) 1 − 1 p = lim m→∞ Nm(ϕ−1(U )) pm(a+1)(cid:16)1 − 1 m→∞ p(cid:17) = lim Nm(U ) pma . (cid:3) Corollary 26. Let ϕ−1(A) = X ⊂ Sn be the pullback of an algebraic set of dimension a. De- note by sm(X) the set of analytically smooth points of X in Zn+1 . Then vol(X) = vol(sm(X)). p Proof. Let Z be the singular locus of A. Since the dimension of Z is strictly smaller than dim(X) = a + 1, we have lim m→∞ p−m(a+1)Nm(X) = lim m→∞ = lim m→∞ p−m(a+1)(cid:0)Nm(sm(X)) + Nm(ϕ−1(Z))(cid:1) p−m(a+1)Nm(sm(X)). (cid:3) 3.3. Quantitative estimates and the Weil canonical measure. For an embedded p-adic analytic set, we can enhance [Oes82, Theorem 2] and give explicit quantitative estimates to compute the limit in Definition 25. Our explicit approach enables us to compute the volume of some specific algebraic sets in Section 5 (see Proposition 38 and Corollary 41). Proposition 27. Let X ⊂ Zn point. Then there exists mx > 0 such that for all ℓ1 ≥ ℓ2 ≥ mx we have: p be an algebraic set of dimension d and x ∈ X be a smooth Nℓ1(X ∩ B(x; p−ℓ2)) = p(ℓ1−ℓ2)d Furthermore, if f is a set of local equations for X at x, then any mx ≥ − logp Jf (x)2 Proof. By Proposition 20 there is a bianalytic isometry p suffices. π : X ∩ B(x; p−mx ) → TxX ∩ B(x; p−mx) p. The result holds for the image of π, so the claim follows. (cid:3) where mx = − logp Jf (x)2 Corollary 28. Let X ⊂ Zn then there exists m0 = m0(X) > 0 such that for all m ≥ m0 we have p be a compact open set of a smooth algebraic set of dimension d, Nm(X) = p(m−m0)dNm0 (X). Proof. Using Proposition 27 and compactness of X, we see that there exists m0(X) such that for every x ∈ X and every ℓ2 ≥ ℓ1 ≥ m0(X), we have Nℓ2(B(x, p−ℓ1)∩X) = p(ℓ2−ℓ1)d. Covering X with Nm0(X) disjoint balls of the form B(xi, p−m0), with each xi ∈ X, we see Nm(X) = Nm0 (X)Xi=1 Nm(B(xi, p−m0 ) ∩ X) = Nm0 (X)Xi=1 p(m−m0)d = Nm0(X)p(m−m0)d. (cid:3) As a first application of our quantitative estimates, we have: Theorem 29. Let X be a subscheme of Pn which is smooth over Spec Zp. Then the Weil canonical volume of X = X (Zp) is equal to the volume as defined above. Proof. The Weil canonical volume of X is #X(Fp) [Wei82, Theorem 2.2.5]. However, we also have that the Jacobian matrix of the defining equations is non-zero modulo p at any point in X, so the limit in Corollary 25 stabilizes in the first term. This proves the result. (cid:3) p P -- ADIC INTEGRAL GEOMETRY 13 4. The p-adic integral geometry formula Lemma 30. Let x ∈ Sn be a point, let B ⊆ Sn be a ball of radius p−m containing x, with m > 0. Finally, let U :=Sa∈Z× p acts in the usual way. Then the subgroup Fix(GLn+1(Zp), U ) := {g ∈ GLn+1(Zp) : gU = U} aU , where Z× p is a finite index subgroup of GLn+1(Zp) containing Z× p . Proof. The action of GLn+1(Zp) on Sn commutes with reduction to Rm. The result is imme- diate since GLn+1(Rm) is a finite group. (cid:3) The following technical lemma is simply the result that if X, Y are random linear subspaces of complementary codimension, the probability densities P ({x} = X ∩ Y x ∈ X), P ({y} = X ∩ Y y ∈ Y ) are uniform on X and Y (resp.), and furthermore are independent. Lemma 31. Let X, Y, H ⊆ Pn be linear varieties such that codim X +codim Y +codim H = n. Let Ux ⊆ X, Uy ⊆ Y be relatively open balls (i.e, open balls of the subspaces X, Y ). Then ZGLn+1(Zp)ZGLn+1(Zp) #(gxUx ∩ gyUy ∩ H)dgydgx = vol(Ux) vol(X) · vol(Uy) vol(Y ) . Proof. The strategy of the proof is to replace the integrand with a constant function by using the fact that X (resp. Y ) is the disjoint union of finitely many copies of Ux (resp. Uy). Let GX := Fix(GLn+1(Zp), X) and let GY = Fix(GLn+1(Zp), Y ). Furthermore, let A := Fix(GX , Ux) and B := Fix(GY , Uy). Note that GX acts on X through GLdim X+1(Zp), so we have the commutative diagram GX GLdim X+1(Zp) . Fix(GX , Ux) Fix(GLdim X+1(Zp), Ux) In particular, A has finite index in GX by the previous lemma. Denote a := [GX : A], b := [GY : B] and let gx,1, . . . , gx,a and gy,1, . . . , gy,b be coset representatives for A ⊆ GX and B ⊆ GY respectively. j=1 gx,jUx. As GX acts transitively on the balls in H of a fixed radius, #(gxgx,jUx ∩ gygy,iUy ∩ H)dgydgx. bXi=1ZGXZGY #(gxgx,jUx ∩ gygy,iUy ∩ H)dgydgx In particular, if #(X ∩ Y ∩ H) = 1, we have Note that X =`a we have for any i, j that ZGX abZGXZGY #(gxUx ∩ gyUy ∩ H)dgydgx = #(gxUx ∩ gyUy ∩ H)dgydgx =ZGX aXj=1 =ZGXZGY =ZGXZGY =ZGXZGY #(gxX ∩ gyY ∩ H)dgydgx #(X ∩ Y ∩ H)dgydgx dgydgx 14 AVINASH KULKARNI AND ANTONIO LERARIO We now write GLn+1(Zp) ∼= GX × (GX\ GLn+1(Zp)) as a product measure space, with cor- responding differential dgx = dsxdtx, using the orbit-stabilizer theorem. Thus, by Fubini's theorem we have #(gxUx ∩ gyUy ∩ H)dgydgx abZGLn+1(Zp)ZGLn+1(Zp) = abZGX \ GLn+1(Zp)ZGXZGY \ GLn+1(Zp)ZGY = abZGX \ GLn+1(Zp)ZGY \ GLn+1(Zp)ZGXZGY =ZGX \ GLn+1(Zp)ZGY \ GLn+1(Zp)(cid:18)abZGXZGY #(gxUx ∩ gyUy ∩ H)dsydtydsxdtx #(gxUx ∩ gyUy ∩ H)dsydsxdtydtx #(gxUx ∩ gyUy ∩ H)dsydsx(cid:19) dtydtx Since the set of (gx, gy) in GLn+1(Zp) where gxX, gyY, H do not meet transversely has measure zero, we have that our expression above equals ZGX \ GLn+1(Zp)ZGY \ GLn+1(Zp)ZGXZGY dsydsxdtydtx = 1. Finally, since the disjoint union of a copies of Ux is X, we have that 1 Corollary 32. Let X, H ⊆ Pn be linear varieties such that codim X + codim H = n. Let Ux ⊆ X be a relatively open ball. Then a = vol(Ux) vol(X) (cid:3) ZGLn+1(Zp) #(gxUx ∩ H)dgx = vol(Ux) vol(X) . Proof. Let Y := Pn and Uy := Y . By Lemma 4.1 we have that But we see that the inner integral is constant in gy and equal to #(Ux∩H) since gyUy = Pn. (cid:3) We define some notation to state the next lemma. Let f1, f2 ⊆ Zp[x0, x1, . . . , xn] be sets of homogeneous polynomials of size n − r, n respectively, and let x, y ∈ Pn. We define J (f1(x), f2(y)) :=(cid:20)Jf1(x) Jf2(y)(cid:21) and J (f1(x), f2(y))p = det J (f1(x), f2(y))p kxkδ1 p kykδ2 p deg f . Additionally, if g ∈ GLn+1(Zp) and f = (f1, . . . , fn−r), we denote where δj = Pf ∈fj f g := (f1 ◦ g, . . . , fn−r ◦ g). Lemma 33 (Linear Approximation Lemma). Let X, Y ⊆ Pn be algebraic sets of complemen- tary codimension. Let x ∈ X, y ∈ Y be smooth points, contained in open balls Ux, Uy of Pn of radius p−m, with local equations fx, fy (respectively). If J (fx(x), fy(y))2 or then #(X ∩ Ux ∩ Y ∩ Uy) = #(TxX ∩ Ux ∩ TyY ∩ Uy). Proof. If Ux ∩ Uy = ∅, we have #(X ∩ Ux ∩ Y ∩ Uy) = #(TxX ∩ Ux ∩ TyY ∩ Uy) = 0. Ux ∩ Uy = ∅, p > p−m Otherwise, note that Ux, Uy are balls of radius p−m with nontrivial intersection, so Ux = Uy. In particular, πm(x) = πm(y). Choosing representatives on Sn for x, y, we have J (fx(x), fy(y)) ≡ J (fx(x), fy(x)) (mod p−m). ZGLn+1(Zp)ZGLn+1(Zp) #(gxUx ∩ gyUy ∩ H)dgydgx = vol(Ux) vol(X) · vol(Uy) vol(Y ) = vol(Ux) vol(X) . P -- ADIC INTEGRAL GEOMETRY 15 The right hand side is just the Jacobian matrix for fx ∪ fy at x. By Hensel's lemma, there is a unique projective solution to the system fx ∪ fy which is contained in Ux = Uy. On the other hand, the equations defining the two tangent spaces TxX, TyX also satisfy the conditions of Hensel's lemma, and have the common interesction point x ≡ y (mod p−m). Thus, the conclusion of the lemma holds in this case as well. (cid:3) Theorem 34. Let A ⊂ Sn be a homogeneous algebraic set of dimension dim(A) = a + 1 and let X ⊂ Pn be its projectivization. If U ⊆ X is a compact open subset and L ⊂ Pn is a projective subspace of dimension n − a, then for almost all g ∈ GLn+1(Zp) the intersection U ∩ gL is transversal and: vol(ϕ−1(U )) = vol(Pa) ·ZGLn+1(Zp) # (U ∩ gL) dg. Proof. Let Z ⊂ GLn+1(Zp) be the set defined by: Z = {g : gL is not transversal to X}. and denote Zℓ :=Sg∈Z B(g, p−ℓ). Since Z is contained in a proper algebraic set, by Corol- lary 25 there is a constant C > 0 such that µ[g∈Z B(g, p−ℓ) ≤ Cp−ℓ dim(Z). With ℓ fixed, we have for every u ∈ U there exists mu,ℓ > 0 such that for all g ∈ GLn+1(Zp)\Zℓ and for every m ≥ mu,ℓ : (4.1) #(cid:0)gL ∩ U ∩ B(u, p−m))(cid:1) = #(cid:0)gL ∩ TuU ∩ B(u, p−m))(cid:1) . Observe that Z1 ⊃ Z2 ⊃ . . . and that Tℓ∈N Zℓ = Z. Since Z has measure zero, we see from the Monotone Convergence Theorem that ZGLn+1(Zp) f (g)dg = lim f (g)dg ℓ→∞ZGLn+1(Zp)\Zℓ for every measurable function f : GLn+1(Zp) → R. With this in mind, we evaluate now: ZGLn+1(Zp) # (U ∩ gL) dg = lim # (U ∩ gL) dg ℓ→∞ZGLn+1(Zp)\Zℓ ℓ→∞ZGLn+1(Zp)\Zℓ (B(ui, p−m) ∩ U ) with Nm(U ) disjoint subsets of the form # (Ui ∩ gL) dg = (∗), Nm(U )Xi=1 = lim where we have covered U =SNm(U ) Ui = B(ui, p−m) ∩ U , with i=1 m = mℓ ≥ sup u∈U mu,ℓ ≥ 0 large enough (such an m exists by compactness of U ). 16 AVINASH KULKARNI AND ANTONIO LERARIO Using (4.1), we can continue with: (∗) = lim ℓ→∞ZGLn+1(Zp)\Zℓ Nm(U )Xi=1 #(cid:0)B(ui, p−m) ∩ TuiUi ∩ gL(cid:1) dg ℓ→∞Z GLn+1(Zp) #(cid:0)B(ui, p−m) ∩ TuiUi ∩ gL(cid:1) dg + O(p−ℓ dim(Z)) Nm(U )Xi=1 ℓ→∞ZGLn+1(Zp) Nm(U )Xi=1 #(cid:0)B(ui, p−m) ∩ TuiUi ∩ gL(cid:1) dg = lim = lim = lim ℓ→∞ Nm(U ) · = lim m→∞ Nm(U ) · vola(B(u, p−m) ∩ Pa) vola(B(u, p−m) ∩ Pa) vola(Pa) vola(Pa) = vol(U ) vol(Pa) by Corollary 25. The conclusion follows now from the definition (3.2). Theorem 35 (The p-adic Integral Geometry Formula). Let U, V ⊂ Pn be open and compact subsets of algebraic sets of dimensions dim(U ) = a, dim(V ) = b. Then for almost all g ∈ GLn+1(Zp) the intersection U ∩ gV is transversal, of dimension c = n− (n− a)− (n− b), and: (cid:3) ZGLn+1(Zp) vol (U ∩ gV ) vol (Pc) dg = vol(U ) vol(Pa) · vol(V ) vol(Pb) . Proof. The proof proceeds similarly to the previous one. We observe first that, whenever the intersection U ∩ gV is transversal, picking L ≃ Pc ⊂ Pn and using Theorem 34, we can write: vol (U ∩ gV ) vol (Pc) # (U ∩ gV ∩ g2L) dg2, =ZGLn+1(Zp) dg =ZGLn+1(Zp)ZGLn+1(Zp) and consequently: (4.2) ZGLn+1(Zp) vol (U ∩ gV ) vol (Pc) # (U ∩ g1V ∩ g2L) dg2dg1. Denote by G = GLn+1(Zp) and consider the set: Z = {(g1, g2) ∈ G × G : U ∩ g1V ∩ g2L is not transversal}. Given ℓ > 0, for every y = (u, v) ∈ U × V there exists my,ℓ > 0 such that for all (g1, g2) ∈ G × G\(cid:0)Sz∈Z B(z, p−ℓ)(cid:1) and for all m ≥ my,ℓ we have: U ∩ B(u, p−m) #(cid:0) ∩ g2L(cid:1) = #(cid:0) TuU ∩ B(u, p−m) ∩ g1 (TvV ∩ B(v, p−m)) ∩ g2L(cid:1). g1 (V ∩ B(v, p−m)) ∩ As in the proof of Theorem 34, since Z is contained in a proper algebraic set, we also have: µ [z∈Z As before, we denote Zℓ =Sz∈Z B(z, p−ℓ). B(z, p−ℓ)! ≤ O(p−ℓ). P -- ADIC INTEGRAL GEOMETRY 17 By compactness of U × V , we choose some m = mℓ ≥ supy∈U ×V my,ℓ > 0. We cover now U = Nm(U )[i=1 (cid:0)B(ui, p−m) ∩ U(cid:1) , V = Nm(V )[i=1 (cid:0)B(vi, p−m) ∩ V(cid:1) with Nm(U ) (resp. Nm(V )) disjoint projective balls of radius p−m. In the sequel we denote Ui,m = B(ui, p−m)∩U and Vi,m = B(vi, p−m)∩V ; we will also write T Ui,m = B(ui, p−m)∩TuiU and T Vj,m = B(vj, p−m) ∩ Tvj V . ZGLn+1(Zp) dg =ZGLn+1(Zp)ZGLn+1(Zp) We proceed now to evaluate the integral in the statement, using (4.2): # (U ∩ g1V ∩ g2L) dg2dg1 vol (U ∩ gV ) vol (Pc) = lim = lim = lim # (U ∩ g1V ∩ g2L) dg2dg1 # (Ui,m ∩ g1Vj,m ∩ g2L) dg2dg1 ℓ→∞ZG×G\Zℓ ℓ→∞ZG×G\Zℓ Nm(U )Xi=1 Nm(V )Xj=1 ℓ→∞ZG×G\Zℓ Nm(U )Xi=1 Nm(V )Xj=1 # (T Ui,m ∩ g1T Vj,m ∩ g2L) dg2dg1 + O(p−ℓ) ℓ→∞ZG×G Nm(U )Xi=1 Nm(V )Xj=1 ℓ→∞ZG×G Nm(U )Xi=1 Nm(V )Xj=1 # (T Ui,m ∩ g1T Vj,m ∩ g2L) dg2dg1 = (∗) # (T Ui,m ∩ g1T Vj,m ∩ g2L) dg2dg1 = lim = lim We use now Lemma 31 and continue with: (∗) = lim ℓ→∞ Nm(U ) vola(B(u, p−m) ∩ Pa) vola(Pa) · Nm(V ) volb(B(v, p−m) ∩ Pb) volb(Pb) = lim m→∞ Nm(U ) vola(B(u, p−m) ∩ Pa) vola(Pa) · Nm(V ) volb(B(v, p−m) ∩ Pb) volb(Pb) = vola(U ) vola(Pa) · volb(V ) volb(Pb) . (cid:3) 5. How many zeroes of a random p-adic polynomial are in Zp? Theorem 36. Let {f0, . . . , fℓ} be a basis for Qp[x0, . . . , xn](d) and for every i = 1, . . . , n consider the random polynomial: Fi(x) = ℓXk=0 ξi,kfk(x) where {ξi,k}i=1,...,n,k=0,...,ℓ is a family of independent uniformly distributed random variables in Zp. Let ν(Pn) ⊂ Pℓ be the image of the "Veronese map" ν : Pn → Pℓ (5.1) ν([x]) = [f0(x), . . . , fℓ(x)]. 18 AVINASH KULKARNI AND ANTONIO LERARIO Then, the expectation of the number of solution of the random system of equations F1 = ··· = Fn = 0 in Pn equals: E#{F1 = ··· = Fn = 0} = vol(ν(Pn)) vol(Pn) . Proof. The result is an easy consequence of the Integral Geometry Formula. We use the fact that {f0, . . . , fℓ} is a linearly independent and spanning set to ensure that the map ν is an embedding and that ν(Pn) is a smooth algebraic subset. Now for every i = 1, . . . , n we can consider the random hyperplane Li ⊂ Pℓ: Li =( ℓXk=0 ξi,kyi = 0) . The distribution of this hyperplane is invariant under the group GLℓ+1(Zp) and can be alter- natively described as fixing a hyperplane L0 ⊂ Pℓ, sampling gi ∈ GLℓ+1(Zp) uniformly and setting Li = giL0. Consequently: E#{F1 = ··· = Fn = 0} = E#ν(Pn) ∩ L1 ··· ∩ Ln = E#ν(Pn) ∩ g1L0 ··· ∩ gnLn = Eg1,...,gn−1 (Egn#ν(Pn) ∩ g1L0 ··· ∩ gnL0) = Eg1,...,gn−1 ZGLℓ+1(Zp) = Eg1,...,gn−1(cid:18) vol(ν(Pn) ∩ g1L0 ··· ∩ gn−1L0) #ν(Pn) ∩ g1L0 ··· ∩ gnL0 dgn! (cid:19) = (∗), vol(P1) where in the last step we have used the Integral Geometry Formula from Theorem 35. Re- peating this process iteratively we get: (∗) = Eg1,...,gn−2(cid:18) vol(ν(Pn) ∩ g1L0 ··· ∩ gn−1L0) = Eg1(cid:18) vol(ν(Pn) ∩ g1L0) (cid:19) = vol(ν(Pn)) vol(Pn−1) = ··· vol(Pn) . vol(P2) (cid:19) (cid:3) The previous theorem reduces the question of the expected number of zeros of a random polynomial systerm to computing the volume of the "Veronese variety" ν(Pn) ⊂ Pℓ (i.e. the image of ν from (5.1)). In order to actually calculate the volume of the image of the Veronese, it is tremendously convienient to have available a coarea formula. Specifically, a statement which allows us to calculate the 1-dimensional volume of a curve in Sd. Lemma 37 (Arc length formula). Let U ⊂ Qp be an open and compact set and γ : U → Qn be an analytic embedding such that for all u ∈ U we have Jγ(u) = pb for some b ∈ Z, then vol1(γ(U )) = pbvol1(U ). p Proof. Using Taylor's formula for every u ∈ U and ǫ ∈ Qp sufficiently small, we have: Because Qn p is an ultrametric space, this implies that for ǫ small enough: γ(u + ǫ) − γ(u) = Jγ(u)ǫ + O(kǫk2 p). kγ(u + ǫ) − γ(u)kp = Jγ(u)p kǫkp . P -- ADIC INTEGRAL GEOMETRY 19 In particular, denoting Jγ(u) = pb, for every u ∈ U there is mu ∈ N such that for all m ≥ m0 (5.2) Since U is assumed to be compact, then there exists m0 > 0 such that for all m ≥ m0 equation (5.2) is true for all u ∈ U . Now, covering U with Nm disjoint balls γ(B(u, p−m)) = γ(U ) ∩ B(γ(u), pb−m). B(um,1, p−m), . . . , B(um,Nm, p−m), we can write vol1(U ) = p−mNm for m large enough: At the same time the balls B(γ(um,1), pb−m), . . . , B(γ(um,Nm ), pb−m) cover γ(U ) and for large enough m are disjoint. In particular, pbvol1(U ) = pb lim m→∞ Nm pm = lim m→∞ Nm−b(γ(U )) pm−b = lim m′=m−b→∞ Nm′(γ(U )) pm′ = vol(γ(U )). (cid:3) 5.1. The standard Veronese random model. Proposition 38. Let Pℓ be the projectivization of the space of homogeneous polynomials with coefficients in Qp, of degree d and in n + 1 variables (so that ℓ =(cid:0)n+d be the Veronese map d (cid:1) − 1). Let ν : Pn → Pℓ νn,d : [x0, . . . , xn] 7→ [(xα0 0 ··· xαn n )α=d]. Then vol(ν(Pn)) = vol(Pn). Proof. We will prove that ν is an isometry onto its image: since Proposition 23 establishes that the volume in our sense is a metric invariant. from this the result will follow, Observe first that the linear representation ρ : GLn+1(Zp) → GLℓ(Qp) by change of variables has image in GLℓ(Zp), hence for every g ∈ GLn+1(Zp), the map ρ(g) : Pℓ → Pℓ acts by isometries (as well as the map g : Pn → Pn). Let now x, y ∈ Pn be any two points. Pick an element g ∈ GLn+1(Zp) such that: g[x] = [1, 0, . . . , 0] and g[y] = [a0, a1, 0, . . . , 0] with (a0, a1, 0, . . . , 0) ∈ Sn. Then, because g is an isometry, using the definition (2.2) of the distance in projective spaces: (5.3) dPn([x], [y]) = dPn([1, 0, . . . , 0], [a0, a1, 0, . . . , 0]) = a1p. Let us evaluate now the distance d([ν(x)], [ν(y)]): dPℓ([ν(x)], [ν(y)]) = dPℓ(ρ(g)[ν(x)], ρ(g)[ν(y)]) = dPℓ([ν(gx)], [ν(gy)]) = dPℓ([1, 0, . . . , 0], ν[a0, a1, 0, . . . , 0]). Since (a0, a1, 0, . . . , 0) was in the unit sphere, the vector ν(a0, a1, 0, . . . , 0) ∈ Qℓ+1 the unit sphere and its components are (in lexicographic order): d−1, a1 ν(a0, a1, 0, . . . , 0) = (a0 d−1a1, . . . , a0a1 d, 0, . . . , 0). d, a0 p is also on As a consequence, using again the definition (2.2), we see that dPℓ([1, 0, . . . , 0], ν[a0, a1, 0, . . . , 0]) equals the p-adic norm of the vector: (a0 d−1a1, . . . , a0a1 d−1, a1 d, 0, . . . , 0) = a1(a0 d−1, . . . , a0a1 = a1(ν1,d−1(a0, a1) d−2, a1 , 0, . . . , 0 d−1, 0, . . .) ). ∈Qp d {z } ∈Qℓ−1−d {z } p 20 AVINASH KULKARNI AND ANTONIO LERARIO But kν1,d−1(a0, a1)kp = 1, so using (5.3) we see: dPℓ([1, 0, . . . , 0], ν[a0, a1, 0, . . . , 0]) = ka1(ν1,d−1(a0, a1), 0, . . . , 0)kp = ka1(ν1,d−1(a0, a1))kp = dPn([x], [y]) (cid:3) Corollary 39. Let {ξ1,α}i=1,...,n,α=d be a family of independent uniformly distributed random variables in Zp and for every i = 1, . . . , n define the random polynomial: gi(x) = Xα=d ξi,αxα0 0 ··· xαn n . Then E#{g1 = ··· = gn = 0} = 1. Proof. Proposition 38 gives the volume of the standard Veronese. We apply Theorem 36. (cid:3) 5.2. Evans' result and the Mahler Veronese. The Mahler basis for the space of (nonho- moegeneous) polynomials of degree d in n variables is defined as follows. For every k1, . . . , kn with k1 + ··· + kn ≤ d consider the polynomial: Tk1,...,kn(t1, . . . , tn) :=(cid:18)t1 k1(cid:19)···(cid:18)tn kn(cid:19) = t1 ··· (t1 − k1 + 1) k1! tn ··· (tn − kn + 1) kn! . ··· The family {Tk1,...,kn} forms a basis for Qp[t1, . . . , tn]d. Evans considers the random system of equations: (5.4) f1 = . . . = fn = 0, where fi(t1, . . . , tn) = Xk1+···kn≤d ξi,k · Tk1,...kn(t1, . . . , tn) and proves [Eva06, Theorem 1.2] that the expected number of solutions of (5.4) in Zn p is E#{f1 = . . . = fn = 0} ∩ Zn p = pn⌊logp d⌋(1 + p−1 + ··· + p−n)−1. We give a proof of Evans' result for the expected number of zeros in Zp using the integral geometry formula in the univariate case. We also extend Evans' result and compute the expected number of zeros in Qp\Zp. The argument for both results follow the same pattern. First, we will prove an elementary technical lemma regarding the size of binomial expressions. Next, we use the Arc length formula to compute the image of the appropriate Veronese map on the unit sphere and also the image in projective space. Finally, we apply the Integral Geometry formula to obtain results about the random model. Lemma 40. Let a ∈ Zp and let F : (t, 1) → (cid:0)1,(cid:0)t map. Then the image of F : Zp → Qd+1 is contained in the unit sphere, and 1≤k≤dk−1 1(cid:1), . . . ,(cid:0)t JF (a)p = max d(cid:1)(cid:1) be the affine Mahler Veronese p = p⌊logp d⌋. p Proof. Since Z is dense in Zp and Zp is a discrete valuation domain, we may assume that a ∈ Z. The first claim is obvious since the first coordinate of F (a) is a unit, and binomial coefficients are integers. Next, let g(t) := d dt(cid:18) t k(cid:19) = 1 t − j(cid:18)t k(cid:19) k−1Xj=0 P -- ADIC INTEGRAL GEOMETRY 21 and note each term of g(t) is a polynomial. Let α be the largest (in the archimedean sense) p-th power in {1, . . . , k} and let β ∈ {a− k + 1, . . . , a} be an element which is maximally p-divisible. Since both intervals of integers have length k, we have that βp ≤ αp. Furthermore, If follows for each a ∈ Z and j ∈ {0, . . . , k − 1} that yp . y6=β x6=α xp ≥ kYx=1  α(cid:12)(cid:12)(cid:12)(cid:12)p aYy=a−k+1 aYy=a−k+1 yp y6=a−j 1  1 (cid:12)(cid:12)(cid:12)(cid:12) k(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)p a − j(cid:18)a kYx=1 since the left hand side is maximal when ommiting the factor a− j = β of minimal size. Thus, each term of the k-th entry of JF (a) is bounded by α−1 p ≤ p⌊logp d⌋. =(cid:12)(cid:12)(cid:12)(cid:12) ≤ α−1 To demonstrate equality, we let α be the largest p-th power in {1, . . . , d} and we consider the (α + 1)-th entry of JF (a). In this case, we have the equality x6=α p −1 xp αYx=1 x6=α xp = aYy=a−α+1 y6=β yp . since both sets {1, . . . , α}, {a−α+1, . . . , a} are a complete set of representatives of the integers modulo the p-th power α. Thus, d dt(cid:18) t α(cid:19)(a) = has a unique term of maximal size α−1 Corollary 41. The length of the image of the affine Mahler Veronese map F : (Zp×{1}) → Sd a − j(cid:18)a α(cid:19) p = p⌊logp d⌋. α−1Xj=0 1 (cid:3) 1(cid:19), . . . ,(cid:18)t d(cid:19)(cid:19) is exactly p⌊logp d⌋. Furthermore, with ϕ : Sd → Pd, we have that vol1(F (Zp)) = vol1(ϕ(F (Zp))). F : (t, 1) →(cid:18)1,(cid:18)t Proof. The first part of the statement, follows immediately from Lemma 37 (with the choice U = Zp and γ(t) = F (1, t)) and Lemma 40: vol1(F (Zp)) = p⌊logp d⌋vol1(Zp) = p⌊logp d⌋. We now prove the second part of the claim. In fact, we prove that the Hopf map restricted to the affine Mahler Veronese F (Zp) is an isometry. Let a, b ∈ F (Zp) ⊆ Sd, and write a := (1, a1, . . . , ad) b := (1, b1, . . . , bd). We have that the standard metric distance d(a, b) is max1≤i≤n ai − bip. On the other hand, the distance between the projective points ϕ(a), ϕ(b) is given by the maximum absolute value of the 2×2 minors of(cid:20)1 a1 1 b1 . . . ad . . . bd(cid:21) . The maximum absolute value of the minors is invariant under GLd+1(Zp), so substracting integral multiples of the first column from the others yields 22 AVINASH KULKARNI AND ANTONIO LERARIO the matrix(cid:20)1 0 1 b1 − a1 . . . . . . 0 bd − ad(cid:21) . From which it is clear that d(ϕ(a), ϕ(b)) = ka ∧ bkp = d(a, b), showing that the Hopf map restricted to F (Zp) is an isometry. (cid:3) We now turn our attention to the complement of Zp in Qp. We will make extensive use of the simple observation that if tp > 1, then t − jp = tp for all j ∈ Z. Proof. For k > 0, we have that Lemma 42. For a fixed tp > 1, the function g(k) :=(cid:12)(cid:12)(cid:0)t (t − j)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)p =(cid:12)(cid:12)(cid:12)(cid:12) k−1Yj=0 (k − 1)!(cid:12)(cid:12)(cid:12)(cid:12)p ·(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Since kp ≤ 1 and tp > 1, we are done. (cid:12)(cid:12)(cid:12)(cid:12)(cid:18)t k(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)p k(cid:12)(cid:12)(cid:12)(cid:12)p ·(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12) 1 1 k(cid:1)(cid:12)(cid:12)p is strictly increasing for k ∈ N. t − k + 1 k (cid:12)(cid:12)(cid:12)(cid:12)p ·(cid:12)(cid:12)(cid:12)(cid:12)(cid:18) t k − 1(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)p . (cid:3) Define for m ≥ 1 the m-th annulus Am := B(0; pm)\B(0; pm−1) in Zp. We show how to map Am to the unit sphere using a scaling of the Mahler Veronese map and we calculate the volume of the image. Definition 43. We extend the affine Mahler Veronese to all of Qp by F : Qp\Zp → (t, 1) Sd 7→ (cid:16)(cid:0)t d(cid:1)−1 d(cid:1)−1 1(cid:1)(cid:0)t ,(cid:0)t , . . . , 1(cid:17) . By Lemma 42, we have that the image of F is contained in Sd, so the map is well-defined. We also have that F is injective, since it the (d − 1)-th component is invertible on Qp\Zp. By our choice of scaling, the map F , together with the previously defined affine Mahler Veronese descends to a well-defined morphism νmahler : P1 → Pd. Lemma 44. If 0 ≤ j < d, then The result follows from the product rule. Corollary 45. If t ∈ Am, then JF (t)p = dp · p−2m. Proof. We see that the d-th component of the Jabobian is 0, and for j < d that d! j!(cid:12)(cid:12)(cid:12)(cid:12)p ·(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) d−1Yr=j when j = d − 1 and equals dp ·(cid:12)(cid:12)t−1(cid:12)(cid:12)p ·(cid:12)(cid:12)t−1(cid:12)(cid:12)p. (JF (t))jp =(cid:12)(cid:12)(cid:12)(cid:12) 1 t − r(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)p ·(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) d−1Xr=j 1 t − r(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)p . We see that each of the three factors is maximal when j = d − 1, so the maximum is attained (cid:3) (cid:3) Proof. We observe that for 0 ≥ j < d that d dt(cid:18)t j(cid:19)(cid:18)t d(cid:19)−1 = d! j! · d−1Yr=j 1 t − r · d−1Xr=j (cid:18)t j(cid:19)(cid:18)t d(cid:19)−1 = d! j! ·Qd−1 Qd−1 r=j (t − r) r=0(t − r) = d! j! · 1 t − r . d−1Yr=j 1 t − r . P -- ADIC INTEGRAL GEOMETRY 23 Corollary 46. For m ≥ 1, the image of F : Am → Sd has volume pm (1 − p−1) (pm − pm−1) · dp · p−2m = dp and the image of F : Qp\Zp → Sd has volume F (Qp\Zp) is an isometry. Proof. Since F is injective on Am, and the Jacobian has constant absolute value, the first claim follows from Lemma 37. For the second, we have dp p . Finally, the Hopf fibration restricted to vol1 (F (Qp\Zp)) = vol1[m≥1 F (Am) = dp (1 − p−1) · 1 pm = dp p . ∞Xm=1 Finally, as the last coordinate of F (t) is 1 for any t ∈ Qp\Zp, we see as before that the Hopf fibration restricts to an isometry on F (Qp\Zp). (cid:3) With all of the calculation details in place, we can give several results at once about the expected number of zeros of random polynomials in the Mahler random model. of i.i.d. random varuable unifomrly distributed on Zp. Then: Theorem 47. Let f (t) be the random polynomial f (t) =Pd (1) (Evans:) E#{f = 0} ∩ Zp = p⌊logp d⌋ 1+p−1 . (2) E#{f = 0} = p⌊logp d⌋+dpp−1 1+p−1 . k=0 ξk(cid:0)t k(cid:1) with {ξk}k=0,...,d a family Proof. Reasoning as in the proof of Theorem 36, we see that, through the p-adic Integral Geometry Formula and the calculation of Corollary 41 we have: E#{f = 0} ∩ Zp = vol1(ϕ(F (Zp)) vol1(P1) = p⌊logp d⌋ 1 + p−1 . For the second part, we use the calculation from Corollary 46 in a similar argument, and combine the result with part (1). (cid:3) References [AB15] Dennis Amelunxen and Peter Bürgisser. Probabilistic analysis of the Grassmann condition number. [Ale03] Found. Comput. Math., 15(1):3 -- 51, 2015. Semyon Alesker. Hard Lefschetz theorem for valuations, complex integral geometry, and unitarily invariant valuations. J. Differential Geom., 63(1):63 -- 95, 2003. [BC13] Peter Bürgisser and Felipe Cucker. Condition, volume 349 of Grundlehren der Mathematischen Wis- senschaften [Fundamental Principles of Mathematical Sciences]. Springer, Heidelberg, 2013. The ge- ometry of numerical algorithms. [BF11] Andreas Bernig and Joseph H. G. Fu. Hermitian integral geometry. Ann. of Math. (2), 173(2):907 -- 945, 2011. [BKL18] Paul Breiding, Khazhgali Kozhasov, and Antonio Lerario. On the Geometry of the Set of Symmetric Matrices with Repeated Eigenvalues. Arnold Math. J., 4(3-4):423 -- 443, 2018. [Cho00] Kwok-Kwong Choi. On the distribution of points in projective space of bounded height. Trans. Amer. [Con] Math. Soc., 352(3):1071 -- 1111, 2000. Keith https://kconrad.math.uconn.edu/blurbs/gradnumthy/multivarhensel.pdf. multivariable Conrad. hensel's lemma. A Available at [EK95] Alan Edelman and Eric Kostlan. How many zeros of a random polynomial are real? Bull. Amer. Math. Soc. (N.S.), 32(1):1 -- 37, 1995. [Eva06] Steven N. Evans. The expected number of zeros of a random system of p-adic polynomials. Electron. Comm. Probab., 11:278 -- 290, 2006. 24 AVINASH KULKARNI AND ANTONIO LERARIO [Fis97] Benji Fisher. A note on Hensel's lemma in several variables. Proc. Amer. Math. Soc., 125(11):3185 -- 3189, 1997. [How93] Ralph Howard. The kinematic formula in Riemannian homogeneous spaces. Mem. Amer. Math. Soc., [Igu00] 106(509):vi+69, 1993. Jun-ichi Igusa. An introduction to the theory of local zeta functions, volume 14 of AMS/IP Stud- ies in Advanced Mathematics. American Mathematical Society, Providence, RI; International Press, Cambridge, MA, 2000. [Oes82] Joseph Oesterlé. Réduction modulo p n des sous-ensembles analytiques fermés de ZN p . Invent. Math., 66(2):325 -- 341, 1982. [San04] Luis A. Santaló. Integral geometry and geometric probability. Cambridge Mathematical Library. Cam- bridge University Press, Cambridge, second edition, 2004. With a foreword by Mark Kac. [Sch11] Peter Schneider. p-adic Lie groups, volume 344 of Grundlehren der Mathematischen Wissenschaften [Ser81] [Fundamental Principles of Mathematical Sciences]. Springer, Heidelberg, 2011. Jean-Pierre Serre. Quelques applications du théorème de densité de Chebotarev. Inst. Hautes Études Sci. Publ. Math., (54):323 -- 401, 1981. [SW93] Rolf Schneider and John A. Wieacker. Integral geometry. In Handbook of convex geometry, Vol. A, B, pages 1349 -- 1390. North-Holland, Amsterdam, 1993. [SW08] Rolf Schneider and Wolfgang Weil. Stochastic and integral geometry. Probability and its Applications (New York). Springer-Verlag, Berlin, 2008. [Wei79] Wolfgang Weil. Kinematic integral formulas for convex bodies. In Contributions to geometry (Proc. Geom. Sympos., Siegen, 1978), pages 60 -- 76. Birkhäuser, Basel-Boston, Mass., 1979. [Wei82] André Weil. Adeles and algebraic groups, volume 23 of Progress in Mathematics. Birkhäuser, Boston, Mass., 1982. With appendices by M. Demazure and Takashi Ono. Max Planck Institute MIS Leipzig E-mail address: [email protected] SISSA (Trieste) E-mail address: [email protected]
1803.03548	2	1803	2018-09-04T19:20:22	Counting lines on surfaces, especially quintics	[ "math.AG" ]	We introduce certain rational functions on a smooth projective surface X in IP^3 which facilitate counting the lines on X. We apply this to smooth quintics in characteristic zero to prove that they contain no more than 127 lines, and that any given line meets at most 28 others. We construct examples which demonstrate that the latter bound is sharp.	math.AG	math	COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS S LAWOMIR RAMS AND MATTHIAS SCH UTT Abstract. We introduce certain rational functions on a smooth pro- jective surface X ⊂ P3 which facilitate counting the lines on X. We apply this to smooth quintics in characteristic zero to prove that they contain no more than 127 lines, and that any given line meets at most 28 others. We construct examples which demonstrate that the latter bound is sharp. . G A h t a m [ 2 v 8 4 5 3 0 . 3 0 8 1 : v i X r a 1. Introduction Recently, there has been considerable progress in the understanding of configurations of lines on smooth quartic surfaces in P3. If the ground field K has char(K) 6= 2, 3, the maximal number of lines on a smooth X4 is 64, and a line can be met by at most 20 other lines (see [18], [14]). For char(K) = 3 the maximal number is 112 and a line meets at most 30 other lines ([15]). The cases char(K) = 2 as well as K = R, or even K = Q were studied in [5], [6], [16]. In fact, the former two papers deal more generally, and to great extent, with smooth quartic surfaces with many lines, not just the maximum. Most of this progress was made possible by the fact that smooth quartics are K3 surfaces, and lines endow them with elliptic fibrations (or genus one fibrations). Moreover, examples can be constructed with the help of lattice-theoretical techniques based on the Torelli theorem. In contrast, for smooth surfaces Xd ⊂ P3 of degree d ≥ 5 very little seems to be known, even if K = C. Although the question what is the maximal number of lines appears in various places in the modern literature (see e.g. [9], [10], [2]), the best known bound essentially stems from work of Salmon (see [17], [4]) and has since then been only slightly improved by Segre ([18]): (1.1) M (d) ≤ d(11d − 28) + 12 where M (d) is the maximum number of lines on a smooth degree-d surface (at least over C). In comparison, the Fermat surface of degree d contains exactly 3d2 lines (over any algebraically closed field of characteristic zero or Date: September 4, 2018. 2010 Mathematics Subject Classification. Primary: 14J25; Secondary 14J70. Key words and phrases. algebraic surface, quintic surface, line, fibration, flecnodal divisor. Research partially supported by the National Science Centre, Poland, Opus grant no. 2017/25/B/ST1/00853 (S. Rams) and by ERC StG 279723 (SURFARI) (M. Schutt). 1 2 S LAWOMIR RAMS AND MATTHIAS SCH UTT exceeding the degree d) and there are examples with more lines known only in degrees 6, 8, 12, 20 (see [3]). Even less seems to be known about the maximum valency v(ℓ) of a line ℓ where v(ℓ) = #{lines ℓ′ ⊂ Xd; ℓ′.ℓ = 1} and Xd is again assumed to be smooth. By [18] if ℓ is a line of the first kind (to be defined below, see Def. 2.1), v(ℓ) cannot exceed (8d − 14) (see Proposition 2.2), but this bound does not generalize to the so-called lines of the second kind (cf. [14], [11]). In the case of quartics, the proof of the sharp bound v(ℓ) ≤ 20 for ℓ ⊂ X4 in [14] (outside characteristics 2, 3) is based on the study of the elliptic fibration given by the pencil of cubics residual to ℓ in X4. Already for ℓ ⊂ X5, i.e. on quintics, this approach leads to a genus 3-fibration, that is much more difficult to control. Instead, for a line ℓ ⊂ Xd we will define certain rational functions b0, . . ., bd−3 on the line ℓ (Def. 3.3) which form the main novelty of this paper. The key feature of these functions, especially on the explicit and computational side, is that their common zeroes encode the points where ℓ is met by other lines on Xd (Prop. 3.7). A line of the second kind on a quartic surface is always met by at least twelve other lines (see [18], [13]), whereas one can construct examples of quintic surfaces that contain only one line and the latter is of the second kind. In this note we give another geometric interpretation of the notion of a line of the second kind (see Prop 3.9) that explains why special properties of lines of the second kind on quartics do not carry over to lines on surfaces of degree d ≥ 5. As an application, we sketch an elementary proof of the valency bound for lines on quartics from [14] in Example 3.11. After those general preparations we shall focus on the case d = 5, i.e. of quintics X5 ⊂ P3. Our first main result concerns the valency on smooth quintics: Theorem 1.1. Let K be a field of characteristic zero. If X5 ⊂ P3 smooth quintic containing a line ℓ, then K is a (1.2) v(ℓ) ≤ 28 . More precisely, the maximal value can possibly be attained only for lines of three certain ramification types (Cor 6.2). For one of these configurations, all surfaces containing it can be exhibited explicitly using our methods (Ex- ample 6.3), so the bound from Theorem 1.1 is indeed sharp. In particular, this implies that a quintic surface with a fivetuplet of coplanar lines can contain at most 125 lines (Corollary 6.1). Combined with a technique using the flecnodal divisor and some basic topological and intersection-theoretical arguments, this leads to the following bound for the maximum number of lines on X5 which is our second main result: COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 3 Theorem 1.2. Let K be a field of characteristic zero. Any smooth quintic X5 ⊂ P3 K may contain at most 127 lines. In other words, M (5) ≤ 127. We have to admit that we do not expect Theorem 1.2 to be sharp, or even close to it. In fact, the current record for the number of lines on a smooth quintic surface (outside characteristic 2, 5) stands at 75, attained both by the Fermat quintic and Barth's quintic ([12]). Yet the given bound provides a substantial improvement compared to (1.1). The organization of the paper is as follows. At first we recall Segre's argu- ment for lines of the first kind (see Sect. 2). In Sect. 3 we define the rational functions b0, . . ., bd−3 for a line on a smooth degree-d surface and describe their relevant properties. In the subsequent three sections we study lines of the second kind on quintics to complete the proof of Theorem 1.1. Finally, in Sections 8, 9 we investigate quintics without fivetuplets of coplanar lines in order to derive Theorem 1.2. Convention 1.3. Since the statements of this paper remain valid under base extension (or restriction), we work over an algebraically closed field K of characteristic zero. Acknowledgement. We thank the anonymous referee for helpful com- ments. 2. Segre's bound for the lines of the first kind Let d ≥ 4 and let Xd ⊂ P3 be a smooth degree-d surface that contains a line ℓ (i.e. a degree-one curve). The linear system OXd(1) − ℓ endows the surface in question with a fibration, i.e. a morphism (2.1) π : Xd → P1 whose fibers are plane curves of degree d − 1. We let FP denote the fiber of (2.1) contained in the tangent space TP Xd for a point P ∈ ℓ. The restriction of the fibration (2.1) to the line ℓ defines the degree-(d−1) map (2.2) πℓ : ℓ → P1. Let Rℓ be the ramification divisor of (2.2). By the Hurwitz formula, one has deg(Rℓ) = 2d − 4 . The line ℓ is said to be of ramification type (12d−4) if and only if all rami- fication points are simple, i.e. no ramification point occurs in Rℓ with mul- tiplicity greater than one. The lines of other ramification types are defined in an analogous way; e.g. type (2, 12d−6) means that only one double point appears in Rℓ. We highlight that in the case of d ≥ 5 a phenomenon occurs that was impossible for quartics: two ramification points of the map (2.2) 4 S LAWOMIR RAMS AND MATTHIAS SCH UTT may belong to the same fiber (see 4.3 for an explicit example). To simplify our notation we say that the line ℓ is of ramification type (12d−6, [1, 1]) if it is of the type (12d−4) and exactly two ramification points of index 1 lie in the same fiber of (2.2). Moreover, a line ℓ′ 6= ℓ, ℓ′ ⊂ Xd is called ℓ- unramified if and only if it meets ℓ away from the set of ramification points of the map (2.2). The general fiber of (2.1) is a smooth planar curve with 3(d − 1)(d − 3) In his work on lines on surfaces Segre introduces the inflection points. following notion (cf. [18, p. 87]): Definition 2.1 (Lines of Second Kind). We call the line ℓ a line of the second kind if and only if it is contained in the closure of the flex locus of the smooth fibers of the fibration (2.1). Otherwise, ℓ is called a line of the first kind. Obviously each line ℓ′ 6= ℓ on Xd that meets ℓ is a component of a fiber of (2.1). In particular, it meets ℓ in a point where both the equation of the degree-(d − 1) curve (= the fiber of the fibration (2.1)) and its hessian vanish. The resultant of the restrictions of both polynomials to the line ℓ is of degree (8d − 14) in the homogeneous coordinates of ℓ (see [18, p. 88], [14, Lemma 5.2]). After verifying that multiple lines meeting ℓ in the same point result in a multiple zero of the resultant, this yields the following bound for the valency: Proposition 2.2 (Segre). If ℓ ⊂ Xd is a line of the first kind, then it is met by at most (8d − 14) other lines on Xd: (2.3) v(ℓ) ≤ 8d − 14. For quintics, for instance, the bound reads v ≤ 26 which is a little better than what we stated in Theorem 1.1. Indeed, examples of smooth degree-d surfaces that contain a line met by (d(d − 2) + 2) other lines (thus eventually violating (2.3)) can be found in [11], whereas quartic surfaces that violate the above bound are studied in [14]. It is due to those examples that in this note we will mostly deal with lines of the second kind; in particular, this will be necessary and sufficient to complete the proof of Theorem 1.1. Remark 2.3. Assume that the ideal of a degree-d surface Xd ⊂ P3 is gener- ated by αi,j · xi 3xj 4 = αi,j,k,l · xi 3xj 4xk 1xl 2 ∈ K[x1, x2, x3, x4] , d Xi,j,k,l=0 (2.4) f = d Xi,j=0 where ℓ = V (x3, x4) is a line of the second kind, αi,j ∈ K[x1, x2] is homoge- neous of degree (d−i−j), and αi,j,k,l ∈ K. The degree-(d−1) curve residual to ℓ in the intersection Xd ∩ V (x4 − λ · x3) is given by the polynomial (2.5) fλ(x1, x2, x3) := f (x1, x2, x3, λx3)/x3 COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 5 We put hλ(x1, x2, x3) to denote the determinant of the hessian of fλ. Let r ∈ K[x1, λ] be the remainder of division of hλ(x1, 1, 0) by fλ(x1, 1, 0). We consider the expansion (2.6) By definition, we have that ri,jxi 1λj r =Xi,j (2.7) ℓ is of the second kind iff all ri,j ∈ K[αi,j,k,l] vanish identically. 3. Counting lines with rational functions In this section we introduce the main new tool of this note -- the rational functions bk (see Def. 3.3). The definition is preceded by some elementary lemmata which we need in order to show that the functions we introduce are in fact well-defined. Let Xd ⊂ P3 be a smooth surface of degree d that contains a line ℓ. To simplify our notation we assume as in Remark 2.3 that ℓ = V (x3, x4) and let f denote a generator of the ideal I(Xd) with expansion (2.4). We define ∂2f ∂xj∂x3 ∂f ∂x3 ∂2f ∂xj ∂x4 ∂f ∂x4 dj(f ) :=(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Let P ∈ ℓ be a point such that ∂f ∂x4 can be parametrized by the map (P ) 6= 0. Since the tangent space TP Xd (3.1) P2 (x1 : x2 : x3) TP Xd → 7→ (x1 : x2 : x3 : C(P ) · x3) where C(P ) := − ∂f /∂x3(P ) ∂f /∂x4(P ) , the fiber FP is given by the vanishing of fC(P ) (i.e. (2.5) with λ = C(P )). One can easily check that for j = 1, 2 the following equality holds (3.2) ∂f ∂x4 (P ) · ∂fC(P ) ∂xj ℓ = ∂2f ∂xj∂x3 ℓ · ∂f ∂x4 (P ) − ∂2f ∂xj∂x4 ℓ · ∂f ∂x3 (P ). Note that the right-hand side gives dj(f )(P ) upon substituting P . As an immediate consequence we obtain the following lemma (which has an anal- ogous statement for d1(f ) as the proof reflects). Lemma 3.1. Suppose that P := (1 : p2 : 0 : 0) ∈ ℓ is a point such that ∂f ∂x4 (P ) 6= 0. Then the following conditions are equivalent: • d2(f )(P ) = 0, • P is a ramification point of the map (2.2). Proof. The point P is a ramification point of the map (2.2) if and only if the line ℓ is tangent to the fiber FP in the point P , i.e. (3.3) ∂fC(P ) ∂x1 (P ) = ∂fC(P ) ∂x2 (P ) = 0 . 6 S LAWOMIR RAMS AND MATTHIAS SCH UTT Suppose d2(f )(P ) = 0. From (3.2) we obtain ∂fC(P ) ∂x2 (P ) = 0. The Euler identity yields ∂fC(P ) ∂x1 (P ) = 0. Assume P is a ramification point of the map (2.2). From (3.2) and (3.3) (cid:3) we obtain d2(f )(P ) = 0. Lemma 3.2. Let Xd ⊂ P3 be a smooth surface that contains the line ℓ = V (x3, x4) and let I(Xd) = hf i. Then f satisfies the following conditions (3.4) ∂f ∂x4 /∈ I(ℓ) and d2(f ) /∈ I(ℓ) . Proof. Suppose to the contrary that the formal partial derivative ∂f ∂x4 to the ideal I(ℓ). Since f ∈ I(ℓ), we can write belongs f = x3 · g(x1, x2, x3, x4) + x4 · h(x1, x2, x4) Thus, we have ∂f ∂x4 that the partials ∂f ∂x1 are singularities of Xd. That contradiction shows that ∂f ∂x4 the ideal of the line ℓ. ∈ I(ℓ) if and only if h(x1, x2, x4) ∈ I(ℓ). The latter implies vanish along ℓ. Hence, the zeroes of g on ℓ cannot belong to , ∂f ∂x4 , ∂f ∂x2 Suppose that d2(f ) ∈ I(ℓ). By Lemma 3.1 every point of the line ℓ where does not vanish is a ramification point of the map (2.2). Contradiction. (cid:3) ∂f ∂x4 After those preparations we can define the main tool of this note (recall that the monomial xi 3xj 4xk 1xl 2 appears in (2.4) with the coefficient αi,j,k,l). Definition 3.3. Let ℓ = V (x3, x4) be a line on a smooth surface Xd ⊂ P3 and let f be a generator of the ideal I(Xd) given as (2.4). For k = 0, . . . , d−3 we define the k-th function bk := bk(f ) ∈ K(x1, x2) as the rational function given by the formula (cid:18)d − (j1 + j2 + j3) k + 3 − (j1 + j2)(cid:19)αj1,j2,j3,d−(j1+j2+j3)Ad−(k+3)−j3B(k+3)−(j1+j2)C j2 Xj1,j2,j3≥0 j1+j2≤k+3 j3≤d−(k+3) where A := − d1(f ) d2(f ) ℓ, B := (∂f /∂x3)·d4(f )−(∂f /∂x4)·d3(f ) 2 d2(f )·(∂f /∂x4) ℓ, C := − ∂f /∂x3 ∂f /∂x4 ℓ . Originally, we introduced the functions b0, . . ., bd−3 to find quartic sur- faces with interesting configurations of lines. Indeed, for an explicitly given surface, the b-functions will provide a simple tool to check whether a line is met by many lines (see Proposition 3.7 and Example 3.11). For quintic surfaces, the number of possible ramification types will still be small enough to successfully apply a similar approach. This will be crucial for the proof of Theorem 1.1. The main feature of the rational functions bk will be presented in Propo- sition 3.7. Before we get there, we discuss some basic properties. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 7 Remark 3.4. Given a pair (Xd, ℓ) where ℓ ⊂ Xd is a line, the functions bk do depend on the choice of four ordered linear forms h1, h2, h3, h4 that constitute a system of homogenous coordinates on P3 such that hh3, h4i = I(ℓ). Indeed, one can easily see that some poles of bk(f ) move when we change the homogenous coordinates. Still, as we are about to show, both the set of common zeroes of b0(f ), . . ., bd−3(f ) and the vanishing of b0(f ), . . ., bj(f ) along the line ℓ have purely geometric meanings for the pair (Xd, ℓ). We prefer to define bk for fixed homogenous coordinates (x1, x2, x3, x4) and a polynomial f ∈ hx3, x4i to keep our exposition concise. Remark 3.5. In the definition of the functions bk we assumed the surface Xd to be smooth. Obviously, one can define the functions bk as soon as there exists a system of homogenous coordinates (x1, x2, x3, x4) on P3 such that the pair (Xd, ℓ) satisfies the conditions (3.4). This can be used in study of configurations of lines on singular surfaces (cf. [19]). For later use, we continue with another interpretation of the polynomials d1(f ), . . . , d4(f ). For a point P ∈ Xd := V (f ), we follow [15] to define the Hessian quadric VP := VP Xd as the quadric in P3 given by the quadratic form (3.5) qP = 1 2 (x1, x2, x3, x4)(cid:16) ∂2f ∂xi∂xj (P )(cid:17)1≤i,j≤4 t(x1, x2, x3, x4). Suppose that P ∈ ℓ is a point such that ∂f (P ) 6= 0. As one can easily ∂x4 check, the pre-image of the intersection of the quadric VP with the tangent space TP Xd under the parametrization (3.1) is the quadric given by the (3 × 3)-matrix (3.6) 0 0 0 0 d1(f )(P ) d2(f )(P )   d1(f )(P ) d2(f )(P ) (−B(P ))d2(f )(P )  In particular, we obtain the following useful observation: Observation 3.6. If P ∈ ℓ is a point such that ∂f (P ) 6= 0 and d2(f )(P ) 6= 0 ∂x4 then the Hessian quadric VP Xd does not contain the tangent space TP Xd. Finally we can state the proposition that justifies our interest in the ra- tional functions b0(f ), . . ., bd−3(f ) (and the way we defined them). Proposition 3.7. Let Xd ⊂ P3 be a smooth surface that contains the line ℓ = V (x3, x4) and let I(Xd) = hf i. Suppose that P := (1 : p2 : 0 : 0) is a point such that ∂f ∂x4 (P ) 6= 0 and d2(f )(P ) 6= 0. Then there is at most one line ℓ′ following conditions are equivalent: 6= ℓ such that P ∈ ℓ′ ⊂ Xd, and the (a) there exists a line ℓ′ 6= ℓ such that P ∈ ℓ′ ⊂ Xd, 8 S LAWOMIR RAMS AND MATTHIAS SCH UTT (b) all functions bj vanish at P , i.e. b0(P ) = . . . = bd−3(P ) = 0 . Proof. We assumed the surface Xd to be smooth, so if we have two lines ℓ′ 6= ℓ′′ on Xd such that ℓ′, ℓ′′ 6= ℓ and P ∈ ℓ′, ℓ′′, then the lines in question are coplanar, so P is a ramification point of the map (2.2). The latter is impossible by Lemma 3.1. By Observation 3.6 the Hessian quadric VP does not contain the tangent space TP Xd, so the latter and VP meet along two lines, one of which is the line ℓ. An elementary computation (see (3.1) and (3.6)) shows that the other line can be parametrized by the map Φ := ΦP given as P1 ∋ (t1 : t3) 7→ (t1 : A(P ) · t1 + B(P ) · t3 : t3 : C(P ) · t3) Observe that we have in particular Φ(1 : 0) = P (e.g. by (3.6)). The line Φ(P1) meets Xd with multiplicity at least 3 in the point P , so the composition (f ◦ Φ) has a triple root in the point (1 : 0). By direct check, one has (3.7) f ◦ Φ = t3 3 · d−2 Xj=0 bj(P ) · tj 3 · td−3−j 1 so the line Φ(P1) lies on Xd if and only if (b) holds. (cid:3) In particular, since the surface Xd contains only finitely many lines for any d ≥ 3, Proposition 3.7 shows that the bk-functions cannot all vanish identically: Observation 3.8. If Xd is smooth and d ≥ 3, then there is an integer k ∈ {0, . . . , d − 3} such that bk 6≡ 0. The proof of Prop. 3.7 enables us to reinterpret Segre's notion of the line of the second kind (Def. 2.1). Proposition 3.9. Let Xd ⊂ P3 be a smooth surface that contains a line ℓ. The following conditions are equivalent: (a) ℓ is a line of the second kind, (b) the function b0(f ) vanishes along the line ℓ. Remark 3.10. Observe that the claim of Prop. 3.9 does not depend on the choice of homogenous coordinates on P3 (cf. Remark 3.4). Proof. We maintain the notation of the proof of Prop. 3.7. Let us assume that the map (2.2) is not ramified at P . By definition, we have ΦP (P1) 6= ℓ and the line ΦP (P1) is tangent to Fp in the point P . The point P is an inflection point of the fiber in question if and only if (f ◦ ΦP ) has a zero of multiplicity ≥ 4 in the point (1 : 0). By (3.7) the latter means that b0(P ) = 0. This completes the proof of the proposition. (cid:3) COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 9 Obviously, for a fixed line ℓ ⊂ Xd we can choose the homogenous coordi- nates (x1, x2, x3, x4) in such a way that the hyperplane section V (x3) ∩ Xd, say, contains no lines and no lines on Xd run through the intersection point V (x1) ∩ ℓ (which is missing from the description in Prop. 3.7). Once we make such a choice, by Prop. 3.7, the number of ℓ-unramified lines on Xd cannot exceed the minimum of the degrees of the numerators of the non-zero functions bk(f ), where k = 0, . . . , d − 3. Thus the bound on the number of ℓ-unramified lines on Xd depends on the presentations (3.8) bj(1, x2) = bj cj . where bj, cj ∈ K[x2] are relatively prime. As an illustration, we compute the functions b0, b1 for the quartic surfaces from the family Z which was central for [14, § 4]. Example 3.11. Let X4 be a quartic that contains a line ℓ of the second kind of ramification type (22). For now, let K denote an algebraically closed field of characteristic 6= 2, 3. By an elementary computation (see [14, Lemma 4.5]) we can assume that X4 is given by the equation x3x3 j=0 qjxj 1 + x4x3 3x2−j 2 + x1x2q(x3, x4) + p(x3, x4), where q = P2 and ℓ := V (x3, x4). The ramification points are P1 := (1 : 0 : 0 : 0), P2 := (0 : 1 : 0 : 0) and we have TPj X4 = V (xj+2). We obtain b0 = 0 and j=0 pjxj 3x4−j 4 4 , p = P4 2 +(cid:0)27 p4 + 3 q1 q2 b1 = 1 1x15 27x3 2 · (−q3 2x18 2 x3 1 2(cid:1) x15 1 +(cid:0)q3 1 +(cid:0)3 q2 2 + 3 q2 1 + 27 p1 + 3 q2 2x9 1 0x18 1 ) By Prop. 3.7 the line ℓ is met by exactly 18 ℓ-unramified lines provided q2q0 6= 0. 1 + 6 q0 q1 q2 + 27 p2(cid:1) x9 0q1 + 27 p0(cid:1) x3 −(cid:0)27 p3 + 3 q0 q2 −(cid:0)3 q0 q2 1q2(cid:1) x12 0q2(cid:1) x6 2 x6 2x12 2x15 1 − q3 We want to check what happens when the ramified fibers of (2.1) contain lines. One can easily see that the fiber FP1 contains exactly one (resp. three) lines if and only if p0 = 0 (resp. q0 = 0). A similar argument shows that FP2 contains exactly one (resp. three) lines if and only if p4 = 0 (resp. q2 = 0). Whenever a ramified fiber (resp. both ramified fibers) acquires a triplet of lines, Prop. 3.7 implies that the number of ℓ-unramified lines drops by three (resp. six) and we arrive at exactly 18 lines that meet ℓ if p0p4 6= 0, or at least 19 if p0p4 = 0. The maximal number 20 is attained precisely when both p0, p4 vanish, but no q0, q2. In this case, the line ℓ is met by 20 other lines on the quartic X4: exactly one line through each of ramification points and 18 ℓ-unramified lines. In Example 3.11 we gave an elementary proof of the most important part of [14, Prop. 1.1]. The computations we just carried out depict a useful property. Apparently the presentation (3.8) can be used only to count the 10 S LAWOMIR RAMS AND MATTHIAS SCH UTT ℓ-unramified lines. Still, the functions b0, . . ., bd−3 detect the existence of many lines through the ramified points of (2.2) as well (i.e. the degree of the numerator of bj decreases when many lines on Xd run through a ramification point of the map (2.2)). On the other hand, the functions b0, . . ., bd−3 can also fail to detect the existence of a line 6= ℓ through a ramification point (see 6.3.1). 4. Lines of the second kind on quintic surfaces Let X5 ⊂ P3 be a smooth quintic surface that contains a line ℓ. In this case we obtain the degree-4 map (2.2) and the ramification divisor Rℓ is of degree six, so there are seven possible ramification types of a line (plus the subcases when two ramification points lie in the same fiber). In this section we discuss important properties of quintic surfaces with lines of the second kind. If P ∈ ℓ is an n-fold ramification point, then generally there can be at most n + 1 other lines on X5 containing P . For simple ramification points on a line of the second kind, there is a useful strengthening (which is a consequence of (2.7)). Lemma 4.1. Let ℓ be a line of the second kind on a smooth quintic X5 and let P occur in the divisor Rℓ with multiplicity one. Then at most two lines on X5 run through the point P (one of them, by assumption, being ℓ). 1(x1 − c1x2)(x1 − c2x2), α0,1 = x2 Proof. We maintain the notation of Remark 2.3, put P = (1 : 0 : 0 : 0), (4.1) α1,0 = x2 and assume, after some linear transformation, that the line ℓ′ := V (x2, x3) lies on the quintic surface X5. By assumption, we have d1d2 6= 0, since otherwise P would not be a simple ramification point. Hence (2.7) yields α0221 = α0311 = 0. One can easily check that the line ℓ′ is the (reduced) tangent cone CP FP . (cid:3) 2(x2 − d1x1)(x2 − d2x1) . The condition (2.7) reduces the problem of finding surfaces with lines of the second kind to solving (relatively large) systems of polynomial equations. In the remainder of this section we explain how one can find all quintics with a line ℓ of the second kind such that Rℓ ∈ {(22, 12), (23), (3, 13), (3, 2, 1), (32)} . (4.2) The remaining two ramification types, namely (16) and (2, 14), are harder to parametrize, so we will use different methods to deal with them (in the next section) in order to prove Theorem 1.1. Let us assume that X5 is given by (2.4) with d = 5, ℓ = V (x3, x4) and r is the remainder defined in Remark 2.3. One can easily check that each ri,j • is of degree at most 2 with respect to αi,j,k,l, when i + j = 2, • is linear with respect to αi,j,k,l, for i + j = 3, • is constant with respect to αi,j,k,l with i + j ≥ 4. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 11 By a linear coordinate change that does not alter the ramification type of P (of the form xi = xi +Pj=3,4 bi,jxj where i = 1, 2, j = 3, 4), we can assume that four of the coefficients αi,2−i,j,k vanish (the most convenient choice of the four coefficients will depend on the ramification type we study -- see e.g. (4.4)). Moreover, a direct computation shows that α2003, α0230 are zero (a geometric interpretation of this fact for some ramification types can be found in [18, § 5]). Throughout the remainder of this section, we denote two of the ramifica- tion points (after some linear transformation) by P1 := P = (1 : 0 : 0 : 0), P2 := (0 : 1 : 0 : 0). Then we fix the polynomials α1,0, α0,1 appropriate for the given ramification type. A determinant computation shows that for all quintic surfaces we consider one can choose twelve ri,j such that the linear system of equations they define (with the indeterminates αi,3−i,k,l) has a unique solution. In this way we obtain a map (4.3) K 6 × K 16 → OP3(5) that associates to αi,j,k,l, where (i + j) ∈ {2, 4, 5}, the quintic given by (2.4). The substitution of (4.3) into all ri,j's yields several affine quadrics in the affine parameter space K 6 which corresponds to αi,2−i,k,l. Let us denote their (set-theoretic) intersection by W. By definition, the quintics with a line of the second kind of the considered ramification type can be parametrized by the image of the restriction of (4.3) to W × K 16 It should be pointed out that in all cases we study W ⊂ K 6 happens to be an affine subspace. Below we discuss the ramification types from (4.2) one-by-one in a series of subsections. We omit certain formulae to keep our exposition compact, but all formulae can be obtained with the help of any computer algebra system without difficulty. Remark 4.2. A careful analysis, for instance based on a parameter count, reveals that, just like for quartics [13], quintics with a line of the second kind do not form a single irreducible family, as different ramification types can yield distinct families. This can also be inferred from the degree of a certain surface in P3 (Sr 11 to be investigated in Section 5, see Remark 5.8). 4.1. Two triple ramification points. Let α1,0 = x4 2. In particular we have Rℓ = 3(P1 + P2) and TPiX5 = V (xi+2) for both ramifi- cation points. After a linear change of coordinates we can assume that 1 and α0,1 = x4 (4.4) α1130 = α1103 = α2030 = α0203 = 0. We consider the system of equations given by r0,2, . . ., r0,5 and ri,j where i = 2, 3, j = 1, 2, 3, 4, to arrive at (4.3). One can easily check that W = V (α2012, α0221) in this case. 12 S LAWOMIR RAMS AND MATTHIAS SCH UTT Consequently, every smooth quintic X5 with a line of the second kind of ramification type (32) can be given by (2.4) where (4.5) α3002 = 1 α2120 = 1 8 α0212 8 α1121 and the other coefficients αi,3−i,k,l vanish. 2, α0320 = 1 2, α1202 = 1 8 α2021 8 α1112 2, α2102 = 1 2, α1220 = 1 4 α1121α2021, 4 α0212α1112 Remark 4.3. Suppose that ℓ is of ramification type (32). Computing the (formal) Taylor expansion of the equation of FP1 (resp. FP2) around P1 (resp. P2) one checks that the line ℓ is the tangent cone of FP1 (resp. FP2) provided either α0212 6= 0 or α0410 6= 0 (resp. either α2021 6= 0 or α4001 6= 0). 4.2. One triple ramification point. In order to study ramification types with a single triple point we put α1,0 = x4 1, α0,1 = x2 2(x2 − x1)(x2 − cx1) in (2.4). As in 4.1, we can assume that (4.4) holds and check that the same system of twelve linear equations has a unique solution for every c ∈ K. A discriminant computation reveals that for three values of c (namely c = 0 and 9c2 − 14c + 9 = 0), we obtain the ramification type (3, 2, 1), whereas c = 1 uniquely leads to (3, 1, [1, 1]). Otherwise the line ℓ is of the (non-degenerate) type (3, 13) (i.e. no pair of ramification points lies in one and the same fiber). One can check that for c = 0 (i.e. ramification type (3, 2, 1)) the set W is given by vanishing of α2012 and α0212 = − 9 64 α1121 − 4 3 α0221 − 243 16384 α2021 − 27 256 α1112 For the values of c such that ℓ is not of ramification type (3, 2, 1), W is a 3-dimensional affine subspace of K 6 given, among others, by the vanishing of α2012. In the remark below we maintain the notation of 4.1. Recall that the fiber FP1 (resp. FP2) is the quartic curve residual to ℓ in the intersection of X5 with the plane V (x3) (resp. V (x4)). Remark 4.4. a. Suppose that ℓ is of the type (3, 2, 1). Without loss of generality we can assume that P1 (resp. P2) is a double, (resp. a triple) ramification point and the fiber FP1 meets the line ℓ in the point (1 : 1 : 0 : 0). This amounts to the choice c = 0 in 4.2 (in particular, we do not have to consider the other values of c for the ramification type (3, 2, 1)). Then Rℓ = 2P1 + 3P2 + P3, with P3 = (4 : 3 : 0 : 0). As in Remark 4.3, one can check that if the fiber FP2 contains a line through the point P2, then α2021, α4001 vanish. By Lemma 4.1 the quintic X5 contains at most one line 6= ℓ through the point P3. b. For the ramification (3, 13) (including the case (3, 1, [1, 1])) we have the same condition for FP2 to have a degree-1 component that contains the triple COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 13 ramification point P2. Lemma 4.1 shows that X5 contains at most one line 6= ℓ through each of the other ramification points. 4.3. Double ramification points. Let us assume that α1,0 = x3 1(x1 − x2), α0,1 = x3 2(x2 − cx1) and c 6= 0, 1. A discriminant computation shows that if c 6= 0, 1, 4, −8, the line ℓ is of the non-degenerate type (22, 12). Obviously, c = 1 yields a singular quintic. For c = 4 we have Rℓ = 2(P1 + P2 + P3), where P3 = (1 : 2 : 0 : 0), so the line ℓ is of ramification type (23). Finally, if c = −8, then the fiber of the parametrization (2.2) defined by the plane V (x3 − 64x4) consists of exactly two (different) ramification points. Hence, the line ℓ is of the ramification type (22, [1, 1]). After some linear transformation, we can assume that α2021, α2030, α1130, α0212 vanish. In order to determine the map (4.3) for a fixed c 6= 0, we solve the system of equations given by the vanishing of r3,0, . . ., r3,4 and ri,j where i = 1, 2 and j = 1, 2, 4, 5. For c = 4 the intersection of quadrics W is given by the single equation (4.6) If c = −8 the set W is a codimension-2 vector subspace of K 6 cut out by: α0203 = −128α2012 + 16α1103 + 8α1112 + 4α1121 − α0221/4. α0203 = −64α1103 + 16α1112 + 1024α2012, α0221 = −128α1112 − 64α1121 − 8192α2012. For c 6= 0, 1, 4, −8, we obtain a codimension-2 affine subspace again. Indeed, one can easily check that α0203, α1103 can be uniquely expressed as elements of K[c, 1/c][α0221, α1112, α1121, α2012]. Remark 4.5. a. By a tangent cone argument if a0221 6= 0 (resp. a2012 6= 0) then at most one line 6= ℓ on X5 runs through the point P1 (resp. P2). b. A similar argument shows that if a0221, a0410, a0401 (resp. a2012, a4010, a4001) vanish simultaneously, then the fiber FP1 (resp. FP2) contains no lines at all through the point P1 (resp. P2). 5. Segre's surface of principal lines along a line Having discussed the five ramification types from (4.2), it remains to analyse the ramification types (16) and (2, 14) (which, as we pointed out before, are relatively hard to parametrize explicitly). Their analysis is the content of the present section using an idea that implicitly appears in [18] to count the lines that meet a line of the second kind of the given ramification type. Sℓ To this end, we consider the following subset of the Grassmannian Gr(1, 3), := {ℓ : ℓ is a line, ∃ P ∈ ℓ such that ℓ ⊂ TP X5 and i(ℓ, FP ; P ) ≥ 2 }, where FP denotes the residual fiber through P and i(·) stands for the in- tersection multiplicity of the planar curves ℓ and FP in the tangent plane 14 S LAWOMIR RAMS AND MATTHIAS SCH UTT TP X5 computed in the point P . We have the following lemma (maintaining the notation of Remark 2.3 and [13, Lemma 2.3]). Lemma 5.1. The total space ∪{ℓ : ℓ ∈ Sℓ} =: S11 is an algebraic surface of degree at most eleven. Proof. Let P = (p1 : p2 : 0 : 0) ∈ ℓ be a point such that α0,1(p1, p2) 6= 0. The residual quartic curve FP ⊂ TP X5 is given by the polynomial g(x1, x2, x3) := f (α0,1(P )x1, α0,1(P )x2, α0,1(P )x3, −α1,0(P )x3)/x3 . Indeed, this is just (α0,1(P )5 · fC(P )), where fC(P ) is defined by (2.5). By direct computation, there exist h1, h2 ∈ K[x1, x2] (resp. h3) of degree 7 (resp. 11) such that ∂g ∂x3 ∂g ∂xj (p1, p2, 0) = α3 (p1, p2, 0) = α4 0,1(p1, p2) · h3(p1, p2), 0,1(p1, p2) · hj(p1, p2) for j = 1, 2. We define bihomogenous polynomials H11 ∈ K[z1, z2][x1, . . . , x4] (resp. H4) of bidegree (11, 1) (resp. (4, 1)) H11 H4 := α0,1(z1, z2) · h1(z1, z2) · x1 + α0,1(z1, z2) · h2(z1, z2) · x2 + h3(z1, z2) · x3 , := α1,0(z1, z2) · x3 + α0,1(z1, z2) · x4 , and repeat almost verbatim the proof of [13, Lemma 2.3], dehomogenizing by putting z2 = 1 and computing the resultant with respect to z1. (In [13] one deals with a quartic surface X4 and considers bihomogeneous polynomials H8, H3 instead of H11, H4, but that's about the only difference.) (cid:3) Definition 5.2. The surface S11 will be called Segre's surface of principal lines along ℓ. Remark 5.3. Suppose P ∈ ℓ and the tangent space TP X5 meets the hessian quadric VP along two lines (obviously one of them is ℓ). Then, by definition both lines are contained in S11. Segre's surface of principal lines has the following useful properties (the proofs of claims a) and b) use the standard topology of P3(C), so here the initial characteristic-0 assumption of the paper is relevant). Lemma 5.4. Let ℓ ⊂ X5 be a line of the second kind. a) If a point P appears in Rℓ with multiplicity one or two, then the tangent plane TP X5 is a component of the ruled surface S11. b) If P appears in Rℓ with multiplicity one, then the surface residual to TP X5 in S11 contains all lines on X5 that run through P . c) If the support supp(Rℓ) consists of at least five points and a fiber FP2 contains two ramification points, then TP2 X5 appears with mul- tiplicity at least two in S11. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 15 Proof. a) Let FP be a fiber of (2.1). According to [18, § 6], if i(ℓ, FP ; P ) ≥ 2, then either P ∈ sing(FP ) or i(ℓ, FP ; P ) ≥ 4. The latter is ruled out by the assumption on multiplicity, so the claim of the lemma follows directly from the definition of the family Sℓ. b) We maintain the notation of the proof of Lemma 4.1 and assume that the line ℓ′ runs through P on X5. One can easily check that for each point P from a (punctured) neighbourhood of P on ℓ (in the complex analytic topology) the hessian quadric V P meets the tangent plane T P X5 properly (i.e. along two lines: ℓ and ℓ P ). Moreover, the line ℓ′ lies in the closure of the union of the lines ℓ P . c) The claim results from an elementary but tedious computation. We maintain the notation of the proof of Lemma 4.1, assume that the points P := (1 : 0 : 0 : 0), P2 := (0 : 1 : 0 : 0) belong to supp(Rℓ) and that the fiber FP (resp. FP2) is the quartic curve residual to ℓ in the intersection of X5 with the plane V (x3) (resp. V (x4)). At first we assume that P , P2 are simple ramification points and P3 := (1 : 1 : 0 : 0) ∈ supp(Rℓ) ∩ FP2 (i.e. c1 = c2 = 1, d1 6= d2 and d1d2 6= 0 in (4.1)). Then we apply (2.7) to compute the map (4.3), and we find the defining polynomial h11 of S11 using the explicit resultant approach from the proof of Lemma 5.1. Finally, we check that h11 is divisible by x2 4, so the claim follows for the ramification types (14, [1, 1]), (12, [1, 1]2), (2, 12, [1, 1]). To deal with the ramification type ([1, 1]3) we assume c1 = c2 = 1 and 0 6= d1 = d2 in (4.1). Using the discriminant one can check that ([1, 1]3) occurs only for d1 = 2, whereas the ramification type (2, [1, 1]2) is impossible. Then for d1 = 2 we repeat the reasoning we used to deal with the other cases, to complete the proof of the claim. (cid:3) Lemma 5.4 has an immediate consequence for the valency of a line ℓ on a smooth quintic X5 granted that all ramification points (relative to ℓ) are simple: Corollary 5.5. If ℓ ⊂ X5 is a line of the second kind of ramification type (16), then ℓ is met by at most 24 other lines on X5: v(ℓ) ≤ 24. Proof. By Lemma 5.4 all lines that meet ℓ on X5 lie on a degree-5 surface contained in S11 (residual to the six distinct tangent planes at the ramifica- tion points). The bound then follows straight from Bezout's theorem. (cid:3) In the remaining case of ramification type (2, 14), we shall see that Segre's surface of principal lines still yields a bound that is sufficient for our pur- poses. However, one has to carry out a more detailed analysis of its be- haviour. For this purpose, we put Sr 11 to denote the surface residual in S11 to the union of tangent planes of X5 in ramification points. Lemma 5.6. Let ℓ ⊂ X5 be a line of the second kind of ramification type (2, 14) and let P appear in Rℓ with multiplicity two. 16 S LAWOMIR RAMS AND MATTHIAS SCH UTT a) If there are more than two lines through P on X5, then deg(Sr 11) ≤ 5. b) If deg(Sr 11) = 6, then Sr plicity ≥ 3. 11 and X5 meet along the line ℓ with multi- Proof. We maintain the notation of the proof of Lemma 4.1, put d1 = 0 (i.e. P = (1 : 0 : 0 : 0) is the double ramification point), and apply (2.7) to find the parametrization (4.3). We use the resultant to compute the polynomial h11 that defines Segre's surface S11 (as indicated in the proofs of Lemmata 5.1, 5.4). It is divisible by x3x4, so set h9 = h11/(x3x4) and define the auxiliary surface S9 = V (h9) ⊃ Sr 11. a) The assumption on the number of lines through P implies that α0221 vanishes (cf. Remark 4.5.a). By direct check, the polynomial h9 is divisible by x3 (resp. by x3x4 when V (x4) contains two ramification points). The claim thus follows from Lemma 5.4.a. b) In general, one checks that the polynomial h9 belongs to the fifth power of the ideal hx3, x4i. Thus, if Sr 11 is a sextic, then it is singular along the line ℓ (this gives intersection multiplicity at least two). The claim will be proved by studying the intersection number of appropriate hyperplane sections of X5 and S9. In detail, we cut X5 and S9 with the hyperplane Ht := V (x1 − tx2) that meets ℓ transversally in the point P ′ = (t : 1 : 0 : 0)). By direct check, the line TP ′X5 ∩ Ht is always a component of the tangent cone of the curve Ht∩S9, so the intersection multiplicity of X5 and S9 along ℓ does exceed five. Since Sr 11 is a sextic by assumption, the hypersurface S9 contains exactly three planes, each of which meets X5 along ℓ with multiplicity one. Thus the claim of Lemma 5.4.b follows. (cid:3) It should be pointed out that for a double ramification point the statement of Lemma 5.4.b does not carry over (i.e. sometimes the surface Sr 11 does not contain all lines through the point in question). Yet Bezout's theorem (using Lemma 5.6) immediately gives the following for ramification type (2, 14). Corollary 5.7. If ℓ ⊂ X5 is a line of the second kind of ramification type (2, 14), then ℓ is met by at most 28 other lines on X5: v(ℓ) ≤ 28. For completeness, we emphasize that together with Corollary 5.5 this covers the two cases missing from Section 4 for the proof of Theorem 1.1. Remark 5.8. As explored for quartics in [13], one can construct examples of quintics with lines of other ramification types and Sr 11 of degree up to 9. For such surfaces the approach we just followed yields weaker bounds for the valency than those needed for the proof of Theorem 1.1. In the next section we will overcome this subtlety using the results of Sect. 3. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 17 6. Counting lines along a line on a quintic surface In this section we can finally come to the problem that was our original motivation. We complete the proof of Theorem 1.1 and show that the given bound for the valency is in fact sharp (Example 6.3). Before the proof, let us formulate an immediate consequence of Theorem 1.1 which already points towards Theorem 1.2: Corollary 6.1. If a smooth quintic surface contains five coplanar lines, then it contains at most 125 lines. Proof of Corollary 6.1. Any line in P3 meets the hyperplane which decom- poses into the 5 given lines when intersected with the quintic X5. Thanks to Theorem 1.1, the total number of lines on X5 is thus bounded by 5 + 5 · (28 − 4) = 125 as claimed. (cid:3) 6.1. Proof of Theorem 1.1. Let ℓ ⊂ X5 be a line. By Prop. 2.2 and Corollaries 5.5, 5.7, we can assume that ℓ is of the second kind and supp(Rℓ) consists of at most four points (as given in (4.2)). The proof is based on the case-by-case study of the bk-functions from Def. 3.3 for each possible ramification type. Without loss of generality we assume that the quintic surface X5 is given by (2.4) and ℓ = V (x3, x4). Recall that Lemma 3.9 implies that the rational function b0 vanishes identically. For j = 1, 2 we define the polynomials bj, cj ∈ K[x2] by the formula (3.8). In order to apply Prop. 3.7, we have to check the condition ∂f /∂x4 6= 0 on ℓ. Presently, one finds that V (cid:18) ∂f ∂x4(cid:19) ∩ ℓ = FP1 ∩ ℓ ⊂ V (x3). Hence all the lines on X5 intersecting ℓ are either • contained in the plane V (x3) or • ℓ-ramified outside the plane V (x3) (so that restrictions imposed by the index of ramification apply, including Lemma 4.1) or • ℓ-unramified outside the plane V (x3) (so that the total number is bounded by both deg(b1) and deg(b2)). In what follows, we will balance out these three cases against each other in order to prove the valency bound from Theorem 1.1. 6.2. Triple ramification points. At first we study the ramification types with a triple point and assume that X5 lies in the image of W × K 16 under the parametrization (4.3) where W ⊂ K 6 is one of the affine subspaces we discussed in 4.1 and 4.2. 6.2.1. Ramification type (32). We maintain the notation of 4.1. Observe that the fiber FP1 meets ℓ only in P1, so FP1 contains no ℓ-unramified lines. On the other hand, all ℓ-ramified lines (if any) are contained in the fibres FP1, FP2. 18 S LAWOMIR RAMS AND MATTHIAS SCH UTT Suppose that b1 6≡ 0. Since, by direct computation, we have deg(b1) ≤ 17, and there are at most four lines in the fibers FP1 (resp. FP2), Prop. 3.7 yields that the line ℓ is met by at most 25 other lines on X5. Assume that b1 ≡ 0. We have deg(b2) ≤ 28, so the valency bound stated in Theorem 1.1 follows directly from Prop. 3.7 provided the fibers FP1, FP2 contain no lines. To complete the proof in this case, it remains to study b2 more precisely. We have (with short-hand notation t.o.d. = terms of degree) b2 = α2021 · (t.o.d. 28, 26) + (t.o.d. 24, . . . , 4) + α0212 · (t.o.d. 3, 0) Therefore, if one of the fibers FP1, FP2 contains a line ℓ 6= ℓ, then by Re- mark 4.3 and Prop. 3.7, we have α0212 = 0 or α2021 = 0, and the line ℓ is met by at most 24 ℓ-unramified lines (resp. at most 20 if both fibers contain a line). This completes the proof of Theorem 1.1 for ramification type (32). 6.2.2. Ramification type (3, 2, 1). We consider the case c = 0 in 4.2 and check that the denominator of bj divides the product (x4 2 ) for j = 1, 2 (here the factor (x1 − x2)5 comes from the vanishing of the partial ∂f /∂x4, whereas the factor given by the simple ramification point P3 (cf. Remark 4.4.a) cancels out). 1(x1 − x2)5x19 Assume b1 6≡ 0. One can easily check that deg(b1) ≤ 20. Moreover, if FP1 contains an ℓ-unramified line (i.e. a line through (1 : 1 : 0 : 0)) then deg(b1) cannot exceed 19. Since the fiber FP1 (resp. FP2) contains at most three (resp. four) lines through the point P1 (resp. P2), Prop. 3.7 combined with Lemma 4.1 and Remark 4.4.a implies that ℓ is met by at most 28 other lines on X5. If b1 ≡ 0, then α0221 = 0 and the degree of b2 is at most 24. Moreover, vanishing of α2021, α4001 results in deg(b2) ≤ 20, whereas the existence of an ℓ-unramified line in FP1 diminishes deg(b2) by one. The required bound thus results from Prop. 3.7 combined with Remark 4.4.a. 6.2.3. Ramification type (3, 13). Again, we maintain the notation of 4.2 and fix any c ∈ K that does not result in the ramification type (3, 2, 1). We obtain deg(b1) ≤ 17 and deg(b2) ≤ 24. If b1 6≡ 0, then the claim follows directly from Lemma 4.1 (indeed, we have at most eight lines in the fibers FP1, FP2 and at most one line through each of the other two simple ramification points). Otherwise, we use Remark 4.4.b to check that deg(b2) ≤ 20 whenever FP2 contains a line through P2. 6.3. Double ramification points. It remains to deal with the ramification types with two double points. As in 6.2, we assume that X5 lies in the image of W × K 16 under the parametrization (4.3), where W ⊂ K 6 is one of the affine subspaces we discussed in 4.3. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 19 We maintain the notation of 4.3. Observe that the partial derivative ∂f /∂x4 vanishes in P1 and in the point (1 : c : 0 : 0) ∈ FP1 (thus the latter is the only unramified point where the first assumption of Prop. 3.7 is not fulfilled (but the second is)) . 6.3.1. Ramification type (23). We assume c = 4 in the formulas from 4.3. Suppose that b1 6≡ 0. By direct computation we have deg(b1) ≤ 26 (resp. deg(b1) ≤ 25 when there is an extra line through (1 : 4 : 0 : 0)). Moreover, if there are at least two extra lines through P2, then Remark 4.5.a yields α2012 = 0 and one gets deg(b1) ≤ 23. More generally, one can check that if there are at least two lines 6= ℓ through each of k ramification points, than deg(b1) ≤ 26 − 3k. Consequently, Prop. 3.7 yields the required bound for the valency from Theorem 1.1 provided b1 6≡ 0 and either ℓ is the unique line through the ramification point P1 or there are at least three lines on X5 which contain the point P1. It remains to discuss the case when there are exactly 26 ℓ-unramified lines and exactly one line 6= ℓ runs through each point Pj for j = 1, . . . , 3. By Prop. 3.7 every root of b1 must be a root of b2, so the remainder of the division of b2 by b1 vanishes. An elementary (but tedious) computation yields α2012 = 0 and at most 23 ℓ-unramified lines. Contradiction. Finally suppose that b1 ≡ 0. Then one can easily see that α2012, α0221, α4001, α4010, α0410, α0401 vanish. By Remark 4.5.b none of the fibers FP1, FP2 contains a line. Moreover, we have deg(b2) ≤ 24, so this completes the proof of Theorem 1.1 for ramification type (23). 6.3.2. Ramification type (22, 12). We assume that c 6= 0, 1, 4, −8 in 4.3 and put P3, P4 to denote the ramification points of index 1 (they can be found with the help of d2(f )). Recall that, by Lemma 4.1, there is at most one line 6= ℓ through P3 (resp. P4). Moreover, one can easily check that once there is a line (different from ℓ) on X5 through (1 : c : 0 : 0), the degree of b1 drops by one. Suppose that b1 6≡ 0. We have deg(b1) ≤ 23. If neither α2012 nor α0221 vanishes, the bound v ≤ 28 follows from Remark 4.5.a. In the remaining cases, a direct computation shows that if α2012 = 0 (resp. α0221 = 0), then we have deg(b1) ≤ 20 (resp. b1 is divisible by x3 2). Prop. 3.7 thus yields the claim again. Eventually we are led to assume b1 ≡ 0. Then α2012, α0221, α4001, α4010, α0410, α0401 vanish and we have deg(b2) ≤ 20. Remark 4.5.b yields that there are no lines 6= ℓ through P1, P2, which completes the proof of Theorem 1.1 for non-degenerate ramification type (22, 12). The proof for the degenerate type follows the same lines, so we omit it for the sake of brevity. (cid:3) 6.4. Record valency. In the above part of the proof of Thm 1.1 we did not analyze (very precisely) the behaviour of b2 under the assumption that b1 vanishes along ℓ. Such an analysis (the details of which we omit) yields the following corollary. 20 S LAWOMIR RAMS AND MATTHIAS SCH UTT Corollary 6.2. Let X5 ⊂ P3 be a smooth quintic surface and let ℓ ⊂ X5 be a line. If ℓ is met by 28 other lines on X5, then one of the following holds: (28a) the line ℓ is of the type (32) and b1 vanishes along ℓ, (28b) the line ℓ is of the type (23), and b1 does not vanish along ℓ, (28c) the line ℓ is of the type (2, 14). Finite field experiments suggest that the configurations (28b), (28c) might not exist, but proving this conjecture exceeds the scope of this paper. Below we describe the family of all quintics that carry the configuration (28a). In particular this shows that Thm 1.1 is sharp. Example 6.3 (Quintics of type (28a)). Let us maintain the notation of 4.1 and assume that X5, endowed with a line with two triple ramification points, satisfies the conditions (4.5). We solve the system of equations given by the vanishing of b1 along the line ℓ. An elementary computation yields that α4001, α0410 vanish, we have α4010 = − 1 α3101 = − 1 16 α2021 16 α2021α1112 and the following equalities hold 2α1112, 2, α0401 = − 1 α1310 = − 1 16 α1121α2 16 α1121 0212, 2α0212, α3110 = − 1 α2201 = − 1 α2210 = − 1 α1301 = − 1 8 α2021α1112α1121 − 1 8 α2021α1112α0212 − 1 8 α2021α1121α0212 − 1 8 α1112α1121α0212 − 1 16 α2021 16 α1112 16 α1112α1121 16 α2021α0212 2α0212, 2α1121, 2, 2 . In this way we may define a map (6.1) Z5 : K 4 × K 6 → OP3(5). Consider the image of Z5, a family of quintic surfaces. A standard Groebner basis computation shows that the quintic surface Z = Z5(1, . . . , 1) is smooth. By Remark 4.3 none of the (ramified) fibers FP1, FP2 contain a line. More- over, for the line ℓ := V (x3, x4) on Z we have b1 = 0, deg(b2) = 28 and the discriminant of b2 does not vanish. By Prop. 3.7, the line ℓ meets exactly 28 other lines on Z. Remark 6.4. As we will see in Section 8 (by Lemma 8.1), if a line on a smooth quintic surface X5 is met by at least 26 other lines, then X5 contains a fivetuplet of coplanar lines (including the given line). Thus no smooth surface in the family Z5 may contain more than 125 lines by Corollary 6.1. 7. Lines vs. conics or twisted cubics Before we can come to the proof of Theorem 1.2, we need a few more preparations, but now in a slightly different direction. Namely we want to get a rough idea of the number of lines on a quintic meeting a given (smooth) COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 21 rational curve C -- here either an irreducible conic or a twisted cubic. To this end, we start by following fairly closely the lines of [15]. At any point P ∈ C, we can compare the tangent plane to X5 = V (f ) ⊂ P3, TP = TP X5, and the Hessian quadric VP ⊂ P3 defined by (3.5). The intersection TP ∩ VP generically consists of the two 3-contact lines of X5 at P ; in particular, if some line ℓ ⊂ X5 contains P , then ℓ ⊂ TP ∩ VP . Depending on the chosen curve C, we now turn to the total space (7.1) Z = ZC = ∪P ∈C(TP ∩ VP ) ⊂ P3. 7.1. Conic case. Proposition 7.1. Assume that C is an irreducible conic. Then Z is a surface of degree 22 in P3. It does not contain X5, but the surface Z contains any line in X5 meeting C. Proof. The proof follows essentially word by word that of [15, Prop. 2.2]. (cid:3) Corollary 7.2. A conic on a smooth quintic surface X5 ⊂ P3 meets at most 88 lines. Proof. If the conic C is geometrically reducible, then the claim follows from Theorem 1.1. Else consider the effective divisor D = Z ∩ X5 ∈ OX5(22) and let m denote the multiplicity of C in D. Then, since all lines meeting C are contained in D, we obtain two inequalities (7.2) #{lines on X5 meeting C} ≤(deg(D − mC) = 110 − 2m, C.(D − mC) = 44 + 4m. The estimate in Corollary 7.2 arises from m = 11. (cid:3) Remark 7.3. If there are lines on X5 planar with the conic C, then the estimate for total number of lines meeting C can be improved. What's more important for us: the bound C. M Xi=1 ℓi! ≤ 88. still holds where the second divisor comprises (all) lines contained in X5. Corollary 7.2 and Remark 7.3 will prove quite useful in the proof of Theo- rem 1.2 as they facilitate simplifications of some of the arguments involved. 22 S LAWOMIR RAMS AND MATTHIAS SCH UTT 7.2. Twisted cubic case. Twisted cubic curves show a rather different be- havior than conics -- notably because they are not plane. We first investigate the surface Z from (7.1) arising from the twisted cubic C. Proposition 7.4. Assume that C is a twisted cubic on a smooth quintic X5 ⊂ P3. Then Z is a surface of degree 33 in P3. It does not contain X5, but the surface Z contains any line in X5 meeting C. Proof. By a linear transformation, we can assume C to be parametrized by ϕ : P1 → C [s, t] 7→ [t3, s3, t2s, ts2] The homogeneous ideal of C is thus generated by the three quadrics q1 = x1x2 − x3x4, q2 = x1x4 − x2 3, q3 = x2x3 − x2 4. At the point P = ϕ(s, t), the tangent plane to the quintic X5 is thus given by a bihomogeneous polynomial g ∈ K[s, t][x1, . . . , x4] which is linear in the xi and of degree 12 in s, t. Meanwhile the quadratic form q defining VP has degree 9 in s, t. Z is defined by the resultant of g and q with respect to t which thus has degree 33 = 2 · 12 + 1 · 9. To conclude, we note as in [15] that any component of Z is automatically (cid:3) ruled by lines. In particular, neither component can equal X5. Contrary to Corollary 7.2, Proposition 7.4 alone does not give a serious estimate for the number of lines on X5 intersecting the twisted cubic C. Arguing as in the proof of Corollary 7.2 with the divisor (7.3) D = Z ∩ X5 ∈ OX5(33), we see at least that a plentitude of adjacent lines forces C to have large multiplicity in D. The combination with the following result will prove quite useful in the sequel. Proposition 7.5. Assume that some line ℓ on the smooth quintic X5 ⊂ P3 meets the twisted cubic C with multiplicity two. Then ℓ has multiplicity at least two in D, i.e. D − 2ℓ > 0. Proof. Assume that ℓ and C intersect in two distinct point P1, P2. On top of the above normalization, we can assume that I(ℓ) = (x3, x4) and P1 = [1, 0, 0, 0], P2 = [0, 1, 0, 0]. By assumption, f ∈ I(C) ∩ I(ℓ), i.e. K[x1, . . . , x4] such that there are cubic polynomials ci ∈ f = q1c1 + q2c2 + q3c3 with c1 ∈ (x3, x4). COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 23 In the end, it will suffice to consider polynomials modulo st (such that only the pure powers in s and in t remain); for this purpose, the following observation concerning the evaluation at P = ϕ(s, t) will be useful: c1, q2, q3 ∈ (x3, x4) =⇒ st c1(P ), q2(P ), q3(P ). Moreover, the same holds true for the (iterated) partial derivates with re- spect to x1 and x2. With this in mind, it is easy to get a hand on the defining polynomials g of TP (S) and q of VP . For instance, g ≡ s3c3(P )x3 + t3c2(P )x4 mod (st). In other words, g ∈ (x3, x4, st), and the same holds for q (as one easily checks by considering the double partials with respect to x1, x2). We continue by computing the resultant of g and q via the Sylvester matrix with respect to s or t. The above argument shows that the first and the last column of this matrix has all (two) entries in (x3, x4). Hence the determinant of the matrix is in (x3, x4)2, and the line ℓ has multiplicity at least two in D as claimed. If the line ℓ meets C tangentially, say at [1, 0, 0, 0], then I(ℓ) = (x2, x4), and the analogous argument shows that g, q ∈ (x2, x4, s2). Hence the first two (or the last two) columns of the Sylvester matrix have entries in (x2, x4), and we conclude as before. (cid:3) 8. Bounding the total number of lines on quintics This section has a slightly different flavour than the previous sections of the paper as the methods are rather different. Our aim is to complete the proof of our second main result, Theorem 1.2. Recall from Corollary 6.1 that if a smooth quintic surface X5 ⊂ P3 admits a hyperplane section splitting into five lines, then X5 contains at most 125 lines. In order to prove Theorem 1.2, it thus remains to study the other possible configurations of lines. Recall that the locus of points P such that there exists a line that meets X5 with multiplicity at least 4 in P is the support of a divisor in OX5(31) (see [4], [8, § 8]). Following classical terminology, we call it the flecnodal divisor of X5 and denote it by F. Observe that each line on X5 is a component of F. A curve on X5 is called flecnodal if and only if it is a component of the flecnodal divisor F. We record the following simple observation. Lemma 8.1. If a line ℓ ⊂ X5 is not contained in any hyperplane splitting into lines, then v(ℓ) ≤ 25. Proof. This is a consequence of the Euler number formula [1, Prop. III.11.4] for the fibration (2.1) and the Euler number computation for singular curves (see [7, Cor. V.4.4.ii]). (cid:3) 24 S LAWOMIR RAMS AND MATTHIAS SCH UTT In order to start working towards the proof of Theorem 1.2, we make from now the following assumption: Assumption 8.2. The smooth quintic X5 contains M > 127 lines. In particular, Corollary 6.1 implies that X5 does not admit any hyperplane splitting into five lines. It thus follows from Lemma 8.1 that v(ℓ) ≤ 25 for any line on X5. Numbering the lines on X5 from ℓ1 to ℓM , we introduce the following auxiliary effective divisor: (8.1) F ′ := F − M Xi=1 ℓi, deg(F ′) = 155 − M ≤ 27. Note that we can compute the intersection number of F ′ and any line ℓ ⊂ X5 as follows: (8.2) ℓ.F ′ = ℓ. F − M Xi=1 ℓi! = 31 − (v(ℓ) − 3) = 34 − v(ℓ) ≥ 9 where the last estimate follows from Lemma 8.1. The overall idea of our approach goes as follows: For large M , the inter- section number in (8.2) becomes too big relative to the degree of F ′. Indeed, for a single line, such a phenomenon might well occur, but not for all lines at the same time as we shall see soon. For later reference, we compute the self-intersection number (F ′)2 in two ways. We know that F 2 = 5 · 312. Let us compare this intersection number with the decomposition (8.1). Clearly we have ℓi!2 M Xi=1 = M Xi=1 while ℓj  = (v(ℓi) − 3) = M Xi=1 v(ℓi) − 3M M Xi=1 M Xj=1  ℓi. Xi=1 M M Xi=1 ℓi.F ′ = (34 − v(ℓi)) = 34M − v(ℓi). M Xi=1 Together this yields, using Lemma 8.1, (8.3) F 2 = (F ′)2 + 65M − v(ℓi) ≥ (F ′)2 + 40M. M Xi=1 Assumption 8.2 thus gives (8.4) (F ′)2 ≤ 5 · 312 − 40M ≤ −315. Note that, if some lines have valency less than 25, then the bounds (8.3), (8.4) improve accordingly (as we exploit occasionally), and similarly for M > 128. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 25 If F ′ does not contain any multiple components, then (8.4) directly leads to a contradiction as follows: Write F ′ = Ci r Xj=1 for some distinct irreducible curves Ci ⊂ X5. Since KX5 = H, adjunction gives (8.5) Applied to the above decomposition of F ′, this gives i ≥ −2 − deg(Ci). C 2 (F ′)2 =  r Xj=1 2 Cj  ≥ r Xj=1 C 2 j ≥ −2r − deg F ′ ≥ −3 deg(F ′) ≥ −81. This contradicts (8.4) by far, so F ′ has to admit multiple components. We shall now study the multiple components more precisely, but before going into the details, we eliminate the lines from much of what is to follow. Lemma 8.3. Any line contained in the support of F ′ contributes positively to (F ′)2. Proof. Write F ′ = F ′′ + L, where L is the sum of lines contained in the support of F ′ (with multiplicities), L = ℓj, s Xj=1 and F ′′ is an effective divisor on X5 not containing any line in its support. We compute, using (8.2), (8.6) that is, (8.7) as claimed. (F ′)2 = F ′.(F ′′ + L) = (F ′′)2 + F ′′.L {z}≥0 + s Xj=1 ; ℓj.F ′ {z}≥9 (F ′)2 ≥ (F ′′)2 + 9 deg(L) (cid:3) Thanks to Lemma 8.3, we can reduce our considerations for F ′ to satisfy (8.4) to an investigation of the non-linear components of F ′, i.e. we shall often work with F ′′. Sometimes we will even use linear components to our advantage, using the positive contribution in (8.7). In fact, this can be sharpened substantially: Remark 8.4. More precisely, any line ℓ contained in the support of F ′ sat- isfies 9 ≤ ℓ.F ′ = ℓ.L + ℓ.F ′′ =⇒ ℓ.F ′′ ≥ 13 − deg(L). ≤deg(L)−4 {z} 26 S LAWOMIR RAMS AND MATTHIAS SCH UTT In particular, L.F ′′ ≥ deg(L)(13 − deg(L)). Plugging into (8.6) causes (8.7) to improve to (8.8) (F ′)2 ≥ (F ′′)2 + max{9 deg(L), deg(L)(22 − deg(L))}. The next result will be crucial in determining the top multiplicity of the components of F ′. For later reference, we first note it in the general set-up. Lemma 8.5. Let D denote an effective divisor with D2 < 0. Then the support of D contains a component C whose multiplicity N satisfies (8.9) N · C 2 ≤ ǫ deg(C) where ǫ = D2 deg(D) < 0. Proof. Write D =Xi NiCi for distinct irreducible curves Ci. Assume that (8.9) does not hold for either component of D. Then D2 ≥Xi N 2 i C 2 i >Xi Ni(ǫ deg(Ci)) = ǫ deg(D) = D2, giving the required contradiction. (cid:3) We shall use this lemma repeatedly. For our key case of D = F ′, we note the following consequence. Lemma 8.6. The support of F ′ contains a non-linear component C whose multiplicity N satisfies (8.10) N · C 2 < −11 deg(C). Proof. By Lemma 8.3 and its proof, it suffices to consider the divisor F ′′ without linear components of degree at most 27. Then apply Lemma 8.5 to F ′′ and simplify. (cid:3) Remark 8.7. In special situations, this argument can be strengthened using (8.7), (8.8). Also we could insert any constant ǫ < 35/3 instead of 11 in (8.10), but presently this would not have any impact on our arguments. Lemma 8.8. The support of F ′ contains an irreducible conic of multiplicity N ≥ 6 or a twisted cubic curve of multiplicity N ≥ 7. Proof. Obviously the two given cases satisfy the inequality (8.10) from Lemma 8.6. In comparison, the only remaining case in degree at most 3, a plane cubic curve C ⊂ X5, has C 2 = −3 by adjunction, so (8.10) would give multiplicity at least 12, exceeding the degree of F ′. For deg(C) ≥ 4, the inequality (8.5) eliminates all cases in combination (cid:3) with (8.10). Lemma 8.9. Any conic has multiplicity at most 9 in F ′. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 27 Proof. For degree reasons, the multiplicity N of a conic Q in F ′ is at most 13. Recall from Corollary 7.2 that the valency of Q satisfies v(Q) ≤ 88, so there certainly is a line ℓ off Q. But then N ≥ 10 would imply ℓ.F ′ = ℓ.(F ′ − N Q) ≤ deg(F ′ − N Q) ≤ 27 − 2N ≤ 7, contradicting (8.2). (cid:3) Before continuing, we record the following useful consequence: Lemma 8.10. Any line has multiplicity at most 3 in F ′. Proof. Let ℓ ⊂ X5 be of multiplicity N in F ′. First assume that N ≥ 5. Then use (8.2) and compute 9 ≤ ℓ.F ′ = −3N + ℓ.(F ′ − N ℓ) ≤ −3N + deg(F ′ − N ℓ) ≤ 27 − 4N. The claim follows immediately. It remains to discuss the case N = 4. Write F ′ = 4ℓ + R with deg(R) ≤ 23 and R2 ≤ −315 + 48 − 72 = −339 by (8.8). Then Lemma 8.5 implies that either R contains a line of multi- plicity at least 5 (contradicting the above), or a conic of multiplicity m ≥ 8. Hence we obtain F ′ = 4ℓ + mQ + R′, deg(R′) ≤ 23 − 2m ≤ 7. By Corollary 7.2 and Lemma 8.1, there is a line ℓ′ ⊂ X5 which intersects neither ℓ nor Q. Hence ℓ′.F ′ ≤ deg(R′) ≤ 7, contradicting (8.2). This completes the proof of Lemma 8.10. (cid:3) Remark 8.11. With a little more work, one can improve Lemma 8.10 to show that any line has multiplicity at most 2 in F ′, but we will not need this in what follows. 9. Proof of Theorem 1.2 We are now in the position to attack the proof of Theorem 1.2. To this end, we have to rule out all possible configurations with an N -fold irreducible conic Q supported on F ′ (N = 6, . . . , 9), or with an N -fold twisted cubic C (N = 7, 8, 9). We proceed by a case-by-case analysis of the residual effective divisor R = F ′ −N Q, deg(R) ≤ 27−2N, resp. R = F ′ −N C, deg(R) ≤ 27−3N. 9.1. Multiple conic case. Following Lemma 8.8 and Lemma 8.9, we start by assuming that F contains an irreducible conic Q of multiplicity N = 6, . . . , 9 (and no conic of higher multiplicity). We proceed by a case-by-case analysis. 28 S LAWOMIR RAMS AND MATTHIAS SCH UTT 9.1.1. N = 9. Recall from Remark 7.3 that Q. M Xi=1 ℓi ≤ 88. Compared with Q.F = 62, this yields Q.R ≥ 10 and (F ′)2 ≥ −324 + 180 + R2 = −144 + R2. Hence (8.4) implies R2 ≤ −171. One might like to continue disregarding the linear components as in Lemma 8.6, but here the positive contribution from linear components might already have been captured in Q.R ≥ 10, so we have to content ourselves with applying Lemma 8.5 to R. It follows that there is a seven-fold line ℓ, but this contradicts Lemma 8.10. 9.1.2. N = 8. We deduce as in 9.1.1 that Q.R ≥ 6. Hence R2 ≤ −155, and Lemma 8.5 shows as before that R contains a five-fold line which is ruled out by Lemma 8.10. 9.1.3. N = 7. Consider the residual divisor R with deg(R) ≤ 13. As be- fore, we infer Q.R ≥ 2 and R2 ≤ −147. It follows from Lemma 8.5 that R contains either a 4-fold line ℓ (which is ruled out by Lemma 8.10) or a 6-fold conic Q′ 6= Q. In the latter case, either M = 129 and the bound (8.4) improves to (F ′)2 ≤ −355 which is impossible to attain, or F ′ = 7Q + 6Q′ + ℓ for some line ℓ whose positive contribution by Remark 8.4 cannot be compensated for. 9.1.4. N = 6. The residual divisor R has degree deg(R) ≤ 15 and R2 ≤ −171. As in 9.1.3, Lemma 8.5 implies that R contains a line of multiplicity 4 (which again is excluded by Lemma 8.10) or another conic Q′ 6= Q of multiplicity m ≥ 6. In the latter case, since m = 7 has already been covered in 9.1.3, it remains to realize that for m = 6, Lemma 8.5 applied to the residual divisor again leads to 3-fold line and subsequently to a contradiction by the positive contribution from (8.7). 9.2. Multiple twisted cubic case. Continuing the present line of argu- ments based on Lemma 8.8, we distinguish three cases depending on the multiplicity N = 7, 8, 9 of the twisted cubic C in F ′. 9.2.1. N = 7. Here R2 ≤ −70 and deg(R) ≤ 6, so Lemma 8.5 implies that R contains a 4-fold line. Hence Lemma 8.10 gives the required contradiction. COUNTING LINES ON SURFACES, ESPECIALLY QUINTICS 29 9.2.2. N = 8. In this case, deg(R) ≤ 3 with no obvious further restrictions. We proceed by distinguishing how R decomposes. If R contains a line, then its positive contribution following Remark 8.4 gives a contradiction. If R is an irreducible conic, then M = 129, and the improved bound F ′2 ≤ −355 from (8.4) cannot be reached. Similarly, of course, if R = 0. Hence R is an irreducible cubic (plane or twisted) which thus meets no more than 98 lines (by looking at R.F). All the remaining lines (at least 30 in number) have to meet C with multiplicity two for (8.2) to hold. But then their valency drops by 7 to v(ℓ) = 18. As this occurs for at least 30 lines, we get a big correction term of −210 to (8.4) which is impossible to beat. 9.2.3. N = 9. We have F ′ = 9C and M = 128. Then (8.2) implies that any line ℓ ⊂ X5 meets C, and for C.F = 93 to hold, C has to intersect exactly 10 lines with multiplicity two, say ℓ1 . . . , ℓ10. We turn to the effective divisor D from 7.2. By Proposition 7.5, there is a decomposition D = ℓi + ℓj + mC + D′. M Xi=1 10 Xj=1 For degree reasons, this implies m ≤ 9. product reads In comparison, the intersection whence m ≥ 12, contradiction. 99 = C.D = 158 − 5m + C.D′ 9.3. Proof of Theorem 1.2. To conclude, let us wrap up the proof of Theorem 1.2. Assuming that M > 127, we deduced in Lemma 8.8 that F contains a conic with multiplicity N ≥ 6, or a twisted cubic with multi- plicity N ≥ 7. On the other hand, N ≤ 9 by Lemma 8.9 resp. for degree reasons. Then the considerations in Sections 9.1, 9.2 successively ruled out all configurations which might have fitted (8.4), thus completing the proof of Theorem 1.2. (cid:3) Remark 9.1. By inspection, the results of the preceding four sections are totally geometric, i.e. they essentially only require that (2.1) has a smooth fiber, and that the flecnodal divisor does not degenerate. References [1] Barth, W., Hulek, K., Peters, C., van de Ven, A.: Compact complex surfaces. Second edition, Erg. der Math. und ihrer Grenzgebiete, 3. Folge, Band 4. Springer, Berlin, 2004. [2] Eisenbud, D., Harris, J., 3264 and All That: A Second Course in Algebraic Geometry. Cambridge University Press, 2016. [3] Boissi´ere, S., Sarti, A.: Counting lines on surfaces. Ann. Sc. Norm. Super. Pisa, Cl. Sci. 6 (2007), 39 -- 52. [4] Clebsch, A.: Zur Theorie der algebraischen Flachen. Journal reine angew. Math. 58 (1861), 93 -- 108. [5] Degtyarev, A. I.: Lines in supersingular quartics. Preprint (2016), arXiv:1604.05836. 30 S LAWOMIR RAMS AND MATTHIAS SCH UTT [6] Degtyarev, A. I., Itenberg, I., Sertoz, A. S.: Lines on quartic surfaces. Math. Ann. 368 (2017), 753 -- 809. [7] Dimca, A.: Singularities and topology of hypersurfaces. Universitext. Springer-Verlag, New York, 1992. [8] Koll´ar, J.: Szem´eredi-Trotter-type theorems in dimension 3. Adv. Math. 271 (2015), 30 -- 61. [9] Miyaoka, Y: The Maximal Number of Quotient Singularities on Surfaces with Given Numerical Invariants. Math. Ann. 268 (1984), 159 -- 171. [10] Miyaoka, Y: Counting Lines and Conics on a Surface. Publ. RIMS, Kyoto Univ. 45 (2009), 919 -- 923. [11] Rams, S.: Projective surfaces with many skew lines. Proc. Amer. Math. Soc. 133 (2005), 11 -- 13. [12] Rams, S., Schutt, M.: The Barth quintic surface has Picard number 41. Ann. Sc. Norm. Super. Pisa Cl. Sci 13 (2014), 533 -- 549. [13] Rams, S., Schutt, M.: On quartics with lines of the second kind. Adv. Geom. 14 (2014), 735 -- 756. [14] Rams, S., Schutt, M.: 64 lines on smooth quartic surfaces. Math. Ann. 362 (2015), 679 -- 698. [15] Rams, S., Schutt, M.: 112 lines on smooth quartic surfaces (characteristic 3). Quart. J. Math. 66 (2015), 941 -- 951. [16] Rams, S., Schutt, M.: At most 64 lines on smooth quartic surfaces (characteristic 2). Nagoya Math. J. (to appear). [17] Salmon, G.: A treatise on the analytic geometry of three dimensions. Vol. II. Fifth edition. Longmans and Green, London, 1915. [18] Segre, B.: The maximum number of lines lying on a quartic surface. Quart. J. Math., Oxford Ser. 14 (1943), 86 -- 96. [19] Veniani, D. C.: The maximum number of lines lying on a K3 quartic surface. Math. Z. 285 (2017), 1141-1166. Institute of Mathematics, Jagiellonian University, ul. Lojasiewicza 6, 30- 348 Krak´ow, Poland E-mail address: [email protected] Institut fur Algebraische Geometrie, Leibniz Universitat Hannover, Welfen- garten 1, 30167 Hannover, Germany Riemann Center for Geometry and Physics, Leibniz Universitat Hannover, Appelstrasse 2, 30167 Hannover, Germany E-mail address: [email protected]
1102.3443	3	1102	2012-04-23T22:21:21	On the order of an automorphism of a smooth hypersurface	[ "math.AG" ]	In this paper we give an effective criterion as to when a positive integer q is the order of an automorphism of a smooth hypersurface of dimension n and degree d, for every d>2, n>1, (n,d)\neq (2,4), and \gcd(q,d)=\gcd(q,d-1)=1. This allows us to give a complete criterion in the case where q=p is a prime number. In particular, we show the following result: If X is a smooth hypersurface of dimension n and degree d admitting an automorphism of prime order p then p<(d-1)^{n+1}; and if p>(d-1)^n then X is isomorphic to the Klein hypersurface, n=2 or n+2 is prime, and p=\Phi_{n+2}(1-d) where \Phi_{n+2} is the (n+2)-th cyclotomic polynomial. Finally, we provide some applications to intermediate jacobians of Klein hypersurfaces.	math.AG	math	ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO Abstract. In this paper we give an effective criterion as to when a positive integer q is the order of an automorphism of a smooth hypersurface of dimension n and degree d, for every d ≥ 3, n ≥ 2, (n, d) 6= (2, 4), and gcd(q, d) = gcd(q, d − 1) = 1. This allows us to give a complete criterion in the case where q = p is a prime number. In particular, we show the following result: If X is a smooth hypersurface of dimension n and degree d admitting an automorphism of prime order p then p < (d − 1)n+1; and if p > (d − 1)n then X is isomorphic to the Klein hypersurface, n = 2 or n + 2 is prime, and p = Φn+2(1 − d) where Φn+2 is the (n + 2)-th cyclotomic polynomial. Finally, we provide some applications to intermediate jacobians of Klein hypersurfaces. Introduction Let n ≥ 2 and d ≥ 3 be integers. The smooth hypersurfaces of degree d of the projective space Pn+1 = Pn+1(C) are classical objects in algebraic geometry. In the following we assume that (n, d) 6= (2, 4). In this case, the group of regular automorphisms of any such hypersurface X is finite and equal to the group of linear automorphisms i.e., every automorphism of X extends to an automorphism of Pn+1 [8]. In this paper we study the possible orders of automorphisms of smooth hypersurfaces. In Section 1 we give the following criterion for the order of an automorphism of a smooth hypersurface: an integer number q ∈ Z>0, with gcd(q, d) = gcd(q, d − 1) = 1, is the order of an automorphism of a smooth hypersurface of dimension n and degree d if an only if ∃ℓ ∈ {1, . . . , n + 2} such that (1 − d)ℓ ≡ 1 mod q . (1) In Section 2 we show that if q is a prime number then q is the order of an automorphism of a smooth hypersurface of dimension n and degree d if and only if q divides d − 1 or (1) holds. This is a generalization of our previous result for cubic hypersurfaces [3]. In Section 3 we show that if X is a smooth hypersurface of dimension n and degree d admitting an automorphism of prime order p then p < (d − 1)n+1; and if p > (d − 1)n then X is isomorphic to the Klein hypersurface, n = 2 or n + 2 is prime, and p = Φn+2(1 − d) where Φn+2 is the (n + 2)-th cyclotomic polynomial. If we restrict to linear automorphism, the results in Sections 1, 2 and 3 hold also in the excluded cases where n = 1 and d ≥ 3, or (n, d) = (2, 4). Finally, in Section 4 we provide some applications to principally polarized abelian varieties that are the intermediate jacobians of Klein hypersurfaces. The authors are grateful to A.G. Gorinov for pointing us to his article, to Pierre-Marie Poloni for showing us Zsigmondy's Theorem, and to the referee for valuable suggestions. Date: August 23, 2018. 2000 Mathematics Subject Classification: Primary 14J40; Secondary 14J30. Key words: Automorphisms of hypersurfaces, linear automorphisms, cyclotomic polynomials, principally polarized abelian varieties. The first author was partially supported by the Fondecyt project 1110516 and the Dgip of the UTFSM. 1 2 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO 1. Orders of automorphisms of smooth hypersurfaces of Pn+1 Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). In this section we give a criterion as to when a positive integer q, with gcd(q, d) = gcd(q, d − 1) = 1, appears as the order of an automorphism of some smooth hypersurface of degree d in Pn+1. Let V be a vector space over C of dimension n + 2, n ≥ 2 with a fixed basis, and let Pn+1 = P(V ) be the corresponding projective space. We also let {x0, . . . , xn+1} be the dual basis of the linear forms on V so that {xi1 · · · xid 0 ≤ i1 ≤ . . . ≤ id ≤ n + 1} is a basis of the vector space Sd(V ∗) of forms of degree d on V . For a form F ∈ Sd(V ∗), we denote by X = V (F ) ⊆ Pn+1 the corresponding hypersurface of dimension n and degree d. We denote by Aut(X) the group of regular automorphisms of X and by Lin(X) the subgroup of Aut(X) that extends to automorphisms of Pn+1. Since (n, d) 6= (2, 4), by [8, Theorems 1 and 2] if X is smooth then Aut(X) = Lin(X) and Aut(X) < ∞. In this setting Aut(X) < PGL(V ) and for any automorphism in Aut(X) we can choose a representative in GL(V ). This automorphism induces an automorphism of Sd(V ∗) such that ϕ(F ) = λF , λ ∈ C∗. These three automorphisms will be denoted by the same letter ϕ. In this paper we consider automorphisms of finite order q. In this case, multiplying by an appropriate constant, we can assume that ϕq = IdV , so that ϕ is also a linear automorphism of order q of V and ϕ(F ) = ξaF where ξ is a primitive q-th root of unity. Furthermore, we can apply a linear change of coordinates on V to diagonalize ϕ, so that ϕ : V → V, (α0, . . . , αn+1) 7→ (ξσ0 α0, . . . , ξσn+1αn+1), 0 ≤ σi < q. Definition 1.1. Letting Zq be the ring of integers modulo q, we define the signature σ of an automorphism ϕ as above by σ = (σ0, . . . , σn+1) ∈ Zn+2 q , where we identify σi with its class in the ring Zq. We also denote ϕ = diag(σ) and we say that ϕ is a diagonal automorphism. For every F ∈ Sd(V ∗) and i ∈ {0, . . . , n + 1}, we let degi(F ) denote the degree of F seen as a polynomial in xi. The following simple lemma is a key ingredient in the proof of Theorem 1.3. Lemma 1.2. Let X be a hypersurface of dimension n and degree d, given by the homogeneous form F ∈ Sd(V ∗). If degi(F ) ≤ d − 2, for some i ∈ {0, . . . , n + 1}, then X is singular. Proof. After a linear change of coordinates, we may and will assume that deg0(F ) ≤ d − 2 so that F = xd−2 0 L2 + xd−3 0 L3 + . . . + x0Ld−1 + Ld , where Lj is a form of degree j in the variables {x1, . . . , xn+1}. Hence, ∂F ∂x0 ∂F ∂xi = (d − 2)xd−3 0 L2 + (d − 3)xd−4 0 L3 + . . . + Ld−1, = xd−2 0 dL2 dxi + xd−3 0 dL3 dxi + . . . + dLd dxi , i ∈ {1, . . . , n + 1} . Now, the Jacobian criterion shows that the point (1 : 0 : . . . : 0) is singular. (cid:3) The following is the main result of this section. ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 3 Theorem 1.3. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). A positive integer q, with gcd(q, d) = gcd(q, d − 1) = 1 is the order of an automorphism of a smooth hypersurface of dimension n and degree d if and only if there exists ℓ ∈ {1, . . . , n + 2} such that (1 − d)ℓ ≡ 1 mod q . Proof. To prove the "only if" part, suppose that F ∈ Sd(V ∗) is a form of degree d such that the hypersurface X = V (F ) ⊆ Pn+1 is smooth and admits an automorphism ϕ of order q, with gcd(q, d) = gcd(q, d − 1) = 1. Without loss of generality, we assume that ϕ is diagonal and we let σ = (σ0, . . . , σn+1) ∈ Zn+2 be its signature. We have ϕ(F ) = ξaF , where ξ is a primitive q-th root of unity. Let b be such that d·b ≡ −a mod q, such a b always exits since gcd(q, d) = 1. Consider the automorphism ψ = ξbϕ of GL(V ). Clearly, ψ and ϕ induce the same automorphism in Pn+1. Furthermore, for the form F of degree d we have ψ(F ) = ξdbϕ(F ) = ξdb+aF = F . Hence, we may and will assume that ϕ(F ) = F . q Let now k0 be such that σk0 6≡ 0 mod q. By Lemma 1.2, F contains a monomial xd−1 k0 xk1 for some k1 ∈ {0, . . . , n+1} (not necessarily with coefficient 1). The form F is invariant by the diagonal automorphism ϕ so the monomial xd−1 xk1 is also invariant by ϕ i.e., (d−1)σk0 +σk1 ≡ 0 mod q, and so k0 σk1 ≡ (1 − d)σk0 mod q . (2) Furthermore, since gcd(q, d − 1) = 1 we have σk1 6≡ 0 mod q, and since gcd(q, d) = 1 we have k1 6= k0. Applying the above argument with k0 replaced by k1, we let k2 be such that the monomial xk2 is invariant by ϕ and is contained in F (not necessarily with coefficient 1). Iterating xki is a xd−1 k1 this process, for all i ∈ {3, . . . , n + 2} we let ki ∈ {0, . . . , n + 1} be such that xd−1 ki−1 monomial in F (not necessarily with coefficient 1) invariant by ϕ. By (2), we have σki ≡ (1 − d)σki−1 ≡ (1 − d)2σki−2 ≡ (1 − d)iσk0 mod q, ∀i ∈ {2, . . . , n + 2} , (3) and all of the σki are non-zero. Since ki ∈ {0, . . . , n + 1} there are at least two i, j ∈ {0, . . . , n + 2}, i > j such that ki = kj. Thus σki = σkj , and since σki ≡ (1 − d)iσk0 mod q and σkj ≡ (1 − d)j σk0 mod q, we have (1 − d)i−j ≡ 1 mod q , and so the "only if" part of the theorem follows. To prove the converse statement, let q be a positive integer such that gcd(q, d) = gcd(q, d − 1) = 1, and assume that there exists ℓ ∈ {1, . . . , n + 2} such that (1 − d)ℓ ≡ 1 mod q. We let F ∈ Sd(V ∗) be the form F = ℓ−1X i=1 xd−1 i−1 xi + xd−1 ℓ−1 x0 + n+1X i=ℓ xd i . By construction, the form F form admits the automorphism ϕ = diag(σ), where σ = (cid:0)1, 1 − d, (1 − d)2, . . . , (1 − d)ℓ−1, n+2−ℓ times z } { 0, . . . , 0 (cid:1) ∈ Zn+2 q . An easy modification of the argument in Example 3.5 below shows that X = V (F ) is smooth, proving the theorem. (cid:3) 4 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO Remark 1.4. Let ϕ = diag(σ) be an automorphism of order q of the smooth hypersurface X = V (F ), with gcd(q, d) = gcd(q, d − 1). As in the proof of Theorem 1.3, we may and will assume that ϕ(F ) = F and we let ℓ be as in Definition 2.1. If σ0 6= 0 is a component of the signature σ, then by (3) we have that (1−d)iσ0 is also a component of σ, ∀i < ℓ. Furthermore, if q is a prime number then replacing ϕ by ϕa, where a ∈ Zq is such that a · σ0 ≡ 1 mod q, we may assume that σ0 = 1. 2. Automorphisms of prime order of smooth hypersurfaces In this section we study the particular case of automorphisms of prime order p. In this case we are able to give a full characterization of the prime numbers that appear as the order of an automorphism of some smooth hypersurface of dimension n and degree d. We also show that the order of such an automorphism is bounded by (d − 1)n+1. Definition 2.1. We say that a prime number p is admissible in dimension n and degree d if either p divides (d − 1) or there exists ℓ ∈ {1, . . . , n + 2} such that This definition is justified by the following proposition. (1 − d)ℓ ≡ 1 mod p . Proposition 2.2. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). A prime number p is the order of an automorphism of a smooth hypersurface of dimension n and degree d if and only if p is admissible in dimension n and degree d. Proof. In the case where p does not divide d or d − 1, the proposition follows directly form Theorem 1.3. Let p be a prime number that divides d. The prime number p is admissible with ℓ = 2. Indeed, (1 − d)2 = 1 − 2d + d2 ≡ 1 mod p. On the other hand, for every n ≥ 2 and d ≥ 3, let X be the Fermat hypersurface i.e., X = V (F ) with F = xd 0 + xd 1 + . . . + xd n + xd n+1 . The hypersurface X is smooth and admits the automorphism or order d given by ϕ : V → V, (α0, α1, . . . , αn+1) 7→ (ξα0, α1, . . . , αn+1) , where ξ is a primitive d-th root of unity. Hence, X also admits an automorphism of order p. Let now p be a prime number that divides d − 1. The prime number p is admissible by definition. On the other hand, for n ≥ 2 and d ≥ 3, let X = V (F ) be the hypersurface given by F = xd−1 0 x1 + xd 1 + . . . + xd n + xd n+1 . A routine computation shows that the hypersurface X is smooth and admits the automor- phism or order d − 1 given by ϕ : V → V, (α0, α1, . . . , αn+1) 7→ (ξα0, α1, . . . , αn+1) , where ξ is a primitive (d − 1)-th root of unity. Hence, X also admits an automorphism of order p. (cid:3) Remark 2.3. Let n ≥ 3 and d ≥ 3. It is a trivial consequence of Zsigmondy's Theorem [16] that there exists at least one prime number admissible in dimension n and degree d that is not admissible in dimension n′ and degree d with n′ < n. See Theorem 1.1 in [11] for a modern statement of Zsigmondy's Theorem. Proposition 2.2 allows us to give, in the following corollary, a first bound for the prime numbers that appear as the order of an automorphism of some smooth hypersurface. This bound will be improved in Corollary 3.3. ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 5 Corollary 2.4. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). If a prime number p is the order of an automorphism of a smooth hypersurface of dimension n and degree d, then p < (d − 1)n+1. Proof. Suppose that p > (d − 1)n+1. By Proposition 2.2, p is admissible in dimension n and degree d, and so (1 − d)n+1 ≡ 1 mod p, or (1 − d)n+2 ≡ 1 mod p . This yields p = (1 − d)n+1 − 1, or k · p = (1 − d)n+2 − 1, k ∈ {1, . . . , d − 1} . Since 1 − d ≡ 1 mod d, we have that d is a divisor of p or k · p. In both cases, this yields gcd(p, d) 6= 1, which provides a contradiction since p > d. (cid:3) Remark 2.5. In [12] it is shown that the order of a linear automorphism of an n-dimensional projective variety of degree d is bounded by dn+1. Hence, in the particular case of prime orders, our bound above is already sharper. Remark 2.6. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). Let X = V (F ) be a smooth hypersurface of degree d of Pn+1. In [5, Theorem 2] it is shown that the order of Aut(X) divides B = 1 n + 1 nY i=0 1 i (cid:1) (cid:0)(−1)n−i+1 + (d − 1)n−i+2(cid:1) · lcm(cid:0)(cid:0)n+2 (cid:0)n+2 i (cid:1)(d − 1)i, (n + 2)(d − 1)n)(cid:1) . And in [5, Page 24, line 4] it is conjectured that every prime number p that divides B is the order of an automorphism of a smooth hypersurface of dimension n and degree d. We can prove this conjecture. Indeed, if a prime p divides B, then p divides (cid:0)(−1)n−i+1 + (d − 1)n−i+2(cid:1), p divides (d−1), or p ≤ n + 2. In the first two cases p is clearly admissible in dimension n and degree d. Assume now p ≤ n + 2 and p does not divide (d − 1). Since p does not divide (d − 1), we have (1 − d)ℓ 6≡ 0 mod p, for all ℓ ∈ Z≥0. Since p ≤ n + 2 there exists ℓ, ℓ′ ∈ {1, . . . n + 2}, mod p. Hence (1 − d)ℓ−ℓ′ ≡ 1 mod p and so p is ℓ > ℓ′ such that (1 − d)ℓ ≡ (1 − d)ℓ′ admissible in dimension n and degree d. Now the conjecture follows from Proposition 2.2. The criterion in Theorem 1.3 is easily computable. As en example, in Table 1 we give all the admissible prime numbers in degree 4 for different values of the dimension n. admissible primes 2, 3, 5, 7, 13, 61, 547 2, 3, 5, 7, 13, 41, 61, 547 2, 3, 5, 7, 61 2, 3, 5, 7, 13, 61 n 3 4 5 6 7 8 9 10 2, 3, 5, 7, 11, 13, 19, 37, 41, 61, 67, 73, 547, 661 2, 35, 7, 11, 13, 19, 37, 41, 61, 67, 547, 661 2, 3, 5, 7, 11, 13, 19, 37, 41, 61, 547 2, 3, 5, 7, 13, 19, 37, 41, 61, 547 Table 1. Admissible prime numbers in degree 4 for 3 ≤ n ≤ 10. In Table 2 we give the maximal admissible prime number p for small n and d. 6 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO n\d 2 3 4 5 6 7 8 9 4 - 61 61 547 547 547 547 9 3 19 5 331 11 331 11 5419 43 5419 43 87211 43 43 87211 683 661 2113 5281 51828151 10746341 87211 8 43 191 191 911 1201 117307 117307 7 37 101 101 197 1297 46441 46441 6 13 521 521 521 521 5167 5167 5 17 41 41 113 257 257 257 Table 2. Maximal admissible prime for 2 ≤ n ≤ 9, 3 ≤ d ≤ 9, and (n, d) 6= (2, 4). 3. Smooth hypersurfaces admitting an automorphism of prime order p > (d − 1)n In this section we show the following result: A smooth hypersurface X of dimension n and degree d admits an automorphism of prime order p > (d − 1)n if and only if X is the Klein hypersurface, n = 2 or n + 2 is prime, and p = Φn+2(1 − d), where Φn+2 is the (n + 2)-th cyclotomic polynomial. First, we recall some results about cyclotomic polynomials, see [7, Ch. VI, §3] for proofs. Definition 3.1. For every m ∈ Z>0, the m-th cyclotomic polynomial is defined as Φm(t) = Y (t − ξ) , where the product is over all primitive m-th roots of unity ξ. ξ It is well known that Φm(t) is irreducible over Q and has integer coefficients. Furthermore, a routine computation shows that Φ1(t) = t − 1 and for every q prime and r ≥ 1 Φq(t) = tq−1 + tq−2 + . . . + 1, and Φqr (t) = Φq(tqr−1 ) . The main result about cyclotomic polynomials that we will need in the sequel is the fol- lowing factorization tn − 1 = Y Φr(t) , rn where rn means r is a divisor of n. Our next theorem gives a criterion for the existence of a smooth hypersurface of dimension n and degree d admitting an automorphism of prime order p > (d − 1)n. For the proof we need the following simple inequalities. Φ1(1 − d) = −d, Φ2(1 − d) = 2 − d, Φ4(1 − d) = d2 − 2d + 2 > (d − 1)2, (d − 1)q−2 < Φq(1 − d) = (1 − d)q−1 + . . . + 1 < (d − 1)q−1, for all q ≥ 3 prime . and Lemma 3.2. Let n ≥ 2, d ≥ 3 be integers, and (n, d) 6= (2, 4). There exists a smooth hypersurface of dimension n and degree d admitting an automorphism of prime order p > (d − 1)n if and only if n = 2 or n + 2 is prime, and p = Φn+2(1 − d). Proof. We prove first the "only if" part of the lemma. Assume that there exists a smooth hypersurface of degree n and dimension d admitting an automorphism of prime order p > (d−1)n. By Proposition 2.2, p is admissible in dimension n and degree d and by Corollary 2.4, p is not admissible in dimension n − 1 and degree d. Hence (1 − d)n+2 ≡ 1 mod p and so (1 − d)n+2 − 1 = k · p, for some k ∈ {−(d − 1)2, . . . , (d − 1)2} . ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 7 If n = 2 then (1 − d)n+2 − 1 = Φ1(1 − d) · Φ2(1 − d) · Φ4(1 − d) , and the only possibility is k = Φ1(1 − d)Φ2(1 − d) = d(d − 2) and p = Φ4(1 − d) = d2 − 2d + 2. If n + 2 is prime, then (1 − d)n+2 − 1 = Φ1(1 − d) · Φn+2(1 − d) , and the only possibility is k = Φ1(1 − d) = −d and p = Φn+2(1 − d) = (1 − d)n+1 + . . . + 1. To finish this direction of the proof, we have to show that these two are the only possible cases. If n 6= 2 and n + 2 is not prime, then n + 2 = q · n′, or n + 2 = 2i , where q ≥ 3 is a prime number,n′ ≥ 2, and i ≥ 3. Assume fist that n + 2 = q · n′. In this case (1 − d)n+2 − 1 = Φ1(1 − d) · Φq(1 − d) · P (1 − d) , for some polynomial P (t). Let k′ = Φ1(1 − d) · Φq(1 − d). Since k′ < d(d − 1)q−1 < p, k is a multiple of k′. But k′ > d(d − 1)q−1 > (d − 1)2 which provides a contradiction. Finally, assume that n + 2 = 2i, i ≥ 3. In this case (1 − d)n+2 − 1 = Φ4(1 − d) · P (1 − d) , for some polynomial P (t). Let k′ = Φ4(1 − d) = (d − 1)2 + 1. Since k′ < (d − 1)3 < p, k is a multiple of k′. but k′ > (d − 1)2 which provides a contradiction. To prove the "if" part, let n = 2 or n + 2 be prime, and assume that Φn+2(1 − d) is prime. In both cases Φn+2(1 − d) ≥ (1 − d)n. If n + 2 is prime, then (1 − d)n+2 − 1 = Φ1(1 − d) · Φn+2(1 − d) ≡ 0 mod Φn+2(1 − d) , and so Φn+2(1 − d) is admissible in dimension n and degree d. If n = 2, then (1 − d)4 − 1 = Φ1(1 − d) · Φ2(1 − d) · Φ4(1 − d) ≡ 0 mod Φ4(1 − d), . Hence, Φ4(1 − d) is admissible in dimension 2 and degree d. This completes the proof. (cid:3) In the following corollary, that follows directly form Lemma 3.2, we give a sharp bound for the order of an automorphism of a smooth hypersurface of dimension n and degree d. Corollary 3.3. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). Assume that a smooth hypersurface of dimension n and degree d admits an automorphism of prime order p. (i) If n = 2 or n + 2 is prime, and Φn+2(1 − d) is prime, then p ≤ Φn+2(1 − d). This bound is sharp. (ii) In any other case, p < (d − 1)n. Remark 3.4. (i) The condition in Lemma 3.2 that Φn+2(1 − d) is prime is fulfilled, for instance, in the cases where (n, d) is (2, 3), (2, 5), (2, 7), (3, 3), (3, 4), (3, 6), (5, 3),(5, 4), (9, 3), and (9, 7). See Table 2. (ii) Assume that (n, d) is such that Φn+2(1 − d) is prime and n 6= 2. Then p = (1 − d)n+2 − 1 (1 − d) − 1 . Prime numbers of this form are usually known as generalized Mersenne primes or repunit primes. For d = −1 they correspond to the classical Mersenne primes and for d = 3 they are usually called Wagstaff primes. It is conjectured that there are infinitely many such primes [15, 9]. In the following example we define the classical Klein hypersurfaces that will be the subject of the remaining of this section. 8 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO Example 3.5. For any n ≥ 1 and d ≥ 2, we define the Klein hypersurface of dimension n and degree d as X = V (F ) ∈ Pn+1, where F = xd−1 0 x1 + xd−1 1 x2 + . . . + xd−1 n xn+1 + xd−1 n+1x0 . (4) It is well known that X is smooth except in the case where d = 2 and n ≡ 2 mod 4. In lack of a good reference we provide a short argument. Proof. Assume that α = (α0 : . . . : αn+1) ∈ X is a singular point i.e., F (α) = 0, and ∂F ∂xi (α) = 0 . It is clear from the equations ∂F ∂xi (α) = 0 imply the equations xi ∂F ∂xi (α) = 0 that αi 6= 0, for all i ∈ {0, . . . , n + 1}. Furthermore, αd−1 i αi+1 = (1 − d)αd−1 i+1 αi+2 = (1 − d)2αd−1 i+2 αi+3 = . . . = (1 − d)n−iαd−1 n αn+1 = (1 − d)n−i+1αd−1 n+1α0 . Hence, F (α) = R · αn+1α0, where R = n+1X (1 − d)i . i=0 If d 6= 2, then R 6= 0 and so F (α) = 0 implies α0 = 0 or αn+1 = 0 which provides a contradiction. In the case where d = 2 a routine computation shows that the quadratic form F is singular if and only if n ≡ 2 mod 4. (cid:3) This kind of hypersurfaces were first introduced by Klein who studied the automorphism group of the Klein hypersurface of dimensions 1, 3 and degree 3 [6]. For the proof of the theorem below, we need the following simple lemma that follows from the uniqueness of the decomposition of an integer in base (1 − d). Lemma 3.6. Let d ≥ 3 and ai ∈ {1, . . . , d − 2}, 0 ≤ i ≤ n + 1. If Pi ai(1 − d)i = 0, then ai = 0 for all i. The following is the main result of this section. Theorem 3.7. Let n ≥ 2 and d ≥ 3 be integers, and (n, d) 6= (2, 4). A smooth hypersurface X = V (F ) of dimension n and degree d admits an automorphism ϕ of prime order p > (d−1)n if and only if X is isomorphic to the Klein hypersurface, n = 2 or n + 2 is prime, and p = Φn+2(1 − d). Proof. Since p > (d − 1)n, by Corollary 2.4, p is not admissible in dimension n − 1 and degree d. Hence, by Remark 1.4, we can assume that ϕ(F ) = F and ϕ = diag(σ), where σ = (σ0, . . . , σn+1) = (cid:0)1, (1 − d), (1 − d)2, . . . , (1 − d)n+1(cid:1) . The Klein hypersurface defined by the form in (4) admits the automorphism ϕ above. This together with Lemma 3.2 proves the "if" part. Assume now that X = V (F ) is a smooth hypersurface of dimension n and degree d admitting the automorphism ϕ of prime order p > (d−1)n. Let E ⊂ Sd(V ∗) be the eigenspace associated to the eigenvalue 1 of the linear automorphism ϕ : Sd(V ∗) → Sd(V ∗), so that F ∈ E. In the following we compute a basis for E. Let xα be a monomial in Sd(V ∗) i.e. xα := xα0 0 · · · xαn+1 n+1 , n+1X i=0 αi = d, and αi ≥ 0 . ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 9 We have xα ∈ E ⇔ L := α0 + α1(1 − d) + . . . + αn+1(1 − d)n+1 ≡ 0 mod p . Since αn+1 = d − Pn i=0 αi, we have L = d(1 − d)n+1 + = d(1 − d)n+1 + d · nX i=0 αi(cid:0)(1 − d)i − (1 − d)n+1(cid:1) nX αi(cid:0)(1 − d)i + . . . + (1 − d)n(cid:1) . i=0 Letting βi = Pn j=0 αj, for all 0 ≤ i ≤ n, we have and 0 ≤ βi ≤ βj ≤ d, for all i < j, and L = d · M, where M = β0 + β1(1 − d) + . . . + βn(1 − d)n + (1 − d)n+1 . Since d is invertible in Zp we have xα ∈ E ⇔ L ≡ 0 mod p ⇔ M ≡ 0 mod p . By Lemma 3.2 we know that p = Φn+2(1 − d) and n = 2 or n + 2 is prime. We divide the proof in two cases. n + 2 is prime: In this case p = 1 + (1 − d) + . . . + (1 − d)n+1. If βn < d − 1 then n + 2 Case n + 2 M = p and Lemma 3.6 shows that βi = 1, ∀i. This corresponds to xα = xd−1 n+1x0. If βn = d − 1 then M = 0 and Lemma 3.6 shows that βi = 0, ∀i < n. This corresponds to xα = xd−1 n xn+1. If βj = d, ∀j > k + 1 and βk < d, for some k < n then M ≡ β0 + . . . + βk(1 − d)k + (1 − d)k+1 mod p This gives βk = (d − 1) and βi = 0, for all i < k. This corresponds to xα = xd−1 k xk+1. Hence, E = hxd−1 n+1x0, xd−1 k xk+1; 0 ≤ k ≤ ni and F = a0 · xd−1 0 x1 + a1 · xd−1 1 x2 + . . . + an · xd−1 n xn+1 + an+1 · xd−1 n+1x0 . Since X = V (F ) is smooth, by Lemma 1.2 ai coordinates we can put 6= 0, ∀i and applying a linear change of F = xd−1 0 x1 + xd−1 1 x2 + xd−1 2 x3 + xd−1 3 x0 . n = 2: In this case p = (d − 1)2 + 1 and so M = (β0 − β2) + (β1 − 1)(1 − d). If M = 0 n = 2 Case n = 2 then β0 = β1 = β2 = 1, or β0 = β1 = 0 and β2 = d − 1. This corresponds to xα = xd−1 3 x0 and xα = xd−1 2 x3, respectively. If M = p then β0 = d − 1 and β1 = β2 = d, or β0 = 0, β1 = d − 1 and β2 = d. This corresponds to xα = xd−1 Hence, E = hxd−1 0 x1 and xα = xd−1 2 x3, xd−1 0 x1 + a1 · xd−1 0 x1, xd−1 F = a0 · xd−1 1 x2, xd−1 1 x2, respectively. 3 x0i and 1 x2 + a2 · xd−1 2 x3 + a3 · xd−1 3 x0 . With the same argument as above, we can apply a linear change of coordinates to put F = xd−1 0 x1 + xd−1 1 x2 + xd−1 2 x3 + xd−1 3 x0 . (cid:3) Let now (n, d) be a pair satisfying the condition of Theorem 3.7 and let ϕ be the automor- phism of order p = Φn+2(1 − d) of the Klein hypersurface X = V (F ). In the remaining of this section, we study the geometry of the action of the cyclic group hϕi ≃ Z/pZ on X. Recall first that ϕ = diag(σ), where σ = (cid:0)1, (1 − d), (1 − d)2, . . . , (1 − d)n+1(cid:1). Since σi 6≡ σj mod p for all i 6= j, the only fixed points of ϕ are the images of the n + 2 standard basis vectors of V = Cn+2 in Pn+1. 10 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO For our next result, we say that a cyclic quotient singularity is of type 1 p (a1, . . . , an) if it is locally isomorphic to the singularity at the vertex of the quotient of Cn by Z/pZ, where the (Z/pZ)-action is generated by the automorphism (x1, . . . xn) 7→ (ξa1x1, . . . , ξan xn), with ξp = 1 and ξ 6= 1 . Proposition 3.8. Let n, d, p be as in Theorem 3.7. The quotient Y = X/(Z/pZ) of the Klein hypersurface by the cyclic group generated by ϕ has n + 2 singular points of singularity type 1 p (cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1). Proof. The set of singular points of Y is contained in the image under the quotient map of the set of fixed points of the (Z/pZ)-action on X given by ϕ. Furthermore, the Klein hypersurface X admits the automorphism ψ : Pn+1 → Pn+1, (x0 : . . . : xn+1) 7→ (x1 : . . . : xn+1 : x0) , and since the orbit of the fixed point α = (1 : . . . : 0) by hψi is the n + 2 fixed points, the singularity type of all the singular points of Y is the same. To compute the singularity type of the image of the point α in Y , we pass to the invariant affine open set U = {x0 6= 1} ≃ Cn+1 in Pn+1 with coordinates x1, . . . xn+1. Now, the fixed point α corresponds to ¯0 ∈ U , the Klein hypersurface XU is given by the equation x1 + xd−1 1 x2 + . . . + xd−1 n xn+1 + xd−1 n+1 = 0 , and the automorphism ϕU is given by ϕU = diag(cid:0)(1 − d) − 1, (1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1) . Since α ∈ XU is a smooth point, the singularity type of the image of α in the quotient, is the same as the singularity type at the image of the origin of the quotient of the tangent space TαX by the linear action eϕ induced by ϕ. The tangent space TαX is given by {x ∈ Cn+1 x1 = 0} ≃ Cn and eϕ = diag(cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1) . p (cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1). To Hence, the singularity type of the image of α is 1 complete the proof, we remark that eϕ is not a pseudo-reflection and so the image of p is indeed a singular point. (cid:3) A cyclic quotient singularity is always Cohen-Macaulay and rational but not necessarily Gorenstein. By [14], a cyclic quotient singularity Cn/G is Gorenstein if and only if the acting group G is a subgroup of SL(Cn). In the following corollary, we apply this result to prove that the singularities of Y are never Gorenstein. Corollary 3.9. Let n, d, p be as in Theorem 3.7. Then the singularities of Y = X/(Z/pZ) are not Gorenstein. p (cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1), we only Proof. Since the singular points of Y are of type 1 have to show that the automorphism of Cn given by diag(cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1) does not belong to SL(Cn). This happens if and only if (1 − d)2 − 1 + . . . + (1 − d)n+1 − 1 6≡ 0 mod p . Since p = Φn−2(1 − d), we have (1 − d)2 − 1 + . . . + (1 − d)n+1 − 1 = p + d − n − 2 . and so Y has Gorenstein singularities if and only if d = n + 2. Finally, a straightforward computation shows that if d = n + 2, then n divides Φn+2(1 − d) = p, which provides a contradiction. (cid:3) ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 11 Remark 3.10. The condition p > (d − 1)n coming from Theorem 3.7 in the above corollary is essential. Indeed, if we let n = 3 and d = 5, then the Klein hypersurface X admits an automorphism ϕ of order p = 41 < (d−1)n = 64 with the same signature σ as in Theorem 3.7, σ = (cid:0)1, (1 − d), (1 − d)2, . . . , (1 − d)n+1(cid:1) = (1, 37, 16, 18, 10) . By the proof Theorem 3.7, the singular points of quotient of X by hϕi are of type 41 (15, 17, 9). Now, this quotient singularity is Goren- p (cid:0)(1 − d)2 − 1, . . . , (1 − d)n+1 − 1(cid:1) = 1 1 stein since 15 + 17 + 9 = 41 ≡ 0 mod 41. 4. An application to intermediate jacobians of Klein hypersurfaces Letting (n, d) satisfy Theorem 3.7 we let X be the Klein hypersurface of dimension n and degree d. Let also ϕ be the automorphism of X of prime order Φn+2(1 − d). In [10] Shioda studies the action of ϕ on the Lieberman Jacobian of X. In particular, the author proves that the spectrum of the action of ϕ in H n(X, C) is composed by all the primitive roots of unity with multiplicity one. We remark that the "naive" question [10, Section 3 (b)] is not so naive by our Remark 3.4 (ii). Here, we study in more detail the particular cases where the intermediate jacobian J (X) admits a principal polarization. Let X be a smooth hypersurface of degree d of Pn+1. It is known [2, Page 22], that the intermediate jacobian J (X) is a non trivial principally polarized abelian variety (p.p.a.v.) if and only if n = 1 and d ≥ 3, n = 3 and d = 3, 4, or n = 5 and d = 3. In this case, the dimension of the cohomology H n(X, C) is given by [13] dim H n(X, C) = (d − 1)n+2 + (−1)n(d − 1) d . It is also possible using the residue calculus of Griffiths to give an explicit representa- tion of H n(X, C). Indeed, let X = V (F ) where F is a form of degree d and let Sl = H 0(Pn+1, OPn+1(l)), for all l ≥ 0 so that S = Ll Sl is the polynomial ring in n + 2 variables. We also let JF = Ll J l F the homogeneous jacobian ideal of F generated by the partial deriva- tives ∂F F be the l-component of the jacobian ring RF = S/JF . With all ∂xi this definitions, we have [13] F = Sl/J l , and Rl H n+1−r,r−1(X, C) ≃ Rrd−n−2 F . (5) Let now X be the Klein hypersurface given by V (F ), where F is as in (4). In the following, we will study the p.p.a.v. J (X) in the case where (n, d) = (3, 3), (n, d) = (3, 4), and (n, d) = (5, 3). Letting k = n−1 2 we let ι be the canonical injection ι : H n(X, Z) ֒→ H k,k+1(X, C) so that J (X) = H k,k+1(X, C)/ι(H n(X, Z)) . By Theorem 3.7, X admits an automorphism ϕ of order p = 11, p = 61, and p = 43, respectively. The automorphism ϕ preserves the Hodge structure of X and since this Hodge structure is of level 1, ϕ induces an automorphism eϕ of the p.p.a.v. J (X). Let Ag be the moduli space of p.p.a.v. of dimension g and let Sing Ag its singular locus. If a prime number p is the order of an automorphism A ∈ Ag, then p ≤ 2g +1. In the case where p = 2g + 1, the p.p.a.v. A is called extremal and corresponds to 0-dimensional irreducible components of Sing Ag in the sense of [4]. As we will see below, J (X) is extremal. 4.1. The cubic Klein threefold. This is the case where (n, d) = (3, 3). In this case J (X) is p.p.a.v of dimension 5 and since eϕ is of order 11, J (X) is indeed extremal. By (5) for r = 3 we have H 1,2(X, C) ≃ R4 F , and a basis for H 1,2(X, C) via this isomorphism is given by {x1x2x3x4, x0x2x3x4, x0x1x3x4, x0x1x2x4, x0x1x2x3} . 12 V´ICTOR GONZ ´ALEZ-AGUILERA AND ALVARO LIENDO The automorphism ϕ of order p = 11 of the Klein hypersurface is given by diag(1, 9, 4, 3, 5). Hence, the spectrum of the induced isomorphism ϕ : T0J (X) ≃ H 1,2(X, C) → T0J (X) is given by C = {ξ10, ξ2, ξ7, ξ8, ξ6}. Since C ∩ C = ∅ and C ∪ C corresponds to all the primitive 11-th roots of unity, J (X) is p.p.a.v. of complex multiplication type [1] and is a 0-dimensional component of the singular locus of A5 [4]. Furthermore ψ(a) = a5 stabilize the complex multiplication type C of J (X) and induce a permutation of coordinates on T0J (X) given by σ = (10, 6, 8, 7, 2). Thus, J (X) is contained in a component of Sing A5 corresponding to p.p.a.v. admitting the automorphisms σ of order 5. Let us denote this component by A5(5, σ). The spectrum of σ is {ζ 0, ζ 1, ζ 2, ζ 3, ζ 4}, where ζ is a primitive 5-th root of unity. Then it follows from [4, Page 1] that A5(5, σ) is a 3-dimensional subvariety of Sing A5 that contain J (X). 4.2. The cubic Klein fivefold. This is the case where (n, d) = (5, 3). In this case J (X) is By (5) for r = 3 we have H 2,3(X, C) ≃ R2 p.p.a.v of dimension 21 and since eϕ is of order 43, J (X) is indeed extremal. F , and a basis for H 2,3(X, C) via this isomorphism is given by {xixj ∈ S2(V ∗) 0 ≤ i < j ≤ 6}. The automorphism ϕ of order p = 43 of the Klein hypersurface X is given by ϕ = diag(1, 41, 4, 35, 16, 11, 21) . Hence, the spectrum of the induced isomorphism ϕ : T0J (X) ≃ H 2,3(X, C) → T0J (X) is given by C = {ξ2, ξ3, ξ5, ξ8, ξ9, ξ12, ξ13, ξ14, ξ15, ξ17, ξ19, ξ20, ξ22, ξ25, ξ27, ξ32, ξ33, ξ36, ξ37, ξ39, ξ42} . Since C ∩ C = ∅ and C ∪ C corresponds to all the primitive 43-th roots of unity, J (X) is p.p.a.v. of complex multiplication type [1] and is a 0-dimensional component of the singular locus of A21 [4]. Furthermore ψ(a) = a11 stabilize the complex multiplication type C of J (X) and induce a permutation of coordinates on T0J (X) of order 7 given by σ = (2, 22, 27, 39, 42, 32, 8)(3, 33, 19, 37, 20, 5, 12)(9, 13, 14, 25, 17, 15, 36) Thus, J (X) is contained in a component of Sing A21 corresponding to p.p.a.v. admitting the automorphisms σ of order 7. Let us denote this component by A21(7, σ). The spectrum of σ is {ζ 0, ζ 0, ζ 0, ζ 1, ζ 1, ζ 1, ζ 2, ζ 2, ζ 2, ζ 3, ζ 3, ζ 3, ζ 4, ζ 4, ζ 4, ζ 5, ζ 5, ζ 5, ζ 6, ζ 6, ζ 6} , where ζ is a primitive 7-th root of unity. Then it follows from [4, Page 1] that A21(7, σ) is a 33-dimensional subvariety of Sing A21 that contain J (X). 4.3. The quartic Klein threefold. This is the case where (n, d) = (3, 4). In this case J (X) is p.p.a.v of dimension 30 and since eϕ is of order 61, J (X) is indeed extremal. By (5) for r = 3 we have H 1,2(X, C) ≃ R3 F , and a basis for H 1,2(X, C) via this isomorphism is given by {xixjxk ∈ S3(V ∗) 0 ≤ i ≤ j ≤ k ≤ 4, and i 6= k} . The automorphism ϕ of order p = 61 of the Klein hypersurface X is given by ϕ = diag(1, 58, 9, 34, 20). Hence, the spectrum of the induced isomorphism ϕ : T0J (X) ≃ H 1,2(X, C) → T0J (X) is given by {ξ60, ξ11, ξ36, ξ22, ξ56, ξ7, ξ32, ξ18, ξ19, ξ44, ξ30, ξ8, ξ55, ξ41, ξ3 ξ28, ξ14, ξ15, ξ40, ξ26, ξ4, ξ51, ξ37, ξ52, ξ38, ξ16, ξ2, ξ49, ξ27, ξ13} . ON THE ORDER OF AN AUTOMORPHISM OF A SMOOTH HYPERSURFACE 13 Since C ∩ C = ∅ and C ∪ C corresponds to all the primitive 61-th roots of unity, J (X) is p.p.a.v. of complex multiplication type [1] and is a 0-dimensional component of the singular locus of A30 [4]. Furthermore ψ(a) = a9 stabilize the complex multiplication type C of J (X) and induce a permutation of coordinates on T0J (X) of order 5 given by σ = (2, 18, 40, 55, 7)(3, 27, 60, 52, 41)(4, 36, 19, 49, 14) (8, 11, 38, 37, 28)(13, 56, 16, 22, 15)(26, 51, 32, 44, 30) . Thus, J (X) is contained in a component of Sing A30 corresponding to p.p.a.v. admitting the automorphisms σ of order 5. Let us denote this component by A30(5, σ). The spectrum of σ is 6 times z { ζ 0, . . . , ζ 0, { } 6 times z { ζ 1, . . . , ζ 1, } 6 times z { ζ 2, . . . , ζ 2, } 6 times z { ζ 3, . . . , ζ 3, } 6 times z { ζ 4, . . . , ζ 4} , } where ζ is a primitive 5-th root of unity. Then it follows from [4, Page 1] that A30(5, σ) is a 93-dimensional subvariety of Sing A30 that contain J (X). References [1] Bennama, H. and Bertin, J. Remarque sur les vari´et´es ab´eliennes avec un automorphisme d'ordre premier. Manuscripta Math. 94 (1997), 409 -- 425. [2] Deligne, P. Cohomologie des intersections compl`etes. Groupes de Monodromie en G´eom´etrie Alge- brique. SGA 7 II, Expos´e XI. Springer Lectures Notes. 340, 1973. [3] Gonz´alez-Aguilera, V. and Liendo, A. Automorphisms of prime order of smooth cubic n-folds. Arch. Math, 97 2011, 25 -- 37. [4] Gonz´alez-Aguilera, V., Munoz-Porras, J. and R. Zamora, A. On the 0-dimensional irreducible compo- nents of the singular locus of Ag. Archiv der Mathematik, 84, (2005), 298 -- 303. [5] Gorinov, A.G. Division theorems for the rational cohomology of some discriminant complements. arXiv:math/0511593v1 [math.AG], 27 pages. [6] Klein, F. Uber die Ausflsung Gleichungen siebenten und achten Grades. Math. Ann., 15, (1879), 251 -- 282. [7] Lang, S. Algebra. Third edition. Graduate Texts in Mathematics 211, Springer-Verlag, 2002. [8] Matsumura, H. and Monsky, P. On the automorphisms of hypersurfaces. J. Math. Kyoto Univ. (1964), 3, 347 -- 361. [9] Melham, R. Probable prime tests for generalized Mersenne numbers. Bol. Soc. Mat. Mexicana, 14, (2008), no. 1, 7 -- 14. [10] Shioda, T. A note on a theorem of Griffiths on the Abel-Jacobi map. Invent. Math., 82, (1985), no. 3, 461 -- 465. [11] Steidl, G, Hanler, M and Tasche, M. On a number-theoretic result of Zsigmondy in domains of quadratic integers. Arch. Math., Vol 54, (1989), 30 -- 39. [12] Szabo, E. Bounding automorphism groups. Mathematische Annalen, 304, (1996), 801 -- 811. [13] Voisin, C. Th´eorie de Hodge et g´eom´etrie alg´ebrique complexe. Cours sp´ecialis´es, 10, (2002) Soci´et´e Math´ematique de France. [14] Watanabe, K. Certain invariant subrings are Gorenstein. I, II. Osaka J. Math., 11, (1974), 1 -- 8 and 379 -- 388. [15] Williams, H. C., Seah, E. Some primes of the form (an − 1)/(a − 1). Math. Comp. 33 (1979), no. 148, 1337 -- 1342. [16] Zsigmondy, K. Zur Theorie der Potenzreste. Monatsh. Math. Phys., 3, (1892), 265 -- 284. Departamento de Matem´aticas, Universidad T´ecnica Federico Santa Mar´ıa, Casilla 110-V, Valpara´ıso, Chile. E-mail address: [email protected] Mathematisches Institut, Universitat Bern, Sidlerstrasse 5, CH -- 3012 Bern, Switzerland. E-mail address: [email protected]
1107.2717	1	1107	2011-07-14T02:48:40	Compact moduli spaces of surfaces of general type	[ "math.AG" ]	We give an introduction to the compactification of the moduli space of surfaces of general type introduced by Koll\'ar and Shepherd-Barron and generalized to the case of surfaces with a divisor by Alexeev. The construction is an application of Mori's minimal model program for 3-folds. We review the example of the projective plane with a curve of degree d > 3. We explain a connection between the geometry of the boundary of the compactification of the moduli space and the classification of vector bundles on the surface in the case H^{2,0}=H^1=0.	math.AG	math	Compact moduli spaces of surfaces of general type Paul Hacking Abstract. We give an introduction to the compactification of the moduli space of surfaces of general type introduced by Koll´ar and Shepherd-Barron and generalized to the case of surfaces with a divisor by Alexeev. The con- struction is an application of Mori's minimal model program for 3-folds. We review the example of the projective plane with a curve of degree d ≥ 4. We explain a connection between the geometry of the boundary of the compact- ification of the moduli space and the classification of vector bundles on the surface in the case H 2,0 = H 1 = 0. 1. Introduction The moduli space Mg of curves of genus g ≥ 2 admits a compactification Mg given by the moduli space of stable curves [DM69]. Here a stable curve C is a connected compact curve of arithmetic genus g having only nodal singularities (xy = 0) ⊂ C2 and finite automorphism group. Components of the boundary ∂Mg := Mg \ Mg correspond to the topological types of degenerations of a curve of genus g obtained by contracting a simple loop to a point. The space Mg is an orbifold of dimension 3g − 3 and the boundary is a normal crossing divisor, that is, the pair (Mg, ∂Mg) is locally isomorphic to C3g−3 together with a collection of coordinate hyperplanes modulo a finite group. The moduli space MK 2,χ of surfaces of general type with fixed topological invariants admits an analogous compactification MK 2,χ, the moduli space of stable surfaces [KSB88]. The definition of a stable surface is derived from Mori's minimal model program for 3-folds. Indeed, given a family of smooth surfaces of general type over a punctured disc, results of the minimal model program produce (after a finite base change) a distinguished extension of the family over the disc (the canonical model given by the projective spectrum of the canonical ring of a smooth model of the total space with special fiber a reduced normal crossing divisor). This is the exact analogue of the stable reduction theorem for curves, which uses the classical theory of minimal models of surfaces, cf. [DM69, 1.12]. A stable surface is then, roughly speaking, a surface which can arise as a limit of smooth surfaces in this way. The moduli space of stable surfaces gives a compactification of the moduli 2000 Mathematics Subject Classification. Primary 14J10. The author was partially supported by NSF grant DMS-0968824. I would like to thank T. Bridgeland, W. Chen, A. Corti, I. Dolgachev, D. Huybrechts, J. Koll´ar, J. Tevelev, R. Thomas, G. Urz´ua, and J. Wahl for helpful discussions and correspondence. 1 2 PAUL HACKING space of smooth surfaces, because it satisfies the valuative criterion of properness by construction. Very little is known about the moduli spaces MK 2,χ in general. In particular they can be essentially arbitrarily singular [V06], and have arbitrarily many con- nected components [C86]. However, we expect that there are many examples of connected components of the moduli space that are well behaved. A particularly attractive example is the case of surfaces of general type such that H 2,0 = 0 and π1 = 0 discussed in §9. We want to study the compactification given by moduli of stable surfaces in these cases. Perhaps the most basic question is: what are the codimension 1 com- ponents of the boundary? Here by definition the boundary is the complement of the locus of surfaces with Du Val singularities (as these are the canonical models of smooth surfaces of general type). An important type of boundary divisor is the locus of equisingular deformations of a normal surface with a unique cyclic quotient singularity of type C2 u,v/(Z/n2Z), where the action is given by (u, v) 7→ (ζu, ζna−1v), ζ = exp(2πi/n2) for some n, a with (a, n) = 1. These singularities were first studied by J. Wahl [W81, 5.9.1] so we refer to them as Wahl singularities. The smoothing component of the deformation space of a Wahl singularity rele- vant to moduli of stable surfaces is a smooth curve germ. So, at least in the absence of local-to-global obstructions, Wahl singularities occur in codimension 1. The Mil- nor fiber of the smoothing is a rational homology ball. Thus if Y X is a global degeneration which is locally of this form, specialization defines an isomorphism of rational homology H∗(Y, Q) → H∗(X, Q). For this reason it seems that it is not possible to predict the existence of the degeneration Y X based on the topology of Y alone. In this paper we explain a relation between boundary components of the moduli space of Wahl type and the classification of rigid holomorphic vector bundles on the smooth surface Y in the case H 2,0 = H 1 = 0 [H11]. We also describe in some detail a motivating example: the case of the projective plane Y = P2 together with a curve D of degree d ≥ 4 [H04]. This is an instance of the moduli space of stable surfaces with boundary, analogous to the moduli spaces of stable pointed curves Mg,n. In this example we are able to describe the components of the boundary of the moduli space fairly explicitly. In particular, the boundary divisors associated to Wahl singularities display a beautiful combinatorial structure that is intimately related to the classification of rigid vector bundles on the projective plane. Notation: We work over the complex numbers. We write Cr/ 1 n (a1, . . . , ar) or just 1 n (a1, . . . , ar) for the cyclic quotient singularity Cr/(Z/nZ) where the action is given by (x1, . . . , xr) 7→ (ζa1 x1, . . . , ζar xr), ζ = exp(2πi/n). 2. Moduli of stable surfaces 2.1. Semi log canonical singularities. Let X be a reduced Cohen-Macaulay surface. We assume that the set S ⊂ X of singularities which are not of double normal crossing type (xy = 0) ⊂ C3 is finite. The double curve ∆ ⊂ X of X is the union of the one dimensional components of the singular locus of X. Let ν : X ν → X be the normalization of X and ∆ν ⊂ X ν the inverse image of the COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 3 double curve. Thus the map ∆ν → ∆ is finite of degree 2. The dualizing sheaf ωX of the surface X is the correct generalization of the canonical line bundle for smooth surfaces. It has the following properties: (1) ωX is invertible away from the finite set S, (2) ωX is reflexive, (3) (ν∗ωX )∨∨ = ωX ν (∆ν ). Here F ∨∨ denotes the double dual (or reflexive hull ) of a coherent sheaf F . The dualizing sheaf ωX corresponds to a Weil divisor class denoted KX . Of course if X is normal this is the usual canonical divisor class. Let X be a normal surface and D a Q-divisor on X. We say the pair (X, D) is log canonical if KX + D is Q-Cartier and the following condition holds. Let π : X → X be a resolution of singularities of X such that the sum of the exceptional divisor E = P Ei of π and the strict transform D′ of D is a normal crossing divisor. Then K X + D′ = π∗(KX + D) + X aiEi in Pic( X) ⊗ Q for some ai ∈ Q. We require that ai ≥ −1 for each i. (It suffices to check the condition for one such resolution.) See [KM98, 4.1] for a detailed classification of log canonical surface-divisor pairs. Definition 2.1. Let X be a surface. We say X has semi log canonical (slc) singu- larities if the following conditions hold. (1) The surface X is reduced, Cohen-Macaulay, and has only double normal crossing singularities (xy = 0) ⊂ C3 away from a finite set of points. (2) With notation as above, the pair (X ν, ∆ν ) consisting of the normalization of X together with the inverse image of the double curve has log canonical singularities. (3) For some N > 0 the N th reflexive tensor power ω[N ] X := (ω⊗N X )∨∨ of the dualizing sheaf ωX is invertible. Semi log canonical surface singularities were classified in [KSB88, §4]. We describe the two examples we need in §4 and §5 below. Remark 2.2. The definition of semi log canonical singularities for X of dimension d is the same except that the Cohen-Macaulay condition is replaced by Serre's condition S2 and we require that X has only double normal crossing singularities (xy = 0) ⊂ Cd+1 away from a codimension 2 subset. 2.2. Stable surfaces. Definition 2.3. A stable surface is a connected projective surface X such that X has semi log canonical singularities and the dualizing sheaf ωX is ample. 2.3. Index one cover. Let (P ∈ X) be an slc surface germ. The index X is invertible. Fix an isomorphism of P ∈ X is the least integer such that ω[N ] θ : ω[N ] X → OX and define a morphism p : Z → X by Z := Spec X (OX ⊕ ωX ⊕ ω[2] X ⊕ · · · ⊕ ω[N −1] X ) where the multiplication on OZ is defined using θ. Then p : Z → X is a cyclic cover of degree N such that (1) the inverse image of P is a single point Q ∈ Z, (2) the morphism p is ´etale over X \ {P }, and (3) the surface Z is Gorenstein, that is, Z is Cohen-Macaulay and the dualizing sheaf ωZ is invertible. Moreover, the germ (Q ∈ Z) is slc. The covering p is called the index one cover of (P ∈ X). It is uniquely determined locally in the analytic topology. 4 PAUL HACKING 2.4. Q-Gorenstein deformations. Let (P ∈ X) be a slc surface germ. We say a deformation (P ∈ X )/(0 ∈ S) is Q-Gorenstein if it is induced by an equivari- ant deformation of the index one cover of (P ∈ X). Equivalently, let ωX/S denote the relative dualizing sheaf of X/S. The sheaf ωX/S is flat over S and commutes with base change because X/S is flat with Cohen- Macaulay fibers [C00, 3.6.1]. Then X/S is Q-Gorenstein iff every reflexive power ω[N ] X/S := (ω⊗N X/S)∨∨ of the relative dualizing sheaf ωX/S commutes with basechange [H04, 3.5]. (The definition of Q-Gorenstein deformations first appeared in this form in [K91, §6].) 2.5. Moduli stack. Definition 2.4. Let S be a scheme of finite type over C. A family of stable surfaces over S is a flat family X/S such that each fiber is a stable surface and X/S is Q- Gorenstein in the sense of §2.4, that is, everywhere locally on X the family X/S is induced by an equivariant deformation of the index one cover of the fiber. Let M denote the moduli stack of stable surfaces so defined. For K 2, χ ∈ Z let MK 2,χ denote the moduli stack of stable surfaces X with K 2 X = K 2, χ(OX ) = χ (thus MK 2,χ is a union of connected components of M). Theorem 2.5. [KSB88],[K90],[AM04]. The stack MK 2,χ is a proper Deligne- Mumford stack of finite type over C with projective coarse moduli space. 3. Infinitesimal study of the moduli stack Let X be a stable surface. By definition, the moduli stack M of stable surfaces near [X] is identified with the quotient stack [Def QG(X)/ Aut X] of the versal Q- Gorenstein deformation space Def QG(X) of X by the automorphism group of X (a finite group). In this section we explain how to compute Def QG(X). Let X be a slc surface. Recall that the Q-Gorenstein deformations of X are by definition those deformations which are locally induced by an equivariant deforma- tion of the index one cover. To treat global Q-Gorenstein deformations in a robust way it is best to rephrase this definition as follows. The data of index one covers everywhere locally on X defines a Deligne- Mumford stack X with coarse moduli space X, the index one covering stack. Let q : X → X denote the induced map. If p : V → U is an index one cover of an open set U ⊂ X with group G ≃ Z/N Z, then q : X → X is locally given by the map [V /G] → U = V /G from the quotient stack [V /G] to its coarse moduli space. The deformations of X are identified with the Q-Gorenstein deformations of X [H04, 3.7], [AH09]. Note that the map q : X → X is an isomorphism away from the finite set of points S ⊂ X where the dualizing sheaf ωX is not invertible, that is, the stack structure is trivial away from S. We consider coherent sheaves F on the ´etale site of X. If p is a local index one cover as above, the restriction of F to V is a G-equivariant sheaf FV on V , and F is determined by these restrictions together with gluing data on the overlaps. The pushforward q∗F of F may be defined locally by (q∗F )U = (p∗FV )G. Note that q∗ is exact. The deformations of X may be described using the cotangent complex LX/C for the structure sheaf OX on the ´etale site of X [I71], [I72]. We refer to [H10, §3] for an exposition of the local theory of the cotangent complex following [LS67]. We COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 5 QG,X := q∗ Exti(LX/C, OX) on X and C-vector spaces Ti QG,X := define sheaves T i Exti(LX/C, OX). Then the basic result is the following. Proposition 3.1. Let X be a slc surface. (1) The first order infinitesimal Q-Gorenstein deformations of X are identified with the C-vector space T1 QG,X . (2) Let A be a finitely generated Artinian local C-algebra. Let X /A be a Q-Gorenstein deformation of X over A. Let A′ → A be an infinitesimal extension of A with kernel I. Let m′ denote the maximal ideal of A′ and assume that m′ · I = 0 (so I is a A/m′ = C-vector space). Then there is QG,X ⊗ I such that o(X /A, A′) = 0 an obstruction class o(X /A, A′) ∈ T2 iff there exists a Q-Gorenstein deformation X ′/A′ of X /A over A′. If o(X /A, A′) = 0 then isomorphism classes of such deformations form a torsor for T1 QG,X ⊗ I. In particular, if X is compact, the versal Q-Gorenstein deformation space of QG,X and is smooth if X is an analytic germ (0 ∈ Def QG(X)) with tangent space T1 T2 QG,X = 0. V )G where T i QG,X U = (p∗T i Let p : V → U be an index one cover of an open set U ⊂ X with group G. V = Exti(LV /C, OV ). We recall that, V = TV := Hom(ΩV , OV ) is the tangent sheaf of V , V = 0 if V is a local complete intersection. See for QG,X = TX is the QG,X is supported on the singular locus of X and can be QG,X is supported on Then we have T i for any reduced scheme V , T 0 V = Ext1(ΩV , OV ) and T 2 T 1 example [A76, p. 33], [H10, 3.6, 4.13]. tangent sheaf of X, (2) T 1 computed locally by T 1 the locus where the index one cover is not a local complete intersection. QG,X U = (p∗ Ext1(ΩV , OV ))G, (3) T 2 It follows that (1) T 0 We have the local-to-global spectral sequence Ep,q 2 = H p(T q QG,X ) ⇒ Tp+q QG,X . In particular, we obtain the exact sequence 0 → H 1(TX ) → T1 Moreover, the obstruction space T2 and the local index one covers are complete intersections. QG,X → H 0(T 1 QG,X vanishes if H 2(TX ) = 0, H 1(T 1 QG,X ) → H 2(TX ). QG,X ) = 0, 4. Wahl singularities Let P ∈ X denote the cyclic quotient singularity C2 n2 (1, na − 1), where (a, n) = 1. These singularities were first studied by J. Wahl [W81, 5.9.1] so we refer to them as Wahl singularities. The index one cover (Q ∈ Z) → (P ∈ X) is the Du Val singularity of type An−1 u,v/ 1 C2 u,v/ 1 n (1, −1) = (xy = zn) ⊂ C3 x,y,z where x = un, y = vn, z = uv. Thus (P ∈ X) = (xy = zn) ⊂ 1 n (1, −1, a) and a Q-Gorenstein smoothing is given by (4.1) (P ∈ X ) = (xy = zn + t) ⊂ 1 n (1, −1, a) × C1 t . The one parameter deformation (P ∈ X )/(0 ∈ C1 t ) is the versal Q-Gorenstein In particular, if X is a stable surface with a unique deformation of (P ∈ X). 6 PAUL HACKING singularity P ∈ X of Wahl type and there are no local-to-global obstructions to Q-Gorenstein deformations of X, then the moduli stack of stable surfaces M is smooth near [X] and equisingular deformations of X determine a codimension 1 component of the boundary. 4.1. Topology of the smoothing. If P ∈ X is a normal surface singularity then the link L of P ∈ X is the smooth 3-manifold obtained as the intersection of X with a small sphere about the point P in some embedding X ⊂ CN . For example the link of the cyclic quotient singularity 1 r (1, a) is the lens space L = L(r, a) = S3/ 1 r (1, a) ⊂ C2/ 1 If (P ∈ X )/(0 ∈ T ) is a one-parameter smoothing of a normal surface singu- larity P ∈ X, the Milnor fiber M of the smoothing is defined as follows. Choose an embedding X ⊂ CN × T , say P 7→ (0, 0), and let B(δ) := {kzk ≤ δ} ⊂ CN be the closed ball with center 0 ∈ CN and radius δ. We define M = Xǫ ∩ B(δ) for 0 < ǫ ≪ δ ≪ 1. Then M is a smooth 4-manifold with boundary the link L of the singularity P ∈ X. r (1, a). Wahl observed that the Milnor fiber M of the smoothing (4.1) is a rational homology ball. More precisely, π1(M ) is cyclic of order n and Hi(M, Z) = 0 for i > 1. Indeed, the Milnor fiber M is the quotient of the Milnor fiber MZ of the smoothing of the An−1 singularity (Q ∈ Z) by a free Z/nZ-action. Now MZ is homotopy equivalent to a bouquet of n − 1 copies of the 2-sphere, in particular MZ is simply connected and e(MZ) = n. Hence π1(M ) = Z/nZ and e(M ) = e(MZ)/n = 1. Finally M is a Stein manifold of complex dimension 2 so has the homotopy type of a cell complex of real dimension 2. We deduce that Hi(M ) = 0 for i > 1 as claimed. The boundary of M is the link L = S3/ 1 π1(L) → π1(M ) is the surjection Z/n2Z → Z/nZ. n2 (1, na − 1) of P ∈ X and the map Let Y be a smooth compact surface. We would like to determine the possible degenerations of Y to a singular surface X which are locally of the form (4.1). It seems that topological considerations provide very little information because the topology of M is almost trivial. We return to this question in §§8–10 below, where we describe an approach in the case H 2,0(Y ) = H 1(Y ) = 0. 5. Orbifold double normal crossing singularities Let X be a surface which is a union of two smooth components X1, X2 meeting transversely along a curve C. That is, X has double normal crossing singularities (xy = 0) ⊂ C3 along C. Then the sheaf T 1 X is the line bundle on C given by the tensor product of the normal bundles of C in X1 and X2 X = Ext1(ΩX , OX ) = NC/X1 ⊗ NC/X2 . T 1 See [F83, 2.3]. If U ⊂ X is a local chart of the form (xy = 0) ⊂ C3 then T 1 X U is generated by x−1 ⊗ y−1, and a local section s = f (z) · x−1 ⊗ y−1 ∈ H 0(U, T 1 X ) corresponds to the first order infinitesimal deformation U = (xy = t · f (z)) ⊂ C3 x,y,z × (Spec C[t]/(t2)). A necessary condition for smoothability of X is H 0(T 1 X ) 6= 0 (because otherwise all deformations are locally trivial). Suppose X /(0 ∈ C1 t ) is a one-parameter deforma- tion of X and s ∈ H 0(T 1 X ) is the induced element (the local part of the induced first order infinitesimal deformation). Then the total space X is smooth away from COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 7 the zeroes of s and has ordinary double points (xy + zt = 0) at simple zeroes of s. In particular, if s has only simple zeroes then the general fiber is smooth. The orbifold double normal crossing singularity (xy = 0) ⊂ C3 x,y,z/ 1 n (1, −1, a), (a, n) = 1, is semi log canonical and occurs frequently on stable limits of smooth surfaces. The index one cover is the double normal crossing singularity (xy = 0) ⊂ C3. A Q-Gorenstein smoothing is given by (xy = t) ⊂ 1 n (1, −1, a) × C1 t . Let X be a surface that is a union of two normal components X1 and X2 meeting along a smooth curve C such that X has orbifold double normal crossing singularities along C and is smooth elsewhere. Then the Q-Gorenstein deformations of X are, by definition, locally induced by equivariant deformations of the index one cover, which is a double normal crossing singularity. The sheaf T 1 QG,X of local first order Q-Gorenstein deformations is a line bundle on C and may be computed as follows. Let CiC denote the Q-divisor on C, well defined up to linear equivalence, obtained by moving C on Xi and restricting to C. Then the sum C1C + C2C is a Z-divisor on C, and (5.1) T 1 QG,X = OC(C1C + C2C ). QG,X is a line bundle on C of degree (C2)X1 + (C2)X2 equal to the In particular T 1 sum of the self-intersection numbers of C on X1 and X2. If U ⊂ X is a local chart of the form (xy = 0) ⊂ 1 QG,XU is generated by x−1 ⊗ y−1, and a local section s = f (zn) · x−1 ⊗ y−1 corresponds to the first order deformation n (1, −1, a), then T 1 (xy = t · f (zn)) ⊂ 1 n (1, −1, a) × Spec(C[t]/(t2)). In particular, if s has a simple zero at the orbifold point (P ∈ C) = (0 ∈ C1 zn ) then locally this deformation may be extended to a one-parameter deformation with general fiber the Wahl singularity 1 n2 (1, na − 1). 6. Surfaces with boundary The moduli space of pairs (X, D) consisting of a smooth surface X and an effective Q-divisor D such that (X, D) is log canonical and KX + D is ample admits a compactification given by the moduli space of stable pairs [A96], [K11]. These compactifications are analogous to the moduli spaces of pointed stable curves Mg,n and their generalizations given by assigning weights to the marked points [H03]. Definition 6.1. Let (X, D) be a pair consisting of a surface X and an effective Q-divisor D. We say the pair (X, D) is a stable pair or stable surface with boundary if the following conditions hold. (1) The surface X is reduced, Cohen-Macaulay, and has double normal cross- ing singularities away from a finite set. No component of the double curve is contained in the support of the divisor D. (2) The pair (X ν, ∆ν + Dν), consisting of the normalization of X together with the sum of the inverse images of the double curve and the divisor D, is log canonical. (3) The Q-divisor KX + D is Q-Cartier and ample. 8 PAUL HACKING The definition of a family of stable pairs (X, D) over a scheme S involves technical difficulties in general [A08], [K10, §6]. These difficulties do not occur if D is a Z-divisor or if we use floating coefficients D = (α + ǫ)B where α ∈ [0, 1) ∩ Q and 0 < ǫ ≪ 1, ǫ ∈ Q. We give a precise definition in the example described in §7 below, where floating coefficients are used. 7. Plane curves A compactification of the moduli space of plane curves was obtained in the author's PhD thesis advised by A. Corti [H04]. It is an instance of the moduli space of stable surfaces with boundary. In this section we review the description of this example as it displays many features we expect to observe in the general case. Let Pd denote the moduli stack of smooth plane curves of degree d ≥ 4. Ex- plicitly Pd is the quotient stack [U/ PGL(3)] where U ⊂ PN is the complement of the discriminant locus in the projective space of homogeneous polynomials of degree d in 3 variables, N = 1 2 d(d + 3). We regard Pd as the moduli stack of pairs (X, D) consisting of a surface X isomorphic to P2 and a divisor D on X given by a smooth plane curve of degree d. The pair (X, αD) has log canonical singularities and KX +αD is ample for 3/d < α ≤ 1, α ∈ Q. One can thus define a compactifica- α tion P d of Pd by taking the closure of the image of Pd in the moduli stack of stable α pairs under the map [D] 7→ [(X, αD)]. Roughly speaking, the compactification P d becomes more complicated as α increases. So our approach is to consider the case where α is "as small as possible". More precisely, there exists ǫ0 > 0 such that our compactification P d coincides with P α d for 3/d < α < 3/d + ǫ0. As noted in §6, the definition of the moduli stack of stable pairs in general is rather technical. So we define the moduli stack P d directly as follows. Definition 7.1. A stable plane pair of degree d is a pair (X, D) consisting of a surface X together with an effective Weil divisor D such that the pair (X, (3/d+ǫ)D) has semi log canonical singularities and the Q-Cartier divisor KX + (3/d + ǫ)D is ample for 0 < ǫ ≪ 1, ǫ ∈ Q. Let (X, D) be a stable plane pair. Then in particular X is slc and D is Q- Cartier. Let P ∈ X be a point and p : (Q ∈ Z) → (P ∈ X) be the index one cover. Let DZ be the divisor on Z given by the inverse image of D on X. We say (X, D) is pre-smoothable if DZ is Cartier for each P ∈ X. (This condition simplifies the deformation theory of pairs considered below. One can show that it is satisfied by the stable limits of pairs (P2, C) consisting of the plane together with a curve [H04, 3.13].) Let (X, D) be a pre-smoothable stable plane pair. We regard D as a closed subscheme D ⊂ X of codimension 1 (without embedded points). By a deformation of the pair (X, D) over a germ (0 ∈ S) we mean a pair (X , D)/(0 ∈ S) consisting of a deformation X /(0 ∈ S) of X together with a closed subscheme D ⊂ X such that D0 = D and D is flat over S. We say a deformation of (P ∈ X, D) is Q-Gorenstein if it is induced by an equivariant deformation of (Q ∈ Z, DZ). A family of pre-smoothable stable plane pairs over a scheme S is a flat fam- ily (X , D)/S such that each fiber is a pre-smoothable stable plane pair and the family is Q-Gorenstein, that is, everywhere locally on X induced by an equivari- ant deformation of the index one cover of the fiber. Let P denote the stack of pre-smoothable stable plane pairs so defined. Then P is a Deligne-Mumford stack COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 9 locally of finite type over C. (Note: we do not assert that P is proper or that its connected components are of finite type.) Let P d ⊂ P denote the closure of Pd, the stack of smoothable stable plane pairs of degree d. Theorem 7.2. The stack P d is a proper Deligne-Mumford stack of finite type over C. If d is not a multiple of 3 then P d is smooth and is a connected component of P. 7.1. Classification of stable planes. Proposition 7.3. Let (X, D) be a smoothable stable plane pair of degree d. Then −KX is ample and dKX + 3D is linearly equivalent to zero. We sketch the proof of the proposition. Since [(X, D)] lies in the closure of Pd, there is a Q-Gorenstein deformation (X , D)/(0 ∈ T ) of (X, D) over the germ of a curve such that the general fiber is the plane together with a curve of degree d. The restriction of dKX + 3D to a general fiber of the family X /(0 ∈ T ) is linearly equivalent to zero, hence dKX + 3D is linearly equivalent to a sum of components of the special fiber X. But also dKX + 3D = (dKX + 3D)X is nef because KX + (3/d + ǫ)D is ample for 0 < ǫ ≪ 1. It follows that dKX + 3D ∼ 0. Now −KX is ample because KX + (3/d + ǫ)D is ample. If d is not divisible by 3 then, since dKX + 3D ∼ 0, the canonical class KX is divisible by 3 in the class group of X. This condition is very restrictive, for example, it implies that X has at most 2 irreducible components [H04, 7.1]. (On the other hand, already for d = 6 there are surfaces X with 18 components.) Remark 7.4. It is instructive to compare with the one dimensional case. We consider the moduli space M of pairs (X, D) such that X ≃ P1 and D is a divisor of degree d ≥ 3. Let us order the points D = P1 +· · ·+ Pd for simplicity. We construct a compactification M as a moduli space of pairs (X, D) such that (X, ( 2 d + ǫ)D) is slc and KX + (2/d + ǫ)D is ample for 0 < ǫ ≪ 1. The first condition means simply that X is a nodal curve, the points Pi are smooth points of X, and the divisor ( 2 d + ǫ)D has coefficients ≤ 1 . Then dKX + 2D ∼ 0, and X is either a copy of P1, or two copies of P1 meeting transversely in a single point. Moreover, in the second case d is even and there are d/2 points on each component. We deduce that there is a birational morphism M → M to the symmetric GIT quotient = (P1)d// SL(2) which is an isomorphism if d is odd and is a resolution of the M singularities of M corresponding to the strictly semistable points if d is even. GIT GIT GIT We describe the degenerate surfaces X in case d is not divisible by 3. There are two types A and B. Surfaces of type A are normal and have quotient singularities of Wahl type. Surfaces of type B have two irreducible components meeting along a smooth rational curve. We have a complete description of the surfaces of type A: Theorem 7.5. [HP10, 1.2] Let X be a normal surface with quotient singularities which admits a smoothing to the projective plane. Then X is obtained from a weighted projective plane P(a2, b2, c2) by a Q-Gorenstein deformation that smoothes some subset of its singularities, where (a, b, c) is a solution of the Markov equation a2 + b2 + c2 = 3abc. 10 PAUL HACKING The solutions of the Markov equation are easily described: (1, 1, 1) is a solution, and all solutions are obtained from (1, 1, 1) by a sequence of mutations of the form (7.1) (a, b, c) 7→ (a, b, c′ = 3ab − c). We can define a graph G with vertices labelled by solutions of the Markov equation and edges corresponding to pairs of solutions related by a single mutation. Then G is an infinite tree such that every vertex has degree 3. 1 The surface P(a2, b2, c2) has cyclic quotient singularities of types a2 (b2, c2), b2 (c2, a2), 1 c2 (a2, b2). Using the Markov equation one sees that these are Wahl singularities (note that a, b, c are coprime and not divisible by 3 by the inductive description of the solutions of the Markov equation above). Moreover there are no equisingular deformations and no local-to-global obstructions because H 1(TP) = H 2(TP) = 0. So there is one Q-Gorenstein deformation parameter for P associated to each singularity. A surface X obtained as a Q-Gorenstein deformation of P is uniquely determined by the subset of singularities that is smoothed. 1 If X is a surface of type B then X is the union of two normal surfaces X1 and X2 meeting along a smooth rational curve C. The surface X has orbifold normal crossing singularities (xy = 0) ⊂ 1 n (1, −1, a) along the double curve C and Wahl singularities away from C. The Picard numbers of the components of X are given by either ρ(X1) = ρ(X2) = 1 or {ρ(X1), ρ(X2)} = {1, 2}. Example 7.6. Consider the two parameter family X = (X0X2 = sX 2 1 + tY ) ⊂ P(1, 1, 1, 2) × C2 s,t The special fiber X := X0 is a surface of type B obtained by glueing two copies of the weighted projective plane P(1, 1, 2) (the quadric cone) along a ruling. It has an orbifold normal crossing singularity of type (xy = 0) ⊂ 1 2 (1, 1, 1). The fibers Xs,t for s 6= 0, t = 0 are isomorphic to P(1, 1, 4), with the embedding in P(1, 1, 1, 2) being the 2-uple embedding P(1, 1, 4) → (X0X2 = X 2 (U0, U1, V ) 7→ (X0, X1, X2, Y ) = (U 2 The surface P(1, 1, 4) has a Wahl singularity of type 1 are isomorphic to P2. The deformation X /(0 ∈ C2 deformation of X. 1 ) ⊂ P(1, 1, 1, 2) 0 , U0U1, U 2 4 (1, 1). The fibers Xs,t for t 6= 0 s,t) is the versal Q-Gorenstein 1 , V ). We remark that in general it is difficult to describe the versal Q-Gorenstein deformation of P := P(a2, b2, c2) explicitly, because for example the embedding in weighted projective space defined by − 1 3 KP = OP(abc) (corresponding to OP2(1) on the general fiber P2) has high codimension. Example 7.7. Here we describe a two parameter family of surfaces which "con- nects" the weighted projective planes P := P(a2, b2, c2), P′ := P(a2, b2, c′2) associ- ated to two solutions of the Markov equation related by a single mutation (7.1). The family is given by X = (XY = sZ c′ + tT c) ⊂ P(a2, b2, c, c′) × C2 s,t. The special fibre X := X0 is the union of two weighted projective planes P(a2, c, c′), P(b2, c, c′) glued along the coordinate lines of degree a2 and b2. It has two Wahl singularities of indices a and b and two orbifold normal crossing singularities of COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 11 indices c and c′. The fibers Xs,t for s = 0, t 6= 0 are isomorphic to P(a2, b2, c2), with the embedding being the c-uple embedding P(a2, b2, c2) → (XY = T c) ⊂ P(a2, b2, c, c′) (U, V, W ) 7→ (X, Y, Z, T ) = (U c, V c, W, U V ). Similarly, the fibers Xs,t for s 6= 0, t = 0 are isomorphic to P(a2, b2, c′2). The fibers Xs,t for s 6= 0, t 6= 0 are obtained from P or P′ by smoothing the singularity of index c or c′ respectively. Note that Example 7.6 is the special case a = b = c = 1. Example 7.8. Here we describe surfaces X of type B such that ρ(X1) = 2, ρ(X2) = 1. We begin with the trivial family Y = P2 × C1 t . Let m, n be positive integers such that (m, n) = 1. Let π : X → Y be the blowup of the point P = ((1 : 0 : 0), 0) with weights (m, n, 1) with respect to the coordinates x1 = X1/X0, x2 = X2/X0, t. The special fiber X := X0 is reduced and has two components X1 and X2 given by the strict transform of Y := Y0 and the exceptional divisor E = P(m, n, 1) of π respectively. The restriction p : X1 → Y = P2 is the weighted blowup of the point (1 : 0 : 0) in P2 with weights (m, n) with respect to the coordinates x1, x2. The components X1 and X2 are glued along the smooth rational curve C given by the exceptional divisor of p on X1 and the coordinate line of degree 1 on X2 = P(m, n, 1). The surface X has two orbifold normal crossing singularities of indices m and n. A toric calculation shows that the Q-Cartier divisor −KX is ample iff 1 2 m < n < 2m. In this case the surface X occurs in P d for d divisible by 3mn. The canonical divisor −KX is divisible by 3 in the class group of X iff m + n is divisible by 3. In this case X occurs in P d for d divisible by mn. 7.2. Boundary divisors. Let P ′ d ⊂ P d denote the locus of pairs (X, D) such that the surface X is isomorphic to P2, and let ∂P d := P d \ P ′ d. Suppose d is not a multiple of 3. Then ∂P d ⊂ P d is a normal crossing divisor with irreducible components corresponding to surfaces of the following types. (1) Surfaces of type A with a unique Wahl singularity. (2) Surfaces X = X1 ∪ X2 of type B such that {ρ(X1), ρ(X2)} = {1, 2} and X is smooth away from C := X1 ∩ X2. This follows from an analysis of the versal Q-Gorenstein deformations of surfaces of type A and B, together with the following fact: for [(X, D)] ∈ P d, the forgetful map Def QG(X, D) → Def QG(X) from deformations of the pair to deformations of the surface is smooth. See [H04, 3.12, 8.2, 9.1]. (Note: The surfaces X which occur in a given degree d are those for which there exists a divisor D ∈ − (d/3)KX such that the pair (X, (3/d + ǫ)) is slc for 0 < ǫ ≪ 1. A necessary condition is that the index of each singularity P ∈ X is at most d [H04, 4.5].) Remark 7.9. Our stability condition for a pair (X, D) such that X ≃ P2 is a natural strengthening of GIT stability for the plane curve D to a local analytic condition [H04, §10]. In particular, the locus P ′ d is contained in the moduli stack of GIT stable plane curves of degree d. 8. Exceptional vector bundles associated to degenerations of surfaces Let Y be a projective surface. A vector bundle F on Y is exceptional if End F = C and H 1(End F ) = H 2(End F ) = 0. In particular, an exceptional vector bundle F is indecomposable, rigid (no infinitesimal deformations), and unobstructed in 12 PAUL HACKING families. So, if Y/(0 ∈ S) is a deformation of Y over a germ (0 ∈ S), then there exists a unique vector bundle F on Y such that FY = F . Theorem 8.1. [H11] Let X be a projective normal surface with a unique singular- ity P ∈ X of type 1 n2 (1, na − 1). Let X /(0 ∈ T ) be a one parameter Q-Gorenstein deformation of X such that the general fiber is smooth. Let Y denote a general fiber of X /T . (1) Assume that H1(Y, Z) is finite of order coprime to n. Then the special- ization map sp : H2(Y, Z) → H2(X, Z) is injective with cokernel isomorphic to Z/nZ. (2) Assume in addition that H 2,0(Y ) = 0. Then, after a base change T ′ → T of degree a, there exists a reflexive sheaf E on X ′ := X ×T T ′ satisfying the following properties. (a) F := EY is an exceptional vector bundle of rank n on Y . (b) E := EX is a torsion-free sheaf on X such that its reflexive hull E∨∨ is isomorphic to the direct sum of n copies of a reflexive rank 1 sheaf A, and the quotient E∨∨/E is a torsion sheaf supported at P ∈ X. If H is a line bundle on X /T which is ample on fibers, then the vector bundle F is slope stable with respect to H := HY . The Chern classes of F are given by c1(F ) = nc1(A) ∈ H2(Y, Z) ⊂ H2(X, Z), c2(F ) = n − 1 2n (c1(F )2 + n + 1). c1(F ) · KY = ±a mod n, H2(X, Z) = H2(Y, Z) + Z · (c1(F )/n). Moreover and Remark 8.2. Note that the bundles obtained from F by dualizing or tensoring by a line bundle arise in the same way. (Indeed, if L is a line bundle on Y then L extends to a reflexive rank one sheaf L on X . Now E ∨ and (E ⊗ L)∨∨ satisfy the properties 8.1(2a,b) and restrict to F ∨ and F ⊗ L on Y .) We sketch the proof of Theorem 8.1 in the case n = 2. In this case P ∈ X is 4 (1, 1). Let us assume for simplicity that the Q-Gorenstein a singularity of type 1 deformation (P ∈ X )/(0 ∈ T ) is versal, that is, isomorphic to 2 (1, 1, 1)(cid:1) × C1 t . (xy = z2 + t) ⊂ (cid:0)C3 x,y,z/ 1 (In general, we obtain our construction by pullback from the versal case.) The germ (P ∈ X ) is a cyclic quotient singularity of type 1 2 (1, 1, 1). Let π : X → X denote the blowup of P ∈ X . Then the exceptional locus W of π is a copy of the projective plane with normal bundle O(−2). The total space X is smooth and the special fiber X := X0 is a normal crossing divisor with two smooth components given by the strict transform X ′ of X ⊂ X and the exceptional divisor W . The restriction p : X ′ → X of π is the minimal resolution of (P ∈ X). The surfaces X ′ and W meet along the smooth rational curve C which is the exceptional curve of p : X ′ → X and is embedded as a conic in W ≃ P2. COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 13 Let B denote the intersection of X with a small ball with center at the singu- larity P ∈ X in some embedding. Let L denote the boundary of B (the link of the singularity) and write X o := X \ B. Our assumption on H1(Y, Z) implies that the map H 2(X o, Z) → H 2(L, Z) is surjective, by a local analysis near the singularity (see §4) and a Mayer–Vietoris argument. (See [H11], proof of Theorem 1.1, for more details.) It follows that the restriction map (8.1) is surjective. We have H 2(X ′, Z) → H 2(C, Z) H i(OX ′) = H i(OX ) = H i(OY ) because the singularity P ∈ X is a quotient singularity, and by assumption H i(OY ) = 0 for i > 0. Thus H i(OX ′ ) = 0 for i > 0 and the map c1 : Pic X ′ → H 2(X ′, Z) is an isomorphism. So by the surjectivity of (8.1) above, there is a line bundle A′ on X ′ such that the restriction of A′ to C has degree 1. Let G be the vector bundle TP2(−1) on W = P2. Then G is an exceptional vector bundle on W , and GC ≃ OC (1)⊕2, where OC (1) denotes the line bundle of degree 1 on the smooth rational curve C. Since X = X ′ ∪ W is a normal crossing surface with double curve C = X ′ ∩ W , we have the exact sequence of sheaves on X (8.2) Thus we can define a vector bundle E over X by glueing A′⊕2 over X ′ and G over W along OC (1)⊕2 over C. 0 → O X → OX ′ ⊕ OW → OC → 0. We show that E is exceptional. Tensoring the exact sequence (8.2) with End E we obtain the exact sequence 0 → End E → O2×2 X ′ ⊕ End G → O2×2 C → 0. Now H 1(OC ) = 0 because C is a smooth rational curve and H i(OX ′ ) = 0 for i > 0 as noted above, so we obtain H i(End E) = H i(End G). Thus E is exceptional because G is exceptional. Let E denote the (unique) vector bundle over X obtained by deforming the exceptional bundle E. Then the restriction F of E to a general fiber Y of X /T is exceptional by upper semicontinuity of cohomology. Let E := (π∗E)∨∨ be the reflexive hull of the pushforward of E to X . Similarly let A := (p∗A′)∨∨ be the reflexive hull of the pushforward of A′ to X. Then E := EX is torsion-free because E is reflexive, and by construction EX\{P } = A⊕2X\{P }, so E∨∨ = A⊕2 and E∨∨/E is supported at P . 9. Boundary divisors of the moduli space of stable surfaces Here we formulate a precise conjecture relating the components of the boundary of the moduli space of stable surfaces and exceptional vector bundles. Let Y be a smooth surface such that KY is ample, H 2(TY ) = 0, H 2,0(Y ) = 0, Y ≤ 5 and Y is homeomorphic and π1(Y ) = 0. These conditions imply that 1 ≤ K 2 14 PAUL HACKING (but not diffeomorphic) to the blowup of P2 in 9 − K 2 surfaces Y are known for K 2 Y = 1, 2, 3, 4 [B85], [LP07], [PPS09a], [PPS09b]. Y points. Examples of such Let M denote the irreducible component of the moduli stack of stable surfaces containing [Y ]. Let M ⊂ M denote the locus of surfaces with Du Val singularities and ∂M := M \ M. Note that M is smooth of the expected dimension 10 − 2K 2 Y near [Y ] by our assumption H 2(TY ) = 0. If M is not normal, we replace it by its normalization. Let Γ ⊂ Aut(H 2(Y )) denote the monodromy group. Then Γ preserves KY . Y is negative definite because H 2,0 = 0 and KY is ample. Thus Γ is The lattice K ⊥ a finite group. For F a vector bundle on Y we define its slope vector v(F ) := c1(F )/ rk(F ) ∈ H 2(Y, Q). Then v(F ⊗ L) = v(F ) + c1(L) and v(F ∨) = −v(F ). Note that c1 : Pic(Y ) → H 2(Y, Z) is an isomorphism because H 1(OY ) = H 2(OY ) = 0 by assumption. Consider exceptional vector bundles F of rank greater than 1 on Y that are slope stable with respect to KY and of degree c1(F ) · KY coprime to rk(F ). Let S denote the set of associated slope vectors v(F ) = c1(F )/ rk(F ) ∈ H 2(Y, Q) modulo translation by H 2(Y, Z), multiplication by ±1, and the action of the monodromy group Γ. Let T denote the set of codimension one components D of the boundary ∂M such that the general fiber over D is a normal surface X with a unique singularity P ∈ X of Wahl type. Theorem 8.1 defines a map of sets Φ : T → S, D 7→ v(F ). We hope that the map Φ is "close" to a bijective correspondence. In particular, we formulate the following conjecture. Conjecture 9.1. The set S is finite. Equivalently, the ranks of exceptional bundles F on Y which are slope stable with respect to KY (and of degree coprime to the rank) are bounded. One can show that the analogue of the map Φ for Y = P2 is bijective using the classification of degenerations of P2 given by Theorem 7.5 and the classification of exceptional vector bundles on P2 [DLP85],[R89]. More precisely: Theorem 9.2. [H11, §6] There is a bijective correspondence between isomorphism types of degenerations X of P2 with a unique Wahl singularity and isomorphism types of exceptional bundles F of rank greater than 1 on P2 modulo dualizing and tensoring by line bundles, given by the construction of Theorem 8.1. 10. Relation with Donaldson theory We note that the proposal of §9 is closely related to the Donaldson theory of invariants of smooth 4-manifolds [DK90], [K05], [M09]. Donaldson showed that the classification of simply connected smooth 4-manifolds up to diffeomorphism is much richer than the classification up to homeomorphism obtained by Freedman [F82]. This was achieved by defining new invariants of smooth 4-manifolds as follows. Let X be a simply connected compact oriented smooth 4-manifold. Let b+ 2 denote the dimension of a maximal positive definite subspace of H 2(X, R); we assume b+ (If X is a Kahler manifold, then 2 > 0. COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 15 b+ 2 = 2 dim H 2,0 + 1 by Hodge theory.) Fix a Riemannian metric g on X and consider a smooth complex vector bundle E of rank n on X together with a Her- mitian metric on E. (Note that the isomorphism type of E is determined by its Chern classes because X has real dimension 4 [W52].) One considers the moduli space M of instantons: anti-self-dual connections on the principal PU(n)-bundle associated to E (where PU(n) = U(n)/ U(1) denotes the projective unitary group). One defines numerical invariants of (X, g) as intersection numbers of certain natural cohomology classes on M . If b+ 2 > 1 then the Donaldson invariants are independent of the choice of the metric g, that is, they are invariants of the smooth manifold X. If b+ 2 = 1 then they depend on the metric g through the associated line of self-dual harmonic 2-forms in H 2(X, R) via a chamber structure and wall crossing formulas. If now X is a complex surface and g is a Kahler metric, then the moduli space of irreducible anti-self-dual connections is identified with the moduli space of holomorphic structures on E which are slope stable with respect to the class of the Kahler form. Thus, in the case of an algebraic surface, Donaldson invariants can be computed by intersection theory on the moduli space of stable vector bundles. Moreover, in case b+ 2 = 1, the line of self dual harmonic 2-forms is spanned by the Kahler form, and the wall crossing formulas are determined by the variation of the moduli of stable bundles with the polarization. 1 ∪ X o 2 where X o 1 and X o 1 and X o One of the first applications of Donaldson theory was the analysis of connected sum decompositions of smooth 4-manifolds. Let X = X1#X2 be a connected sum decomposition of a 4-manifold X. Thus X = X o i is the complement of a ball about a point in Xi, and X is obtained by identifying the boundaries of X o 1 and X o 2 . We choose a Riemannian metric g on X such that the "neck" connecting 2 is isometric to S3 × (−R, R) for R ≫ 0. In this situation one proves X o that instantons on X are obtained by gluing instantons on X o 2 which decay exponentially to a flat connection at the boundary. In this way one obtains gluing formulas for the Donaldson invariants of X in terms of relative Donaldson invariants of X o 1 and X o 2 . The situation we consider above is similar. Recall that we begin with a smooth surface Y and consider a degeneration to a normal surface X with a unique Wahl sin- gularity P ∈ X. We have a generalized connected sum decomposition Y = X o#LM of the 4-manifold Y . That is, writing X o for the complement of a small ball about the singular point P ∈ X, L for the link of the singularity, and M for the Milnor fiber of the smoothing, the 4-manifold Y is obtained by identifying the boundaries of X o and M , which are copies of the lens space L. Such generalized connected sum decompositions may be studied using Donaldson theory in the same way as above [D02]. In particular, the rank 2 Donaldson invariants for the decomposition corresponding to a Wahl singularity of type 1 n2 (1, n − 1) were analyzed in [FS97]. (We note that the surgery of smooth 4-manifolds given by passing from the minimal resolution of a Wahl singularity to its smoothing is called a rational blowdown in differential and symplectic geometry.) We expect that our construction admits an interpretation in these terms. More precisely, we expect that for an appropriate choice of metric g, the exceptional bundle we construct admits a unique anti-self-dual connection which is projectively trivial over X o (and nontrivial over M ). In the Donaldson theory it is common to restrict attention to bundles of rank 2. In fact it is conjectured that the higher rank invariants are determined by the 16 PAUL HACKING rank 2 invariants [MM98]. However, as we have seen, for our application to the study of the boundary of the moduli space of complex surfaces, it is crucial to consider vector bundles of arbitrary rank n, because they are directly connected to the geometry of the degeneration. On the other hand, we restrict our attention to rigid vector bundles, that is, we only consider the case in which the moduli space of stable vector bundles is zero dimensional. 11. Other examples of boundary divisors We describe some types of codimension 1 boundary components of the moduli space of stable surfaces different from the Wahl type described in §4. Let X be a stable surface. If X satisfies the conditions set out in one of the examples below, then equisingular deformations of X determine a codimension 1 component of the boundary of the moduli space of stable surfaces. Example 11.1. The surface X has a unique singularity P ∈ X of one of the following types, and there are no local-to-global obstructions. (1) The minimal resolution π : X → X has exceptional locus a union of 4 smooth rational curves E1, E2, E3, and F . The Ei are disjoint, each meets the curve F transversely in a single smooth point, and (−E2 3; −F 2) = (3, 3, 3; 4), (2, 4, 4; 3), or (2, 3, 6; 2). 1 , −E2 2 , −E2 See [W10, §8]. (Note: The index one cover p : (Q ∈ Z) → (P ∈ X) is the quotient of the cone over an elliptic curve of degree 9, 8, or 6 by a cyclic group of order 3, 4, or 6, respectively [K88, 9.6(3)].) (2) A cone over an elliptic curve of degree 9 [S98, 4.5]. (3) A smoothable cusp singularity such that, writing E = E1 + · · · + Er for the exceptional locus of the minimal resolution, −E2 = 9 + r [W81, 5.6]. (Note: The smoothability criterion for cusp singularities conjectured by Looijenga [L81, III.2.11] is proven in [GHK11]. The smoothable cusps with r ≤ 3 are listed in [FM83, §4].) Example 11.2. The surface X is the union of two components X1 and X2 meeting transversely along a smooth curve C, X has orbifold normal crossing singularities (xy = 0) ⊂ 1 QG,X is the line bundle on C given by (5.1). We require that dim H 0(T 1 QG,X ) = 0, and H 2(TX ) = 0. For example if C has genus zero we require T 1 QG,X ≃ OC . (See [T09] for more results on Q-Gorenstein deformations of non-normal surfaces.) n (1, −1, a) along C and is smooth elsewhere. Then T 1 QG,X ) = 1, H 1(T 1 References AH09. D. Abramovich, B. Hassett, Stable varieties with a twist, preprint arXiv:0904.2797v1 [math.AG], 2009. A96. V. Alexeev, Moduli spaces Mg,n(W ) for surfaces, in Higher-dimensional complex varieties (Trento, 1994), 1-22, de Gruyter, Berlin, 1996. A08. V. Alexeev, Limits of stable pairs, Pure Appl. Math. Q. 4 (2008), no. 3, 767-783. AM04. V. Alexeev, S. Mori, Bounding singular surfaces of general type, in Algebra, arithmetic and geometry with applications, 143-174, Springer, 2004. A76. M. Artin, Lectures on deformations of singularities, Lectures on Mathematics and Physics 54, Tata Inst. Fund. Res., 1976. B85. R. Barlow, A simply connected surface of general type with pg = 0, Invent. Math. 79 (1985), no. 2, 293-301. COMPACT MODULI SPACES OF SURFACES OF GENERAL TYPE 17 C86. F. Catanese, Connected components of moduli spaces, J. Differential Geom. 24 (1986), no. 3, 395-399. C00. B. Conrad, Grothendieck duality and base change, Lecture Notes in Math. 1750, Springer, 2000. DM69. P. Deligne, D. Mumford, The irreducibility of the space of curves of given genus, Inst. Hautes ´Etudes Sci. Publ. Math. No. 36 (1969), 75-109. D02. S. Donaldson, Floer homology groups in Yang-Mills theory, Cambridge Tracts in Math. 147, C.U.P., 2002. DK90. S. Donaldson, P. Kronheimer, The geometry of four-manifolds, O.U.P., 1990. DLP85. J.-M. Drezet, J. Le Potier, Fibr´es stables et fibr´es exceptionnels sur P2, Ann. Sci. ´Ecole Norm. Sup. (4) 18 (1985), no. 2, 193-243. FS97. R. Fintushel, R. Stern, Rational blowdowns of smooth 4-manifolds, J. Differential Geom. 46 (1997), no. 2, 181-235. F82. M. Freedman, The topology of four-dimensional manifolds, J. Differential Geom. 17 (1982), no. 3, 357-453. F83. R. Friedman, Global smoothings of varieties with normal crossings. Ann. of Math. (2) 118 (1983), no. 1, 75-114. FM83. R. Friedman, R. Miranda, Smoothing cusp singularities of small length, Math. Ann. 263 (1983), no. 2, 185-212. GHK11. M. Gross, P. Hacking, and S. Keel, Mirror symmetry for log Calabi–Yau surfaces I, preprint arXiv:1106.4977v1 [math.AG], 2011. H04. P. Hacking, Compact moduli of plane curves, Duke Math. J. 124 (2004), 213–257. H11. P. Hacking, Exceptional bundles associated to degenerations of surfaces, preprint arXiv:1107.2644v1 [math.AG], 2011. HP10. P. Hacking and Y. Prokhorov, Smoothable del Pezzo surfaces with quotient singularities, Compos. Math. 146 (2010), no. 1, 169-192. H10. R. Hartshorne, Deformation theory, Grad. Texts in Math. 257, Springer, 2010. H03. B. Hassett, Moduli spaces of weighted pointed stable curves, Adv. Math. 173 (2003), no. 2, 316-352. I71. L. Illusie, Complexe cotangent et d´eformations, I, Lecture Notes in Math., Vol. 239. Springer, 1971. I72. L. Illusie, Complexe cotangent et d´eformations, II, Lecture Notes in Math., Vol. 283. Springer, 1972. K88. Y. Kawamata, Crepant blowing-up of 3-dimensional canonical singularities and its applica- tion to degenerations of surfaces, Ann. of Math. (2) 127 (1988), no. 1, 93-163. K90. J. Koll´ar, Projectivity of complete moduli, J. Differential Geom. 32 (1990), no. 1, 235-268. K91. J. Koll´ar, Flips, flops, minimal models, etc., in Surveys in differential geometry (Cambridge, MA, 1990), 113-199, Lehigh Univ., 1991. K10. J. Koll´ar, Moduli of varieties of general type, preprint arXiv:1008.0621v1 [math.AG] . K11. J. Koll´ar, Moduli of varieites of general type, book in preparation, preliminary version available at www.math.princeton.edu/∼kollar/ . KM98. J. Koll´ar, S. Mori, Birational geometry of algebraic varieties, C.U.P., 1998. KSB88. J. Koll´ar, N. Shepherd-Barron, Threefolds and deformations of surface singularities, Inv. Math. 91 (1988), 299–338. K05. P. Kronheimer, Four-manifold invariants from higher-rank bundles, J. Differential Geom. 70 (2005), no. 1, 59-112. LP07. Y. Lee, J. Park, A simply connected surface of general type with pg = 0 and K 2 = 2, Invent. Math. 170 (2007), no. 3, 483-505. LS67. S. Lichtenbaum, M. Schlessinger, The cotangent complex of a morphism, Trans. Amer. Math. Soc. 128 (1967), 41-70. L81. E. Looijenga, Rational surfaces with an anticanonical cycle, Ann. of Math. (2) 114 (1981), no. 2, 267-322. MM98. M. Marino, G. Moore, The Donaldson–Witten function for gauge groups of rank larger than one, Comm. Math. Phys. 199 (1998), no. 1, 25-69. M09. T. Mochizuki, Donaldson type invariants for algebraic surfaces, Lecture Notes in Math. 1972, Springer, 2009. PPS09a. H. Park, J. Park, D. Shin, A simply connected surface of general type with pg = 0 and K 2 = 3, Geom. Topol. 13 (2009), no. 2, 743-767. 18 PAUL HACKING PPS09b. H. Park, J. Park, D. Shin, A simply connected surface of general type with pg = 0 and K 2 = 4, Geom. Topol. 13 (2009), no. 3, 1483-1494. R89. A. Rudakov, Markov numbers and exceptional bundles on P2, Math. USSR-Izv. 32 (1989), no. 1, 99–112. S98. J. Stevens, Degenerations of elliptic curves and equations for cusp singularities, Math. Ann. 311 (1998), no. 2, 199-222. T09. N. Tziolas, Q-Gorenstein deformations of non-normal surfaces, Amer. J. Math. 131 (2009), no. 1, 171-193. V06. R. Vakil, Murphy's law in algebraic geometry: badly-behaved deformation spaces, Invent. Math. 164 (2006), no. 3, 569-590. W81. J. Wahl, Smoothings of normal surface singularities, Topology 20 (1981), no. 3, 219–246. W10. J. Wahl, Construction of QHD smoothings of valency 4 surface singularities, preprint arXiv:1005.2199v2 [math.AG] (2010). W52. W-T. Wu, Sur les classes caract´eristiques des structures fibr´ees sph´eriques, Publ. Inst. Math. Univ. Strasbourg 11 (1952), 5-89. Department of Mathematics and Statistics, Lederle Graduate Research Tower, Box 34515, University of Massachusetts, Amherst, MA 01003-9305, USA E-mail address: [email protected]
1909.06714	1	1909	2019-09-15T02:19:37	An algorithm for a Massey triple product of a smooth projective plane curve	[ "math.AG", "math.NT" ]	We provide an explicit algorithm to compute a Massey triple product relative to a defining system for a smooth projective plane curve $X$ defined by a homogeneous polynomial $G(\underline x)$ over a field. The main idea is to use the description (due to Carlson and Griffiths) of the cup product for $H^1(X,\mathbb{C})$ in terms of the multiplications inside the Jacobian ring of $G(\underline x)$ and the Cech-deRham complex of $X$. Our algorithm gives a criterion whether a Massey triple product vanishes or not in $H^2(X)$ under a particular non-trivial defining system of the Massey triple product and thus can be viewed as a generalization of the vanishing criterion of the cup product in $H^2(X)$ of Carlson and Griffiths. Based on our algorithm, we provide explicit numerical examples by running the computer program.	math.AG	math	AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE YOUNGGI LEE, JEEHOON PARK, JUNYEONG PARK, AND JAEHYUN YIM Abstract. We provide an explicit algorithm to compute a Massey triple product relative to a defining system for a smooth projective plane curve X defined by a homogeneous polynomial G(x) over a field. The main idea is to use the description (due to Carlson and Griffiths) of the cup product for H 1(X, C) in terms of the multiplications inside the Jacobian ring of G(x) and the Cech-deRham complex of X. Our algorithm gives a criterion whether a Massey triple product vanishes or not in H 2(X) under a particular non-trivial defining system of the Massey triple product and thus can be viewed as a generalization of the vanishing criterion of the cup product in H 2(X) of Carlson and Griffiths. Based on our algorithm, we provide explicit numerical examples by running the computer program. Contents Introduction 1. 1.1. Acknowledgement 2. Main computations 2.1. Griffiths' description for H 1(X, C) 2.2. Main computations 3. An explicit algorithm for the Massey triple product 3.1. Main theorem 3.2. Other possible Massey triple products 4. Computer algorithm and experiment References 1 3 4 4 4 8 8 10 10 12 1. Introduction The Massey product is an important invariant in topology and appears in several areas of mathematics as a higher order structure which captures more refined informations than the usual cup product of the cohomology. Let M be a topological manifold. Assume that there are three classes xi ∈ H mi(M, C), i = 0, 1, 2 such that x0 ∪ x1 = 0 = x1 ∪ x2. Choose cochain representatives Xi for the xi. Since X0 ∪ X1 is a coboundary, choose X01 so that d(X01) = X0 ∪ X1. Similarly choose X12 so that d(X12) = X1 ∪ X2. Form the cochain X01 ∪ X2 − (−1)m0X0 ∪ X12 which becomes a cocycle. The choices of X01 and of X12 are not unique. Precisely either can be altered by a cocycle. Define the ideal x0,x2 = x0 ∪ H •−m0(M, C) + H •−m2(M, C) ∪ x2 ⊆ H •(M, C) J • The cocycle X01 ∪ X2 − (−1)m0X0 ∪ X12 can be altered by (a cochain representative of) any element in J m0+m1+m2−1 x0,x2 . For any x1 ∈ H m2(M, C) such that x0 ∪ x1 = 0 = x1 ∪ x2, the Massey triple product is defined to be the class of X01 ∪ X2 − (−1)m0X0 ∪ X12 in this quotient group. hx0; x1; x2i ∈ H m0+m1+m2−1(M, C)/J m0+m1+m2−1 x0,x2 2000 Mathematics Subject Classification. 18G55(primary), 14J70, 14D15, 13D10 (secondary). Key words and phrases. Massey triple product, Cech-deRham complex, smooth projective curve. 1 AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 2 When M is a compact Riemann surface (the main object of interest in the current paper) and m0 = x0,x2 = 0 for nonzero x0, x2 ∈ H 1(M, C) by the non-degeneracy of the m1 = m2 = 1, then H 2(M, C)/J 2 cup product pairing for H 1(M, C) and the Massey triple product vanishes. In general, when M is a compact Kähler manifold, then the Massey triple product is known to vanish by the result of [3]. But in this article we will be interested in computing the lifted Massey triple product which is defined to be the cohomology class (depending on X01 and X12) [x0, x1, x2] := [X01 ∪ X2 − (−1)m0 X0 ∪ X12] ∈ H m0+m1+m2−1(M, C), for a particular explicit choice of X01 and X12. In other words, we will provide an algorithm for the value of the Massey triple product relative to a certain defining system (see Definition 1.1 below). Definition 1.1. The Massey product [f0, ..., fn] is said to be defined if there is an array A of Cech-de Rham 1-cochains A = {Fi1···ik ∈ C1 0 ⊕ C0 1 : ip+1 = ip + 1 (∀p)}0≤k≤n−1 such that Fi is a 1-cocycle representing fi for 0 ≤ i ≤ n, and D(Fi1 ···ik ) = Fi1···ik−1 • Fik + Fi1 · · · ik−2 • Fik−1ik + · · · + Fi1 • Fi2···ik , for 0 ≤ k ≤ n − 1, ip+1 = ip + 1 (∀p). An array A is called a defining system for [f0, · · · , fn]. The value of the Massey (n + 1)−th product relative to a defining system A, denoted by [f0, · · · , fn]A, is then defined to be the cohomology class of H2(XG) represented by the 2-cocycle c(A) := F0···n−1 • Fn + F0···n−2 • Fn−1n + · · · + F0 • F1···n. We will provide an explicit defining system A for the Massey triple product of a projective smooth plane curve (based on Carlson-Griffiths' Jacobian vanishing criterion of the cup product and computations in Cech-deRham complex with respect to a Jacobian covering) and an algorithmic formula (see the main theorem 3.1) for the C-valued Massey triple product relative to A. Now we explain a more precise set up of this paper. Let XG be a smooth projective curve in P2 defined by a homogeneous polynomial G(x) ∈ k[x] = k[x0, x1, x2] over a field k. Let P2 k be a projective 2-space over k. Assume that k is the complex field C for the simplicity of presentation.1 Let X = XG(C) (respectively, P2 = P2 C(C)) and be the complex analytic space (the compact Riemann surface) associated to XG (respectively, P2 C). For each i = 0, 1, 2, set Gi := ∂G(x) ∂xi , Ui = {x ∈ P2 : Gi(x) 6= 0} = P2 \ XGi(C). Then the Griffiths theorem, [4], asserts a canonical isomorphism between the middle-dimensional singular cohomology H 1(X, C) and the degree deg G−3 and 2 deg G−3 part of the quotient C-vector space C[x]/JG where JG is the Jacobian ideal of C[x] generated by G0, G1, G2. If we let U0, U1 ∈ C[x] correspond to singular cocycles W0, W1 (whose cohomology class [Wi] ∈ H 1(X, C)) respectively under the Griffiths isomorphism, then the theorem 2, [2], says that [W0 ∪ W1] vanishes in H 2(X, C) if and only if U0U1 belongs to the Jacobian ideal JG = hG0(x), G1(x), G2(x)i. Our main result will be in the following shape: if U0, U1, U2 ∈ C[x] correspond to W0, W1, W2 such that U0U1, U1U2 ∈ JG (i.e. U0, U1, U2 satisfies the condition to define a Massey triple product), then the Massey triple product [W0, W1, W2] under a particular non-trivial defining system (i.e. a choice of W01 and W12: see Proposition 2.2) vanishes in H 2(X, C) if and only if a certain explicit quantity (see the quantity A − B in Theorem 3.1) belongs to the ideal generated by G0(x)2, G1(x)2, G2(x)2. 1Since the theory is algebraic, our algorithm has a chance to be applied to more general fields, e.g. the p-adic Dwork cohomology defined over a p-adic field. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 3 Remark 1.2. The similarity of vanishing criterions of the cup product and the triple Massey product suggests us to conjecture that the n-Massey product vanishes (under a particular non-trivial defining sys- tem) if and only if some explicit quantity belongs to the ideal generated by G0(x)n−1, G1(x)n−1, G2(x)n−1. But due to technical difficulties arising in the computation of higher (n ≥ 4) Massey products, we were not able to provide such general vanishing criterion. We consider the Cech-deRham double complex of X for the open cover U = {Ui}i=0,1,2: q := Cq(UX , Ωp Cp X ), where Ωp X is the sheaf of holomorphic differential p-forms on X. We will compute the Massey triple product for the first degree cohomology of the associated single complex (so called, the hypercohomology) with respect to the total differential D = d + (−1)pδ, where d is the exterior differential and δ is the Cech differential. Remark that the smoothness of the XG is essential, since otherwise U do not cover P2. We shall refer to U as the Jacobian cover of P2 relative to G(x). Let H•(XG) = Lp+q=• H p(UX , Ωq X ) be the hypercohomology of the Cech-deRham complex of XG with respect to the Jacobian cover U. Note that the hypercohomology H•(XG) of this associated single complex is isomorphic to the singular cohomology H •(X, C). Let us briefly recall the natural twisted product structure on the Cech-deRham complex. Let ∧ denote the natural product given by the "front q-face, back s-face" rule, followed by exterior multiplication of forms. Then the twisted product Cp q × Cr s → Cp+r q+s. is skew-commutative and satisfies the Leibnitz rule: αp q • αr s := (−1)qrαp q ∧ αr s αr s • αp D(αp q • αr q = (−1)(p+q)(s+r)αp s) = D(αp q ) • αr q • αr s, s + (−1)p+qαp q • D(αr s). Because the twisted product reduces to exterior multiplication of forms, it represents the topological cup-product on the level of hypercohomology. Therefore, the Massey triple product for the hypercoho- mology H1(X) gives the Massey triple product for H 1(X, C). Let w0, w1, w2 ∈ H1(XG) such that w0 ∪ w1 = w1 ∪ w2 = 0 in H2(XG). More precisely, we consider W0, W2 ∈ C0 1 and W1 ∈ C1 0 with the associated cohomology class [Wi] = wi such that D(W01) = W0 • W1, D(W12) = W1 • W2 in C1 1 for some W01, W12 ∈ C1 the Massey triple product: 0. Note that [W0 • W1] = w0 ∪ w1 and [W1 • W2] = w1 ∪ w2. Then we can form [w0, w1, w2] := [W0 • W12 + W01 • W2] ∈ H 1(UX , Ω1 X ). Since there is an isomorphism θ : H 1(UX , Ω1 X ) ≃−→ C, we can define a C-valued Massey triple product: hw0, w1, w2i := θ([w0, w1, w2]) ∈ C. We will find explicit formulas for a non-trivial defining system A = {W0, W1, W2, W01, W12} and the corresponding C-valued Massey triple product, which can be implemented in computer algebra system. We will also show (see Corollary 3.3) that vanishing of the Massey triple product does not depend on choices of cohomology representatives of [W0], [W1], [W2] under our particular non-trivial defining system. 1.1. Acknowledgement. Jeehoon Park would like to thank Minhyong Kim for a useful related discus- sion and this project was motivated from the conversation with him. The work of Jeehoon Park was partially supported by BRL (Basic Research Lab) through the NRF (National Research Foundation) of South Korea (NRF-2018R1A4A1023590). AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 4 2.1. Griffiths' description for H 1(X, C). We introduce a new variable y and consider the polynomial ring k[x0, x1, x2, y]. Let us define the charge (additive) grading on k[x0, x1, x2, y]: 2. Main computations Then k[x0, x1, x2, y] has charge decomposition: ch(xi) = 1, ch(y) = − deg G. k[x0, x1, x2, y] =Mch k[x0, x1, x2, y]ch. Let S(y, x) = yG(x) and cX := deg G − 3. Let JS be the Jacobian ideal of k[x0, x1, x2, y] generated by ∂S(x) ∂y Then there is a canonical isomorphism due to Griffiths, [4], (for example, , i = 0, 1, 2, and ∂S(x) ∂xi see the Proposition 4.10 and section 4 of [5] for a detailed explanation) k[x0, x1, x2, y]cX /JS ∩ k[x0, x1, x2, y]cX ≃ H 2(P2 \ X) ResX−−−→ H 1(X, C), where the first map is given by yk−1u(x) 7→ (−1)k−1(k − 1)! u(x) G(x)k · (x0dx1 ∧ dx2 − x1dx0 ∧ dx2 + x2dx0 ∧ dx1), k ≥ 1 and ResX is the residue map (e.g. see p 298, [5] for its precise description). Let JG be the Jacobian ideal of k[x] = k[x0, x1, x2] generated by ∂G(x) , i = 0, 1, 2. Then the inclusion k[x] ֒→ k[x, y] induces an ∂xi isomorphism (2.1) Mm≥0 k[x]m deg G+cX /JG ∩ k[x]m deg G+cX ≃ k[x, y]cX /JS ∩ k[x, y]cX , where k[x]α is the submodule of k[x] consisting of homogeneous polynomials of degree α. Note that k[x]m deg G+cX = k[x](m+1) deg G−3. Let us consider the Hodge decomposition H 1(X, C) ≃ H 0(X, Ω1 X ) ⊕ H 1(X, Ω0 X ). Then it is known (the pole order filtration with respect to G(x) is sent to the Hodge filtration under ResX due to Griffiths, Theorem 8.3, [4]) that k[x]deg G−3/JG ∩ k[x]deg G−3 ≃ H 0(X, Ω1 k[x]m deg G−3/JG ∩ k[x]m deg G−3 ≃ H 0(X, Ω1 X ), X ) ⊕ H 1(X, Ω0 X ), M1≤m≤2 k[x]m deg G−3/JG ∩ k[x]m deg G−3 = 0, if m ≥ 3 (Macaulay's theorem), the multiplication map followed by the projection to degree 3 deg G − 6-part: k[x](m+1) deg G−3 to H 1(X, C). Consider under the isomorphism fromLm≥0 M1≤m≤2 k[x]m deg G−3/JG ∩ k[x]m deg G−3 × M1≤m≤2 k[x](m+1) deg G−3/JG∩Lm≥0 k[x]m deg G−3/JG ∩ k[x]m deg G−3 → k[x]3 deg G−6/JG ∩ k[x]3 deg G−6. In [2], it is shown that this multiplication matches with the cup product H 1(X, C) × H 1(X, C) → H 2(X, C) ≃ C under the above isomorphism and the isomorphism (the corollary (iv), p 19, [2]) k[x]3 deg G−6/JG ∩ k[x]3 deg G−6 ≃ C. 2.2. Main computations. Fix a nonzero section of Ω2 P2 (3) ≃ OP2 , say, Ω := x0dx1 ∧ dx2 − x1dx0 ∧ dx2 + x2dx0 ∧ dx1. Given a vector field Z on k3, let i(Z) denote the operation of contraction, and let Kj = i( ∂ ∂xj multi-index J = (j0, ..., jq) of size q, let ). Given a KJ = Kjq · · · Kj0 , ΩJ = KJ Ω, GJ = Gj0 · · · Gjq . AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 5 Let w0 = [W0], w2 = [W2] correspond to anti-holomorphic cohomology classes and w1 = [W1] correspond to a holomorphic cohomology class.2 Then we can express the Cech-deRham cocycle Wi with respect to the open covering U as (2.2) W0 = { W1 = { W2 = { U0 G01 U1 G0 U2 G01 Ω01, U0 G02 Ω02, U0 G12 Ω01} ∈ C0 1, Ω0, U1 G1 Ω1, U1 G2 Ω2} ∈ C1 0, Ω01, U2 G02 Ω02, U2 G12 Ω01} ∈ C0 1, for some polynomials Ui ∈ k[x] with deg U0 = deg U2 = 2 deg G − 3 and deg U1 = deg G − 3. Then W0 • W1 = (−1)1{ U0U1 G0G2 1 Ω01Ω1, U0U1 G0G2 2 Ω02Ω2, U0U1 G1G2 2 Ω01Ω2} ∈ C1 1. W1 • W2 = { U1U2 G0G2 1 Ω01Ω1, U1U2 G0G2 2 Ω02Ω2, U1U2 G1G2 2 Ω01Ω2} ∈ C1 1. The following theorem is one of the main technical results in [2]. Theorem 2.1. The cohomology class wi ∪ wj = [Wi • Wj] of the cocycle Wi • Wj is trivial if and only if UiUj belongs to the Jacobian ideal JG of G(x). By the above theorem, if U0U1 =P2 ∈ k[x], then W0•W1 and W1 • W2 are coboundaries. Our first observation is that there is a simple cochain level formula for W01 such that D(W01) = W1 • W2: A simple Cech type computation confirms the following proposition. i Gi and U1U2 =P2 i Gi for R(01) i=0 R(01) i=0 R(12) , R(12) i i Proposition 2.2. If we define W01 := { x1R(01) 2 − x2R(01) 1 G2 0 Ω0, −x0R(01) 2 + x2R(01) 0 G2 1 Ω1, x0R(01) 1 − x1R(01) 0 G2 2 Ω2} ∈ C1 0, then D(W01) = W1 • W2 in C1 1. Similarly, if we define W12 = {− x1R(12) 2 − x2R(12) 1 G2 0 Ω0, − −x0R(12) 2 + x2R(12) 0 G2 1 Ω1, − x0R(12) 1 − x1R(12) 0 G2 2 Ω2} ∈ C1 0, then D(W12) = W2 • W3 in C1 1. Proof. Because Ω2 d(W01) = 0 ∈ C2 identity: X is the sheaf of holomorphic differential 2-forms on X and dimC X = 1, C2 0 = 0. Thus 0 and it is enough to check that δ(W01) = W0 • W1. Then this reduces to proving the following (2.3) − U0U1 G0G2 1 Ω01Ω1 = x1R(01) 2 − x2R(01) 1 G2 0 Ω0 − −x0R(01) 2 + x2R(01) 0 G2 1 Ω1, since the desired identities for other indices will follow from the same computation by the symmetry of indices. Note that Ω0 = x2dx1 − x1dx2, Ω1 = x0dx2 − x2dx0, Ω01 = x2. We will use the following identities to prove (2.3), which follows from the defining equation G(x) = 0 of XG: x0G0 + x1G1 + x2G2 = deg(G) · G(x) = 0, G0dx0 + G1dx1 + G2dx2 = 0. x1R(01) 2 − x2R(01) 1 Ω0 − G2 0 2 + x2R(01) 0 )G2 −(−x0R(01) −x0R(01) 2 + x2R(01) 0 G2 1 Ω1 RHS = = = −(−x0R(01) 2 + x2R(01) 0 )G2 2 − x2R(01) 1 )G2 1(x2dx1 − x1dx2) 2 − x2R(01) 0Ω1 + (x1R(01) G2 0(x0dx2 − x2dx0) + (x1R(01) 0G2 1 )G2 1 1Ω0 G2 0G2 1 2There are 23 = 8 choices for {w0, w1, w2} depending on wi holomorphic or anti-holomorphic. See Table 3.1 for an explanation why we choose this way. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 6 LHS = − = − = − U1U2G0Ω01Ω1 0G2 1 G2 (R(01) 0 G0 + R(01) (R(01) 0 G0 + R(01) = − (R(01) 0 G0 + R(01) 1 G1 + R(01) 2 G2)G0Ω01Ω1 G2 0G2 1 1 G1 + R(01) 2 G2)G0x2(x0dx2 − x2dx0) G2 0G2 1 2 G2)x2(x0G0)dx2 − (R(01) 1 G1 + R(01) G2 0G2 1 0 G0 + R(01) 1 G1 + R(01) 2 G2)x2 2(G0dx0) (G2 0G2 = R(01) = R(01) 1)(LHS − RHS) (x1x2G1G2 + x2 (x1x2G1G2 + x2 1G2 1G2 2 2 1 − x0G0x0G0)dx2 − R(01) 1 − x0G0(−x1G1 − x2G2))dx2 (x2 2 2G1G2 + x1x2G2 1)dx1 + R(01) 2 x0x2G0G0dx0 2 = R(01) −R(01) (x2 2G1G2 + x1x2G2 1)dx1 + R(01) 2 x0x2G0(−G1dx1 − G2dx2) 2 x1G1(x2G2 + x1G1 + x0G0)dx2 − R(01) 2 x2G1(x2G2 + x1G1 + x0G0)dx1 = 0. The k-valued Massey triple product hw0, w1, w2i of w0, w1, and w2 is, by definition, the image of the cohomology (cid:3) class [W01 • W2 + W0 • W12] ∈ H 2(XG) ≃ H 2(⊕p+q=•Cp Let η : XG → P2 be a given closed embedding. q , D) under an isomorphism θ : H 2(XG) In order to compute hw0, w1, w2i, we use the connecting ≃→ k. coboundary map induced from the Poincare residue exact sequence of sheaves on P2: (2.4) 0 → Ω• P2 → Ω• P2 (log XG) Res→ η∗Ω•−1 XG → 0, P2 is a sheaf of regular p-forms on P2 and Ωp where Ωp P2 (log XG) is a sheaf of rational p-forms ω on P2 such that ω and dω are regular on P2 \ X and have at most a pole of order one along X. See the page 444 in [1] for more details. In [2], the authors gave an explicit formula of the coboundary map δ in the Poincaré residue sequence induced from 2.4 δ : H 1(UXG , Ω1 XG ) → H 2(P2, Ω2 P2 ). For the computation of δ, we need to compute the lift of W01 • W2 + W0 • W12, namely, (W01 • W2 + W0 • W12) ∧ dG G ∈ C1(U, Ω2 P2 (log XG)), and then apply the Cech coboundary map δ: hδ(cid:16)(W01 • W2 + W0 • W12) ∧ dG G (cid:17)i = δ([W01 • W2 + W0 • W12]) inside H 2(P2, Ω2 P2 ). Following [2], consider the complexes C• ℓ (U, Ω2 P2 ) where a typical cochain φ has the form R012Ω where R012 ∈ Gℓ 012 k[x] with deg(R012) = 3ℓ(deg G − 1) − 3. If one denotes the cohomology groups of this complex by H • then one has ℓ (P2, Ω2 P2 ), H •(P2, Ω2 P2 ) = lim −→ H • ℓ (P2, Ω2 P2 ), where the map from C• P2 ) is given by multiplication of the numerator in the cocyle against G012. According to the lemma ([2], p18), the Grothedieck residue map (see 3.2 our explicit normalization of Res0) P2 ) to C• ℓ+1(U, Ω2 ℓ (U, Ω2 Res0 : H 2 ℓ (P2, Ω2 P2 ) −→ k is an isomorphism which commutes with multiplication by G012. If we let θ = Res0 ◦δ, then For simplicity, we use the following notation: hw0, w1, w2i = Res0(cid:0)δ([W01 • W2 + W0 • W12])(cid:1) . (2.5) A(ij) 0 := x1R(ij) 2 − x2R(ij) 1 , A(ij) 1 := −x0R(ij) 2 + x2R(ij) 0 , A(ij) 2 := x0R(ij) 1 − x1R(12) 0 , for (ij) = (01) or (12). One can easily check that Ωij Ωi Gi = Ωij Ωj Gj . for i, j = 0, 1, 2 with i < j. Then the above proposition 2.2 says that (2.6) − UkUℓΩij Ωj Gij Gj = − UkUℓΩij Ωi Gij Gi = (−1)i A(kℓ) i Ωi G2 i + (−1)j A(kℓ) j Ωj G2 j for (k, ℓ) = (0, 1) or (1, 2) and i, j = 0, 1, 2 with i < j. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 7 Lemma 2.3. We have the following formula δ(cid:16)(W01 • W2 + W0 • W12) ∧ dG G (cid:17) = deg G · A − B G2 012 where · Ω ∈ C2 2(U, Ω2 P2 ), A = U0x1R(12) B = A(01) 0 G0G1 + U2x0R(01) 2 G0G2 + U0x0R(12) 1 G0G2 + U0U1U2x0G0. 0 G0G0, 0 U2G1G2 + U0A(12) Proof. Note that W01 • W2 = W01 ∧ W2 = { A(01) 0 U2 G2 0G12 Ω0Ω01, A(01) 0 U2 G2 0G02 Ω0Ω02, A(01) 1 U2 G2 1G12 Ω1Ω01} ∈ C1 1, and U0A(12) G2 2G02 By using the lemma ([2], p 14), we get the following formulae: W0 • W12 = −W0 ∧ W12 = { U0A(12) G2 1G12 Ω1Ω01, 1 2 Ω2Ω02, U0A(12) G2 2G12 2 Ω2Ω01} ∈ C1 1. (W01 • W2 + W0 • W12) ∧ dG G = {S01 Ω G , S02 Ω G , S01 Ω G } ∈ C1(U, Ω2 P2 (log XG)), where S01 = S02 = S12 = 1 x2 −A(01) A(01) 0 U2x2 G0G12 0 U2x1 G0G02 −A(01) 1 U2x0 G1G12 + + −U0A(12) G12G1 2 x1 −U0A(12) G02G2 −U0A(12) G12G2 + , 2 x0 , . Therefore δ(cid:0)(W01 • W2 + W0 • W12) ∧ dG −(cid:20)(cid:18) A(01) (cid:19) x2G2 +(cid:18) A(01) 0 U2G1 + U0A(12) G (cid:1) is equal to G12G012 1 G0 G02G012 0 U2G2 + U0A(12) 2 G0 (cid:19) x1G1 +(cid:18) A(01) 1 U2G2 + U0A(12) 2 G1 G12G012 (cid:19) x0G0(cid:21) Ω G . Using the relation −x2G2 = x0G0 + x1G1 − deg G · G in P2, we continue the computation and is equal to dG δ(cid:16)(W01 • W2 + W0 • W12) ∧ G (cid:17) (cid:19) (x0G0 + x1G1 − deg G · G) (cid:19) x1G1 −(cid:18) A(01) 2 G0 Ω G 0 U2G1 + U0A(12) 1 G0 G12G012 0 U2G2 + U0A(12) (cid:18) A(01) −(cid:18) A(01) G02G012 0 U2G1 + U0A(12) 1 G0 = − deg G(cid:18) A(01) G12G012 0 U2G1G2 − A(01) +(cid:18) A(01) G2 012 1 U2G2 + U0A(12) 2 G1 G12G012 1 G0G2 − U0A(12) 2 G1G0 (cid:19) Ω +(cid:18) U0A(12) G2 012 (cid:19) x1G1 Ω G (cid:19) x0G0 Ω G (cid:19) x0G0 Ω G . 1 U2G2G0 + U0A(12) 1 G0G2 − U0A(12) 2 G1G0 On the other hand, we have that U0A(12) 1 G0G2 − U0A(12) 2 G1G0 = U0G0(A(12) 1 G2 − A(12) 2 G1) 2 + x2R(12) = U0G0(cid:16)(−x0R(12) = U0G0(cid:16)(x2G2 + x1G1)R(12) = U0G0(cid:16)deg G · G · R(12) 0 )G2 − (x0R(12) 1 − x1R(12) 0 − x0(R(12) 2 G2 + R(12) 1 0 − x0(R(12) 0 G0 + R(12) 1 G1 + R(12) 0 )G1(cid:17) )G1)(cid:17) 2 G2)(cid:17) = deg G · U0R(12) 0 G0G − U0U1U2x0G0. Similarly, we have that A(01) 0 U2G1G2 − A(01) 1 U2G2G0 = deg G · U2R(01) 2 G2G − U0U1U2x2G2. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 8 By substituting these terms in the equation regarding δ(cid:0)(W01 • W2 + W0 • W12) ∧ dG computation confirms that the numerator involved is divisible by G and this proves the lemma. G (cid:1), a straightforward (cid:3) 3. An explicit algorithm for the Massey triple product 3.1. Main theorem. Let g be the genus of X so that dimC H 0(X, Ω1 X ) = g. We choose all the possible U1 ∈ k[x]deg G−3 (modulo JG) and all the possible U0, U2 ∈ k[x]2 deg G−3 (modulo JG) such that (this can be done by using a computer algebra program) X ) = dimC H 1(X, Ω0 (3.1) U0U1 = 2Xi=0 R(01) i Gi, U1U2 = 2Xi=0 R(12) i Gi, for R(01) i , R(12) i ∈ k[x]. Then W0 • W1 and W1 • W2 are coboundaries by Theorem 2.1. The Massey triple product is not defined for all the elements of the first singular cohomology group H 1(X, C). Choosing U0, U1, U2 satisfying 3.1 amounts to finding a maximal isotropic subspace of H 1(X, C). Define AX := {W0, W1, W2, W01, W12 ∈ C1 0 ⊕ C0 1 : U0, U1, U2 satisfies 3.1}, where Wi are given in 2.2 and Wij are given in Proposition 2.2. Then Proposition 2.2 implies that AX is a defining system (Definition 1.1) for a Massey triple product for wi = [Wi], i = 0, 1, 2. The key point of Proposition 2.2 is that it gives an explicitly computable expression for the Cech representative Wij in terms of polynomials in k[x]. We compute the Massey triple product of w0, w1, w2 with respect to AX by the formula: using hw0, w1, w2iAX := Res0(cid:0)δ([W01 • W2 + W0 • W12])(cid:1) , θ : H 1(UX , Ω1 X ) δ−→ H 2 2 (UP2 , Ω2 P2 ) Res0−−−→ C. Lemma 2.3 gives an explicitly computable expression for δ: δ ([W01 • W2 + W0 • W12]) = [deg G · A − B G2 012 where A, B are given explicitly in Lemma 2.3. · Ω] ∈ H 2 2 (UP2 , Ω2 P2 ), The final thing is to compute Res0 : H 2 2 (P2, Ω2 P2 ) ≃ k explicitly. For this, let J be the ideal of k[x] generated by G0(x)2, G1(x)2, G2(x)2. Then there is an isomorphism (see Corollary (iv), p 18, [2]) Thus, if we put (k[x]/J)6 deg G−9 ≃ k. DetG(x) := det(cid:18) ∂Gi(x)2 ∂xj (cid:19) ∈ k[x]6 deg G−9, then DetG(x) does not belong to J and, for any given F (x) ∈ k[x]6 deg G−9, there is a unique RF ∈ k such that F (x) = 2Xi=0 vi(x)Gi(x)2 + RF · DetG(x), vi(x) ∈ k[x]. The fact that Res0 factors through F 7→ RF (see Lemma and its proof, p 18, [2]) Res0 : H 2 2 (P2, Ω2 P2 ) ≃ (k[x]/J)6 deg G−9 ≃ k implies that, Then (3.2) So our final formula is the following: Res0(cid:18) F (x) G2 012 · Ω(cid:19) = RF . Theorem 3.1. Let wi = [Wi] correspond to Ui satisfying 3.1. Assuming all the notations so far, we have that where hw0, w1, w2iAX = Rdeg G·(A−B), A = U0x1R(12) B = A(01) 0 U2G1G2 + U0A(12) 0 G0G1 + U2x0R(01) 2 G0G2 + U0x0R(12) 1 G0G2 + U0U1U2x0G0. 0 G0G0, See 2.5 and 3.1 for the notations R(jk) i and A(jk) i . This theorem implies that the Massey triple product are completely computable using a computer algebra program. In particular, we have AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 9 Corollary 3.2. The Massey triple product hw0, w1, w2iAX relative to AX vanishes if and only if A − B ∈ J = hG0(x)2, G1(x)2, G2(x)2i. Corollary 3.3. Vanishing of hw0, w1, w2iAX does not depend on choices of cohomology representatives of [W0], [W1], [W2]. Proof. Since a cohomology representative Wi of wi corresponds to Ui via the Griffiths isomorphism 2.1, the statement that Wi is cohomologous tofWi is equivalent to that Ui −eUi belongs to the Jacobian ideal JG (we use for quantities corresponding tofWi). It suffices to prove the corollary for each Ui, i = 0, 1, 2, one the notation e· by one. We start from U0. Let eU0 := U0 + T0G0 + T1G1 + T2G2 for Ti ∈ C[x]. In this case, eR(01) eA(01) k + Tk+2xk+1U1 − Tk+1xk+2U1, i = R(01) k = A(01) k ≡ 0, 1, 2 mod 3. eR(12) i = R(12) i + TiU1, i = 0, 1, 2, , i eA ≡ eU0R(12) 0 G0G1 + U2x0eR(01) ≡ (U0 + T2G2)x1R(12) k , k = A(12) eA(12) 2 + T2U1(cid:17) G0G2 mod J 0 G0G1 + U2x0(cid:16)R(01) 2 G0G2 mod J ≡ A + T2x1R(12) 0 G0G1G2 + T2x0R(12) 1 G0G1G2 mod J. eB ≡ eA(01) ≡(cid:16)A(01) 0 U2G1G2 + eU0A(12) 0 + T2x1U1 − T1x2U1(cid:17) U2G1G2 + (U0 + T1G1)A(12) 1 G0G2 + eU0U1U2x0G0 mod J 1 G0G2 + (U0 + T1G1 + T2G2)U1U2x0G0 mod J 0 G0G1G2 + T1A(12) ≡ B + (T2x1 − T1x2)R(12) 1 G0G1G2 + T1x0R(12) 2 G0G1G2 + T2x0R(12) 0 G0G1G2 + T2x0R(12) ≡ B + T2x1R(12) 1 G0G1G2 mod J 1 G0G1G2 mod J Therefore, eA − eB ≡ A − B mod J. If we let eU2 := U2 + T0G0 + T1G1 + T2G2 for Ti ∈ C[x], then a similar computation again confirms that eA − eB ≡ A − B mod J. Finally, let eU1 := U1 + T0G0 + T1G1 + T2G2. Since degeU1 = deg U1 = deg G − 3, the only possibility is Therefore, eR(ij) Proof. Since deg G = 3, we can assume that U1 = 1, U0 =P2 . Hence eA − eB ≡ A − B mod J in this case, too. i Gi, and U2 =P2 Corollary 3.4. If deg G = 3, then A − B ∈ J, i.e., hw0, w1, w2iAX = 0 for all w0, w1, w2. and eA(ij) T0G0 + T1G1 + T2G2 = 0 i Gi. By Theorem k = R(ij) k = A(ij) i=0 R(01) i=0 R(12) 3.1, (cid:3) k k A = U0x1R(12) 0 G0G1 + U2x0R(01) 2 G0G2 + U0x0R(12) 0 G0G0, = (cid:16) 2Xi=0 R(01) i Gi(cid:17)x1R(12) 0 G0G1 +(cid:16) 2Xi=0 R(12) i Gi(cid:17)x0R(01) 2 G0G2 +(cid:16) 2Xi=0 ≡ x1R(01) 2 R(12) 0 G0G1G2 + x0R(01) 2 R(12) 1 G0G1G2 mod J, B = A(01) 0 U2G1G2 + U0A(12) 1 G0G2 + U0U1U2x0G0, R(01) i Gi(cid:17)x0R(12) 0 G0G0, = (x1R(01) 2 − x2R(01) 1 )(cid:16) 2Xi=1 i Gi(cid:17)(cid:16) 2Xi=1 R(12) i Gi(cid:17)G1G2 +(cid:16) 2Xi=1 i Gi(cid:17)x0G0, R(12) R(01) +(cid:16) 2Xi=1 R(01) i Gi(cid:17)(−x0R(12) 2 + x2R(12) 0 )G0G2 )R(12) 0 G0G1G2 + R(01) 2 + x2R(12) 0 )G0G2 ≡ (x1R(01) 1 2 − x2R(01) 1 G1R(12) 2 R(12) +R(01) ≡ x1R(01) 2 G2x0G0 + R(01) 0 G0G1G2 + x0R(01) 2 G2R(12) 2 R(12) 1 G1(−x0R(12) 1 G1x0G0 mod J, 1 G0G1G2 mod J. Therefore, A − B ≡ 0 mod J, i.e., A − B ∈ J. (cid:3) AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 10 Table 3.1. Defining systems for the Massey triple product: A = {W0, W1, W2, W01, W12} W0 W1 W2 W01 W12 hW0, W1, W2iA anti-holomorphic anti-holomorphic holomorphic holomorphic anti-holomorphic holomorphic holomorphic anti-holomorphic holomorphic anti-holomorphic anti-holomorphic anti-holomorphic anti-holomorphic anti-holomorphic holomorphic holomorphic holomorphic holomorphic anti-holomorphic holomorphic anti-holomorphic holomorphic anti-holomorphic holomorphic * 0 * * 0 0 0 * * * 0 * 0 0 * 0 * * * 0 0 0 0 0 3.2. Other possible Massey triple products. We explain why we chose w0, w2 anti-holomorphic and w1 holomorphic in order to get a non-trivial defining system. The first line in Table 3.1 is the case we considered in Theorem 3.1. The second and third lines in Table 3.1 can be considered in the same way as Theorem 3.1. For other remaining choices, we do not know how to get some non-zero Wij such that D(Wij ) = Wi ∪ Wj and the Massey triple product is zero for trivial reasons (the cup product of two holomorphic forms (respectively, two anti-holomorphic forms) is zero). Note that we do not know an explicit expression3 Wij ∈ C0 1 such that D( Wij ) = Wi • Wj ∈ C1 1 when wi ∪ wj = 0. The asterisk * in Table 3.1 means that it is not zero in general. 4. Computer algorithm and experiment The goal of this section is to provide an explicit computer algorithm and to calculate triple Massey products 2 as n of some examples. Our example is a family of smooth projective curve XG given by G(x) = xn varies. Let us use the following simplified notations: 1 + xn 0 + xn hU0, U1, U2i = hw0, w1, w2iAX , where the relations between Ui and wi are give in (2.2). We use the program SageMath8.3 for running the algorithms. First, we have to find U0, U1, U2 for computing the triple Massey product such that U0U1, U1U2 ∈ JG. Our method for choosing U0, U1 such that U0U1 ∈ JG is to generate repeatedly homogeneous polynomials U0, U1 randomly until the multiplication U0U1 is in JG. During this repeated process, we analyze the relation between the number of monomial terms of selected polynomials U0, U1 and vanishing behavior of the cup product. For this, we choose 50,000 pairs of homogeneous polynomials U0 and U1 such that degU0 = 2n − 3 and degU2 = n − 3 and then calculate the ratio of pairs U0, U1 among 50,000 pairs such that U0U1 ∈ JG. Define M1(n) = (n − 1)(n − 2) 2 , M2(n) = (2n − 1)(2n − 2) 2 where M1(n) (respectively, M2(n)) is the number of monomials of degree n − 3 (respectively, 2n − 3). Then we calculate the ratio of vanishing cup product when the number of monomial terms of randomly chosen U0 (respectively, U1) is less than or equal to p M2(n) q) as we vary ℓ = 1, 2, · · · . The results are recorded in the graphs of Figure 1. The x-axis means the degree n of G(x) and the y-axis means the ratio of vanishing cup product among the 50,000 pairs. When there is no limitation of number of monomial terms (the bottom graph), the ratio of vanishing cup product converges to zero as the degree n grows. However, we see that the vanishing ratio grows as the ℓ increases (i.e. the number of monomial terms of U0, U1 becomes smaller). Moreover, in the top graph case (U0 and U1 themselves are monomials), it converges to 100 as the degree n grows. q (respectively, p M1(n) ℓ ℓ 3We only know Wij ∈ C1 0 such that D(Wij ) = Wi • Wj in Proposition 2.2. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 11 (%) 100 80 60 40 20 0 monomial ℓ = 50 ℓ = 30 ... ℓ = 16 ℓ = 14 ℓ = 12 ℓ = 10 ℓ = 8 ℓ = 6 ℓ = 1 ℓ = 2 ℓ = 4 5 10 15 20 25 30 (degree) Figure 1. The graphs of ratios of vanishing cup product Therefore, for simplification of computation of the triple Massey product (we need vanishing of the cup product), we choose U0, U1 so that their number of monomial terms is less than or equal to three. Algorithm 1: Making Ui corresponds to wi = [Wi] for triple massey product randomly. Input: Polynomial G(x) Output: Homogeneous polynomials U0, U1, U2 such that U0U1 ∈ JG and U1U2 ∈ JG begin repeat U0 ← homogeneous random polynomial in k[x] with degree=2 deg G − 3 and number of terms ≤ 3 U1 ← homogeneous random polynomial in k[x] with degree=deg G − 3 and number of terms ≤ 3 U2 ← homogeneous random polynomial in k[x] with degree=2 deg G − 3 and number of terms ≤ 3 until U0U1 ∈ JG and U1U2 ∈ JG return U0, U1, U2 end The following examples are random polynomials which came from Algorithm 1 where G(x) = x5 0 + x5 1 + x5 2: (4.1) (4.2) U0 = x3 U0 = − 1 1x2 2 0x4 1 + x5 6 x3 0x2 1x2 2 4 x2 U1 = 1 U1 = x2 2 2 U2 = 1 U2 = 2 3 x4 9 x4 0x1x2 2, 0x3 2. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 12 Next, let us calculate the triple Massey product. Consider the following algorithm: Algorithm 2: Calculate hU0, U1, U2i. Input: Polynomials G, U0, U1, U2 such that U0U1 ∈ JG and U1U2 ∈ JG Output: Triple massey product hU0, U1, U2i begin i Gi i Gi i ←Polynomials such that U0U1 =P R(01) i ←Polynomials such that U1U2 =P R(12) /* Calculate A,B in Theorem 3.1 R(01) R(12) A, B ← By Theorem 3.1 /* Make (k[x]/J)6 deg G−9 for Calculation. M ←Ordered set of all of mononmials whose degree is 6 deg G − 9 N ←Set of all af monomials whosee degree is 4 deg G − 7 L ← {G2 J ← ∅ for n ∈ N do 1, G2 2} 0, G2 / / for ℓ ∈ L do J ← J ∪ {nℓ} end end K ← k-vector space dimension=M JV ← Span(vector representations of elements in J) K/J ← Qutient space K/JV hU0, U1, U2i ←(Image of deg G · (A − B) in K/J)/(Image of DetG(x) in K/J) return hU0, U1, U2i /* 1 dimensional vectorspace */ end 1 2 3 4 5 The main step of Algorithm 2 is to realize (k[x]/JG)6 deg G−9 in a computer. We will use a finite dimensional k-vector space with the standard basis in order to handle the polynomial ring in a computer4. We take k- vecotr space K in line 4 with standard basis {e1, . . . , es} which corresponds to M = {m1, . . . , ms} in line 1. Then we take J as the basis of J6 deg G−9 (line 2∼3), and make JV in line 5 which corresponds to J under the correspondence between {e1, . . . , es} and {m1, . . . , ms}. This enables us to compute the quotient space and the remaining calculation using a computer. As a result of Algorithm 2, we provide two numerical examples with zero (4.1) and nonzero (4.2) Massey triple products: (4.1) (4.2)     1 · G2, 0x2 20 x5 20 x3 2 · G1 + 1 U0U1 = 0 · G0 + 1 U1U2 = 0 · G0 + 0 · G1 + 1 01 x4 A − B = 5 deg G · (A − B) = 0 · DetG(x) + 1 hU0, U1, U2i = 0. 60 x4 0x1 0x1 · G2, 11x6 01 x7 2 + 5 01 x0x6 2 − 5 01 x9 0x6 1x6 0x5 1x6 1x9 2, 2 · G2 0 − 1 01 x4 0x3 1x6 2 · G2 1 + 1 01 x7 0x5 1x2 · G2 2, U0U1 = 0 · G0 + 0 · G1 − 1 U1U2 = 0 · G0 + 0 · G1 + 2 1x7 A − B = − 5 2 + 5 1 8640000 · DetG(x) − 1 deg G · (A − B) = 1 hU0, U1, U2i = 0x2 1 · G2, 0x2 · G2, 0x7 30 x3 45 x4 27 x7 27 x0 01x2 1x7 2, 8640000 . 27 x4 0x2 1x7 2 · G2 0 + 0 · G2 1 + 0 · G2 2, where DetG(x) = 8000000x7 0x7 1x7 2. References [1] Peters, C., Steenbrink, J.; Infinitesimal variations of Hodge structure and the generic Torelli problem for pro- jective hypersurfaces (after Carlson, Donagi, Green, Griffiths, Harris), Classification of algebraic and analytic manifolds (Katata, 1982), 399 -- 463, Progr. Math., 39, Birkhauser Boston, Boston, MA, 1983. [2] Carlson J., Griffiths P.; Infinitesimal variations of Hodge structure and the global Torelli problem, in Journees de geometrie algebrique, Angers, juillet 1979, Sijthoff and Noordhoff, Alphen aan den Rijn, 1980, 51 -- 76. 4The program Sagemath has a function which calculates a quotient space of a finite dimensional vector space. AN ALGORITHM FOR A MASSEY TRIPLE PRODUCT OF A SMOOTH PROJECTIVE PLANE CURVE 13 [3] Pierre Deligne, Phillip Griffiths, John Morgan, and Dennis Sullivan: Real homotopy theory of Kähler manifolds, Invent. Math. 29 (1975), no. 3, 245 -- 274. [4] Griffiths, Phillip A.: On the periods of certain rational integrals. I, II, Ann. of Math. (2) 90 (1969), 460-495; ibid. (2) 90 (1969), 496-541. [5] Park, Jae-Suk; Park, Jeehoon: Enhanced homotopy theory for period Integrals of smooth projective hypersurfaces, Communications in Number theory and Physics, Volume 10 (2016), Number 2, Pages 125-337. E-mail address: [email protected] E-mail address: [email protected] E-mail address: [email protected] E-mail address: [email protected] Department of Mathematics, POSTECH, 77 Cheongam-Ro, Namgu, Pohang, Gyeongbuk, 37673, South Korea.
1511.04677	1	1511	2015-11-15T09:50:27	Connected components of closed affine Deligne-Lusztig varieties in affine Grassmannians	[ "math.AG" ]	We determine the set of connected components of closed affine Deligne-Lusztig varieties for hyperspecial maximal parahoric subgroups of unramified connected reductive groups. This extends the work by Viehmann for split reductive groups, and the work by Chen-Kisin-Viehmann on minuscule affine Deligne-Lusztig varieties.	math.AG	math	CONNECTED COMPONENTS OF CLOSED AFFINE DELIGNE-LUSZTIG VARIETIES IN AFFINE GRASSMANNIANS SIAN NIE Abstract. We determine the set of connected components of closed affine Deligne-Lusztig varieties for hyperspecial maximal parahoric subgroups of unramified connected reductive groups. This extends the work by Viehmann for split reductive groups, and the work by Chen-Kisin-Viehmann on minuscule affine Deligne-Lusztig varieties. 1. Introduction 1.1. Let Fq be a finite field with q elements and let k be an algebraic closure of Fq. Let k[[t]] (resp. Fq[[t]]) be the ring of formal power series over k (resp. Fq), whose fractional field is denoted by L (resp. F ). Let σ be the Frobenius automorphism of L/F . Let G be a (connected) reductive group over Fq[[t]]. We also de- note by σ the induced automorphism on G(L). Since G is unramified and quasi-split over Fq[[t]], there exist a Borel subgroup B ⊆ G and a maximal torus T ⊆ B (over Fq[[t]]), which split over k[[t]]. Denote by Y the (absolute) cocharacter group of T and by Y + ⊆ Y the set of G-dominant cocharacters defined with respect to B. We have Cartan decomposition G(L) = ⊔µ∈Y +KtµK, where K = G(k[[t]]) is hyperspe- cial maximal parahoric subgroup of G(L). For λ ∈ Y + and b ∈ G(L), the affine Deligne-Lusztig variety X G λ (b) = Xµ(b) is defined as Xλ(b) = {g ∈ G(L)/K; g−1bσ(g) ∈ KtλK}. The closed affine Deligne-Lusztig variety X(cid:22)λ(b) is defined as X(cid:22)λ(b) = ∪λ′(cid:22)λXλ′(b). Here λ′ (cid:22) λ means Ktλ′K/K is contained in the closure of KtλK/K in G(L)/K. Note that Xλ(b) = X(cid:22)λ(b) if µ is minuscule. The main purpose of this paper is to determine the set π0(X(cid:22)λ(b)) of connected components of X(cid:22)λ(b). When G is split over Fq[[t]] or λ is minuscule, this problem has been solved by Viehmann [13] and by Chen-Kisin-Viehmann [2] respectively. Their description of π0(X(cid:22)λ(b)) has essential applications on the connected components of unramified This work is supported in part by the National Natural Science Foundation of China (No. 11321101). 1 2 SIAN NIE Rapoport-Zink spaces (see [1] and [2]) and on Langlands-Rapoport conjecture for mod p points on Shimura varieties (see [10]). In this paper, we finish the computation of π0(X(cid:22)λ(b)) in general case. 1.2. To state our main result, we need more notations. Let [b] = {g−1bσ(g); g ∈ G(L)} be the G(L)-conjugacy class of b. Due to Kottwitz, [b] is determined by two invariants. One is the Kottwitz point κG [b], the image of b under the natural projection κG : G(L) → π1(G)σ := π1(G)/(1 − σ)(π1(G)). The other is the Newton point νG [b], the G-dominant cochracter conjugate to the Newton cocharater of b. By [11] and [3], X(cid:22)λ(b) 6= ∅ if and only if κG(b) = κG(tλ) and λ⋄ −νG [b] is a linear combination of coroots occurring in B with nonnegative rational coefficients. Here λ⋄ is the average of σ-conjugates of λ. From now on we assume X(cid:22)λ(b) 6= ∅. Following [2], we say the pair (λ, b) is irreducible with respect to the Hodge-Newton decomposition (HN irreducible for short) if the coefficient of any simple coroot in λ⋄ − νG [b] is strictly positive. Thanks to [2, §2.4 & §2.5], to study π0(X(cid:22)λ(b)), it suffices to consider the case that G is simple and adjoint, and (b, λ) is HN-irreducible. Thus the following main result of the paper, which is conjectured by Chen-Kisin-Viehmann, completes the computation of connected components of closed affine Deligne-Lusztig varieties. Theorem 1.1. Assume G is adjoint and simple. If (λ, b) is HN- irreducible, then the natural projection ηG : G(L) → π1(G) induces a bijection π0(X(cid:22)λ(b)) ∼= (σ − 1)−1(ηG(tλ) − ηG(b)) ⊆ π1(G). The following is a byproduct of Theorem 1.1, which holds without any restrictions on G, λ and b. Theorem 1.2. Let Jb(F ) = {g ∈ G(L); g−1bσ(g) = b} be the σ- centralizer of b. Then Jb(F ) acts transitively on π0(X(cid:22)λ(b)) by left multiplication. Remark. Theorem 1.1 and Theorem 1.2 are proved if G splits over k[[t]] (see [13]) or λ is minuscule (see [2]). The proofs of the main results follow the strategy of [2]. Firstly, we prove Theorem 1.1 when b is superbasic, that is, b is not σ-conjugate to an element of any proper Levi subgroup of G. As a consequence, Jb(F ) acts on π0(X(cid:22)λ(b)) transitively. Secondly, we find a suitable standard parabolic subgroup P = MN ⊆ G and a superbasic σ-conjugacy class [b′]M ⊆ [b] of M(L) with νM [b] such that each connected com- µ (b′), where ponent of X(cid:22)λ(b′) contains a point in (Jb′(F ) ∩ N(L))X M = νG [b′]M 3 M ⊇ T (resp. N) is the Levi subgroup (resp. unipotent radical) of P and µ ranges over the set ¯Iλ,M,b′ = {µ ∈ Y ; µ (cid:22) λ, µ is M-minuscule and κM (b′) = κM (tµ)}. Thirdly, we show the image of the natural map ϕµ : π1(M)σ ∼= π0(X M µ (b′)) → π0(X(cid:22)λ(b′)) dose not depend on the choice of µ ∈ ¯Iλ,M,b′. Here π1(M)σ is the set of σ-fixed points on π1(M). From the first three steps, we see immediately Jb′(F ) acts on π0(X(cid:22)λ(b′)) transitively and Theorem 1.2 follows. To deduce Theorem 1.1 (i.e., π0(X(cid:22)λ(b′)) ∼= π1(G)σ), we need to show, under the assumption that (λ, b) is HN-irreducible, that ϕµ is surjective, and moreover, any two points P, P ′ ∈ X M µ (b′) are in the same connected component of X(cid:22)λ(b′) if and only if ηG(P ) − ηG(P ′) lies in the kernel of the natural surjective homomorphism π1(M)σ ։ π1(G)σ. This is the last step of the whole proof. The first two steps are already established in [2]. In this paper, we develop the techniques used in [2] systematically and finish the last two steps. Based on Deligne-Lusztig reduction methods, we also provide a new conceptual proof for the first step, avoiding concrete computations in superbasic case. Compared to the original proof for the case that λ is minuscule, there are two main difficulties for the general case. The first lies in the third step. We need to construct affine lines in X(cid:22)λ(b′) connecting X M µ (b′) µ′ (b′) for any µ, µ′ ∈ ¯Iλ,M,b. When λ is minuscule, the original with X M construction of affine lines relies on the property that each element of ¯Iλ,M,b is conjugate to λ (under the Weyl group of T ). In general case, ¯Iλ,M,b becomes much more complicated. To overcome the difficulty, we introduce a new algorithm for the construction via defining the key set Θ(µ, µ′, λ) (see §6 for definition). The other difficulty is to prove the sufficiency direction of the "moreover" part in the last step. There is a gap in the original proof in [2]. However, this gap can be filled in (for minuscule λ) by Miaofen Chen via a case-by-case analysis. For general case, we figure out a conceptual proof using weakly dominant cocharacters (see §4 for definition), extending Chen's arguments. The paper is organized as follows. In §2, we collect basic notations and properties which are used frequently in the paper. In §3, we prove Theorem 1.1 for superbasic case. In §4, we show the existence of the Levi subgroup M as above such that ¯Iλ,M,b′ contains a weakly domi- nant cocharacter. In §5, we outline the proof of Theorem 1.1, where the surjectivity of ϕµ in the fourth step is redundant. In §6 and §7, we in- troduce the set Θ(µ, µ′, λ) and establish the third step (see Proposition 5.1). In §8 and §9, we finish the last step (see Proposition 5.3). Acknowledgement. We would like to thank Xuhua He for his preprint [7]. We are grateful to Miaofen Chen for sharing her essential ideas on 4 SIAN NIE the proof of Proposition 5.3. Various definitions and arguments, espe- cially in Section 5, 7, 8 and 9, are inspired or borrowed form the work of Chen-Kisin-Viehmann [2]. Their work has a foundational influence on the present paper. Part of the work was done during the author's visit to the Institut Mittag-Leffler. We are grateful to the institute for the excellent working atmosphere. 2. Preliminaries 2.1. Let G ⊇ B ⊇ T be as above. We denote by R = (Y, Φ∨ G, X, ΦG, ΠG) the root datum of G, where X (resp. Y ) is the absolute character (resp. cocharacter) group of T ; ΦG = Φ ⊆ X (resp. Φ∨ ⊆ Y ) is the set of roots (resp. coroots); ΠG = Π0 ⊆ Φ is the set of simple roots occurring in B. Moreover, there exists a bijection α 7→ α∨ between Φ and Φ∨. Let NT ⊆ G be the normalizer of T in G. The quotients W0 = NT (L)/T (L) and W = NT (L)/T (k[[t]]) are called the Weyl group and the extended affine Weyl group of G respectively. We have WG = W = Y ⋊ W0 = {tµw; µ ∈ Y, w ∈ W0}. We can embed W into the group of affine transformations of YR, where the action of w = tµw is given by v 7→ µ + w(v). Here W0 acts on YR naturally by scalar extension. Let ΦG = Φ = Φ×Z be the set of (real) affine roots. Let a = (α, k) ∈ Φ. We can view a as an affine function such that a(v) = −hα, vi + k for v ∈ YR, where h, i : XR × YR → Z is the scalar extension of the natural pairing between X and Y . The induced action of W on Φ is given by ( w(a))(v) = a( w−1(v)) for w ∈ W . Let sa = tkα∨sα ∈ W be the corresponding affine reflection. Then {sa; a ∈ Φ} generates the affine Weyl group W a = ZΦ∨ ⋊ W0 = {tµw; µ ∈ ZΦ∨, w ∈ W0}. Let Φ+ = Φ ∩ NΠ0 be the set of positive roots and let Y + R = {v ∈ YR; hα, vi > 0, α ∈ Φ+} be the set of dominant vectors. For v ∈ YR, we denote by ¯v the unique dominant vector in the W0-orbit of v. Let ∆ = {v ∈ YR; 0 < hα, vi < 1, α ∈ Φ+} be the base alcove. Note that W = W a⋊Ω, where Ω = {x ∈ W ; x(∆) = ∆}. Set Φ+ = {a ∈ Φ; a(∆) > 0} and Φ− = − Φ+. Then Φ = Φ+ ⊔ Φ−. The associated length function ℓ : W → N is defined by ℓ( w) = Φ− ∩ w( Φ+). Let Sa = {sa; a ∈ Φ, ℓ(sa) = 1} be the set of simple affine reflections and let S0 = Sa ∩ W0. Then (W a, Sa) and (W0, S0) are Coxeter systems. For w, w′ ∈ W , we say w ≤ w′ if there exists a sequence w = w1, . . . , wr = w′ such that ℓ( wk) < ℓ( wk+1) and wk w−1 k+1 is an affine reflection for k ∈ [1, r − 1]. We call this partial order ≤ the Bruhat order on W . 5 Let J ⊆ Sa. We denote by WJ the parabolic subgroup of W a gen- erated by J. Let W ⊇ WJ be a subgroup of W . We set W J = {min( wWJ ); w ∈ W }, J W = (W J )−1 and J W J = J W ∩ W J . 2.2. Let σ be the Frobenius automorphism of G(L). We also denote by σ the induced automorphism on the root datum R. Then σ acts on YQ as a linear transformation of finite order. For w ∈ W , there exists n ∈ N such that ( wσ)n = tξ for some ξ ∈ Y . We define ν w = ξ/n, which dose not depend on the choice of n. Let ηG : G(L) → π1(G) = Y /ZΦ∨ ∼= Ω be the natural homomor- phism such that ηG(KtµK) = µ ∈ Y /ZΦ∨ for µ ∈ Y . Let κG : G(L) → π1(G)σ be the Kottwitz map obtained by composing ηG with the quo- tient map π1(G) → π1(G)σ = π1(G)/(1 − σ)(π1(G)). Since K lies in the kernel of ηG, we also denote by ηG (resp. κG) the induced map on W or on G(L)/K. [b] and the Newton point νG For b ∈ G(L), we denote by [b] = {g−1bσ(g); g ∈ G(L)} the σ- conjugacy class of G(L) containing b. It is known that [b] ∩ NT (L) 6= ∅. Then the Kottwitz point κG [b] of [b] are given by κ(τ ) and ¯νp(τ ) respectively for any τ ∈ [b]∩NT (L). Here p : NT (L) → W is the natural projection. Note that σ(νG [b] since σ([b]) = [b]. We say b or [b] is basic if νG [b]i = 0 for any α ∈ Φ. We say b or [b] is superbasic if there dose not exists a proper Levi subgroup M ⊆ G (over Fq[[t]]) such that [b] ∩ M(L) 6= ∅. [b] is central on Φ, that is, hα, νG [b]) = νG We say ω ∈ Ω is σ-superbasic in W if each orbit of the action s 7→ ωσ(s)ω−1 on Sa is a union of connected components of the affine Dynkin diagram of Sa. Lemma 2.1. Let b ∈ G(L). Then [b] is a superbasic σ-conjugacy of G(L) if and only if there exits τ ∈ [b] ∩ NT (L) such that p(τ ) ∈ Ω is σ-superbasic in W . Proof. It follows from [8, Proposition 3.5] and [2, Lemma 3.1.1]. (cid:3) 2.3. Let M ⊆ G be a Levi subgroup containing T . Replacing the triple T ⊆ B ⊆ G with T ⊆ M ∩ B ⊆ M, we can define, as in §2.1 and §2.2, ΠM , ΦM , WM , Sa M , ≤M and so on. Let J ⊆ S0 with σ(J) = J. We denote by MJ ⊆ G the Levi subgroup generated by T and the root subgroups Uα such that sα ∈ WJ . We write ΠJ = ΠMJ , ΦJ = ΦMJ , WJ = WMJ and so on for simplicity. Moreover, for v ∈ YR, we denote by ¯vJ the unique J-dominant elements in the WJ -orbit of v. For µ, λ ∈ Y , we write µ 6J λ if λ − µ ∈ NΠ∨ J and write µ (cid:22)J λ if ¯µJ 6J ¯λJ . If J = S0, we abbreviate 6J (resp. (cid:22)J ) to 6 (resp. (cid:22)) for simplicity. The following two properties will be used frequently in the paper. (a) µ (cid:22) λ if µ (cid:22)J λ; 6 SIAN NIE (b) Let α ∈ ΦJ . Then µ − α∨ (cid:22)J µ if hα, µi > 1. Moreover, µ − α∨ ≺J µ if hα, µi > 2. Let KJ = MJ (k[[t]]) = K ∩ MJ (L). Via the natural embedding MJ (L)/KJ ֒→ G(L)/K, we can view X MJ (cid:22)J λ(b), the closed Deligne- Lusztig variety define with respect to MJ , as a closed subset of X(cid:22)λ(b) for b ∈ MJ (L). 2.4. Let J ⊆ S0 with σ(J) = J and b ∈ G(L). We say the pair (J, b) is admissible if the following two statements holds: [b] is central on ΦJ , that is, hα, νG (a) νG (b) [b] ∩ MJ (L) contains a (unique) σ-conjugacy class [b]J,dom of [b]i = 0 for any α ∈ ΦJ ; MJ (L) such that νMJ [b]J,dom = νG [b]. Assume (J, b) is admissible. We define ¯Iλ,MJ ,b = ¯Iλ,J,b = {x ∈ π1(MJ ); κJ (x) = κMJ [b]J,dom , µx (cid:22) λ}. Here µx ∈ Y is the unique J-dominant and J-minuscule coweight such that ηJ (tµx) = ηMJ (tµx) = x. Via the bijection x 7→ µx, we also view ¯Iλ,J,b as the set of J-dominant J-minuscule cocharacters µ such that µ (cid:22) λ and κJ (tµ) = κJ (b). Furthermore, we say b is superbasic in MJ if [b]J,dom is a superbasic σ-conjugacy class of MJ (L). For x ∈ π1(MJ ), we denote by bx = tµxwx ∈ ΩJ the unique element such that ηJ (bx) = x, where wx = uxwJ and ux (resp. wJ ) is the unique maximal element in parabolic subgroup generated by {s ∈ J; s(µx) = µx} (resp. J). Lemma 2.2. Let J ⊆ S0 with σ(J) = J and b ∈ G(L) such that (J, b) is admissible. If ¯Iλ,J,b 6= ∅, then νbx = νG [b] for any x ∈ ¯Iλ,J,b. Proof. By definition, there exists τ ∈ [b]J,dom ∩ NT (L) such that νp(τ ) = νG [b] is central on ΦJ . Let x0 = ηJ (τ ) ∈ π1(MJ ). Applying [8, Theorem 3.3] to WJ , we have νbx0 = νp(τ ) = νG [b]. Let x ∈ ¯Iλ,J,b. We show νbx = νbx0 [b]J,dom = κJ (x0). We have x = x0 + (1 − σ)(x′) ∈ π1(MJ ) for some x′ ∈ π1(MJ ). Then bx = bx′bx0σ(b−1 [b] is central on ΦJ , we have νbx = wx′(νbx0 (cid:3) x′ ) ∈ ΩJ . Since νbx0 . Note that κJ (x) = κMJ as desired. ) = νbx0 = νG Terminology. Let g, h ∈ G(L) and Z ⊆ G(L). We put gh = ghg−1 (resp. gZ = gZg−1) and gσh = gσ(h)g−1 (resp. gσZ = gσ(Z)g−1). For λ ∈ Y +, J ⊆ S0 and b ∈ G(L), we write Q ∼λ,J,b P if P and If J = S0, we Q are in the same connected component of X MJ abbreviate ∼J,λ,b to ∼λ,b for simplicity. (cid:22)J λ(b). For w ∈ WJ with J ⊆ S0, we denote by w a lift of w in NT (L)∩M(L) under the natural projection p : NT (L) → W . 7 3. The superbasic case In this section, we use the Deligne-Lusztig reduction method to study the connected components of unions of affine Deligne-Lusztig varieties in affine flag varieties. As an application, we obtain a new conceptual proof of Theorem 1.1 in the superbasic case. Let B− be the opposite Borel subgroup of B such that B ∩ B− = T . Let I ⊆ G(L) be the pre-image of B−(k) under the natural mod t projection K = G(k[[t]]) → G(k). We call I an Iwahori subgroup of G(L). We have Bruhat decomposition G(L) = ⊔ w∈ W I wI. For a = (α, k) ∈ Φ with α ∈ Φ and k ∈ Z, let Ua denote the corresponding affine root subgroup of the loop group LG. We have Ua(k) = Uα(ktk). Here Uα ⊆ G is the root subgroup of α. Then I = T (k[[t]])Qa∈ Φ+ Ua(k). Lusztig variety (in affine flag variety) by For w ∈ W and b ∈ G(L), we define the corresponding affine Deligne- X w(b) = {gI ∈ G(L)/I; g−1bσ(g) ∈ I wI}. We say w is reducible (with respect to b) if the following two statements holds: (a): X w(b) 6= ∅; (b): for any Q ∈ X w(b), there exists a morphism φ : P1 → G(L)/I and an open subset 0 ∈ U ⊆ A1 = P1 − {∞} such that φ(U) ⊆ X w(b), Q = φ(0) and P = φ(∞) ∈ Xu(b) for some u ∈ W such that u < w in the sense of Bruhat order. Lemma 3.1. w is reducible if so is σ( w). Lemma 3.2. If w ∈ W is non-reducible, then it is of minimal length in its σ-conjugacy classes in W . To prove the lemma, we need some preparation. For w, w′ ∈ W and s ∈ Sa, we write w s−→σ w′ if w′ = swσ(s) and ℓ(w′) 6 ℓ(w). We write w →σ w′ if there is a sequence w = w0, w1, · · · , wn = w′ of elements in sk−→σ wk for some sk ∈ Sa. We W such that for each k ∈ [1, n], wk−1 write w ≈σ w′ if w →σ w′ and w′ →σ w. Theorem 3.3. [8] Let O be a σ-conjugacy class in W and let Omin be the set of minimal length elements in O. Then for each w ∈ O, there exists w′ ∈ Omin such that w →σ w′. Let Q, P ∈ G(L)/I. Then there exists a unique w ∈ W such that gQ = I and gP = wI for some g ∈ G(L). In this case, we write P w→ Q. For example, Q ∈ X w(b) if and only if Q w→ bσ(Q). Define Z w = {(Q, P ) ∈ (G(L)/I)2; Q w→ P } Lemma 3.4. Let w, w1, w2 ∈ W . We have the following properties: 8 SIAN NIE (1) If P w1→ H w2→ Q and Q w′ (2) If Q w→ P and w = w1 w2 with ℓ( w) = ℓ( w1) + ℓ( w2), then there → P , then I wI ⊆ I w1I w2I; exists a unique H ∈ G(L)/I such that P w1→ H w2→ Q; (3) Z w = ∪u6 wZu, where Z w denotes the closure of Z w in (G(L)/I)2. Proof of Lemma 3.2. Assume w is not of minimal length in its con- jugacy classes. We show it is reducible. Let b ∈ G(L) such that X w(b) 6= ∅. By [8, Theorem 1.1], there exist w′ ∈ W with w′ ≈σ w and s ∈ Sa such that s w′σ(s) < w′. First we show (a) w′ is reducible. Since w′ ≈σ w, X w′(b) 6= ∅. We choose Q′ ∈ X w′(b). Let g′ ∈ G(L) such that Q′ = g′I and g′−1bσ(g′) ∈ IsUas(x0) w for some x0 ∈ k. Here as ∈ Φ+ is the positive affine root corresponding to s and Is = T (k[[t]])Qas6=a∈ Φ+ Ua(k). Define φ : A1 → G(L) by x 7→ g′U−αs(x)I. Set U = A1 if x0 = 0 and U = A1 − {−x−1 0 } otherwise. One checks that φ(U) ⊆ X w′(b), Q′ = φ(0) and P ′ := φ(∞) = g′sI. Then P ′ ∈ X w′σ(s)(b) ∪ Xs w′σ(s)(b). Note that w′σ(s), s w′σ(s) < w′. So w′ is reducible and (a) is proved. By Theorem 3.3, to finish the proof, it suffices to show the following statement: Let s ∈ Sa such that w and w′ = s wσ(s) are of the same length. Then w is reducible if so is w′. 1) Without loss of generality, we assume s w < w < wσ(s). By Lemma 3.1, it suffices to show that σ( w) is reducible. Let Q1 ∈ Xσ( w)(b). By σ(s) Lemma 3.4(2), there exists unique Q′ −→ σ(Q′ ℓ(s w) + ℓ(σ(s)), we have σ(Q′ 1) is, Q′ σ(s w) −→ bσ(Q1). Note that bσ(Q1) σ(s wσ(s)) 1 ∈ G(L)/I such that Q1 σ2(s) → bσ2(Q′ 1) and ℓ(s wσ(s)) = 1) by Lemma 3.4(1), that → bσ2(Q′ 1 ∈ X w′(σ−1(b)). By assumption, there exists φ′ : P1 → G(L)/I and an open subset 0 ∈ U ⊆ A1 such that φ′(U) ⊆ X w′(σ−1(b)), Q′ 1 := φ′(∞) ∈ Xu′(σ−1(b)) for some u′ < w′. Using Lemma 3.4(2), we can define a morphism φ : U → G(L)/I such that φ′ w′σ(s) σ(s) −→ σ−1(b)σ(φ′). One checks that φ −→ bσ(φ). So φ(U) ⊆ Xσ( w)(b) and φ(0) = Q1. Now we extend φ to P1 ⊇ U, which is still denoted by φ. Let u ∈ W such that P1 := φ(∞) ∈ Xσ(u)(b). By Lemma 3.4(3), u 6 w. We have to show that u < w. Assume otherwise, that 1 = φ′(0) and P ′ −→ σ−1(b)φ σ(s) −→ σ(φ′) σ( w′σ(s)) is, u = w and P1 (b) P1 σ(s) −→ σ(P ′ σ( w) −→ bσ(P1). We claim 1). 9 Note that φ(x) σ(s) −→ σ(φ′(x)) for x ∈ U. So either P1 σ(s) −→ σ(P ′ 1) If the latter case occurs, then 1) ∈ Xσ(u′)(b) and hence u′ = w, which is impossible because or P1 = σ(P ′ P1 = σ(P ′ ℓ(σ(u′)) < ℓ( w′) = ℓ( w). Therefore, (b) is proved. 1) by Lemma 3.4(3). Now we have (∗) P1 σ(s) −→ σ(P ′ 1) σ(u′) −→ bσ2(P ′ 1) σ2(s) −→ bσ(P1). Therefore, ℓ(u′) > ℓ( w′)−2 by Lemma 3.4(1). If ℓ(u′) = ℓ( w′)−2, then w = su′σ(s) and hence wσ(s) < w, which contradicts our assumption on w. So ℓ(u′) = ℓ( w′) − 1 and w = su′ or w = u′σ(s). In the former case, we have ℓ(su′σ(s)) = ℓ(u′) + 2, which means, by (∗) and Lemma 3.4(1), w = su′σ(s), a contradiction because ℓ(su′σ(s)) 6= ℓ( w). In the latter case, we have u′ = wσ(s) and ℓ(u′) > ℓ( w′), which is also impossible. Therefore σ( w) is reducible and the proof is finished. (cid:3) Let v ∈ YR. Denote by Mv ⊆ G the Levi subgroup (over k[[t]]) generated by T and the root subgroups Uα with hα, vi = 0. Denote by Nv ⊆ G the unipotent subgroup (over k[[t]]) generated by the root subgroups Uβ with hβ, vi > 0. We put IMv = Mv(L) ∩ I and INv = Nv(L) ∩ I. Let w = tµw ∈ W with µ ∈ Y and w ∈ W0. We say w is a (v, σ)-alcove element if wσ(v) = v and wσINv ⊆ INv. We say w is (v, σ)-fundamental if w is a (v, σ)-alcove element and wσIMv = IMv. Theorem 3.5 ([4], [5]). Let v ∈ YR and w ∈ W such that w is a (v, σ)-alcove elements. Then the inclusion IMv wσ(IMv) ֒→ I wI induces a bijection between IMv-σ-conjugacy classes in IMv wσ(IMv) and I-σ- conjugacy classes in I wI. Theorem 3.6 ([4], [6], [12]). Let w ∈ W . The following are equivalent: (a) w is (ν w, σ)-fundamental; (b) w is σ-fundamental, that is, I wI lies in a single I-σ-conjugacy class; (c) w is σ-straight, that is, (I wσI)n = I( wσ)nI for n ∈ N, or equiv- alently, ℓ( w) = 2hρ, ¯ν wi. Here ρ is the half sum of positive roots. Proposition 3.7. Let b ∈ G(L) and let C ⊆ W be closed under the Bruhat order. Then any connected component of ∪u∈CXu(b) intersects with Xx(b) for some σ-straight element x ∈ C. Proof. Let w ∈ C and Q ∈ X w(b). We show the connected component of ∪u∈CXu(b) containing Q intersects with Xx(b) for some σ-straight element x ∈ C. Since C is closed under the Bruhat order, we may assume, by Lemma 3.2, that w is of minimal length in its σ-conjugacy class. Thanks to [6, Theorem 3.5], I wI lies in a single σ-conjugacy class of G(L). Therefore, w is a (ν w, σ)-alcove element by [12, Theorem 5.2]. 10 SIAN NIE M with WJ finite, x ∈ W which is (ν w, σ)- Moreover, there exist J ⊆ Sa fundamental such that xσ(J)x−1 = J and w = ux for some u ∈ WJ . M is the set of simple affine refections in WM Here M = Mν w and Sa defined in §2.3. Write Q = gI for some g ∈ G(L). Then g−1bσ(g) ∈ I wI. By The- orem 3.5, we can assume g−1bσ(g) ∈ IM wσ(IM ) since w is a (ν w, σ)- alcove element. Let PJ ⊆ M(L) be the standard parahoric subgroup of type J and let UJ ⊆ PJ be its pro-unipotent radical. Since xσIM = IM and xσ(J)x−1 = J, we see that Lang's map ψ : PJ → PJ define by f 7→ f −1 xσ(f ) x−1 is surjective. So there exists h0 ∈ PJ such that h−1 xσ(h0) = g−1bσ(g), which means g = jh0 for some j ∈ G(L) satis- 0 fying j−1bσ(j) = x. For w ∈ WJ , we define X PJ w ( x) = {hIM ∈ PJ /IM ; h−1 xσ(h) x−1 ∈ IM wIM } ≤M u( x) = ∪w∈WJ ,w≤M uX PJ and X PJ w ( x). Here ≤M is the Bruhat order on WM defined in §2.3. Then the map hIM 7→ jhI induces a natural morphism w,b : X PJ ≤M u( x) → ∪u∈CXu(b). Since the image of ∪w∈WJ ,w≤M uIM wIM under the natural projection PJ → PJ/UJ is connected and ψ induces a flat and finite endomorphism on PJ /UJ , there exists h1 ∈ PJ such that h1IM ∈ X PJ id ( x) and h0IM ∈ u ( x) are in the same connected component of X PJ X PJ ≤M u( x). Therefore, via the morphism w,b, we deduce that Q = gI = jh0I and P = jh1I ∈ Xx(b) are connected in ∪u∈CXu(b). (cid:3) Remark. There is an alternative proof of Proposition 3.7 due to Xuhua He [7]. The main idea is to show the following two facts: (1) X w(b) is locally quasi-affine; (2) each irreducible component of X w(b) is of dimension > 1 if w is not σ-straight. We say b ∈ G(L) is basic if κG [b] is central on Φ. Corollary 3.8. Let λ ∈ Y + and b ∈ G(L) which is basic. Then (i) each connected component of X(cid:22)λ(b) intersects with Xµ(b), where µ ∈ Y + is the unique dominant minuscule cocharacter such that µ (cid:22) λ; (ii) Jb(F ) = {g ∈ G(L); g−1bσ(g) = b} acts transitively on π0(X(cid:22)λ(b)); (iii) if b is superbasic, then the natural projection X(cid:22)λ(b) → π1(G) induces an isomorphism π0(X(cid:22)λ(b)) ∼= (σ − 1)−1(η(tλ) − η(b)). Proof. Let π : G(L)/I → G(L)/K be the natural projection. Note that π−1(X(cid:22)λ(b)) = ∪u∈CXu(b), where C = ∪λ′(cid:22)λW0tλ′W0 is closed under the Bruhat order. By Lemma 3.7, each connected component of π−1(X(cid:22)λ(b)) intersects with Xx(b) for some σ-straight element x ∈ C. So, we have [ x] = [b] by [6, Theorem 3.7]. In particular, ¯νx = νG [b] and 11 hence ℓ(x) = 2hρ, νG of Ω ∩ C. [b]i = 0. Therefore, x must be the unique element Now (i) follows by noticing that x ∈ tµW0 and π(Xx(b)) ⊆ Xµ(b). Since x ∈ Ω, each element of I xI is σ-conjugate to x by I. In particular, the map g 7→ gI induces a surjective map: Jb, x(F ) = {g ∈ G(L); g−1bσ(g) = x} ։ Xx(b). Then (ii) follows by noticing that Jb(F ) acts (by left multiplication) on Jb, x(F ) transitively. Since [ x] = [b], we can assume b = x ∈ NT (L). Since x ∈ Ω is superbasic, Jb(F ) is generated by NT (L)bσ = Jb(F ) ∩ NT (L) and I bσ = Jb(F ) ∩ I. Moreover, by [6, Proposition 5.3], the image of NT (L)bσ under the natural projection p : NT (L) → W is W xσ = { w ∈ W ; xσ( w)x−1 = w} = { w ∈ Ω; σ( w) = w} ∼= π1(G)σ. So Jb(F ) = NT (L)bσI bσ and Jb(F )/I bσ = W xσ. By (ii), the map g 7→ gK induces a surjection π1(G)σ = W xσ = Jb(F )/I bσ → π0(X(cid:22)λ(b)). On the other hand, the map is obviously an injection and hence a bijection as desired. (cid:3) Corollary 3.9. Let λ ∈ Y +, b ∈ G(L) and J ⊆ S0 with σ(J) = J such that (J, b) is admissible. Assume b ∈ NT (L) ∩ [b]J,dom such that p(b) ∈ ΩJ . Then each connected component of X(cid:22)λ(b) contains a point µ (b) for some µ ∈ ¯Iλ,J,b. Here jQ, where j ∈ Jb(F ) ∩ N(L) and Q ∈ X MJ N =Qα∈Φ+−ΦJ Uα. Proof. By [2, Proposition 3.4.1], each connected component of X(cid:22)λ(b) contains a point jQ′, where j ∈ Jb(F )∩N(L) and Q′ ∈ X MJ µ′ (b) for some µ′ (cid:22) λ. Applying Corollary 3.8 (i), there exist a J-minuscule and J- µ (b) such that Q′ ∼J,µ′,b Q. dominant cocharacter µ (cid:22)J µ′ and Q ∈ X MJ Hence we have jQ′ ∼λ,b jQ. Since X MJ µ (b) 6= ∅, we have κJ (tµ) = κJ (b). So µ ∈ ¯Iλ,J,b and the proof is finished. (cid:3) 4. The choice of J Let λ ∈ Y + and b ∈ G(L). In view of Corollary 3.9, we are faced with the question of the existence of J ⊆ S0 such that (a) σ(J) = J, (J, b) is admissible and b is superbasic in MJ . It is known such J always exists. However, due to technical reasons (see Lemma 8.4), we need to add the following additional requirement. (b) ¯Iλ,J,b contains a weakly dominant cocharacter. Here we say a cocharacter µ ∈ Y is weakly dominant if hα, µi > −1 for any α ∈ Φ+. 12 SIAN NIE The main purpose of this section is to show the existence of J that satisfies (a) and (b), using the different characterizations of σ-straight elements in Theorem 3.6. Notice that if λ is minuscule, then each element ¯Iλ,J,b is minuscule and (b) is a priori satisfied. For v ∈ Y + Q , we define Jv = {s ∈ S0; s(v) = v}. Let w ∈ W be a σ-straight element. We set w⋄ = z wσ(z), where z ∈ J¯ν w W0 is the unique element such that z(ν w) = ¯ν w. Lemma 4.1. Let w ∈ W be σ-straight. Then (a) w⋄ ∈ ΩJ¯ν w (b) w⋄ ∈ tχWJ¯ν w with σ(J¯ν w) = J¯ν w; with χ ∈ Y weakly dominant. , that is, = IMν w Proof. (a) The equality σ(J¯ν w) = J¯ν w follows immediately from the fact that ¯ν w is dominant. Note that zIMν w . By Theorem 3.6(a), we w⋄σIM¯ν w have wσIMν w . So w⋄ ∈ ΩJ¯ν w = IM¯ν w = IM¯ν w (b) We show that hα, χi > −1 for any α ∈ Φ+. If α ∈ ΦJ¯ν w , then hα, χi ∈ {0, 1} since w⋄ ∈ ΩJ¯ν w . Then hz−1σ−1(u)−1(α), ν wi = hσ−1(u)−1(α), ¯ν wi = hα, ¯ν wi > 0, where u = t−χ w⋄ ∈ WJ¯ν w and z ∈ J¯ν w W0 is the unique element such that z(ν w) = wσ(Uz−1σ−1(u−1)(α)(L) ∩ I) ⊆ I, ¯ν w. Since w is a (ν w, σ)-alcove element, which implies hz−1(α), z−1(χ)i = hα, χi > −1 and the proof is finished. (cid:3) . Assume α ∈ Φ+ − ΦJ¯ν w . Lemma 4.2. Let w ∈ W be σ-straight. Then there exist J ⊆ J¯ν w and z0 ∈ WJ¯ν w such that (a) σ(J) = J and w0 = z0 w⋄σ(z0)−1 is σ-superbasic in WJ ; (b) w0 ∈ tχ0WJ with χ0 ∈ Y weakly dominant; (c) ν w0 = ν w⋄ = ¯ν w. Proof. Let V w⋄ = {q ∈ YQ; w⋄σ(q) = q + ¯ν w}, which is a nonempty affine subspace. Following [9, Section 3], we set V ′ w⋄ = {q − e; q ∈ V w⋄}, where e is an arbitrary point of V w⋄. Then V ′ w⋄ is the (linear) subspace parallel to V w⋄. We choose a point v0 ∈ V ′ w⋄ lying in a sufficiently small neighborhood of ν w⋄ = ¯ν w which is also generic in V ′ w⋄, that is, for any root α ∈ Φ, hα, v0i = 0 implies that hα, v′i = 0 for each v′ ∈ V ′ w⋄. Let J = J¯v0 and let z0 be the unique element in J¯v0 W0 such that z0(v0) = ¯v0. Since v0 lies in a sufficient small neighborhood of the dominant vector ¯ν w, z0(¯ν w) is also dominant. So z0(¯ν w) = ¯ν w and z0 ∈ WJ¯ν w . Then (a) follows from [9, Lemma 3.1]. (a) Note that w⋄ ∈ ΩJ¯ν w (b) Assume w⋄ ∈ tχW0 for some χ ∈ Y . Then w0 ∈ tz0(χ)W0. By Lemma 4.1, χ is weakly dominant and J¯ν w-minuscule. So χ0 = z0(χ) is weakly dominant as z0 ∈ WJ¯ν w . . (c) It follows directly from the inclusion z0 ∈ WJ¯ν w . (cid:3) 13 Corollary 4.3. Let λ ∈ Y + and b ∈ G(L). If X(cid:22)λ(b) 6= ∅. Then there exists J ⊆ S0 with σ(J) = J such that (i) (J, b) is admissible and b is superbasic in MJ ; (ii) ¯Iλ,J,b contains a weakly dominant coweight. Proof. Since X(cid:22)λ(b) 6= ∅, there exists u ∈ W0tλ′W0 with λ′ (cid:22) λ such that Xu(b) 6= ∅. By the proof of [12, Proposition 1.5], there exists a σ-straight element w ∈ W such that w 6 u and [ w] = [b]. Let J ⊆ S0 and w0 ∈ ΩJ be defined as in Lemma 4.2 (with respect to w). Then [b]. So [ w0]J = [b]dom,J and w0 is σ-superbasic in WJ with ν w0 = ¯ν w = νG (i) follows. Let χ0 ∈ Y be such that w0 ∈ tµWJ , which is J-dominant, J-minuscule and weakly dominant by Lemma 4.2. Moreover, χ0 (cid:22) λ since W0 w0W0 = W0 wW0 and w 6 u. So χ0 ∈ ¯Iλ,J,b and (ii) follows. (cid:3) 5. Proof of Theorem 1.1 The aim of this section is to prove Theorem 1.1. The strategy is the same as in [2, §4.1]. We rephrase it here for completeness. Throughout the section, we assume G is adjoint and simple. We fix λ ∈ Y +, b ∈ G(L) such that X(cid:22)λ(b) 6= ∅. 5.1. In this subsection, we fix J ⊆ S0 such that σ(J) = J, (J, b) is admissible and b is superbasic in MJ . By Lemma 2.1, we can assume: [b] and p(b) ∈ ΩJ is (A) b ∈ NT (L) ∩ [b]J,dom such that νp(b) = νG σ-superbasic in WJ . By lifting the natural projection N MJ T (A), Ωσ J MJ b J = {x ∈ ΩJ ; σ(x) = x} = π1(MJ )σ as a subset of N MJ T is the normalizer of T in MJ . (L), where N MJ (F ) ∩ N MJ (L) → WJ , we also view, by (L)bσ = T T Proposition 5.1. (cf. [2, Theorem 4.1.12]) The image of the natural map π0(X MJ µx (b)) → π0(X(cid:22)λ(b)) induced by the embedding X MJ choice of x ∈ ¯Iλ,J.b. µx (b) ֒→ X(cid:22)λ(b) dose not depend on the The proof will be given in Section 7. For α ∈ Φ, we set yα = Pβ∈Oα β∨ ∈ Y /ZΦ∨ J = π1(MJ )σ. Here Oα = {σk(α); k ∈ Z}. Let Aλ,J,b,x be the subset of Φ such that α ∈ Aλ,J,b,x if and only if there exist Q, P ∈ X MJ µx (b) satisfying Q ∼λ,b P and ηJ (Q) − ηJ (P ) = yα. Lemma 5.2. Let x ∈ ¯Iλ,J,b. Then J = Ωσ (i) Aλ,J,b,x = −Aλ,J,b,x; (ii) Aλ,J,b,x′ = Aλ,J,b,x for any x′ ∈ ¯Iλ,J,b; (iii) If α ∈ Aλ,J,b,x, then for any Q ∈ X MJ µx (b) such that Q ∼λ,b P and ηJ (Q) − ηJ (P ) = yα ∈ π1(MJ )σ. µx (b), there exists P ∈ X MJ 14 SIAN NIE Proof. (i) it follows directly by definition. (ii) By symmetry, it suffices to show Aλ,J,b,x ⊆ Aλ,J,b,x′. Let α ∈ Aλ,J,b,x. Then there exist Q, P ∈ X MJ µx (b) such that Q ∼λ,b P and ηJ (Q) − ηJ (P ) = yα ∈ π1(MJ )σ. By Proposition 5.1, there exists Q′ ∈ X MJ µx′ (b) such that Q′ ∼λ,b Q. Note that y−1 α Q ∼J,µx,b P by Corollary 3.8 (iii). Thus Q′ ∼λ,b Q ∼λ,b P ∼J,λ,b y−1 α Q′ and the proof is finished. α Q ∼λ,b y−1 (iii) It follows from the fact that the action of J MJ b is transitive, see Corollary 3.8 (ii). (F ) on π0(X MJ µx (b)) (cid:3) Thanks to Lemma 5.2, we can define Aλ,J,b = Aλ,J,b,x, which is inde- pendent of the choice x ∈ ¯Iλ,J,b. Let Lλ,b,J ⊆ π1(MJ )σ be the sublattice spanned by yα for α ∈ Aλ,J,b}. Proposition 5.3. Assume furthermore that (λ, b) is HN-irreducible (see Introduction) and ¯Iλ,J,b contains a weakly dominant cocharacter. Then Lλ,b,J is the kernel of the natural surjective projection π1(MJ )σ ։ π1(G)σ. The proof will be given in Section 8. 5.2. Now we are ready to prove our main result. Proof of Theorem 1.1. By Corollary 4.3, there exists J with σ(J) = J such that (J, b) is admissible, b is superbasic in MJ (L) and ¯Iλ,J,b contains a weakly dominant cocharacter. Moreover, we can assume b = bx for some x ∈ ¯Iλ,J,b. So §5.1 (A) holds. Since ηG(b) = ηG(tλ), it suffices to show the map g 7→ ηG(g) induces a bijection π0(X(cid:22)λ(b)) ∼= π1(G)σ. By Corollary 3.8 (iii), the map g 7→ gKJ induces a π1(MJ )σ-equivariant surjection π1(MJ )σ ∼= J MJ b (F )/(J MJ b (F ) ∩ IMJ ) ։ π0(X MJ µx (b)). Let C be the image of the natural map π0(X MJ Then we obtain a natural π1(MJ )σ-equivariant map µx (b)) → π0(X(cid:22)λ(b)). ϕx : π1(MJ )σ ։ π0(X MJ µx (b)) ։ C. Since π1(MJ )σ is an abelian group, the stabilizer stab(ϕx) := ϕ−1 independent of the choice of c ∈ C. We claim that x (c) is (i) stab(ϕx) = ker(π1(MJ )σ ։ π1(G)σ). (ii) C = π0(X(cid:22)λ(b)). Assuming (i) and (ii), we see that ϕx induces a bijection from π1(G)σ = π1(MJ )σ/stab(ϕx) to π0(X(cid:22)λ(b)), as desired. We prove (i) first. Let Q ∈ X MJ µx (b). By Proposition 5.3, it suffices to show Q ∼λ,b yαQ for each α ∈ Aλ,J,b = Aλ,J,b,x. Applying Lemma 5.2 (iii), there exist P ∈ X MJ µx (b) such that Q ∼λ,b P and ηJ (Q) − ηJ (P ) = 15 −yα ∈ π1(MJ )σ. On the other hand, since ηJ (P ) = ηJ (yαQ), we have yαQ ∼J,µx,b P by Corollary 3.8 (iii). So Q ∼λ,b yαQ and (i) is proved. Now we show (ii). By Corollary 3.9 and Proposition 5.1, ϕx is sur- jective modulo the action of Jb(F ) ∩ N(L) on π0(X(cid:22)λ(b)). So it suffices to show Q ∼λ,b jQ for any j ∈ Jb(F ) ∩ N(L) and any Q ∈ X MJ µx (b). Choose δ ∈ ZΦ∨ (depending on j and Q) such that δ is central on ΦJ , σ(δ) = δ and tδjt−δQ = Q. Note that δ ∈ ker and tδ ∈ J MJ (F ). Then we have tδQ ∼λ,b Q by (i). Therefore, b tδjQ = tδjt−δtδQ ∼λ,b tδjt−δQ = Q ∼λ,J tδQ, which means jQ ∼λ,b Q as desired. (cid:3) 6. The set Θ(µ, µ′, λ) This section is devoted to the study of the set Θ(µ, µ′, λ) define below, which plays a crucial role in the proof of Proposition 5.1. Let µ, µ′ ∈ Y and λ ∈ Y + such that µ, µ′ (cid:22) λ. We Define Θ(µ, µ′, λ) = {α ∈ Φ; hα, µ − µ′i > 2, µ − α∨, µ′ + α∨ (cid:22) λ}; Ξ(µ, µ′) = {α ∈ Φ; hα, µi, −hα, µ′i > 1}; Ξ1(µ, µ′) = {α ∈ Φ; hα, µi = −hα, µ′i = 1}. Lemma 6.1. We have the following properties: (a) Ξ(µ, µ′) ⊆ Θ(µ, µ′, λ); (b) Θ(µ, µ′, λ) = −Θ(µ′, µ, λ); (c) w(Θ(µ, µ′, λ)) = Θ(w(µ), w(µ′), λ), w(Ξ(µ, µ′)) = Ξ(w(µ), w(µ′)) and w(Ξ1(µ, µ′)) = Ξ1(w(µ), w(µ′)), where w ∈ W0. Lemma 6.2. If Ξ(µ, µ′) = ∅, then there exists w ∈ W0 such that w(µ), w(µ′) are dominant. Lemma 6.3. Let δ, δ′ ∈ NΠ∨ such that 0 . Then there exists an index set Λ of roots Proof. We argue by induction on the partial order 6 on NΠ∨ k ; (3) hγi, γ∨ j i > 0 for any i, j ∈ Λ. k 6 δ′ and −Pk∈Σ γ∨ (1) δ′ − δ =Pk∈Λ γ∨ (2) Pk∈Σ γ∨ Lemma 4.2.5], we can write δ = −Pi∈Γ α∨ that αi ∈ Φ− (resp. βi′ ∈ Φ+) and hαi, α∨ any i, j ∈ Γ (resp. i′, j′ ∈ Γ′). If hαi, β∨ we may take Λ = Γ⊔Γ′. Otherwise, there exists i0 ∈ Γ and i′ that hαi0, β∨ i′ 0 i 6 −1. Then θ∨ := α∨ i0 + β∨ i′ 0 of generality, we may assume θ∨ > 0. Set δ1 = −Pi∈Γ\{i0} α∨ 1 = θ∨ +Pj ′∈Γ′\{i′ δ′ now follows from the induction hypothesis on the pair (δ1, δ′ j ′ < δ′. Note that δ′ 0} β∨ k 6 δ for any subset Σ ⊆ Λ; j i > 0 (resp. hβi′, β∨ i (resp. δ′ =Pj∈Γ′ β∨ 0 . By [2, j ) such j ′i > 0) for i′ i > 0 for any i ∈ Γ and i′ ∈ Γ′, 0 ∈ Γ′ such ∈ Φ∨ ∪ {0}. Without loss j < δ and 1 − δ1 = δ′ − δ. The statement (cid:3) 1). 16 SIAN NIE Lemma 6.4. Let µ ∈ Y , λ ∈ Y + and α ∈ Φ. If µ + α∨ 6 λ and µ + α∨ (cid:14) λ, then there exists γ ∈ Φ+ such that hγ, µ + α∨i 6 −2. If, moreover, µ ∈ Y +, then hγ, µxi 6 1, hγ, α∨i 6 −2 and either (a) γ = −α; or (b) γ is a long root of Φ and µ + α∨ + γ∨ 6 λ. Proof. Set µ0 = µ + α∨. We define Φ+(µ0) = {θ ∈ Φ+; hθ, µ0i 6 −2}. Let w ∈ W0 be the unique minimal element satisfying w(µ0) = ¯µ0. Let w = sr · · · s1 be a reduced expression of w with each si ∈ S0. Let µk = sk · · · s1(µ0). Note that hαk, µk−1i 6 −1 and hence µk−1+α∨ k 6 µk for k ∈ [1, r], where αk denotes the simple root corresponding to sk. Since µ0 6 λ and µr = ¯µ0 (cid:10) λ, there exists i ∈ [1, r] such that µi−1 6 λ and µi = si(µi−1) (cid:10) λ. Write λ = µi−1 +Pα∈Π nαα∨, where each nβ ∈ N. We have nαi > 1 since hαi, λi > 0. So µi−1 + α∨ i 6 λ. If hαi, µi−1i = −1, then µi = µi−1 + α∨ i 6 λ, which contradicts the choice of i. Therefore, hαi, µi−1i 6 −2, that is, γ := s1 · · · si−1(αi) ∈ Φ+(µ0). Assume, moreover, µ is dominant. We have hγ, µi 6 1 and hγ, α∨i 6 −2 because hγ, α∨i > −3. If γ 6= −α, then γ and αi are long roots of Φ. Therefore, µ + α∨ + γ∨ = µ0 + γ∨ 6 µ0 + α∨ 6 µ1 + α∨ 6 · · · 6 µi−1 + α∨ 1 + · · · + α∨ 2 + · · · + α∨ i 6 λ. i i The proof is finished. (cid:3) Lemma 6.5. [3, Proposition 2.2] Let µ ∈ Y and λ ∈ Y +. If µ 6 λ and µ is weakly dominant, then µ (cid:22) λ. In particular, If µ ∈ Y + and γ ∈ Φ+ is a positive long root such that µ + γ∨ 6 λ, then µ + γ∨ (cid:22) λ. Lemma 6.6. Let µ ∈ Y be a weakly dominant coweight and α ∈ Π0 such that hα, µi > 1 (resp. hα, µi = −1). Then µ − α∨ (resp. µ + α∨) is weakly dominant. Proof. Assume hα, µi > 1. Let β ∈ Φ+. If β = α, then hα, µ − α∨i > −1. Otherwise, we have hβ, µi > −1 and hβ, sα(µ)i = hsα(β), µi > −1 since sα(β) ∈ Φ+ and µ is weakly dominant. Note that µ − α∨ lies in the convex hull of µ and sα(µ), which means hβ, µ − α∨i = −1. Therefore, µ − α∨ is weakly dominant. The case that hα, µi = −1 can be handled similarly. (cid:3) Lemma 6.7. Let µ 6= µ′, λ ∈ Y + with µ, µ′ (cid:22) λ. Then Θ(µ, µ′, λ) 6= ∅. Proof. Let δ, δ′ ∈ NΠ∨ 0 such that λ = µ + δ = µ′ + δ′. We write k , where Λ is as in Lemma 6.3. Since µ 6= µ′, Λ is nonempty. Choose k0 ∈ Λ and set α = γk0. Then hα, δ′ − δi = k i > 2 and µ − α∨, µ′ + α∨ 6 λ. If µ − α∨, µ′ + α∨ (cid:22) λ, the µ − µ′ = δ′ − δ =Pk∈Λ γ∨ hα,Pk∈Λ γ∨ 17 proof is finished. Otherwise, by symmetry, we may assume µ′ +α∨ (cid:14) λ. Applying Lemma 6.4, there exists γ ∈ Φ+ such that hγ, µ′ + α∨i 6 −2. In particular, hγ, α∨i 6 −2 and hγ, µ′i 6 1. We claim that γ 6= −α. Otherwise, we have 2 6 hα, µ − µ′i = −hγ, µi + hγ, µ′i 6 0 + 1 = 1, which is a contradiction and the claim is proved. Therefore, by Lemma 6.4 (b), γ is a long coroot of Φ and µ + α∨ + γ∨ 6 λ. Define β = sα(γ), which is again a long root. We show that β ∈ Θ(µ, µ′, λ). First note that hβ, µ − µ′i = hγ − hγ, α∨iα, µ − µ′i > hγ, µi − hγ, µ′i + 2hα, µ − µ′i > 0 − 1 + 4 = 3. It remains to show µ − β∨, µ′ + β∨ (cid:22) λ. Note that µ′ + β∨ = µ′ + α∨ + γ∨ 6 λ and µ − β∨ 6 µ − α∨ 6 λ. Thanks to Lemma 6.5, it suffices to show µ − β∨, µ′ + β∨ are weakly dominant. Let γ′ ∈ Φ+ − {±β}, then hγ′, µ′ − β∨i > 0 − 1 = −1 since β is a long root. Similarly, hγ′, µ+β∨i > −1. Without loss of generality, we may assume −β ∈ Φ+, the other case can be handled similarly. Then h−β, µ − β∨i > 0 + 2 = 2 and h−β, µ′ + β∨i = h−β, µ + β∨i + h−β, µ′ − µi > −2 + 3 = 1. So µ − β∨, µ′ + β∨ are weakly dominant as desired. Therefore, β ∈ Θ(µ, µ′, λ). (cid:3) Proposition 6.8. Let µ 6= µ′ ∈ Y and λ ∈ Y + such that µ, µ′ (cid:22) λ. Then Θ(µ, µ′, λ) 6= ∅. Proof. If Ξ(µ, µ′) 6= ∅, then Θ(µ, µ′, λ) ⊇ Ξ(µ, µ′) is nonempty. Other- wise, by Lemma 6.2, there exists w ∈ W0 such that w(µ), w(µ′) ∈ Y +. Then by Lemma 6.7, Θ(µ, µ′, λ) = w−1(Θ(w(µ), w(µ′), λ)) is again nonempty. (cid:3) Let D ⊆ Φ be a subset. We say D is orthogonal if hα, β∨i = 0 for any α 6= β ∈ Dµ. Lemma 6.9. Assume Φ is simply laced. Let µ ∈ Y be weakly dominant. Then there exists an orthogonal subset Dµ ⊆ Φ+ such that (a) hγ, µi = −1 for any γ ∈ Dµ; (b) ¯µ − µ =Pα∈Dµ α∨. Proof. We argue by induction on ¯µ − µ ∈ NΠ∨ 0 . If ¯µ − µ = 0, then Dµ = ∅ and there is nothing to prove. Otherwise, there exists a simple root α ∈ Π0 such that hα, µi 6 −1, and hence hα, µi = −1 since µ is weakly dominant. Let υ = sα(µ) = µ + α∨, which is also weakly dominant by Lemma 6.6. Noticing that ¯υ − υ < ¯µ − µ. So there exists, by induction hypothesis, an orthogonal subset Dυ ⊆ Φ+ with properties (a) and (b). Let Dυ,− = {β ∈ Dυ; hβ, µi = −1} = {β ∈ Dυ; hβ, α∨i = 0} 18 and SIAN NIE Dυ,0 = {β ∈ Dυ; hβ, µi = 0} = {β ∈ Dυ; hβ, α∨i = −1}. Since µ is weakly dominant and Φ is simply laced, we have Dυ = β ∈ Φ+. (cid:3) Dυ,− ∪ Dυ,0. By the orthogonality of Dυ, γ := α +Pβ∈Dυ,0 Now we set Dµ = Dυ,− ∪ {γ}, which satisfies (a) and (b). Lemma 6.10. Assume Φ is simply laced. Let µ, υ, λ ∈ Y with µ, υ (cid:22) λ. If Θ(µ, υ, λ) = Ξ1(µ, υ), then µ and υ are conjugate under W0. Moreover, there exists an orthogonal subset ∆µ,υ ⊆ Ξ1(µ, υ) such that µ − υ =Pβ∈∆µ,υ β∨. Proof. By Lemma 6.1, the lemma dose not change if we replace the pair (µ, υ) with (w(µ), w(υ)) for any w ∈ W0. So we can assume µ ∈ Y + and {α ∈ Φ+; hα, υi 6 −1} ⊆ Ξ(µ, υ). On the other hand, we have, by assumption, Ξ(µ, υ) ⊆ Θ(µ, υ, λ) = Ξ1(µ, υ). Therefore, {α ∈ Φ+; hα, υi 6 −1} = Ξ1(µ, υ). In particular, υ is weakly dominant. We have to show µ = ¯υ. Assume otherwise. By lemma 6.9, there exists an orthogonal subset Dυ ⊆ {α ∈ Φ+; hα, υi = −1} = Ξ1(µ, υ) such that ¯υ − υ =Pβ∈Dυ β∨. By Proposition 6.8, Θ(µ, ¯υ, λ) 6= ∅. Choose γ ∈ Θ(µ, ¯υ, λ). We claim (a) ±γ /∈ Dυ ⊆ Ξ1(µ, υ). Indeed, if −γ ∈ Dυ, one computes that h−γ, µ − υi = h−γ, µ − ¯υi + h−γ, ¯υ − υi = h−γ, µ − ¯υi + h−γ, Xβ∈Dυ β∨i = h−γ, µ − ¯υi + 2 6 −2 + 2 = 0, contradicting the inclusion γ ∈ Ξ1(µ, υ). If γ ∈ Dυ, one computes similarly that hγ, µ−υi > 2+2 = 4, which also contradicts the inclusion γ ∈ Ξ1(µ, υ). So (a) is proved. Thanks to (a) and that Φ is simply laced, we have Dυ,γ := {β ∈ Dυ; hγ, β∨i 6 −1} = {β ∈ Dυ; hγ, β∨i = −1}. Moreover, by the orthogonality of Dυ, we have α := γ +Pβ∈Dυ,γ β ∈ Φ. Note that hβ, α∨i > 0 for β ∈ Dυ − Dυ,γ and hβ′, α∨i = 1 for β′ ∈ Dυ,γ. We claim that (b) υ + α∨, µ − α∨ (cid:22) λ. We number the roots in Dυ − Dυ,γ as β1, β2, . . . , βr for some r ∈ N. k . Note that υr = ¯υ + γ∨ (cid:22) λ since γ ∈ Θ(µ, ¯υ, λ). To show υ + α∨ (cid:22) λ, it suffices to show υi−1 (cid:22) υi for i ∈ [1, r], which follows from the observation: hβi, υii = hβi, υ + α∨ + We set υi = υ + α∨ +Pi k=1 β∨ 19 β∨ i i > −1 + 0 + 2 = 1. Now we show µ − α∨ (cid:22) λ. Similarly, we number ∨. the roots in Dυ,γ as β′ Note that µr′ = µ − α∨ and µ0 = µ − γ∨ (cid:22) λ. Now it suffices to show µi−1 (cid:23) µi for i ∈ [1, r′], which follows from the observation: hβ′ r′ and set µi = µ − γ∨ −Pi i, µ − γ∨i = 2. The claim (b) is proved. 1, β′ 2, . . . , β′ k=1 β′ k i, µi−1i = hβ′ Now one computes that (c) hα, µ − υi = hα, µ − ¯υi + hα, ¯υ − υi = hγ, µ − ¯υi + ♯Dυ,γ + Xβ∈Dυ−Dυ,γ hα, β∨i, where the second equality follow from the fact that hβ, µi = hβ, ¯υi = hβ′, α∨i = 1 for any β ∈ Dυ and any β′ ∈ Dυ,γ. So hα, µ − υi > hγ, µ − ¯υi > 2 and α ∈ Θ(µ, υ, λ) = Ξ1(µ, υ). Thus hα, υi = −1 and hα, µ − υi = 2. By (c), we deduce that hβ, γ∨i = 0 for any β ∈ Dυ. Therefore, α = γ and hγ, µi = −hγ, υi = −hγ, ¯υi = 1, which contradicts that µ, ¯υ ∈ Y +. Therefore, we must have µ = ¯υ. Now the "Moreover" part follows from Lemma 6.9 by noticing that υ is weakly dominant and that {α ∈ Φ+; hα, υi 6 −1} = Ξ1(µ, υ). (cid:3) 7. Proof of Proposition 5.1 The aim of this section is to prove Proposition 5.1. Through out this section, we assume G is adjoint and simple; the root system Φ of G has h connected components, on which σ acts transitively. Lemma 7.1. Let x, x′ ∈ ¯Iλ,J,b. Then there exist some WJ -conjugates µ, µ′ ∈ Y of µx and µx′ respectively such that µ − µ′ ∈ (1 − σ)ZΦ∨. Proof. The proof is the same as that of [2, Proposition 4.3.1], noticing that we only use the property that µx, µx′ are J-minuscule. (cid:3) Let α ∈ Φ, we denote by Oα the σ-orbit of α. Lemma 7.2. [2, Lemma 4.2.1 & Example 4.2.2] We have the following properties: (i) Let α ∈ Φ and β 6= γ ∈ Oα, then hβ, γ∨i 6 0. Moreover, the equality holds unless each connected component of Φ is of type Am with m ∈ N even. (ii) Assume α, β are in the same connected component of Φ and hσh(α), α∨i = hσh(β), β∨i = −1. Then hα, β∨i 6= 0. [2, Proposition 4.3.2]) Let µ 6= υ ∈ Y and λ ∈ Y + Lemma 7.3. (cf. such that µ − υ ∈ (1 − σ)Y and µ, υ (cid:22) λ. Then there exist α ∈ Φ and r ∈ [1, h − 1] if ♯Oα = h (resp. r ∈ [1, h] if ♯Oα = 2h and r ∈ [1, 2h − 1] if ♯Oα = 3h) such that (a) σr(α) 6= α; (b) either µ, µ − α∨, µ + σr(α∨), µ − α∨ + σr(α∨) (cid:22) λ or υ, υ + α∨, υ − σr(α∨), υ + α∨ − σr(α∨) (cid:22) λ; (c) µ − υ − α∨ + σr(α∨) ≺ µ − υ. 20 SIAN NIE Proof. By Proposition 6.8, Θ(µ, υ, λ) 6= ∅. Let γ ∈ Θ(µ, υ, λ). since µ − υ ∈ (1 − σ)Y , we have P♯Oγ Eγ,µ,υ = {k ∈ [1, ♯Oγ]; hσk(γ), µ − υi 6 −1}. k=1hσk(γ), µ − υi = 0. Define Since hγ, µ − υi > 2, we have Eγ,µ,υ 6= ∅. Case(1): there exists γ ∈ Θ(µ, υ, λ), j ∈ Eγ,µ,υ such that hσj(γ), γ∨i = 0. Then σj(γ) 6= ±γ. If hσj(γ), υi > 1, then hσj(γ), υi = hσj(γ), υ + γ∨i > 1, which implies υ − σj(γ∨) (cid:22) υ (cid:22) λ and υ + γ∨ − σj(γ∨) (cid:22) υ + γ∨ (cid:22) λ. Otherwise, we have hσj(γ), µi 6 −1. Then hσj(γ), µi = hσj(γ), µ − γ∨i 6 −1 and we have µ + σj(γ∨) (cid:22) µ (cid:22) λ and µ − γ∨ + σj(γ∨) (cid:22) µ−γ∨ (cid:22) λ similarly. Moreover, note that hσj(γ), µ−υ−γ∨i = hσj(γ), µ − υi 6 −1 and hγ, µ − υi > 2. One deduces that µ − υ − γ∨ + σj(γ∨) (cid:22) µ − υ − γ∨ ≺ µ − υ. If ♯Oα = h, then Eγ,µ,υ ⊆ [1, h − 1], and we take α = γ and r = j ∈ [1, h − 1]. If ♯Oγ = 2h (resp. ♯Oγ = 3h). We set α = γ and r = j if j 6 h (resp. j 6 2h − 1) and set α = −σj(γ) and r = ♯Oγ − j otherwise. Now one checks that the lemma follows in this case. Case(2): Case(1) fails. By Lemma 7.2 (i), each connected component of Φ is of type Am with m even. Moreover, we have Eγ,µ,υ = {h}, ♯Oγ = 2h and hγ, σh(γ)i = −1 for any γ ∈ Θ(µ, υ, λ). We take r = h and choose α from Θ(µ, υ, λ). The precise choice of α depends on the following subcases. However, (a) always holds since hσh(γ), µ − υi 6 −hγ, µ − υi 6 −2 for any γ ∈ Θ(µ, υ, λ). Case(2.1): there exists some γ ∈ Θ(µ, υ, λ) − Ξ1(µ, υ). Note that hσh(γ), µ − υi 6 −2 since Eγ,µ,υ = {h}. So one of the following two cases occurs: (2.1.1): hσh(γ), µi 6 −2 or hσh(γ), υi > 2; (2.1.2): hσh(γ), µi = −hσh(γ), υi = −1, that is, −σh(γ) ∈ Ξ1(µ, υ). In (2.1.1), we take α = γ. Without loss of generality, we may assume hσh(α), υi > 2. Then hσh(α), υi, hσh(α), υ + α∨i > 1 and hence υ − σh(α∨), υ + α∨ − σh(α∨), υ + α∨ (cid:22) λ. Therefore, (b) holds. In (2.1.2), we take α = −σh(γ) ∈ Ξ1(µ, υ) ⊆ Θ(µ, υ, λ). Then either hσh(α), µi 6 −2 or hσh(α), υi > 2 since hσh(α), µ − υi 6 −2 and −σh(α) = γ /∈ Ξ1(µ, υ). Therefore, (b) follows similarly as in (2.1.1). Case (2.2): Θ(µ, υ, λ) = Ξ1(µ, υ). (2.2.1): hσi(γ), µ − υi > 1 for some γ ∈ Θ(µ, υ, λ) and some i ∈ [1, 2h − 1] − {h}. Then hσh(γ), µ − υi 6 −3, which implies either hσh(γ), µi 6 −2 or hσh(γ), υi > 2. Then (b) follows similarly as in (2.1.1) if we take α = γ. (2.2.2): hσk(γ), µ − υi = 0 for any γ ∈ Θ(µ, υ, λ) and any k ∈ [1, 2h − 1] − {h}. We show this case dose not occur. By Corollary 6.10, there exists an orthogonal subset ∆µ,υ ⊆ Ξ1(µ, υ) such that µ − υ = Pβ∈∆µ,υ β∨. Since 0 6= µ − υ ∈ (1 − σ)Y , we have ♯∆µ,υ > 2. Choose If γ1, γ2 are in the same connected component of γ1 6= γ2 ∈ ∆µ,υ. 21 Φ, it contradicts Lemma 7.2 (ii) in view of the assumption of Case (2). So each connected component of Φ contains at most one element of ∆µ,υ. Now let k0 ∈ [1, 2h − 1] − {h} such that σk0(γ1), γ2 are in the same connected component of Φ. By the assumption of (2.2.2), hσk0(γ1), γ∨ 2 i = hσk0(γ1), µ − υi = 0, which again contradicts Lemma 7.2 (ii). Finally, it remains to show (c) for Case (2). Since α ∈ Θ(µ, υ, λ) and Eα,µ,υ = {r} (with r = h), we have hσr(α), µ − υi 6 −2 and hence hσr(α), µ − υ − α∨i 6 −1 by (a). Therefore, µ − υ − α∨ + σr(α∨) (cid:22) µ − υ − α∨ ≺ µ − υ and the proof is finished. (cid:3) Corollary 7.4. Let µ, υ ∈ Y and λ ∈ Y + such that µ − υ ∈ (1 − σ)Y and µ, υ (cid:22) λ. Then there exist µj ∈ Y , αj ∈ Φ, and rj ∈ [1, h − 1] if ♯Oαj = h (resp. rj ∈ [1, h] if ♯Oαj = 2h and rj ∈ [1, 2h − 1] if ♯Oαj = 3h) such that µ0 = µ, µm = υ, µj+1 − µj = α∨ j − σrj (α∨ j − σrj (α∨ j ) (cid:22) λ µj, µj + α∨ j ) and j , µj − σrj (α∨ j ), µj + α∨ for j ∈ [0, m − 1]. Proof. We argue by induction on µ − υ. If µ − υ = 0, then µ = υ and there is nothing to prove. We assume the lemma holds for any pair (µ′, υ′) such that µ′, υ′ (cid:22) λ and µ′ − υ′ ≺ µ − υ. We show it also holds for the pair (µ, υ). Applying Lemma 7.3, there exists α ∈ Φ and r ∈ [1, h − 1] if ♯Oα = h (resp. r ∈ [1, h] if ♯Oα = 2h and r ∈ [1, 2h − 1] if ♯Oα = 3h) such that µ − υ + α∨ − σr(α∨) ≺ µ − υ and one of the following two statements holds: (a) µ, µ + α∨, µ − σr(α∨), µ + α∨ − σr(α∨) (cid:22) λ (b) υ, υ − α∨, υ + σr(α∨), υ − α∨ + σr(α∨) (cid:22) λ. We set η = µ + α∨ − σr(α∨) (resp. η = υ − α∨ + σr(α∨)) if (a) (resp. (b)) occurs. Then η (cid:22) λ and η − υ ≺ µ − υ (resp. µ − η ≺ µ − υ). The lemma then follows by induction hypothesis on the pair (η, υ) (resp. (µ, η)). (cid:3) Inspired by [2, Definition 4.4.8], we introduce the following notations, which play a crucial role throughout the paper. Let x, x′ ∈ π(MJ ). We write x (α,r) → x′ for some α ∈ Φ − ΦJ and some r ∈ N if x′ − x = α∨ − σr(α∨) and µx, µx+α∨, µx−σr(α∨), µx′ (cid:22) λ. We write x (α,r) ֌ x′ if x (α,r) → x′ and neither (α,i) → x + α∨ − σi(α∨) x (σi(α),r−i) → x′ nor (σi(α),r−i) → x + σi(α∨) − σr(α∨) (α,i) → x′ x for any i ∈ [1, r − 1]. 22 SIAN NIE (α,r) Lemma 7.5. Let x, x′ ∈ ¯Iλ,J,b such that x → x′ for some α ∈ Φ − ΦJ and some r ∈ N. Then there exist xj ∈ ¯Iλ,J,b, αj ∈ Oα and rj ∈ [1, r] such that x0 = x, xm = x′ and xj (αj ,rj) ֌ xj+1 for j ∈ [0, m − 1]. Moreover, we have Pr−1 i=0 σi(α∨) =Pm−1 j=0 Prj−1 Proof. It follows directly by induction on r. k=0 σk(α∨ j ). (cid:3) Lemma 7.6. (cf. two statements holds. [2, Lemma 4.4.1]) Let J ⊆ S0. Then the following (a) For β ∈ Φ − ΦJ , there exists a unique βJ ∈ Φ − ΦJ such that J − β∨ ∈ ZΦ∨ β∨ (b) Let λ ∈ Y + and µ ∈ Y be a J-dominant and J-minuscule J is J-anti-dominant and J-minuscule. J and β∨ J (cid:22) λ. coweight. If µ + β∨ (cid:22) λ, then µ + β∨ Proof. Denote by β′ J the unique J-anti-dominant conjugate of β under ∨ (cid:22) µ + β∨ (cid:22) λ since µ is J-dominant. If β′ ∨ is WJ . Then µ + β′ J J J-minuscule, we set βJ = β′ J and the proof is finished. Otherwise, ∨i 6 −2 for some γ ∈ Φ+ hγ, β′ J . Since β /∈ ΦJ , β 6= −γ and hence γ is J ∨i 6 1 − 2 = −1. (γ). Note that hγ, µ + β′ a long root in Φ. Let α = sβ ′ J ∨ + γ∨ (cid:22) λ. Set βJ = α′ We have µ + α∨ = µ + β′ J is J J-minuscule since α is a long root of Φ. Moreover, we have, we have µ + β∨ (cid:3) [2, Proposition 4.4.10]) Let x 6= x′ ∈ ¯Iλ,J,b. Then Lemma 7.7. (cf. there exist xj ∈ ¯Iλ,J,b, αj ∈ Φ − ΦJ and rj ∈ N for j ∈ [0, m − 1] such that J (cid:22) λ (since µ + α∨ (cid:22) λ) and β∨ J − β∨ ∈ ZΦ∨ J . Then β∨ J , as desired. J j is J-anti-dominant and J-minuscule; (1) α∨ (2) rj ∈ [1, h] if ♯Oαj ∈ {h, 2h} and rj ∈ [1, 2h − 1] if ♯Oαj = 3h; (3) x0 = x, xm = x′ and xj → xj+1. (αj ,rj) Proof. By Lemma 7.1, there exist some WJ -conjugates µ, υ ∈ Y of µx, µx′ respectively such that µ − υ ∈ (1 − σ)ZΦ∨. Note that µ, υ (cid:22) λ. By Corollary 7.4, there exist µj ∈ Y , βj ∈ Φ and tj ∈ [1, h − 1] tj ∈ [1, h] if ♯Oβj = 2h, and tj ∈ [1, 2h − 1] if if ♯Oβj = h (resp. ♯Oβj = 3h) such that µ0 = µ, µm′ = υ, µj+1 − µj = β∨ j ) and µj, µj + β∨ j ) (cid:22) λ for j ∈ [0, m′ − 1]. j , µj − σtj (β∨ j − σtj (β∨ Set x0 = µ0 ∈ π1(MJ ). Assuming xk and αk−1 are already con- structed for k ∈ [0, j], we now construct xj+1 ∈ π1(MJ ) and αj ∈ Φ−ΦJ (αj ,tj ) → xj+1. If xj = µm′ ∈ π1(MJ ), the construction is fin- such that xj ished. Otherwise, let ij = max{i ∈ [1, m′]; µi = xj ∈ π1(MJ )} and set xj+1 = µij +1 and αj = (βij )J , where (βij )J is defined (for βij ) as in Lemma 7.6. In this way, we obtain x = x0 (α0,t0) → x1 (α1,t1) → · · · (αm−1,tm−1) → xm = x′ such that x0, x1, . . . , xm are distinct elements of π1(MJ ). In particular, αj ∈ Φ − ΦJ for i ∈ [0, m]. By Lemma 7.6 (a), σi(αj) = αj if σi(βj) = βj. Thus either ♯Oαj = ♯Oβj or ♯Oαj = h < ♯Oβj . Define rj = tj if tj 6 h or ♯Oαj = ♯Oβj and define rj = tj − h otherwise. One checks that (2) and (3) are satisfied. (cid:3) [2, Proposition 4.5.4]) Let x, x′ ∈ ¯Iλ,J,b, α ∈ Proposition 7.8. (cf. Φ − ΦJ and r ∈ N such that 23 (1) α∨ is J-anti-dominant and J-minuscule; (2) r ∈ [1, h] if ♯Oα ∈ {h, 2h} and r ∈ [1, 2h − 1] if ♯Oα = 3h; (3) x Then for any P ∈ X MJ (α,r) → x′. µx (b), there exists P ′ ∈ X MJ i=0 σi(α∨) ∈ π1(MJ ). P ∼λ,b P ′ and ηJ (P ) − ηJ (P ′) =Pr−1 Proof of Proposition 5.1. It follows by combining Lemma 7.7 with Propo- sition 7.8. (cid:3) µx′ (b) satisfying The rest of this section is devoted to the proof of Proposition 7.8. Lemma 7.9. [2, Lemma 4.4.7] Let x ∈ π1(MJ ) and β ∈ Φ − ΦJ such that β∨ is J-anti-dominant. (a) hw(β), µxi 6 hwx(β), µxi for any w ∈ WJ ; (b) wx(β) is the unique minimal element in {w(β); w ∈ WJ , hw(β), µxi = hwx(β), µxi}; (c) If wx(β) = β, then µx is central on the connected components of ΦJ on which β∨ is not central; (d) in particular, (c) happens if hβ, µxi = hwx(β), µxi. Here wx ∈ WJ satisfies tµxwx ∈ ΩJ , see §2.4. Lemma 7.10. [2, Lemma 4.4.5] Let x ∈ π1(MJ ) and δ, δ′ ∈ Y which are J-anti-dominant and J-minuscule. Then µx + δ, µx − wx(δ′) and µx + δ − wx(δ′) are J-minuscule. In particular, they are conjugate to µx+δ,µx−δ′ and µx−δ+δ′ under WJ respectively. Lemma 7.11. [2, Lemma 4.5.1] Let x ∈ π1(MJ ) and β, γ ∈ Φ such that µx+β∨, µx−γ∨ (cid:22) λ. If α ∈ Φ − ΦJ satisfies (i) µx+β∨−α∨, µx+α∨−γ∨ (cid:14) λ; (ii) neither β nor γ lies in the connected component of Φ containing α. Then hw(α), µxi = 0 for any w ∈ WJ . In particular, wx(α) = α and hα, µxi = 0. Corollary 7.12. (cf. [2, Remark 4.5.2]) Let x, x′ ∈ ¯Iλ,J,b such that (α,r) ֌ x′ for some α ∈ Φ−ΦJ and some r ∈ N. Then wx(σi(α)) = σi(α) x and hσi(α), µxi = 0 if i ∈ [1, r − 1] and i, i − r /∈ Zh. Lemma 7.13. Let β ∈ Φ, µ ∈ Y and x ∈ k[[t]]×. Then Uβ(xt−1)tµ lies in KtµK if hβ, µi 6 −1 and lies in Ktβ∨+µK otherwise. Similarly, tµUβ(xt−1) lies in KtµK if hβ, µi > 1 and lies in Ktµ−β∨K otherwise. 24 SIAN NIE Proof. We only need to prove the first statement. If hβ, µi 6 −1, then Uβ(xt−1)tµ = tµUβ(xt−hβ,µi−1) ∈ KtK. Otherwise, we have Uβ(xt−1)tµ ∈ U−β(x−1t)sββ∨(x−1)tβ∨ U−β(x−1t)tµK ⊆ Ktµ+β∨ K as desired. (cid:3) Lemma 7.14. Let µ ∈ Y , λ ∈ Y + and α, β ∈ Φ such that the root system Φ ∩ (Zα + Zβ) is of type A2 or A1 × A1 or A1, and µ, µ + α∨, µ − β∨, µ + α∨ − β∨ (cid:22) λ. Then Uα(yt−1)tµUβ(zt−1) ∈ ∪λ′(cid:22)λKtλ′K, where y, z ∈ k[[t]]. Proof. Without loss of generality, we may assume Φ ∩ (Zα + Zβ) is of type A2. In particular, α 6= ±β. First, we suppose either hα, µi 6 −2 or hα, µi = −1 and hα, β∨i > 0. Note that Φ ∩ (Zα + Zβ) is of type A2. One computes that Uα(yt−1)tµUβ(zt−1) = tµUα(yt−1−hα,µi)Uβ(zt−1) =(tµUβ(zt−1)Uα(yt−1−hα,µi), tµUβ(zt−1)Uα(yt−1−hα,µi)Uα+β(cα,βyzt−2−hα,µi), if hα, β∨i > 0; otherwise. ⊆ KtµK ∪ Ktµ−β∨ K, Here cα,β ∈ k[[t]] is some constant and the inclusion follows from Lemma 7.13. Now we suppose otherwise, that is, either hα, µi > 0 or hα, µi = −1 and hα, β∨i = −1. One checks that Uα(yt−1)tµUβ(zt−1) = U−α(y−1t)sαα∨(y−1)tµ+α∨ U−α(y−1t1+hα,µi)Uβ(zt−1) ∈(Ktµ+α∨Uβ(zt−1)U−α(y−1t1+hα,µi), Ktµ+α∨Uβ(zt−1)U−α(y−1t1+hα,µi)Uβ−α(c−α,βy−1zthα,µi), if hα, β∨i 6 0; otherwise. ⊆ Ktµ+α∨ K ∪ Ktµ+α∨−β∨ K, where c−α,β ∈ k[[t]] is some constant and the last inclusion follows from Lemma 7.13. The proof is finished. (cid:3) Lemma 7.15. Let x, x′ ∈ ¯Iλ,J,b, α ∈ Φ and r ∈ N such that x′ − x = α∨ − σr(α∨) ∈ π1(MJ ). Then the following two statements are equivalent: µx (b), there exists P ′ ∈ X MJ i=0 σi(α∨) ∈ π1(MJ ). µx′ (b) satisfying P ∼λ,b (a) For any P ∈ X MJ (b) There exist P ∈ X MJ P ′ and ηJ (P ) − ηJ (P ′) =Pr−1 and ηJ (P ) − ηJ (P ′′) =Pr−1 µx (b) and P ′′ ∈ X MJ i=0 σi(α∨) for some µ′ ∈ Y . µ′ (b) such that P ∼λ,b P ′′ 25 b Proof. We only need to show (b) ⇒ (a). Since J MJ on π0(X MJ µx (b)), it suffices to show there exists P ′ ∈ X MJ (F ) acts transitively µx′ (b) such that i=0 σi(α∨) ∈ π1(MJ ). In view of the conditions of (b), we have µ′ = µx′ ∈ π1(MJ ) and µ′ (cid:22) λ. By Corollary µx′ (b) with P ′′ ∼J,µ′,b P ′. Then we have 3.8 (i), there exists P ′ ∈ X MJ (cid:3) P ∼λ,b P ′ and ηJ (P ) − ηJ(P ′) =Pr−1 P ∼λ,b P ′ and ηJ (P ) − ηJ (P ′) =Pr−1 Now we are ready to prove Proposition 7.8. i=0 σi(α∨) as desired. Proof of Proposition 7.8. Thanks to Lemma 7.5, we can and do assume x (α,r) ֌ x′. Moreover, by Lemma 7.15, it suffices to show (i) there exist P ∈ X MJ µx (b) and P ′′ ∈ X MJ i=0 σi(α∨) for some µ′ ∈ Y . and ηJ (P ) − ηJ (P ′′) =Pr−1 Let gx ∈ MJ (L) such that g−1 x bσ(gx) = bx = tµx wx, where bx = µ′ (b) such that P ∼λ,b P ′′ tµxwx ∈ ΩJ is define in §2.4. By Corollary 7.12, we have (a) wx(σi(α)) = σi(α), hσi(α), µxi = 0 and hence hσi(α), βi = 0 (by Lemma 7.9 (c)) if i ∈ [1, r − 1] and i, i − r /∈ Zh. Case(1): r ∈ [1, h]. Define g : P1 = A1 ∪ {∞} → G(L)/K such that g(z) = gx Uα(zt−1) bxσUα(zt−1) · · · (bxσ)r−1 Uα(zt−1)K. One computes that g(z)−1bσg(z) Uα(zt−1)K Uα(zt−1)K x bσgxUα(zt−1) · · · (bxσ)r−1 Uα(−zt−1)) · · · Uα(−zt−1)g−1 Uα(−zt−1) · · · Uα(−zt−1)bxσUα(zt−1) · · · (bxσ)r−1 = K (bxσ)r−1 = K (bxσ)r−1 = KUα(−zt−1) bxσ (bxσ)r−1 = KUα(−zt−1)tµxUwx(σr (α))(cσr(z)t−1) wxσK ∈ ∪λ′(cid:22)λKtλ′ where c ∈ k[[t]]× is some constant; the third equality follows from that α, σi(α) lie in different connected components of Φ for i ∈ [1, r − 1]; the fourth equality follows from (ii); the last inclusion follows from Lemma 7.10 and Lemma 7.14. Moreover, one computes that Kσ ⊆ K\G(L)/K, Uα(zt−1)K r−1 P ′′ := g(∞) = lim z→∞ gx U−σi(α)(ciσi(z−1)t)t−σi(α∨)K Yi=0 = gx t− Pr−1 i=0 σi(α∨)K ∈ MJ (L)/(MJ (L) ∩ K), where ci ∈ k[[t]]× is some constant. So P ′′ ∈ X MJ µx′ (b), P = g(0) ∼λ,b g(∞) = P ′′ and ηJ (P ) − ηJ (P ′′) =Pr−1 i=0 σi(α∨) as desired. Case(2): r ∈ [h + 1, 2h − 1]. Then ♯Oα = 3h and each connected component of Φ is of type D4. Since α∨ is J-anti-dominant and J- minuscule, we have either J = ∅ or ΠJ = Oβ, where β is the unique 26 SIAN NIE simple root in ΠJ such that σh(β) = β and β, α are in the same con- nected component of Φ. We claim that (2-a) If hβ, α∨i = −1 and hσr(β), µxi = 1, then hσr(β), µx′i = 0. Indeed, if hσr(β), µxi = 1, then wx(σi(α)) = σi(α) + σr(β) for i ∈ r + Zh. Using the relation x′ = x + α∨ − σr(α∨), we have hσr(β), µx′i = hσr(β), sβ(µx + α∨ − wx(σr(α∨)))i = hσr(β), µx − σr(α∨) − σr(β∨)i = 0 Note that x′ and (2-a) is verified. (−α,r) ֌ x is equivalent to x (α,r) ֌ x′. Therefore, by Lemma 7.15, the proposition dose not change if we exchange the roles of x and x′. Thus, we may and do assume (2-b) hσr(β), µxi = 0 if hβ, α∨i = −1. Then we show that (2-c) hσr−h(α), µxi = hσh(α), µxi = 0. Note that hβ, α∨i ∈ {0, −1}. So by (2-b) if it applies, we always have µx − σr(α) = µx − wx(σr(α)) (cid:22) λ. Then hσr−h(α), ζi = hσr−h(α), µxi, where ζ ∈ A(x, α, r) := {µx, µx + α∨, µx − wx(σr(α∨)), µx + α∨ − wx(σr(α∨))}. By Lemma 7.10, we know that A(x, α, r) (cid:22) λ. Thus one checks that (α,r−h) → x + α∨ − σr−h(α∨) (σr−h(α),h) → x′ if hσr−h(α), µxi > 1; (σr−h(α),h) → x + σr−h(α∨) − σr(α∨) (α,r−h) → x′ if hσr−h(α), µxi 6 −1. x x Hence hσr−h(α), µxi = 0 since x follows similarly and (2-c) is proved. (α,r) ֌ x′. The equality hσh(α), µxi = 0 Case(2.1): hβ, α∨i = −1 and hβ, µxi = 1. To show (i), we define g : P1 → G(L)/K such that g(z) = gx (bxσ)r−1 Uα(zt−1) (bxσ)r−2 Uα(zt−1) · · · Uα(zt−1)K. Let P ′′ = g(∞) ∈ MJ (L)/(K ∩ MJ (L)). By computations on on page In −1bσ(g) ⊆ ∪λ′(cid:22)λKtλ′K ⊆ K\G(L)/K, 45-47 of [2], we have ηJ (P ) − ηJ (P ′) = Pr−1 view of (i), it remains to show g which holds, by (a), (2-b) and (2-c) (see page 45 of [2]), if i=0 σi(α∨) ∈ π1(MJ ). (2.1.1) Uα(yt−1)tµx ∈ ∪λ′(cid:22)λKtλ′K for y ∈ k[[t]]; (2.1.2) Uσr−h(α)(yt−1)Uσr (α)+σr (β)(zthσr (α),µxi)Uσr−h(α)(−yt−1) ∈ K for y, z ∈ k[[t]]. Since µx, µx + α∨ (cid:22) λ, (2.1.1) follows from Lemma 7.14. To verify (2.1.2), it suffices to show hσr(α), µxi > 1. Indeed, assume hσr(α), µxi 6 0. Then hσr(α + β), µx − σr(α∨) − σr(β∨) − σr−h(α∨)i 6 −1, which implies µx−σr−h(α∨) (cid:22) µx−σr(α∨)−σr(β∨)−σr−h(α∨) = sσr−h(β+α)(µx−σr(α∨)) (cid:22) λ, 27 where the second equality follows from hσr(β)+σr−h(α), µx−σr(α∨)i = 1. Thus µx + α∨ − σr−h(α∨), µx + α∨ − σr(α∨) − σr(β∨) − σr−h(α∨) (cid:22) λ since α, σr(α) are not in the same connected component of Φ. Then (α,r−h) → x + α∨ − σr−h(α∨) x (σr−h(α),h) → x′, contradicting the assumption x desired. (α,r) ֌ x′. So we have hσr(α), µxi > 1 as Case(2.2): either hβ, α∨i or hβ, µxi = 0. By (2-b) if it applies, wx fixes σr−h(α) and σh(α). Combining (a) and (2-c), we have hσj(α), µxi = 0 and wx(σj(α)) = σj(α) for j ∈ [1, r−1]. Noticing that hσi−h(α), σi(α∨)i = 0 for i ∈ Z, the proposition follows similarly as in Case(1). (cid:3) 8. Proof of Proposition 5.3 The main aim of this section is to prove Proposition 5.3. Through out this section, we assume G is adjoint and simple; the root system Φ of G has h connected components, on which σ acts transitively. For µ, λ ∈ Y , we define Υ+(µ, λ) = {α ∈ Φ+; µ + α∨ (cid:22) λ}. [2, Proposition 4.1.9]) Let J ⊆ S0 with σ(J) = J, Lemma 8.1. (cf. b ∈ G(L) and λ ∈ Y + such that (J, b) is admissible and (λ, b) is HN- irreducible. For x ∈ ¯Iλ,J,b, we define Cλ,J,b,x = {β ∈ Υ+(µx, λ); β∨ is J-minuscule and J-anti-dominant}. Then the coroot lattice ZΦ∨ is spanned by Φ∨ J and ∪β∈Cλ,J,b,x O∨ β . If, moreover, Φ is a union root systems of type G2 and J = Πshort,G2. Then ZΦ∨ is still generated by Π∨ α. Here Πshort,G2 (resp. Πlong,G2) denotes the single σ-orbit of simple short (resp. long) roots in Π0. J and ∪α∈Cλ,J,b,x∩Πlong,G2 O∨ Proof. By Lemma 2.2 and that (λ, b) is HN-irreducible, we have λ⋄ − νbx = λ⋄ − νG cαα∨, [b] = Xα∈Π0 where bx ∈ ΩJ is defined in §2.4 and each cα ∈ Q is strictly positive. Therefore, (a) for each α ∈ Π0 − ΠJ , there exists θ ∈ Oα such that µx + θ∨ 6 λ. We claim that (b) ZΦ∨ is generated by Φ∨ Assume (b). For α ∈ Υ+(µx, λ) − ΦJ , let αJ be defined as in Lemma J . Therefore, ZΦ∨ is also β for α ∈ Υ+(µx, λ), as desired. 7.6. Then αJ ∈ Cλ,J,b,x and α∨ − α∨ spanned by ZΦ∨ αJ ⊆ ∪β∈Cλ,J,b,x J and ∪α∈Υ+(µx,λ)O∨ α. J ∈ ZΦ∨ O∨ J and O∨ 28 SIAN NIE It remains to show (b). Let A be the lattice spanned by Φ∨ J and and ∪α∈Υ+(µx,λ)O∨ α. Let J0 ⊆ S0 be the maximal subset such that Φ∨ J ⊆ A. Note that J ⊆ J0 and σ(J0) = J0. We have to show that J0 = S0. Assume otherwise, we choose w ∈ WJ0 such that µ = w(µx) is J0-dominant. Case(1): µ /∈ Y +. Then there exists β ∈ Π − ΠJ0 such that hβ, µi 6 J0 ⊆ A, −1, which implies w−1(β) ∈ Υ+(µx, λ). Since w ∈ WJ0 and Φ∨ we deduce that β∨ ∈ A, contradicting the maximality of J0. Case(2): µ ∈ Y +. Choose α ∈ Π − ΠJ0. Since µ − µx ∈ ZΦ∨ J0 and µ 6 λ, there exists, by (a), θ ∈ Oα such that µ + θ∨ 6 λ. Subcase(2.1): µ + θ∨ (cid:22) λ. Then we have w−1(θ) ∈ Υ+(µx, λ) and θ∨ ∈ A as in Case(1), a contradiction. Subcase(2.2): µ + θ∨ (cid:14) λ. Notice that θ ∈ Φ+. Thus Lemma 6.4 (b) applies, that is, there exists γ ∈ Φ+ such that µ + γ∨ 6 µ + θ∨ + γ∨ 6 λ and γ∨, sθ(γ∨) = θ∨ + γ∨ are short positive coroots. By Lemma 6.5, µ + γ∨, µ + θ∨ + γ∨ (cid:22) λ and hence w−1(γ), w−1(sθ(γ)) ∈ Υ+(µx, λ). Therefore, we deduce that θ∨ ∈ A, a contradiction. So (b) is proved. Π∨ long,G2, it suffices to show Now we turn to the "moreover" part. Since ZΦ∨ is spanned by short,G2 and Π∨ (c) Cλ,J,b,x ∩ Πlong,G2 6= ∅. If µx /∈ Y +, (c) follows similarly as in Case(1) above. Assume µx ∈ Y +. By (a), there exists θ ∈ Πlong,G2 such that µx + θ∨ 6 λ. Hence µx + θ∨ (cid:22) λ by Lemma 6.5. So (c) still holds. (cid:3) Corollary 8.2. Keep the assumptions of Lemma 8.1. We modify the definition of Cλ,J,b,x slightly by resetting Cλ,J,b,x = {α ∈ Πlong,G2; α is J-antidominant and µx + α∨} if Φ is a union of root systems of type G2 and ΠJ = Πshort,G2. Then ZΦ∨ is still generated by Π∨ J and ∪α∈Cλ,J,b,x O∨ α. This modified definition will be used to prove Proposition 8.3 below. Proposition 8.3. (cf. [2, Proposition 4.1.10]) Let J ⊆ S0 with σ(J) = J, b ∈ G(L) and λ ∈ Y + such that (J, b) is admissible. For x ∈ ¯Iλ,J,b we have Cλ,J,b,x ⊆ Aλ,J,b, where Aλ,J,b is defined in §5.1. The proof will be given in Section 9. Now we are ready to prove Proposition 5.3. The statement (b) and its usage in the proof below are suggested by Miaofen Chen, which play a fundamental role in our arguments. Proof of Proposition 5.3. By [2, Corollary 2.5.12], the kernel of the nat- ural map π1(MJ )σ → π1(G)σ is spanned by yα =Pβ∈Oα β∨ ∈ π1(MJ )σ for α ∈ Φ. Let J0 = {sα ∈ S0; α ∈ Π0, yα ∈ Lλ,J,b}, where Lλ,J,b is 29 the lattice spanned by yα for α ∈ Aλ,J,b. It suffices to show J0 = S0. Note that σ(J0) = J0 and J ⊆ J0. By assumption of the proposition, there exists µ ∈ ¯Iλ,J,b, which is J-dominant, J-minuscule and weakly dominant. Since σ acts transitively on the h connected components of Φ, {♯Oγ; γ ∈ Π0} = {nΦ, mΦ} for some nΦ 6 mΦ ∈ N with mΦ/nΦ ∈ {1, 2, 3}. Let α ∈ Φ0. Then ♯Oα = h if ♯Oα < mΦα. First we claim (a) if ♯Oα = mΦ, then α ∈ ΠJ0. By Corollary 8.2, α∨ lies in the lattice generated by ZΦ∨ J and ∪α∈Cλ,J,b,µ i=1 yγi ∈ π1(MJ )σ with each γi ∈ ±Cλ,J,b,µ. Hence yα ∈ Lλ,J,b by Proposition 8.3. Therefore, α ∈ ΠJ0 and (a) is proved. Moreover, since ♯Oα = mΦ, we have yα = Pr If J0 = S0, there is nothing to prove. Assume J0 6= S0. We show this assumption will lead to some contradiction, and this complete our proof. By (a), σ is of order 2h or 3h and each simple root of Π − ΠJ0 is fixed by σh. In particular, Φ is simply laced. We claim that O∨ α. (b) there exist α ∈ Π0 − ΠJ0, ϑ ∈ Φ+ with σh(ϑ) = ϑ such that J0 and either ϑ∨ − α∨ ∈ ZΦ∨ (b1) ϑ ∈ Cλ,J,b := ∪x∈ ¯Iλ,J,bCλ,J,b,x; or (β,h) → x + β∨ − σh(β∨) and x (b2) x β ∈ ΦJ0 and x ∈ ¯Iλ,J,b. (β+ϑ,h) → x + β∨ − σh(β∨) for some Assuming (b), we now can finish the proof of the proposition. J0 and ♯Oα = ♯Oϑ = h, we have yα − yϑ ∈ (ZΦ∨ If (b1) occurs, we have, by Proposition 8.3, yϑ ∈ Lλ,J,b. Since ϑ∨ − α∨ ∈ ZΦ∨ J0)σ ⊆ Lλ,J,b, which means yα ∈ Lλ,J,b and hence α ∈ ΠJ0, a contradiction. If (b2) occurs, by Proposition 7.8, there exist Q ∈ X MJ µx (b) and P1, P2 ∈ X MJ (b) such that Q ∼λ,b P1 (resp. Q ∼λ,b P2) and ηJ (Q) − i=0 σi(ϑ∨ + β∨)). Therefore, P1 ∼λ,b P2 and ηJ (P1) − ηJ (P2) = yϑ, that is, ϑ ∈ Aλ,J,b and hence yϑ ∈ Lλ,J,b, which again is a contradiction by the same argument in the case (b1). i=0 σi(β∨) (resp. ηJ (Q) − ηJ (P2) = Ph−1 µx+β∨−σh(β∨) ηJ (P1) = Ph−1 It remains to prove (b). Case (1): σ is of order 3h. Then each connected component of Φ is of type D4. We can write S0 = F ⊔ H, where F and H are the two distinct σ-orbits on S0. Assume σh acts trivially on H. Then F = J0 by (a) and the assumption J0 6= S0. Note that (λ, b) is NH-irreducible. So there exists α ∈ H = Π0 − ΠJ0 such that (1-a) µ + α∨ 6 λ. We remark that (1-b) Let µ′ ∈ Y such that µ′ (cid:22) λ and µ′ + α∨ 6 λ. Assume hα, µ′i 6 −1 or µ′ is J0-dominant, then µ′ + α∨ (cid:22) λ. Indeed, Since µ′ (cid:22) λ, We can assume Φ has only one connected component. If hα, µ′i 6 −1, the statement is obvious. Now we assume µ is J0-dominant and hα, µ′i > 0, that is, µ′ is dominant (since Φ is 30 SIAN NIE of type D4). Then the statement follows from Lemma 6.5. So (1-b) is proved. Case (1.1): J = F = J0. Then µ is J0-dominant and (b1) follows from (1-b) by taking ϑ = α. Case (1.2): J = ∅. Choose β ∈ ΠJ0 such that hβ, α∨i = −1. We claim that (1.2-a) there exists µ0 ∈ ¯Iλ,J,b such that µ0 +α∨ 6 λ and hβ +σh(β)+ σ2h(β), µ0i > 0. In particular, Pγ∈Oβ Let E = {µ′ ∈ ¯Iλ,J,b; µ′ is weakly dominant and µ′ + α∨ 6 λ}. Then µ ∈ E 6= ∅. Let c0 = max{hβ + σh(β) + σ2h(β), µ′i; µ′ ∈ E}. We have to show c0 > 0. Assume otherwise. Let µ0 ∈ E such that hβ + σh(β) + σ2h(β), µ0i = c0 6 −1. By Lemma 2.2, νtµ0 = 1 i=0 σi(µ0) hγ, µ0i > 0. So there exist γ′ ∈ is dominant. {β, σh(β), σ2h(β)} and γ′′ ∈ Oβ −{β, σh(β), σ2h(β)} such that hγ′, µ0i 6 −1 and hγ′′, µ0i > 1. Let µ′ 0 (cid:22) λ 0 + α∨ 6 λ since is weakly dominant by Lemma 6.6. Moreover, µ′ µ0 + α∨, µ′ 0 6 λ and α is a simple root. Thus, µ′ 0 ∈ E. However, hβ + σh(β) + σ2h(β), µ′ 0i = c0 + 2, a contradiction. So (1.2-a) is proved. Thanks to (1-b), we can assume, by (1.2-a) and the symmetry among β, σh(β) and σ2h(β), that hµ0, αi > 0, hβ, µ0i = −1 and one of the following cases occurs: 0 = µ0 + γ′∨ − γ′′∨ ∈ ¯Iλ,J,b. Then µ′ 3hP3h−1 (1.2.1): hσ2h(β), µ0i = −1 and hσh(β), µ0i > 2; (1.2.2): hσh(β), µ0i, hσ2h(β), µ0i > 0. Assume (1.2.1) occurs. Then µ′ 0 − σ2h(β), µ′ 0i > 1. Thus we deduce that µ′ 0 + β∨ − σ2h(β∨) 0 + β∨ (cid:22) λ is J0-dominant. So µ′ 0 − σ2h(β) − α∨ (cid:22) λ since hσ2h(β), µ′ 0 := µ0 + σ2h(β∨) − σh(β∨) (cid:22) λ. 0 + β∨ + α∨ 6 λ since 0 + β∨ + α∨ (cid:22) λ by (1-b). Moreover, we 0i, hσ2h(β) + → µ′ 0 and Note that µ′ µ0 + α∨ 6 λ. Therefore, µ′ have µ′ α, µ′ µ′ 0 + β∨ − σ2h(β∨) Assume (1.2.2) occurs. Again we have µ0 + β∨ + α∨ (cid:22) λ by (1-b). → µ0 and µ0 + β∨ − (β,h) → µ0 + β∨ − σh(β∨) If hσ2h(β), µ0i > 1, then µ0 + β∨ − σ2h(β∨) σ2h(β∨) → µ0. If hσh(β), µ0i > 1, then µ0 (α+σ2h(β),h) → µ′ 0. (σ2h(β),h) (σ2h(β),h) (α+σ2h(β),h) and µ0 (α+β,h) → µ0 + β∨ − σh(β∨). So (b) holds when σ is of order 3h. Case (2): σ is of order 2h. As in the proof of Lemma 8.1, there exists α ∈ Π − ΠJ0 such that ¯µJ0 + α∨ (cid:22) λ (by using Lemma 6.5 and that Φ is simply laced when ¯µJ0 is dominant). In particular, µ + α∨ 6 λ. We claim that (2-a) There exists an orthogonal subset D ⊆ Φ+ J0 satisfying (2-a-1) hγ, µi = hγ, α∨i = −1 for any γ ∈ D; (2-a-2) µ + α∨ +Pγ∈D γ∨ (cid:22) λ. 31 Indeed, since µ is weakly dominant, there exists, by Lemma 6.9, an β∨ and orthogonal subset Dµ,J0 ⊆ Φ+ hβ, µi = −1 for any β ∈ Dµ,J0. Let J0 such that ¯µJ0 − µ = Pβ∈Dµ,J0 D = {β ∈ Dµ,J0; hβ, α∨i 6 −1} = {β ∈ Dµ,J0; hβ, α∨i = −1}. Note that Dµ,J0 − D = {β ∈ Dµ,J0; hβ, α∨i > 0}. So µ + α∨ +Xγ∈D γ∨ (cid:22) µ + α∨ +Xγ∈D γ∨ + Xβ∈Dµ,J0 −D β∨ = ¯µJ0 + α∨ (cid:22) λ. Thus (2-a) is proved. Now (b) follows, in the case that σ is of order 2h, from (2-a) by Lemma 8.4 below. In a word, the proof of (b) is finished. (cid:3) Lemma 8.4. Let λ ∈ Y + and J ⊆ J0 ⊆ S0 with σ(J0) = J0. Assume Φ is simply laced and σ2h = id and there exist µ ∈ ¯Iλ,J,b which is weakly dominant, θ ∈ Φ+ − ΦJ0 with σh(θ) = θ and an orthogonal subset D ⊆ Φ+ J0 such that (1) hγ, µi = hγ, θ∨i = −1 for any γ ∈ D; (2) µ + θ∨ +Pγ∈D γ∨ (cid:22) λ. one of the following holds: Then there exists ϑ ∈ Φ+ such that σh(ϑ) = ϑ, ϑ∨ − θ∨ ∈ ZΦ∨ J0 and (i) either ϑ ∈ Cλ,J,b := ∪x∈ ¯Iλ,J,bCλ,J,b,x or (α+ϑ,h) (ii) x (α,h) → x + α∨ − σh(α∨) and x α ∈ ΦJ0 and some x ∈ ¯Iλ,J,b. Proof. We employ the notation used in Lemma 7.6. → x + α∨ − σh(α∨) for some Case(0): σh = id. Let β = (Qγ∈D sγ)(θ) and take ϑ = βJ , where βJ is defined as in Lemma 7.6. Then ϑ ∈ Cλ,J,b,µ as desired. Case(1): σh is of order 2. Case(1.0): D = ∅. Then µ + θ∨ (cid:22) λ. Let ϑ = θJ , then σh(ϑ) = ϑ since σh(θ) = θ. Then the lemma follows similarly as in Case(0). Case(1.1): ♯D = 1. Write D = {γ} for some γ ∈ Φ+ J0. If γ = σh(γ), then we return to Case(1.0) by replacing θ with θ + γ. Now If hσh(γ), µ + γ∨i 6 0, then hσh(γ), µ + θ∨ + assume γ 6= σh(γ). γ∨i 6 hσh(γ), θ∨i = hγ, θ∨i = −1. Thus θ + γ + σh(γ) ∈ Φ+ and µ + θ + γ + σh(γ) (cid:22) λ. So we return to Case(1.0) by replacing θ with θ+γ+σh(γ). Assume hσh(γ), µ+γ∨i > 1. In particular, hσh(γ), µi > 1. We have µ − σh(γ∨) (cid:22) µ (cid:22) λ and µ + γ∨ − σh(γ∨) (cid:22) µ + γ∨ (cid:22) µ (cid:22) λ. Moreover, hσh(θ + γ), µi = hθ + σh(γ), µi > hθ + γ, µi + 2 > −1 + 2 = 1, where the last inequality follows from the assumption that µ is weakly dominant. So µ−σh(θ∨+γ∨) (cid:22) µ (cid:22) λ. Therefore, µ → µ+γ−σh(γ) (θ+γ,h) and µ (γ,h) → µ + γ − σh(γ). So (ii) is satisfied. 32 SIAN NIE j i = hσh(γj), γ∨ Case(1.2): ♯D = 2. Write D = {γ1, γ2}. We have σh(γi) 6= γj since otherwise, we would return to Case(1.0) or Case(1.1). As a conse- quence, ǫi,j := hσh(γi), γ∨ i i ∈ {−1, 0, 1} for i, j ∈ {1, 2}, and moreover, ǫi,i = 0 since −1 6 hσh(γi), θ∨ + γ∨ i i = −1 + ǫi,i 6 −1. If ǫ1,2 6 0, then hσh(θ + γ1 + γ2), θ∨ + γ∨ 2 i 6 0 + 2(ǫ1,2 − 1) 6 −2, which is impossible. So ǫ1,2 = 1. By symmetry, we may assume δ := γ1 − σh(γ2) ∈ Φ+. Then θ + γ1 + γ2 = θ + γ2 + σh(γ2) + δ and 1 + γ∨ hδ, µ + θ∨ + γ∨ 2 + σh(γ∨ 2 ) + δ∨i = hδ, µi + 1. If hδ, µi > 0, then hδ, µ + θ∨ + γ∨ 2 + σh(γ∨ 2 ) + δ∨i > 1 and hence µ+θ∨ +γ∨ 2 +σh(γ∨ 2 ) (cid:22) µ+θ∨ +γ∨ 2 +σh(γ∨ 2 )+δ∨ = µ+θ∨ +γ∨ 1 +γ∨ 2 (cid:22) λ. So we return to Case(1.0). Assume hδ, µi = −1 (since µ is weakly dominant). We have hδ, θ∨ + γ∨ 2 )i = −1 and hence return to Case(1.1). 2 + σh(γ∨ Case(1.3): ♯D = 3. Write D = {γ1, γ2, γ3}. As in Case(1.2), we j i = i i ∈ {−1, 0, 1} for i, j ∈ {1, 2, 3}. Then ǫi,i = 0 for i ∈ can assume σh(γi) 6= γj for i, j ∈ {1, 2, 3}. Set ǫi,j = hσh(γi), γ∨ hσh(γj), γ∨ {1, 2, 3} as above. One computes that hθ+σh(γi), θ∨ +γ∨ 1 +γ∨ 2 +γ∨ 3 i = −2+hσh(γi), γ∨ 1 +γ∨ 2 +γ∨ 3 i ∈ {−1, 0, 1}, which implies ǫi,j > 0 if i 6= j. If ǫi,j = 0 for any i 6= j ∈ {1, 2, 3}, then hθ+γ1+γ2+γ3, θ∨+γ∨ 2 + γ∨ 3 i 6 −4, which is impossible. By symmetry, we can assume ǫ1,2 = 1 and δ := γ1−σh(γ2) ∈ Φ+. Then θ+γ1+γ2+γ3 = θ+γ2+σh(γ2)+γ3+δ. Since µ is weakly dominant, we have 1 +γ∨ (b) hδ, µi = −1 − hσh(γ2), µi 6 0 and hδ, γ∨ Case(1.3.1): hδ, γ∨ 3 i = −ǫ2,3 = 0. One computes that 3 i = −ǫ2,3 6 0. hδ, µ + θ∨ + γ∨ 2 + σh(γ∨ 2 ) + γ∨ 3 + δ∨i = hδ, µi + 1. We apply the arguments in Case(1.2). If hδ, µi > 0, then µ + θ∨ + γ∨ σh(γ∨ 3 (cid:22) λ and we return to Case(1.1). Otherwise, 2 ) + γ∨ 2 + hγ3, θ∨ + γ2 + σh(γ2)i = hδ, θ∨ + γ2 + σh(γ2)i = hδ, µi = −1 and we return to Case(1.2). Case(1.3.2): hδ, γ∨ 3 i = −1. Then ǫ2,3 = 1. Since µ is weakly domi- nant, hδ + γ3, µi > −1, which implies hδ, µi = 0 by (b). One computes that hγ3 + δ, θ∨ + γ∨ 2 )i = hγ3 + δ, µi = −1 and we return to Case(1.1). 2 + σh(γ∨ Case(1.4): ♯D > 4. This case dose not occur since hθ+Pγ∈D γ, θ∨i 6 −2, contradicting the fact that Φ is simply laced. (cid:3) 33 9. Proof of Proposition 8.3 The aim of this section is to prove Proposition 8.3, which completes the proof of Proposition 5.3. We follow Chen-Kisin-Viehmann [2, §4.7] closely (but not exactly) and use their ideas and constructions freely. We will see that most of the results therein can be generalized in our setting. However, the arguments become more subtle. Throughout this section, we assume G is adjoint and simple; the root system Φ of G has h connected components, on which σ acts transitively. We also fix λ ∈ Y +, b ∈ G(L) and a σ-stable subset J ⊆ S0 such that (J, b) is admissible and b is superbasic in MJ (L). For α ∈ Φ − ΦJ , we set ΦJ,α = Φ ∩ Z(ΦJ ∪ Oα). Thanks to [2, Proposition 4.2.11], the set of simple root of ΦJ,α is ΠJ ∪ Oα if Φ is simply laced. Following [2, §4.7], we say Oα is of type I, II, or III (with respect to J) if ♯Oα is equal to d, 2d or 3d respectively. Here d is the minimal positive integer such that α, σd(α) are in the same connected component of ΦJ,α. We define α, α + β + σd(α), α + σd(α) + σ2d(α) + β, if Oα is of type I; if Oα is of type II; if Oα is of type III, α =  where β is the unique common neighbor (might be zero in type II case) of Oα in he Dynkin diagram of ΦJ,α. We emphasize that α is root. Moreover, α is J-anti-dominant and J-minuscule if so is α. Now we fix x ∈ ¯Iλ,J,b and α ∈ Cλ,J,b,x. Set n = ♯Oα. To prove Proposition 8.3, we show α ∈ Aλ,J,b separately according to the type of Oα. Lemma 9.1. Let x ∈ ¯Iλ,J,b and α ∈ Cλ,J,b,x. If there do not exist x′ ∈ (θ,j) ¯Iλ,J,b, θ ∈ Oα and j ∈ [1, n − 1] such that x → x′. Then wx(σk(α)) = σk(α) and hσk(α), µxi = 0 for k ∈ [1, n] − Nd. Proof. Let k ∈ [1, n] − Nd. Then α, σk(α) are in different connected components of ΦJ,α. First we claim that (a) hwx(σk(α)), µxi 6 0 for k ∈ [1, n] − Nd. Indeed, otherwise, hwx(σk(α)), µx + α∨i = hwx(σk(α)), µxi > 1, (α,k) → x + α∨ − σk(α∨), contradicting the assumption which means x of the lemma. So (a) is proved. Assume hα, µxi > 0. Then µx − α∨ (cid:22) λ since µx, µx + α∨ (cid:22) λ. Thus, (σk(α),n−k) we have hσk(α), µxi > 0. Indeed, otherwise we would have x x + σk(α∨) − α∨, a contradiction. By (a) and Lemma 7.9, we have hwx(σk(α)), µxi = hσk(α), µxi = 0 and wx(σk(α)) = σk(α) as desired. → 34 SIAN NIE Now assume hα, µxi 6 −1. We adopt the following argument in [2, Lemma 4.7.10]. Since νbx ∈ YQ is dominant (see Lemma 2.2), for any positive m ∈ Nd with (wxσ)m = 1, we have m−1 m−1 0 6 mhα, νbxi = h(wxσ)−i(α), µxi = hwi(σi(α)), µxi Xi=0 Xi=0 for some wi ∈ WJ with w0 = 1. However, by (a), we have hwi(σi(α))i 6 hwx(σi(α)), µxi = 0 for i ∈ [0, m − 1] −Nd. Using Lemma 7.9 (a) again, we obtain 0 6 Xi∈[0,m−1]∩Nd hwi(α), µxi 6 hα, µxi+( which implies m d −1)hwx(α), µxi 6 m d hwx(α), µxi, (b) hwx(σi(α)), µxi > 1 and hence µx − wx(σi(α∨)) (cid:22) λ. (σk(α),i−k) By (b), we have x (σk (α),n+i−k) x similarly as in the above paragraph. → x + σk(α∨) − σi(α∨) if k < i and x + σk(α∨) − σi(α∨) if i < k. Now the lemma follows (cid:3) → Proof of Proposition 8.3 when Oα is of type I. Case(1): x (θ,r) → x′ for some x′ ∈ ¯Iλ,J,b, θ ∈ Oα and r ∈ [1, d − 1]. We claim that µx (b) and Q′ ∈ X MJ i=0 σi(θ∨). µx′ (b) such that P ∼λ,b Q (1-a) there exist Q ∈ X MJ and ηJ (P ) − ηJ (Q) =Pr−1 metry, we have x′ (σr (α),d−r) on π0(X MJ First we show the proposition follows from (1-a). Indeed, by sym- → x. Thus by (1-a) and that J MJ µx′ (b)) transitively, there exist P ′ ∈ X MJ (F ) acts µx (b) σi+r(θ∨). Therefore, Q ∼λ,b P and ηJ (Q) − ηJ (P ) = yθ = yα ∈ π1(MJ )σ. So α ∈ Aλ,J,b as desired. such that Q′ ∼J,λ,b P ′ ∼λ,b P and ηJ (P ′) − ηJ (P ) = Pd−r−1 µx′ (b) and P ∈ X MJ i=0 b Now we prove (1-a) by induction on r. If r = 1, then x (θ,r) ֌ x′ and the lemma follows from Proposition 7.8. Assume the lemma holds for r < r0 ∈ [1, d]. We show it holds for r = r0. Case(1.1): there exist u ∈ WJ and j ∈ [1, r−1] such that hu(σj(θ)), µxi 6= 0. Without loss of generality, we may assume that hu(σj(θ)), µxi 6 −1, the other case can be handled similarly. Let x′′ = x + σj(θ∨) − σr(θ∨) ∈ π1(MJ ). Since Oα is of type I, we have hσj(θ), wx(σr(θ∨))i = hσj(θ), θ∨i = 0 and hence hσj(θ), µx + θ∨ − wx(σr(θ∨))i = hσj(θ), µxi 6 hu(σj(θ)), µxi 6 −1, which means x exist P ′′ ∈ X MJ (−σj (θ),r−j) → x′′ (−θ,j) µx′′ and P ′ ∈ X MJ ηJ (P )−ηJ (P ′′) =Pr−j−1 i=0 → x′. By induction hypothesis, there µx′ such that P ∼λ,b P ′′ ∼λ,b P ′ and i=0 σi(θ∨). σi+j(θ∨) and ηJ (P ′′)−ηJ (P ′) =Pj−1 35 as desired. Therefore, we deduce that P ∼λ,b P ′ and ηJ (P )−ηJ (P ′) =Pr−1 Case(1.2): hu(σi(θ)), µxi = 0 for any u ∈ WJ and any i ∈ [1, r − 1]. By Lemma 7.9, wx(σi(θ)) = σi(θ) and hσi(θ), µxi = 0 for any i ∈ [1, r − 1]. Define g : P1 → G(L)/K by i=0 σi(θ∨) g(z) = gx Uθ(zt−1) bxσUθ(zt−1) · · · (bxσ)r−1 Uθ(zt−1)K, x bσ(gx) = bx = tµx wx and bx = tµxwx ∈ where gx ∈ MJ (L) such that g−1 ΩJ is defined in §2.4. Note that θ, σi(θ) are not in the same connected component of ΦJ,α for i ∈ [1, r − 1] since Oα is of type I. One checks that ηJ (g(0)) − ηJ (g(∞)) =Pr−1 k=0 σk(θ∨) and g(z)−1bσ(g(z)) = KUθ(zt−1)tµxUwx(σr (θ))(cσr(z)t−1) wxK, where c ∈ k[[t]]× is some constant. In view of Lemma 7.15, to show −1bσ(g) ⊆ ∪λ′(cid:22)λKtλ′K. By Lemma 7.14, it (1-a), it remains to verify g suffices to show (1.2-a) Φ ∩ (Zθ + Zwx(σr(θ))) is of type A1 or A1 × A1 or A2. If Φ is simply laced, (1.2-a) follows since all the roots of Φ are of the same length. Otherwise, (1.2-a) follows since θ, wx(σr(θ))) are in different connected components of Φ. Therefore, (1-a) is proved. Case(2): There do not exist x′ ∈ ¯Iλ,J,b, θ ∈ Oα and r ∈ [1, d − 1] satisfying x (θ,r) → x′. By Lemma 9.1, we have (2-a) wx(σk(α)) = σk(α) and hwx(σk(α)), µxi = 0 for k ∈ [1, d − 1]. We define g : P1 → G(L)/K by g(z) = gx Uα(zt−1) bxσUα(zt−1) · · · (bxσ)d−1 x bσ(gx) = bx = tµx wx as usual. Note that α, σi(α) are not in where g−1 the same connected component of ΦJ,α for i ∈ [1, r − 1] since Oα is of k=0 σk(α∨) = type I. As in Case(1.2), we have ηJ (g(0)) − ηJ (g(∞)) =Pr−1 yα ∈ π1(MJ ) and Uα(zt−1)K, g(z)−1bσ(g(z)) = KUα(zt−1)tµxUwx(α)(cσd(z)t−1) wxK, where c ∈ k[[t]]× is some constant. Again, it remains to verify g ∪λ′(cid:22)λKtλ′K ⊆ K\G(L)/K. So it suffices to show −1bσ(g) ⊆ (2-b) Uα(yt−1)tµxUw−x(α)(zt−1) ∈ ∪λ′(cid:22)λKtλ′K for y, z ∈ k[[t]]. Similar to (a) in the proof of Lemma 9.1 (using (2-a)), we have (2-c) hwx(α), µxi > 0. Moreover, hwx(α), µxi > 1 if hα, µxi 6 −1. Case(2.1): Φ∩(Zα +Zwx(α)) is of type A2 or A1 ×A1 or A1. Thanks to Lemma 7.14, to prove (2-b), it suffices to show that µx + α∨, µx − wx(α∨), µx + α∨ − wx(α∨) (cid:22) λ. Indeed, we already know µx + α∨ (cid:22) λ (since α ∈ Cλ,J,b,x) and µx + α∨ − wx(α∨) (cid:22) µx (cid:22) λ (by Lemma 7.10). If hwx(α), µxi > 1, we have µx − wx(α∨) (cid:22) λ as desired. Otherwise, 36 SIAN NIE hwx(α), µxi = hα, µxi = 0 by (2-c). Thus wx(α) = α by Lemma 7.9 and µx − wx(α∨) = µx − α∨ = sα(µx + α∨) (cid:22) λ as desired. Case(2.2): Φ ∩ (Zα + Zwx(α)) is of type B2. We show this case dose not occur. First note that α 6= ±wx(α) and hwx(α), α∨i = 0 since α, wx(α) are of the same length. By the assumption of Case(2.2), we have α − wx(α) ∈ ΦJ . However, hwx(α) − α, α∨i = −2, which contradicts that α∨ is J-minuscule. Case(2.3): Φ ∩ (Zα + Zwx(α)) is of type G2. We show this case also dose not occur. First note that Φ ∩ (Zα + Zwx(α)) is a connected component of Φ. So ΦJ ∩ (Zα + Zwx(α)) = {±ξ} for some ξ ∈ Φ+ J such that hξ, α∨i 6= 0 since wx(α) 6= α. Thus hξ, α∨i = −1 since α∨ is J-anti-dominant and J-minuscule. If ξ is a long root of Φ, so is α since hξ, α∨i = −1. Thus Φ ∩ (Zα ⊕ Zwx(α)) is of type A2, a contradiction. If ξ is a short root of Φ, then α is again a long root by the modified construction of Cλ,J,b,x (since Φ is a union of root systems of type G2, see Corollary 8.2), leading the same contradiction. (cid:3) [2, Lemma 4.7.3]) Assume Oα is of type II and Lemma 9.2. (cf. µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ N. If there exist θ ∈ Oα, r ∈ [d + 1, 2d − 1] and x′ ∈ ¯Iλ,J,b such that x (θ,r) ֌ x′, then we have (a) wx(σr−d(θ)) = σr−d(θ), hσr−d(θ), µxi = hσr(γ), µxi = 0 and wx(σr(γ)) = σr(γ); (b) wx(σd(θ)) = σd(θ) + γ and hwx(σd(θ)), µxi = 1; (c) γ 6= 0 if and only if hγ, µxi = 1; (d) wx(σi(θ)) = σi(θ), hσi(γ), µxi = 0 and hσi(θ), µxi = 0 for i ∈ [1, r − 1] − {r − d, d}; (e) hwx(σr−d(wx(σd(θ)))), µxi = hwx(σr(θ)) + σr(γ), µxi > 1. Here γ is the unique common neighbor of θ and σd(θ) in the Dynkin diagram of ΠJ ∪ Oα. Proof. (a) Since µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ N, we have (i) hσk( α), µyi > 0 for any y ∈ ¯Iλ,J,b and any k ∈ N. We claim that (ii) hwx(σr−d(θ)), µxi 6 0. If hwx(σr−d(θ)), µxi > 1, then hwx(σr−d(θ)), µx+θ∨−wx(σr(θ∨))i > 1 and (θ,r−d) → x + θ∨ − σr−d(θ∨) x (σr−d(θ),d) → x′, contradicting the assumption of the lemma. So (ii) is proved. Case(a1): hσr−d(θ), µxi 6 −1. Then µx+σr−d(θ∨), µx+θ∨+σr−d(θ∨) (cid:22) λ. Moreover, hσr(θ), µxi > 0 by (i), where θ = θ + γ + σd(θ). So we obtain (a1.0): hσr(θ), µxi > 0. Firstly, we show 37 (a1.1): hwx(σr(θ)), µx + σr−d(θ∨))i 6 0. Hence hwx(σr(θ)), µxi 6 1. Indeed, otherwise, we have µx + σr−d(θ∨) − wx(σr(θ∨)) (cid:22) µx + σr−d(θ∨) (cid:22) λ, and similarly, µx + θ∨ + σr−d(θ∨) − wx(σr(θ∨)) (cid:22) λ, which implies (σr−d(θ),d) → µx + σr−d(θ∨) − σr(θ∨) (θ,r−d) → x′, µx contradicting the assumption of the lemma. Thus (a1.1) is proved. Secondly, we show (a1.2): hwx(σr(θ)), µxi = 1. Assume otherwise, then 0 6 hσr(θ), µxi 6 hwx(σr(θ)), µxi 6 0 by (a1.0) and (a1.1), that is, hσr(θ), µxi = hwx(σr(θ)), µxi = 0. Applying Lemma 7.9 (d) and (c), wx(σr(θ)) = σr(θ) and hσr(γ), µxi = 0 since σr(γ) ∈ ΠJ and hσr(γ), σr(θ∨)i = −1. Thus hσr(θ), µxi 6 −1, which contradicts (i). So (a1.2) is proved. Thirdly, we show (a1.3): hwx(σr(θ)), µxi = hσr(θ) + σr(γ), µxi = 1 and wx(σr(θ)) ∈ {σr(θ), σr(θ) + σr(γ)}. If hσr(θ), µxi = 0, then hσr(γ), µxi = 1 since hσr(θ), µxi > 0 and µx is J-minuscule. Otherwise, we have hσr(θ), µxi = 1 = hwx(σr(θ)), µxi by (a1.0) and (a1.2), then hσr(γ), µxi = 0 by Lemma 7.9 (c). Moreover, If wx(σr(θ)) 6= σr(θ)+σr(γ), then by Lemma 7.9 (b), σr(θ) 6 wx(σr(θ)) < sσr(γ)(σr(θ)) = σr(θ) + σr(γ). Hence wx(σr(θ)) = σr(θ) since σr(γ) is a simple root. So (a1.3) is proved. Now we compute by (a1.3) that hσr(θ), µx′i = hσr(θ), µx − wx(σr(θ∨))i 6 −1, which is a contradiction. Therefore, Case(a1) dose not occur. So, by (ii), we have hσr−d(θ), µxi = hwx(σr−d(θ)), µxi = 0. Now (a) follows from Lemma 7.9 by noticing that hσr−d(θ), σr(γ)i = −1 if γ 6= 0. (b) If hwx(σd(θ)), µxi > 2, then x (θ,d) → x + θ∨ − σd(θ∨) (σd(θ),r−d) → x′, contradicting the assumption of the lemma. So we have (b1) hwx(σd(θ)), µxi 6 1. We claim that (b2) hσd(θ) + γ, µx + θ∨i > 0. Otherwise, we have µx + θ∨ (cid:22) µx + θ∨ (cid:22) λ, which contradicts the assumption of the lemma. So (b1) is proved. As a consequence, we have (b3) hσd(θ), µxi > 0. If γ = 0, (b3) follows directly from (b2). Otherwise, hγ, θ∨i = −1 and hγ, µxi 6 1 since µx is J-minuscule. So (b3) still follows from (b2). If hwx(σd(θ)), µxi 6 0, then hσd(θ), µxi = hwx(σdθ), µxi = 0 by (b3). Hence hγ, µxi = 0 by Lemma 7.9 (c). One computes hσd(θ) + γ, µx + θ∨i 6 −1, which contradicts (b2). Therefore, hwx(σd(θ)), µxi = 1 38 SIAN NIE by (b1). Combining with (b3), we have 0 6 hσd(θ), µxi 6 1. hσd(θ), µxi = 0, then hγ, µxi = 1 by (b2). hwx(σd(θ)), µxi, then hγ, µxi = 0 by Lemma 7.9. In a word, we have If If hσd(θ), µxi = 1 = (b4) hwx(σd(θ)), µxi = hσd(θ) + γ, µxi = 1. It remains to show wx(σd(θ)) = σd(θ) + γ. Assume otherwise. Simi- larly as (a1.3), we deduce that wx(σd(θ)) = σd(θ) and hγ, µxi = 0. So γ 6= 0 and hσd(θ), θ∨i = 0. Then hwx(σd(θ)), µx + θ∨ − wx(σr(θ∨))i = hwx(σd(θ)), µx + θ∨i = 1 by (b4), which implies (θ,d) → x + θ∨ − σd(θ∨) x (σd(θ),r−d) → x′, which is a contradiction and (b) is proved. (c) Assume hγ, µxi = 0 and γ 6= 0. Then hσd(θ), θ∨i = 0. By (b), we → x′, (θ,d) → x + θ∨ − σd(θ∨) (σd(θ),r−d) have hσd(θ), µxi = 1, which implies x a contradiction. (d) It follows from Corollary 7.12 and Lemma 7.9. (e) First note that wx(σr−d(wx(σd(θ)))) = wx(σr(θ)) + σr(γ) by (b) and (a). Assume hwx(σr−d(wx(σd(θ)))), µxi = hwx(σr(θ)), µxi 6 0. We claim that µx + σr(θ∨) (cid:22) λ. If hσr(θ), µxi = 0, then hwx(σr(θ)), µxi = hσr(θ), µxi = 0, which means, by Lemma 7.9, wx(σr(θ)) = σr(θ) and µx + σr(θ∨) = sσr(θ)(µx − wx(σr(θ∨))) (cid:22) λ. If hσr(θ), µxi 6 −1, we also have µx + σr(θ∨) (cid:22) µx (cid:22) λ and the claim is proved. By (a), hσr−d(γ) + σr−d(θ), µx + σr(θ∨)i = −1. Thus µx + σr−d(θ∨) (cid:22) µx + σr(θ∨) (cid:22) λ, which is a contradiction. So we have hwx(σr(θ)), µxi > 1 as desired. (cid:3) Lemma 9.3. (cf. [2, Lemma 4.7.12]) Assume Oα is of type II and µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ N. If there do not exist θ ∈ Oα, r ∈ [1, 2d − 1] and x′ ∈ ¯Iλ,J,b satisfying x (θ,r) → x′, then (a) hwx(σd(α)), µxi = 1; (b) wx(σd(α)) = σd(α) + β; (c) β = 0 if and only if hβ, µxi = 1; (d) hwx(σd(α)), α∨i = −1. (e) hα, µxi > −1. Here β is the unique common neighborhood of α and σd(α) in the Dynkin diagram of ΠJ ∪ Oα. Proof. First we have hwx(σd(α)), µxi 6 1 and (i): hσd(α)+β, µx+α∨i > 0 as in the proof of Lemma 9.2 (b). In particular, hσd(α), µxi > 0. If hwx(σd(α)), µxi 6 0, then hσd(α), µxi = hwx(σd(α)), µxi = hβ, µxi = 0 by Lemma 7.9, which contradicts (i). So we have hwx(σd(α)), µxi = 1 and 0 6 hσd(α), µxi 6 1. Now (a), (b) and (c) follow similarly as in the proof Lemma 9.2 (b). (d) follows from (b). Note that h α, µxi = hα, µxi + 1 > 0. So (e) follows. (cid:3) Lemma 9.4. (cf. [2, Lemma 4.7.13]) Keep the assumptions of Lemma 9.3. Then hwx( α), α∨i > −1 and hwx( α), µx + α∨i > 1. 39 Proof. First we have hwx( α), α∨i > −1 since wx( α) 6= −α and Φ is simply laced. Note that α is J-anti-dominant and J-minuscule. By [2, Lemma 4.6.1], there exists an orthogonal subset D ⊂ Φ+ J such that wx( α)− α =Pγ∈D γ∨. Moreover, hγ, µxi = 1 for each γ ∈ D by Lemma 7.9 (a) (see [2, page 55 of 66]). Note that h α, µxi = 1 + hα, µxi > 0 (by Lemma 9.3 (a), (b), (e)) and h α, α∨i = 1. We have hwx( α), µx + α∨i = h α +Xγ∈D as desired. γ∨, µx + α∨i > h α, µx + α∨i > 1 (cid:3) Proof of Proposition 8.3 when Oα is of type II. Note that α is J-anti- dominant and J-minuscule since Φ is simply laced. If µy + σk( α∨) (cid:22) λ for some y ∈ ¯Iλ,J.b and k ∈ Z, then σk( α) ∈ Cλ,J,b,y and Oα is of type I by noticing that yσk(α) = yα ∈ π1(MJ ). So the proposition follows from the type I case. We assume from now on that µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J.b and any k ∈ Z. Case(1): There do not exist θ ∈ Oα, r ∈ [1, 2d − 1] and x′ ∈ ¯Iλ,J,b satisfying x (θ,r) → x′. Case(1.1): wx( α) 6= α. Let g : P1 → G(L)/K be defined as in [2, §4.7.14]. Following the computation in loc. cit, we obtain, by Lemma 9.1, Lemma 9.3 and Lemma 9.4, that ηJ (g(0)) − ηJ (g(∞)) = yα and g(z)−1bσ(g(z)) ∈ KUα(−zt−1)Uwx(α)(−c′ where c′ K\G(L)/K, it suffices to show dσd(y)σ2d(y)thwx(α),µxi−1)tµxK, −1bσ(g) ⊆ ∪λ′(cid:22)λKtλ′K ⊆ d ∈ k[[t]]× is some constant. To verify g (a) Uα(yt−1)Uwx(α)(zthwx(α),µxi−1)tµx ∈ ∪λ′(cid:22)λKtλ′K for y, z ∈ k[[t]]. If hwx( α), α∨i > 0, then Uwx(α) and Uα commutes and Uα(yt−1)Uwx(α)(zthwx(α),µxi−1)tµx = Uwx(α)(zthwx(α),µxi−1)Uα(yt−1)tµx ∈ KtµxK ∪ Ktµx+α∨ K, where the inclusion follows from hwx( α), µxi > 1 by Lemma 9.4. Oth- erwise, we have hwx( α), α∨i = −1 and hwx( α), µxi > 2 by Lemma 9.4. One checks that Uα(yt−1)Uwx(α)(zthwx(α),µxi−1)tµx ∈ KUwx(α)+α(cyzthwx(α),µxi−2)Uα(zt−1)tµx ⊆ KUα(zt−1)tµx ⊆ KtµxK ∪ Ktµx+α∨ K, where c ∈ k[[t]] is some constant. Therefore, (a) is proved. Case(1.2): wx( α) = α. If hα, µxi = −1, the proposition follows from the construction in [2, §4.7.16]. Otherwise, h α, µxi > 1 by Lemma 9.3 (b) (e), and let g : P1 → G(L)/K be defined as in [2, §4.7.14]. 40 SIAN NIE Following the computation in loc. cit, we obtain, by Lemma 9.1 and Lemma 9.3, that ηJ (g(0)) − ηJ (g(∞)) = yα and g(z)−1bσ(g(z)) = KUα(−zt−1)U α(−c′ dσd(z)σ2d(z)thα,µxi−1)tµxK = KUα(−zt−1)tµxK ∈ KtµxK ∪ Ktµx+α∨ K, where c′ from that Uα commutes with U α (since h α, α∨i = 1). d ∈ k[[t]]× is some constant, and the second equality follows Case(2): x (θ,r) → x′ for some θ ∈ Oα, r ∈ [1, 2d − 1] and x′ ∈ ¯Iλ,J,b. Similar to the Case(1) in the proof of the type I case, it suffices to show that µx (b) and Q ∈ X MJ i=0 σi(θ∨). µx′ (b) such that P ∼λ,b Q (b) there exist P ∈ X MJ and ηJ (P ) − ηJ (Q) =Pr−1 By Lemma 7.5, we may assume x (θ,r) ֌ x′. If r ∈ [1, d], (b) follows from Proposition 7.8. Assume r ∈ [d + 1, 2d − 1]. Let g : P1 → G(L)/K be as in the proof of [2, Lemma 4.7.2]. Following the computation in [2, §4.7.5], we obtain by Lemma 9.2 that ηJ (g(0)) − ηJ (g(∞)) = Pr−1 i=0 σi(θ∨) and g(z)−1bσ(g(z)) = KUα(−zt−1)Uσr−d(α)(−cr−dσr−d(z)t−1)Uwx(σr (α))+σr (γ)(crthwx(σr (α))+σr (γ),µxi) × Uσr−d(α)(cr−dσr−d(z)t−1)tµx wxK K, ∈ ∪λ′(cid:22)λKtλ′ where cr−d, cr ∈ k[[t]]× are some constants and the last inclusion follows from Lemma 9.2 (e). (cid:3) Lemma 9.5. (cf. [2, Lemma 4.7.6]) Assume Oα is of type III and µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ N. If there exist θ ∈ Oα, r ∈ [2d, 3d − 1] and x′ ∈ ¯Iλ,J,b such that x (θ,r) ֌ x′, then (a) If r = 2d, then hσd(θ), µxi = 0 and hσ2d(θ), µxi > 1; (b) If r ∈ [2d + 1, 3d − 1], then hγ, µxi = 1, hσr(γ), µxi = 0, hσr(θ), µxi > 1 and hσi(θ), µxi = 0 for i ∈ {d, 2d, r − d, r − 2d}; (c) wx(σi(θ)) = σi(θ) and hσi(γ), µxi = hσi(θ), µxi = 0 for i ∈ [1, r − 1] − {d, 2d, r − d, r − 2d}. Here γ ∈ ΠJ is the common neighbor of θ, σd(θ) and σ2d(θ). Proof. Note that Φ = ΦJ,α, whose connected components are of type D4. Thus {θ, σd(θ), σ2d(θ)} is an orthogonal set. We set θ = θ+σd(θ)+ σ2d(θ) + γ. (a) First we claim that (i) µx + θ∨ − σ2d(θ∨) (cid:22) λ. Indeed, one checks that µx+θ∨−σ2d(θ∨) = µx+θ∨ −wx(σ2d(θ∨)) (cid:22) λ if hγ, µxi = 0 and µx + θ∨ − σ2d(θ∨) = sγ(µx + θ∨ − wx(σ2d(θ∨))) (cid:22) λ if hγ, µxi = 1. So (i) is proved. 41 (σd(θ),d) → x + σd(θ∨) − σ2d(θ∨) If hσd(θ), µxi 6 −2, then x (θ,d) → x′, which contradicts the assumption of the lemma. So hσd(θ), µxi > −1. Similarly, using (i) we have hσd(θ), µxi 6 0. Assume hσd(θ), µxi = −1. If hσ2d(θ), µxi 6 0, then hσd(θ) + σ2d(θ) + γ, µx + θ∨i 6 −1 and µx + θ∨ (cid:22) µx + θ∨ (cid:22) λ, a contradiction. So we have hσ2d(θ), µxi > 1 (σd(θ),d) and µx − σ2d(θ∨) + σd(θ∨) (cid:22) µx − σ2d(θ∨) (cid:22) λ. Then x x + σd(θ∨) − σ2d(θ∨) (θ,d) → x′, a contradiction. Thus hσd(θ), µxi = 0. → Assume hσ2d(θ), µxi 6 0. Then hσ2d(θ), µx + θ∨ + σd(θ∨) + γ∨i 6 −1 and hσd(θ) + σ2d(θ) + γ, µx + θ∨ − σ2h(θ∨)i 6 −1, which implies µx + θ∨ (cid:22) µx + θ∨ + σd(θ∨) + γ∨ (cid:22) µx + θ∨ − σ2h(θ∨) (cid:22) λ, which is a contradiction. So hσ2d(θ), µxi > 1 as desired. (b) Similar to the proof of (a), we have hσd(θ), µxi = hσ2d(θ), µxi = 0. If hγ, µxi = 0, then hσd(θ) + σ2d(θ) + γ, µx + θ∨i = −1, which means µx + θ∨ (cid:22) µx + θ∨ (cid:22) λ, a contradiction. So hγ, µxi = 1. Assume hσr(γ), µxi = 1. If hσr−d(θ), µxi > 0, then hwx(σr−d(θ)), µxi = hσr−d(θ) + σr(γ), µxi > 1 and µx + θ∨ − wx(σr(θ∨)) − wx(σr−d(θ∨)) (cid:22) µx + θ∨ − wx(σr(θ∨)) (cid:22) λ, (θ,r−d) (σr−d(θ),d) → x + θ∨ − σr−d(θ∨) → x′, a contradiction. which implies x So hσr−d(θ), µxi 6 −1. Similarly, hσr−2d(θ), µxi 6 −1. If hσr(θ), µxi > → x′, a 1, then one checks that x contradiction. So hσr(θ), µxi 6 0 and hµx, σr(θ)i = −1, which is a contradiction. Therefore, the case hσr(γ), µxi = 1 dose not occur. → x + σr−d(θ∨) − σr(θ∨) (σr−d(θ),d) (θ,r−d) Now we have hσr(γ), µxi = 0 and wx(σi(θ)) = σi(θ) for i ∈ r + Zd. Then one deduces that hσr−d(θ), µxi = hσr−2d(θ), µxi = 0 as in the proof of (a). If hσr(θ), µxi 6 0, then hσr(θ), µx′i = hσr(θ), µx − σr(θ∨)i 6 −1, a contradiction. So we have hσr(θ), µxi > 1 as desired. (c) It follows from Corollary 7.5 and Lemma 7.9. (cid:3) Lemma 9.6. (cf. Case 3 in the proof of [2, Lemma 4.7.10]) Assume Oα is of type III and µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ N. If there do not exist θ ∈ Oα, r ∈ [1, 3d − 1] and x′ ∈ ¯Iλ,J,b (θ,r) → x′, then hβ, µxi = 1, hσd(α), µxi = hσ2d(α), µxi = 0 satisfying x and hα, µxi > −1. Here β is the common neighborhood of α, σd(α) and σ2d(α) in the Dynkin diagram of ΠJ ∪ Oα. Proof. As in the proof of Lemma 9.5, we have hσd(α), µxi, hσ2d(α), µxi 6 0. If hσd(α)+σ2d(α), µxi 6 −1, then hσd(α)+σ2d(α)+β, µx+α∨i 6 −1 and µx + α∨ (cid:22) µx + α∨ (cid:22) λ, a contradiction. So hσd(α), µxi = hσ2d(α), µxi = 0. Again, by the inequality hσd(α) + σ2d(α) + β, µx + 42 SIAN NIE α∨i > 0, we have hβ, µxi = 1. Finally, we have hα, µxi > −1 since h α, µxi > 0. (cid:3) Proof of Proposition 8.3 when Oα is of type III. If µy + σk( α∨) (cid:22) λ for some y ∈ ¯Iλ,J,b and k ∈ Z, the proposition follows similarly as in the type II case. We assume from now on that µy + σk( α∨) (cid:14) λ for any y ∈ ¯Iλ,J,b and any k ∈ Z. Case(1): There do not exist θ ∈ Oα, r ∈ [1, 3d − 1] and x′ ∈ ¯Iλ,J,b (θ,r) → x′. Let g : P1 → G(L)/K be defined as [2, page 60]. satisfying x Following the computation in loc. cit., we obtain, by Lemma 9.1 and Lemma 9.6, that ηJ (g(0)) − ηJ (g(∞)) = yα and −1(z)bσ(g(z)) g = KUα+β(c3dσ3d(z)thα+β,µxi) × Uα+σd(α)+σ2d(α)+2β(−c′′σd(z)σ2d(z)σ3d(z)thα+β,µxi) × Uα+σ2d(α)+β(c′σ2d(z)σ3d(z)thα+β,µxi)Uα(−zt−1)tµxK = KUα(−zt−1)tµxK ∈ ∪λ′(cid:22)λKtλ′ K ⊆ K\G(L)/K, where c3d, c′, c′′ ∈ k[[t]] are some constants, and the second inclusion follows from hα + β, µxi > 0 by Lemma 9.6. Case(2): x (θ,r) → x′ for some θ ∈ Oα, r ∈ [1, 3d − 1] and x′ ∈ ¯Iλ,J,b. As in the type I case, it suffices to show that (a) there exist P ∈ X MJ µx (b) and Q ∈ X MJ µx′ (b) such that ηJ (P ) − i=0 σi(θ∨). ηJ (Q) =Pr−1 By Lemma 7.5, we may assume x If r ∈ [1, 2d − 1], (a) follows from Proposition 7.8. Now we assume r ∈ [2d, 3d − 1]. Let g : P1 → G(L)/K be defined by (θ,r) ֌ x′. g(z) = gx (bxσ)r−1 Uθ(zt−1)(bxσ)r−2 Uθ(zt−1) · · · Uθ(zt−1)K, where gx ∈ MJ (L) satisfying g−1 one computes that ηJ (g(0)) − ηJ (g(∞)) =Pr−1 −1bσ(g) ⊆ ∪λ′(cid:22)λKtλ′K ⊆ K\G(L)/K. show g x bσ(gx) = bx = tµx wx as usual. Then k=0 σk(θ∨). It remains to Case(2.1): r = 2d. If hγ, µxi = 1, by Lemma 9.1 and Lemma 9.5 (a), we have −1(z)bσ(g(z)) g = KUθ(−zt−1)Uσ2d(θ)+σd(θ)+γ(cσd(z)σ2d(z)thσ2d (θ),µxi)Uσ2d(θ)(c2dσ2d(z)thσ2d (θ),µxi)tµx wxK = KUθ(−zt−1)tµxK ∈ ∪λ′(cid:22)λKtλ′ K. If hγ, µxi = 0, then −1(z)bσ(g(z)) = KUθ(−zt−1)Uσ2d(θ)(c2dσ2d(z)thσ2d (θ),µxi−1)tµx wxK g = KUθ(−zt−1)tµxK ∈ ∪λ′(cid:22)λKtλ′ K. 43 Here c, c2d ∈ k[[t]] are some constants; the second equality in both cases follows from hσ2d(θ), µxi > 1 by Lemma 9.5 (a). Case(2.2): r ∈ [2d + 1, 3d − 1]. Then we have (see [2, page 53]) −1(z)bσ(g(z)) g = KUθ(−zt−1)Uσr (θ)+σr−d(θ)+σr−2d(θ)+σr (γ)(c′′σr(z)σr−d(z)σr−2d(z)thσr (θ),µxi−1) × Uσr(θ)+σr−d(θ)+σr (γ)(c′σr(z)σr−d(z)thσr (θ),µxi)Uσr(θ)(crσr(z)thσr (θ),µxi)tµx wxK = KUθ(−zt−1)tµxK ∈ ∪λ′(cid:22)λKtλ′ where c′, c′′, cr ∈ k[[t]] and the second equality follows from hσr(θ), µxi > 1 by Lemma 9.5 (b). (cid:3) K, References [1] M. Chen, Composantes connexes g´eom´etriques d'espaces de modules de groupes p-divisibles, Ann. Sci. ´Ec. Norm. Sup´er. (4) 47 (2014), 723 -- 764. [2] M. Chen, M. Kisin and E. Viehmann, Connected components of affine Deligne-Lusztig varieties in mixed characteristic, Compositio Math. 1697 -- 1762. [3] Q. Gashi, On a conjecture of Kottwitz and Rapoport, Ann. Sci. ´Ecole Norm. Sup. 43 (2010), 1017 -- 1038. [4] U. Gortz, T. Haines, R. Kottwitz, D. Reuman, Affine Deligne-Lusztig varieties in affine flag varieties, Compositio Math. 146 (2010), 1339-1382. [5] U. Gortz, X. He and S. Nie, P -alcoves and nonemptiness of affine Deligne- Lusztig varieties, Ann. Sci. ´Ecole Norm. Sup. 48 (2015), 647 -- 665. [6] X. He, Geometric and homological properties of affine Deligne-Lusztig varieties, Ann. of Math. (2) 179 (2014), no. 1, 367 -- 404. [7] , On the connected components of closed affine Deligne-Lusztig varieties, preprint. [8] X. He and S. Nie, Minimal length elements of extended affine Weyl groups, Compositio Math. 150 (2014), 1903 -- 1927. [9] , On the acceptable elements, arXiv:1408.5836. [10] M. Kisin, Mod p points on Shimura varieties of abelian type, preprint. [11] R. Kottwitz and M. Rapoport, On the existence of F -crystals, Comment. Math. Helv. 78 (2003), 153 -- 184. [12] S. Nie, Fundamental elements of an affine Weyl group, Math. Ann. 362 (2015), 485 -- 499. [13] E. Viehmann, Connected components of affine Deligne-Lusztig varieties, Math. Ann. 340 (2008), 315 -- 333. Institute of Mathematics, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, 100190, Beijing, China E-mail address: [email protected]
1101.0868	1	1101	2011-01-05T02:08:16	Terminal Resolutions of Brauer Pairs	[ "math.AG" ]	A Brauer pair is a pair (X, {\alpha}) where X is a quasi-projective variety over an algebraically closed field and {\alpha} is an element in the 2-torsion part of the Brauer group of the function field of X. A Brauer pair (Y, {\alpha}) is a terminal pair if the Brauer discrepancy of (Y, {\alpha}) is positive. We show that given a Brauer pair (X, {\alpha}), there is a terminal pair (Y, {\alpha}) with a birational morphism Y -> X. In short, any Brauer pair admits a terminal resolution.	math.AG	math	Terminal Resolutions of Brauer Pairs Basil Nanayakkara [email protected] Abstract A Brauer pair is a pair (X, α) where X is a quasi-projective variety over an algebraically closed field and α is an element in the 2-torsion part of the Brauer group of the function field of X. A Brauer pair (Y, α) is a terminal pair if the Brauer discrepancy of (Y, α) is positive. We show that given a Brauer pair (X, α), there is a terminal pair (Y, α) with a birational morphism Y −→ X. In short, any Brauer pair admits a terminal resolution. 1 Introduction The problem considered in this article is central to research in the area of maximal orders on algebraic varieties. The works of Daniel Chan, Colin Ingalls and Rajesh Kulkarni ([2], [3], [4], [8]), considered various questions regarding orders on algebraic surfaces. The main idea was to build a minimal model program for maximal orders on surfaces. This was successfully estab- lished in a seminal article in the area by Daniel Chan and Colin Ingalls [1]. In this article, they define the notion of terminal orders (which are analogs of smooth surfaces) and then prove the main theorem: any maximal order has a terminal resolution. This led to several articles in which terminal models of orders on surfaces were classified. This classification has been a significant achievement of the past decade. In Section 2, we explain how an element in the Brauer group of the function field of a variety induces a boundary divisor, via a complex that appears in the coniveau spectral sequence of the variety. In Section 3, we describe birational geometry of Brauer pairs after introducing the notion of Brauer discrepancy of a pair. Finally, in Section 4, we prove the main result that any Brauer pair admits a terminal resolution. 2 Logarithmic pairs from Brauer pairs Level 1 terms of the coniveau spectral sequence for a smooth algebraic variety X over an algebraically closed field k were written by Grothendieck [6]: One has Ei,j 1 = Mx∈X (i) H j−i(k(x), µ⊗(1−i) n ) where µn is the group of nth roots of unity in k, and X (i) is the set of all irreducible subvarieties of X of codimension i. The cohomology mentioned is Galois cohomology. The tensor product is over Z/nZ. By definition, µ−1 n := Hom(µn, Z/nZ), and we write µ⊗(−m) n )⊗m when m is positive. For more details, see Section 3.5 of [10]. for (µ−1 n Accordingly, if X is an irreducible 3-fold, we get Figure 1 on page 15 for the first quadrant of level 1. In that figure, D, C, pt are prime divisors, irreducible curves and points of X respectively. The row j=2 has the same form for any irreducible n-fold where n ≥ 2 with the interpretation that C and pt are irreducible subvarieties of codimension 2 and 3 respectively. Now, from row j = 2, we obtain the complex H 2(k(X), µn) −→MD H 1(k(D), Z/nZ) −→MC H 0(k(C), µ−1 n ) −→ 0. We know that H 2(k(X), µn) ∼= Brn(k(X)), where Brn(k(X)) is the n- torsion part of the Brauer group Br(k(X)) of k(X). (See section 4.4 of [5]). Therefore, we get the complex Brn(k(X)) −→MD H 1(k(D), Z/nZ) −→MC µ−1 n −→ 0. This tells us, in particular, that any element α in Br2(k(X)), induces a (possibly ramified) 2-sheeted cover or a 1-sheeted cover on each irreducible divisor D. Note that the ramifications must cancel on the irreducible sub- varieties C, since the sequence above is a complex. This simple observation will play an important role in our study. Definition 2.1 A Brauer pair is a pair (X, α) where X is a quasiprojective variety over an algebraically closed field and α is an element in Br2(k(X)). Let (X, α) be a Brauer pair. Then α induces a boundary divisor ∆X,α on X as follows: 2 Consider the complex Brn k(X) a−→MD H 1(k(D), Z/nZ) −→MC µ−1 n −→ 0 (2.2) that we obtained above, where D runs through all the irreducible divisors of X and C runs through all the irreducible subvarieties of codimension 2 of X. For a given irreducible divisor D, let a(α)D be the image of α indexed by D. Since H 1(k(D), Q/Z) classifies cyclic covers of D, a(α)D determines a ramified cover of D. Let eD be the degree of this cover. We define the boundary divisor ∆X,α to be, ∆X,α :=XD (cid:18)1 − 1 eD(cid:19) D where D runs through all the prime divisors of X such that a(α)D 6= 0. (See Section 3.3 of [1]). 3 Birational Geometry of Brauer Pairs In the following, by a divisor over X we mean an irreducible divisor E ⊆ Y where Y is a normal variety with a birational morphism Y −→ X. Two divisors D1, D2 of X are said to be numerically equivalent if D1 ·C = D2 · C for all irreducible curves C ⊆ X. In the definition below, by KZ,α we mean KZ + ∆Z,α, and ≡ denotes numerical equivalence. Definition 3.1 Let E be an irreducible exceptional divisor over X and α ∈ Br k(X). The Brauer discrepancy of the pair (X, α) along E, denoted by b(E, X, α), is the coefficient of E in the formula KY,α ≡ f ∗KX,α +Xi b(Ei, X, α)Ei where E is one of the exceptional divisors Ei of f where f : Y −→ X is a birational morphism. Definition 3.2 Let f : Y −→ X and g : Y ′ −→ X be birational morphisms. Suppose E ⊆ Y and E ′ ⊆ Y ′ are f -exceptional and g-exceptional divisors 3 respectively. We say that E and E ′ are isomorphic as divisors if there exist open sets U, U ′ containing E and E ′ respectively and an isomorphism φ : U −→ U ′ such that φE : E −→ E ′ is an isomorphism and φ = g−1 ◦ f on U \ {f -exceptional divisors}. The following lemma is the reason why we suppress the birational mor- phism f and the variety Y in the notation b(E, X, α) for the Brauer discrep- ancy of the pair (X, α) along E. The analogous remark for usual discrepancy (see below for the definition), is mentioned in Remark 2.23 of [9]. Lemma 3.3 b(E, X, α) does not depend on the particular birational mor- phism. Proof. Suppose f : Y −→ X and g : Y ′ −→ X are birational morphisms, E ⊆ Y and E ′ ⊆ Y ′ are f −exceptional and g−exceptional divisors re- spectively, that are isomorphic as divisors. Then there is an isomorphism φ : U −→ U ′ on a neighborhood U of E such that φ = g ◦ f −1 almost everywhere on U, and U ′ is a neighborhood of E ′. Now, we have, KY,α ≡ f ∗KX,α +Xi b(Ei, X, α)Ei + b(E, X, α)E where Ei are f −exceptional divisors, Ei 6= E. Similarly, KY ′,α ≡ g∗KX,α +Xj b(E ′ j, X, α)E ′ j + b(E ′, X, α)E ′ where E ′ Now, j are g−exceptional divisors, E ′ j 6= E ′. (KY,α − f ∗(KX,α))U = φ∗((KY ′,α − g∗(KX, α))U ′). Therefore, X b(Ei, X, α)EiU + b(E, X, α)E = φ∗(cid:16)X b(E ′ Hence b(E, X, α) = b(E ′, X, α). j, X, α)E ′ jU(cid:17) + b(E ′, X, α)E. For comparison purposes, and also for later use, we give the definition of the (usual) discrepancy of a log pair (X, ∆), which chronologically preceded the notion of Brauer discrepancy. (cid:3) 4 Definition 3.4 Let (X, ∆) be a logarithmic pair and E an exceptional divi- sor over X. The discrepancy of the pair (X, ∆) along E, denoted by a(E, X, ∆), is the coefficient of E in KY + f −1 ∗ ∆ ≡ f ∗(KX + ∆) +XEi a(Ei, X, ∆)Ei where E is one of the exceptional divisors Ei and f : Y −→ X is a birational morphism. This is the usual notion of discrepancy in algebraic geometry. The lemma below tells us how Brauer discrepancy and the (usual) dis- crepancy are related. Lemma 3.5 Let (X, α) be a Brauer pair, E an exceptional divisor over X, and ∆X,α the boundary divisor induced by α. Then, b(E, X, α) = a(E, X, ∆X,α) + 1 − 1 e where e is the degree of the cover on E induced by α. For a proof, see the proof of Proposition 3.15 of [1]. We define the Brauer discrepancy of the Brauer pair (X, α) to be, bdiscrep(X, α) := inf{e(E, X, α)·b(E, X, α) : E is an exceptional divisor over X}, where e(E, X, α) is the degree of the cover on E induced by α. Definition 3.6 A pair (X, α) is a terminal pair if bdiscrep(X, α) > 0. Definition 3.7 A terminal resolution of (X, α) is a birational morphism Y −→ X from a smooth variety Y such that the pair (Y, α) is a terminal pair. Now we can state our main theorem. Theorem 3.8 Any Brauer pair (X, α) has a terminal resolution. We will see that we can arrive at a terminal pair by successively blowing up (X, ∆X,α). Using Hironaka's theorem on resolution of singularities [7], we can im- prove (X, ∆X,α) such that X is smooth and ∆X,α is simple normal crossing. 5 Thus, ´etale-locally, the improved X has the form X = spec k{x1, x2, . . . , xn}. The boundary divisor is of the form, ∆X,α = n Xi=1 (cid:18)1 − 1 ei(cid:19) V (xi) where V (xi) is the hyperplane defined by xi = 0. I. e. V (xi) = spec k{x1, . . . , xi−1, xi+1, . . . , xn} with V (xi) ֒→ X the dual of the map k{x1, . . . , xn} → k{x1, . . . , xi−1, xi+1, . . . , xn} arrived at by setting xi = 0. The number ei is the degree of the cover on V (xi) induced by α. Since α is in the 2-torsion part of Br k(X), we have ei ∈ {1, 2}, by complex 2.2 of Section 2. Suppose D1 and D2 are prime divisors of X with covers on them. Let C = D1 ∩ D2. If one of the covers ramifies on C then the other also must ramify on C, since the sequence (2.2) is a complex. 4 Proof of the main result In this section we will prove the main result that given a Brauer pair (X, α), there is a smooth variety Y and a birational morphism Y −→ X such that (Y, α) is a terminal pair. First, we present a few lemmas. Lemma 4.1 Let X be a smooth variety, Z ⊆ X a smooth subvariety of codimension c and α ∈ Br k(X). Suppose p : BZX = Y −→ X is the blow- up of X along Z, E is the exceptional divisor. Then, b(E, X, α) = c − 1 e −Xi ai · multZDi where e is the degree of the cover on E determined by α and Pi aiDi is the boundary divisor on X determined by α. Proof. By the definition of Brauer discrepancy, b(E, X, α)E = KY + ∆Y,α − p∗(KX + ∆X,α) = KY − p∗KX + ∆Y,α − p∗∆X,α. 6 Now, KY − p∗KX = (c − 1)E and ∆Y,α − p∗(∆X,α) = (cid:18)1 − = (cid:18)1 − = (cid:18)1 − = (cid:18)1 − Thus, 1 1 e(cid:19) E + p−1 e(cid:19) E + p−1 e(cid:19) E −Xi e(cid:19) E −Xi 1 1 ∗ (∆X,α) − p∗∆X,α ∗ Xi aiDi! −Xi ai(p∗Di) ai(p∗Di − p−1 ∗ Di) ai · (multZDi)E. b(E, X, α)E = (c − 1)E +(cid:18)1 − b(E, X, α) = c − 1 e(cid:19) E − Xi ai · multZDi! E, and so ai · multZDi. (cid:3) 1 e −Xi The following lemma, which is the analog of Lemma 4.1 for usual discrep- ancy, appears as Lemma 2.29 in [9] without proof. Lemma 4.2 (Analogous to Lemma 4.1) Let X be a smooth variety and P aiDi is a boundary divisor on X. Let Z ⊆ X be a smooth subvariety of codimension c. Suppose p : BZX = Y −→ X is the blow up of X along Z, and E denotes the exceptional divisor. Then, a(E, X, ∆) = c − 1 −Xi ai · multZDi. Proof is similar to the proof of Lemma 4.1, and therefore omitted. Lemma 2.45 of [9] tells us that any exceptional divisor over a variety X can be reached by finitely many blow ups. This encourages the following definitions. 7 Definition 4.3 Let E be an exceptional divisor over a variety X. The divi- sor E is called a level n exceptional divisor, if n = inf(cid:8)m ∈ Z+ : E can be reached by m successive blow ups starting from X(cid:9) . Definition 4.4 A Brauer pair (X, α) is called level n Brauer terminal if b(E, X, α) > 0 for all level m exceptional divisors E over X for 1 ≤ m ≤ n. 4.1 Blowing up along a subvariety of codimension ≥ 3 In this section we show that if E is an exceptional divisor generated by a blow up of a subvariety of codimension ≥ 3, then b(E, X, α) is positive. Lemma 4.5 Let (X, α) be an n-dimensional Brauer pair and Z ⊆ X be a subvariety of codimension c. Let E be the exceptional divisor generated in the blow up of X along Z. If c ≥ 3, then b(E, X, α) is positive. Proof. As described towards the end of Section 3, ´etale locally (X, ∆X,α) is of the form (cid:16)spec k{x1, . . . , xn},Pi(cid:16)1 − 1 −Xi (cid:18)1 − b(E, X, α) = c − 1 eE ei(cid:17) V (xi)(cid:17) . Thus, by Lemma 4.1, ei(cid:19) · multZV (xi). 1 Since Z is of codimension c, it lies on at most c of V (xi). Thus, Xi (cid:18)1 − Therefore, 1 ei(cid:19) · multZV (xi) ≤ c −(cid:18) 1 ei1 + 1 ei2 + · · · + 1 eic(cid:19) . b(E, X, α) ≥ c − 1 eE −(cid:26)c −(cid:18) 1 + · · · + 1 eic(cid:19)(cid:27) ei1 1 eE 1 eic − ≥ 1 ei1 ≥ c · + · · · + 1 2 − 1 eE since eij = 1 or 2 > 0 since eE = 1 or 2 and c ≥ 3. (cid:3) 8 4.2 Blowing up along a subvariety of codimension 2 Here we show that if X is blown up along a subvariety of codimension 2, then either the Brauer discrepancy of (X, α) along the resulting exceptional divisor is positive, or X can be blown up to obtain X ′ so that (X ′, α) is level 1 Brauer terminal. Lemma 4.6 Let (X, α) be an n-dimensional Brauer pair with α2 = 1 and Z ⊆ X be a subvariety of codimension 2. Let E be the f -exceptional divi- sor where f : BlZ(X) −→ X is the blow up of X along Z. Then, either (1) b(E, X, α) > 0, or (2) X can be blown up to obtain a level 1 Brauer terminal pair (X ′, α). Proof. Locally, we have (cid:16)spec k{x1, . . . , xn},Pi(cid:16)1 − 1 codimension 2, it lies on at most two of V (xi). If Z lies on none or exactly one of V (xi), then by Lemma 4.1, ei(cid:17) V (xi)(cid:17). Since Z is of b(E, X, α) = 2 − 1 eE or b(E, X, α) = 2 − 1 eE −(cid:18)1 − 1 ei(cid:19) = 1 − 1 eE + 1 ei respectively. In either case, we have b(E, X, α) > 0, because eE, ei ∈ {1, 2}. Now, suppose Z lies on two of V (xi). Assume, without loss of generality, that Z lies on V (x1) and V (x2). Then, again by Lemma 4.1, b(E, X, α) = 2 − −(cid:18)1 − 1 e1 + 1 − 1 e2(cid:19) 1 eE 1 e2 + − = 1 e1 1 eE . Thus, we see that b(E, X, α) is positive except when e1 = e2 = 2 and eE = 1, in which case b(E, X, α) = 0. But this latter situation can occur only when the covers on V (x1) and V (x2) are double covers that do not ramify on V (x1, x2). (If the double covers ramify, then the induced cover on E must be double cover which makes eE = 2, a contradiction.) Whenever this situation occurs, we can obtain a new variety X ′ by blowing up along V (x1, x2). The resulting variety X ′ will not have two prime divisors with double covers on 9 them without the covers ramifying on the intersection of the prime divisors. Therefore, in the new variety this undesirable situation (e1 = e2 = 2 and eE = 1) does not occur. Hence (X ′, α) is level 1 Brauer terminal. (cid:3) 4.3 Completion of the Proof In this section we show that if (X, α) is level 1 Brauer terminal, then it is indeed Brauer terminal and complete the proof that any Brauer pair admits a terminal resolution. In the following lemma, a(E, X, ∆) denotes the (usual) discrepancy of the logarithmic pair (X, ∆) along an exceptional divisor E over X. Lemma 4.7 Let (X, α) be a Brauer pair such that X is smooth and ∆X,α is simple normal crossing. Let Z ⊆ X be an irreducible subvariety of codimen- sion c, where c ≥ 2. Suppose p : BZX −→ X is the blow up of X along Z and E ⊆ BZX the exceptional divisor. Then, a(E, X, ∆X,α) ≥ 0. Proof. ´Etale locally, we have ∆X,α = n Xi=1 (cid:18)1 − 1 ei(cid:19) V (xi). Applying Lemma 4.2, we see that a (E, X, ∆X,α) = c − 1 − n Xi=1 (cid:18)1 − 1 ei(cid:19) · multZV (xi). Since Z is of codimension c, it lies on maximum of c prime divisors V (xi). Thus, a(E, X, ∆X,α) ≥ c − 1 − c(cid:18)1 − 1 ei(cid:19) = c ei − 1 ≥ 0, since c ≥ 2 and ei ∈ {1, 2}. (cid:3) The following lemma can be considered as a Brauer version of the com- position of Lemma 2.29 and Lemma 2.30 of [9]. 10 Lemma 4.8 Let f : Y −→ X be a birational morphism, E is an f -exceptional divisor in Y , E0 an irreducible subvariety of E. Suppose g : Z −→ Y is the blow up of Y along E0 and F the g-exceptional divisor. Suppose b(E ′, X, α) ≥ 0 for all f -exceptional divisors E ′ and a(F, Y, ∆Y,α) ≥ 0. Let λ be a real number. If b(E, X, α) ≥ λ, then b(F, X, α) ≥ λ. Proof. Let ∆Y = f −1 ∗ ∆X,α −XE ′ a(E ′, X, ∆X,α)E ′ where the sum runs through all the f −exceptional divisors E ′. Then f∗∆Y = ∆X,α and KY + ∆Y ≡ f ∗(KX + ∆X,α). Therefore, we can apply Lemma 2.30 of [9], which gives, Now, by Lemma 4.2, we get, a(F, X, ∆X,α) = a(F, Y, ∆Y ). a(F, Y, ∆Y ) = c−1−"X ai · multE0(f −1 ∗ Di) −XE ′ where c = codimY E0 and ∆X,α =Pi aiDi. Now, ∗ Di) +XE ′ (cid:18)1 − ∆Y,α =Xi ai(f −1 a(E ′, X, ∆X,α) · multE0E ′# 1 eE ′(cid:19) E ′ where eE ′ is the degree of the cover on E ′ induced by α ∈ Br(k(X)) ∼= Br(k(Y )). Again, by Lemma 4.2, a(F, Y, ∆Y,α) = c−1−"X ai · multE0(f −1 ∗ Di) +XE ′ (cid:18)1 − 1 eE ′(cid:19) · multE0E ′# . Thus, we get, a(F, Y, ∆Y ) − a(F, Y, ∆Y,α) = XE ′ (cid:20)a(E ′, X, ∆X,α) +(cid:18)1 − 1 eE ′(cid:19)(cid:21) multE0E ′ b(E ′, X, α)multE0E ′. = XE ′ Since b(E ′, X, α) ≥ 0, b(E, X, α) ≥ λ and multE0E = 1, we have, a(F, Y, ∆Y ) − a(F, Y, ∆Y,α) ≥ λ. 11 Since a(F, Y, ∆Y,α) ≥ 0 by hypothesis, we get a(F, Y, ∆Y ) ≥ λ. But we proved earlier that a(F, X, ∆X,α) = a(F, Y, ∆Y ). Thus, a(F, X, ∆X,α) ≥ λ. (cid:3) This gives b(F, X, α) = a(F, X, ∆X,α) +(cid:16)1 − 1 eF(cid:17) ≥ λ. Theorem 4.9 Any Brauer pair admits a terminal resolution. Proof. Let (X, α) be a Brauer pair. Using Hironaka's desingularization theorem [7], we can assume that X is smooth and ∆X,α is simple normal crossing, where ∆X,α is the boundary divisor on X induced by α. If (X, α) is level 1 Brauer terminal then Lemma 4.8 with Lemma 4.7 show that (X, α) is Brauer terminal. If (X, α) is not level 1 Brauer terminal, then X can be blown up using Lemma 4.6 to obtain a level 1 Brauer terminal pair. This pair is Brauer terminal, again by Lemmas 4.7 and 4.8. (cid:3) In summary, we have shown that given a Brauer pair (X, α), we can associate to it a Brauer pair (Y, α) with the following properties: 1. Y is nonsingular; 2. There is a birational morphism f : Y −→ X; 3. The boundary divisor ∆Y,α induced on Y by α is simple normal crossing; 4. The Brauer discrepancy of any exceptional divisor over Y is positive. In short, any Brauer pair (X, α) admits a terminal resolution (Y, α) −→ (X, α). Remark. Note that in our analysis, we restricted α to be in the 2-torsion part of the Brauer group Br k(X). If we require α to be in Br3 (k(X)) instead, the analogous statement to the main result we proved here may not be true. For example, consider a 3-fold X with α ∈ Br3 k(X) that induces a simple normal crossing divisor ∆X,α that has the local form, 1 3 V (x1) + 1 3 V (x2) + 1 3 V (x3). Suppose the 3-sheeted covers on V (x1) and V (x2) ramify on V (x1, x2), but there is no ramification on V (x2, x3) and V (x1, x3). This pair (X, α) cannot 12 be improved by blowing up. Now, let E and Y be the exceptional divisor and the variety generated respectively, when X is blown up along V (x1, x3). Then, b(E, X, α) = 2 − 1 3 −(cid:18)1 − 1 3 + 1 − 1 3(cid:19) = 1 3 , by Lemma 4.1. Now blow up Y along E ∩V (x1), and let F be the exceptional divisor generated. Then, b(F, Y, α) = 2 − 1 eF −(cid:18)1 − 1 3 + 1 − 1 3(cid:19) = 2 3 − 1 eF where eF is the degree of the cover on F induced by α. We can show, using Brauer versions of Lemmas 2.29 and 2.30 of [9], that Then, we get b(F, X, α) = b(F, Y, α) + b(E, X, α). b(F, X, α) = 2 3 − 1 eF + 1 3 = 1 − 1 eF . Note that it is not posible to determine the degree eF of the cover on F induced by α without carrying out a detailed ramification computation in- volving roots of unity. If it happens that eF = 1, then we get B(F, X, α) = 0, indicating that the pair (X, α) may not admit a terminal resolution. How- ever, to determine the degree eF of the cover definitely, one must carry out ramification computations involving roots of unity. Acknowledgement: This paper is based on the author's PhD thesis. The author wishes to thank his supervisor, Colin Ingalls, for suggesting this prob- lem and also for teaching the necessary background material. References [1] D. Chan, C. Ingalls, The minimal model program for orders over surfaces, Invent. Math. 161, 427-452 (2005). [2] D. Chan, C. Ingalls, P. Hacking, Canonical sigularities of orders over surfaces, Proc. of the LMS, Vol. 98, p.83-115, (2009) 13 [3] D. Chan, R. Kulkarni, Numerically Calabi-Yau Orders on sur- faces, J. London Math Soc., Vol. 72(3), (2005), 571-584. [4] D. Chan, R. Kulkarni, Del Pezzo orders on projective surfaces, Advances in Math., 173, (2003), 144-177. [5] P. Gille, T. Szamuely, Central simple algebras and Galois coho- mology, Cambridge Studies in Advanced Mathematics 101, Cam- bridge University Press (2006). [6] A. Grothendieck, On the de Rham cohomology of algebraic vari- eties, Pub. Math. IHES 29, (1966). [7] H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic zero, Annals of Math. 79 (1964), 109-326. [8] C. Ingalls, Quantizable orders over surfaces, J. of Algebra, Vol. 207, Issue 2 (1998), 616-656. [9] J. Koll´ar, S. Mori, Birational geometry of algebraic varieties, Cam- bridge Tracts in Math. 134, Cambridge University Press (1998). [10] L. Le Bruyn, Noncommutative geometry and Cayley-smooth or- ders, Monographs and Textbooks in Pure and Applied Mathe- matics, Taylor and Francis group, Florida (2008). 14 Figure 1: Level 1 of the coniveau spectral sequence for an irreducible 3-fold X j 3 2 1 0 0 0 0 0 ✻ H 3(k(X), µn) LD H 2(k(D), Z/nZ) LC H 1(k(C), µ−1 n ) Lpt H 0(k(pt), µ⊗−2 n ) H 2(k(X), µn) LD H 1(k(D), Z/nZ) LC H 0(k(C), µ−1 n ) H 1(k(X), µn) LD H 0(k(D), Z/nZ) H 0(k(X), µn) 0 0 1 0 0 2 15 0 0 0 3 0 0 0 0 0 ✲ i
1711.09077	1	1711	2017-11-24T18:46:17	Semistability of Lazarsfeld-Mukai bundles via parabolic structures	[ "math.AG" ]	Our aim in this article is to produce new examples of semistable Lazarsfeld- Mukai bundles on smooth projective surfaces $X$ using the notion of parabolic vector bundles. In particular, we associate natural parabolic structures to any rank two (dual) Lazarsfeld-Mukai bundle and study the parabolic stability of these parabolic bundles. We also show that the orbifold bundles on Kawamata coverings of $X$ corresponding to the above parabolic bundles are themselves certain (dual) Lazarsfeld-Mukai bundles. This gives semistable Lazarsfeld-Mukai bundles on Kawamata covers of the projective plane and of certain K3 surfaces.	math.AG	math	SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES POORNAPUSHKALA NARAYANAN Abstract. Our aim in this article is to produce new examples of semistable Lazarsfeld- Mukai bundles on smooth projective surfaces X using the notion of parabolic vector bundles. In particular, we associate natural parabolic structures to any rank two (dual) Lazarsfeld-Mukai bundle and study the parabolic stability of these parabolic bundles. We also show that the orbifold bundles on Kawamata coverings of X corresponding to the above parabolic bundles are themselves certain (dual) Lazarsfeld-Mukai bundles. This gives semistable Lazarsfeld-Mukai bundles on Kawamata covers of the projective plane and of certain K3 surfaces. 7 1 0 2 v o N 4 2 ] . G A h t a m [ 1 v 7 7 0 9 0 . 1 1 7 1 : v i X r a 1. Introduction In recent years, the study of certain vector bundles called Lazarsfeld-Mukai bundles (LM bundles) has gained prominence. Investigating the properties of these naturally occurring bundles is of increasing interest. This can be attributed to the deep applications of these bundles in Brill-Noether theory of curves, especially those lying on K3 surfaces. LM bundles were first used by Lazarsfeld [11] to prove Petri's conjecture and by Mukai [16] in the classification of certain Fano manifolds. They have also been useful in studying the constancy of gonality, Clifford index and Clifford dimension of smooth projective curves belonging to ample or globally generated linear systems on K3 surfaces [5, 6, 9]. Voisin's proof of the generic Green's conjecture employed these bundles [21, 22], and Aprodu and Farkas [1] use LM bundles and their parameter spaces while proving Green's conjecture for curves on a K3 surface. The (semi)stability properties of LM bundles over K3 surfaces were studied by Lelli-Chiesa [13], and of certain LM bundles over Jacobian surfaces and higher dimensional varieties by us [18, 19]. In this article, our aim is to produce new examples of semistable Lazarsfeld-Mukai bundles via the general theory of parabolic vector bundles. In particular, we associate certain parabolic vector bundles to a rank two LM bundle and study the related notions of parabolic stability. We refer to § 2 for some preliminaries on parabolic vector bundles. Suppose X is a smooth projective surface over C and C i ֒−→ X is a smooth curve. Consider a globally generated line bundle A on C, with dim H 0(C, A) = 2. Then the rank two LM bundle associated to the pair (C, A) is the dual of the vector bundle F , where F is defined by the following short exact sequence: 0 → F → H 0(C, A) ⊗ OX → i∗A → 0 . Key words and phrases. (Semi)stability, Parabolic bundles. Mathematics Classification numbers: 14C20, 14E20, 14J60. 1 2 P. NARAYANAN Note that F comes in-built with a parabolic structure given by the following filtration with associated weights 0 ≤ a1 < a2 < 1: FC : F C ⊃a1 A ⊗ OX (−C)C ⊃a2 0 . We thus obtain a parabolic vector bundle F∗ = (F, FC, a1, a2) which we call the (dual) parabolic LM bundle. Our second parabolic bundle arises as follows. Consider a point q of multiplicity one on 0 ≤ b1 < b2 < 1: C. Consider the blow up of X at q, say eX. Let π : eX → X denote the blow down map and E ⊂ eX, the exceptional divisor. The pulled back rank two vector bundle eF := π∗F on eX admits a parabolic structure along the exceptional divisor E by associating weights Here, M = kernel(eF E → H 0(C, A) ⊗ OE). We call eF∗ := (eF ,eFE, b1, b2) the (dual) blown eFE : eF E = F (q) ⊗ OE ⊃b1 M ⊃b2 0 . up parabolic LM bundle. Refer § 3 for details. In § 4, we study the parabolic stability of the two parabolic vector bundles mentioned above. Let L be an ample line bundle on X. Then we have the following theorem. Theorem 1.1. Suppose that the vector bundle F is µL-stable. Then the (dual) parabolic LM bundle F∗ on X is parabolic µL-stable if the weights a1, a2 are such that: (a) a2 − a1 < 2 (b) a2 − a1 < 1 (C·L), if the intersection number (C · L) is even, or (C·L), if (C · L) is odd. sufficiently large. In this case, we have the following theorem. Next, consider the line bundle Ln := nπ∗L ⊗ O eX (−E) on eX which is ample for n all n ≥ n0, eF∗ is parabolic µLn-stable for all weights b1 and b2. Theorem 1.2. Suppose that F is µL-stable. Then, there is an integer n0 such that for In [3], Biswas established a two way relationship between parabolic vector bundles and orbifold vector bundles. Motivated by this, in § 5, we obtain the orbifold bundle on Kawamata covers of X associated to the (dual) parabolic LM bundle F∗ for weights of the form a1 = 0 and a2 = N −m N , where 1 ≤ m < N are positive integers. Let p : Y → X be the Kawamata covering of X such that p∗C is a non-reduced divisor NC ′ with C ′ := (p∗C)red a smooth curve on Y . Then we have the following theorem. Theorem 1.3. The orbifold vector bundle on Y corresponding to the (dual) parabolic LM bundle (F, FC, 0, N −m N ) on X is the dual LM bundle on Y , say F ′, associated to the triple (mC ′, A′, H 0(C, A)) given by the short exact sequence 0 → F ′ → H 0(C, A) ⊗ OY → j′ ∗A′ → 0 . Here A′ is the pullback of the line bundle A from C to the curve mC ′ mC ′ is non-reduced when m > 1). j ′ ֒−→ Y (the curve It is interesting to observe that the orbifold bundles are themselves certain dual LM bundles on the Kawamata covers. This further highlights that LM bundles are vector bundles with desirable properties. SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 3 As a consequence of the above correspondence, we obtain new examples of semistable LM bundles on the Kawamata covers of P2 and K3 surfaces in § 6. Theorem 1.4. Suppose X is a smooth projective surface and C is a smooth curve on X. Let A be an ample line bundle on C and V ⊂ H 0(C, A) be a general two dimensional subspace with the linear series PV base-point free. For any positive integer N, consider the Kawamata covering p : Y → X of X such that p∗C is a non-reduced divisor NC ′ with C ′ := (p∗C)red a smooth curve on Y . Let 1 ≤ m < N be an integer and A′ denote the pullback of A from C to (the possibly non-reduced curve) mC ′. (a) Consider X = P2 and C ∈ O(d), where d is odd. Set A to be of the form O(rd)C d. Then the LM bundle on Y associated for r ≥ 1. Choose m such that N −m to the triple (mC ′, A′, V ) is µp∗O(1)-semistable. N < 1 (b) Let X be a smooth projective K3 surface and L be an ample line bundle on X such that a general curve C ∈ L has genus g, Clifford dimension 1 and maximal gonality. Let A be a complete base-point free g1 d on C where the Brill Noether number ρ(g, 1, d) > 0. Set V = H 0(C, A). Let m be such that N −m g−1. Then the LM bundle on Y associated to the triple (mC ′, A′, H 0(C, A)) is µp∗L-semistable. N < 1 We have the following as particular cases of (a) and (b) respectively. Let l = deg A′. (a') Let d = 1, i.e. C ∈ O(1) is a line and set m = 1. Then there is an ir- l (OY (C ′)) corresponding to reducible component of the Brill-Noether variety G1 µp∗O(1)-semistable LM bundles. (b') Suppose X be a smooth projective K3 surface and L is as above which in addition satisfies (L2) = 2. Let m = 1. Then there is an irreducible component of the Brill-Noether variety G1 l (OY (C ′)) corresponding to µp∗L-semistable LM bundles. Acknowledgements. I thank Dr. Jaya NN Iyer for introducing me to this problem and for guiding me during the course of this project. I also thank Dr. T. E. Venkata Balaji and Prof. D. S. Nagaraj for helpful discussions. 2. Preliminaries 2.1. Parabolic sheaves. Parabolic bundles were first introduced over curves by Seshadri [20] and Mehta [15]. This was generalized to higher dimensions by Bhosle [2] and by Maruyama-Yokogawa [14]. Consider a connected smooth projective variety X and an effective divisor D on X. Definition 2.1. Let E be a coherent torsion-free OX -module. A quasi-parabolic structure on E along D is defined by giving a filtration of ED of the form: FD : ED = F1(ED) ⊃ F2(ED) ⊃ · · · ⊃ Fl(ED) ⊃ Fl+1(ED) = 0 . A parabolic structure on E is a quasi-parabolic structure as above together with a system of parabolic weights a1, a2, · · · , al such that 0 ≤ a1 < a2 < · · · < al < 1. We denote E with the parabolic structure by (E, FD, a∗) or by E∗. If E is locally free, then E∗ is called a parabolic bundle. 4 P. NARAYANAN In the following definitions, let E be a torsion-free coherent sheaf with parabolic struc- ture (E, FD, ai(E)) where i = 1, 2, . . . , l and FD : ED = F1(ED) ⊃ F2(ED) ⊃ · · · ⊃ Fl(ED) ⊃ Fl+1(ED) = 0 . Definition 2.2. [2, Definition 1.10] Suppose that G is another torsion-free coherent sheaf with quasi-parabolic structure along D prescribed by the filtration F′ D of GD: D : GD = F1(GD) ⊃ F2(GD) ⊃ · · · ⊃ Fr(GD) ⊃ Fr+1(GD) = 0 . F′ together with weights {ai(G)}. A morphism of parabolic sheaves G → E is a morphism of sheaves g : G → E such that whenever ai(G) > aj(E), then gD(Fi(GD)) ⊂ Fj+1(ED). Consider now a subsheaf F ֒→ E such that the quotient Q = E/F is torsion-free. In this case, F D → ED is an injection (cf. [2, § 1.8]). Then we have the following induced parabolic structure on the subsheaf F . Definition 2.3. [2, § 1.8] Let e : F D ֒→ ED denote the inclusion. We define a quasi- parabolic structure on F along D by pulling back the flag FD by e. FD : ED = F1(ED) F2(ED) · · · Fl(ED) 0? e−1(F1(ED)) = F D e−1(F2(ED)) · · · e−1(Fl(ED)) 0? Define the flag F′ D of F D by choosing the irredundant terms of the filtration and call them Fj(F D) where j ∈ J ⊂ {1, 2, . . . , l}. The weight aj(F ) associated to Fj(F D) is determined as follows. If i is the largest integer such that Fj(F D) = e−1(Fi(ED)), then define aj(F ) := ai(E). This parabolic structure on F is called the induced parabolic structure. With this definition, the inclusion F ֒→ E is in fact a morphism of parabolic sheaves (as in Definition 2.2). Hence F is parabolic subsheaf of E. The induced structure on F is the "maximum parabolic structure" making the inclusion morphism a morphism of parabolic sheaves. We next define the concepts of parabolic slope and parabolic stability. Let L be an ample line bundle on X with dim X = n. Definition 2.4. The parabolic weight of E denoted by µ-wt(E) is defined as: µ-wt(E) = lXi=1 ai(E)[r(Fi(ED)) − r(Fi+1(ED))]D · Ln−1. Here r(Fi(ED)) denotes the rank of Fi(ED). The motivation for the definition is the following. Let i : D ֒→ X denote the inclusion. Then c1(i∗Fi(ED)) = r(Fi(ED))D, cf. [2, Lemma 2.2]. Thus the above definition is the same as: µ-wt(E) = lXi=1 ai(E)[c1(i∗Fi(ED)) · Ln−1 − c1(i∗Fi+1(ED)) · Ln−1] . Definition 2.5. The parabolic slope of E with respect to L is: c1(E) · Ln−1 + µ-wt(E) parµL(E) = rank E . ? _ o o ? _ o o _ o o ? O O ? _ o o ? O O ? _ o o ? O O _ o o SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 5 Definition 2.6. Let E be a parabolic sheaf as above. Then E is said to be parabolic µL-semistable (resp. parabolic µL-stable) if for every subsheaf F of E with torsion-free quotient such that 0 < rank F < rank E equipped with the induced parabolic structure, one has the inequality of slopes, parµL(F ) ≤ parµL(E) (resp. parµL(F ) < parµL(E)) . 2.2. Orbifold sheaves. Let Y be a connected smooth projective variety of dimension n. Let Aut(Y ) denote the group of algebraic automorphisms of Y . Let G be a finite group and ρ : G → Aut(Y ) be an injective group homomorphism. Definition 2.7. An orbifold sheaf on Y is a coherent torsion-free sheaf E on Y together with a lift of the action of G to E. That is, G acts on the total space of stalks of E and the automorphism of the space of stalks for the action of any g ∈ G is a coherent sheaf isomorphism between E and ρ(g−1)∗E. If E is locally free, then E is called an orbifold bundle. A subsheaf F of an orbifold sheaf E with torsion-free quotient E/F is said to be G- saturated if F is left invariant by the action of G. Hence F has an induced orbifold sheaf structure. Let eL ∈ Pic(Y ) be an ample line bundle which is also an orbifold bundle. Recall that the usual Mumford-Takemoto slope of E with respect to eL is µ eL(E) = c1(E)· eLdim Y −1 Definition 2.8. An orbifold sheaf E is said to be orbifold semistable (resp. orbifold stable) if for all G-saturated subsheaves F of E such that 0 < rank F < rank E one has µ eL(F ) ≤ µ eL(E) (resp. µ eL(F ) < µ eL(E)). rank E . Note that if E is an orbifold sheaf which is µ eL- (semi)stable, then clearly E is orbifold (semi)stable. There is a converse statement in case of semistability which is as follows. Proposition 2.9. [3, Lemma 2.7] An orbifold bundle is orbifold semistable if and only if it is semistable in the sense of Mumford-Takemoto. Let p : Y → Y /G =: X be the projection. Assume that the quotient X is a smooth variety. Let E be an orbifold sheaf on Y . The direct image sheaf p∗E on X naturally has a G-action compatible with the trivial action of G on the quotient X. Let (p∗E)G denote the subsheaf of G-invariants in p∗E. Definition 2.10. Let Av denote the averaging Reynolds operator Av : p∗E → p∗E given by Av = 1 The image of Av is the invariant subsheaf (p∗E)G. Thus we have a short exact sequence: GPh∈G th. Here th is the endomorphism of p∗E induced by the action of h ∈ G. 0 → (p∗E)G → p∗E → p∗E (p∗E)G → 0 . When E is an orbifold bundle, the direct image sheaf p∗E is also locally free. In this case, since the above exact sequence splits, the subsheaf of invariants (p∗E)G is locally free as well. 3. Parabolic structures associated to LM bundles Consider a smooth projective surface X over the field C. For any smooth curve C i ֒−→ X, d on C, say A (i.e. A ∈ Picd(C) is base-point free consider a complete base-point free g1 6 P. NARAYANAN with h0(A) = 2). We then have the following short exact sequence corresponding to the linear series (A, H 0(C, A)) on C: (1) 0 −→ A∨ −→ H 0(C, A) ⊗ OC −→ A −→ 0 . The dual LM bundle FC,A =: F associated to (C, A) on X is given by the short exact sequence: (2) 0 −→ F −→ H 0(C, A) ⊗ OX −→ i∗A −→ 0 . Since F is a rank two vector bundle with determinant OX(−C), we have F ∨ ≃ F ⊗ (det F )∨ ≃ F ⊗ OX (C). Note that F ∨ is the LM bundle associated to (C, A) and sits in the following sequence: (3) 0 −→ H 0(C, A)∨ ⊗ OX −→ F ⊗ OX(C) −→ i∗(A∨ ⊗ OX (C)C) −→ 0 . We now associate parabolic vector bundles to the dual LM bundle F in two ways. 3.1. Parabolic LM bundle. We first associate a parabolic structure to the vector bundle F along the divisor C on X. Restrict (2) to the curve C to get the following exact sequence on C: 0 −→ A ⊗ OX(−C)C −→ F C −→ H 0(C, A) ⊗ OC −→ A −→ 0 . From the above sequence and (1), we also get (4) 0 −→ A ⊗ OX (−C)C −→ F C −→ A∨ −→ 0 . This gives a quasi-parabolic structure on F along the curve C, the flag being given by: FC : F C ⊃a1 A ⊗ OX (−C)C ⊃a2 0 . Here we have associated weights 0 ≤ a1 < a2 < 1 to get a parabolic vector bundle (F, FC, a1, a2). Let us denote F along with the above parabolic structure by F∗. We call this parabolic vector bundle the (dual) parabolic LM bundle. Remark 3.1. Dualizing the exact sequence (4), we get the following exact sequence on C, where EC,A = F ∨ is the LM bundle: 0 −→ A −→ (EC,A)C −→ (A ⊗ OX (−C)C)∨ −→ 0 . This gives a parabolic structure on the LM bundle EC,A, by associating weights 0 ≤ a1 < a2 < 1 given by (EC,A)C ⊃a1 A ⊃a2 0 . We call the above parabolic vector bundle the parabolic LM bundle. However, in this paper, we work with the dual parabolic LM bundle F∗. 3.2. Parabolic Blown up LM bundle. Next, we obtain a parabolic structure on the pullback of F to a blown up surface, along the exceptional divisor. Let X, C, A and F be as in the § 3.1. Consider a point q of multiplicity one on C. Consider the blow up of X at q, say eX. Let π : eX → X denote the blow down map, and E ⊂ eX, the exceptional divisor. Pull back the exact sequence (2) by π to the blown-up surface eX: 0 → π∗F =: eF → H 0(C, A) ⊗ O eX → π∗i∗A → 0 . SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 7 Restricting the above sequence to the exceptional divisor E, we get: (5) 0 → M → eF E ≃ F (q) ⊗ OE → H 0(C, A) ⊗ OE → A(q) ⊗ OE → 0 . Here M is the kernel of the restriction. On eX, we have the following commutative diagram, where eC is the strict transform of C in eX. E / eC + E = π∗C πC eX π C i / X Pulling back the short exact sequence (1) on C to π∗C = eC + E and restricting to E, we 0 → A∨(q) ⊗ OE → H 0(C, A) ⊗ OE → A(q) ⊗ OE → 0 . get: From the above short exact sequence and from (5), we get: 0 → M → F (q) ⊗ OE → A∨(q) ⊗ OE → 0 . The kernel M is in fact a line bundle of the form M = W ⊗ OE where W is a one dimensional vector space. We thus have the following parabolic structure on eF along E by associating weights 0 ≤ b1 < b2 < 1: eFE : eF E = F (q) ⊗ OE ⊃b1 M = W ⊗ OE ⊃b2 0 We denote the corresponding parabolic vector bundle by eF∗ and call it the (dual) blown up parabolic LM bundle. 4. Stability of the parabolic bundles In this section, we discuss the parabolic stability of the parabolic vector bundles F∗ on X and eF∗ on eX. 4.1. Parabolic stability of F∗. Consider a rank one coherent subsheaf G of F with torsion-free quotient F/G. Then, G is a line bundle, cf. [7, Prop. 1.1 & 1.9]. We give the induced parabolic structure on G, induced from the parabolic vector bundle F∗ defined by the parabolic structure F C ⊃a1 A ⊗ OX (−C)C ⊃a2 0. By the procedure explained in Definition 2.3, the induced structure is given as follows. Lemma 4.1. The induced parabolic structure on G along C is given by GC ⊃a1 0 or GC ⊃a2 0 . Proof. Let e′ : G ֒→ F denote the inclusion. Since F/G is torsion-free, the restriction e′C : GC ֒→ F C is also an injection. Denote e := e′C. The induced quasi-parabolic structure on G is given as under. 0 0 / A ⊗ OX(−C)C / F C / e−1(A ⊗ OX (−C)C) =: G′ e−1(F C) = GC / / / / / / / / / ? O O ? O O 8 P. NARAYANAN Since G′ = e−1(A ⊗ OX (−C)C) is a subsheaf of the line bundle GC, G′ is either of rank 0 or rank 1. • If G′ is of rank 0, then since G′ ֒→ A ⊗ OX(−C)C, it is torsion-free and hence G′ = 0. In this case, the parabolic structure on G is given by GC ⊃ 0 with weight a1. • If G′ is of rank 1, then the quotient GC G′ quotient the parabolic structure on G is given by GC ⊃ 0 with weight a2. A⊗OX (−C)C is of rank 0, and is a subsheaf of the ≃ A∨. Hence the quotient is 0 and G′ ≃ GC. In this case, F C Hence the induced parabolic structure on G is given by GC ⊃ai 0 for i = 1, 2. (cid:3) We now prove Theorem 1.1 regarding the stability of F∗ with respect to an ample line bundle L on X. Proof of Theorem 1.1. From Definition 2.6, we recall that in order to check parabolic sta- bility of F∗, it is enough to check the inequality of parabolic slopes for rank one subsheaves G of F with torsion-free quotient equipped with the induced structure. By Lemma 4.1, the induced parabolic structure G∗ is given by GC ⊃ai 0 for i = 1, 2. . Since F is µL-stable, µL(G) < µL(F ). Also µL(G) = We know that, µL(F ) = −C·L 2 c1(G) · L ∈ Z. Thereby, µL(G) ≤(cid:26) µL(F ) − 1 if (C · L) is even, if (C · L) is odd. µL(F ) − 1 2 The parabolic weights of F∗ and G∗ are given by (cf. Definition 2.4): µ-wt(F ) = (a1 + a2)(C · L) and µ-wt(G) = ai(C · L) for i = 1, 2 . Next we compute the parabolic slopes of F∗ and G∗. parµL(F ) = c1(F ) · L + µ-wt(F ) 2 = µL(F ) + (a1 + a2)(C · L) 2 and, parµL(G) = µL(G) + ai(C · L) where i = 1 or 2 . Since a1 < a2, it is enough to prove that µL(G) + a2(C · L) < parµL(F ) in order to show that F∗ is parabolic µL-stable. (a) C.L is even - In this case, our condition on the ai-s, i.e. a2 − a1 < 2 (C·L) gives −1 < (a1−a2)(C·L) 2 . Thus, µL(G) + a2(C · L) ≤ µL(F ) − 1 + a2(C · L) < µL(F ) + (a1 + a2)(C · L) 2 = parµL(F ) . (b) C.L is odd - In this case, the condition on ai-s is a2 − a1 < 1 (C·L). This gives − 1 2 < (a1−a2)(C·L) 2 . Thus, µL(G) + a2(C · L) ≤ µL(F ) − < µL(F ) + + a2(C · L) 1 2 (a1 + a2)(C · L) 2 = parµL(F ) . SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES By parts (a) and (b), F∗ is parabolic µL-stable. 9 (cid:3) Corollary 4.2. Let X be a smooth, projective K3 surface and L ∈ Pic(X) an ample line bundle such that a general curve C ∈ L has genus g, Clifford dimension 1 and maximal gonality k. Let the Brill-Noether number ρ(g, 1, d) > 0. For a general C ∈ L , consider the rank 2 dual LM bundle F := FC,A associated with a general complete, base-point free g1 d, say A on C. Then the dual Parabolic LM bundle (F, FC, a1, a2) is parabolic µL-stable, whenever a2 − a1 < 1 . g − 1 Proof. Lelli-Chiesa [13, Theorem 4.3] proves that such an F = FC,A is µL-stable. The proof follows from Theorem 1.1 by observing that on a K3 surface C · L = (C 2) = 2g − 2, which is even. (cid:3) Corollary 4.3. Let X = P2 C, the projective plane. Consider O(d) for an odd d > 0. Let C ∈ O(d) be a general smooth curve, and A = O(rd)C (for r > 0) on C. Let V ⊂ H 0(C, A) be a general two dimensional subspace of global sections. Consider the dual LM bundle F := FC,A,V associated to the triple (C, A, V ), and the associated parabolic structure on F , say (F, FC, a1, a2) where a2 − a1 < 1 d. Then (F, FC, a1, a2) is parabolic µO(1)-stable. Proof. From [19, Theorem 1.2, Remark 4.10], we observe that the dual LM bundle F is µO(1)-stable. Since L = O(1), for C ∈ O(d), we get C · L = d (which is odd in our case). The result thus follows. (cid:3) 4.2. Parabolic stability of eF∗. The line bundle π∗L is not ample on eX because it is trivial when restricted to the exceptional divisor. Consider the line bundle Ln on eX which is ample (by [4, Lemma 3]) for n sufficiently large, Ln = nπ∗L ⊗ O eX (−E) . [17, Prop. 2.2 (1)] for a proof. Remark 4.4. If F is a µL-stable vector bundle on X, then there exists an n0 ∈ Z such We study the parabolic stability of eF∗ with respect to the ample line bundle Ln. We first note the following about the µLn-stability of eF in the following Remark. that for all n ≥ n0, the vector bundle eF = π∗F is a µLn-stable vector bundle on eX. Refer Consider a rank 1 coherent subsheaf eG of eF with torsion-free quotient. By similar arguments as before, eG is a line bundle. We give the induced parabolic structure on eG induced from eF∗ which is defined by eF E ⊃b1 M = W ⊗ OE ⊃b2 0 . Lemma 4.5. The induced parabolic structure on eG along E is given either by eGE ⊃b1 0 or eGE ⊃b2 0 . Proof. The proof is exactly on the same lines as the proof of Lemma 4.1. (cid:3) We now prove Theorem 1.2 about the µLn-stability of eF∗ for n ≥ n0, where n0 is as in Remark 4.4). 10 P. NARAYANAN Proof of Theorem 1.2. Since the dual LM bundle F is µL-stable, by Remark 4.4, there is an integer n0 such that for all n ≥ n0, eF is µLn-stable. Note that, c1(eF ) · Ln If eG is any rank 1 subsheaf of eF with torsion-free quotient, then µLn(eG) < µLn(eF ) . Since µLn(eG) = c1(eG) · Ln ∈ Z, we get: µLn(eF ) = π∗OX(−C) · (nπ∗L − E) if (C · L) and n are odd , −n(C · L) ≃ = 2 2 2 . µLn(eG) ≤(cid:26) µLn(eF ) − 1 2 µLn(eF ) − 1 otherwise. As discussed in the previous theorem, in order to check for parabolic stability it is enough equipped with the induced parabolic structure. The induced parabolic structure on such to check the slope inequality for rank 1 subsheaves eG ⊂ eF with torsion-free quotient a eG is eGE ⊃bi 0 for i = 1, 2 . So the parabolic weights of eF and eG are: µ-wt(eF ) = (b1 + b2)E · (nπ∗L − E) = −(b1 + b2)E2 . Since (E2) = −1, we get µ-wt(eF ) = b1 +b2. Similarly, we get that µ-wt(eG) = biE ·Ln = bi. parµLn(eG) = µLn(eG) + µ-wt(eG) = µLn(eG) + bi . parµLn(eF ) = µLn(eF ) + We now compute the parabolic slopes, Since b1 < b2, it is enough to check that µLn(eG) + b2 < parµLn(eF ). Since 0 ≤ b1 < b2 < 1, we have b2 − b1 < 1. Again we have two cases. = µLn(eF ) + µ-wt(eF ) b1 + b2 2 2 . (a) Both n and (C · L) are odd - Note that, − 1 2 < b1−b2 2 . We now have µLn(eG) + b2 ≤ µLn(eF ) − < µLn(eF ) + + b2 1 2 b1 + b2 2 = parµLn(eF ) . (b) At least one of n or (C · L) is even - Note that b2−b1 2 < b2 − b1 < 1, which gives −1 < b1−b2 2 . We have µLn(eG) + b2 ≤ µLn(eF ) − 1 + b2 b1 + b2 < µLn(eF ) + 2 = parµLn(eF ) . Thus from parts (a) and (b), we have that the (dual) parabolic blown up LM bundle eF∗ is parabolic µLn-stable for all weights b1 and b2 and for n ≥ n0, where n0 is as in Remark 4.4. (cid:3) Corollary 4.6. Let X be a smooth, projective K3 surface and L ∈ Pic(X) an ample line bundle such that a general curve C ∈ L has genus g, Clifford dimension 1 and maximal gonality k. Let the Brill Noether number ρ(g, 1, d) > 0. For a general C ∈ L , consider the rank 2 dual LM bundle F := FC,A associated with a general complete, base-point free g1 d say A on C. Then, there is an integer n0 such that the dual parabolic blown up LM bundle (eF ,eFE, b1, b2) is parabolic µLn-stable, for all choices of weights and for all n ≥ n0. SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 11 Proof. This result again follows from Lelli-Chiesa [13, Theorem 4.3], whereby such an FC,A is µL-stable. (cid:3) Corollary 4.7. Let X = P2 C, the projective plane. Consider O(d) for an odd number d > 0. Let C ∈ O(d) be a general smooth curve, and A = OX(rd)C for r > 0 on C. Let V ⊂ H 0(C, A) be a general two dimensional subspace of global sections. Consider the dual LM bundle F := FC,A,V associated to the triple (C, A, V ), and the associated parabolic structure on eF , say (eF ,eFE, b1, b2). Then there is an integer n0 such that (eF ,eFE, b1, b2) is parabolic µLn-stable, where L = O(1) and for all n ≥ n0. Proof. The result follows from [19, Theorem 1.2] which proves that the dual LM bundle FC,A,V is µO(1)-stable. (cid:3) 5. Orbifold LM bundle Recall that the quasi-parabolic structure on the parabolic vector bundle F∗ is given by F C = F1(F C) ⊃ A ⊗ OX (−C)C = F2(F C) ⊃ 0 = F3(F C) . By [14, Definition 1.1], the quasi-parabolic structure can be alternatively described by a filtration of the vector bundle F itself of the form The elements of the two filtrations are related as follows (cf. [3, § 3]): F = F1(F ) ⊃ F2(F ) ⊃ F3(F ) = F (−C) . Fi(F C) = Fi(F ) F (−C) . We obtain the equivalent formulation of the quasi-parabolic structure on F∗ from the following lemma. Lemma 5.1. The quasi-parabolic structure on the (dual) parabolic LM bundle F∗ is equiv- alent to the quasi-parabolic structure given by the following filtration of F : F : F ⊇ H 0(C, A) ⊗ OX(−C) ⊇ F (−C) . Proof. By the discussion above, we just need to determine F2(F ). We also have, F2(F C) = F2(F )/F (−C) . Hence F2(F ) has to sit in the following short exact sequence: 0 → F (−C) → F2(F ) −→ i∗(A ⊗ OX (−C)C) → 0 . Tensoring the exact sequence (2) which defines F by OX(−C), we get 0 −→ F (−C) −→ H 0(C, A) ⊗ OX (−C) −→ i∗(A ⊗ OX(−C)C) −→ 0 . Thus, F2(F ) = H 0(C, A) ⊗ OX (−C). (cid:3) So, the parabolic vector bundle (F, FC, a1, a2) from § 3.1 can alternatively be described as (F, F, a1, a2) where the flag F is described in Lemma 5.1. Remark 5.2. Set a0 = a2 − 1 and a3 = 1, where F∗ = (F, F, a1, a2) is the (dual) parabolic LM bundle. Let [t] denote the integral part of any t ∈ R. For each t ∈ R, consider the following locally free sheaves Et = Fi(F )(−[t]C), where ai−1 < t − [t] ≤ ai. Note that the locally free sheaves {Et} give a filtration which is decreasing, i.e. if t ≥ t′, then Et ⊆ Et′. This filtration is left continuous. Further, this filtration has a jump at some t, i.e. for all 12 P. NARAYANAN ǫ > 0, Et+ǫ 6= Et if and only if t − [t] = ai where i = 1, 2. Finally, for all t ∈ R, we have Et+1 = Et(−C). This filtration {Et}t∈R completely determines the parabolic vector bundle (F, F, a1, a2), cf. [14] and [3]. In particular, the parabolic vector bundle F∗ is completely determined by the filtration of vector bundles Et = F for t ∈ [0, a1] = H 0(C, A) ⊗ OX (−C) for t ∈ (a1, a2] = F (−C) for t ∈ (a2, 1]. We recall the Kawamata covering Theorem [12, Prop. 4.1.12]. Given a positive integer N > 0, there is a non-singular connected projective variety Y and a finite Galois morphism p : Y → X with Galois group Γ = Gal(Rat(Y )/Rat(X)) such that p∗C is a non-reduced divisor of the form NC ′ with (p∗C)red = C ′. Here C ′ is a smooth curve on Y . We then have the following commutative diagram, for any m such that 1 ≤ m < N. C ′ g ❙ ❙ mC ′ NC ′ ❙ ❙ ❙ ❙ ❙ ● ● ● ● ● ❙ ❙ ❙ ψ ❙ ❙ ❙ ❙ f φ ● ● ● ❙ ❙ ● ❙ #● )❙ p C j i Y p / X Pulling back the exact sequence (1) to mC ′ by φ, we get: 0 → φ∗A∨ → H 0(C, A) ⊗ OmC ′ → φ∗A → 0 . Denote A′ = φ∗A. Then (A′, H 0(C, A)) is a base-point free linear system on the curve mC ′ (which is non-reduced when m > 1). We can thus consider the dual LM bundle F ′ on Y associated to the triple (mC ′, A′, H 0(C, A)) given by the following short exact sequence: (6) 0 → F ′ → H 0(C, A) ⊗ OY → j∗f∗A′ → 0 . Remark 5.3. The dual LM bundle F ′ on Y associated to the triple (mC ′, A′, H 0(C, A)) has a natural orbifold structure. Indeed, H 0(C, A) ⊗ OY → j∗f∗A′ is a morphism between orbifold sheaves on Y compatible with the action of Γ on both the sheaves. We call this orbifold bundle F ′ on Y "the orbifold LM bundle". By the construction of Biswas [3, § 2c], this orbifold LM bundle corresponds to a uniquely determined parabolic bundle on X. We now determine this parabolic vector bundle. For any t ∈ R, define, as in [3], (7) Et =(cid:16)p∗(cid:16)F ′ ⊗ OY ([−t · N]C ′)(cid:17)(cid:17)Γ . Just as in Remark 5.2, this defines a parabolic sheaf on X, cf. [3, § 2] and [14, § 1]. We first have the following preliminary lemma. Lemma 5.4. Let A be a line bundle on C. For any m such that 1 ≤ m < N, consider the line bundle L′ := ψ∗A ⊗ OY (−mC ′)C ′ on C ′. Then, with notations as in the commutative diagram below, the invariant sheaf (i∗ψ∗L′)Γ = (p∗j∗f ′ ∗g∗L′)Γ = 0. ∗h′ / / ) / / # / / / SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 13 C ′ g mC ′ h′ (m + 1)C ′ f ′ +❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳ ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ ψ χ φ %▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ NC ′ p C j i Y p / X Proof. Corresponding to the closed immersion h′ : mC ′ ֒→ (m+1)C ′, for m = 1, 2, · · · N − 1, we have the surjective morphism of sheaves O(m+1)C ′ ։ h′ ∗OmC ′. The ideal sheaf of this closed subscheme is h′ ∗g∗OY (−mC ′)C ′. This gives, 0 → h′ ∗g∗OY (−mC ′)C ′ → O(m+1)C ′ → h′ ∗OmC ′ → 0 . Tensor by χ∗A (and use projection formula) to get 0 → h′ ∗g∗(ψ∗A ⊗ OY (−mC ′)C ′) ≃ h′ ∗g∗L′ → χ∗A → h′ ∗(φ∗A) → 0 . Pushforward this sequence to X by applying p∗j∗f ′ ∗: 0 → p∗j∗f ′ ∗h′ ∗g∗L′ → p∗j∗f ′ ∗χ∗A → p∗j∗f ′ ∗h′ ∗φ∗A → 0 . That is, 0 −→ i∗ψ∗L′ −→ i∗χ∗χ∗A −→ i∗φ∗φ∗A −→ 0 . Note that the above is a sequence of pushforwards to X by p of Γ-sheaves from Y . Therefore, we can take Γ-invariants of each of these sheaves. The sequence of Γ-invariant sheaves continues to be exact, cf. [10, §1.4.3]: 0 −→ (i∗ψ∗L′)Γ −→ (i∗χ∗χ∗A)Γ −→ (i∗φ∗φ∗A)Γ −→ 0 . However, (i∗χ∗χ∗A)Γ ≃ i∗A ≃ (i∗φ∗φ∗A)Γ. Thus, (i∗ψ∗L′)Γ = 0. (cid:3) We now prove Theorem 1.3 stated in the introduction. Proof of Theorem 1.3. We need to prove that the orbifold bundle on Y corresponding to the dual parabolic LM bundle (F, FC, 0, N −m N ) on X is the dual LM bundle F ′. We do that by showing that the parabolic bundle on X corresponding to F ′ is the required one. We recall the commutative diagram where 1 ≤ m < N: C ′ g ❙❙ mC ′ NC ′ ❙ ❙ ❙ ❙ ❙ ● ● ❙ ❙ ψ ❙ ❙ ● ● ● ❙ ❙ ❙ f φ ● ● ❙ ❙ ● ● ❙ #● )❙ p C j i Y p / X For all a = 1, 2, . . . , N, we have the following exact sequence which we will use repeatedly in the proof: 0 → F ′(−aC ′) → F ′(−(a − 1)C ′) → j∗f∗g∗(cid:0)F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1) → 0 . (8) Note that, just as in the short exact sequence (4), F ′mC ′ fits in the following short exact sequence of sheaves on mC ′: 0 −→ φ∗A ⊗ OY (−mC ′)mC ′ −→ F ′mC ′ −→ φ∗A∨ −→ 0 . We restrict this to C ′: (9) 0 −→ ψ∗A ⊗ OY (−mC ′)C ′ −→ F ′C ′ −→ ψ∗A∨ −→ 0 . / / + / / / / % / / / / / ) / / # / / / 14 P. NARAYANAN By equation (7), we have E0 = (cid:16)p∗F ′(cid:17)Γ a = 1, 2, · · · , N − 1, N, then, . We further have that, if t ∈ ( a−1 N , a N ] where Et =(cid:16)p∗(cid:16)F ′ ⊗ OY (−aC ′)(cid:17)(cid:17)Γ . We now compute Et as we vary t ∈ [0, 1], and this filtration of sheaves will completely determine the parabolic bundle on X corresponding to F ′. t = 0 1. Pushforward the exact sequence (6) to X and consider the sequence of Γ-invariant sheaves (Taking Γ-invariants is an exact functor). Since p∗j∗f∗A′ ≃ i∗φ∗φ∗A, we get 0 → (p∗F ′)Γ → H 0(C, A) ⊗ (p∗OY )Γ → (i∗φ∗φ∗A)Γ → 0 . Thus we get, 0 → (p∗F ′)Γ → H 0(C, A) ⊗ OX → i∗A → 0 . Therefore (p∗F ′)Γ = E0 = F . 2. N , a . We we get from (9): t ∈(cid:0)0, N −m N (cid:3) N(cid:3) for a = 1, 2, . . . , N − m. Then, Et =(cid:16)p∗(cid:16)F ′ ⊗ OY (−aC ′)(cid:17)(cid:17)Γ Consider t ∈(cid:0) a−1 consider the pushforward of the sequence (8) to X by p, noting that p∗j∗f∗g∗(cid:0)F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1) ≃ i∗ψ∗(cid:0)F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1), to get: 0 → p∗F ′(−aC ′) → p∗F ′(−(a − 1)C ′) → i∗ψ∗F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′ → 0 . Now the sheaf i∗ψ∗(cid:0)F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1) fits in the following short exact sequence that (10) 0 → i∗ψ∗(cid:0)ψ∗A ⊗ OY(cid:0)(−(a − 1) − m)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1) → i∗ψ∗F ′(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′ → i∗ψ∗(cid:0)ψ∗A∨ ⊗ OY(cid:0) − (a − 1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1) → 0 . For a = 2, . . . , N − m, we have (cid:0)i∗ψ∗(cid:0)ψ∗A ⊗ OY(cid:0)(−(a − 1) − m)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1)(cid:1)Γ (cid:0)i∗ψ∗(cid:0)ψ∗A∨⊗OY(cid:0)−(a−1)C ′(cid:1)(cid:12)(cid:12)C ′(cid:1)(cid:1)Γ = 0 and = 0 by Lemma 5.4. Hence, (i∗ψ∗F ′(−(a−1)C ′)C ′)Γ = 0 and thus (p∗F ′(−aC ′))Γ = (p∗F ′(−(a − 1)C ′))Γ for a = 2, 3, · · · , N − m. Now it is sufficient to compute (p∗F ′(−C ′))Γ which is in fact Et for t ∈ (0, 1 N ] i.e. the case a = 1 above. Set a = 1 in (8) and pushforward the sequence of Γ-sheaves on Y to X: (11) 0 → p∗F ′(−C ′) → p∗F ′ → p∗j∗f∗g∗(F ′C ′) ≃ i∗ψ∗(F ′C ′) → 0. Next, setting a = 1 in the short exact sequence (10), we get: (12) 0 → i∗ψ∗(ψ∗A ⊗ OY (−mC ′)C ′) → i∗ψ∗F ′C ′ → i∗ψ∗ψ∗A∨ → 0 . By Lemma 5.4, (i∗ψ∗(ψ∗A⊗OY (−mC ′)C ′))Γ = 0 since m < N. Also, we have (i∗ψ∗ψ∗A∨)Γ = i∗A∨. Hence, (i∗ψ∗F ′C ′)Γ ≃ i∗A∨. Considering the invariants of the short exact sequence (11), we get 0 → (p∗F ′(−C ′))Γ → F → i∗A∨ → 0. SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 15 From the short exact sequence (3), it follows that (p∗F ′(−C ′))Γ ≃ H 0(C, A) ⊗ OX (−C). Thus Et = H 0(C, A) ⊗ OX (−C) for t ∈ (0, 1/N] and (p∗F ′(−aC ′))Γ = (p∗F ′(−C ′))Γ = H 0(C, A) ⊗ OX (−C) for a = 2, . . . , N − m. Thereby, Et = H 0(C, A) ⊗ OX(−C) for all N (cid:3) . t ∈(cid:0)0, N −m t ∈(cid:0) N −m N , N −m+1 N (cid:3) 3. In this case, we have Et =(cid:16)p∗(cid:16)F ′ ⊗ OY (−(N − m + 1)C ′)(cid:17)(cid:17)Γ m)C ′)C ′) = i∗ψ∗(F ′(−(N − m)C ′)C ′), from (8) we have . Since p∗j∗f∗g∗(F ′(−(N − (13) 0 → p∗F ′(−(N − m + 1)C ′) → p∗F ′(−(N − m)C ′) → i∗ψ∗(F ′(−(N − m)C ′)C ′) → 0 . From (9), we get 0 → (i∗ψ∗(ψ∗A ⊗ OY (−NC ′)C ′))Γ → (i∗ψ∗F ′(−(N − m)C ′)C ′)Γ → (i∗ψ∗(ψ∗A∨ ⊗ OY (−(N − m)C ′)C ′))Γ → 0 . As before, the right hand side term of the above sequence is zero. But note that (i∗ψ∗(ψ∗A ⊗ OY (−NC ′)C ′))Γ = (i∗ψ∗ψ∗(A ⊗ OX(−C)C))Γ ≃ i∗(A ⊗ OX(−C)C) and hence (i∗ψ∗F ′(−(N − m)C ′)C ′)Γ ≃ i∗(A ⊗ OX(−C)C). Finally, taking Γ-invariants of (13), we get: 0 → (p∗F ′(−(N − m + 1)C ′))Γ → H 0(C, A) ⊗ OX (−C) → i∗(A ⊗ OX (−C)C) → 0 . N , N −m+1 N From the exact sequence (2), we get (p∗F ′(−(N − m + 1)C ′))Γ ≃ F ⊗ OX (−C). Thus for of the flag for the parabolic structure, we deduce, based on the known properties of Et (cid:3), we get Et = F ⊗ OX (−C) = F (−C). Since this is the expected end t ∈ (cid:0) N −m from Biswas [3], that Et = F (−C) for t ∈(cid:0) N −m Thus, we have the following sheaves Et for t ∈ [0, 1]: N , 1(cid:3). (1) E0 = F , (2) Et = H 0(C, A) ⊗ OX (−C) for t ∈(cid:0)0, N −m N (cid:3), (3) Et = F (−C) for t ∈(cid:0) N −m N , 1(cid:3). As there is a jump at t = 0 and at t = N −m the parabolic vector bundle on X corresponding to F ′ is: N , these give the parabolic weights. Indeed, F ⊃0 H 0(C, A) ⊗ OX (−C) ⊃ N −m N F (−C) or equivalently the (F, FC, 0, N −m abolic sheaf. N ). This is precisely the description of the required par- (cid:3) The above theorem shows that the LM bundles behave well under the parabolic-orbifold bundle correspondence. 16 P. NARAYANAN 6. Proof of Theorem 1.4 We keep notations as in previous sections. Let L be an ample line bundle on X. Corollary 6.1. The dual parabolic LM bundle (F, FC, a1, a2) with a1 = 0 and a2 = N −m on X is parabolic µL-semistable if and only if the corresponding Orbifold LM bundle F ′ associated to the triple (mC ′, φ∗A, H 0(C, A)) on Y is µp∗L-semistable. N Proof. This is a direct consequence of [3, Lemma 3.13] and the fact that an orbifold bundle is orbifold semistable with respect to p∗L if and only if it is µp∗L-semistable on Y , cf. [3, Lemma 2.7]. (cid:3) Consequently, we prove Theorem 1.4. Proof of Theorem 1.4. Keep notations as in the statement of the Theorem. (a) In this case, X = P2 and C ∈ O(d) is a general smooth degree d curve, where d is odd. Let A = O(rd)C for any r ≥ 1, and V ⊂ H 0(C, A) be a general two dimensional subspace. Let F be the dual LM bundle on X corresponding to (C, A, V ) which is µO(1)-stable, cf. [19, Theorem 1.2]. Given that m is an integer such that N −m d. Consider the dual parabolic LM bundle F∗ = (F, FC, a1, a2) where the weights a1 = 0 and a2 = N −m d. By Corollary 4.3, F∗ is parabolic µO(1)-stable. By Corollary 6.1, the dual LM bundle on Y corresponding to the triple (mC ′, A′, V ) on Y is µp∗O(1)-semistable. N < 1 N < 1 (b) Here X is a smooth projective K3 surface and L is an ample line bundle on X such that a general curve C ∈ L has genus g, Clifford dimension 1 and maximal gonality k. We have ρ(g, 1, d) > 0. For a general smooth C ∈ L , consider the rank 2 dual LM bundle F := FC,A associated with a general complete, base-point free g1 d, say A on C. By [13], F is µL-stable. We have an integer m such that N −m g−1. Then, by Corollary 4.2, the dual parabolic LM bundle (F, FC, a1, a2) with weights a1 = 0 and a2 = N −m g−1 is parabolic µL-stable. Then by Corollary 6.1, the LM bundle on Y corresponding to the triple (mC ′, A′, H 0(C, A)) is µp∗L-semistable. N < 1 N < 1 (a') As a particular case of (a), let C ∈ O(1) be a line, A is a globally generated line bundle on C and V be as earlier. Again, the dual LM bundle F correspond- ing to (C, A, V ) is µO(1)-stable. By Corollary 4.3, the dual parabolic LM bundle (F, FC, a1, a2) is parabolic µO(1)-stable for any weights ai such that a2 − a1 < 1. Set a1 = 0 and a2 = N −1 N , i.e. m = 1 in part (a). By part (a) the dual LM bundle F ′ on Y corresponding to the triple (C ′, A′, V ) is µp∗O(1)-semistable on Y . Let l = deg A′. There is a flat family of LM bundles on Y parametrized by an open subset of the elements of the Brill-Noether vari- ety G1 [18, § 3]. Since semistability of vector bundles is an open condition in flat families [8, Prop. 2.3.1], there is an irreducible component of G1 l (OY (C ′)) corresponding to µp∗O(1)- semistable LM bundles. l (OY (C ′)), cf. (b') We consider a particular case of (b), where L additionally satisfies L2 = 2. Then a general curve C ∈ L has genus 2, Clifford dimension 1 and maximal gonality SEMISTABILITY OF LAZARSFELD-MUKAI BUNDLES VIA PARABOLIC STRUCTURES 17 k. Suppose A is a complete base-point free g1 d on C where d is a positive integer such that the Brill-Noether number ρ(2, 1, d) > 0 (note that d = 3 satisfies the requirement that ρ(2, 1, d) > 0 and completeness). Then the dual LM bundle F := FC,A associated to a general C and A is µL-stable. By Corollary 4.2, the dual parabolic bundle (F, FC, a1, a2) is parabolic µL-stable for any weights ai such that a2 − a1 < 1 g−1 = 1. Set weights a1 = 0 and a2 = N −1 N , i.e. m = 1 in case (b). Then (F, FC, 0, N −1 N ) is parabolic µL-stable. Just as in part (b), the dual LM bundle on Y corresponding to the triple (C ′, A′, H 0(C, A)) is µp∗L-semistable. Consider the flat family of LM bundles parametrized by the elements of an open subset of the Brill-Noether variety G1 [18, § 3]. Here l = deg A′. Since semistability of vector bundles is an open condition in flat families [8, Prop. 2.3.1], there is an irreducible component of G1 (cid:3) l (OY (C ′)) corresponding to µp∗L-semistable LM bundles. l (OY (C ′)), cf. Remark 6.2. Consider σ : X → P2, a K3 surface which is a double cover of the projective plane branched along a smooth sextic curve, and L = σ∗OP2(1). Then X and L satisfy the requirements of part (b') of the above theorem. That is, (L2) = 2 and a general curve C ∈ L has genus 2, Clifford dimension 1 and constant gonality 2, cf. [5, Theorem A and Prop. 3.3]. References [1] M. Aprodu and G. Farkas, Green's conjecture for curves on arbitrary K3 surfaces, Compos. Math. 147(3), 839-851, 2011. [2] U. Bhosle, Parabolic sheaves on higher dimensional varieties, Math. Ann. 293, 177-192, 1992. [3] I. Biswas, Parabolic bundles as orbifold bundles, Duke Math. J., 88(2), 305-325, 1997. [4] R. Brussee, Stable bundles on blown up surfaces, Math. Z., 205, 551-565, 1990. [5] C. Ciliberto and G. Pareschi, Pencils of minimal degree on curves on a K3 surface, J. Reine Angew. Math., 460, 15-36, 1995. [6] M. Green and R. Lazarsfeld, Special divisors on curves on a K3 surface, Invent. Math., 89(2), 357-370, 1987. [7] R. Hartshorne, Stable Reflexive Sheaves, Math. Ann. 254 (2), 121-176, 1980. [8] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves, 2nd ed., Cambridge Uni- versity Press, Cambridge, 2010. [9] A. L. Knutsen, On two conjectures for curves on K3 surfaces, Int. J. Math. 20(12), 1547-1560, 2009. [10] A. Krug, Extension groups of tautological sheaves on Hilbert schemes, J. Algebraic Geom., 23(3), 571-598, 2014. [11] R. Lazarsfeld, Brill-Noether-Petri without degenerations, J. Diff. Geom. 23, 299-307, 1986. [12] R. Lazarsfeld, Positivity in algebraic geometry I & II, Ergebnisse der Mathematik und ihrer Gren- zgebiete, vol. 48 & 49, Springer-Verlag, Berlin, 2004. [13] M. Lelli-Chiesa, Stability of rank-3 Lazarsfeld-Mukai bundles on K3 surfaces, Proc. Lond. Math. Soc. 107 (3), 451-479, 2013. [14] M. Maruyama and K. Yokogawa, Moduli of parabolic stable sheaves, Math. Ann., 293(1), 77-99, 1992. [15] V. B. Mehta and C. S. Seshadri, Moduli of vector bundles on curves with parabolic structures, Math. Ann., 248(3), 205-239, 1980. [16] S. Mukai, Biregular classification of Fano threefolds and Fano manifolds of coindex 3, Proc. Nat. Acad. Sci. USA, 86, 3000-3002, 1989. [17] T. Nakashima, Moduli of stable bundles on blown up surfaces, J. Math. Kyoto Univ., 571-581, 1993. [18] P. Narayanan, On the semistability of certain Lazarsfeld–Mukai bundles on abelian surfaces, Annali dell'Universita di Ferrara, 2017, doi:10.1007/s11565-017-0277-z. 18 P. NARAYANAN [19] P. Narayanan, Lazarsfeld-Mukai reflexive sheaves and their stability, Communications in Algebra, 2017, doi:10.1080/00927872.2017.1354009. [20] C. S. Seshadri, Moduli of vector bundles on curves with parabolic structures, Bull. Amer. Math. Soc., 83(1), 124-126, 1977. [21] C. Voisin, Greens generic syzygy conjecture for curves of even genus lying on a K3 surface, J. European Math. Soc. 4, 363-404, 2002. [22] C. Voisin, Greens canonical syzygy conjecture for generic curves of odd genus, Compositio Math. 141, 1163-1190, 2005. Department of Mathematics, Indian Institute of Technology Madras, Chennai - 600036. E-mail address: [email protected]
1011.5551	3	1011	2011-08-29T17:40:42	Local models of Shimura varieties, I. Geometry and combinatorics	[ "math.AG", "math.NT", "math.RT" ]	We survey the theory of local models of Shimura varieties. In particular, we discuss their definition and illustrate it by examples. We give an overview of the results on their geometry and combinatorics obtained in the last 15 years. We also exhibit their connections to other classes of algebraic varieties such as nilpotent orbit closures, affine Schubert varieties, quiver Grassmannians and wonderful completions of symmetric spaces.	math.AG	math	Local models of Shimura varieties, I. Geometry and combinatorics Georgios Pappas, Michael Rapoport, and Brian Smithling Abstract. We survey the theory of local models of Shimura varieties. In particular, we discuss their definition and illustrate it by examples. We give an overview of the results on their geometry and combinatorics obtained in the last 15 years. We also exhibit their connections to other classes of algebraic varieties. . G A h t a m [ 3 v 1 5 5 5 . 1 1 0 1 : v i X r a Introduction Local models of Shimura varieties are projective algebraic varieties over the spectrum of a discrete valuation ring. Their singularities are supposed to model the singularities that arise in the reduction modulo p of Shimura varieties, in the cases where the level structure at p is of parahoric type. The simplest case occurs for the modular curve with Γ0(p)-level structure. In this example the local model is obtained by blowing up the projective line P1Zp over Spec Zp at the origin 0 of ×Spec Zp Spec Fp. Local models for Shimura varieties are the special fiber P1Fp defined in terms of linear algebra data inside the product of Grassmann varieties, at least as far as type A, or C, or some cases of type D are concerned. Another version of these varieties arises as closures of Schubert varieties inside the Beilinson- Drinfeld-Gaitsgory deformation of affine flag varieties. It is the aim of this survey to discuss local models from various points of view, exhibit their connections to other classes of algebraic varieties, and give an overview of the results on them obtained in the last 15 years. = P1Zp Why does such a survey have a place in the handbook of moduli? The reason is that Shimura varieties are often moduli spaces of abelian varieties with additional structure. Therefore, determining the singularities of their reduction modulo p is an inherent part of the theory of such moduli spaces. The archetypical example is the Shimura variety attached to the group of symplectic similitudes (and its canonical family of Hodge structures). In this case, the Shimura variety represents the moduli functor on Q-schemes of isomorphism classes of principally Received by the editors October 31, 2018. 2000 Mathematics Subject Classification. Primary 14G35, 11G18; Secondary 14M15. Key words and phrases. Shimura variety, local model, affine flag variety. Partially supported by NSF grants DMS08-02686 and DMS11-02208. 2 Local models of Shimura varieties, I. polarized abelian varieties of a fixed dimension, equipped with a level structure. In case the p-component of this level structure is of parahoric type, there is an obvious way to extend the moduli functor to a moduli functor on Z(p)-schemes, which however will have bad reduction, unless the p-component of the level structure is hyperspecial. Local models then serve to analyze the singularities in the special fibers of the Z(p)-models thus defined. For instance, natural questions of a local nature are whether the Z(p)-schemes that arise in this way are flat over Z(p), or Cohen-Macaulay, or what the set of branches through a point in the reduction is. All these questions, exactly because they are of a local nature, can be transferred to questions on the corresponding local models. We will not give a sketch of the historical development of the theory here. We only mention that the origin of these ideas lies in the work of Deligne and Pappas [DP], of Chai and Norman [CN], and of de Jong [J] on specific Shimura varieties. The definitions of local models in the examples considered in these papers were formalized to some degree in the work of Rapoport and Zink in [RZ] with the introduction of what were subsequently termed naive local models. The paper [P1] of Pappas pointed to the fact that naive local models are not always flat. Whereas the examples of Pappas arise due to the fact that the underlying group is non-split (in fact, split only after a ramified extension), it was later pointed out by Genestier [Ge2] that a similar phenomenon also occurs for split orthogonal groups. This then led to the definition of local models in the papers [PR1, PR2, PR4], as it is presented here. The local structure of local models was considered in papers by Gortz [Go1, Go2, Go4], Faltings [Fa1, Fa2], Arzdorf [A], Richarz [Ri2], and others. At the same time, the combinatorics of the special fiber of local models (in particular, the {µ}-admissible set in the Iwahori-Weyl group), was considered in papers by Kottwitz and Rapoport [KR], Haines and Ngo [HN2], Gortz [Go4], and Smithling [Sm1, Sm2, Sm3, Sm4, Sm5]. Finally, we mention the papers by Gaitsgory [Ga], Haines and Ngo [HN1], Gortz [Go3], Haines [H1, H2, HP], and Kramer [Kr], addressing the problem of determining the complex of nearby cycles for local models (Kottwitz conjecture). It is remarkable that local models also appear in the study of singularities of other moduli schemes. In [Ki] Kisin constructs a kind of birational modification scheme of the universal flat deformation of a finite flat group scheme over a discrete valuation ring of unequal characteristic (0, p), and shows that the singularities in characteristic p of these schemes are modeled by certain local models that correspond to Shimura varieties of type A. Another context in which local models appear is in the description of Faltings [Fa3] of the singularities of the moduli space of vector bundles on semi-stable singular algebraic curves. The theory of local models falls fairly neatly into two parts. The first part is concerned with the local commutative algebra of local models, and the combina- torics of the natural stratification of their special fibers. This part of the theory G. Pappas, M. Rapoport, and B. Smithling 3 is surveyed in the present paper. The second part is concerned with the cohomol- ogy of sheaves on local models, and will be presented in a sequel to this paper. More precisely, we will survey in a second installment the cohomology of coherent sheaves on local models (and in particular will explain the coherence conjecture of [PR3]), as well as the cohomology of (cid:96)-adic sheaves on local models, and in par- ticular the determination of the complex of nearby cycles. Of course, both parts are interrelated by various links between them, and we will try to make this plain in the sequel to this first installment. This survey consists of three parts of a rather distinct nature. In the first part (§1), we give two approaches to local models, each with a different audience in mind. It should be pointed out that only one of these approaches is the one with which we actually work, and which relates directly to the theory of Shimura varieties, especially those which are of PEL type. The other approach points to a more general theory and shows the ubiquity of local models in other contexts, but is not completely worked out at this stage (although there is relevant work in preparation by Pappas and Zhu). In the second part (§§2 -- 4), we give an account of the results on local models that have been obtained in the last 15 years, and we highlight open questions in this area. In the third part (§§5 -- 8) we explain the relation of local models to other classes of algebraic varieties, such as nilpotent orbit closures, matrix equation varieties, quiver Grassmannians, and wonderful compactifications of symmetric spaces, that have been established in some cases. Especially as concerns the last section, this is still largely uncharted territory, which explains why this part is of a more informal nature. We are happy to acknowledge the important contributions of K. Arzdorf, C.-L. Chai, P. Deligne, V. Drinfeld, G. Faltings, D. Gaitsgory, A. Genestier, U. Gortz, T. Haines, J. de Jong, R. Kottwitz, L. Lafforgue, P. Norman, T. Richarz, and Th. Zink to the subject of the survey. In addition, we thank U. Gortz, T. Richarz, J. Schroer, and X. Zhu for their help with this survey, G. Farkas and I. Morrison for inviting us to include our text in their handbook of moduli and for putting on us just the right amount of pressure for a (un)timely delivery, and the referee for his/her suggestions. 1. An object in search of a definition In this motivational section, we sketch two possible approaches to local mod- els. It is the first approach that is directly related to the original purpose of local models, which is to construct an elementary projective scheme over the ring of integers of a p-adic localization of the reflex field of a Shimura variety, whose singularities model those of certain integral models of a Shimura variety. Unfortu- nately we cannot make the corresponding definition in as great a generality as we 4 Local models of Shimura varieties, I. would like. It is the second approach which is most easily related to the theory of algebraic groups. It is also the most elegant, in the sense that it is uniform. In a preliminary subsection we list the formal properties that we have come to expect from local models. 1.1. Local models in an ideal world The ideal situation presents itself as follows. Let F be a discretely valued field. We denote by OF its ring of integers and by k = kF its residue field which we assume to be perfect. Let G be a connected reductive group over F , and let {µ} be a geometric conjugacy class of one-parameter subgroups of G, defined over an algebraic closure F of F . Let E be the field of definition of {µ}, a finite extension of F contained in F (the reflex field of the pair (G,{µ})). Finally, let K be a parahoric subgroup of G(F ) in the sense of [BTII], see also [T]. These subgroups are ("up to connected component", see [BTII] for a precise definition) the stabilizers of points in the Bruhat-Tits building of the group G(F ). We denote by G the smooth group scheme over OF with generic fiber G and with connected special fiber such that K = G(OF ). The existence of a canonical group scheme G with these properties is one of the main results of [BTII]. To these data, one would like to associate the local model, a projective scheme M loc = M loc(G,{µ})K over SpecOE, at least when {µ} is a conjugacy class of minuscule1 coweights. It should be equipped with an action of GOE = G ⊗OF OE. At least when {µ} is minuscule, M loc should have the following properties. (i) M loc is flat over SpecOE with generic fiber isomorphic to G/Pµ. Here G/Pµ denotes the variety over E of parabolic subgroups of G of type {µ}. (ii) There is an identification of the geometric points of the special fiber, M loc(kE) =(cid:8) g ∈ G(L)/(cid:101)K(cid:12)(cid:12) (cid:101)Kg(cid:101)K ∈ Adm(cid:101)K({µ})(cid:9) . Here L denotes the completion of the maximal unramified extension of F in F , and (cid:101)K = G(OL) the parahoric subgroup of G(L) corresponding to K. Finally, Adm(cid:101)K({µ}) ⊂ (cid:101)K\G(L)/(cid:101)K is the finite subset of {µ}-admissible elements [R], cf. Definition 4.3.2 below. (iii) For any inclusion of parahoric subgroups K ⊂ K(cid:48) of G(F ), there should be a morphism (1.1.1) K −→ M loc K(cid:48) , M loc which induces the identity (via (i)) on the generic fibers. For a central isogeny G → G(cid:48), and compatible conjugacy classes {µ} and {µ(cid:48)}, and compatible parahoric subgroups K ⊂ G(F ), resp. K(cid:48) ⊂ G(cid:48)(F ), one should have an identification (1.1.2) M loc(G,{µ})K = M loc(G(cid:48),{µ(cid:48)})K(cid:48) . 1Recall that a coweight µ is minuscule if (cid:104)α, µ(cid:105) ∈ {−1, 0, 1} for every root α of GF . G. Pappas, M. Rapoport, and B. Smithling 5 More generally, if ϕ : G → G(cid:48) is a homomorphism, and {µ(cid:48)} = {ϕ ◦ µ}, and if ϕ(K) ⊂ K(cid:48), there should be a morphism M loc(G,{µ})K −→ M loc(G(cid:48),{µ(cid:48)})K(cid:48) ⊗OE(cid:48) OE , (1.1.3) which induces in the generic fiber the natural morphism G/Pµ → (G(cid:48)/Pµ(cid:48))⊗E(cid:48) E. Here E(cid:48) ⊂ E is the reflex field of (G(cid:48),{µ(cid:48)}). (iv) Let F (cid:48) be a finite extension of F contained in F . Let G(cid:48) = G ⊗F F (cid:48), and regard {µ} as a geometric conjugacy class of one-parameter subgroups of G(cid:48). Let K(cid:48) ⊂ G(cid:48)(F (cid:48)) be a parahoric subgroup with K = K(cid:48) ∩ G(F ). Note that the reflex field of (G(cid:48),{µ}) is equal to E(cid:48) = F (cid:48)E. Under these circumstances one should expect a morphism of local models (1.1.4) M loc(G,{µ})K ⊗OE OE(cid:48) −→ M loc(G(cid:48),{µ})K(cid:48) , which induces the natural morphism (G/Pµ) ⊗E E(cid:48) −→ G(cid:48)/P (cid:48) µ in the generic fibers. Furthermore, if F (cid:48)/F is unramified, then the morphism (1.1.4) should be an isomorphism. i=1 µi are all products. i=1 Gi and the reflex fields Ei, 1 ≤ i ≤ n, generate the reflex field E. We then expect an equivariant isomorphism of local models i=1 Ki, and µ =(cid:81)n i=1 Gi, K =(cid:81)n ∼−→(cid:89) M loc(Gi,{µi})Ki ⊗OEi (cid:89) (Gi/Pµi) ⊗Ei E i (G/Pµ) = i OE , (v) Suppose that G =(cid:81)n Then G =(cid:81)n (1.1.5) M loc(G,{µ})K which induces the natural isomorphism in the generic fibers. Here we should point out that it is not clear that the above listed properties are enough to characterize the local models M loc(G,{µ})K up to isomorphism. In fact, a general abstract (i.e. "group theoretic") definition of local models is still lacking, although, as we will explain in §1.3, there is now some progress on this problem. We now sketch two different approaches to the concept of local models. 1.2. Local models arising from Shimura varieties Let ShK = Sh(G,{h}, K) denote a Shimura variety [D1] attached to the triple consisting of a connected reductive group G over Q, a family of Hodge structures h and a compact open subgroup K ⊂ G(Af ). We fix a prime number p and assume that K factorizes as K = K p · Kp ⊂ G(Ap f ) × G(Qp). In fact we assume in addition that K = Kp is a parahoric subgroup of G(Qp). Let E ⊂ C denote the reflex field of (G,{h}), i.e., the field of definition of the geometric conjugacy class of one-parameter subgroups {µ} = {µh} attached 6 Local models of Shimura varieties, I. to {h}, cf. [D1]. Then E is a subfield of the field of algebraic numbers Q, of finite degree over Q. Fixing an embedding Q → Q p determines a place ℘ of E above p. We denote by the same symbol the canonical model of ShK over E and its base change to E℘. It is then an interesting problem to define a suitable model SK of ShK over SpecOE℘. Such a model should be projective if ShK is (which is the case when Gad is Q-anisotropic), and should always have manageable singularities. In particular, it should be flat over SpecOE℘, and its local structure should only depend on the localized group G = G ⊗Q Qp, the geometric conjugacy class {µ} over Q p, and the parahoric subgroup K = Kp of G(Qp). Note that, due to the definition of a Shimura variety, the conjugacy class {µ} is minuscule. More precisely, we expect the local model M loc(G,{µ})K to model the sin- gularities of the model SK, in the following sense. We would like to have a local model diagram of OE℘-schemes, in the sense of [RZ], (cid:101)ϕ &MMMMMMMMMMM M loc(G,{µ})K , yttttttttttt (cid:101)SK π SK = G ⊗Zp OE℘, and in which (cid:101)ϕ is smooth of relative dimension dim G. Equiv- in which π is a principal homogeneous space (p.h.s.) under the algebraic group GOE℘ alently, using the language of algebraic stacks, there should be a smooth morphism of algebraic stacks of relative dimension dim G to the stack quotient, SK −→ [M loc(G,{µ})K/GOE℘ ] . In particular, for every geometric point x ∈ SK(Fp), there exists a geometric point x ∈ M loc(G,{µ})K(Fp), unique up to the action of G(Fp), such that the strict henselizations of SK at x and of M loc at x are isomorphic. Note that the generic fiber G/Pµ = M loc(G,{µ})K ⊗OE℘ E℘ is nothing but the compact dual of the hermitian symmetric domain corresponding to the Shimura variety Sh(G,{h}, K) (after extending scalars from E to E℘). From this perspec- tive, the local model M loc(G,{µ})K is an OE℘-integral model of the compact dual of the Shimura variety Sh(G,{h}, K). The problems of defining a model of ShK over OE℘ and of defining a local model M loc(G,{µ})K are closely intertwined (although not completely equivalent, as the example of a ramified unitary group shows [PR4]). Let us explain this and also briefly review the general procedure for the construction of local models M loc(G,{µ})K in some cases where the Shimura variety is of PEL type. Recall that in the PEL cases treated in [RZ] one first constructs a "naive" integral model S naive of the Shimura variety ShK; this is given by a moduli space description and affords a corresponding "naive local model" M naive together with a smooth K y & G. Pappas, M. Rapoport, and B. Smithling 7 morphism S naive K −→ [M naive/GOE℘ ]. As we mentioned in the introduction, these naive models M naive and SK are often not even flat over OE℘ [P1, Ge2]. Then, in most cases, the (non-naive) local model -invariant closed subscheme M loc := M loc(G,{µ})K of M naive with the is a GOE℘ same generic fiber which is brutally defined as the flat closure. The general idea then is that, from M loc(G,{µ})K, one also obtains a good (i.e at least flat) integral model SK of the Shimura variety via the cartesian diagram SK [M loc/GOE℘ ] S naive K / [M naive/GOE℘ ]. Unfortunately, in general, the schemes M loc(G,{µ})K and SK, do not have a reasonable moduli theoretic interpretation. Nevertheless, there are still (proven or conjectural) moduli descriptions in many interesting cases [P1, Go1, Go2, PR1, PR2, PR4]. All these issues are explained in more detail in §2. We mention here that taking the expected functorialities (i) -- (v) of local mod- els into account, we may, in constructing a local model for the data (G,{µ}, K), make the following hypotheses. We may assume that the adjoint group of G is simple; we may extend scalars to an unramified extension F of Qp. If we insist that {µ} be minuscule, this reduces the number of possible cases to an essentially finite list. Let us explain this in more detail. We assume Gad is simple and denote by µad the corresponding minuscule cocharacter of Gad(Q p be the com- pletion of the maximal unramified extension of Qp; by Steinberg's theorem every reductive group over Qun is quasi-split. We can write p). Let Qun p Gad/Qun p = ResL/Qun p (H) , µ = {µσ}σ: L→Q p , where H is absolutely simple adjoint over L, σ runs over embeddings of L over Qun p , and µσ are minuscule cocharacters of H(Q p). The group H over L is also quasi-split. The possible cases for the pairs (H, µσ) are given in the table below which can be obtained by combining the table of types of quasi-split, residually split, absolutely simple groups from [T, p. 60 -- 61] with the lists of minuscule coweights in [B] which are dominant relative to the choices of positive roots in [B]. In the local Dynkin diagrams, h denotes a hyperspecial vertex, s a special (but not hyperspecial) vertex, and • a nonspecial vertex. We refer to [T, 1.8] for the explanation of the notation in the diagrams. There are n + 1 vertices in each diagram that explicitly depends on n, i.e. aside from the diagrams for A1, A(2) 2 , D(3) 4 , E6, E(2) 6 , and E7. / / / 8 Local models of Shimura varieties, I. Type of H(L) Local Dynkin diagram A1 h h An, n ≥ 2 tttt JJJJ h h h h h A(2) 2 (C-BC1) s s< Nonzero dominant minuscule coweights for H(Q p) ∨ 1 i , 1 ≤ i ≤ n ∨ ∨ 1 , ∨ 2 A(2) 2n (C-BCn), n ≥ 2 A(2) 2n−1 (B-Cn), n ≥ 3 Bn, n ≥ 3 Cn, n ≥ 2 Dn, n ≥ 4 D(2) n+1 (C-Bn), n ≥ 2 • s < • > • • < • h • > LLL • rrr h h • s < • • • • • • • • • • • • • < s i , 1 ≤ i ≤ 2n ∨ s rrr LLL • s rrr LLL • h h • h < rrr LLL • h h • > s i , 1 ≤ i ≤ 2n − 1 ∨ ∨ 1 ∨ n ∨ 1 , ∨ n−1, ∨ n ∨ 1 , ∨ n , ∨ n+1 D(3) 4 , D(6) 4 (GI 2) • •< s ∨ 1 , ∨ 3 , ∨ 4 E6 h • rrr • h rrrr • LLLL • LLL h E(2) 6 (F I 4) • • < • • s ∨ 1 , ∨ 6 ∨ 1 , ∨ 6 G. Pappas, M. Rapoport, and B. Smithling 9 E7 h • • • • • • h ∨ 1 Note that the minuscule coweights are for H(Q p) and so they only depend on the absolute type over Q p. There are no nonzero minuscule coweights for E8, F4, G2 types. Of course, there is no simple description of the local model for ResL/Qun (H) in terms of a local model for H. For example, see the case of H = GLn in §2.4 below. However, we expect that most properties of local models for a group which is the restriction of scalars ResL/Qun (H) will only depend on H, p , the combinatorial data describing {µσ}σ and the type the degree of L over Qun (conjugacy class) of the parahoric subgroup K ⊂ H(L) (and not on the particular choice of the field L). p p Recall that to each such pair (H, µ) with H absolutely simple adjoint and µ minuscule as above, we associate a homogeneous space H/Pµ. Following Satake, in [D2, 1.3] Deligne studies faithful symplectic representations ρ : H(cid:48) → GSp2g, where H(cid:48) → H is a central isogeny, with the property that the coweight ρad◦µad is the (unique) minuscule coweight ∨ g in type Cg. Such symplectic representations exist for all the pairs in the table, except for those corresponding to exceptional groups. Hence, for all classical pairs, we can obtain an embedding of H/Pµ in the Grassmannian of Lagrangian subspaces of rank g in symplectic 2g-space. As we will see in the rest of the paper, the local model is often defined using such an embedding. By loc. cit., Shimura varieties of "abelian type" [M, M2] produce pairs that support such symplectic representations. Among them, the pairs (Bn, ∨ 1 ), (Dn, ∨ 1 ) do not appear, when we are just considering Shimura varieties of PEL type. For these pairs, the corresponding homogeneous spaces H/Pµ are forms of quadric hypersurfaces in projective space. So far, local models involving these pairs and the exceptional pairs have not been the subject of a systematic investigation. The construction in §1.3 applies to some of these local models, but we will otherwise omit their discussion in this survey. 1 ), (D(3) 1 ), (D(2) n , ∨ 4 , ∨ 1 ), (D(6) 4 , ∨ Example 1.2.1. Let us consider the Siegel case, i.e., the Shimura variety of prin- cipally polarized abelian varieties of dimension g with level K-structure, where the p-component Kp of K is the parahoric subgroup of Gp2g(Qp) which is the stabilizer of a selfdual periodic lattice chain L in the standard symplectic vector space of dimension 2g over Qp. In this case the Shimura field is equal to Q, and a model SK over Z(p) is given as the moduli scheme of principally polarized chains of abelian varieties of dimension g of type corresponding to L, with a level structure prime to p. In this case, the local model is given inside the product of finitely many 10 Local models of Shimura varieties, I. copies of the Grassmannian of subspaces of dimension g in a 2g-dimensional vec- tor space, which satisfy two conditions: a periodicity condition, and a self-duality condition. This example is discussed in §2.2. Example 1.2.2. Let us consider Shimura varieties related to the Picard moduli schemes of principally polarized abelian varieties of dimension n with complex multiplication of the ring of integers Ok in an imaginary-quadratic field k of signa- ture (r, s) (cf. [KuR, §4] for precise definitions). Here the Qp-group G is the group of unitary similitudes for the quadratic extension k⊗Qp of Qp. Three alternatives present themselves. (i) p splits in k. Then G (cid:39) GLn × Gm, and {µ} is the conjugacy class of a cocharacter of the form (cid:0)(1(r), 0(s)); 1(cid:1). Here, for n = r + s, we write (1(r), 0(s)) for the cocharacter x (cid:55)−→ diag(x, . . . , x (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) , 1, . . . , 1 ) r s p × Z× p is a p , where K 0 of GLn. The parahoric subgroup Kp is of the form K 0 parahoric supgroup of GLn(Qp). (ii) p is inert in k. Then G becomes isomorphic to GLn ×Gm after the unramified base extension ⊗Qp kp. Hence, by the expected general property (iv) of local models (which is true in the case at hand), the local models in cases (i) and (ii) become ℘ = kp · E℘. Note that, if r (cid:54)= s, isomorphic after extension of scalars from E℘ to E(cid:48) then E℘ can be identified with kp, and hence E(cid:48) (iii) p is ramified in k. Again G ⊗Qp kp = GLn × Gm. There is a morphism of local models ℘ = E℘. M loc(G,{µ})K ⊗OE℘ OE(cid:48) ℘ −→ M loc(GLn × Gm,{µ})K(cid:48) , for any parahoric subgroup K(cid:48) ⊂ GLn(kp) × k× However, in general this is not an isomorphism. p with intersection K with G(Qp). The Picard moduli problems lead to local models defined in terms of linear algebra, similar to the Siegel case above. The local models relating to the first two cases are discussed in §2.1; the local models of the last case is discussed in §2.6. As is apparent from this brief discussion, the definitions of the local models related to the last two kinds of Shimura varieties strongly use the natural repre- sentations of the classical groups in question (the group of symplectic similitudes in the Siegel case, the general linear group in the Picard case for unramified p, and the group of unitary similitudes in the Picard case for ramified p). They are therefore not purely group-theoretical. In the next section, we will give, in some case, a purely group-theoretical construction of local models. G. Pappas, M. Rapoport, and B. Smithling 11 1.3. Local models in the Beilinson-Drinfeld-Gaitsgory style (cid:26) iso-classes of pairs (cid:12)(cid:12)(cid:12)(cid:12) The starting point of the construction is a globalized version of the affine Grassmannian as in [BD]. Let O be a complete discrete valuation ring, with fraction field F and residue field k. Let X = SpecO[t] be the affine line over O. Let G be a split reductive algebraic group. We consider the following functor on (Sch/X). Let S ∈ (Sch/X), with structure morphism y : S → X, and define F a G-bundle on X × S, (1.3.1) GrG,X (S) = Here Γy ⊂ X × S denotes the graph of y, and the fiber products are over SpecO. Then GrG,X is representable by an ind-scheme over X. The relation of this β a trivialization of F(X×S)\Γy (F, β) . ind-scheme to the usual affine Grassmannian is as follows. Recall that to G and any field κ, there is associated its positive loop group L+G over κ, its loop group LG, and its affine Grassmannian GrG = LG/L+G (quotient of f pqc-sheaves on κ-schemes). Here L+G is the affine group scheme on Spec κ representing the functor on κ-algebras (cid:27) R (cid:55)−→ L+G(R) = G(R[[T ]]) , R (cid:55)−→ LG(R) = G(cid:0)R((T ))(cid:1) , and LG is the ind-group scheme over Spec κ representing the functor and GrG is the ind-scheme over Spec κ representing the functor (cid:26) iso-classes of pairs (F, β) (cid:12)(cid:12)(cid:12)(cid:12) F a G-bundle on Spec R[[T ]], β a trivialization of FSpec R((T )) (cid:27) , R (cid:55)−→ GrG(R) = comp. [BL], cf. also §3.1 below. When we wish to emphasize that we are working over the field κ, we will denote the affine Grassmannian by GrG,κ. Lemma 1.3.2 (Gaitsgory [Ga, Lem. 2]). Let x ∈ X(κ), where κ is either the residue field of O, or the fraction field of O, and identify the completed local ring Ox of X × Spec κ with κ[[T ]], using the local parameter T = t − x. Then the restriction morphism induces an isomorphism of ind-schemes over Spec κ, x : GrG,X ×X,x Spec κ −→ GrG,κ . i∗ Here GrG,κ denotes the affine Grassmannian of G over κ. (cid:3) We next construct a degeneration of GrG,F to the affine flag variety F lG,k = LG/B over k, where B denotes the Iwahori subgroup scheme of L+G given as the inverse image under the reduction map of a fixed Borel subgroup B of G, G L+G ⊂B ⊂ /____ B. / / / 12 Local models of Shimura varieties, I. (cid:26) (cid:12)(cid:12)(cid:12)(cid:12) (cid:27) iso-classes of triples (F, β, ε) Denote by 0 ∈ X(O) the zero section. Let F lG,X be the ind-scheme over X which represents the following functor on X-schemes, (1.3.3) S (cid:55)−→ . Let πX : F lG,X → GrG,X be the forgetful morphism, which is a smooth proper morphism with typical fiber G/B. Now fix a uniformizer π ∈ O. We denote by δ the section of X over O defined by δ∗(t) = π. Let F lG,O, resp. GrG,O, be the pull-back via δ of F lG,X , resp. GrG,X , to SpecO, and let (1.3.4) ε a reduction of F{0}×S to B (F, β) ∈ GrG,X (S), πO : F lG,O −→ GrG,O be the pull-back of πX . Note that the section δ gives by Lemma 1.3.2 identifications of the generic fiber of GrG,O with GrG,F , and of the special fiber of GrG,O with GrG,k. Lemma 1.3.5. The morphism πO induces • over F a canonical isomorphism F lG,O × Spec F (cid:39) GrG,F × G/B , • over k a canonical isomorphism F lG,O × Spec k (cid:39) F lG,k . Proof. (cf. [Ga, Prop. 3]) If S is a F -scheme, then X × S \ Γy = SpecS OS[t, (t − π)−1], and the trivialization β induces a trivialization of F along the section t = 0. Hence the reduction ε to B corresponds to a section of G/B over S, which provides the claimed identification of the generic fiber. If S is a k-scheme, then the identification of GrG,O × Spec k with GrG,k is via the origin t = 0, in the sense of Lemma 1.3.2. Hence the reduction ε to B corresponds to the choice of a compatible flag in the non-constant G-bundle Ft=0 over S, hence the triple (F, β, ε) corresponds to a lifting of the S-valued point (F, β) of GrG,k to an S-valued point of F lG,k, which gives the claimed (cid:3) identification of the special fiber. Next we recall that the orbits of L+G on GrG are parametrized by the dom- inant coweights, cf. Remark 4.1.9. More precisely, if A denotes a maximal split torus in B, and X∗(A)+ denotes the dominant coweights with respect to B, then the map λ (cid:55)−→(cid:0)L+G · λ(T ) · L+G(cid:1)(cid:14)L+G = Oλ defines a bijection between X∗(A)+ and the set of orbits. Furthermore, Oλ is a quasi-projective variety of dimension (cid:104)2, λ(cid:105), and Oµ ⊂ Oλ if and only if µ≤λ (i.e., G. Pappas, M. Rapoport, and B. Smithling 13 λ − µ is a non-negative integral sum of positive co-roots), cf. Proposition 4.2.11. In particular, Oλ is a projective variety if and only if λ is a minuscule coweight. Now we may define a version of local models in this context. Definition 1.3.6. The local model attached to λ ∈ X∗(A)+ in the Beilinson- Drinfeld-Gaitsgory context is the scheme-theoretic closure MG,λ in F lG,O of the locally closed subset Oλ × {e} of GrG,F ×Spec F G/B. This definition is essentially independent of the choice of the uniformizer π of O. Indeed, any two uniformizers differ by a unit, which may be used to construct a canonical isomorphism between the corresponding local models. It follows from the definition that MG,λ is a projective scheme flat of relative dimension (cid:104)2, λ(cid:105) over O. If λ is minuscule, then the generic fiber of MG,λ is projective and smooth. The theory of local models is concerned with the structure of the schemes MG,λ. Natural questions that arise in this connection are the following. When is the special fiber MG,λ⊗O k reduced? What are its singularities, and how can one enumerate its irreducible components? Variants 1.3.7. (i) Replacing the Borel subgroup B by a parabolic subgroup P containing B, and the Iwahori subgroup B by the parahoric subgroup P corre- sponding to P under the reduction morphism, we obtain a scheme F lG,P,O with generic fiber equal to GrG,F ×G/P and with special fiber equal to the affine partial flag variety LG/P. Correspondingly we define local models MG,P,λ over SpecO for λ ∈ X∗(A)+, with generic fiber contained in GrG,F , and with special fiber contained in the partial flag variety LG/L+P over k. For an inclusion P ⊂ P (cid:48) of two standard parabolic subgroups of G, we obtain a morphism between local models MG,P,λ −→ MG,P (cid:48),λ , which induces an isomorphism in the generic fibers. In the extreme case P = G, the scheme F lG,P,O has generic fiber GrG,F and special fiber GrG,k, and the local model MG,G,λ "looks constant" over SpecO, with generic fiber the Schubert variety Oλ in GrG,F , and special fiber the Schubert variety Oλ in GrG,k. If λ is minuscule, then MG,G,λ is projective and smooth over O. (ii) The preceding considerations generalize without substantial changes to the case when G is a quasi-split reductive group over O. (iii) An alternative definition of MG,P,λ can be given as follows. Starting from the Chevalley form of G over O and a parabolic subgroup P as above, we can construct a smooth "parahoric group scheme" (cid:98)G over SpecO[[t]]. The generic, resp. special, fiber of (cid:98)G → SpecO is isomorphic to the smooth affine "parahoric group scheme" (cid:98)Gκ over Spec κ[[t]] with κ = F , resp. k, given by Bruhat-Tits (These are characterized by requiring that (cid:98)Gκ(κsep[[t]]) is equal to the theory. group of elements of G(κsep[[t]]) with reduction modulo t contained in P (κsep).) 14 Local models of Shimura varieties, I. to the two dimensional base SpecO[[t]] by picking appropriate schematic root data given by ideals generated by powers of t, see also [PR4, p. 147]. The base change For example, (cid:98)G can be obtained by applying the constructions of [BTII, 3.2, 3.9.4] (cid:98)G ×Spec O[[t]] SpecO((t)) is identified with the Chevalley group scheme G ×Spec O with (cid:98)G over SpecO[[t]] to produce a "Bruhat-Tits group scheme" G over the affine SpecO((t)). We can now glue the "constant" group scheme G over SpecO[t, t−1] line X = SpecO[t], cf. [PR5, He]. Let us define the functor GrG,X exactly as in (1.3.1) above, except that G-torsors are now replaced by G-torsors. Also as above, set GrG,O = GrG,X ×X,δ SpecO where δ : SpecO → X is given by t (cid:55)→ π. Note that G ×X,δ SpecO is the parahoric group scheme associated to the subgroup of elements of G(Oun) with reduction modulo π contained in P (ksep). Similar to Lemma 1.3.5, we can see that GrG,O × Spec F = GrG,F is the affine Grassmannian of the loop group of G over F , while GrG,O×Spec k = LG/L+(cid:98)Gk is the affine flag variety corresponding to the parahoric subgroup (cid:98)G(k[[t]]) over k. The rest of the construction proceeds the same way: we define MG,P,λ = MG,λ to be the Zariski closure of the orbit Oλ. This construction extends beyond the split case and is used in [PZ] to provide a definition of local models M loc(G,{µ})K under some rather general assumptions. Indeed, one can deal with all reductive groups G that split over a tamely ramified extension of F and with general parahoric subgroups. The technical details of the construction of the group scheme G over X = SpecO[t] and of the global affine Grassmannian GrG,X in the general (tamely ramified) case are quite involved and we will not attempt to report on them here. Instead, we will refer the reader to the forthcoming article [PZ]. In the rest of this survey we will only discuss the models that are directly related to (mostly PEL) Shimura varieties, as sketched in §1.2. However, especially after Gaitsgory's paper [Ga], it is likely that methods from elsewhere, such as from the theory of the Geometric Langlands Correspondence, will have an impact on the problems discussed in this report.2 We hope that our loose discussion above can help in this respect to attract people from these other areas to the theory of local models. 2. Basic examples In this section we make explicit the definition of the local model in the style of §1.2 in the most basic cases. Let F be a discretely valued field, OF its ring of integers, π ∈ OF a uniformizer, and k = OF /πOF its residue field which we assume is perfect. Let n be a positive integer. A lattice chain in F n is a collection 2In this respect, we refer to very recent work of X. Zhu [Zh] on the coherence conjecture [PR3] and to the forthcoming [PZ]. G. Pappas, M. Rapoport, and B. Smithling 15 of OF -lattices in F n totally ordered under inclusion. A lattice chain L is periodic if aΛ ∈ L for every Λ ∈ L and a ∈ F ×. For i = na + j with 0 ≤ j < n, we define the OF -lattice j(cid:88) n(cid:88) (2.0.1) Λi := π−a−1OF el + π−aOF el ⊂ F n, l=1 l=j+1 where e1, . . . , en denotes the standard ordered basis in F n. Then the Λi's form a periodic lattice chain (2.0.2) ··· ⊂ Λ−2 ⊂ Λ−1 ⊂ Λ0 ⊂ Λ1 ⊂ Λ2 ⊂ ··· , which we call the standard lattice chain. Given a partition n = r + s, we recall the cocharacter (cid:0)1(r), 0(s)(cid:1) of GLn defined in Example 1.2.2(i); we shall also regard this as a cocharacter of certain subgroups of GLn (e.g. GSp2g, GO2g, . . . ), as appropriate. In each case except for §§2.4 and 2.5, we give the types of the adjoint group and nontrivial minuscule coweights under consideration, in the sense of the table in §1.2. 2.1. Split unitary, i.e. GLn (types (An−1, ∨ r ), 1 ≤ r ≤ n − 1) split unitary group relative to an unramified quadratic extension together. teger r with 0 ≤ r ≤ n, let µ denote the cocharacter(cid:0)1(r), 0(n−r)(cid:1) of the standard We refer to §1.2 for an explanation of why we lump the cases GLn and the Let G := GLn over F and let L be a periodic lattice chain in F n. Fix an in- maximal torus of diagonal matrices in G, and let {µ} denote the geometric conju- G,{µ},L attached to the triple (G,{µ},L) gacy class of µ over F . The local model M loc is the functor on the category of OF -algebras that assigns to each OF -algebra R the set of all families (FΛ)Λ∈L such that (i) (rank ) for every Λ ∈ L, FΛ is an R-submodule of Λ⊗OF R which Zariski-locally on Spec R is a direct summand of rank n − r; (ii) (functoriality) for every inclusion of lattices Λ ⊂ Λ(cid:48) in L, the induced map Λ ⊗OF R → Λ(cid:48) ⊗OF R carries FΛ into FΛ(cid:48): Λ ⊗OF R Λ(cid:48) ⊗OF R ⊂ FΛ /_____ ⊂ FΛ(cid:48); (iii) (periodicity) for every a ∈ F × and every Λ ∈ L, the isomorphism Λ a−→∼ aΛ identifies FΛ ∼−→ FaΛ. It is clear that M loc G,{µ},L is representable by a closed subscheme of a product of finitely many copies of Gr(n − r, n)OF , the Grassmannian of (n − r)-planes in G,{µ},L has generic fiber isomorphic to Gr(n − r, n)F . The n-space; and that M loc fundamental result of Gortz's paper [Go1] is the following. / / / 16 Local models of Shimura varieties, I. Theorem 2.1.1 (Gortz [Go1, 4.19, 4.21]). For any µ =(cid:0)1(r), 0(n−r)(cid:1) and periodic when µ =(cid:0)1, 0(n−1)(cid:1). G,{µ},L is flat over SpecOF with reduced special fiber. The irre- lattice chain L, M loc ducible components of its special fiber are normal with rational singularities, so in particular are Cohen-Macaulay. Furthermore, M loc G,{µ},L has semi-stable reduction (cid:3) Here a normal variety having "rational singularities" is meant in the strongest sense, i.e., there exists a birational proper morphism from a smooth variety to it such that the higher direct images of the structure sheaf and of the dualizing sheaf vanish. Example 2.1.2. The simplest nontrivial example occurs for n = 2, µ = (1, 0), and L the standard lattice chain (the Iwahori case). The most interesting point on the local model is the k-point x specified by the lines ke1 ⊂ Λ0 and ke2 ⊂ Λ1, where we use a bar to denote reduction mod π. In terms of standard affine charts of the Grassmannian, we find that x has an affine neighborhood U in the local model consisting of all points of the form FΛ0 = span{e1 + Xe2} and FΛ1 = span{Y π−1e1 + e2} such that XY = π. Hence U ∼= SpecOF [X, Y ]/(XY − π). Hence U visibly satisfies the conclusions of Theorem 2.1.1; its special fiber consists of two copies of A1 k meeting transversely at x. In fact, globally the special fiber of the local k meeting at x. By contrast, taking L to be the model consists of two copies of P1 homothety class of Λ0 or of Λ1 (the maximal parahoric case), the local model is tautologically isomorphic to P1OF In §6.1 we shall consider various analogs of the scheme U both for higher rank and for other groups, which we broadly refer to as schemes of matrix equations. Note that U is exactly the scheme Z1,2 appearing in Theorem 6.1.2. . Remark 2.1.3. In light of Theorem 2.1.1, it is an interesting question whether the special fiber, as a whole, of the local model is Cohen-Macaulay; by the flat- ness result, this is equivalent to the local model itself being Cohen-Macaulay. If Cohen-Macaulayness holds, then, since by the theorem above the special fiber is generically smooth, we can apply Serre's criterion to deduce that the local model is also normal. In [Go1, §4.5.1], Gortz proposes to attack the question of Cohen-Macaulayness of the special fiber by means of a purely combinatorial problem in the affine Weyl group, which however appears to be difficult, at least when L is the full standard lattice chain. In this way he has found that the special fiber is Cohen-Macaulay for n ≤ 4 by hand calculations, and for n ≤ 6 by computer calculations. Cohen- Macaulayness can also be shown via his approach for any n, in the case that the lattice chain L consists of the multiples of only two lattices. Similar remarks ap- ply to local models for any group, whenever the special fiber of the local model G. Pappas, M. Rapoport, and B. Smithling 17 can be identified with a union of Schubert varieties in an affine flag variety. See Remark 2.2.5 for another case where this property can be shown. By contrast, we know of no experimental evidence for Cohen-Macaulayness of the special fiber in any other Iwahori, i.e. "full lattice chain," cases. We shall discuss embedding the special fiber of local models in affine flag varieties in §3.3. The question of Cohen-Macaulayness and normality of local models is a major open problem in the field. 2.2. Split symplectic (types (Cg, ∨ g )) Let n = 2g, and let (cid:104) , (cid:105) denote the alternating F -bilinear form on F 2g whose −Hg matrix with respect to the standard ordered basis is (2.2.1) Jn := where Hg denotes the g × g matrix (cid:33)  . Given a lattice Λ in L, we denote by (cid:98)Λ its (cid:104) , (cid:105)-dual, (cid:98)Λ :=(cid:8) x ∈ F 2g(cid:12)(cid:12) (cid:104)Λ, x(cid:105) ⊂ OF (cid:32)  Hg := (2.2.2) . . . 1 Hg , 1 (cid:9). Then (cid:104) , (cid:105) induces a perfect bilinear pairing of OF -modules (2.2.3) We say that a lattice chain L in F 2g is self-dual if (cid:98)Λ ∈ L for all Λ ∈ L. Let G := GSp2g := GSp(cid:0)(cid:104) , (cid:105)(cid:1) over F , let µ denote the cocharacter(cid:0)1(g), 0(g)(cid:1) Λ ×(cid:98)Λ −→ OF . of the standard maximal torus of diagonal matrices in G, and let {µ} denote its geometric conjugacy class over F . Let L be a periodic self-dual lattice chain in G,{µ},L is the closed OF -subscheme of M loc F 2g. The local model M loc GL2g,{µ},L whose R-points, for each OF -algebra R, satisfy the additional condition (iv) (perpendicularity) for all Λ ∈ L, the perfect R-bilinear pairing (cid:0)Λ ⊗OF R(cid:1) ×(cid:0)(cid:98)Λ ⊗OF R(cid:1) −→ R obtained by base change from (2.2.3) identifies F⊥ Λ ⊂(cid:98)Λ ⊗OF R with F(cid:98)Λ. This time the local model M loc G,{µ},L has generic fiber LGr(2g)F , the Grass- mannian of Lagrangian subspaces in F 2g. The fundamental result of Gortz's paper [Go2] is the following. Theorem 2.2.4 (Gortz [Go2, 2.1]). For any periodic self-dual lattice chain L, G,{µ},L is flat over SpecOF with reduced special fiber. The irreducible compo- M loc nents of its special fiber are normal with rational singularities, so in particular are (cid:3) Cohen-Macaulay. 18 Local models of Shimura varieties, I. Remark 2.2.5. In the case that the lattice chain L consists of multiples of two lattices Λ and Λ(cid:48) such that (cid:98)Λ = Λ and (cid:98)Λ(cid:48) = πΛ, one can obtain a better result, namely that the whole special fiber is Cohen-Macaulay and that the local model is normal. This was first shown in [CN]. See Theorem 6.1.6 and the discussion after its statement. 2.3. Split orthogonal (types (Dg, ∨ g−1), (Dg, ∨ g )) In this example we assume char k (cid:54)= 2. Let n = 2g, and let ( , ) denote the symmetric F -bilinear form on F 2g whose matrix with respect to the standard ordered basis is H2g (2.2.2). Let (cid:98)Λ denote the ( , )-dual of any lattice Λ in F 2g. Analogously to the previous subsection, ( , ) induces a perfect pairing Λ×(cid:98)Λ → OF Let G := GO2g := GO(cid:0)( , )(cid:1) over F ,3 let µ denote the cocharacter(cid:0)1(g), 0(g)(cid:1) for any lattice Λ. We again say that a lattice chain in F 2g is self-dual if it is closed under taking duals. of the standard maximal torus of diagonal matrices in G, and let {µ} denote its G(F )-conjugacy class over F . Let L be a periodic self-dual lattice chain F 2g. The naive local model M naive GL2g,{µ},L defined in exactly the same way as for GSp2g, that is, we impose condition (iv) with the understanding that all notation is taken with respect to ( , ). G,{µ},L is the closed OF -subscheme of M loc Analogously to the symplectic case, M naive G,{µ},L has generic fiber OGr(g, 2g)F , the orthogonal Grassmannian of totally isotropic g-planes in F 2g. But contrary to the symplectic and linear cases -- and the reason here for the adjective "naive" -- the naive local model is typically not flat over OF , as was first observed by Genestier [Ge2]. A major source of trouble is the fact that the orthogonal Grassmannian is not connected, but has two connected components. To fix ideas, let us suppose then (cid:99)Λ(cid:48)(cid:48) = πΛ(cid:48)(cid:48). Given an R-point (FΛ)Λ∈L of M naive that L contains a self-dual lattice Λ(cid:48) and a lattice Λ(cid:48)(cid:48) ⊃ Λ(cid:48) with dimk Λ(cid:48)(cid:48)/Λ(cid:48) = g; and periodicity conditions require that FΛ(cid:48)(cid:48) be totally isotropic for (cid:0)π−1( , )(cid:1) G,{µ},L, the perpendicularity condition requires that FΛ(cid:48) be totally isotropic for ( , )R, and the perpendicularity R, where we use a subscript R to denote base change to R. Hence we get a map (2.3.1) M naive G,{µ},L (FΛ)Λ∈L / OGr(g, 2g)OF × OGr(g, 2g)OF / (FΛ(cid:48),FΛ(cid:48)(cid:48)). Now, quite generally, a scheme X over a regular, integral, 1-dimensional base scheme is flat if and only if the scheme-theoretic closure in X of the generic In our present situation, the target space in (2.3.1) fiber of X is equal to X. has 4 connected components, 2 of which contain the image of the 2 connected G,{µ},L. But for any g ≥ 1, the image of components of the generic fiber of M naive 3Note that G is disconnected, so that it does not honestly fit into the framework of §1. See the discussion after Remark 2.3.4. / / G. Pappas, M. Rapoport, and B. Smithling 19 G,{µ},L always meets another component; see [PR4, 8.2] for a simple example G,{µ},L is M naive which is easy to generalize to higher rank. Hence the generic fiber of M naive not dense in M naive G,{µ},L, so that M naive G,{µ},L is not flat. G,{µ},L to be the scheme-theoretic closure in M naive To correct for non-flatness of the naive local model, one simply defines the true local model M loc G,{µ},L of its generic fiber. Then M loc G,{µ},L is flat essentially by definition, but a priori it car- ries the disadvantage of not admitting a ready moduli-theoretic description. In [PR4] a remedy for this disadvantage is proposed in the form of a new condi- tion, called the spin condition, which is added to the moduli problem defining M naive G,{µ},L. Unfortunately the spin condition is a bit technical to formulate; we In the simple case that L refer to [PR4, §§7.1, 8.2] or [Sm1, §2.3] for details. consists of the homothety classes of a self-dual lattice Λ(cid:48) and a lattice Λ(cid:48)(cid:48) satisfy- ing (cid:99)Λ(cid:48)(cid:48) = πΛ(cid:48)(cid:48), the map (2.3.1) is a closed embedding, and the effect of the spin condition is simply to intersect M naive OGr(g, 2g)OF × OGr(g, 2g)OF marked by the generic fiber of M naive general L, the spin condition becomes more complicated. G,{µ},L with the two connected components of G,{µ},L. For more In general, let M spin G,{µ},L denote the closed subscheme of M naive G,{µ},L that clas- sifies points satisfying the spin condition. The inclusion M spin G,{µ},L is shown in [PR4] to be an isomorphism on generic fibers, and we then have the following. Conjecture 2.3.2 ([PR4, Conj. 8.1]). For any periodic self-dual lattice chain L, M spin G,{µ},L is flat over SpecOF . G,{µ},L ⊂ M naive G,{µ},L, that is, M spin G,{µ},L = M loc Hand calculations show that M spin G,{µ},L is indeed flat with reduced special fiber for n ≤ 3; see [PR4, §8.3] for some explicit examples for n = 1 and 2. The main result of [Sm1] is the following weakened form of the conjecture (for arbitrary n), the full version of which is still open. Recall that a lattice chain is complete if all successive quotients have k-dimension 1, and that a scheme over a regular, integral, 1-dimensional base scheme is topologically flat if its generic fiber is dense. Theorem 2.3.3 ([Sm1, 7.6.1]). For any complete periodic self-dual lattice chain L, M spin G,{µ},L is topologically flat over SpecOF ; or in other words, the underlying (cid:3) topological spaces of M spin Remark 2.3.4. For sake of unformity assume g ≥ 4. In proving Theorem 2.3.3, it suffices to take L to be the standard lattice chain Λ• (2.0.2). But from a building- theoretic perspective, it is more natural to instead consider the periodic self-dual lattice oriflamme G,{µ},L and M loc G,{µ},L coincide. ··· ⊂ Λ0 ⊂ Λ0(cid:48) ⊂ ⊂ Λg ⊂ Λg(cid:48) ⊂ ⊂ Λ2 ⊂ ··· ⊂ Λg−2 ⊂ Λg+2 ⊂ ··· ⊂ Λ2g−2 ⊂ Λ2g ⊂··· ⊂ Λ2g(cid:48) ⊂ , 20 Local models of Shimura varieties, I. (cid:19) π−aOF el (cid:18)2g−1(cid:88) l=2 + π−a+1OF e2g where, for a ∈ Z, Λ2ga(cid:48) := π−a−1OF e1 + and Λ(g+2ga)(cid:48) := (cid:18)g−1(cid:88) l=1 (cid:19) π−a−1OF el + π−aOF eg + π−a−1OF eg+1 + 2g(cid:88) l=g+2 π−aOF el. Then the lattice-wise fixer of Λ• in G(F ) is the same as that for the displayed oriflamme, namely the Iwahori subgroup B of elements in G(OF ) which are upper triangular mod π;4 and the parahoric subgroups of G◦(F ) that contain B are pre- cisely the parahoric stabilizers of periodic, self-dual subdiagrams of the displayed oriflamme. One can define a naive local model for the displayed oriflamme just as we have done for lattice chains, namely by specifying a locally direct summand of rank g for each lattice in the oriflamme, subject to functoriality, periodicity, and perpendicularity conditions. However this naive local model again fails to be flat: this time the four lattices Λ0, Λ0(cid:48), Λg, and Λg(cid:48) are all self-dual up to scalar, and one can see in a way very similar to what we discussed on p. 18 that the naive local model is not even topologically flat. One can see as in [PR4, §8.2] that it is necessary to impose a version of the spin condition, and we conjecture that the resulting spin local model is flat. As already noted, the treatment of the local model in this subsection does not honestly fall under the framework set out in §1, since GO2g is disconnected. But if we take the philosophy of §1 seriously, then we should expect to have local models for the connected group GO◦ 2g) and each of its minuscule coweights g−1 and g. Here we can simply define these two local models to be the respective Zariski closures of each of the two components of OGr(g, 2g)F in M naive G,{µ},L for GO2g is just the disjoint union of these two local models for GO◦ 2g. 2g (or its adjoint quotient P GO◦ G,{µ},L. In this way the local model M loc 2.4. Weil restriction of GLn We now begin to consider the simplest examples of local models for nonsplit groups. Let F0 be a discretely valued field with ring of integers OF0 and residue field k0. We suppose that F is a finite extension of F0 contained in a separable closure F sep of F0. Let d := [F : F0] and let e denote the ramification index of F/F0, so that e d. In this subsection we generalize our discussion of GLn in §2.1 to the group G := ResF/F0 GLn over F0. As in §2.1, we place no restrictions on the character- istic of k. Let L be a periodic OF -lattice chain in F n. Let F Gal denote the Galois 0 4One verifies easily that in fact B ⊂ G◦(OF ). G. Pappas, M. Rapoport, and B. Smithling 21 . Then we have the standard splitting upon base change to closure of F in F sep F Gal, 0 (2.4.1) GF Gal ∼= GLn, ϕ : F→F sep 0 (cid:89) (cid:88) where the product runs through the set of F0-embeddings ϕ : F → F sep such ϕ, choose an integer rϕ with 0 ≤ rϕ ≤ n; let 0 . For each let µϕ denote the cocharacter (cid:0)1(rϕ), 0(n−rϕ)(cid:1) of the standard maximal torus of r := rϕ; ϕ 0 0 diagonal matrices in GLn; let µ denote the geometric cocharacter of G whose ϕ- component, in terms of (2.4.1), is µϕ; and let {µ} denote the geometric conjugacy class of µ. Let E denote the reflex field of {µ}; this is easily seen to be the field of definition of µ, that is, the fixed field in F sep of the subgroup of the Galois group (cid:8) σ ∈ Gal(F sep 0 /F0)(cid:12)(cid:12) rσ◦ϕ = rϕ for all ϕ : F → F sep (cid:9). The naive local model M naive Plainly E ⊂ F Gal. Let OE denote the ring of integers in E. G,{µ},L attached to the triple (G,{µ},L) is the func- tor on the category of OE-algebras that assigns to each OE-algebra R the set of all families (FΛ)Λ∈L such that (i) for every Λ ∈ L, FΛ is an (OF ⊗OF0 locally on Spec R is a direct summand as an R-module of rank dn − r; (ii) for every inclusion of lattices Λ ⊂ Λ(cid:48) in L, the induced map Λ ⊗OF0 Λ(cid:48) ⊗OF0 (iii) for every a ∈ F × and every Λ ∈ L, the isomorphism Λ a−→∼ aΛ identifies FΛ (iv) (Kottwitz condition) for every a ∈ OF and every Λ ∈ L, the element a ⊗ 1 ∈ OF ⊗OF0 R)/FΛ as an R-linear endomorphism R acts on the quotient (Λ ⊗OF0 R)-submodule of Λ ⊗OF0 R carries FΛ into FΛ(cid:48); ∼−→ FaΛ; and R which Zariski- R → with characteristic polynomial (cid:0) a ⊗ 1(cid:12)(cid:12) (Λ ⊗OF0 R)/FΛ (cid:1) = (cid:89) (cid:0)X − ϕ(a)(cid:1)rϕ. charR ϕ : F→F sep Note that in the statement of the Kottwitz condition the polynomial(cid:81) (cid:0)X− ϕ(a)(cid:1)rϕ can first be regarded as a polynomial with coefficients in OE by definition of E, and then as a polynomial with coefficients in R via its OE-algebra structure. We remark that in [Ko1, §5] and [RZ, §3.23(a)], the Kottwitz condition is formu- lated in a different (but equivalent) way as a "determinant" condition. As always, G,{µ},L is plainly representable by a projective OE-scheme. M naive ϕ 0 When F is unramified over F0, upon base change to F Gal, M naive G,{µ},L becomes isomorphic to a product of local models for GLn of the form considered in §2.1. Hence (2.1.1) implies the following. 22 Local models of Shimura varieties, I. Theorem 2.4.2 (Gortz [Go1, 4.25]). Suppose F is unramified over F0. Then for any µ as above and any periodic OF -lattice chain L, M naive G,{µ},L is flat over SpecOE with reduced special fiber. The irreducible components of its special fiber are normal with rational singularities, so in particular are Cohen-Macaulay. (cid:3) G,{µ},L = M naive In general, i.e. in the presence of ramification, the naive local model need not be flat. As in the orthogonal case, the honest local model M loc G,{µ},L is then defined to be the scheme-theoretic closure in M naive G,{µ},L of its generic fiber; thus M loc G,{µ},L when F is unramified. Unfortunately, in contrast to the orthogonal case, it appears to be unreasonable to hope to give a simple, explicit, purely moduli-theoretic description of M loc G,{µ},L in general; see however [PR1, Th. 5.7], where just such a description is given in special cases under the hypothesis of the correctness of the conjecture of DeConcini and Procesi on equations defining the closures of nilpotent conjugacy classes in gln. To better focus on the issues at hand, we shall suppose henceforth that F is totally ramified over F0, i.e. that e = d. Although there seems to be no simple moduli-theoretic description of M loc G,{µ},L, there are at least two nontrivial descriptions of it that bear mention. For the first, note that whenever L(cid:48) is a subchain of L, there is a natural forgetful morphism ρL(cid:48) : M naive G,{µ},L −→ M naive G,{µ},L(cid:48). G,{µ},L → M naive In particular, for every lattice Λ ∈ L, we may consider its homothety class [Λ] ⊂ L and the projection M naive G,{µ},[Λ]; here the target space corresponds to the maximal parahoric case, as the stabilizer of Λ in G(F0) is a maximal parahoric subgroup. In [PR1, §8] it is proposed to describe M loc G,{µ},L by first taking the local model M loc G,{µ},[Λ] in the sense of the previous paragraph for each homothety class [Λ] ⊂ L, and then defining5 M vert G,{µ},L := (cid:92) [Λ]⊂L (cid:0)M loc ρ−1 [Λ] G,{µ},[Λ] (cid:1). In the maximal parahoric case, the special fiber of M loc G,{µ},[Λ] is integral and normal with only rational singularities [PR1, 5.4]. On the other hand, Gortz [Go4, Prop. 1] has shown that M vert G,{µ},L is topologically flat. These results can be combined to yield the following. Theorem 2.4.3 (Gortz [Go4, §1 Th.]; [PR2, 7.3], [PR1, 5.4]). For any µ as above and any periodic OF -lattice chain L, M vert G,{µ},L is flat over SpecOE. The special fiber of M loc G,{µ},L is reduced and its irreducible G,{µ},L, that is, M vert G,{µ},L = M loc 5Note that [PR1] uses the notation M loc to denote what we call M vert G,{µ},L, which a priori is different from our definition of M loc G,{µ},L. G. Pappas, M. Rapoport, and B. Smithling 23 components are normal with rational singularities, so in particular are Cohen- Macaulay. When L consists of a single lattice homothety class, the special fiber of (cid:3) M loc G,{µ},L is moreover irreducible. G,{µ},L. The definition of M vert Note that if a moduli-theoretic description of the local models M loc G,{µ},[Λ] can be found, then there would clearly also be a moduli-theoretic description of M vert G,{µ},L is closely related to the combinatorial notion of vertexwise admissibility, which we shall take up in §4.5. G,{µ},L makes use of the splitting model MG,{µ},L defined in [PR2, §5]. We shall not recall the details of the definition here. Roughly speaking, MG,{µ},L is a projective scheme defined over the ring of integers OF Gal in F Gal which represents a rigidified version of the moduli problem defining M naive G,{µ},L. There are canonical morphisms The second description of M loc MG,{µ},L −→ M naive G,{µ},L ⊗OE OF Gal −→ M naive G,{µ},L. The canonical local model M can G,{µ},L is defined to be the scheme-theoretic image in M naive G,{µ},L of the composite. It is shown in [PR2] that the first displayed arrow is an isomorphism on generic fibers (the second is trivially an isomorphism after base change to F Gal, of course) and that MG,{µ},L can be identified with a certain twisted product of local models for GLn over SpecOF Gal , so that MG,{µ},L is flat. One then obtains the following. Theorem 2.4.4 ([PR2, 5.1, 5.3]). For any µ as above and any periodic OF -lattice chain L, M can (cid:3) G,{µ},L = M loc G,{µ},L. Note that although M loc G,{µ},L itself does not appear to admit a ready moduli- theoretic description, the theorem exhibits it as the image of a canonical morphism between schemes that do. As pointed out by Haines, the splitting model can be used to give a second G,{µ},L that bypasses part of the proof of topological flatness proof of flatness for M vert of Gortz. See [PR2, 7.5] and [Go4, §5 Rem.]. 2.5. Weil restriction of GSp2g In addition to Weil restrictions of GLn, local models for Weil restrictions of GSp2g have also been studied in some detail. In this subsection we shall very briefly survey their theory, outsourcing essentially all of the details to the papers [PR2] and [Go4]. Let G := ResF/F0 GSp2g, and otherwise continue with the assumptions and notation of the previous subsection. For each F0-embedding ϕ : F → F sep , let rϕ := g. Let µ denote the resulting geometric cocharacter of ResF/F0 GL2g, regard µ as a geometric cocharacter for G, and let {µ} denote the geometric conjugacy class of µ for G. Then the reflex field of {µ} is F0. Let L be a periodic OF - lattice chain in F 2g which is "self-dual" in the sense of [PR4, §8] or [Go4, §6]. 0 24 Local models of Shimura varieties, I. G,{µ},L attached to (G,{µ},L) is the closed subscheme of The naive local model M naive M naive ResF /F0 GL2g,{µ},L whose points satisfy a perpendicularity condition relative to every pair of dual lattices in L, in close analogy with the perpendicularity condition in §2.2; again see [PR4, §8] or [Go4, §6]. Essentially all the results in the previous subsection are known to carry over to the present setting. For unramified extensions we have the following. Theorem 2.5.1 (Gortz [Go2, §2 Rem. (ii)]). Suppose F is unramified over F0. Then for any self-dual periodic OF -lattice chain L, M naive G,{µ},L is flat over SpecOF0 with reduced special fiber. The irreducible components of its special fiber are normal (cid:3) with rational singularities, so in particular are Cohen-Macaulay. Let us suppose for the rest of the subsection that F/F0 is totally ramified. G,{µ},L is flat (but see Conjecture 2.5.4 below), and G,{µ},L of its generic In [PR2, Display 12.2], there is defined a natural "vertexwise" analog of in loc. Then it is not known whether M naive we define M loc fiber. M vert cit.). We then have the following. G,{µ},L to be the scheme-theoretic closure in M naive ResF /F0 GL2g,{µ},L, which we denote by M vert G,{µ},L (this is denoted by N loc I Theorem 2.5.2 (Gortz [Go4, Prop. 3]; [PR2, Ths. 12.2, 12.4]). For any self-dual periodic OF -lattice chain L, M vert G,{µ},L = M loc G,{µ},L is flat over SpecOF0. The special fiber of M loc G,{µ},L is reduced and its irreducible components are normal with rational singularities, so in particular are Cohen-Macaulay. When L is a minimal self-dual periodic lattice chain, the special fiber of M loc G,{µ},L is (cid:3) moreover irreducible. G,{µ},L, that is, M vert Gortz's contribution to Theorem 2.5.2 is to show that M vert G,{µ},L is topologi- cally flat. In fact he proves the following stronger result. Theorem 2.5.3 (Gortz [Go4, Prop. 3]). For any self-dual periodic OF -lattice chain L, M naive (cid:3) G,{µ},L is topologically flat over SpecOF0. G,{µ},L, M vert G,{µ},L, and M naive Thus M loc G,{µ},L all coincide at the level of topological spaces. Gortz furthermore conjectures that they are equal on the nose. Conjecture 2.5.4 (Gortz [Go4, §6 Conj.]). For any self-dual periodic OF -lattice chain L, M naive G,{µ},L is flat over SpecOF0. Note that the conjecture stands in contrast to the case of ResF/F0 GLn, where the naive local model may even fail to be topologically flat. We finally mention that, in analogy with the previous subsection, the notions of splitting model and canonical local model are also developed in [PR2] in the setting of local models for G. We refer to loc. cit. for details, where, in particular, it is shown that the canonical local model equals M loc G,{µ},L. G. Pappas, M. Rapoport, and B. Smithling 25 1 ); (A(2) n−1, ∨ s ), 1 ≤ s ≤ 2.6. Ramified, quasi-split unitary (types (A1, ∨ n − 1)6 In this subsection we take up another typical example of a group that splits only after a ramified base extension, namely ramified, quasi-split GUn. We suppose n ≥ 2 and char k (cid:54)= 2. We continue with the notation of the previous subsection, but we now restrict to the special case that F/F0 is ramified quadratic. To simplify matters, assume that π (cid:55)→ −π under the nontrivial automorphism of F/F0, so that π0 = π2 is a uniformizer of F0. Let φ denote the F/F0-Hermitian form on F n whose matrix with respect to the standard ordered basis is Hn (2.2.2). We attach to φ the alternating F0-bilinear form (cid:104) , (cid:105) : V × V (x, y) (cid:0)π−1φ(x, y)(cid:1). / 1 2 TrF/F0 Given an OF -lattice in F n, we denote by (cid:98)Λ its common (cid:104) , (cid:105)- and φ-dual, (cid:9). As usual, (cid:104) , (cid:105) induces a perfect OF0-bilinear pairing (cid:98)Λ :=(cid:8) x ∈ F n(cid:12)(cid:12) (cid:104)Λ, x(cid:105) ⊂ OF0 (cid:9) =(cid:8) x ∈ F n(cid:12)(cid:12) φ(Λ, x) ⊂ OF / F0 Λ ×(cid:98)Λ −→ OF0 for any OF -lattice Λ; and we say that an OF -lattice chain in F n is self-dual if it is closed under taking duals. Let G := GUn := GU (φ) over F0, and let L be a periodic self-dual OF -lattice chain in F n. Although we shall define local models for any such L, when n is even, facets in the building only correspond to L with the property that (∗) if L contains a lattice Λ such that πΛ ⊂ (cid:98)Λ ⊂ Λ and dimk(cid:98)Λ/πΛ = 2, then L Such a Λ(cid:48) then satisfies (cid:98)Λ(cid:48) = πΛ(cid:48). See [PR3, §4.a], [PR4, §1.2.3]. also contains a lattice Λ(cid:48) ⊃ Λ with dimk Λ(cid:48)/Λ = 1. Over F we have the standard splitting (f,c)−−−→∼ GLn × Gm, GF (2.6.1) where c : GF → Gm is the similitude character and f : GF → GLn is given on R-points by the map on matrix entries R ⊗F0 F x ⊗ y / R / xy for an F -algebra R. Let D denote the standard maximal torus of diagonal matrices in GLn. Choose a partition n = r + s; we refer to the pair (r, s) as the signature. Let µ denote the cocharacter (cid:0)1(s), 0(r); 1(cid:1) of D × Gm. Then we may regard µ 6Note that type A(2) 3 does not actually appear in the table in §1.2. Rather the adjoint group P GU4 is of type D(2) 3 . / / / / 26 Local models of Shimura varieties, I. The naive local model M naive as a geometric cocharacter of G via (2.6.1), and we denote by {µ} its geometric conjugacy class. We denote by E the reflex field of {µ}; then E = F0 if r = s and E = F otherwise. Let OE denote the ring of integers in E. G,{µ},L is the functor on the category of OE-algebras that assigns to each OE-algebra R the set of all families (FΛ)Λ∈L such that (i) for every Λ ∈ L, FΛ is an (OF ⊗OF0 (ii) for every inclusion of lattices Λ ⊂ Λ(cid:48) in L, the induced map Λ ⊗OF0 Λ(cid:48) ⊗OF0 (iii) for every a ∈ F × and every Λ ∈ L, the isomorphism Λ a−→∼ aΛ identifies FΛ (iv) for every Λ ∈ L, the perfect R-bilinear pairing locally on Spec R is a direct summand as an R-module of rank n; R)-submodule of Λ ⊗OF0 R carries FΛ into FΛ(cid:48); R which Zariski- ∼−→ FaΛ; R(cid:1) −→ R induced by (cid:104) , (cid:105) identifies F⊥ (v) (Kottwitz condition) for every a ∈ OF and every Λ ∈ L, the element a ⊗ 1 ∈ OF ⊗OF0 R)/FΛ as an R-linear endomorphism (cid:0)Λ ⊗OF0 Λ ⊂(cid:98)Λ ⊗OF0 R acts on the quotient (Λ ⊗OF0 (cid:0) a ⊗ 1(cid:12)(cid:12) (Λ ⊗OF0 R(cid:1) ×(cid:0)(cid:98)Λ ⊗OF0 R with F(cid:98)Λ; (cid:1) = (X − a)r(X − a)s, with characteristic polynomial R)/FΛ charR R → where we use a bar to denote the nontrivial automorphism of F/F0. (cid:0)X 2 − (a + a)X + aa(cid:1)s When r = s, the right-hand side of the last display is to be interpreted as . The Kottwitz condition is equivalent to requiring the "determinant" condition that for every Λ ∈ L, we have an equality of polynomials with coefficients in R (cid:0) X(1 ⊗ 1) + Y (π ⊗ 1)(cid:12)(cid:12) (Λ ⊗OF0 detR (cid:1) = (X + Y π)r(X − Y π)s, R)/FΛ where 1 ⊗ 1, π ⊗ 1 ∈ OF ⊗OF0 requiring that the single element π ⊗ 1 acts on (Λ ⊗OF0 polynomial (X − π)r(X + π)s. R; and these conditions are mutually equivalent to R)/FΛ with characteristic As always, the naive local model is representable by a closed subscheme If we denote by V the n- of a finite product of Grassmannians over SpecOE. dimensional F -vector space then the map V := ker( π ⊗ 1 − 1 ⊗ π F n ⊗F0 F ), (FΛ)Λ (cid:55)−→ ker( π ⊗ 1 − 1 ⊗ π FΛ ) (independent of Λ) defines an isomorphism from the F -generic fiber M naive F onto the Grassmannian Gr(s, V )F . It was observed in [P1] that M naive G,{µ},L fails to be flat in general; historically, this was the first time it was found that the Rapoport -- Zink local model can fail to G,{µ},L⊗OE G. Pappas, M. Rapoport, and B. Smithling 27 R. Hence, when R is reduced, π⊗1 necessarily acts on (Λ⊗OF0 be flat. The key point is that the Kottwitz condition fails to impose a condition on the reduced special fiber. Indeed, if R is a k-algebra, then π ⊗ 1 is nilpotent in OF⊗OF0 R)/FΛ with characteristic polynomial X n, in accordance with the Kottwitz condition. Thus the reduced special fiber is independent of the signature. Hence by Chevalley's theorem (EGA IV.13.1.5), the special fiber has dimension ≥ max{dim M naive G,{µ},L ⊗OE E}0≤s≤n = max{dim Gr(s, V )}0≤s≤n = (cid:106) n (cid:107)(cid:108) n (cid:109) 2 2 . The max in the display is achieved for r − s ≤ 1. Thus M naive G,{µ},L is not flat for r − s > 1, as its generic and special fibers have different dimension. We note that the analogous argument given in the proof of [P1, Prop. 3.8(b)] should be amended to use the reduced special fiber in place of the honest special fiber. G,{µ},L to be the scheme-theoretic closure in M naive As always, one remedies for non-flatness of the naive local model by defining the honest local model M loc G,{µ},L of its generic fiber. Although less is known about M loc G,{µ},L for ramified GUn than for ramified ResF/F0 GLn and ResF/F0 GSp2g, there are by now a number of results that have been obtained in various special cases. In low rank, the case n = 3 has been completely worked out. Theorem 2.6.2 ([P1, 4.5, 4.15], [PR4, §6]). Let n = 3 and (r, s) = (2, 1). (i) Let L be the homothety class of the lattice Λ0 = On G,{µ},L = M loc G,{µ},L is normal and Cohen-Macaulay, it is smooth outside a single point y in its special fiber, and its special fiber is integral and normal and has a rational singularity at y. The G,{µ},L at y is regular with special fiber a reduced union of G,{µ},L is flat over SpecOF . Moreover, M naive blowup (cid:102)M loc F ⊂ F n. Then M naive G,{µ},L, that is, M naive G,{µ},L → M loc two smooth surfaces meeting transversely along a smooth curve. (ii) Let L = [Λ1, Λ2], the lattice chain consisting of the homothety classes of Λ1 and G,{µ},L is smooth over SpecOF with geometric special fiber isomorphic Λ2. Then M loc to P2. (iii) Let L be the standard maximal lattice chain in F 3. Then M loc G,{µ},L is normal and Cohen-Macaulay. Its special fiber is reduced and consists of two irreducible components, each normal and with only rational singularities, which meet along (cid:3) two smooth curves which, in turn, intersect transversally at a point. We shall discuss the case n = 2 at the end of the subsection in Remark 2.6.13. Remark 2.6.3. In each case in the theorem, the stabilizer of L in G(F0) is a parahoric subgroup. In this way the three cases correspond to the three conjugacy classes of parahoric subgroups in G(F0). See [PR4, §1.2.3(a)]. Remark 2.6.4. Quite generally, for any fixed n and L, it is elementary to verify that M naive G,{µ},L) is unchanged up to isomorphism if we replace the signature (r, s) with (s, r). Moreover, it is easy to see from the wedge condition G,{µ},L (and hence M loc 28 Local models of Shimura varieties, I. G,{µ},L is just SpecOF itself in case r or s is 0. So the discussed below that M loc theorem covers all cases of interest when n = 3. For larger n, results on M loc G,{µ},L are known for cases of simple signature and for cases of simple lattice chains L. One important tool for proving reducedness of the special fiber is Hironaka's Lemma (EGA IV.5.12.8). G,{µ},L is integral and normal and has only rational singularities. Theorem 2.6.5. (i) ([PR4, Th. 5.1]; Arzdorf [A, Th. 2.1], Richarz [Ri2, Cor. 5.6]) Let n ≥ 3. Suppose that n is even and L = [Λn/2], or that n = 2m + 1 is odd and L = [Λ0] or L = [Λm, Λm+1]. Then for any signature (r, s), the special fiber of M loc (ii) ([P1, 4.5]) Let n ≥ 2, (r, s) = (n − 1, 1), and L = [Λ0]. Then M loc G,{µ},L is normal and Cohen-Macaulay, and it is smooth over SpecOE outside a single point y in the special fiber. For n = 2, M loc G,{µ},L is regular and its special fiber is a G,{µ},L divisor with simple normal crossings. For n ≥ 3, the blowup (cid:102)M loc G,{µ},L → M loc G,{µ},L is smooth. at y is regular with special fiber a divisor with simple normal crossings. (iii) ([PR4, §5.3]; Richarz [A, Prop. 4.16]) Let n ≥ 3 and (r, s) = (n−1, 1). Suppose that n is even and L = [Λn/2] or that n = 2m + 1 is odd and L = [Λm, Λm+1]. (cid:3) Then M loc Remark 2.6.6. In (i), the cases n even, L = [Λn/2] and n odd, L = [Λ0] are in [PR4], and the other is due to Arzdorf. A different proof is due to Richarz [Ri2]. The significance of the assumptions on n and L is that, up to G(F0)-conjugacy, these are all the cases that correspond to special maximal parahoric level structure, i.e. the parahoric stabilizer of L in G(F0) is the parahoric subgroup corresponding to a vertex in the building which is special in the sense of Bruhat -- Tits theory. See [PR4, §1.2.3]. Remark 2.6.7. The blowup (cid:102)M loc G,{µ},L occurring in (ii) is described explicitly by Kramer in [Kr] in terms of a moduli problem analogous to the Demazure resolution (cid:102)M loc of a Schubert variety in the Grassmannian. She shows that the special fiber of G,{µ},L consists of two smooth irreducible components of dimension n − 1 -- one of which, the fiber over y, being isomorphic to Pn−1 , the other one being a P1 k-bundle over a smooth quadric -- which intersect transversely in a smooth irreducible variety of dimension n − 2. k Remark 2.6.8. In (iii), the case of n even is in [PR4], and the case of n odd is due to Richarz. The result in the former case is not directly stated in [PR4], but it follows from the cited reference, where it is shown that M loc G,{µ},[Λn/2] has an open neighborhood around its "worst point" isomorphic to An−1OF (note that in the last sentence of [PR4, §5.3], 1Ur,s should be replaced by 1U∧ r,s). In the cases n even, L = [Λn/2] and n odd, L = [Λ0], the local models are never smooth outside the cases enumerated in (iii) (up to switching (r, s) and (cid:94)s+1 R (cid:94)r+1 R G. Pappas, M. Rapoport, and B. Smithling 29 (s, r), cf. Remark 2.6.4), provided that the signature is nontrivial, i.e., r (cid:54)= 0 or s (cid:54)= 0, see [Ri1, Th. 3.15]. Probably the same holds for the case n = 2m + 1 is odd and L = [Λm, Λm+1]. In light of the failure of M naive G,{µ},L to be flat in general, it is an interesting problem to obtain a moduli-theoretic description of M loc G,{µ},L. Motivated by the Kottwitz condition's failure to impose a condition on the reduced special fiber, in [P1] the following additional condition is introduced to the moduli problem defining M naive (vi) (wedge condition) if r (cid:54)= s, then for every Λ ∈ L, we have G,{µ},L: (π ⊗ 1 + 1 ⊗ π Fi) = 0 and (π ⊗ 1 − 1 ⊗ π Fi) = 0. (There is no condition when r = s.) The wedge local model M∧ G,{µ},L is the closed subscheme of M naive G,{µ},L that classifies points satisfying the wedge condition. It is easy to see that the wedge and naive local models have common generic fiber, and under the special hypotheses of Theorem 2.6.5(ii) it has been shown that the wedge condition cuts out the flat closure M loc Proposition 2.6.9 ([P1, 4.5]). Let n ≥ 2, (r, s) = (n − 1, 1), and L = [Λ0]. Then (cid:3) M∧ G,{µ},L. G,{µ},L = M loc G,{µ},L. More generally, it is conjectured in [P1] that M∧ G,{µ},[Λn/2] is not even topologically flat. G,{µ},[Λ0] is flat for any n and any signature.7 But for more general lattice chains, the wedge condition turns out to be insufficient [PR4, Rems. 5.3, 7.4]. For example, for n = 3 and G,{µ},L for L = [Λ1, Λ2] and L the standard lattice (r, s) = (2, 1), the schemes M∧ chain are topologically flat but not flat. And for n even and r (cid:54)= 0 or s (cid:54)= 0, the scheme M∧ Remark 2.6.10. When n is even, [PR4] only shows that M∧ G,{µ},[Λn/2] is not topologically flat for r and s odd. But the same holds for r and s even, provided neither is 0: for example, the point denoted F1 in [PR4, §5.3] is not in the closure of the generic fiber in this case. Accordingly, for n even, [PR4, Rem. 5.3(a)] should be corrected to say that Conjecture 5.2 in loc. cit. implies that M∧ G,{µ},[Λn/2] contains G,{µ},[Λn/2] as an open subscheme for any signature, not that M∧ M loc G,{µ},[Λn/2] = M loc G,{µ},[Λn/2] for r and s even. (Here the corrected statement allows for r and s to be odd as well as even, since the odd case reduces to the even case, as follows from [PR4, §5.3].) See Conjecture 6.1.13 below for a statement of Conjecture 5.2 in loc. cit. G,{µ},[Λ0] coincides with M∧ 7We mention that it is also conjectured that M naive G,{µ},[Λ0] for r − s ≤ 1. This is proved for n equal to 2 and 3 in [P1, 4.5, 4.15]. M loc G,{µ},[Λ0] and 30 Local models of Shimura varieties, I. More precisely, one verifies at once that the perfect pairing Λn/2 × Λn/2 id×π−−−→∼ Λn/2 × Λ−n/2 (cid:104) , (cid:105)−−→ OF0 is split symmetric. Hence M naive G,{µ},[Λn/2] naturally embeds as a closed subscheme of OGr(n, 2n)OE . Then [PR4, Conj. 5.2], together with the topological flatness result Theorem 2.6.12 below, implies that M loc G,{µ},[Λn/2] with the connected component of OGr(n, 2n)OE marked by the common generic fiber of M naive G,{µ},[Λn/2] is the intersection of M∧ G,{µ},[Λn/2] and M∧ G,{µ},[Λn/2]. G,{µ},L, one can still hope to describe M loc Although the wedge condition is not sufficient in general to cut out the local model inside M naive G,{µ},L via a further refine- ment of the moduli problem. In [PR4] it is shown that, in addition to the wedge condition, M loc G,{µ},L satisfies a close analog of the spin condition that arose in the setting of even orthogonal groups in §2.3, which is again called the spin condition. In the setting of Remark 2.6.10, with n even and L = [Λn/2], the spin condition amounts exactly to intersecting M naive G,{µ},[Λn/2] with the connected component of OGr(n, 2n)OE marked by the generic fiber of M naive G,{µ},[Λn/2]. In general the spin condition is more complicated, and we shall just refer to the source papers for its formulation: see [PR4, §7.2] or [Sm3, §2.5] (the latter contains a correction to a minor sign error in the former). As in the orthogonal case, we denote by M spin G,{µ},L that classifies points satisfying the spin condition, and we have the following. Conjecture 2.6.11 ([PR4, Conj. 7.3]). Let L be a periodic self-dual OF -lattice chain, satisfying property (∗) from the beginning of the subsection if n is even. Then for any n ≥ 3 and any signature, M spin G,{µ},L is flat over SpecOE. G,{µ},L the closed subscheme of M∧ G,{µ},L, that is, M spin G,{µ},L = M loc Although the conjecture remains open, there is the following result, in anal- ogy with the orthogonal case. Theorem 2.6.12 ([Sm3, Main Th.], [Sm4]). For any n ≥ 3, any signature, and G,{µ},L is topologically flat over SpecOE. If n any L as in Conjecture 2.6.11, M spin (cid:3) is odd, then M∧ In the special case n = 2m + 1 is odd, (r, s) = (n− 1, 1), and L = [Λm, Λm+1] G,{µ},L also follows from mentioned in Theorem 2.6.5(iii), topological flatness of M∧ [A, Prop. 4.16]. G,{µ},L is also topologically flat. We emphasize that for odd n, although M spin G,{µ},L coincide as topological spaces, their scheme structures really do differ in general, and it is only the spin local model that is conjectured to be flat. By contrast, for even n, M spin G,{µ},L typically do not even agree at the level of topological spaces; see [Sm4]. G,{µ},L and M∧ G,{µ},L and M∧ G. Pappas, M. Rapoport, and B. Smithling 31 Remark 2.6.13 (GU2). To be able to treat the Bruhat-Tits-theoretic aspects of GUn in a uniform way, the paper [PR4] omits the case n = 2. Let us briefly discuss it now. The only nontrivial signature to worry about is (r, s) = (1, 1). In this case the naive and wedge local models coincide and are defined over SpecOF0. The derived group SU2 is isomorphic to SL2, which is split. Each alcove in the building has two vertices, both of which are special and GU2(F0)-conjugate. Thus there are essentially two cases to consider: the (special) maximal parahoric case and the Iwahori case. First take L = [Λ1]. Then the stabilizer in GU2(F0) of Λ1 is a maximal parahoric subgroup. The naive local model is a closed subscheme of Gr(2, Λ1)OF0 , and by restricting standard open affine charts of the Grassmannian, [PR4, §5.3] computes two affine charts on M naive GU2,{µ},[Λ1]. (Although, strictly speaking, [PR4] makes the blanket assumption n ≥ 3, the calculations in loc. cit. still go through for n = 2.) The first chart U1,1 identifies with the scheme of all 2 × 2 matrices X such that X 2 = π0 · Id, X t = −JXJ, where J = (cid:32) charX (T ) = T 2 − π0, . and (cid:33) 0 −1 1 0 One easily solves these equations to find that U1,1 = Spec k. The second affine chart 1U1,1 identifies with the scheme of all 2 × 2 matrices X such that X t = −JXJ, which, as noted in loc. cit., is the scheme of all scalar matrices X. Hence 1U1,1 A1OF0 to the local model, one finds that globally . By restricting the remaining standard affine charts on the Grassmannian ∼= GU2,{µ},[Λ1] = M∧ M naive GU2,{µ},[Λ1] ∼= P1OF0 (cid:113) Spec k. Of course this scheme is not flat because of the copy of Spec k. As described in Remark 2.6.10 and the paragraph preceding Theorem 2.6.12, M naive GU2,{µ},[Λ1] is inside Gr(2, Λ1)OF0 actually contained in OGr(2, Λ1)OF0 , and imposing the spin condition amounts to intersecting M naive GU2,{µ},[Λ1] with the connected component of OGr(2, Λ1)OF0 F0. In this way the spin condition visibly eliminates the extraneous copy of Spec k. For the Iwahori case we take L to be the standard lattice chain. To warm up, let us consider the naive local model associated just to the homothety class [Λ0], without worrying about the functoriality conditions attached to the inclusions Λ0 ⊂ Λ1 and Λ1 ⊂ π−1Λ0. Let x denote the k-point on M naive GU2,{µ},[Λ1] ⊗OF0 marked by M naive GU2,{µ},[Λ0] given by (π ⊗ 1) · (Λ0 ⊗OF0 k) ⊂ Λ0 ⊗OF0 k. 32 Local models of Shimura varieties, I. An affine chart for M naive scheme of all 2 × 2 matrices X such that8 X 2 = π0 · Id, X t = H2XH2, and charX (T ) = T 2 − π0, GU2,{µ},[Λ0] around x is described in [P1, p. 596 -- 7]: it is the where as always H2 is the antidiagonal unit matrix (2.2.2). Writing (cid:32) (cid:33) X = x11 x12 x21 x22 , one finds that this chart is given by SpecOF0[x12, x21]/(x12x21 − π0). Thus we find semistable reduction; in fact the global special fiber consists of two copies of P1 k meeting at the point x. GU2,{µ},L can now be obtained from M loc The full local model M loc GU2,{µ},[Λ0] and M loc GU2,{µ},[Λ1] by imposing the functoriality conditions attached to the inclu- sions Λ0 ⊂ Λ1 and Λ1 ⊂ π−1Λ0. We leave it to the reader to verify that in fact GU2,{µ},L ∼= M loc GU2,{µ},[Λ0], i.e. the submodule FΛ0 uniquely determines FΛ1, M loc without constraint. This fact admits a building-theoretic interpretation: the sta- bilizer of Λ0 in GU2(F0) is not a maximal parahoric subgroup, but rather, after passing to its connected component, the Iwahori subgroup fixing the entire stan- dard chain. Finally, note that the local model obtained in each of our two cases is isomorphic to the local model for GL2 in the analogous case, cf. Example 2.1.2. g , ∨ 2.7. Quasi-split but nonsplit orthogonal (types (D(2) g )) Again assume char k (cid:54)= 2 and let n = 2g − 2, n ≥ 4. Let V be the 2g- dimensional F -vector space on the ordered basis e, f , e1, . . . , e2g−2, and let ( , ) denote the symmetric F -bilinear form on V whose matrix with respect to this basis is9 g−1), (D(2) g , ∨  π 1  , H2g−2 with H2g−2 the anti-diagonal matrix (2.2.2). As always, we denote by (cid:98)Λ the ( , )- dual of any OF -lattice Λ in V , and ( , ) induces a perfect pairing Λ ×(cid:98)Λ → OF . Let G := GO(cid:0)( , )(cid:1) over F . Then G is quasi-split but not split. Consider the cocharacter(cid:0)1(g), 0(g)(cid:1) of GF (cid:39) GO2g given as in §2.3, and let {µ} denote its G(F )- conjugacy class over F . Let L be a periodic lattice chain in F 2g which is self-dual G,{µ},L is the closed OF -subscheme of for the form ( , ). The naive local model M naive M loc GL2g,{µ},L defined in the exactly the same way as for GSp2g and split GO2g, that is, we impose the duality condition (iv) with the understanding that all notation 8Note that [P1] has the condition X t = X instead of X t = H2XH2, owing to how the form ψp is defined there. 9It follows from Springer's Theorem [Lam, Th. VI.1.4] that, after passing to a sufficiently big unramified extension of F , any symmetric bilinear form on F 2g becomes isomorphic to ( , ) or to the split form ( , ) considered in §2.3. G. Pappas, M. Rapoport, and B. Smithling G,{µ},L has generic fiber OGr(cid:0)( , )(cid:1) 33 can see [PR4, 8.2.1] that OGr(cid:0)( , )(cid:1) is taken with respect to ( , ). Once again, M naive F , the orthogonal Grassmannian of totally isotropic g-planes in F 2g for the form ( , ). This Grassmannian is now connected although not geometrically connected. We F supports a canonical morphism to Spec K D) is the ramified quadratic extension of F obtained by extracting where K = F ( a square root of the discriminant D = (−1)gπ. (Let us remark here that the form G,{µ},L ⊗OF K is the split orthogonal ( , ) splits over K.) The base change M naive Grassmannian OGr(g, 2g)K which has two (geometrically) connected components. √ There is enough computational evidence to suggest the following. Conjecture 2.7.1. The scheme M naive G,{µ},L is topologically flat over SpecOF . This is in contrast to the case of split GO2g. However, the naive local model M naive G,{µ},L is still typically not flat and a version of the spin condition is needed. This condition is explained in [PR4, 8.2] where the reader can also find the conjecture that the corresponding spin local model M spin In fact, in this case, G,{µ},L is naturally an OK-scheme. The local model M loc M spin G,{µ},L is by definition G,{µ},L ⊗OF OK and is also naturally an the flat closure of M naive OK-scheme. Except for the results of a few calculations not much is known in this case. For more details, we refer to loc. cit. G,{µ},L ⊗OF K in M naive G,{µ},L is flat. Let us remark here that, similarly to the example in the split case of §2.3, the group G does not fit neatly into our framework of §1 since G is not connected. To define corresponding local models for the connected quasi-split group P GO◦(cid:0)( , )(cid:1) g of type D(2) and the cocharacters given as above, we can argue as follows: First note that the reflex field in this case is the quadratic ramified extension K of G,{µ},L ⊗OF F supports a canonical F as above. As above, the generic fiber M naive morphism to Spec K. We can now consider the flat closures of the two components G,{µ},L⊗OF of the orthogonal Grassmannian OGr(g, 2g)K = M naive OK. These two schemes over SpecOK give by definition the local models for P GO◦(cid:0)( , )(cid:1) and the two PEL minuscule cocharacters ∨ G,{µ},L⊗OF K in M naive g−1, ∨ g . 3. Local models and flag varieties for loop groups A basic technique in the theory of local models, introduced by Gortz [Go1], is to embed the special fiber of the local model into an appropriate affine flag variety. In this section we discuss this and related matters, focusing on the representative examples of the linear and symplectic groups. Throughout this section we denote by k a field, by K := k((t)) the field of Laurent series in t with coefficients in k, and by OK := k[[t]] the subring of K of power series. 34 Local models of Shimura varieties, I. 3.1. Affine flag varieties For any contravariant functor G on the category of K-algebras, we denote by LG the functor on k-algebras LG : R (cid:55)−→ G(cid:0)R((t))(cid:1), where we regard R((t)) as a K-algebra in the obvious way. Similarly, for any contravariant functor P on the category of OK-algebras, we denote by L+P the functor on k-algebras L+P : R (cid:55)−→ P(cid:0)R[[t]](cid:1). If P is an affine OK-scheme, then L+P is an affine k-scheme. If G is an affine K- scheme, then LG is an ind-scheme expressible as the colimit of a filtered diagram of closed immersions between affine k-schemes. For our purposes, we shall be interested in the case that G and P are group- valued functors; then we call LG and L+P the loop group and positive loop group attached to G and P , respectively. Given such P , let Pη denote its generic fiber, and consider the fpqc quotient or in other words, the fpqc sheaf on k-algebras associated to the presheaf FP := LPη/L+P, R (cid:55)−→ P(cid:0)R((t))(cid:1)(cid:14)P(cid:0)R[[t]](cid:1). When P is smooth and affine, the following is a basic structure result on FP . It generalizes results of Faltings [Fa4]. Theorem 3.1.1 ([PR3, Th. 1.4]). Let P be a smooth affine group scheme over OK. Then FP is representable by an ind-scheme of ind-finite type over k, and the quotient morphism LPη → FP admits sections locally in the ´etale topology. (cid:3) Recall that an ind-scheme over k is of ind-finite type if it is expressible as a filtered colimit of k-schemes of finite type. For applications to local models, we are mainly interested in the sheaf FP in the case that G = Pη is a connected reductive group over K and P is a parahoric group scheme. Let us elaborate. Let G be a connected reductive group over the t-adically valued field K, let Gad denote its adjoint group, and consider the Bruhat -- Tits building B := B(cid:0)Gad(K)(cid:1). Let f be a facet in B. Then Bruhat -- Tits theory attaches to the pair (G, f ) the parahoric group scheme Pf ; this is a smooth affine OK-scheme with generic fiber G, with connected special fiber, and whose OK-points are identified with the corresponding parahoric subgroup of G(K). We make the following definition. Definition 3.1.2. Given a facet f in the building of the adjoint group of the connected reductive K-group G, the affine flag variety relative to f (or to Pf , or G. Pappas, M. Rapoport, and B. Smithling 35 to Pf (OK)) is the ind-scheme over k Ff := FPf = LG/L+Pf . In some cases, this mirrors the closely related constructions of (partial) affine flag varieties in the setting of the theory of Kac-Moody Lie algebras ([Kac], [Ku2]). See Remark 3.2.4 for more details on this relation. In concrete examples involving classical groups, one can often identify the affine flag variety with a space of lattice chains; this fact is crucial to the embedding of the special fibers of local models mentioned at the beginning of this section. Let R be a k-algebra, and consider the R((t))-module R((t))n for some n ≥ 1. Recall that a lattice in R((t))n is an R[[t]]-submodule L ⊂ R((t))n which is free as an R[[t]]-module Zariski-locally on Spec R, and such that the natural arrow L ⊗R[[t]] R((t)) → R((t))n is an isomorphism. We leave it as an exercise to check that it is equivalent to say that L is an R[[t]]-submodule of R((t))n such that tN R((t))n ⊂ L ⊂ t−N R((t))n for N sufficiently big, and such that t−N R((t))n/L is projective as an R-module for one, hence any, such N . All of the terminology for lattices from §2 admits an obvious analog in the present setting. A collection of lattices in R((t))n is a chain if it is totally ordered under inclusion and all successive quotients are projective R-modules (necessarily of finite rank). A lattice chain is periodic if t±1L is in the chain for every lattice L in the chain. In analogy with the definition of Λi (2.0.1), for i = na + j with 0 ≤ j < n, we define the OK-lattice j(cid:88) n(cid:88) (3.1.3) λi := t−a−1OKel + t−aOKel ⊂ K n, l=1 l=j+1 where now e1, . . . , en denotes the standard ordered basis in K n. The λi's form a periodic lattice chain ··· ⊂ λ−1 ⊂ λ0 ⊂ λ1 ⊂ ··· , which we again call the standard chain. More generally, let I ⊂ Z be any nonempty subset which is closed under addition by n; or in other words, I is the inverse image under the canonical projection Z → Z/nZ of a nonempty subset of Z/nZ. Then we denote by λI the periodic subchain of the standard chain consisting of all lattices of the form λi for i ∈ I. the Bruhat -- Tits building B(cid:0)P GLn(K)(cid:1) are in bijective correspondence with the Example 3.1.4 (GLn and SLn). Let G = GLn over K. Then the facets in periodic OK-lattice chains in K n, and the parahoric group schemes for G can be described as automorphism schemes of these lattice chains. More precisely, let us consider the chain λI for some nonempty I closed under addition by n; of course, every periodic OK-lattice chain in K n is G(K)-conjugate to λI for some such I. Let PI denote the automorphism scheme of λI as a periodic lattice chain over OK. 36 Local models of Shimura varieties, I. Then for any OK-algebra A, the A-points of PI consist of all families (3.1.5) GLA(λi ⊗OK A) (gi) ∈(cid:89) i∈I such that the isomorphism λi ⊗ A ta⊗idA−−−−−→∼ all i ∈ I and all a ∈ Z, and such that the diagram λi−na ⊗ A identifies gi with gi−na for λi ⊗OK A gi ∼ λi ⊗OK A λj ⊗OK A gj∼ / λj ⊗OK A commutes for all i < j in I. The scheme PI is a smooth OK-scheme with evident generic fiber G = GLn and whose OK-points identify with the full fixer in G(K) of the facet f corresponding to λI . Moreover, it is not hard to see that PI has connected special fiber. Hence PI is the parahoric group scheme Pf attached to f ; see [BTII, 1.7, 4.6, 5.1.9, 5.2.6]. For R a k-algebra, let Latn(R) denote the category whose objects are the R[[t]]-lattices in R((t))n and whose morphisms are the natural inclusions of lattices. Of course, any R[[t]]-lattice chain may be regarded as a full subcategory of Latn(R). We define FI to be the functor on k-algebras that assigns to each R the set of all functors L : λI → Latn(R) such that (C) (chain) the image L(λI ) is a lattice chain in R((t))n; (P) (periodicity) L(tλi) = tL(λi) for all i ∈ I, so that the chain L(λI ) is periodic; and (R) (rank ) dimk λj/λi = rankR L(λj)/L(λi) for all i < j. In more down-to-earth terms, an R-point of FI is just a periodic lattice chain in R((t))n indexed by the elements of I, such that the successive quotients have the same rank as the corresponding quotients in λI . on R((t))n, and it follows that the LG-equivariant map LG → FI specified by The loop group LG acts on FI via the natural representation of G(cid:0)R((t))(cid:1) taking the tautological inclusion(cid:0)λI (cid:44)→ L(K n)(cid:1) ∈ FI (k) as basepoint defines an LG-equivariant morphism ϕ : Ff −→ FI . In fact ϕ is an isomorphism: it is plainly a monomorphism, and it is an epimor- phism because every periodic lattice chain in R((t))n admits a so-called "normal form" Zariski-locally on Spec R, as is proved in [RZ, Ch. 3 App.]. Similar remarks apply to SLn over K. Up to conjugacy, the parahoric group schemes for SLn are again given by certain automorphism schemes P (cid:48) I of the chains λI for nonempty I closed under addition by n, where this time we consider families (gi) as in (3.1.5) satisfying the same conditions as above and such that det(gi) = 1 for all i. For given such I, let f(cid:48) denote the facet associated to P (cid:48) I and / / / F(cid:48) f(cid:48) = LSLn/L+P (cid:48) induces a monomorphism F(cid:48) for GLn. To describe F(cid:48) G. Pappas, M. Rapoport, and B. Smithling 37 I the associated affine flag variety. The inclusion SLn ⊂ GLn f(cid:48) (cid:44)→ Ff , where f again denotes the associated facet f(cid:48) as a space of lattice chains, we call a functor L : λI → Latn(R) special if (S)(cid:86)n R[[t]] L(λi) = t−iR[[t]] as a submodule of (cid:86)n that a point L ∈ Ff (R) is special as soon as(cid:86)n i ∈ I. Then the isomorphism ϕ above identifies F(cid:48) f(cid:48) with the subfunctor of FI of special points L; this is easy to check directly, or see [Go1, 3.5]. As a consequence, note R[[t]] L(λi) = t−iR[[t]] for a single R((t)) R((t))n = R((t)) for all i ∈ I. is r-special if (cid:86)n For applications to local models, it is convenient to consider not just the (cid:44)→ Ff , but the following variant, involving a simple canonical embedding F(cid:48) f(cid:48) generalization of the notion of special. For r ∈ Z, we say that a point L ∈ Ff (R) R[[t]] L(λi) = tr−iR[[t]] as a submodule of R((t)) for one, hence every, i ∈ I. Let I − r := { i − r i ∈ I }. Then the functor λi (cid:55)→ λi−r is an r-special point in Ff (k), and, taking it as basepoint, it determines an LSLn-equivariant isomorphism from F(cid:48) f(cid:48)(cid:48) onto the subfunctor in Ff of r-special points, where f(cid:48)(cid:48) is the facet for SLn corresponding to λI−r. Example 3.1.6 (GSp2g and Sp2g). Let φ denote the alternating K-bilinear form on K 2g whose matrix with respect to the standard basis is (cid:32) (cid:33) Hg , J2g = −Hg as in (2.2.1). We denote by G the K-group GSp2g := GSp(φ). To describe the parahoric group schemes for G, let I be a nonempty subset of Z closed under addition by 2g and multiplication by −1. For any i ∈ I and any OK-algebra R, the pairing φ induces a perfect R-bilinear pairing (λi ⊗OK R) × (λ−i ⊗OK R) φR−−→ R, where we use a subscript R to denote base change from OK to R. Let PI denote the OK-group scheme whose R-points consist of all families (gi) ∈(cid:89) GLR(λi ⊗OK R) i∈I satisfying the same conditions as in the GLn case and such that, in addition, there exists c ∈ R× such that (3.1.7) φR(gix, g−iy) = c · φR(x, y) 38 Local models of Shimura varieties, I. for all i ∈ I and all x ∈ λi ⊗OK R, y ∈ λ−i ⊗OK R. Then, analogously to the GLn case, PI is a parahoric group scheme for G, and up to conjugacy all parahoric group schemes arise in this way. Given nonempty I closed under addition by 2g and multiplication by −1, let f denote the associated facet in the Bruhat -- Tits building for Gad. To describe the affine flag variety attached to f , let R be a k-algebra and recall the lattice category Lat2g(R) from Example 3.1.4. For an R[[t]]-lattice Λ in R((t))2g, let (cid:98)Λ denote the φ-dual of Λ, that is, the R[[t]]-module (cid:98)Λ :=(cid:8) x ∈ R((t))2g(cid:12)(cid:12) φR((t))(Λ, x) ⊂ R[[t]](cid:9). We define FI to be the functor on k-algebras that assigns to each R the set of all functors L : λI → Lat2g(R) satisfying conditions (C), (P), and (R) from Example 3.1.4 and such that, in addition, (D) (duality) Zariski-locally on Spec R, there exists c ∈ R((t))× such that (cid:92)L(λi) = c · L(cid:0)(cid:98)λi the tautological inclusion (cid:0)λI (cid:44)→ L(K n)(cid:1) ∈ FI (k) as basepoint specifies an LG- Analogously to the GLn case, the loop group LG acts naturally on FI , and taking (cid:1) for all i ∈ I. equivariant isomorphism Ff ∼−→ FI . This description of FI is plainly equivalent to the lattice-theoretic description of the affine flag variety for GSp2g given in [PR2, §10] (except that the scalar denoted a there should only be required to exist Zariski-locally on Spec R). For the group Sp2g := Sp(φ) over K, up to conjugacy, the parahoric group schemes are again given by certain automorphism schemes P (cid:48) I of the chains λI for nonempty I closed under addition by 2g and multiplication by −1, namely, we now take the closed subscheme of PI of points for which c = 1 in (3.1.7). For given such I, let f(cid:48) denote the associated facet for Sp2g and F(cid:48) f(cid:48) the associated affine f(cid:48) (cid:44)→ Ff , flag variety. The inclusion Sp2g ⊂ GSp2g induces a monomorphism F(cid:48) with the subfunctor of FI of points L such that (cid:92)L(λi) = L((cid:98)λi) for all i ∈ I, i.e. where f again denotes the associated facet for GSp2g. In this way F(cid:48) f(cid:48) identifies such that c can be taken to equal 1 in (D). This subfunctor can also be described as the subfunctor of all special L such that the lattice chain L(λI ) is self-dual. As in the linear case, it is convenient to consider other embeddings besides the standard one F(cid:48) f(cid:48) (cid:44)→ Ff . This time we consider r-special L only for r ∈ gZ. Then the functor λi (cid:55)→ λi−r is an r-special point in Ff (k) whose image lattice chain is self-dual, and just as in the linear case, it specifies an LSp2g-equivariant isomorphism from F(cid:48) f(cid:48)(cid:48) onto the subfunctor in Ff of r-special points L such that L(λI ) is self-dual, where f(cid:48)(cid:48) is the facet for Sp2g corresponding to λI−r. Note that if L is r-special, then the scalar c appearing in (D) can be taken to equal t−r/g. G. Pappas, M. Rapoport, and B. Smithling 39 The affine flag varieties for other groups discussed in §2 can all be described similarly. For example, see [Sm1, §6.2] for GO2g (at least in the Iwahori case) and [PR4, §3.2] and [Sm3, §4.2] for ramified GUn. Returning to the general discussion, we conclude this subsection with a couple of further structure results from [PR3]. The first describes the connected compo- nents of loop groups and affine flag varieties in the case k is algebraically closed. Let G be a connected reductive group over K with k = k. Let π1(G) denote the fundamental group of G in the sense of Borovoi [Bor]; this can be described as the group X∗(T )/Q∨ of geometric cocharacters of T modulo coroots, where T is any maximal torus in G defined over K. Let K sep denote a separable closure of K. Then the inertia group I := Gal(K sep/K) acts naturally on π1(G), and we may consider the coinvariants π1(G)I . In [Ko2], Kottwitz constructs a functorial surjective homomorphism (3.1.8) which turns out to parametrize the connected components of LG and Ff as follows. G(K) (cid:16) π1(G)I Theorem 3.1.9 ([PR3, Th. 5.1]). Assume that k is algebraically closed. Then for any facet f , the Kottwitz homomorphism induces isomorphisms π0(LG) ∼−→ π0(Ff ) ∼−→ π1(G)I . (cid:3) In the special case that G is split we have π1(G)I = π1(G). Then the theorem may be regarded as an avatar of the familiar statement in topology, where LG plays the role of the loop space of G. The final result of the subsection (a generalization of a result of Faltings [Fa4]) concerns the (ind-)scheme structure on LG and Ff . Recall that an ind-scheme is reduced if it is expressible as a filtered colimit of reduced schemes. Theorem 3.1.10 ([PR3, Th. 6.1]). Assume that k is perfect and let G be a con- nected semi-simple K-group. Suppose that G splits over a tamely ramified exten- sion of K and that the order of the fundamental group π1(Gder) of the derived group Gder is prime to the characteristic of k. Then the ind-schemes LG and Ff , (cid:3) for any facet f , are reduced. We note that Theorem 3.1.10 is only an existence theorem. In [PR1, Prop. 6.6], in the case G = SLn, a candidate is proposed for writing the affine Grass- mannian LG/L+G as an increasing union of reduced projective subschemes. This candidate indeed works if char k = 0, or if n ≤ 2, cf. loc. cit. This is related to Remark 5.2.6 below. By contrast, if G is reductive but not semi-simple, then LG and Ff are necessarily non-reduced [PR3, Prop. 6.5]. We do not know if the assumption in the theorem that G splits over a tamely ramified extension of K is necessary. On the other hand, the assumption on the order of π1(Gder) appears to be: for 40 Local models of Shimura varieties, I. example, π1(P GL2) = Z/2Z and LP GL2 is non-reduced in characteristic 2 [PR3, Rem. 6.4]. 3.2. Schubert varieties cells, and analogously for f(cid:48)-Schubert varieties. In this subsection we discuss Schubert cells and varieties in affine flag vari- eties; these are the analogs in the context of loop groups of the usual notions for ordinary flag varieties. Let G be connected reductive over K, and let f and f(cid:48) be facets in B(Gad) contained in a common alcove. Definition 3.2.1. For g ∈ G(K) = LG(k), the associated f(cid:48)-Schubert cell in Ff , denoted Cg, is the reduced, locally closed subscheme of Ff whose underlying topological space is the image of L+Pf(cid:48) in Ff under the L+Pf(cid:48)-equivariant map sending 1 to the class of g.10 The associated f(cid:48)-Schubert variety in Ff , denoted Sg, is the Zariski closure of Cg in Ff endowed with its reduced scheme structure. We also refer to f(cid:48)-Schubert cells as Pf(cid:48)-Schubert cells or Pf(cid:48)(OK)-Schubert The f(cid:48)-Schubert cell Cg and the f(cid:48)-Schubert variety Sg in Ff only depend on the image of g in the double coset space Pf(cid:48)(OK)\G(K)/Pf (OK). For k al- gebraically closed, we shall see later in Proposition 4.1.8 that this double coset space can be identified with the Iwahori-Weyl group of G when f and f(cid:48) are a common alcove, and with a certain double coset space of the Iwahori-Weyl group in general. Note that, since Pf(cid:48) is smooth over OK with connected special fiber, it follows from [Gr, p. 264 Cor. 2] that L+Pf(cid:48) is reduced and irreducible. Hence each Schubert cell is irreducible. Hence each Schubert variety is reduced and irre- ducible (reducedness being imposed by definition). In general, the Schubert cells and Schubert varieties are subschemes of Ff of finite type over k; moreover the Schubert varieties are proper over k. The following theorem gives important information on the structure of Schu- bert varieties. Theorem 3.2.2 ([PR3, Th. 8.4]). Suppose that G splits over a tamely ramified extension of K and that the order of the fundamental group π1(Gder) of the derived group Gder is prime to the characteristic of k. Then all f(cid:48)-Schubert varieties in Ff are normal and have only rational singularities. If k has positive character- istic, then all f(cid:48)-Schubert varieties contained in a given f(cid:48)-Schubert variety are (cid:3) compatibly Frobenius split. We refer to [BK] for the notion of a scheme X in characteristic p being Frobenius split, and for a family of closed subschemes of X being compatibly Frobenius split. This property has important consequences for the local structure: 10Note that Cg is not a topological cell when f is not an alcove, i.e., it is not isomorphic to an affine space. G. Pappas, M. Rapoport, and B. Smithling 41 if X is Frobenius split, then X is reduced and weakly normal, cf. [BK, §1.2]. Also, if {X1, . . . , Xn} is a family of compatibly split closed subschemes of X, then their (reduced) union X1 ∪ ··· ∪ Xn and their intersection X1 ∩ ··· ∩ Xn are also compatibly split; in particular, X1 ∩ ··· ∩ Xn is reduced. Frobenius splitness also has interesting global consequences, such as strong forms of the Kodaira vanishing property, cf. loc. cit. After introducing the Iwahori-Weyl group in §4, we will give in Propositions 4.2.10 and 4.2.8, respectively, the dimension of Schubert varieties and their inclu- sion relations in terms of the combinatorics of the Iwahori-Weyl group. Remark 3.2.3. In [PR3] f(cid:48)-Schubert cells and varieties are only defined, and Theorem 3.2.2 is only formulated and proved, in the case that f = f(cid:48). But the method of proof involves a reduction to the case that f = f(cid:48) is an alcove, and this reduction step works just as well for any f(cid:48)-Schubert variety in Ff in the sense defined here. See [PR3, Rem. 8.6, §8.e.1]. We shall see how f(cid:48)-Schubert varieties in Ff with f(cid:48) (cid:54)= f arise naturally in the context of local models in the next subsection. Remark 3.2.4. In the case that the group G is split semi-simple and simply connected, Theorem 3.2.2 is due to Faltings [Fa4]. Let us mention here that there are also corresponding results in the theory of affine flag varieties for Kac-Moody Lie algebras. To explain this, assume that G is quasi-split, absolutely simple and simply connected and splits over a tamely ramified extension. The local Dynkin diagram of G (as in the table of §1) is also the Dynkin diagram of a uniquely determined affine (or twisted affine) Kac-Moody Lie algebra g = gKM(G) (see [Kac]). In the Kac-Moody setting, there is an affine flag variety Fg and Schubert varieties Sg w (see [Ku2], [Ma]; here w is an element of the affine Weyl group). Their definition is given by using an embedding into the infinite dimensional projective space associated to a highest weight representation of the Kac-Moody algebra g; it is a priori different from our approach. The normality of Schubert varieties Sg w in the Kac-Moody setting is a well-known cornerstone of the theory; it was shown by Kumar [Ku1] in characteristic 0 and by Mathieu [Ma] and Littelmann [Li] in all characteristics. It is not hard to show that when g = gKM(G), the Kac-Moody Schubert varieties Sg w are stratawise isomorphic to the Schubert varieties Sw in FB = LG/L+B that we consider here. As a result, we can see a posteriori, as a consequence of Theorem 3.2.2 and the results of Mathieu and Littelmann, that the Schubert varieties Sg w and Sw are isomorphic. This also implies that the affine flag variety FB is isomorphic to the affine flag variety Fg for the corresponding Kac-Moody Lie algebra. See [PR3, 9.h] for more details. 3.3. Embedding the special fiber of local models We now come to the key application of affine flag varieties to the theory of local models, namely the embedding of the special fiber of the local model into an 42 Local models of Shimura varieties, I. appropriate affine flag variety. Since we do not know how to define the local model in general (cf. §1), we can only describe the embedding in particular examples. Here we do so for GLn and GSp2g. Note that for the Beilinson -- Gaitsgory local model in §1.3 such an embedding is tautological. We resume the notation of §2. In particular, we take k to be the residue field of F and we recall the OF -lattices Λi from (2.0.1). In analogy with our notation for the λi's, for nonempty I ⊂ Z closed under addition by n, we denote by ΛI the periodic lattice chain in F n consisting of the lattices Λi for i ∈ I. For any OF -scheme X, we write X for its special fiber X ⊗OF k. n, let µ denote the cocharacter (cid:0)1(r), 0(n−r)(cid:1) of the standard diagonal maximal Example 3.3.1 (GLn). Let I ⊂ Z be nonempty and closed under addition by torus in GLn and {µ} its geometric conjugacy class, and recall the local model M loc in the affine flag variety FI for GLn (see Example 3.1.4) as follows. GLn,{µ},ΛI Let R be a k-algebra and (FΛi)i∈I an R-point of M loc over SpecOF from §2.1. We embed the special fiber M loc GLn,{µ},ΛI . For i ∈ I, we GLn,{µ},ΛI identify Λi ⊗OF k (cid:39) λi ⊗OK k by identifying the standard ordered bases on the two sides. In this way we get an isomorphism of lattice chains ΛI ⊗OF k (cid:39) λI ⊗OK k. Via this isomorphism, we regard FΛi ⊂ Λi ⊗OF R as a submodule of λi ⊗OK R, and we define Li to be the inverse image of FΛi under the reduction-mod-t-map λi ⊗OK R[[t]] (cid:16) λi ⊗OK R. Then Li is an R[[t]]-lattice in R((t))n. Denoting the elements of I by ··· < i−1 < i0 < i1 < ··· , we get a diagram of lattices in R((t))n ··· ⊂ λi−1 ⊗OK R[[t]] ⊂ ⊂ ··· ⊂ λi0 ⊗OK R[[t]] ⊂ ⊂ λi1 ⊗OK R[[t]] ⊂ ⊂ ··· ··· . (3.3.2) ⊂ Li−1 ⊂ ⊂ Li0 ⊂ ⊂ Li1 ⊂ ··· ⊂ tλi−1 ⊗OK R[[t]] ⊂ tλi0 ⊗OK R[[t]] ⊂ tλi1 ⊗OK R[[t]] ⊂ ··· It is easy to verify that the collection (Li)i∈I specifies a point in FI (R), and we define the morphism ι : M loc GLn,{µ},ΛI −→ FI by the rule (FΛi)i (cid:55)→ (Li)i. Plainly ι is a monomorphism, and it is therefore a closed immersion of ind-schemes since M is proper. loc GLn,{µ},ΛI G. Pappas, M. Rapoport, and B. Smithling 43 Moreover, it is easy to see that (Li)i is r-special as defined in Example 3.1.4. Hence we get an embedding M loc GLn,{µ},ΛI G G ι FI C , #G F(cid:48) I−r where F(cid:48) the unlabeled solid arrow is the embedding discussed in Example 3.1.4. I−r is the affine flag variety for SLn corresponding to the set I − r and loc GLn,{µ},ΛI into FI and into F(cid:48) The embeddings of M I−r enjoy an important equivariance property which we now describe. Let A denote the OF -group scheme of automorphisms of the lattice chain ΛI , defined in the obviously analogous way to the OK-group scheme PI in Example 3.1.4. Then A acts naturally on M loc . GLn,{µ},ΛI Now consider the positive loop group L+PI over Spec k. The tautological action of PI on λI furnishes a natural action of L+PI on the chain λI ⊗OK k. The isomorphism λI ⊗OK k (cid:39) ΛI ⊗OF k then yields a homomorphism L+PI → A. Hence L+PI acts on M It is now easy to see that the embedding (cid:44)→ FI is L+PI -equivariant with respect to the natural L+PI -action M on FI . As a consequence, we see that M decomposes into a union of PI -Schubert cells inside FI . loc GLn,{µ},ΛI loc GLn,{µ},ΛI loc GLn,{µ},ΛI . Entirely similar remarks apply to the embedding M loc GLn,{µ},ΛI I acts naturally on M positive loop group L+P (cid:48) embedding into F(cid:48) decomposes into a union of P (cid:48) I−r is then L+P (cid:48) Example 3.3.3 (GSp2g). Now let n = 2g, µ =(cid:0)1(g), 0(g)(cid:1), and {µ} its geometric I -Schubert cells inside F(cid:48) I -equivariant, and we conclude that M I−r. (cid:44)→ F(cid:48) loc GLn,{µ},ΛI I−r: the in an analogous way, the loc GLn,{µ},ΛI conjugacy class in GSp2g, and suppose that I is, in addition, closed under multi- plication by −1. Then we may consider the local model M loc for GSp2g as in §2.2. The embedding of the special fiber M loc GSp2g,{µ},ΛI into the affine flag variety FI for GSp2g is completely analogous to the situation just considered for GLn. GSp2g,{µ},ΛI loc GL2g,{µ},ΛI More precisely, under the embedding of M into the affine flag variety for GL2g from the previous example, it is easy to see that the closed subscheme M loc GSp2g,{µ},ΛI ⊂ M loc GL2g,{µ},ΛI is carried into the locus of points satisfying condition (D) in Example 3.1.6 (where for any R-valued point, the scalar c in (D) can be taken to equal t−1 globally). into the affine flag variety Hence we get the desired embedding of M for GSp2g. loc GSp2g,{µ},ΛI / / q # 1 C 44 Local models of Shimura varieties, I. M loc GSp2g,{µ},ΛI since the image of M loc GL2g,{µ},ΛI Example 3.1.6 again acts naturally on M Continuing the analogy, the parahoric group scheme for GSp2g denoted PI in , and we again conclude that decomposes into a union of PI -Schubert cells inside FI . Moreover, consists of g-special points in FI , there is an (cid:44)→ F(cid:48) I−g denotes the affine flag decomposes I−g, where F(cid:48) induced embedding M variety for Sp2g corresponding to the set I − g; and M into a union of P (cid:48) I -Schubert cells inside F(cid:48) I−g. loc GL2g,{µ},ΛI loc GL2g,{µ},ΛI loc GSp2g,{µ},ΛI The embeddings of the special fibers of the local models for other groups discussed in §2 can all be described similarly. See [Sm1, §7.1] for GO2g (at least in the Iwahori case), [PR2, §4] for totally ramified ResF/F0 GLn, [PR2, §11] for totally ramified ResF/F0 GSp2g, and [PR4, §3.3] and [Sm3, §4.4] for ramified GUn. In all cases, the image of the special fiber of the local model decomposes into a union of Schubert cells inside the affine flag variety. In fact, since the local model is proper, the image is a union of Schubert varieties. It then becomes an interesting problem to determine which Schubert varieties occur in the union. This is a problem of an essentially combinatorial nature to which we turn in §4. 4. Combinatorics In all known examples -- and as we saw explicitly for GLn and GSp2g in §3.3 -- the special fiber of the local model admits an embedding into an affine flag variety, with regard to which it decomposes into a union of Schubert varieties. It is then a basic problem to determine which Schubert varieties occur in the union. Arising from this are a number of considerations of an essentially combinatorial nature to which we turn in this section. Much of our discussion is borrowed from [R, §§2 -- 3] and [PR4, §§2.1 -- 2.2]. We shall work over a complete, discretely valued field L, which we suppose in addition is strictly Henselian. For applications to local models, we are especially interested in the setting L = k((t)), where k is an algebraic closure of the residue field k as denoted in §2; this setting implicitly corresponds to working with the geometric special fiber of the local model. We write OL for the ring of integers in L. Given a connected reductive group G over L, we denote by κG its Kottwitz homomorphism, as encountered earlier in (3.1.8); recall that this is a functorial surjective map G(L) → π1(G)I , where I := Gal(Lsep/L). To be clear about signs, we take κG to be exactly the map defined by Kottwitz in [Ko2, §7] (which makes sense over any complete, discretely valued, strictly Henselian field), without the intervention of signs. This is opposite to the sign convention taken in Richarz's article [Ri2], to which we shall refer in several places. The only practical effect of this difference is that we shall be led to make use of dominant coweights wherever Richarz makes use of antidominant coweights. G. Pappas, M. Rapoport, and B. Smithling 45 4.1. Iwahori-Weyl group Let G be a connected reductive group over L, let S be a maximal split torus in G, let Gad denote the adjoint group of G, and let Sad denote the image of S in Gad. Then Sad is a maximal split torus in Gad, and we let A := X∗(Sad) ⊗Z R denote the apartment in the building of Gad attached to Sad. Let T be the centralizer of S in G. Then T is a maximal torus, since by Steinberg's theorem G is quasi-split. Let N be the normalizer of T in G, and let T (L)1 denote the kernel of the Kottwitz homomorphism κT : T (L) (cid:16) π1(T )I = X∗(T )I for T . Definition 4.1.1. The Iwahori-Weyl group of G associated to S is the group (cid:102)WG,S :=(cid:102)WG :=(cid:102)W := N (L)/T (L)1. Observe that the evident exact sequence 0 −→ T (L)/T (L)1 −→(cid:102)W −→ N (L)/T (L) −→ 1 exhibits (cid:102)W as an extension of the relative Weyl group W0 := N (L)/T (L) by (via the Kottwitz homomorphism) X∗(T )I ∼= T (L)/T (L)1. In fact this sequence splits, by splittings which depend on choices. More pre- cisely, for any parahoric subgroup K ⊂ G(L) attached to a facet contained in the apartment for S, let11 W K :=(cid:0)N (L) ∩ K(cid:1)(cid:14)T (L)1. (4.1.2) decomposition Proposition 4.1.3 ([HR, Prop. 13]). Let K be the maximal parahoric subgroup of G(L) attached to a special vertex in A. Then the subgroup W K of (cid:102)W projects isomorphically to the factor group W0, so that (cid:102)W admits a semidirect product (cid:102)W = X∗(T )I (cid:111) W K ∼= X∗(T )I (cid:111) W0. (cid:3) We typically write tµ when we wish to regard an element µ ∈ X∗(T )I as an element in (cid:102)W , and we refer to X∗(T )I as the translation subgroup of (cid:102)W . affine Weyl group (cid:102)W (R) of R to be the semidirect product X∗ (cid:111) W (R), where In the case that G is split, (cid:102)W canonically identifies with the extended affine Weyl group(cid:102)W (R) of the root datum R :=(cid:0)X∗(S), X∗(S), Φ, Φ∨(cid:1) of G. Indeed, in Remark 4.1.4. Let R = (X∗, X∗, Φ, Φ∨) be a root datum. We define the extended W (R) denotes the (finite) Weyl group of R. 11Here we follow the convention of [HR, PR3, PR4] by using a superscript K in (4.1.2). Some authors would instead denote the group (4.1.2) by WK , and then use W K to denote the set of elements w in the affine Weyl group such that w has minimal length in the coset wWK . 46 Local models of Shimura varieties, I. this case S = T , the action of I on X∗(T ) is trivial, and T (L)1 = S(OL). Taking G(OL) as the special maximal parahoric subgroup in Proposition 4.1.3, we have (cid:102)W = X∗(S) (cid:111) W0, where W0 = W is the absolute Weyl group, which identifies with W (R). Then (cid:102)W contains the affine Weyl group Wa(R) := Q∨ (cid:111) W0 as a Even if G is nonsplit,(cid:102)W can be identified with a generalized extended affine normal subgroup with abelian factor group π1(G) = X∗(S)/Q∨. Here Q∨ ⊂ X∗(S) denotes the subgroup generated by the coroots Φ∨. Weyl group of a reduced root system, as is explained in Remark 4.2.5 below. Example 4.1.5 (GLn). Let G = GLn in Remark 4.1.4 and take for S = T the standard split maximal torus of diagonal matrices in G. Then where Zn ∼−→ X∗(S) by sending the ith standard basis element to the cocharacter x (cid:55)→ diag(cid:0)1(i−1), x, 1(n−i)(cid:1), and where the symmetric group Sn of permutation matrices maps isomorphically to the Weyl group. (cid:102)W ∼= Zn (cid:111) Sn, Example 4.1.6 (GSp2g). Let G = GSp2g in Remark 4.1.4 and take S = T to be the standard split maximal torus of diagonal matrices in G. Let where i∗ := 2g+1−i for any i ∈ {1, . . . , 2g}. Then S∗ 2g identifies with the subgroup of permutation matrices in G and maps isomorphically to the Weyl group. We obtain 2g :=(cid:8) σ ∈ S2g S∗ (cid:12)(cid:12) σ(i∗) = σ(i)∗ for all i(cid:9), (cid:102)W ∼= X∗ (cid:111) S∗ 2g, where, in terms of the natural embedding of S into the maximal torus for GL2g, and in terms of the identification of the previous example, we have X∗ :=(cid:8) (x1, . . . , x2g) ∈ Z2g(cid:12)(cid:12) x1 +x2g = x2 +x2g−1 = ··· = xg +xg+1 (cid:9) ∼−→ X∗(S). Remark 4.1.7 (GO2g). Although the orthogonal similitude group GO2g is not connected, we can give an ad hoc definition of its Iwahori-Weyl group by following the recipe in Definition 4.1.1 in the most literal way, where we take the normalizer in the full group. We find just as in the previous example. The Iwahori-Weyl group (cid:102)WGO◦ 2g is naturally a subgroup of (cid:102)WGO2g of index 2. Explicitly, component GO◦ ∼= X∗ (cid:111) S∗ 2g, of the identity 2g where field case(cid:102)WGO2g continues to parametrize the Schubert cells in the Iwahori affine It turns out that, just as we shall see for connected groups, in the function 2g flag variety for GO2g. See [Sm1]. (cid:102)WGO2g (cid:102)WGO◦ 2g :=(cid:8)σ ∈ S∗ 2g S◦ ∼= X∗ (cid:111) S◦ 2g, (cid:12)(cid:12) σ is even in S2g (cid:9). G. Pappas, M. Rapoport, and B. Smithling 47 The following group-theoretic result provides the key link between the Iwahori- Weyl group and local models. Proposition 4.1.8 ([HR, Prop. 8]). Let B be an Iwahori subgroup of G(L) at- tached to an alcove in A. Then the inclusion N (L) ⊂ G(L) induces a bijection (cid:102)W ∼−→ B\G(L)/B. More generally, let K and K(cid:48) be parahoric subgroups of G(L) attached to facets in A. Then the inclusion N (L) ⊂ G(L) induces a bijection W K(cid:48)\(cid:102)W /W K ∼−→ K(cid:48)\G(L)/K. (cid:3) Remark 4.1.9. Assume in Proposition 4.1.8 that K is a parahoric subgroup attached to a special vertex. Then (cid:102)W ∼= X∗(T )I (cid:111) W K and, since W K ∼= W0, W K\(cid:102)W /W K ∼= X∗(T )I /W0. This last set may in turn be identified with the set of dominant elements in X∗(T )I (any element in X∗(T )I is conjugate under W0 to a unique dominant element, cf. [Ri2, Remark before Cor. 1.8]; recall that we use dominant elements where Richarz uses antidominant ones). The notion of dominant elements in X∗(T )I arises after identifying (cid:102)W with a generalized extended affine Weyl group of a certain root system Σ in the sense of Remark 4.2.5 below, and then choosing a basis for Σ; comp. Remark 4.2.7. If we assume in addition that G is split, then K is hyperspecial, X∗(T )I = X∗(S), the notion of a dominant coweight in X∗(S) is more standard. Remark 4.1.10. Let K be a parahoric subgroup of G(L) and P the corresponding parahoric group scheme over SpecOL. Then W K can be identified with the Weyl group of the special fiber P of P [HR, Prop. 12]. 4.2. Bruhat order The force of Proposition 4.1.8 in the context of local models is that, when L is a field of Laurent series and K and K(cid:48) are as in the proposition, the set of double classes W K(cid:48)\(cid:102)W /W K (which is (cid:102)W itself in the Iwahori case) parametrizes in this subsection we shall introduce the Bruhat order on(cid:102)W . This will lead us to the K(cid:48)-Schubert cells in the affine flag variety for K. To better exploit this fact, some much heavier usage of Bruhat-Tits theory than we have yet encountered, but non-experts should be able to safely treat many of the external results we appeal to as black boxes, with little impairment to overall understanding. We continue with the notation of the previous subsection. 48 Local models of Shimura varieties, I. Consider the apartment A = X∗(Sad) ⊗Z R ∼= X∗(Tad)I ⊗Z R. We obtain from the Kottwitz homomorphism a map12 T (L)/T (L)1 κT−−→∼ X∗(T )I −→ X∗(Tad)I −→ A. Let Aff(A) denote the group of affine transformations on A. The relative Weyl group W0 acts naturally on A by linear transformations, and regarding A (cid:111) W0 as a subgroup of Aff(A), the displayed map extends to a map of exact sequences (cid:102)W (4.2.1) 1 1 / T (L)/T (L)1 / A ν / A (cid:111) W0 W0 / 1 / W0 / 1, in which ν is unique up to conjugation by a unique translation element in Aff(A); see [T, §1.2] or [Lan, Prop. 1.6 and 1.8]. Having chosen ν, the set of affine roots Φa, which consists of certain affine functions on A, is defined in Tits's article [T, §1.6]. There then exists a unique reduced root system Σ on A with the properties that • every root α ∈ Σ is proportional to the linear part of some affine root; and • for any special vertex v ∈ A [T, §1.9], translation by −v A t−v−−→ A carries the vanishing hyperplanes of the affine roots to precisely the vanishing hyperplanes of the functions on A u (cid:55)−→ α(u) + d for α ∈ Σ, d ∈ Z; (4.2.2) see [BTI, 1.3.8], [T, §1.7], and [B, VI §2.5 Prop. 8]. For G absolutely simple and simply connected, see Remark 4.2.5 below for a description of Φa and Σ. The affine Weyl group Wa is the group of affine transformations on A gener- ated by the reflections through the affine root hyperplanes. Analogously, the affine Weyl group Wa(Σ) of Σ [B, VI §2.1 Def. 1] is the group of affine transformations on A generated by the reflections through the vanishing hyperplanes of the functions (4.2.2). Thus for any special vertex v, we have Wa = tvWa(Σ)t−1 v . Hence Wa is a Coxeter group generated (as a Coxeter group) by the reflections through the walls of any fixed alcove. The affine Weyl group for Σ admits the 12Note that the displayed composite differs from the analogous map defined in Tits's article [T, §1.2] by a sign of −1. However this discrepancy will make no difference in any of our subsequent appeals to [T]. It matters only in that, as we have mentioned before, we systematically work with dominant elements where Richarz [Ri2] uses antidominant elements; see especially Proposition 4.2.11. / / / / / / / / / / G. Pappas, M. Rapoport, and B. Smithling 49 semidirect product decomposition Wa(Σ) = Q∨(Σ) (cid:111) W (Σ), where Q∨(Σ) is the coroot lattice for Σ and W (Σ) is the Weyl group of Σ. To apply the preceding discussion to the Iwahori-Weyl group, consider the diagram (cid:102)W We shall show that Wa lifts canonically to (cid:102)W . Indeed, let Wa ⊂ Aff(A). ν G(L)1 := ker κG and (cid:0)G(L)1, B, N (L)1, Π(cid:1) N (L)1 := N (L) ∩ G(L)1. (4.2.3) Let B ⊂ G(L) be the Iwahori subgroup attached to an alcove in A, and let Π be the set of reflections through the walls of this alcove. Then, taking into account that N (L) ∩ B = T (L)1 [HR, Lem. 6] and that G(L)1 is the subgroup of G(L) generated by the parahoric subgroups of G(L) [HR, Lem. 17], the quadruple (4.2.4) via ν with Wa by [BTII, 5.2.12]. is a double Tits system [BTI, 5.1.1] whose Weyl group N (L)1/T (L)1 ⊂(cid:102)W identifies The affine Weyl group can also be realized as a subgroup of(cid:102)W via the simply and normalizer Nsc. Let(cid:102)Wsc := Nsc(L)/Tsc(L)1 denote the Iwahori-Weyl group of Prop. 5.2.10],(cid:0)Gsc(L), Bsc, Nsc(L), Π(cid:1) is a double Tits system whose Weyl group (cid:102)Wsc (we again use [HR, Lem. 6]) identifies with Wa via the composite connected cover Gsc of the derived group Gder of G. To explain this, let Ssc, Tsc, and Nsc denote the respective inverse images of S ∩ Gder, T ∩ Gder, and N ∩ Gder in Gsc. Then Ssc is a maximal split torus in Gsc with centralizer Tsc Gsc, let Bsc ⊂ Gsc(L) be the Iwahori subgroup attached to an alcove in A, and let Π again be the set of reflections through the walls of this alcove. Then by [BTII, (cid:102)Wsc −→(cid:102)W ν−→ A (cid:111) W0. ∼= X∗(Tsc)I identifies with the translation elements In this way Tsc(L)/Tsc(L)1 in Wa, or in other words, with the coroot lattice in Wa(Σ). Moreover, for any parahoric subgroup Ksc ⊂ Gsc(L) attached to a special vertex v, the composite (cid:102)Wsc −→(cid:102)W ν−→ A (cid:111) W0 conjugation by t−v −−−−−−−→ A (cid:111) W0 50 carries W Ksc ⊂(cid:102)Wsc isomorphically to the Weyl group of Σ, which identifies with Local models of Shimura varieties, I. W0. In other words, the composite isomorphism conjugation (cid:102)Wsc ν(cid:102)Wsc−−−→∼ Wa by t−v −−−−−−−→∼ Wa(Σ) and Wa(Σ) = Q∨(Σ) (cid:111) W (Σ). is compatible with the semidirect product decompositions(cid:102)Wsc ∼= X∗(Tsc)I (cid:111) W Ksc Remark 4.2.5. Just as (cid:102)Wsc can in this way be identified with the affine Weyl group for the root system Σ, so can (cid:102)W be identified with a generalized extended affine Weyl group for Σ via push-out by the canonical injection X∗(Tsc)I (cid:44)→ X∗(T )I . Here "generalized" means that the abelian group X∗(T )I may have tor- sion. For any absolutely simple, simply connected group G over a discretely valued field with algebraically closed residue field, the affine root system Φa and root system Σ admit the following descriptions (up to a choice of normalization of the valuation, and of a special vertex as origin), which are given by Prasad -- Raghunathan [PrRa, §2.8]; we thank J.-K. Yu and X. Zhu for pointing this out to us. Let Φ denote the relative roots of S in G. If G is split over L, then Φ is necessarily reduced, Φa = { a + Z a ∈ Φ}, If G is nonsplit over L and Φ is reduced, then and Σ = Φ. Φa = a + (a, a) (cid:26) (cid:26) (cid:27) 2 Z (cid:12)(cid:12)(cid:12)(cid:12) a ∈ Φ (cid:12)(cid:12)(cid:12)(cid:12) a ∈ Φ, a 2 and Σ = Φ∨ ∼= (cid:26) (cid:27) (a, a) (cid:26) 2a (cid:12)(cid:12)(cid:12)(cid:12) a ∈ Φ, , (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) a ∈ Φ (cid:27) ∈ Φ , a 2 where ( , ) is a nondegenerate W0-invariant inner product. If G is nonsplit and Φ is nonreduced, then G is an outer form of type A2n, Φa = a + 1 + 2Z and Σ is the subset of Φ of roots a for which 2a /∈ Φ. a + Z /∈ Φ ∪ 1 −→ Wa −→(cid:102)W −→ X∗(T )I /X∗(Tsc)I −→ 1. We return to the main discussion. Since the target of the Kottwitz homo- morphism is abelian, it is immediate from the definition (4.2.3) of N1(L) that it is a normal subgroup of N (L). Hence Wa (cid:39) N (L)1/T (L)1 is a normal subgroup of (cid:102)W , and we get an exact sequence transitively on the alcoves in A [T, 1.7], (cid:102)W is the semidirect product of Wa with the normalizer Ω ⊂(cid:102)W of the base alcove, ∼= X∗(T )I /X∗(Tsc)I This sequence splits canonically after choosing a base alcove: since Wa acts simply (cid:102)W = Wa (cid:111) Ω, ∼= π1(G)I . ∼−→(cid:102)W /Wa with Ω (4.2.6) G. Pappas, M. Rapoport, and B. Smithling The semidirect product decomposition (4.2.6) for(cid:102)W has the important con- sequence of endowing (cid:102)W with a Bruhat order and length function. Again let Π 51 denote the subset of Wa of reflections across the walls of the base alcove. As we have already recalled, Π is a set of Coxeter generators for Wa. We then get a Bruhat order ≤ and a length function (cid:96) on Wa as for any Coxeter group: for s, s(cid:48) ∈ Wa, (cid:96)(s) is the smallest nonnegative integer r such that s is expressible as a product s1s2 ··· sr with s1, s2, . . . , sr ∈ Π; and s(cid:48) ≤ s if there exists an expression s = s1 ··· sr with (cid:96)(s) = r and the si's in Π such that s(cid:48) can be obtained by deleting some of the si's from the product. These definitions naturally extend to exactly when ω(cid:48) = ω and s(cid:48) ≤ s in Wa. (cid:102)W via (4.2.6): for s, s(cid:48) ∈ Wa and ω, ω(cid:48) ∈ Ω, we have (cid:96)(sω) := (cid:96)(s), and s(cid:48)ω(cid:48) ≤ sω and f(cid:48) of the base alcove, the Bruhat order on (cid:102)W induces one on W K(cid:48)\(cid:102)W /W K. For parahoric subgroups K, K(cid:48) ⊂ G(L) attached to respective subfacets f Indeed, let X and X(cid:48) denote the respective subsets of Π of reflections fixing f and f(cid:48), let WX and WX(cid:48) denote the respective subgroups of Wa generated by X and X(cid:48), and recall from [BTII, 5.2.12] that the parahoric subgroups of G(L) are precisely the parahoric subgroups of the Tits system (4.2.4). Then K = BWX B and K(cid:48) = BWX(cid:48)B, and = WX(cid:48). W K = WX and W K(cid:48) Hence by [B, IV §1 Ex. 3] each double coset w ∈ W K(cid:48)\(cid:102)W /W K contains a unique element in(cid:102)W , which we denote (cid:101)w, with the property that (cid:101)w ≤ x for all x ∈ w. For w, w(cid:48) ∈ W K(cid:48)\(cid:102)W /W K, we then define w(cid:48) ≤ w if (cid:101)w(cid:48) ≤ (cid:101)w in(cid:102)W . The Bruhat order on W K(cid:48)\(cid:102)W /W K has the property that if x ≤ y in (cid:102)W , then W K(cid:48) in W K(cid:48)\(cid:102)W /W K. Remark 4.2.7. The Bruhat order on W K(cid:48)\(cid:102)W /W K can be expressed in a par- tify (cid:102)Wsc with the affine Weyl group of the reduced root system Σ, and (cid:102)W with ticularly simple way when K = K(cid:48) is a maximal parahoric subgroup attached to a special vertex v of the base alcove. Indeed, the choice of v allows us to iden- xW K ≤ W K(cid:48) yW K a generalized extended affine Weyl group of Σ, as in Remark 4.2.5 above. The affine root hyperplanes passing through v determine chambers in A, and we take as positive chamber the chamber opposite the unique chamber containing our base alcove. This determines a notion of positive roots in X∗(Tsc)I , of positive coroots in X∗(Tsc)I , and of dominant elements in X∗(T )I (those elements that pair non- negatively with positive roots). The dominance order on X∗(T )I is defined by λ ≤ λ(cid:48) if λ(cid:48) − λ is a nonnegative Z-linear combination of positive coroots. Then, after identifying W K\(cid:102)W /W K with the set of dominant elements in X∗(T )I as in Remark 4.1.9, the Bruhat order on W K\(cid:102)W /W K is identified with the restric- tion of the dominance order to the set of dominant elements, cf. [Ri2, Cor. 1.8]. 52 Local models of Shimura varieties, I. (In contrast to [Ri2], we use dominant elements instead of antidominant elements because we have taken the base alcove to be in the negative chamber.) In the function field setting, and in analogy with the case of ordinary flag varieties, the Bruhat order carries important geometric content about Schubert varieties. Proposition 4.2.8 (Richarz [Ri2, Prop. 2.8]; [PR3, Prop. 9.6]). Suppose that L = k((t)) with k algebraically closed. Let f and f(cid:48) be subfacets of the base alcove in A, let K and K(cid:48) denote the respective associated parahoric subgroups of G(L), and consider the associated affine flag variety Ff over Spec k, cf. Definition 3.1.2. For w ∈ W K(cid:48)\(cid:102)W /W K, let Sw denote the associated f(cid:48)-Schubert variety in Ff . Then for all w, w(cid:48) ∈ W K(cid:48)\(cid:102)W /W K, By choosing good representatives in (cid:102)W for double cosets, the inclusion rela- Sw(cid:48) ⊂ Sw in Ff ⇐⇒ w(cid:48) ≤ w in W K(cid:48)\(cid:102)W /W K. (cid:3) tions between Schubert varieties can be phrased in a somewhat more precise way, which is sometimes useful. We first state the following lemma. Lemma 4.2.9 (Richarz, Waldspurger [Ri2, Lem. 1.9]). Let f and f(cid:48) be subfacets of the base alcove in A, and let K and K(cid:48) denote the respective associated parahoric subgroups of G(L). Let w ∈(cid:102)W . (i) There exists a unique element wK of minimal length in wW K. (ii) There exists a unique element K(cid:48)wK of maximal length in {(vw)K v ∈ W K(cid:48)}. (cid:3) We introduce the following subset of (cid:102)W , K(cid:48)(cid:102)W K :=(cid:8) K(cid:48)wK (cid:12)(cid:12) w ∈(cid:102)W(cid:9). Then K(cid:48)(cid:102)W K maps bijectively to the set of double classes W K(cid:48)\(cid:102)W /W K, and we Proposition 4.2.10 (Richarz [Ri2, Prop. 2.8]). Let w ∈ K(cid:48)(cid:102)W K. Then the f(cid:48)- may phrase the inclusion relations between Schubert varieties in terms of these special representatives of double classes as follows. Sw =(cid:83) Schubert variety Sw in Ff satisfies (i) {w(cid:48)∈K(cid:48)(cid:102)W Kw(cid:48)≤w} Sw(cid:48); and (ii) dim Sw = (cid:96)(w). Now let us specialize to the case of special maximal parahorics. Proposition 4.2.11 (Richarz [Ri2, Cor. 1.8]). Let K = K(cid:48) be a special maxi- mal parahoric subgroup attached to a vertex of the base alcove, and consider the dominant elements of X∗(T )I as in Remark 4.2.7. (i) K(cid:102)W K = { tλ λ ∈ X∗(T )I is dominant}. (cid:3) G. Pappas, M. Rapoport, and B. Smithling dim Sw = (cid:104)λ, 2ρ(cid:105) for w corresponding to tλ ∈ K(cid:102)W K. (ii) Here ρ denotes the halfsum of positive roots for Σ. 4.3. The {µ}-admissible set 53 (cid:3) In this subsection we come to the key notion of {µ}-admissibility. We con- tinue with the notation of the previous two subsections. Let {µ} ⊂ X∗(T ) be a set of cocharacters whose images in X∗(T ) ⊗Z R are contained in some (absolute) Weyl chamber corresponding to a Borel subgroup of G containing T and defined W -conjugacy class of geometric cocharacters of T . Let (cid:101)Λ{µ} ⊂ {µ} be the sub- over L.13 Then (cid:101)Λ{µ} forms a single W0-conjugacy class, since all such Borels are W0-conjugate. Let Λ{µ} denote the image of (cid:101)Λ{µ} in X∗(T )I . Let a be an alcove in the apartment A, and consider the associated Bruhat order ≤ on (cid:102)W . We first state a conjecture. Conjecture 4.3.1. Let {µ} denote the image of the W -conjugacy class {µ} in X∗(T )I . Then the set of maximal elements in {µ} with respect to the Bruhat order is precisely the set Λ{µ}. Of course the conjecture only has content for nonsplit G. We have verified it for Weil restrictions of split groups and for unitary groups. The validity of the conjecture not being known to us in general, we define the {µ}-admissible set as follows. Definition 4.3.2. An element w ∈ (cid:102)W is {µ}-admissible if w ≤ tλ for some λ ∈ Λ{µ}. We denote the set of µ-admissible elements in (cid:102)W by Adm({µ}). In other words, w ∈ (cid:102)W is {µ}-admissible if and only if w ≤ σtµσ−1 = tσ·µ for some σ ∈ W0, where µ is the image in X∗(T )I of a cocharacter µ ∈(cid:101)Λ{µ}. Since W0 can be lifted to the affine Weyl group inside(cid:102)W , all elements in Adm({µ}) are subfacets of a, and consider the set of double cosets W K(cid:48)\(cid:102)W /W K. Definition 4.3.3. An element w ∈ W K(cid:48)\(cid:102)W /W K is {µ}-admissible if We denote the set of {µ}-admissible elements in W K(cid:48)\(cid:102)W /W K by AdmK(cid:48),K({µ}), More generally, let K and K(cid:48) be parahoric subgroups of G(L) attached to w ≤ W K(cid:48) tλW K for some λ ∈ Λ{µ}. congruent modulo Wa. or just by AdmK({µ}) when K = K(cid:48). Note that if Conjecture 4.3.1 holds true, then the notion of {µ}-admissibility is just that w ≤ W K(cid:48) tµW K for some µ in the image of {µ} in X∗(T )I . 13Note that such Borel subgroups always exists since G is quasi-split. 54 Local models of Shimura varieties, I. Example 4.3.4. Suppose that K is a special maximal parahoric subgroup. Then the Bruhat order on W K\(cid:102)W /W K identifies with the dominance order on the set of set in W K\(cid:102)W /W K identifies with the dominant elements in X∗(T )I that are dominant elements in X∗(T )I , as in Remark 4.2.7. In this way the {µ}-admissible ≤ µdom in the dominance order, where µdom denotes the unique dominant element in Λ{µ}. It is also worth making explicit the notion of {µ}-admissibility in the setting of root data, which amounts to working in the special case that G is split, cf. Remark 4.1.4. Let R = (X∗, X∗, Φ, Φ∨) be a root datum and {µ} ⊂ X∗ a W (R)- conjugacy class of cocharacters. Again choose a base alcove and consider the induced Bruhat order on (cid:102)W (R). Then we define the {µ}-admissible set Adm({µ}) :=(cid:8) w ∈(cid:102)W (R)(cid:12)(cid:12) w ≤ tµ for some µ ∈ {µ}(cid:9). More generally, let f and f(cid:48) be subfacets of the base alcove, and let X (resp. X(cid:48)) be the set of reflections across the walls of the base alcove containing f (resp. f(cid:48)). As on p. 51, let WX (resp. WX(cid:48)) be the subgroup of Wa(R) generated by X (resp. X(cid:48)). Then we define Admf ,f(cid:48) ({µ}) :=(cid:8) w ∈ WX(cid:48)\(cid:102)W (R)/WX (cid:12)(cid:12) w ≤ WX(cid:48)tµWX for some µ ∈ {µ}(cid:9). When f = f(cid:48), we write Admf ({µ}) := Admf ,f ({µ}). Remark 4.3.5. Let R be a root datum, choose a positive chamber, and take the base alcove to be the unique alcove contained in the negative chamber and whose closure contains the origin. Let µ be a dominant cocharacter. He and Lam [HL, Th. 2.2] have recently given a description of the partially ordered set Adm(cid:0)W (R)·µ(cid:1)∩W (R)tµW (R) in terms of the combinatorics of projected Richard- ample 4.3.4 that Adm(cid:0)W (R) · µ(cid:1) ⊂ W (R)tµW (R). Thus He and Lam's result son varieties. Note that in the special case when µ is minuscule, µ is minimal among dominant cocharacters in the dominance order, and it follows from Ex- describes the full admissible set in the minuscule case. 4.4. Relation to local models We continue with the notation of the previous three subsections. Let us now return to the problem we posed at the beginning of §4, namely that of identifying the Schubert cells that occur in the geometric special fiber of a local model upon a suitable embedding into an affine flag variety. More precisely, let F be a discretely valued field with residue field k, let (cid:101)G be a connected reductive group over F , let {µ} be a conjugacy class of geometric cocharacters of (cid:101)G, let E denote the reflex have attached a (flat) local model M loc(cid:101)G,{µ} to (cid:101)G and {µ} over SpecOE; in each field of {µ}, and let OE and kE denote the respective ring of integers in, and residue field of, E. Suppose that for some choice of parahoric level structure we example we encountered in §2, this was taken to be the scheme-theoretic closure G. Pappas, M. Rapoport, and B. Smithling 55 of the generic fiber of the naive local model. Let F un denote the completion of W K\(cid:102)WG/W K via the above bijection (cid:102)WG in every example we know,14 • there exists a connected reductive group G over L ("a function field analog of a maximal unramified algebraic extension of F , let (cid:101)S be a maximal split torus in (cid:101)GF un := (cid:101)G ⊗F F un, let (cid:101)T be the centralizer of (cid:101)S in (cid:101)GF un, and regard {µ} as an absolute conjugacy class of geometric cocharacters of (cid:101)T . Let L = kE((t)). Then (cid:101)GF un") such that G and (cid:101)G are forms of the same split Chevalley group defined over Z, and whose Iwahori-Weyl group (cid:102)WG naturally identifies with (cid:102)W(cid:101)GF un ; loc(cid:101)G,{µ}⊗kE kE embeds L+P -equivariantly in the affine • the geometric special fiber M flag variety FP for G, where P is a parahoric group scheme for G corresponding to the original choice of parahoric level structure; loc(cid:101)G,{µ} ⊗kE • and with regard to this embedding, the Schubert cells occurring in M ∼= (cid:102)W(cid:101)GF un , where K denotes the para- kE are parametrized by precisely the {µ}-admissible set, regarded as a subset of horic subgroup P (OL) ⊂ G(L). loc(cid:101)G,{µ} ⊗kE kE, which Note that this says that the irreducible components of M loc(cid:101)G,{µ} ⊗kE kE that are maximal for the Example 4.4.1 (GLn). Let (cid:101)G = GLn over F , and let {µ} be the conjugacy class of µ =(cid:0)1(r), 0(n−r)(cid:1), as in §2.1. Then E = F . Let L be the standard lattice chain Iwahori-Weyl group (cid:102)W =(cid:102)WG,S as in Example 4.1.5, and we embed as in Example 3.3.1. Then an element w ∈(cid:102)W = N (L)/S(OL) specifies a Schubert GLn,{µ},ΛZ ⊗k k exactly when w · λZ is contained GLn,{µ},ΛZ ⊗k k, that is, exactly when the lattice chain w · λZ correspond to the Schubert cells in M inclusion relation of their closures, are exactly parametrized by the elements of Λ{µ}. ΛZ in F n. We take G := GLn over L with split maximal diagonal torus S and loc GLn,{µ},ΛZ ⊗k k −→ FZ M cell contained in the image of M loc in the image of M satisfies loc (1) λi ⊃ w · λi ⊃ tλi for all i; and (2) dimk(w · λi/tλi) = n − r for all i. To translate these conditions into more explicit combinatorics, let us identify each OL-lattice of the form ti1OL ⊕ ··· ⊕ tinOL with the vector (i1, . . . , in) ∈ Zn. Then with regard to our identifications, the natural action of N (L)/S(OL) on lattices translates to the natural action of Zn(cid:111)Sn on Zn by affine transformations, with Zn acting by translations and Sn acting by permuting coordinates. For 14This will be addressed more systematically in [PZ]. 56 Local models of Shimura varieties, I. i = nd + j with 0 ≤ j < n, the lattice λi translates to the vector ωi :=(cid:0)(−1)(j), 0(n−j)(cid:1) − d, where for any d we write boldface d to denote the vector (d, . . . , d). Conditions (1) and (2) become equivalent to (1(cid:48)) 0 ≤ w · ωi − ωi ≤ 1 for all i; and (2(cid:48)) for all i, the sum of the entries of the vector w · ωi − ωi is r. Note that µ and all its Weyl conjugates, regarded as translation elements in (cid:102)W , of all w ∈ (cid:102)W satisfying (1(cid:48)) and (2(cid:48)) is precisely the set Adm({µ}), where the trivially satisfy (1(cid:48)) and (2(cid:48)). The main result for GLn in [KR] is that the set Bruhat order is taken with respect to the alcove determined by the ωi's.15 Entirely analogous remarks hold for any subchain ΛI of ΛZ. Let us return to the general discussion, with L again an arbitrary complete, discretely valued, strictly Henselian field. Taking note that, in the previous exam- ple, the images of the ωi's in the standard apartment for P GLn are the vertices of the base alcove, the papers [KR, R] abstract conditions (1(cid:48)) and (2(cid:48)) to any Iwahori-Weyl group as follows. Let Tad denote the image of T in Gad. Consider the composition X∗(T )I −→ X∗(Tad)I −→ X∗(Tad)I ⊗Z R ∼= X∗(Sad) ⊗Z R = A, and let P{µ} denote the convex hull in A of the image of the set Λ{µ}. Definition 4.4.2. An element w ∈(cid:102)W is {µ}-permissible if • w ≡ tµ mod Wa for one, hence any, µ ∈ Λ{µ}; and • w · x − x ∈ P{µ} for all x ∈ a. More generally, for any subfacet f of a with associated parahoric subgroup K, an element w ∈ W K\(cid:102)W /W K is {µ}-permissible if w ≡ tµ mod Wa for any µ ∈ Λ{µ} permissible elements in(cid:102)W and PermK({µ}) for the set of {µ}-permissible elements in W K\(cid:102)W /W K. Note that for w ∈ W K\(cid:102)W /W K, the condition w ≡ tµ mod Wa is well-defined and w · x − x ∈ P{µ} for all x ∈ f . We write Perm({µ}) for the set of {µ}- because W K ⊂ Wa, and the condition (4.4.3) is well-defined by [R, §3, p. 282]. By convexity, (4.4.3) is equivalent to requiring that w · x − x ∈ P{µ} for all vertices x of f . In the case of GLn and {µ} the conjugacy class of µ = (cid:0)1(r), 0(n−r)(cid:1) from Example 4.4.1, one sees almost immediately that the set of elements in(cid:102)W satisfying w · x − x ∈ P{µ} for all x ∈ f 15Note that this alcove is the alcove contained in the negative Weyl chamber (relative to the standard choice of positive roots) and whose closure contains the origin. This is the motivation for our convention in defining the positive chamber in Remark 4.2.7. G. Pappas, M. Rapoport, and B. Smithling 57 (1(cid:48)) and (2(cid:48)) is precisely Perm({µ}). Thus the main result for GLn in [KR] is to establish the equality Adm({µ}) = Perm({µ}) for such µ. In many (but not all!) cases known to us, the Schubert cells in the special fiber of the local model turn out to be parametrized by the {µ}-permissible set, i.e. one has an equality between {µ}-admissible and {µ}-permissible sets. And in explicit computations it is easier to determine the {µ}-permissible set than the {µ}-admissible set. Thus it is of interest to understand the relationship between {µ}-admissibility and {µ}-permissibility. The first results in this direction are the following. Proposition 4.4.4. (i) ([KR, §11]) For any G and any {µ}, Perm({µ}) is closed in the Bruhat order and Adm({µ}) ⊂ Perm({µ}). (ii) (Haines -- Ngo [HN2, 7.2]) The reverse inclusion can fail. More precisely, sup- pose that G is split over L with irreducible root datum of rank ≥ 4 and not of type A. Then Adm({µ}) (cid:54)= Perm({µ}) for {µ} the conjugacy class of any sufficiently (cid:3) regular cocharacter µ. (cid:26) w ∈(cid:102)W (R) Perm({µ}) := In (ii), we refer to the proof of the cited result for the precise meaning of "sufficiently regular." We also note that in [Sm4] it is shown that Adm({µ}) (cid:54)= Perm({µ}) for {µ} the Weyl orbit of the coweight(cid:0)1(r), 0(m−r)(cid:1) for Bm (using the standard coordinates, as in [B, Pl. II]) for m, r ≥ 3. While {µ}-admissibility and {µ}-permissibility are not equivalent in general, the following result gives a summary of most situations in which they are known to coincide. We shall formulate the results for extended affine Weyl groups attached to root data; in the most literal sense one may regard this as an assumption that G is split over L, as in Remark 4.1.4, but see Remark 4.4.7 below for the relevance of this to the nonsplit case. Given a root datum R = (X∗, X∗, Φ, Φ∨) and a W (R)-conjugacy class {µ} ⊂ X∗ of cocharacters, we define Perm({µ}) in obvious analogy with Definition 4.4.2, (cid:12)(cid:12)(cid:12)(cid:12) w ≡ tµ mod Wa(R) for any µ ∈ {µ} and w · x − x ∈ (cid:101)P{µ} for all x in the base alcove (cid:27) , where (cid:101)P{µ} denotes the convex hull of {µ} in X∗ ⊗Z R. Proposition 4.4.5. Let (cid:102)W be the extended affine Weyl group attached to a root datum R, as in Remark 4.1.4, and take the Bruhat order on (cid:102)W corresponding to a base alcove a. (i) (Haines -- Ngo [HN2, 3.3]; [KR, 3.5]) If R involves only type A, then Adm({µ}) = Perm({µ}) for any W (R)-conjugacy class {µ}. (ii) (Haines -- Ngo [HN2, 10.1]; [KR, 4.5, 12.4]) Suppose that(cid:102)W is the Iwahori-Weyl group of GSp2g and that {µ} = W (R) · µ for µ a sum of dominant minuscule cocharacters. Then Adm({µ}) = Perm({µ}). 58 Local models of Shimura varieties, I. (iii) ([KR, 3.5, 4.5], [Sm1, 7.6.1], [Sm2, Main Theorem]) Suppose that R involves only types A, B, C, and D and that {µ} is a W (R)-conjugacy class of minuscule cocharacters. Then Adm({µ}) = Perm({µ}). (cid:3) In (ii), a cocharacter µ is a sum of dominant minuscule cocharacters (with respect to the standard choice of positive Weyl chamber) exactly when it is of the form (cid:0)n(g), 0(g)(cid:1) + d for some n ∈ Z≥0 and d ∈ Z, in the notation of Example 4.1.6. Remark 4.4.6. As stated, the proposition covers only the Iwahori case, but it is known to generalize to the general parahoric case. To be precise, let f be a subfacet of a, let X be the set of reflections across the walls of the base alcove containing f , and let WX be the subgroup of Wa(R) generated by X. Then, in analogy with Definition 4.4.2, we define Permf ({µ}) to be the set of all w ∈ WX\(cid:102)W (R)/WX such that w ≡ tµ mod Wa(R) for any µ ∈ {µ}, and such that (cid:101)w · x − x ∈ (cid:101)P{µ} for all x ∈ f , where (cid:101)w is any representative of w in (cid:102)W (R) (this is again independent of the choice of (cid:101)w by [R, §3, p. 282]). Then in (i), we have Admf ({µ}) = Permf ({µ}) for any {µ} and any f when R involves only type A; this was proved in the case of minuscule {µ} in [KR, 9.6], and the general case is an immediate consequence of Gortz's result [Go4, Cor. 9] (which itself makes crucial use of the cited result in the Iwahori case of Haines -- Ngo). In (ii), we have Admf ({µ}) = Permf ({µ}) inside WX\(cid:102)WGSp2g /WX for any f and any {µ} which is the conjugacy class of a sum of dominant minuscule coweights; this was proved in the case of minuscule {µ} in [KR, 10.7], and the general case is an immediate consequence of Gortz's result [Go4, Cor. 13] (which again relies on the Iwahori case established in [HN2]). It follows that the parahoric version of (iii) holds for any f and any minuscule {µ} provided R involves only types A and C. On the other hand, the general parahoric version of (iii) for types B and D will be deduced in [Sm5] from the Iwahori case for these types, via arguments along the lines of those in [KR] or [Go4]. Remark 4.4.7. Proposition 4.4.5 is useful for more than just the case that G is split. Indeed, for any group G, questions of admissibility and permissibility in (cid:102)W can always be reduced to the case of an extended affine Weyl group attached to a root datum. The link is made via the reduced root system Σ on A attached to the affine root system Φa for G, as discussed in §4.2, p. 48. Consider the group X∗(Tad)I . By [BTII, 4.4.16], X∗(Tad) is an induced Galois module. Hence X∗(Tad)I is torsion-free. And by [HR, Lem. 15], we have Q∨(Σ) ⊂ X∗(Tad)I ⊂ P ∨(Σ), G. Pappas, M. Rapoport, and B. Smithling 59 where Q∨(Σ) and P ∨(Σ) denote the respective coroot and coweight lattices for Σ. Hence is a root datum. For any v ∈ A which is a special vertex relative to the affine root system for G, the image of the composition R :=(cid:0)X∗(Tad)I , X∗(Tad)I , Σ, Σ∨(cid:1) (cid:102)W ν−→ A (cid:111) W0 −−−−−−−→ A (cid:111) W0 conjugation by t−v subgroup attached to v and W K is the subgroup (4.1.2). For the present discussion it is necessary to understand the map f in terms Then, on translation elements f restricts to the natural map X∗(T )I → X∗(Tad)I , is contained in (cid:102)W (R) = X∗(Tad)I (cid:111) W0; let us write f : (cid:102)W −→(cid:102)W (R). and f carries W K ⊂(cid:102)W isomorphically to W0, where K ⊂ G(L) is the parahoric of the semidirect product decomposition (cid:102)W = Wa (cid:111) Ω (4.2.6), where Ω is the stabilizer of the base alcove a inside (cid:102)W . Inside (cid:102)W (R) is the affine Weyl group Wa(R) = Q∨(Σ) (cid:111) W0, and we denote by Ω(R) the stabilizer in (cid:102)W (R) of the translate a− v, which is an alcove in A for Σ. Then(cid:102)W (R) = Wa(R) (cid:111) Ω(R), and ∼−→ Wa(R) and a map Ω → Ω(R). Endow (cid:102)W • w(cid:48) ≤ w in (cid:102)W =⇒ f (w(cid:48)) ≤ f (w) in (cid:102)W (R), with the converse holding exactly • f carries the subset Adm({µ}) ⊂ (cid:102)W bijectively onto Adm({µad}) ⊂ (cid:102)W (R), • f carries the subset Perm({µ}) ⊂(cid:102)W bijectively onto Perm({µad}) ⊂(cid:102)W (R). where {µad} denotes the image of Λ{µ} in X∗(Tad)I ; and • Adm({µ}) = Perm({µ}) in (cid:102)W ⇐⇒ Adm({µad}) = Perm({µad}) in (cid:102)W (R). when w(cid:48) ≡ w mod Wa; and that for any W -conjugacy class {µ} ⊂ X∗(T ), f restricts to an isomorphism Wa with the Bruhat order corresponding to a − v. Then it is clear that Moreover, we have Remark 4.4.8. The following variant of the preceding remark, which uses the building for G in place of the building for Gad, is sometimes more convenient in practice. Let (cid:101)A := X∗(S) ⊗ R ∼= X∗(T )I ⊗ R, and consider the natural map T (L)/T (L)1 Then there exists an extension of the displayed composite to a map(cid:102)W (cid:101)ν−→ (cid:101)A(cid:111) W0. More precisely, replacing A with (cid:101)A everywhere in the diagram (4.2.1), there exists a map(cid:102)W (cid:101)ν−→ (cid:101)A (cid:111) W0 making the diagram commute, and any two extensions differ ∼−→ X∗(T )I −→ (cid:101)A. by conjugation by a translation element, but this translation element is no longer uniquely determined. 60 Local models of Shimura varieties, I. Let Φa denote the affine root system for G relative to the composite (cid:102)W (cid:101)ν−→ (cid:101)A (cid:111) W0 −→ A (cid:111) W0. Let Σ denote the associated reduced root system on A, as in the preceding remark. Then we can regard the elements of Σ as linear functions on (cid:101)A, and the W0-action on (cid:101)A allows us to canonically lift the coroots to (cid:101)A: for each root α ∈ Σ we have the associated reflection sα ∈ W0, and this determines the associated coroot α∨ via the formula sα(x) = x − (cid:104)α, x(cid:105)α∨ for x ∈ (cid:101)A.16 (cid:101)R :=(cid:0)X∗(T )I , X∗(T )I /N, Σ, Σ∨(cid:1) Finally let N denote the torsion subgroup of X∗(T )I . Then is a root datum, and everything carries over from the previous remark with (cid:101)R in place of R. Remark 4.4.9. Although we are interested in minuscule conjugacy classes of cocharacters for applications to Shimura varieties, we caution that, in the context of the previous two remarks, the image of a minuscule(cid:101)Λ{µ} in X∗(T )I or X∗(Tad)I need not be minuscule for Σ. In this way the study of admissibility for non- minuscule cocharacters in root data is relevant to the study of admissibility for minuscule cocharacters in nonsplit groups. Remark 4.4.10. It is conjectured in [R, §3, p. 283] that Adm({µ}) = Perm({µ}) for any Weyl orbit {µ} of minuscule cocharacters in any extended affine Weyl group attached to a based root datum. Thus part (iii) of Proposition 4.4.5 is a partial confirmation of this conjecture. In fact, [R] formulates the more optimistic conjec- ture that Adm({µ}) = Perm({µ}) whenever {µ} is the conjugacy class attached to a sum µ of dominant minuscule cocharacters. However, this more optimistic version of the conjecture can fail, cf. [Sm4]. In particular, Adm({µ}) (cid:54)= Perm({µ}) for µ the sum of dominant minuscule coweights (1, 1, 1, 0) =(cid:0) 1 (cid:1) +(cid:0) 1 2 ,− 1 in D4 (using the standard coordinates, as in [B, Pl. IV]). 2 , 1 2 , 1 2 , 1 2 , 1 2 , 1 2 (cid:1) 2 Let us conclude this subsection by making more explicit the relation of the {µ}-admissible and {µ}-permissible sets to the local models discussed in this article and elsewhere in the literature. For all of the local models attached to GLn in §2.1 [Go1], GSp2g in §2.2 [Go2], ResF/F0 GLn in §2.4 [Go4, PR2], ResF/F0 GSp2g in §2.5 [Go4, PR2], and ramified, quasi-split GUn in §2.6 [PR4, Sm3, Sm4], the geometric special fiber of the local model M loc G,µ,L admits an embedding into an affine flag variety -- constructed very much in the spirit of §3.3 -- with regard to 16Of course, we can also canonically lift the coroots to (cid:101)A via the embedding X∗(Tsc)I (cid:44)→ X∗(T )I discussed in §4.2, p. 49. G. Pappas, M. Rapoport, and B. Smithling 61 which it decomposes into a union of Schubert cells indexed by exactly the {µ}- In §2.3, the orthogonal group GO2g is disconnected, so that as admissible set. in Remark 4.1.7 the present discussion does not literally apply. Nevertheless, the special fiber of the local model, which has two connected components, can still be embedded into an affine flag variety for GO2g, where it is found to contain the Schubert cells indexed by two admissible sets for GO◦ 2g: one for the conjugacy class of the cocharacter (cid:0)1(g), 0(g)(cid:1), and the other for the conjugacy class of the cocharacter(cid:0)1(g−1), 0, 1, 0(g−1)(cid:1). Note that these cocharacters are GO2g-conjugate but not GO◦ 2g-conjugate. See [Sm1]. In all of the examples mentioned in the previous paragraph, one also has an equality between {µ}-admissible and {µ}-permissible sets with the single exception of ramified, quasi-split GUn for n even and ≥ 4, in which case {µ}-admissibility and {µ}-permissibility are typically not equivalent. See [Sm4]. 4.5. Vertexwise admissibility We continue with the notation of the previous four subsections. Let K and K(cid:48) be parahoric subgroups of G(L) attached to subfacets of the base alcove a, and let {µ} ⊂ X∗(T ) be a W -conjugacy class. It is an immediate consequence of the properties of the Bruhat order that the canonical projection (cid:102)W → W K(cid:48)\(cid:102)W /W K induces a surjective map (4.5.1) Adm({µ}) (cid:16) AdmK(cid:48),K({µ}). If f is a subfacet of a with associated parahoric subgroup K, then for each vertex x of f , let Kx denote the associated parahoric subgroup and ρx : W K\(cid:102)W /W K (cid:16) W Kx\(cid:102)W /W Kx (cid:0)AdmKx ({µ})(cid:1). K ({µ}) := (cid:92) ρ−1 x Admvert the canonical projection. We make the following definition. Definition 4.5.2. The {µ}-vertexwise admissible set in W K\(cid:102)W /W K is the subset vertices x of f In other words, an element w ∈ W K\(cid:102)W /W K is {µ}-vertexwise admissible if W Kx wW Kx ∈ AdmKx ({µ}) for all vertices x of f . It is an obvious consequence of the map (4.5.1) that AdmK(µ) ⊂ Admvert K (µ), and we conjecture the following. Conjecture 4.5.3. Let {µ} ⊂ X∗(T ) be a W -conjugacy class of minuscule cochar- acters, and let f be a subfacet of a with associated parahoric subgroup K. Then the inclusion AdmK({µ}) ⊂ Admvert K ({µ}) is an equality. We do not know if the assumption that {µ} be minuscule is necessary, but the examples that we have studied all arise from local models, where the assumption holds by definition. 62 Local models of Shimura varieties, I. We note that in cases where {µ}-admissibility and {µ}-permissibility are K ({µ}) ⊂ equivalent, the conjecture is automatic. Indeed, for any {µ} we have Admvert PermK({µ}) because AdmK(cid:48)({µ}) ⊂ PermK(cid:48)({µ}) for any K(cid:48), and in particular for K(cid:48) of the form Kx, and because {µ}-permissibility is manifestly a vertex- wise condition. Hence the equality AdmK({µ}) = PermK({µ}) implies the equal- ity AdmK({µ}) = Admvert K ({µ}). Because of this, the conjecture may in some sense be regarded as a version for arbitrary groups of the conjecture in [R] that Adm({µ}) = Perm({µ}) for minuscule cocharacters in split groups; see Remark 4.4.10. We also note that the conjecture holds in all examples that we know of arising from local models. More precisely, for all of the local models attached to GLn in §2.1; GSp2g in §2.2; GO2g in §2.3; ResF/F0 GLn in §2.4; and ResF/F0 GSp2g in §2.5, the conjecture holds because {µ}-admissibility and {µ}-permissibility are equivalent by Proposition 4.4.5 and Remark 4.4.6. For the local models attached to ramified, quasi-split GUn for n odd in §2.6, the conjecture is known via the equivalence of {µ}-admissibility and {µ}-permissibility [Sm3], but these cases are not covered by Proposition 4.4.5. Finally, for the local models attached to ramified, quasi-split GUn for n even in §2.6, {µ}-admissibility and {µ}-permissibility are typically not equivalent, but the conjecture still holds in these cases [Sm4]. 5. Local models and nilpotent orbits In a few cases, the special fibers of local models can be described via nilpo- tent orbits and their closures. As was first observed in [PR1], this connection is especially tight in the case of the (ramified) group ResF/F0GLn. This also gives a connection between affine Schubert varieties for SLn and nilpotent orbit closures. In this section we discuss this relation in a somewhat informal manner. 5.1. Nilpotent orbits Let G be a reductive group over a field k and denote by g its Lie algebra, which we think of as an affine space. Recall that an element x of g is called nilpotent if its adjoint endomorphism ad(x) : g → g is nilpotent. The property of being nilpotent is invariant under the adjoint action of G on g; a nilpotent orbit Nx = {ad(g) · x g ∈ G} is the orbit of a nilpotent element x under the adjoint action. Here we consider Nx as the reduced subscheme with underlying topological space the orbit of x. We will denote by Nx the Zariski closure of Nx in the affine space g. The varieties Nx have been the subject of intense study ([KP, BC, deC-P2], etc.) The most classical example of course is when G = GLr and g = Matr×r. Then NA is the conjugation orbit of the nilpotent matrix A. These orbits are parametrized by partitions r = (r1 ≥ r2 ≥ ··· ≥ rs) of r; the numbers ri are the sizes of the blocks in the Jordan decomposition of A. G. Pappas, M. Rapoport, and B. Smithling 63 5.2. Relations to local models We consider the situation of §2.4, i.e take G = ResF/F0GLn, where F/F0 is a totally ramified separable extension of degree e. Let π be a uniformizer of OF , and let Q(T ) ∈ OF0 [T ] be the Eisenstein polynomial satisfied by π. Recall from loc. cit. that the minuscule cocharacter µ is determined by choos- ing rϕ with 0 ≤ rϕ ≤ n, for each embedding ϕ of F in a fixed algebraic closure F 0 of F0. We choose the lattice chain L = {πkΛ0}k∈Z to be given by the multi- ples of the standard OF -lattice in F n. (Then the corresponding parahoric group is maximal and special). We denote by M the naive local model M naive G,{µ},L for these choices (defined in §2.4) and write M loc for the corresponding local model G,{µ},L given as the flat closure of M ⊗OE E over the ring of integers OE of the M loc reflex field E. Denote by kE the residue field of OE. ϕ rϕ. Then the subspace F := FΛ0 ⊂ Λ0,S := Λ0⊗Zp OS occurring the GLr-torsor (cid:102)M over M defined by in the definition of M is locally on S free of rank n− r. This allows us to consider (cid:102)M (S) =(cid:8) (F, α)(cid:12)(cid:12) F ∈ M (S), α : Λ0,S/F ∼−→ Or Set r =(cid:80) (cid:9) , S and construct a GLr-equivariant morphism (cid:101)q : (cid:102)M −→ N , (cid:12)(cid:12)(cid:12)(cid:12) det(X · I − A) ≡(cid:89) with (cid:26) (cid:27) where the GLr-action on the target is via conjugation. The morphism(cid:101)q is smooth (cid:0)X − ϕ(π)(cid:1)rϕ, Q(A) = 0 A ∈ Matr×r (5.2.1) N := ϕ , [PR1, Th. 4.1], and hence we obtain a smooth morphism of algebraic stacks q : M −→ [GLr\N ]. (5.2.2) Note that the special fiber N ⊗OE kE is the GLr-invariant subscheme of the nilpo- tent matrices Nilpr×r over kE, given as (cid:12)(cid:12) det(X · I − A) ≡ X r, Ae = 0(cid:9) . N ⊗OE kE =(cid:8) A ∈ Matr×r (5.2.3) Recall the dual partition t of the decomposition {rϕ}ϕ of r defined by t1 = #{ ϕ rϕ ≥ 1}, t2 = #{ ϕ rϕ ≥ 2}, etc. We have t1 ≥ t2 ≥ ··· ≥ tn. Consider the (reduced by definition) closed nilpotent orbit N t that corresponds to the partition t. All matrices in this closure N t have Jordan blocks of size at most e. Hence we have a GLr-equivariant closed immersion i : N t (cid:44)→ N ⊗OE kE. From [PR1], Theorem 5.4 and the above, we now deduce that the special fiber M loc ⊗OE kE of the local model M loc is isomorphic to the pull-back of i along q. This gives the following: 64 Local models of Shimura varieties, I. Theorem 5.2.4. There is a smooth morphism of algebraic stacks qloc : M loc ⊗OE kE −→ [GLr\N t]. (cid:3) Corollary 5.2.5. The special fiber M loc ⊗OE kE of the local model M loc for the choice of µ determined by {rϕ}ϕ is smoothly equivalent to the closed nilpotent orbit (cid:3) N t. In particular, M loc ⊗OE kE is reduced. By [MvdK] the closed orbits N t are normal and Frobenius split (when kE has positive characteristic), and so we conclude that the same properties are true for M loc ⊗OE kE. Remark 5.2.6. Note that if all rϕ differ amongst themselves by at most 1, then N t = (N ⊗OE kE)red. In [PR1], it is conjectured that N ⊗OE kE is in fact reduced. This holds by a classical result of Kostant when r ≤ e, and this is proved by Weyman in [W] in the cases where either char kE = 0, or where e = 2, comp. Theorem 6.1.8 below. Remark 5.2.7. The fact that M loc ⊗OE kE is reduced and normal has found an interesting application in the theory of deformations of Galois representations by Kisin [Ki]. This application is based on the following lemma, comp. [Ki, Cor. 2.4.10]: Let X be a scheme which is proper and flat over the spectrum S of a complete discrete valuation ring. We denote by Xη, resp. Xs the generic, resp. the special fiber. If Xs is reduced, then there are bijections between the sets of connected components π0(Xs) = π0(X) = π0(Xη). Consider the dominant cocharacter λ of GLn that corresponds to t, and de- note by Oλ the corresponding Schubert variety in the affine Grassmannian for GLn. Now we can see that the embedding of the special fiber of the local model in the affine Grassmannian (cf. §3.3, and for this example [PR1]) gives an isomor- phism (5.2.8) M loc ⊗OE kE ∼−→ Oλ. Since by varying the data t we can obtain all dominant cocharacters λ, this ob- servation together with Theorem 5.2.4 also shows Theorem 5.2.9 ([PR1, Th. C]). Any Schubert variety in the affine Grassmannian of GLn is smoothly equivalent to a nilpotent orbit closure for GLr, for suitable (cid:3) r. This has also been shown independently by Mirkovi´c and Vybornov [MVy]. Recall that earlier Lusztig [Lu] interpreted certain Schubert varieties in the affine Grassmannian of GLr as compactifications of the nilpotent variety of GLr (namely the Schubert variety corresponding to the coweight (r, 0, . . . , 0)), compatible with G. Pappas, M. Rapoport, and B. Smithling 65 the orbit stratifications of both varieties. In particular, as used by Lusztig in his paper, all singularities of nilpotent orbit closures occur in certain Schubert varieties in the affine Grassmannians. The above goes in the opposite direction. Remark 5.2.10. This tight connection between local models (or affine Schubert varieties) and nilpotent orbits does not persist for other groups. There are, how- ever, some isolated instances of such a correspondence in other cases. For example, the reduced special fibers of the local models for the ramified unitary groups and special parahoric subgroups are smoothly equivalent to nilpotent orbit closures in the classical symmetric pairs (sln, son), resp. (sl2n, sp2n) which have been studied by Kostant-Rallis [KosR], and Ohta [Oh]. See [PR4, §5], and especially Theorem 5.4 and its proof in [PR4], for more details. However, not all such nilpotent orbits appear in this correspondence. 6. Local models and matrix equations In some cases, local charts around points of local models can be described via the spectra of affine rings given by generators and relations, in shorthand matrix form (matrix equations). We have already seen some instances of this in Example 2.1.2, Remark 2.6.13, and §5. Rather than giving a formal definition of what we mean by matrix equations, we list in this section a few examples. Obviously, structure results on matrix equations have consequences for local models. What is more surprising is that sometimes results on local models imply structure results on matrix equations. 6.1. Matrix equations related to naive local models Our first example is as follows. Let O be a discrete valuation ring with uniformizer π. We fix positive integers r and n, and consider the following closed subscheme of affine space of dimension nr2 over SpecO, (cid:12)(cid:12)(cid:12)(cid:12) A1A2 ··· An = A2A3 ··· A1 = ··· = = AnA1 ··· An−1 = π · I (cid:27) (cid:26) (A1, . . . , An) ∈ Matn r×r (6.1.1) Zr,n = . In the special case r = 1 there is only one equation X1X2 ··· Xn = π in the n unknowns X1, X2, . . . , Xn, which describes the semistable reduction case. The special fiber Zr,n ⊗O k is called the generalized circular variety over the residue field k. The scheme Zr,2 ⊗O k is called the variety of circular complexes, and has been considered long before local models were defined, cf. [MT, St]. Theorem 6.1.2 (Gortz [Go1, 4.4.5]). The scheme Zr,n is flat over O, with reduced special fiber. The irreducible components of its special fiber are normal with rational (cid:3) singularities, so in particular are Cohen-Macaulay. The matrix equation (6.1.1) arises in the analysis of local charts for local models for the triple consisting of GLn, the Iwahori subgroup, and the minuscule 66 Local models of Shimura varieties, I. cocharacter µ = (1(r), 0(n−r). Recall from Theorem 2.1.1 that, in this case, the local model coincides with the naive local model. More precisely, and similarly to what happened in §5.2, we define a scheme (cid:102)M loc over M loc which parametrizes, terms of these bases, we obtain a morphism q : (cid:102)M loc → Zr,n, which turns out to be in addition to a point (Fi i ∈ Z/nZ) of M loc(S), a basis of Λi,S/Fi. Then associating to the transition morphisms Λi,S/Fi → Λi+1,S/Fi+1 their matrices in smooth, cf. [Fa1, PR1]. Hence the properties claimed in the theorem follow from Theorem 2.1.1, locally at each point of Zr,n in the image of q. Something similar holds for any parahoric subgroup corresponding to a partial periodic lattice chain L. Now apply this result to the local model for the triple consisting of GLrn, the parahoric subgroup corresponding to the periodic lattice chain L = {Λi i ∈ rZ}, and the minuscule coweight (1(r), 0(rn−r)). It is easy to see that in this case the morphism q : (cid:102)M loc → Zr,n is surjective, and this proves the claim, cf. [Go5]. The next example arises in the analysis of the naive local model for the triple consisting of a symplectic group, a non-special maximal parahoric subgroup, and the unique conjugacy class of nontrivial minuscule coweights µ. Let n be even, and define Z = { A ∈ Matn×n AJ tA = tAJA = π · I }. (6.1.3) Here, as in the beginning of §2.2, J = Jn denotes the matrix describing the standard symplectic form. Theorem 6.1.4 (Gortz [Go2, §5]). The scheme Z is flat over O, with reduced (cid:3) irreducible normal special fiber, which has only rational singularities. The proof of Gortz of this theorem uses local model techniques, combined with the theory of deConcini [deC1] of doubly symplectic tableaux which gives a good basis of the coordinate ring of Z ⊗O k as a k-vector space. Similarly, in the analysis of the naive local model for the triple consisting of a symplectic group, the parahoric subgroup which stabilizes a pair of lattices Λ, Λ(cid:48) where Λ is self-dual and Λ(cid:48) is self-dual up to scalar π, and the unique (nontrivial dominant) minuscule coweight µ, the following matrix equations arise, (6.1.5) Z = { A, B ∈ Matn×n AB = BA = π · I, tA = A, tB = B }. More precisely, Z is locally around the origin isomorphic to an open neighborhood of the 'worst point' of the local model in question. Theorem 6.1.6 (Chai -- Norman [CN], [DP], Gortz [Go2, 2.1]). The scheme Z is flat, normal and Cohen-Macaulay over O, with reduced special fiber. The irre- ducible components of its special fiber are normal with rational singularities. (cid:3) Whereas Gortz' proof of this theorem uses local models (in particular, the embedding of the special fiber in the affine Grassmannian) and Frobenius splitting methods, the proof of Chai and Norman uses techniques from the theory of algebras G. Pappas, M. Rapoport, and B. Smithling 67 with straightening laws (and the proof in [DP] is a simplification of this proof). The Cohen-Macaulay property of Z is shown directly in [CN], but it can also be derived by the methods of Gortz (see [Go1, §4.5.1]). We refer to [Go2] and [DP] for further discussion of other methods in the literature. Another example of a matrix equation we have seen already in the previous section, cf. (5.2.1). For better comparison with the matrix equations appearing right after it, let us recall it. As in the beginning of §5.2, let F/F0 be a totally ramified separable extension of degree e. Let π be a uniformizer of OF , and let Q(T ) ∈ OF0[T ] be the Eisenstein polynomial satisfied by π. As in loc. cit., we fix a tuple r = (rϕ). Then (6.1.7) which is a scheme over SpecOE, where E is the reflex field corresponding to r. ϕ(X − ϕ(π))rϕ , det(X · I − A) ≡(cid:81) Q(A) = 0 and (cid:26) N = Nr = A ∈ Matr×r (cid:12)(cid:12)(cid:12)(cid:12) field kE is zero, or that e ≤ 2, or that (cid:80) (cid:27) Theorem 6.1.8 (Weyman [W]). Assume that all e integers rϕ differ amongst each other by at most 1. Assume further that either the characteristic of the residue ϕ rϕ ≤ e. Then N is flat over O, with (cid:3) reduced special fiber, which is normal with rational singularities. As explained in the previous section, the scheme N occurs in relation to the naive local model for the triple consisting of the group G := ResF/F0 GLn, the natural special maximal parahoric subgroup, and the minuscule cocharacter µ determined by r, cf. §2.1. If the conclusion of Theorem 6.1.8 were true without the "further" restrictions listed (as is conjectured in [PR1]), then the local model and the naive local model would coincide in this case. For the triple consisting of G := ResF/F0 GSp2n, the natural special parahoric subgroup, and the natural minuscule cocharacter µ, one obtains in the analogous way the following matrix equation, cf. [PR2, 12.5], (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) a, b ∈ Matne×ne, tb = −b, Q(a) = 0, det(X · I − a) ≡(cid:81) ϕ(X − ϕ(π))n (cid:41) . (6.1.9) P = ∈ Mat2ne×2ne (cid:40)(cid:18)a b (cid:19) 0 ta Conjecture 6.1.10 ([PR2, 12.5]). The scheme P is flat over SpecOF0 , with reduced special fiber. If this conjecture held true, it would follow that in this case the naive local model is flat, i.e., coincides with the local model -- which would constitute a special case of Conjecture 2.5.4. Our next examples are related to the case of a ramified unitary group. Let F/F0 be a ramified quadratic extension obtained by adjoining the square root of a uniformizer π0 of F0. The following matrix equations arise in connection with the triple consisting of a group of unitary similitudes of size n for F/F0, a special maximal parahoric subgroup (in the case when n is odd, the parahoric subgroup 68 Local models of Shimura varieties, I. fixing a self-dual lattice, and in the case when n is even, the parahoric subgroup fixing a lattice which is selfdual up to a scalar π0), and a minuscule cocharacter given by (r, s) with r + s = n. Consider the following schemes of matrices. When n is odd, √ A ∈ Matn×n A ∈ Matn×n (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) charA(T ) = (T − √ A2 = π0 · I, At = HAH, √ π0)s(T + ∧s+1A = 0, ∧r+1A = 0 charA(T ) = (T − √ A2 = π0 · I, At = −JAJ, √ π0)s(T + ∧s+1A = 0, ∧r+1A = 0  .  , π0)r, π0)r, (6.1.11) N = When both n = 2m and s are even, (6.1.12) N = where the conditions on wedge powers are imposed only when r (cid:54)= s. Here, as in the beginning of §2.2, H = Hn denotes the antidiagonal unit matrix, and J = Jn the skew-symmetric matrix with square blocks 0m on the diagonal and Hm, resp. −Hm, above the diagonal, resp. below the diagonal. Conjecture 6.1.13 ([PR4, §5]). The schemes N above are flat over OE, with reduced special fiber (which is then normal, with rational singularities). If this conjecture were true, it would follow that for the local models men- tioned above, the wedge local model contains the local model as an open sub- scheme, cf. Remark 2.6.10 (a corrected version of [PR4, Rem. 5.3]). Remark 6.1.14. There should be similar matrix equations related to local models for orthogonal groups. This does not seem to have been investigated so far. 7. Local models and quiver Grassmannians In a few cases, the special fibers of local models can be identified with certain quiver Grassmannians in the sense of Zelevinsky and others, cf. [Z]. In this section we discuss this in rough outline. 7.1. Quiver Grassmannians Let Q be a quiver, with set of vertices Q0 and set of arrows Q1. Then Q is in the obvious way a category. Let (V, ϕ) be a representation of Q over the field k, in other words, a functor from the category Q to the category of finite-dimensional vector spaces over k. To any such representation there is associated its dimension vector d(V ) ∈ (Z≥0)Q0 with d(V )i = dim Vi. Let e ∈ (Z≥0)Q0 such that e ≤ d, i.e., each component of e is less than or equal to the corresponding component of d. The quiver Grassmannian associated to these data is the projective variety (comp. e.g. [CR, §1]) (7.1.1) Gre(V ) =(cid:8)Fi ∈ Gr(ei, Vi), ∀i ∈ Q0 (cid:12)(cid:12) ϕi,j(Fi) ⊂ Fj, ∀(i, j) ∈ Q1 (cid:9) . G. Pappas, M. Rapoport, and B. Smithling 69 The subgroup GV of elements in(cid:81) GL(Vi) which respect the homomorphisms {ϕi,j}(i,j)∈Q1 acts in the obvious way on Gre(V ). Most often, there are infinitely many orbits. i∈Q0 7.2. Relations to local models We consider the situation of §2.4, i.e., take G = GLn over F , µ = (1(r), 0n−r)), and a periodic lattice chain in F n extracted from the standard lattice chain Λi, i ∈ Z, by keeping those Λi with i congruent to an element in a fixed non-empty subset I ⊂ Z/nZ. Let Λi = Λi ⊗O k, with the linear maps Λi → Λi+1 induced by the inclusions Λi ⊂ Λi+1. Using the identification Λi+n = πΛi, we may identify Λi+n with Λi, and therefore define unambiguously Λi for i ∈ Z/nZ. By keeping only those Λi with i ∈ I, we obtain a representation ΛI of the quiver of type (cid:101)AI. Here an extended Dynkin diagram of type A defines a quiver by choosing the clockwise orientation of its bonds. This representation is characterized up to isomorphism by the following conditions: (i) dim Λi = n for all i ∈ I. (ii) dim Ker ϕi,i(cid:48) = i(cid:48) − i for all i ≤ i(cid:48) ≤ i + n. From §2.1 it is plain that a point of the local model M = M loc G,µ,I with values in a k-scheme S corresponds to a S-valued point Fi of the Grassmannian of subspaces of dimension n − r of Λi, one for each i ∈ I such that, under ϕi,i(cid:48), the subspace Fi is carried into a subspace of Fi(cid:48). Comparing with (7.1.1), we see that M⊗OF k can be identified with the quiver representation ΛI of the quiver of type (cid:101)AI. Furthermore, under this identification, Grassmannian Grn−r(ΛI ) of subspaces with scalar dimension vector n − r of the the action of the loop group L+PI on M ⊗OF k from Example 3.3.1 coincides with the action of the automorphism group GV of the quiver ΛI from §7.1. In particular, in this case the GV -action has only finitely many orbits. From this perspective, the local model M is a deformation over OF of a quiver Grassmannian over k. Remark 7.2.1. In [CR] and other papers in the area of representations of alge- bras, quiver Grassmannians are considered as varieties, i.e., nilpotent elements are neglected. It follows from Gortz's Theorem 2.1.1 that the quiver Grassmannians of type (cid:101)A are reduced. For other quiver Grassmannians this question does not seem to have been considered in the literature. Remark 7.2.2. It is not clear which local models can be described in this way. (i) In §2.4, we mentioned the splitting model M = MG,{µ},L from [PR2] for G = ResF/F0(GLn), where F/F0 is a totally ramified extension. Similar to the above identification, the special fiber of M can be described as a subvariety of a quiver flag variety of a representation of a quiver of type (cid:101)A (defined by the 70 Local models of Shimura varieties, I. condition that the nilpotent operator induced by π induces the zero endomorphism on a certain associated graded vector space), cf. [PR2]. (ii) Recall from §2.2 the local model corresponding to the triple (GSp2g,{µ},L), where {µ} is the unique conjugacy class of nontrivial minuscule coweights of GSp2g, and where L is a self-dual periodic lattice chain. In fact, to simplify matters, let us assume that L is maximal. By choosing the symplectic form as in §2.2, and taking for L the standard lattice chain, we see that (cid:98)Λi = Λ−i. Using again the notation Λi for Λi ⊗OF k, we see that we obtain a non-degenerate pairing (7.2.3) Now a point of the special fiber M ⊗OF k is given by a subspace Fi of dimension g of Λi, one for each i, such that, under each map ϕi,i+1 : Λi → Λi+1, the subspace Fi is mapped into a subspace of Fi+1, and such under the natural pairing (7.2.3) the subspaces Fi and F−i are perpendicular to each other, for all i. However, this kind of object has apparently not been considered in the context of quiver Grassmannians. Λi × Λ−i −→ k. 8. Local models and wonderful completions In this section, which is of an (even) more informal nature, we will ex- plain various relations between the theory of local models and the so-called won- derful compactifications of symmetric spaces. This extends to also give a rela- tion of local models for GLn with Lafforgue's compactifications of the quotients (P GLr)s/P GLr. At the moment we do not have a very good understanding of the scope of these connections between the theory of local models and those theories; they appear somewhat sporadic. As a result we will mainly concentrate on several illustrative examples and explanations that, we hope, are enough to explain why one should expect such a connection in the first place. We also hope that this will motivate readers to explore possible generalizations. An instance of a connection between some local models and wonderful com- pletions was first observed by Faltings [Fa1] (also [Fa2]). Faltings starts by con- sidering certain schemes given by matrix equations. These schemes are given by embedding symmetric spaces in projective spaces defined by homogeneous line bundles and considering their closures. In several cases, these give affine charts of local models in the sense of §6. Faltings then uses constructions from the theory of wonderful completions of symmetric spaces to produce birational modifications of these schemes. In many cases, these also give modifications of the corresponding local models which are regular and have as special fiber a divisor with (possibly non-reduced) normal crossings. In this section, after a quick review of wonderful completions (§8.1), we will explain (in §§8.2, 3) a different and more direct relation between local models and wonderful completions, based on some unpublished notes [P2] by the first G. Pappas, M. Rapoport, and B. Smithling 71 author. This was inspired by Faltings' paper. The goal in this approach is to relate local models to closures of orbits of parabolic subgroups in the wonderful completion; such parabolic orbit closures have been studied by Brion and others [Br2, BrP, BrTh]. In some cases this gives an alternative construction of the local models. Then, in §8.4, we give some comments on Faltings' methods. Contrary to our notation earlier in the paper, in this section we shall use the symbols Λ0, Λ1, . . . to denote arbitrary lattices in a vector space, not lattices in the standard lattice chain (2.0.2). 8.1. Wonderful completions For a more complete overview of this "wonderful" theory and its connections to classical algebraic geometry we refer the reader to [deC, deC-P]. We also refer to [deC-S] and [Fa1, §2] for details on the actual constructions in the generality we require. The basic set-up is as follows. Let G be an adjoint semi-simple algebraic group over a field F of characteristic (cid:54)= 2 which is equipped with an involution θ defined over F . Let H = Gθ be the fixed points of the involution which is then a reductive group over F ; it is connected when G is simply connected, cf. [deC-S, §1]. The corresponding symmetric space is the affine quotient X = G/H over F . The wonderful completion X of X is a smooth projective variety over F which contains X as a dense Zariski open subset. It supports a left action of G that extends the translation action on X. In addition, it has the following property: The complement X − X is a divisor with normal crossings which is the union of a finite set of smooth irreducible G-stable divisors such that all their partial intersections are transversal; the closures of the G-orbits in X are precisely these intersections. One basic example is obtained by taking the group to be the product G × G with θ(g1, g2) = (g2, g1), so that X = (G × G)/G (cid:39) G. Then X = G is a compactification of the group G. Another well-studied example is given by taking the group P GLn with involution given by θ(g) = (tg)−1. Then H = P GOn, and X is the variety of invertible symmetric matrices and X is the classical variety of "complete quadrics," see [deC-P, Lak, dCGMP]. 8.2. The example of the general linear group In this subsection, we explain the method of [P2]. We will concentrate on two classes of examples. For simplicity, we only consider the equal characteristic case17, i.e., the local models will be schemes over the discrete valuation ring O = k[[t]] with uniformizer π = t. Let F = k((t)). Suppose that Λ0, Λ1 are two O-lattices in the vector space V = F n such that Λ0 ⊂ Λ1 ⊂ t−1Λ0. Choose a O-basis of 17An extension to the mixed characteristic case depends on defining wonderful completions over Zp. This should not present any problems (provided p is odd) and is roughly sketched in [Fa1] and [deC-S], but we prefer to leave the details for another occasion. 72 Local models of Shimura varieties, I. e1, . . . , en of Λ1 such that Λ0 has O-basis formed by e1, . . . , em, tem+1, . . . , ten, for some m ≤ n − 1. Fix an integer 0 < r < n. Recall from §2.1 that the naive local model M = M naive corresponding to the triple consisting of GLn, of the minuscule cocharacter µ = (1(r), 0(n−r)), and of the above lattice chain, is the scheme over SpecO whose S-points parametrize pairs (F0,F1) of OS-subbundles of rank n − r of Λ0 ⊗O OS, resp. Λ1 ⊗O OS, such that the following diagram commutes (8.2.1) Λ0 ⊗O OS Λ1 ⊗O OS t−1Λ0 ⊗O OS ⊂ F0 ⊂ / F1 ⊂ / t−1F0. Here the horizontal maps on the top row are induced by the inclusions Λ0 ⊂ Λ1 ⊂ ∼−→ t−1Λ0 and t−1F0 stands for image of F0 under the isomorphism Λ0 ⊗O OS t−1Λ0⊗OOS induced by multiplication by t−1. (Recall that in this case, by Gortz's Theorem 2.1.1, the naive local model is equal to the local model M loc = M .) Of course, M is realized as a closed subscheme of the product of Grass- mannians Gr(n − r, n)O ×O Gr(n − r, n)O. The generic fiber is isomorphic to Gr(n − r, n)F , and in the generic fiber the embedding of M ⊗O F in the prod- uct Gr(n − r, n)F ×F Gr(n − r, n)F is described by F1 (cid:55)→ (A · F1,F1) with A = diag(1(m), t(n−m)) (m copies of 1 and n − m copies of t placed along the diagonal.) This allows us to view M as a deformation over O of what is essen- tially the diagonal embedding Gr(n − r, n)F (cid:44)→ Gr(n − r, n)F ×F Gr(n − r, n)F . Such deformations have been considered by Brion [Br1] (following work by Thad- deus [Th] and others on Kapranov's "Chow quotients" [Ka]). Brion views such deformations as parametrized by a part of the Hilbert scheme of subschemes of Gr(n − r, n)O ×O Gr(n − r, n)O. Here is how this is related to the wonderful completion of G = P GLn. Set P = Pµ for the standard parabolic subgroup of G such that Gr(n − r, n) = G/P . Let G be the wonderful completion of G and denote by P ⊂ G the Zariski closure of P in G. The product G × G acts on G in a way that extends the action (g1, g2)· g = g1gg−1 of G× G on G. This also restricts to an action of P × P on P . Over G, we can construct a family of closed subschemes of Gr(n−r, n)×Gr(n−r, n) as follows. Consider the commutative diagram 2 ι G/P × G/P × G M φ (QQQQQQQQQQQQQQQQQ π G/P × G/P prG G. / / / / / / / / ( G. Pappas, M. Rapoport, and B. Smithling 73 Here M = (G×G) of G × G × P by the right action of P × P by P×P× P is the "contracted product" which is given as the quotient (cid:0)(g1, g2), x(cid:1) · (p1, p2) =(cid:0)g1p1, g2p2, (p−1 ι(cid:0)(g1, g2), x(cid:1) =(cid:0)g1P, g2P, (g1, g2) · x(cid:1) 1 , p−1 2 ) · x(cid:1). The morphism ι is given by and the morphism φ by φ(cid:0)(g1, g2), x(cid:1) = (g1P, g2P ). It is easy to see that ι is a closed immersion; hence we can view π : M → G as a family of closed subschemes of G/P × G/P over the base G. It follows from [Br1, BrP] that this is a flat family. Now the matrix A gives, by the valuative criterion of properness, a well- defined point [A] : Spec k[[t]] → G. Then it is not hard to see that the base change M ×G,[A] O of π along [A] can be identified with the flat closure of the generic fiber M ⊗O F in the naive local model M . By definition, this is the local model M loc for our situation, and so we obtain the following result. Theorem 8.2.2. In the situation described above, there is an isomorphism M loc (cid:39) M ×G,[A] SpecO. Indeed, using the flatness result above, it is enough to check that this base change is a closed subscheme of the naive local model M . This can be easily verified (see the proof of Theorem 8.3.3 below for a more detailed explanation of a more interesting case). Remark 8.2.3. a) As was pointed out above, the naive local model M is flat by Gortz's theorem, and so M loc = M . However, the above construction of the local model is independent of Gortz's result. b) Note that the morphism φ above is a (Zariski) locally trivial fibration with fibers isomorphic to P . Hence, the singularities of the total space M are smoothly equivalent to the singularities of P . The singularities of parabolic orbit closures in complete symmetric varieties (such as G) have been studied by Brion and others (e.g [Br2, BrP]); we can then obtain results on the singularities of the local models in question. For example, one can deduce from this approach that the special fibers of these local models are reduced and Cohen-Macaulay. Of course, these results can also be obtained by the method of embedding the local models in affine flag varieties described in the previous sections (see [Go1]), cf. Theorem 2.1.1. (In the case of this lattice chain, which consists of multiples of two lattices, one can obtain that the special fiber, as a whole, is Cohen-Macaulay, cf. Remark 2.1.3. See also Remark 2.2.5 and Theorem 6.1.6 for similar results in the symplectic case.) 8.3. Other examples, some symplectic and orthogonal cases Here we will explain how the method of §8.2 can be extended to relate cer- tain local models for maximal parahoric subgroups in the cases of symplectic and 74 Local models of Shimura varieties, I. even orthogonal groups to wonderful completions of the corresponding symmetric spaces. assume that the form h is split and "standard" i.e it satisfies(cid:0)h(ei, ej)(cid:1) resp.(cid:0)h(ei, ej)(cid:1) Suppose that n = 2m is even. We assume that V = F n = k((t))n is equipped with a perfect form h : V × V → F which is alternating, resp. symmetric, the two cases leading to a description of local models with respect to the symplectic, resp. the orthogonal group. When h is symmetric, we assume char(k) (cid:54)= 2. We will i,j = Jn, i,j = Hn, for the standard basis {ei}i of V = F n, with the matrices Jn, resp. Hn as in (2.2.1) and (2.2.2). Denote by S the matrix of the form h so that S = Jn, resp. S = Hn. Recall that we denote by G(V, h) the group of similitudes of the form h. By the above, this is GSpn(F ), resp. GOn(F ). We consider the minuscule cocharacter µ = (1(m), 0(m)) for G(V, h) expressed as a cocharacter for the standard torus in GL(V ) = GLn. For 0 ≤ r ≤ m, we consider the lattice πOej + r(cid:88) Λ = n(cid:88) Oej. j=1 j=r+1 Then Λ ⊂(cid:98)Λ. We will denote by α the inclusion Λ ⊂(cid:98)Λ. The form h restricts to give an O-bilinear form Λ × Λ → O and a perfect O-bilinear form Λ ×(cid:98)Λ → O; we will also write h for these forms. We also denote by h(cid:48) the (different) alternating, resp. symmetric, form on Λ given on the standard basis {πe1, . . . , πer, er+1, . . . , en} of Λ by the matrix S . Denote by L the O-submodule of rank m of Λ generated by the first m standard basis elements of Λ as listed above; it is totally isotropic for both forms h(cid:48) and h. In this case, the local model M loc can be described as follows. Let us first give the "naive" local model M = M naive for this situation. Consider the functor over SpecO whose points with values in an O-scheme S are given by OS-submodules F ⊂ Λ ⊗O OS, which are OS-locally free direct summands of rank m such that Here the perpendicular F⊥ ⊂ (cid:98)Λ ⊗O OS is by definition the kernel of the OS- homomorphism (Λ ⊗O OS)∗ =(cid:98)Λ ⊗O OS → F∗ which is the dual of the inclusion (αOS )(F) ⊂ F⊥. F ⊂ Λ ⊗O OS. This condition is equivalent to (h ⊗O OS)(F,F) ≡ 0. This functor is representable by a projective scheme M over SpecO which is a closed subscheme of the Grassmannian Gr(m, n) over SpecO. The generic fiber of M can be identified with the Langrangian, resp. (disconnected) orthogonal Grassmannian of isotropic m-subspaces in n-space. The local model M loc is by definition the (flat) Zariski closure of the generic fiber in M . G. Pappas, M. Rapoport, and B. Smithling 75 Consider the involution θ on G = PGLn given by θ(g) = S−1(tg)−1S. The n can be identified with the groups P GSp2m, resp. P GOn. fixed points H = PGLθ Let us consider the symmetric space X = G/H. The morphism identifies X with the quotients gH (cid:55)−→ Ag = (tg)−1 · S · g−1. (cid:8) A ∈ Matn×n (cid:8) A ∈ Matn×n (cid:12)(cid:12) tA = −A, det(A) (cid:54)= 0(cid:9)(cid:14)Gm , (cid:12)(cid:12) tA = A, det(A) (cid:54)= 0(cid:9)(cid:14)Gm of antisymmetric, resp. symmetric n × n invertible matrices up to homothety. Consider the wonderful completion X of the symmetric space X = G/H. By the construction of X, it follows that there is a morphism T : X −→ Pn2−1 = (Matn×n − {0})/Gm which extends the natural inclusion G/H (cid:44)→ (Matn×n −{0})/Gm. The morphism T factors through the closed subscheme given by matrices which are antisymmetric, resp. symmetric. Now let us consider the parabolic P of G that corresponds to µ, so that Gr(m, n) = G/P . Let us also consider the Zariski closure P mod H = P/P ∩ H of the orbit of 1 · H by the action of P ⊂ G in X. There is a diagram q←− G ×P (P mod H) π−→ G/H. G/P (8.3.1) Here G ×P (P mod H) = (G × P mod H)/P where the quotient is for the right P -action given by (g, x) · p = (gp, p−1 · x). We have q(g, x) = gP and π is given by π(g, x) = g · x, via the action of G on G/H. There is also a morphism (8.3.2) given by ι(z) =(cid:0)q(z), π(z)(cid:1). These fit in a diagram ι : G ×P (P mod H) −→ G/P × G/H , M ι G/P × G/H pr2 G/H, %KKKKKKKKKKK π q G/P where M = G ×P (P mod H). As in [Br1], one can see that: a) The morphism q is an ´etale locally trivial G-equivariant fibration with fibers isomorphic to P mod H. b) The morphism ι is a closed immersion which identifies G ×P (P mod H) with the closed subscheme of G/P × G/H whose points (gP, x) satisfy the "incidence" condition x ∈ g · (P mod H). c) The morphism π : M → G/H is flat. / / % 76 Local models of Shimura varieties, I. Now consider the matrix HΛ =(cid:0)h(ei, ej)(cid:1) ij ∈ Matn×n(O) obtained by the restriction of our form h to Λ × Λ. Since h ⊗O F is perfect, this matrix HΛ gives an F -valued point of G/H. By properness, this extends to a point [HΛ] : SpecO −→ G/H. After these preparations we can finally give the description of the local model. Theorem 8.3.3 ([P2]). Under our assumptions, there is an isomorphism M loc (cid:39) M ×G/H,[HΛ] SpecO. Proof. (Sketch) Denote by M(cid:48) the base change in the statement of the theorem: (8.3.4) M(cid:48) G ×P (P mod H) π SpecO [HΛ] / G/H. By c), M(cid:48) → SpecO is flat. Since, by definition, the local model M loc is the Zariski closure of the generic fiber M ⊗O F in the naive local model M , it remains to show that M(cid:48) is a closed subscheme of M and has the same generic fiber, i.e., M(cid:48) ⊗O F = M ⊗O F . Let us identify G/P with the Grassmannian using gP (cid:55)→ F = gL. Recall that we can identify M with a closed subscheme of G/P : this is the subscheme of points gP for which [HΛ] ∈ g · (P mod H), or equivalently g−1 · [HΛ] ∈ P mod H. Using that L is an isotropic subspace for the form h(cid:48), we now obtain that the image of P mod H under the morphism T is contained in the closed subscheme with affine cone the antisymmetric or symmetric matrices A for which tv · A · w = 0 v, w ∈ L . (8.3.5) Now suppose that gP is in M(cid:48), i.e., g−1 · [HΛ] ∈ P mod H. By applying T we find that A = (tg)T ([HΛ])g satisfies (8.3.5). Since, by definition, the O-valued point T ([HΛ]) is equal to HΛ =(cid:0)h(ei, ej)(cid:1) for all ij, we obtain that tv ·t gT ([HΛ])g · w = h(gv, gw) = 0 (8.3.6) Since F = gL, this shows that F is isotropic for h. Hence M(cid:48) is a closed subscheme of M . Now it is not hard to show that the generic fibers of M(cid:48) and M are equal, (cid:3) and the claim follows. for all v, w ∈ L . Remark 8.3.7. This approach can also be applied to the local model studied by Chai and Norman ([CN], cf. (6.1.5)) and to certain orthogonal local models corresponding to pairs of lattices. The relevant symmetric space is the one corre- sponding to the symplectic, resp. the orthogonal group. In the interest of brevity we omit discussing these examples. / / / G. Pappas, M. Rapoport, and B. Smithling 77 8.4. Wonderful completions and resolutions In what follows, we will first explain in rough outline some of the constructions of [Fa1] in the case of the local model M = M loc for GLn considered in §8.2 above. Faltings' approach also applies to cases of other groups, see Remark 8.4.7. Then we sketch the method of [Fa2] to include more general parahoric level structures (defined by more than two lattices). Similar constructions also appear in the work of Genestier [Ge1, Ge3]. The main goal of all these papers is to produce resolutions of a local model M loc which are regular and have as special fiber a divisor with normal crossings. More precisely, this goal may be formulated as follows. Recall that, in all cases that it is successfully constructed, the local model G,{µ},L supports an action of the parahoric group scheme G ⊗O OE. M loc Definition 8.4.1. An equivariant modification of M loc G,{µ},L consists of a proper OE-scheme that supports an action of G⊗O OE and a G⊗O OE-equivariant proper birational morphism π : N → M loc G,{µ},L, which is an isomorphism on the generic fibers. We can obtain such modifications by blowing-up G ⊗O OE-invariant sub- schemes of M loc G,{µ},L which are supported in the special fiber. It is reasonable to conjecture that there always exists an equivariant modi- G,{µ},L such that N is regular and has as special fiber a divisor fication N → M loc with (possibly non-reduced) normal crossings [P1]. Let us return to the local model M = M loc for GLn and µ = (1(r), 0(n−r)) given by choosing bases for Λ0/F0 and Λ1/F1, considered in §8.2. Consider, as in §6, the GLr × GLr-torsor (cid:102)M −→ M (cid:102)M (S) =(cid:8)(F0,F1) ∈ M (S), α0 : Λ0,S/F0 (cid:9) . The scheme (cid:102)M affords a morphism q : (cid:102)M → Y , where Y is the O-scheme of Y =(cid:8) (A, B) ∈ Matr×r × Matr×r (cid:12)(cid:12) A · B = B · A = π · I(cid:9), (8.4.2) comp. (6.1.1). The morphism q is given by sending (F0,F1; α0, α1) to the pair of matrices that describe the maps Λ0,S/F0 → Λ1,S/F1, resp. Λ1,S/F1 → Λ0,S/F0 induced by Λ0⊗O OS → Λ1⊗O OS, resp. Λ1⊗O OS → Λ0⊗O OS. It is not hard to Y supports an action of GLr × GLr given by see that the morphism q : (cid:102)M → Y is smooth, comp. [PR1, Th. 4.2]. The scheme S, α1 : Λ1,S/F1 matrices ∼−→ Or ∼−→ Or S (g1, g2) · (A, B) = (g1Ag−1 2 , g2Bg−1 1 ) , such that q is GLr × GLr-equivariant. Hence we obtain a smooth morphism of algebraic stacks (8.4.3) M −→ [(GLr × GLr)\Y ]. 78 Local models of Shimura varieties, I. Now consider the following variant of Y , Y1 =(cid:8) (A, B, a)(cid:12)(cid:12) A · B = B · A = a · Id(cid:9) ⊂ Matr×r × Matr×r × A1 , regarded as a k-variety with GLr × GLr-action. Following [Fa1, p. 194], (see also [Ge3, §2.2]), we can now see that the open subset Y1 − {a = 0} is a G2 m-bundle over P GLr (cid:39) (P GLr × P GLr)/P GLr and actually Y1 can be viewed as a double affine cone over the projective variety X in P(Matr×r) × P(Matr×r) given by the closure of the image of the map A (cid:55)→ (A, Aadj), where Aadj denotes the adjugate matrix of A. As in loc. cit. we see that the total space of the corresponding affine bundle obtained by pulling back by P GLr → X provides a resolution Y1 → Y1. By intersecting Y1 with a − π = 0, we obtain a resolution (cid:101)Y → Y . Explicitly, Y can be obtained by successively blowing up ideals obtained from minors of A and B. This can now be used to obtain that in this very special case: Theorem 8.4.4. There exists an equivariant modification π : N → M loc such that (cid:3) N is regular and has as special fiber a divisor with simple normal crossings. Remark 8.4.5. One can attempt to generalize this method of resolution, as well as the method of [P2], to general parahoric level subgroups. Let us start with a lattice chain Λ0 ⊂ Λ1 ⊂ ··· ⊂ Λs−1 ⊂ t−1Λ0 in V = F n and consider the corresponding local model for G = GLn and µ = (1(r), 0(n−r)). The natural replacement for P GLr = (P GLr × P GLr)/P GLr to accomodate more than two lattices would be a suitable completion X of the quo- tient X = (P GLr)s/P GLr. Unfortunately, there is no easy "wonderful" choice for such a completion. Indeed, the equivariant compactifications of such quo- tients have a very complicated theory, which was developed by Lafforgue [Laf]. To transpose the theory of [P2], one can then consider the corresponding closures in the completion of a product of parabolics and attempt to obtain local models as pull-backs of the corresponding universal families. The details of such a general construction have not been worked out. On the other hand, as far as construct- ing resolutions of local models in the style of Theorem 8.4.4 are concerned, an approach using completions of (P GLr)s/P GLr is given in [Fa2], see also [Ge3]. Remark 8.4.6. A somewhat different but related point of view which also con- nects with Lafforgue's completions is explained in [Fa2]. Using it, Faltings con- structs a resolution of the local models of Remark 8.4.5 when r = 2 (when r = 1 the local models themselves have the desired properties, cf. the second part of The- orem 2.1.1). We will not attempt to fully reproduce his (ingenious!) construction in this survey, but here is an idea. Faltings starts with the following observation: If R is a discrete valuation ring, then an R-valued point F(R) := {Fi(R)}i of the local model M loc gives a G. Pappas, M. Rapoport, and B. Smithling 79 sequence of free R-modules F0 ⊂ F1 ⊂ ··· ⊂ Fs−1 of rank r. This sequence can be viewed as a lattice arrangement in a vector space of dimension r. Note that the local model M loc and all its proper modifications N that share the same generic fiber, also share the same set of R-valued points for a discrete valuation ring R. We can now view the search for a suitable bi- rational modification N → M loc as a search for a suitable parameter space of lattice arrangements as above. Observe that to any such lattice arrangement we can associate its corresponding Deligne scheme D = D(F(R)) over Spec R (which is a type of local model on its own) cf. [Fa2, §5], [Mu]. A suitable blow-down of the Deligne scheme gives a "minimal" model Dmin with toroidal singularities. Now parameters for a space of lattice arrangements can be obtained by looking at moduli of these minimal Deligne schemes. More specifically, Faltings constructs a universal family of lattice arrangements that supports a universal minimal Deligne scheme. The base of this family is a projective equivariant embedding of the ho- mogeneous space (P GLr)s/P GLr. It turns out that this embedding is of the kind considered by Lafforgue. As explained in Remark 8.4.5 above, this can then be used to obtain modifications of the local models. For example, when r = 2, the Deligne scheme is a projective flat curve over Spec R with generic fiber P1 and spe- cial fiber a chain of P1's intersecting transversely (our first local model for Γ0(p) in Example 2.1.2 is such an example of a Deligne scheme). The minimal model Dmin now gives a semi-stable curve over Spec R and we can parametrize the lattice arrangement by a corresponding point of the moduli space of genus 0 semi-stable marked curves. In this case, Faltings' construction produces a smooth compact- ification of (P GL2)s/P GL2 and hence also a regular equivariant modification of the corresponding local model for µ = (1(2), 0(n−2)) and the periodic lattice chain with s members. See also [KT] for some more recent developments in this circle of ideas. Remark 8.4.7. In [Fa1], Faltings gives a construction of resolutions of local mod- els in some cases related to other groups. This is done by working with explicit schemes of matrices that give affine charts for M loc, and relating those to won- derful completions. This then leads to resolutions for these affine charts. One can then obtain equivariant resolutions of the corresponding local models as in Theorem 8.4.4 (note however that the special fibers of these resolutions are not always reduced). We conclude this section with a list of some matrix equations which are among those investigated by Faltings [Fa1, §4]. Before doing so, we make two remarks. First of all, even though some of the matrix equations that Faltings writes down are among the ones discussed in §6 (e.g., Z in (6.1.5) appears in the middle of p. 194 in [Fa1], and N of (6.1.11), resp. (6.1.12) occurs in the middle of p. 195 in 80 Local models of Shimura varieties, I. [Fa1]), and therefore are closely related to local models, this is less clear for others. In fact, his list arises from embedding symmetric spaces in projective spaces via homogeneous line bundles, and considering the singularities which occur in their closures -- so there is a priori no connection to local models. Secondly, Faltings is less interested in questions of flatness, but rather allows himself to pass to the flat closure of the generic fiber, i.e., to the affine variety with coordinate ring obtained by dividing out by π-torsion, and then tries to construct resolutions of those. Again we fix O with uniformizer π. One matrix equation considered in [Fa1] is (8.4.8) Z = { A ∈ Matn×n AAad = AadA = π · I }, where Aad is the adjoint of A with respect to a symmetric or a symplectic form. Faltings proves that, when n is even, the flat closure of Z ⊗O F inside Z is Cohen- Macaulay with rational singularities. When n is odd, the flat closure of Z ⊗O F inside Z is not Cohen-Macaulay, but its normalization is, with rational singulari- ties. Furthermore, he gives equivariant resolutions of these flat O-schemes which have a normal crossings divisor as their special fibers, and computes the multiplic- ities of the irreducible components. In a similar vein, Faltings also analyzes the intersection of Z with the locus where A = Aad, i.e., (8.4.9) { A ∈ Matn×n A = Aad, A2 = π · I }. When Aad is the adjoint of A for a symmetric form, this matrix equation relates to local models for the ramified unitary group, and the maximal parahoric subgroup fixing a selfdual lattice, comp. (6.1.11). Similarly, he also considers the matrix equation (8.4.10) We refer to [Fa1, §4] for further matrix equations, and results concerning them. { A ∈ Matn×n A = −Aad, A2 = π · I }. In [Fa2, Th. 13], Faltings constructs resolutions of local models in the case of the symplectic group of genus 2, for more general parahoric subgroups (see also [Ge3]). References [A] [BC] [BD] K. Arzdorf: On local models with special parahoric level structure, Michigan Math. J. 58 (2009), 683 -- 710. 2, 28, 30 P. Bala, R.W. Carter: Classes of unipotent elements in simple algebraic groups. I. Math. Proc. Cambridge Philos. Soc. 79 (1976), no. 3, 401 -- 425. 62 A. Beilinson, V. Drinfeld: Quantization of Hitchin's integrable system and Hecke eigensheaves. Preprint. 11 [BL] [Bor] [B] [Br1] [Br2] [BrTh] [BrP] [BK] [BTI] [BTII] [CR] [CN] [deC1] [deC] G. Pappas, M. Rapoport, and B. Smithling 81 A. Beauville, Y. Laszlo: Conformal blocks and generalized theta func- tions. Comm. Math. Phys. 164 (1994), no. 2, 385 -- 419. 11 M. Borovoi: Abelian Galois cohomology of reductive groups, Mem. Amer. Math. Soc. 132 (1998), no. 626. 39 N. Bourbaki: Lie groups and Lie algebras. Chapters 4 -- 6, Elements of Mathematics, Springer-Verlag, Berlin, 2002. Translated from the 1968 French original by Andrew Pressley. 7, 48, 51, 57, 60 M. Brion: Group completions via Hilbert schemes. J. Algebraic Geom. 12 (2003), no. 4, 605-626. 72, 73, 75 M. Brion: On orbit closures of spherical subgroups in flag varieties. Comment. Math. Helv. 76 (2001), no. 2, 263-299. 71, 73 M. Brion, J. F. Thomsen. F -regularity of large Schubert varieties. Amer. J. Math. 128 (2006), no. 4, 949 -- 962. 71 M. Brion, P. Polo: Large Schubert varieties. Represent. Theory 4 (2000), 97-126. 71, 73 M. Brion, S. Kumar: Frobenius splitting methods in geometry and rep- resentation theory. Progress in Mathematics, 231. Birkhauser Boston, Inc., Boston, MA, 2005. x+250 pp. 40, 41 F. Bruhat, J. Tits: Groupes r´eductifs sur un corps local. Inst. Hautes ´Etudes Sci. Publ. Math. 41 (1972), 5 -- 251. 48, 49 F. Bruhat, J. Tits: Groupes r´eductifs sur un corps local. II. Sch´emas en groupes. Existence d'une donn´ee radicielle valu´ee. Inst. Hautes ´Etudes Sci. Publ. Math. 60 (1984), 197 -- 376. 4, 14, 36, 49, 51, 58 P. Caldero, M. Reineke: On the quiver Grassmannian in the acyclic case. J. Pure Appl. Algebra 212 (2008), no. 11, 2369 -- 2380. 68, 69 C. Chai, P. Norman: Singularities of the Γ0(p)-level structure. J. Alg. Geom. 1 (1992), 251 -- 278. 2, 18, 66, 67, 76 C. de Concini: Symplectic standard tableaux, Adv. in Math. 34 (1979), 1 -- 27. 66 C. De Concini: Equivariant embeddings of homogeneous spaces. Pro- ceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif., 1986), 369 -- 377, Amer. Math. Soc., Providence, RI, 1987. 71 [deC-S] [dCGMP] C. De Concini, M. Goresky, R. MacPherson, C. Procesi: On the ge- ometry of quadrics and their degenerations. Comment. Math. Helv. 63 (1988), no. 3, 337 -- 413. 71 C. De Concini, T. A. Springer: Compactification of symmetric varieties. Dedicated to the memory of Claude Chevalley. Transform. Groups 4 (1999), no. 2-3, 273 -- 300. 71 C. De Concini, C. Procesi: Complete symmetric varieties. Invariant the- ory (Montecatini, 1982), 1 -- 44, Lecture Notes in Math., 996, Springer, [deC-P] 82 Local models of Shimura varieties, I. Berlin, 1983. 71 [deC-P2] C. De Concini, C. Procesi: Symmetric functions, conjugacy classes and [D1] [D2] [DP] [Fa1] [Fa2] [Fa3] [Fa4] [Ga] [Ge1] [Ge2] [Ge3] [Go1] [Go2] [Go3] the flag variety. Invent. Math. 64 (1981), no. 2, 203 -- 219. 62 P. Deligne: Travaux de Shimura. (French) S´eminaire Bourbaki, 23`eme ann´ee (1970/71), Exp. No. 389, pp. 123 -- 165. Lecture Notes in Math., Vol. 244, Springer, Berlin, 1971. 5, 6 P. Deligne: Vari´et´es de Shimura: interpr´etation modulaire, et tech- niques de construction de mod`eles canoniques. (French) Automorphic forms, representations and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 2, pp. 247 -- 289, Proc. Sympos. Pure Math., XXXIII, Amer. Math. Soc., Providence, R.I., 1979. 9 P. Deligne, G. Pappas: Singularit´es des espaces de modules de Hilbert, en les caract´eristiques divisant le discriminant. Compositio Math. 90 (1994), 59 -- 79. 2, 66, 67 G. Faltings: Explicit resolution of local singularities of moduli-spaces. J. Reine Angew. Math. 483 (1997), 183 -- 196. 2, 66, 70, 71, 77, 78, 79, 80 G. Faltings: Toroidal resolutions for some matrix singularities. Moduli of abelian varieties (Texel Island, 1999), 157 -- 184, Progr. Math., 195, Birkhauser, Basel, 2001. 2, 70, 77, 78, 79, 80 G. Faltings: Moduli-stacks for bundles on semistable curves. Math. Ann. 304 (1996), no. 3, 489 -- 515. 2 G. Faltings: Algebraic loop groups and moduli spaces of bundles. J. Eur. Math. Soc. (JEMS) 5 (2003), no. 1, 41 -- 68. 34, 39, 41 D. Gaitsgory: Construction of central elements in the affine Hecke al- gebra via nearby cycles. Invent. Math. 144 (2001), 253 -- 280. 2, 11, 12, 14 A. Genestier: Un mod`ele semi-stable de la vari´et´e de Siegel de genre 3 avec structures de niveau de type Γ0(p). Compositio Math. 123 (2000), 303 -- 328. 77 A. Genestier. Letter to M. Rapoport. (Jan. 22, 2001) 2, 7, 18 A. Genestier: A toroidal resolution for the bad reduction of some Shimura varieties. math.AG/9912054 77, 78, 80 U. Gortz: On the flatness of models of certain Shimura varieties of PEL-type. Math. Ann. 321 (2001), 689 -- 727. 2, 7, 15, 16, 22, 33, 37, 60, 65, 67, 73 U. Gortz: On the flatness of local models for the symplectic group. Adv. Math. 176 (2003), 89 -- 115. 2, 7, 17, 24, 60, 66, 67 U. Gortz: Computing the alternating trace of Frobenius on the sheaves of nearby cycles on local models for GL4 and GL5. J. Algebra 278 G. Pappas, M. Rapoport, and B. Smithling 83 (2004), 148 -- 172. 2 U. Gortz: Topological flatness of local models in the ramified case. Math. Z. 250 (2005), 775 -- 790. 2, 22, 23, 24, 58, 60 U. Gortz: Matrixgleichungen und Familien von abelschen Varietaten in positiver Charakteristik. Mitt. der DMV 17 (2009), no. 1, 23 -- 30. 66 M. Greenberg: Schemata over local rings. II, Ann. of Math. (2) 78 (1963), 256 -- 266. 40 T. Haines: Test functions for Shimura varieties: the Drinfeld case. Duke Math. J. 106 (2001), no. 1, 19 -- 40. 2 T. Haines: The combinatorics of Bernstein functions. Trans. Amer. Math. Soc. 353 (2001), no. 3, 1251 -- 1278 2 T. Haines, A. Pettet: Formulae relating the Bernstein and Iwahori- Matsumoto presentations of an affine Hecke algebra. J. Algebra 252 (2002), no. 1, 127 -- 149. 2 T. Haines, Ngo, B. C.: Nearby cycles for local models of some Shimura varieties. Compositio Math. 133 (2002), 117 -- 150. 2 T. Haines, Ngo, B. C.: Alcoves associated to special fibers of local models. Amer. J. Math. 124 (2002), no. 6, 1125 -- 1152. 2, 57, 58 T. Haines, M. Rapoport: On parahoric subgroups (Appendix to [PR3]). Adv. Math. 219 (2008), 188 -- 198. 45, 47, 49, 58 X. He, T. Lam: Projected Richardson varieties and affine Schubert varieties. arXiv:1106.2586v1 [math.AG]. 54 J. Heinloth: Uniformization of G-bundles. Math. Ann. 347 (2010), no. 3, 499 -- 528. 14 J. de Jong: The moduli spaces of principally polarized abelian varieties with Γ0(p)-level structure. J. Alg. Geom. 2 (1993), 667 -- 688. 2 V. Kac: Infinite-dimensional Lie algebras. Third edition. Cambridge University Press, Cambridge (1990). 35, 41 M. Kapranov: Chow quotients of Grassmannians. I. I. M. Gelfand Sem- inar, 29 -- 110, Adv. Soviet Math., 16, Part 2, Amer. Math. Soc., Provi- dence, RI, 1993. 72 M. Kisin: Moduli of finite flat group schemes, and modularity. Annals of Math. Vol. 170, no. 3, (2009), 1085 -- 1180. 2, 64 B. Kostant, S. Rallis: Orbits and representations associated with sym- metric spaces. Amer. J. Math. 93 (1971), 753 -- 809. 65 R. Kottwitz: Points on some Shimura varieties over finite fields. J. Amer. Math. Soc. 5 (1992), 373 -- 444. 21 R. Kottwitz: Isocrystals with additional structure, II. Compos. Math. 109 (1997) 255 -- 339. 39, 44 R. Kottwitz, M. Rapoport: Minuscule Alcoves for Gln and GSp2n. Manuscripta Math. 102 (2000), 403 -- 428. 2, 56, 57, 58 [Go4] [Go5] [Gr] [H1] [H2] [HP] [HN1] [HN2] [HR] [HL] [He] [J] [Kac] [Ka] [Ki] [KosR] [Ko1] [Ko2] [KR] 84 [Kr] [KP] [KT] [Ku1] [Ku2] [KuR] [Laf] [Lak] [Lam] [Lan] [Lu] [Li] [Ma] Local models of Shimura varieties, I. N. Kramer: Local models for ramified unitary groups. Abh. Math. Sem. Univ. Hamburg 73 (2003), 67 -- 80. 2, 28 H. Kraft, C. Procesi: Closures of conjugacy classes of matrices are normal. Invent. Math. 53 (1979), no. 3, 227 -- 247. 62 S. Keel, J. Tevelev: Geometry of Chow quotients of Grassmannians. Duke Math. J. 134 (2006), no. 2, 259 -- 311. 79 S. Kumar: Demazure character formula in arbitrary Kac-Moody set- ting. Invent. Math. 89 (1987), 395 -- 423. 41 S. Kumar: Kac-Moody groups, their flag varieties and representation theory. Progress in Mathematics 204, Birkhauser Boston Inc., Boston, MA (2002). 35, 41 S. Kudla, M. Rapoport: Special cycles on unitary Shimura varieties, II: global theory. arXiv:0912.3758 10 L. Lafforgue: Chirurgie des grassmanniennes. CRM Monograph Series, 19. American Mathematical Society, Providence, RI, 2003. xx+170 pp. 78 D. Laksov: Completed quadrics and linear maps. Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), 371 -- 387, Proc. Sympos. Pure Math., 46, Part 2, Amer. Math. Soc., Providence, RI, 1987. 71 T. Y. Lam: Introduction to quadratic forms over fields. Graduate Stud- ies in Mathematics, 67. American Mathematical Society, Providence, RI, 2005. xxii+550 pp. 32 E. Landvogt: A compactification of the Bruhat-Tits building. Lecture Notes in Mathematics, 1619. Springer-Verlag, Berlin, 1996. viii+152 pp. 48 G. Lusztig: Green polynomials and singularities of unipotent classes, Adv. in Math. 42 (1981), 169 -- 178. 64 P. Littelmann: Contracting modules and standard monomial theory for symmetrizable Kac-Moody algebras. J. Amer. Math. Soc. 11 (1998), 551 -- 567. 41 O. Mathieu: Formules de caract`eres pour les alg`ebres de Kac-Moody g´en´erales. Ast´erisque 159-160 (1988). 41 [MT] [MvdK] V. B. Mehta, W. van der Kallen: A simultaneous Frobenius splitting for closures of conjugacy classes of nilpotent matrices. Compositio Math. 84 (1992), no. 2, 211 -- 221. 64 V. Mehta, V. Trivedi: The variety of circular complexes and F- split- ting, Invent. math. 137 (1999), 449-460. 65 J. Milne: Introduction to Shimura varieties. Harmonic analysis, the trace formula, and Shimura varieties, 265 -- 378, Clay Math. Proc., 4, Amer. Math. Soc., Providence, RI, 2005. 9 [M] [M2] [MVy] [Mu] [Oh] [P1] [P2] [PR1] [PR2] [PR3] [PR4] [PR5] [PZ] [PrRa] [R] [RZ] [Ri1] G. Pappas, M. Rapoport, and B. Smithling 85 J. Milne: Shimura varieties and motives. Motives (Seattle, WA, 1991), 447 -- 523, Proc. Sympos. Pure Math., 55, Part 2, Amer. Math. Soc., Providence, RI, 1994. 9 I. Mirkovi´c, M. Vybornov: On quiver varieties and affine Grassmanni- ans of type A. C. R. Math. Acad. Sci. Paris 336 (2003), no. 3, 207 -- 212. 64 G. A. Mustafin: Non-Archimedean uniformization. (Russian) Mat. Sb. (N.S.) 105(147) (1978), no. 2, 207 -- 237, 287. 79 T. Ohta : The singularities of the closures of nilpotent orbits in certain symmetric pairs. Tohoku Math. J. (2) 38 (1986), 441 -- 468. 65 G. Pappas: On the arithmetic moduli schemes of PEL Shimura vari- eties, J. Alg. Geom. 9 (2000), 577 -- 605. 2, 7, 26, 27, 28, 29, 32, 77 G. Pappas: Local models and wonderful completions. Notes. 70, 71, 76, 78 G. Pappas, M. Rapoport: Local models in the ramified case. I. The EL-case. J. Alg. Geom. 12 (2003), 107 -- 145. 2, 7, 22, 39, 62, 63, 64, 66, 67, 77 G. Pappas, M. Rapoport: Local models in the ramified case. II. Split- ting models. Duke Math. Journal 127 (2005), 193 -- 250. 2, 7, 22, 23, 24, 38, 44, 60, 67, 69, 70 G. Pappas, M. Rapoport: Twisted loop groups and their affine flag varieties. With an appendix by T. Haines and Rapoport. Advances in Math. 219 (2008), no. 1, 118 -- 198. 3, 14, 25, 34, 39, 40, 41, 45, 52, 83 G. Pappas, M. Rapoport: Local models in the ramified case. III. Uni- tary groups. J. Inst. Math. Jussieu 8 (2009), no. 3, 507 -- 564. 2, 6, 7, 14, 19, 20, 23, 24, 25, 27, 28, 29, 30, 31, 33, 39, 44, 45, 60, 65, 68 G. Pappas, M. Rapoport: Some questions about G-bundles on curves. in Algebraic and Arithmetic Structure of Moduli Spaces, Advanced Studies in Pure Mathematics 58, (2010), 159-171. 14 G. Pappas, X. Zhu: Local models and a conjecture of Kottwitz. In preparation. 14, 55 G. Prasad, M. S. Raghunathan: Topological central extensions of semisimple groups over local fields. Ann. of Math. (2) 119 (1984), no. 1, 143 -- 201. 50 M. Rapoport: A guide to the reduction modulo p of Shimura varieties. Ast´erisque 298 (2005), 271 -- 318. 4, 44, 56, 58, 60, 62 M. Rapoport, Th. Zink: Period spaces for p -- divisible groups. Ann. of Math. Studies 141, Princeton University Press (1996). 2, 6, 21, 36 T. Richarz: Local models and Schubert varieties in twisted affine Grass- mannians. Diploma thesis Bonn, 2010. 29 86 [Ri2] [Sm1] [Sm2] [Sm3] [Sm4] [Sm5] [St] [Th] [T] [W] [Z] [Zh] Local models of Shimura varieties, I. T. Richarz: Schubert varieties in twisted affine flag varieties and local models. arXiv:1011.5416v1 [math.AG]. 2, 28, 44, 47, 48, 51, 52 B. Smithling: Topological flatness of orthogonal local models in the split, even case. I. Math. Ann. 350 (2011), 381 -- 416. 2, 19, 39, 44, 46, 58, 61 B. Smithling: Admissibility and permissibility for minuscule cocharac- ters in orthogonal groups. To appear in Manuscripta Math. 2, 58 B. Smithling: Topological flatness of local models for ramified uni- tary groups. I. The odd dimensional case. Adv. Math. 226 (2011), 3160 -- 3190. 2, 30, 39, 44, 60, 62 B. Smithling: Topological flatness of local models for ramified unitary groups. II. The even dimensional case. In preparation. 2, 30, 57, 60, 61, 62 B. Smithling: Orthogonal Shimura varieties of PEL type and local models. In preparation. 2, 58 E. Strickland: On the Conormal Bundle of the Determinantal Variety, J. Alg. 75 (1982), 523-537. 65 M. Thaddeus. Complete collineations revisited. Math. Ann. 315 (1999), no. 3, 469 -- 495. 72 J. Tits: Reductive groups over local fields. Automorphic forms, repre- sentations and L-functions. (Proc. Sympos. Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1, pp. 29 -- 69, Proc. Sympos. Pure Math., XXXIII, Amer. Math. Soc., Providence, R.I. (1979). 4, 7, 48, 50 J. Weyman: Two results on equations of nilpotent orbits. J. Algebraic Geom. 11 (2002), no. 4, 791 -- 800. 64, 67 A. Zelevinsky: Quiver Grassmannians and their Euler characteristics: Oberwolfach talk, May 2010. arXiv:1006.0936v1 [math.AG]. 68 X. Zhu: On the coherence conjecture of Pappas and Rapoport. arXiv:1012.5979v2 [math.AG]. 14 Department of Mathematics, Michigan State University, East Lansing, MI 48824- 1027, USA E-mail address: [email protected] Mathematisches Institut der Universitat Bonn, Endenicher Allee 60, 53115 Bonn, Germany E-mail address: [email protected] University of Toronto, Department of Mathematics, 40 St. George St., Toronto, ON M5S 2E4, Canada E-mail address: [email protected]
1610.04558	1	1610	2016-10-14T17:53:57	Syzygies of Quadratic Veronese Embedding	[ "math.AG", "math.RT" ]	In this work we explicitly calculate syzygies of quadratic Veronese embedding $\mathbb{P}(V)\subset\mathbb{P}(\operatorname{Sym}^2V)$ as representations of the group $\operatorname{GL}(V)$. Also resolutions of the sheaves $\mathscr{O}_{\mathbb{P}(V)}(i)$ are constructed in the category $D(\mathbb{P}(\operatorname{Sym}^2V))$.	math.AG	math	0 SYZYGIES OF QUADRATIC VERONESE EMBEDDING I. V. NETAY Abstract. In this work we explicitly calculate syzygies of quadratic Veronese embedding P(V ) ⊂ P(Sym2 V ) as representations of the group GL(V ). Also resolutions of the sheaves OP(V )(i) are constructed in the category D(P(Sym2 V )). 1. Introduction Let V be a vector space with coordinates (x0, . . . , xn) over a field k, P(V ) be its projectivization with homogeneous coordinates (x0 : . . . : xn). We assume that the ground field k has zero characteristic and is algebraically closed. Consider the space M of matrices of rank 1 in the space Sym2 V of symmetrical matrices with coordinates (x2 n), where 0 6 i 6 j 6 n. Clearly, it is a cone. It is well known that P(M ) ≃ P(V ) ⊂ P(Sym2 V ). This embedding is called quadratic Veronese embedding. It is also well known that the minimal set of equations defining this variety is a set of all (2 × 2)-minors of the matrix (see [1, Lecture 2, example 2.6]). Denote by S the algebra of polynomials k[Sym2 V ] on the projective space P(Sym2 V ) and by A the projective coordinate algebra of the image of the Veronese embedding P(M ). The algebra A is a S-module. 0 : . . . : xixj : . . . : x2 Given a projective variety X ⊂ P(V ), denote by A(X) it projective coordinate algebra and by I(X) the ideal of functions in S vanishing on X. We get the f0−→ A → 0. Let us choose a set of homogeneous exact triple 0 → I(X) → S generators g1, . . . , gm1 of the ideal I(X). We get the surjection f1 : S⊕m1 → I(X) defined by the row (g1, . . . , gm1) such that the following sequence of S-modules is exact: S⊕m1 f1−→ S f0−→ A → 0. The map f1 can have a kernel. Choosing a set of homogeneous generators of ker f1, we extend the sequence to the left: S⊕m2 f2−→ S⊕m1 f1−→ S f0−→ A → 0. Iterating this operation, we obtain finally an exact sequence of graded S-modules: · · · → F2 → F1 → S → A → 0, where Fp = Rp ⊗S, and Rp are graded vector spaces over the field k. We put Rp,q = (Rp)q. Such exact sequence of S-modules is called a free resolution of the module A. Note that there is no way to choose generators in ker fp canonically. Nevertheless under some minimality conditions the spaces Rp,q are defined canonically. Since on each step of the construction we take a homogeneous basis, the matrices of maps fp consist of homogeneous elements. A resolution F• = {· · · → F2 → F1 → F0 = S} 0The author is partially supported by AG Laboratory HSE, RF government grant, ag. 11.G34.31.0023, grant MK-2858.2014.1, RFBR-2969.2014.1, by a grant from Dmitry Zimin's "Dy- nasty" foundation, RFBR 14-01-31398 1 2 I. V. NETAY is called minimal, if we tame minimal possible number of generators on each step. Obviously, if the matrix defining the map fp contains an element of degree zero, then the resolution F• is not minimal. Conversely, if all matrix elements have positive degrees, then the resolution is minimal. Definition 1. If a resolution F• is minimal, then the space Rp,q is called syzygy space of order p and degree q of a graded S-module A. From Hilbert syzygy theorem (see [2]) each module over the ring of polynomials in n variables over a field has a free resolution of length at most n. In this work we deal with only modules over rings of polynomials. Computation of syzygies is a very complicated problem in general case. It is solved algorithmically, although the algorithm requires a huge time amount. Here we consider some modules arising from geometrical problems. Syzygy spaces of a projective embedding of an algebraic variety are very im- portant invariant of the embedding. Many useful information on this topic can be found in the book [3]. Let us consider the following natural situation: a group G ⊂ GL(V ) acts linearly on the projective space P(V ) preserving an algebraic variety X ⊂ P(V ). Then the natural action of the group G on projective coordinate algebra A and on all the syzygy spaces appears (see §2.1). This allows us to apply representation theory to computation of the spaces Rp,q. It is the most simple and convenient if the category of representations of the group is semisimple. We consider projective homogeneous spaces of reductive algebraic groups. In this work we deal mainly with the quadratic Veronese embedding. The following questions about syzygies are interesting: • which of the spaces Rp,q are zero? • how to construct the spaces Rp,q? • how to construct the maps fp? In this work we consider only the first two questions. Actually, the first question is interesting and has some open problems in particular for homogeneous spaces. Definition 2. An embedding X ⊂ P(V ) satisfies Np-property, if • dim R0,q =(1, 0 if q = 0; otherwise; • Rk,q = 0, if q 6= k + 1 for 1 6 k 6 p. The famous conjecture about Np-property for Veronese embeddings had been stated in [4]. In [4] it is proved for the case of quadratic Veronese embedding and for the cubic embedding of projective plane. It is also proved that the Np-property does not hold in all the cases, where it should not hold according to the conjecture. Recently in [5] the conjecture was proved for cubic Veronese embedding of projective space of any dimension. The question about vanishing of syzygy spaces Rp,q of Veronese embeddings is still far from a complete solution. Moreover, the problem of description of these spaces seems to be unsolvable in a reasonable form. Nevertheless, some cases can have some reasonable answer. Consider a projective rational normal curve in the projective space P(V ) not lying in any hyperplane. It is well known that it is a projective line P1 embedded into P(V ) by the Veronese embedding of degree dim V . All the syzygy spaces of SYZYGIES OF QUADRATIC VERONESE EMBEDDING 3 this embedding are explicitly calculated (see for example [3]) as one of the simplest non-trivial examples of calculation of syzygies. The syzygy spaces of projective embeddings of projective plane are calculated for only in few cases. Now let us restrict the degree of embedding. The syzygy spaces of quadratic Veronese embedding are known from [6]. In the case of cubic Veronese embedding the problem is open. Here we consider quadratic Veronese embedding and S-modulesLk>0 Sym2k+a V ∗ and calculate their syzygies for all a. Geometrically this construction leads us to construction of minimal resolution of invertible sheaves on the image of quadratic Veronese embedding (see §2.3). Let us introduce some notation. We will draw Young diagrams to the right and below from the box (0, 0) using non-negative numbers as the coordinates of boxes. We call a hook the subset of boxes of Young diagram λ composed of a box (vertex), all the boxes below it, and all the boxes to the right of it in diagram λ. We call a hook main, if its vertex lies on the diagonal, i. e. has the coordinates (k, k) for some k. Denote the length of the diagonal (number of boxes of the form (k, k)) in λ by l(λ). Definition 3. Let λ be a Young diagram. It is a union of its main hooks. Denote their width-height pairs by (a1, b1), . . . , (al, bl), where l = l(λ). We will denote the diagram λ by (a1, . . . , akbk, . . . , b1). Let λ be a dominant weight a reductive group G. Denote by Vλ its irreducible representation of highest weight λ. For more details about representations with highest weight see [7]. The function C puts any Young diagram λ into a correspondence with {1, . . . , l(λ)} and is defined in 4. Theorem 1. Let X = P(V ) ⊂ P(Sym2 V ) be the quadratic Veronese embedding. Then there exists a resolution . . . Ra 1,k+1 ⊗ O(−k − 1) Ra 0,k ⊗ O(−k) / OX (k) / 0, / Lk>0 Remark 1. In the case a < 0 such a resolution can be obtained from the resolution for a = 0 by the grading shift by −a. Therefore, it is consistent only for a > 0. In section 3 we prove this theorem and describe the structure of the corresponding Koszul complex as a complex of representation of GL(V ). After we classify all the cases that admit analogous computations of syzygy spaces of projectivizations of highest wight vector orbits (see §4) and that admit an analogous description of Koszul complex structure. Possibility of such computations is based on the following two properties of a dominant weight λ of reductive group G. / Lk>0 where n = dim V and (1.1) Ra p,p+q = Lω=ω′ wt(ω)=q l(ω)=2q−a L ω=(a1 ,...,ak bk ,...,b1) bi 6ai6bi−1 +1, i=1,...,k wt(ω)=2p+a  where s = 1 2(cid:16)Pk i=1 ai − bi − 1 − a(cid:17). V ∗ ω , q > 0 or p = q = 0, (V ∗ ω )⊕(C(ω)−1 s ), p > 0, q = 0, s ∈ Z, / / / / 4 I. V. NETAY Property 1. For any k the representation ΛkVλ of G has no multiple subrepresen- tations. We will denote the property by SMF(from ¡¡skew multiplicity free¿¿ follow- ing [8]). Property 2. For any dominant weight µ and any n the representation Vµ ⊗ Vnλ has no multiple subrepresentations. We will denote this property by STMF. As it will be shown in the section 4, these cases consist of two infinite series: Segre embedding (see results in [9], where the computation is proceeded in an analogous way), quadratic Veronese embedding and a finite set of cases listed in Proposition 5. See the classification in section 4. Consider the following property to simplify the classification. Property 3. For any dominant weight µ the representation Vµ ⊗Vλ has no multiple subrepresentation. This property is a particular case of STMF. We will denote it by TMF. Con- versely, the property 3 for any multiplicity of the weight λ coincides with the prop- erty 2 for the weight λ. In section §4 we classify all the representations of highest weights that are SMFand STMF. Some analogous classifications form the core of paper [8], where representations with some property similar to SMFare classified. In this work the cases are classi- fied, where exterior algebra of a representation of a reductive group has no multiple subrepresentations. We need some more weak property: each exterior power of a representation has no multiple subrepresentations. These properties coincide for so called saturated1 representations (for example, the tautological representation of GL(V ) is saturated and the tautological representation of SL(V ) is not). In [8] more general representations are considered, not only irreducible. For convenience of the reader we give the classification concerning only interesting for us cases. This allows to make it more simple. To proof to we need to look over a big number of particular cases of representations. We use computer algebra systems LiE and SCHUR. After if the Appendix A some examples are given, in the Appendix B questions related to weighted projective spaces are considered. The author is grateful for S. O. Gorchinskiy, E. B. Vinberg for useful discussions and also for V. V. Ostrik for the idea of the proof of Lemma 4.2 and for a useful reference. This paper is to be published in Matematicheskii Sbornik. 2. Preliminaries 2.1. Koszul complex. Let X ⊂ P(W ) be a projective variety the in projective space, Aq = H0(X, O(q)) be q-th homogeneous component in the projective coor- dinate algebra of the projective variety X. Let ı : ΛpW ∗ → Λp−1W ∗ ⊗ W ∗ be the 1Saturatedness is a condition on the action of the central torus of the reductive group; in the framework of this paper we can assume it to be satisfied. See the definition (not only for irreducible representations) in [8, §2]. SYZYGIES OF QUADRATIC VERONESE EMBEDDING 5 map dual to the exterior multiplication map Λp−1W ⊗ W → ΛpW . Using the mul- tiplication map πq : A1 ⊗ Aq → Aq+1, let us define the map dp,q as a composition: (2.1) Λp−1W ∗ ⊗ W ∗ ⊗ Aq / Λp−1W ∗ ⊗ A1 ⊗ Aq ı⊗Id Id⊗πq ΛpW ∗ ⊗ Aq dp,q / Λp−1W ∗ ⊗ Aq+1, where the upper arrow is induced by the natural restriction map W ∗ = H0(P(W ), O(1)) → H0(X, O(1)) = A1. Lemma 2.1. The sequence of groups and morphisms (2.2) . . . / Λp+1W ∗ ⊗ Aq−1 dp+1,q−1 / ΛpW ∗ ⊗ Aq dp,q / Λp−1W ∗ ⊗ Aq+1 / . . . is a complex. Moreover, its cohomology groups are the syzygy spaces Rp,p+q = ker(dp,q) im(dp+1,q−1) . Proof. Consider the Koszul complex Λ•W ∗ ⊗Sym• W ∗ = Λ•W ∗ ⊗S (see [10], ch. 1). It is quasi-isomorphic to the trivial S-module k. Tensoring it over S with A, we obtain the quasi-isomorphism Λ•W ∗ ⊗ A ∼= k morphisms A. Therefore, the sequence of L ⊗ S . . . → Λp+1W ∗ ⊗ A → ΛpW ∗ ⊗ A → Λp−1W ∗ ⊗ A → . . . is a complex, and its cohomology groups equal graded vector spaces TorS p (k, A). Since the differential d of the complex (2.2) is homogeneous of degree 0, we can decompose the Koszul complex into the sum of subcomplexes: . . . / Λp+1W ∗ ⊗ Aq−1 dp+1,q−1 / ΛpW ∗ ⊗ Aq dp,q / / Λp−1W ∗ ⊗ Aq+1 / . . . Finally, we get Rp,p+q = (TorS p (A, k))p+q = ker(dp,q) im(dp+1,q−1) . (cid:3) 2.2. Projective coordinate algebras. Let G be a reductive algebraic group. Denote by W = Vλ the irreducible representation of G of highest weight λ. Let X be the G-orbit of the point w ∈ P(W ) corresponding to highest weight vector in the representation W . In [11] it is proved that any such variety X is an intersection of quadrics in P(W ). Remark 2. Recall that X = G · w ∼= G/P is a projective variety, where P is a parabolic subgroup. Denote by I the set of simple roots orthogonal to λ. Each set I of simple roots corresponds to a parabolic subgroup PI ⊂ G containing some fixed Borel subgroup B of the group G. Recall that linear bundles on the variety G/PI correspond to weights of the group G orthogonal to all roots in the set I. In particular, the bundle OX (1) = OP(W )(1)X corresponds to the weight λ. The following proposition with more detailed proof and other results about high- est weight orbits and more general quasi-homogeneous spaces can be found in [12]. / O O / / / / / / / / 6 I. V. NETAY Proposition 1. In the notation above the projective coordinate algebra of the va- riety X equals V ∗ nλ. AX =Mn>0 Γ(G/P, Lnλ) =Mn>0 Proof. By the Borel -- Bott -- Weyl theorem (see [13]) we get AX =Mn>0 Γ(X, OX (n)) =Mn>0 Γ(G/B, Lnλ) =Mn>0 V ∗ nλ. If we have a representation Vλ of G, then it has a unique closed G-orbit isomorphic to G/P in the projectivization. Then the bundle L on G/P can be defined as the restriction of O(1) on P(Vλ) onto G/P . (cid:3) Corollary 1. In the notation above the complex (q−1)λ → ΛpW ∗ ⊗k V ∗ . . . → Λp+1W ∗ ⊗k V ∗ qλ → Λp−1W ∗ ⊗k V ∗ (q+1)λ → . . . of representations of the group G computes the syzygy spaces of the variety X = G · w ⊂ P(W ). Denote by Σλ the Schur functor Vect → Vect. It is a polynomial functor on the category of vector spaces such that for the action of the group G = GL(V ) the representation ΣλV is the unique representation of the group G with the highest weight λ. The symbol ΣλV ′ denotes the same as Vλ if V ′ = V is the tautological representation of the group GL(V ). The explicit construction of the Schur functor can be found, for example, in [14]. Corollary 2. Let Wi = Σλi Vi, W1 ⊗ . . . ⊗ Wm be an irreducible GL(V1) × . . . × GL(Vm)-representation, X ⊂ P(W ) be the highest weight orbit of weight λ1⊕. . .⊕λm in W = W1 ⊗ . . . ⊗ Wm. Then the projective coordinate algebra of the variety X equals AX =Mn>0 Σnλ1V ∗ 1 ⊗ . . . ⊗ Σnλm V ∗ m as a GL(V1)×. . .×GL(Vm)-representation. Therefore, the Koszul complex Λ•(W1 ⊗ . . . ⊗ Wn) ⊗ AX computes the syzygy spaces of X. 2.3. Localization. After in Section §3 we will compute cohomology groups of the Koszul complex and therefore we will construct minimal free resolutions for some graded modules over the polynomial ring k[W ]. Here we explain how it is related to resolutions of sheaves on the projective space P(W ). For a sheaf F on P(W ) we can construct a graded k[W ]-module applying the functor α : F 7→Mn>0 Γ(P(W ), F (n)). We will consider its image as an object of the category C of graded k[W ]-modules of finite type, where objects are the same graded k[W ]-modules and morphisms are defined in the following way: HomC (M, N ) = lim −→ m0 Homk[W ]-grmod Mm>m0 Mm, Mm>m0 Nm . The functor α : k[W ] → C is an equivalence of categories. It takes morphisms to morphisms and is exact. The opposite to α functor ∼ can be constructed as a /Lk /Lk /Lk /Lk SYZYGIES OF QUADRATIC VERONESE EMBEDDING 7 generalization of the localization functor for the affine case. (There given a module the proof can be found in [15, lecture 7]. M over a ring A one obtains a sheaf fM over Spec A). Detailed construction and Nevertheless the functor α is not exact as a functor to the category of graded modules. We will construct below free resolutions over the polynomial ring k[W ]. But if we construct a free resolution for the module M in the category k[W ]-grmod, obviously it remains a free resolution in the category C , because the functor of localization of an abelian category by an abelien subcategory is exact. Therefore, applying the functor ∼, we construct a resolution for fM in the category of sheaves This implies that by results of [15] the following proposition holds. on P(W ). Proposition 2 (([15])). Suppose that a module M has a resolution in the category of graded modules of finite type over k[W ]: . . . k[W ](k + 1) ⊗ R1,k k[W ](k) ⊗ R0,k / M / 0. Then in the category of coherent sheaves on P(W ) there is a resolution . . . OP(W )(−k − 1) ⊗ R1,k OP(W )(−k) ⊗ R0,k / 0. / fM We will consider objects of the form F = OP(V )(k) with support equal to the image of the quadratic Veronese embedding P(V ) ⊂ P(Sym2 V ). To construct a resolution of the sheaf F , firstly we construct a resolution of the k[V ]-module α(F ) and then using the fact that it remains a resolution under the action of the functor of localization by the subcategory of finite-dimensional objects, we apply the functor β and obtain a resolution for the sheaf F in terms of the sheaves OP(Sym2 V )(i). So we need to construct resolutions of modules of the form α(O(k)) =Lm>0 Sym2m+k V ∗ over α(O) = Sym• Sym2 V ∗. 2.4. Graphs in the lattice. Fix a dominant weight λ of a reductive group G. Then the isotypic component Kλ of weight λ in the Koszul complex K = Λ•V ∗ λ ⊗ A(X) admits the natural decomposition into the sum of subrepresentations Kµ,n, where Kµ,n is isotypic component of weight λ in the representation (Λ•V ∗ λ )µ ⊗ Vnλ and (Λ•V ∗ λ of the group G. Applying the Schur Lemma, we can pass from complexes of represen- tations of G to complexes of vector spaces. λ )µ is the isotypic component of weight µ in the representation Λ•V ∗ LetK • be a complex of vector spaces. Then we can choose a basis in each space K i and compose a graph with vertices corresponding to vectors in these bases and arrows corresponding to non-zero differential maps. For example, the graph on the fig. 1 corresponds to the complex {k → k} with non-zero differential map. It is always acyclic. Figure 1. For example, consider the graph on fig. 2. The condition of correspondence to a complex implies that xy + uv = 0 and means that d2 = 0. By a replacement of bases in one-dimensional vector spaces corresponding to vertices, we can put x = y = u = 1, v = −1. Therefore the complex corresponding to this graph is / / / / / / / / 8 I. V. NETAY x y u v Figure 2. unique up to isomorphism. Nevertheless not all graphs corresponding to complexes have the same property. For example, consider the graph on fig. 3 corresponding to a complex of the form {k2 → k2}, where in the chosen basis the 2 × 2-matrix of the differential d consists of non-zero elements. Rank of such matrix can be equal to 1 or 2. Therefore, this graph correspond to two different isomorphism classes of complexes. Figure 3. At the same time this graph is a full subgraph of the graph on fig. 4. This one corresponds to one isomorphism class of complexes and therefore the cohomologies of such complex are uniquely defined. Figure 4. Under the restrictions given by the conditions STMF and SMF the set of vertices of the graph corresponding to any isotypic component of the Koszul complex com- puting the syzygy spaces of the highest weight orbit is a subset of points of some lattice (namely, X ⊕ Z, where X is the weight lattice of the reductive group) and ar- rows between vertices can be easily described using some additional combinatorics on Young diagrams. 2.5. Truncated cubes. Suppose that the complex K • of vector spaces corresponds to a graph with vertices at vertices of the cube {0, 1}n and arrows correspond to edges of the cube and are directed to the vertex where the sum of coordinates is more. We will call such complexes n-dimensional combinatorial cubes. In accor- dance with results of [9] for any dimension there is a unique combinatorial cubes of this dimension up to isomorphism. In particular, for positive dimension the cor- responding complex is always acyclic. For example, the complex corresponding to the graph on fig. 2. Consider the complex obtained from the combinatorial cube by the stupid k-th truncation (we always assume the truncation from below here). Then it corresponds to a full subgraph in the cube of vetices with the sum of coordinates not less than k. We will call Such a complex truncated combinatorial cube. According to results SYZYGIES OF QUADRATIC VERONESE EMBEDDING 9 of [9] there exists a unique k-th truncated combinatorial n-cube. From here it is cohomologies vanish. obvious that zero cohomologies of this complex equal k6⊕(cid:0)n−1 syzygies of modules of the formLn>0 Sym2n+k Sym2 V ∗ over Sym• Sym2 V ∗. Below we will need such types of complexes to describe isotypic components of the Koszul complex computing syzygies of quadratic Veronese embedding and k (cid:1) and that all other 3. Computation of syzygies Let us introduce some notations needed below in this Section. Let λ be a Young diagram. Definition 4. Denote by C(λ) the set of such 1 6 l 6 k that • al > bl, i. e. l-th hook has width more than its height; • al+1 6 bl+1 + 1 or l = k, i. e. the following hook has width at most its height plus one or there is no a following hook, i. e. l = k. We will be interested in values of C on diagrams λ, where any hook is not more wide than high. Therefore the condition l ∈ C can be reformulated in the following way: • diagram λ contains the box (l + bl, l − 1), • diagram λ does not contain the box (l + bl, l). Figure 5 contains an example of a diagram λ = (8, 8, 6, 6, 4, 3) = (8, 7, 4, 32, 4, 5, 6). Here C(λ) = {2, 4}. In terms of the reformulation for each l ∈ {1, 2, 3, 4} the Figure 5. box (l + bl, l − 1) contains a bold dot. Element l is included into C(λ) if the corresponding dot is in the diagram and the following is not. Definition 5. Denote by Z(w, a) the set of Young diagrams θ with some of boxes shaded, where • there are 2w not shaded boxes, they form a Young diagram θ0, each it main hook has width equal to height plus one; • there are a shaded boxes, • there are no two shaded boxes in one column. Denote by ω(θ) the diagram of the same form as θ without shaded boxes. Figure 6 contains an example of an element θ ∈ Z(4, 4). Denote by Zn(w, a) the subset of elements in Z(w, a) with at most n columns. 10 I. V. NETAY Figure 6. Lemma 3.1. There is an isomorphism Λw Sym2 V ∗ ⊗ Syma V ∗ = Mθ∈Zn(w,a) where n = dim V . Proof. From well known formula (see [16]) we have Λw Sym2 V ∗ = M λ=(a1 ,...,an an−1,...,a1 −1) a1>a2 >...>an a1 +...+an =w ΣθV ∗, ΣλV ∗. Here the sum is taken over the set Z(w, 0). mula (see [14]) tensoring with Syma V ∗ and shading added boxes. It remains to apply the Pieri for- (cid:3) The main goal of this Section is to describe isotypic components of the Koszul complex (3.1) . . . → Λp+1 Sym2 V ∗ ⊗ Syma+q−2 V ∗ → → Λp Sym2 V ∗ ⊗ Syma+q V ∗ → → Λp−1 Sym2 V ∗ ⊗ Syma+q+2 V ∗ → . . . , which computes syzygies of Sym• Sym2 V ∗-moduleLm>0 Sym2m+a V ∗. Denote by prω the projection to isotypic component corresponding to the Young diagram ω. Lemma 3.2. Let ω = (a1, . . . , akbk, . . . , b1) be a Young diagram. Suppose one can shade some of its boxes and obtain a diagram θ ∈ Zn(w, a). Then there is an isomorphism of complexes prω(K •) = Vλ ⊗ Λ>skC(ω)(r), where s = a 0. 2 −Pk and r =Pk ai−bi−1 2 i=1 ai−bi+1 2 −C(ω). Otherwise prω(K •) = i=1 Proof. Let us check the conditions necessary for prω(K •) 6= 0. By Lemma 3.1 there is a bijection between irreducible G-modules in Λw Sym2 V ∗ ⊗ Syma V ∗ and the set Zn(w, a). Therefore the complex prω(K •) is isomorphic to (3.2) . . . → Mθ∈Zn(w+1,a−2) ΣθV ∗ → Mθ∈Zn(w,a) ΣθV ∗ → Mθ∈Zn(w+1,a−2) ΣθV ∗ → . . . Clearly, prω(K •) is a subcomplex in (3.1) and consists of modules ΣθW ∗ with ω(θ) = ω. It remains to check that this subcomplex is a truncated combinatorial cube. There is a graded vector space V • such that prω(K •) ≃ ΣωW ∗ ⊗ V •. Since the SYZYGIES OF QUADRATIC VERONESE EMBEDDING 11 representation ΣωW ∗ of G is irreducible, from the Schur Lemma (see [13]) we have End(ΣωW ∗) = k. Hence, d ∈ EndG(ΣωW ∗ ⊗ V •) = End(V •). Now let us prove that the complex prω(K •) is a truncated combinatorial cube. We prove it by two steps. At first we choose a basis in V •. At second we find all non-zero matrix elements of d in this basis. After it remains only to apply Lemma 3.5 from [9]. Take any diagram ω. If we can not shade ω to obtain an element of Zn(w, a), then prω(K •) = 0. We assert hereafter that we can shade the diagram ω in this way. For any shaded diagram θ ∈ Zn(w, a) its non-shaded part consists of hooks of the form (kk − 1). All these diagrams differ only in shading of some pairs of boxes. Namely, these pairs are (l − 1, l − 1 + bl) and (l + bl, l − 1) for l ∈ C(ω). Therefore, shaded diagrams of the form ω are in bijection with subsets of C(ω), i. e. with a basis of the exterior algebra Λ•E of the vector space E spanned by vectors ξ1, . . . , ξk indexed by elements of C(ω). A vector ξi1 ∧ . . . ∧ ξil with 1 6 i1 < . . . < il 6 k corresponds to the diagram θ shaded in the following way. At first, for each main hook in ω let us shade all boxes except the maximal subhook of the form (mm− 1). After for each i ∈ {i1, . . . , il} in the hook number i let us shade a pair of boxes such that non-shaded boxes form again a hook of the form (mm − 1). Therefore we should shift the degree of the complex by 1 hook in non-shaded part have the form (mm−1), we shadePk 2Pk Minimal number of shaded boxes in diagram equalsPk Clearly, if in ω we shade maximal possible number of boxes such that each main i=1(ai −bi +1) boxes. i=1(ai − bi + 1) − C(ω). i=1 ai − bi − 1. In the same time the complex contain only diagrams θ, where at least a boxes are shaded. Therefore the exterior power ΛsE contributes to the complex if and only if s > 2 − 1 a The differential d in the Koszul complex (3.1) is a composition of the comulti- i=1 ai − bi − 1. 2Pk plication map Λp+1 Sym2 V ∗ ⊗ Syma+q V ∗ ι−→ Λp Sym2 V ∗ ⊗ Sym2 V ∗ ⊗ Syma+q V ∗ in the exterior algebra and the multiplication map Λp Sym2 V ∗ ⊗ Sym2 V ∗ ⊗ Syma+q V ∗ π−→ Λp Sym2 V ∗ ⊗ Syma+p+2 V ∗ in the projective coordinate algebra. Take θ ∈ Zn(p + 1, a). By Lemma 3.1 it cor- responds to an irreducible G-subrepresentaion in Λp+1 Sym2 V ∗ ⊗ Syma V ∗. The restriction of the comultiplication map onto the summand corresponding to ΣθV ∗ can be decomposed into the sum ΣθV ∗ → Σν V ∗ ⊗ Sym2 V ∗, where ν ∈ Zn(p, a) is obtained from θ by removement of a pair of boxes from different columns. Note that the composition of the comultiplication map restricted to the summand cor- responding θ and the multiplication map does not vanish under projection to the summand corresponding η ∈ Zn(p, a + 2) if and only if η can be obtained from θ by shading of two boxes, because diagrams must have the same form, because all the map are G-equivariant. So, if a diagram θ corresponds to a vector ξi1 ∧. . .∧ξil , then differential has non-zero restriction maps from ΣθV ∗ to all ΣηV ∗, where diagrams η correspond to non-zero vectors ±ξm ∧ ξi1 ∧ . . . ∧ ξil for m ∈ C(ω). From [9, Lemma 3.5] pr(K•) is a truncated combinatorial cube. This gives the required isomorphism of complexes. (cid:3) 12 I. V. NETAY Proposition 3. The isotypic component of weight ω in the complex (3.1) is a combinatorial 0-cube if and only if the diagram ω is symmetric, i. e. ω = ω′, and parity of wt(ω) is the same as parity of a. Proof. Note that in the complex (3.1) each diagram θ has a + 2k shaded boxes for different k. Each main hook of a diagram θ in the complex can have at most one shaded box at the bottom. Therefore height of any hook does not exceed its width. Let θ = (a1, . . . , akbk, . . . , b1). Suppose that the diagram θ is not symmetric. Let l be maximal such that al 6= bl. This means that al > bl. But either k = l or al+1 = bl+1 < bl. Therefore the box (bl +l, l−1) is contained in θ, hence l ∈ C(θ). Consequently, the isotypic component of weight θ is not a combinatorial 0-cube. This implies that each hook in a diagram corresponding to a 0-cube is the sym- metric. This means that each main hook contains one shaded box at the bottom. The number of main hooks coincides with the diagonal length. Its parity coincides with parity of a. It is also obvious that for a diagram θ satisfying these conditions the corresponding isotypic component is a 0-cube. (cid:3) Lemma 3.3. Suppose a 6 0. Then the complex K • does not contain an isotypic component beeing a k-truncated combinatorial cube for k > 0. Proof. By Lemma 3.2 for a 6 0 the isotypic component of weight ω is isomor- phic to Vω ⊗ Λ>sE(r) for some k, for s 6 a 6 0 and for some vector space E. Therefore Vω ⊗ Λ>sE(r) = Vω ⊗ Λ•E(r). (cid:3) Now let us prove Theorem 1. Proof. By Lemma 2 the syzygy spaces Ra p,q are calculated by the Koszul com- plex (3.1). By Lemma 3.2 all the isotypic components of this complex are trun- cated combinatorial cubes. Combinatorial 0-cubes of complex (3.1) are described by Proposition 3 and correspond to the first row in (1.1). Now let us decribe truncated combinatorial cubes. 2(cid:16)a −Pk degree zero. Take a diagram ω. It must havePk Each truncated combinatorial cube has the component of minimal degree at i=1 ai − bi − 1 shaded boxes. The diagram corresponding to a representaion lying in degree zero has totally a shaded boxes. The set C(ω) corresponds to the hooks where we can shade some other boxes. Except the boxes that must be shaded, there are also s = 1 hooks, where we can shade some additional boxes. Note that if the number s is not integer, then the complex has no representations corresponding to a shaded diagram of form ω. To get a correct shaded diagram, we need to shade some boxes in diagram ω corresponding to a subrepresentation in the Koszul complex in such a way that there is at most one shaded box in each column. In the same time non- shaded boxes must form a Young diagram of the form (c1, . . . , ckck − 1, . . . , c1 − 1) for some ci. This means that in a non-shaded diagram we can add to i-th hook at most one shaded box to the bottom and at at most bi−1 + 1 − ai shaded boxes to the right. This is equivalent to inequalities bi 6 ai 6 bi−1 + 1 for i = 1, . . . , k. It remains to note that we have obtained s-th truncated combinatorial C(ω)-cube, therefore the representation V ∗ ω gives the contribution to cohomology groups with (cid:3) i=1 ai − bi − 1(cid:17) multiplicity(cid:0)C(ω)−1 (cid:1). s SYZYGIES OF QUADRATIC VERONESE EMBEDDING 13 4. Classification Fix a dominant weight λ of a reductive group G. Put X = G · [vhw] ⊂ P(Vλ), where vhw is the highest weight vector in the representation Vλ of G. Syzygies of the embedding X ⊂ P(Vλ) are calculated by the Koszul complex Λ•V ∗ λ ⊗ A(X). To calculate the objects in the Koszul complex as a complex of representations of G, we need calculate exterior powers of representation V ∗ λ and tensor products of appearing in the previous calculation representations and representations of the form V ∗ nλ, where n ∈ N. Calculation of morphisms in the Kosul complex in much more sophisticated problem. Nevertheless, in in some cases where all the multiplici- ties equal 0 or 1 and some of morphisms are known to be non-zero, the isomorphism classes of isotypic components of Koszul complex are uniquely defined. The following properties of dominant weights are important for the description of isotypic components of Koszul complex in terms of Section §2.4. Property 4. A dominant weight λ of a reductive group G satisfies the property SMF (is ¡¡skew multiplicity-free¿¿, following [8]), if for each k the representa- tion ΛkVλ of G has no multiple subrepresentations. Property 5. A dominant weight λ of reductive group G satisfies the property STMF (is ¡¡strongly tensor multiplicity-free¿¿), if for any dominant weight µ and for each n the representation Vnλ ⊗ Vµ of G has no multiple subrepresentations. Remark 3. In [8] the property SMF of a weight λ was defined as absence of multiple subrepresentations in the representation Λ•Vλ. These two properties are some bit different. For example, the tautological representation of GL(V ) is SMF. In the same time the tautological representation of SL(V ) is SMF in our definition, but is not SMF in the definition of [8]. These variants of definition coincide under so called saturatedness condition of the representation. In can be defined for any finite-dimensional representation, but we need it only for the case of irreducible representations. In this case it is equivalent to the condition that the central torus of the reductive group G act non-trivially. Since we always pass to the projectivization of the representation, we can assume that the reductive group group always contain a central torus with non-trivial action. The following property of a dominant weight of a reductive group is more simple for classification and more natural than STMF. Property 6. A dominant weight λ of reductive group G satisfies the property TMF (is ¡¡tensor multiplicity-free¿¿), if for any dominant weight µ the representation Vλ ⊗ Vµ of G has no multiple subrepresentations. The following property is useful for the classification of representations being STMF and TMF. Lemma 4.1. The property to be STMF(and the same for TMF) for weights λ1 of a group G1 and λ2 of a group G2 is equivalent to the same property for the weight λ1 ⊗ λ2 of the group G1 × G2. Applying this lemma, we deduce the classification to simple algebraic groups. Lemma 4.2. Let λ, µ and ν be dominant weights of a reductive group G. Then there is an embedding Vλ ⊗ Vµ ⊆ Vλ+ν ⊗ Vµ−w◦ν. 14 I. V. NETAY Remark 4. Note that the lemma implies that there is an embedding. This means that for any irreducible representation Vθ its multiplicity in the left hand side does not exceed its multiplicity in the right hand side. Proof. Given some dominant weights λ and µ, we define the multiplicities mθ subrepresentations by the formula λ,µ of Vλ ⊗ Vµ =Mθ ⊕mθ V θ λ,µ . Then from [17, Theorem 3.1] it follows that mθ λ,µ = dim V +(θ; λ − µ∗, µ∗), where Therefore it remains to prove that mθ V +(θ, µ, γ) := {v ∈ V (θ)µ : eγ(hi)+1 v = 0 ∀i ∈ I}. i λ,µ 6 mθ λ+ν,µ+ν∗ , i. e. dim V +(θ; λ − µ∗, µ∗) 6 dim V ∗(θ; λ − µ∗, ν + µ∗). This follows from V +(θ; λ, µ) ⊆ V +(θ; λ, µ + ν) for any dominant weights θ, λ and µ. This inclusion directly follows from the definition of these spaces. (cid:3) We have proved that there is an embedding, but there is a canonical embedding. The following lemma proves it. Lemma 4.3. There is a natural embedding for any dominant weights λ, µ and ν of a reductive group G. Vλ ⊗ V ∗ µ ⊆ Vλ+ν ⊗ V ∗ µ+ν Proof. Consider the variety X = G/B. The space of global sections of the sheaf Fµ can be naturally identified with V ∗ µ by the Borel -- Bott -- Weyl Theorem [13, §23.3, Claim 23.57]. Analogously, since the weight −λ is dominant with respect to B−, −λ on X ′ = G/B− can be identified the space of global sections of the sheaf F ′ with Vλ. Hence, we have Γ(X × X ′, Fµ ⊠ F ′ −λ) = V ∗ µ ⊗ Vλ. Take the identical section ιν ∈ V ∗ ν ⊗ Vν = Γ(X × X ′, Fν ⊠ F ′ −ν ). Multiplication by the section ιν defines the sheaf morphism Fµ ⊠ F ′ −λ → Fµ+ν ⊠ F ′ −λ−ν . Since the section ιν vanish nowhere, this morphism is an embedding and induce the required embedding of spaces of global sections. (cid:3) Lemma 4.4. If a dominant weight λ of a group G is not TMFand a weight µ is dominant, then the weight λ + µ also is not TMF. Proof. Let us apply the following two properties of representations of any group G: HomG(U, V ⊗ W ) = HomG(U ⊗ V ∗, W ), k ⊆ V ⊗ V ∗ SYZYGIES OF QUADRATIC VERONESE EMBEDDING 15 for any representations U, V, W . If λ is not TMF, then for some weights θ and ν the following holds by 4.2: V ⊕2 θ ⊆ Vλ ⊗ Vν ⇔ dim HomG(Vθ, Vλ ⊗ Vν ) > 2 ⇔ ⇔ dim HomG(Vθ ⊗ V ∗ λ , Vν ) > 2 ⇒ dim HomG(Vθ+µ ⊗ V ∗ ⇔ dim HomG(Vθ+µ, Vλ+µ ⊗ Vν ) > 2 ⇔ V ⊕2 λ+µ, Vν) > 2 ⇔ θ+µ ⊆ Vλ+µ ⊗ Vν. This means that the weight λ + µ is not TMF. (cid:3) By this lemma it is sufficient for classification of all TMF weights for each simple algebraic group to find all TMF weights and to check that all minimal weights greater are not TMF. Let G be a semisimple algebraic group. Denote by πi its i-th fundamental weight (see. [18]). Proposition 4. Let G be a simple non-commutative algebraic group, and π be its dominant weight such that for any dominant weight of G the representation Vπ ⊗ Vµ of G has no multiple subrepresentations. Then the pair (G, π) is one of the listed in the following table up to dualization: SL(n) SL(n) SO(2n + 1) SO(2n) Sp(2n) Sp(2n) E6 E7 G2 kπ1, k > 1 πk, k > 1 π1, πn π1, πn−1, πn π1 πn, n = 2, 3 π1, π6 π7 π1 Proof. Let us list the minimal weights greater than TMF weights and apply Lemma 4.4. The list is given in the following table. G SL(n) SO(2n + 1) Sp(2n) SO(2n) E6 E7 E8, F4 G2 π kπ1 + πi, i > 1, πi + kπn−1, i < n − 1, πi + πj, i < j 2π1, 2πn, π1 + πn, πi, 1 < i < n 2π1, πi, i > 1 πi + πn, 2πn, n 6 3 2π1, 2πn−1, 2πn, π1 + πn−1, π1 + πn, πn−1 + πn, πi, 1 < i < n − 1 2π1, 2π6, π1 + π6, πi, 1 < i < 6 2π7, πi, i 6= 7 πi 2π1, π2 At first, let us check all the TMF weights. In the case of the group SL(n) and the representations with highest weight kπ1 (or kπn−1 by duality) and πk tensor 16 I. V. NETAY product can be calculated by the well known Pieri formula for all k. In the case of the group G = SO(n) or Sp(2n) the tensor product of the representation Vπ1 and a representation Vλ equals the sum of irreducible representations Vλ′ with weights λ′ that can be obtained from λ by addition or subtraction of any prime root. For other listed in the table above representations the property TMF follows from fact that the maximal torus of the group has no multiple weights in the representation. Such representations are classified in [19]. This corollary is proved in [20]. Now it remains only to check that there are no TMF weights greater than listed above. • For SL(n) we have m(k−1)π1+2π2+πm kπ1+πm,π1+π2 πl+πm,π1+π2 = 2, l < m. duality), m2π1+πl+1+πm = 2 (analogously, for πm + kπn−1 by • For SO(2n+1) we have mπ1+π2 2π1,π1+π2 = 2, mπ1+πn π1+πn,π1+π2 = 2, mπ1+2πn 2πn,π1+π2 = 2, mπ1+πm πm,π1+π2 = 2, 1 < m < n. • For Sp(2n) we have mπ1+π2 2π1+2π2,2π2 = 2, m2π1 2π3,π1+π2+π3 = 2 for Sp(6). • For SO(2n) we have mπ1+π2 have m2π1+2π2 mπ1+π2+π3 π1+πn−1,π1+π2 = 2, mπ1+πn−1 mπ1+πn 2, 1 < m < n − 1. 2π1,π1+π2 = 2, mπ1+πm π1+π2,π1+π2 = 2 for Sp(4) and mπ1+π2 πm,π1+π2 = 2, 1 < m < n. Also we π1+π3,π2+π3 = 2, 2π1,π1+π2 = 2, mπ1+πn−1+πn π1+πn,π1+π2 = 2, mπ1+πn−1+πn 2πn−1,π1+π3 = 2, mπ1+πn−1+πn πn−1+πn,π1+π2 = 3, mπ1+πm 2πn,π1+π3 = 2, πm,π1+π2 = • For E6 we have mπ1+π3 2π1,π1+π2 = 2, mπ2+π6 2π6,π1+π2 = 2, mπ5 π1+π6,π1+π2 = 2, mπ1+π2 π2,π1+π2 = 2, mπ4 • For E7 we have mπ1+π2 π3,π1+π2 = 2, mπ5 π1,π1+π2 = 2, mπ3 π4,π1+π2 = 2, mπ3 π2,pi1+π2 = 2, mπ1+π2 π5,π1+π2 = 2. π3,π1+π2 = 4, mπ1+π2 π4,π1+π2 = 6, mπ6 π5,π1+π2 = 2, mπ1+π2 • For E8 we have mπ1+π8 π6,π1+π2 = 3, mπ1+π2 π1,π2 = 2, mπ1+π8 2π8,π1+π2 = 2. π3,π4 = 4, mπ1+π8 π5,π6 = 6, mπ1+π2 π7,π1+π2 = 6, mπ1+π2 π8,π1+π2 = 2. 2. • For F4 we have mπ1+π2 π1,π1+π2 = 2, mπ1+π2 π2,π1+π2 = 5, mπ1+π3+π4 π3,π1+π2 = 3, mπ1+π2+π3+π4 π4,π1+π2+π3+π4 = • For G2 we have mπ1+π2 2π1,2π1 = 2 and mπ1+π2 π2,π1+π2 = 2. Definition 6. Irreducible representations V and W of the group G are geometri- cally equivalent, if in the projectivizations the G-orbits of highest weight vectoror- bits are isomorphic embeddings of algebraic varieties into a projective space. Proposition 5. Each STMF and SMF representation of semisimple algebraic group is equivalent to one of the listed in the following table: (cid:3) X P(V ) P(V ) P1 P2 P(W ) P(V ) P(Sym2 V ) Pn, n 6 6 P9 P(U ) × P(V ) P(U ⊗ V ) P2 × P1 P1 × Pk, k 6 4 P1 × P1 P1 × P1 × P1 P11 P3k+2 P7 P7 OP(W )(1)X O(1) O(2) O(n) O(3) O(1, 1) O(2, 1) O(2, 1) O(3, 1) O(1, 1, 1) SYZYGIES OF QUADRATIC VERONESE EMBEDDING 17 Proof. The property of the representation to be STMF deduces the classification to Segre -- Veronese varieties. This fact follows from the fact that a weight λ is STMF if and only if nλ is TMF for any n. Classification of TMF weights is done above. There only the weights of SL(V ) can be again TMF after multiplication by any in- teger. For convenience we will consider representations of the group GL(V ) instead of SL(V ). It remains only to find SMF representations among the representations of the form Syma1 V1 ⊗ . . .⊗ Syman Vn of the group GL(V1)× . . .× GL(Vn). One can easily check the cases listed in the table applying the Cauchy formulas in infinite cases and directly in the remaining finite number of cases. Let us show that there are no other cases. Each Segre -- Veronese embedding is defined by the dimenstions mi = dim Vi and the degrees ai. Let us write them into two rows: (cid:18)m1 a1 . . . mn . . . an(cid:19) Note that there are three ways to "enlarge" a Segre -- Veronese embedding: (i) to add a column(cid:0)1 1(cid:1), (ii) to inscrease a dimension ni, (iii) to inscrease a degree ai. Let us show that the property of a weight not to be SMF is preserved. The case (i) trivially follows from the isomorphism HomG(W, ΛkV ) ∼= HomG×k×(W ⊗ k, Λk(V ⊗ k)) for each representations V and W of G, where k is the tautological representation of k× = GL(1). Clearly, if we increase any ni, then all the multiplicities of subrepre- sentations can not decrease. Therefore the case (ii) is obvious. The case (iii) follows from the fact that if dim HomG(Vθ, ΛkVλ) > 2, then dim HomG(Vθ+kν,Λk Vλ+ν ) > 2, where λ, µ and ν are dominant weights of a reductive group G. It is a particular case of the following lemma. Lemma 4.5. Let G be a reductive group, λ, ν and θ are dominant weights. Then there is a canonical embedding HomG(Vθ, ΛkVλ) ⊆ HomG(Vθ+kν , ΛkVλ+ν ) for any k. In particular, if Vλ is not SMF, then Vλ+ν is not SMF for any domi- nant ν. Proof. Consider the variety X = G/B. Global sections of the sheaf Fθ can be identified with V ∗ θ . Since −λ is dominant with respect to B−, global sections of F−λ on X ′ = G/B− can be identified with Vλ. Therefore global sections of ΛkF−λ can be identified with ΛkVλ. So we have Γ(X × X ′, Fθ ⊠ ΛkF−λ) = V ∗ Consider the identical section ıν,k ∈Nk(V ∗ −ν ). On X multiplication by ıν,k induces the sheaf morphism Fθ → Fθ+kν. On X ′ it induces the morphism ΛkF−λ → ΛkF−λ−ν. Since ıν,k vanishes nowhere, multipli- cation by it induces the required G-equivariant embedding on global sections: ν ⊗ Vν ) = Γ(X × X ′, F ⊗k ν ⊠ F ⊗k θ ⊗ ΛkVλ. V ∗ θ ⊗ ΛkVλ ·ık,ν−−−→ V ∗ θ+kν ⊗ ΛkVλ+ν . (cid:3) 18 I. V. NETAY Now, from (i -- iii) one can see that it remains to check a finite set of Segre -- Veronese embeddings. The set of minimal non-SMF representations to check is given by the following dimension-degree matrices: (cid:18)2 2 1 1 2 2 1 1(cid:19) ,(cid:18)2 2 1 1 2 2(cid:19) ,(cid:18)2 1 2 3 1 1(cid:19) ,(cid:18)2 2 (cid:18)2 3 1 2(cid:19) ,(cid:18)3 2 2 2(cid:19) ,(cid:18)2 2 1 4(cid:19) , 1 3(cid:19) ,(cid:18)5 2 1 2(cid:19) ,(cid:18)4 3(cid:19) ,(cid:18)3 4(cid:19) ,(cid:18)2 7(cid:19) . For each of them it is easy to find corresponding exterior power with a multiple subrepresentation, using, for example, LiE computer algebra system. (cid:3) Appendix A. Examples Example 1. Consider the Veronese embedding X = P3 ⊂ P9. There exist 8 symmetric Young diagrams with even diagonal of height at most 4: diagram representation syzygy space dimension ∅ k Σ2,2V ∗ Σ3,2,1V ∗ Σ3,3,2V ∗ Σ4,2,1,1V ∗ Σ4,3,2,1V ∗ Σ4,4,2,2V ∗ R0,0 R1,2 R2,3 R3,4 R4,5 R5,6 1 20 64 45 140 64 20 Σ4,4,4,4V ∗ In particular, there exists a resolution: 0 → O(−8) → O(−6)⊕20 → O(−5)⊕64 → O(−4)⊕185 → O(−3)⊕64 → O(−2)⊕20 → O → OX → 0. R6,8 1 Example 2. Consider the Veronese embedding X = P2 ⊂ P5 and the sheaf F = OX (1). There exist 4 symmetric Young diagrams of height at most 3 with odd diagonal length. R0,0 V ∗ R1,1 Σ2,1V ∗ R2,2 Σ3,1,1V ∗ R3,4 Σ3,3,3V ∗ SYZYGIES OF QUADRATIC VERONESE EMBEDDING 19 In particular, there exists a resolution: 0 → O(−4) → O(−2)⊕6 → O(−1)⊕8 → O ⊕3 → F → 0. Appendix B. Weighted projective spaces Another interesting question is how to calculate syzygies of weighted projective spaces. This problem is surprisingly hard. For example, if we take the space P(1l, km) and consider its embedding into a projective space of minimal dimension, then the problem of calculation of syzygies is equivalent to calculation of syzygies of the Veronese embedding of Pl of degree k. This demonstrates why the problem is so hard in the general case. But in the case P(1l, 2m) the problem can be deduced to the quadratic Veronese embedding, where the answer is known. Proposition 6. Let Rp,q be the syzygy spaces of weighted projective space P(1l, 2m) ⊂ P(l+1 2 )+m. Then there exist an isomorphism of representations of GL(l): Rp,q = Mλ=λ′ Σλkl. wt(λ)=2q l(λ)=2q−2p Actually, this embedding can be obtained from the quadratic Veronese embed- ding by the following operation applied few times: we can embed it into a projective space of dimension greater by 1 and there take a cone over previous variety. Here the set of equation is preserved, but some new variables appear. Obviously, the syzygy spaces are the same. Difficulty of calculation for more general projective embeddings of weighted pro- jective spaces is related to the fact that their automorphism groups are usually significantly less. Correspondingly, theory of representations of these groups does not provide so powerful instrument for calculation of syzygies. References [1] J. Harris, Algebraic geometry. A first course., Springer, 1992. [2] D. Hilbert, " Uber die Theorie der algebraischen Formen", Ges. Abh. II 2 (1970), 199 -- 257. [3] D. Eisenbud, The Geometry of Syzygies, Springer, 2005. [4] G. Ottaviani, R. Paoletti, "Syzygies of Veronese embeddings", Compositio Math., 125:1 (2001), 31 -- 37 [5] Thavn Vu, "N6 property for third Veronese embedding", http://arxiv.org/abs/1303.5532v1 [6] V. Reiner, J. Roberts, "Minimal resolutions and the homology of matching and chessboard complexes", Journal of Algebraic Combinatorics, 11:2 (2000), 135 -- 154. [7] James E. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer, 1978. [8] Tobias Pecher, "Classification of skew multiplicity-free modules", Transformation Groups, 17:1 (2012), 233 -- 257. [9] I. V. Netay, "Syzygy Algebras for Segre Embeddings" Funktsional. Anal. i Prilozhen. 47:3 (2013), 54 -- 74; Funct. Anal. Appl. 47:3 (2013), 210 -- 226 [10] Gelfand, Sergei I., Manin, Yuri I., Methods of Homological Algebra, Springer, 2003. [11] G.Lancaster, J.Towber, "Representation-functors and flag-algebras for the classical groups", J. Algebra, 59 (1979), 16 -- 38. [12] `E. B. Vinberg, V. L. Popov, "On a class of quasihomogeneous affine varieties", Izv. Akad. Nauk SSSR Ser. Mat., 36:4 (1972), 749 -- 764; Math. USSR-Izv., 6:4 (1972), 743 -- 758. [13] W. Fulton, J. Harris, "Representation theory. A first course", Graduate Texts in Mathematics, 129, Springer-Verlag, 1991. [14] W. Fulton, Young tableaux, CUP, 1997. [15] D. Mumford, Lectures on Curves on an Algebraic Surface, Princeton University Press, 1966. 20 I. V. NETAY [16] I. Macdonald, Symmetric Functions and Hall Polynomials, Oxford University Press, 1995. [17] Apoorva Khare, "Representations of complex semi-simple Lie groups and Lie algebras", arxiv.1208:0416v2, 2012 [18] Al L. Onishchik, E. B. Vinberg, Lie Groups and Algebraic Groups, Springer, 1995. [19] R. Howe, "Perspectives on Invariant Theory: Schur duality, multiplicity-free actions and be- yond", Israel Math. Conf. Proceedings, 8 (1995), 1 -- 182. [20] Toshiyuki Kobayashi, "Multiplicity-free Representations and Visible Actions on Complex Manifolds", RIMS, Kyoto Univ., 41 (2005), 497 -- 549. Institute for information Transmission Problems, Laboratory of Algebraic Geome- try and its Applications, National Research University Higher School of Economics E-mail address: [email protected]
1802.09556	2	1802	2019-04-18T22:49:22	Formal moduli problems and formal derived stacks	[ "math.AG", "math.AT" ]	This paper presents a survey on formal moduli problems. It starts with an introduction to pointed formal moduli problems and a sketch of proof of a Theorem (independently proven by Lurie and Pridham) which gives a precise mathematical formulation for Drinfeld's derived deformation theory philosophy, which gives a correspondence between formal moduli problems and differential graded Lie algebras. The second part deals with Lurie's general theory of deformation contexts, which we present in a slightly different way than the original paper, emphasising the (more symmetric) notion of Koszul duality contexts and morphisms thereof. In the third part, we explain how to apply this machinery to the case of non-split formal moduli problems under a given derived affine scheme; this situation has been dealt with recently by Joost Nuiten, and requires to replace differential graded Lie algebras with differential graded Lie algebroids. In the last part, we globalize this to the more general setting of formal thickenings of derived stacks, and suggest an alternative approach to results of Gaitsgory and Rozenblyum.	math.AG	math	Formal moduli problems and formal derived stacks Damien Calaque and Julien Grivaux To the memory of Jean-Louis Koszul A mathematician, like a painter or a poet, is a maker of patterns. If his patterns are more permanent than theirs, it is because they are made with ideas. G. H. Hardy -- A Mathematician's Apology Abstract This paper presents a survey on formal moduli problems. It starts with an introduction to pointed formal moduli problems and a sketch of proof of a Theorem (independently proven by Lurie and Pridham) which gives a precise mathematical formulation for Drinfeld's derived deformation theory philosophy, which gives a correspondence between formal moduli problems and differential graded Lie algebras. The second part deals with Lurie's general theory of deformation contexts, which we present in a slightly different way than the original paper, emphasizing the (more symmetric) notion of Koszul duality contexts and morphisms thereof. In the third part, we explain how to apply this machinery to the case of non-split formal moduli problems under a given derived affine scheme; this situation has been dealt with recently by Joost Nuiten, and requires to replace differential graded Lie algebras with differential graded Lie algebroids. In the last part, we globalize this to the more general setting of formal thickenings of derived stacks, and suggest an alternative approach to results of Gaitsgory and Rozenblyum. Contents 1 Introduction to pointed formal moduli problems for commutative algebras 1.1 Small augmented algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 6 8 9 1.3 A glimpse at the description of FMPk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The∞-category of formal moduli problems . . . . . . . . . . . . . . . . . . . . . . . . . . 2 General formal moduli problems and Koszul duality 12 2.1 Deformation contexts and small objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.2 Dual deformation contexts and good objects . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.3 Koszul duality contexts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.4 Morphisms of (weak) Koszul duality contexts . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.5 Description of general formal moduli problems . . . . . . . . . . . . . . . . . . . . . . . . 23 2.6 The tangent complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 3 DG-Lie algebroids and formal moduli problems under Spec(A) 26 . . . . . . . . . . . . . . . . . . . . . . . . . 27 3.1 Split formal moduli problems under Spec(A) 3.2 DG-Lie algebroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.3 Formal moduli problems under Spec(A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.4 The relative tangent complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.5 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4 Global aspects 35 4.1 Formal derived prestacks and formal thickenings . . . . . . . . . . . . . . . . . . . . . . . 35 4.2 DG-Lie algebroids and formal thickenings . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 4.3 Formal moduli problems under X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 1 A Appendix 43 A.1 Recollection on graded mixed stuff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 A.2 Graded mixed cdgas and dg-Lie algebroids . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Introduction The aim of these lecture notes is to present a recent work, due independently to Lurie and Pridham, concerning an equivalence of infinity-categories between formal moduli problems and differential graded Lie algebras. The link between deformation theory and dg-Lie algebras has a long history, which is as old as deformation theory itself. One of the first occurrence of a relation between these appears in Kodaira -- Spencer's theory of deformations of complex compact manifolds (see [21]): if X is such a manifold, then infinitesimal deformations of X over a local artinian C-algebra A with maximal ideal mA correspond to Maurer -- Cartan elements of the dg-Lie algebra gX ⊗C mA modulo gauge equivalence, where gX is the dg-Lie algebra Γ (X, A0,• X ⊗ TX), the differential being the ∂ operator. This very concrete example illustrates the following general principle: to any sufficiently nice dg-Lie algebra it is possible to attach a deformation functor (see e.g. [22, §3.2]), which is defined on local artinian algebras as follows: Defg(A) = MC(g ⊗ mA)/ gauge equivalence This correspondence was carried out by many people, including Quillen, Deligne and Drinfeld. In a letter to Schechtman [6], Drinfeld introduced in 1988 the Derived Deformation Theory (DDT) philosophy: Every (dg/derived) deformation problem is controlled by a dg-Lie (or L∞-) algebra. Since then, there has been a lot of work confirming this philosophy. We can refer the interested reader to the expository paper [24] for more details. Let us give now a few examples, which are related to derived algebraic geometry. In 1997 Kapranov [19] studied deformations of local systems: given a affine algebraic group G and a G-local system E on a manifold S, deformations of E are encoded by a formal dg-scheme RDef(E) whose tangent complex T[E]RDef(E) at the closed point [E] is RΓ (S, ad(E))[1]. The next result is that T[E]RDef(E)[−1] is naturally an L∞-algebra, the corresponding Lie algebra structure on the cohomology groups Hi(S, ad(E)) being induced by the natural Lie structure on ad(E). This last observation is not at all a hazard, and reflects a far more general phenomenon: in loc. cit., Kapranov proved1 that for any smooth dg-scheme X, the shifted tangent complex TxX[−1] at a point rephrasing of Kapranov's result can be presented as follows: for any derived affine scheme X = Spec(A) x : ∗→ X carries an L∞-algebra structure that determines the formal geometry of X around x. A modern with a k-point x = Spec(A→ k), we have that TxX[−1] ≃ TA ⊗A k[−1] ≃ Tk/A ≃ RDerA(k, k) and thus TxX[−1] is a derived Lie algebra. Let us explain another folklore calculation2 confirming this result in another concrete example. Let X = [G0/G1] be a 1-stack presented by a smooth groupoid G1 G0, where G0 and G1 are smooth affine algebraic varieties over a field k. If x is a k-point of G0, then the tangent complex T[x]X of X at G1 sits in degree −1, W = TxG0 sits in degree 0, and the arrow is given by the differential of the target map t at idx. the k-point [x] is a 2-step complex V→ W where V = T s idx According to Kapranov's result, there should exist a natural L∞-structure on T[x]X[−1]. Let us sketch the construction of such a structure. Observe that L := (TsG1)G0 has the structure of a (k, A) Lie algebroid, where G0 = Spec(A). kL → L and an anchor map L → TG0 = Derk(A). Very roughly, Taylor components at x of the anchor give us maps Sn(W) ⊗ V → W, and Taylor components of the bracket at x give us maps Sn(W) ⊗ Λ2(V)→ V. These 1This idea is also present in Hinich's [14]. 2We don't know a precise reference for that, but it seems to be known among experts. In particular we have a k-linear Lie bracket Λ2 s −→−→t 2 are the only possible structure maps for an L∞-algebra concentrated in degrees 0 and 1. Equations for the L∞-structure are guaranteed from the axioms of a Lie algebroid. As a particular but illuminating case, we can consider the stack BG for some affine algebraic group G. Here G0 = {∗} and G1 = G. Then T∗BG[−1] ≃ g where g is the Lie algebra of G, and the L∞-structure is simply the Lie structure of g. Let us explain how all this fits in the DDT philosophy. Any derived affine stack Spec(A) endowed with a k-point x defines a representable deformation functor B → Hom(B, A), the Hom functor being taken in the category of k-augmented commutative differential graded algebras (cdgas). Hence it should be associated to a dg-Lie (or indifferently a L∞-) algebra. This dg-Lie algebra turns out to be exactly Tx Spec(A)[−1], endowed with Kapranov derived Lie structure. In this way, we get a clear picture of the DDT philosophy for representable deformation functors. However, the work of Lurie and Pridham goes way beyond that; what they prove is the following statement: between formal moduli problems and dg-Lie algebras. Theorem ([23, 27]). Over a base field of characteristic zero, there is an equivalence of ∞-categories A very nice exposition of the above Theorem, with several examples and perspectives, is in Toen's [30]. Remark. Lurie's work [23] extends to En-deformation problems. Pridham's work [27] has some extension to the positive characteristic setting. In a forthcoming paper, Brantner and Mathew [2] actually generalize problems and so-called partition Lie algebras. the above result over any field, proving that there is an equivalence of∞-categories between formal moduli In these lectures, we will work in characteristic 0, and our goal is to provide a version of Lurie -- Pridham Theorem in families. In other words, we are aiming at first at a statement "over a cdga A", and build an extension from the affine case X = Spec(A) to an arbitrary derived Artin stack X. One shall be very careful as there are different meanings to "over a base". We will provide two variants. The second one, which is the one we are interested in, will actually rather be named "under a base". Split families of formal moduli problems In the same year 1997, Kapranov [18] considered the family of all formal neighborhood of points in a ^X × X of the diagonal in X × X. smooth algebraic variety X, which is nothing but the formal neighborhood He showed that the sheaf TX[−1] ≃ TX/X×X is a Lie algebra object in Db(X), whose Chevalley -- Eilenberg ^X × X. It is important to observe here that we have a kind of cdga gives back the structure sheaf of ^X × X) that lives both over X and under X. This example is a prototype for split families formal stack ( of formal moduli problems: a generalization of Lurie -- Pridham Theorem has recently been proven by Benjamin Hennion [11] in 2013, in the following form: Theorem ([11]). Let A be a noetherian cdga concentrated in nonpositive degrees, and let X be a derived Artin stack of finite presentation. -- A-pointed A-linear formal moduli problems are equivalent (as an∞-category) to A-dg-Lie algebras. -- X-pointed X-families of formal moduli problems are equivalent to Lie algebra objects in QCoh(X). Formal moduli problems under a base In 2013, Caldararu, Tu and the first author [4] looked at the formal neighborhood ^Y of a smooth closed subvariety X into a smooth algebraic variety Y. They prove that the relative tangent complex TX/Y is a dg-Lie algebroid whose Chevalley -- Eilenberg cdga gives back the structure sheaf of ^Y. Similar results had been previously proven by Bhargav Bhatt in a slightly different formulation (see [1]). We also refer to the work of Shilin Yu [33], who obtained parallel results in the complex analytic context. Note that this time, the derived scheme ^Y doesn't live anymore over X (although it still lives under X). Keeping this example in mind, we expect the following generalization of Hennion's result: -- A-pointed k-linear formal moduli problems are equivalent to dg-Lie algebroids over A, which has recently been proven by Joost Nuiten in [25]. This appears as Theorem 3.9 of the present survey. -- X-pointed formal moduli problems are equivalent to dg-Lie algebroids over X. A weaker version of this appears as Theorem 4.12 in the present paper. 3 Remark. These results are somehow contained in the recent book [8, 9] of Gaitsgory -- Rozenblyum, though in a slightly different formulation3. We finally observe that it would be very interesting to understand the results from [10], describing the algebroids on X. derived geometry of locally split first order thickenings X ֒→ S of a smooth scheme X, in terms of Lie Description of the paper §1 We present the basic notions necessary to understand Lurie -- Pridham result, relate them to more classical constructions in deformation theory and provide some examples. We finally give a glimpse of Lurie's approach for the proof. §2 A general framework for abstract formal moduli problems is given in details. In §2.1, we recall Lurie's deformation contexts. In §2.2, we deal with dual deformation contexts. In §2.3, we introduce the useful notion of Koszul duality context, which is a nice interplay between a deformation context and a dual deformation context. In §2.4 we talk about morphisms between these, which is a rather delicate notion. In §2.5 we restate some results of Lurie using the notion of Koszul duality context, making his approach a bit more systematic. In §2.6 we discuss tangent complexes. §3 We extend former results from dg-Lie algebras to dg-Lie algebroids. We prove Hennion's result [11] in §3.1. Most of the material in §3.2, §3.3 and §3.4 happens to be already contained in the recent preprints [26, 25] of Joost Nuiten. §3.5 presents some kind of base change functor that will be useful for functoriality in the next Section. §4 We explain how to globalize the results of §3. In §4.1 we introduce formal (pre)stacks and formal thickenings, mainly following [8, 9] and [5], and compare these with formal moduli problems. In §4.2 we state a consequence, for Lie algebroids, of the previous §, and we sketch an alternative proof that is based on Koszul duality contexts and morphisms thereof. In §4.3 we show that formal thickenings of X fully faithfully embed in Lie algebroids on X, and we conjecture that this actually is an equivalence. Acknowledgements The first author thanks Mathieu Anel, Benjamin Hennion, Pavel Safronov, and Bertrand Toen for numer- ous discussions on this topic. This survey paper grew out of lecture notes, taken by the second author, of a 3 hours mini-course given by the first author at the session DAGIT of the ´Etats de la Recherche, that took place in Toulouse in June 2017. We both would like to thank the organizers for a wonderful meeting, as well as the participants to the lectures for their enthusiasm. Damien Calaque acknowledges the financial support of the Institut Universitaire de France, and of the ANR grant "SAT" ANR-14-CE25-0008. Julien Grivaux acknowledges the financial support of the ANR grant "MicroLocal" ANR-15-CE40-0007, and ANR Grant "HodgeFun" ANR-16-CE40-0011. Credits -- §1 actually doesn't contain more than the beginning of [23], and §2 is somehow a nice re-packaging of the general construction presented in loc.cit. However, we believe the formalism of §2 is way more user-friendly for the reader wanting to apply [23] in concrete situations. -- Even though we thought they were new at the time we gave these lectures, Joost Nuiten indepen- dently obtained results that contain and encompass the material that is covered in §3 (see the two very nice papers [26, 25], that appeared while we were writing this survey). These results must be attributed to him. -- Global results presented in §4 rely on the general theory of formal derived (pre)stacks from [9] (see also [5]). Some of these results seem to be new. 3Actually, Gaitsgory and Rozenblyum define Lie algebroids as formal groupoids. Our results, as well as Nuiten's one, thus somehow indirectly give a formal exponentiation result for derived Lie algebroids. 4 Notation Below are the notation and conventions we use in this paper. M[W−1 -- Going from M to M is harmless regarding (co)limits: homotopy (co)limits in M correspond to associated with the model category qcat of (small) quasi-categories. stabilization (see §2.1), it is the category of spectra. -- Abusing notation, we will denote by the same letter an ordinary (i.e. discrete) category and its -- If M is a given model category, we write WM for its subcategory of weak equivalences, and M := M ] for the associated∞-category (unless otherwise specified, localization is always understood Categories, model categories and ∞-categories -- We make use all along of the language of∞-categories. We will only use (∞, 1)-categories, that is categories where all k-morphisms for k ≥ 2 are invertible. By∞-categories, we will always mean (∞, 1)-categories. associated∞-category. as the∞-categorical localization, that is Dwyer-Kan simplicial localization). ∞-categorical (co)limits in M. -- Conversely, any presentable∞-category can be strictified to a model category (i.e. is of the form M for some model category M). In the whole paper, we will only deal with presentable ∞-categories. It can be obtained as the ∞-category -- Let cat∞ be the ∞-category of (small) ∞-categories. The associated∞-category is denoted by sSet, it is equivalent to∞Grpd. We denote by spa its an adjunction as F : C −→←− D : G, F is the left adjoint. -- When writing an adjunction horizontally (resp. vertically), we always write the left adjoint above (resp. on the left ) and the right adjoint below (resp. on the right ). This means that when we write -- We denote by sSet the model category of simplicial sets (endowed with Quillen's model structure). Complexes -- The letter k will refer to a fixed field of characteristic zero. -- The category modk is the model category of unbounded (cochain) complexes of k-modules from [13] (it is known as the projective model structure), and modk is its associated∞-category. For this model structure, weak equivalences are quasi-isomorphisms, and fibrations are componentwise surjective morphisms. -- The category mod≤0 k is the model category of complexes of k-modules sitting in nonpositive degrees, and mod≤0 surjective morphisms in degree ≤ −1. k is its associated∞-category. For this model structure, fibrations are componentwise -- In the sequel, we will consider categories of complexes with an additional algebraic structure (like commutative differential graded algebras, or differential graded Lie algebras). They carry model structures for which fibrations and weak equivalences are exactly the same as the ones for complexes. It is such that the "free-forget" adjunction with modk (or, mod≤0 k ) is a Quillen adjunction. We refer the reader to the paper [13] for more details. Differential graded algebras -- The category cdgak denotes the model category of (unbounded) unital commutative differential graded k-algebras (that is commutative monoids in modk), and cdgak denotes its associated ∞-category. -- A (unbounded) unital commutative differential graded k-algebra will be called a "cdga". 5 -- The category Calgaug k is equivalent to the category Calgnu equivalence is actually a Quillen equivalence. is the slice category cdgak/k, i.e. the category of augmented k-algebras. It k of non-unital commutative differential graded algebras. This -- The category cdga≤0 k denotes the model category of cdgas over k sitting in nonpositive degree. k . -- For any A ∈ cdgak we write modA for the model category of left A-modules, and modA for its The associated∞-category is denoted by cdga≤0 associated∞-category. -- Note that modA is a stable ∞-category. This can be deduced from the fact that modA is a triangulated dg category. The looping and delooping functors Ω∗ and Σ∗ are then simply given by degree shifting (∗)[−1] and (∗)[1], respectively. -- We define cdgaA is then the model category of A-algebras (commutative monoids in modA), and cdgaA is the corresponding∞-category. -- All these definitions have relative counterparts: if A is a cdga and if B is an object of modA, then cdgaA/B is the slice category (cdgaA)/B of relative A-algebras over B. Differential graded Lie algebras and L∞-algebras -- We denote by Liek the model category of (unbounded) diffential graded Lie algebras over k (i.e. -- A differential graded Lie algebra over k will be called a "dgla". Lie algebra objects in modk). The associated∞-category is denoted by Liek. being∞-morphisms, with respect to∞-quasi-isomorphisms (see e.g. [31]). -- Remark that Liek is equivalent to the localization of the category of L∞-algebras, with morphisms 1 Introduction to pointed formal moduli problems for commu- tative algebras 1.1 Small augmented algebras For any dg-algebra A and any A-module M, we can form the square zero extension of A by M; we denote it by A ⊕ M where there is no possible ambiguity. We set first some crucial definitions for the rest of the paper: Definition 1.1. Recall that Calgaug k -- A morphism in Calgaug where k→ k ⊕ k[n] is the square zero extension of k by k[n]. -- A morphism in Calgaug is the∞-category of augmented cdgas. k is called small if it is a finite composition of elementary morphisms. is called elementary if it is a pull-back4 of k→ k ⊕ k[n] for some n ≥ 1, is called small if the augmentation morphism ǫ : A→ k is small. the full5 sub-∞-category of small objects of Calgaug . Small objects admits k k -- An object in Calgaug k We denote by Calgsm k various concrete equivalent algebraic characterizations: Proposition 1.2 ([23, Proposition 1.1.11 and Lemma 1.1.20]). An object A of Calgaug only if the three following conditions hold: k is small if and -- Hn(A) = {0} for n positive and for n sufficiently negative. -- All cohomology groups Hn(A) are finite dimensional over k. 4Since we are working in the ∞-categorical framework, pullback means homotopy pullback. 5Hence we allow all morphisms in the category Calgsm k , not only small ones. 6 -- H0(A) is a local ring with maximal ideal m, and the morphism H0(A)/m→ k is an isomorphism. Moreover, a morphism A→ B between small objects is small if and only if H0(A)→ H0(B) is surjective. Remark 1.3. Let us make a few comments on this statement, in order to explain its meaning and link it to classical results in commutative algebra and deformation theory. -- Observe that small algebras are nothing but dg-artinian algebras concentrated (cohomologically) in nonpositive degree. -- To get a practical grasp to the definitions of elementary and small morphisms, it is necessary to be . This is a tractable problem able to compute homotopy pullbacks in the model category Calgaug k since the model structure on cdgak is fairly explicit, and Calgaug is a slice category of cdgak. k -- If a small object A is concentrated in degree zero, the theorem says that A is small if and only if A is a local artinian algebra with residue field k. Let us explain concretely why this holds (the argument is the same as in the general case). If A is a local artinian algebra, then A can be obtained from the residue field as a finite sequence of (classical) small extensions, that is extensions of the form where R1, R2 are local artinian with residue field k, and (t) is the ideal generated by a single element t annihilated by the maximal ideal of R2 (hence it is isomorphic to the residue field k). In this way we get a cartesian diagram6 0→ (t)→ R2→ R1→ 0 R2 k k in Calgaug fibration. Therefore this diagram is also cartesian in Calgaug diagram of the form . Since the bottom horizontal map is surjective in each degree, it is in particular a , and is isomorphic to a cartesian k cone ((t)→ R2) / k ⊕ k[1] R2 R1 k / k ⊕ k[1] in Calgaug k . Hence the morphism R2→ R1 is elementary. -- An important part of classical deformation theory in algebraic geometry is devoted to formal deformations of algebraic schemes. For a complete account, we refer the reader to the book [29]. Following the beginning of [23], we will explain quickly how small morphisms fit in this framework. Given a algebraic scheme Z over k (that will be assumed to be smooth for simplicity), the formal deformation theory of Z deals with equivalence classes of cartesian diagrams Z Z Spec(k) / Spec(A) where A is a local artinian algebra with residue field k. This construction defines a deformation functor Def Z from the category of local artinian algebras to sets. It is possible to refine the functor Def Z to a groupoid-valued functor that associates to any A the groupoid of such diagrams. Since any groupoid defines a homotopy type (by taking the nerve), we 6The cdga structure we put on the cone is the natural one: we require that (t) has square zero. 7 / / / / / / / / / get an enriched functor DefZ from local artinian algebras to the homotopy category of topological spaces, such that π0(DefZ) = Def Z. The first important case happens when A = k[t]/t2. In this case, Kodaira-Spencer theory gives a bijection between isomorphism classes of deformations of X over Spec(k[t]/t2) and the cohomology group H1(Z, TZ). In other words, H1(Z, TZ) is the tangent space to the deformation functor Def Z. The next problem of the theory is the following: when can an infinitesimal deformation θ in H1(Z, TZ) be lifted to Spec(k[t]/t3)? The answer is: exactly when [θ, θ] vanishes in H2(Z, TZ). This can be interpreted in the framework of derived algebraic geometry as follows: the extension yields a cartesian diagram 0→ (t2)→ k[t]/t3→ k[t]/t2→ 0 k[t]/t3 k k[t]/t2 / k ⊕ k[1] in Calgaug k . This gives a fiber sequence of homotopy types hence a long exact sequence DefZ(k[t]/t3)→ DefZ(k[t]/t2)→ DefZ(k ⊕ k[1]) · · ·→ π0(DefZ(k[t]/t3))→ π0(DefZ(k[t]/t2))→ π0(DefZ(k ⊕ k[1]). It turns out that the set π0(DefZ(k ⊕ k[1]) of equivalence classes of deformations of Z over the derived scheme Spec(k ⊕ k[1]) is isomorphic to H2(Z, TZ). Hence the obstruction class morphism Def Z(k[t]/t2) ∼= H1(Z, TZ) −→ H2(Z, TZ) θ 7−→ [θ, θ] can be entirely understood by writing k[t]/t3→ k[t]/t2 as an elementary morphism. We start by introducing formal moduli problems in the case of cdgas: 1.2 The ∞-category of formal moduli problems Definition 1.4. A formal moduli problem (we write fmp) is an∞-functor X : Calgsm the following two properties: k → sSet satisfying -- X(k) is contractible. -- X preserves pull-backs along small morphisms. The second condition means that given a cartesian diagram in Calgsm k where A→ B is small, then is cartesian. N M A / B X(N) X(A) X(M) / X(B) 8 / / / / / / / / / Remark 1.5. Observe that the second condition is stable under composition and pullback. Hence it is equivalent to replace in this condition small morphisms with elementary morphisms. We claim that we Indeed, consider a cartesian diagram can even replace elementary morphisms with the particular morphisms k→ k ⊕ k[n] for every n ≥ 1. / A N where f is elementary, that is given by a cartesian diagram f / B M A f B / k / k ⊕ k[n] If we look at the diagram X(N) X(A) X(∗) X(M) / X(B) / X(k ⊕ k[n]) cartesian, so the left square is cartesian if and only if the big square is cartesian (which is the case). and assume that X preserves pullbacks along the morphisms k → k ⊕ k[n], then the right square is k → sSet is a fmp if and only if X(k) is contractible and preserves Corollary 1.6. A functor X : Calgsm pull-backs whenever morphisms in the diagram are surjective on H0. Proof. Assume that X preserves pull-backs whenever morphisms in the diagram are surjective on H0. Consider a cartesian diagram of the following type, with n ≥ 1: N k M / k ⊕ k[n] Since M is an augmented k-algebra, the map M→ k ⊕ k[n] is surjective on H0. Hence X(N) X(k) X(M) / X(k ⊕ k[n]) is cartesian. According to the preceding remark, this implies that X is a fmp. The reverse implication is obvious thanks to the last sentence of Proposition 1.2. problems. We write FMPk for the full sub ∞-category of Fun(Calgsm k , sSet) consisting of formal moduli 1.3 A glimpse at the description of FMPk In this section, we explain some heuristical aspects of the proof of the following theorem: Theorem 1.7 ([23], [27]). There is an equivalence of ∞-categories Liek→ FMPk. Remark 1.8. Again, we discuss various points in this theorem related to more classical material. 9 / / / / / / / / / / / / / / / / -- Let us first give a naive idea about how the∞-functor can be defined on a "sufficiently nice" dgla. The procedure is rather classical; we refer the reader to [15] and [16] for further details. If g is a dgla of over k, we can consider its (discrete) Maurer -- Cartan set MC(g) = {x ∈ g1 such that dx + [x, x]g = 0}. The "good" object attached to g is the quotient of MC(g) under gauge equivalence. This can be better formulated using a simplicial enrichment as follows: for any n ≥ 0, let Ω•(∆n) be the cdga of polynomial differential forms on the n-simplex ∆n. The collection of the Ω•(∆n) defines a simplicial cdga. Then we define the simplicial set MC(g) as follows7: This being done, we can attach to g a deformation functor Def g : Calgsm MC(g)n = MC(g ⊗ Ω•(∆n)). Def g(A) = MC(g ⊗k IA) → sSet defined by where IA is the augmentation ideal of A. This defines (again in good cases) a formal moduli problem. -- B The main problem of the Maurer -- Cartan construction is that the functor g→ Def g does not always preserve weak equivalences. For dglas that satisfy some extra conditions (like for instance nipoltence conditions), Def g will be exactly the fmp we are seeking for. We will see very soon how Lurie circumvents this problem using the Chevalley -- Eilenberg complex. -- To illustrate an example where the Maurer -- Cartan construction appears, let us come back to deformation theory of algebraic schemes in a slightly more differential-geometric context: instead of algebraic schemes we deform compact complex manifolds. We can attach to a complex compact manifold Z the Dolbeault complex of the holomorphic tangent bundle TZ, which is the complex 0→ C∞(TZ) ∂ −→ A0,1(TZ) ∂ −→ · · · We see this complex as a dgla over C, the Lie structure being given by the classical Lie bracket of vector fields and the wedge product on forms. Then it is well known (see e.g. [17, Lemma 6.1.2]) that deformations of Z over an artinian algebra A yield IA-points in the set-valued deformation functor associated with the dgla (A0,•(TZ), ∂). Let us now explain the good construction of the equivalence from Liek to FMPk. First we start by recalling the following standard definitions: Definition 1.9. For any dgla g, we define the homological and cohomological Chevalley -- Eilenberg complexes CE•(g) and CE•(g) as follows: -- As a graded vector space, CE•(g) = S (g[1]) is the (graded) symmetric algebra of g[1]. The differential is obtained by extending, as a degree 1 graded coderivation, the sum ot the differential g[1] → g[2] with the Lie bracket S2(g[1]) → g[2]. Jacobi identity and Leibniz rule ensure that this coderivation squares to zero. The complex CE•(g) is actually an (coaugmented, counital, and conilpotent) cocommutative coalgebra object in the category of complexes. -- CE•(g) is the linear dual of CE•(g), it is an augmented cdga. Remark 1.10. -- Observe that the above definition still makes sense for an L∞-algebra g. Indeed, an L∞-algebra structure on g is defined as a degree 1 graded codifferential that makes S (g[1]) a coaugmented counital cocommutative differential graded coalgebra. -- It is possible to prove that CE•(g) ≃ k L ⊗U(g) k and CE•(g) ≃ RHomU(g)(k, k). 7The π0 of MC(g) is exactly the quotient of MC(g) under gauge equivalence. 10 -- Let V be an object of modk, and let free (V) be the free dgla generated by V. Then CE•(free (V)) and CE•(free (V)) are quasi-isomorphic to the square zero extensions k ⊕ V[1] and k ⊕ V ∗[−1] respectively. -- For any cdga A and any dgla g of finite dimension over k, there is a map Homcdgak(CE•(g), A)→ MC(g ⊗ IA) which is in good cases an isomorphism (where IA is the augmentation ideal of A). To see how the map is constructed, let ϕ be an element in Homcdgak (CE•(g), A). Forgetting the differential, it defines an algebra morphism from the completed algebra bS (g∗[−1]) to A, and in fact to IA (since the morphism is pointed). In particular, we have a map φ : g∗[−1] → IA, and since g is finite-dimensional, we can see the morphism φ as a map k[−1] → g ⊗ IA, hence an element x of (g ⊗ IA)1. Now we have a commutative diagram φ dA (1) g∗[−1] [ . , . ]∗ g∗[−1] ⊗k g∗[−1] A φ⊗φ / A ⋆A / A ⊗k A Unwrapping what it means expressing φ with x, we end up exactly with the Maurer -- Cartan equation dx + [x, x] = 0. The reason why the map (1) is not always bijective is that CE•(g) is not the symmetric algebra of g∗[−1], but it is the completed symmetric algebra. -- There is a simplicially enrichement of (1), given by Apart from the completion issue that we have already discussed, this morphism may not be an equivalence as CE•(g) may not be a cofibrant object in cdgak. Homcdgak (CE•(g), A)→ MC(g ⊗ IA). Sketch of the construction of the equivalence. -- The Chevalley -- Eilenberg construction preserves weak equivalences, hence defining an∞-functor CE• : Lieop . The functor (CE•)op commutes with small colimits (see [23, Proposition 2.2.17]), so since Liek is presentable, CE• admits a left adjoint. We call this adjoint D. Hence we have an adjunction k k → Calgaug −→←− Lieop k : CE• D : Calgaug k that can be seen as some version of Koszul duality. The main point in this step is that the Chevalley -- Eilenberg functor does only commute with small homotopy limits, not usual small limits. Hence adjunction). the adjoint functor is only defined in the∞-categorical setting (i.e. it does not come from a Quillen -- We define an∞-functor from Liek to Fun(Calgaug k (cid:0)g, D(−)(cid:1) = HomLiek(cid:0)D(−), g(cid:1). ∆(g) = HomLieop , sSet) as follows: k The functor ∆ will define the equivalence we are seeking for. -- Let us explain why ∆ factors through FMPk. We introduce the notion of good object: a dg-Lie these morphisms appears in a pushout diagram algebra L is good if there exists a finite chain 0 = L0 → L1 → . . .→ Ln = L such that each of free k[−ni − 1] Li {0} / Li+1 11 / / / / O O / / / in Liek, or equivalently a pullback diagram Li+1 {0} Li / free k[−ni − 1] k . We denote by Liegd in Lieop k consisting of good objects. We see that good objects are formally the same as small ones in Lieop k , using the sequence of objects free k[−n − 1] instead of k ⊕ k[n]. This will be formalized using the various notions of deformation contexts developed in the next section. k the full subcategory of Lieop is an equivalence. This is the crucial technical input and will be proved in Proposition 2.30. By -- The next step consists of proving that if g is good, the counit morphism DCE•(g)→ g in Lieop purely formal arguments (see Proposition 2.22), this implies that the adjunction D −→←− CE• defines an equivalence of categories between Calgsm and Liegd k . k -- Using this, if we have a cartesian diagram k N k M / k ⊕ k[n] D(N) {0} D(M) / D(k ⊕ k[n]) where N and M are small, then is cartesian in Lieop k , and therefore ∆(g)(N) ∗ ∆(g)(M) / ∆(g)(k ⊕ k[n]) is also cartesian in sSet. This implies that ∆ is an object of FMPk. Hence ∆ factors through the category FMPk. 2 General formal moduli problems and Koszul duality 2.1 Deformation contexts and small objects In this section, we will explain how the notions of small and elementary morphism make sense in a broader categorical setting. We start by some general facts on∞-categories. -- If C is an ∞-category with finite limits8, its stabilization Stab(C) can be described as the ∞- sequence E = (En)n∈Z of pointed objects9 together with weak equivalences En → ΩEn+1, where category of spectrum objects (also called infinite loop objects) in C. An object of Stab(C) is a Ω := Ω∗ denotes the based loop functor: Ω(c) = ∗ × c ∗. We often write En = Ω∞−nE. 8This hypothesis will be always implicit in the sequel. It implies among other things that C has a terminal object ∗. 9The ∞-category of pointed objects is the coslice ∞-category ∗/C under the terminal object. 12 / / / / / / / / / / / / -- If C is stable, then Stab(C) is naturally equivalent to C via the map sending (En)n to E0. in C and c/C are themselves equivalent. -- If C is an ∞-category with finite limits and c is an object of C, then the stabilization Stab(c/C) of its coslice∞-category c/C is equivalent to Stab(C). Indeed, the∞-categories of pointed objects -- If C is an∞-category with finite limits and c is an object of C, then the stabilization Stab(C/c) of its slice ∞-category C/c is the category of spectrum objects in the ∞-category idc/(C/c) of sequences c→ d→ c such that the composition is the identity. -- If C = sSet, then Stab(C) is the∞-category spa of spectra (that is spectrum objects in spaces). Definition 2.1. A pair (C, E), where C is a presentable∞-category with finite limits and E is an object -- A morphism in C is elementary if it is a pull-back of ∗→ Ω∞−nE for n ≥ 1 (where ∗ is a terminal -- An object c is small if the morphism c→ ∗ is small. We let (C, E)sm be the full subcategory of C spanned by the small objects. When it is clear from the context, we may abuse notation and write Csm := (C, E)sm. -- A morphism in C is small if it can be written as a finite sequence of elementary morphisms. of Stab(C), is called a deformation context. Given a deformation context (C, E): object in the category C). Let us give two examples of deformation contexts: Example 2.2. If C = modk, which is already stable, then we have an equivalence In this context we will mainly consider the spectrum object E = (k[n + 1])n∈Z. modk −→ Stab(C) M 7−→ (M[n])n. Remark 2.3. Instead of working over the ground field k, we can work over an arbitrary cdga A. Then we can take C = modA and E = (A[n + 1])n∈Z in Stab(C). Example 2.4. The category Stab(Calgaug Calgaug a non-unital algebra R to R[−1] equipped with the trivial product; hence the equivalence ) is equivalent to modk. Indeed, we have an equivalence k (0 is the terminal nonunital cdga) sends k , and the based loop functor Ω0 in Calgnu k ≃ Calgnu k modk ) k −→ Stab(Calgaug M 7−→ (k ⊕ M[n])n∈Z. In this context we will mainly consider the spectrum object E = (k ⊕ k[n])n∈Z. cdgaA/A of A-augmented A-algebras, where A is in cdgak, is equivalent to modA. Moreover, if Remark 2.5. One can prove in a similar way that the stabilization Stab(cdgaA/A) of the ∞-category cdgak/A is the∞-category of A-augmented k-algebras, then Stab (cdgak/A) ≃ Stab (cdgaA/A) ≃ modA. In this case a natural spectrum object to consider is (A ⊕ A[n])n∈Z. We have the following obvious, though very useful, lemma, which allows to transfer deformation contexts along adjunctions: Lemma 2.6. Suppose (C, E) is a deformation context. If we are given an adjunction T ′ : C ′ −→←− C : T ,, then T preserves small limits, so that (C ′, T (E)) is a deformation context whenever C ′ is presentable. Besides, in this case T induces a functor from Csm to (C ′)sm. ✷ 13 Let us give four examples of this transfer principle10 Example 2.8. The relative cotangent complex functor LA/− : Calgaug Hence the deformation context from Example 2.2 and Remark 2.3 can be transferred to L/modA. Example 2.7. Given an A-module L, the push-out functor −`L 0 : L/modA→ modA along the zero map L→ 0 admits a right adjoint, being the functor sending an A-module M to the zero map L → M. A → modA admits a right adjoint, being the split square zero extension functor M 7→ A ⊕ M. Hence the deformation context from Example 2.9. The relative cotangent complex functor LA/− : cdgak/A→ Example 2.4 and Remark 2.5 can be obtained by transfer from the one given in Example 2.2 and Remark 2.3. M[−1] that it classifies. Hence the deformation context from Example 2.7 (with adjoint, being the functor sending a morphism LA/k zero extension A ⊕ d LA/k /modA admits a right d 0 Example 2.10. The forgetful functor Calgaug Hence the deformation context from Example 2.9 can also be obtained by transfer from the one given in Example 2.8. → M in modA to the non-necessarily split square A → cdgak/A is a right adjoint: its left adjoint is A ⊗ −. L = LA/k) can be transferred to cdgak/A. The main observation is that formal moduli problems make sense with Calgaug being replaced by k any deformation context (C, E). We write FMP(C, E) for the ∞-category of formal moduli problems associated with it. Then FMP(cid:0)Calgaug , (k ⊕ k[n])n(cid:1) = FMPk. k 2.2 Dual deformation contexts and good objects We introduce the dual concept of a deformation context: is called a dual deformation context. Definition 2.11. A pair (D, F), where D is a presentable∞-category and F is an object of Stab(Dop), B If (D, F) is a dual deformation context, (Dop, F) is not in general a deformation context because Dop is almost never presentable. Example 2.12. The first easy example is for D = modA. Nevertheless, this example is of high interest (see Lemma 2.16 below). Since modA is stable, so is modop A , and the looping and delooping functors are swapped when passing to the opposite category. Hence spectrum objects in modop A are of the form (M[−n])n for some object M. In particular, (cid:0)modA, (A[−n − 1])n(cid:1) is a dual deformation context. assume to be given a dual deformation context (D, F) as well as an adjunction T : D −→←− D ′ : ι. Then it is easy to see that (D ′, T (F)) is a dual deformation context on D ′ whenever D ′ is presentable, since T preserves colimits (hence limits in the opposite category). Using this, one can build a lot of dual deformation contexts starting from modA. Just like deformation contexts, dual deformation contexts can be transported using adjunctions: Example 2.13. For instance, the adjunction free : modA −→←− LieA : forget yields the dual deformation context (cid:0)LieA, free(A[−n − 1])n(cid:1). Example 2.14. Let T be an A-module. Then the pull-back functor − ×T 0 : modA/T → modA along the zero morphism 0 → T admits a left adjoint: → T . This yields the dual deformation context (cid:0)modA/T , (A[−n − 1] 0 10In all these examples, though, the use of transfer is not strictly necessary. Indeed, all the categories involved have a it is the functor sending an A-module S to the zero → T )n(cid:1). stabilization that is equivalent to modA. morphism S 0 14 Definition 2.15. Given a dual deformation context (D, F), an object (resp. morphism) of D is good if it is small when considered as an object (resp. morphism) of Dop. More explicitely, if ∅ denotes the initial element of D, an object b of D is good if there is a finite sequence of morphisms that are pushouts along Fn→ ∅ in D: i.e. each fi fits into a pushout square / bi Fn fm−−→ · · · f2−→ b1 f1−→ b0 ∅ = bm ∼= b fi ∅ / bi−1 in D for some n ≥ 1. We let (D, F)gd be the full subcategory of D spanned by the good objects, which we might simply denote Dgd if there is no ambiguity. Let us give a somehow nontrivial example. Let A be a k-algebra11, and consider the dual deformation context (cid:0)modA, (A[−n − 1])n(cid:1). Lemma 2.16. An element of modA is good for the dual deformation context (cid:0)modA, (A[−n − 1])n(cid:1) if and only if it is perfect and cohomologically concentrated in positive degrees. Proof. Recall that a perfect complex is the same as a dualizable object, which means a complex quasi- isomorphic to a finite complex consisting of projective A-modules of finite type. Let K be a perfect complex concentrated in positive degree and prove, by induction on the amplitude, that K is good. If the amplitude is 0 then K is quasi-isomorphic to P[−n], for n ≥ 1, with Ar = P ⊕ Aq . We have the following push-out square in modA: Aq[−n − 2] / 0 0 / Aq[−n − 1] Hence Aq[−n − 1] is good. We then also have another push-out square: Ar[−n − 1] / Aq[−n − 1] 0 / P[−n] well. Hence the morphism Aq[−n − 1]→ P[−n] is good, and since Aq[−n − 1] is good, then P[−n] is good as Performing the induction step is now easy: let K be a positively graded complex of finitely generated projective A-modules of some finite amplitude d > 0, let n > 0 be the index where K starts and let P = Kn. We have a push-out square: P[−n − 1] / τ>nK 0 / K where τ>n is the stupid truncation functor, and using again that Ar = P ⊕ Aq we get another push-out square: Ar[−n − 1] / Aq[−n − 1] ⊕ τ>nK 11A is concentrated in degree 0. 0 / K 15 / / / / / / / / / / and Aq[−n − 1] is good as well, thus so is K. This finishes the induction step. Hence the morphism Aq[−n−1]⊕τ>nK→ K is good. But τ>nK is good (by induction on the amplitude) For the converse statement, it suffices to observe that, given a push-out square A[−n − 1] 0 K / L in modA, where n ≥ 1, and K is perfect and concentrated in positive degrees, then so is L. . Similarly, one can prove that small objects for the deformation context (cid:0)modA, (A[n + 1])n(cid:1) are perfect complexes of A-modules cohomologicaly concentrated in negative degrees. We leave it as an exercise to the reader. Remark 2.17. If we replace the k-algebra A by a bounded cdga concentrated in non-positive degrees, then one can still prove that good objects are perfect A-modules that are cohomologically generated in positive degree. In other words, a good object is quasi-isomorphic, as an A-module, to an A-module P having the following property: as a graded A-module, P is a direct summand of A ⊗ V, where V is a finite dimensional positively graded k-module. 2.3 Koszul duality contexts We now introduce the main notion that is needed to state Lurie's theorem on formal moduli problems in full generality: Definition 2.18. (1) A weak Koszul duality context is the data of : -- a deformation context (C, E), -- a dual deformation context (D, F), (2) A Koszul duality context is a weak Kozsul duality context satisfying the two additional properties: such that for every n ≥ 0 there is an equivalence En ≃ D ′Fn. -- an adjoint pair D : C −→←− Dop : D ′, -- For every good object d of D, the counit morphism DD ′d→ d is an equivalence. -- The functor is conservative and preserves small sifted colimits. Θ : HomD(Fn, −) : D→ Stab(sSet) = spa With a slight abuse of notation, we will denote the whole package of a (weak) Koszul duality context by D : (C, E) −→←− (Dop, F) : D ′ Remark 2.19. The name Kosul duality context has been chosen in agreement with the first non-trivial example we will deal with (see Proposition refopinel), as it reflects the well-known Koszul duality between commutative and Lie algebras. Similarly, the characterization of associative formal moduli problems [23, §3], resp. En formal moduli problems [23, §4], can be proven with the help of a Koszul duality context that reflects the Koszul duality for associative algebras, resp. En-algebras. We actually expect that a pair (O, P) of Koszul dual operads always lead to a Koszul duality context between O-algebras and P-algebras, allowing to show that formal moduli problems for O-algebras are P-algebras. Observe that there may be more objects of D for which the counit morphism is an equivalence, than just good objects. We call them reflexive objects. 16 / / / Example 2.20. The only elementary Koszul duality context we can give at this stage is the following: if A is a bounded cdga concentrated in nonpositive degrees, (−)∨ = HomA(−, A) : (cid:0)modA, (A[n + 1])n(cid:1) −→←−(cid:0)modop To prove that it is indeed a Koszul duality context, we use Lemma 2.16 and Remark 2.17: a good object in modop A is a perfect A-module (generated in non-negative degrees), so it is isomorphic to its bidual. Lastly, the functor Θ is simply the forgetful functor A , (A[−n − 1])n(cid:1) : HomA(−, A) = (−)∨. modA→ modk ≃ Stab(modk)→ Stab(sSet) ≃ spa which is conservative. Note that there are strictly more reflexive objects than good ones: for instance, ⊕n≥0A[−n] is reflexive, but not good (it is an example of an almost finite cellular object in the termi- nology of [11]). There is a peculiar refinement of the above example, that will be useful for later purposes. Example 2.21. Let A be a bounded cdga concentrated in nonpositive degrees, and let L be an A-module. We have a weak Koszul duality context12 (−)∨ : (cid:0)L/ modA, (L 0 k → A[n + 1])n(cid:1) −→←−(cid:0)(modA/L∨ )op, (A[−n − 1] 0 −→ L∨ to the composed morphism L → (L∨)∨ k∨ → L∨)n(cid:1) : (−)∨ , Lastly, the functor Θ is the composition of the pull-back functor − × L∨ where the right adjoint sends K that this is actually a Koszul duality context. Indeed, one first observes that good objects are given by −→ K∨. We claim morphisms K→ L∨ where K is a good object in modA. Hence they are again isomorphic to their biduals. 0 : (modA/L∨ )op→ modA along the zero morphism 0→ L∨, with the forgetful functor modA→ spa from the previous example. They There are a few properties that can be deduced from the definition, which are absolutely crucial. are both conservative, hence Θ is. Proposition 2.22 ([23, Proposition 1.3.5]). Assume that we are given a Koszul duality context (A) For every n ≥ 0, DEn ≃ Fn. D : (C, E) −→←− (Dop, F) : D ′ (B) For every small object M in C, the unit map M→ D ′D(M) is an equivalence. (C) There is an equivalence of categories (C, E)sm / (Dop, F)gd. D / D ′ (D) Consider a pullback diagram N M A f / B where f is small and M is small. Then the image of this diagram by D is still a pullback diagram. 12Observe that, contrary to what one could think, it is not required that L is perfect, or even just reflexive. Indeed, for ℓ → M and K k → L∨, the following commuting diagrams completely determine each other: L f∨ M∨ and }④④④④④④④④ ℓ∨ K k L∨ L ℓ M (L∨)∨ k∨ f / K∨ 17 o o / / / / / } / / / Proof. The proof is clever but completely formal, and doesn't require any extra input. (A) This is straightforward: DEn ≃ DD ′Fn ≃ Fn since Fn is good. (B) We proceed in two steps. First we prove it if M is for the form D ′(N) for some good object N. This is easy: the composition D ′ ∼= id ◦ D ′⇒ (D ′ ◦ D) ◦ D ′ ∼= D ′ ◦ (D ◦ D ′)⇒ D ′ ◦ id ∼= D ′ equivalence. is an equivalence (it is homotopic to the identity). This means that the composition D ′D(D ′N)→ D ′N is obtained up to equivalence by applying D ′ to the equivalence DD ′(N) → N, hence it is also an Now we prove that every small object can be written as D ′(X) where X is good. This will in particular imply that D maps small objects to good objects13. Every small morphism can be written as a composition of a finite number i of elementary morphisms, we will argue by induction on the number i. If i = 0, then since D ′ is a right adjoint it preserves limits. Hence D ′(∅) = ∗, where ∅ is the initial element of D. Since ∅ is good, we are done. We now deal with the induction step. Let us consider a cartesian diagram Mi+1 / ∗ Mi / En in C such that Mi → ∗ is obtained by composition of at most i elementary morphisms. By induction, Mi+1 = D ′(N) for some good object N. This implies that D(Mi) is good. Let X be the pullback the object of D making the diagram X ∗ cartesian in Dop. Since D(Mi) is good and D(En) ≃ Fn, X is good. Now we apply D ′, which preserves limits. We get a cartesian diagram isomorphic to D(Mi) / D(En) D ′(X) ∗ Mi / En This proves that Mi+1 is isomorphic to D ′(X). (C) This is a direct consequence of (B). (D) We can reduce to diagrams of the form N ∗ M / En Then the property follows from (C). 2.4 Morphisms of (weak) Koszul duality contexts We will now introduce the notion of morphisms between weak Koszul duality contexts. extremely useful in the sequel. It will be 13This is nontrivial: D preserves colimits, and small objects are defined using limits. 18 / / / / / / / / / / / Definition 2.23. Let and be two weak Koszul duality contexts. D1 : (C1, E1) −→←− (Dop D2 : (C2, E2) −→←− (Dop 1 , F1) : D ′ 1 2 , F2) : D ′ 2 A weak morphism between these two duality contexts consists of two additional pairs of adjoint functors appearing (vertically) in the diagram C1 S T C2 D1 D ′ 1 D2 D ′ 2 / Dop 1 Y Z / Dop 2 such that: (A) The diagram consisting of right adjoints C1 T C2 commutes. D ′ 1 D ′ 2 Dop 1 Z Dop 2 (B) We have equivalences Z(F2,n) ≃ F1,n. Example 2.24. Let L be an A-module. We then have the following weak morphism from the Koszul duality context of Example 2.21 to the one of Example 2.20: L/modA (−)∨ (−)∨ / (modA/L∨ )op cofib M7→(L 0 →M) fib K7→(K 0 →L∨) modA (−)∨ (−)∨ / modop A where cofib = 0`L − and fib = 0 × L∨ −. Notice that the natural equivalence we define pictorially as the composition from condition (A) above has a mate θ which where id the mate θ is the composition SD ′ and id are used to depict units and counits of vertical adjunctions. In other words, Remark 2.25. The commutativity of the square of right adjoints implies the commutativity of the square of left adjoints. Hence there is a natural equivalence D2S ∼= YD1 that pictorially reads , and the mate θ can also be identified with the following composition: 1ZY ∼= ST D ′ 1⇒ SD ′ 2Y. 2Y⇒ D ′ 2D2SD1 ∼= D ′ 2YD1D ′ or, pictorially . SD ′ 1⇒ D ′ 2Y 1⇒ D ′ Definition 2.26. One says that the above commuting square of right ajoints satisfies the Beck -- Chevalley condition locally at d, where d is an object of D1, if the mate θd : SD ′ 2Y(d) is an equivalence. Definition 2.27. A morphism of weak Koszul duality contexts is a weak morphism such that, borrowing the above notation: 1(d)→ D ′ 19 / o o O O / o o O O o o O O O O o o / o o O O / o o O O (C) The functor Y is conservative, preserves small sifted limits, and sends good objects to reflexive objects. (D) The commuting square of right adjoints satisfies the Beck -- Chevalley condition locally at good objects. The main feature of this definition is a result allowing to transfer Koszul duality contexts along morphisms: Proposition 2.28 (Transfer theorem). Assume to be given a morphism between two weak Koszul duality contexts. If the target deformation context is a Koszul duality context, then the source is also a Koszul duality context. Proof. We take the notation of Definition 2.23. We have two functors Θ1 and Θ2, defined on D1 and D2 respectively and with values in spa, defined by (cid:14)Θ1(d)n = HomD1 (F1,n, d) Θ2(d)n = HomD2 (F2,n, d) For any object d in D1, we have Θ1(d)n = HomD1 (F1,n, d) ≃ HomDop 1 (d, Z(F2,n)) (Y(d), F2,n) ≃ HomD2 (F2,n, Yop(d)) ≃ HomDop ≃ Θ2(Yop(d))n 2 So Θ1 = Θ2 ◦ Yop, and thus Θ1 is conservative (because Y and Θ2 are). Moreover, Y preserves small sifted limits, so Yop preserves small sifted colimits, and hence so does Θ1 (because Θ2 does too). Lastly, we consider the following diagram in Dop 2 , where d is still an object in D1: YD1D ′ 1 ∼ D2SD ′ 1 Y◦ co-unit 3 Y co-unit ◦Y D2◦θ 3 D2D ′ 2Y. We claim that it commutes. Indeed, writing the mate explicitely gives a diagram Y◦ co-unit YD1D ′ 1 ∼ i ❏❏❏❏❏❏❏❏❏❏ ❏❏❏❏❏❏❏❏❏❏ D2SD ′ 1 co-unit ◦YD1D ′ 1 D2D ′ 2Y 3 Y co-unit ◦Y D2◦ unit ◦SD ′ 1 D2D ′ 2D2SD ′ 1 k ∼ D2D ′ 2Y◦ co-unit 3 D2D ′ 2YD1D ′ 1 ❏❏❏❏❏❏❏❏❏❏❏ ❏❏❏❏❏❏❏❏❏❏❏ where the two triangles commute. Given now a good object d in D1, we have that Y(d) is reflexive (after (C) in Definition 2.27): thus the co-unit morphism of the adjunction (D2, D ′ 2) is an equivalence on Y(d). Besides, θd is also an equivalence (thanks to (D) in Definition 2.27). Therefore Y ◦ co-unit(d) is an equivalence and thus, by conservativity of Y, the co-unit morphism D1D ′ Example 2.29. Going back to Example 2.24, one sees that the weak morphism is a morphism if and only if L is reflexive (in which case, L∨ is so as well). Indeed: 1(d)→ d is an equivalence. -- The functor fib is conservative. -- If K→ L∨ is good then K itself is good, thus perfect, in modA, and thus fib(K→ L∨) is reflexive as soon as L∨ is so. 20 K S + + K S K S + K S s + K S a -- the mate θK→L∨ is the natural morphism cofib(L→ K∨)→ fib(K→ L∨)∨, which is an equivalence if and only if the unit morphism L→ (L∨)∨ is an equivalence. Hence the Koszul duality context from Example 2.21 is in general not obtained by transfer from the one of 2.20, but it is in the case when L is reflexive. Proposition 2.30. The adjunction D : (Calgaug k defines a Koszul duality context. , k ⊕ k[n]) −→←− (Lieop k , free k[−n − 1]) : CE• Proof. We consider the diagram (see Remark 1.10): Calgaug k CE• V7→k⊕V[−1] modk Hom(−,k) Lieop k free modop k We can fill it with left adjoints everywhere. This gives the following (nice!) diagram: Calgaug k D CE• / Lieop k L V7→k⊕V[−1] forget free modk Hom(−,k) Hom(−,k) / modop k where L(R) = Lk/R ≃ LR/k ⊗R k[1]. We claim that these four adjunctions define a morphism of weak Koszul duality contexts. Properties (A) and (B) are true, so that we have a weak morphism. We will now prove that it is actually a morphism, so that we get the result, using Proposition 2.28 and the fact that the bottom adjunction is a Koszul duality context (Example 2.20). The functor forget is conservative. Let us now prove that, if g is good, then it is equivalent to a very good dgla: a good dgla with underlying graded Lie algebra being generated by a finite dimensional graded vector space sitting in positive degrees. Lemma 2.31. Any good dgla g is quasi-isomorphic to a very good one. Proof of the lemma. We first observe that 0 is very good. We then proceed by induction: assume that g is very good, and consider a dgla g ′ obtained by a pushout free k[−n − 1] g 0 / g ′ in Liek. The first trick is that free k[−n − 1] is cofibrant and g is fibrant (every object in Liek is). Hence the morphism free k[−n − 1] in the ∞-category Liek can be represented by a honest morphism in Liek. The next step consists in picking a cofibrant replacement of the left vertical arrow. A cofibrant replacement is given by the morphism free k[−n − 1]→ free (cid:10)cone (k 21 id −→ k)[−n − 1](cid:11) . o o O O o o O O / o o O O / o o O O / / / Hence our pushout in Liek is represented by the following honest non-derived pushout in Liek: free k[−n − 1] free (cid:10)cone (k id −→ k)[−n − 1](cid:11) g / g ′ This pushout is obtained by making the free product of g and free (cid:10)cone (k taking the quotient by the image of the ideal generated by free k[−n − 1]. This completes the induction step: g ′ is still very good. −→ k)[−n − 1](cid:11), and then by id This in particular shows that forget sends good objects to reflexive objects for the dual deformation context (modk, k[−n − 1]). Lastly, thanks to [23, Proposition 2.1.16], the forgetful functor from Lieop k to modop k preserves small sifted limits. Thus (C) holds. It remains to prove (D), which is the main delicate point of the proof. Recall for that purpose that, for a dgla g, θg is defined as the following composition, where we omit the forget functor (its appearance being obvious): L(cid:0)CE•(g)(cid:1)→ L(cid:0)CE•(free g)(cid:1)→ L(k ⊕ g∗[−1])→ g∗ . At this point it is important to make a rather elementary observation: the composed cdga morphism CE•(g)→ CE•(free g)→ k ⊕ g∗[−1] is nothing but the projection onto the quotient by the square I2 of the augmentation ideal I = ker(CE•(g)→ k) whenever g is very good14. We now introduce the uncompleted Chevalley -- Eilenberg cdga fCE (g) as a sub-cdga S(g∗[−1]) ⊂ CE•(g) (it is obviously a graded subalgebra, and it can be easily checked as an exercise that it is stable under the differential). The quotient by the square of its augmentation ideal is still k ⊕ g∗[−1], and we have the following commuting diagram and its image through the left-most vertical adjunction of our square: • (g) • fCE ιg %▲▲▲▲▲▲▲▲▲▲▲ CE•(g) / k ⊕ g∗[−1] • (g)(cid:1) #●●●●●●●●●● L(cid:0)fCE L(cid:0)CE•(g)(cid:1) L(ιg) eθg θg / g∗ Let us start with the following lemma: Lemma 2.32. Let A be an augmented cdga, with augmentation ideal J, that is cofibrant as a cdga. Then In order to prove that θg is an equivalence when g is good, we will prove that both eθg and L(ιg) are. the morphism L(A)→ J/J2[1] associated with the projection A→ k ⊕ J/J2 is an equivalence. Proof. First of all, the projection A→ k ⊕ J/J2 is an actual morphism in the category cdgak, so that we have a factorization where: L(A) ≃ LA/k ⊗A k[1]→ Ω1 A/k ⊗A k[1]→ J/J2[1] , -- The morphism LA/k ⊗A k→ Ω1 -- The second morphism is an isomorphism in modk. A/k ⊗A k is an equivalence because A is cofibrant. The lemma is proved. 14Indeed, in this case the natural map Sk(g∗[−1]) → Sk(g[1])∗ is an isomorphism. Thus, forgetting the differential, CE•(g) = ^S(g∗[−1]). 22 / / / % / # / We then observe that when g is very good (which we can always assume without loss of generality when dealing with good dglas), then fCE Lastly, it can be shown that CE•(g) is flat over fCE • (g) is cofibrant and thus eθg is an equivalence15. (g), which implies that the natural map • is also an equivalence, which shows (after applying − ⊗CE•(g) k) that L(ιg) is an equivalence. L fCE • (g)/k ⊗ fCE • (g) CE•(g) −→ LCE•(g)/k 2.5 Description of general formal moduli problems Assume that we are given a Koszul duality context D : (C, E) −→←− (Dop, F) : D ′ We can define a functor Ψ : D→ Fun(C, sSet) by the composition −→ Fun(Dop, sSet) ◦ D where Y is the Yoneda functor d→ HomDop(d, −). Theorem 2.33 (Lurie [23]). Given a Koszul duality context Ψ : D Y −−→ Fun(C, sSet). The functor Ψ factors through FMP(C, E) and the induced functor D : (C, E) −→←− (Dop, F) : D ′, Ψ : D→ FMP(C, E) is an equivalence. Sketch of proof. The first point is a direct consequence of Proposition 2.22 (D). The proof that Ψ is an equivalence proceeds on several steps. -- The first step consists of proving that Ψ is conservative. This follows almost immediately from the have Ψ(X) = HomD(D(∗), X) and similarly for Y. Since all En are small and D(En) ≃ Fn, f induces an equivalence of spectra hypotheses. Indeed, let f : X→ Y an arrow in D inducing isomorphic formal moduli problems. We Since HomD(Fn, −) : D→ spa is conservative, f is an equivalence. -- The next step consists in proving that Ψ commutes with limits and accessible colimits, so it has a left adjoint. This part is elementary, and uses the same techniques as in Proposition 2.22. We denote this left adjoint by Φ. HomD(Fn, X) ≃ HomD(Fn, Y). -- The functor Ψ being conservative, it suffices to prove that the unit idFMP(C,E) ⇒ Ψ ◦ Φ is an equivalence. This is the most technical part in the proof: it involves hypercoverings to reduce to pro-representable moduli problems; this is where the condition on sifted colimits plays a role. Here pro-representable means small limit of fmp representables by small objects. 15Here is another approach, avoiding the use of very good models. If g is a dgla then there is an L∞-structure on H∗(g) that makes it equivalent to g in Liek. If moreover g is good then H∗(g) is finite dimensional and concentrated in positive degree, so that the L∞-structure has only finitely many non-trivial structure maps. Thus the uncompleted Chevalley-Eilenberg cdga fCE •(cid:0)H∗(g)(cid:1) can be defined, is cofibrant, and is equivalent to fCE (g). • 23 -- In the representable case, we can explain what happens: we have an adjunction diagram Φ Ψ D a❈❈❈❈❈❈❈❈ Dop Cop / FMP(C, E) 9rrrrrrrrr Y that is on (Csm)op. Indeed, for any objects c and d of Csm and D respectively, where Y is the Yoneda functor c → HomC(c, −). We claim that Φ ◦ Y ≃ Dop on small objects, HomD(Φ(Yc), d) ≃ HomFMP(C,E)(Yc, Ψ(d)) = HomFun(Csm,E)(Yc, Ψ(d)) = ψ(d)(c) = HomDop (d, D(c)) = HomD(Dop(c), d) Using this, the unit map of Yc is given by the unit map of the adjunction between D and D ′ via the natural equivalence Yc→ Ψ ◦ Φ(Yc) ≃ Ψ(Dop(c)) = HomDop(D(c), D(−)) ≃ HomC(c, D ′D(−)). Using Proposition 2.22 (B), we see that this map is an equivalence of formal moduli problems. 2.6 The tangent complex In this section, we introduce the tangent complex associated to a formal moduli problem. We start with a very general definition. Definition 2.34. Let (C, E) be a deformation context. For any fmp X in FMP(C, E), its tangent complex TX is the spectrum X(E) = X(En)n. Remark 2.35. There is a slight subtelty in the definition of TX, as X(En) is only defined for n ≥ 0. However, this suffices to define it uniquely as a spectra, by putting (TX)−m := Ωm ∗ (cid:0)(TX)0(cid:1). Assume now to be given a Koszul duality context Then we have the following result: D : (C, E) −→←− (Dop, F) : D ′. Proposition 2.36. The following diagram commutes: D Ψ FMP(C, E) !❇❇❇❇❇❇❇❇❇ Θ spa yssssssssss T Proof. This is straightforward: given an object d in D, Ψ(d) = HomDop(d, D(∗)), so TΨ(d) = HomDop (d, D(En)) = HomD(Fn, d) = Θ(d). Remark 2.37. If D is k-linear (resp. A-linear), then Θ actually lifts to modk (resp. modA): replacing HomD(Fn, −) by its enriched version HOMD(Fn, −) gives us the lift of Θ. indeed, 24 / o o 9 a / / ! y Example 2.38. Going back to Example 2.20 we get that, for an A-module M, Θ(M) = HommodA (A[−n − 1], M)n . In the modA-enriched version, we have Θ(M) = (M[n+1])n which gives Θ(M) ≃ M[1] via the canonical identification Stab(modA) ≃ modA. We now deal with functoriality: Proposition 2.39. Assume to be given a weak morphism C1 S T C2 D1 D ′ 1 D2 D ′ 2 / Dop 1 Y Z / Dop 2 between two Koszul duality contexts. Then there is an induced commuting diagram ∼ ∼ D1 Y op D2 FMP (C1, E1) ◦T spa T )❘❘❘❘❘❘❘❘❘❘❘❘ 5❧❧❧❧❧❧❧❧❧❧❧❧❧ T / FMP (C2, E2) (1) First, for every n ≥ 0: T (En,2) ≃ T D ′ In particular, the functor FMP (C1, E1)→ FMP (C2, E2) is conservative. Proof. We begin by proving that T ∗ := − ◦ T : Fun(C1, sSet)→ Fun(C2, sSet) indeed defines a functor FMP (C1, E1)→ FMP (C2, E2): (2) Then, T being a right adjoint it preserves in particular pull-backs along ∗→ En,2, and thus send them to pull-backs along ∗→ En,1 whenever n ≥ 0. Hence it sends small objects to small objects. is an fmp), and F ◦ T preserves pull-back along ∗→ En,2 (thanks to the second point and that F is (3) Finally, let F be an fmp for (C1, E1). Then F ◦ T (∗) ≃ F(∗) ≃ ∗ (because T is a right adjoint and F Note that Z also sends good objects to good objects (the proof is the same as for the second point above). We now come to the proof of the commutativity of the square, which is essentially based on the 1Z(Fn,2) ≃ D ′ 1(Fn,1) ≃ En,1. 2(Fn,2) ≃ D ′ an fmp). following Lemma: Lemma 2.40. There is a natural transformation D1T ⇒ ZD2 that is an equivalence on small objects. Proof of the Lemma. The commutativity of the square of right adjoints tells us there is a natural equiv- alence T D ′ 2 ∼= D ′ 1Z. Hence we have a mate D1T ⇒ ZD2 defined as the composition ∼= D1D ′ 2D2 D1T ⇒ D1T D ′ 1ZD2⇒ ZD2 , . We then observe that 2D2 is an equivalence. that can also be depicted as -- on small objects, the unit id⇒ D ′ and Z preserves the goods). -- ZD2 sends small objects to good objects (D2 realizes an equivalence between smalls and goods, -- on good objects, the co-unit D1D ′ 1⇒ id is an equivalence. 25 / o o O O / o o O O / / ) / 5 Hence the mate D1T ⇒ ZD2 is an equivalence on small objects. of small objects): The commutativity of the square then reads as follows (recall that we are reasoning on the category T ∗Ψ1 = HomDop ≃ HomDop ≃ HomDop = Ψ2Yop 1 (cid:0)−, D1T (−)(cid:1) 1 (cid:0)−, ZD2(−)(cid:1) 2 (cid:0)Y(−), D2(−)(cid:1) (by definition). (by definition) (using the mate) (by adjunction) Finally, we have to prove that the triangle commutes. This is obvious: X ◦ T (En,2) ≃ X(En,1). Example 2.41. Let us see what it implies for our preferred morphism of Koszul duality context: Calgaug k D CE• / Lieop k L V7→k⊕V[−1] forget free modk Hom(−,k) Hom(−,k) / modop k Note that we are in the k-linear situation, hence viewing the tangent complex as an object in modk. Let ϑ denote the functor V 7→ k ⊕ V[−1]. If X is in FMPk, then X ◦ ϑ belongs to FMP(cid:0)modk, (k[n + 1])n(cid:1), and thus TX = TX◦ϑ. Now: TX◦ϑ = Θ2Φ2(X ◦ ϑ) = Φ2(X ◦ ϑ)[1] = forget Φ1(X)[1] . Hence we obtain the following beautiful result, explaining one of the phenomenon mentionned in the introduction: The underlying complex of the dgla gX := Φ1(X) attached to X is TX[−1]. The following proposition describes the tangent complex of a representable moduli problem: Proposition 2.42. Let A be a representable element in FMPk. Then its tangent complex TA, considered in modk, is equivalent to TA ⊗A k, with TA := (LA/k)∨. Proof. It is a simple calculation: (TA)n := HomCalgaug k (A, k ⊕ k[n]) ≃ Hommodk (Lk/A, k[n + 1]) ≃ Hommodk (LA/k ⊗A k[1], k[n + 1]) ≃ TA ⊗A k[n] Hence, using the canonical identification Stab(modk) ≃ modk we get that TA ≃ TA ⊗A k. As a consequence, we get that Tk/A ≃ TA ⊗A k[−1] carries a dgla structure. 3 DG-Lie algebroids and formal moduli problems under Spec(A) In this part, A will denote a fixed cdga over k that is concentrated in non-positive degree and cohomo- logically bounded. In what follows the boundedness hypothesis is important; we will explain precisely where it has to be used. 26 / o o O O / o o O O 3.1 Split formal moduli problems under Spec(A) One of the main purpose of the work [11] (in the local case) is to prove that the equivalence provided by Theorem 2.22 can be extended when replacing the ground field k by A. We have (see Lemma 2.4) Stab (cdgaA/A) ≃ modA. We consider the deformation context (cid:0)cdgaA/A, (A ⊕ A[n])n(cid:1). Then Hennion's result runs as follows: Theorem 3.1 (Hennion [11]). If A is a cohomologically bounded cdga concentrated in non-positive degree, there is an isomorphism where FMPA/A := FMP(cid:0)cdgaA/A, (A ⊕ A[n])n(cid:1). FMPA/A ≃ LieA , Remark 3.2. In [11], there is the additional assumption that H0(A) is noetherian, but it does not appear to be necessary. Hints of proof. The strategy is to produce a Koszul duality context (cdgaA/A, A ⊕ A[n]) −→←− (Lieop A , free A[−n − 1]) and then to apply Theorem 2.33. The Chevalley -- Eilenberg construction can be performed over A, but Concretely, assuming that A is cofibrant, the Chevalley Eilenberg is defined in the usual way on A-dg-Lie algebras that are projective as A-modules. Since every element in LieA is isomorphic to a dg-Lie algebra whose underlying A-module is projective, we get a functor CE• left adjoint DA. We consider again a diagram of the form we must take its derived version (because the functor M→ SA(M) does not respect weak equivalences). A → cdgaA/A, and it admits a A : Lieop cdgaA/A DA CE• A / Lieop A LA V→A⊕V[−1] forget free modA HomA(−,A) HomA(−,A) / modop A where LA(R) = LA/R and try to construct a morphism between weak Koszul duality contexts. This works exactly as in the proof of Proposition 2.30, but some aspects have to be taken care of in condition (C) of morphisms between weak Koszul duality contexts: the fact that theforgetful functor commutes with with sifted limits is explained in the proof of [11, prop. 1.2.2]. On the other hand, the cohomological boundedness of A is crucial, otherwise the forgetful functor (although conservative) wouldn't send good objects to reflexive objects. More precisely, even if Lemma 2.31 still holds in this context16, very good dglas over A are not necessarily reflexive as A-modules (for instance if A is not a bounded k-algebra, then free A[−2] is not quasi-isomorphic to its double dual as a complex of A-modules). But they are whenever A is bounded. All results of the previous Section remain true if one replaces k with a bounded A. For instance, we have the following analog of Proposition 2.42: Proposition 3.3. Let B be a representable element in FMPA/A. Then its tangent complex TB, consid- ered in modA, is equivalent to TB/A ⊗B A, with TB/A := (LB/A)∨. ✷ 16A very good dgla over A is a good dgla over A such that -- the underlying A-module is projective. -- the underlying graded Lie algebra is freely generated over A by finitely many generators in positive degree. 27 / o o O O / o o O O In particular if B = A ⊗k A, with A-augmentation being given by the product, then LA⊗kA/A ⊗A⊗kA A ≃ LA and thus TA⊗kA ≃ TA. Corollary 3.4. The A-module TA[−1] is a Lie algebra object in modA. In [11], Hennion proves global versions of the above results, and shows in particular that if X is an algebraic derived stack locally of finite presentation, then TX[−1] is a Lie algebra object in QCoh(X). This again explains (and generalizes) a phenomenon that we mentioned in the introduction. 3.2 DG-Lie algebroids In this section, we introduce the notion of dg-Lie algebroids, which is the dg-enriched version of Lie algebroids. This will be used to construct a Koszul duality context for A-augmented k-algebras in the next section. Informally, a (dg)-Lie algebroid is a Lie algebra L over k such that L is an A-module, A is a L-module, both structures being compatible. More precisely: Definition 3.5. A (k, A)-dg-Lie algebroid is the data of a dgla L over k endowed with an A-module structure, as well as an action of L on A, satisfying the following conditions: -- L acts on A by derivations, meaning that the action is given by a A-linear morphism of k-dglas ρ : L→ Derk(A), called the anchor map. -- The following Leibniz type rule holds for any a ∈ A and any ℓ1, ℓ2 ∈ L: [ℓ1 , aℓ2] = (−1)a×ℓ1 a[ℓ1 , ℓ2] + ρ(ℓ1)(a)ℓ2 Morphisms of (k, A)-dg-Lie algebroids are A-linear morphisms of dglas over k commuting with the anchor map. Remark 3.6. -- Every A-linear dgla defines a dg-Lie algebroid: it suffices to keep the same underlying object and to set the anchor map to zero. On the other hand, if L is a (k, A)-dg-Lie algebroid, the kernel of the anchor map is a true A-linear dgla. These two constructions are adjoint. -- Given a pair (k, A), it is possible (see [20]) to attach to any object V of modA/Derk(A) a free forgetful functor from (k, A)-dg-Lie algebroids to A-modules lying over Derk(A). (k, A)-dg-Lie algebroid, denoted by free (V). The functor V → free (V) is the left adjoint to the -- The Chevalley -- Eilenberg construction can be performed for (k, A)-dg-Lie algebroids. If L is a k/A(L) is the A-dual of SA(L[1]) and the (k, A)-dg-Lie algebroid, then the A-module underlying CE• differential reads as follows (omitting signs): dCEω(ℓ0, . . . , ℓn) = ± ± nXi=0 nXi=0 ω(cid:0)ℓ0, . . . , dL(ℓi), . . . , ℓn(cid:1) ρ(ℓi)(cid:0)ω(ℓ0, . . . ^ℓi, . . . , ℓn)(cid:1) ± X0≤i<j≤n ω([ℓi, ℓj], ℓ0, . . . ^ℓi, . . . , ^ℓj, . . . , ℓn) . However, the same problem appearing for the Chevalley-Eilenberg complex of A-dg-Lie algebras happens: this functor must be derived. Hence, the above definition is set only on (k, A)-dg-Lie algebroids that are projective as A-modules (again A is assumed to be cofibrant), and since every (k, A)-dg-Lie algebroid is quasi-isomorphic to one whose underlying A-module is projective, this defines a functor CE• k/A from (k, A) dg-Lie algebroids to A-augmented k-cdgas. is impossible to expect this kind of result for (k, A)-dg-Lie algebroids, due to the presence of the -- If two cdgas A and A ′ are equivalent, then the ∞-categories LieA and LieA ′ are equivalent. It derivations of A (which is not an∞-functor). This motivates the next forthcoming definition. 28 Definition 3.7. A derived (k, A)-dg-Lie algebroid is a (k, QA)-dg-Lie algebroid, where QA is a cofibrant replacement of A. We denote by Liek/A the category of (k, QA)-dg-Lie algebroids. It can be proved that the category Liek/A has a naturel model structure obtained by transferring the model structure of modQA/Derk(QA) via the adjunction free : modQA/Derk(QA) −→←− Liek/A : forget (see [32]). Hence we get an∞-category Liek/A together with an adjunction modA/TA −→←− Liek/A . where TA ≃ Derk(QA) is the tangent complex of A. Fibrant (k, A)-dg-Lie algebroids are those with surjective anchor map. Remark 3.8 (Corrigendum). Actually, the transferred structure is only a semi-model structure17, as shown in [26]. One can consider the under-category h/Liek/A, where h is a fibrant-cofibrant replacement of the initial Lie algebroid 0. It happens to be a genuine combinatorial model category that is obviously Quillen equivalent to Liek/A (see [25, Remark 2.5]). Hence, from our point of view everything is still In the sequel, we will always assume that A is cofibrant (which is possible after taking a cofibrant replacement). In this way, derived (k, A)-dg-Lie algebroids will be usual (k, A)-dg-Lie algebroids. Observe that Liek/A can be made into a dual deformation context: namely, we transfer the dual fine, as the∞-category Liek/A is indeed presentable. deformation context (cid:0)modA/TA , (A[−n − 1] 0 forget : Lieop → TA)n(cid:1) from Example 2.14 along the adjunction k/A : −→←− modop : free . A/TA 3.3 Formal moduli problems under Spec(A) In this section, we are looking at a more general situation as in §3.1: we look at A-augmented k-algebras, where A is a bounded cdga in non-positive degrees, which we assume to be cofibrant. Recall from Remark 2.5 that we have a deformation context (cid:0)cdgak/A, (A ⊕ A[n])n(cid:1). The aim of this section is to give a description of formal moduli problems for A-augmented cdgas over k. The motto behind what follows will be the following one: Going from A-augmented A-algebras to A-augmented k-algebras corresponds via Koszul duality to go from Lie algebras over A to (k, A) Lie algebroids. The precise result we want to prove is: Theorem 3.9 (Nuiten [25]). Given a cdga A as above, there is a Koszul duality context This in particular implies that there is an equivalence (cid:0)cdgak/A, (A ⊕ A[n])n(cid:1) −→←−(cid:0)Lieop k/A, (free (A[−n − 1] 0 → TA)n) FMP(cdgak/A, A ⊕ A[n]) ≃ Liek/A . Proof. The first step consists in building the weak Koszul duality context. A reasonable candidate for the right adjoint functor is the Chevalley -- Eimenberg functor Here are a few properties of the Chevalley -- Eilenberg functor that carry on to the Lie algebroid setting: CE• k/A : Lieop k/A −→ cdgak/A . 17The main problem is that the initial object 0 is not fibrant. 29 -- CE• k/A(L) ≃ RHomU(A,L)(A, A), where U(A, L) is the universal envelopping algebra of L (see e.g. [28] for the definition of U(A, L))18. -- To every M f square zero extension A ⊕ df → TA = Der(A) one can associate a derivation df : A→ M∨ and thus a non-split k/A(cid:0)free (f)(cid:1) ≃ A ⊕ M∨[−1]. It is a fact that CE• M∨[−1]. df -- CE• k/A commutes with small limits (viewed as a functor on Lieop k/A) and thus, since Liek/A is presentable, admits a left adjoint, that we denote by Dk/A. Thus we have a weak Koszul duality context Dk/A : (cid:0)cdgak/A, (A ⊕ A[n])n(cid:1) −→←−(cid:0)Lieop k/A, (free (A[−n − 1] 0 → TA)n) : CE• k/A The next step is to prove that it is a Koszul duality context, by following the strategy of Proposition 2.30. We first consider the diagram cdgak/A CE• k/A (LA d →M)7→A⊕ d M[−1] Lieop k/A free LA/modA (−)∨ (modA/TA )op that one can fill everywhere with left adjoints, giving rise to the following weak morphism of weak Koszul duality contexts: cdgak/A Dk/A CE• k/A / Lieop k/A L:(B→A)7→(LA→LA/B) d7→A⊕ d M[−1] forget free LA/modA (−)∨ (−)∨ (modA/TA )op Knowing, from Example 2.21, that the bottom weak Kozul duality context is actually a Koszul duality context, according to Proposition 2.28 it remains to prove that this weak morphism is a morphism. The functor forget is conservative, and it is proved in [26] that it preserves sifted limits. Let us now show that, if L is good, then it is equivalent to a very good dg-Lie algebroid: i.e. a good dg-Lie algebroid that is of the form g 0 Lemma 3.10. Any good dg-Lie algebroid L is quasi-isomorphic to a very good one. → TA for a very good dgla g over A. Proof of the lemma. We first observe that 0 is very good. We then proceed by induction: assume that L is very good, and consider a dg-Lie algebroid L ′ obtained by a pushout free (A[−n − 1] 0 → TA) 0 L / L ′ in Liek/A. Since free (A[−n − 1] 0 Liek/A is represented by a morphism free (A[−n − 1] 0 → TA) is cofibrant, then a morphism free (A[−n − 1] 0 → TA)→ L in → TA)→ eL in Liek/A, where eL is a chosen fibrant 18In [28], (k, A)-Lie algebroids are named (k, A)-Lie algebras, and everything that is done makes sense in the differential graded setting as well. 30 o o O O o o O O / o o / / O O o o O O / / / replacement of L. Now, we observe that one can choose eL = L ⊕ h, where h is a fibrant replacement of 0, so that the replacement morphism L→ eL splits. All in all, any morphism free (A[−n − 1] 0 → TA)→ L in Liek/A can be represented by an actual morphism free (A[−n − 1] 0 → TA) → L in Liek/A. Next, a cofibrant replacement of the left vertical arrow is given by the morphism → TA(cid:11) . Hence our pushout in Liek/A is represented by the following honest non-derived pushout in Liek/A: free (A[−n − 1] 0 id L free (A[−n − 1] 0 −→ A)[−n − 1] 0 → TA)→ free(cid:10)cone (A → TA) → TA)(cid:11) −→ A)[−n − 1] 0 → TA) for some very good dgla g over A. We finally observe that → TA : LieA −→ Liek/A → TA) for a very good dgla over A. (−) 0 / L ′ id free(cid:10)cone (A Now recall that, by induction, L = (g 0 the functor commutes with colimits, so that L ′ = (g ′ 0 This in particular shows that, again under the boundedness assumption on A, forget sends good objects to reflexive objects; hence (C) holds. It remains to prove (D). We consider the natural morphism θL defined as the following composition: L(cid:0)CE• k/A(L)(cid:1)→ L(cid:0)CE• k/A(free forget L)(cid:1)→ L(cid:0)A ⊕ dρ (forget L)∨[−1](cid:1)→ (forget L)∨ , where ρ is the anchor map19 and L is the left adjoint to the twisted square zero extension functor given by the formula L(B→ A) = (LA→ LA/B). Again, the cdga morphism k/A(free forget L)→ A ⊕ k/A(L)→ CE• CE• dρ L∨[−1] is nothing but the obvious projection (onto the quotient by the square of the augmentation ideal whenever L is very good). As in the proof of Proposition 2.30 we consider the uncompleted Chevalley -- Eilenberg sub-cdga • L∨[−1], and k/A(L). The quotient by the square of its augmentation ideal is again A ⊕ dρ we have the following commuting diagram and its image through the left-most vertical adjunction of our square: fCE k/A(L) ֒→ CE• fCE • k/A(L) ιL '❖❖❖❖❖❖❖❖❖❖❖❖❖❖ / A ⊕ dρ CE• k/A(L) (forget L)∨[−1] L(cid:0)fCE L(cid:0)CE• L(ιL) • k/A(L)(cid:1) eθL &◆◆◆◆◆◆◆◆◆◆◆ k/A(L)(cid:1) θL / (forget L)∨ In order to prove that θL is an equivalence when L is good, it suffices to prove that both eθL and L(ιL) are. Let us start with the following variation on Lemma 2.32: Lemma 3.11. Let B be an A-augmented k-cdga, with augmentation ideal J, that is cofibrant as a k-cdga, and such that the augmentation morphism is a fibration. Then J/J2[1] is naturally an A-module under LA, and there is an equivalence L(B) ≃ (LA→ J/J2[1]) in LA/modA. 19In the above, (forget L)∨ shall be understood as LA ρ∨ → L∨. 31 / / / ' / & / Proof of the lemma. First of all, observe that the exact sequence splits in modk. Hence there is a k-derivation δ : A→ J/J2[1] such that B/J2 is isomorphic to A ⊕ in cdgak/A. Hence, δ 0→ J/J2→ B/J2→ A→ 0 J/J2 Homcdgak/A(B, B/J2) = Homcdgak/A (B, A ⊕ δ J/J2) ≃ Hom LA/modA (L(B), J/J2[1]) . Therefore, the quotient morphism B → B/J2 gives us a morphism L(B) → (LA → J/J2[1]). To show that this is an equivalence it is sufficient to prove that the morphism LA/B→ J/J2[1] is an equivalence. Now, since A and B are cofibrant, and B→ A is fibrant, B ⊗B A→ Ω1 B ⊗B A→ Ω1 The morphism LA/B→ J/J2[1] is thus modeled by the shift of the natural map B ⊗B A→ Ω1 A)→ J/J2 , which is an isomorphism. LA/B ≃ cone(Ω1 A) ≃ ker(Ω1 ker(Ω1 A)[1] . We then observe that -- when L is very good (which we can always assume without loss of generality when dealing with • -- the augmentation map fCE good dg-Lie algebroids), then fCE k/A(L)→ A is obviously a fibration. • k/A(L) is cofibrant. Thus eθL is an equivalence if L is good. We again conclude with the very same flatness argument as in the proof of Proposition 2.30 in order to get that L(ιL) is an equivalence. 3.4 The relative tangent complex We want to compute the underlying anchored module of the dg-Lie algebroid associated with a repre- sentable fmp under Spec(A). We have an equivalence Φk/A : FMP(cdgak/A) −→Liek/A Proposition 3.12. Let B be the representable fmp associated with an A-augmented k-algebra B. Then Sketch of proof. Using Proposition 2.39 together with Theorem 3.9, one sees that the image of Φk/A(B) the image of Φk/A(B) along the forgetful functor Liek/A→ modA/TA is TA/B→ TA. along the forgetful functor is equivalent to the image of the representable fmp LA→ LA/B on LA/modA along the equivalence FMP(LA/modA) →modA/TA . One easily sees that the image of this representable fmp is indeed (LA→ LA/B)∨ = TA/B→ TA. functor Liek/A→ modA/TA deserves to be denoted TSpec(A)/X and called the "relative tangent complex According to the above proposition, the image of Φk/A(X) of an fmp under Spec(A) along the forgetful Remark 3.13. Unsurprisingly, an easy calculation allows to prove that the tangent complex to an fmp under Spec(A) is of Spec(A) over X". where TX can be viewed in modA because Liek/A is A-linear. In the case A = k, we get back the Lie structure20 on TX ≃ cofib(TSpec(A)/X→ TA) , 20A dg-Lie (k, k)-algebroid is nothing but a dgla. TX[−1] ≃ fib(0→ TX) = fib(Tk→ TX) ≃ TSpec(k)/X . 32 Even if A is not k itself, we have a fiber sequence of Lie algebroids TX[−1]→ TSpec(A)/X→ TA in modA, where TSpec(A)/X and TA are (k, A)-dg-Lie algebroids, and TSpec(A)/X→ TA is a morphism in Liek/A. The forgetful functor Liek/A → modA/TA being a right adjoint, there is a (k, A)-dg-Lie algebroid structure on TX[−1] that allows to upgrade the above sequence into a fiber sequence of (k, A)- dg-Lie algebroids. One can then show that the functor "taking the fiber of the anchor" actually factors through dglas over A. Here is a question we would like to answer : What is the split formal moduli probem under Spec(A) having TX[−1] as associated dgla? We consider the commuting diagram: cdgak/A CE• k/A Lieop k/A forget g7→(g 0 →TA) cdgaA/A CE• A Lieop A It can be filled with left adjoints in the following manner: cdgak/A V→V⊗kA forget cdgaA/A Dk/A CE• k/A DA CE• A / Lieop k/A Ξ g7→(g 0 →TA) Lieop A Here the right-most vertical adjunction is obtained from a Quillen adjunction: left21 derived functor of the functor given on models by the kernel of the anchor map. in particular Ξ is the Remark 3.14. Note that the forgetful functor maps cdgasm Typically, non-split square-zero extensions do not belong to cdgasm k/A; however it is not an equivalence: cdgasm A/A. A/A to cdgasm k/A is way bigger than cdgasm A/A. cdgaA/A→ cdgak/A Lemma 3.15. The above square of adjunctions is a weak morphism of Koszul duality contexts. Proof. The only thing left to prove is condition (B), which is obvious: precisely A ⊕ A[n], viewed as a k-algebra. indeed, forget (A ⊕ A[n]) is Consequently, Proposition 2.39 tells us that we have the following commuting diagram Ψk/A Liek/A Ξ LieA ΨA FMPk/A / FMPA/A ◦forget modA T (◗◗◗◗◗◗◗◗◗◗ 6♠♠♠♠♠♠♠♠♠♠♠ T 21Recall that we are working with opposite categories: Ξ : Lieop k/A → Lieop A . 33 o o O O o o O O / o o / / O O o o O O / / ( / 6 Let X be a formal moduli problem under Spec(A), and let LX := Φk/A(X) be its relative tangent Lie algebroid. We then have: Ξ(LX) ≃ gX◦forget . Hence the Lie dgla structure on TX[−1] obtained by taking the fiber of the anchor map of the relative tangent Lie algebroid LX coincide with the one given on TX◦forget. This answers our question: the split formal moduli problem under Spec(A) having Ξ(LX) as tangent Lie algebra is X ◦ forget. Remark 3.16. Geometrically, and following the intuitive ideas presented in the introduction, one shall understand X as a formal thickening Spec(A)→ X of Spec(A). The geometric interpretation of X ◦ forget is then as the split formal thickening Spec(A)→ Spec(A) × X→ Spec(A) given by the graph22 of the previous one. Making this idea more precise is a bit complicated as Spec(A) is initial for the two ∞- categories of fmp's that we are considering. Hence Spec(A) × − does nothing, and shall be understood as composing with the forgetful functor. 3.5 Generalizations Before going to the last, more geometric, Section of this survey, let us mention a few unnecessary assumptions that we have made for the sake of exposition: -- one can replace k with any cdga sitting (cohomologically) in non-positive degree and containing Q (as a sub-cdga). -- one can replace A with any k-cdga with a reflexive cotangent complex LA/k in modA. -- one can replace k and A with presheaves of such cdgas on some given∞-category. Remark 3.17. One could ask if all this still works for commutative algebra objects in a gentle enough stable model category M, e.g. satisfying the standing assumptions of [5, §1.1]. Ideally, we would like to be able to allow M to be the category of graded mixed modules (see §A.1), and cover the following situation: k = DXDR is the mixed crystalline structure sheaf of a derived Artin stack X and A = BX is the mixed principal parts of X (after [5, Definition 2.4.11]), which are graded mixed cdgas (see §A.1, again). We will come back to the comparison with the [5] in the next Section. In particular, let k→ B→ A be a sequence of cdgas, with k and B sitting cohomologically in non-positive degree, and both LA/B and LA/k reflexive in modA. One can show in a very similar fashion that the weak morphism of Koszul duality contexts from the previous subsection generalizes as follows: cdgak/A Dk/A CE• k/A / Lieop k/A V→V⊗BA forget − × T A/k TA/B (L→TA/B)7→(L→TA/k) cdgaB/A DB/A CE• B/A Lieop B/A This leads (again after Proposition 2.39) to a commuting square Ψk/A Liek/A − × T A/k TA/B ΨB/A LieB/A FMPk/A / FMPB/A ◦forget modA T (◗◗◗◗◗◗◗◗◗◗ 6♠♠♠♠♠♠♠♠♠♠♠ T 22Actually, the formal neighborhood of the graph. 34 / o o / / O O o o O O / / ( / 6 above that the relative tangent Lie algebroid23 TSpec(B)/X◦forget of X ◦ forget, viewed as a (k, A)-dg-Lie If X lies in FMPk/A, then it has a relative tangent Lie algebroid TSpec(A)/X→ TA/k. It follows from the algebroid using the Lie algebroid morphism TA/B→ TA/k, is equivalent to the pull-back TA/B . TSpec(A)/X × TA/k In the spirit of Remark 3.16, one geometrically interprets X ◦ forget as the following "Spec(B)-split formal thickening of Spec(A)": Remark 3.18. One can actually show that LieB/A ≃ (Liek/A)/TA/B . FMPB/A ≃ (FMPk/A)/B. Spec(A)→ Spec(B) × X . In particular, this shows that This shall be interpreted geometrically as follows. If we have a Spec(B)-split formal thickening of There is yet another geometric situation one may want to understand in terms of Lie algebroids: Spec(A), then the map X → Spec(B) uniquely factors through X → B, which is terminal among such formal thickenings (being the formal completion of the map Spec(A)→ Spec(B)). given a fmp Spec(B) → X under Spec(B), one can look at it under Spec(A) by just composing24 with Spec(A)→ Spec(B). On the level of anchored modules, we have a functor modB/TB/k −→ modA/TA/k L 7−→ L ⊗B A ⊕ TB/k⊗BA TA/k . Hence the above fiber product in modA can be computed as an ordinary fiber product in modA: more Assume that B→ A is a cofibration betwen cofibrant cdgas. Hence the morphism TB/k ⊗B A→ TA/k in modA is represented by dφ : Derk(B) ⊗ A → Derk(A) in modA, which is moreover a fibration. precisely, it is ker(ρ ⊗B A − dφ), where ρ : L → Derk(B) is the anchor map. In this situation, it has algebroid25. We therefore get a functor Liek/B→ Liek/A that actually factors as been shown by Higgins and Mackenzie [12] that one can put a Lie bracket on L that turns it into a Lie The main upshot of this sketchy discussion is that if we are given a commuting square of cdgas Liek/B −→ TA/B/Liek/A −→ Liek/A B1 B2 A1 / A2 where A1 and B1 cohomologically sit in non-positive degree and contain Q, and LA2/B1 and LA2/A1 are reflexive in modA2 , then we have a pull-back functor 4 Global aspects LieB1/B2 → LieB1/A2 → LieA1/A2 . 4.1 Formal derived prestacks and formal thickenings We start with several definitions and statements, mainly extracted from [5, 8, 9]. Let us denote by dAfff.p. k almost finitely presented : H0(A) is finitely generated as a k-algebra and Hi(A) is a finitely presented the opposite ∞-category of non-positively graded k-cdgas A that are 23Note that we slightly abuse the notation here. As we have seen in Proposition 3.12, TA/B is the underlying A-module of the (k, A)-dg-Lie algebroid Ψk/A(B). 24And, maybe, completing afterwards. 25In [12] the authors work in the differential setting and nothing is dg. Nevertheless, their construction carry on without any problem. Their only technical assumption is that the map along which they pull-back is submersive, which is fine in our setting thanks to the cofibrancy of B → A. 35 / / / H0(A)-module. We then define the ∞-category dPrStf.p. prestacks over k to be the∞-category of presheaves on dAfff.p. terminology of [8]) if the canonical map k of locally almost finitely presented derived k . We say that a locally almost finitely presented derived prestack F is nilcomplete (convergent in the F(A≤n) n F(A)→ lim is an equivalence, where A≤n denotes the n-th Postnikov truncation of A. spanned by nilcomplete prestacks26 is equivalent k ⊂ dAfff.p. spanned by eventually coconnective cdgas (see [8, Proposition 1.4.7 of Chapter 2]), where a cdga A is called eventually coconnective if Hn(A) = 0 for n << 0. The full sub-∞-category dPrStf.t. to the essential image of the right Kan extension from the full sub-∞-category dAfff.t. A formal derived prestack is a nilcomplete locally almost finitely presented derived prestack27 F with of dPrStf.p. k k k the following to properties: -- F is infinitesimally cohesive: for any cartesian square of almost finitely presented non-positively graded k-cdgas A A1 A2 / A0 such that each H0(Ai)→ H0(A0) is surjective with nilpotent kernel, then the induced square F(A1) F(A) is cartesian as well. F(A2) / F(A0) -- F admits a procotangent complex. Here we say that F admits a procotangent complex if it admits a procotangent complex LF,x at every point x : Spec(A)→ F, and for every morphism of points Spec(A) Spec(B) u x #●●●●●●●●● {✇✇✇✇✇✇✇✇✇ y F co-prorepresentable. the induced morphism u∗ : LF,y ⊗B A → LF,x is an equivalence. We say that F admits a pro- cotangent complex at a given point x : Spec(A) → F if the functor DerF(x, −) of derivations is We denote by FdPrStk the∞-category of formal derived prestacks over k. For a derived prestack F, we denote by Fred its associated reduced prestack Fred := colim where Ared denotes the quotient of H0(A) by its maximal nilpotent ideal. We also consider the de Rham prestack FDR of F, that is defined as follows: FDR(A) := F(Ared). We now summarize the properties of reduced and de Rham prestacks: Spec(A)→F Spec(Ared), -- there are canonical maps Fred→ F→ FDR, inducing equivalences (Fred)DR →FDR and Fred →(FDR)red. This implies in particular that a morphism F → G induces an equivalence Fred →Gred between their reduced prestacks if and only if it induces an equivalence FDR →GDR between their de Rham prestacks. -- (−)red is left adjoint to (−)DR. 26The superscript "f.t." stands for (locally almost of ) finite type. 27Nilcomplete locally almost finitely presented prestacks are called said to be locally almost of finite type (laft) in [8]. 36 / / / / / / / / # { Proposition 4.1 ([9], Chapter 5, Proposition 1.4.2). Let A be a non-positively graded cdga that is almost finitely presented and eventually coconnective (i.e. A is almost of finite type). Then there is an For any formal derived prestack X, one can consider the sub-∞-category Thick(X) of X/FdPrSt spanned by those morphisms X → F that induce an equivalence Xred → Fred between the associated reduced prestacks. We call it the∞-category of formal thickenings of X. equivalence of ∞-categories Idea of the proof. We have a pull-back functor Thick(cid:0)Spec(A)(cid:1)→ FMPk/A along that consists in evaluating our prestack on Spec(A)→ Spec(B) with B a small A-augmented k-cdga. This pull-back functor has a left adjoint, given by a left Kan extension28. One can prove (using infinitesimal cohesiveness) that a formal thickening is completely determined by its restriction on Spec(B)'s for which Bred ≃ Ared. We conclude by observing that the functor cdgasm obviously factors through the category of those cdgas of finite type with reduced algebra equivalent to Ared, and that this has a left adjoint: sending a B such that Bred ≃ Ared to A ×Ared B. FMPk/A ≃ Thick(cid:0)Spec(A)(cid:1) . k/A→ cdgaf.t. k/A → cdgaf.t. := (cid:0)dAfff.t. k (cid:1)op k cdgasm k , As a consequence, we have the following useful Corollary: Corollary 4.2. For any non-positively graded cdga A almost of finite type, there is an equivalence FMPk/A ≃ A/FMPk/Ared . thickening of X, we have an equivalence Moreover, for any formal derived prestack X→ Spec(A) over Spec(A) that exhibits Spec(A) as a formal Thick(X)/Spec(A) ≃ X/FMPA/Ared . In the second part of the Corollary, we abusively denote by the same symbol the formal derived prestack X and its associated formal moduli problem under Spec(Ared). tween their associated reduced prestacks. Then we obviously have an equivalence Thick(Y) ≃ Y/Thick(X). To prove the first part of the Corollary, we begin with the obvious observation that the map Proof. Let X→ Y be a morphism of formal derived prestacks that induces an equivalence Xred →Yred be- Spec(Ared)→ Spec(A) is a formal thickening of Spec(Ared). Hence we have an equivalence Thick(cid:0)Spec(A)(cid:1) ≃ Spec(A)/Thick(cid:0)Spec(Ared)(cid:1) . FMPk/A ≃ Thick(cid:0)Spec(A)(cid:1) ≃ Spec(A)/Thick(cid:0)Spec(Ared)(cid:1) ≃ A/FMPk/Ared . Then we notice that the fmp A ∈ FMPk/Ared is the pullback of Spec(A) along cdgasm We finally use the above Proposition to get a chain of equivalences k/Ared → cdgaf.t. For the second part of the statement, we observe that X → Spec(A) being a formal thickening means that the induced map Xred→ Spec(Ared) is an equivalence. This in turns teaches us that Spec(Ared) ≃ Xred → X is a formal thickening, and thus can be understood as a fmp under Spec(Ared). Hence we Thick(X) ≃ X/Thick(cid:0)Spec(Ared)(cid:1) ≃ X/FMPk/Ared , have k . As a consequence we get that Thick(X)/Spec(A) ≃ X/(FMPk/Ared )/A. We conclude by observing that (FMPk/Ared )/A ≃ FMPA/Ared (see Subsection 3.5). 28Observe that the left Kan extension of a functor on small A-augmented cdgas is a priori just a derived prestack locally almost of finite type. It is an exercise to check that left Kan extensions of fmps are sent to formal thickenings. The only non-obvious property to check that it admits a procotangent complex, which follows from the prorepresentability of fmps. 37 4.2 DG-Lie algebroids and formal thickenings The consequences of Corollary 4.2 are of particular interest. Let B→ A be a morphism in cdgaf.t. assume that the induced morphism Bred→ Ared is an isomorphism. Proposition 4.3. We have equivalences: k , and -- FMPk/A ≃ TAred/A/Liek/Ared. -- FMPB/A ≃ TBred/A/LieB/Bred. Proof. This is a direct consequence of (the proof of) Corollary 4.2, Theorem 3.9, and Proposition 3.12. Indeed, we have two chains of equivalences: and FMPk/A ≃ A/FMPk/Ared ≃ TAred/A/Liek/Ared FMPB/A ≃ A/FMPB/Bred ≃ TBred/A/LieB/Bred . In the rest of this subsection we will provide (sketches of) two alternative proofs that do not rely on formal thickenings and Proposition 4.1. We actually prove the following more general result: trated in non-positive degree, and with B being small in cdgak/A. Then we have an equivalence Theorem 4.4. Consider a sequence of k-cdgas K→ B→ A, all cohomologically bounded and concen- FMPK/B ≃ B/FMPK/A . Observe that B is small in cdgak/A if and only if it is so in cdgaK/A. Also note that if B is almost of finite type and A = Bred, then B is small in cdgak/Bred and thus we recover in particular Proposition 4.3. Sketch of proof of the Theorem. We first observe that there is an adjunction where ι(C) = C and p(C) = C × A B. ι : cdgaK/B −→←− cdgaK/A : p , Since B is small, and p commutes with limits (being a right adjoint), then p preserves small morphisms and p∗ : Fun(cdgaK/B, sSet) → Fun(cdgaK/A, sSet) sends fmps to fmps. Moreover, for every X ∈ FMPK/B, there is a natural morphism B→ p∗X: Hence we obtain a functor FMPK/B→ B/FMPK/A, which we abusively still denote p∗. This functor has a right adjoint p∗, which we describe now. Since B is small, and ι commutes with pull-backs, then ι preserves small morphisms. But it does not preserves fmps, as it doesn't preserve terminal objects. To cure this we define, for an X ∈ B/FMPK/A, B = HomcdgaK/A (B, −) ≃ HomcdgaK/A (B, −) × X(B)→ X(− × B) = p∗X . A p∗X := ι∗X × X(B) ∗ ∈ FMPK/B , where ∗→ X(B) is given by {idB}→ B(B)→ X(B). of the adjunction: for every X ∈ FMPK/B, we have Let us prove that the adjoint pair (p∗, p∗) is actually an equivalence. We start by describing the unit p∗p∗X = ι∗p∗X × p∗X(B) ∗ ≃ X(− × A B) × X(B× A B) X(B) . 38 Hence there is an obvious natural morphism X → p∗p∗X. We claim that it is an equivalence. Indeed, for every small C→ B, the square C B is cartesian, and its bottom horizontal morphism is small. Therefore X preserves this cartesian square, B C × A / B × A B and thus the unit morphism X→ p∗p∗X is an equivalence. We now turn to describing the counit of the adjunction. For every X ∈ B/FMPK/A, we have p∗p∗X = p∗X(− × A B) = X(− × A B) ∗ X(B) , and the counit is the composed morphism p∗p∗X→ X(− × composition of pull-back squares A p∗p∗X(C) B)→ X. Thus we have, for every C→ A, a ∗ X(C × A B) X(B) X(C) / X(A) ≃ ∗ Theorem. This shows that the counit morphism p∗p∗X(C) → X(C) is an equivalence, and ends the proof of the In the remainder of this subsection, we work under the hypothesis of Theorem 4.4. As a consequence of the Theorem we get (using Theorem 3.9 and Proposition 3.12) an equivalence LieK/B ≃ TA/B/LieK/A . TA/B/LieK/A can be proven to be the derived version of Higgins -- Mackenzie pull- back of Lie algebroids (see the end of §3.5). The inverse functor is not so easy to describe purely in terms of Lie algebroids, but there is nevertheless a nice way of directly proving that there is an equivalence The functor LieK/B→ that we explain now. Our strategy is (again) as follows: FMPK/B ≃ TA/B/LieK/A -- we first show that there is a weak Koszul duality context involving cdgak/B and TA/B/LieK/A. -- we then show that it comes with a weak morphism to a Koszul duality context having the form of Example 2.21. -- we finally claim that this weak morphism is a morphism, showing that the original weak Koszul duality context is a Koszul duality context (after Proposition 2.28). -- we conclude using Theorem 2.33. The weak Koszul duality context We begin with the following rather evident result: K/A(TA/B) ≃ B. In particular, we have an adjunction Lemma 4.5. We have that DK/A(B→ A) ≃ TA/B and CE• DK/A : cdgaK/B LieK/A(cid:17)op : CE• K/A . −→←−(cid:16)TA/B/ 39 / / / / / / / / Proof. The first part of the statement is just Proposition 3.12, and the second part follows from the fact that B is small. Finally, one concludes using Theorem 3.9 and the obvious observation that cdgaK/B ≃ (cdgaK/A)/B→A. The deformation context is still the one from Example 2.9: (cid:0)cdgaK/B, (B ⊕ B[n])n(cid:1), and the dual deformation context is TA/B/LieK/A equipped with the spectrum object TA/B ⊕ free (A[−n − 1] 0 → TA/K) . for its opposite∞-category. Remark 4.6. The reader can check directly that this is indeed a spectrum object, but this will also be a consequence of our construction of the weak morphism below (see Lemma 4.7). Together, the deformation context and the dual deformation context define a weak Koszul duality context. Indeed: K/A(cid:0)TA/B ⊕ free (A[−n − 1] 0 → TA/K)(cid:1) A(cid:0)free (A[−n − 1])(cid:1) CE• K/A(TA/B) × A (A ⊕ A[n]) CE• ≃ CE• ≃ B × A ≃ B ⊕ B[n] . The weak morphism We already have the following adjunctions: cdgaK/B DK/A CE• K/A / (cid:16)TA/B/ LieK/A(cid:17)op (C→B)7→(LB/K→LB/C) d7→B⊕ d M[−1] LB/K/modB (−)∨ (−)∨ (modB/TB/K)op Lemma 4.7. There is an adjunction L : modB/TB/K such that Sketch of proof. We define L as the composition of the following two functors: L(B[−n − 1] 0 → TB/K) ≃ TA/B ⊕ free (A[−n − 1] 0 → TA/K) . TA/B/LieK/A : R −→←− TA/B/modA/TA/K , sending M→ TB/K to -- modB/TB/K → Note that 0 ′ = fib (TA/K→ TB ⊗B A) ≃ TA/B. -- TA/B/modA/TA/K → M ′ := M ⊗B A × TA/B/LieK/A, sending TA/B→ N→ TA/K to free N . TA/K . TB⊗BA TA/B afree TA/B One easily checks that L(B[−n − 1]) ≃ TA/B ⊕ free (A[−n − 1] 0 We also define R as the composition of two functors: 40 → TA/K). / o o / / O O o o the (k, A)-dg-Lie algebroid TA/B (which are just modules over its universal enveloping algebra). -- first a functor TA/B/LieK/A→ modTA/B , where modTA/B denotes the∞-category of modules over Here the functor is defined as cofib (TA/B→ −), which naturally lands in TA/B-modules29. -- then the functor modTA/B → modB/TB, sending a TA/B-module E to its Chevalley -- Eilenberg TA/B-module structure on cofib (TA/B → TA/K) ≃ TA/K ⊗B A is given by the action of TA/B on complex CE•(TA/B, E) (see e.g. [3, §2.2] for a definition). Here the reader has to check that the R(TA/K) ≃ CE•(TA/B, TB/K ⊗B A) ≃ TB/K ⊗B CE•(TA/B) ≃ TB/K . A, so that The details of the construction, and the proof that R is right adjoint to L, will appear elsewhere. Lemma 4.8. The square of right adjoints cdgaK/B CE• K/A (cid:16)TA/B/ LieK/A(cid:17)op d7→B⊕ d M[−1] Z:=Lop LB/K/modB (−)∨ (modB/TB/K)op commutes. Sketch of proof. The functor CEK/A is right adjoint, and thus preserves limits. In particular, it sends push-outs in LieK/A to pull-backs in cdgaK/A. Hence we have that CE• K/ALop(M) ≃ (A ⊕d ′ (M ′)∨[−1]) × B , A⊕can LA/B[−1] where d ′ : LA/K → (M ′)∨ is adjoint to the pullback of M ⊗B A → TB/K ⊗B A, and along TA/K → TB/K ⊗B A, and can : LA/K → LA/B is the canonical map. Hence CE• where d : LB/K→ M∨ is adjoint to M→ TB/K. LieK/A(cid:17)op Together, Lemma 4.7 and Lemma 4.8 imply that we have a weak morphism of weak Koszul duality K/ALop(M) ≃ B ⊕d M∨[−1], contexts from cdgaK/B The proof that this weak morphism is indeed a morphism will appear elsewhere. to LB/K/modB −→←− (modB/TB/K)op. −→←−(cid:16)TA/B/ 4.3 Formal moduli problems under X Proof. First of all observe that formal derived prestacks are stable by pullbacks, so that XA is a formal Let X be a formal derived prestack, and let X→ Y be a formal thickening of X. Lemma 4.9. For every Spec(A)→ Y, XA := X ×Y Spec(A)→ Spec(A) is a formal thickening of XA. derived prestack. The only thing left to prove is that XA→ Spec(A) induces an equivalence on reduced This allows us to build a presheaf of ∞-categories A 7→ Thick(XA)/Spec(A) on dAfff.t., and a prestacks, which is equivalent to require that it induces an equivalence on de Rham prestacks. The functor (−)DR being a right adjoint, we have that (XA)DR ≃ XDR ×YDR Spec(A)DR ≃ Spec(A)DR. commuting diagram of functors Thick(X)/Y lim Thick(XA)/Spec(A) Spec(A)→Y X/(dPrStf.t. k )/Y lim Spec(A)→Y XA/(dPrStf.t. k )/Spec(A) 29To see this, one can use models and compute the cofiber in the case when the map out of TA/B is injective. In this case the TA/B-module structure is rather classical (see e.g. [3, §3.1]). 41 o o O O O O o o / / / / The bottom horizontal functor has a left adjoint k )/Y lim Spec(A)→Y XA/(dPrStf.t. k )/Spec(A) −→ X/(dPrStf.t. that sends a diagram (cid:0)XA→ Z(A)→ Spec(A)(cid:1)Spec(A)→Y to its colimit Z(A)→ colim Spec(A)(cid:1)→ Z XA→ colim Spec(A)→Y(cid:0)Z × The counit of the adjunction X ≃ colim Spec(A)→Y colim Y Spec(A)→Y Spec(A)→Y Spec(A) ≃ Y . is an equivalence, and thus the bottom horizontal functor is fully faithful. Since the two vertical functors are also fully faithful, therefore so is the remaining horizontal one. The next Lemma tells us how far it is from being an equivalence. Lemma 4.10. The essential image of lim Spec(A)→Y Thick(XA)/Spec(A) lies in the essential image of X/(dPrStf.t. Proof. The first part of the statement is obvious. Let (ZA)Spec(A)→Y be a diagram of formal derived k )/Y . Moreover, it is included in the sub-∞-category spanned by (images of ) those X → Z→ Y inducing equivalences on associated reduced stacks and such that Z is infinitesimally cohesive. stacks lying over the tautological diagram (cid:0)Spec(A)(cid:1)Spec(A)→Y , and such that (ZA)red→ Spec(Ared) is an equivalence. The functor (−)red being left adjoint, it commutes with colimits, and thus Zred→ Yred and thus with pullbacks in particular, so that Z is infinitesimally cohesive because every single ZA is. ZA). Moreover, filtered colimits do commute with finite limits is an equivalence (where Z = colim Spec(A)→Y If we could prove that derived prestacks in the essential image have a procotangent complex, then Thick(XA)/Spec(A) is an equivalence. This is probably not true in full generality, but we think that it is for several examples: we would get that the functor Thick(X)/Y → lim Conjecture 4.11. If the functor IndCoh(Y)→ QCoh(Y) is an equivalence, then the functor Thick(XA)/Spec(A) Spec(A)→Y Spec(A)→Y is an equivalence as well. Thick(X)/Y → lim Combining the above Lemma for Y = XDR, together with the results of the previous Subsections, we get the following Theorem: Theorem 4.12. There is a fully faithful functor Thick(X)→ LieXDR/X := lim Spec(A)→XDR TSpec(Ared )/XA /LieA/Ared . Proof. We simply have to prove that Thick(X) ≃ Thick(X)/XDR . This is true as XDR happens to be the terminal formal thickening of X. We strongly believe that the functor appearing in the Theorem is actually an equivalence. This is in (see [7, Proposition 2.4.4]). fact a consequence of our conjecture, as we know that IndCoh(XDR)→ QCoh(XDR) is an equivalence Comparison with formal localization Following Remark 3.17, let us pretend that our results do hold in the more general graded mixed setting from [5] (see §A.1 for a recollection), and consider the following situation: K = DXDR is the mixed crystalline structure sheaf of a derived Artin stack X and A = BX is the mixed principal parts of X (after [5, Definition 2.4.11]). These are prestacks over XDR with values in graded mixed cdgas, and they are defined as follows: 42 7−→ D(B) := DR(Bred/B) ≃ CEǫ(TBred/B), with D(B) ≃ B; 7−→ D(XB) := DR(XB/B). -- (cid:0)Spec(B)→ XDR(cid:1) K -- (cid:0)Spec(B)→ XDR(cid:1) A We would thus have a prestack of∞-categories LieK/A over XDR, of which the global sections are denoted also been defined as global sections of a prestack of∞-categories. Let us explain why do believe so. XDR/X. Presumably, this would be an ind-coherent version of LieXDR/X defined above, which has -- First observe that, according to our results, we shall have an equivalence Lief LieD(B)/D(XB) ≃ TBred/D(XB )/LieD(B)/Bred . -- Then, the realization functor · and its left adjoint shall produce two adjunctions, related by a forgetful functor as follows: TSpec(Bred)/XB /LieB/Bred / TBred/D(XB )/LieD(B)/Bred modB / modD(B) -- Finally, it has been proven in [5, Proposition 2.2.8] that, restricted to suitable subcategories of perfect modules, the adjunction restricts to an equivalence. Moreover, it is conjectured that modD(XB) is equivalent to IndCoh(XB) (see e.g. [5, Remark 2.2.6]). D(XB) ⊗Bred − : modB −→←− modD(B) : · sequence of graded mixed cdgas Let us conclude with the following observation. Let X→ Y be a formal thickening for which Y is also a derived geometric stack. In particular we know that XDR → YDR is an equivalence, and we thus get a and hence graded mixed (K, A)-Lie algebroid TA/B. We therefore get a functor ThickGeom(X) → XDR/X, which shall fit into the following conjectural commuting diagram of functors: K→ B := BY → A , Lief ThickGeom(X) Thick(X) Lief XDR/X LieXDR/X IndCoh(X) / QCoh(X) where ThickGeom(X) denotes the subcategory of Thick(X) consisting of geometric thickenings. A Appendix In this appendix our aim is to compare the Chevalley-Eilenberg functor with the de Rham functor that is used in the formal localization procedure of [5]. This is a very first step for a future comparison between the various approaches to formal geometry in the derived setting. 43 / o o / o o / / / / / A.1 Recollection on graded mixed stuff We fix a base cdga A (not assuming anything at the moment), and start with Definitions. Definition A.1. -- A graded module is a sequence (Mn)n∈Z of A-modules. We refer to the index n as the weight. -- We denote by modgr A the category of graded modules. It comes equipped with the weight shifting functor M 7→ M(1), defined by M(1)n = Mn+1. -- A graded mixed module is a pair (M, ǫ), with M a graded module and ǫ : M→ M[1](1) a morphism of graded modules such that ǫ[1](1) ◦ ǫ = 0 (which we will abusively write ǫ2 = 0). We call ǫ the mixed differential. the category of graded mixed modules. The weight shifting functor obvi- -- We denote by modǫ−gr ously lifts to modǫ−gr . A and modǫ−gr -- Both categories (modgr A A ) are symmetric monoidal, so all categories of algebraic structures on modules that we considered for modA can be considered as well within them. We will add superscripts gr and ǫ − gr to the already introduce notation when emphasize which underlying symmetric monoidal category we are working with. For instance, cdgaǫ−gr denotes the category of commutative monoids in modǫ−gr A A . A We equip modgr A with the objectwise model structure from the one of modA. We can transport this model structure along the forgetful functor modǫ−gr A (−)♯ A −→ modgr and get a model structure30 on modǫ−gr for which the monoidal unit A is cofibrant. This model structure is nice enough so that all model category theoretic constructions based on modA still make sense on modǫ−gr , see e.g. [5, §1.1, §1.2 & Appendix A] and [26, Variant 3.10]. Thus we have for instance an A Then recall from [5, §1.3] that the Quillen adjunction of graded mixed A-cdgas. A ∞-category cdgaǫ−gr A induces an adjunction modA −→←− modǫ−gr A : HOMmodǫ−gr A (A, −) modA −→←− modǫ−gr A : − for which the right adjoint − , called (standard) realization, is lax symmetric monoidal. This implies in particular that this realization functor is also well-defined on the other aforementioned categories carrying the superscript ǫ − gr. It has a very explicit description: Proposition A.2 ([5]). For a graded mixed module (M, ǫ), M ≃ (Qn≥0 Mn, dM + ǫ), where dM : M→ M[1] is the (weight preserving) differential. every object is fibrant. Moreover, one has the following explicit cofibrant replacement eA for A, which is Proof. Let us (temporarily) use the projective model structure on modǫ−gr quasi-free (as a graded mixed A-module): . For this model structure, A -- let us consider the free A-module generated by xi's for i ≥ 0 and yj's for j ≥ 1. -- assign to xi cohomological degree 0 and weight i, and to yj cohomological degree 1 and weight j. -- define ǫ(xi) = yi+1. -- modify the differential31 by imposing that d(xi) = yi (by convention, y0 = 0). Finally observe that HOMmodǫ−gr A ( A, M) = (Qn≥0 Mn, dM + ǫ). 30The injective one (though we will soon use the projective one, with which it is easier to compute mapping spaces, but for which the monoidal unit is not cofibrant). 31Before doing that, our graded mixed module was free. 44 A.2 Graded mixed cdgas and dg-Lie algebroids Let B→ A be a morphism of cdgas which, as usual, we assume to be cofibrant. Proposition A.3. The Chevalley -- Eilenberg functor CE• B/A factors through a commuting diagram cdgaǫ−gr B/A CEǫ B/A 9ssssssssss − &▲▲▲▲▲▲▲▲▲▲▲ cdgaB/A CE• B/A Lieop B/A Proof. For a dg-(B, A)-algebroid L we set CEǫ B/A(L)n := Sn(L[1])∨, together with ǫ(ω)(ℓ0, . . . , ℓn) := ± nXi=0 ρ(ℓi)(cid:0)ω(ℓ0, . . . ^ℓi, . . . , ℓn)(cid:1) ± X0≤i<j≤n ω([ℓi, ℓj], ℓ0, . . . ^ℓi, . . . , ^ℓj, . . . , ℓn) . We let the reader check that we indeed have that CEǫ B/A(L) ≃ CE• B/A(L) using PropositionA.2.32 Hence, composing with the forgetful functor cdgaǫ−gr , we get a functor B/A → cdgaǫ−gr B , Lieop B/A→ cdgaǫ−gr B equivalence C0 ≃ A, which has an initial object, denoted DR(A/B) (an explicit description of which is is left adjoint to the functor which lands in the full sub-∞-category consisting of graded mixed cdgas C such that C0 ≃ A. It even factors through the ∞-category consisting of graded mixed cdgas C together with a the data of an provided in [5]). Actually, the functor DR(−/B) : cdgaB → cdgaǫ−gr Lemma A.4. The morphism DR(A/B)→ CEǫ sending a graded mixed B-cdga C to its weight zero component C0. B/A(TA/B) is an equivalence whenever LA/B is coherent Proof. Equivalences between graded mixed objects are checked componentwise. On weight components the map is given by and (cohomologically) bounded above. B which is an equivalence if LA/B is coherent and (cohomologically) bounded above. DR(A/B)n ≃ Sn A(TA/B[1])∨ ≃ CEǫ B/A(TA/B)n , A(LA/B[−1])→ Sn If B is a non-positively graded k-cdga (k being a field, for simplicity) that is almost finitely presented, then we know that DR(Bred/B) ≃ B after [5, Lemma 2.2.4]. Hence we get that CEǫ k/Bred (TBred/B) ≃ CEǫ B/Bred (TBred/B) ≃ B , which we already knew for B almost of finite type. References [1] B. Bhatt. Completions and derived de Rham cohomology. arXiv 1207.6193, July 2012. [2] Lukas Brantner and Akhil Mathew. Deformation theory and partition Lie algebras. arXiv:1904.07352, April 2019. [3] Damien Calaque. A PBW theorem for inclusions of (sheaves of) Lie algebroids. Rendiconti del Seminario Matematico della Universit`a di Padova, 131:23 -- 47, 2014. 32We simply observe that the discrepancy between dCE and ǫ is resolved precisely thanks to the specific form of the realization functor. 45 & / / 9 [4] Damien Calaque, Andrei Caldararu, and Junwu Tu. On the Lie algebroid of a derived self- intersection. Adv. Math., 262:751 -- 783, 2014. [5] Damien Calaque, Tony Pantev, Bertrand Toen, Michel Vaqui´e, and Gabriele Vezzosi. Shifted Poisson structures and deformation quantization. J. Topol., 10(2):483 -- 584, 2017. [6] V. Drinfeld. A letter from Kharkov to Moscow. EMS Surv. Math. Sci., 1(2):241 -- 248, 2014. Trans- lated from the Russian by Keith Conrad. [7] Dennis Gaitsgory and Nick Rozenblyum. Crystals and D-modules. arXiv:1111.2087, October 2014. [8] Dennis Gaitsgory and Nick Rozenblyum. A study in derived algebraic geometry. Vol. I. Correspon- dences and duality, volume 221 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, 2017. [9] Dennis Gaitsgory and Nick Rozenblyum. A study in derived algebraic geometry. Vol. II. Defor- mations, Lie theory and formal geometry, volume 221 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, 2017. [10] Julien Grivaux. Derived geometry of the first formal neighbourhood of a smooth analytic cycle. arXiv:1505.04414, November 2015. [11] Benjamin Hennion. Tangent Lie algebra of derived Artin stacks. Journal fur die reine und ange- wandte Mathematik. [12] Philip J Higgins and Kirill Mackenzie. Algebraic constructions in the category of Lie algebroids. Journal of Algebra, 129(1):194 -- 230, 1990. [13] Vladimir Hinich. Homological algebra of homotopy algebras. Comm. Algebra, 25(10):3291 -- 3323, 1997. [14] Vladimir Hinich. DG coalgebras as formal stacks. JPAA, 162(2-3):209 -- 250, 2001. [15] Vladimir Hinich and Vadim Schechtman. Deformation theory and Lie algebra homology. I. Algebra Colloq., 4(2):213 -- 240, 1997. [16] Vladimir Hinich and Vadim Schechtman. Deformation theory and Lie algebra homology. II. Algebra Colloq., 4(3):291 -- 316, 1997. [17] Daniel Huybrechts. Complex geometry. Universitext. Springer-Verlag, Berlin, 2005. An introduction. [18] M. Kapranov. Rozansky-Witten invariants via Atiyah classes. Compositio Mathematica, 115(1):71 -- 113, 1999. [19] M. Kapranov. Injective resolutions of BG and derived moduli spaces of local systems. J. Pure Appl. Algebra, 155(2-3):167 -- 179, 2001. [20] Mikhail Kapranov. Free Lie algebroids and the space of paths. Selecta Mathematica, 13(2):277 -- 319, October 2007. [21] Kunihiko Kodaira. Complex manifolds and deformation of complex structures. Classics in Mathe- matics. Springer-Verlag, Berlin, english edition, 2005. [22] Maxim Kontsevich. Deformation quantization of Poisson manifolds. Letters in Mathematical Physics. A Journal for the Rapid Dissemination of Short Contributions in the Field of Mathematical Physics, 66(3):157 -- 216, 2003. [23] Jacob Lurie. Derived Algebraic Geometry X: Formal Moduli Problems. preprint available at http: // www. math. harvard. edu/ ~ lurie/ papers/ DAG-X. pdf , 2011. [24] Marco Manetti. Differential graded Lie algebras and formal deformation theory. In D. Abramovich, A. Bertram, L. Katzarkov, R. Pandharipande, and M. Thaddeus, editors, Proceedings of Symposia in Pure Mathematics, volume 80.2, pages 785 -- 810. American Mathematical Society, Providence, Rhode Island, 2009. 46 [25] Joost Nuiten. Koszul duality for Lie algebroids. arXiv:1712.03442, December 2017. [26] Joost Nuiten. Homotopical algebra for Lie algebroids. Appl Categor Struct, pages 1 -- 42, 2019. [27] J. P. Pridham. Unifying derived deformation theories. Adv. Math., 224(3):772 -- 826, 2010. [28] G. S. Rinehart. Differential forms on general commutative algebras. Trans. Amer. Math. Soc., (108):195 -- 222, 1963. [29] Edoardo Sernesi. Deformations of algebraic schemes, volume 334 of Grundlehren der Mathematis- chen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 2006. [30] Bertrand Toen. Probl`emes de modules formels. Ast´erisque, (390):Exp. No. 1111, 199 -- 244, 2017. S´eminaire Bourbaki. Vol. 2015/2016. Expos´es 1104 -- 1119. [31] Bruno Vallette. Homotopy theory of homotopy algebras. arXiv:1411.5533, November 2014. [32] Gabriele Vezzosi. A model structure on relative dg-Lie algebroids. In Stacks and categories in geometry, topology, and algebra, volume 643 of Contemp. Math., pages 111 -- 118. Amer. Math. Soc., Providence, RI, 2015. [33] Shilin Yu. Dolbeault dga and L∞-algebroid of the formal neighborhood. Adv. Math., 305:1131 -- 1162, 2017. D.C.: IMAG, Univ Montpellier, CNRS, Montpellier, France & Institut Universitaire de France Email: [email protected] J.G.: CNRS, I2M (Marseille) & IH´ES Email: [email protected] 47
1912.07518	1	1912	2019-12-16T17:28:25	Large unions of generalized integral sections on elliptic surfaces	[ "math.AG" ]	Let $f \colon X \to B$ be a nonisotrivial complex elliptic surface and let $\mathcal{D} \subset X$ be an integral divisor dominating $B$. We study finiteness related properties of generalized $(S, \mathcal{D})$-integral sections $\sigma \colon B \to X$ of $X$. These integral sections $\sigma$ correspond to rational points in $A(K)$ which satisfy the set-theoretic condition $f ( \sigma(B) \cap \mathcal{D})\subset S$, where $S \subset B$ is an arbitrary given subset. For $S \subset B$ finite, the set of $(S, \mathcal{D})$-integral sections of $X$ is finite by the well-known Siegel theorem. In this article, we establish a general quantitative finiteness result of several large unions of $(S, \mathcal{D})$-integral sections in which both the subset $S$ and the divisor $\mathcal{D}$ are allowed to vary in families where notably $S$ is not necessarily finite nor countable. Some applications to generalized unit equations over function fields are also given.	math.AG	math	LARGE UNIONS OF GENERALIZED INTEGRAL SECTIONS ON ELLIPTIC SURFACES XUAN KIEN PHUNG Abstract. Let f : X → B be a nonisotrivial complex elliptic surface and let D ⊂ X be an integral divisor dominating B. We study finiteness related properties of generalized (S,D)-integral sections σ : B → X of X. These integral sections σ correspond to rational points in A(K) which satisfy the set-theoretic condition f (σ(B) ∩ D) ⊂ S, where S ⊂ B is an arbitrary given subset. For S ⊂ B finite, the set of (S,D)-integral sections of X is finite by the well-known Siegel theorem. In this article, we establish a general quantitative finiteness result of several large unions of (S,D)-integral sections in which both the subset S and the divisor D are allowed to vary in families where notably S is not necessarily finite nor countable. Some applications to generalized unit equations over function fields are also given. Contents Introduction 1. 2. The hyperbolic-homotopic height machinery 3. A key technical lemma 4. Proof of the main result 5. Applications to generalized unit equations over function fields 6. Appendix: Proof of Theorem 5.10 References 1 4 6 9 11 19 21 1. Introduction Notations. We fix throughout a compact Riemann surface B of genus g. Denote K = C(B) its function field. For every complex space X, the pseudo Kobayashi hyperbolic metric on X is denoted by the symbol dX (cf. Definition 2.3). Discs in Riemannian surfaces are assumed to have sufficiently piecewise smooth boundary. The symbol # stands for cardinality. 1.1. Generalized integral points. In this paper, we investigate finiteness related prop- erties of certain unions of generalized integral points in elliptic curves over function fields. The general definition is as follows. Key words and phrases. Siegel's theorem, elliptic surfaces, unit equations, generalized integral sections, function fields, hyperbolicity. 1 9 1 0 2 c e D 6 1 ] . G A h t a m [ 1 v 8 1 5 7 0 . 2 1 9 1 : v i X r a 2 XUAN-KIEN PHUNG Definition 1.1 ((S,D)-integral point and section). Let f : X → B be a proper flat mor- phism. Let S ⊂ B be a subset (not necessarily finite) and let D ⊂ X be a subset (not necessarily an effective divisor). A section σ : B → X is said to be (S,D)-integral if it satisfies the set-theoretic condition (cf. Figure 1): (1.1) For every P ∈ XK(K), let σP : B → X be the corresponding section. Then P is called an (S,D)-integral point of XK if the section σP is (S,D)-integral. f (σ(B) ∩ D) ⊂ S. Figure 1. An (S,D)-integral section σ The above notion of generalized integral points and sections are introduced and studied in a more general context of families of abelian varieties in [22]. It is not hard to see that Definition 1.1 generalizes the notion of integral solutions of a system of Diophantine equations when the subsets S ⊂ B are finite (cf., for example, [21, Introduction]). Let d be a Riemannian metric on B, the following general quantitative result is shown in [22] which generalizes a theorem of Parshin in [20]. Theorem 1.2. Let A/K be an abelian variety with a model f : A → B. Let D ⊂ A be the Zariski closure of an effective ample divisor D ⊂ A. Assume that D does not contain any translate of nonzero abelian subvarieties of A. Let ε > 0. There exists a finite union of disjoint closed discs Zε ⊂ B with vold Zε < ε satisfying the following. Let W be a finite union of disjoint closed discs in B disjoint to Z, let B0 = B \ (W ∪ Zε). There exists m > 0 such that: (*) For Is (s ∈ N) the union of (S,D)-integral points of A over all subsets S ⊂ B such that # (S ∩ B0) ≤ s, we have: for every s ∈ N. (1.2) #Is < m(s + 1)2 dim A. rank π1(B0), Here, a model of A means a N´eron model or a proper flat morphism A → B such that AK (cid:39) A and rank π1(B0) ∈ N denotes the minimal number of generators of π1(B0). Let the notations be as in Theorem 1.2. The trace Trk( ¯C)/k(A)(C) of A (cf. [3], [4]) acts naturally on A(K) = A(B) by translations. If A/K is a nonisotrivial elliptic curve, its trace is zero and the condition D not containing any translate of nonzero abelian subvarieties is empty. In this case, Theorem 1.2 shows that the union Is of (S,D)-integral points is BσXfDSLARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 3 finite and grows at most as a polynomial in terms of s. We remark moreover that the growth order 2 dim A. rank π1(B0) in Theorem 1.2 is a reasonably optimal order that one can possibly expect in general (cf. [22, Remark 1.10]). Classical approaches via height theory can establish the finiteness as well as the polyno- mial growth of Is only when the set W ∪ Zε in Theorem 1.2 is finite. This is the content of the generalized Siegel theorem over function fields (cf. [16]) for integral points of bounded denominators (cf., for example, [23, Corollary 1.7], [25, 2.9]). While the intersection height theory works over an arbitrary field (cf. [9]), the best we can obtain with this tool is a uniform bound on the intersection multiplicities (as in [2], [13, Theorem 0.6], [19]). For this reason, whenever S is infinite, the height with respect to the divisor D of an (S,D)-integral point, which is the sum of the intersection multiplicities over S, cannot be simply bounded using the uniform bound on the intersection multiplicities. 1.2. Main result. In this article, we develop the technique studied in [22] in the case of elliptic curves. It turns out that we can obtain in this situation a very strong property on the finiteness of certain unions Js "twice larger" than Is (cf. (1.2)) consisting of (S,D)- integral points in which both the set S and the divisor D are allowed to vary in families. Moreover, it is shown that the growth of Js in terms of s is still at most polynomial of degree 2π1(B0) as in Theorem 1.2. Recall that an effective divisor D on a fibered variety X over B is called horizontal if the induced map D → B is dominant. Otherwise, D is said to be vertical. Our setting is as follows. Setting (E). Fix a nonisotrivial elliptic surface f : X → B. Denote by T ⊂ B the type of X, i.e., the finite subset above which the fibres of f are not smooth. Let Z be a smooth complex algebraic variety. Let D ⊂ X × Z be an algebraic family of relative horizontal effective Cartier divisors. Assume that D → B × Z is flat. Let Z ⊂ Z be a relatively compact subset with respect to the complex topology. The first main result of the article is the following (cf. Section 4). Theorem A. Let the notations and hypothesis be as in Setting (E). Consider any finite union of disjoint closed discs V ⊂ B containing T such that distinct points in T are contained in different discs. For each s ∈ N, the following union of integral points Js := ∪z∈Z ∪S⊂B,#(S\V )≤s {(S,Dz)-integral points of XK} ⊂ XK(K) is finite. Moreover, there exists m > 0 such that for every s ∈ N, we have: #Js ≤ m(s + 1)2 rank π1(B\V ). Remark 1.3. In fact, the exact same proof of Theorem A presented in this article shows that the conclusion of Theorem A also holds when X → B is an isotrivial elliptic surface, up to replacing #Js by #(Js mod TrK/C(XK)(C)). The proof of Theorem A is a combination of the hyperbolic-homotopic method developed in [22] with a technical Lemma 3.1 which controls locally the hyperbolic metric on certain 4 XUAN-KIEN PHUNG smaller subsets of X \ Dz when z varies. In particular, our proof does not use the usual height theory and thus no height bound is established. Remark 1.4. By the miracle flatness theorem (cf. [27, Lemma 00R4], [18, Theorem 23.1]), the condition requiring D → B × Z to be flat in Setting (E) is equivalent to the condition saying that for every z ∈ Z, the divisor Dz contains no vertical components (or also by [17, Proposition 3.9] since B is an algebraic curve). Theorem A can be seen as a certain generalization of the following theorem of Hindry- Silverman (cf. [13, Theorem 0.6])) which is obtained using the N´eron-Tate height theory. Theorem (Hindry-Silverman). Let (O) be the zero section in X and r = rank XK(K). There exists an explicit function c(g) > 0 such that: #{(S, (O))-integral points of XK} ≤ c(g)((#S)1/2 + 1)r. The above result of Hindry-Silverman holds when B is defined over any algebraically closed field k of characteristic 0. But in the case k = C, our Theorem A gives a much stronger finiteness result without losing the polynomial bound of reasonable degree order. To conclude, some applications to generalized integral sections on ruled surfaces (cf. Theorem B) and to generalized unit equations over function fields (cf. (5.2), Corollary 5.6, Corollary 5.8) are also given in Section 5. 2. The hyperbolic-homotopic height machinery We collect and formulate in this section the tools that are necessary for the article. 2.1. The homotopy reduction step of Parshin. We continue with the notations as in Setting (E). We can suppose without loss of generality that XK[n] ⊂ XK(K) for some integer n ≥ 2. Let U be a disjoint union of closed discs in B such that any two distinct points in T are contained in different discs. Let b0 ∈ B0 := B \ U . Denote Γ = H1(Xb0, Z) and G = π1(B \ U, b0). Since XB0 → B0 is a proper submersion, it is a fibre bundle by Ehresmann's fibration theorem (cf. [5]). It follows that we have an exact sequence of fundamental groups induced by the fibre bundle Xb0 → XB0 → B0 of K(π, 1)-spaces: 0 → π1(Xb0, w0) = H1(Xb0, Z) → π1(XB0, w0) (2.1) ρ =f∗−−−→ π1(B0, b0) → 0. To fix the ideas, w0 is chosen here and in the rest of the article to be the zero point of Xb0, which also lies on the zero section of XB0. We fix a collection of smooth geodesics lw0,w : [0, 1] → Xb0 on the torus Xb0 such that lw0,w(0) = w0 and lw0,w(1) = w. Every (analytic or algebraic) section σP : B0 → XB0 gives rise to a section iP : π1(B0, b0) → π1(XB0, w0) of the exact sequence (2.1) as follows. Given any loop γ of B0 based at b0, we define the section iP by the following formula: (2.2) w0,σP (b0) ◦ σP (γ) ◦ lw0,σP (b0)] ∈ π1(XB0, w0). iP ([γ]) = [l−1 LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 5 Note that as oppose to the composition of homotopy classes, we concatenate oriented paths as maps as above, so the order reverses. The quantitative version of the homotopy reduction step of Parshin can now be stated in our context as follows: Proposition 2.1. Let the notations be as in Setting (E) and let U ⊂ B be a disjoint union of closed discs in B such that any two distinct points in T are contained in different discs. Every section i of the exact sequence (2.1) is induced by at most tX = #XK(K)tors rational points P ∈ XK(K) as in the definition (5.7). Proof. cf. [22, Proposition 9.2], see also [20, Proposition 2.1]. More details and description (cid:3) can be found in [21]. 2.2. Hyperbolic height on Riemann surfaces. One of the key ingredients in the proof of Theorem 1.2 in [22] is the following linear bound on the hyperbolic length of loops in various complements of a Riemann surface. Let U be any finite union of disjoint closed discs in the Riemann surface B and denote B0 := B \ U . It is shown in [22] that: Theorem 2.2. For every free homotopy class α ∈ π1(B0), there exists L > 0 with the following property. For any finite subset S ⊂ B0, there exists a piecewise smooth loop γ ⊂ B0 \ S which represents the free homotopy class α in B0 and satisfies: (2.3) (γ) ≤ L(#S + 1). lengthdB0\S Recall that dB0\S denotes the intrinsic Kobayashi hyperbolic metric on B0 \ S (cf. Defi- nition 2.3) 2.3. Metric properties of hyperbolic manifolds. In this section, we collect some fun- damental properties of the pseudo Kobayashi hyperbolic metric and of hyperbolic man- ifolds due to Green. Let X be a complex manifold. The pseudo Kobayashi hyperbolic metric dX : X × X → X is defined as follows. Let ρ be the Poincar´e metric on the unit disc ∆ = {z ∈ C : z = 1}. Let x, y ∈ X. Consider the data L consisting of a finite sequence of points x0 = x, x1, . . . , xn = y in X, a sequence of holomorphic maps fi : ∆ → X and of pairs (ai, bi) ∈ i=0 ρ(ai, bi). ∆2 for i = 0, . . . , n such that fi(ai) = xi and f (bi) = xi+1. Let H(x, y; L) =(cid:80)n Definition 2.3 (cf. [14]). For x, y ∈ X, we define dX(x, y) := inf L H(x, y; L). X is called a hyperbolic manifold if dX(x, y) > 0 for all distinct points x, y ∈ X, i.e., when dX is a metric. The fundamental distance-decreasing property (cf. [15, Proposition 3.1.6]) can be stated as follows: Lemma 2.4. Let f : X → Y be a holomorphic map of complex manifolds. Then for all x, y ∈ X, dY (f (x), f (y)) ≤ dX(x, y). In particular, if X ⊂ Y , dY X ≤ dX. A complex space X is said to be Brody hyperbolic if it does not contains entire curves, i.e., there is no nonconstant holomorphic maps C → X. 6 XUAN-KIEN PHUNG Theorem 2.5 (Green). Let X be a relatively compact open subset of a complex manifold M . Let D ⊂ X be a closed complex subspace. Denote by ¯X, ¯D the closures of X and D in M . Assume that ¯D and ¯X \ ¯D are Brody hyperbolic. Then X \ D is hyperbolic and we have dX\D ≥ ρX\D for some Hermitian metric ρ on M . In particular, if M is compact and λ is any Riemannian metric on M then there exists c > 0 such that dX\D ≥ cλX\D. (cid:3) Proof. See [11, Theorem 3]. Remark 2.6. Let X be a smooth complex manifold. Let ∆(0, R) = {z ∈ C : z < R} for every R > 0. The infinitesimal Kobayashi-Royden pseudo metric λX on X corresponding to the Kobayashi pseudo hyperbolic metric dX can be defined as follows. For x ∈ X and every vector v ∈ TxX, λX(x, v) := inf 2/R, where the minimum is taken over all R > 0 for which there exists a holomorphic map f : ∆(0, R) → X such that f(cid:48)(0) = v. Thanks to the distance-decreasing property of the pseudo-Kobayashi hyperbolic metric, we have the following important property of sections. Lemma 2.7. Let f : X → Y be a holomorphic map between complex spaces. Suppose that σ : Y → X is a holomorphic section. Then σ(Y ) is a totally geodesic subspace of X, i.e., for all x, y ∈ Y , we have dY (x, y) = dX(σ(x), σ(y)). Proof. By Lemma 2.4, dY (x, y) = dY (f (σ(x)), f (σ(y))) ≤ dX(σ(x), σ(y)) ≤ dY (x, y). (cid:3) 3. A key technical lemma For the proof of Theorem A, we fix a Hermitian metric ρ on the smooth surface X. It is clear that we can assume Z integral. Define B0 := B \ V and fix a system generators α1, . . . , αk consisting of free homotopy classes of the fundamental groupoid π1(B0). For a complex space Y , the symbol dY always denotes the Kobayashi hyperbolic pseudo-metric on Y (cf. Definition 2.3). The proof of Theorem A is based on the following key technical lemma which is of local nature. The main idea is the following. For each z ∈ Z, the hyperbolic metric on (X \ Dz)B0 dominates the Hermitian metric ρ by Green's theorem 2.5 up to a certain strictly positive factor. When z varies in Z, these factors vary as well and may a priori tend to 0. The point of Lemma 3.1 below is that for z in a small neighborhood of Z, we can, up to restricting further (X \ Dz)B0 over some fixed nice complement of B0 (cf. Property (b) below), these factors are in fact bounded below by a strictly positive constant (cf. Property (P) below). Lemma 3.1. Let the notations be as in Theorem A. Let ε > 0. Then there exists M > 0 such that for each zi ∈ Z, we have the following data: (a) an analytic open neighborhood Ui of zi in Z; (b) a disjoint union Vi ⊂ B0 consisting of ≤ M closed discs each of radius ≤ ε; (c) a constant ci > 0; LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 7 with the following property: (P) for each z ∈ Ui, we have d(X\Dz)(B0\Vi) ≥ ciρ(X\Dz)(B0\Vi). We remark first a standard lemma. Lemma 3.2. Let f : X → Y be a proper flat morphism of integral complex algebraic varieties of the same dimension. Assume that Y is a smooth manifold and for some y ∈ Y , the fibre Xy is reduced and finite. Then there exists an analytic open neighborhood U ⊂ Y of y such that f−1(U ) → U is a finite ´etale cover. Proof. Since Xy is finite and reduced and f is flat, every point x ∈ Xy is a smooth point of f since we are in characteristic 0 (cf. [17, Lemma 3.20]). Since Y is regular, we deduce from the open property of smooth morphisms (cf. [10, Definition 6.14]) that every point [17, Theorem 4.3.36]). Since dim X = dim Y and Y of Xy is a regular point of X (cf. is a smooth manifold, [10, Definition 6.14] implies that the map f is submersive at each point of the fibre Xy. Let S ⊂ X be the set of singular points of X (i.e., where X is not locally a manifold) and let S(cid:48) ⊂ X be the set where f is not submersive. Since f is proper, A = f (S) ∪ f (S(cid:48)) is a closed subset of Y . Then by the proof of [12, Theorem 21, Chapter III], X \ f−1(A) → Y \ A is an ´etale cover. We have seen that Xy ∩ (S ∪ S(cid:48)) = ∅. Therefore, y /∈ A and there exists an analytic open neighborhood U ⊂ Y \ A of y such that f−1(U ) → U is a finite ´etale cover. (cid:3) For ease of reading, we mention here the following key theorem of Brody: Theorem 3.3 (Brody's reparametrization lemma). Let M be a complex manifold with (possibly empty) boundary. Let H be a Hermitian metric on M . Suppose that f : ∆R → M be a holomorphic map with df (0)H > c form some c > 0. Then there exists a holomorphic map g : ∆R → M satisfying the following conditions: (a) dg(0)H = c; (b) dg(z)H ≤ cR2 (c) Im(g) ⊂ Im(f ). R2−z2 for all z ∈ ∆R; Proof. See Brody's reparametrization lemma, page 616 in [11] or [1]. (cid:3) For the proof of Lemma 3.1, we claim first that there exists N > 0 such that the total number of irreducible components (counted with multiplicities) of each effective divisor Dz is at most N for every z ∈ Z. Indeed, let H be any ample divisor on X then C · H ≥ 1 for every irreducible curve C ⊂ X. Remark that the divisors Dz are all algebraic equivalent (since Z is integral hence connected by curves) thus numerically equivalent. Therefore, N := Dz · H is a constant independent of z ∈ Z. By the linearity of the intersection pairing, the above two remarks clearly show that the total number of irreducible components (counted with multiplicities) of Dz is at most N as claimed. For each z ∈ Z, notice that the effective divisor Dz contains only horizontal components with respect to the fibration f : X → B by Remark 1.4. We denote by (Dz)red the induced 8 XUAN-KIEN PHUNG reduced scheme structure of Dz. By the adjunction formula, we find that p1 := pa(Dz) = 2 Dz(Dz + KX) + 1 ≥ 0 is a constant independent of z ∈ Z. Now, since the arithmetic genus of Dz are uniformrly bounded, a version of the Riemann-Hurwitz theorem (cf. [17, Propositions 7.4.16, 7.5.4]) for the ramified cover of algebraic curves πz : (Dz)red → B implies that for all z ∈ Z: 2(p1 − 2) + 2N ≥ #{ramification points of πz} + #{singular points of (Dz)red}. Let Tz ⊂ B be the image in B of the union of the ramification points of πz and of the singular points of (Dz)red. It follows that Tz is finite and we have: for all z ∈ Z. (3.1) #Tz ≤ M := 2(p1 − 2) + 2N, We can now return to the proof of Lemma 3.1. Proof of Lemma 3.1. Let us fix ε > 0 sufficiently small and zi ∈ Z. Clearly, we can choose a finite disjoint union Vi of at most M (defined in (3.1)) nonempty open discs in B0 of ρ-radius ≤ ε to cover the points of Tzi \ V . Recall that Tzi ⊂ B is the image in B of the union of the ramification points of πzi : (Dzi)red → B and of the singular points of (Dzi)red. This gives the data Vi for (b). Denote Yz := (X \ Dz)B0\Vi for each z ∈ Z. To show the existence of Ui and ci so that (i) is satisfied, we suppose the contrary. Hence, by the continuity of ρ on the compact unit tangent space of X, there would exist a sequence (zn)n≥1 ⊂ Z such that zn → zi in the analytic topology and a sequence of holomorphic maps hn : ∆Rn → Yzn n/(R2 gn : ∆Rn → Yzn such that Rn → ∞ and dhn(0)ρ > 1. Here, ∆R ⊂ C denotes the open disc of radius R in the complex plane. By Brody's reparametrization lemma (Theorem 3.3), we obtain a sequence of holomorphic maps n − z2) for all z ∈ ∆Rn. In particular, such that dgn(0)ρ = 1 and dgn(z)ρ ≤ R2 dgn(z)ρ ≤ 4/3 for all z ∈ ∆Rn/2. It follows that the family (gn)n≥1 is equicontinuous with image inside the compact space XB0\Vi. By the Arzela-Ascoli theorem, we deduce, up to passing to a subsequence, that (gn)n≥1 converges uniformly on compact subsets of C to a map g : C → XB0\Vi. A standard argument using Cauchy's theorem and Morera's theorem shows that g is a holomorphic map and we also have dg(0)ρ = 1. Consider the holomorphic composition map π◦g : C → B0\Vi. Since B0\Vi is hyperbolic, it is Brody hyperbolic and thus the map π◦g is a constant b∗ ∈ B0\Vi. Remark that B× Z is smooth and D → B× Z is a proper flat morphism of relative dimension 0 by hypotheses. Since moreover the fibre of Dred over (b∗, zi) is reduced and finite, we can apply Lemma 3.2. It follows that there exists a constant λ ∈ N, a small analytic open disc ∆ ⊂ B0 \ Vi containing b∗ and a small analytic connected open neighborhood Ui of zi in Z such that D∆×Ui → ∆ × Ui is an ´etale cover of degree λ. Up to shrinking ∆ and Ui, we can suppose LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 9 that D∆×Ui consists of λ disjoint connected components. In particular, Dz∆ → ∆ is an ´etale cover consisting of disjoint λ-sheets for every z ∈ Ui. Fix a connected component D0 of D∆×Ui. We thus obtain a family of 1-sheeted covers z → z → ∆ with z ∈ Ui. Each D0 z is a complex submanifold of X∆ via the inclusion ιz : D0 D0 z → ∆ is thus holomorphic and bijective for every z ∈ Ui. X∆. The composition π ◦ ιz : D0 Since every injective holomorphic map is biholomorphic to its image (cf. [24, Theorem 2.14, Chapter I]), we deduce that the map sz : ∆ → D0 z(t) = π−1(t)∩ D0 is holomorphic for every z ∈ Ui. z → X∆ given by t (cid:55)→ D0 We can write X∆ ⊂ P2 × ∆ as defined by the Weierstrass equation: z y2 = x3 + A(t)x + B(t) where A(t), B(t) are holomorphic functions on ∆ such that the discriminant 4A3 + 27B2 does not vanish on ∆ (since ∆ ∩ T = ∅ where we recall that T ⊂ B is the finite subset above which the fibres of f are not smooth). We can thus write sz(t) = (u(t, z), v(t, z)) where u, v : ∆ × Ui → C are holomorphic functions. Since the translation maps on the elliptic fibration X∆ are algebraic thus holo- morphic, the maps Ψz : X∆ → X∆ given by the translations by sz − szi: Ψz(x) = x + sz(f (x)) − szi(f (x)), x ∈ X∆, form a smooth family of biholomorphisms commuting with the map f : X∆ → ∆. Since gn → g uniformly on compact subsets of C and Im(g) ⊂ f−1(b∗), we can, up to passing to a subsequence with suitable restrictions of domains of definitions, suppose that Im(gn) ⊂ X∆ and that the holomorphic maps gn : ∆Rn → X∆ \ Dzn still satisfy the properties: into the fixed space X∆ \ Dzi. Then by the smoothness of the family of biholomorphisms (ψz)z∈Ui and the compactness of the ρ-unit tangent bundle of X, there exists a constant c > 1 such that Since Rn → ∞, Remark 2.6 then implies immediately a contradiction to the fact that X∆ \ Dzi is hyperbolically embedded in X∆ = f−1(∆) (for the Hermitian metric ρf−1(∆)) by Green's theorem 2.5. To check the latter fact, it suffices to remark that ¯∆, thus Dzi ¯∆ are Brody hyperbolic and moreover, the fibres of X ¯∆ \ (Dzi ¯∆) are also Brody hyperbolic. (cid:3) Hence, the existence of the data in (a), (b) and (c) such that (P) is satisfied. c−1 ≤ dfn(0)ρ ≤ c, for all n ≥ 1. 4. Proof of the main result We can now return to the proof of Theorem A. Proof of Theorem A. We have defined B0 = B \ V at the beginning of Section 3 and we have fixed a system of simple generators α1, . . . , αk of the fundamental group π1(B0) with Consider now the sequence of holomorphic maps Rn → ∞, dgn(0)ρ = 1. fn := ψzn ◦ gn : ∆Rn → X∆ \ Dzi 10 XUAN-KIEN PHUNG fixed based point. Recall that Z is a compact subset of Z with respect to the complex topology. Fix ε > 0 sufficiently small. We can enlarge slightly the discs in V if necessary. Then we obtain a constant M > 0 and a set of data (Ui, Vi, ci) for each element zi ∈ Z as in Lemma 3.1. We have obviously an open covering Z ⊂ ∪zi∈ZUi of Z. Since Z is compact, there exists a finite subset Z∗ ⊂ Z such that Z ⊂ ∪zi∈Z∗Ui. As ci > 0 for every zi ∈ Z by Lemma 3.1, we can define ci > 0. c∗ := min zi∈Z∗ (4.1) Denote Wi := V ∪ Vi ⊂ B and Bi := B \ Wi = B0 \ Vi for each zi ∈ Z∗. Since the Vi's are union of disjoint closed discs, they are contractible. Hence, we have a canonical inclusion π1(B0) ⊂ π1(Bi) for each zi ∈ Z∗. For every zi ∈ Z∗, we denote by Li > 0 the maximum of the constants given by Theorem 2.2 applied to the the compact Riemann surface B and the disjoint union of closed discs Wi and to each free homotopy classes α1, . . . , αk viewed as elements of π1(Bi). We define: (4.2) Li ∈ R+. Now let P ∈ Js, i.e., P ∈ XK(K) is an (S,Dz)-integral point for some z ∈ Z and for some S ⊂ B such that #(S ∩ B0) ≤ s. Since Z ⊂ ∪zi∈Z∗Ui, there exists zi ∈ Z∗ such that z ∈ Ui. By the definition of the constant Li and by Theorem 2.2 applied to Bi, there exists bi ∈ Bi and a system of loops γ1, . . . , γk based at bi representing respectively the homotopy classes α1, . . . , αk up to a single conjugation such that γj ⊂ Bi \ S and that (4.3) (γj) ≤ Li(s + 1). Now, let σP : B → X be the corresponding section of the rational point P . For every lengthdBi\S L = max zi∈Z∗ j ∈ {1, . . . , k}, we have σP (γj) ⊂ (X \ Dz)Bi\S ⊂ (X \ Dz)Bi. This is true because P is (S,Dz)-integral so that σP (γj) cannot intersect Dz outside of f−1(S) and because γj ⊂ Bi \ S. It follows that: i i (σP (γj)) lengthρ(σP (γj)) ≤ c−1 lengthd(X\Dz )B0\Vi ≤ c−1 lengthd(X\Dz )Bi\S ≤ c−1∗ lengthdBi\S (γj) ≤ c−1∗ Li(s + 1) ≤ c−1∗ L(s + 1). Remark that the second inequality in the above follows from Lemma 2.4 and from the definition Bi = B0 \ V0 ⊃ Bi \ S. (by Lemma 3.1) (as (X \ Dz)Bi\S ⊂ (X \ Dz)Bi) (by (4.1) and Lemma 2.7) (by (4.2)) (by (4.3)) (σP (γj)) LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 11 Now let δ be the diameter of XB0∪∂B0 with respect to the metric ρ. Remark that by hypothesis, distinct points of the non smooth locus T ⊂ B of f are contained in distinct discs of V . It follows that the homotopy section iP (cf. (5.7)) associated to the rational point P of the short exact sequence (2.1) 0 → π1(Xb0, w0) → π1(XB0, w0) → π1(B0, b0) → 0 sends the basis (αj)1≤j≤k of π1(B0, b0) to the homotopy classes in π1(XB0, w0) which admit representative loops of ρ-lengths bounded by H(s) := c−1∗ L(s + 1) + 2δ. The constant 2δ in the definition of H(s) corresponds to the bound on the length of the extra paths induced by the change of base points from σP (bi) to w0 = σO(b0). From the above bound H(s) on the length of the image loops by (S,Dz)-integral sections for every z ∈ Z and every S ⊂ B such that #(S ∩ B0) ≤ s, the rest of the proof follows exactly the same lines as in the proof of [22, Theorem A] which uses the homotopy reduction Proposition 2.1 and the geometry of the fundamental groups π1(Xb, wb) for b ∈ B0 and wb ∈ Xb as a counting lemma (cf. [22, Lemma 13.10]). Therefore, we obtain a constant m > 0 (independent of s) such that: #Js ≤ m(s + 1)2 rank π1(B0), for every s ∈ N. (cid:3) 5. Applications to generalized unit equations over function fields 5.1. Statement of the main result. The goal of the present section is to apply the approach used to prove Theorem A to obtain similar results in the case of ruled surfaces. Throughout this section, we will fix a compact connected Riemann surface B of function fields K = C(B), and a finite subset S ⊂ B. Moreover, the following definitions and notations are used: associated to the points 0,∞ ∈ P1(K); (1) BS := B \ U where U is a finite disjoint union of closed discs centered at points of S; (2) B0 := BS \ V where V ⊂ BS is a finite disjoint union of closed discs; (3) X = P1C × B and f : X → B is the second projection; (4) ρ is a fixed Hermitian metric on X; (5) σx : B → X denotes the section induced by x ∈ K and let (x) := σx(B); (6) F = Gm = P1 \ {0,∞} is the fibre of f and (0), (∞) ⊂ X are the constant sections (7) Y = F × B0 = f−1(B0) \ ((0) ∪ (∞)) ⊂ X0 = P1 × B0. Remark 5.1. The set of sections of F × (B \ S) → B \ S is canonically identified with the set O∗ S of S-units of K. Indeed, sections of the surface F × (B \ S) are exactly sections of X which do not intersect (0) and (∞) at points lying above B \ S. On the other hand, each element x ∈ K∗ corresponds canonically to a non zero section denoted σx ⊂ X and vice versa by the valuative critera for properness. The condition x ∈ OS (resp. x−1 ∈ OS) means exactly that σx(B) does not intersect (∞) (resp. (0)) at points lying above B \ S. Therefore, x ∈ O∗ S if and only if σx is a section of F × (B \ S) → B \ S as claimed. 12 XUAN-KIEN PHUNG Definition 5.2. Let D ⊂ X be an effective divisor and let R ⊂ B be a subset. We say that a point x ∈ K∗ is (R, D)-integral if it satisfies f (σx(B) ∩ D) ⊂ R. Definition 5.3. For each subset R ⊂ B, the generalized ring of R-integers of K is denoted by OR = {x ∈ K : valν(x) ≥ 0, ∀ν ∈ B \ R} ⊂ K. Definition 5.4. For each ε ≥ 0, we define the (B0, ε)-interior subset of O∗ S by (cid:9) . S(B0, ε) :=(cid:8)x ∈ O∗ O∗ S : x(t),x(t)−1 > ε, ∀t ∈ B0 Equivalently, the image of every meromorphic function x ∈ O∗ the ε-neighborhoods of 0 and ∞ in P1. Remark that O∗ Moreover, we have O∗ S = ∪ε>0O∗ S(B0, ε). S(B0, 0) = O∗ S(B0, ε) on B does not meet S since 0,∞ /∈ B0. The main result of this section is the following quantitative finiteness result for large unions of generalized integral points on rational curves over function fields. Theorem B. Let Z be a smooth complex algebraic variety and Z ⊂ Z a compact subset with respect to the complex topology. Suppose that D ⊂ X × Z is a family of effective divisors such that D → B × Z is flat and Dz is not contained in (0)∪ (∞) for every z ∈ Z. For every r ∈ N and ε > 0, the following union of integral points: Jr,ε := ∪z∈Z ∪R⊂B,#R∩B0≤r {x ∈ O∗ S(B0, ε) : x is (R,Dz)-integral in X} ⊂ O∗ S, is finite modulo C∗. Moreover, there exists a constant m > 0 such that: #(Jr,ε mod C∗) ≤ m(r + 1)2 rank π1(B0) for every r ∈ N. The proof of Theorem B given in Section 5.3 will follow closely the steps in the proof of Theorem A. It might be helpful to first consider the following counter-example explaining why we cannot take the whole set of S-units O∗ S, i.e., ε = 0, in the union Jr,ε. For every n ∈ Z, we define cn =(cid:0)2 supt∈B0 tn(cid:1)−1. Then cn > 0 and is a finite number Example 5.5. Let the notations and hypotheses be as in Theorem B. Assume moreover that B = CP1 is the Riemann sphere and that S = {0,∞}. Let t be the inhomogenous S = C∗.{tn : n ∈ Z} = {ctn : c ∈ C∗, n ∈ Z}. Suppose also that coordinate on P1 then O∗ Z = Z = {·} and D = (1) ⊂ X is the section induced by 1 ∈ K. since B0 ⊂ CP1 is a complement of a finite union U ∪ V of closed discs with nonempty interior containing 0 and ∞. It follows that 0 < cntn < 1 for all t ∈ B0. Therefore, for each n ∈ Z, xn := cntn ∈ O∗ S is (R,D)-integral in X in the sense of Definition 5.3 where R = U ∪ V . In particular, {xn : n ∈ Z} ⊂ Jr,0 for all n ∈ Z and r ∈ N. However, {xn : n ∈ Z} modulo C∗ is Z and thus Jr,0 is infinite. Remark that there exists t0 ∈ B0 ⊂ C such that t0 (cid:54)= 0, 1. Hence, if t0 > 1, we see that cn → 0 when n → −∞. Otherwise, if 0 < t0 < 1 then cn → 0 as n → +∞. Therefore, it is necessary to restrict to the union of integral points Jr,ε where ε > 0 to obtain a finiteness result as in Theorem B. LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 13 Let DS :=(cid:80) 5.2. Some applications to generalized unit equations. We illustrate in this section several applications of Theorem B in a context generalizing the classical S-unit equations (cf. [6], [7], [8]). b∈S[b] be the effective divisor of B associated to the finite subset S ⊂ B. Let r ∈ N, consider the following subset of K: OB0,r := ∪R⊂B0,#R≤r{x ∈ K : valν(x) ≥ 0, for every ν /∈ R ∪ B0} (5.1) = ∪R⊂B0,#R≤rOR. and OB0,r is not a ring unless r = 0. Remark that OB0,r ⊂ OB0,r+1 for every r ≥ 0 and K = ∪r≥0OB0,r. Moreover, OS ⊂ OB0,0 For integers n ≥ 1, r ≥ 0 and a real number ε > 0, we consider the Diophantine equation x + y = z S(B0, ε); (5.2) with (x, y, z) ∈ K 3 satisfying the following conditions: (i) x ∈ O∗ (ii) y−1 ∈ OB0,r; (iii) z ∈ L(nDS) \ {0} = {h ∈ K∗ : div(h) + nDS ≥ 0} ⊂ K∗. In other words, we consider the union of solutions (x, y) ∈ O∗ S(B0, ε) × (OB0,r \ {0})−1 In the case r = ε = 0 and B0 = B \ S, we recover the usual S-unit equation x + y = 1 S by setting z = 1 ∈ L(nDS) \ {0}. Indeed, r = 0 and B0 = B \ S imply S and y−1 ∈ OS such that x + y = 1, then S. Therefore, (5.2) generalizes the usual S with x/y /∈ C∗ of the unit equation It turns out that Theorem B actually implies that a similar of the parametrized equations x + y = z with z varying in the space ∈ L(nDS) \ {0}. with x, y ∈ O∗ OB0,r = OS. On the other hand, if x ∈ O∗ y = 1 − x ∈ OS since OS is a ring and thus y ∈ O∗ S-unit equation over function fields. The finiteness of the numbers of solutions x, y ∈ O∗ x + y = 1 is well-known. property still holds for the generalized equation (5.2). Corollary 5.6. There are only finitely many x ∈ O∗ equation (5.2) admits a solution. Proof of Corollary 5.6. Denote d = dim L(nDS) = dim H 0(B,O(nDS)). Fixing a basis (z1, . . . , zd) of the complex vector space L(nDS), we consider the compact unit sphere S(B0, ε) modulo C∗ such that the (cid:41) (cid:40) d(cid:88) (cid:16)(cid:80)d k=1 Sd = akzk ∈ L(nDS) : (a1, . . . , ak) = 1 ⊂ L(nDS) (cid:17)1/2 . Recall that DS :=(cid:80) b∈S[b]. where (a1, . . . , ak) = k=1 ak Define Z := L(nDS) \ {0} (cid:39) Cd \ {0} then Z is a integral smooth algebraic variety. We have a canonical valuation morphism val : B × Z → P1, (b, z) (cid:55)→ z(b). 14 XUAN-KIEN PHUNG Consider the flat family of divisors D ⊂ X × Z given by the image of the algebraic section Σ : B × Z → P1 × B × Z, (b, z) (cid:55)→ (z(b), b, z). For each z ∈ Z ⊂ K∗, let (z) ⊂ X be the induced section of the projection f : X → B. It is clear that (z) = Dz (cid:54)⊂ (0) ∪ (∞) for every z ∈ Z by the construction of D. Now let z ∈ Z ⊂ K∗, x ∈ O∗ S(B0, ε) and y = z − x. It is not hard to see that the condition y−1 ∈ OR for a certain subset R ⊂ B verifying #(R ∩ B0) ≤ r means exactly that (5.3) We cannot apply directly Theorem B since Z = L(nDS) \ {0} is not compact. However, it suffices to restrict ourselves to the case z ∈ Z := Sd since the compact subspace Sd contains all classes modulo C∗ of Z. Therefore, Theorem B says that the set S(B0, ε) : x(cid:48) is (R,Dz)-integral in X}. x ∈ ∪R⊂B,#R∩B0≤r{x(cid:48) ∈ O∗ Jr,ε = ∪z∈Sd ∪R⊂B,#R∩B0≤r {x(cid:48) ∈ O∗ S(B0, ε) : x(cid:48) is (R,Dz)-integral in X} is finite. Combining with (5.3), the proof of Corollary 5.6 is completed. (cid:3) Following the general idea that parametrized Diophantine equations have no or very few integral solutions under a general choice of parameters, we mention below a remarkable theorem on unit equations in the case of number fields. Theorem 5.7 (Evertse-Gyory-Stewart-Tijdeman). Let K be a number field and S a finite number of places. There exists a finite set of triples A ⊂ (K∗)3 with the following property. For every α = (α1, α2, α3) ∈ (K∗)3 whose class [α] ∈ (K∗)3/(K∗(O∗ S)3) does not belong to [A] ⊂ (K∗)3/(K∗(O∗ (5.4) S)3), the S-unit equations α1x + α2y = α3 has at most 2 solutions. Proof. See [8, Theorem 1]. Note that the natural action of K∗(O∗ (c, (u, v, w)) · (α1, α2, α3) = (cuα1, cvα2, cwα3). S)3 on (K∗)3 is given by: (cid:3) Theorem 5.7 implies that almost all equations of the form (5.4) have no more than 2 unit solutions. As an analogous result for certain Diophantine equations in function fields, Corollary 5.6 can be directly reformulated as follows. Corollary 5.8. Given ε > 0, let ω ∈ O∗ (5.5) with unknowns x, y ∈ K∗ satisfying x ∈ L(nDS) \ {0} and y−1 ∈ OB0,r (cf. (5.1)). There exists a finite subset A ⊂ O∗ S(B0, ε) such that whenever ω /∈ C∗A, the equation (5.5) has no solutions. Moreover, there exists m > 0 such that we can take A having no more than m(r + 1)2 rank π1(B0) elements for every r ∈ N. S(B0, ε) and n ≥ 1. Consider the equation x + y = ω Note that the above last statement follows immediately from Theorem B. 5.3. Proof of Theorem B. LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 15 5.3.1. Preliminaries. We begin with an easy analogue of the Lang-N´eron theorem for the multiplicative group Gm. Recall that S is a finite subset of B. Proposition 5.9. O∗ S/C∗ is a torsion-free abelian group of rank ≤ #S. Proof. Consider the following homomorphism of groups ρ : O∗ S → ⊕ν∈SZ, f (cid:55)→ (multν(div f ))ν∈S. We claim that Ker ρ = C∗. Indeed, suppose that f ∈ O∗ S satisfies ρ(f ) = 0. Since f ∈ O∗ S, all poles and zeros of f belongs to S. However, ρ(f ) = 0 implies that these poles and zeros are all of order 0. It follows that the corresponding morphism f : B → P1 must be constant and thus f ∈ C∗ as claimed. Therefore, we have an injective homomorphism of groups O∗ (cid:3) S = Gm(OS) induces a section σz of the projection Y → B0 and thus a S/C∗ is a torsion-free abelian group of rank ≤ #S. S/C∗ → ⊕ν∈SZ. O∗ Now each z ∈ O∗ section iz of the following exact sequence of fundamental groups: 0 → π1(Yb0, w0) → π1(Y, w0) η−→ π1(B0, b0) → 0, (5.6) where we fix w0 = 1 ∈ Yb0 = C∗ above the fixed point b0 ∈ B0. Fix a collection of geodesics lw0,w : [0, 1] → Yb0 on Yb0 such that lw0,w(0) = w0 and lw0,w(1) = w ∈ Yb0. Every x ∈ O∗ S induces a section σx : B0 → YB0 which in turn gives rise to a section ix : π1(B0, b0) → π1(YB0, w0) of the exact sequence (2.1) as follows. For every loop γ of B0 based at b0, we define: Denote G = π1(B0, b0) and let (cid:98)G be the profinite completion of G. As in the case of (5.7) w0,σx(b0) ◦ σx(γ) ◦ lw0,σx(b0)] ∈ π1(YB0, w0). ix([γ]) = [l−1 elliptic surfaces (Proposition 2.1), we have the following reduction result. Theorem 5.10. Let n ≥ 2 be an integer. We have the following commutative diagram of homomorphisms of groups: O∗ S/(O∗ S)n δ H 1((cid:98)G, µn) (cid:39) O∗ Moreover, two elements of O∗ tal groups (5.6) if and only if they differs by a factor in C∗. H 1(G, Z) α S β H 1(G, µn). S induces the same section of the exact sequence of fundamen- The proof of Theorem 5.10 will be given in Appendix 6. 5.4. Key lemma. Recall that dM means the Kobayashi hyperbolic pseudo-metric on the complex space M and ρ is a fixed Hermitian metric on the smooth surface X. As in the case of elliptic fibrations, the main additional ingredient in the proof of Theorem B for ruled surfaces is the following analogous technical lemma of Lemma 3.1: 16 XUAN-KIEN PHUNG Lemma 5.11. Let ε > 0. Then there exists M > 0 such that for each zi ∈ Z, we have the following data: (a) an analytic open neighborhood Ui of zi in Z; (b) a disjoint union Vi consisting of ≤ M discs each of ρ-radius ≤ ε in B0; (c) a constant ci > 0; with the following property: (Q) for each z ∈ Ui, we have d(Y \Dz)(B0\Vi) ≥ ciρ(Y \Dz)(B0\Vi). The proof of Lemma 5.11 applies, mutatis mutandis, the proof of Lemma 3.1 with some minor modifications. The same remark at the beginning of the proof of Lemma 3.1 shows that there exists N(cid:48) > 0 such that the total number of irreducible components (counted with multiplicities) of each effective divisor Dz is at most N(cid:48) for every z ∈ Z. Moreover, since D → B × Z is flat, every divisor Dz, z ∈ Z, contains no vertical components with respect to the projection f : X → B. The divisors Dz are numerically equivalent and are not contained in the curve (0)∪ (∞) ⊂ X. In particular, it follows that for some constant N(cid:48)(cid:48) > 0, we have #D ∩ ((0) ∪ (∞)) ≤ N(cid:48)(cid:48), for all z ∈ Z. By the adjunction formula, p1 := pa(Dz) = Dz(Dz + KX) + 1 ≥ 0 2 is a constant independent of z ∈ Z. Let T (cid:48) z ⊂ B be the image in B of the union of the ramification points of the ramified cover of algebraic curves πz : (Dz)red → B, and of the singular points of (Dz)red. We have as in the relation (3.1) that: #T (cid:48) z ≤ M(cid:48) := 2N(cid:48) + 2(p1 − 2), for all z ∈ Z. z ∪ f (D ∩ ((0) ∪ (∞))) ⊂ B then it follows from the above discussion that #Tz ≤ M := M(cid:48) + N(cid:48)(cid:48), for all z ∈ Z. Define Tz = T (cid:48) (5.8) We return to the proof of Lemma 5.11. Again, the idea of the proof is the same as in Lemma 3.1 but we indicate in details the needed modifications. Proof of Lemma 5.11. Fix ε > 0 and zi ∈ Z. Recall that B0 = B \ (U ∪ V ). Let M be the constant defined in (5.8). We can clearly choose a finite disjoint union Vi of at most M nonempty closed discs in B0 of ρ-radius ≤ ε to cover the points of Tzi \ V . Thus, we obtain the data Vi for Lemma 5.11.(b). Define D = D ∪ (((0) ∪ (∞)) × Z) ⊂ X × Z and for each z ∈ Z, let Yz := (Y \ Dz)B0\Vi = (X \ Dz)B0\Vi ⊂ X. To show the existence of Ui, ci satisfying (i), we suppose the contrary. By the continuity of ρ on the compact unit tangent space of X, there would exist a sequence (zn)n≥1 ⊂ Z LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 17 such that zn → zi in the analytic topology and a sequence of holomorphic maps hn : ∆Rn → Yzn gn : ∆Rn → Yzn such that Rn → ∞ and dhn(0)ρ > 1. Here, ∆R ⊂ C denotes the open disc of radius R in the complex plane. By Brody's reparametrization lemma (Theorem 3.3), we obtain a sequence of holomorphic maps such that dgn(0)ρ = 1 and dgn(z)ρ ≤ R2 n − z2) for all z ∈ ∆Rn. In particular, dgn(z)ρ ≤ 4/3 for all z ∈ ∆Rn/2. It follows that the family (gn)n≥1 is equicontinuous with image inside the compact space XB0\Vi. Up to passing to a subsequence, (gn)n≥1 converges uniformly on compact subsets of C to a holomorphic map g : C → XB0\Vi with dg(0)ρ = 1. Since B0\Vi is hyperbolic, f ◦g : C → B0\Vi is a constant b∗ ∈ B0\Vi. As in Lemma 3.1, we can apply Lemma 3.2 to find a constant λ ∈ N, a small analytic open disc ∆ ⊂ B0 \ Vi containing b∗ and a small analytic connected open neighborhood Ui of zi in Z with: n/(R2 D∆×Ui → ∆ × Ui an ´etale cover of degree λ. Shrinking ∆ and Ui if necessary, D∆×Ui consists of λ disjoint connected components including (0)∆ × Ui and (0)∆ × Ui. Fixing a connected component D0 of D∆×Ui, we obtain a family of 1-sheeted covers z for z ∈ Ui is a z → ∆ with z ∈ Ui satisfying D0 D0 z → ∆ z → Y∆. The composition f◦ιz : D0 complex submanifold of Y∆ via the inclusion ιz : D0 is thus holomorphic and bijective for every z ∈ Ui. As injective holomorphic maps are z → Y∆ biholomorphic to their images ([24, Theorem 2.14, Chapter I]), the map sz : ∆ → D0 given by t (cid:55)→ D0 z is holomorphic for every z ∈ Ui. Moreover, the map z ∩ ((0) ∪ (∞)) = ∅. Hence, each D0 z(t) = f−1(t) ∩ D0 ∆ × Ui → Y∆ × Ui, (t, z) (cid:55)→ (D0 z(t), z) = (f−1(t) ∩ D0 z, z) is holomorphic. As Y∆ = F × ∆, we can write sz(t) = (u(t, z), t) where u : ∆ × Ui → C∗ is a holomorphic function. The maps Ψz : Y∆ → Y∆, z ∈ Ui, given by the fibrewise multiplication by szs−1 , i.e., zi Ψz(x, t) = (xu(t, z)u(t, zi)−1, t), (x, t) ∈ Y∆ = C∗ × ∆, form a smooth family of biholomorphisms commuting with the projection Y∆ → ∆. By passing to a subsequence with suitable restrictions of domains of definitions, we can assume that Im(gn) ⊂ X∆ and that the holomorphic maps gn : ∆Rn → X∆\ Dzn still satisfy Rn → ∞, dgn(0)ρ = 1. Since D0 z ⊂ Dz, we can consider the sequence of holomorphic maps fn := ψzn ◦ gn : ∆Rn → Y∆ \ D0 = X∆ \ (D0 zi zi ∪ (0) ∪ (∞)) into the fixed space Y∆\ D0 and the compactness of the ρ-unit tangent bundle of X, there exists c > 1 such that . By the smoothness of the family of biholomorphisms (ψz)z∈Ui zi (5.9) c−1 ≤ dfn(0)ρ ≤ c, for all n ≥ 1. 18 XUAN-KIEN PHUNG Now, consider any holomorphic map h : C → Y∆ \ D0 . The composition f ◦ h must be a ∪ (0) ∪ (∞)). constant since ∆ is hyperbolic. Thus, h factors through a fibre of X∆ \ (D0 However, as each such fibre is the complement of at least 3 points in P1 and thus is hyperbolic, h must be constant. Similarly, it is clear that each holomorphic map C → z ∪ (0)∪ (∞)) ¯∆ is constant. Green's theorem 2.5 implies that Y∆ \Dzi is hyperbolically (D0 embedded in X∆ = f−1(∆) (with respect to the metric ρX∆). But since Rn → ∞, we clearly obtain a contradiction using (5.9) and Remark 2.6. We have therefore proved the (cid:3) existence of the data in (a), (b) and (c) such that (Q) is satisfied. zi zi c∗ := min zi∈Z∗ 5.5. Proof of Theorem B. We can now return to the main result which will be very similar to the proof of Theorem A. Let (α1, . . . , αk) be a fixed system of generators of the fundamental group π1(BS, b0) with a fixed based point b0 ∈ B0. Let w0 = 1 ∈ Yb. Proof of Theorem B. Fix ε > 0. We can enlarge slightly the discs in V if necessary. Then we obtain a constants M > 0 and a set of data (Ui, Vi, ci) for each element zi ∈ Z as in Lemma 5.11. Consider the open covering Z ⊂ ∪zi∈ZUi. Since Z is compact, there exists Z∗ ⊂ Z finite such that Z ⊂ ∪zi∈Z∗Ui. As ci > 0 for every zi ∈ Z by Lemma 5.11, we have (5.10) ci > 0. For each zi ∈ Z∗, denote by Li > 0 the maximum of the constants given by Theorem 2.2 applied to Bi := B \ (V ∪ Vi) = B0 \ Vi and to each free homotopy classes α1, . . . , αk regarded as elements of π1(Bi) ⊃ π1(B0). Let L = max (5.11) zi∈Z∗ Li > 0. such that #R ∩ B0 ≤ r. As Z ⊂ ∪zi∈Z∗Ui, there exists zi ∈ Z∗ such that z ∈ Ui. Now let x ∈ Jr,ε, that is, x ∈ O∗ S(B0, ε) is (R,Dz)-integral in X for some z ∈ Z, R ⊂ B By Theorem 2.2 applied to Bi, there exists bi ∈ Bi and a system of loops γ1, . . . , γk based at b representing respectively the homotopy classes α1, . . . , αk up to a single conjugation such that γj ⊂ Bi \ R for every j = 1, . . . , k and that (5.12) The second inequality follows from #R ∩ Bi ≤ #R ∩ B0 ≤ r. Let σx : B → X be the section induced by x. For every j ∈ {1, . . . , k}, we find that σx(γj) ⊂ (Y \Dz)Bi\R ⊂ (Y \Dz)Bi by the definition of (R,D)-integral points and because x ∈ O∗ (γj) ≤ Li(#R ∩ Bi + 1) ≤ Li(r + 1). lengthdBi\R i S. It follows that: lengthρ(σx(γj)) ≤ c−1 lengthd(Y \Dz )B0\Vi = c−1 lengthd(Y \Dz )Bi ≤ c−1 lengthd(Y \Dz )Bi\R ≤ c−1∗ lengthdBi\R (γj) ≤ c−1∗ L(r + 1). i i (σx(γj)) (σx(γj)) (σx(γj)) (by Lemma 5.11) (as Bi := B0 \ Vi ⊂ B0) (as (Y \ Dz)Bi\R ⊂ (Y \ Dz)Bi) (by (5.10) and Lemma 2.7) (by (5.11) and (5.12)) LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 19 Let δ be the ρ-diameter of X. Then the homotopy section ix associated to x of the short exact sequence (cf. the sequence (5.6) in Theorem 5.10) 0 → π1(Yb0, w0) → π1(YB0, w0) → π1(B0, b0) → 0 sends the basis (αj)1≤j≤k of π1(B0, b0) to the classes in π1(YB0, w0) which admit represen- tative loops of ρ-lengths bounded by H(r) := c−1∗ L(r + 1) + 2δ. As of Theorem A, the rest of the proof follows tautologically the same lines of the proof of [22, Theorem A]. We can thus obtain a finite number m > 0 such that: for every r ∈ N. #(Jr,ε mod C∗) ≤ m(r + 1)2 rank π1(B0), The only needed modification is the following. We can clearly assume that 0 < ε < 1. Consider the compact bordered manifold Eε := {z ∈ C : ε ≤ z ≤ ε−1}. Remark that π1(Eε, w0) = Z (recall that w0 = 1 ∈ Yb0 = C∗). S(B0, ε), it actually induces a homotopy section ix,ε of the short exact Since x ∈ O∗ 0 → π1(Eε, w0) → π1(Eε × B0, w0) → π1(B0, b0) → 0. sequence Since π1(Eε × B0, w0) = π1(Eε, w0) × π1(B0, b0), the homotopy section ix,ε is determined by the π1(Eε, w0)-component of ix,ε(αj) for every j = 1, . . . , k. But we have shown above that the induced ρ-length of certain representative loops of these components are bounded by H(r) := c−1∗ L(r + 1) + 2δ. The representative loops are in fact pr1(σx(γj)) where pr1 : Eε × B0 → Eε is the first projection. As Eε is compact and π1(Eε, w0) = Z, we can thus conclude as in the proof of [22, (cid:3) Theorem A] by applying [22, Lemma 13.10]. The proof is thus completed. 6. Appendix: Proof of Theorem 5.10 Recall that Y = C∗ × B0. We regard f : Y → B0 as a constant sheaf of (multiplicative) abelian groups over B0. There is a canonical short exact sequence of sheaves over B0 induced by the exponential map: (6.1) 0 → (R1f∗Z)∨ → TY → Y → 0. Consider the n-th power B0-morphism ∧n : Y → Y . The induced differential map d(∧n) : TY → TY is an isomorphism since we are in characteristic 0 so that n is invertible. The group of global algebraic sections Y (B0) is naturally identified with a subgroup of YK(K) = K∗. Clearly, Y [n] := Ker(∧n) is the constant sheaf µn on B0. The map ∧n and the sequence (6.1) induce a commutative digram: 20 XUAN-KIEN PHUNG (6.2) 0 0 0 Y [n] ∧n Y Y 0 0 TY d(∧n) TY 0 0 0 0 (R1f∗Z)∨ n (R1f∗Z)∨ Q We have a natural isomorphism Y [n] (cid:39) Q by the snake lemma. The cohomology long 0 exact sequences induced by Diagram (6.2) give a natural commutative diagram: Y (B0) n Y (B0) H 1(B0, (R1f∗Z)∨) n H 1(B0, (R1f∗Z)∨) H 1(B0, TY ) (cid:39) H 1(B0, TY ) (6.3) H 1(B0, Y [n]) (cid:39) H 1(B0, Q) 0 We decompose the map Y (B0) → H 1(B0, Y [n]) in the first column as: Y (B0) → Y (B0)/Y (B0)n (cid:44)→ H 1(B0, Y [n]). Hence, we obtain a natural commutative diagram from (6.3): (6.4) O∗ S/(O∗ S)n O∗ S Y (B0)/Y (B0)n H 1(B0, µn) (cid:39) Y (B0) H 1(B0, (R1f∗Z)∨) H 1(B0, Q). Since Y and B0 are K(π, 1)-spaces, the cohomology of local systems coincide with the group cohomology. We thus have canonical isomorphisms: H 1(B0, µn) (cid:39) H 1(G, µn), H 1(B0, (R1f∗Z)∨) (cid:39) H 1(G, Γ), LARGE UNIONS OF GENERALIZED INTEGRAL POINTS ON ELLIPTIC SURFACES 21 (cid:39) Z. The actions of the group G on where G = π1(B0, b0) and Γ = H1(Yb0, Z) (cid:39) (R1f∗Z)∨ there is a natural isomorphism H 1((cid:98)G, Γ) (cid:39) H 1(G, Γ) induced by the injection G → (cid:98)G. Γ and on µn are trivial (the monodromy is trivial in a trivial fibration). By [26, I.2.6.b], b0 Hence, (6.4) implies now easily the commutative diagram in Theorem 5.10. The detailed descriptions of the involved homomorphisms can be found in [21]. The last statement of Theorem 5.10 follows directly from Proposition 5.9 and the fact S → H 1(G, Z) is a homomorphism of groups defined by α(x) = ix − i1 that the map α : O∗ (cf. (5.7)). References [1] R. Brody, Compact Manifolds and Hyperbolicity, Transactions of the American Mathematical Society Vol. 235 (Jan., 1978), pp. 213 -- 219. [2] A. Buium, The abc Theorem for Abelian Varieties, International Mathematics Research Notices, No. 5 (1994), pp. 219 -- 233. [3] L. Chow, Abelian varieties over function fields, Trans. AMS 78 (1955), pp. 253 -- 275. [4] L. Chow, On abelian varieties over function fields, Proc. Natl. Acad. Sci. USA 41 (1955), pp. 582 -- 586. [5] C. Ehresmann, Les connexions infinit´esimales dans un espace fibr´e diff´erentiable, Colloque de topolo- gie (espaces fibr´es), Bruxelles, 1950, Georges Thone, Li`ege; Masson et Cie., Paris, pp. 29 -- 55, MR 0042768 [6] J.H. Evertse, On equations in two S-units over function fields of characteristic 0, Acta Arithmetica, vol. 47, iss. 3 (1986), pp. 233 -- 253. [7] J.H. Evertse, K. Gyory, Stewart and Tijdeman, On the numbers of solutions of weighted unit equations, Compositio Mathematica, tome 66, no. 3 (1988), pp. 329 -- 354 [8] J.H. Evertse, K. Gyory, C.L. Stewart and R. Tijdeman, On S-unit equations in two un- knowns, Inventiones mathematicae, vol. 92 (1988), pp. 461 -- 477. [9] W. Fulton, Intersection Theory, Second Edition, Springer-Verlag New York (1998), DOI: 10.1007/978-1-4612-1700-8. [10] U. Gortz and T. Wedhorn, Algebraic geometry I, Advanced Lectures in Mathematics, Vieweg + Teubner, Wiesbaden, 2010. Schemes with examples and exercises. [11] M.L. Green, Holomorphic maps to complex tori, American Journal of Mathematics 100, no. 3 (1978), pp. 615 -- 20. doi:10.2307/2373842. [12] R.C Gunning and H. Rossi, Analytic functions of several complex variables, Prentice-Hall, Inc., Englewood Cliffs, N.J. (1965). [13] M. Hindry and J.H. Silverman, The canonical height and integral points on elliptic curves, Inven- tiones mathematicae, vol. 93 (1988), pp. 419 -- 450. [14] S. Kobayashi, Invariant distances on complex manifolds and holomorphic mappings, J. Math. Soc. Japan, Volume 19, Number 4 (1967), 460 -- 480. [15] S. Kobayashi, Hyperbolic complex spaces, vol. 318 of Grundlehren der Mathematischen Wis- senschaften [Fundamental Principles of Mathematical Sciences], Springer-Verlag, Berlin, 1998. doi: 10.1007/978-3-662-03582-5. [16] S. Lang, Integral points on curves, Publ. Math. Inst. Hautes Etudes Sci., vol. 6 (1960), pp. 27 -- 43. [17] Q. Liu, Algebraic geometry and arithmetic curves, vol. 6 of Oxford Graduate Texts in Mathematics, Oxford University Press, Oxford, 2002. Translated from the French by Reinie Ern´e, Oxford Science Publications. [18] H. Matsumura, Commutative Ring Theory, Cambridge University Press (1986). [19] J. Noguchi and J. Winkelmann, Bounds for curves in abelian varieties, Journal fur die reine und angewandte Mathematik vol. 572 (2004), pp. 27 -- 47. 22 XUAN-KIEN PHUNG [20] A.N. Parshin, Finiteness Theorems and Hyperbolic Manifolds, The Grothendieck Festschrift: A Collection of Articles Written in Honor of the 60th Birthday of Alexander Grothendieck, Birkhauser Boston, Boston, MA, (1990), pp. 163 -- 78. doi: 10.1007/978-0-8176-4576-2-6. [21] X.K. Phung, Points entiers g´en´eralis´es sur les vari´et´es ab´eliennes. PhD thesis, in preparation, Uni- versity of Strasbourg. [22] X.K. Phung, Generalized integral points on abelian varieties and the Geometric Lang-Vojta conjec- ture. Preprint 2019. [23] X.K. Phung, On Parshin-Arakelov theorem and uniformity of S-integral sections on elliptic surfaces. Preprint 2019. [24] R.M. Range, Holomorphic Functions and Integral Representations in Several Complex Variables, Graduate texts in mathematics, vol. 108 (1986), Springer-Verlag, New York. [25] M. Schutt and T. Shioda, Elliptic surfaces, Advanced Studies in Pure Mathematics, vol. 60 (2010), Algebraic Geometry in East Asia - Seoul 2008, pp. 51-160. [26] J.P. Serre, Cohomologie galoisienne, Cinqui`eme ´edition, Springer-Verlag Berlin Heidelbergs (1994). [27] Stack Projects Authors, Stacks Project, stacks.math.columbia.edu. Universit´e de Strasbourg, CNRS, IRMA UMR 7501, F-67000 Strasbourg, France E-mail address: [email protected]
1407.5906	3	1407	2015-09-15T15:34:01	A period map for global derived stacks	[ "math.AG", "math.CT" ]	We develop the theory of Griffiths period map, which relates the classification of smooth projective varieties to the associated Hodge structures, in the framework of Derived Algebraic Geometry. We complete the description of the local period map as a morphism of derived deformation functors, following the path marked by Fiorenza, Manetti and Martinengo. In the end we show how to lift the local period map to a (non-geometric) morphism of derived stacks, in order to construct a global version of that.	math.AG	math	A Period Map for Global Derived Stacks Carmelo Di Natale Abstract We develop the theory of Griffiths period map, which relates the classification of smooth projective varieties to the associated Hodge structures, in the framework of Derived Algebraic Geometry. We complete the description of the local period map as a morphism of derived deformation functors, following the path marked by Fiorenza, Manetti and Martinengo. In the end we show how to lift the local period map to a (non-geometric) morphism of derived stacks, in order to construct a global version of that. Contents 1 The Period Map as a Holomorphic Function 1.1 Griffiths Period Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The Differential of the Period Mapping . . . . . . . . . . . . . . . . . . . . . . . . 2 The Period Map as a Morphism of Deformation Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Deformations of k-Schemes 2.2 Mapping Cones and Deformations of Filtered Complexes . . . . . . . . . . . . . . 2.3 Cartan Homotopies and Period Maps . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Flag Functors and the Fiorenza-Manetti Period Map . . . . . . . . . . . . . . . . 3 The Period Map as a Morphism of ∞-Groupoids 3.1 Quick Review of Derived Deformation Theory . . . . . . . . . . . . . . . . . . . . 3.2 The Algebraic Fiorenza-Manetti-Martinengo Period Map . . . . . . . . . . . . . . 3.3 Affine DG≥0-Categories and the Dold-Kan Correspondence . . . . . . . . . . . . 3.4 Derived Deformations of Filtered Complexes . . . . . . . . . . . . . . . . . . . . . 3.5 Derived Deformations of k-Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 The Geometric Fiorenza-Manetti-Martinengo Period Map . . . . . . . . . . . . . 4 The Period Map as a Morphism of Derived Stacks 4.1 A Non-Geometric Global Period Map . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Towards a Derived Analytic Period Mapping . . . . . . . . . . . . . . . . . . . . 5 6 7 8 8 10 11 14 16 17 21 23 31 37 42 46 46 47 1 Introduction Let X be a smooth complex projective variety of dimension d and consider a family of deformations X → S of X over some contractible base S; in 1968 Griffiths observed that any such Kuranishi family induces canonically a variation of Hodge structures on X. More formally let 0 = F k+1H k (X) (cid:44)→ F kH k (X) (cid:44)→ ··· (cid:44)→ F 1H k (X) (cid:44)→ F 0H k (X) = H k (X) be the Hodge filtration on cohomology and set bp,k := dim F pH k (X); define Grass (H∗ (X)) := Grass(cid:0)bp,k, H k (X)(cid:1) (cid:89) k which is a complex projective variety as so are the Grassmannians involved. Griffiths constructed the morphism Pp : S −−→ Grass (H∗ (X)) F pH k (Xt) (cid:55)−→(cid:89) t (0.1) k where Xt is the fibre of the family X → S over t ∈ S; map (0.1) is said to be the pth local period map associated to X → S. In [15] Griffiths proved that such a map is well-defined and holomorphic; he also computed its differential and showed that it is the same as the contraction map on the space H 1 (X, TX ) of first-order deformations of X. Moreover it is possible to use map (0.1) to derive some constraints on the obstructions of X. It is important to notice that, despite connecting two algebraically defined objects, map (0.1) is not algebraic: as a matter of fact one can show that its image lies in the so-called period domain (see [49] Section 10.1.3), which is a subspace of the Grassmannian determined in general by transcendental equations. The existence and holomorphicity of the local period map says that for any given Kuranishi family of a projective manifold X there is a canonical way to construct a variation of its Hodge structures; moreover such a correspondence seems to be compatible with the general deformation theory of the variety X: prompted by this observation, in 2006 Fiorenza and Manetti described Griffiths period map in terms of deformation functors. Let Def X : ArtC −−−−−−−−−−→ Set A (cid:55)→ {deformations of X over A} isomorphism be the functor of Artin rings parametrising the deformations of the variety X and recall that such a deformation functor is isomorphic to the deformation functor associated to the Kodaira- Spencer dgla KSX := RΓ (X, TX ); in a similar way for all non-negative p define the functor of Artin rings GrassF pH∗(X),H∗(X) : ArtC −−−−−−−−−−−−−−−−−−→ Set (cid:55)→ {A-deformations of F pH∗ (X) inside H∗ (X)} A isomorphism This work was supported by the Engineering and Physical Sciences Research Council [grant number EP/I004130/1]. 2 which describes the deformations of the complex F pH∗ (X) as a subcomplex of H∗ (X): this functor precisely encodes variations of Hodge structures on X. In [8], [9] and [10] Fiorenza and Manetti proved the following facts: • GrassF pH∗(X),H∗(X) is a deformation functor (in the sense of Schlessinger) and GrassF pH∗(X),H∗(X) (cid:39) Def Cχ where Cχ is the L∞-algebra defined as the cone of the inclusion of dgla's χ : EndF p (H∗ (X)) (cid:44)→ End ∗ (H∗ (X)) (H∗ (X)) :=(cid:8)f ∈ End ∗ (H∗ (X))(cid:12)(cid:12)f (F pH∗ (X)) ⊆ F pH∗ (X)(cid:9) ; with EndF p • the map FMp : KSX −−−−→ Cχ (cid:55)−→ (lξ, iξ) ξ where i is the contraction of differential forms with vector fields and l stands for the holomorphic Lie derivative, is a L∞-morphism, thus it induces a morphism of deformation functors FMp : Def KSX −→ Def Cχ; 1 • the natural transformation ∀A ∈ ArtC (cid:16)OA Pp : Def X ξ→ OX (cid:17) −−→ GrassF pH∗(X),H∗(X) (cid:55)−→ F pH∗ (X, OA) (0.2) is a morphism of deformation functors extending Griffiths period map2 and the two mor- phisms FMp and Pp are canonically isomorphic. Fiorenza and Manetti's work shows that the pth local period map is actually a morphism of de- formation theories, thus it commutes with all deformation-theoretic constructions: in particular all results of Griffiths about the differential of map (0.1) are easily recovered as purely formal corollaries of the preceding statements. Moreover Fiorenza and Manetti's construction works for any proper smooth scheme of dimension d over a field k of characteristic 0. As we are able to interpret Griffiths period map as a natural transformation of deformation functors, the next step would be to look at it in the context of Derived Deformation Theory: more formally one could ask whether there exist derived enhancements of the functors Def X and GrassF pH∗(X),H∗(X) for which it is possible to find some natural derived extension of morphism (0.2). In 2012 Fiorenza and Martinengo approached this problem, tackling it from an entirely algebraic viewpoint. As a matter of fact they observed that the contraction of differential forms 1Here, by a slight abuse of notation, the symbol FMp is denoting both the L∞-map and the induced morphism of deformation functors. 2This is why, by a slight abuse of notation, we are using the same symbol for both. 3 (cid:18) (cid:19) with vector fields i (seen in the most general way, i.e. as a morphism of complexes of sheaves over X) and the Lie derivative l give rise to a morphism of differential graded Lie algebras FMM : RΓ (X, TX ) (l,ei) −−−−−→ holim←− ≥0 (RΓ (X, Ω∗ End X )) incl.−−−→−−−→ 0 ∗ End (RΓ (X, Ω∗ X )) (0.3) ≥0 (RΓ (X, Ω∗ X )) is the dgla made of non-negatively graded maps of the complex where End RΓ (X, Ω∗ X ) in itself, which can be interpreted as the dgla of all filtration-preserving maps. Notice also that the codomain of map (0.3) is nothing but the homotopy fibre over 0 of the inclusion of X )). In [11] Fiorenza and Martinengo showed that map End (0.3) induces a morphism of derived deformation functors X )) into End (RΓ (X, Ω∗ ≥0 (RΓ (X, Ω∗ FMM : RDefRΓ(X,TX ) (l,ei) −−−−−→ RDef (cid:32) holim←− End≥0(RΓ(X,Ω∗ X)) (cid:33) (0.4) incl.−−−→−−−→ 0 End∗(RΓ(X,Ω∗ X)) whose 0-truncation FM is very close to FMp (actually FM is even more interesting than FMp as it does not depend on the degree of the filtration, so it can be interpreted as a universal version of Griffiths period map). The ultimate goal of this paper is to lift Fiorenza, Manetti and Martinengo's work to a global level, i.e. to find a morphism of derived stacks whose restriction to formal neighbourhoods gives back map (0.4). A first crucial step in order to do so consists of finding a more geometric description of such a map, thus we will define the morphism of derived deformation functors (cid:104)OA,∗ ϕ−→OX ∀A∈dgArt ≤0 k  (cid:105)≈ P: RDef X X X 2h Spec (k) / RSpec (A)  (cid:55)→ −→ hoFlag(RΓ(X,Ω∗ X ),F •RΓ(X,Ω∗ X )) [((RΓ(π0X,Ω∗ X/A),F •), ϕ)] (0.5) X ) , F •RΓ (X, Ω∗ and prove that it is naturally isomorphic to map (0.4); in formula (0.5) RDef X parametrises derived deformations of the scheme X (i.e. homotopy flat families of derived schemes deforming the underived scheme X), while hoFlag (RΓ (X, Ω∗ X )) encodes derived defor- X ) , F •). Although it is intuitively quite clear what mations of the filtered complex (RΓ (X, Ω∗ such functors should be, giving a careful definition of them reveals to be non-trivial at all and has actually lead us to develop the notions of affine differential graded category and affine simplicial category, which are probably interesting objects themselves to study. RDef X and hoFlag (RΓ (X, Ω∗ X )) are by construction formal neighbourhoods of interesting derived stacks: the former is the formal neighbourhood at X of the derived (non-geometric) stack DSchd/k of derived schemes of dimension d, which has been recently studied by Pridham in [40], while the latter is the formal neighbourhood at of the homotopy flag variety DFlagk The above tools provide us with a path towards a sensible global version of map (0.5), i.e. a geometric morphism of derived geometric stacks inducing the latter on formal neighbourhoods. A very partial answer to such a question is discussed in the end of this paper, where we consider X ) , F •RΓ (X, Ω∗ , which is studied in [6]. , F •(cid:17) (cid:16)RΓ (cid:16)RΓ (cid:17)(cid:17) X, Ω∗ X, Ω∗ X/k (cid:16) (cid:17) (cid:16) X/k 4 / / / the non-geometric morphism of derived stacks ∀A∈dgAlg ≤0 k (cid:16)(cid:16) (cid:16) X/k)} X/k)⊗A] RP: DSchd/k×hRPerfk{RΓ(X,Ω∗ ∼→RΓ(X,Ω∗ [Y,θ:RΓ(π0Y,Ω∗ (cid:17) (cid:17)(cid:17)(cid:17) (cid:16) Y /A) (cid:16)RΓ , F •(cid:17) and observe that it actually induces map (0.5) on formal neighbourhoods at and RP map (0.6) cannot be geometric for at least two important reasons: X, RΓ X, Ω∗ X, Ω∗ X/k X/k = −→ DFlagk(RΓ(X,Ω∗ (cid:55)−→ (RΓ(π0Y,Ω∗ X/k)) Y /A),F •) X, RΓ (cid:16) (cid:16) (0.6) (cid:17)(cid:17) X, Ω∗ X/k . However it is important to notice that 1. a global geometric morphism lifting Griffiths period map to the derived world cannot be algebraic as the classical underived map is already non-algebraic in general; 2. if we move the problem to the context of the newly-born Derived Analytic Geometry, we will have to take care of monodromy issues: in fact, given a family p : X → S of derived schemes (or even derived analytic spaces) globally deforming X, the presence of a (possibly higher) monodromy group acting non-trivially on the fibres of Rp∗Ω∗ X/S may represent a serious obstruction to make map (0.6) into a derived analytic morphism of derived analytic stacks. The problem of constructing a derived geometric global period map in the context of Derived Analytic Geometry seems to be very interesting: it is now being jointly studied by Holstein and the author and will appear in [7]. Acknowledgements -- The author does wish to thank his PhD supervisor Jonathan P. Prid- ham for suggesting the problem and for his constant support and advise along all the preparation of this paper. The author is also deeply indebted to Domenico Fiorenza, Ian Grojnowski, Ju- lian V. S. Holstein, Donatella Iacono, Dominic Joyce, Marco Manetti and Elena Martinengo for several inspiring discussions about the period mapping in Deformation Theory. 1 The Period Map as a Holomorphic Function Let X be a compact connected complex Kähler manifold of dimension d and consider a family of deformations ϕ : X → S, i.e. a proper holomorphic submersion of complex manifolds (where the base S is contractible) admitting a distinguished fibre ϕ−1 (0) =: X0 (cid:39) X. Recall that a famous result due to Ehresmann says that any such family is C∞-trivial, i.e. there exists a diffeomorphism T : X ∼−→ X0 × S (cid:39) X × S (1.1) over S (see [49] Theorem 9.3). For all t ∈ S let Xt := ϕ−1 (t): Ehresmann's trivialisation (1.1) clearly induces a diffeomorphism Xt (cid:39) X for all t, thus we can think of the morphism ϕ as a collection of complex structures over the C∞-manifold underlying the complex variety X. This situation is the prototypical example of all deformation problems and was originally studied by Kodaira and Spencer. A very natural question to ask is how the standard Hodge structures over X vary with respect to the family ϕ; more formally, consider the cohomology algebra of X H∗ (X, C) := H k (X, C) (cid:77) 0≤k≤d 5 and recall that each cohomology group H k (X, C) is endowed with a Hodge structure of weight k defined by the Hodge decomposition H k (X, C) = H p.q (X) H p,q (X) (cid:39) H q (X, Ωp X ) (cid:77) or, equivalently, by the Hodge filtration p+q=k 0 = F k+1H k (X, C) (cid:44)→ F kH k (X, C) (cid:44)→ ··· (cid:44)→ F 1H k (X, C) (cid:44)→ F 0H k (X, C) = H k (X, C) where F mH k (X, C) := H p,q (X) . (cid:77) p≥m The question we want to address is whether the family ϕ induces any interesting structure on the cohomology of the fibres. 1.1 Griffiths Period Map Observe that Ehresmann's trivialisation (1.1) provides us with a diagram of isomorphisms of vector spaces H k (X, C) ∼ / / H k (X, C) (cid:111) H k (X, C) ∼ / / H k (Xt, C) which commutes for all k ≥ 0 and ∀t ∈ S; actually much more is true, as for all k ≥ 0 the sheaf Rkϕ∗C -- where ϕ∗ : Sh (X) → Sh (S) is the push-forward functor -- is seen to be a local system over S isomorphic to the constant sheaf H k (X, C) (see [49] Section 9.2 for a more detailed explanation), thus the diagram above does not depend on the choice of the trivialisation. Denote hk := dim H k (X, C) hp,q := dim H p,q (X) bp,k := dim F pH k (X, C) . A standard argument based on the E1-degeneration of the Hodge-to-De Rham spectral sequence of X shows that there exists a neighbourhood of 0 ∈ S such that dim H p,q (Xt) = dim H p,q (X) =: hp,q thus the Hodge numbers of X are invariant under (infinitesimal) deformation; moreover this immediately implies that the Hodge-to-De Rham spectral sequence of such fibres degenerates at its first page3, as well (see [49] Proposition 9.20). Definition 1.1. (Griffiths) In the above notations define the (p, k)th local period map to be Pp,k : S −−→ Grass(cid:0)bp,k, H k (X, C)(cid:1) (cid:55)−→ F pH k (Xt, C) . t (1.2) 3Up to shrinking the base S, the fibres of ϕ are Kähler manifolds themselves (see [49] Theorem 9.23). 6 Since H k (Xt, C) is canonically isomorphic to H k (X, C) and the Hodge numbers of X are invariant under deformation, map (1.2) is well-defined; the following is a famous result of Grif- fiths. Theorem 1.2. (Griffiths) The (p, k)th local period map (1.2) is holomorphic ∀p ≤ k. Proof. See [49] Theorem 10.9. 1.2 The Differential of the Period Mapping Griffiths deeply studied the differential of map (1.2), as well: in order to state his result let us review what the contraction of differential forms with vector fields and the (holomorphic) Lie derivative are. Recall that the tangent sheaf TX is endowed with a natural structure of sheaf of Lie algebras (which can be considered as dgla's concentrated in degree 0) induced by the canonical isomorphism TX (cid:39) Der (OX , OX ), while End∗ (Ω∗ X ) comes with a structure of sheaf of differential graded Lie algebras through the standard differential on Hom complexes and the standard Lie bracket. Now the contraction morphism is defined to be the "shifted" map of sheaves of differential graded Lie algebras X ) [−1] i : TX −→ End∗ (Ω∗ ξ (cid:55)−→ iξ such that iξ (ω) := ξ(cid:121) ω (on local sections) while the differential of map (1.3) (as an element of the complex Hom∗ (TX , End∗ (Ω∗ is by definition the Lie derivative l : TX −→ End∗ (Ω∗ X ) ξ (cid:55)−→ lξ such that lξ (ω) := d (ξ(cid:121) ω) + ξ(cid:121) (dω) (on local sections) (1.4) which is a genuine morphism of sheaves of dgla's4. Theorem 1.3. (Griffiths) The differential dPp,k of map (1.2) factors through the (cohomology) contraction map (1.3) X ) [−1])) i : H 1 (X, TX ) −→ Hom F pH k (X, C) , (cid:18) (cid:16) (cid:19) . H k (X, C) F pH k (X, C) (1.5) (cid:17) .5 Moreover map (1.5) actually takes values in Hom F pH k (X, C) , F p−1H k(X,C) F pH k(X,C) Proof. The theorem has been stated in relatively modern terms, but a complete proof of it is given in [49] Proposition 10.12, Lemma 10.19 and Theorem 10.21. The (p, k)th local period map (1.2) depends by definition on two parameters, a cohomology one -- that is k -- and a filtration one -- that is p; we would like to encode all cohomologi- cal information about the variations of Hodge structures induced by the family φ in a single morphism. 4By a slight abuse of notation we will tend to denote by i and l the morphisms that maps (1.3) and (1.4) induce on global sections and derived global sections, as well. 5This last property is generally known as Griffiths transversality. 7 Definition 1.4. In the above notations define the pth local period map to be Pp : S −−→ Grass (H∗ (X, C)) F pH k (Xt, C) . t (1.6) Notice that map (1.6) is holomorphic and that its differential dPp still factors through a (cid:55)−→(cid:89) (cid:18) k contraction morphism i : H 1 (X, TX ) −→(cid:77) k Hom F pH k (X, C) , . (1.7) (cid:19) H k (X, C) F pH k (X, C) 2 The Period Map as a Morphism of Deformation Functors The work of Griffiths which has been described in Section 1 relates deformations of a com- plex smooth projective variety (or more generally complex Kähler manifold) to variations of its Hodge structures. Unfortunately the local period map (1.6) is not really a morphism of defor- mation theories, as it depends on a given deformation of a complex variety X, nonetheless its differential (1.7) is very "deformation-theoretic" in nature, as it relates the space H 1 (X, TX ), i.e. the tangent space to the deformation functor parametrising all deformations of X, to an- other cohomological invariant which depends only on X rather than the special Kuranishi family over X determining map (1.6). Observations like these led Fiorenza and Manetti to believe that Griffiths period map could be described as a morphism of deformation functors (in the sense of Schlessinger) whose induced tangent mapping coincided with map (1.7). 2.1 Deformations of k-Schemes Let k be any (non-necessarily algebraically closed) field of characteristic 0 and consider a smooth proper scheme X of dimension d: these assumptions over X just algebraically resemble the analytic framework in which Griffiths studied map (1.6), while the fact that the theory we are about to summarise works for any field of characteristic 0 is a consequence of Deligne's views on Hodge Theory (for more details see [2], [3] and [4]). Notice also that by [2] Theorem 5.5 the Hodge-to-De Rham spectral sequence of the scheme X degenerates at its first page: such a property will be used several times in this paper. Recall that the functor of deformations of X is the functor of Artin rings Def X : Artk −−−−−−−−−−→ Set A (cid:55)→ {deformations of X over A} isomorphism (2.1) where a deformation6 of X over A is a Cartesian diagram in Schk X i 2 X p Spec (k) / Spec (A) 6From now on by deformation we will always mean infinitesimal deformation, i.e. a deformation over an Artinian base. 8 / / / with i a closed immersion and p flat and proper; equivalently a deformation of X over A can be viewed as a morphism of sheaves of A-algebras OA → OX such that OA is flat over A and OA⊗A k (cid:39) OX. Of course, two A-deformations X1 → Spec (A) and X2 → Spec (A) of X are said to be isomorphic if there is an isomorphism X1 −→X2 of schemes over A inducing the identity on X: it is well-known that functor (2.1) is a deformation functor in the sense of Schlessinger (see [31] or [46] for a definition). Now let (l,(cid:104)···(cid:105)n)n>0 be a L∞-algebra over k (see [31] for a definition) and recall that the deformation functor associated to (l,(cid:104)···(cid:105)n)n>0 is defined to be Def l : Artk −−−−−−−−→ Set where MCl (A) := homotopy equivalence MCl (A) (cid:55)→ A x ∈ l0 [1] ⊗ mA s.t. (cid:88) n≥1 (cid:104)x(cid:12)n(cid:105)n n!  = 0 is the set of solutions of the Maurer-Cartan equation and two elements x0, x1 ∈ MCl (A) are said to be homotopy equivalent if there exists a "path" x (t, dt) ∈ MCl[t,dt] (A) such that x (0) = x0 and x (1) = x1; again, it is not hard to verify that Def l is a deformation functor in the sense of Schlessinger. Notice that, if the higher products (cid:104)···(cid:105)n = 0 for all n ≥ 3, i.e. if the L∞-algebra is actually a differential graded Lie algebra (see [5] or [31] for a definition), we recover the more classical notion of deformation functor associated to a dgla. A fundamental fact in Deformation Theory -- essentially due to Kodaira, Kuranishi and Spencer and developed in many ways by several other authors -- states that the functor of deforma- tions Def X associated to a scheme X which satisfies the above conditions is isomorphic to the deformation functor associated to the Kodaira-Spencer dgla of X, which is defined to be the differential graded Lie algebra (KSX , [−,−] , D) where (cid:17) (cid:16) (cid:16) (cid:104) f d¯zI KSX := RΓ (X, TX ) (cid:39) Γ := d¯zI ∧ d¯zJ , gd¯zJ ∂ ∂zj ∂ ∂zi X, A 0,∗ X (TX ) − g ∂f ∂ ∂zj ∂zj ∂ ∂zi (cid:17) (cid:17) f ∂g ∂zi := − ¯∂ (ω) ∂ ∂zi D ω ∂ ∂zi (cid:105) (cid:16) . (2.2) (cid:16) (cid:17) X, A 0,∗ Warning 2.1. In this paper the algebra KSX will always correspond to the specific resolution computing RΓ (X, TX ), equipped with the Lie structure (2.2). This becomes Γ very relevant in comparisons with the work of Fiorenza, Manetti and Martinengo. X (TX ) Now consider the natural transformation ∀A ∈ Artk Def KSX (A) (cid:51) ξ (cid:55)−→ (Oξ → OX ) ∈ Def X (A) . (2.3) where for all open U ⊆ X Oξ (U ) := and the map Oξ → OX is induced by the projection A 0,0 O : Def KSX −−−−−→ Def X (cid:110) (cid:111) X (U ) ⊗ A s.t. ¯∂f = ξ(cid:121)∂f X ⊗ A → A 0,0 X . f ∈ A 0,0 9 Theorem 2.2. (Kodaira-Spencer, Kuranishi, [...]) isomorphism of deformation functors. In the above notations, map (2.3) is an Proof. There is a variety of different proofs of this result in the literature: we refer to [19] Theorem II.7.3 for a very detailed algebraic one; see also [20] Theorem 3.4. 2.2 Mapping Cones and Deformations of Filtered Complexes The functor Def X is the most natural candidate for the domain of a purely "deformation- theoretic" version of Griffiths period map; now we wish to understand what the codomain of such a morphism should be, i.e. we seek a deformation functor which parametrises variations of Hodge structures over X. Let (V, d) be a differential graded k-vector space and (W, d) a subcomplex of its; for any A ∈ Artk, consider the groups7 A (V ⊗ A, V ⊗ A) s.t. f ≡ Id(V,d) (mod mA)(cid:9) := (cid:8)f ∈ Hom0 := (cid:8)f ∈ AutV (A) s.t. f d = df(cid:9) := (cid:8)f ∈ AutV (A) s.t. f (W ⊗ A) = W ⊗ A(cid:9) (cid:110) f ∈ AutV,d (A) s.t. H∗ (f ) is the identity on H∗ (V ⊗ A, d) := (cid:111) AutV (A) Aut(V,d) (A) AutW,V (A) (V,d) Aut (A) and define the functor of deformations of (W, d) inside (V, d) to be the functor of Artin rings GrassW,V : Artk −−−−−−−−−−−−−−−→ Set (cid:8)f ∈ AutV (A) s.t. df (W ⊗ A) ⊆ f (W ⊗ A)(cid:9) A (cid:55)→ (V,d) Aut (A) × AutW,V (A) . (2.4) Remark 2.3. Formula (2.4) is the original definition of the functor of deformations of the sub- complex (W, d) as we find it in [8]; although it is quite elegant, it may not seem very intuitive, as there is no explicit reference to what a deformation of (W, d) over a local Artinian k-algebra A should be. Anyway a more careful look at it immediately shows that a deformation of (W, d) over A inside (V, d) is a complex of free A-modules (V ⊗ A, dA) such that its residue modulo mA equals (V, d) and dA (W ⊗ A) ⊆ W ⊗ A (this is exactly what the "numerator" in formula (2.4) parametrises); on the other hand two such deformations (V ⊗ A, dA) and (V ⊗ A, d(cid:48) A) are isomorphic if there exists an isomorphism of cochain complexes ϕ between them such that ϕ (W ⊗ A, dA) = ϕ (W ⊗ A, d(cid:48) A) and H i (ϕ) = IdH i(V ⊗A,d) for all i (this is exactly what the "denominator" in formula (2.4) parametrises). Now consider the graded vector spaces ∗ End ((V, d)) EndW ((V, d)) := Hom∗ ((V, d) , (V, d)) := {f ∈ End ∗ ((V, d)) s.t. f (W ) ⊆ W} . They are endowed with natural structures of differential graded Lie algebras and there is an obvious inclusion χW,V : EndW ((V, d)) (cid:44)→ End ∗ ((V, d)) 7In this section, by a slight abuse of notation, the symbol d may indifferently denote the differential of the complex V , the differential of the twisted complex V ⊗ A and the differential of the endomorphism complex End ((V, d)). 10 which is a morphism of dgla's; recall also that the mapping cone (cid:0)CχW,V , δ(cid:1) of the morphism χW,V is defined to be its homotopy cokernel, i.e. the complex (cid:19) (cid:18) holim←− EndW ((V, d)) χW,V−−−−→−−−−→ 0 ∗ End ((V, d)) . More concretely, the mapping cone is given by the formulae C i χW,V δ ((f, g)) := EndW ((V, d)) := (df, χ (f ) − dg) . i ⊕ Endi−1 ((V, d)) (2.5) Proposition 2.4. (Fiorenza-Manetti) In the above notations, there is a canonical L∞-structure on the mapping cone CχW,V . Proof. See [9] Section 4 and Section 5. Remark 2.5. Fiorenza and Manetti gave two different proofs of Proposition 2.4: the first one is a very elegant but non-constructive proof based on the Homotopy Transfer Theorem (see [23] and [24], while [48] provides a gentler introduction), while the second proof relies on a careful explicit description of all the higher products defining the L∞-structure of CχW,V ; anyway, we are not reporting such formulae since they are not really needed for the sake of this paper. Consider the natural transformation ∀A ∈ Artk ΨχW,V : Def CχW,V (A) (cid:51) η Def CχW,V −−−−−→ GrassW,V (cid:55)−→ (η (W ⊗ A) , d) ∈ GrassW,V (A) . Theorem 2.6. (Fiorenza-Manetti) In the above notations, map (2.6) is an isomorphism of de- formation functors; in particular GrassW,V is a deformation functor in the sense of Schlessinger. Proof. See [8] Proposition 9.2. 2.3 Cartan Homotopies and Period Maps The work of Fiorenza and Manetti, especially Theorem 2.6, suggests that a good candidate for the codomain of a purely deformation-theoretic version of Griffiths pth local period map should be the functor GrassF pH∗(X,k),H∗(X,k), where H∗ (X, k) := H∗(cid:16) (cid:17) X, Ω∗ X/k is the algebraic De Rham cohomology of the scheme X and F • is the Hodge filtration over it. Now we are almost ready to describe the actual morphism that Fiorenza and Manetti constructed in order to translate Griffiths period map in terms of deformation functors. Definition 2.7. (Fiorenza-Manetti) Let (g, d, [−,−]) and (l, d, [−,−]) be two differential graded Lie algebras over k; a linear map i ∈ Hom−1 (g, l) is said to be a Cartan homotopy if ∀a, b ∈ g i ([a, b]) = [i (a) , di (b)] and [i (a) , i (b)] = 0.8 8Again, we are denoting by the same symbol the differential and the bracket of the dgla's g, l and Hom∗ (g, l). 11 Remark 2.8. The following facts directly follow from Definition 2.7 1. The differential of a Cartan homotopy is a morphism of differential graded Lie algebras (i.e. it preserves grading and differentials); 2. The notion of Cartan homotopy is stable under composition with a dgla map and under tensorisation with a differential graded commutative algebra; 3. The notion of Cartan homotopy generalises to maps of sheaves of dgla's; 4. Let i : g → l [−1] be a Cartan homotopy and l : g → l its differential: ei is an homotopy between l and the zero dgla morphism 0. Example 2.9. The contraction map associated to the scheme X is a Cartan homotopy of sheaves of dgla's (see Section 1.2 for a definition in the context of complex manifolds), while its derived globalisation provides us with an honest Cartan homotopy of dgla's: the latter will be a key ingredient of this paper (see Section 3.2). The reason why we are interested in Cartan homotopies is that they behave very well with respect to mapping cones. Proposition 2.10. (Fiorenza-Manetti) In the notations of Definition 2.7, let l := di; then the linear map i : g −−−−→ Cl a (cid:55)−→ (a, i (a)) is a L∞-morphism; in particular it induces a morphism between the associated deformation functors. Proof. See [8] Proposition 7.4. Now, in the notations of Proposition 2.4, for all p > 0 set X, Ω∗ W := F pRΓ V := RΓ X, Ω∗ X/k (cid:16) (cid:17) χp := χV,W X/k (cid:16) (cid:17) while denote by i the contraction map associated to X and by l its differential, i.e. the Lie derivative. Warning 2.11. As we did in Warning 2.1 in the case of the Kodaira-Spencer dgla, we will always fix a specific choice of functor for RΓ , i.e. the one given by the Dolbeaut resolution: in other words throughout the paper we will have X, Ω∗ (cid:16) X/k (cid:17) (cid:17) (cid:39) Γ(cid:0)X, A ∗,∗ X (cid:1) . (cid:16) RΓ X, Ω∗ X/k Theorem 2.12. (Fiorenza-Manetti) The linear map fmp : KSX −−−−→ Cχp (cid:55)−→ (lξ, iξ) ξ is a L∞-morphism; in particular it induces a morphism of deformation functors fmp : Def KSX −−−−→ Def Cχp .9 (2.6) (2.7) 9Here, by a slight abuse of notation, the symbol fmp is denoting both the L∞-map and the induced morphism of deformation functors. 12 Proof. See [8] Theorem 12.1. (cid:16) (cid:16) (cid:16) (cid:17) (cid:17) (cid:17)(cid:17) → H∗ (X, k) is (cid:44)→ RΓ Remark 2.13. Recall that, as a consequence of the E1-degeneration of the Hodge-to De Rham spectral sequence of X, the canonical inclusion of complexes F pRΓ F pRΓ descends to cohomology, i.e. the induced linear map H∗(cid:16) X, Ω∗ X, Ω∗ X, Ω∗ X/k X/k injective. This is equivalent to say that for all p there is a quasi-isomorphism of complexes and F pH∗ (X, k)10, which in turn induces a quasi-isomorphism of between F pRΓ dgla's between EndF pRΓ(X,Ω∗ and EndF pH∗(X,k) (H∗ (X, k)). X/k)(cid:16)RΓ (cid:17)(cid:17) X, Ω∗ X, Ω∗ (cid:16) X/k (cid:16) (cid:17) X/k X/k Now denote χp := χH∗(X,k),F pH∗(X,k): Remark 2.13 entails in particular the existence of a homotopy equivalence of L∞-algebras h : Cχp −→ C χp which induces, by the Basic Theorem of Deformation Theory (see [31]), an isomorphism h : Def Cχp −→ Def C χp 11 (2.8) between the corresponding deformation functors. In the same fashion, the natural transforma- tion (cid:16) (cid:17) (cid:16) F pRΓ X,Ω∗ X/k ,RΓ X,Ω∗ X/k (cid:17) −−−−→ GrassF pH∗(X,k),H∗(X,k). H∗ : Grass (2.9) induced by the algebraic De Rham cohomology functor is an isomorphism: for a proof see [8] Theorem 10.6. Definition 2.14. For all p > 0 define the algebraic pth Fiorenza-Manetti local period map to be the morphism FMp : Def KSX −−−−→ Def C χp given by the composition of maps (2.8) and (2.7). Definition 2.15. For all p > 0 define the geometric pth Fiorenza-Manetti local period map to be the morphism Def X (A) (cid:51)(cid:16)OA Pp : Def X (cid:17) −−→ GrassF pH∗(X),H∗(X) (cid:55)−→ ξ→ OX ∀A ∈ Artk F pH∗ (X, OA) ∈ GrassF pH∗(X),H∗(X) (A) . Now we are finally ready to lift Griffiths period map to a morphism of deformation functors. Theorem 2.16. (Fiorenza-Manetti) There is a natural isomorphism between maps FMp and Pp, meaning that the diagram Def KSX O (cid:111) Def X FMp Def C χp (cid:111) Ψ χp Pp / GrassF pH∗(X),H∗(X) commutes. Moreover the tangent morphism to the functor Pp is the same as map (1.7). Proof. See [8] Theorem 12.3 and Corollary 12.5. 10Notice that the case p = 0 is trivial. 11Again, the symbol h is denoting both the L∞-map and the induced morphism of deformation functors. 13 / / / 2.4 Flag Functors and the Fiorenza-Manetti Period Map Both FMp and Pp depend on a filtration parameter: we would like to get rid of it, in order to define universal versions of the algebraic and geometric Fiorenza-Manetti period map. Observe that the target functor of any universal version of the geometric Fiorenza-Manetti period map should not be simply the product of the deformation functors GrassF pH∗(X,k),H∗(X,k), because the only deformations of the sequence (F pH∗ (X, k))p of subcomplexes of H∗ (X, k) which may belong to its image are those preserving the property that F • is a filtration. For this reason, let (V, F•) be a filtered complex and define the flag functor associated to (V, F•) to be FlagF• V : Artk −−−−−−−−−−−−−−−−−−−−−−−→ Set ((U, Gp))p s.t. (U, Gp) ∈ GrassFpV,V (A) , GpU (cid:44)→ Gp−1U (cid:55)→(cid:110) A Consider the complex EndF• (V ) := EndFpV (V ) (cid:92) p (cid:111) . (2.10) which may be seen as the subcomplex of End (V ) made of filtration-preserving endomorphisms. Example 2.17. Notice that if our base filtered complex (V, F•) is the algebraic De Rham complex (or cohomology) of a scheme X equipped with the Hodge filtration, than the complex of filtration-preserving endomorphisms is nothing but the complex of non-negatively graded endomorphisms. In other words, in the notations of formula (2.10) we have that EndF •(cid:16)RΓ (cid:16) (cid:17)(cid:17) ≥0(cid:16)RΓ (cid:16) X, Ω∗ X/k = End X, Ω∗ X/k (cid:17)(cid:17) . Again EndF• (V ) is endowed with a natural structure of differential graded Lie algebra: this comes with a natural inclusion of dgla's χ : EndF• (V ) (cid:44)→ End (V ) . (cid:18) and let be its homotopy cokernel. C F• V := holim←− EndF• (V ) χ−→−→ 0 ∗ End (cid:19) (V ) . Proposition 2.18. (Fiorenza-Martinengo) In the above notations there is an isomorphism of functors In particular FlagF• FlagF• V is a deformation functor. V (cid:39) Def C F• V Proof. See [11] Section 5 and Section 6. Now consider the functors FlagF • RΓ (cid:16) X,Ω∗ X/k deal with map (2.9) imply that the morphism H∗ : FlagF • RΓ (cid:16) X,Ω∗ X/k (cid:17) and FlagF • (cid:17) −−−−→ FlagF • H∗(X,k) H∗(X,k): the same arguments used to (2.11) is well-defined and an isomorphism. 14 X, Ω∗ (cid:16) (cid:16) (cid:16)RΓ Remark 2.19. In the language of [11] a pair of differential graded Lie algebras (g, l) is said to be a formal pair if there is an inclusion of dgla's g (cid:44)→ l inducing an injective morphism H∗ (g) (cid:44)→ H∗ (l) on cohomology: in particular by Remark 2.13 we have that for any smooth proper k-scheme X and for all p ≥ 0 the pair is formal. Moreover The formality argument of Remark (2.13) is uniform in p, therefore there and (H∗ (X, k) , F •), providing in is a filtered quasi-isomorphism between . turn a week equivalence between the dgla's End It is easy to see that if (g, l) is a formal pair of dgla's then the homotopy fibre holim←− X/k is formal as well. incl.−−−→−−−→ l is quasi-abelian and in fact a model for it is given by the complex l/g [−1] endowed with the trivial bracket: see [11] Section 5 for a more detailed explanation. In particular, if we apply this to the De Rham-theoretic case we have that (cid:16) (cid:17)(cid:17) (cid:16)RΓ and H∗(cid:16) End≥0(cid:16) (cid:17)(cid:17) (cid:17)(cid:17)(cid:17) (cid:18) (cid:16) EndF p(cid:16)RΓ , F •(cid:17) (cid:17) ≥0(cid:16)RΓ (cid:16) (cid:16)RΓ (cid:16) (cid:17)(cid:17) (cid:17)(cid:17)(cid:17) (cid:17)(cid:17) (cid:19) ≥0(cid:16)RΓ It follows that the pair X, Ω∗ X, Ω∗ X, Ω∗ X, Ω∗ X, Ω∗ , End , End (cid:16) (cid:16) Ω∗ End X/k X/k X/k X/k X/k X/k g 0 (cid:17)(cid:17) −−→−−→ End (cid:16) (cid:16)RΓ X, Ω∗ X/k (cid:17)(cid:17)(cid:19) (cid:16) CRΓ (cid:17) (cid:16) X,F pΩ∗ X/k ,RΓ X/k X, Ω∗ (cid:17)(cid:17) (cid:17)(cid:17) [−1] (cid:17)(cid:17) (cid:17)(cid:17) [−1] (cid:17)(cid:17) −−→−−→ End X/k X/k X/k X/k X,Ω∗ (cid:16) X, Ω∗ X, Ω∗ Ω∗ Ω∗ (cid:18) EndF p(cid:16)RΓ (cid:17) (cid:39) holim←− (cid:16) ∗(cid:16)RΓ EndF p(cid:16)RΓ (cid:16) (cid:39) End End∗(cid:16) (cid:39) H∗(cid:16) H∗(cid:16) EndF p(cid:16) (cid:18) (cid:16) ≥0(cid:16)RΓ (cid:17) (cid:39) holim←− (cid:17)(cid:17) (cid:16) ∗(cid:16)RΓ (cid:17)(cid:17) [−1] (cid:16) ≥0(cid:16)RΓ (cid:39) End End∗(cid:16) (cid:39) H∗(cid:16) (cid:17)(cid:17) (cid:17)(cid:17) [−1] H∗(cid:16) (cid:16) and EndF p(cid:16) , End≥0(cid:16) (cid:17) X, Ω∗ X, Ω∗ Ω∗ Ω∗ End≥0 X, Ω∗ Ω∗ Ω∗ End End X/k X/k X/k X/k X/k X/k (cid:17) (cid:17) (cid:16) (cid:16)RΓ X, Ω∗ X/k (cid:17)(cid:17)(cid:19) (2.12) (2.13) and (cid:16) C F • RΓ X,Ω∗ X/k where End∗(cid:16) Ω∗ X/k X/k denote respectively the endomorphism X/k, the sheaf of its non-negatively graded endomorphisms and the sheaf of those sheaf of Ω∗ endomorphisms preserving the pth piece of the Hodge filtration. Summing up the preceding considerations, we obtain a very explicit description of the flag functor associated to the algebraic De Rham complex of X. X/k Corollary 2.20. There is a chain of isomorphism of deformation functors FlagF • H∗(X,k) (cid:39) FlagF • End∗(cid:16)RΓ (cid:16) X,Ω∗ X/k (cid:17)(cid:17) (cid:39) Def End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) . [−1] (2.14) Proof. The first isomorphism in chain (2.14) is map (2.11), the other one follows combining Proposition 2.18 and formula (2.13). 15 Now we are ready to define some universal version of the Fiorenza-Manetti morphism. Definition 2.21. Define the universal geometric Fiorenza-Manetti period map to be the natural transformation P : ∀A ∈ Artk (cid:16)OA Def X ξ→ OX −−−−→ FlagF • H∗(X,k) (cid:17) (cid:55)−→(cid:16) Pp(cid:16)(cid:16)OA ξ→ OX (cid:17)(cid:17)(cid:17) . p Notice that Definition 2.15 ensures that P is a well-defined morphism of functors. Map P is a good universal version of the geometric Fiorenza-Manetti period map; we would like to complete the picture with a natural universal version of the algebraic Fiorenza-Manetti map, that is we would like to construct a morphism of differential graded Lie algebras ∗(cid:16)RΓ (cid:16) (cid:16) ≥0(cid:16)RΓ X, Ω∗ X, Ω∗ X/k X/k (cid:17)(cid:17) (cid:17)(cid:17) [−1] FM : KSX −→ End End such that the diagram Def KSX (cid:111) Def X FM P Def End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) (cid:111) [−1] / FlagF • H∗(X,k) commutes: we will construct it in Section 3.2. 3 The Period Map as a Morphism of ∞-Groupoids Theorem 2.16 attests two very interesting facts: the first one is that Definition 2.14 and Definition 2.15 are naturally equivalent (and this enables us to simply talk about the Fiorenza- Manetti local period map, dropping any further adjective) and the second one is that map (2.10) really extends the period mapping (1.6) to a morphism of deformation theories, as the tangent maps are the same. In this perspective, the period map is seen to play a remarkable unifying role in Deformation Theory and Hodge Theory: as a matter of fact a number of highly non-trivial classical results such as Kodaira Principle and Bogomolov-Tian-Todorov Theorem are recovered as corollaries of Theorem 2.16 (see [8], [10], [11] and [21] for more details). Anyway the contemporary viewpoint on Deformation Theory claims that Schlessinger's defor- mation functors are not the most suitable tools in order to study general local moduli problems, as they are often unable to capture most of the hidden geometry of such problems. As a matter of fact Schlessinger's functors do not generally take into account automorphisms and higher autoequivalences of the objects they classify and in most cases they do not give a proper de- scription of obstructions, either. Moreover the correspondence between differential graded Lie algebras and deformation functors in the context of classical Deformation Theory is not fully satisfying12, but the most important drawback of Schlessinger's functors for the sake of this 12Notice that an instance of such a drawback has already appeared in Section 2.2, since the mapping cone (2.5) is endowed with a non-trivial L∞-structure. 16 / / / paper is that in general they are not formal neighbourhoods of any global moduli space; this is precisely the case of the functor Def X defined in Section 2.1: there does not exist any (classical) moduli space of proper smooth schemes of dimension d > 1, thus for a general choice of the scheme X the functor Def X cannot be describing infinitesimally any algebraic space. 3.1 Quick Review of Derived Deformation Theory The critical aspects we have briefly listed above mark some of the reasons that have been leading to the development of Derived Deformation Theory: the rough idea behind this subject is that Deformation Theory is not really a "categorical" subject, but rather an "(∞, 1)-categorical" one, meaning that its constructions and invariants should be homotopical (or derived) in nature. In particular the basic objects of Derived Deformation Theory should be homotopy analogues of Schlessinger's functors -- i.e. functors defined over (some subcategory of) dgArtk rather than Artk -- satisfying homotopical versions of Schlessinger's axioms and preserving the homotopical structure of the category of Artinian dg-algebras. Foundational work on Derived Deformation Theory includes [12], [17], [22], [26], [29], [30], [37] and [45], while a gentle introduction to the subject can be found in [5]: here we quickly review some of the main concepts just to fix notations. There are several different ways to enhance a classical deformation functor to a derived one, giving rise to various consistent derived deformation theories; in [37] Pridham proved that all these variants are homotopy equivalent13, thus in this paper by derived deformation functor we will always mean a Hinich derived deformation functor 14. The latter is a functor F : dgArt ≤0 k −→ sSet satisfying weaker versions of Schlessiger's axioms for classical deformation problems: for a precise definition see [5] or the original paper [17], but essentially F is required to be homotopic -- i.e. to map quasi-isomorphisms in dgArt to weak equivalences in sSet -- and homotopy-homogeneous -- i.e. such that for all surjections A (cid:16) B and all maps C → B in dgArt the natural map ≤0 k ≤0 k F (A ×B C) −→ F (A) ×h F(B) F (C) 15 ≤0 k is a weak equivalence. In case F is only homotopy-surjecting -- i.e. for all tiny acyclic extension A → B in dgArt the induced map π0 (F (A)) → π0 (F (B)) is surjective -- we will say that it is a derived pre-deformation functor. All the geometry of Hinich functors is captured by certain cohomological invariants which gen- eralise tangent spaces and obstruction theories for classical deformation functors: let us briefly ≤0 k −→ sSet, recall how to construct them. Given a derived deformation functor F : dgArt consider as in [37] Section 1.6 the functor tan F : dgVect V ≤0 k −−→ sVectk (cid:55)−→ F (k ⊕ V ) 13All approaches to Derived Deformation Theory are described by a well-defined (∞, 1)-category: Pridham proved that all such (∞, 1)-categories are equivalent; for more details see [37]. 14In the literature people also refer to such functors as formal moduli problems or formal stacks. 15The symbol − ×h− − denotes the homotopy fibre product in sSet. 17 and recall that the j-th generalised tangent space of F is said to be the group H j (F) := πi (tan F (k [−n])) where n − i = j and the definition is well-given because of [37] Corollary 1.46. Generalised tangent spaces extend the underived notions of tangent and obstruction spaces in the sense that if F is a derived deformation functor, the group H j (F) parametrises infinitesimal j-automorphisms associated to it; in particular H 0 (F) encodes first-order derived deformations and H 1 (F) encodes second- order derived deformations, i.e. all obstructions (see [37] Section 1.6). One of the properties of derived deformation functors which are most interesting to us is that they provide the right notion of formal stack, i.e. that they describe derived geometric stacks infinitesimally. Foundational work on higher stacks and Derived Algebraic Geometry includes [26], [27], [38], [43] and [45]: here we only recall that given a (possibly non-geometric) derived stack over k F : dgAlg ≤0 k → sSet a point x over it, the formal neighbourhood of F at x is defined as Fx : ≤0 dgArt k A −−−−−→ sSet (cid:55)−→ F (A) ×h F(k) {x} . A well-known folklore result in Derived Algebraic Geometry is that Fx is a derived deformation functor: a proof of it is hidden somewhere in [26] and [45]; see also [43] and [37]. Now denote by Ω∗ differential forms, given in simplicial level n by DR (∆∗) the simplicial differential graded commutative algebra of polynomial Ω∗ DR (∆n) := k [x0, x1, . . . , xn, dx0, dx1, . . . , dxn] (cid:80) xi = 1,(cid:80) dxi = 0 where x0, x1, . . . , xn live in cochain degree 0 and dx0, dx1, . . . , dxn in cochain degree 1; more generally, given a simplicial set S, the symbol Ω∗ DR (S) will stand for the simplicial differential graded commutative algebra of polynomial differential forms on S, which is defined in dg level p by DR (S) := HomsSet (S, Ωp Ωp DR (∆∗)) . Also recall that the Hinich nerve of a dgla g is defined to be the derived deformation functor RDef g : dgArt ≤0 k −−−−−−→ sSet A (cid:55)−−−−−→ RMCg⊗Ω∗ DR(∆∗) (A) where RMCg⊗Ω∗ DR(∆∗) (A) is the simplicial set determined in level n by the set RMCg⊗Ω∗ DR(∆n) (A) := x ∈ (g ⊗ Ω∗ DR (∆n) ⊗ mA)1 s.t. d (x) + 1 2 [x, x] = 0 . (cid:26) (cid:27) Theorem 3.1. (Hinich, Lurie, Pridham) The functor RDef : dgLiek −−→ DefHin (cid:55)−→ RDef g g k 18 is an equivalence of (∞, 1)-categories, thus it induces an equivalence on the homotopy categories Ho (dgLiek) (cid:39) Ho Proof. See [37] Corollary 4.56. (cid:16) (cid:17) DefHin k . Despite its great theoretical properties, the Hinich nerve is seldom handy enough to make concrete computations. For this reason, recall that the (derived) Deligne groupoid associated to a differential graded Lie algebra g is defined to be the formal groupoid Delg : dgArt≤0 −−−−−→ Grpd (cid:55)−→(cid:2)(cid:103)MCg(A)/(cid:102)Ggg(A)(cid:3) A where (cid:103)MCg : dgArt ≤0 k A (cid:26) −−−−−−−−−−−−−−−→ Set (cid:55)→ x ∈ (g ⊗ mA)1 s.t. d (x) + (cid:102)Ggg : ≤0 dgArt k A −−−−−→ Grp (cid:55)→ exp (g ⊗ mA)0(cid:17) (cid:16) and let BDelg : dgArt ≤0 k −→ sSet (cid:27) [x, x] = 0 1 2 (3.1) (3.2) denote its nerve. Remark 3.2. Notice that formula (3.1) and formula (3.2) are just straightforward generalisations of the notions of Maurer-Cartan and gauge functor in underived Deformation Theory; these objects are used to define extended deformation functors in the sense of Manetti (see [30] or In [37] Pridham also proved that there is an equivalence of (∞, 1)-categories between [31]). and DefHin DefMan Warning 3.3. The nerve of the Deligne groupoid associated to a differential graded Lie algebra is a derived pre-deformation functor but not a derived deformation functor: as a matter of fact it is not homotopic in general. Moreover, although it might be a bit confusing, we will tend to refer to both Delg and BDelg as the Deligne groupoid associated to the differential graded Lie algebra g. . k k Fix g ∈ dgLiek: we can define the functor BDelg : dgArt ≤0 k −−−−−−−−−−−−−−−−−−−−−−−−−→ sSet / BDelg⊗Ω∗(∆1) (A) BDelg (A) A (cid:55)−→ diag (cid:16) / BDelg⊗Ω∗(∆2) (A) (cid:17) / ··· which is sometimes called the simplicial Deligne groupoid of g. Theorem 3.4. (Pridham) Let g be a differential graded Lie algebra concentrated in non-negative degrees; we have that 19 / o o o o / / / o o o o o o / / / / / o o o o o o o o • the functor BDelg is a derived deformation functor; • the functor BDelg is the universal derived deformation functor under BDelg; • the functors BDelg and RDef g are weakly equivalent. Proof. See [41] Section 3 for the proof of the first two claims, while the last statement is proved in [18] Section 3. As a consequence of Theorem 3.4 we have that all geometric and homotopy-theoretic in- formation concerning the Hinich nerve of a differential graded Lie algebra g are completely determined by its associated Deligne groupoid, which is a much more down-to-earth object as it is essentially a formal groupoid. Unfortunately, as BDelg does not map quasi-isomorphisms to weak equivalences, the description of higher tangent spaces we gave above in this section is no longer valid; nonetheless Pridham found a coherent way to define good cohomology theories for derived pre-deformation functors. As a matter of fact fix a derived pre-deformation functor ≤0 k → sSet and define as in [39] Section 3.3 the j-th generalised tangent space of F to F : dgArt be (cid:16) F (cid:16) k[ε] (cid:17)(cid:17) (ε2) π−j H j (F) := π0(tan(F(k[i])))/π0(tan(F(cone(k[i])))) if j ≤ 0 otherwise which is seen to be consistent with the definition given above in this section in case F is also homotopic (see [39] Lemma 3.15). Now fix g to be a differential graded Lie algebra over k concentrated in non-negative degrees and apply the above definitions to its Deligne groupoid. We have that H−1 (BDelg) = π1 π1 (cid:16)(cid:104) (cid:16) k[ε] (ε2) (cid:16) (cid:18) BDelg (cid:19) (cid:16) k[ε] (cid:18) (ε2) (cid:17)(cid:17) (cid:16)(cid:104)(cid:103)MCg (cid:19)(cid:105)(cid:17) (cid:39) Stab = π1 k[ε] (ε2) (cid:18) (cid:19) (cid:18) /(cid:102)Ggg (cid:19) (0) k[ε] (ε2) (cid:18) Ggg k[ε] (ε2) MCg k[ε] (ε2) /Ggg (cid:19)(cid:105)(cid:17) k[ε] (ε2) = (cid:17) = Ggg x ⊗ ε ∈ g1 ⊗ k[ε] (cid:111) = exp(cid:0)g0 ⊗ (ε)(cid:1) (cid:39) Id + g0ε (ε2) s.t. d (x) = 0 (cid:110) (cid:16) k[ε] (ε2) (cid:17) = Z 1 (g) ε but MCg and notice that the gauge action just reduces to Id + g0ε × Z 1 (g) ε (Id + aε, xε) ∗−−−−−→ Z 1 (g) ε (cid:55)−→ (x + d (a)) ε (3.3) therefore we get Stab Ggg (cid:18) k[ε] (ε2) (cid:19) (0) =(cid:8)(Id + aε) ∈ Id + g0ε s.t. (Id + aε) ∗ 0 = 0(cid:9) (cid:39) (cid:8)a ∈ g0 s.t. d (a) = 0(cid:9) = Z 0 (g) (cid:39) H 0 (g) 20 where the last identification follows from the fact that g lives in non-negative degrees. Similarly we see that (cid:16) (cid:18) (cid:17)(cid:17) (cid:18) (cid:16) k[ε] (cid:19) (ε2) /Ggg (cid:19)(cid:105)(cid:17) H 0 (BDelg) = π0 BDelg k[ε] (ε2) = MCg k[ε] (ε2) = (cid:19) (cid:39) Z1(g)ε/Id+g0ε k[ε] (ε2) (cid:16)(cid:104)(cid:103)MCg (cid:18) π0 (cid:18) k[ε] (ε2) (cid:19) /(cid:102)Ggg (cid:17) (cid:16) k[ε] thus the quotient of MCg At last observe that cone (k ⊕ k [j − 1]) is a path object for k ⊕ k [j], so the same kind of computation gives us that for all j ≥ 0 under the gauge action (3.3) is given by H 1 (g). (ε2) H j (BDelg) = π0(tan(BDelg(k[i])))/π0(tan(BDelg(cone(k[i])))) = Z0(g⊗(k⊕k[j]))/Z0(g⊗(k⊕cone(k⊕k[j−1]))) = Zj+1(g)/gj and again gj acts on Z j+1 (g) by differentials, so the quotient is H j+1 (g). Remark 3.5. Let g be any differential graded Lie algebra; by combining Theorem 3.4 and the above observations we have that H i (RDef g) (cid:39) H i (BDelg) = H−1 (RDef g) (cid:39) H−1 (BDelg) (cid:39) Stab Zi+1/gi (cid:39) H i+1 (g) (cid:19) (0) (cid:39) H 0 (g) d(cid:120) (cid:18) Ggg0 k[ε] (ε2) ∀i ≥ 0. 3.2 The Algebraic Fiorenza-Manetti-Martinengo Period Map The above considerations give many motivations to try to lift the period map from a mor- phism of classical deformation functors to the context of Derived Deformation Theory; Fiorenza and Martinengo started to address such a question, tackling it from an entirely algebraic view- point. Let X still be a proper smooth scheme of dimension d over a field k of characteristic 0 and, again, take the Cartan homotopy defined by the contraction of differential forms with vector fields and the Lie derivative [−1] X/k Ω∗ Ω∗ (cid:17) i : TX −→ End∗(cid:16) l : TX −→ End∗(cid:16) (cid:17) ∗(cid:16)RΓ (cid:16) ∗(cid:16)RΓ (cid:17)(cid:17) → End Ω∗ X/k X/k (cid:16) X, Ω∗ X/k [−1] (cid:17)(cid:17) (cid:17)(cid:17) X, Ω∗ X/k which corresponds to the differential of i in the Hom complex. Now consider the linear map of dgla's i : KSX (cid:39) RΓ (X, TX ) −→ End defined as the composition of RΓ (X, i) with the map (cid:16) X, End∗(cid:16) RΓ induced by the action of derived global sections of the endomorphism sheaf of Ω∗ X/k on de- rived global sections of X: this is still a Cartan homotopy; denote by l the associated mor- phism of dgla's, which is essentially the derived globalisation of the Lie derivative. Recall that 21 ≥0(cid:16)RΓ (cid:16) X, Ω∗ (cid:17)(cid:17) X/k is the differential graded Lie subalgebra of End con- End sisting of non-negatively graded endomorphisms of the (derived global sections of the) algebraic De Rham complex and that this is the same as the subalgebra of endomorphisms preserving the Hodge filtration. Notice also that the image of l is contained in End , therefore in the end there is a diagram of dgla's (cid:17)(cid:17) X, Ω∗ (cid:16) X/k X/k X, Ω∗ ≥0(cid:16)RΓ (cid:16) (cid:17)(cid:17) incl. X, Ω∗ X/k / End 0 X, Ω∗ X/k KSX (cid:39) RΓ (X, TX ) l / End ∗(cid:16)RΓ (cid:16) ≥0(cid:16)RΓ (cid:16) ∗(cid:16)RΓ (cid:17)(cid:17) (cid:17)(cid:17) where 0 stands for the zero map. Since i is a Cartan homotopy, by Remark 2.8 ei gives an homotopy between l and the zero map, thus there is an induced morphism of dgla's to the homotopy fibre ∗(cid:16)RΓ (cid:16) X, Ω∗ X/k (cid:17)(cid:17)(cid:19) (3.4) (cid:16)l,e i(cid:17) −−−−−→ holim←− (cid:18) End KSX as observed in [11] Section 6; moreover, recall from formula (2.13) that a model for the above homotopy fibre is the abelian dgla X, Ω∗ X/k ≥0(cid:16)RΓ (cid:16) (cid:16) ∗(cid:16)RΓ ≥0(cid:16)RΓ (cid:16) End End (cid:17)(cid:17) −−→−−→ End (cid:17)(cid:17) (cid:17)(cid:17) [−1] . X, Ω∗ X, Ω∗ X/k X/k Definition 3.6. Define the universal algebraic Fiorenza-Manetti local period map to be the morphism of deformation functors FM : Def KSX −−−−−→ Def End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) [−1] induced by map (3.4). Definition 3.7. Define the (universal) algebraic Fiorenza-Manetti-Martinengo local period map to be the morphism of derived deformation functors FMM : RDef KSX −−−−−→ RDef End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) [−1] induced by map (3.4). In Section 2.4 we described a universal version of the geometric period map (see Definition 2.21), but we did not construct its Lie-theoretic counterpart: Fiorenza and Martinengo showed that this is precisely given by map (3.4). Theorem 3.8. (Fiorenza-Martinengo) The diagram RDef KSX π0π≤0 Def KSX (cid:111) Def X FMM FM P 22 RDef End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X π0π≤0 X/k)) /k)) [−1] Def End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) (cid:111) [−1] / FlagF • H∗(X,k) / / / / / / / / / is well-defined and commutes. Proof. See [11] Section 6. 3.3 Affine DG≥0-Categories and the Dold-Kan Correspondence Theorem 3.8 says that morphism FMM is the correct derived enhancement of the universal Fiorenza-Manetti local period map; however the geometric interpretation of such a result is somehow indirect, thus it would be worth to find an equivalent morphism of derived deformation functors having more evident geometric meaning. Of course the key step in order to do this consists of finding the right domain and codomain for such a morphism, i.e. defining two derived deformation functors RDef X and hoFlagF • RΓ (cid:17) such that (cid:16) X,Ω∗ X/k • RDef X is weakly equivalent to RDef KSX and similarly hoFlagF • RΓ (cid:16) X,Ω∗ X/k (cid:17) is weakly equiv- alent to RDef End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) ; [−1] • RDef X and hoFlagF • RΓ tively. (cid:16) X,Ω∗ X/k (cid:17) are derived enhancements of Def X and FlagF • H∗(X,k), respec- In order to construct such functors we need some homotopy-theoretic background. Warning 3.9. In this section we will deal with non-negatively graded differential graded chain structures rather than non-positively graded cochain ones, though the pictures they provide are largely equivalent; the reason for this lies in the fact that -- at least in the framework of this paper -- the codomain of a derived deformation functor is the simplicial model category of simplicial sets, which is more directly related to chain structures than cochain ones. First of all, recall that the normalisation of a simplicial k-vector space (V•, ∂i, σj) is defined to be the non-negatively graded chain complex of k-vector spaces (NV, δ) where (NV )n := ker (∂i : Vn → Vn−1) (3.5) i and δn := (−1)n ∂n. Moreover, given a map f : V• → W• of simplicial k-vector spaces, we can define the chain map N (f ) : NV• −→ NW• identified by the relation N (f )n := fnNVn; notice that this construction gives us a well-defined morphism of chain complexes. In the end, there is a normalisation functor N : sVectk −→ Ch≥0 (Vectk) . (cid:92) at our disposal. On the other hand, let V be a chain complex of k-vector spaces and recall that its denormalisation is defined to be the simplicial vector space ((KV )• , ∂i, σj) given in level n by the vector space (Vp [η] (cid:39) Vp) . (cid:89) (KV )n := Vp [η] η∈Hom∆([p],[n]) η surjective 23 Remark 3.10. Notice that (KV )n (cid:39) V0 ⊕ V ⊕n 1 ⊕ V ⊕(n 2) 2 ⊕ ··· ⊕ V ⊕(n k) k ⊕ ··· ⊕ V ⊕(n n) n . In order to complete the definition of the denormalisation of V we need to define face and degeneracy maps: we will describe a combinatorial procedure to determine all of them. For all morphisms α : [m] → [n] in ∆, we want to define a linear map K (α) : (KV )n → (KV )m; this will be done by describing all restrictions K (α, η) : Vp [η] → (KV )m, for any surjective non-decreasing map η ∈ Hom∆ ([p] , [n]). For all such η, take the composite η ◦ α and consider its epi-monic factorisation16 ◦ η(cid:48), as in the diagram α [m] η(cid:48) [n] η [q] / [p] . Now • if p = q (in which case is just the identity map), then set K (α, η) to be the natural identification of Vp [η] with the summand Vp [η(cid:48)] in (KV )m; • if p = q + 1 and is the unique injective non-decreasing map from [p] to [p + 1] whose image misses p, then set K (α, η) to be the differential dp : Vp → Vp−1; • in all other cases set K (α, η) to be the zero map. The above constructions determine the whole of the simplicial vector space ((KV )• , ∂i, σj). As done for normalisation, for any chain map f : V → W we can define a morphism of simplicial k-vector spaces K (f ) : KV −→ KW by setting (cid:18) V0 × V ⊕n v0,(cid:0)vi (cid:1) 1 × V i , 1 ⊕(n 2) 2 (cid:16) (cid:17) vj 2 × ··· × V , . . . , vn j ⊕(n n) n (cid:19) (cid:18) K(f )n −−−−−→ (cid:55)−→ W0 × W ⊕n f0 (v0) ,(cid:0)f1 (cid:0)vi (cid:16) (cid:1)(cid:1) 1 × W i , 1 ⊕(n 2) 2 (cid:16) f2 vj 2 (cid:17)(cid:17) × ··· × W ⊕(n n) n (cid:19) , . . . , fn (vn) . j Again, there is a denormalisation functor K : Ch≥0 (Vectk) −→ sVectk. at our disposal. Theorem 3.11. (Dold, Kan) The functors N and K form an equivalence of categories between sVectk and Ch≥0 (Vectk). Proof. See [14] Corollary 2.3 or [50] Theorem 8.4.1. 16The existence of such a decomposition is one of the key properties of the category ∆. 24 / / / The Dold-Kan correspondence described in Theorem 3.11 is known to induce a number of very interesting ∞-equivalences; for instance the Eilenberg-Zilber shuffle product and the Alexander-Whitney map, which we will discuss in more details later in this section, allow us to extend normalisation and denormalisation to a pair of functors N : sAlgk (cid:29) dg≥0Algk : K which is seen to be a Quillen equivalence. Moreover recall that • a dg≥0-category over k is a category enriched in Ch≥0 (Vectk); • a k-simplicial category is a category enriched in sVectk; • a simplicial category is a category enriched in sSet; • a simplicial groupoid is a simplicial object in Grpd: equivalently a simplicial groupoid is a simplicial category in which all 1-morphisms are invertible. All the above structures form well-understood model categories; furthermore it is well-known in the homotopy-theoretic folklore that Theorem 3.11 induces a Quillen equivalence N : sCatk (cid:29) dg≥0Catk : K.17 (3.6) Tabuada also constructed an explicit Quillen equivalence between dg≥0Catk and sCat (see [42], where an explicit proof of formula (3.6) can be found, as well). We will use slightly more general versions of the Dold-Kan correspondence provided by Theorem 3.11 and its corollaries, so we need to develop a few technical tools. Define Affk to be the category whose objects are k-vector spaces and whose morphisms are affine maps between k-vector spaces, i.e. HomAffk (V, W ) := {v (cid:55)→ f (v) + b s.t. f linear, b ∈ W} (cid:39) HomVectk (V, W ) × W. Affk can be thought of as the category of affine spaces over k and affine maps. Given V, W ∈ Affk, define their tensor product to be (3.7) where the tensor product V ⊗W in the right-hand side of formula (3.7) is just the tensor product as vector spaces; in a similar way, given two affine maps V ⊗W := V ⊕ W ⊕ (V ⊗ W ) φ ∈ HomAffk (V, W ) where φ (v) := f (v) + b ψ ∈ HomAffk (U, Z) where ψ (u) := g (u) + d the tensor product map is given by φ ⊗ ψ : V ⊕ W ⊕ (V ⊗ W ) −→ U ⊕ Z ⊕ (U ⊗ Z) (u, v, x ⊗ y) (cid:55)→ (u + b, v + d, f (x) ⊗ g (y)) . (3.8) Formula (3.7) and formula (3.8) determine a monoidal structure on Affk: we will be more precise about this a little bit further in this section, when dealing with dg≥0-affine spaces. 17There is some abuse of notation in this statement. 25 Definition 3.12. Define a (chain) differential graded affine space over k in non-negative degrees (dg≥0-affine space for short) to be a pair (A0, V ) where V : d←− V1 d←− V2 d←− ··· d←− Vn d←− V0 is a non-negatively graded chain complex of k-vector spaces and A0 is an affine space over k whose difference vector space is V0. If (A0, V ) and (B0, W ) are dg≥0-affine spaces over k, a morphism φ : (A0, V ) → (B0, W ) will be a chain map which is affine in degree 0 and linear in higher degrees: more formally the (cid:111) set of morphisms between (A0, V ) and (B0, W ) is defined to be (cid:110) v (cid:55)→ f (v) + b s.t. f ∈ HomCh≥0(Vectk) (V, W ) , b ∈ W0 HomCh≥0(Affk) ((A0, V ) , (B0, W )) := . In the end we have a well-defined category of dg≥0-affine spaces over k, which we will denote as Ch≥0 (Affk). Remark 3.13. We have defined the objects of Ch≥0 (Affk) as pairs where the first term is an affine space and the second term is a chain complex of vector spaces just to make the affine structure explicit; an equivalent and more compact characterisation of Ch≥0 (Affk) is Ob(cid:0)Ch≥0 (Affk)(cid:1) := Ob(cid:0)Ch≥0 (Vectk)(cid:1) HomCh≥0(Affk) ((A0, V ) , (B0, W )) (cid:39) HomCh≥0(Vectk) (V, W ) × W0. In particular Ch≥0 (Affk) is a k-linear category. The category Ch≥0 (Affk) is both complete and cocomplete: limits and colimits are con- structed from those in Ch≥0 (Vectk). For example if (A0, V ) and (B0, W ) are dg≥0-affine spaces their product will be just (A0 × B0, V × W ), where V × W is the product of V and W in Ch≥0 (Vectk) and A0 × B0 is the affine space over k whose difference vector space is V0 × W0. We can also put a tensor structure over Ch≥0 (Affk): given two dg≥0-affine spaces (A0, V ) and (B0, W ), define their tensor product (A0, V ) ⊗ (B0, W ) to be the dg≥0-affine space determined by the chain complex (3.9) (3.10) V ⊕ W ⊕ (V ⊗ W ) . Similarly, given φ ∈ HomCh≥0(Affk) ((A0, V ) , (B0, W )) where φ (v) := f (v) + b ψ ∈ HomCh≥0(Affk) ((C0, U ) , (D0, Z)) where ψ (u) := g (u) + d the tensor product map is given by φ ⊗ ψ : V ⊕ W ⊕ (V ⊗ W ) −→ U ⊕ Z ⊕ (U ⊗ Z) (cid:0)u, v, x ⊗ y(cid:1) (cid:55)→(cid:0)u + b, v + d, f (x) ⊗ g(cid:0)y(cid:1)(cid:1) . Formula (3.9) and formula (3.10) determine a monoidal structure on Ch≥0 (Affk): in particular the unit is given by the object ({∗}, 0), the associator is induced by the monoidal structure on Ch≥0 (Vectk) and the unitors are simply given by (V ⊕ 0 ⊕ (V ⊗ 0)) ∼−→ V 26 (0 ⊕ V ⊕ (0 ⊗ V )) ∼−→ V. The reader can check that the above definitions verify the pentagon and the triangle identities. Remark 3.14. Let (A0, V ) and (B0, W ) be dg≥0-affine spaces: notice that (A0, V ) ⊗ (B0, W ) (cid:39) A0 ⊗B0 (3.11) where formula (3.11) is a canonical identification in Affk; an analogous coherence statement holds for morphisms. Definition 3.15. Define a simplicial affine space over k to be just a simplicial object in Affk. Let sAffk be the category of simplicial affine spaces over k, i.e. sAffk := Aff∆op Remark 3.16. There is a natural linearisation functor k . L : sAffk −→ sVectk which just deletes the non-linear part in the face and degeneracy maps defining a simplicial affine space, as well as the non-linear part of morphisms between simplicial affine spaces; in the same fashion there is a forgetful functor U : sVectk −→ sAffk which just takes (maps of) simplicial vector spaces and looks at them as (maps of) simplicial affine ones. Warning 3.17. The pair of functors (U, L) does not provide an adjunction between sAffk and sVectk. The category sAffk has all small limits and colimits, which are just taken levelwise; moreover define the tensor product in sAffk to be constructed by simply taking the tensor product in Affk in all levels: it is straightforward to check that this equips such a category with a monoidal structure. Now define the normalisation of a dg≥0-affine space over k to be the functor N : sAffk −−−−−→ Ch≥0 (Affk) A• A• (A0, N (L (A•))) N (L (A•)) (cid:55)−→  A0 B0 φ0 (cid:55)→ φ B•  , N(L(φ)) N (L (B•)) and observe that such a definition is well-given as the 0-th term of the chain complex N (L (A•)) is precisely the difference vector space of A0; in other words, the normalisation of a simplicial vector space does not affect the object in degree 0, as follows from formula (3.5). Analogously, define the denormalisation of a simplicial affine space over k to be the functor K : Ch≥0 (Affk) −−−−−−−−−−−−−−−−−−→ sAffk (A0, V ) (cid:55)−→ A0 / A0 × V1 / A0 × V ⊕2 1 × V2 / ··· 27 / o o o o / / / o o o o o o / / / / / o o o o o o o o where the maps involving A0 and A0 × V1 are A0 × V1 (cid:51) (a, v) (cid:55)→ a + d (v) ∈ A0 A0 × V1 (cid:51) (a, v) (cid:55)→ a ∈ A0 A0 (cid:51) a (cid:55)→ (a, 0) ∈ A0 × V1 and all other faces and degeneracies -- which do not involve the affine space A0, but rather only the vector spaces Vi -- are defined as done for classical denormalisation (see Remark 3.10 and subsequent discussion). In a similar way, given φ ∈ HomCh≥0(Affk) ((A0, V ) , (B0, W )) where φ (v) := f (v) + b the morphism K (φ) of simplicial affine spaces is defined in level n by the affine map ⊕(n n) n × ··· × W × ··· × V A0 × V ⊕n ⊕(n n) n ⊕(n 2) 2 ⊕(n 2) 2 (cid:18) −→ (cid:55)→ B0 × W ⊕n f0 (a0) + b,(cid:0)f1 (cid:0)vi (cid:1)(cid:1) 1 × W i , 1 (cid:16) (cid:16) (cid:17)(cid:17) f2 vj 2 (cid:19) (cid:18) a0,(cid:0)vi (cid:1) 1 × V i , 1 (cid:16) (cid:17) vj 2 j , . . . , vn , . . . , fn (vn) j (cid:19) We are ready to describe the generalisation of Theorem 3.11 we mentioned before. Proposition 3.18. The functors N and K form an equivalence of categories between sAffk and Ch≥0 (Affk). Proof. The arguments used in [50] Theorem 8.4.1 to prove the classical Dold-Kan correspondence given by Theorem 3.11 carry over to this context. As follows for instance from the discussion in [47] Section 2.3, the normalisation functor N can be made into a lax monoidal functor via the Eilenberg-Zilber shuffle map, which is the natural transformation EZ : N (−) ⊗ N (−) −→ N (− ⊗ −) determined for all A•, B• ∈ sAffk by the morphisms EZp,q A•,B• : N (A•)p ⊗ N (B•)q −−−−−−−−→ N (A• ⊗ B•)p+q a ⊗ b sign (µ, ν) σν (a) ⊗ σµ (b) (cid:55)−→ (cid:88) where the sum runs over all (p, q)-shuffles (µ,ν) (µ, ν) := (µ1, . . . , µp, ν1, . . . , νq) and the corresponding degeneracy maps are sµ := sµp ◦ . . . sµ1 sν := sνq ◦ . . . sν1 . In the same fashion, again from [47] Section 2.3, the denormalisation functor K can also be made into a lax monoidal functor by means of the Alexander-Whitney map. The latter is defined to be the natural transformation AW : N (− ⊗ −) −→ N (−) ⊗ N (−) 28 given for all A•, B• ∈ sAffk by the morphisms AWn A•,B• : N (A• ⊗ B•)n −−−−−−−−→(cid:16) N (A•) ⊗ N (B•) (cid:17) (cid:16) dp (a) ⊗ dq (cid:55)−→ (cid:77) a ⊗ b (cid:17) n (3.12) 0 (b) p+q=n "back face" dq "front face" dp and the 0 are induced respectively by the injective where the monotone maps δp : [p] → [p + q] and δp 0 : [q] → [p + q]; in particular the Alexander-Whitney map makes the normalisation functor N into a comonoidal one (again, see [47] Section 2.3, whose considerations adapt to these context). Notice also that by setting A(cid:48) := N (A) and B(cid:48) := N (A) in formula (3.12) and using the equivalence provided by Proposition 3.18, we get a version of the Alexander-Whitney transformation AW : K (−) ⊗ K (−) −→ K (− ⊗ −) 18 which makes the denormalisation K into a lax monoidal functor. Also we have that the composite AW◦ EZ is the same as the identity, while the transformation EZ◦ AW is chain homotopic to the identity: in particular the Dold-Kan equivalence provided by Proposition 3.18 is lax monoidal. Now we are ready to introduce the notions of affine dg≥0-category and affine simplicial category, which will be crucial technical tools to develop a good derived version of the period map. Definition 3.19. An affine differential graded category over k (C, C•) (affine dg≥0-category for short) is a category C enriched over Ch≥0 (Affk). We will denote by dg≥0CatAff k the ∞-category of affine dg≥0-categories. Let (C, C•) be an affine dg≥0-category and denote by H0 ((C, C•)) the (honest) category defined by the relations ∀X, Y ∈ C Ob (H0 ((C, C•))) HomH0((C,C•)) (X, Y ) := C := H0 (C• (X, Y )) . Definition 3.20. An affine differential graded groupoid over k (affine dg≥0-groupoid for short) will be an affine dg≥0-category (C, C•) such that the category H0 ((C, C•)) is a groupoid. We will denote by dg≥0GrpdAff k the ∞-category of affine dg≥0-groupoids. Remark 3.21. The notion of dg≥0-affine space allows us to define a notion of ∞-groupoid in the differential graded context: as a matter of fact a more naive notion of dg≥0-groupoid -- intended as a dg≥0-category where all morphisms in level 0 are isomorphism -- would not really make sense as every dg≥0-category comes with a zero morphism, which is seldom an isomorphism. Definition 3.22. An affine simplicial category over k (C, C•) is a category C enriched over sAffk We will denote by sCatAff k the ∞-category of affine simplicial categories. Let (C, C•) be an affine simplicial category and denote by π0 ((C, C•)) the (honest) category defined by the relations ∀X, Y ∈ C Ob (π0 ((C, C•))) Homπ0((C,C•)) (X, Y ) := C := π0 (C• (X, Y )) . 18There is some abuse of notation in this formula 29 Definition 3.23. An affine simplicial groupoid over k will be an affine simplicial category (C, C•) such that the category π0 ((C, C•)) is a groupoid. We will denote by sGrpdAff k the ∞-category of affine simplicial groupoids. Remark 3.24. The notion of simplicial affine space allows us to define a notion of ∞-groupoid in the k-simplicial context, just like dg≥0-affine spaces give rise to a good notion of differential graded groupoid, as observed in Remark 3.21 Of course any simplicial affine space has an underlying simplicial set, so an affine simplicial category over k is in particular a simplicially enriched category: more formally, there is a natural forgetful functors from sCatAff The slightly extended version of the Dold-Kan equivalence given by Proposition 3.18 induces a pair of functors to sCat. k N : sCatAff k (cid:29) dg≥0CatAff k : K19 (3.13) where N : ∀P, Q ∈ C ∀P, Q, R ∈ C ∀P, Q ∈ C ∀P, Q, R ∈ C and K : sCatAff k −−−−−−−−−−−−−−−→ dg≥0CatAff C (cid:55)−−−−−−−−−−−−−−−−−−−−→ C k C• (P, Q) (cid:55)−−−−−−−−−−−−−→ N (C• (P, Q))  C• (P, Q) ⊗ C• (Q, R) ◦  (cid:55)→ C• (P, R) N (C• (P, Q)) ⊗ N (C• (Q, R)) EZC• (P,Q),C• (Q,R) N (C• (P, Q) ⊗ C• (Q, R)) N(◦) N (C• (P, R)) dg≥0CatAff k −−−−−−−−−−−−−−−→ sCatAff k C (cid:55)−−−−−−−−−−−−−−−−−−−−→ C C• (P, Q) (cid:55)−−−−−−−−−−−−−→ K (C• (P, Q))  C• (P, Q) ⊗ C• (Q, R) ◦  (cid:55)→ C• (P, R) K (C• (P, Q)) ⊗ K (C• (Q, R)) AWC• (P,Q),C• (Q,R) K (C• (P, Q) ⊗ C• (Q, R)) K(◦) K (C• (P, R))     Notice also that the ∞-equivalence given by formula (3.13) restricts to an ∞-equivalence 19There is some abuse of notation in this formula. N : sGrpdAff k (cid:29) dg≥0GrpdAff k : K. 30 At last, let us recall that there is a natural functor ¯W : sCat −→ sSet (3.14) given by the right adjoint to Illusie's Dec functor; we are not describing it explicitly as its construction is slightly technical and not really needed for the sake of this paper: the definition of ¯W can be found in [14] Section V.7 or [40] Section 1. Moreover in [1] Cegarra and Remedios proved that ¯W is weakly equivalent to the diagonal of the simplicial nerve functor. Functor (3.14) is also known to induce a right Quillen equivalence ¯W : sGrpd −→ sSet and -- as a corollary of the results in [40] Section 1 -- we also have that functor (3.14) restricts to an equivalence ¯W : sGrpdAff k −→ sSet. In Section 3.4 we will apply the functor ¯W to interesting affine simplicial groupoids in order to define rigorously the derived deformation functor hoFlagF • RΓ will be constructed in Section 3.5 by using different techniques. (cid:17), while the functor RDef X X,Ω∗ (cid:16) X/k 3.4 Derived Deformations of Filtered Complexes In this section we will study in detail how to enhance the formal Grassmannian and the flag functor analysed in Section 2.2 and Section 2.4 to the world of derived deformation functors; in particular we will define the notions of derived formal total Grassmannian and formal homotopy flag variety and make comparisons with the corresponding derived stacks studied in [6]. Throughout this section fix V to be a complex of k-vector spaces, S a subcomplex of it and F• a filtration on V . Also recall from [6] that for any R in Algk or even dgAlg there is a model structure on FdgModR modelled on the projective model structure over dgModR. Definition 3.25. A derived deformation of V over A ∈ dgArt complex of A-modules VA and a surjective quasi-isomorphism ϕ : VA ⊗A k → V . Definition 3.26. A derived deformation of S inside V over A ∈ dgArt derived deformation (VA, ϕ) of V and a subcomplex SA ⊆ VA such that is the datum of a cofibrant is the datum of a ≤0 k ≤0 k ≤0 k ϕ(cid:12)(cid:12)SA : SA ⊗A k −→ S is still a surjective quasi-isomorphism. Definition 3.27. A derived deformation of (V, F•) over A ∈ dgArt where (VA, F• surjective morphism such that the maps Fp A for all p. A) is a cofibrant filtered complex of A-modules and ϕ : (VA, F• A) , ϕ), A) → (V, F•) is a VA⊗Ak → FpV induced by ϕ are quasi-isomorphisms is a pair ((VA, F• ≤0 k Remark 3.28. Notice that the key technical assumptions in Definition 3.25, Definition 3.26 and Definition 3.27 are 1. restricting to cofibrant objects while defining derived deformations of (filtered) derived modules; 31 2. requiring that the structure morphism ϕ is surjective. Assumption (1) allows us to use tensor products instead of derived tensor products. Assumption (2) makes objects in te slice category dgModA/V (or FdgModA/(V,F•)) fibrant. Actually in the main geometric example we will consider, that is derived deformations of (filtered) perfect complexes of modules over a scheme, the perfectness condition will imply cofibrancy wherever relevant. Derived deformation of complexes, subcomplexes and filtrations are governed by nice derived deformation functors, already hidden in the Derived Deformation Theory folklore: we now propose a rigorous way to define them via the language of affine dg≥0-categories. In the above notations, let (cid:0)hDefV (A) , hDefV (A)• (cid:1) be the affine dg≥0-category defined by the formulae hDefV (A) := {derived A-deformations of V } and for all (VA, ϕ) , (WA, φ) ∈ hDefV (A) hDefV (A)0 :=(cid:8)Ψ ∈ Hom0 (VA, WA) s.t. φ ◦ Ψ = ϕ(cid:9) hDefV (A)1 :=(cid:8)Ψ ∈ Hom−1 (VA, WA) s.t. φ ◦ Ψ = 0(cid:9) hDefV (A)n :=(cid:8)Ψ ∈ Hom−n (VA, WA) s.t. φ ◦ Ψ = 0(cid:9) ... ... with the differential induced by the standard differential on Hom complexes. Similarly, let (cid:16) (cid:16) hDef F• V (A) , hDef F• V (A)• 32 hDefS,V (A) , hDefS,V (A)• be the affine dg≥0-category defined by the formulae hDefS,V (A) := {(derived) A-deformations of S inside V } and for all ((SA, VA) , ϕ) , ((TA, WA) , φ) ∈ hDefS,V (A) hDefS,V (A)0 :=(cid:8)Ψ ∈ Hom0 ((SA, VA) , (TA, WA)) s.t. φ ◦ Ψ = ϕ(cid:9) hDefS,V (A)1 :=(cid:8)Ψ ∈ Hom−1 ((SA, VA) , (TA, WA)) s.t. φ ◦ Ψ = 0(cid:9) hDefS,V (A)n :=(cid:8)Ψ ∈ Hom−n ((SA, VA) , (TA, WA)) s.t. φ ◦ Ψ = 0(cid:9) ... ... with the differential induced by the standard differential on Hom complexes. Finally let (cid:17) (cid:17) be the affine dg≥0-category defined by the formulae hDef F• V (A) := {(derived) A-deformations of (V, F•)} and for all ((VA, F• hDef F• hDef F• V (A) A) ϕ) , ((WA, G• A) , φ) ∈ hDef F• V (A)0 :=(cid:8)Ψ ∈ Hom0 ((VA, F• V (A)1 :=(cid:8)Ψ ∈ Hom−1 ((VA, F• V (A)n :=(cid:8)Ψ ∈ Hom−n ((VA, F• ... hDef F• A) , (WA, G• A) , (WA, G• A) , (WA, G• A)) s.t. φ ◦ Ψ = ϕ(cid:9) A)) s.t. φ ◦ Ψ = 0(cid:9) A)) s.t. φ ◦ Ψ = 0(cid:9) ... with the differential induced by the standard differential on Hom complexes. Remark 3.29. The 2-out-of-3 property implies that morphisms in hDefV (A)0, hDefS,V (A)0 and hDef F• V (A)0 are all weak equivalences; in particular (cid:1)(cid:1) (cid:0)(cid:0)hDefV (A) , hDefV (A)• (cid:16)(cid:16) (cid:16)(cid:16) hDefS,V (A) , hDefS,V (A)• hDef F• V (A) , hDef F • V (A)• (cid:17)(cid:17) (cid:17)(cid:17) • H0 • H0 • H0 are groupoids, so the affine dg≥0-categories (cid:1) • (cid:0)hDefV (A) , hDefV (A)• • (cid:16) • (cid:16) hDefS,V (A) , hDefS,V (A)• hDef F• V (A) , hDef F• V (A)• are really affine dg≥0-groupoids. (cid:17) (cid:17) Now define the derived deformation functors RDef V : ≤0 dgArt k A RDef S,V : ≤0 dgArt k A RDef F• V : dgArt ≤0 k A −−−−−−−−−−−−−−−→ sSet (cid:55)−→ ¯W (cid:16) K(cid:0)(cid:0)hDefV (A) , hDefV (A)• (cid:16)(cid:16) (cid:16) K (cid:16)(cid:16) (cid:16) K V (A) , hDef F• −−−−−−−−−−−−−−−→ sSet (cid:55)−→ ¯W hDef F• −−−−−−−−−−−−−−−→ sSet (cid:55)−→ ¯W hDefS,V (A) , hDefS,V (A)• V (A)• . (cid:1)(cid:1)(cid:17) (cid:17)(cid:17)(cid:17) (cid:17)(cid:17)(cid:17) . . Lemma 3.30. In the above notations we have: 1. RDef V is weakly equivalent to the derived deformation functor RDef End∗(V ) associated to ∗ the dgla End (V ); 33 2. RDef S,V is weakly equivalent to the derived deformation functor RDef EndS (V ) associated to the dgla EndS (V ); 3. RDef F• V is weakly equivalent to the derived deformation functor RDef EndF• (V ) associated to the dgla EndF• (V ); Proof. This result is well-known in the Derived Deformation Theory folklore: we just recall the morphism RDef End∗ (V ) −→ RDef V giving Claim (1). Note that by Theorem 3.4 it suffices to determine such a map on the (derived) Deligne groupoid ∗ BDelEnd∗(V ) associated to the dgla End for all A ∈ dgArt ≤0 (V ), so define ν : BDelEnd∗(V ) −−−−−→ RDef V k (cid:103)MCEnd∗(V ) (A) (cid:51) σ (cid:55)−→ [(V ⊗ A, d + σ)]   (V ⊗ A, d + σ1) (cid:102)GgEnd∗(V ) (A) (cid:51) ξ (cid:55)−→ eξ (V ⊗ A, d + σ2) Map ν is known to be a weak equivalence: a very rigorous but quite abstract proof can be found in [36] Section 4.1, while a simpler one can be found in [31]; see also [11] Section 6. Claim (2) and Claim (3) are proved in an entirely analogous way. Now we are ready to define coherent derived version of formal Grassmannians and flag functors. Definition 3.31. Define the derived formal total Grassmannian associated to S inside V to be the functor hoGrassS,V = holim←− RDef S,V −−−−−−−−−→ forgetful map −−−−−−−−−→ 0 RDef V . In particular hoGrassS,V is a well-defined derived deformation functor. Definition 3.32. Define the formal homotopy flag variety associated to (V, F•) to be the functor (cid:18) (cid:18) (cid:19) (cid:19) . hoFlagF• V = holim←− RDef F• V −−−−−−−−−→ forgetful map −−−−−−−−−→ 0 RDef V In particular hoFlagF• Proposition 3.33. In the above notation we have that: V is a well-defined derived deformation functor. 20Recall from formula (2.12) that 1. The functors hoGrassS,V and RDef CS,V are weakly equivalent.20 (cid:19) (cid:18) CS,V := holim←− EndS (V ) χ−→−→ 0 ∗ End (V ) . 34 2. The functors hoFlagF• V and RDef C F• V are weakly equivalent;21 Proof. We only prove Claim (1) as Claim (2) is proved in an entirely similar way. We want to show that the functor hoGrassS,V is weakly equivalent to the Hinich nerve of (cid:18) CS,V := holim←− EndS (V ) incl.−−−→−−−→ 0 ∗ End (V ) . (cid:19) (cid:19) . As proved in [12] and [16], the functor RDef (homotopically) commutes with homotopy limits, thus we have that RDef (cid:32) (cid:33) EndS (V )−−→−−→End∗(V ) holim←− is weakly equivalent to (cid:18) holim←− RDef EndS (V ) −−→−−→ RDef End∗(V ) By Lemma 3.30 we have that RDef V is weakly equivalent to RDef End∗(V ), while the functor RDef S,V is weakly equivalent to RDef EndS (V ), thus the statement follows by applying Definition 3.31. As a consequence of Proposition 3.33 we have that functors hoGrassS,V and hoFlagF• V : this is exactly the content of the next result. really derived versions of GrassS,V and FlagF• Proposition 3.34. In the above notations we have that: V are 1. the derived formal total Grassmannian hoGrassS,V is a derived enhancement of the functor GrassS,V , i.e. π0π≤0hoGrassS,V (cid:39) GrassS,V ; 2. the formal homotopy flag variety hoFlagF• V variety FlagF• V , i.e. π0π≤0hoFlagF• V (cid:39) FlagF• V . is a derived enhancement of the formal flag Proof. Again, we only prove Claim (1) as Claim (2) is proved in an entirely similar way. Theorem 2.6 and Proposition 3.33.1 state that the (homotopy) dgla CS,V represents both the deformation functor GrassS,V and the derived deformation functor hoGrassS,V ; Theorem 3.1 implies that the latter has to be a derived enhancement of the former. At last recall from [6] that there exist two (locally geometric) derived stacks DGrassk (V ) and DFlagk (V ) which respectively enhance the total Grassmannian variety Grass (H∗ (V )) and the (total) flag variety Flag (H∗ (V )) to the derived world (see [6] Theorem 2.42 and [6] Theorem 2.45). As a conclusion of this section we want to show that functors hoGrassS,V and hoFlagF• are formal neighbourhoods of the above derived stacks. V 21Recall from formula (2.13) that (cid:18) C F• V := holim←− EndF• (V ) χ−→−→ 0 ∗ End (cid:19) (V ) . 35 Proposition 3.35. In the above notations we have that: 1. the derived deformation functor hoGrassS,V is the formal neighbourhood of the derived stack DGrassk (V ) at [S (cid:44)→ V ]; 2. the derived deformation functor hoFlagF• DFlagk (V ) at (V, F•). V is the formal neighbourhood of the derived stack (cid:33) RPerfk (cid:91) Proof. Again, we only prove Claim (1) as Claim (2) is proved in an entirely similar way. As a first step, recall that -- by [6] Proposition 2.41 -- the stack DGrassk (V ) is an open derived substack of the big total derived Grassmannian (cid:32) DGRASSk (V ) := holim←− RSubk [U (cid:44)→W ](cid:55)→W −−−−−−−−−→ −−−−−−−−−→ constV where RPerfk = (cid:91) RPerf n k RSubk = RSubn k n n are respectively the locally geometric derived stack of perfect complexes and the locally geometric derived stack of perfect subcomplexes: for more details see [6] Section 2.2 and 2.3. As a result, it suffices to show that hoGrassS,V is the formal neighbourhood of DGRASSk (V ) at [S (cid:44)→ V ], which by Definition 3.31 amounts to prove that RDef S,V is the formal neighbourhood of RSubk at [S (cid:44)→ V ] and RDef V is the formal neighbourhood of RPerfk at V ; we show only the first assertion, as the other one is proved analogously. Observe from [6] Remark 2.34, [6] Corollary 2.32 and [6] Theorem 2.33 that the derived stack RSubn k is obtained as follows: given for all A ∈ Algk the functorial simplicial category sub,k (A) := full simplicial subcategory of FdgModA made of pairs (E , C ) of perfect complexes of length at most n and for which E ⊆ C Mn construct for all A ∈ dgbNil ≤0 k the functorial simplicial category Mn sub,k (A) := full simplicial subcategory of W (F dCARTk (A)) 22 A H 0 (A) , C ⊗L made of pairs (E , C ) such that (E ⊗L is weakly equivalent to an object in Mn (cid:0)H 0 (A)(cid:1) sub,k A H 0 (A)) and end up with a functor ¯W Mn sub,k which turns out to be via Lurie-Pridham Representability ≤0 k of the derived geometric stack RSubn (see [26] or [39] for more details) the restriction to dgbNil k, which is fully determined then. On the other hand recall from the beginning of this section that the derived deformation functor RDef S,V is obtained from the formal affine dg≥0-groupoid (cid:16) (cid:17) hDefS,V (A) , hDefS,V (A)• A ∈ dgArt ≤0 k by applying functor ¯W and (affine) Dold-Kan denormalisation K. Now fix A ∈ dgArt ≤0 k ; clearly we have sub,k(k) {(S, V )} (cid:39) K (cid:16)(cid:16) Mn sub,k (A) ×h Mn hDefS,V (A) , hDefS,V (A)• (cid:17)(cid:17) A ∈ dgArt ≤0 k (3.15) 22For the precise meaning of F dCARTk see [6] formula (2.29). 36 thus the equivalence of RDef S,V and the formal neighbourhood of RSubn follows by applying functor ¯W to formula (3.15). k ⊆ RSubk at (S, V ) 3.5 Derived Deformations of k-Schemes Now we want to describe the functor RDef X which parametrises derived deformations of the scheme X: the idea consists of deforming the scheme X through derived schemes instead of ordinary schemes. There are a variety of equivalent definitions of derived scheme (in particular see [26] Definition 4.5.1 and [45] Chapter 2.2 for the two most standard ways to look at it); the one we are about to recall probably is not the most elegant, but it is definitely the handiest one to make actual computations. As a matter of fact, by [38] Theorem 6.42 a derived scheme S over k can be seen as a pair(cid:0)π0S, OS,∗(cid:1), where π0S is an honest k-scheme and OS,∗ is a presheaf of differential graded commutative algebras in non-positive degrees on the site of affine opens of π0S such that: • the (cohomology) presheaf H0 (OS,∗) (cid:39) Oπ0S; • the (cohomology) presheaves Hn (OS,∗) are quasi-coherent Oπ0S-modules. Also, recall from [38] that a morphism f : A → B in dgAlg ≤0 k is said to be homotopy flat if is flat and the maps H 0 (f ) : H 0 (A) −→ H 0 (B) H i (A) ⊗H 0(A) H 0 (B) −→ H i (B) are isomorphisms for all i; moreover a very useful characterisation says that f is homotopy flat if and only if B ⊗L A H 0 (A) is (weakly equivalent to) a discrete flat H 0 (A)-algebra: for a proof see [40] Lemma 3.13.23 Now define a derived deformation of the scheme X over A ∈ dgArt≤0 to be a homotopy pull-back diagram of derived schemes X i 2h X p Spec (k) / RSpec (A) where the map p is homotopy flat; equivalently such a deformation can be seen as a morphism OA,∗ → OX of presheaves of differential graded commutative algebras over A such that: 1. OA,∗ is homotopy flat; 2. the induced k-linear morphism OA,∗ ⊗L 3. the morphism OA,∗ → OX is surjective; 4. OA,∗ is cofibrant. A k → OX is a weak equivalence; 23[40] and [38] actually deal with homotopy flatness in terms of simplicial and dg≥0 chain algebras; nevertheless all definitions and arguments readily adapt to cochain algebras in non-positive degrees. 37 / / / Remark 3.36. In the above notations, Condition (1) and Condition (2) are proper deformation- theoretic conditions, which resemble the ones characterising underived deformations of schemes (see Section 2.1), while Condition (3) and Condition (4) are fibrancy-cofibrancy conditions, which are needed in order to ensure that certain maps of derived deformation functors which will arise in the rest of the paper are well-defined. Now consider the formal groupoid DelX : dgArt A ≤0 k −−−−−→ Grpd (cid:55)−→ DelX (A) defined by the formulae (cid:16)OA,∗ ϕ→ OX (cid:17) (cid:16)O(cid:48) and for all HomDelX (A) (ϕ, φ) :=(cid:8)Ψ ∈ Hom0 A,∗ , φ→ OX (cid:17) ∈ DelX (A) (cid:0)OA,∗, O(cid:48) DelX (A) := {(derived) deformations of X over A} A,∗(cid:1) s.t. φ ◦ Ψ = ϕ, Ψ ≡ Id (mod mA)(cid:9) A (3.16) Remark 3.37. In the notations of formula (3.16), notice that the condition Ψ ≡ Id (mod mA) ≤0 ensures that DelX (A) is a groupoid for all A ∈ dgArt k ; roughly speaking, the formal groupoid DelX can be thought as some sort of (derived) Deligne groupoid associated to the scheme X, meaning that its role is intended to formally resemble the one played by the (derived) Deligne groupoid associated do a differential graded Lie algebra, which we described in Section 3.1. Now consider the functor BDelX : dgArt ≤0 k −−−−−→ sSet given by the nerve of DelX and define RDef X : dgArt ≤0 k −−−−−→ sSet to be the right derived functor of BDelX. The definition of DelX implies immediately that this is a derived pre-deformation functor, thus -- by [39] Theorem 3.16 -- RDef X turns to be a derived deformation functor. Now let us briefly look at global derived moduli of schemes; consider the assignment Stack0 n/k : Algk −−−−−→ sCat (cid:55)−→ Stack0 A n/k (A) := simplicial category of algebraic n-spaces over A which is the simplicial (underived) moduli functor classifying (underived) 0-stacks of dimension n over k. Pridham has shown that such a functor induces a derived stack DSchn/k parametrising derived schemes over k of dimension n: see [40] Example 3.36 for a detailed construction. Un- fortunately the stack DSchn/k is far too large to be geometric, nonetheless Pridham constructed many interesting geometric substacks of its: see [40] Section 3 for more details. Remark 3.38. The derived deformation functor RDef X is the formal neighbourhood of the derived stack DSchd/k at X; in particular, it follows -- using either [38] Theorem 8.8 or [40] Theorem 10.8 -- that (cid:0)LX/k, OX (cid:1) H i (BDelX ) (cid:39) H i (RDef X ) (cid:39) Exti+1 OX 38 GgKSX (A) (cid:51) ξ (cid:55)−→ (3.17)  τ≤0ROA (x1) eξ OX τ≤0ROA (x2)  Theorem 3.39. The functors RDef X and RDef KSX are weakly equivalent. Proof. We want to construct a natural transformation RDef KSX −−−−−→ RDef X providing a weak equivalence between such derived deformation functors, i.e. an isomorphism on the level of homotopy categories. Again, by Theorem 3.4 and the definition of RDef X it is enough to define such a morphism on BDelKSX , thus define the map for all A ∈ dgArt ≤0 k µ : BDelKSX −−−−−−−−→ BDelX (cid:55)−→(cid:2)τ≤0ROA (x) → τ≤0ROk (x) (cid:39) OX MCKSX (A) (cid:51) x (cid:3) (cid:16)A 0,∗ (cid:17) X ⊗ A, ¯∂ + lx where ROA (x) := , l being the Lie derivative (i.e. the differential of the contraction map) and the map τ≤0ROA (x) → τ≤0ROk (x) is induced by A (cid:16) A/mA (cid:39) k. Notice that the complex ROA (x) is cofibrant, since it is bounded above and the underlying graded module A 0,∗ X ⊗A is projective. Furthermore observe that the surjectivity of the natural map A (cid:16) A/mA together with the surjectivity of the canonical morphism τ≤0ROA (x) (cid:16) H 0(cid:0)τ≤0ROA (x)(cid:1) -- due in turn to the fact that the complex τ≤0ROA (x) lives in non-positive degrees -- ensures that the morphism τ≤0ROA (x) → OX is surjective. In order to show that map 3.17 is well-defined, we have to check that (cid:2)τ≤0ROA (x) → τ≤0ROk (x) (cid:39) OX (cid:3) actually determines a derived deformation of the scheme X, i.e. we need to prove that τ≤0ROA (x) is homotopy flat over A and τ≤0ROA (x) ⊗L A k is weakly equivalent to OX as complexes of presheaves of differential graded commutative k-algebras: this essentially means to verify that ROA (x) ⊗L Let us first prove that τ≤0ROA (x) is weakly equivalent to ROA (x). Filter the latter complex by powers of the maximal ideal mA of A, i.e. define the filtered complex (ROA (x) , F•) through the relation A H 0 (A) is flat over H 0 (A). FpROA (x) := mp ROA (x) A and take the associated graded object Grp (F•) := Notice that formula (3.18), so mp A mp+1 ROA (x) ROA (x) A . (3.18) Grp (F•) (cid:39) A 0,∗ X ⊗ mp A mp+1 A . 39 % % 9 9 H p+q (cid:32) (cid:17) H j(cid:16) mp H i(cid:16)A 0,∗ (cid:17) A mp+1 X Now consider the spectral sequence (cid:33) (cid:39) (cid:77) i+j=p+q (cid:32) H i(cid:16)A 0,∗ X (cid:17) ⊗ H j (cid:32) (cid:33)(cid:33) mp A mp+1 A =⇒ H p+q (ROA (x)) (3.19) X ⊗ mp A 0,∗ A mp+1 A = 0 when j > 0 and, since the which converges by the Classical Convergence Theorem (see [50] Theorem 5.5.1); note that X ←(cid:45) OX provides a X and OX in the category of OX-modules in complexes, also = 0 when i > 0: this means that at least one of these two terms vanishes whenever weak equivalence between A 0,∗ "Dolbeaut" resolution A 0,∗ A i + j > 0, so the convergence of spectral sequence (3.19) implies that ∀n > 0. In particular ROA (x) and τ≤0ROA (x) are weakly equivalent. Now we want to prove that τ≤0ROA (x) ⊗L H n (ROA (x)) = 0 A H 0 (A) is flat over H 0 (A); first notice that τ≤0ROA (x) ⊗L A H 0 (A) ≈ ROA (x) ⊗L A H 0 (A) ≈ ROA (x) ⊗A H 0 (A) 24 so it is enough to show that ROA (x) ⊗A H 0 (A) is flat over H 0 (A). In order to prove this let M be any H 0 (A)-module, consider the complex (cid:0)ROA (x) ⊗A H 0 (A)(cid:1) ⊗H 0(A) M (cid:39) ROA (x) ⊗A M. and filter it by powers of the maximal ideal mH 0(A) of H 0 (A), i.e. define the filtered complex (ROA (x) ⊗A M, F•) through the relation Fp (ROA (x) ⊗A M ) := mp H 0(A) (ROA (x) ⊗A M ) . As before, the associated graded object is Grp (F) := H 0(A) (ROA (x) ⊗A M ) mp H 0(A) (ROA (x) ⊗A M ) mp+1 (cid:39) A 0,∗ X ⊗ mp mp+1 H 0(A)M H 0(A)M and there is a spectral sequence (cid:32) H p+q H 0(A)M H 0(A)M X ⊗ mp A ∗,0 mp+1 (cid:33) (cid:39) (cid:77) i+j=p+q X ⊗ mp A ∗,0 mp+1 H 0(A)M H 0(A)M (cid:32) (cid:19) (cid:33) =⇒ H p+q (ROA (x) ⊗A M ) (3.20) (cid:32) H i(cid:16)A 0,∗ X (cid:17) ⊗ H j (cid:32) mp H 0(A)M H 0(A)M mp+1 (cid:33)(cid:33) = 0 for all j (cid:54)= 0, thus the convergence of spectral sequence (3.20) implies which still converges because of the Classical Convergence Theorem. Of course H p+q (cid:18) mp and H j M H0(A) mp+1 H0(A) M TorA n (ROA (x) , M ) (cid:39) H−n (ROA (x) ⊗A M ) = 0 ∀n (cid:54)= 0 24The symbol ≈ stands for "weakly equivalent". 40 which gives us the flatness of τ≤0ROA (x) ⊗L putation also ensures that the map (cid:2)τ≤0ROA (x) → τ≤0ROk (x) (cid:39) OX (cid:3) A H 0 (A) over H 0 (A). Notice that the same com- is quasi-smooth. Now we want to prove that map 3.17 is a weak equivalence of derived deformation functors; by [37] Corollary 1.49 it suffices to check that such a map induces isomorphisms on generalised tangent spaces, so consider the morphisms H i (µ) : H i (BDelKSX ) −→ H i (BDelX ) i ≥ −1 and notice that higher tangent maps in larger negative degrees vanish as KSX lives only in non-negative degrees. For all i ≥ 0 we have the chain of canonical identifications H i (RDef KSX ) (cid:39) H i (BDelKSX ) (cid:39) H i+1 (KSX ) (cid:39) H i+1(cid:16) (cid:17) (cid:39) H i+1 (X, TX ) X, A 0,∗ X (TX ) where the first and the second isomorphism come from Remark 3.5, the third one is true just by definition and the last one is given by the Dolbeaut Theorem. In the same fashion, there is also a chain of canonical isomorphisms (cid:16)OX , Hom H i (BDelX ) (cid:39) Exti+1 OX X/k, OX Ω1 (cid:16) (cid:0)LX/k, OX (cid:17) [−i − 1] (cid:16) (cid:17) (cid:39) (cid:1) (cid:39) HomD(X) (cid:17) (cid:39) HomD(X) (OX , TX [−i − 1]) (cid:39) Exti+1 X/k ⊗ OX , OX [−i − 1] Ω1 HomD(X) (OX , TX ) where the first isomorphism -- as we discussed before -- comes from the fact that RDef X is the formal neighbourhood of a derived stack of schemes, the third one is true by adjunction, while all the other ones directly follow from definitions. Finally, for all i ≥ 0 we see that the map H i (µ) is OX H i (µ) : H i+1 (X, TX ) −→ Exti+1 (OX , TX ) OX ξ−→ TX [−i − 1] (cid:55)−→(cid:16)OX ξ (cid:17) (3.21) ξ−→ TX [−i − 1] is nothing but the where the (cohomology class of the) degree i morphism OX map25 induced in D (X) by the cocycle ξ; on the other hand -- again by using Remark 3.5 -- the map H−1 (µ) turns out to be H−1 (µ) : H 0 (KSX ) (cid:39) Stab (cid:19) (0) −→ H 0 (X, TX ) (cid:39) Ext1 (OX , TX ) (cid:18) OX Gg Γ(X,A 0,0 X )) X (T Id + ξ k[ε] (ε2) (cid:55)−→ ξ. (3.22) Both map (3.22) and map (3.21) are clearly isomorphisms, so this completes the proof. Remark 3.40. RDef X is a derived enhancement of Def X, i.e. 25Of course, there is some abuse of notation in this sentence. π0π≤0RDef X (cid:39) Def X . 41 3.6 The Geometric Fiorenza-Manetti-Martinengo Period Map Now we have all the ingredients to give a geometric interpretation of the map FMM described in Definition 3.7. Definition 3.41. Define the (universal) geometric Fiorenza-Manetti-Martinengo local period map to be the morphism of derived deformation functors RP : RDef X −−−−→ hoFlagF • RΓ X,Ω∗ (cid:17) (cid:16) X/k identified by the universal morphism of derived deformation functors under the map of derived pre-deformation functors given for all A ∈ dgArt ≤0 k by (cid:104)OA,∗ ϕ−→ OX (cid:105) ≈ BDelX  X Spec (k) O(cid:48) A,∗ Ψ O(cid:48)(cid:48) A,∗ ϕ(cid:48) ϕ(cid:48)(cid:48) OX  X / RSpec (A) 2h  −−−−→  (cid:55)→(cid:104)(cid:16)(cid:16)RΓ (cid:16)  (cid:16)(cid:16)RΓ (cid:16) (cid:16) (cid:16)(cid:16)RΓ (cid:55)→ hoFlagF • RΓ X,Ω∗ π0X, Ω∗ X/A (cid:16) (cid:17) , ϕ X/k (cid:17) , F •(cid:17) , F •(cid:17) (cid:17) (cid:17) , F •(cid:17) (cid:17)(cid:105) , ϕ(cid:48)(cid:17) , ϕ(cid:48)(cid:48)(cid:17)  π0X, Ω∗ X(cid:48)/A Ψ π0X, Ω∗ X(cid:48)(cid:48)/A where • ϕ is the derived globalisation of the natural A-linear map extending ϕ to the algebraic De Rham complex; • Ψ is constructed by using the same universal property; • the complex is sometimes known as derived de Rham complex and the Hodge filtration over it is just (cid:16) (cid:16) (cid:17) (cid:17) (cid:89) (cid:32) i(cid:94) LX/A (cid:33) (cid:32) i(cid:94) LX/A (cid:89) i (cid:33) . i≥p RΓ π0X, Ω∗ X/A := F pRΓ π0X, Ω∗ X/A := Remark 3.42. The fibrant-cofibrant replacement properties pointed out in Remark 3.28 and Remark 3.36 ensure that the geometric Fiorenza-Manetti-Martinengo local period map described in Definition 3.41 is well-defined. In the end all constructions and results we have discussed so far sum up in the following theorem. 42 / / / " " < < Theorem 3.43. The diagram of derived deformation functors and (Schlessinger's) deformation functors RDef KSX FMM RDef End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ X/k)) X/k)) [−1] ∼ ∼ RP π0π≤0 Def KSX P RDef X π0π≤0 Def X (cid:16) hoFlagF • RΓ (cid:17) X,Ω∗ X/k π0π≤0 FM / FlagF • H∗(X,k) ∼ ∼ π0π≤0 Def End∗ (H∗ (X,k)) End≥0(H∗ (X,k)) [−1] commutes up to isomorphism; in particular the morphisms RP and FMM are equivalent. Proof. Notice that: • the commutativity of the bottom diagram follows from Theorem 2.16 and Theorem 3.8; • the commutativity (up to isomorphism) of the back diagram corresponds to Theorem 3.8; • the commutativity (up to isomorphism) of the front diagram follows immediately from Remark 3.40, Proposition 3.34.2 and the definitions of the maps P and RP; • the commutativity (up to isomorphism) of the left hand diagram is obtained by combining Theorem 2.2, Theorem 3.39 and Remark 3.40; • the commutativity (up to isomorphism) of the right hand diagram is obtained by combining Corollary 2.20, Proposition 3.33.2 and Proposition 3.34.2. As regards the top diagram, again by Theorem 3.4 it suffices to verify its commutativity up to isomorphism on BDelKSX ; moreover Proposition 3.33.2 and Definition 3.32 say that the derived (cid:17) are homotopy fibres, so deformation functors RDef End∗(RΓ(X,Ω∗ End≥0(RΓ(X,Ω∗ and hoFlagF • RΓ X,Ω∗ X/k)) X/k)) [−1] (cid:16) X/k it is enough to check that the diagrams26 RDef(l) (X,OA,∗)=:X(cid:55)→RΓ(π0X,Ω∗ X/A) / BDelKSX (cid:111)µ RDef X BDel End∗(cid:16)RΓ (cid:16) (cid:16) / RDefRΓ X,RΓ (cid:111) ν (cid:16) (3.23) (cid:17)(cid:17) (cid:17)(cid:17) X,Ω∗ X/k X,Ω∗ X/k 26here maps µ and ν are the morphisms defined in Theorem 3.39 and Lemma 3.30 respectively. 43 / / x x v v / / / / v v u u / / / and BDelKSX (cid:111)µ RDef X RDef(l) BDel (cid:17)(cid:17) X,Ω∗ X/k End≥0(cid:16)RΓ (cid:16) (cid:17)(cid:17) (cid:16) (cid:111) ν X,Ω∗ X/k (X,OA,∗)=:X(cid:55)→(RΓ(π0X,Ω∗ X/A),F •) / RDef (cid:16) F •RΓ X,RΓ (cid:16) (cid:16) ,RΓ X,RΓ X,Ω∗ X/k (cid:17)(cid:17) (3.24) commute up to isomorphism. We are only going to show the commutativity of diagram (3.23), as the commutativity of diagram (3.24) is verified by a similar argument. Let us walk along its arrows: for all A ∈ dgArtk an element x ∈ MCKSX (A) maps through µ to [X → RSpec (A)] -- where X = (X, ROA (x)) -- and in turn this is sent to the complex (cid:17) which is an honest derived deformation over A of the algebraic De Rham complex RΓ ; on the other side, the vector x is sent to the derivation lx and -- proceeding down along map ν -- this determines the complex (cid:17) ≈ RΓ π0X, Ω∗ ROA(x)/A X, Ω∗ X, Ω∗ (3.25) (cid:16) (cid:16) (cid:16) (cid:17) RΓ X/A X/k We claim that complexes (3.25) and (3.26) are quasi-isomorphic: more precisely, we assert that the natural zig-zag (cid:17) (cid:16) X, Ω∗ (cid:16)RΓ (cid:17) ⊗ A, d + lx (cid:1)(cid:1) X ⊗ A(cid:1) , ∂ +(cid:0) ¯∂ + lx (cid:0)Γ(cid:0)X, A ∗,∗ X/k . (3.26) (cid:16) (cid:16)RΓ X, Ω∗ X/k (cid:17) ⊗ A, d + lx (cid:17) (cid:16) RΓ X, Ω∗ ROA(x)/A (cid:17) (3.27) is a chain of quasi-isomorphisms. The right-hand morphism in diagram (3.27) is essentially given by the resolution (2.6). As regards the left-hand one, this is constructed in the following way: consider the standard Dolbeaut resolution A 0,∗ X ←(cid:45) OX and twist it through the derivation lx, so get a map ROA (x) ← OX ⊗ A and hence a morphism Ω∗ X ⊗ A; now just recall that A ∗,∗ X ⊗ A, whose globalisation finally gives us the left-hand map in diagram (3.27). Now denote A 0,∗ X : this provides us with a natural map Ω∗ ROA(x) ← A ∗,∗ X (cid:39) Ω∗ ROA(x)/A ← Ω∗ (cid:1) ROA (x) (n) :=(cid:0)A n,∗ X ⊗ A, ¯∂ + lx X ⊗OX and observe that to show that the zig-zag (3.27) is really a chain of quasi-isomorphisms it suffices to prove that the complexes ROA (x) (n) and ΩnROA(x) are weakly equivalent. As already done in the proof of Theorem 3.39, filter them by powers of the maximal ideal mA, i.e. consider the filtrations27 Fp(cid:16) Fp (ROA (x) (n)) := mp ΩnROA(x) := mp (cid:17) A ROA (x) (n) ⇒ Grp (F) (cid:39) A n,∗ AΩnROA(x) ⇒ Grp (F) (cid:39) Ωn X ⊗ mp ⊗ mp A mp+1 A 0,∗ A A mp+1 X A which kill the twisting lx. Now observe that A 0,∗ ≈ Ωn Ωn OX ≈ A n,∗ X 27There is some abuse of notation in these formulae. 44 / / / o o (cid:17) (cid:1) where the first quasi-isomorphism is induced by the Dolbeaut resolution A 0,∗ X ←(cid:45) OX and the second one is true basically by definition of Ωn A 0,∗ for all m. Finally, look at the induced spectral sequences: we have = H m(cid:0)A n,∗ Ωn OX X X (cid:32) X ⊗ mp A n,∗ A mp+1 A H p+q and H p+q (cid:32) Ωn A 0,∗ X ⊗ mp A mp+1 A (cid:33) (cid:39) (cid:77) (cid:33) (cid:39) (cid:77) i+j=p+q X (cid:32) H i(cid:0)A n,∗ (cid:32) H i(cid:0)A n,∗ X i+j=p+q (cid:32) (cid:33)(cid:33) , so in particular H m(cid:16) (cid:1) ⊗ H j (cid:1) ⊗ H j mp A mp+1 (cid:33)(cid:33) (cid:32) A mp A mp+1 A =⇒ H p+q (ROA (x) (n)) =⇒ H p+q(cid:16) (cid:17) ΩnROA(x) so the complexes ΩnROA(x) and ROA (x) (n) are quasi-isomorphic as their cohomologies are com- puted by the same spectral sequence. Now let us look at diagram (3.23) on the level of morphisms; a gauge element ξ in the Kodaira- Spencer differential graded Lie algebra associated to X maps through RDef (l) to lξ, which in turn induces by ν the morphism of complexes (cid:17) ⊗ A, d + lx1 (cid:17) −→(cid:16)RΓ (cid:16) X, Ω∗ X/k (cid:17) ⊗ A, d + lx2 (cid:17) . RΓ(cid:0)X, elξ(cid:1) : (cid:16) (cid:16)RΓ X, Ω∗ X/k In a similar way, the gauge ξ determines via µ the morphism of complexes eξ : ROA (x1) −→ ROA (x2) (cid:16) (cid:17) RΓ (X, Ω∗ X, Ω∗ ROA(x2)/A which in turn induces through the bottom arrow in diagram (3.23) the morphism X, Ω∗ ROA(x1)/A eξ ) : RΓ (cid:16) (cid:17) −→ RΓ (cid:1)(cid:1) X ⊗ A(cid:1) , ∂ +(cid:0) ¯∂ + lx1 (cid:0)Γ(cid:0)X, A ∗,∗ X ⊗ A(cid:1) , ∂ +(cid:0) ¯∂ + lx2 (cid:0)Γ(cid:0)X, A ∗,∗ (cid:1)(cid:1) Γ(X,elξ ) RΓ X, Ω∗ (cid:16) (cid:16) (cid:17) (cid:17) ROA(x1)/A therefore we end up with a diagram (cid:17) (cid:17) ⊗ A, d + lx1 (cid:17) ⊗ A, d + lx2 (cid:17) Notice that the right hand square of diagram (3.28) commutes because the morphism Γ(cid:0)X, elξ(cid:1) is induced by RΓ(cid:0)X, elξ(cid:1) via the standard Dolbeaut resolution; as regards the left hand square, (cid:16) (cid:16)RΓ (cid:16) (cid:16)RΓ ROA(x2)/A RΓ(X,elξ ) RΓ(X,Ω∗ X, Ω∗ X, Ω∗ X, Ω∗ (3.28) RΓ eξ ) X/k X/k . consider for all n the unglobalised diagram ΩnROA(x1) Ωn eξ Ω∗ ROA(x2) (cid:0)A n,∗ (cid:0)A n,∗ X ⊗ A, ¯∂ + lx1 elξ X ⊗ A, ¯∂ + lx2 (cid:1) (cid:1) (3.29) and again filter all complexes by powers of the maximal ideal mA in order to kill the derivations 45 o o o o o o o o o o o o lx1, lx2 and hence the gauge lξ: we end up with a sequence of commutative diagrams Ωn A 0,∗ X Ωn A 0,∗ X ⊗ mp A mp+1 A ⊗ mp A mp+1 A (cid:16)A n,∗ (cid:16)A n,∗ X ⊗ mp A mp+1 A , ¯∂ X ⊗ mp A mp+1 A , ¯∂ (cid:17) (cid:17) therefore diagram (3.29) has to commute and so does diagram (3.28), as well. This observation completes the proof. 4 The Period Map as a Morphism of Derived Stacks Theorem 3.43 gives the ultimate picture of the local period map as a deformation-theoretic morphism, since it explains how the Fiorenza-Manetti map lifts naturally to the context of Derived Deformation Theory. Anyway, despite being entirely canonical, the Fiorenza-Manetti- Martinengo map28 is still a local morphism: concretely this means that it provides a fully satisfying description of the behaviour of "derived variations of the Hodge structures" associated to some nice k-scheme X with respect to the infinitesimal derived deformations of the scheme itself, but this map is not able to give us any global information, i.e. it does not provide significant relations between the associated global (derived) moduli stacks. In this section we will set a path towards a coherent notion of global derived period map, which will be further studied in [7]. 4.1 A Non-Geometric Global Period Map Fix again X to be a smooth proper scheme over k of dimension d and define the (non- geometric) derived stack DSchX d/k := DSchd/k ×hRPerfk where the map defining the homotopy fibre product in formula (4.1) is (cid:17)(cid:111) X, Ω∗ X/k (cid:16) (cid:110)RΓ (cid:16) DSchd/k −−−−−→ RPerfk Y (cid:55)−→ RΓ π0Y, Ω∗ Y /A (cid:17) (4.1) ≤0 k . for all A ∈ dgAlg Definition 4.1. Define the non-geometric (universal) global period map to be the morphism of derived stacks −−−−−→ DFlag0 (cid:55)−→ (cid:16)RΓ (cid:16) k (cid:16) (cid:16)RΓ (cid:17)(cid:17) (cid:17) X, Ω∗ X/k (cid:17) π0Y, Ω∗ Y /A , F (4.2) RP : (cid:104) Y, θ : RΓ (cid:16) for all A ∈ dgAlg ≤0 k . DSchX (cid:17) ∼→ RΓ d/k π0Y, Ω∗ Y /A (cid:16) X, Ω∗ X/k (cid:17) ⊗ A (cid:105) 28Again, Theorem 3.43 allows us to drop any further adjective. 46 o o o o Proposition 4.2. Consider the diagram of derived stacks and derived deformation functors DSchX d/k RP / / DFlag0 k X, Ω∗ X/k (cid:16) (cid:16)RΓ (cid:17)(cid:17) RDef X RP (cid:16) / hoFlagF RΓ X,Ω∗ X/k (cid:17) where the right-hand-side vertical arrow is the formal neighbourhood inclusion and the left-hand- side one is the composite (cid:32) (cid:33) IdRDefX ,constRΓ(X,Ω∗ X/k) RDef X / RDef X ×h RDefRΓ(X,Ω∗ X/k) (cid:16) (cid:16)RΓ X, Ω∗ X/k (cid:17)(cid:17)f ormal neighbourhood inclusion/ / DSchX d/k. Then the above square is well-defined and commutes. Proof. The fact that the diagram is well-defined is precisely the content of Remark 3.38 and Theorem 3.35.2; the commutativity is readily verified just walking along the arrows, as done in the proof of Theorem 3.43. 4.2 Towards a Derived Analytic Period Mapping Proposition 4.2 is certainly a first step towards a globalisation of Griffiths period map, but it is certainly not sufficiently satisfying since map (4.2) is highly non-geometric; let us outline here some of the ideas that are to be developed in [7] and which should allow us to identify a reasonable global derived period map in the complex-analytic context. Fix k to be the field of complex numbers C and the base scheme X to be smooth and projective; consider a family p : X → S of derived schemes globally deforming it, i.e. a homotopy pull-back diagram X i 2h Spec (C) X p / S (cid:16)RΓ where p is flat and H 0 (p) is smooth. For any point s ∈ S, we can consider the (homotopy) fibre of p at s, which is a derived scheme we will denote as Xs. Unlike the infinitesimal case, the associated filtered (derived) De Rham complex does not determine a point in the homotopy flag variety associated to RΓ (cid:17) , F •(cid:17) , because the group Ω(cid:0)π0S(cid:1) -- where Ω(cid:0)π0S(cid:1) stands for the (simplicial) loop group attached to the topological space π0S -- (cid:17)(cid:17) acts by (higher) monodromy on the fibres of Rp∗ΩX/S and thus on DFlagC . Therefore we would like to define the global period map associated to the family p as a morphism of derived stacks π0Xs, Ω∗ X, Ω∗ (cid:16)RΓ (cid:16) (cid:16) Xs/C X, Ω∗ X/C (cid:16) (cid:17) X/C (cid:55)−→(cid:104)(cid:16)RΓ S −→ DFlagC(RΓ(X,Ω∗ π0Xs, Ω∗ (cid:16) s Xs/S X/C))/Ω(π0S) (cid:17) , F •(cid:17)(cid:105) . (4.3) 47 O O / O O / / / / However it is not clear that the quotient DFlagC(RΓ(X,Ω∗ X/C))/Ω(π0S) exists as a derived geometric stack in the algebraic sense -- and in general we do not expect it to do -- but it is very likely to exist in the analytic setting. As a matter of fact in the underived case the monodromy action on the Grassmannian (or flag variety) is properly discontinuous and this ensures that the quotient has a complex structure; it is then reasonable to expect that the action of Ω(cid:0)π0S(cid:1) on X/C (or a suitable open substack of it recovering the "analytification" of DFlagC the notion of period domain in the derived setting) should be regular enough for the quotient to exist as a derived analytic stack and map (4.3) to be (derived) holomorphic. Thus the ultimate study of the global derived period mapping involves massively the newly-born theory of Derived Analytic Geometry, as well as a good notion of analytification functor in the derived context (see [28], [32], [33], [34], [35] for foundational work). The goal of [7] is to analyse thoroughly the above ideas. (cid:16) (cid:16)RΓ X, Ω∗ (cid:17)(cid:17) Notations and conventions • If i ≥ 0 ∆i is the i-th standard simplicial simplex • diag (−) = diagonal of a bisimplicial set • k = fixed field of characteristic 0, unless otherwise stated • If A is a (possibly differential graded) local Artin ring, mA will be its unique maximal (possibly differential graded) ideal • R = fixed (possibly differential graded) commutative unital k-algebra, unless otherwise stated • If (V ∗, d) is a cochain complex (in some suitable category) then(cid:0)V [n] • If R is a commutative unital ring then R is the constant sheaf of stalk R ∗ (cid:1) will be the , d[n] cochain complex such that V [n]j := V j+n and dj [n] = dj+n X = "Dolbeaut" complex of X X = double complex of " ¯k"-valued forms on X X/k = algebraic De Rham complex of X • Gm = multiplicative group scheme over k • X = smooth proper scheme over k of finite dimension, unless otherwise stated • OX = structure sheaf of X • TX = tangent sheaf of X • A 0,∗ • A ∗,∗ • Ω∗ • F • = Hodge filtration on Ω∗ • LX/k = (absolute) cotangent complex of X over k • Sh (X) = category of sheaves of abelian groups over X • D (X) = derived category of X • ∆ = category of finite ordinal numbers • Algk = category of commutative associative unital algebras over k X/k or cohomology, unless otherwise stated 48 • Affk = category of (linear) affine spaces over k • Artk = category of local Artin algebras over k • Ch≥0 (Vectk) = model category of chain complexes of vector spaces over k in non-negative degrees • Ch≥0 (Affk) model category of (chain) dg≥0-affine spaces over k • DefHin • DefMan • dg≥0Algk = model category of (chain) differential graded commutative algebras over k in k = ∞-category of Hinich derived deformation functors (over k) k = ∞-category of Manetti extended deformation functors (over k) non-negative degrees • dg≥0Catk = model category of (chain) differential graded categories over k • dg≥0CatAff • dg≥0GrpdAff • dgAlg k = ∞-category of affine (chain) differential graded categories over k k = ∞-category of affine (chain) differential graded groupoids over k ≤0 k = model category of (cochain) differential graded commutative algebras over k in non-positive degrees • dgArtk = model category of (cochain) differential graded local Artin algebras over k • dgArt ≤0 k = model category of (cochain) differential graded local Artin algebras over k in non-positive degrees • dgLiek = model category of (cochain) differential graded Lie algebras over k • dgModR = model category of R-modules in (cochain) complexes • dgbNil ≤0 k = ∞-category of bounded below differential graded commutative k-algebras in non-positive degrees such that the canonical map A → H 0 (A) is nilpotent • dgVect ≤0 k = model category of (cochain) differential graded vector spaces over k in non- positive degrees • FdgModR = model category of filtered R-modules in (cochain) complexes • Grpd = 2-category of groupoids • Set = category of sets • Schk = category of schemes over k • sAffk = model category of simplicial affine spaces over k • sAlgk = model category of simplicial commutative associative unital algebras over k • sCat = model category of simplicial categories • sCatk = ∞-category of k-simplicial categories over k • sCatAff • sGrpd = model category of simplicial groupoids • sGrpdAff • sSet = simplicial model category of simplicial sets • sVectk = model category of simplicial vector spaces over k • Vectk = category of vector spaces over k k = ∞-category of affine simplicial categories over k k = ∞-category of affine simplicial groupoids over k 49 References [1] A. M. Cegarra, J. Remedios, The relationship between the diagonal and the bar constructions on a bisimplicial set, Topology and its Applications, 153 (1) (2005): 21-51. [2] P. Deligne, Théorème de Lefschetz et critères de dégénerescence de suites spectrales, Pub- lications mathématiques de l'I.H.É.S., 35 (1968): 107-126. [3] P. Deligne, Théorie de Hodge, II, Publications Mathématiques de l'I.H.É.S., 40 (1971): 5-57. [4] P. Deligne, Théorie de Hodge, III, Publications Mathématiques de l'I.H.É.S., 44 (1974): 5-77. [5] C. Di Natale, An Overview of Derived Deformation Theory, https://www.dpmms.cam.ac.uk/ cd475/GAeL.pdf, 2013. [6] C. Di Natale, Derived Moduli of Complexes and Derived Grassmannians, arXiv : 1407.5900, 2014. [7] C. Di Natale, J. Holstein, A Global Derived Analytic Period Map, [8] D. Fiorenza, M. Manetti, L∞-Algebras, Cartan Homotopies and Period Maps, in preparation. arXiv : math/0605297, 2006. [9] D. Fiorenza, M. Manetti, L∞-Structures on Mapping Cones, Algebra and Number Theory, 1 (2007): 301-330. [10] D. Fiorenza, M. Manetti, A Period Map for Generalised Deformations, Journal of Non- commutative Geometry (3), 4 (2009): 579-597. [11] D. Fiorenza, E. Martinengo, A Short Note on ∞-Groupoids and the Period Map for Pro- jective Manifolds, Publications of the nLab (2), 1 (2012): 1-13. [12] E. Getzler, Lie Theory for Nilpotent L∞-Algebras, Annals of Mathematics (2), 170 (1) (2009): 271-301. [13] P. G. Goerss, K. Schemmerhon, Model Categories and Simplicial Methods, in "Interac- tions Between Homotopy Theory and Algebra": 3-49, Contemporary Mathematics, 436, American Mathematical Society, Providence, RI, 2007. [14] P. G. Goerss, J. F. Jardine, Simplicial Homotopy Theory, Birkäuser, 1999. [15] P. A. Griffiths, Periods of Integrals on Algebraic Manifolds II - Local Study of the Period Mapping, American Journal of Mathematics, 90 (1968): 805-865. [16] V. Hinich, Descent of Deligne Groupoids, International Mathematical Research Notices, 5 (1997): 223-239. [17] V. Hinich, DG Coalgebras as Formal Stacks, Journal of Pure and Applied Algebra, 162 (2001): 209-250. [18] V. Hinich, Deformations of Homotopy Algebras, Communications in Algebra, 32 (2004): 473-494. [19] D. Iacono, Differential Graded Lie Algebras and Deformations of Holomorphic Maps, PhD Thesis, arXiv : math.AG/0701091, 2006. 50 [20] D. Iacono, L∞-Algebras and Deformations of Holomorphic Maps, matical Research Notices, 2008 (2008):article ID:mn013, 36 pages. International Mathe- [21] D. Iacono, M. Manetti, An Algebraic Proof of Bogomolov-Tian-Todorov Theorem, in De- formation Spaces, 39, Vieweg Verlag, 2010. [22] M. Kontsevich, Deformation Quantization of Poisson Manifolds, Letters of Mathematical Physics, 66 (3) (2003): 157-216. [23] M. Kontsevich, Y. Soibelman, Deformations of Algebras over Operads and Deligne's Con- jecture, G. Dito and D. Sternheimer (eds) Conférence Moshé Flato 1999, Vol. I (Dijon 1999), Kluwer Academic Publications, Dordrecht (2000): 255-307. [24] J.-L. Loday, B. Vallette, Algebraic Operads, Grundlehren der Mathematischen Wis- senschaften 346, Springer, Heidelberg, 2012. [25] M. Lieblich, Moduli of Complexes on a Proper Morphism, Journal of Algebraic Geometry, 15 (2006): 175-206. [26] J. Lurie, Derived Algebraic Geometry, PhD thesis, www.math.harvard.edu/ lurie/papers/DAG.pdf, 2004. [27] J. Lurie, Higher Topos Theory, Annals of Mathematics Studies, 2009. [28] J. Geometry Algebraic Derived Lurie, V -- Structured Spaces, [29] J. www.math.harvard.edu/ lurie/papers/DAG-V.pdf, 2011. Derived Algebraic Geometry X -- www.math.harvard.edu/ lurie/papers/DAG-X.pdf, 2011. Lurie, Formal Moduli Problems, [30] M. Manetti, Extended Deformation Functors, International Mathematical Research No- tices, 14 (2002): 719-756. [31] M. Manetti, Lectures on Deformations of Complex Manifolds, Rendiconti di Matematica, VII, 4 (2004):1-183. [32] M. Porta, Derived Complex Analytic Geometry I: GAGA Theorems, arXiv : 1506.09042, 2015. [33] M. Porta, Comparison Results for Derived Deligne-Mumford Stacks, arXiv : 1507.00573, 2015. [34] M. Porta, Derived Complex Analytic Geometry II: Square-Zero Extensions, arXiv : 1507.06602, 2015. [35] M. Porta, T. Yue Yu Higher Analytic Stacks and GAGA Theorems, arXiv : 1412.5166, 2014. [36] J. P. Pridham, Derived Deformations of Artin Stacks, arXiv : 0805.3130, 2008. [37] J. P. Pridham, Unifying Derived Deformation Theories, Advances in Mathematics, 224 (3) (2010): 772-826. [38] J. P. Pridham, Presenting Higher Stacks as Simplicial Schemes, Advances in Mathematics, 238 (2013): 184-245. [39] J. P. Pridham, Representability of Derived Stacks, Journal of K-Theory (10), 2 (2012): 413-453 51 [40] J. P. Pridham, Derived Moduli of Schemes and Sheaves, Journal of K-Theory (10), 1 (2012): 41-85, arXiv : 1011.2742v6. [41] J. P. Pridham, Constructing Derived Stacks, Geometry and Topology, 17 (3) (2013): 1417- 1495. [42] G. Tabuada, Differential Graded Versus Simplicial Categories, Topology and its Applica- tions, 157 (3) (2010): 563-593. [43] B. Toën, Higher and Derived Stacks: a Global Overview, in Algebraic Geometry - Seattle 2005, Proceedings of Symposia in Pure Mathematics, 80, Part 1, American Mathematical Society, Providence, RI, 2009. [44] B. Toën, M. Vaquié, Moduli of Objects in DG-Categories, Annales Scientifiques de l' École Normale Supérieure (4) 40 (2007): 387-444. [45] B. Toën, G. Vezzosi, Homotopical Algebraic Geometry II: Geometric Stacks and Applica- tions, Memoirs of the American Mathematical Society, 193 (902), 2008. [46] M. J. Schlessinger, Functors of Artin Rings, Transactions of the American Mathematical Society, 130 (1968): 208-222. [47] S. Schwede, B. Shipley, Equivalences of Monoidal Model Categories, Algebraic and Geo- metric Topology, 3 (2003): 287-334. [48] B. Vallette, Algebra + Homotopy = Operad, to appear in the Proceedings of the Mathemati- cal Science Research Institute and Research Institute for Mathematical Science "Symplectic Geometry, Noncommutative Geometry and Physics", arXiv : 1202.3245, 2012. [49] C. Voisin, Hodge Theory and Complex Algebraic Geometry Vol. I, Cambridge University Press, 2002. [50] C. Weibel, An Introduction to Homological Algebra, Cambridge University Press, 1994. 52
1201.1646	3	1201	2012-07-16T14:33:02	Classical curves via one-vertex maps	[ "math.AG" ]	One-vertex maps (a type of dessin d'enfant) give a uniform characterization of certain well-known algebraic curves, including those of Klein, Wiman, Accola-Maclachlan and Kulkarni. The characterization depends on a new classification of one-vertex (dually, one-face or unicellular) maps according to the size of the group of map automorphisms. We use an equivalence relation appropriate for studying the faithful action of the absolute Galois group on dessins, although we do not pursue that line of inquiry here.	math.AG	math	Classical curves via one-vertex maps Department of Mathematics and Computer Science Anthony Weaver Bronx Community College City University of New York 212-568-0669 [email protected] May 1, 2012 Abstract One-vertex maps (a type of dessin d'enfant) give a uniform characterization of certain well-known algebraic curves, including those of Klein, Wiman, Accola-Maclachlan and Kulkarni. The characterization depends on a new classification of one-vertex (dually, one-face or unicellular) maps according to the size of the group of map automorphisms. We use an equivalence relation appropriate for studying the faithful action of the absolute Galois group on dessins, although we do not pursue that line of inquiry here. 1 1 Introduction Klein's quartic curve, w3z + z3 + w = 0, admits a tesselation by 336 hyperbolic triangles, and has PSL2(7) as its full group of conformal automorphisms [21]. It is characterized as the curve of smallest genus realizing the upper bound 84(g − 1) on the order of a group of conformal automorphisms of a curve of genus g > 1, given by A. Hurwitz [13] in 1893. Around the same time, A. Wiman characterized the curves w2 = z2g+1 − 1 and w2 = z(z2g − 1), g > 1, as the unique curves of genus g admitting cyclic automorphism groups of largest and second-largest possible order (4g + 2 and 4g, respectively) [36]. Two other curve families, identified much later, share a similar characterization: the Accola-Maclachlan and Kulkarni curves of genus g realize the sharp lower bound 8g + 8 on the maximum order of an automorphism group of a curve of genus g > 1 ([1], [29], [23]). Collectively, we call these curves (together with one other exceptional curve discovered by Wiman) the classical curves. Further details are given in Section 5.2. The triangular tesselation on the Klein quartic is an example of a map, or imbedding of a finite connected graph on a compact oriented surface, so that the complement is a union of simply connected faces. Maps are a special type of dessin d'enfant ("child's drawing") as defined by G. Grothendieck [10]. The topological surface underlying a map has a canonical complex structure, making it a compact Riemann surface, and hence, a complex algebraic curve (see, e.g., [35], [16]). It turns out that any curve with a map (or dessin) admits a Belyı function (a meromorphic function with at most three critical values) and is therefore, by Belyı's theorem ([2], [22], [37]), defined over an algebraic number field. Conversely, any curve defined over a number field admits a map which is geometric in the sense that its edges are geodesics in a canonical complex structure [17]. The absolute Galois group acts faithfully on maps (or dessins) via its action on the coefficients of the equations defining the curves and the formulae for the Belyı functions [18]. In this paper, we show that the classical curves can be uniformly characterized as curves admitting a strictly edge-transitive one-vertex map whose automorphisms are a proper subgroup of the full group of conformal automorphisms of the curve (see Section 2.1 for definitions). The main result is Theorem 6.3. The paper includes three expository sections (Sections 2, 3 and 5) and two sections containing new results (Sections 4 and 6). Section 2 gives the definition of a map and its automorphisms purely combinatorially in terms of permutation groups. Section 3 constructs the canonical complex structure on a surface with 1 2000 Mathematics Subject Classification: Primary 14J50, 14H57, 11G32 1 a map, making the map geometric and its automorphisms conformal. Section 5 discusses conformal group actions on Riemann surfaces, and the question of whether and how a finite cyclic action can extend to the action of a larger group. Details on the classical curves, all of which admit large cyclic automorphism groups, are also given there. In Section 4 we enumerate equivalence classes of one-vertex maps according to the size of the group of map automorphisms, staying within the purely combinatorical framework. In Section 6 we specialize to regular and strictly edge-transitive one-vertex maps, whose automorphism groups are, respectively, maximal and second-maximal (with respect to the number of edges). The classical curves (and some others) arise when the second-maximal group of map automorphisms is not the full group of conformal automorphisms of the curve (Theorem 6.3). One-face or unicellular maps are the duals of one-vertex maps, and have been studied from various points of view. They were enumerated by means of a recurrence relation on the number of maps of a given genus [11], and more recently by a different combinatorial identity [8]. Neither enumeration gives any explicit information about the automorphism groups, as ours does. In [33], D. Singerman characterized Riemann surfaces admitting strictly edge-transitive uniform unicellular dessins, a less general class of maps than the one we focus on. One-face maps of genus 0 (plane trees) have been extensively studied (see, e.g., [18], [25] and references therein), since they are easily visualized, and the action of the absolute Galois group, restricted to these simple dessins, remains faithful. The final paragraph of the paper summarizes the work most closely related to our main result. The recent book [25] is an excellent general reference on dessins; see also [17]. The first complete exposition of the theory of maps on surfaces was given by G.A. Jones and D. Singerman [15], and it remains an excellent reference. The paper [30] describes an analytic approach to maps using quadratic differentials. 2 Maps and permutation groups A map is a finite connected graph embedded on a compact oriented surface so that the complement is a union of simply connected faces. Loops and multiple edges are allowed. A directed edge is called a dart, symbolized by an arrowhead on the edge pointing toward one of the two incident vertices. We usually assume that all edges carry two darts, although it is possible and sometimes useful to relax this assumption. The set of darts pointing toward a given vertex receives a counter-clockwise cyclic ordering from the orientation of the ambient surface. If the map has k edges, and the darts are labelled arbitrarily with the symbols 0, 1, . . . , 2k − 1, the cyclic ordering of the darts at each vertex gives a collection of disjoint cycles in the symmetric group S2k; we define x ∈ S2k to be the product of these cycles. We define y ∈ S2k to be the permutation that interchanges the dart labels on each edge. With the assumption that all edges carry two darts, y is a free involution. The monodromy group GM of a map M is the subgroup hx, yi ≤ S2k. By the connectedness of the underlying graph, GM acts transitively on the darts. Figure 1 shows a map with k = 24 darts imbedded on the torus of modulus eπi/3. (The lower left corner of the parallelogram is 0 ∈ C and the upper left is eπi/3.) There are 8 vertices, 12 edges, and 4 hexagonal faces. Evidently x = (0 1 2) (3 4 5) (6 7 8) (9 10 11) (12 13 14) (15 16 17) (18 19 20) (21 22 23), y = (0 10) (1 17) (2 3) (4 6) (5 13) (7 23) (8 9) (11 19) (12 22) (14 15) (16 18) (20 21) ∈ S24. We will revisit this example throughout the paper. For any map with monodromy group hx, yi, the cycles of (xy)−1 = yx−1 describe closed oriented circuits (paths of darts arranged tip-to-tail) which bound the faces of the map. To see this, start with a dart δ1 pointing toward a given vertex v1. x−1δ1 is the next dart after δ1 in the clockwise cyclic ordering of the darts pointing toward v1. Let δ2 = yx−1δ1, the "reversal" of x−1δ1, pointing along the same edge, toward an adjacent vertex, v2. (It may happen that v2 = v1; in this case the edge carrying δ2 is a loop at v1). Repeat the process staring with δ2, obtaining δ3 = yx−1δ2, pointing toward an adjacent vertex, v3 (possibly v3 = v2 or = v1). After finitely many steps, say, n ≥ 1 of them, δn = δ1, that is, we arrive again at the original dart. At no step does an edge incident with a vertex vi intervene between the edge carrying δi and the edge carrying δi+1. Thus the circuit δ1, δ2, . . . , δn−1 traverses the boundary of a unique face of M, namely, the face which is on the left from the point of a view of a traveller following the circuit. With this convention, each dart belongs to the boundary of a unique face. For the map in Figure 1, yx−1 = (0 3 13 22 20 11) (1 10 8 23 12 15) (2 17 18 21 7 4) (5 6 9 19 16 14) ∈ S24. 2 A bc 23 21 b 22 bc C A bc 14 12 b 13 3 b 5 4 bc B 17 15 b 16 bc D 6 b 8 7 B bc 20 18 b 19 11 9 b 10 bc A C bc 0 b 2 1 bc A Figure 1: A map on a torus. Opposite sides of the parallelogram are identified. The four cycles specify the boundaries of the hexagonal faces centered at C, A, B, D, respectively. The topological genus g of the surface underlying a map is easily determined from the formula for the Euler characteristic, 2g − 2 = {vertices} − {edges} + {faces}, where the number of vertices is the number of cycles in x, and the number of faces is the number of cycles in (xy)−1 (1-cycles are counted). The number of edges is the number of 2-cycles in a free involution on 2k symbols, that is, k. If there is a permutation in S2k which simultaneously conjugates the monodromy generators x1, y1 ∈ GM1 of a map M1 to, respectively, the monodromy generators x2, y2 ∈ GM2 of another map M2, we say that the groups GM1, GM2 are strongly conjugate. Definition 1. Two maps M1, M2 are equivalent if their monodromy groups are strongly conjugate. A map M is (non-trivially) equivalent to itself if and only if there is a permutation in S2k, not contained in GM, which commutes with both monodromy generators and hence centralizes GM. Since conjugate subgroups have conjugate centralizers, the automorphism group of a map, Aut(M) = CentS2k (GM) = {π ∈ S2k πg = gπ for all g ∈ GM}, is well-defined (up to isomorphism) on equivalence classes of maps. The equivalence relation we adopt is appropriate for studying the faithful action of the absolute Galois group on maps: if maps from distinct equivalence classes are in the same Galois orbit, their monodromy groups are conjugate but not strongly conjugate (see, e.g., [18]). We do not pursue this since we focus on maps on surfaces defined over Q, which are fixed by the Galois action. 2.1 Regular and edge-transitive maps Definition 2. A map M has type (n, r), if n is the lcm (least common multiple) of the vertex valencies and r is the lcm of the face valencies. (The valence of a face is the number of darts in its bounding circuit.) The map is uniform if all vertices have valence equal to n and all faces have valence equal to r. The map M is called regular if Aut(M) acts transitively on the set D of darts. This implies that the map is uniform, and in particular, that every dart has the same local incidence relations. The map in Figure 1 is uniform of type (3, 6). It is also regular: Aut(M) ≃ Z6 ⋉ (Z2 ⊕ Z2), where Z2 ⊕ Z2 is the factor group generated by translations carrying A 7→ B and A 7→ C, modulo the normal subgroup generated by their squares. The study of regular maps goes back at least to Klein and Dyck, and perhaps as far as Kepler (see [7], Chapter 8); for more recent work see, e.g., [6], [19]. A map for which Aut(M) is transitive on edges (but not necessarily on darts) is called edge-transitive. In this case there are at most two sets of local incidence relations since a dart might not be equivalent (via an automorphism) to its reversal. Thus there are at most two distinct face-valences, and at most two distinct vertex valences. We call a map which is edge-transitive but not regular strictly edge transitive ("half-regular" in [33]). 3 3 Uniformization of maps and surfaces Let G = GM = hx, yi be the monodromy group of a map M of type (n, r). Let Γ = Γ(n, r) be the group with presentation Γ(n, r) = hξ1, ξ2, ξ3 ξn 1 = ξ2 2 = ξr 3 = ξ1ξ2ξ3 = 1i. There is an obvious surjective homomorphism θ : Γ → G, defined by θ : ξ1 7→ x, ξ2 7→ y, ξ3 7→ (xy)−1. (1) (2) For δ ∈ D, where D is the set of darts of M, let Gδ denote the isotropy subgroup {g ∈ G δg = δ}. Definition 3. The subgroup M = θ−1(Gδ) ≤ Γ, is called the canonical map subgroup for M. Transitivity of the permutation group (G, D) implies all Gδ are conjugate, so M is well-defined (up to conjugacy) independent of δ. The core of M in Γ is the normal subgroup γ−1M γ. M ∗ = \γ∈Γ Γ/M ∗ acts on the set Γ/M of cosets M in Γ, as follows: M ∗γ1 : M γ 7→ M γγ1, γ, γ1 ∈ Γ. One may verify that the action is well-defined, faithful and transitive. Lemma 3.1. The permutation groups (G, D) and (Γ/M ∗, Γ/M ) are isomorphic. Proof. It suffices to verify that the diagram G × D → D Γ iy M ∗ × (cid:12)(cid:12)(cid:12)(cid:12) Γ by M(cid:12)(cid:12)(cid:12)(cid:12) Γ by M(cid:12)(cid:12)(cid:12)(cid:12) → (cid:12)(cid:12)(cid:12)(cid:12) is commutative, where arrows denote group actions, and the maps i and b are defined as follows: i : G → Γ/M ∗ maps g 7→ M ∗θ−1(g); for fixed but arbitrary δ ∈ D, b : D → Γ/M maps b : δg 7→ M θ−1(g), g ∈ G. To show that i is a well-defined isomorphism, one first verifies that M ∗ = ker(θ); then i is simply the inverse of the canonical isomorphism Γ/ker(θ) → G. b is a well-defined bijection according to the following argument, whose steps are reversible: for g1, g2 ∈ G, δg1 = δg2 ⇐⇒ g2g−1 1 ) ⊆ M ⇐⇒ M θ−1(g2)(θ−1(g1))−1 = M ⇐⇒ M θ−1(g1) = M θ−1(g2). 1 ∈ Gδ ⇐⇒ θ−1(g2g−1 3.1 Groups with signature The map subgroup M , and its overgroup Γ, are examples of groups with signature. These groups act properly discontinuously by conformal isometries on one of the three simply connected Riemann surfaces: the Riemann sphere (P1), the complex plane (C), or the Poincaré upper half plane (H). This fact allows us to obtain a canonical complex structure on the topological surface underlying a map. A group Λ = Λ(σ) with signature σ = (h; r1, r2, . . . , rs) has presentation Λ(σ) = ha1, b1, . . . , ah, bh, ξ1, . . . ξs ξr1 1 = · · · = ξrs s = a−1 i b−1 i ab h Yi=1 s Yj=1 ξj = 1i, (3) and acts by conformal isometries on P1, C, or H, depending on whether the number µ(σ) = 2h − 2 + s Xi=1(cid:0)1 − 1 ri(cid:1) is, respectively, negative, 0, or positive. If s > 0, the ri are called the periods of Λ. All periods are > 1, and the number of occurrences of a given period is the number of conjugacy classes of elements of that order in Λ. Signatures which differ only by a permutation of the periods determine isomorphic groups and are considered the same. If s = 0, Λ is a torsion-free surface group, with signature (h; −), isomorphic to the fundamental 4 a π n b π r T c a π n π n eξ1 c aξ2ξ1 e aξ2 Figure 2: Construction of Mn,r, the universal map of type (n, r) group of a surface of genus h. If ∆ is any subgroup of finite index n in Λ(σ), then ∆ is also a group with signature σ′ and the Riemann-Hurwitz relation states that µ(σ′) = nµ(σ). The orbit space U/Λ, where U denotes P1, C, or H as appropriate, is a compact Riemann surface of genus h with s distinguished points, over which the projection U → U/Λ branches. The complex structure on U/Λ is the one which makes the branched covering map U → U/Λ holomorphic. The uniformization theorem of Klein, Poincaré and Koebe states that every compact surface of genus h can be obtained in this way; moreover one may choose the uniformizing group Λ to be a surface group, so that there are no distinguished points, and the covering is regular. When µ(σ) > 0, which is the case except for finitely many signatures, Λ(σ) is a co-compact Fuchsian group (discrete subgroup of PSL(2, R) containing no parabolic elements) and 2πµ(σ) is the hyperbolic area of a fundamental domain for the action of Λ(σ) on H. The normalizer NPSL(2,R)(Λ) of a Fuchsian group Λ is itself a Fuchsian group, containing Λ with finite index. Lemma 3.2. The automorphism group of H/Λ is isomorphic to the finite quotient group NPSL(2,R)(Λ)/Λ. For proofs and further details, see, e.g., [15], [16], [20]. 3.2 The canonical Riemann surface of a map Let M ≤ Γ = Γ(n, r) be the canonical map subgroup of a map M of type (n, r). Since M is a group with signature, the quotient space U/M is a compact Riemann surface, known as the canonical Riemann surface for M. The canonical Riemann surface contains the map M geometrically, that is, the edges of M are geodesics, faces are regular geodesic polygons, and face-centers are well-defined points. To see this, we pass to the universal map Mn,r of type (n, r). Mn,r is a regular map imbedded on one of the simply connected Riemann surfaces U = P1, C, or H. It can be constructed as follows. Let T ∈ U be a geodesic triangle with vertices a, b, c, at which the interior angles are π/n, π/2, π/r respectively (see Figure 2). Reflections in the sides of T generate a discrete group of isometries of U whose sense-preserving subgroup (of index 2) is isomorphic to Γ = Γ(n, r). The generators ξ1, ξ2, ξ3 ∈ Γ correspond to the rotations about a, b, c through angles of, respectively, 2π/n, π, 2π/r; the product of the three rotations (considered as a product of six side reflections) is easily seen to be trivial. Let e be the geodesic segment ab ∪ ξ2(ab), directed toward the vertex a. The map with dart set D = {γ(e) γ ∈ Γ} and vertex set V = {γ(a) γ ∈ Γ} is the universal map Mn,r. Darts meet in sets of n at each vertex and the angle between any two consecutive edges is 2π/n. The faces are regular geodesic r-gons centered at the Γ-images of c. The permutation group (Γ, D) is isomorphic to (Γ, Γ), the right regular representation of Γ on itself. Thus the map is regular and Aut( Mn,r) = Aut(Γ, Γ) = Γ. 1 2 1 r − − 1 n If the number is negative, 0, or positive, Mn,r lies on P1, C, or H, respectively. In the negative case, Mn,r is a central projection of one of the Platonic solids onto the circumscribed sphere; in the positive case, it is one of infinitely many regular tesselations of the hyperbolic plane. In the 0 case, it is a tesselation of C by squares, hexagons, or equilateral triangles. (Figure 1 is a portion of M3,6.) See, e.g., [7]. The orbifold map U → U/M is a local isometry. Hence the image of D in the canonical Riemann surface U/M is a set of directed geodesic segments. These segments are in bijection with the set Γ/M of right cosets 5 b b b b c of M in Γ, and Γ acts by right multiplication on Γ/M . If γ ∈ Γ fixes every coset, it belongs to the core M ∗ of M in Γ. Hence the universal map Mn,r projects to a map Mn,r/M on U/M -- the promised geometric map -- with monodromy group (Γ/M ∗, Γ/M ). With Lemma 3.1, we conclude that every map of type (n, r) is a quotient of the universal map Mn,r, by a subgroup M of finite index in the group Γ = Γ(n, r), and may be assumed to lie on its canonical Riemann surface. The general theory of covering spaces specializes to the category of maps of type (n, r). Thus, if M1 and M2 are two finite maps with canonical map subgroups M1, M2 ≤ Γ = Γ(n, r), then M1 is a finite covering of M2 if and only if M1 is conjugate within Γ to a subgroup of finite index in M2. In particular, if M2 = NΓ(M1), the normalizer of M1 in Γ, then Aut(M1) ≃ M2/M1. This leads to the following Lemma. Lemma 3.3. Let M be a map of type (n, r) with canonical map subgroup M ≤ Γ = Γ(n, r). Then Aut(M) ≤ Aut(X), where X = U/M is the canonical Riemann surface for M, and Aut(X) is the group of conformal automorphisms of X. Proof. Aut(M) is isomorphic to NΓ(M )/M . Since Γ, M and NΓ(M ) are discrete subgroups of Isom+(U), the sense-preserving isometries of U, The equality on the right is a consequence of Lemma 3.2 if U = H. NΓ(M ) M ≤ NIsom+(U )(M ) M = Aut( U /M ). In the local geometry of the canonical Riemann surface, an automorphism α ∈ Aut(M) which preserves a face acts as a rotation about the fixed face-center through an angle 2π/r′, where r′ is a divisor of the face-valence (and a divisor of r if the type is (n, r)). Similarly, if α preserves an edge while transposing its darts, the geometric action is a rotation through an angle π about the fixed edge midpoint. These are consequences of corresponding facts about the universal map Mn,r, together with the fact that the covering projection U → U/M is a local isometry. See [15]. 4 One-vertex maps with automorphisms We now enumerate one-vertex maps, up to equivalence, according to the size of their automorphism groups. Let M be a map with one vertex and k edges. The monodromy generator x ∈ S2k is a 2k-cycle specifying the counterclockwise cyclic ordering of the darts (arbitrarily labelled 0, 1, 2, . . . , 2k − 1) around the vertex. The other monodromy generator, y ∈ S2k, is a free involution whose k disjoint transpositions specify how the darts pair off into edges. Since all 2k-cycles are conjugate in S2k, we may assume, up to map equivalence, that x is the standard 2k-cycle (0 1 2 . . . 2k − 1). (4) Then M is determined by the free involution y, which can be chosen in k−1 Yi=0(cid:18)2k − 2i 2 (cid:19) = (2k − 1) · (2k − 3) · · · · · 3 · 1 = (2k − 1)!! ways. Figure 3 shows three of the 15 one-vertex maps with three edges. The number of equivalence classes of one-vertex maps with k edges is smaller than (2k − 1)!!, since some of the maps have non-trivial automorphisms. Theorem 4.1. Let M be a one-vertex map with k edges and monodromy group GM = hx, yi, where x is a 2k-cycle and and y is a free involution. Then: (i) Aut(M) ≤ hxi ≃ Z2k. (ii) If Aut(M) = hxpi, p a divisor of 2k, the monodromy groups hx, ysi, where ys = x−syxs, s = 0, . . . p−1, determine distinct but equivalent maps. Conversely, if M is equivalent to M′ with monodromy group GM′ = hx, y′i, then Aut(M) ≃ Aut(M′) = hxpi for some divisor p of 2k, and y′ = ys for some s, 0 ≤ s ≤ p − 1. Proof. The centralizer of GM is contained in the centralizer of each of its generators; in particular, it is contained in the centralizer of hxi. The latter is an abelian transitive permutation group which, by a well- known result in the theory of permutation groups, is its own centralizer (see, e.g., [34], Section 10.3). It follows that Aut(M) ≤ hxi ≃ Z2k. Now suppose p is a divisor of 2k and Aut(M) = hxpi. The permutations xs, s = 1, . . . , p−1 do not commute with y (otherwise they would be automorphisms of M), hence there are p distinct 6 free involutions ys = x−syxs, s = 0, 1, . . . , p − 1, belonging to strongly conjugate monodromy groups hx, ysi, determining distinct but equivalent maps. Conversely, if M is equivalent to M′, the monodromy groups GM = hx, yi and GM′ = hx, y′i are strongly conjugate, with conjugate centralizers and hence isomorphic automorphism groups. Since both automorphism groups are contained in hxi, Aut(M) = Aut(M′) = hxpi for some divisor p of 2k. By strong conjugacy of GM and GM′, there exists a permutation σ ∈ S2k, such that σ−1xσ = x and σ−1yσ = y′. In particular σ ∈ CentS2k (hxi) = hxi. If y′ 6= y, σ ∈ hxi \ hxpi. Thus σ = xd, for d a non-multiple of p. If d > p, σ−1yσ = σ′−1yσ′, where σ′ = xd−p, since xp commutes with y. It follows that all possible y′ are obtained by taking σ ∈ {x, x2, . . . , xp−1}. 4.1 Free involutions with prescribed commutation property If x ∈ S2k is a 2k-cycle, and p is a divisor of 2k, then xp is a product of p cycles of length d = 2k/p. Assuming x is the standard 2k-cycle (4), the cycles are Cj = (j j + p . . . j + 2k − p), j = 0, 1, . . . , p − 1. (5) Let cy : S2k → S2k denote conjugation by a free involution y ∈ S2k. If y commutes with xp, then hcyi ≃ Z2 block-permutes the p d-cycles at (5), while preserving their internal cyclic orderings. The permutation cy either preserves a given cycle or interchanges two distinct cycles in a block-orbit of length 2. Suppose cy preserves the cycle Cj, as well as its internal cyclic order. Then for some positive integer t and for every symbol a ∈ Cj, cy(a) = a + tp (mod 2k). Because cy has order 2, 2tp ≡ 0 (mod 2k) or, equivalently, tp ≡ 0 (mod k), which has the unique (mod k) solution t = k/p = d/2, implying that d is even. On the other hand, if cy : S2k → S2k interchanges two cycles Ci and Cj, i < j, preserving their internal cyclic orders, there exists an integer t ≥ 0 such that every symbol i + ap ∈ Ci, is mapped to cy(i + ap) = j + (a + t)p, and, reciprocally, every symbol j + bp ∈ Cj, is mapped to cy(j + bp) = (i + (b − t)p (all symbols are reduced mod 2k). The shift parameter t is not uniquely determined, but may take any value 0, 1, . . . , d − 1. There is no requirement that d be even. It follows that a free involution commuting with xp is uniquely determined by the following data: • the set of cycles Cj fixed by cy (empty if d = 2k/p is odd); • the (possibly empty) set of cycle pairs [Ci, Cj], i < j, transposed by cy; • for each transposed pair, a choice of shift parameter t, 0 ≤ t ≤ d − 1. Let q = ⌊p/2⌋, the greatest integer less than or equal to p/2. If cy transposes m ≤ q cycle pairs, those pairs can be chosen, up to reordering, in 1 m!(cid:18)p 2(cid:19)(cid:18)p − 2 2 (cid:19) . . .(cid:18)p − (2m − 2) 2 2(cid:19)m (cid:19) =(cid:18) 1 p! m!(p − 2m)! ways. Each pair requires a choice of shift parameter from among 0, 1, 2, . . . , d − 1, so the number of choices is multiplied by dm. If d is odd, there are no unpaired cycles and m = p/2 = q, an integer since p is even. If d is even, unpaired cycles are possible so the number of choices is a sum over all possible m. This yields the following Lemma. Lemma 4.2. Let ν p be the number of free involutions in S2k commuting with xp, where x is a 2k-cycle and p a divisor of 2k. Then νp =  where d = 2k/p, and q = ⌊ p 2 ⌋. q 2(cid:19)m Xm=0(cid:18) d 2(cid:19)q p! (cid:18) d q! p! m!(p − 2m)! if d is even if d is odd, The Lemma yields ν2k = (2k − 1)!!, as it should, since every free involution commutes with x2k = {1}, and ν1 = 1, since xk is the unique (free) involution in CentS2k (x) = hxi. 7 (a) (b) (c) 2 3 1 4 0 5 2 3 1 4 0 5 2 3 1 4 0 5 x = (0 1 2 3 4 5) y = (0 1)(2 3)(4 5) yx−1 = (0 4 2) (1) (3) (5) x = (0 1 2 3 4 5) y = (0 3)(1 4)(2 5) yx−1 = (0 2 4) (1 3 5) x = (0 1 2 3 4 5) y = (1 2)(3 4)(5 0) yx−1 = (0) (2) (4) (1 3 5) Figure 3: One-vertex maps with three edges. (a) and (c) are equivalent and strictly edge-transitive; (b) is regular. 4.2 One-vertex maps with prescribed automorphism If y commutes with xp, but no lower power of x, then hxpi is the full automorphism group of a one-vertex map whose monodromy group is strongly conjugate to hx, yi. Theorem 4.3. Let νp be the number of one-vertex maps with k edges whose full automorphism group is hxpi, p a divisor of 2k. Then νp = νp +Ps i=1(−1)iσi, where σ1 = X1≤j≤s νp/pj , σ2 = X1≤j<k≤s νp/pj pk , σ3 = X1≤j<k<l≤s ν p/pj pkpl, . . . , and pi, i = 1, . . . , s, are the prime divisors of p. Proof. Let p′ be any proper divisor of p. If p is a prime, p′ = 1 and ν1 = 1. Hence νp = ν p − 1 = νp − σ1, proving the theorem in this case. If p is not prime, p = p′pǫ1 s , where at least one of the ǫi > 0. Reordering the pi if necessary we may assume that the first t ≤ s of the ǫi > 0, while the remaining s − t are 0. Hence, without loss of generality, p = p′pǫ1 t , with all ǫi > 0, i = 1, 2, . . . , t. Each map with automorphism group hxp′ i contributes 1 to νp; it suffices to show that such a map contributes i=1(−1)iσi, so that its total contribution to the formula in the theorem is 0. We use a standard inclusion/exclusion argument. If t = 1, the map contributes −1 to −σ1 and 0 to the remaining summands. If t = 2, the map contributes −2 to −σ1, 1 to σ2, and 0 to the remaining summands. In general, for arbitrary t ≤ s, the contribution is −( t 2 ) to σ2, etc. It follows from the binomial theorem that the total contribution is −1 toPs 1 ) to −σ1, ( t 1 pǫ2 2 . . . pǫt 1 pǫ2 2 . . . pǫs 1 + t Xi=1 (−1)i(cid:18)t i(cid:19) = (1 − 1)t = 0. Thus the formula counts only those maps whose automorphism group is equal to hxpi. Corollary 4.4. The number of equivalence classes of one-vertex maps with k edges whose full automorphism group is hxpi, p a divisor of 2k, is νp/p. Proof. Theorem 4.1(ii) combined with Theorem 4.3. In particular, the number of equivalence classes of strictly edge-transitive, one-vertex maps with k edges is ν2/2 =(k/2 if k is even; (k − 1)/2 if k is odd. (6) Since ν1 = 1, there is a unique regular one-vertex map with k edges. Figure 3 illustrates the case k = 3. 8 b b b 5 Conformal automorphisms of Riemann surfaces A finite group G acts by conformal automorphisms on a compact Riemann surface X of genus g if and only if there is a group Λ(σ), with signature σ, containing a normal surface subgroup Mg, with signature (g; −), and a surface-kernel epimorphism ρ : Λ → G making the sequence of homomorphisms {1} → Mg ֒→ Λ(σ) ρ → G → {1} (7) exact (see [26], or [16], Theorem 5.9.5). The surface X is conformally equivalent to U/Mg, where U is the universal covering space. G is isomorphic to Λ(σ)/Mg, and the Riemann-Hurwitz relation requires 2g − 2 = Gµ(σ). We say that G acts in genus g with signature σ. Definition 4. A σ-generating vector for a finite group G is an ordered set of elements, one for each generator of Λ(σ), which generate G and fulfill the corresponding relations in Λ(σ) (and possible other relations). A σ-generating vector for G determines a surface kernel epimorphism ρ satisfying (7), and conversely. Hence Lemma 5.1. G acts in genus g with signature σ if and only if G has a σ-generating vector. Let σ = (0; n, m, r), so that Λ(σ) is a triangle group , and let G = Zn. By a general theorem of Maclachlan [28] (see also Harvey [12]), an abelian group of order n has an (0; n, m, r)-generating vector if an only if lcm(m, r) = n. We establish a normal form for such a generating vector. Let Zn = hx xn = 1i, and let hxa, xb, xci be a (0; n, m, r)-generating vector. Then a + b + c ≡ 0 (mod n) since the product of the elements must be the identity. For the elements to have the appropriate orders, n/(a, n) = n, n/(b, n) = m, and n/(c, n) = r, where ( , ) denotes the gcd (greatest common divisor) of two integers. If ρ in (7) is post- composed with an automorphism of G, an equivalent generating vector, specifying a topologically equivalent action, is obtained (see [3], Prop. 2.2). Hence we may assume a = 1. Writing Zn additively, with the generator 1, we define Definition 5. The normal form for an (0; n, m, r)-generating vector for Zn = h{0, 1, . . . , n}, +i is h 1, b, c i, 1 + b + c ≡ 0 (mod n), n/(b, n) = m, n/(c, n) = r. (8) 5.1 Finite maximality and extendability Definition 6. A group with signature is finitely maximal if it is not a subgroup of finite index in any other group with signature. If Λ(σ) in (7) is finitely maximal, then Λ(σ)/Mg is necessarily the full automorphism group of X = U/Mg. Finite maximality is not generally a property of the abstract group Λ(σ), but depends on the particular imbedding of the group in Isom+(U). However, there are some pairs of signatures (σ, σ′) such that every imbedding of a group Λ(σ) with signature σ in Isom+(U) extends to an imbedding of an overgroup Λ(σ′). Signature pairs of this type were considered by Greenberg in [9], and fully classified by Singerman [32]. For each such pair, there arises a delicate problem in finite group theory: for each subgroup-overgroup pair (G, G′), with [G′ : G] = [Λ(σ′) : Λ(σ)], and such that G′ has a σ′-generating vector and G has a σ-generating vector, determine conditions under which the G′-action is an extension of the G-action on the same surface. Equivalently, determine conditions under which the surface-kernel epimorphism ρ in (7) is the restriction of a surface-kernel epimorphism (with the same kernel) ρ′ : Λ(σ′) → G′. This problem is treated comprehensively in the papers [5] and [4]. The signature pairs (σ, σ′) where both signatures are triangular and the first is cyclic-admissable -- that is, Λ(σ) admits surface-kernel epimorphisms onto a cyclic G -- are given in Table 1, which uses the case nomenclature established in [5]. The index [Λ(σ′) : Λ(σ)] = [G′ : G] is easily computed using the Riemann-Hurwitz relation and the fact that Mg is a subgroup of both groups. In cases N6 and N8 in the table, Λ(σ) is a normal subgroup of Λ(σ′). We give complete descriptions of these cases below. The necessary numerical conditions on k are implicit but not stated in [5]. Lemma 5.2 (Case N6). A Zk action with signature (0; k, k, k) has an extension of type N6 to a G action with signature (0; 3, 3, k) only if 32 ∤ k and p ≡ 1 (mod 3) for every prime divisor p 6= 3 of k. The numerical conditions imply that Zk has an automorphism β of order 3, acting non-trivially on each p-Sylow subgroup, p 6= 3. Then G ≃ Z3 ⋉β Zk and the normal form of the generating vector for the Zk action is h1, β(1), β2(1)i. 9 Case N6 N8 T1 T4 T8 T9 T10 σ (0; k, k, k) (0; k, k, u) (0; 7, 7, 7) (0; 8, 8, 4) (0; 4k, 4k, k) (0; 2k, 2k, k) (0; 3k, k, 3) σ′ [Λ(σ′) : Λ(σ)] Conditions (0; 3, 3, k) (0; 2, k, 2u) (0; 2, 3, 7) (0; 2, 3, 8) (0; 2, 3, 4k) (0; 2, 4, 2k) (0; 2, 3, 3k) 3 2 24 12 6 4 4 k ≥ 4 uk, k ≥ 3 - - k ≥ 2 k ≥ 3 k ≥ 3 Table 1: Cyclic-admissible signatures (σ) and possible extensions (σ′) Proof. Since Zk is a normal subgroup, G has the semi-direct product structure Z3 ⋉β Zk, where β is a (possibly trivial) automorphism of Zk of order a divisor of 3. In addition, any non-trivial subgroup of G, being a homomorphic image of the group with signature (0; 3, 3, k), is generated by at most two elements of order 3. Let pG denote a p-Sylow subgroup of G. 3G has the structure Z3 ⋉β Z3s, s ≥ 0, but since it is generated by a most two elements of order 3, s ≤ 1. Hence 32 ∤ k. For a prime divisor p 6= 3 of k, pG is cyclic, and β induces an automorphism βp on pG. The subgroup Z3 ⋉βp (pG) ≤ G is generated by two elements of order 3 only if βp is a non-trivial automorphism of order 3. This implies that p ≡ 1 (mod 3), and yields the p(1) ∈p G stated necessary conditions on k. If 1 is the additive generator of pG, then the element 1 + βp(1) + β2 is fixed by βp and hence must be the identity (equivalently, 1 + βp(1) + β2 p(1) ≡ 0 (mod pG)). Now, Zk is the direct sum of its p-Sylow subgroups; hence β = ⊕pβp, where the sum is over the prime divisors of k. (βp is trivial if p = 3.) It follows that hβi is a subgroup of order 3 in the automorphism group of Zk which acts non-trivially on each p-Sylow subgroup, p 6= 3. Taking 1 as the additive generator of Zk, we have 1 + β(1) + β2(1) ≡ 0 (mod k). Thus h1, β(1), β2(1)i is the normal form of a (0; k, k, k) generating vector for a Zk action extendable to a G action. Lemma 5.3 (Case N8). A Zk action with signature (0; k, k, u) has an extension of type N8 to a G action with signature (0; 2, k, 2u) if and only if either 1. k is odd, G ≃ Z2k, u = k, and the normal form of the Zk generating vector is h1, 1, k − 2i; 2. k is even, G = Z2 ⊕ Zk, u = k/2, and the normal form of the Zk generating vector is h1, 1, k − 2i; 3. k 6= 2, 4, ps, 2ps, (p an odd prime), G ≃ Z2 ⋉α Zk, (non-abelian, non-dihedral), where α is an automorphism of Zk of order 2, α(1) 6= −1, and the normal form of the Zk generating vector is h1, α(1), k − (1 + α(1))i. Proof. The group G has the semi-direct product structure Z2 ⋉α Zk, where α is a (possibly trivial) automor- phism of Zk of order a divisor of 2. If G is non-abelian, α is non-trivial and the generating vector is of the form h1, α(1), ci. α(1) 6= −1, for then c = 0, and u = 1. Such an α exists if and only if k 6= 2, 4, ps, 2ps, (p an odd prime) [14]. G has the (non-dihedral) semi-direct product structure Z2 ⋉α Zk. If G is abelian, α is trivial, and by Maclalchlan's lcm condition [28], either u = k and G ≃ Z2k, or u = k/2 and G = Z2 ⊕ Zk. 5.2 Cyclic actions on the classical curves The classical curves admit automorphism groups which are "large" with respect to the genus g; in particular, the maximal cyclic subgroups of automorphisms have order at least ≥ 2g + 1. The paper [24] gives a precise definition of "large," and has interesting points of contact with our work. (See, e.g., Section 3 of [24], where cyclic group actions with a fixed point are described in terms of side-pairings of hyperbolic polygons; being one-face maps, the polygons can be viewed as the duals of one-vertex maps in which side-pairings correspond to dart-pairings on edges.) The Wiman curves of genus g > 1 are w2 = z2g+1 − 1 w2 = z(z2g − 1) (type I) (type II) w3 = z4 + 1 (g = 3) (type III). 10 Their maximal cyclic automorphism groups are (w, z) 7→ (−w, ze2πi/2g+1) of order 4g + 2 (type I) (w, z) 7→ (weπi/2g, zeπi/g) of order 4g (type II) (w, z) 7→ (we2πi/3, ze2πi/4) of order 4g = 12 (type III). Signatures for the cyclic actions can be deduced by analyzing the ramifications of the branched covering projection Pz : (w, z) 7→ z. For example, for the Wiman curve of type I, Pz : (w, z) 7→ z has degree 4g + 2 with ramification of order 2g + 1 over z = 0, order 4g + 2 over z = ∞, and order 2 over the (2g + 1)st roots of unity. The unique normal form of a generating vector for Z4g+2 with a (0; 4g + 2, 2g + 1, 2) signature is easily seen to be h1, 2g, 2g + 1i. By similar arguments one sees that the signatures of the Type II and III actions are (0; 4g, 4g, 2) and (0; 12, 4, 3), respectively. Corresponding generating vectors, unique in normal form, are given in Table 2. The type I signature is finitely maximal (it does not appear in Table 1), so Z4g+2 is the full automorphism group of the type I curve. The type II and type III signatures are not: the cyclic actions extend to larger conformal actions according to cases N8 and T10 of Table 1, with k = 4g, k = 4, respectively. The full2 automorphism groups are SD8g = ha, b a4g = b2 = 1, b−1ab = a2g−1i acting with signature (0; 2, 4g, 4) (type II) H48 = hKulkarni's group of order 48i acting with signature (0; 2, 3, 12) (type III). A presentation of H48 is given in [24]. In the late 1960's R.D.M. Accola [1] and C. Maclachlan [29] independently showed that the genus g curve with equation w2 = z2g+2 − 1 has full automorphism group AM8g+8 = ha, b a2g+2 = b4 = 1, (ab)2 = [a, b2] = 1i, of order 8g + 8, and that, for infinitely many g ≥ 2, this curve realizes the maximal order of an automorphism group of a surface of genus g. (A. M. Macbeath had shown in 1961 [27] that the Hurwitz bound 84(g − 1) is not attained for infinite sequences of genera.) Thus, for every g ≥ 2, there exists a surface Xg of genus g such that Aut(Xg) ≥ 8g + 8, and the bound is sharp for infinitely many g. If g ≡ −1 (mod 4), there is a second surface of genus g, identified by Kulkarni in 1991 [23], having full automorphism group of order 8g + 8, acting with the same signature as AM8g+8. The full3 automorphism group of the Kulkarni surface is K8g+8 = ha, b a2g+2 = b4 = 1, (ab)2 = 1, b2ab2 = ag+2i. Both AM8g+8 and K8g+8 act with signature (0; 2, 4, 2g + 2) and contain the cyclic subgroup ha a2g+2 = 1i of index 4. The distinct normal forms of the generating vectors for the cyclic actions, given in Table 2, show that the Accola-Maclachlan and Kulkarni curves, when they exist in the same genus, are distinct. The index 4 extensions to AM8g+8 and K8g+8 realize Case T9 in Table 1. The Klein quartic is uniquely determined by the normal form of the generating vector for its maximal cyclic group of automorphisms (Z7) given in Table 2. This action has an index 3 extension (type N6) to the group Z3 ⋉β Z7, β(1) = 2, and a further extension (type T1) to PSL2(7). It is worth noting that there is another Z7 action in genus 3, with the same signature (0; 7, 7, 7) as the one determining the Klein quartic. This action has generating vector h1, 1, 5i and determines the Wiman curve of type I via an N8 extension, with u = k = 2g + 1 = 7. 6 The classical curves and edge-transitive one-vertex maps We have seen that a one-vertex map M with k edges has Aut(M) = hxpi ≤ Z2k = hxi, where p is a divisor of 2k. If M is edge-transitive, Aut(M) must have at least k elements, hence p = 1 or p = 2. If p = 1, M is regular. If p = 2, M is strictly edge transitive, i.e., edge transitive but not regular. The following theorem (stated in terms of unifacial dessins) was proved in [33], and completely charac- terizes regular one-vertex maps. Theorem 6.1. Let M be a regular, one-vertex map on a surface X of genus g > 0. Then X is the Wiman curve of type I and M has 2g + 1 edges, or X is the Wiman curve of type II and M has 2g edges. 2If g = 2, SD16 extends to an action of GL2(Z3), of order 48, with signature (0; 2, 3, 8) (Case T11 in [5]). 3If g = 3, K32 extends to an action by a group of order 96 acting with signature (0; 2, 3, 8) (Case T11 in [5]). 11 Group Z4g+2 Z4g Curve Wiman type I Wiman type II Accola-Maclachlan Z2g+2 Kulkarni Z2g+2 Wiman type III Z12 Klein Z7 Signature (0; 4g + 2, 2g + 1, 2) (0; 4g, 4g, 2) (0; 2g + 2, 2g + 2, g + 1) (0; 2g + 2, 2g + 2, g + 1) (0; 12, 4 , 3) (0; 7, 7, 7) Gen. Vector h1, 2g, 2g + 1i h1, 2g − 1, 2gi h1, 1, 2gi h1, g + 2, g − 1i h1, 3, 8i h1, 2, 4i Table 2: Maximal cyclic actions on the classical curves A B B 1 0 2 5 3 4 A A B A A 1 2 0 3 A A Figure 4: Regular one-vertex maps in genus 1 The regular one-vertex maps in genus 1 are shown in Figure 4. The type I map on the left has two faces, centered at A and B, and is the geometrization of the map in Figure 3 (b). It lies on the elliptic curve of modulus e2πi/3. The type II map on the right has one face, centered at A, and lies on the elliptic curve of modulus i. These curves are the natural genus 1 analogues of the Wiman curves of type I, II , characterized (up to conformal equivalence) by admitting an automorphism of order greater than 2 (6, 4, respectively) which fix a point. Note that the map in Figure 1 also lies on the type I elliptic curve. 6.1 Strictly edge-transitive one-vertex maps If M is strictly edge-transitive with k edges, the edge midpoints are permuted regularly in a k-cycle (with trivial isotropy subgroup) and the face centers fall into two orbits, each having isotropy subgroup of order equal to the corresponding face-valence. By Wiman's bound on the order of an automorphism (or an automorphism with a fixed point if g = 1), k ≤ 4g + 2. k = 4g + 1 is not possible, by the Riemann-Hurwitz relation. Theorem 6.2. Let M be strictly edge-transitive one-vertex map on a surface X of genus g ≥ 1. The action of Aut(M) ≃ Zk has signature and generating vector Zk (0; k, l1, l2) h1, t, k − (t + 1)i, l1 = k/(t, k), l2 = k/(t + 1, k), (9) for some t, 0 < t < k/2, t 6= (k − 1)/2. The notation ( , ) indicates the gcd (greatest common divisor) of two integers. Proof. In the monodromy group GM = hx, yi, the free involution y commutes with x2 but not with x, hence it is obtained from a pairing of the two k-cycles C0 = (0 2 . . . 2k − 2) and C1 = (1 3 . . . 2k − 1) with a shift parameter t, 0 ≤ t ≤ k − 1 (cf. Section 4.1). If k is odd, t = (k − 1)/2 is excluded, since in that case, y = xk and the map would be regular. The permutation (xy)−1 = yx−1 consists of cycles of at most two different lengths. If a cycle of yx−1 contains an odd symbol 2c + 1, then 2c + 1 x−1 7→ 2c y 7→ 2c + 2t + 1, which implies that the cycle has length l1 where l1 is the minimal positive integer for which 2l1t ≡ 0 (mod 2k). It follows that l1 = k/(t, k). Similarly, if a cycle contains an even symbol 2c, then 2c x−1 7→ 2c − 1 y 7→ 2c − 2 − 2t, 12 b b c b c b c b c b c b c b c b c b c b c b which implies that the cycle has length l2, where l2 is the minimal positive integer such that −l2(2 + 2t) ≡ 0 (mod 2k). This yields l2 = k/(t + 1, k). It follows that (xy)−1 is a product of (t, k) cycles of length l1 and (t + 1, k) cycles of length l2. The single vertex of M is fixed by Aut(M). The edge midpoints are permuted regularly in a k-cycle, and the centers of the two face types are permuted in orbits of lengths (t, k) and (t + 1, k), with isotropy subgroups of orders l1 and l2, respectively. It follows that Aut(M) acts with signature (0; k, l1, l2). By Theorem 4.1, hx, yi is strongly conjugate to hx, y1i, where y1 = x−1yx. Moreover y1 is obtained from y by replacing the shift parameter t with the shift parameter k − (t + 1). We note that since (t + 1, k) = (k − (t + 1), k), the signature of the Zk action is not changed if t is replaced by k − (t + 1). It follows that up to map equivalence, we may assume t < k/2. If t = 0, the signature reduces to (0; k, k) which places the map (and the Zk action) on the Riemann sphere. The lcm of l1 and l2 is k, so a (k, l1, l2)-generating vector for Zk exists and may be taken in the normal form h1, t, k − (t + 1)i (cf. (8)). In most cases, if a curve X of genus g > 1 has a strictly edge-transitive, one-vertex map M, then Aut( M) = Aut( X). The exceptions are characterized in our final theorem, below. If AutM = Aut( X), an equation for X is wk = zk/l1 (z − 1)k/l2, where Aut(M) ≃ Zk is generated by (w, z) 7→ (w, e2πi/kz). Theorem 6.3. Let M be a strictly edge-transitive one-vertex map with k edges, where Aut( M) = Zk has signature and generating vector (9). Let X be the canonical Riemann surface of genus g > 1. If Aut( X) > Aut( M), then 1. k = 12, t = 3, and X is the Wiman type III curve; 2. k = 2g + 1, t = 1, and X is the Wiman type I curve; 3. k = 2g + 2, t = 1, and X is the Accola-Maclachlan curve; 4. k = 2g + 1, t = β(1), and Aut(X) ≃ Z3 ⋉β Zk as in Lemma 5.2; except • if k = 7, t = β(1) = 2, X is the Klein quartic. 5. 2g + 2 ≤ k ≤ 4g, t = α(1), and Aut(X) ≃ Z2 ⋉α Zk as in Lemma 5.3; except • if k = 2g + 2, g ≡ −1 (mod 4), α(1) = g + 2, X is the Kulkarni curve; or • if k = 12 or 24, g = 4 or 10, α(1) = 7 or 19 (resp.), Aut(X) contains Z2 ⋉α Zk with index 3. In case 5, with k = 4g and α(1) = 2g − 1, Z2 ⋉α Zk ≃ SD8g and X is the Wiman curve of type II. Proof. If Aut( X) > Aut( M), the signature in (9) must coincide with one of the cyclic-admissible signatures σ in Table 1. For an extension of type N6, by the Riemann-Hurwitz relation, k = 2g + 1; if g = 3, the N6 extension is subsumed by the T1 extension, which yields the Klein quartic. For an extension of type N8, by the Riemann-Hurwitz relation, k = 2g(u/u − 1). For an abelian extension, there are two possibilities: (i) u = k/2 (equivalently, k = 2g + 2) and t = 1; or (ii) u = k (equivalently, k = 2g + 1) and t = 1. In the first case we obtain the Zk action (see Table 2) which determines the Accola-Maclachlan curve. (The N8 extension to Z2 ⊕ Z2g+2 is subsumed by the T9 extension.) In the second case, we obtain the cyclic action which determines the Wiman curve of type I, extending to Z4g+2. For a non-abelian extension of type N8, u ≤ k/2 (otherwise the extension would be cyclic of order 2k), hence 2g + 2 ≤ k ≤ 4g. Both the lower and upper bounds, which occur when l2 = k/2, l2 = 2, respectively, satisfy the necessary conditions on k given in Lemma 5.3 for a nonabelian, non-dihedral extension to Z2 ⋉α Zk. If g ≡ −1 (mod 4), α(1) = g + 2 determines an automorphism of Z2g+2 of order 2 since (g + 2)2 ≡ g2 ≡ 1 (mod 2g + 2). Taking k = 2g + 2, we have k − (1 + α(1)) = g − 1 which has has order g + 1 (mod k) since (g + 1)(g − 1) = (2g + 2)(g − 1)/2 ≡ 0 (mod k). This yields the Zk action which determines the Kulkarni surface. (The N8 extension is subsumed by the T9 extension, or by the T4 extension if g = 3). In the last two exceptional cases (k = 12, 24) the N8 extension is subsumed by the T8 extension. The Wiman type III curve arises from the T10 extension. Subsets of the curves mentioned in Theorem 6.3 have been characterized in ways related to our charac- terization. A. Melekoğlu and D. Singerman ([31], Section 6) characterized the Wiman curves of type I and II, and the Accola-Maclachlan curves (all of which are hyperelliptic) as the unique curves of genus g > 1 admitting double-star maps. These are two-sheeted covers (via the hyperelliptic involution) of a one-vertex map on the sphere having only free edges (carrying a single dart). We note that double-star maps are the "exceptional" ribbon graphs of [30] (Definition 1.9). Melekoğlu and Singerman ([31], Theorems 8.2 and 13 8.3) also characterized the Accola-Maclachlan and Wiman type II curves as the unique Platonic M - and (M − 1)-surfaces, respectively. These are curves of genus g > 1 admitting an anti-conformal involution with the maximal and second-maximal (g + 1, resp., g) number of fixed simple closed curves. D. Singerman ([33], Theorem 6.2) classified the surfaces admitting uniform strictly edge-transitive one-vertex maps; this is the special case l1 = l2 in our Theorem 6.2. References [1] Accola, R.D.M., On the number of automorphisms of a closed Riemann surface, Trans. Am. Math. Soc. 131 (1968) 398-408. [2] G. Belyı, On Galois extensions of a maximal cyclotomic field, Math. USS Izv. 14, 2, (1980) 247-256. [3] S.A. Broughton, Classifying finite group actions on surfaces of low genus, J. Pure and Applied Algebra 69 (1990) 233-270. [4] E. Bujalance, M. Conder, On cyclic groups of automorphisms of Riemann surfaces, J. London Math. Soc. (2) 59 (1999) 573-584. [5] E. Bujalance, F.J. Cirre, M. Conder, On extendability of group actions on compact Riemann surfaces, Trans. Am. Math. Soc. 355 (4) (2002) 1537-1557. [6] Conder, M.D.E, Regular maps and hypermaps of Euler characteristic −1 to 200, J. Combin. Theory Ser. B 99 (2009) 455-459. [7] H.S.M. Coxeter and W.O.J. Moser, Generators and relations for discrete groups, 3rd ed., Springer-Verlag, New York Heidelberg Berlin, 1972. [8] G. Chapuy, A new combinatorial identity for unicellular maps, via a direct bijective approach, Adv. Appl. Mathematics, 47 (4) (2011) 874-893. [9] L. Greenberg, Maximal groups and signatures, in: Discontinuous groups and Riemann surfaces, Ann. of Math. Studies, No. 79 (Princeton, 1974), 207-226. [10] A. Grothendieck, Esquisse d'un programme, in: Geometric Galois Actions, 1, L. Schneps and P. Lochak, eds., LMS Lecture Notes Series 242, Cambridge U. Press, 1997. [11] J. Harer and D. Zagier, The Euler characteristic of the moduli space of curves, Invent. Math., 85 (1986) 457-485. [12] W. Harvey, Cyclic groups of automorphisms of a compact Riemann surface, Quart. J. Math. Oxford (2) 17 (1966) 86-97. [13] A. Hurwitz, Uber algebraische Gebilde mit eindeutigen Transformationen in sich, Math. Ann. 41 (1893) 403-442. [14] Ireland, K., Rosen, M., A classical introduction to modern number theory, Springer, GTM 84, 1982. [15] G.A. Jones and D. Singerman, Theory of maps on orientable surfaces, Proc. London Math. Soc. 3, (1978) 273-307. [16] G.A. Jones and D. Singerman, Complex Functions: an algebraic and geometric viewpoint, Cambridge U. Press, 1987. [17] G.A. Jones and D. Singerman, Belyı functions, hypermaps and Galois groups, Bull. London Math. Soc. 28, no. 6 (1996) 561-590. [18] G.A. Jones and M. Streit, Galois groups, monodromy groups and cartographic groups, in: Geometric Galois Actions, 2, L. Schneps and P. Lochak, eds., LMS Lecture Notes Series 243, Cambridge U. Press, 1997. [19] G.A. Jones, M. Streit and J. Wolfart, Wilson's map operations on regular dessins and cyclotomic fields of definition, Proc. London Math. Soc. 100, (2010) 510-532. 14 [20] S. Katok, Fuchsian groups, Chicago Lectures in Mathematics, U. of Chicago Press, 1992. [21] F. Klein, Uber die Transformationen siebenter Ordnung der elliptischen Funktionen, Math. Ann. 14 (1879) 428-71. [22] B. Koeck, Belyı's theorem revisited, Beiträge zur Algebra und Geometrie 45 (1) (2004) 253-265. [23] Kulkarni, R.S., A note on Wiman and Accola-Maclachlan surfaces, Ann. Acad. Sci. Fenn., Ser. A.I. Mathematica 16 (1991) 83-94. [24] Kulkarni, R.S., Riemann surfaces admitting large automorphism groups, in Extremal Riemann surfaces, Contemporary Mathematics Series 201, J.R. Quine and P. Sarnak, eds. Amer. Math. Soc. (1997) 63-79. [25] S.K. Lando and A.K. Zvonkin, Graphs on surfaces and their applications, Springer-Verlag, 2004. [26] A. M. Macbeath, Discontinuous groups and birational transformations, in: Proceedings of the Summer School in Geometry and Topology, Queens College, Dundee, 1961, 59-75. [27] A.M. Macbeath, On a theorem of Hurwitz, Proc. Glasgow Math. Assoc. 5 (1961) 90-96. [28] C. Maclachlan, , Abelian groups of automorphisms of compact Riemann surfaces, Proc. London Math Soc. (3) 15 (1965) 669-712. [29] C. Maclachlan, A bound for the number of automorphisms of a compact Riemann surface, J. London Math. Soc. 44 (1969) 265-272. [30] M. Mulase and M. Penkava, Ribbon graphs, quadratic differentials on Riemann surfaces, and algebraic curves defined over Q ∗, Asian J. Math. 2 (4) (1998) 875-920. [31] A. Melekoğlu and D. Singerman, Reflections of regular maps and Riemann surfaces, Rev. Mat. Iberoamer- icana 24 (2008), 3, 921-939. [32] D. Singerman, Finitely maximal Fuchsian groups, J. London Math. Soc. (2) 6 (1972) 29-38. [33] D. Singerman, Unicellular dessins and a uniqueness theorem for Klein's Riemann surface of genus 3, Bull. London Math. Soc. 33 (2001) 701-710. [34] W.R. Scott, Group Theory, Dover Publications, New York, 1987. [35] G. Springer, Introduction to Riemann surfaces, Addison-Wesley, Reading, Massachusetts, 1957. [36] A. Wiman, Über die Hyperelliptischen Curven und diejenigen von Geschlechte p = 3 welche eindeutige Transformationen in sich zulassen, Bighang Till K. Svenska Vet.-Akad. Handlingar (Stockholm 1895-6) bd. 21 1-23. [37] J. Wolfart, The 'obvious' part of Belyı's theorem and Riemann surfaces with many automorphisms, in: Geometric Galois Actions, 1, L. Schneps and P. Lochak, eds., LMS Lecture Notes Series 242, Cambridge U. Press, 1997. 15
1301.1822	1	1301	2013-01-09T12:00:26	Reflexive and spanned sheaves on $\mathbb{P}^3$	[ "math.AG" ]	We investigate the reflexive sheaves on $\PP^3$ spanned in codimension 2 with very low first Chern class $c_1$. We also give the sufficient and necessary conditions on numeric data of such sheaves for indecomposabiity. As a by-product we obtain that every reflexive sheaf on $\PP^3$ spanned in codimension 2 with $c_1=2$ is spanned.	math.AG	math	REFLEXIVE AND SPANNED SHEAVES ON P3 E. BALLICO, S. HUH AND F. MALASPINA Abstract. We investigate the reflexive sheaves on P3 spanned in codi- mension 2 with very low first Chern class c1. We also give the sufficient and necessary conditions on numeric data of such sheaves for indecom- posabiity. As a by-product we obtain that every reflexive sheaf on P3 spanned in codimension 2 with c1 = 2 is spanned. 1. Introduction Reflexive sheaves play an important role in algebraic geometry as a nat- ural generalization of vector bundles and the investigation on them over projective spaces was deeply initiated by Hartshorne in [13]. One of the basic questions on vector bundles is their globally generatedness and their classification problem for lower first Chern classes was recently answered by several authors in many ways [2][14][9][11][16][17]. The authors of this arti- cle investigated the globally generated vector bundles on a smooth quadric threefold and give a numerical criterion for indecomposability [5][6]. So the natural question is on the classification of reflexive sheaves on projective spaces with lower first Chern class with certain extra conditions, for example spannedness in codimension 2. In this paper we investigate the existence of a rank r reflexive sheaf on the projective variety Pn with n ≥ 3 and very low first Chern class c1 = c1(F) that is spanned in codimension 2, i.e. such that the evaluation map eF : H 0(F) ⊗ OX −→ F is surjective outside finitely many subvariety of codimension at least 3. Our main ingredient in this article is the sheaf Fk,n,L defined by Fk,n,L := coker( j : OPn(−1) −→ O⊕(k+1) Pn ) where the map j is given by k + 1 linearly independent sections of OPn(1) with a linear subspace L ⊂ Pn as their common zero locus. It is a torsion- free sheaf on Pn with rank k and singular locus L (see Section 3). When L = {P } is a single point space, then we will simply denote it by Fk,n,P . The main result of this article is the following theorem: 1991 Mathematics Subject Classification. Primary: 14F99; Secondary: 14J99. Key words and phrases. reflexive sheaf, spanned, projective space. The second author is supported by Basic Science Research Program 2012-0002904 through NRF funded by MEST. 1 2 E. BALLICO, S. HUH AND F. MALASPINA Theorem 1.1. There exists an indecomposable and non-locally free reflexive sheaf F of rank r ≥ 2 on P3 with c1 ≤ 2 and spanned in codimension 2 if and only if (c1, c2, r) is one of the followings: (1, 1, 2) ; F ∼= F2,3,P with P ∈ P3, (2, 2, 2) ; F ∈ A, a 6-dimensional irreducible variety, (2, 3, r) ; r ∈ {2, 3, 5} and (2, 4, r) ; r ∈ {2, 3, 4, 5, 6, 7, 8}. The exceptional case is (c1, c2, r) = (2, 3, 4) when there are only F2,3,P ⊕ F2,3,P ′ with P, P ′ ∈ P3. Indeed, we do not only give a numerical criterion for the existence of indecomposable and non-locally free sheaves, but also give a complete de- scription of reflexive sheaves spanned in codimension 2 including the locally free sheaves. In particular, we obtain that there is no indecomposable reflex- ive sheaf of rank higher than 7 on P3 with c1 ≤ 2 and spanned in codimension 2. As an automatic consequence, we obtain the following result : Corollary 1.2. Let F be a reflexive sheaf on P3 with c1(F) = 2 and dim V (F) ≤ 0. Then we have V (F) = ∅, i.e. F is spanned. Based on the classification on P3, we also prove that every vector bundle on Pn, n ≥ 3 with c1 = 2 and spanned in codimension 2, is also spanned everywhere (see Section 5). 2. Preliminaries Let F be a reflexive sheaf of rank r on a projective space Pn. For our technical reason, we define two subvaieties associated to F : Sing(F) := {P ∈ Pn F is not locally free at P } V (F) := {P ∈ Pn eF is not surjective at P } where eF : H 0(F) ⊗ OPn −→ F is the evaluation map of F. Note that the codimension of Sing(F) is at least 3 since F is reflexive. In particular, every reflexive sheaf on P2 is locally free. We also get that F is a line bundle when r = 1 due to Proposition 1.9 in [13]. Thus let us assume that n ≥ 3 and r ≥ 2. Set c1 = c1(F) ∈ Pic(Pn) and then we can identify c1 with an integer, i.e. c1 ∈ Z. Moreover we assume that V (F) has codimension at least 3 in Pn, which implies that c1 ≥ 0. Taking r − 1 general sections of F, we get an inclusion j : O⊕(r−1) Lemma 2.1. J has no torsion and J ∼= IC(c1) for some closed subscheme C ⊂ Pn with codimension at least 2. −→ F and let us set J := coker(j). Pn Proof. The definition of J gives the exact sequence (1) 0 −→ O⊕(r−1) Pn j −→ F u −→ J −→ 0. REFLEXIVE SHEAVES 3 Pn Since F is reflexive, we have depth(FP ) ≥ 2 for every P ∈ Pn. Since the depth of OPn at P ∈ Pn is at least 2, the sequence (1) gives that depth(JP ) ≥ 1 at each point of Pn. Let τ ⊂ J be the torsion of J and set A := u−1(τ ). J has no torsion if and only if A = j(O⊕(r−1) ). Fix a general complete intersection smooth curve C ⊂ Pn, e.g. take as C a general line. For such a general curve C, we have C ∩ V (F) = Sing(FC ) = ∅. Let VC be the image of H 0(F) by the restriction map H 0(F) −→ H 0(C, FC ). By the choice of C, VC spans the vector bundle FC . We fix C and then take r − 1 general sections of F to get u. For general sections, we get that j(O⊕(r−1) ) is a subbundle of FC . Hence τ ∩ C = ∅. Since C is general, we get that τ has support of codimension at least 2, i.e. A/j(O⊕(r−1) ) has support of codimension at least 2. Since j(O⊕(r−1) ) is reflexive and A has no torsion, so we get A = j(O⊕(r−1) Pn C Pn ). Pn Hence J has no torsion, i.e. the natural map J −→ J ∨∨ is injective. Since J ∨∨ has rank 1 and it is reflexive, it is a line bundle by Proposition 1.9 in [13]. Hence J ∼= IC(c1) for some closed subscheme C ⊂ Pn with codimension at least 2. (cid:3) Lemma 2.1 gives an exact sequence for a reflexive sheaf F : (2) where C is a subscheme of Pn with codimension at least 2. −→ F −→ IC(c1) −→ 0 Pn 0 −→ O⊕(r−1) Lemma 2.2. The subscheme C is locally Cohen-Macaulay with no embed- ded component such that Cred has pure codimension 2. Moreover, we have Sing(C) ⊆ V (F) ∪ Sing(F). If n ≥ 3, then C is reduced. Proof. From the sequence (2), the sheaf IC has projective dimension at least 2. Since Pn is a regular scheme, the Serre formula gives that OC has depth at least n − 2. Since Cred contains no hypersurface, the local ring OC,P also has depth at least n − 2 for each P ∈ Cred. Since OC,P has dimension at most n − 2, it must be a Cohen-Macaulay local ring of dimension n − 2. Hence C has pure dimension n − 2 and no embedded component. A form of the Bertini theorem says that C is smooth outside V (F) ∪ Sing(F) and a subscheme of codimension ≥ 4 ([8] and Part (F) at page 4 in [2]). If n ≥ 3, we get that C is reduced because it has pure dimension n − 2 and no embedded component. (cid:3) When n = 3 and certain cohomology vanishing is true, then the converse also holds. In other words, given a locally Cohen-Macaulay curve C ⊂ P3 and r − 1 sections of ωC(c1(P3) − c1) ≃ ωC(4 − c1) spanning it, then we get a unique reflexive sheaf fitting in an exact sequence (2) ([13], Theorem 4.1 and its generalizations to higher ranks in [4]). Remark 2.3. Let F be any sheaf fitting into the sequence (2). Then F has no torsion and C ⊇ Sing(F). If F is spanned outside finitely many points, then IC(c1) is spanned outside finitely many points. Since h1(OPn) = 0, we 4 E. BALLICO, S. HUH AND F. MALASPINA have V (F) = V (IC(c1)). In other words, F is spanned outside an algebraic set V if and only if IC(c1) is spanned outside V . Although C depends on the choice of r − 1 sections of F, the set V (IC(c1)) depends only on F and hence it may be used to check if two sheaves are isomorphic. Lemma 2.4. Let F be a coherent sheaf of rank r on Pn with c1 = c1(F). (1) For a line L ⊂ Pn with L * Sing(F), FL is the direct sum of a torsion sheaf τ and a vector bundle E on L with rank(E) = r. (2) We have c1 = deg(E) + deg(τ ) and deg(τ ) ≥ ♯(L ∩ Sing(F)). If L * V (F), then we have deg(E) ≥ 0. Proof. Note that every coherent sheaf on a smooth curve is isomorphic to the direct sum of its torsion part and a locally free sheaf. Hence FL is the direct sum of a torsion sheaf τ and a vector bundle E with rank(E) = r. Let us fix a general hyperplane H ⊂ Pn containing L such that H + Sing(F). Let χF (t) and χFH (t) denote the Hilbert polynomials of F and FH, respectively, and then from the exact sequence 0 −→ F(t − 1) −→ F(t) −→ F(t)H −→ 0 (3) we get χFH (t) = χF (t) − χF (t − 1). Since L * Sing(F), we have dim(H ∩ Sing(F)) = dim(Sing(F)) − 1. If n = 2, the we get χF (t) = rt + r + c1. If n > 2 we get r = rank(FH ) and c1 = c1(FH ). Then we use induction on n and the Riemann-Roch formula on L to get χ(FL) = rt + r + c1. For a point P ∈ Sing(F) ∩ L, let FP (resp. FP,L) be the germ of the sheaf F (resp. FL) at P . We also let m (resp. mL) be the maximal ideal of the local ring OPn,P (resp. OL,P ). Since the tensor product is a right exact functor, we get that the vector space FP /mFP has dimension > r and that this is the dimension of the vector space FP,L/mLFP,L. The latter dimension is greater than r if and only if P is in the support of τ . If L * V (F), then E is generically spanned and hence deg(E) ≥ 0. Hence we get the second part of the lemma. (cid:3) Remark 2.5. Assume c1 > 0 and so det(F) is not trivial. Then the rank of any map O⊕r X −→ F drops on a set of codimension at least 1. Thus if V (F) has codimension at least 2, then we have h0(IC(c1)) ≥ 2. Lemma 2.6. Let E be a vector bundle of rank r on Pn with c1(E) = c1 such that V (E) 6= Pn. For a line L * V (E), we have c1 ≥ ♯(L ∩ V (F)). Proof. Let V ⊆ H 0(L, EL) denote the image of the restriction map H 0(E) −→ H 0(L, EL). Use that EL has degree c1 and that the cokernel of the evalua- tion map V ⊗ OL −→ EL has cokernel supported at the points L∩ V (F). (cid:3) 3. Reflexive sheaves with c1 = 1 Let us fix two integers n and k such that 2 ≤ k ≤ n. From k + 1 linearly independent forms u1, . . . , uk+1 on Pn, we can construct an injective map of sheaves j : OPn(−1) −→ O⊕(k+1) . Let L be an (n − k − 1)-dimensional linear Pn REFLEXIVE SHEAVES 5 subspace of Pn defined by {u1 = · · · = uk+1 = 0} and then the cokernel of j only depends on the choice of L, i.e. it does not depend on the choice of the forms u1, . . . , uk+1 defining L. Thus we can denote it by Fk,n,L. (4) Fk,n,L := coker( j : OPn(−1) −→ O⊕(k+1) Pn ). For simplicity, we often write Fn−1,n,P instead of Fn−1,n,{P } if L = {P } is a single point space. Remark 3.1. Let L′ ⊂ Pn be another linear subspace of Pn with codimen- sion k. Then there is a map ϕ ∈ Aut(Pn) such that ϕ(L) = L′ and thus we have ϕ∗(Fk,n,L′) ∼= Fk,n,L, i.e. Fk,n,L and Fk,n,L′ are projectively equivalent. As an example, we have Fn,n,∅ = T Pn(−1) from the Euler sequence. By definition, Fk,n,L is a spanned sheaf of rank k. Since h0(Fk,n,L) = k + 1, we have h0(Fk,n,L ⊕ O⊕(r−k) ) = r + 1 for all r ≥ k. Pn Lemma 3.2. The sheaf Fk,n,L has no torsion and Sing(Fk,n,L) = L. More- over Fk,n,L is reflexive if and only if k ≥ 2. Proof. At each point P ∈ Pn \ L, the map j defining Fk,n,L is a vector bundle embedding and so we have Sing(Fk,n,L) ⊆ L and Fk,n,L has rank k. Conversely, the rank of the map j drops at P ∈ L and so Fk,n,L is not locally free at P . Thus we have Sing(Fk,n,L) = L. In particular, Fn,n,∅ is locally free and we have Fn,n,∅ ∼= T Pn(−1) from the Euler sequence of Pn. Let T be the torsion of Fk,n,L and let π : O⊕(k+1) −→ Fk,n,L/T denote the obvious quotient map. Since O⊕(k+1) is locally free and Fk,n,L/T has no torsion, so the sheaf ker(π) is reflexive ([13], Corollary 1.5). Since ker(π) has rank 1, it is a line bundle ([13], Proposition 1.9). Since T is supported in L, the inclusion OPn −→ ker(π) is the identity map outside a subspace of codimension ≥ 2. Thus this inclusion is the identity map everywhere and in particular we have T = 0, i.e. Fk,n,L has no torsion. Pn Pn Thus if Fk,n,L is reflexive, then we have k ≥ 2 ([15], Corollary II.1.1.10). Conversely, let us assume that k ≥ 2. Since Fk,n,L has no torsion, the natural map i : Fk,n,L −→ F ∨∨ k,n,L is injective. Set ε := coker(i) and then Fk,n,L is reflexive if and only if ε = 0. Assume ε 6= ∅. Since Sing(Fk,n,L) = L, the support Supp(ε) of ε is contained in L. Since ϕ∗(Fk,n,L) ∼= Fk,n,L for every ϕ ∈ Aut(Pn) such that ϕ(L) = L, we get Supp(ε) = L. The reflexive sheaf F ∨∨ k,n,L is spanned outside L, i.e. outside a subset of codimension at least 3. By Proposition 3.3, we have either F ∨∨ or F ∨∨ M ⊂ Pn with codimension k + 1. for some l ∈ {3, . . . , k} and some linear subspace ∼= OPn(1) ⊕ O⊕(k−1) ∼= Fl,n,M ⊕ O⊕(k−l) k,n,L k,n,L Pn Pn First assume F ∨∨ k,n,L is spanned k,n,L) = k + 1. Since h0(Fk,n,L) = k + 1 and Fk,n,L is a proper . We get that F ∨∨ ∼= Fl,n,M ⊕ O⊕(k−l) k,n,L Pn and h0(F ∨∨ subsheaf of F ∨∨ k,n,L, we get a contradiction. 6 E. BALLICO, S. HUH AND F. MALASPINA Now assume that F ∨∨ k,n,L ∼= OPn(1)⊕O⊕(k−1) Pn and then we have an inclusion Pn Pn Pn Pn ∼= J ⊕ O⊕(k−1) /J . Since h0(Fk,n,L) = k + 1, we get J ∼= IL(1) and so Fk,n,L i1 : Fk,n,L ֒→ OPn(1) ⊕ O⊕(k−1) with ε isomorphic to coker(i1). Since Fk,n,L is spanned, it implies that Fk,n,L with L as the support of O⊕(k−1) ∼= IL(1)⊕O⊕(k−1) . For a fixed point P ∈ L, let mP be the maximal ideal of the local ring OPn,P . Since the tensor product is a right exact functor and Fk,n,L is a quotient of O⊕(k+1) , the vector space Fk,n,L/mP Fk,n,L has dimension k + 1. But the vector space (OPn(1) ⊕ O⊕(k−1) ) has dimension (n + 1) + (k − 1) > k + 1 and it is a contradiction. (cid:3) Proposition 3.3. A reflexive sheaf F of rank r ≥ 3 on Pn, n ≥ 3 with c1 = 1 has codim V (F) ≥ 3 if and only if F is isomorphic to either )/mP (OPn(1) ⊕ O⊕(k−1) Pn Pn Pn (1) OPn(1) ⊕ O⊕(r−1) (2) Fk,n,L ⊕ O⊕(r−k) Pn Pn , or for some integer k such that 2 ≤ k ≤ min{n, r} and an (n − k − 1)-dimensional linear subspace L ⊂ Pn. . Pn Proof. Let us take r − 1 general sections of F to have an exact sequence (2). If C = ∅, we have F ∼= OPn(1) ⊕ O⊕(r−1) Let us assume that C 6= ∅ and let M denote the linear span of C in Pn, the intersection of all linear subspaces containing the scheme C. By convention, we set M = Pn if there is no such a linear subspaces. Since C has pure codimension 2 by Lemma 2.2 and IC(1) is spanned outside V (F), we have h0(IC(1)) ≥ 2. Since H 0(IC(1)) = H 0(IM (1)), we get dim(M ) ≤ n − 2. So the only possibility is that dim(M ) = n − 2 and C = M . Thus the scheme C is a linear subspace of codimension 2 and h0(F) = r + 1. Since IC(1) is spanned and h1(OPn) = 0, so F is spanned. Since F is torsion-free and O⊕(r+1) is locally free, the sheaf ker(eF ) is a reflexive sheaf of rank 1 ([13], Proposition 1.1). Thus ker(eF ) is a line bundle ([13], Proposition 1.9) and F fits into an exact sequence Pn ϕ −→ O⊕(r+1) Pn Pn −→ F −→ 0 0 −→ OPn(−1) (5) with ϕ = (u1, . . . , ur+1) and each ui ∈ H 0(OPn(1)). Up to an automorphism of O⊕(r+1) , we may assume that u1, . . . , uk+1 are linearly independent and ui = 0 for all i > k + 1. Hence k is a positive integer with k ≤ min{n, r}. Hence O⊕(r−k) ⊕ F ′ for some reflexive sheaf F ′ of rank k. Let L be the common zero-locus of u1, . . . , uk+1 which is an (n − k − 1)-dimensional subspace of Pn. Then we have V (F) = L and F ′ ∼= Fk,n,L. Now by Lemma 3.2, F is reflexive if and only if k ≥ 2. (cid:3) As an automatic consequence of this proof, we get that F is locally free is a direct factor of F, i.e. F ∼= O⊕(r−k) Pn Pn if and only if F ∼= T Pn(−1) ⊕ O⊕(r−n) Remark 3.4. About the cohomology groups of Fk,n,L, we can observe the followings: with r ≥ n. Pn REFLEXIVE SHEAVES 7 (1) From (5) we get hi(Fk,n,L(t)) = 0 for all t ∈ Z and all i ∈ {1, . . . , n − 2} and hn−1(Fk,n,L(t)) = 0 for all t ≥ −n + 1. (2) Fix integers n, k such that n ≥ 3 and 2 ≤ k ≤ n − 1. Let L ⊂ Pn be an (n − k − 1)-dimensional linear subspace of Pn and let u1, . . . , uk+1 be linear forms such that L = {u1 = · · · = uk+1 = 0}. Then there is a family of maps ϕλ : OPn(−1) −→ O⊕(n+1) , λ ∈ ∆, ∆ a smooth and connected affine curve, and o ∈ ∆ such that ϕo = (u1, . . . , uk+1, 0, . . . , 0), while ϕλ is induced by the choice of a basis of H 0(OPn(1)) for all λ ∈ ∆ \ {o}. We get that Fk,n,L ⊕ O⊕(n−k) is the flat limit of a family of vector bundles isomorphic to T Pn(−1). Hence Fk,n,L and T Pn(−1) have the same Segre classes. By the semicontinuity theorem for cohomology ([12], III.12.8) we get hn−1(Fk,n,L(−n)) > 0. Since (5) gives hn−1(Fk,n,L(−n)) ≤ 1, we get hn−1(Fk,n,L(−n)) = 1. Pn Pn As an example, let us take a look at F2,3,P with P a point in P3 for our later use. The sheaf F2,3,P is a spanned sheaf of rank 2 with h0(F2,3,P (−1)) = 0 and h0(F2,3,P ) = 3. By Remark 2.5.1 in [13], we have c3(F2,3,P ) = 1. Lemma 3.5. Some cohomological properties of F2,3,P are as follows: (1) h0(Ext1(F2,3,P , OP3(r)) = 1 for each r ∈ Z. (2) h1(F2,3,P (t)) = h1(F ∨ (3) h2(F2,3,P (t)) ≤ 1 and dim(Ext1(F2,3,P , OP3(t))) ≤ 1 for all t ∈ Z. (4) h2(F2,3,P (−4)) = dim(Ext1(F2,3,P , OP3 )) = 1. (5) dim(Ext1(F2,3,P , T P3(−1)) ≥ 3. 2,3,P (t)) = 0 for all t ∈ Z. Proof. P and thus we have (1) Note that Ext1(F2,3,P , OP3 ) is a skyscraper sheaf supported by h0(Ext1(F2,3,P , OP3)(r)) = h0(Ext1(F2,3,P , OP3)(s)) for all r, s ∈ Z. The number is c3(F2,3,P ) = 1 by Proposition 2.6 in [13]. (2) Now from the Euler sequence 0 −→ OP3(−1) −→ O⊕4 P3 −→ T P3(−1) −→ 0 we get h1(F2,3,P (t)) = 0 for all t ∈ Z. Since F ∨ Proposition 1.10), we get h1(F ∨ 2,3,P (t)) = 0 for all t ∈ Z. 2,3,P ∼= F2,3,P (−1) ([13], (3),(4) The Serre duality gives h2(F2,3,P (t)) = dim(Ext1(F2,3,P , OP3(−4− t))) for all t ∈ Z. Since h0(Ext1(F2,3,P , OP3 )(r)) = 1 for each r ∈ Z, the last part of Theorem 2.5 in [12], gives h2(F2,3,P (t)) ≤ 1 and that h2(F2,3,P (t)) = 1 if and only if h2(F ∨ if and only if h2(F2,3,P (−t − 5)) = 0. Taking t = −4 we get h2(F2,3,P (−4)) = 1 by Remark 3.4. 2,3,P (−t − 4)) = 0, i.e. (5) Taking Hom(F2,3,P , ·) functor to the Euler sequence, we obtain the exact sequence Ext1(F2,3,P , OP3(−1)) −→ Ext1(F2,3,P , O⊕4 P3 ) −→ Ext1(F2,3,P , T P3(−1)). 8 E. BALLICO, S. HUH AND F. MALASPINA (cid:3) Use dim(Ext1(F2,3,P , OP3(−1))) ≤ 1 and dim(Ext1(F2,3,P , OP3)) = 1. Remark 3.6. Fix an integer r ∈ {2, . . . , n − 1} and any F = Fn−1,n,P where P is a point in Pn. Let R ⊂ Pn be a line. If P /∈ R, then FR is a spanned vector bundle of rank n−1 with degree 1 and thus FR has splitting type (1, 0, · · · , 0). If P ∈ R, then FR is a direct sum of a non-zero torsion sheaf τ supported by P and a spanned rank r vector bundle E on R with deg(E) = 1 − deg(τ ) by Lemma 2.4. Hence deg(τ ) = 1, i.e. τ = OP and E ∼= O⊕r R . In general, let us consider an arbitrary F = Fk,n,L. If R ⊂ Pn is a line such that R ∩ L = ∅, then the vector bundle FR has splitting type (1, 0, . . . , 0). If R ∩ L is a point, say P , then FR is isomorphic to the direct sum of OP and a trivial vector bundle O⊕r R . 4. Reflexive sheaves on P3 with c1 = 2 The main goal of this section is to prove the following result. Proposition 4.1. Let F be an indecomposable and reflexive sheaf of rank r on P3 with c1(F) = 2 and dim V (F) ≤ 0. Then we have r ≤ 8, c2(F) ≤ 4 and ♯(Sing(F)) ≤ 8. In fact we investigate each cases in more detail and give complete descrip- tion in some cases. Remark 4.2. For a non-locally free, reflexive and decomposable sheaf F without trivial factor on P3 with c1 = 2, we have ♯(Sing(F)) ≤ 2. Indeed, by Proposition 3.3, F is isomorphic to either F2,3,P ⊕ OP3(1), F2,3,P ⊕ T P3(−1) or F2,3,P ⊕ F2,3,P ′ for some P, P ′ ∈ P3 (possibly P = P ′). In each case, we have ♯(Sing(F)) ≤ 2 and only when F ∼= F2,3,P ⊕ F2,3,P ′ with P 6= P ′, we have ♯(Sing(F)) = 2. In other words, if ♯(Sing(F)) ≥ 3, then F is indecomposable. Let F be a reflexive sheaf of rank r on P3 with c1(F) = 2 and dim V (F) ≤ 0. Then it admits an exact sequence (2) and C is either empty or locally Cohen-Macaulay with Sing(C) ⊆ V (F)∪Sing(F) such that IC(2) is spanned outside V (F). Since Sing(C) is finite and C is locally Cohen-Macaulay, so C is reduced. Remark 4.3. Fix a linear subspace V ⊆ H 0(C, ωC (2)) spanning ωC(2) outside finitely many points, but not spanning ωC(2) at exactly α > 0 points P1, . . . , Pα. Then the reflexive sheaf F associated to (2) is not locally free and Sing(F) = {P1, . . . , Pα}. The extension (2) is given by r − 1 sections of ωC(2) and we set V ⊆ H 0(C, ωC (2)) to be their linear span. By our assumptions and Remark 4.3, V spans ωC(2) outside Sing(F). Since IC(2) is spanned outside the finite set V (F), then we have h0(IC(2)) ≥ 2 and so c2(F) = deg(C) ≤ 4. Moreover, there is no line L with deg(L∩C) ≥ 3, which is not a component . Thus of C. Note that we have C = ∅ if and only if F ∼= OP3(2) ⊕ O⊕(r−1) let us assume C 6= ∅ from now on. P3 REFLEXIVE SHEAVES 9 . P3 Lemma 4.4. If deg(C) = 1, then we have F ∼= OP3(1)⊕2 ⊕ O⊕(r−2) Proof. Note that C is a line. Since IC(2) is spanned and h1(OP3) = 0, so F is spanned. Any such F is given by r − 1 sections s1, . . . , sr−1 of ωC(2) ∼= OC with only finitely many common zeros. Since h0(C, OC ) = 1, by an automorphism of O⊕(r−1) we can assume that one of them is nowhere vanishing and the others are the zero-section. In other words, F is locally free and it has O⊕(r−2) as its direct factor. So we can assume rank(F) = 2. From the main theorem in [16], the only spanned vector bundle F of rank 2 with h0(F(−1)) = h0(IC(1)) = 2 is OP3(1)⊕2. (cid:3) P3 P3 Now assume that c2(F) = deg(C) = 2. Then C is either the disjoint union of two lines or a reduced conic. Proposition 4.5. Let F be a reflexive sheaf of rank r on P3 with (c1, c2) = (2, 2) and dim V (F) ≤ 0. Then F is one of the followings: (1) F ∼= NP3(1) ⊕ O⊕(r−2) P3 (2) 0 −→ OP3 −→ OP3(1) ⊕ T P3(−1) ⊕ O⊕(r−3) −→ F −→ 0. (3) F ∼= G ⊕ O⊕(r−2) or F ∼= ΩP3(2) ⊕ O⊕(r−3) P3 P3 . P3 with G ∈ A, a 6-dimensional irreducible variety. P3 Proof. When C is the disjoint union of two lines, say C = L1 ⊔ L2, we have ωC(2) ∼= O⊕2 C and IC(2) is spanned. In particular F is also spanned, i.e. V (F) = ∅. Note that h0(C, ωC (2)) = 2 and thus O⊕(r−3) is a direct factor of F if r ≥ 4. Hence it is sufficient to analyze the cases r ∈ {2, 3}. If r = 2, then the sequence (2) is induced by a nowhere zero section of ωC(2) and so F is locally free. Thus F is isomorphic to NP3(1), a null-correlation bundle of P3 twisted by 1 from the list in [16]. Now assume that r = 3. If P is a point in Li, then any section of ωC(2) vanishing at P also vanishes at every point of Li. It implies again that F is locally free (see Theorem 1 in [4]). Assuming that F has no trivial factor, we have F ∼= ΩP3(2) (e.g., by [16]). Now assume that C is a reduced conic. Since IC(2) is spanned, so is F. Note that h0(C, ωC (2)) = 3 and thus F has O⊕(r−4) as its direct factor if r ≥ 5. So it is sufficient to analyze the cases 2 ≤ r ≤ 4. The case of locally free sheaves turns out to satisfy the resolution (2) in [16], so let us assume that F is non-locally free. Since ωC(2) is spanned by 3 linearly independent sections of H 0(ωC(2)), we have r ≤ 3. If r = 3, we have F ∼= F2,3,P ⊕ OP3(1) for some P ∈ P3 by Proposition 4.10. From the defining sequence of F2,3,P and the Euler sequence of T P3(−1), we have a surjection T P3(−1) −→ F2,3,P . Thus F also admits a resolution (2). P3 The remaining case of r = 2 is induced from Proposition 4.9 giving us the (cid:3) last case in the list. In particular, such a sheaf with no trivial factor has r ≤ 3, c3(F) ∈ {0, 2} and ♯(Sing(F)) ≤ 2. 10 E. BALLICO, S. HUH AND F. MALASPINA Lemma 4.6. Let F be a reflexive sheaf of rank 2 on P3 fitting into an exact sequence (6) 0 −→ OP3 −→ F −→ IC(2) −→ 0 with C a reduced conic. Then there is an effective Cartier divisor Z ⊂ C of degree 2 such that OC(1) ∼= OC(Z), where Z is the scheme-theoretic zero- locus of a section of OC(1) with only finite zeros and OZ ∼= Ext1(F ∨, OP3 (−2)). The scheme Z is uniquely determined by F. If C is reducible, say C = L1 ∪ L2 with L1 and L2 lines, then we have deg(Z ∩ L1) = deg(Z ∩ L2) = 1. Proof. We have ωC(2) ∼= OC(1). As in [13], Theorem 4.1, F is uniquely determined by s ∈ H 0(C, OC (1)) which vanishes only at finitely many point, i.e. it does not vanish identically in one of the irreducible components of C. Hence the scheme-theoretic zero-locus of s is a degree 2 effective divisor Z of C. If C is reducible, say C = L1 ∪ L2 with L1 and L2 lines, then we have deg(Z ∩ L1) = deg(Z ∩ L2) = 1. As in the equation (2) in the proof of Theorem 4.1 in [13], we have an exact sequence 0 −→ OP3(−2) −→ F(−2) −→ Ext1(F ∨, OP3 (−2)) −→ 0. f −→ OP3 g −→ Ext1(IC, OP3 (−2)) Here we have IC = Im(f ) and the sheaf Ext1(IC, OP3 (−2)) is identified with ωC(2) ([13], proof of Theorem 4.1). Thus we have OZ ∼= OP3/ker(g) ∼= Ext1(F ∨, OP3(−2)). Lemma 4.7. Fix a reduced conic C ⊂ P3 and a Cartier divisor Z ⊂ C of degree 2. If C is reducible, say C = L1 ∪ L2 with L1 and L2 lines, assume deg(Z∩L1) = deg(Z∩L2) = 1. Then there is a unique reflexive sheaf FC,Z of rank 2 fitting into an extension (6) associated to a section of ωC(2) ∼= OC(1) with Z as its zero-locus. Proof. Since ωC(2) ∼= OC(1), the lemma is a particular case of Theorem 4.1. in [13]. (cid:3) (cid:3) Lemma 4.8. For a fixed 0-dimensional subscheme Z ⊂ P3 of degree 2, let C1 be the set of all reduced conics containing Z and let C2 be the set of all C ∈ C1 such that Z is the zero-locus of a section of OC (1). (a) Both C1 and C2 are irreducible varieties of dimension 5. (b) For a fixed C ∈ C1, we have C ∈ C2 if and only if either C is smooth or C is reducible, say C = L1 ∪ L2 with L1 and L2 lines, and deg(Z ∩ L1) = deg(Z ∩ L2) = 1. Proof. Note that Z is contained in a unique line, say L. Thus we have h0(IZ (1)) = h0(IL(1)) = 2 and so part (a) is a simple dimension counting. If C is smooth, then every 0-dimensional subscheme of C is a Cartier divisor and we have h0(C, OC (1) ⊗ OC (−Z)) = h0(OC ) = 1. Now assume that C = L1∪L2 is reducible with {O} = L1∩L2. Let Π be the plane spanned REFLEXIVE SHEAVES 11 by L1 ∪ L2. Since h0(Li, OLi(1)) = 2, we need to have deg(Z ∩ Li) ≤ 1 for all i. Hence if Z is reduced, then we have ♯(Z ∩ (Li \ {O})) = 1, i = 1, 2, while if Z is not reduced, then Zred = {O}. Every 0-dimensional subscheme W ⊂ Π of degree 2 such that Wred = {O} spans a unique line LW . Any such W is contained in L1 ∪ L2. W is a Cartier of L1 ∪ L2 if and only if L 6= L1 and L 6= L2, i.e. deg(W ∩ L1) = deg(W ∩ L2) = 1. (cid:3) Proposition 4.9. Let A be the set of all spanned reflexive sheaves of rank 2 on P3 with (c1, c2, c3) = (2, 2, 2). Then A is parametrized by an irreducible variety of dimension 6. Proof. The set of all 0-dimensional subschemes Z ⊂ P3 of degree 2 is parametrized by an irreducible, smooth and projective variety of dimen- sion 6. Note that A is the set of all reflexive sheaves F fitting into an exact sequence (6) with C a reduced conic. Since ωC(2) ∼= OC (1) is very ample, we have A 6= ∅ by [13], Theorem 4.1. For a fixed F ∈ A, we have h0(F) = 6 and so the non-zero sections of F are parametrized by a 5-dimensional projective space. Lemma 4.6 shows that Ext1(F ∨, OP3(−2)) ∼= OZ for a unique subscheme Z of degree 2. Lemmas 4.6 and 4.8 show that any reduced conic C associated to a non-zero section of F contains Z as a Cartier divisor and that if C is reducible, then no irreducible component of C contains Z. Lemma 4.7 says that any pair (C, Z) as above is associated to a unique F, up to isomorphisms. Since h0(F) = 6, we see that every conic C containing the scheme Z associated to Ext1(F ∨, OP3(−2)) arises from a non-zero section of F. (cid:3) Proposition 4.10. Let F be a non-locally free and reflexive sheaf of rank 3 on P3 whose associated curve C is a reduced conic with dim V (F) ≤ 0. Then we have F ∼= F2,3,P ⊕ OP3(1) for some P ∈ P3. Proof. Let ∆1 denote the set of all pairs (T, P ), where T ⊂ P3 is a reduced conic and P ∈ T . The set ∆1 is an irreducible quasi-projective variety. For each (T, P ) ∈ ∆1, we have h0(T, IP (1)) = 2 and IP (1) has no base points. Since ωT (2) ∼= OT (1) for any conic T , the set ∆ of all non-locally free reflexive sheaves of rank 3 on P3 admitting the sequence (2) with C = T ∈ ∆1, is parametrized by an irreducible quasi-projective variety. Let us fix F ∈ ∆. Since we have h0(F(−1)) = 1 and h0(F(−2)) = 0, there is an exact sequence i −→ F −→ G −→ 0 (7) with G torsion-free, rank(G) = 2, c1(G) = 1 and h0(G) = 3. 0 −→ OP3(1) Now assume that G is not reflexive. Let i : G ֒→ G∨∨ denote the natural inclusion. First assume that coker(j) is finite. In this case V (G∨∨) is finite. By Proposition 1.2 we get G∨∨ ∼= F2,3,P ′ for some P ′. Since h0(F2,3,P ′) = 3 = h0(G) and both G and F2,3,P ′ are spanned, we get G ∼= G∨∨, a contradiction. Now assume that coker(i) is supported by a curve Y , i.e. the one-dimensional part of Sing(G) is not empty. First assume that Y is not a line. There is 12 E. BALLICO, S. HUH AND F. MALASPINA a line R ⊂ P3 such that ♯(R ∩ Y ) ≥ 2 and R is not contained in Y . Hence the torsion part of GR is supported by at least two point, say P1 and P2. At Pi either FR has torsion or the restriction of i to R is not a vector bundle inclusion. In the latter case OR(2) is a subsheaf of FR and hence FR has splitting type (2, 0, 0). We excluded the latter case for any R, while the former case does not occur, because P1 6= P2. Now assume Y = R. As above we get a contradiction if Sing(G) has an isolated point. Hence we may assume that Y = Sing(G) is a line. Since G is spanned, h0(G) = 3 and G has no torsion, the kernel of the evaluation map H 0(G) ⊗ OP3 −→ G is a line bundle. Hence G fits into an exact sequence v −→ O⊕3 0 −→ OP3(−1) P3 −→ G −→ 0 (8) with v = (v1, v2, v3) ∈ H 0(OP3(1))⊕3. Since Sing(G) = Y is a line, the sections {v1, v2, v3} generate a 2-dimensional linear space and G ∼= IY (1) ⊕ OP3. Since OP3 is a direct summand of G and G is a quotient of F, so OP3 is a direct summand of F, a contradiction. Thus G is reflexive and so G ∼= F2,3,P for some P ∈ P3. Now from the sequence (7) it is easy to check that F is not semistable if if and only if F ∼= F2,3,P ⊕ OP3(1) and only if the sequence (7) splits, i.e. for some P ∈ P3. Hence non-split sheaves forms an open subset U of ∆ and we need to prove that U = ∅. Assume that U 6= ∅. We just saw that U is irreducible. Let us fix a general F ∈ U . For any line R ⊂ P3, the sheaf F2,3,P ⊕ OP3(1)R is a vector bundle of splitting type (1, 1, 0) if P /∈ R, while F2,3,P ⊕ OP3(1)R is a direct sum of OP and a trivial vector bundle if P ∈ R by Lemma 2.4. Hence there is no line with either F2,3,P ⊕ OP3(1)R of splitting type (2, 0, 0) or with torsion of degree ≥ 2. Since this condition is open, we get that F satisfies this condition with respect to the point {P } := Sing(F). Since F is spanned, so is G. We get G ∼= F2,3,P if G is reflexive. Since we have dim(Ext1(F2,3,P , OP3(1))) = 1 from Lemma 3.5, so there is a unique sheaf H, up to isomorphism, which is the middle term of a non-trivial extension of F2,3,P by OP3(1) and H is not isomorphic to F2,3,P ⊕ OP3(1). To get a contradiction to the non-emptiness of U it is sufficient to prove that H is locally free at P . We have F ∨ 2,3,P 2,3,P (1)) = 0. ∼= F2,3,P (−1) by Proposition 1.10 in [13] and so we have h1(F ∨ It implies that the non-zero extension giving H gives a non-zero element of H 0(Ext1(F2,3,P , OP3(1))). The latter vector space is 1-dimensional by Lemma 3.5 and so it is sufficient to find a locally free extension of F2,3,P by OP3(−1). Take a general inclusion OP3 −→ T P3(−1). Its cokernel Γ is a reflexive spanned sheaf of rank 2. By the list we get that Γ ∼= F2,3,P ′ for some P ′ ∈ P3. Composing with an automorphism of P3 sending P ′ into P we get that T P3(−1) is an extension of F2,3,P by OP3. (cid:3) Now assume that c2(F) = deg(C) = 3. Since C is reduced and there is no line D * C with deg(D ∩ C) ≥ 3, we can check that C is connected REFLEXIVE SHEAVES 13 by checking possible configurations of C. In this case, we have pa(C) = 0 and C is linearly normal. By the Castelnuovo-Mumford criterion, IC(2) is spanned. In all cases C is arithmetically Cohen-Macaulay and hence h1(F(t)) = 0 for all t ∈ Z. C is not a locally complete intersection if and only if C is the union of three non-coplanar lines through a common point; in this case C is not Gorenstein; even in this case we have deg(ωC) = 2pa(C)− 2 = −2. Thus in all cases we have c3(F) = 4. We also have h1(ωC(2)) = h0(C, OC (−2)) = 0 by the Serre duality for locally Cohen-Macaulay curves and so we have h0(C, ωC (2)) = 5 by Riemann-Roch. Hence F has O⊕(r−6) as a direct factor if r ≥ 7. So let us assume that 2 ≤ r ≤ 6. P3 Lemma 4.11. There is no non-locally free and spanned sheaf F of rank 6 with no trivial factor whose corresponding curve C has deg(C) = 3. Proof. We need to analyze the cases when C is reducible. As the worst case let us assume that C is not Gorenstein, i.e. C = L1 ∪ L2 ∪ L3, the union of 3 lines Li, not coplanar and through a common point, say O. Then O is the only non-Gorenstein point of C. We have h0(C, Hom(ωC , ωC)) = h1(C, ωC ) by III.7.6 with n = 1, i = 0 and F = ωC in [12]. Since C is connected, we have h0(C, OC ) = 1. Hence the case of n = i = 1 and F = OC in III.7.6. [12] gives h1(C, ωC ) = 1. We have 0 = h0(C, OC (−2)) = h1(C, ωC (2)) from the case of n = 1, F = ωC(2), i = 0 in the duality theorem [12], III.7.6. Hence the Riemann-Roch theorem gives h0(C, ωC (2)) = deg(ωC(2)) + 1 = 5. Let us fix a point P ∈ C and let IP ωC(2) denote the image of the natural map IP ⊗ωC −→ ωC induced by the inclusion IP ֒→ OC. To check that ωC(2) is spanned at P , it is sufficient to prove h1(C, IP ωC(2)) = 0. By [12], III.7.6, we have h1(C, IP ωC(2)) = dim Ext0(IP ωC(2), ωC ). Since each element of H 0(C, Hom(ωC , ωC)) is the multiplication by a scalar, while IP (2) has a non-zero section with at least a zero outside P , we get that the dimension of Ext0(IP ωC(2), ωC ) is zero and thus ωC(2) is spanned. (cid:3) Lemma 4.12. There is an indecomposable extension of F2,3,P by T P3(−1). Proof. From Lemma 3.5, we can fix a non-split exact sequence (9) 0 −→ T P3(−1) σ −→ F −→ F2,3,P −→ 0. We claim that F is indecomposable. Assume that F is decomposable. F is spanned, because h1(T P3(−1)) = 0. Since neither T P3(−1) nor F2,3,P have a trivial factor, F has no trivial factor. By the classification c1 = 1 we get F ∼= T P3(−1) ⊕ F2,3,P . We identify F with the latter sheaf. Write σ = (σ1, σ2) with σ1 : T P3(−1) −→ T P3(−1) and σ2 : T P3(−1) −→ F2,3,P . Since σ is injective and rank(F2,3,P ) < rank(T P3(−1)), we have σ1 6= 0. Since T P3(−1) is simple, σ1 is a non-zero multiple of the identity. In particular σ1 is invertible. Composing σ with the automorphism (cid:18) σ−1 −σ2 ◦ σ−1 1 1 0 1(cid:19) 14 E. BALLICO, S. HUH AND F. MALASPINA of T P3(−1)⊕ F2,3,P , we get that (9) splits and so we get a contradiction. (cid:3) Lemma 4.13. Fix P, P ′ ∈ P3 such that P 6= P ′. Let V be a general 3- dimension linear subspace of H 0(F2,3,P ⊕ F2,3,P ′) and C ⊂ P3 be the curve associated to it. Then C is a rational normal curve containing {P, P ′}. is not locally free at P and P ′, so we have Proof. Since F2,3,P ⊕ F2,3,P ′ {P, P ′} ⊂ C. Since F2,3,P ⊕ F2,3,P ′ is spanned and Sing(F2,3,P ⊕ F2,3,P ′) = {P, P ′}, we have Sing(C) ⊆ {P, P ′}. To prove the lemma it is sufficient to prove that C is smooth. Assume that C is singular. Let us define Ω ⊂ Gr(3, H 0(F2,3,P ⊕ F2,3,P ′)) to be the open subset of the Grassmannian parametrizing all V for which C is smooth outside {P, P ′}. We also define subsets Ω(P ) := {W ∈ Ω : C is locally free at P } and similarly Ω(P ′). The subsets Ω(P ) and Ω(P ′) are open in Ω. We assumed Ω(P ) ∩ Ω(P ′) = ∅. Since Ω is irreducible, we have either Ω(P ) = ∅ or Ω(P ′) = ∅. Assume Ω(P ) = ∅ for instance. Since F2,3,P ⊕ F2,3,P ′ ∼= F2,3,P ′ ⊕ F2,3,P , we get Ω(P ′) = ∅ and so Sing(C) = {P, P ′}. Let T be the line spanned by P and P ′. No connected union C with pa(C) = 0 of a line and a smooth conic or of 3 lines through a common point has exactly two singular points. Let C be a connected union of 3 lines with pa(C) = 0 and with Sing(C) = {P, P ′}. We have C = T ∪ L ∪ L′ with L a line through P and L′ a line through P ′. Let us fix a point O ∈ T \ {P, P ′}. Since F2,3,P ⊕ F2,3,P ′ is spanned and O /∈ V (F2,3,P ⊕ F2,3,P ′) for a general V ∈ Ω, so the associated curve, say C1, does not contain O. Hence T * C1 and we get a contradiction. (cid:3) Lemma 4.14. Fix P, P ′ ∈ P3 and a rational normal curve C ⊂ P3. Then F2,3,P ⊕ F2,3,P ′ is the unique spanned reflexive sheaf F of rank 4 with c1 = 2, c3 = 4, Sing(F) = {P, P ′} and associated to C. Proof. Let ∆ be the set of all triples (C1, P1, P ′ 1) with C1 a rational normal curve, P1 ∈ C1, P ′ 1. Any two triples in ∆ are projec- tively equivalent. Hence the lemma is true for one triple (C1, P1, P ′ 1) ∈ ∆ if and only if it is true for all such triples. Since any two elements of ∆ are projectively equivalent, there is an exact sequence (2) with r = 4, F2,3,P ⊕ F2,3,P ′ instead of F, C = C1 a rational normal curve and with {P, P ′} = Sing(F2,3,P ⊕ F2,3,P ′). Hence (C, P, P ′) is realized by some exten- sion (2) with r = 4 and F = F2,3,P ⊕ F2,3,P ′. 1 ∈ C1 and P1 6= P ′ Let us fix a triple (C, P, P ′). A spanned reflexive sheaf F comes from (2) with r = 4 and Sing(F) = {P, P ′} if and only if it is associated to a 3-dimensional linear subspace of H 0(C, ωC (2)) spanning ωC(2) at all points of C \ {P, P ′}, but spanning ωC(2) neither at P nor at P ′. Since ωC(2) has degree 4, we have h0(C, I{P,P ′} ⊗ ωC(2)) = 3. Hence there is a unique reflexive sheaf F obtained in this way and so F ∼= F2,3,P ⊕ F2,3,P ′ by Lemma 4.13. (cid:3) REFLEXIVE SHEAVES 15 Proposition 4.15. F2,3,P ⊕ F2,3,P ′ is the unique spanned reflexive sheaf on P3 with no trivial factor, (c1, c2, c3) = (2, 3, 4) and {P, P ′} as its non-locally free locus. Proof. Let F be a spanned reflexive sheaf of rank 4 with (c1, c2, c3) = (2, 3, 4), Sing(F) = {P, P ′} and no trivial factor. Assume F 6= F2,3,P ⊕ F2,3,P ′ and then F is indecomposable by Proposition 3.3. Take a general 3-dimensional linear subspace of H 0(F) and use it to get the exact sequence (2) with r = 4. The curve C is smooth outside {P, P ′} and we know that F comes from a 3-dimensional linear subspace of H 0(ωC(2)) spanning ωC(2) at all points of C \ {P, P ′}, but spanning C neither at P nor at P ′. If C is a rational normal curve, then we can apply Lemma 4.14 and thus we may assume that C is reducible. The Bertini theorem gives Sing(C) ⊆ {P, P ′}. Exchanging P and P ′ if necessary, we may assume that C is singular at P . (a) First assume that C is not formed by 3 lines through P . Hence C is Gorenstein and ωC(2) is a degree 4 spanned line bundle. First assume h1(C, I{P,P ′} ⊗ ωC(2)) = 0 and so we have h0(C, I{P,P ′} ⊗ ωC(2)) = 3. Then the triple (C, P, P ′) gives a unique bundle G. There is an integral affine curve A with o ∈ A and a flat family of triples {(Ct, Pt, P ′ t )}t∈A with Ct ⊂ P3 a smooth rational normal curve containing {P, P ′} for all t ∈ A\{o} and Pt, P ′ t distinct points of Ct. In this family the function t 7→ h0(Ct, I{P,P ′} ⊗ ωCt(2)) is constant. Hence we find a flat family of spanned reflexive sheaves {Ft}t∈A with Fo ∼= F and Sing(Ft) = {P, P ′} for all t. Lemma 4.14 gives Ft ∼= F2,3,P ⊕ F2,3,P ′ for all t 6= o. By the semicontinuity theorem for cohomology we get the existence of non-zero maps m : F2,3,P −→ F , m′ : F2,3,P ′ −→ F u : F −→ F2,3,P , u′ : F −→ F2,3,P ′ with u′ ◦ m = 0 and u ◦ m′ = 0. Since F2,3,P and F2,3,P ′ are stable and non-isomorphic, we get F ∼= F2,3,P ⊕ F2,3,P ′. (b) Now assume that C is formed by 3 non-coplanar lines through a common point P , which is the unique singular point of C. By step (a) and Lemma 4.14 to find a contradiction to the irreducibility of F it is sufficient to find another exact sequence (2) for F with C not the union of 3 lines through either P or P ′. Let T ⊂ P3 be the line spanned by P and P ′. It is contained in any union C1 of 3 lines through a common point containing both P and P ′. Let us fix O ∈ T \{P, P ′}. Since F is spanned and O /∈ V (F) for a general 3-dimensional subspace of H 0(F), the associated curve, say C1, does not contain O. Thus we have T * C1. Since {P, P ′} = Sing(F) ⊂ C1, we get that C1 is not a union of 3 lines through a common point. (cid:3) Proposition 4.16. For each r ∈ {2, . . . , 5} there exists a non-locally free, spanned and reflexive sheaf F of rank r and with no trivial factor, (c1, c2) = (2, 3) and k := ♯(Sing(F)) = 6−r. Moreover, there exists an indecomposable sheaf in each cases except when (r, k) = (4, 2). 16 E. BALLICO, S. HUH AND F. MALASPINA Proof. Let us take C a normal rational curve of degree 3. For any S ⊂ C with ♯(S) = k, we have h0(C, ωC (2)⊗OC (−S)) = 5−k and the linear system ωC(2)⊗OC (−S) has S as its base locus. Thus V := H 0(ωC(2)⊗OC (−S)) ⊂ H 0(ωC(2)) defines an extension 0 −→ V ⊗ OP3 −→ F −→ IC(2) −→ 0 with Sing(F) = S. Such F is reflexive and spanned since IC(2) is spanned. In other words, we obtain the first statement due to Remark 4.3. For the case of k = 1, we have an indecomposable sheaf of rank 5 with one singular point by Lemma 4.12. When k = 2, i.e. r = 4, we have a unique choice for such sheaves that is F2,3,P ⊕ F2,3,P ′ by Proposition 4.15. Thus we cannot have an indecomposable one in this case. When k ∈ {3, 4, 5}, every such sheaves are indecomposable due to Proposition 3.3 and Remark 4.2. (cid:3) In fact, we can have indecomposable reflexive sheaves F of rank r even with ♯(Sing(F)) < 6 − r. For example, when r = 4, we have the following result : Lemma 4.17. There exists an indecomposable, reflexive and spanned sheaf F of rank 4 with ♯(Sing(F)) = 1. Proof. Let us fix a general map σ : OP3 −→ F2,3,P ⊕ T P3(−1). Then the sheaf F := Coker(σ) is locally free outside P , because F2,3,P ⊕ T P3(−1) is spanned and it is locally free outside P . Since the non-locally free sheaf F2,3,P ⊕ T P3(−1) is an extension of F by OP3, F is not locally free at P . We also get that F is a spanned reflexive sheaf whose corresponding curve C has degree 3. Assume that F is decomposable. Since we have Sing(F) = {P }, the classification in Proposition 3.3 gives that either F ∼= F2,3,P ⊕ F2,3,P or F ∼= F2,3,P ⊕ OP3(1) ⊕ OP3. The latter case is excluded, because F2,3,P and OP3(1) ⊕ OP3 have different Segre classes and hence F and F2,3,P ⊕ OP3(1) ⊕ OP3 have different Segre classes. About the first case, let us note that the fibers of F2,3,P ⊕ F2,3,P and of F2,3,P ⊕ T P3(−1) at P have dimension 6. We fixed the point P and then took a general σ. Since σ is general, the fiber of F at P has dimension 5. Hence F is not isomorphic to F2,3,P ⊕ F2,3,P . Thus we have an indecomposable, reflexive and spanned sheaf of rank 4 with one singular point. (cid:3) Now assume that c2(F) = deg(C) = 4 as the final case. Since IC(2) is spanned outside finitely many points, we get that C is a complete inter- section and so IC(2) is spanned. Thus F is spanned, h0(F) = r + 1 and ωC(2) ∼= OC(2). Since h0(C, ωC(2)) = 8, we have F ∼= G ⊕ O⊕(r−9) for some G if r ≥ 10 and so we can analyze the cases 2 ≤ r ≤ 9. Since the case when F is locally free was already dealt in [16], we will assume that F is non-locally free. If r = 9, then V = H 0(C, ωC (2)) spans ωC(2) and so F is locally free. Thus we can assume that 2 ≤ r ≤ 8. P3 REFLEXIVE SHEAVES 17 Lemma 4.18. Let C be an irreducible curve and L a very ample line bundle on C. For a fixed integer k with 1 ≤ k ≤ h0(L) − 1 and a set A of general k points on C, we have h0(L(−A)) = h0(L) − k and the base locus of the linear system L(−A) is finite. Moreover, if k satisfies 1 ≤ k ≤ h0(L) − 2, then A is exactly the base locus of L(−A) for a general A. Proof. Since L is very ample, we can embed C as a linearly normal curve C ⊂ Pm with m := h0(L) − 1 by the linear system of L. Any set of general k points with k ≤ m on the non-degenerate curve C ⊂ Pm are linearly independent and any hyperplane contains only finitely many points of C. Hence h0(L(−A)) = h0(L) − k and the base locus of the linear system L(−A) is finite. If k ≤ m − 1, then A is the base locus of the linear system L(−A), because a general hyperplane section of C is in linearly general position (see page 109 in [3]). (cid:3) Proposition 4.19. For each r ∈ {2, . . . , 8} there exists a non-locally free, spanned, indecomposable and reflexive sheaf F with (c1, c2) = (2, 4) and ♯(Sing(F)) = (cid:26) 8, 9 − r, if r = 2; if r 6= 2. Proof. Note that ωC(2) is very ample and h0(ωC (2)) = 8. Assume that C is irreducible. Then for each integer k with 1 ≤ k ≤ 6, we can find a finite subset S ⊂ C such that ♯(S) = k and the base locus of the linear system ωC(2)⊗OC (−S) is finite and so V := H 0(ωC (2)⊗OC (−S)) defines a reflexive sheaf F of rank 9−k with no trivial factor. We also have Sing(F) = S by the second statement of Lemma 4.18. In the case of r = 2, the assertion follows by Theorem 4.1 in [13]. Now as in the proof of Proposition 4.16, all such sheaves are indecompos- (cid:3) able due to the classification of sheaves with c1 = 1. So far we give a complete description on reflexive sheaves on P3 with c1(F) = 2 and dim V (F) ≤ 0. As an automatic consequence, we obtain that every such a reflexive sheaf on P3 is really spanned everywhere, i.e. V (F) = ∅ (see Corollary 1.2). Remark 4.20. In [16] we have a complete list of the spanned bundles on Pn with c1 ≤ 2 and so the possibilities for decomposable spanned bundles with no trivial factor and c1 = 2 are OPn(1) ⊕ T Pn(−1) and T Pn(−1)⊕2. Conversely, let E be a spanned bundle of rank 4 on P3 with (c1, c2, c3) = (2, 2, 2) and no trivial factor. Then it fits into the following sequence 0 −→ O⊕3 P3 −→ E −→ IC(2) −→ 0 where C is a smooth conic in P3. Applying Theorem 2.3 in [1] to E ∨ with j = 1 and t = 0, we obtain that ΩP3(1) is a direct summand of E ∨ and so E ∼= OP3(1) ⊕ T P3(−1). Similarly every spanned bundle of rank 6 on P3 with (c1, c2, c3) = (2, 3, 4) and no trivial factor is isomorphic to T P3(−1)⊕2. It gives us the list of indecomposable bundles from Theorem 1.1 of [16]. 18 E. BALLICO, S. HUH AND F. MALASPINA 5. Locally free sheaf with c1 = 2 on Pn, n ≥ 3 In this section, we prove a similar statement to Corollary 1.2 for vector bundles on Pn. Theorem 5.1. Let E be a vector bundle on Pn with n ≥ 3 such that c1(E) = 2 and assume that E is spanned outside a subvariety of Pn with codimension at least 3. Then E is spanned and so it is one of the vector bundles described in [16]. Proof. The case n = 3 is true by Corollary 1.2. Now assume n > 3 and V (E) 6= ∅. For a fixed point P ∈ V (E), let M ⊂ Pn be a general 3-dimensional linear subspace containing P . Since M is general, M ∩ V (E) has dimension max{0, dim(V (E)) − 3}. Set F := EM . By Corollary 1.2 the vector bundle F is spanned and hence it is as listed in [16] for the case n = 3. Let V ⊆ H 0(F) denote the image of the restriction map H 0(E) −→ H 0(F). Taking r − 1 general sections of H 0(E), we get an exact sequence (10) 0 −→ O⊕(r−1) Pn −→ E −→ IK(2) −→ 0 where K is a locally Cohen-Macaulay subscheme of Pn of pure codimension 2 (or the empty set). By part (F) at page 4 in [2] and their references [7][8], K is smooth outside V (E) and a subscheme ∆ of codimension at least 4. Let C ⊂ M denote the scheme-theoretic intersection K ∩ M . Since the tensor product is a right exact functor, the ideal sheaf IC,M (2) is the image of IK(2) ⊗ OM inside OM . We often write IC instead of IC,M when we are working with F and M . Since O⊕(r−1) has no torsion, by tensoring (10) with OM we get the exact sequence on M : Pn (11) 0 −→ O⊕(r−1) M −→ F −→ IC(2) −→ 0 (here we use that E is locally free, otherwise E ⊗ OM may have torsion). Since M is a general 3-dimensional subspace containing P , we have M ∩ (V (E) ∪ ∆) = {P }. Hence the Bertini theorem gives that C is smooth outside P . By (11) the germ IC,P (2) has a free resolution of length 2. Since OM,P is a regular local ring of dimension 3, so OC,P (2) has depth 1, i.e. C is locally Cohen-Macaulay at P . Since C is reduced outside P , we get that C is reduced at P . The integer deg(C) is the degree of the union (counting multiplicities) of the codimension 2 components of K. (a) If F ∼= OP3(2) ⊕ O⊕(r−1) , then by Corollary 1.7 in [10] we have . In particular, it is spanned. Similarly when F ∼= P3 E ∼= OPn(2) ⊕ O⊕(r−1) Pn OP3(1)⊕2 ⊕ O⊕(r−2) P3 , then we have E ∼= OPn(1)⊕2 ⊕ O⊕(r−2) . Pn (b) Now assume F ∼= NP3(1) ⊕ O⊕(r−2) , where NP3 is a null-correlation bundle. Hence C is a degeneration of a family of pairs of disjoint lines. Since C is reduced, it is a disjoint union of two lines. Hence K is (outside P3 REFLEXIVE SHEAVES 19 V (E)) the union of two 2-codimensional linear subspaces, K1 and K2 with K1 ∩ K2 ∩ M = ∅. Since M is general among the 3-dimensional linear subspaces containing P , we get P /∈ K1 ∩ K2. K was fixed. Then we take any P ∈ V (E). We get K1 ∩ K2 ∩ V (E) = ∅ and that K is smooth at each point of V (E). Since E is locally free and for each P ′ ∈ Pn the ring OPn,P ′ is an n-dimensional local ring, so (10) implies that K has depth ≥ n − 2. Fix P ′ ∈ Sing(K). Taking a general 3-dimensional linear subspace M ′ containing P ′ we get as above that K ∩M ′ is a disjoint union of two lines. Hence K ∩M ′ is smooth at P ′. Hence K is smooth at P ′, a contradiction. Since n ≥ 4, we have K1 ∩ K2 6= ∅. Hence Sing(K) 6= ∅. Hence this case does not occur if n ≥ 4. Alternatively, since K is reduced with pure codimension 2 and K ∩M is the disjoint union of two lines, K is a union of two linear subspaces K1 and K2 such that dim(K1) = dim(K2) = n − 2 and dim(K1 ∩ K2) = n − 4. K is not locally Cohen-Macaulay at any point of K1 ∩ K2. (c) Now assume F ∼= ΩP3(2)⊕O⊕(r−3) . As in (b) we get a contradiction for all n ≥ 4, because even in this case C is the disjoint union of two lines. Indeed, we have h1(IC) = h1(ΩP3) = 1 and it implies that h0(OC ) = 2 and so C is a curve of degree 2 with two connected components. P3 (d) Now assume that F fits in an exact sequence 0 −→ OP3(−2) −→ O⊕(r+1) P3 −→ F −→ 0. Note that h0(F) = r + 1 and no proper subspace of H 0(F) spans F out- side finitely many points. Hence the restriction map H 0(E) −→ H 0(F) is surjective and so P /∈ V (E), a contradiction. (e) Now assume that F fits in an exact sequence 0 −→ OP3(−1)⊕2 −→ O⊕(r+2) P3 −→ F −→ 0. Note that h0(F) = r + 2 and the image of V in H 0(IC(2)) has dimension 2. Thus its zero-locus is a degree 4 complete intersection containing C. Since V1 spans IC(2) outside finitely many points, we get a contradiction. (cid:3) References [1] V. Ancona and G. Ottaviani, Some applications of Beilinson's theorem to projective spaces and quadrics, Forum Math. 3 (1991), no. 2, 157 -- 176. MR 1092580 (92e:14039) [2] C. Anghel and N. Manolache, Globally generated vector bundles on Pn with c1 = 3. [3] E. Arbarello, M. Cornalba, P. A. Griffiths, and J. Harris, Geometry of algebraic curves. Vol. I, Grundlehren der Mathematischen Wissenschaften [Fundamental Princi- ples of Mathematical Sciences], vol. 267, Springer-Verlag, New York, 1985. MR 770932 (86h:14019) [4] E. Arrondo, A home-made Hartshorne-Serre correspondence, Rev. Mat. Complut. 20 (2007), no. 2, 423 -- 443. MR 2351117 (2008g:14084) [5] E. Ballico, S. Huh, and F. Malaspina, Globally generated vector bundles of rank 2 on a smooth quadric threefold, Preprint, arXiv : 1211.1100 [math.AG], 2012. [6] E. Ballico, S. Huh, and F. Malaspina, On higher rank globally generated vector bundles over a smooth quadric threefold, Preprint, arXiv : 1211.2593v2 [math.AG], 2012. [7] C. Banica, Smooth Reflexive Sheaves, Rev. Roum. Math. Pures et Appl., 36 (1991), 571 -- 593. MR 1172165 (93g:14027) 20 E. BALLICO, S. HUH AND F. MALASPINA [8] M.-C. Chang, A filtered Bertini-type theorem, J. Reine Angew. Math. 397 (1989), 214 -- 219. MR 993224 (90i:14054) [9] L. Chiodera and P. Ellia, Rank two globally generated vector bundles with c1 ≤ 5, Rend. Istit. Mat. Univ. Trieste 44 (2012), 1 -- 10. [10] G. Elencwajg and O. Forster, Bounding cohomology groups of vector bundles on Pn, Math. Ann. 246 (1979/80), no. 3, 251 -- 270. MR 563404 (81h:32035) [11] P. Ellia, Chern classes of rank two globally generated vector bundles on P2, Preprint, arXiv:1111.5718 [math.AG], 2011. [12] R. Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977, Graduate Texts in Mathematics, No. 52. MR 0463157 (57 #3116) [13] , Stable reflexive sheaves, Math. Ann. 254 (1980), no. 2, 121 -- 176. MR 597077 (82b:14011) [14] N. Manolache, Globally generated vector bundles on P3 with c1 = 3, Preprint, arXiv:1202.5988 [math.AG], 2012. [15] C. Okonek, M. Schneider, and H. Spindler, Vector bundles on complex projective spaces, Progress in Mathematics, vol. 3, Birkhauser Boston, Mass., 1980. MR 561910 (81b:14001) [16] J.-C. Sierra and L. Ugaglia, On globally generated vector bundles on projective spaces, J. Pure Appl. Algebra 213 (2009), no. 11, 2141 -- 2146. MR 2533312 (2010d:14062) [17] J.-C. Sierra and L. Ugaglia, On globally generated vector bundles on projective spaces II, Preprint, arXiv:1203.0185 [math.AG], 2012. Universit`a di Trento, 38123 Povo (TN), Italy E-mail address: [email protected] Department of Mathematics, Sungkyunkwan University, Cheoncheon-dong, Jangan-gu, Suwon 440-746, Korea E-mail address: [email protected] Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy E-mail address: [email protected]
1004.4411	2	1004	2012-01-03T07:46:59	Moduli spaces of irregular singular connections	[ "math.AG" ]	In the geometric version of the Langlands correspondence, irregular singular point connections play the role of Galois representations with wild ramification. In this paper, we develop a geometric theory of fundamental strata to study irregular singular connections on the projective line. Fundamental strata were originally used to classify cuspidal representations of the general linear group over a local field. In the geometric setting, fundamental strata play the role of the leading term of a connection. We introduce the concept of a regular stratum, which allows us to generalize the condition that a connection has regular semisimple leading term to connections with non-integer slope. Finally, we construct a symplectic moduli space of meromorphic connections on the projective line that contain a regular stratum at each singular point.	math.AG	math	MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS CHRISTOPHER L. BREMER AND DANIEL S. SAGE Abstract. In the geometric version of the Langlands correspondence, irregu- lar singular point connections play the role of Galois representations with wild ramification. In this paper, we develop a geometric theory of fundamental strata to study irregular singular connections on the projective line. Funda- mental strata were originally used to classify cuspidal representations of the general linear group over a local field. In the geometric setting, fundamental strata play the role of the leading term of a connection. We introduce the concept of a regular stratum, which allows us to generalize the condition that a connection has regular semisimple leading term to connections with noninte- gral slope. Finally, we construct a moduli space of meromorphic connections on the projective line with specified formal type at the singular points. 1. Introduction A fundamental problem in the theory of differential equations is the classifi- cation of first order singular differential operators up to gauge equivalence. An updated version of this problem, rephrased into the language of algebraic geome- try, is to study the moduli space of meromorphic connections on an algebraic curve C/k, where k is an algebraically closed field of characteristic 0. This problem has been studied extensively in recent years due to its relationship with the geometric Langlands correspondence. To elaborate, the classical Langlands conjecture gives a bijection between automorphic representations of a reductive group G over the ad`eles of a global field K and Galois representations taking values in the Lang- lands dual group G∨. By analogy, meromorphic connections (or, to be specific, flat G∨-bundles) play roughly the same role in the geometric setting as Galois rep- resentations (see [11, 12] for more background). Naively, one would like to find a description of the moduli space of meromorphic connections that resembles the space of automorphic representations of a reductive group. A more precise statement is that the geometric Langlands data on the Galois side does not strictly depend on the connection itself, but rather on the monodromy representation determined by the connection. When the connection has regular singularities, i.e., when there is a basis in which the matrix of the connection has simple poles at each singular point, the Riemann-Hilbert correspondence states that the monodromy representation is simply a representation of the fundamental group. However, when the connection is irregular singular, the monodromy has a more subtle description due to the Stokes phenomenon. The irregular monodromy data consists of a collection of Stokes matrices at each singular point, which characterize 2010 Mathematics Subject Classification. Primary:14D24, 22E57; Secondary: 34Mxx, 53D30. Key words and phrases. meromorphic connections, irregular singularities, moduli spaces, para- horic subgroups, fundamental stratum, regular stratum. The research of the second author was partially supported by NSF grant DMS-0606300 and NSA grant H98230-09-1-0059. 1 2 CHRISTOPHER L. BREMER AND DANIEL S. SAGE the asymptotic expansions of a horizontal section on sectors around each irregular singular point (see [24], or [25] for a modern treatment.) The irregular Riemann-Hilbert map from moduli spaces of connections to the space of Stokes matrices is well understood in the following situation. Let V ∼= On C be a trivializable rank n vector bundle, and let ∇ be a meromorphic connection on V with an irregular singular point at x. After fixing a local parameter t at x, suppose that ∇ has the following local description: ∇ = d + Mr dt tr+1 + Mr−1 dt tr + . . . , where Mj ∈ gln(C) and the leading term Mr has pairwise distinct eigenvalues. Then, we say that ∇ has a regular semisimple leading term at x. Under this as- sumption, Jimbo, Miwa and Ueno [16] classify the deformations of ∇ that preserve the Stokes data by showing that they satisfy a system of differential equations (the so-called isomonodromy equations). In principle, the isomonodromy equations give a foliation of the moduli space of connections, with each leaf corresponding to a sin- gle monodromy representation. Indeed, the Riemann-Hilbert map has surprisingly nice geometric properties for connections with regular semisimple leading terms. Consider the moduli space M of connections on P1 which have singularities with regular semisimple leading terms at {x1, . . . , xm} and which belong to a fixed for- mal isomorphism class at each singular point. Boalch, whose paper [5] is one of the primary inspirations for this project, demonstrates that M is the quotient of a smooth, symplectic manifold fM by a torus action. Moreover, the space of Stokes data has a natural symplectic structure which makes the Riemann-Hilbert map symplectic. However, many irregular singular connections that arise naturally in the geo- metric Langlands program do not have a regular semisimple leading term. In [25, Section 6.2], Witten considers a connection of the form ∇ = d +(cid:18) 0 t−r+1 t−r 0 (cid:19) dt, which has a nilpotent leading term. Moreover, it is not even locally gauge-equivalent to a connection with regular semisimple leading term unless one passes to a ramified cover. A particularly important example is described by Frenkel and Gross in [13]. They construct a flat G-bundle on P1, for arbitrary reductive G, that corresponds to a 'small' supercuspidal representation of G at ∞ and the Steinberg representation at 0. In the GLn case, the result is a connection (due originally to Katz [17]) with a regular singular point at 0 (with unipotent monodromy) and an irregular singular point with nilpotent leading term at ∞. This construction suggests that connections with singularities corresponding to cuspidal representations of G, an important case in the geometric Langlands correspondence, do not have regular semisimple leading term in the sense above. These examples lead us to one of our main questions: is there a natural generalization of the notion of a regular semisimple leading term which allows us to extend the results of Boalch, Jimbo, Miwa and Ueno? The solution to this problem is again motivated by analogy with the classical Langlands correspondence. Suppose that F is a local field and W is a ramified rep- resentation of GLn(F ). Let P ⊂ GLn(F ) be a parahoric subgroup with a decreasing MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 3 filtration by congruence subgroups {P j}, and suppose that β is an irreducible rep- resentation of P r on which P r+1 acts trivially. We say that W contains the stratum (P, r, β) if the restriction of W to P r has a subrepresentation isomorphic to β. In the language of Bushnell and Kutzko [7, 8, 18], the data (P, r, β) is known as a fundamental stratum if β satisfies a certain non-degeneracy condition (see Section 2.4). If we write eP for the period of the lattice chain stabilized by P , an equivalent condition is that (P, r, β) attains the minimal value r/eP over all strata contained in W ([7, Theorem 1]). It was proved independently by Howe and Moy [15] and Bushnell [7] that every irreducible admissible representation of GLn(F ) contains a fundamental stratum. It was further shown in [8] and [18] that fundamental strata play an important role in the classification of supercuspidal representations, especially in the case of wild ramification. As a tool in representation theory, fundamental strata play much the same role as the leading term of a connection in the cases considered above. Therefore, we are interested in finding an analogue of the theory of strata in the context of meromorphic connections in order to study moduli spaces of connections with cuspidal type singularities. In this paper, we develop a geometric theory of strata and apply it to the study of meromorphic connections. We introduce a class of strata called regular strata which are particularly well-behaved: connections containing a regular stratum have similar behavior to connections with regular semisimple leading term. More precisely, a regular stratum associated to a formal meromorphic connection allows one to "diagonalize" the connection so that it has coefficients in the Cartan subalgebra of a maximal torus T . We call the diagonalized form of the connection a T -formal type. In Section 4.4, we show that two formal connections that contain regular strata are isomorphic if and only if their formal types lie in the same orbit of the relative affine Weyl group of T . The perspective afforded by a geometric theory of strata has a number of benefits. (1) The description of formal connections obtained in terms of fundamental strata translates well to global connections on curves; one reason for this is that, unlike the standard local classification theorem [20, Theorem III.1.2], one does not need to pass to a ramified cover. In the second half of the pa- per, we use regular strata to explicitly construct moduli spaces of irregular connections on P1 with a fixed formal type at each singular point. In par- ticular, we obtain a concrete description of the moduli space of connections with singularities of "supercuspidal" formal type. (2) Fundamental strata provide an illustration of the wild ramification case of the geometric Langlands correspondence; specifically, in the Bushnell- Kutzko theory [8, Theorem 7.3.9], refinements of fundamental strata corre- spond to induction data for admissible irreducible representations of GLn. (3) The analysis of the irregular Riemann-Hilbert map due to Jimbo, Miwa, Ueno, and Boalch [16, 5] generalizes to a much larger class of connections. Specifically, one can concretely describe the isomonodromy equations for families of connections that contain regular strata [6]. (4) Since the approach is purely Lie-theoretic, it can be adapted to study flat G-bundles (where G is an arbitrary reductive group) using the Moy-Prasad theory of minimal K-types [21]. 4 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Here is a brief outline of our results. In Section 2, we adapt the classical theory of fundamental strata to the geometric setting. Next, in Section 3, we introduce the notion of regular strata; these are strata that are centralized (in a graded sense defined below) by a possibly non-split maximal torus T . The major result of this section is Theorem 3.8, which states that regular strata split into blocks corresponding to the minimal Levi subgroup L containing T . In Section 4, we show how to associate a stratum to a formal connection with coefficients in a Laurent series field F or, equivalently, to a flat GLn-bundle over the formal punctured disk Spec(F ). By Theorem 4.10, every formal connection (V, ∇) contains a fundamental stratum (P, r, β), and the quantity r/eP for any funda- mental stratum contained in (V, ∇) is precisely the slope. Moreover, Theorem 4.12 states that any splitting of (P, r, β) induces a splitting of (V, ∇). In particular, any connection containing a regular stratum has a reduction of structure to the Levi subgroup L defined above. In Section 4.3, we show that the matrix of any connec- tion containing a regular stratum is gauge-equivalent to a matrix in t = Lie(T ), which we call a formal type. We show in Section 4.4 that the set of formal types associated to an isomorphism class of formal connections corresponds to an orbit of the relative affine Weyl group in the space of formal types. In Section 5, we construct a moduli space M of meromorphic connections on P1 with specified formal type at a collection of singular points as the symplectic reduction of a product of smooth varieties that only depend on local data. By Theorem 5.6, there is a symplectic manifold fM which resolves M by symplec- tic reduction. Finally, Theorem 5.26 relaxes the regularity condition on strata at regular singular points so that it is possible to consider connections with unipo- tent monodromy. In particular, this construction contains the GLn case of the flat G-bundle described by Frenkel and Gross. 2. Strata In this section, we describe an abstract theory of fundamental strata for vector spaces over a Laurent series field in characteristic zero. Strata were originally developed to classify cuspidal representations of GLn over non-Archimedean local fields [7, 8, 18]. We will show that there is an analogous geometric theory with applications to the study of flat connections with coefficients in F . In Section 3, we introduce a novel class of fundamental strata of "regular uniform" type. These strata will play an important role in describing the moduli space of connections constructed in section 5. 2.1. Lattice Chains and the Affine Flag Variety. Let k be an algebraically closed field of characteristic zero. Here, o = k[[t]] is the ring of formal power series in a variable t, p = to is the maximal ideal in o, and F = k((t)) is the field of fractions. Suppose that V is an n-dimensional vector space over F . An o-lattice L ⊂ V is a finitely generated o-module with the property that L ⊗o F ∼= V . If we twist L by powers of t, then every L(m) is an o-lattice as well. L(m) = t−mL, Definition 2.1. A lattice chain L is a collection of lattices (Li)i∈Z with the fol- lowing properties: MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 5 (1) Li ) Li+1; and (2) Li(m) = Li−me for some fixed e = eL . Notice that a shift in indexing (L[j])i := Li+j produces a (trivially) different lattice chain L [j]. The lattice chain L is called complete if e = n; equivalently, Li/Li+1 is a one dimensional k-vector space for all i. Definition 2.2. A parahoric subgroup P ⊂ GL(V ) is the stabilizer of a lattice chain L , i.e., P = {g ∈ GL(V ) gLi = Li for all i}. The Lie algebra of P is the parahoric subalgebra P ⊂ gl(V ) consisting of P = {p ∈ gl(V ) pLi ⊂ Li for all i}. Note that P is in fact an associative subalgebra of gl(V ). An Iwahori subgroup I is the stabilizer of a complete lattice chain, and an Iwahori subalgebra I is the Lie algebra of I. There are natural filtrations on P (resp. P) by congruence subgroups (resp. ideals). For r ∈ Z, define the P-module Pr to consist of X ∈ P such that XLi ⊂ Li+r for all i; it is an ideal of P for r ≥ 0 and a fractional ideal otherwise. The congruence subgroup P r ⊂ P is then defined by P 0 = P and P r = In+Pr for r > 0. Define eP = eL ; then, tP = PeP . Finally, P is uniform if dim Li/Li+1 = n/e for all i. In particular, an Iwahori subgroup I is always uniform. Proposition 2.3 ([7, Proposition 1.18]). The Jacobson radical of the parahoric subalgebra P is P1. Moreover, when P is uniform, there exists an element P ∈ P such that P P = PP = P1. As an example, suppose that V = Vk ⊗k F for a given k-vector space Vk. There is a distinguished lattice Vo = Vk ⊗k o, and an evaluation map ρ : Vo → Vk obtained by setting t = 0. Any subspace W ⊂ Vk determines a lattice ρ−1(W ) ⊂ V . Thus, if F = (Vk = V 0 ⊃ V 1 ⊃ . . . ⊃ V e = {0}) is a flag in Vk, then F determines a lattice chain by LF = (. . . ⊃ t−1ρ−1(V n−1) ⊃ Vo ⊃ ρ−1(V 1) ⊃ . . . ⊃ ρ−1(V n−1) ⊃ tVo ⊃ . . .). We call such lattice chains (and their associated parahorics) standard. Thus, if F0 is the complete flag determined by a choice of Borel subgroup B, then ρ−1(B) is the standard Iwahori subgroup which is the stabilizer of LF0. Similarly, the partial flag in Vk determined by a parabolic subgroup Q gives rise to a standard parahoric subgroup ρ−1(Q) which is the stabilizer of the corresponding standard lattice chain. In particular, the maximal parahoric subgroup GLn(o) is the stabilizer of the standard lattice chain associated to (Vk ⊃ {0}). In this situation, the obvious GLn(F )-action acts transitively on the space of complete lattice chains, so we may identify this space with the affine flag variety GLn(F )/I, where I is a standard Iwahori subgroup. More generally, every lattice chain is an element of a partial affine flag variety GLn(F )/P for some standard parahoric P . For more details on the relationship between affine flag varieties and lattice chains in general, see [22]. For any maximal subfield E ⊂ gl(V ), there is a unique Iwahori subgroup IE such that o× E ⊂ IE. Lemma 2.4. Suppose that P is a parahoric subgroup of GL(V ) that stabilizes a lattice chain L . Let E/F be a degree n = dim V field extension with a fixed 6 CHRISTOPHER L. BREMER AND DANIEL S. SAGE embedding in gl(V ) such that o× LE ⊃ L with stabilizer IE such that o× translation of the indexing. subgroup. E ⊂ P . Then, there exists a complete lattice chain E ⊂ IE and IE ⊂ P ; it is unique up to E is contained in a unique Iwahori In particular, o× Proof. We may identify V with E as an E-module. Since o× E ⊂ P , it follows that oE ⊂ P. Therefore, we may view L as a filtration of E by nonzero oE-fractional ideals. Since oE is a discrete valuation ring, there is a maximal saturation LE of L , unique up to indexing, consisting of all the nonzero fractional ideals, and it is clear that o× E ⊂ IE ⊂ P . The final statement follows by taking P to be the stabilizer of the lattice oE. (cid:3) F/k be the space of one-forms on F , and let Ω× ⊂ Ω1 2.2. Duality. Let Ω1 F/k be the F ×-torsor of non-zero one-forms. If ν ∈ o× dt tℓ ⊂ Ω×, its order is defined by ord(ν) = −ℓ. Any ν ∈ Ω× defines a nondegenerate invariant symmetric k-bilinear form h, iν on gln(F ) by hA, Biν = Res [Tr(AB)ν] , where Res is the usual residue on differential forms. In most contexts, one can take ν to be dt t . Let P be the parahoric subalgebra that preserves a lattice chain L . Proposition 2.5 (Duality). Fix ν ∈ Ω×. Then, (Ps)⊥ = P1−s−(1+ord(ν))eP , and, if r ≤ s, (Pr/Ps)∨ ∼= P1−s−(1+ord(ν))eP /P1−r−(1+ord(ν))eP ; here, the superscript ∨ denotes the k-linear dual. This is shown in Proposition 1.11 and Corollary 1.13 of [7]. In particular, when ord(ν) = −1, (Ps)⊥ = P1−s and (Pr/Pr+1)∨ = P−r/P−r+1. Observe that any element of Pr induces an endomorphism of the associated graded o-module gr(L ) of degree r; moreover, two such elements induce the same endomorphism of gr(L ) if and only if they have the same image in Pr/Pr+1. The following lemma gives a more precise description of the quotients Pr/Pr+1. Let ¯G = GL(L0/tL0) ∼= GLn(k) with ¯g the corresponding Lie algebra. Note that there is a natural map from P → ¯G whose image is a parabolic subgroup Q; its unipotent radical U is the image of P 1. Analogous statements hold for the Lie algebras Lie(Q) = q and Lie(U ) = u. Lemma 2.6. (1) There is a canonical isomorphism of o-modules Pr/Pr+1 ∼= Hom(Li/Li+1, Li+r/Li+r+1). eP −1Mi=0 (2) In the case r = 0, this isomorphism gives an algebra isomorphism between P/P1 and a Levi subalgebra h for Lie(Q) = q (defined up to conjugacy by U ). Moreover, P is a split extension of h by P1. (3) Similarly, if H is a Levi subgroup for Q, then P ∼= H ⋉ P 1. MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 7 i=0 Hom(Li/Li+1, Li+r/Li+r+1); it is an algebra homomorphism when r = 0. It is clear that Pr+1 is the kernel, since any o-module map that takes Li to Li+r+1 for 0 ≤ i ≤ eP − 1 must lie in Pr+1. Proof. There is a natural o-module map Pr →LeP −1 Now, suppose that (φi) ∈ LeP −1 i=0 Hom(Li/Li+1, Li+r/Li+r+1). Let F be the partial flag in L0/tL0 = L0/LeP given by {Li/LeP 0 ≤ i ≤ eP }. We may choose an ordered basis e for L0 that is compatible with F modulo LeP . This means that there is a partition e = e0 ∪ · · · ∪ eeP −1 such that Wj = span(ej ) ⊂ Lj is naturally isomorphic to Lj/Lj+1. In this basis, the groups Hom(Li/Li+1, Li+r/Li+r+1) ap- pear as disjoint blocks in Pr (with exactly one block in each row and column of the array of blocks), so it is clear that we can construct a lift φ ∈ Pr that maps to (φi). Note that when r = 0, the image of this isomorphism is a Levi subalgebra h for the parabolic subalgebra q. The choice of basis gives an explicit embedding h ∼= gl(W1) ⊕ · · · ⊕ gl(WeP −1) ⊂ GL(V ), so the extension is split. The proof in the group case is similar. (cid:3) Remark 2.7. The same proof gives an isomorphism (2.1) Pr/Pr+1 ∼= Hom(Li/Li+1, Li+r/Li+r+1) m+eP −1Mi=m for any m. However, if Hom(Li/Li+1, Li+r/Li+r+1) and Hom(Lj/Lj+1, Lj+r/Lj+r+1) for i ≡ j mod eP are identified via homothety, the image of an element of Pr/Pr+1 is independent of m up to cyclic permutation. Indeed, this follows immediately from the observation that if m = seP + j for 0 ≤ j < eP , then ts+1e0 ∪ · · · ∪ ts+1ej−1 ∪ tsej ∪ tseeP −1 is a basis for Lm. In particular, P/P1 is isomorphic to a Levi subalgebra in gl(Lm/tLm) Remark 2.8. Any element ¯x ∈ P/P1 determines a canonical GL(L0/tL0)-orbit in gl(L0/tL0), and similarly for P/P 1. To see this, note that any choice of ordered basis for L0 compatible with L maps ¯x onto an element of a Levi subalgebra of gl(L0/tL0); a different choice of compatible basis will conjugate this image by an element of Q. In fact, this orbit is also independent of the choice of base point L0 in the lattice chain. Indeed, an ordered basis for L0 compatible with L gives a com- patible ordered basis for Lm by multiplying basis elements by appropriate powers of t and then permuting cyclically. Using the corresponding isomorphism L0 → Lm to identify gl(L0/tL0) and gl(Lm/tLm), the images of ¯x are the same. Accordingly, it makes sense to talk about the characteristic polynomial or eigenvalues of ¯x. Notational Conventions. Let ν ∈ Ω×, and let V be an F -vector space. Suppose that P ⊂ GL(V ) is a parahoric subgroup with Lie algebra P that stabilizes a lattice chain (Li)i∈Z. We will use the following conventions throughout the paper: (1) ¯Li = Li/Li+1. (2) ¯P = P/P 1, ¯P = P/P1. (3) ¯P ℓ = P ℓ/P ℓ+1, ¯Pℓ = Pℓ/Pℓ+1. (4) If X ∈ Ps, then ¯X will denote its image in ¯Ps and the corresponding degree s endomorphism of gr(L ). (5) If α ∈ (Pr)∨, then αν ∈ gl(V ) with denote an element such that α = hαν , ·iν . 8 CHRISTOPHER L. BREMER AND DANIEL S. SAGE (6) If β ∈ ( ¯Pr)∨, then βν will denote an element of Pr−(1+ord(ν))eP such that ¯βν ∈ ¯Pr−(1+ord(ν))eP is the coset determined by the isomorphism in Propo- sition 2.5. (7) Let X ∈ Ps. Then, δX : ¯Pi → ¯Pi+s is the map induced by ad(X). (8) Let a ⊂ End(V ) be a subalgebra. We define ai = a ∩ Pi and ¯ai = ai/ai+1. (9) If A ⊂ GL(V ) is a subgroup, define Ai = A ∩ P i and ¯Ai = Ai/Ai+1. 2.3. Tame Corestriction. In this section, we first suppose that L is a com- plete (hence uniform) lattice chain in V with corresponding Iwahori subgroup I. By Proposition 2.3, I1 is a principal ideal generated by I ; similarly, each frac- tional ideal Iℓ is generated by ℓ I . Choose an ordered basis (e0, . . . , en−1) for V indexed by Zn, so that en+i = ei for all i. Furthermore, we may choose the ba- if r = qn − s, 0 ≤ s < n, then Lr is spanned by sis to be compatible with L : {tq−1e0, . . . , tq−1es−1, tqes, . . . , tqen−1}. Thus, if we let ¯ei denote the image of ei in L0/tL0, then L corresponds to the full flag in L0/tL0 determined by the ordered basis (¯e0, . . . ¯en−1). In this basis, we may take (2.2) I = 0 ... 0 t 1 . . . . . . 0 · · · . . . 0 · · · 0 ... 1 0  . Notice that the characteristic polynomial of I is equal to λn−t, which is irreducible over F . Thus, F [I] is a degree n field extension isomorphic to F [t1/n]. Remark 2.9. If P is a parahoric subgroup stabilizing the lattice chain L , we say that a basis for L0 is compatible with L if it is a basis that is compatible as above for any complete lattice chain extending L . Note that any pullback of a compatible basis for the induced partial flag in L0/tL0 is such a basis. We first examine the kernel and image of the map δr have order −1. We define ψℓ(X) = hX, −ℓ be the induced functional on ¯Iℓ. I : ¯Il → ¯Il−r. Let ν ∈ Ω× I iν . Note that ψℓ(Iℓ+1) = 0; we let ¯ψℓ Let d ⊂ gln(k) be the subalgebra of diagonal matrices. By Lemma 2.6, the Iwahori subgroup and subalgebra have semidirect product decompositions: I = d∗ ⋊ I 1 and I is a split extension of d by I1. Accordingly, any coset in Iℓ/Iℓ+1 has a unique representative xℓ I with x = diag(x0, x1, . . . , xn−1) ∈ d. in ¯Iℓ contains the one-dimensional subspace spanned by ℓ in ¯Iℓ−r is contained in ker( ¯ψℓ−r), and the kernel I . Equality in Lemma 2.10. The image of δ−r of δ−r both cases happens if and only if gcd(r, n) = 1. I I Proof. Take X = xℓ I with x ∈ d as above. By direct calculation, (2.3) [X, −r I ] = x′ℓ−r I , with x′ = diag(x0 − x−r, . . . , xn−1 − xn−1−r). MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 9 Therefore, ψℓ−r(x′ℓ−r I ) = Res(Tr(x′ℓ−r I r−ℓ I )ν) = Res(ν) Tr(x′) = Res(ν) (¯Iℓ) ⊂ ker( ¯ψℓ−r). (xi − xi−r) = 0. n−1Xi=0 It follows that δ−r I I The kernel of δ−r satisfies the equations xi − xi−r = 0 for 0 ≤ i ≤ n − 1. If we set x0 = α, then x−r = x−2r = x−3r = · · · = α. When gcd(r, n) = 1, j ≡ −mr (mod n) is solvable for any j, and it follows that xj = α for all j. Therefore, the kernel is just the span of ℓ I . Otherwise, the dimension of the kernel is at least 2. This implies that the image of δ−r has codimension 1 if and only if gcd(r, n) = 1. (cid:3) I For future reference, we remark that there is a similar formula to (2.3) for Ad. Any element of ¯I ∼= d∗ is of the form ¯p for p = diag(p0, . . . , pn−1). Then, (2.4) Ad(¯p)(−r I ) = p′−r I ∈ ¯I−r, where p′ = diag( p0 p−r , p1 p1−r , . . . , pn−1 pn−1−r ). In particular, when gcd(r, n) = 1, every generator of I−r lies modulo I−r+1 in the Ad(I)-orbit of a−r for some a ∈ k∗. I Next, we consider more general uniform parahorics. Let E/F be a degree m extension; it is unique up to isomorphism. Now, identify V ∼= En/m as an F vector space. We will view E as a maximal subfield of glm(F ): if we define E = I ∈ glm(F ) as in (2.2), then E is the centralizer of E, which is in fact a uniformizing parameter for E. Since mn, define a Cartan subalgebra t ∼= En/m in gl(V ) as the block diagonal embedding of n/m copies of E ⊂ glm(F ). Let gcd(r, m) = 1, and take ξ = (a1−r E ) with the ai's pairwise unequal elements of k. This implies that ξ is regular semisimple with centralizer t. E , . . . , an/m−r corresponding Iwahori subgroup. We define a lattice chain L =Ln/m Let LE be the complete lattice chain in F m stabilized by oE; we let IE be the LE in V with associated parahoric subgroup P . It is clear that P is uniform with eP = m. Moreover, elements of Pℓ are precisely those n/m × n/m arrays of m × m blocks with entries in Iℓ E; in particular, we can take P = (E, . . . , E). Note that t ∩ gl(L) ⊂ P for any lattice L in L . i=1 Proposition 2.11 (Tame Corestriction). There is a morphism of t-bimodules πt : gl(V ) → t satisfying the following properties: (1) πt restricts to the identity on t; (2) πt(Pℓ) ⊂ Pℓ; (3) the kernel of the induced map ¯πt : (t + Pℓ−r)/Pℓ−r+1 → t/(t ∩ Pℓ−r+1) is given by the image of ad(Pℓ)(ξ) modulo Pℓ−r+1; (4) if z ∈ t and X ∈ gl(V ), then hz, Xiν = hz, πt(X)iν; (5) πt commutes with the action of the normalizer N (T ) of T . 10 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Proof. First, take ν = dt E in t and 0 elsewhere. Define ψi t . Let ǫi ∈ gl(V ) be the identity element in the ith copy of s(X) = 1 m h−s E ǫi, Xiν and πt(X) = ∞Xs=−∞ n/mXi=1 ψi s(X)s Eǫi. It is easily checked that for s ≪ 0, ψi s(X) = 0 and that πt is a t-map. A direct calculation shows that for n ∈ N (T ), πt(Ad(n)X) = Ad(n)πt(X). Since πt is defined using traces, it is immediate that it vanishes on the off-diagonal blocks ǫi gl(V )ǫj for i 6= j. Moreover, πt is the identity on t, since ψi Eǫk) = 1, if j = s and i = k, and equals 0 otherwise. s(j We note that πt(Pℓ) ⊂ Pℓ ∩ t, so the induced map ¯πt makes sense. Let V ℓ−r ij = ǫiPℓ−rǫj so that ¯V ℓ−r is an isomorphism whenever i 6= j. This proves that the off-diagonal part of ker(¯πt) is of the desired form. We may now reduce without loss of generality to the case of a single diagonal block, i.e., t = E and P = I. ij ⊂ ker(¯πt) for i 6= j. By regularity, δξ : ¯V ℓ ij → ¯V ℓ−r ij Since πt is the identity on t, the kernel of ¯πt is contained in ¯Iℓ−r = ℓ−r Notice that when s < ℓ − r, −s that ψs(X)s Lemma 2.10, ad(Iℓ)(−r of the proposition. E ∈ Iℓ−r+1 for s > ℓ − r. ¯I. I I ⊂ I1; therefore, ψs(Iℓ−r) = 0. It is trivial It follows that ker(¯πt) = ker( ¯ψℓ−r). By E ) = ker( ¯ψℓ−r). This completes the proof of the third part I ℓ−r I is a t-map, it suffices to prove the fourth part when ν = dt Finally, for arbitrary ν′ = f ν, hz, Xiν ′ = hz, f Xiν. Since f ∈ F ⊂ T , and πt t . Although z ∈ t is an Eǫi for some asi ∈ k, only a finite number Eǫi. of terms contribute to the inner products. Hence, it suffices to consider z = s Observing that hs infinite sum of the formPs≥qPn/m E ǫjiν = mδijδrs, we see that i=1 asis Eǫi, −r hs Eǫi, Xiν = mψi −s(X) = hs Eǫi, ψi −s(X)−s E ǫiiν = hs Eǫi, πt(X)iν, as desired. (cid:3) Remark 2.12. Suppose that P ⊂ GL(V ) is a uniform parahoric that stabilizes a lattice chain L . Let H ⊂ P be the Levi subgroup that splits P → P/P 1 as in Lemma 2.6, and let h ⊂ P be the corresponding subalgebra. We will show that there is a generator P for P1 that is well-behaved with respect to H, akin to I ∈ I1. In the notation used in the proof of Lemma 2.6, h is determined by j=0 ej. Setting an ordered basis e for L0 partitioned into eP equal parts: e =SeP −1 Wj = span ej ∼= ¯Lj, we have h =LeP −1 j=0 gl(Wj) and H =QeP −1 Now, let L ′ be the complete lattice chain determined by the ordered basis for L0 given above, and let I be the corresponding Iwahori subgroup. If I is the generator of I1 constructed in (2.2), define P = m I , where m = n/eP . This matrix is an eP × eP block matrix of the same form as (2.2), but with scalar m × m blocks. Evidently, P (Li) = Li+1, so P generates P1. Furthermore, P (Wj) = Wj+1 for 0 ≤ j < eP − 1, and P (t−1WeP −1) = W0. It follows that P normalizes H, and Ad(P )(h) ⊂ h. In fact, if A = diag(A0, . . . , AeP −1) ∈ h, then Ad(r P )(A) = diag(Ar, . . . , Ar+eP −1), with the indices understood modulo eP . j=0 GL(Wj ). MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 11 2.4. Strata. For the remainder of Section 2, ν ∈ Ω× will be a fixed one-form of order −1. Definition 2.13. Let V be an F vector space. A stratum in GL(V ) is a triple (P, r, β) consisting of • P ⊂ GL(V ) a parahoric subgroup; • r ∈ Z≥0; • β ∈ ( ¯Pr)∨. Proposition 2.5 states that ( ¯Pr)∨ = ¯P−r. Therefore, we may choose a represen- tative βν ∈ P−r for β. Explicitly, a stratum is determined by a triple (L , r, βν), where L is the lattice chain preserved by P , and βν is a degree −r endomorphism of ν) give the stratum if and only if r = r′, L ′ L . The triples (L , r, βν ) and (L ′, r′, β′ ν induce the same maps on gr(L ), i.e., ¯βν = ¯βν . is a translate of L , and βν and β′ We say that (P, r, β) is fundamental if βν +P−r+1 contains no nilpotent elements of gln(F ). By [7, Lemma 2.1], a stratum is non-fundamental if and only if (βν )m ∈ P1−rm for some m. Remark 2.14. A stratum (P, r, β) is fundamental if and only if ¯βν ∈ End(gr(L )) is non-nilpotent in the usual sense. In particular, if βν(Li) = Li−r for all i ∈ Z, then (P, r, β) is necessarily fundamental. Definition 2.15. Let (P, r, β) be a stratum in GL(V ). A reduction of (P, r, β) is a GL(V )-stratum (P ′, r′, β′) with the following properties: (cid:16)β′ (cid:0)βν + P1−r(cid:1) 6= ∅, βν + P1−r ⊂ (P′)−r′ both of the associated lattice chains L and L ′. ν + (P′)1−r′(cid:17) ∩ , and there exists a lattice L that lies in Let (P ′, r′, β′) be a reduction of (P, r, β). The first property allows one to choose βν ∈ gln(F ) to represent both β and β′. The second implies that any representative βν for β determines an element of ( ¯P′)r′ . Note that it is possible to have two 2) with the same P ′ and r′, if P1−r * different reductions (P ′, r′, β′ (P′)1−r′ 1) and (P ′, r′, β′ . An important invariant of a stratum (P, r, β) is its slope, which is defined by slope(P, r, β) = r/eP . The following theorem describes the relationship between slope and fundamental strata. Theorem 2.16. Suppose that r ≥ 1. Then (P, r, β) is a non-fundamental stratum if and only if there is a reduction (P ′, r′, β′) with slope(P, r, β) < slope(P ′, r′, β′). This is proved in Theorem 1 and Remark 2.9 of [7]. Definition 2.17. A stratum (P, r, β) is called uniform if it is fundamental, P is a uniform parahoric subgroup, and gcd(r, eP ) = 1. The stratum is strongly uniform if it is uniform and βν(Li) = Li−r for all Li ∈ L . Remark 2.18. A uniform stratum (P, r, β) is strongly uniform if and only if the induced maps ¯βν : ¯Li → ¯Li−r are isomorphisms for each i. The forward implication follows since the ¯Li's have the same dimension. For the converse, note that if the ¯βi ν 's are isomorphisms, then, in particular, Li−r+j = βν (Li+j) + Li−r+j+1 for j=0 βν (Li+j)+Li−r+eP = βν(Li)+tLi−r, 0 ≤ j < eP . Substituting gives Li−r =PeP −1 so βν(Li) = Li−r by Nakayama's Lemma. Any fundamental stratum has a reduction with gcd(r, eP ) = 1. 12 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Lemma 2.19. If (P, r, β) is a fundamental stratum, there is a fundamental reduc- tion (P ′, r′, β′) with the property that gcd(r′, eP ) = 1. Proof. Let g = gcd(r, eP ) and r′ = r/g. Let (L′)j = Ljg, and set L ′ = ((L′)j). This is the sub-lattice chain of L consisting of all lattices of the form LaeP +br with a, b ∈ Z. If we choose a representative βν for β, βν((L′)j) ⊂ (L′)j−r′ . Thus, βν + . Let β′ ∈ (( ¯P′)r)∨ be the functional determined by the image of βν P1−r ⊂ (P′)−r′ in ( ¯P′)−r′ ν ∈ (P′)−N r′+1, then βN ν ∈ P−N r+1: if βν(L′)j ⊂ (L′)j−N r′+1, then . If βN for any j ∈ Z and 0 < m < g, βν(Ljg+m) ⊂ βν(Ljg) ⊂ Ljg−N r+g ⊂ Ljg−N r+m+1. Thus, if (P, r, β) is fundamental, so is (P ′, r′, β′). (cid:3) 2.5. Split Strata. We now generalize the notion of a 'split stratum' given in [18, Section 2] and [8, Section 2.3] to the geometric setting. Suppose that (P, r, β) is a stratum in V and that L = (Li)i∈Z is the lattice chain stabilized by P . Let V = V1 ⊕ V2 with V1, V2 6= {0}. Define Li j has maximal rank in Vj, so it is indeed a lattice. Let Lj be the lattice chain consisting of (Li j) omitting repeats. We denote the parahoric associated to Lj by Pj . Note that if Li = Li j = Li ∩ Vj for j = 1, 2. Note that Li j is automatically a lattice in Vj. 2 for all i, then each Li 1 ⊕ Li Definition 2.20. We say that (V1, V2) splits P if (1) Li = Li (2) L1 is a uniform lattice chain with eP1 = eP . 2 for all i, and 1 ⊕ Li In addition, (V1, V2) splits β at level r if βν (Li j) ⊂ Li−r j + Li−r+1. Note that the above definition is independent of the choice of representative βν. However, it is possible to choose a 'split' representative for βν . Let πj : V → Vj be the projection, ιj : Vj → V the inclusion, and ǫj = ιj ◦ πj. Set βjν = πj ◦ βν ◦ ιj ν ∈ βν + P1−r. Thus, by and β′ replacing βν by β′ ν, we may assume without loss of generality that the representative βν is "block-diagonal", i.e., it satisfies β′ . If βj is the functional induced by βjν , then (Pj, r, βj) is a stratum in GL(Vj ). ν = β1ν ⊕ β2ν. Whenever (V1, V2) splits β and P , β′ j) ⊂ Li−r ν(Li j 1 ⊕ ¯Li Remark 2.21. If P is uniform and (V1, V2) splits P , then P2 is also uniform with eP2 = eP , since ¯Li ∼= ¯Li 2. Furthermore, if (P, r, β) is strongly uniform, (V1, V2) splits β, and the first part of the splitting condition for P is satisfied, then (V1, V2) splits P . Since gcd(r, eP ) = 1 and βν (Li) = Li−r for all i, we may choose integers ν generates the P-module P1. Thus, if we choose βν a and b such that αν = taβb 1) = ¯Li+1 such that βν(Vj ) ⊂ Vj as above, it is clear that αν (Li . Accordingly, the subquotients ¯Li 1 have the same dimension for all i and there are no repeats in the lattice chain. This implies that L1 is uniform and eP1 = eP . In fact, we see that (P1, r, β1) is strongly uniform. and ¯αν( ¯Li 1) = Li+1 1 1 Define V12 to be the vector space HomF (V2, V1), and let ∂βν be the operator ∂βν : V12 → V12 x 7→ β1νx − xβ2ν . We remark that if we embed V12 in gl(V ) in the obvious way and assume that βν is block-diagonal, then ∂βν (x) = [βν, x] = δβν (x). MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 13 The map ∂βν is a degree −r endomorphism of the o-lattice chain M = (M j) defined by M j = {x ∈ V12 xLi 2 ⊂ Li+j 1 for all i}. By [18, Lemma 2.2], M is a uniform lattice chain with period eP . The functional on ¯P induced by ∂βν is independent of the choice of representative; we denote the corresponding stratum by (P12, r, ∂β). Recall from Remark 2.8 that any element of P/P1 determines a conjugacy class in gln(k). Accordingly, if (P, 0, β) is a stratum with r = 0, it makes sense to refer to the 'eigenvalues' of ¯βν . If the stratum splits at level 0, then the eigenvalues of the diagonal blocks ( ¯β1)ν and ( ¯β2)ν are well-defined. Definition 2.22. We say that (V1, V2) splits the fundamental stratum (P, r, β) if (1) (V1, V2) splits P and β at level r; (2) (P1, r, β1) and (P12, r, ∂β) are strongly uniform; and (3) when r = 0, the eigenvalues of ¯β1ν are distinct from the eigenvalues of ¯β2ν modulo Z. Remark 2.23. The congruence subgroup P 1 acts on the set of splittings of (P, r, β), i.e., if g ∈ P 1 and (V1, V2) splits (P, r, β), then so does (gV1, gV2). First, note that 1⊕gLi P 1 stabilizes this stratum. Next, given g ∈ P 1, it is clear that Li = gLi = gLi 2 and that gL1 is uniform with the same period as L1. Thus, (gV1, gV2) splits P ; it also splits β at level r, since gx ∈ x + Li+1 for any x ∈ Li. It is obvious that the induced strata on gV1 and gV12 are strongly uniform. Finally, note that when viewed as subalgebras of ¯P is the natural way, ¯Pj and gPj/gP1 j are the same. j and ¯βjν are the same, so the last It follows that the eigenvalues of gβjν + gP1 condition also holds. If P1 is a uniform parahoric, then P12 is as well, with the same period. To see this, note that there is an isomorphism (2.5) Since dimk( ¯Lℓ 1) = dimk(L0 1/tL0 ¯M j → eP −1Mℓ=0 Hom( ¯Lℓ 2, ¯Lℓ+j 1 ). dimk( ¯M j) = dimk(L0 1) dimk(L0 2/tL0 2)/eP1, 1)/eP1 for all ℓ, 1/tL0 and P12 is uniform. Furthermore, since dimk(M j/tM j) = dimk(L0 it follows that tM j = M j+eP1 , i.e., eP12 = eP1 = eP . 1/tL0 1) dimk(L0 2/tL0 2), Let V21 = HomF (V1, V2). Define a lattice chain N = {N i} in V21 in the same way as for M . An argument similar to that given above shows that N is uniform with period eP and that the operator ∂′ βν (x) = β2νx − xβ1ν is an endomorphism of N of degree −r. We let (P21, r, ∂′ β) be the associated stratum. βν on V21 defined by ∂′ Lemma 2.24. The stratum (P21, r, ∂′ β) is strongly uniform. Proof. It only remains to show that ∂′ β(N i) = N i−r for all i. First, observe that there is a natural injection ¯M i ֒→ ¯Pi, so by Proposition 2.5, we have a surjection ¯P−i ∼= ( ¯Pi)∨ → ( ¯M i)∨. Since the kernel of this map consists of the image of the "block upper triangular" matrices, we see that ( ¯M i)∨ ∼= ¯N −i. Next, if x ∈ M i and 14 CHRISTOPHER L. BREMER AND DANIEL S. SAGE y ∈ N −i+r, hβ1x − xβ2, yiν = Res(Tr(β1xy)ν − Tr(xβ2y)ν) = Res(Tr(xyβ1)ν − Tr(xβ2y)ν) = −hx, β2y − yβ1iν . Since ¯∂βν ¯∂′ β( ¯N −i+r) = ¯N −i, so ∂β(N i) = N i−r by the same remark. : ¯M i → ¯M i−r is an isomorphism by Remark 2.18, it follows that (cid:3) j Suppose that (V1, V2) splits a uniform stratum (P, r, β) as above. By Remark 2.21, using the indexing con- j)i∈Z. In this setting, it makes sense eP1 = eP2 = eP . Thus, it is never the case that Li vention in Definition 2.20, and indeed Lj = (Li to think of (P, r, β) as the direct sum of (P1, r, β1) and (P2, r, β2). j = Li+1 with each Vj 6= {0}. Let (Pj , r, βj) be a stratum in GL(Vj ) corresponding to a uniform parahoric Pj, and let ePj = ePk for all j, k. Define Li j and LJ = (Li J )i∈Z, and let PJ ⊂ GL(V ) be the parahoric subgroup that stabilizes LJ . In the following, let J be a finite indexing set, and suppose VJ = Lj∈J Vj, Finally, let βJ =Lj∈J βj. Definition 2.25. Under the assumptions of the previous paragraph: J = Lj∈J Li (1) When J = {1, 2}, we say that (PJ , r, βJ ) = (P1, r, β1)⊕ (P2, r, β2) if (V1, V2) splits (PJ , r, βJ ). (2) When J = {1, . . . , m}, we define the direct sum recursively by Note that if we set Jℓ = {ℓ, . . . , m} for ℓ ∈ J, then (Vℓ, VJℓ+1) must split (PJℓ , r, βJℓ) for all ℓ. (3) We say that a uniform stratum (P, r, β) ∈ GL(V ) splits into the direct sum Lj∈J (Pj, r, βj) if there is an isomorphism V ∼= VJ under which (P, r, β) and (PJ , r, βJ ) are equivalent. Remark 2.26. It is clear that the direct sum operation is associative. It is not symmetric because Definition 2.22(2) implies that (Pj , r, βj) is strongly uniform whenever j 6= m. The definition is easily modified to make it symmetric, but we will not do so here. We can determine if a stratum has a splitting by considering the characteristic polynomial of (P, r, β). Fix a parameter t ∈ F and let g = gcd(r, eP ). Define an element (2.6) yβ = βeP /g ν tr/g + P1 ∈ ¯P. Recall from Remark 2.8 that yβ determines a conjugacy class in gln(k). Definition 2.27. We define the characteristic polynomial φβ ∈ k[X] of the stratum (P, r, β) to be the characteristic polynomial of yβ. The local field version of the following proposition is in [18, Proposition 3.4]. Proposition 2.28. Suppose that gcd(r, eP ) = 1 and r > 0. The stratum (P, r, β) is fundamental if and only if φβ(X) has a non-zero root. If (P, r, β) is fundamental, it splits if φβ (X) = g(X)h(X) for g, h ∈ k[X] relatively prime of positive degree. (Pj, r, βj) = (P1, r, β1) ⊕ ((P2, r, β2) ⊕ (. . . ⊕ (Pm, r, βm))) . Mj∈J MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 15 Remark 2.29. Given any fundamental stratum (P, r, β), one can always find a re- duction that satisfies the condition gcd(r, eP ) = 1 by Lemma 2.19. Proof. Note that ¯βν and yβ = tr ¯βeP ν , viewed as endomorphisms of gr(L ), are either simultaneously nilpotent or not. Using the identification of ¯P and a Levi subalgebra of gln(k), we see that the latter is nilpotent if and only its characteristic polynomial φβ(X) equals X n. Since (P, r, β) is not fundamental if and only if ¯βν is nilpotent, we see that (P, r, β) is fundamental if and only if φβ(X) has a nonzero root. Let φβ ∈ F [X] be the characteristic polynomial of y = trβeP ν . Then, φβ nec- essarily has coefficients in o and φβ ≡ φβ (mod p). Hensel's lemma states that φ(X) = g(X)h(X), where h ≡ h (mod p) and g ≡ g (mod p). We take V1 = ker(g(y)) and V2 = ker(h(y)). By Lemmas 3.5 and 3.6 of [18], for all i, and ∂β(M j) = M j−r for (cid:3) (V1, V2) splits P and β at level r, β1(Li all j. Therefore, (V1, V2) splits (P, r, β). 1) = Li−r 1 Corollary 2.30. Suppose that (P, r, β) is a uniform stratum that is not strongly uniform. Then, (P, r, β) splits into the direct sum of two strata (P1, r, β1) and (P2, r, β2), where (P1, r, β1) is strongly uniform and (P2, r, β2) is non-fundamental. Proof. Factor φβ(X) = g(X)h(X) so that h(X) = X m and g(0) 6= 0, and let V1 and V2 be the subspaces from the proof of the previous proposition. Since (P, r, β) is fundamental, deg(g) > 0 and V1 is nontrivial. Moreover, Remark 2.18 implies that V1 6= V ; if not, yβ = tr ¯βeP (and hence ¯βν) would be invertible endomorphisms ν of gr(L ), contradicting the fact that (P, r, β) is not strongly uniform. Since yβ (and hence ¯βν) restricts to an automorphism of gr(L1), the same remark shows that (P1, r, β1) is strongly uniform. On the other hand, since yβ restricts to a nilpotent endomorphism of gr(L2), Remark 2.14 shows that (P2, r, β2) is not fundamental. (cid:3) 3. Regular Strata In this section, we make precise the notion of a stratum with regular semisimple "leading term". We introduce the concept of a regular stratum; this is a stratum which is "graded-centralized" by a maximal torus. Regular strata do not appear in the theory of strata for local fields. However, they play an important role in the geometric theory. 3.1. Classification of regular strata. Consider a stratum (P, r, β). Recall that the congruence subgroups P i act on βν by the adjoint action. In particular, Ad(P i)(βν ) ⊂ βν + Pi−r, since p ∈ P i implies that p and p−1 act trivially on Lj/Li+j. Define Z i(βν) ⊂ ¯P i to be the stabilizer of βν (mod Pi−r+1). Notice that this is independent of the ν = βν + γ, for some γ ∈ P−r+1, then choice of representative βν for β: Ad(P i)(γ) ⊂ γ + Pi−r+1. if β′ Let T ⊂ GL(V ) be a maximal torus. The corresponding Cartan subalgebra t ⊂ gl(V ) is the centralizer of a regular semisimple element and is therefore an as- sociative subalgebra. In particular, since t must have the structure of a commutative semisimple algebra, t is the product of field extensions of F : t = E1 × E2 × . . . × Eℓ and T = E× ℓ . We let oj be the ring of integers of Ej and pj its maximal ideal. Let sj = [Ej : F ]. The field Ej contains a uniformizer which is an 2 × . . . × E× 1 × E× 16 CHRISTOPHER L. BREMER AND DANIEL S. SAGE root of t; we let ωj ∈ t denote this uniformizer supported on the jth summand. sth j We will also denote the identity of the jth Wedderburn component of t by χj. There is a map NT : T → (F ×)ℓ obtained by taking the norm on each summand. j . Similarly, there is a trace map Trt : t → F ℓ, Define T (o) = N −1 and we set t(o) = Tr−1 subalgebra and k-torus: t♭ ⊂ t(o) is the k-linear span of the χj's and T ♭ def= (t♭)× ⊂ T (o). Of course, T ♭ ∼= (k×)ℓ and t♭ ∼= kℓ. (o) = Qj oj. We also define a finite-dimensional k-toral T (o×)ℓ = Qj o× t We will be concerned with tori which are compatible with a given parahoric subgroup in the sense that T (o) ⊂ P or equivalently t(o) ⊂ P. Lemma 3.1. If T (o) ⊂ P , then T ∩ P i = T (o) ∩ P i and t ∩ Pi = t(o) ∩ Pi for all i ≥ 0. Proof. It suffices to show that T ∩ P = T (o) and t ∩ P = t(o); moreover, the first statement follows from the second by taking units. Since the central primitive idempotents χj are contained in t ∩ P, it is enough to check that if xχj ∈ Ej ∩ P, then x ∈ oj. Suppose x /∈ oj, so that xχj = ωq j and q < 0. Since j χj ∈ P ⊂ P, we see that ωq o× j ∈ P. We deduce that tsχj ∈ P for all s ∈ Z, which is absurd. (cid:3) j f for some f ∈ o× j ∈ P. This implies that tqχj = ωqm Definition 3.2. A uniform stratum (P, r, β) is called regular if there exists a max- imal torus T (possibly non-split) with the following properties: • T (o) ⊂ P ; • ¯T i = Z i(βν ) for all i; • yβ ∈ ¯P (defined as in (2.6)) is semisimple; • in the case r = 0 (and thus ep = 1), the eigenvalues of ¯βν ∈ gl( ¯L0) are distinct modulo Z. We say that T centralizes (P, r, β). If T ∼= E× for some field extension E/F , the stratum is called pure. Remark 3.3. Suppose that (P, r, β) is a regular stratum centralized by T , and L is a lattice with P ⊂ GL(L). Then, for any g ∈ GL(L), (gP g−1, r, Ad∗(g)β) is a regular stratum centralized by gT g−1. Remark 3.4. If T centralizes a regular stratum (P, r, β), then any conjugate of T by an element of P 1 also centralizes (P, r, β). Thus, T is not unique. It will be useful to have a variation of Definition 3.2 in terms of the graded action of t on β. Define zi(βν ) ⊂ ¯Pi to be the image of {z ∈ Pi ad(z)(βν) ∈ P−r+i+1}. Proposition 3.5. Let P be a uniform parahoric, and let (P, r, β) be a regular stratum centralized by the torus T . Then ¯T i = Z i(βν) if and only if ¯ti = zi(βν) for each i ≥ 0. Proof. First, we take i = 0. Suppose ¯t0 = z0(βν ). Given z ∈ P , it is clear that zβνz−1 − βν ∈ P−r+1 if and only if ad(z)βν ∈ P−r+1. This immediately gives ¯T 0 ⊂ Z 0(βν ). It also implies that if zP 1 ∈ Z 0(βν ), then z + P1 ∈ z0(βν). By assumption, this means that there exists s ∈ t(o) such that z − s ∈ P1, and we obtain s ∈ P ∩ t(o) = T (o). Consequently, s−1z ∈ P 1, i.e., zP 1 ∈ ¯T 0. Next, suppose that Z 0(βν ) = ¯T 0. Recall that a finite-dimensional k-algebra is spanned by its units. (Let A be such an algebra with Jacobson radical J. Since MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 17 1 + x ∈ A× for x ∈ J, J is in the span of A×. Moreover, a ∈ A× if and only if ¯a ∈ (A/J)×. The result now follows because A/J is a product of matrix algebras, and hence is spanned by its units.) We show that (z0(βν))× = (¯t0)×. Suppose y ∈ (z0(βν ))× ⊂ ¯P×. Since P1 is the Jacobson radical of P, any z ∈ P lifting y is invertible, hence lies in P . The argument above shows that zP 1 ∈ Z 0(βν), so we can assume z ∈ T , i.e., y ∈ (¯t0)×. A similar argument (using the fact that a lift of y ∈ (¯t0)× to t(o) actually lies in T (o)) gives the reverse inclusion. We conclude that z0(βν ) = span(cid:0)(z0(βν ))×(cid:1) = span(cid:0)(¯t0)×(cid:1) = ¯t0. Now suppose i > 0. There is an isomorphism ¯Pi → ¯P i induced by X 7→ 1 + X. Since Ad(1 + X)(βν) ∈ βν + ad(X)(βν) + P−r+i+1 for X ∈ Pi, it is clear that this map restricts to give an isomorphism between zi(β) and Z i(β). Since this same map takes ¯ti to ¯T i, the proof is complete. (cid:3) Remark 3.6. If T centralizes (P, r, β), then in fact ¯ti = zi(βν ) for all i ∈ Z. For i ≥ 0, this has been shown in the proposition. On the other hand, if i < 0 and s is any integer such that i + seP ≥ 0, then the result follows because multiplication by ts induces isomorphisms ¯ti ∼= ¯ti+seP (βν) and zi ∼= zi+seP (βν ). Since ¯βν ∈ z−r(βν ), βν ∈ t ∩ P−r + P−r+1; it follows that we can always choose the representative βν ∈ t ∩ P−r. Corollary 3.7. Let (P, r, β) be a regular stratum centralized by T and let X ∈ Pℓ. If βν ∈ t−r is a representative for β, and ad(X)(βν) ∈ P−r+j, then X ∈ tℓ + Pj. Proof. When ℓ ≥ j, there is no content. We note that the case j = ℓ + 1 follows from Proposition 3.5 and Remark 3.6. By induction on j > ℓ, suppose that the statement is true for j − 1. Let Y ∈ Pj−1 satisfy X − Y ∈ t. Then, X ∈ tℓ + Pj if and only if Y ∈ tj−1 + Pj. The latter statement follows from the base step above. (cid:3) The main goal of this section is to give a structure theorem for regular strata. Theorem 3.8. Let (P, r, β) be a regular stratum. (1) If (P, r, β) is pure, then eP = dim V . (2) (a) If (P, r, β) is strongly uniform, then it splits into a direct sum of pure strata (necessarily of the same dimension). (b) If (P, r, β) is not strongly uniform, then eP = 1 and (P, r, β) is the di- rect sum of a regular, strongly uniform stratum and a non-fundamental stratum of dimension 1. In each case, the splitting coincides with the splitting induced by a P 1- conjugate of T . Corollary 3.9. Let (P, r, β) be a regular stratum centralized by T . Take E/F to be the unique (up to isomorphism) field extension of degree eP . Then, T ∼= (E×)n/eP . Moreover, the maps T ♭ → ¯T 0 and t♭ → ¯t0 are isomorphisms. Remark 3.10. We note that if (P, r, β) is not strongly uniform, Theorem 3.8 implies that eP = 1. By the corollary, this can only happen when T is totally split. The following proposition allows us to make sense of what it means for an element βν to have regular semisimple leading term. 18 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Proposition 3.11. If (P, r, β) is regular, then every representative βν for β is regular semisimple. By Remark 3.6, we may choose βν ∈ t ∩ P−r. Corollary 3.9 implies that βν is a block diagonal matrix with entries in F [E]×. Then, the leading term β′ ν is the matrix consisting of the degree −r/eP terms from each block diagonal entry in βν (after identifying E with t1/eP ). It suffices to check that βν − β′ ν ∈ P−r+1, in which case Proposition 3.11 implies that β′ If eP = 1, this is clear. When eP > 1, we may assume without loss of generality that the splittings for T and (P, r, β) are induced In particular, the jth block-diagonal entry βjν is a by the same splitting of V . representative for the jth summand of (P, r, β). Therefore, by Theorem 3.8, we may reduce to the case where (P, r, β) is pure and eP = dim V . In this case, E generates P1, so it is clear that βν − β′ ν is regular semisimple. ν ∈ P−r+1. We call a maximal torus uniform if it isomorphic to (E×)ℓ for some field exten- sion E. Given a fixed lattice L and a uniform maximal torus T with T (o) ⊂ GL(L), we can associate a corresponding parahoric subgroup PT,L ⊂ GL(L) containing T (o) as follows. The isomorphism t ∼= Eℓ induces splittings V = ⊕Vj and L = ⊕Lj. Lemma 2.4 states that there is a unique complete lattice chain (Li j)i∈Z in Vj up to indexing; we normalize it so that L0 j = Lj. Let LT be the lattice chain with Li = ⊕Li j, and let PT,L be its stabilizer. Since L0 = L, we have PT,L ⊂ GL(L) as desired. It is obvious that T (o) ⊂ PT,L. Note that ePT ,L = n/ℓ = [E : F ]. t; the ith graded piece is given byL ki an [E : F ]th root of t. We denote the corresponding filtration by t(i) =L oEi Given a uniform torus T , there is a canonical Z-grading on its Cartan subalgebra Eχi, where E is a uniformizer in E which is Eχi. There is a corresponding canonical N-filtration on T (o) given by T (0) = T (o) and T (i) = 1 + t(i). Proposition 3.12. Let L be a fixed lattice, and let T ∼= (E×)ℓ be a uniform maximal torus with T (o) ⊂ GL(L). (1) If P ⊂ GL(L) is a parahoric subgroup for which there exists a regular stratum (P, r, β) centralized by T , then P = PT,L. (2) Let r ≥ 0 satisfy (r, n/ℓ) = 1, and suppose x ∈ t−r has regular semisimple leading term. Then, there is a unique regular stratum (P, r, β) with P ⊂ GL(L) which has x as a representative. (3) The canonical filtrations on t and T (o) coincide with the filtrations induced by PT,L, i.e., t(i) = t ∩ Pi T,L and T (i) = T ∩ P i T,L. Proof. Let (P, r, β) be a regular stratum as in the first statement. Since P ⊂ GL(L), we may take L0 = L in the corresponding lattice chain L . By Corollary 3.9, t ∼= En/eP , so eP = n/ℓ. Theorem 3.8 implies that the splitting V = ⊕Vj induced by t ∼= En/eP splits (P, r, β) into a direct sum of pure strata when eP > 1 and a sum of one-dimensional strata (with at most one non-fundamental summand) when eP = 1. In either case, each Vj is an E-vector space of dimension one, and Li = ⊕(Li ∩ Vj). However, Lemma 2.4 states that there is a unique complete lattice chain (Li j = L ∩ Vj. By definition, L = LT , so P = PT,L. j)i∈Z in Vj up to indexing, and we know that L0 The uniqueness part of the second statement is now immediate. For existence, it E ) ∈ Eℓ ∼= t determines is clear from the construction of PT,L that (a1−r E , . . . , aℓ−r MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 19 a regular stratum with parahoric subgroup PT,L if ai 6= aj whenever i 6= j and r is coprime to n/ℓ. Since the leading term of x is regular semisimple, we obtain a regular stratum (PT,L, r, β) with the leading term of x, and hence x itself, as a representative. Finally, tVj T,L = T,L is an immediate consequence when i ≥ 1. The i = 0 case is obtained by taking units in t(0) = t ∩ PT,L. Eχj = t(i). The fact that T (i) = T ∩ P i ) = L oEi L(tVj ∩ Pi j . We deduce that t ∩ Pi It follows that Li ∼= E. j = j EL0 TVj ,L0 j (cid:3) If (P, r, β) is a regular stratum centralized by T , then the proposition shows that {ti} and {T i} are actually the canonical filtrations. Remark 3.13. Given a fixed F -isomorphism V ∼→ F n, we can choose a standard representative of each conjugacy class of uniform maximal tori. Indeed, if the torus is isomorphic to (F ((t1/e))×)n/e, then under the identification GL(V ) ∼= GLn(F ), we can choose a block diagonal representative T (and t) with each uniformizer t1/e mapping to the e × e matrix I from (2.2) in the corresponding block. In this case, PT is the standard uniform parahoric subgroup that is 'block upper-triangular modulo t'. 3.2. Lemmas and proofs. We now give proofs of the results described above. We also include some lemmas that will be needed later. We remark that this section is largely technical in nature. Lemma 3.14. The homomorphism T ♭ → ¯T 0 is an injection, and if U is the unipo- tent radical of ¯T 0, then the induced map T ♭ → ¯T 0/U is an isomorphism. Similarly, the map t♭ → ¯t0 is an injection which induces an isomorphism t♭ ∼= ¯t0/n, where n ⊂ ¯t0 is the Jacobson radical. Proof. It is immediate from the definitions that T ♭ ⊂ P and T ♭ ∩ P 1 = {1}, so T ♭ → ¯T 0 is injective. Moreover, the unipotent radical U of ¯T 0 is the image of Qj(1 + pj), whence the isomorphism T ♭ → ¯T 0/U . A similar proof works for t♭. (cid:3) Lemma 3.15. Suppose that P is a uniform parahoric and (P, r, β) is a non- fundamental stratum in an F -vector space V . If (P, r, β) satisfies the first three conditions of Definition 3.2 and gcd(r, eP ) = 1, then V must have dimension one. ν ∈ P−rm+1. Without loss of generality, we may assume βm Proof. Since (P, r, β) is non-fundamental, it follows that there is a minimal m > 0 such that βm ν = 0. Indeed, after choosing a basis, Lemma 2.6 shows that we may take the representative βν to be the product of −r P with an element D ∈ h. By Remark 2.12, we may assume Ad(P )(h) ⊂ h. Therefore, (D−r for some D′ ∈ h. Since D′−rm ∈ P−rm+1, D′ ∈ P1 ∩ h = {0}. P )m = D′−rm P First, we claim that βν is regular nilpotent. Let z be the centralizer of βν in P. Note that z is a free o-module of rank equal to the dimension of the centralizer of βν in gl(V ), hence is at least n. Since Nakayama's lemma implies that rank(z) = dimk(z/tz), to show that βν is regular, it now suffices to show that dimk(z/tz) ≤ n. By Lemma 3.1, t0/teP = t(o)/tt(o), which clearly has k-dimension n. Also, recalling our convention that ¯zi is the projection of z in ¯Pi, we have ¯zi ⊂ zi(β). It P 20 CHRISTOPHER L. BREMER AND DANIEL S. SAGE follows that dimk(z/tz) ≤ eP −1Xi=0 eP −1Xi=0 dimk(zi(β)) = dimk(¯ti) = dimk(t(o)/tt(o)) = n as desired. Note that this argument actually shows that any coset representative βν is regular. Since the index of nilpotency of a regular nilpotent matrix is n, we have m = n. ν + P1 ∈ ¯P is nilpotent, yβ is semisimple only if ν ∈ P−reP +1, so n ≥ eP ≥ m, i.e., eP = n. Thus, In particular, since yβ = trβeP trβeP P = I is an Iwahori subgroup, and gcd(n, r) = 1. ν ∈ P1. This implies that βeP In the notation of Section 2.3, we write βν = x−r I where x = diag(x0, . . . , xn−1) ∈ d. Define σq(x) = (x−qr, x1−qr, . . . , xn−1−qr) to be the cyclic shift of the coeffi- cients of x by −qr places (with indexing in Zn). It is immediate from (2.3) that Ad(q I )(x) = σq(x). Therefore, ν = x Ad(−r βs I )(x) . . . Ad(−(s−1)r 6= 0. Thus, one of the components x′ )(x)−rs By assumption, βn−1 of x′ = xσ1(x) . . . σn−2(x) is non-zero; since gcd(r, n) = 1, x′ but one of the components of x. Moreover, βn conclude that exactly one component of x is equal to 0. ν = 0, so x′ ν I I = xσ1(x) . . . σs−1(x). j = xjxj−r . . . xj−(n−2)r j is the product of all j xj−(n−1)r = 0 and we Without loss of generality, assume that x0 = 0. Then, if ¯p ∈ Z 0(β), we may choose a representative p = diag(p0, . . . , pn−1) ∈ d∗. Equation (2.4) shows that pi = pi−r for all i; again, since gcd(r, n) = 1, this implies that p0 = p1 = · · · = pn−1. It follows that Z 0(β) has dimension 1. Lemma 3.14 now implies that T ♭ also has dimension 1, so T = E×, where E/F is a field extension of degree n. By Lemma 2.4, L is a saturated chain of oE-lattices, so may assume that I is a uniformizer in E. Applying (2.3), we see that ad(x)(I ) = x′I where x′ = diag(x0 − x1, . . . ). However, x0 = 0 and x1 6= 0 whenever n > 1, so E× only centralizes (P, r, β) when n = 1. (cid:3) Lemma 3.16. Let (P, r, β) be a regular stratum centralized by a torus T . If (V1, V2) is a splitting of (P, r, β), then (P1, r, β1) is regular, and (P2, r, β2) is either regular or non-fundamental. In the latter case, V2 has dimension 1 and eP = 1. Moreover, there exists p ∈ P 1 such that (V1, V2) splits the torus pT p−1 (which also centralizes (P, r, β)) into T1 × T2; here, T1 centralizes (P1, r, β1) and T2 centralizes (P2, r, β2) when this stratum is regular. 1(βν )×Z i 1(βν) × Z i j ; similarly, let zi j(βν) = Z i(βjν ), the centralizer of βjν in ¯P i Proof. Let Z i j (βν) = 2(βν), which is embedded in ¯P i by zi(βjν ). First, we claim that Z i(βν) = Z i diagonal blocks. It is clear that Z i 2(βν) ⊂ Z i(βν). Recall that ∂βν (resp. βν ) has trivial kernel in ¯M i (resp. ¯N i) by Definition 2.22 (resp. by Lemma 2.24). ∂′ If we identify ¯M i and ¯N i with the upper and lower off-diagonal components of ¯Pi, βν ) on ¯M i then δβν preserves each of these subspaces and restricts to ∂βν (resp. ∂′ 1 × ¯Pi (resp. ¯N i). Since δβν also preserves the diagonal blocks, its kernel lies in ¯Pi 2. We handle the cases i > 0 and i = 0 separately. When i = 0, Z 0(βν ) ⊂ ¯P ⊂ ¯P. It is clear that Z 0(βν) lies in the kernel of δβν . Therefore, Z 0(βν ) is supported on the diagonal blocks, so Z 0(βν) = Z 0 2 (βν ). When i > 0, there is an isomorphism ¯Pi → ¯P i induced by x 7→ 1+x. Moreover, Ad(1+x)(βν ) ∈ βν +ad(x)(βν )+Pi−r+1. Thus, Z i(βν )) must lie in 1 + ker(δβν ). The same argument as above implies that 1 (βν )×Z 0 MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 21 Z i(βν) is supported on the diagonal blocks and equals Z i one shows that zi(βν ) = zi 1(βν ) × zi 2(βν ). 1(βν ) × Z i 2(βν ). Similarly, 1 and ǫ′ 1(βν) × z0 j(βν) = ( ¯T ′ j(βν ). When i = 0, this is clear. 2. We claim that Z i j)i maps to Z i Let u be the Jacobson radical of z0(βν ). If ǫj ∈ P is the idempotent corresponding to the identity in Pj, its image ¯ǫj in ¯P lies in z0 j (βν ). Therefore, u = ¯ǫ1u + ¯ǫ2u, and the splitting z0(βν ) = z0 2(βν) induces a splitting on z0(βν)/u. Moreover, this splitting is non-trivial, since ǫj has non-trivial image in z0(βν )/u. Lemma 3.14 implies that t♭ ∼= z0(βν )/u, so t♭ is split. Let ǫ′ 2 be the idempotents corresponding to the identity in each summand of t♭. The same lemma implies that j ∈ ¯P is simply the identity matrix in the corresponding diagonal block; thus, ¯ǫ′ ǫ′ j ∈ ǫj + P1. These idempotents determine a splitting of t, and thus of T . Write j)i. Since ¯T i = Z i(βν ), it suffices to T = T ′ 1 × T ′ show that ( ¯T ′ In the case i > 0, jt ∩ Pi and ǫjPiǫj is the image jt ∩ Pi ⊂ ǫjPiǫj + Pi+1. Since (T ′ ǫ′ j)i maps to Z i of Pi j(βν ) as desired. j = ǫ′ 1 + ǫ2ǫ′ 2, j ) ⊂ Vj . The map p induces the identity map on ¯Li, and since the kernel of so p(V ′ p lies in ∩i∈ZLi = {0}, we deduce that p ∈ P 1. It is clear that pT p−1 centralizes (P, r, β) (indeed, this is true for any p ∈ P 1 by remark 3.4), and that (V1, V2) splits pT p−1 into a product T1 × T2. Moreover, yβ is semisimple if and only if yβ1 and yβ2 are, since yβ = tr(β1ν + β2ν )eP = trβeP 2ν = yβ1 + yβ2. The fact that T1 centralizes (P1, r, β1) follows from the previous paragraph, so (P1, r, β1) is regular. The first part of Remark 2.21 implies that P2 is a uniform parahoric with eP2 = eP . If (P2, r, β2) is fundamental, we conclude in the same way that it is regular and centralized by T2. Finally, if P2 is non-fundamental, (P2, r, β2) satisfies the conditions of Lemma 3.15. Thus, V2 has dimension 1, and moreover eP = eP2 = 1. (cid:3) j × 1) ∩ P i = 1 + ǫ′ j embedded as a diagonal block, we see that ( ¯T ′ 2 ) be the splitting of V determined by V ′ jV . Let p = ǫ1ǫ′ 1ν + trβeP Let (V ′ 1 , V ′ Lemma 3.17. If (P, r, β) is a pure stratum, then eP = n. Proof. Set m = n/eP , and assume that m > 1. Let T = E× be a torus centralizing (P, r, β). By Lemma 2.4, we can find a saturation LE = {Li E} of L that is stabilized E = Li, and let I be the Iwahori subgroup that by oE. We index LE so that Lmi stabilizes LE. We fix a uniformizer E for E; we can assume that P = m E . Recall that I1 , and furthermore j E = EI = IE by Proposition 2.3. Thus, j E ∈ P⌊ j m ⌋. By Proposition 3.5, ad(j E)(βν ) ∈ P−r+⌊ j ELi = Lmi+j E m ⌋+1. First, we show that βν ∈ I−rm. By assumption, βν(Lim E ) ⊂ Lim−rm E . Now, take 0 < j < m. We have βν(Lmi+j E ) = βν j E(Li) = (j Eβν − ad(j E)(βν ))Li ⊂ Lim+j−rm E , since ad(j E)(βν )Li ⊂ Li−r+1 = Lim+m−rm E . Thus, βν ∈ I−rm. We next show that ad(E)(βν) ∈ I−rm+2. The calculation above actually ) for any v ∈ L0 and 0 ≤ j < m. In Eβν v (mod L−rm+j+1 Ev ≡ j E showed that βν j particular, this gives Eβν j Ev ≡ j+1 E βνv ≡ βνEj Ev (mod L−rm+j+2 E ), for 0 < j < m−1, with the first congruence also holding for j = m−1. However, the second congruence is also true when j = m − 1; in this case, j+1 E = P , and the 22 CHRISTOPHER L. BREMER AND DANIEL S. SAGE ⊂ L−rm+(m−1)+2 congruence follows from ad(P )(βν )L0 ⊂ L−r+2 = L−rm+2m The congruences also hold trivially for j = 0, so ad(E)βν ∈ I−rm+2 as desired. E E . E ν ∈ βν + I−rm+1. Let β−rm = α−rm By Lemma 2.10, βν ∈ E + I−rm+1. Thus, there exists β′ ν ∈ I−rm ∩ E such , with α ∈ k×, be the homo- that β′ geneous degree −rm term of β′ ν, so that X = βν − β−rm ∈ I−rm+1. Clearly, ad(β−rm)(βν ) = ad(β−rm)(X). Moreover, since E centralizes (P, r, β), the remark after Proposition 3.5 shows that ad(β−rm)(X) ∈ P−2r+1. It follows that X and β−rm commute up to a term in P−2r+1, so ν = αeP 1 + eP trXβeP −1 −rm + higher order terms. (3.1) If X ∈ I−rm+j for 1 ≤ j < eP , then eP trXβeP −1 terms of (3.1) lie in I2j. such that X ∈ I−rm+j \ I−rm+j+1, and it now follows that N = trβeP I−rm+j \ I−rm+j+1. so yβ has Jordan decomposition yβ = trβeP semisimplicity of yβ, so X ∈ P−r+1. −rm ∈ Ij and the higher order In particular, if X /∈ P−r+1, there exists 1 ≤ j < eP ν − αeP ∈ It is obvious that ¯N ∈ ¯P is a nonzero nilpotent operator, ν = αeP 1 + ¯N . This contradicts the trβeP On the other hand, if X ∈ P−r+1, then ¯t0 = z0(βν) = z0(β−rm). By Lemma 2.10, z0(β−rm) is one-dimensional if and only if m = gcd(−rm, n) = 1, contradicting our assumption that m > 1. Hence, m = 1 and eP = n. (cid:3) Proof of Theorem 3.8. First, assume that (P, r, β) is strongly uniform. Suppose that we have a nontrivial splitting T = T1 × T2, with corresponding idempotents ǫj. Setting V1 = V 1×T2 and V2 = V T1×1, we show that (V1, V2) splits P and β at 1 ⊕ Li level r. Note that ǫj ∈ P, since ǫj ∈ t(o). Therefore, Li 2. By Remark 2.21, in order to see that (V1, V2) splits P and β, it suffices to show that ǫ1βν ǫ2 and ǫ2βνǫ1 are in P−r+1. j = ǫjLi, and Li = Li Note that ǫj ∈ T ♭; indeed, it is a (nonempty) sum of the primitive idempotents χi for t. By Lemma 3.14, a1ǫ1 + a2ǫ2 ∈ Z 0(βν ) for any a1, a2 ∈ k×. This implies that Ad(a1ǫ1 + a2ǫ2)(ǫ1βνǫ2) = a1 ǫ1βνǫ2 ≡ ǫ1βν ǫ2 (mod P)−r+1; accordingly, ǫ1βνǫ2 ∈ a2 P−r+1. Similarly, ǫ2βνǫ2 ∈ P−r+1. Let (Pj, r, βj) be the stratum corresponding to Vj. By Remark 2.21, each (Pj, r, βj) is strongly uniform. We next show that (P12, r, ∂β) is uniform. Using no- tation from the previous section, ∂β determines a map from ¯M j → ¯M j−r. It has al- ready been established in (2.5) that M = {M j} is uniform. It remains to show that ∂β has trivial kernel in ¯M j, say for all j ≥ 0. If x ∈ ker(∂β), then (β1)ν x ≡ x(β2)ν (mod M j−r+1). Therefore, Ad(1 + ι1xπ2)(βν ) ≡ βν (mod Pj−r+1), so 1 + ι1xπ2 ∈ Z j(βν ). However, (T1 × T2) ∩ (1 + ι1M jπ2) = 1, implying that x ∈ M j+1. We note that in the case r = 0, the eigenvalues of ¯βν are pairwise distinct modulo Z by Definition 3.2. A fortiori, the eigenvalues of ( ¯β1)ν are distinct from the eigen- values of ( ¯β2)ν modulo Z. We have thus shown that (P, r, β) is the direct sum of two strongly uniform strata, and Lemma 3.16 shows that these strata are regular (centralized by the Tj's). We can iterate this procedure until (P, r, β) is the direct sum of regular, strongly uniform strata each of which is centralized by a rank one torus, i.e., by the units of a field. Therefore, (P, r, β) splits into a sum of pure strata. Finally, suppose that (P, r, β) is not strongly uniform. When r > 0, Corol- lary 2.30 implies that (P, r, β) splits into a strongly uniform stratum (P1, r, β1) and a non-fundamental stratum (P2, r, β2). By Lemma 3.16, V2 has dimension one and eP = 1. When r = 0, Definition 3.2 implies that the kernel of ¯βν has dimension one MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 23 and that the non-zero eigenvalues of ¯βν are not integers. Write ¯L0 = ¯V1 ⊕ ¯V2, where ¯V2 = ker( ¯βν) and ¯V1 is the span of the other eigenvectors. It is easily checked that any lift of this splitting to L0 induces a splitting V = V1 ⊕ V2, and (V1, V2) splits (P, 0, β). (cid:3) Proof of Corollary 3.9. When (P, r, β) is strongly uniform, Theorem 3.8 states that (P, r, β) splits into a sum of pure strata (Pi, r, βi) with ePi = eP . Therefore, by Lemma 3.17, (Pi, r, βi) is centralized by a torus isomorphic to E×, and each It follows from Lemma 3.16 that T ∼= component Vi ⊂ V has dimension eP . (E×)n/eP . Otherwise, eP = 1 and (P, r, β) splits into a strongly uniform stratum (P1, r, β1) and a one-dimensional non-fundamental stratum (P2, r, β2). In particu- lar, by Lemma 3.16, this gives a splitting of a conjugate of T into T1 × T2, where T2 ∼= F ×. Since eP = 1, it follows from the theorem that T1 also splits into rank one factors. We now prove the last statement. By Lemma 3.14, we know that t♭ ∼= ¯t0/n, where n is the image in ¯P of Q pE. However, we have already seen that (E, . . . , E) generates P1, so n = {0}. The proof for T ♭ is similar. (cid:3) We can now prove Proposition 3.11. First, we need a lemma. Lemma 3.18. Let (P, r, β) be a regular stratum centralized by T , and suppose that βν ∈ t + P−r+m. Then, βν is conjugate to an element of t by an element of P m. Proof. Let E/F be a field extension of degree eP , and let E be a uniformizer in E. By Corollary 3.9, (P, r, β) splits into a sum of pure strata (Pi, r, βi), each of which is centralized by a torus isomorphic to E×. In particular, we can choose a block-diagonal representative β′ . Denote the summands of V by Vi. We may identify the (ℓ − j)th off-diagonal block with HomF (Vj , Vℓ). Let n ⊂ gl(V ) be the subalgebra of matrices in the (ℓ − j)th off-diagonal block corresponding to HomE(Vj, Vℓ). If aℓ = aj, then 1 + n centralizes βν. Since ((1 + n) ∩ P i)P i+1 * T iP i+1, this is a contradiction. Thus, the aj's are pairwise distinct. E ) ∈ Qi P−r E , . . . , an/eP −r ν = (a1−r E , a2−r i By Proposition 2.11, there exists Xm ∈ Pm such that βν −πt(βν ) ≡ ad(β′ ν )(Xm) ≡ ad(βν )(Xm) (mod P2−r). Therefore, Ad(1 + Xm)(βν ) ≡ πt(βν ) (mod P2−r). In- ductively, we can find Xj ∈ Pj+1 so that, setting pj = (1 + Xj)(1 + Xj−1) . . . (1 + Xm), Ad(pj)(βν ) ∈ t + Pj+1−r and pj ≡ pj−1 (mod P j−1). If we let p ∈ P 1 be the inductive limit of the pj's, we see that Ad(p)(βν ) ∈ t. (cid:3) Remark 3.19. We note that, in the argument above, it is not necessarily the case that βν is conjugate to πt(βν ). Proof of Proposition 3.11. It was shown in the proof of Lemma 3.15 that any rep- resentative βν is regular. To show that βν is also semisimple, it suffices to show that it is conjugate to an element of a Cartan subalgebra t. First, suppose (P, r, β) is strongly uniform. By Theorem 3.8 and Corollary 3.9, there exists a splitting V = V1 ⊕ · · · ⊕ Vn/eP with dim Vi = eP for each i and a block ν + P−r+1 and the (Pi, r, βi)'s diagonal β′ are pure strata. Moreover, by Lemma 3.16, we can choose a maximal torus T centralizing (P, r, β) such that the splitting of V induces a splitting T = T1 × · · · × Tn/eP , with Ti centralizing (Pi, r, βi). Since Ti is isomorphic to the units of the ν = (βiν ) ∈Qn/eP i=1 gl(Vi) such that βν ∈ β′ 24 CHRISTOPHER L. BREMER AND DANIEL S. SAGE field extension E/F of degree eP , Lemma 2.10 implies that βiν ∈ Ti ∩ P−r Therefore, βν ∈ t + P1−r. By Lemma 3.18, βν is conjugate to an element of t. i + P1−r i . If (P, r, β) is not strongly uniform, then eP = 1 by the second part of The- orem 3.8. As above, we can choose a splitting V = V1 ⊕ · · · ⊕ Vn, a diagonal representative β′ i=1 gl(Vi), and a compatibly split torus T which centralizes In (P, r, β). particular, βν ∈ t−r + P1−r, and Lemma 3.18 again implies that βν is conjugate to an element of t. (cid:3) In this case, dim Vi = 1 for all i, so t = Qn ν ∈Qn i=1 gl(Vi) and β′ ν ∈ t. We conclude this section with two lemmas that will be needed in Section 5. We recall from Remark 3.6 that if (P, r, β) is a regular stratum centralized by the maximal torus T , then one can choose βν ∈ t. We next show that if two such representatives are conjugate, then they are the same. Lemma 3.20. Suppose that (P, r, β) is a regular stratum. Choose representatives βν, β′ ν for some g ∈ GL(V ), then β′ ν ∈ t for β. If Ad(g)(βν ) = β′ ν = βν. Proof. By Proposition 3.11, βν is regular semisimple. Since Ad(g)(βν) ∈ t, g lies in the normalizer of T . Let w be the image of g in the relative Weyl group W = N (T )/T . It suffices to show that w is the identity. First, we show that Ad(g)(ti) ⊂ ti. Recall from Corollary 3.9 that t ∼=Qn/eP Therefore, ti consists of those (xj ) ∈ Qn/eP i=1 E, so t splits over E. Let ωj be a uniformizer of E supported on the jth summand of t as before; we have ωjPj = P1 j in the splitting determined by Theorem 3.8. joE. These are precisely the F -rational points of tE with eigenvalues of degree at least i/eP . Since the action of W permutes the eigenvalues, it follows that Ad(g)(ti) ⊂ (ti). i=1 E such that xj ∈ ωi Suppose that s ∈ t1−r. The previous paragraph shows that Ad(g)(βν +s) = β′ ν + Ad(g)(s) ∈ βν + t1−r. By induction on i, Ad(gi)(βν ) ∈ βν + t1−r, so Ad(gi)(βν ) is i=0 Ad(gi)(βν )) a representative for β. Let m be the order of w. Then, β′′ is a representative for β fixed by the action of w. Since β′′ ν is regular semisimple, w must be the identity. m (Pm−1 ν = 1 (cid:3) Lemma 3.21. Suppose that (P, r, β) is a regular stratum centralized by T and that βν ∈ t−r. Let A ∈ gl(V )∨ be the functional determined by βν and ν. Then, A determines an element Ai ∈ (Pi)∨ by restriction, and the stabilizer of Ai under the coadjoint action of P i is given by T iP r+1−i whenever r ≥ 2i, and P i whenever r < 2i. Proof. Recall from Proposition 2.5 that (Pi)⊥ = P1−i. Thus, Ad∗(p)(Ai) = Ai if and only if Ad(p)(βν) ∈ βν + P1−i. If r < 2i, then −r + i ≥ −i + 1. Therefore, since Ad(p)(βν ) ∈ βν + P−r+i for any p ∈ P i, P i lies in the stabilizer of Ai in this case. Suppose now that Ad∗(p)(Ai) = Ai and r ≥ 2i. The image of p in ¯P i must lie in Z i(βν ); therefore p = tp′ ∈ T iP i+1. Assume, inductively, that p = tq ∈ T iP j with j < r + 1 − i. Since Ad(t−1)(βν) = βν , q stabilizes Ai. Moreover, Ad(P j)(βν ) ⊂ βν + Pj−r; since j − r < 1 − i, the image of q in ¯P j lies in Z j(βν). Therefore, q ∈ T jP j+1 ⊂ T iP j+1. We conclude that p ∈ T iP r+1−i. When j = r + 1 − i, Ad(P j)(βν ) ⊂ βν + P1−i, so P j stabilizes Ai. It is now clear that T iP r+1−i stabilizes Ai. (cid:3) MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 25 Note that although the functionals Ai depend on the choice of βν ∈ t−r, their stabilizers do not. 4. Connections and Strata In this section, we describe how to associate a stratum to a formal connection. The local theory of irregular singular point connections is well understood; an elegant classification is given in [20, Theorem III.1.2]. The geometric theory of strata provides a Lie-theoretic interpretation of elements in the classical theory. In particular, the combinatorics of fundamental strata may be used to determine the slope of a connection, and the theory of strata makes precise the notion of the leading term of a connection with noninteger slope. Moreover, a split stratum induces a splitting on the level of formal connections. First, we recall some notation and basic facts. As before, k is an algebraically closed field of characteristic 0, o is the ring of formal power series in a parameter t, and F is the field of formal Laurent series. (1) DF (resp. Do) is the ring of formal differential operators on F (resp. o). dt and contains the Lie algebra DF is generated as an F -algebra by ∂t = d of k-derivations (i.e., vector fields) on F : Derk(F ) = F ∂t. F/k is the F ×-torsor of non-zero one forms in Ω1 F/k; if ω, ν ∈ Ω×, (2) Ω× ⊂ Ω1 then ω ν ∈ F × is the unique element such that ω ν ν = ω. (3) If ν ∈ Ω×, there is a unique vector field τν ∈ DF whose inner derivation takes ν to 1, i.e., ιτν (ν) = 1. For example, τ dt t = t∂t, and τdf = dt df ∂. (4) A connection ∇ on an F -vector space V is a k-linear derivation ∇ : V → V ⊗F Ω1 F/k. The connection ∇ gives V the structure of a DF -module: ξ ∈ Derk(F ), then ξ(v) = ∇ξ(v) def= ιξ(∇(v)). if v ∈ V , and (5) A connection ∇ is regular singular if there exists an o-lattice L ⊂ V with o/k(1). Equivalently, if ν ∈ Ω× has order the property that ∇(L) ⊂ L ⊗o Ω1 −1, ∇τ (L) ⊂ L. Otherwise, ∇ is irregular. (6) Suppose V has dimension n. Let V triv = F n be the trivial vector space with standard basis. If we fix a trivialization φ : V ∼→ V triv, then ∇ has the matrix presentation (4.1) ∇ = d + [∇]φ where [∇]φ ∈ gln(F )⊗F Ω1 F/k. The space of trivializations is a left GLn(F )- torsor, and multiplication by g changes the matrix of [∇]φ by the usual gauge change formula (4.2) g · [∇]φ := g[∇]φg−1 − (dg)g−1. Thus, [∇]gφ = g · [∇]. We note that the matrix form of ∇τ is given by [∇τ ]φ = ιτ ([∇]φ), with gauge change formula g · [∇τ ]φ := g[∇τ ]φg−1 − (τ g)g−1. We will drop the subscript φ whenever there is no ambiguity about the trivialization. 26 CHRISTOPHER L. BREMER AND DANIEL S. SAGE 4.1. Strata contained in connections. Let ∇ be a connection on an n-dimensional F -vector space. Fix ν ∈ Ω×, and set τ = τν . Suppose that L is a lattice chain with the property that ∇τ (Li) ⊂ Li−r−(1+ord(ν))eP . We define gri(∇τ ) to be the following map induced by ∇τ : ¯Li+j → ¯Li+j−r−(1+ord(ν))eP . gri(∇τ ) : eP −1Mj=0 eP −1Mj=0 By Lemma 2.6, this determines an element of ¯P−r−(1+ord(ν))eP . Equivalently, if we fix a trivialization φ of Li (viewed as a trivialization of V taking Li to on), then gri(∇τ ) equals the image of φ−1[∇τ ]φ in ¯P−r−(1+ord(ν))eP . Definition 4.1. We say that (V, ∇) contains the stratum (P, r, β) if P stabilizes the lattice chain (Li), ∇τ (Li) ⊂ Li−r−(1+ord(ν))eP for all i, and grj(∇τ ) = ¯βν ∈ ¯P−r−(1+ord(ν))eP for some j. Proposition 4.2. The stratum (P, r, β) is independent of ν ∈ Ω×. Proof. Take ν′ = f ν for some f ∈ F ×, so that τ ′ = τν ′ = 1 f τν . Since ∇τ ′ = f ∇τ , it is clear that ∇τ ′(Li) ⊂ Li−r−(1+ord(ν ′))eP if and only if the analogous 1 inclusion holds for ν; in addition, gri(∇τ ′) = 1 f gri (∇τ ). On the other hand, if βν ∈ P−r−(1+ord(ν))eP is a representative for the functional β on ¯Pr, then hβν , Xiν = hf −1βν, Xiν ′ for all X ∈ Pr implies that one can take βν ′ = 1 f βν . Hence, gri(∇τ ′) = ¯βν ′. (cid:3) For the rest of Section 4, we will fix ν ∈ Ω× of order −1. Proposition 4.3. Suppose that r ≥ 1. Then, the coset in ¯P−r determined by grℓ(∇τ ) under the isomorphism (2.1) is independent of ℓ. Proof. The maps grℓ(∇τ ) and gr0(∇τ ) determine the same element on ¯P−r when they "coincide up to homothety". More precisely, we must show that if 0 ≤ i < eP and ℓ ≤ i + jeP < ℓ + eP , then ∇τ (tjv) ≡ tj∇τ (v) (mod Li+jeP +1) for all v ∈ Li. By the Leibniz rule, ∇τ (tj v) = τ (tj) tj ∈ o, so for r ≥ 1, τ (tj ) (cid:3) tj (tjv) + tj∇τ (v). However, τ (tj ) tj (tjv) ∈ Lℓ+jeP ⊂ Lℓ+jeP −r+1 as desired. In other words, if r ≥ 1, the grℓ(∇τ )'s determine a unique coset gr(∇τ ) ∈ ¯P−r; viewed as a degree −r endomorphism of gr(L ), gr(∇τ )(¯x) = grℓ(∇τ )(¯x) for x ∈ ¯Li and any ℓ with ℓ ≤ i < ℓ + eP . The following lemma will be used in Sections 4.3 and 5. Lemma 4.4. If P ⊂ GLn(o) is a parahoric subgroup, then τ (Pℓ) ⊂ Pℓ. Moreover, if p ∈ P , then τ (p)p−1 ∈ P1. Proof. Let L be the lattice chain stabilized by P such that L0 = on. We may choose a basis e1, . . . , en for on that is compatible with L (as in Remark 2.9). It is clear that τ (Lj) ⊂ Lj for any j, since τ (ei) ⊂ ton. Therefore, if v ∈ Lj and X ∈ Pℓ, τ (X)v = τ (Xv) − Xτ (v) ∈ Pj+ℓ. It follows that τ (X) ∈ Pℓ. By Lemma 2.6, there exists H ⊂ GLn(k) for which P = P 1 ⋉ H. Hence, it suffices to prove the second statement for p ∈ P 1. Since P 1 is topologically unipotent, exp : P1 → P 1 is surjective. If p = exp(X), we obtain τ (exp(X)) exp(−X) = τ (X) exp(X) exp(−X) = τ (X) ∈ P1. MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 27 (cid:3) We now investigate the strata contained in a connection. Lemma 4.5. Every connection (V, ∇) contains a stratum. Proof. Take any lattice L ⊂ V with stabilizer P , and let L be the lattice chain (Li = tiL). It is obvious that ∇τ (L) ⊂ L−r for some r ≥ 0. The Leibniz rule calculation from the proof of Proposition 4.3 shows that ∇τ (Li) ⊂ Li−r, so (V, ∇) contains (P, r, β), where β corresponds to gr0(∇τ ). (cid:3) One of the standard ways to study irregular singular connections is to find a good lattice pair. The theory of good lattice pairs bears a superficial resemblance to the theory of fundamental strata. However, we will see that there are only a few cases in which there is a direct relationship between the two theories. Lemma 4.6. [9, Lemme II.6.21] Given a connection (V, ∇), there exist two o- lattices M 1 ⊂ M 2 ⊂ V with the following properties: (1) ∇(M 1) ⊂ M 2 ⊗o Ω1 (2) For all ℓ > 0, ∇ induces an isomorphism o/k(1) (4.3) ¯∇ : M 1(ℓ)/M 1(ℓ − 1) ∼=hM 2 ⊗o Ω1 o/k(ℓ + 1)i /hM 2 ⊗o Ω1 o/k(ℓ)i . The connection ∇ is regular singular if and only if M 1 = M 2. We call M 1 and M 2 a good lattice pair for (V, ∇). Remark 4.7. In the regular singular case, the data of a good lattice pair ( ¯∇, M 1, M 2) is equivalent to a strongly uniform stratum contained in (V, ∇): take L = (tiM 1)i∈Z, and β such that the image of ¯βν under the appropriate isomorphism (2.1) is ιτ ( ¯∇) = gr−1(∇τ ). However, it is not immediately possible to construct a fundamental stratum from a good lattice pair in general. Given a good lattice pair M 1 and M 2, one might naively construct a lattice chain L as follows. Set L0 = M 1. Choose s ∈ Z≥0 such that L0(s) ⊃ M 2 but L0(s − 1) + M 2. First, we suppose that M 2 = L0(s). Define L to be the In this case, eP = 1. Take β such that ¯βν = gr−1(∇τ ) as chain (Li = tiL0). above. Equation (4.3) implies that (V, ∇) contains (GL(L0), s, β), and this stratum is fundamental (in fact, strongly uniform). The naive generalization of the construction above does not necessarily produce a fundamental stratum. Set L1 = M 2(−s) + L0(−1). Since L0(s) ) M 2, it follows using Nakayama's Lemma that the map M 2(−s) → L0/L0(−1) is not surjective. We conclude that L0 ) L1 ) L0(−1). This extends to a lattice chain L with eP = 2. Finally, it is clear that there exists a minimal r ≥ 0 such that ∇τ (Li) ⊂ Li−r for i = 0, 1. The usual Leibniz rule argument shows that ∇τ (Li) ⊂ Li−r for all i. Choosing βν ∈ P−r whose coset corresponds to grj(∇τ ) (for some fixed j), we have the data necessary to give a stratum contained in (V, ∇). Notice that the stratum constructed above is not necessarily fundamental. For instance, suppose that [∇τ ] =(cid:18)0 t−3 0 (cid:19) 1 in V = V triv. Set M 1 = oe1 + oe2 and M 2 = p−3e1 + oe2. It is easy to check that this is a good lattice pair for (V, ∇), and our construction gives L0 = M 1 and 28 CHRISTOPHER L. BREMER AND DANIEL S. SAGE L1 = oe1 + pe2. However, the coset in ¯P−5 corresponding to gr0(∇τ ) contains the 0 0 (cid:1). nilpotent operator(cid:0) 0 t−3 In Theorem 2.16, we showed that a stratum is fundamental if and only if it can not be reduced to a stratum with smaller slope. The main goal of this section is to show that the slope of a connection is the same as the slope of any fundamental stratum contained in it. First, we define the slope of a connection. Fix a lattice L ⊂ V . If e = {ej} is a finite collection of vectors in V , we define v(e) = m if m is the greatest integer such that e ⊂ tmL. Take e to be a basis for V . An irregular connection (V, ∇) has slope σ, for σ a positive rational number, if the subset of Q given by {(cid:12)(cid:12)(∇i τ e) + σi(cid:12)(cid:12) i > 0} τ e = {∇i is bounded. Here, ∇i τ (ej)}. By a theorem of Katz [9, Theorem II.1.9], every irregular singular connection has a unique slope, and the slope is independent both of the choice of basis e and the choice of ν of order −1. We define the slope of a regular singular connection to be 0. Lemma 4.8. A connection (V, ∇) contains a stratum of slope 0 if and only if it is regular singular, i.e., has slope 0. Moreover, a regular singular connection contains a fundamental stratum with slope 0 and no fundamental stratum with positive slope. Proof. If (V, ∇) contains (P, 0, β), then there exists a lattice L ∈ L such that ∇τ (L) ⊂ L. Thus, ∇ is regular singular. Now, suppose that (V, ∇) is regular singular. Lemma 4.6 gives a lattice L = M 1 = M 2 for which ∇τ (L) ⊂ L. Thus, ∇τ preserves the lattice chain (Li = tiL), and the corresponding stratum has r = 0. Moreover, (4.3) implies that gr−1(∇τ )(L−1/L0) = L−1/L0, so the stratum is fundamental. Suppose that the same connection (V, ∇) contains a stratum (P, r, β) with r > 0. Let (Li) be the associated lattice chain. We will show that ¯β ∈ ¯P−r is a nilpotent operator on gr(L ). With the lattice L as above, choose i and m > 0 such that Li−mr+eP ⊃ L ⊃ Li. For 0 ≤ j < eP , we obtain (∇τ )m(Li+j ) ⊂ L ⊂ Li+j−mr+1. By Proposition 4.3, the endomorphism ¯βν coincides with each grℓ(∇τ ) on the latter's domain, so ¯βm ν ( ¯Li+j) = gri−(m−1)r(∇τ ) ◦ · · · ◦ gri(∇τ )( ¯Li+j ) = 0. Therefore, (P, r, β) is not fundamental. (cid:3) The following corollary describes the relationship between a fundamental stratum contained in (V, ∇) and its slope. Proposition 4.9. If (V, ∇) contains the fundamental stratum (P, r, β), then slope(∇) = r/eP . Proof. When (V, ∇) is regular singular, r/eP = 0 = slope(∇) by Lemma 4.8, so we assume that (V, ∇) is irregular singular. Let L = L0. Choose an ordered basis e for L that is compatible with L . We use the notation from the proof of Lemma 2.6: j=0 ej with Wj = span(ej) ⊂ Lj naturally isomorphic to Lj/Lj+1. For ν (w) + L−ir+j+1 by repeated application of Proposition 4.3. ν(ej) * L−ri+j+1 for e = FeP −1 ν /∈ P−ri+1. Therefore, βi w ∈ Wj, (∇τ )i(w) ∈ βi Since (P, r, β) is fundamental, βi some ej. It follows that ∇i τ ej * L−ri+j+1, and thus ∇i τ e * L−ri+eP . Let ⌈ ri eP ⌉ (resp. ⌊ ri eP ⌋) be the integer ceiling (resp. floor) of ri eP L−ri ⊆ t−⌈ ri eP ⌉L0. However, ∇i τ e * t−⌊ ri eP ⌋+1L0, since t−⌊ ri eP . Then, ∇i τ e ⊆ ⌋+1L0 ⊆ L−ri+eP . We MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 29 ⌉. In particular,(cid:12)(cid:12)(cid:12)v(∇i τ e) + ri (cid:3) eP(cid:12)(cid:12)(cid:12) < conclude that v(∇i 1, which proves the proposition. τ e) is equal to either −⌊ ri eP ⌋ or −⌈ ri eP We are now ready to state our main theorem on the relationship between slopes of connections and fundamental strata. In the context of the representation theory of local fields, the analogous theorem is due to Bushnell [7, Theorem 2]. Theorem 4.10. Any stratum (P, r, β) contained in (V, ∇) has slope greater or equal to slope(∇). Moreover, the set of strata contained in ∇ with slope equal to slope(∇) is nonempty and consists precisely of the fundamental strata contained in ∇. In particular, every connection contains a fundamental stratum. Proof. The regular singular case is dealt with in Lemma 4.8, so assume that (V, ∇) is irregular singular. By Proposition 4.9, any fundamental stratum contained in ∇ has slope equal to slope(∇). Now, let (P0, r0, β0) be any stratum contained in ∇. Assume that this stratum is not fundamental. We show that the stratum has a reduction (P1, r1, β1) with strictly smaller slope which is also contained in ∇. 0) and L1 = (Li By Theorem 2.16, there is a reduction of (P0, r0, β0) to (P1, r1, β′ 1) such that r1/eP1 < r0/eP0. Let L0 = (Li 1) be the lattice chains corresponding to P0 and P1 respectively. By definition, there is a lattice L ∈ L0 ∩ L1; reindexing the lattice chains if necessary, we can assume without loss of generality that L = L0 0 = L0 1. Choose a basis {e1, . . . , en} for L, and write ∇τ (v) = [∇τ ](v) + τ (v). In partic- ular, if v ∈ L, then τ (v) ∈ tL, since τ (f ej) = tf ′ej ∈ tL for f ∈ o. Thus, for any eP0 v ∈ Li 1, 0 ≤ ℓ < eP1), ∇τ (v) − [∇τ ](v) ∈ L (resp. 0 ). Therefore, [∇τ ] is a representative for both gr0 ∇τ (w) − [∇τ ](w) ∈ L 0(∇τ ) and gr0 0, 0 ≤ i < eP0 (resp. w ∈ Lℓ eP1 1 1(∇τ ). By Proposition 4.3, we conclude that [∇τ ] is a representative for β0. By definition of a reduction, 1)∨ be the corresponding functional, it is immediate that (P1, r1, β1) is also a reduction of (P0, r0, β0) with strictly smaller slope. If we let β1 ∈ ( ¯Pr [∇τ ] ∈ P−r1 1 . Finally, consider the collection of all strata contained in ∇; this set is nonempty by Lemma 4.5. Since eP ≤ n (where n = dim V , these slopes are all contained in 1 n! Z. Thus, the set of these slopes has a minimum value s > 0. (This value is positive by Lemma 4.8.) The argument given above shows that any stratum with slope s is fundamental, so ∇ contains a fundamental stratum and s = slope(∇). (cid:3) 4.2. Splittings. Recall that the connection (V, ∇) is split by the direct sum de- composition V = V1 ⊕ V2 if ∇τ (V1) ⊂ V1 and ∇τ (V2) ⊂ V2. In this section, we will show that any time a connection (V, ∇) contains a fundamental stratum (P, r, β) that splits, then there is an associated splitting of the connection itself. We note that, in the language of flat GLn-bundles, this corresponds to a reduction of struc- ture of (V, ∇) to a Levi subgroup. Lemma 4.11. Suppose that (V, ∇) contains a fundamental stratum (P, 0, β) that is split by (V1, V2). Then, there exists q ∈ P 1 and a fundamental stratum (P ′, 0, β′) contained in ∇ such that eP ′ = 1, P ′1 ⊂ P 1, and (qV1, qV2) splits both strata Proof. Let L be the lattice chain stabilized by P , and, without loss of generality assume that β is determined by gr0(∇τ ). Choose a trivialization for L0, and let i = tiL0) [∇τ ] ∈ P be the corresponding F -endomorphism. Now let L ′ = (L′ 30 CHRISTOPHER L. BREMER AND DANIEL S. SAGE with corresponding stabilizer P ′ = GL(L0). It is obvious that P ′1 ⊂ P 1. Setting β′ ∈ ( ¯P′)∨ equal to the functional induced by [∇τ ], we see that (P ′, 0, β′) is a stratum contained in (V, ∇) with eP ′ = 1. It will be convenient to denote [∇τ ] by βν or β′ ν depending on whether it is being viewed as a representative of β or β′. With this convention, it follows from Lemma 2.6 that there is a parabolic subalgebra q ⊂ gl( ¯L0) with Levi subalgebra h such that ¯β′ ν ∈ q and ¯βν is the projection of ¯β′ onto h. ν ν (resp. ¯∂′ νǫ2 ∈ P1 has non-trivial image in ¯P′. Since (P, 0, β) is fundamental, there exists 0 ≤ j < eP such that ¯βν ∈ End( ¯Lj) It follows that (P ′, 0, β′) is fundamental. The splitting V = is non-nilpotent. V1 ⊕ V2 for β does not necessarily split β′ at level 0: using the notation of Sec- tion 2.5, it is possible that ǫ1β′ Identify ¯M j with ǫ1Pjǫ2/ǫ1Pj+1ǫ2 and ¯N j with ǫ2Pjǫ1/ǫ2Pj+1ǫ1. By Definition 2.22 and Lemma 2.22, ¯∂β ′ ) is an automorphism of each ¯M j (resp. ¯M j). In particu- lar, there exists X1 ∈ ǫ1P1ǫ2 and Y1 ∈ ǫ2P1ǫ1 such that ad(X1)(β′ νǫ2+P2 ν )ǫj ∈ P2 and ad(Y1)(β′ for i 6= j. Continuing this process, we construct p = (1 + YeP −1)(1 + XeP −1) . . . (1 + ν )ǫj ∈ PeP for i 6= j. Since PeP ⊂ P′, this Y1)(1 + X1) ∈ P 1 such that ǫi Ad(p)(β′ implies that (V1, V2) splits Ad(p)(β′ ν at level 0. Clearly, (p−1V2, p−1V2) still splits (P, 0, βν ); q = p−1 will be the desired element of P 1. νǫ1 +P2. It follows that ǫi Ad(1+Y1) Ad(1+X1)(β′ ν) at level 0, so (p−1V1, p−1V2) splits β′ ν) ∈ −ǫ2β′ ν ) ∈ −ǫ1β′ β ′ ν Without loss of generality, we may assume that V1 and V2 split L ′ and β′ ν at ν projects to ¯βν ∈ h, it is clear that these matrices have the same level 0. Since ¯β′ eigenvalues (as do ¯βjν and ¯β′ jν for j = 1, 2), so (P ′, 0, β′) satisfies conditions (1) and (3) of Definition 2.22. By assumption (P1, 0, β1) and (P12, 0, ∂β) are strongly uniform. Since (Vi ∩ L ′) is a sub-lattice chain of Li, it is clear that (P ′ 1) is strongly uniform. It remains to show that the induced map ¯∂β ′ ν is an automorphism of End( ¯M ′). 1, 0, β′ Let F i be the image of M i ∩ (M ′)0 in ¯M ′. It is easy to see that (M ′)0 ⊂ M 1−eP and M eP ⊂ (M ′)1, implying that F eP = {0} and ¯M 0 = F −eP +1. Moreover, ν -invariant subspace of ¯M i. Since ∂β ′ (P12, 0, ∂β) is strongly uniform, ¯∂β ′ ν ∈ Aut( ¯M i) for all i, hence the restrictions to F i/F i+1 are also automorphisms. It follows that ¯∂β ′ ν gives an automorphism of ¯M ′. ν preserves the flag {F i}, so F i/F i+1 is a ¯∂β ′ (cid:3) Theorem 4.12. Suppose that (V, ∇) contains a fundamental stratum (P, r, β). Let (V1, V2) split (P, r, β). Then, there exists p ∈ P 1 such that (pV1, pV2) splits both (P, r, β) and ∇. The case when eP = 1 is well known (see [19, Lemma 2]). Proof. We first recall from Remark 2.23 that (pV1, pV2) splits (P, r, β) for any p ∈ P 1. Let V ′ = F n, and let V ′ = V ′ 2 be the standard splitting of F n into subspaces with dim V ′ j ∈ gln(F ) be the corresponding idempotents. By (4.1), (W1, W2) splits ∇ with dim Wj = dim Vj if and only if there exists a trivialization ψ : V → V ′ such that j = dim Vj. Let ǫ′ 1 ⊕V ′ (4.4) and Wj = ψ−1(V ′ j ). ǫ′ 1[∇τ ]ψǫ′ 2 = ǫ′ 2[∇τ ]ψǫ′ 1 = 0, MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 31 Since ∇ contains (P, r, β), there exist j such that grj(∇τ ) = ¯βν ; reindexing the lattice chain if necessary, we can assume that j = 0. Fix a trivialization φ such j . We set P′m = φPm ⊂ gln(F ) and similarly for that φ(L0) = on and φ(Vj ) = V ′ P ′m. Setting β′ ν (mod P′−r+1). By (4.2), it suffices to find h ∈ P ′1 such that h · [∇τ ]φ satisfies (4.4); then (pV1, pV2) splits ∇, where p = φ−1h−1φ ∈ P 1. ν = φβν , we have [∇τ ]φ ≡ β′ Inductively, we construct hm ∈ P ′1 such that hm ≡ hm−1 (mod P′m−1) and ǫ′ j ≡ 0 (mod P′−r+m+1) for i 6= j. The limit h = lim hm ∈ P 1 will i(hm · [∇τ ]φ)ǫ′ then satisfy (4.4). For m = 0, we can take h0 = I, since (V1, V2) splits (P, r, β). Now, suppose m ≥ 1 and we have already constructed hm−1. Let Q = hm−1 · [∇τ ]φ and Qij = ǫ′ 1 ∈ P ′m with X ∈ V12 and Y ∈ V ′ j, so that Q12, Q21 ∈ P ′−r+m. We will find g = I + ǫ′ 2 + ǫ′ 1Xǫ′ 2Y ǫ′ 21 satisfying iQǫ′ (4.5) The element hm = ghm−1 ∈ P 1 will then have the desired properties. (g · Q)12 ≡ (g · Q)21 ≡ 0 (mod P′−r+m+1). Given g as above, the gauge change formula g·Q = gQg−1−τ (g)g−1 immediately leads to the equation (cid:18) I X I (cid:19)(cid:18)Q11 Q12 Q21 Q22(cid:19) − τ (cid:18) 0 X 0(cid:19) = (cid:18)(g · Q)11 (g · Q)21 Y Y (g · Q)12 (g · Q)22(cid:19)(cid:18) I X I (cid:19) . Y Since XQ21 and Y Q12 lie in P′−r+2m ⊂ P′−r+m+1, the congruences (4.5) are equivalent to the system of congruences Q11 − (g · Q)11 ≡ 0 Q22 − (g · Q)22 ≡ 0 −τ X + XQ22 − (g · Q)11X + Q12 ≡ 0 −τ Y + Y Q11 − (g · Q)22Y + Q21 ≡ 0, (mod P′−r+m+1) where the first two automatically hold for any g of the given form. Suppose that r ≥ 1. In this case, τ X and τ Y are in P′m ⊂ P′−r + m + 1, so these terms drop out of the congruences. Substituting using the first two congruences, the problem is reduced to finding X and Y such that Q11X − XQ22 ≡ Q12 (mod P′−r+m+1) (mod P′−r+m+1). Q22Y − Y Q11 ≡ Q21 However, since Q ≡ β′ ν (mod P′−r+1), the first equation is equivalent to ∂β ′(X) ≡ Q12 (mod P′−r+m+1), and a solution X exists since (P12, r, ∂β) is strongly uni- form. Similarly, Lemma 2.24 guarantees the existence of a solution Y to the second equation. When r = 0, Lemma 4.11 shows that there exists q ∈ P 1 such that (qV1, qV2) splits a fundamental stratum ( P , 0, β) with P 1 ⊂ P 1 and e P = 1. We are thus in the classical case of lattice chains with period 1, and there exists q′ ∈ P 1 such that (q′qV1, q′qV2) splits ∇ by [19, Lemma 2]. The desired element of P 1 is thus given by p = q′q. (cid:3) 4.3. Formal Types. Suppose that (V, ∇) is a formal connection which contains a regular stratum. We fix a trivialization φ : V → F n. In this section, we will show that the matrix of (V, ∇) in this trivialization can be diagonalized by a gauge trans- formation into a uniform torus t ⊂ gln(F ). The diagonalization of [∇τ ] determines 32 CHRISTOPHER L. BREMER AND DANIEL S. SAGE a functional A ∈ (t0)∨ called a formal type, and any two connections on V with the same formal type are isomorphic. In the following, let (P, r, β) be a regular stratum in GLn(F ) with P ⊂ GLn(o), and let T ⊂ GLn(F ) centralize (P, r, β). We denote φ−1P φ ⊂ GL(V ) by P φ, and write βφ for the pullback of β to ( ¯Pφ)−r. Suppose that (V, ∇) contains (P φ, r, βφ), and that βφ is determined by gr0(∇τ ). (By Lemma 4.3, the second condition is superfluous when r > 0.) The goal of this section is the following theorem: Theorem 4.13. Fix ν of order −1. There exists p ∈ P 1 and a regular element Aν ∈ t−r such that p · [∇τ ]φ = Aν and Aν is a representative for β. Furthermore, the orbit of Aν under P 1-gauge transformations contains Aν +P1, and Aν is unique modulo t1. The obvious analogue of this theorem holds for an arbitrary ν. Remark 4.14. The above theorem implies that after passing to a ramified cover (specifically, the splitting field for T ), any connection containing a regular stratum is formally gauge equivalent to a direct sum of line bundles of slope less than or equal to r (with equality in all but at most one factor, with inequality only possible when e = 1). Moreover, the associated rank one connections have pairwise distinct leading terms. These properties could be used as an ad hoc way of defining the class of connections which are the primary topic of this paper. However, the perspective gained from our intrinsic approach via regular strata will prove essential below. The Lie-theoretic nature of this approach also suggests that it can be adapted to study flat G-bundles for G a reductive group. We also remark that Aν satisfies a stronger condition than regular semisimplicity. E + + . . . , with aj 6= 0 except possibly for a single j when d = 1. The ν = (a−r ) is a representative for βν , since the ν is regular semisimple, Suppose that Aν = (a1, . . . , an/d) ∈ En/d, where d = eP . Then, aj = aj,−r−r aj,−r+1−r+1 leading term A′ higher order terms lie in P−r+1. By Proposition 3.11, A′ and we see that Aν has regular leading term. n/d−r+1 1 −r+1 E , . . . a−r E E In the following definition, let T ⊂ GLn(F ) be a uniform maximal torus such that T (o) ⊂ GLn(o). We set P = PT,on as defined before Proposition 3.12. We also allow ν to have arbitrary order. Definition 4.15. A functional A ∈ (t0)∨ is called a T -formal type of depth r if (1) tr+1 is the smallest congruence ideal contained in A⊥; and (2) the stratum (P, r, β) is regular and centralized by T , where β ∈ ( ¯Pr)∨ is the functional induced by π∗ t (A) ∈ P∨. We denote the space of T -formal types of depth r by A(T, r) ⊂ (t0/tr+1)∨. A T -formal type is any element of A(T ) = ∪r≥0 A(T, r). We will always use the notation Aν for a representative of A in t−r−(1+ord(ν))eP . Remark 4.16. There is an embedding of A(T, r) into t−r−(1+ord(ν))eP /t1−(1+ord(ν))eP determined by the pairing h, iν. For simplicity, we only describe it when ν = dt t . First, recall that t has a natural grading so that t−r/t1 ∼= ⊕0 ¯ti. If r > 0, then A(T, r) is isomorphic to the open subspace of t−r/t1 with degree −r term regular. If r = 0, then t is split and t0/t1 ∼= t♭ ∼= kn. In this case, A(T, r) corresponds to P aiχi ∈ t♭ with the ai's distinct modulo Z. This is not a Zariski-open subset of t♭. However, if k = C, it is open in the complex topology. i=−r MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 33 formal types A and representatives of the form Aν =Pr To be even more explicit, assume that t is the block-diagonal Cartan subalgebra of gln(F ) as in Remark 3.13. If t is split (and r > 0), there is a bijection between i=0 t−iDi with Di ∈ gln(k) diagonal and Dr regular. In the pure case, there is a similar bijection between formal types and representatives Aν = q(−1 I ) where q ∈ k[X] has degree r. Throughout Section 5, we will assume that t has such a block diagonal embedding into gln(F ). Remark 4.17. An element of A(T, r) may also be viewed as a functional on P/Pr+1 (resp. gln(o)∨) for which the corresponding stratum (P, r, β) is regular and all of whose representatives lie in t−r−(1+ord(ν))eP + P1−(1+ord(ν))eP . The notion of a T -formal type actually depends only on the conjugacy class of T . Indeed, set L = on ⊂ F n. If T and S are conjugate tori with T (o), S(o) ⊂ GLn(o), then Lemma 4.18 below states that there exists h ∈ GL(L) such that hT = S. S,L and htj = sj for all j. Applying Remark 3.3, we conclude that Ad(h)(Aν ) ∈ s−r determines a regular stratum (PS,L, r, β′) centralized by S. It is evident that hPj T,L = Pj We say that a lattice L ⊂ V is compatible with T if T (o) ⊂ GL(L). Lemma 4.18. Suppose that T is a uniform maximal torus. (1) The set of lattices L that are compatible with T is a single N (T )-orbit. (2) If S is conjugate to T in GL(V ), and L is compatible with both S and T , then S is conjugate to T in GL(L). Proof. Suppose that L and L′ are compatible with T , and let g ∈ GL(V ) satisfy gL = L′. In particular, this implies that S = g−1T g is compatible with L. Choose x ∈ t−r with regular leading term. By Proposition 3.12, there exist parahoric subgroups PT,L, PS,L ⊂ GL(L) such that x and Ad(g−1)(x) determine regular strata (PT , r, β) and (PS, r, β′). By Theorem 3.8, ePT = ePS . The same theorem states that (PT , r, β) and (PS, r, β′) induce splittings of L, and it is easily checked that there exists an element of h ∈ GL(L) taking the components of the T -splitting to the S-splitting. Replacing g with gh, we may assume that the splittings induced by S and T are the same. Thus, we may reduce to the pure case. Suppose that (PT , r, β) and (PS, r, β′) are pure. We may choose h′ ∈ GL(L) such that h′PT (h′)−1 = PS, so by a similar argument, we may assume PT = PS = P . By (2.4), there exists p ∈ P such that Ad(p)(Ad(g−1)(x)) ∈ t + P−r+1. Finally, Lemma 3.18 implies that there exists p′ ∈ P 1 such that Ad(p′pg−1)(x)) ∈ t. It follows that p′pg−1 = n−1 ∈ N (T ). It is now clear that nL = L′, since p′ and p are in GL(L). We now prove the second statement. Suppose that S = gT g−1 for g ∈ GLn(F ). Then, L and gL are compatible with S. By the first part, gL = nL for some n ∈ N (S). It follows that there exists h = n−1g ∈ GL(L) such that S = hT h−1. (cid:3) We continue to fix T ⊂ GLn(F ) as in Definition 4.15. Definition 4.19. The set A(V,∇) ⊂ A(T ) of T -formal types associated to (V, ∇) consists of those A for which there is a trivialization φ : V → F n such that (V, ∇) contains the stratum (P φ, r, βφ) and the matrix [∇τ ]φ is formally gauge equivalent to an element of Aν + P1−(1+ord(ν))eP by an element of P 1. T By Theorem 4.13, the last statement is equivalent to the condition [∇τ ]φ is formally gauge equivalent to Aν . 34 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Proposition 4.20. Let ∇ be a connection containing a T -formal type AT . If S is a maximal torus with S(o) ⊂ GLn(o), then ∇ has an S-formal type if and only if S is GLn(F )-conjugate to T . Moreover, if h ∈ GLn(o) conjugates T to S, then Ad∗(h−1) gives a bijection from A(V,∇) to A(V,∇) . T S Proof. Set L = on. By Lemma 4.18, there exists h ∈ GLn(o) such that S = hT h−1. We may choose a trivialization φ : V → F n such that [∇τ ]φ = Aν ∈ t by Theorem 4.13. We now observe that, by Lemma 4.4, h · [∇τ ]φ ∈ Ad(h)(Aν ) + t gln(o) ⊂ s + t gln(o). After dualizing, Aν and Ad(h)(Aν ) determine functionals AT and AS (respectively) in gln(o)∨ and Ad∗(h−1)(AT ) = AS. Moreover, AS is an S formal type correspond- ing to (V, ∇): it is clear that any representative for AS lies in S = Ad(h)(T ) + P1 S, since t gln(o) ⊂ P1 S , r, β′). Here, β′ is the functional on Pr (cid:3) S, and (V, ∇) contains the regular stratum (P hφ determined by AS. S/Pr+1 S Before proving Theorem 4.13, we give two corollaries. Corollary 4.21. Suppose that Aν ∈ t is a representative of a T -formal type. If g ∈ GLn(F ) satisfies g · Aν = Aν, then g ∈ T ♭. Proof. We assume without loss of generality that ν = dt dt . Since g is invertible, it will suffice to show that g ∈ t♭. Choose a regular stratum (P, r, β) corresponding to A, and consider the exhaustive filtration t♭+Pi of gln(F ). Suppose that g /∈ t♭, and let ℓ be the largest integer such that g ∈ t♭ + Pℓ. By assumption, [g, Aν] = τ (g) ∈ Pℓ. t , so τ = t d First, assume that r > 0. Note that ℓ 6= 0, since g ∈ P implies that g ∈ t♭ + P1 by Proposition 3.5. Suppose ℓ < 0. Corollary 3.7 gives g = s + h with s ∈ tℓ and h ∈ Pℓ+r. Since πt([g, Aν]) ∈ tℓ+1 by Proposition 2.11(3), we also get πt(τ (s)) ∈ tℓ+1. Lemma 4.26 now gives s ∈ Pℓ+1, a contradiction. Hence, ℓ ≥ 1, and we have g = s0 + x with s0 ∈ t♭ and x ∈ Pℓ. Since [x, Aν ] = τ (x) ∈ Pℓ, we get the contradiction x ∈ Pℓ+1 by the same argument as in the ℓ < 0 case. When r = 0, we may assume Aν is a regular diagonal matrix in gln(k) satisfying In other words, − ad(Aν ) has no non-zero the last condition of Definition 3.2. integer eigenvectors in gln(k). Write g = tℓgℓ + tℓ+1gℓ+1 + . . . with gj ∈ gln(k). Then, [g, Aν] = τ (g) implies that − ad(Aν )(gj) = jgj. We deduce that gj = 0 except when j = 0. Moreover, since Aν is regular, [g0, Aν] = 0 implies that g0 ∈ t♭. Thus, g ∈ t♭. (cid:3) Corollary 4.22. Let A be a formal type. Any two connections with formal type A are formally isomorphic. Furthermore, the set of formal types associated to a connection is independent of choice of ν ∈ Ω×. Proof. Independence of ν follows by the argument given in Proposition 4.2 and the remark above. Thus, fix ν with order −1. Suppose (V, ∇) and (V ′, ∇′) have formal type A, and let φ (resp. φ′) be the given trivialization for V (resp. V ′). By Theorem 4.13, there exists Aν ∈ t−r and p, p′ ∈ P 1 such that p · [∇τ ]φ = Aν = p′ · [∇′ τ ]φ′. It is easily checked that the composition (φ′)−1 ◦ (p′)−1 ◦ p ◦ φ : V → V ′ takes ∇ to ∇′. (cid:3) MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 35 We begin the proof of Theorem 4.13. Throughout, we will suppress the fixed trivialization φ from the notation. We may assume that ν = dt dt . First, we show that if the result holds for the trivialization φ and the regular stratum (P, r, β) centralized by T , then, for any g ∈ GLn(o), it holds for the trivialization gφ, the regular stratum (gP, r, gβ), and its centralizing torus gT . t , so τ = t d Suppose that g ∈ GLn(o). By Lemma 4.4, τ (g)g−1 ∈ t gln(o). In particular, τ (g)g−1 ∈ P1. Therefore, if the theorem holds for (P, r, β), then there exists p ∈ P 1 such that p · [∇τ ] = Aν + Ad(g−1)(τ (g)g−1). It follows that gp · (g · [∇τ ]) = g · (p · [∇τ ]) = Ad(g)(Aν ). Thus, the first part of the theorem still holds after changing the trivialization by g. The second and third parts follows from a similar argument. Without loss of generality, we henceforth assume that t embeds into the d × d diagonal blocks of gln(F ) and in each diagonal block the matrix I from (2.2) is a uniformizer for the corresponding copy of E. By Theorem 3.8, this splitting of V splits (P, r, β) into pure strata, plus at most one non-fundamental stratum in the case eP = 1. Therefore, Theorem 4.12 shows that ∇ splits into a direct sum of connections containing a pure stratum when eP > 1 and into a direct sum of connections in dimension 1 when eP = 1. Moreover, the splitting for ∇ maps to the splitting determined by T by an automorphism p ∈ P 1. In other words, p · [∇τ ] lies inLn/d First, we consider the case eP = 1 (which includes the case r = 0). By the above discussion, we may reduce to the case where dim V = 1. In this case, [∇τ ] ∈ F and g · [∇τ ] ∈ [∇τ ] + p1 for all g ∈ 1 + p1. This proves the first statement and the statement about uniqueness. It suffices to show that the orbit of [∇τ ] under gauge transformations contains [∇τ ] + p1. Suppose X ∈ p1. Since τ : p1 → p1 and log : (1 + p1) → p1 are surjective, there exists g ∈ 1 + p1 such that τ (log(g)) = X. Therefore, g · ([∇τ ] + X) = [∇τ ], and the assertion follows. j=1 gl(Vj). When eP > 1, it suffices to prove the theorem in the case when (P, r, β) is pure. In particular, P = I is an Iwahori subgroup and T ∼= E×. Take βν = [∇τ ]. By Remark 3.6, βν ∈ t ∩ I−r + I1−r. The following two lemmas prove Theorem 4.13 in the pure case with eP > 1 and thus complete the proof of the theorem. Lemma 4.23. Let ψℓ be defined as in Section 2.3. When ℓ ≥ 1, τ (Iℓ) ⊂ Iℓ and ψℓ(τ (ℓ I )) 6= 0. Furthermore, I )−1 ≡ ατ (ℓ I ) (mod Iℓ+1) for any α ∈ k. Proof. Suppose that ℓ = qn + z, for 0 ≤ z < n. The matrix coefficients of ℓ I are (cid:2)τ (1 + αℓ I )(cid:3) (1 + αℓ I )ij = tq+1 tq 0 (ℓ if j = i + z − n; if j = i + z; if j 6≡ i + z (mod n). I ) = xℓ Let x be the diagonal matrix with xjj = q when j ≤ n − z and q + 1 otherwise. Then, τ (ℓ I . Moreover, τ (I) ⊂ τ (gln(o)) ⊂ t gln(o) ⊂ I. The Leibniz rule I I now imply that τ (Iℓ) ⊂ Iℓ for all ℓ ≥ 1. The first and the fact that Iℓ = ℓ assertion of the lemma follows, since ψℓ(τ (ℓ I )) is the trace of x, which is non-zero for ℓ 6= 0. 36 CHRISTOPHER L. BREMER AND DANIEL S. SAGE To see the second statement, observe that (1 + αℓ I )−1 = 1 − αℓ I + y, with y ∈ Iℓ+1. Therefore, (cid:2)τ (1 + αℓ I )(cid:3) (1 + αℓ I )−1 =(cid:2)τ (1 + αℓ I )(cid:3) (1 − αℓ = ατ (ℓ I )(1 − αℓ ≡ ατ (ℓ I ) I + y) (mod Iℓ+1). I + y) (cid:3) I ) by an element of I 1. If Bν ∈ q(−1 Lemma 4.24. Suppose that (V, ∇) contains the pure stratum (I, r, β) with n ≥ 2 (so r ≥ 1). Then, there is a unique q(x) ∈ k[x] such that [∇τ ] is formally gauge equivalent to q(−1 I ) + I1, then Bν is formally gauge equivalent to [∇τ ] by an element of I 1. Proof. By the remarks made before Lemma 4.23, [∇τ ] = βν = qr−r I + y with y ∈ I−r+1. Moreover, since βν /∈ I−r+1, qr 6= 0. We need to find p ∈ I 1 with the property (4.6) p · βν = q(−1 I ), for q ∈ k[x] as in the statement of the lemma. Inductively, we construct gℓ ∈ I 1 and qℓ ∈ k[x] of degree r such that gℓ ≡ gℓ−1 I ) + Iℓ−r. Moreover, I ) is unique modulo tℓ−r. Note that qℓ is independent of ℓ (mod Iℓ−1), deg(qℓ − qℓ−1) ≤ r − ℓ + 1, and gℓ · βν ∈ qℓ(−1 we will show that qℓ(−1 for ℓ > r + 1. If we set p = lim gℓ and q = qℓ for large ℓ, (4.6) is satisfied. We start by taking g1 = 1 and q1 = qrxr. Suppose that we have constructed gℓ r = qr. We will find g = 1 + X ∈ I ℓ such that gℓ+1 = ggℓ ∈ I 1 and qℓ; note that qℓ has the required properties. Obviously, gℓ+1 ≡ gℓ (mod Iℓ). To construct g, first, note that τ (g)g−1 = τ (X)g−1 ∈ Iℓ by Lemma 4.23. More- over, g−1 ≡ 1 − X (mod Iℓ+1). If ℓ − r ≤ 0, it suffices to find g ∈ I ℓ such that Ad(g)(gℓ · βν) ∈ t + I−r+ℓ+1. We see that Ad(g)(gℓ · βν ) ≡ (1 + X)(gℓ · βν )(1 − X) (mod Iℓ−r+1) ≡ gℓ · βν + qrδX (−r I ) (mod Iℓ−r+1). Thus, we need to solve the equation gℓ · βν + qrδX (−r I ) ≡ Y (mod Iℓ−r+1). for Y ∈ t. Since qr 6= 0, Proposition 2.11(3) implies that a solution for X exists if and only if Y ∈ πt(gℓ · βν ) + Iℓ−r+1. Letting qℓ+1(−1 I ) denote the terms of nonpositive degree in πt(gℓ · βν) (where qℓ+1 ∈ k[x]), we see that deg(qℓ+1 − qℓ) ≤ r − l. Moreover, qℓ+1 is uniquely determined. Now, suppose ℓ−r > 0. The first part of Lemma 4.23 implies that πt(τ (ℓ−r )) /∈ Iℓ−r+1. The argument above implies that we may choose s ∈ I ℓ with the property I Ad(s)(gℓ · βν) ≡ qℓ(−1 I ) + απtτ (ℓ−r I ) (mod Iℓ−r+1). for some α ∈ k. Again, Proposition 2.11 implies that there exists h ∈ I ℓ such that Ad(h)(qℓ(−1 I ) + ατ (ℓ−r I )) ≡ qℓ(−1 I ) + απtτ (ℓ−r I ) (mod Iℓ−r+1). Thus, by the second part of Lemma 4.23, Ad(h−1s)(gℓ · βν) ≡ qℓ(−1 I ) +(cid:2)τ (1 + αℓ−r I )(cid:3) (1 + αℓ−r I )−1 (mod Iℓ−r+1). MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 37 Since 1 + αℓ−r I follows that commutes with qℓ(−1 I ) and τ (h−1s)s−1h ∈ Iℓ ⊂ Iℓ−r+1, it (1 + αℓ I ) ·(cid:2)(h−1s) · (gℓ · βν)(cid:3) ≡ qℓ(−1 I ) (mod Iℓ−r+1). Setting gℓ+1 = (1 + αℓ I )h−1s and qℓ+1 = qℓ completes the induction. The same inductive argument (beginning with ℓ = r + 1) shows that for any I ). This completes I ) + I1, there exists h ∈ I ℓ such that h · Bν = q(−1 Bν ∈ q(−1 the proof of the second statement of the lemma. (cid:3) 4.4. Formal Types and Formal Isomorphism Classes. In this section, we describe the relationship between formal types and isomorphism classes of formal connections. In particular, we show that formal types are the isomorphism classes in the category of framed formal connections. This category is the disjoint union of the categories of T -framed formal connections as T runs over conjugacy classes of uniform maximal tori. Moreover, there is an action of the relative affine Weyl group of T on the set of T -formal types, and the forgetful functor to the category of formal connections sets up a bijection between orbits of T -formal types and isomorphism classes of formal connections containing a regular stratum of the form (PT , r, β). We also exhibit an intermediate category whose isomorphism classes correspond to relative Weyl group orbits. Given a conjugacy class of uniform maximal tori and a fixed lattice L, we can choose a representative T such that T (o) ⊂ GL(L). Setting P = PT,L, we will further have T (o) ⊂ P ⊂ GL(L) and T ∼= (E×)n/eP with E/F a degree eP ramified extension. Upon choosing a basis for L, we can assume without loss of generality that T (o) ⊂ P ⊂ GLn(o) and that T is the standard block diagonal torus described in Remark 3.13. Throughout this section, we will fix a form ν = dt t and the corresponding derivation τ = t d dt . Let WT = N (T )/T and W aff T = N (T )/T (o) be the relative Weyl group and the relative affine Weyl group associated to T . Note that W aff is a semi-direct product T ∼= WT ⋉ T /T (o). Furthermore, of WT with the free abelian group T /T (o), i.e., W aff T if we write Σn/eP for the group of permutations on the E×-factors of T and CeP for the Galois group of E/F , then WT ∼= Σn/eP ⋉ (Cn/eP ). Here, Σn/eP acts on Cn/eP by permuting the factors. We note that N (T ) ∩ GLn(o) ⊂ PT,on , since CeP eP and Σn/eP both preserve the filtration determined in Proposition 3.12. eP Any element of WT has a representative in GLn(k) ⊂ GLn(F ). Therefore, WT ∼= (N (T ) ∩ GLn(k))/T ♭. In fact, WT can be embedded as a subgroup of GLn(k) as follows. The centralizer of T ♭ in GLn(k) is a Levi subgroup isomorphic i=1 GLeP (k). toQn/eP Let Di (resp. di) denote the diagonal subgroup (resp. subalgebra) in each component. Fix a primitive eth P root of unity ξ. We view Σn/ep as the subgroup of permutation matrices that permute the factors of this Levi subgroup while the ith copy of CeP maps to the cyclic subgroup of Di generated by diag(1, ξ, ξ2, . . . , ξeP −1). T on A(T, r). Taking w ∈ GLn(k) a representa- tive for wT ∈ WT , s = (s1, . . . , sn/eP ) ∈ T , and A ∈ (t0/tr+1)∨, we obtain actions We now define an action of W aff 38 CHRISTOPHER L. BREMER AND DANIEL S. SAGE of WT and T (F )/T (o) on (t0/tr+1)∨ via (wT )(A) = Ad∗(w)(A) (sT (o))(A) = A − n/ePXi=1 degE si eP χ∨ i . is the functional induced by χi t , where χi is the identity of the ith Here, χ∨ i component of t. It is easy to see that these two actions give rise to a unique action of W aff T . dt i=1 degE si W aff T x −Pn/eP To check that this action restricts to an action on A(T, r), consider the action of on t defined by the similar formulas ν (wT )(x) = Ad(w)(x) and ν(s)(x) = χi. The induced action on t−r/t1 corresponds to under the isomorphism t−r/t1 ∼= (t0/tr+1)∨ determined by ν = dt t , and for any w ∈ W aff T , ν( w)(Aν ) is a representative for ρ( w)(A). If A ∈ A(T, r), then the leading term of Aν is regular, with distinct eigenvalues modulo Z when r = 0. If r > 0, it is clear that ν ( w)(Aν ) also has regular leading term. If r = 0, then the action permutes and adds integers to the eigenvalues, so again the condition for being a formal type is preserved. eP Proposition 4.25. Suppose that g ∈ N (T ) and A ∈ A(T, r). If Aν ∈ t is a representative for A, then g · Aν is formally gauge equivalent to ν(gT (o))(Aν) by an element of P 1. Furthermore, ν (gT (o))(Aν) ∈ πt(g · Aν ) + t1. Proof. The case when r = 0 is easily checked since T is the usual split torus, so we assume that r > 0. First, consider the case g = s ∈ T , so s · Aν = Aν − (τ s)s−1. Recall that the intersection of P with the block-diagonal Levi subgroup is a product of Iwahori subgroups Ii ⊂ GLeP (F ). Each Pi determines an ordering on the roots of Di. We take Hi ∈ di to be the half sum of positive coroots, and H = (H1, . . . , Hn/eP ) ∈ P. Lemma 4.26. Suppose that s ∈ tr. If H is defined as above, then Moreover, πt(τ (s)) ∈ r eP ad(H)(s) − r eP s ∈ P1+r. τ (s) + 1 eP s + P1+r. Proof. The first statement follows from the observation τ (i We then obtain the second statement from Proposition 2.11(3). E) = i eP i E− 1 eP ad(Hj)(i E). (cid:3) Setting s = (s1, . . . , sn/eP ), the lemma gives (τ s)s−1 ∈ degE si eP χi − 1 eP ad(H)(s)s−1 + P1. n/ePXi=1 Observe that each term on the right of this expression lies in P. Applying Propo- sition 2.11, we obtain X ∈ Pr such that ad(X)(Aν) ∈ πt((τ s)s−1) − (τ s)s−1 + P1. Taking h = 1 − X, we see that h · (s · Aν) ∈ πt(s · Aν ) + P1. By Theorem 4.13, it follows that s · Aν is gauge equivalent to πt(s · Aν). Since πt is a t-bimodule map, we deduce from Lemma 2.10 that πt(ad(H)(s)s−1) ∈ t1. Therefore, πt(Aν − (τ s)s−1) ∈ ν(sT (o))(Aν) + t1. Note that if s ∈ T (o), MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 39 Lemma 4.4 implies that (τ s)s−1 ∈ P1; thus, T (o) does not affect the P 1-gauge equivalence class. For the general case, take g = sn with n ∈ N (T ) ∩ GLn(k) and s ∈ T . The result now follows by applying the case above to the formal type ν(nT (o)(Aν ) = Ad(n)(Aν ) = n · Aν = πt(n · Aν ) ∈ A(T, r). (cid:3) ν ∈ t. If Aν ν are GLn(F )-gauge equivalent modulo t1, then there exists a unique w ∈ W aff T Lemma 4.27. Suppose that A, A′ ∈ A(T, r) with representatives Aν , A′ and A′ such that ( w)(A) = A′. Proof. Since W aff acts freely on A(T, r), it suffices to show existence of w. First, T take r = 0, so t is the space of diagonal matrices. Write Aν (0) for the image of Aν under the evaluation map t 7→ 0. Then, Aν and A′ ν are gauge equivalent modulo t if and only if there exists f ∈ GLn(C) such that Ad(f ) exp(2πiAν (0)) = exp(2πiA′ ν (0)). (This follows from the Riemann-Hilbert correspondence and [24, Theorem 5.5 and Section 17.1]). Therefore, f lies in the normalizer of T , and Ad(f )(Aν (0)) differs from A′ In particular, A and A′ lie in the same W aff ν (0) by a diagonal matrix with integer entries. Now, assume r > 0. Suppose that h · A′ ν = Aν + x for some x ∈ P1. Fix a split torus D with D(o) ⊂ P . Using the affine Bruhat decomposition, we may write h = p1np2, where p1, p2 ∈ P and n ∈ N (D). We see that T orbit. Aν + x = Ad(h)(A′ ν − p−1 2 τ p2) − (τ p1)p−1 1 − Ad(p1)((τ n)n−1). 2 τ p2 and (τ p1)p−1 1 By Lemma 4.4, p−1 both lie in P1. Moreover, there exists d ∈ D and σ ∈ N (D) ∩ GLn(C) such that n = dσ, so (τ n)n−1 = (τ d)d−1 ∈ d(o) ⊂ P. In particular, Ad(h)(A′ 2 τ p2) ∈ Aν + Ad(p1)((τ d)d−1) + P1. By Lemma 3.18, there exist q1 ∈ P r and q2 ∈ P r+1 such that ad(q2)(A′ 2 τ p2) and ad(q−1 1 + Ad(p1)((τ n)n−1)) lie in t. Since A′ 2 τ p2 is regular by Proposition 3.11, it follows that h ∈ q1N (T )q2. 1 )(Aν + x + (τ p1)p−1 ν − p−1 ν − p−1 ν − p−1 Set g = q−1 1 hq−1 2 and w = gT (o). We will show that ( w)(A) = A′. The ν = πt(q2 · Aν ) ∈ t is a valid representative for A, since πt(Ad(q2)(Aν )) ∈ 2 ∈ P1. Moreover, the fact that πt is a N (T )-map implies element A1 Aν + P1 and (dq2)q−1 that πt((gq2) · Aν ) = πt(g · A1 ν ). ν ) ∈ ν( w)(A1 By Proposition 4.25, πt(g·A1 Write q1 = 1 + X for X ∈ Pr so that (gq2) · Aν ∈ A′ that πt((gq2) · Aν ) ∈ A′ Lemma. ν )+t1. ν) + P1. It follows ν + t1 by Proposition 2.11(3). Thus, w = gT (o) satisfies the (cid:3) ν )+t1, so πt((gq2)·Aν ) ∈ ν( w)(A1 ν − ad(X)(A′ Theorem 4.28. Suppose (V, ∇) is a formal connection. If A ∈ A(V,∇) ∩ A(T, r) and A′ ∈ A(V,∇) ∩ A(T ′, r′), then T and T ′ are GLn(o)-conjugate and r = r′. Moreover, if h ∈ GLn(o) satisfies hT = T ′, then there exists a unique w ∈ W aff T such that A′ = Ad∗(h−1)( w)(A). T ′ T Proof. Proposition 4.20 shows that r = r′ and allows us to assume without loss of generality that T = T ′. By definition of formal types, any choice of representatives Aν and A′ ν are formally gauge equivalent modulo P1. The theorem now follows from Lemma 4.27. (cid:3) 40 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Corollary 4.29. Suppose that A ∈ A(V,∇) . Let φ be an associated trivialization, let (P φ, r, βφ) be the associated stratum (in GL(V )), and let L = φ−1(on). Suppose A′ ∈ A(V,∇) such that A′ = T ( w)(A). has associated trivialization φ′, and choose w ∈ W aff T T (1) If φ′−1(on) = L, then w ∈ WT . (2) If, in addition, (P φ, r, βφ) = (P φ′ , r, βφ′ ), then w is the identity. Proof. By Lemma 4.27, there exists w ∈ W aff such that ( w)(A) = A′. Recall T ∼= T /T (o) ⋉ WT , and WT ∼= (N (T ) ∩ GLn(k))/T ♭. Observe that the that W aff T set of trivializations of satisfying φ′−1(on) = L is a single GLn(o)-orbit. Moreover, P φ = P φ′ by Proposition 3.12. Since P is its own normalizer in GLn(o), it follows that Aν and A′ ν are gauge equivalent by an element p ∈ P in both cases. In the second case, the stabilizer of β + P1−r in P is equal to P 1T ♭. Without loss of generality, take p ∈ P 1. Now, the uniqueness statement in Theorem 4.13 implies that A = A′. In the first case, as long as r > 0, we have A′ ν ∈ Ad(n)(Aν ) + P−r+1, where n ∈ N (T ) is a representative for w. Since n ∈ mT (o) for some m ∈ N (T ) ∩ GLn(k) ⊂ P , we also A′ ν ∈ Ad(m)(Aν ) + P−r+1. We deduce that p ∈ mP 1, and ν = Ad(m)(Aν ). This implies that A′ = the same uniqueness argument shows that A′ (mT (o))(A), and simple transitivity of the W aff T -action gives w = (mT (o)) ∈ WT . If r = 0, then let m ∈ GLn(k) be the image of p modulo t. Since A′ ν ∈ Ad(p)(Aν ) + t gln(o), we have A′ ν = Ad(m)(Aν ). We conclude that w = (mT (o)) ∈ WT as before. (cid:3) We can now make precise the relationship between formal types and formal isomorphism classes in terms of moduli spaces of certain categories of formal con- nections. In the following, we consider three related categories of formal connec- tions. We define C to be the full subcategory of formal connections (V, ∇) of rank n such that (V, ∇) contains a regular stratum. Let C lat be the category of triples (V, ∇, L), where V ∈ C and L ⊂ V is a distinguished o-lattice such such that P ⊂ GL(L) for some regular stratum (P, r, β) contained in V . Morphisms in HomC lat((V, ∇, L), (V ′, ∇′, L′)) in C lat consist of homomorphisms φ : V → V ′ (in the category C ) such that L′ ∩ φ(V ) = φ(L). Note that if φ is an isomorphism, this implies that φ(L) = L′. Finally, C f r is the category of framed connections. This consists of objects in C lat where HomC f r ((V, ∇, L), (V ′, ∇′, L′)) is the set of isomorphisms φ ∈ HomC lat((V, ∇, L), (V ′, ∇′, L′)) such that (φ−1(P ′), r, φ∗(β′)) = (P, r, β). Fix a uniform torus T ⊂ GLn(F ) satisfying T (o) ⊂ GLn(o) and an integer r ≥ 0. We denote the full subcategory of C (resp. C lat, C f r) of connections that have formal type in A(T, r) by C (T, r) (resp. C lat(T, r), C f r(T, r)). Proposition 4.20 implies that the subcategory C (T, r) only depends on the conjugacy class of T in GLn(F ). It follows from Theorem 4.28 that the set of objects in C is the disjoint union of objects in C (Ti, r), taken over a set of representatives Ti as above for the conjugacy classes of uniform tori. The analogous statement holds for C lat and C f r. Corollary 4.30. Fix a uniform torus T ⊂ GLn(F ) with T (o) ⊂ GLn(o) and r ∈ Z≥0. Then, A(T, r) is the moduli space for C f r(T, r), A(T, r)/WT is the moduli space for C lat(T, r), and A(T, r)/W aff is the moduli space for the category T C (T, r). MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 41 5. Moduli Spaces In this section, we will describe the moduli space M (A1, . . . , Am) of 'framed' connections on C = P1(C) with singular points {x1, . . . , xm} and formal type Ai at xi. In our explicit construction, we show that this moduli space is the Hamiltonian reduction of a symplectic manifold via a torus action. Set k = C. We denote by Fx ∼= F the field of Laurent series at x ∈ C and ox ⊂ Fx the ring of power series. Let V be a trivializable rank n vector bundle on P1; thus, there is a noncanonical identification of V with the trivial rank n vector bundle V triv ∼= On P1. The space of global trivializations of V is a GLn(C)-torsor, so we will fix a base point and identify each trivialization φ with an element g ∈ GLn(C). Thus, we will write [∇] for the matrix of ∇ in the fixed trivialization, and g · [∇] for [∇]φ. Define Vx = V ⊗OC Fx and Lx = V ⊗OC ox. The inclusion VC = Γ(P1; V ) ⊂ Vx gives Vx a natural C-structure. Furthermore, Lx determines a unique maximal para- horic Gx = GL(Lx) ∼= GLn(o). In particular, by the remarks preceding Lemma 2.6, there is a one-to-one correspondence between parahoric subgroups P ⊂ Gx and par- abolic subgroups Q ⊂ GL(VC), where Q = P/G1 x. Let Tx be a uniform torus in GLn(Fx) such that Tx(ox) ⊂ GLn(ox), and set x . In the following, (V, ∇) is a connection on C, and Ax ∈ A(Tx, r) is Px = PTx,on a formal type associated to (V, ∇) at x. This means that the formal completion (Vx, ∇x) at x has formal type Ax. We denote the corresponding GLn(Fx)-stratum by (Px, r, βx). We may assume, by Proposition 4.20, that Tx has a block-diagonal embedding in GLn(Fx) as in Remark 4.16. We write Ux = P 1 x. Furthermore, if g ∈ GLn(C), P g x are the pullbacks of P and β, respectively, under the corresponding trivialization (as in Section 4.3). x ⊂ GL(V ) and βg x /G1 Definition 5.1. A compatible framing for ∇ at x is an element g ∈ GLn(C) with the property that ∇ contains the GL(Vx)-stratum (P g x) defined above. We say that ∇ is framable at x if there exists such a g. x , r, βg For example, suppose that eP0 = 1. Choose ν ∈ Ω× 0 of order −1. By Re- mark 4.16, A0ν = 1 tr−1 Dr−1 + . . . where Dj ∈ GLn(C) are diagonal matrices and Dr is regular. It follows that g is a compatible framing for (V, ∇) at 0 if and only if tr Dr + 1 g · [∇] = 1 tr Drν + 1 tr−1 Mr−1ν + . . . , with Mj ∈ gln(C). Now, let A = (A1, . . . , Am) be a collection of formal types Ai at points xi ∈ P1. Definition 5.2. The category C ∗(A) of framable connections with formal types A is the category whose objects are connections (V, ∇), where • V is a trivializable rank n vector bundle on P1; • ∇ is a meromorphic connection on V with singular points {xi}; • ∇ is framable and has formal type Ai at xi; and whose morphisms are vector bundle maps compatible with the connections. The moduli space of this category is denoted by M ∗(A). By Corollary 4.22, any two objects in C ∗(A) correspond to connections that are formally isomorphic at each xi. Note that C ∗(A) is not a full subcategory of the category of meromorphic connections. However, the next proposition show that the 42 CHRISTOPHER L. BREMER AND DANIEL S. SAGE moduli space of this full subcategory coincides with M ∗(A), so this moduli space may be viewed as a well-behaved subspace of the moduli stack of meromorphic connections. Proposition 5.3. Suppose that (V, ∇) and (V ′, ∇′) are framable connections in C ∗(A). If they are isomorphic as meromorphic connections, then they are isomor- phic as framable connections. Proof. Choose trivializations for V and V ′. Then, (V, ∇) and (V ′, ∇′) are isomor- phic as meromorphic connections if and only if there exists a meromorphic section g of the trivial GLn(C)-bundle such that g · [∇] = [∇′]. Moreover, g is necessarily regular at all points of P1\{x1, . . . , xm}. It suffices to show that g is regular at each of the the singular points of ∇. Thus, we may reduce to the following local problem: if ∇ and ∇′ are formal framed connections, g · [∇] = [∇′], and ∇ and ∇′ have the same formal type, then g ∈ GLn(o). Fix ν = dt t . By Theorem 4.13, there exist g1, g2 ∈ GLn(o) such that g1 · [∇τ ] = 1 ) · Aν = Aν. By Corollary 4.21, this implies that τ ] = Aν . Therefore, (g2gg−1 g2 · [∇′ g2gg−1 1 ∈ T ♭. It follows that g ∈ GLn(o). We will construct M ∗(A) using symplectic reduction, so in general M ∗(A) will not be a manifold. Following Section 2 of [5], we define an extended moduli space (cid:3) fM ∗(A) that resolves M ∗(A). Definition 5.4. The category eC ∗(A) of framed connections with formal types A has objects consisting of triples (V, ∇, g), where • (V, ∇) satisfies the first two conditions of Definition 5.2; • g = (Ux1g1, . . . , Uxmgm), where gi is a compatible framing for ∇ at xi; • the formal type (A′)i of ∇ at xi satisfies (A′)it1 = Ait1 . A morphism between (V, ∇, g) and (V ′, ∇′, g′) is a vector bundle isomorphism φ : V → V ′ that is compatible with ∇ and ∇′, with the added condition that xi )g′ (φ−1 responding moduli space. i )) = (P gi xi (P ′g′ xi ), r, φ∗ xi((β′ i W aff Txi xi , r, βgi and WTxi xi) for all i. We let fM ∗(A) denote the cor- as in Section 4.4. The groups W =Qxi Remark 5.5. Define W aff Txi and Waff = Qxi act componentwise on Qxi WTxi A(Txi, rxi). We note that a global connection (V, ∇) lies in C ∗(A) if (Vxi , ∇xi) is isomorphic to the diagonalized νν) in C lat. It follows from Corollary 4.30 that the categories connection (F n C ∗(A′) and C ∗(A) have the same objects if and only if A′ = wA for some w ∈ W. In particular, M ∗(wA) ∼= M ∗(A). If we let j denote the injection of these spaces into the moduli space of meromorphic connections, then j(M ∗(wA)) = j(M ∗(A)). On the other hand, if A′ is not in the Waff -orbit of A, then j(M ∗(A′)) and j(M ∗(A)) are disjoint. This is because connections in the corresponding categories are not even formally isomorphic by Theorem 4.28. xi , d + Ai Now, suppose that s = (s1, . . . , sm) ∈ Waff and si ∈ Txi(F )/Txi(o). In this case, C ∗(sA) 6= C ∗(A) unless si is the identity. However, it is clear that (si)(Ai)t1 = Ait1. We deduce that eC ∗( wA) = eC ∗(A), and fM ∗( wA) ∼= fM ∗(A) for all w ∈ Waff . (Indeed, eC ∗(A′) = eC ∗(A) if and only if for every xi there exists wi ∈ W aff such that ( wi)((A′)i)t1 = Ait1 .) Txi MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 43 Let X be a symplectic variety with a Hamiltonian action of a group G. There is a moment map µG : X → g∨. If α ∈ g∨ lies in [g, g]⊥, so that the coadjoint orbit of α is a singleton, then the symplectic reduction X (cid:12)α G is defined to be the quotient µ−1(α)/G. In Section 5.1, we will use the formal type Ai at xi to define extended orbits GLn(C). The following theorem generalizes [5, Proposition 2.1]: Mi and fMi. These are smooth symplectic manifolds with a Hamiltonian action of Theorem 5.6. Let M ∗(A), fM ∗(A) be the moduli spaces defined above. (1) The moduli space M ∗(A) is a symplectic reduction ofQi Mi: (2) Similarly, Mi) (cid:12)0 GLn(C). M ∗(A) ∼= (Yi fM ∗(A) ∼= (Yi fMi) (cid:12)0 GLn(C). Moreover, fM ∗(A) is a symplectic manifold. is naturally a symplectic reduction of fM ∗(A) by the groupQi T ♭ (3) Let Ti = Txi. There is a Hamiltonian action of T ♭ This theorem will be proved in Section 5.2. i on fM ∗(A), and M ∗(A) i . Remark 5.7. We also obtain a version of this theorem when additional singularities corresponding to regular singular points are allowed (Theorem 5.26). In the case of the Katz-Frenkel-Gross connection [17, 13], the moduli space reduces to a point, consistent with the rigidity of this connection. Theorem 5.26 also allows one to construct many other examples of connections with singleton moduli spaces, which are thus plausible candidates for rigidity. Remark 5.8. It is not surprising that these moduli spaces are symplectic: it is con- ceivable that this fact might be proved independently using the abstract methods of [10, Section 6]. The advantage of Theorem 5.6 is that it gives explicit constructions of M ∗(A) and fM ∗(A) in a number of important, novel cases (including connections with 'supercuspidal' type singularities). Moreover, the fact that fM ∗(A) is smooth allows one to generalize the work of Jimbo, Miwa, and Ueno [16] and explicitly calculate the isomondromy equations in these cases (see [6]). 5.1. Extended Orbits. In this section, we will construct symplectic manifolds, called extended orbits, which will be "local pieces" of the moduli spaces M ∗(A) and fM ∗(A). Without loss of generality, we will take our singular point to be x = 0, Our study of extended orbits is motivated by the relationship between coadjoint and we will suppress the subscript x from Fx, Px, Ax, etc. orbits and gauge transformations. In the following, fix ν ∈ Ω× 0 . Proposition 5.9. The map gln(F ) → P∨ determined by ν intertwines the gauge action of P on gln(F ) with the coadjoint action of P on P∨. Proof. Recall from Lemma 2.6 that P = H ⋉ P 1 for H ∼= Q/U a Levi subgroup of GLn(C). Thus, we may write any element of P as p = hu, for h ∈ H and u ∈ P 1. Without loss of generality, we may assume that ν has order −1. Thus, p · X = h · (Ad(u)(X) − τ (u)u−1) = Ad(p)(X) − Ad(h)(τ (u)u−1). 44 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Lemma 4.4 shows that Ad(h)(τ (u)u−1) ∈ P1 = P⊥. Applying Proposition 2.5, we see that Ad∗(p)(hX, ·iν P) = hAd(p)(X), ·iνP = hp · X, ·iνP. (cid:3) From now on, we assume ν has order −1. We suppose that A is a formal type at 0 stabilized by a torus T with T (o) ⊂ P and that the corresponding regular stratum (P, r, β) has r > 0. In particular, any connection with formal type A is irregular singular. Denote the projection of π∗ t (A) onto (P1)∨ by A1. Let G = GLn(o) be the maximal standard parahoric subgroup at 0 with congruence subgroups (resp. fractional ideals) Gi (resp. gi). Then, G1 ⊂ P ⊂ G, and P/G1 ∼= Q. For any subgroup H ⊂ G with Lie algebra h, there is a natural projection πh : g∨ → (h)∨ obtained by restricting functionals to h ⊂ g. Denote the P -coadjoint orbit of π∗ t (A) by O, and the P 1-coadjoint orbit of A1 by O 1. Definition 5.10. Let A be a formal type at 0 with irregular singularity, and let U be the unipotent radical of Q. We define the extended orbits M (A) and fM (A) by • M (A) ⊂ (Q\ GLn(C)) × g∨ is the subvariety defined by M (A) = {(Qg, α) πP(Ad∗(g)(α)) ∈ O)}; (5.1) • fM (A) ⊂ (U \ GLn(C)) × g∨ is defined by fM (A) = {(U g, α) πP1 (Ad∗(g)(α)) ∈ O 1)}; Proposition 5.11. The extended orbits M (A) and fM (A) are isomorphic to sym- plectic reductions of T ∗G × O and T ∗G × O 1 respectively: M (A) ∼= T ∗G × O (cid:12)0 P fM (A) ∼= T ∗G × O 1 (cid:12)0 P 1. In particular, the natural symplectic form on T ∗G × O descends to both M (A) and fM (A). Moreover, M (A) and fM (A) are smooth symplectic manifolds. Remark 5.12. Note that T ∗G is not finite dimensional. However, for ℓ sufficiently large, A ∈ (gℓ)⊥. Since Gℓ ⊂ P 1, we see that G/P 1 ∼= (G/Gℓ)/(P 1/Gℓ). Thus, in Proposition 5.11, it suffices to consider T ∗(G/Gℓ) × O (cid:12)0 P (resp. T ∗(G/Gℓ) × O 1 (cid:12)0 P 1). This fact, although concealed in our notation, ensures that we are al- ways applying results from algebraic and symplectic geometry to finite-dimensional varieties. Proof. The proof in each case is similar, so we will prove the second isomorphism. The group P 1 acts on T ∗G by the usual left action p(g, α) = (pg, α) and on O 1 by the coadjoint action. Moreover, on each factor, the action of P 1 is Hamiltonian with respect to the standard symplectic form. The moment map for the diagonal action of P 1 is the sum of the two moment maps: µP 1 : T ∗G × O 1 → (P1)∨ µP 1 (g, α, β) = πP1 (− Ad∗(g)(α)) + β. In particular, P 1 (0) = {(g, α, β) πP1 (Ad∗(g)(α)) = β}. µ−1 We will show that µ−1 P 1 (0) is smooth. Let ϕ : µ−1 P 1 (0) → G × O 1 be defined by ϕ(g, α, β) = (g, β). Choose a local section f : O 1 → O 1 + (P1)⊥ ⊂ g∨. Then, ϕ−1(g, β) = {(g, Ad∗(g−1)(f (β) + X), β) X ∈ (P1)⊥}. Therefore, µ−1 P 1 (0) is an MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 45 affine bundle over G × O 1 with fibers isomorphic to P/g1. It follows that µ−1 is smooth. P 1 (0) Since U ∼= P 1/G1, U \ GLn(C) ∼= P 1\G. Therefore, the map (g, α, β) 7→ (P 1g, α) takes µ−1 P 1\µ−1 (ζ′(P 1g), α, πP1 (Ad∗(ζ(P 1g))(α))). This shows that µ−1 Finally, choose a local section ζ : P 1\G → G with domain W , and let W ′ = P 1 (0) given by ζ′(P 1g, α) = P 1 (0) to fM (A); moreover, the fibers are P 1 orbits, so the map identifies P 1 (0) ∼= fM (A). fM (A) ∩ (W × g∨). There is a section ζ′ : W ′ → µ−1 P 1 (0) → fM (A) is a principal P 1-bundle, since P 1 acts freely on the fibers. In particular, fM (A) is smooth, and the symplectic form on T ∗G × O 1 descends to fM (A). Let res : g∨ → gln(C)∨ be the restriction map dual to the inclusion gln(C) → g. Notice that if we fix a representative αν ∈ gln(F ) for α ∈ g∨, then gln(C)∨ ∼= gln(C) under the trace pairing and res(α) corresponds to the ordinary residue of αν ν. (cid:3) There is a Hamiltonian left action of GLn(C) on T ∗G defined by (5.2) ρ(h)(g, α) = (gh−1, Ad∗(h)α). The moment map µρ is given by µρ(g, α) = res(α). To see this, observe that ρ is the restriction to GLn(C) of the usual left action of G on T ∗G (via inversion composed with right multiplication). Hence, the map µρ is just the composition of the moment map for right multiplication µ(g, α) = α with res. The action ρ defines left actions of GLn(C) on the first components of T ∗G × O and T ∗G × O 1 respectively. These actions commute with the left actions of P and P 1, and it is clear that µP and µP 1 are GLn(C)-equivariant. Lemma 5.13. Let G1 and G2 act on a symplectic manifold X via Hamiltonian ac- tions, and let µ1 and µ2 be the corresponding moment maps. If µ2 is G1-invariant on µ−1 1 (λ), then there is a natural Hamiltonian action of G2 on X (cid:12)λ G1. Further- more, if ιλ : µ−1 1 (λ) → X (cid:12)λG1 are the natural maps, then the induced moment map ¯µ2 on X (cid:12)λ G1 is the unique map satisfying µ2 ◦ ιλ = ¯µ2 ◦ πλ. 1 (λ) → X and πλ : µ−1 This follows from [1, Theorem 4.3.5]. Thus, ρ descends to natural Hamiltonian actions on M (A) and fM (A). For example, if (Qg, α) ∈ M (A) and h ∈ GLn(C), h(Qg, α) = (Qgh−1, Ad∗(h)(α)). then Proposition 5.14. The moment map for the action of GLn(C) on M (A) is given by µGLn (Qg, α)) = res(α). Proof. This follows directly from Lemma 5.13. The action of GLn(C) on fM (A) has the analogous moment map µGLn . Lemma 5.15. The moment map µGLn : fM (A) → gln(C) is a submersion. Proof. By Proposition 5.11, fM (A) is smooth. We will show that the differential map dµGLn on tangent spaces is surjective. Note that µGLn(U gg′, Ad∗(g′−1)α) = g′−1 µGLn (U g, α). Therefore, it suffices to show that the tangent map is surjective at points s = (U, α) in the subvariety S defined by taking g to be the identity. Let u = Lie(U ), so u⊥ ⊂ gln(C)∨. Indeed, u⊥ ∼= (P1)⊥ ⊂ g∨. If we choose a section f : (P1)∨ → g∨, we see that O 1 × u⊥ ∼= S by the map (γ, y) 7→ (U, f (γ)+ y). (cid:3) 46 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Here, the image of y is identified with its image in g∨. In particular, the image of dµGLn(Ts fM (A)) contains u⊥ ⊂ gln(C)∨. Therefore, it suffices to show that the composition of dµGLn with the quotient gln(C)∨ → u∨ is surjective. Observe that tangent vectors to O 1 are of the form ad∗(X)(α) for X ∈ P1. First, suppose that r > eP . In this case, we will show that ad∗(P1−eP +r)(α) ⊂ TsO 1 surjects onto u∨. More precisely, we will construct a filtration u∨ = u1 ⊃ u2 ⊃ · · · ⊃ ueP = {0} such that the map X 7→ ad∗(X)(α) induces a surjection u : ¯Pj−eP +r → uj/uj+1 for each j. Since u ∼= P1/g1, we see that u∨ ∼= g/P under ¯πj the duality isomorphism. We now obtain the desired filtration on u∨ by subspaces of the form uj ∼= (Pj−eP ∩g)/P. More explicitly, uj is the restriction of (PeP −j+1)⊥ ⊂ (P1)∨ to u. Note that the map ¯πj u = τ j ◦ (−δαν ), where τ j : ¯Pj−eP → uj/uj+1 is the surjection defined by τ j(X) = (hX, ·iνu) + uj+1. Furthermore, ¯πj u depends only on the coset αν + P−r+1. By assumption, αν ∈ Aν + P−r+1. Proposition 2.11 shows that δαν ( ¯Pj−eP +r) = ker(¯πt). Since ¯πt : ¯Pj−eP → ¯tj−eP is a surjection, a diagram chase shows that τ jker(¯πt) is surjective if and only if ¯πtker(τ j ) is surjective. We now verify this last statement. It is easily checked that Yjν = t−1 Res(j−eP dt) corresponds to a non-zero element Yj of ker(¯πj . Therefore, the span of Yjν surjects onto the one-dimensional space ¯tj−eP . A similar proof works for eP < n, using the observation in Corollary 3.9 that t ∼= En/eP . In the case eP = n, recall the description of E from (2.2). u) (since 1 ≤ j < eP ) and πt(Yjν ) = eP −j eP Now, assume that 1 ≤ r ≤ eP . The above argument shows that ¯πj u( ¯Pj−eP +r) = uj/uj+1; however, in this case, we can only conclude that the image of dµGLnTsO 1 contains ueP −r+1. Let wj = Pj−eP +r ∩ gln(C); it follows that wj /wj+1 determines a well-defined subset of ¯Pj−eP +r. We claim that ¯πj u(wj/wj+1) = uj/uj+1 for 1 ≤ j ≤ eP − r. Observe that t−1g ⊃ Pℓ ⊃ Pℓ+1 ⊃ tg for −eP ≤ ℓ ≤ 0. Therefore, we may take a representative βν ∈ t−1 gln(C) + gln(C) for αν + P−r+1. Similarly, choose a representative X + t−1X ′ for ¯X ∈ ¯Pj−eP +r, where X, X ′ ∈ gln(C). It follows that had(X + t−1X ′)βν , Y iν = had(X)βν , Y iν whenever Y ∈ u. This proves the claim, and we conclude that ad∗(w1)(α) surjects onto u∨/ueP −r+1 . E j−eP E Finally, let X ∈ w1. The action of GLn(C) on fM (A)) gives rise to a map gln(C) → Ts fM (A) ⊂ u\ gln(C) × g∨; explicitly, X 7→ (−X, ad∗(X)α), which is sent to res(ad∗(X)(α)) by dµGLn . Therefore, dµGLn maps tangent vectors coming from w1 ⊂ gln(C) surjectively onto u∨/ueP −r+1. It follows that the image of dµGLn contains u∨, so µGLn is a submersion. (cid:3) Lemma 5.16. GLn(C) acts freely on fM (A). Proof. Suppose that h ∈ GLn(C) fixes (U g, α). In particular, U gh = U g, so gh ∈ U . To show that h = 1, it suffices to show that gh = 1, so without loss of generality, we may assume that g = 1 and h ∈ U . By Proposition 2.5, there exists a rep- resentative αν ∈ gln(F ) for α with terms only in nonpositive degrees. The fact that Ad∗(h−1)(α) = α implies that Ad(h−1)(αν ) = αν + X for X ∈ g1. Since h ∈ GLn(C), X = 0, and we see that Ad(h−1)(αν ) = αν. We will show that h is P 1-conjugate to an element of T (o). In particular, since P 1 ∼= U ⋉ G1, we see that h is U -conjugate to an element of T (o)G1 ∩GLn(C) = T ♭. Since T ♭ ∩ P 1 is trivial, Corollary 3.9 implies that h = 1. MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 47 Take p ∈ P 1 such that Ad∗(p)(α) = A1; thus, Ad(p)(αν ) ∈ t+P. By Lemma 3.18, there exists p′ ∈ P 1 and a representative A1 A1 qh ∈ T ∩ G = T (o). ν. Therefore, setting q = p′p ∈ P 1, Ad((qh)−1)A1 ν ∈ t−r of A1 such that Ad(p′)(Ad(p)(αν )) = ν = A1 ν . By Lemma 3.11, (cid:3) Lemma 5.17. If (Qg1, α) and (Qg2, α) both lie in M (A), then g2 = pg1 for some p ∈ Q. Moreover, if (U g1, α) and (U g2, α) both lie in fM (A), then g2 = usg1 for some u ∈ U and s ∈ T ♭. Proof. Notice that (Q, Ad∗(g1)α) and (Qg2(g−1 1 ), Ad∗(g1)α) satisfy the conditions of the first statement. There is a similar reformulation of the second statement. Thus, we may assume without loss of generality that g1 is the identity; we set g2 = g. In the first case, note that by Lemmas 3.18 and 3.20, there exist p1, p2 ∈ P such centralizes 2 ∈ T ∩ G = T (o). We conclude that that Ad(p1g)(αν ) = Aν = Ad(p2)(αν ) for some Aν ∈ t. Since p1gp−1 2 the regular semisimple element Aν , p1gp−1 g ∈ P ∩ GLn(C) = Q. In the second case, the same argument shows that whenever Ad∗(g)(α) ∈ O 1, g = p−1 1 sp2 for some s ∈ T (o) and pi ∈ P 1. Since P 1 is normal in P , g = us for some u ∈ P 1. By Corollary 3.9, we may assume that s ∈ T ♭. This implies that u ∈ GLn(C) ∩ P 1 = U as desired. (cid:3) Lemma 5.18. Let α ∈ g∨ be a functional such that πP1 (α) = A1. Then, if s ∈ T (o), πP1 (Ad∗(s)α) = A1. Proof. Since any representative of A1 lies in t(o), αν ∈ t + P. The lemma is now clear, since T (o) preserves P and stabilizes t. (cid:3) call, from Lemma 3.14, that T ♭ = T (o) ∩ GLn(C). There is a left action of T ♭ We are now ready to describe the relationship between M (A) and fM (A). Re- on fM (A) given by s(U g, α) = (U sg, α). To see this, note that by assumption, πP1 (Ad∗(g)(α)) ∈ O 1, so there exists u ∈ P 1 such that Ad∗(u)(πP1 (Ad∗(g)(α))) = A1. We wish to show that there exists u′ ∈ P 1 such that Ad∗(u′)(πP1 (Ad∗(sg)(α))) = A1. However, (5.3) Ad∗(us)(πP1 (Ad∗(sg)(α))) = πP1 (Ad∗(su) Ad∗(sg)(α)) = πP1 (Ad∗(s) Ad∗(ug)(α)) = A1, We will show that this action is Hamiltonian with moment map µT ♭ defined as where the last equality follows from Lemma 5.18. In particular, s(U g, α) ∈ fM (A). follows. Take (U g, α) ∈ fM (A). There exists u ∈ P 1 such that πP1 (Ad∗(ug)(α)) = A1. (5.4) Define a map µT ♭(U g, α) = −(Ad∗(ug)(α))T ♭ . We need to show that this map is well-defined. Let A = Ad∗(ug)(α). Suppose that u′ ∈ P 1 satisfies (5.4). Observe that Ad∗(u′u−1)(A1) = A1. By Lemma 3.21, u′u−1 ∈ T (o)P r. It suffices to show that whenever s ∈ T (o) and p ∈ P r, (Ad∗(sp)( A))T ♭ = ( A)T ♭ . In fact, we will prove the stronger statement: (5.5) πt∩P(Ad∗(sp)( A)) = πt∩P( A). 48 CHRISTOPHER L. BREMER AND DANIEL S. SAGE Fix a representative Aν ∈ P−r. By Proposition 2.11(4), the projection (gl(F ))∨ → t∨ corresponds to tame corestriction πt : gln(F ) → t after dualizing. Thus, πt(Ad(sp)( Aν )) is a representative of πt∩P(Ad∗(sp) A). Since πt commutes with the action of t, πt(Ad(sp)( Aν )) = πt(Ad(p)( Aν )). By Proposition 2.11(3), πt(Ad(p)( Aν ))− πt( Aν ) ∈ P1 = P⊥, so πt( Aν ) is a representative for both functionals in (5.5) as desired. The following lemma generalizes [5, Lemma 2.3]. The proof is more complicated, due to the absence of a 'decoupling' map in the general case. Proposition 5.19. Let Λ = At♭ . The action of T ♭ on fM (A) is Hamiltonian with moment map µT ♭ . Moreover, M (A) ∼= fM (A) (cid:12)−Λ T ♭. Proof. Recall that O 1 be the P 1-coadjoint orbit containing A1. If β ∈ O 1, we may take α ∈ g∨ such that πP1 (α) = β. The torus T ♭ acts on O 1 by s · β = πP1 ((Ad∗(s)(α))). (One sees that this element is in O 1 by an argument similar to the one used to show (5.3), and it is easily checked that it is independent of the choice of α.) We construct a moment map for this action. Consider the semi-direct product T ♭ ⋉ P 1 ⊂ P , and lift A1 ∈ O 1 to A ∈ (t♭)⊥ ⊂ (t♭ × P1)∨. Let O ⊂ (t♭ × P1)∨ be the coadjoint orbit of A. Since T ♭ stabilizes A1 by Lemma 5.18, it is clear that In particular, P 1 acts transitively on O. We will prove it stabilizes A as well. in Lemma 5.20 that the natural map π : O → O 1 is a T ♭-equivariant symplectic isomorphism. Therefore, the moment map µ : O 1 → (t♭)∨ is given by µ(β) = πT ♭(π−1(β)), where πT ♭ is the projection (t♭ × P1)∨ → (t♭)∨. We remark that if a different lift of A1 is chosen, say A+γ for γ ∈ (P1)⊥ ∼= (t♭)∨, then (5.6) (Ad∗(u)( A + γ))(z) = Ad∗(u)( A)(z) + γ(z) Indeed, if (g, α, β) ∈ T ∗G × O 1, then s(g, α, β) = (sg, α, s · β) defines a O 1. Hamiltonian action; the moment map µ′ is given by the sum of the natural moment map on T ∗G and µ . Moreover, T ♭ preserves µ−1 P 1 (0) → for u ∈ P 1 and z ∈ t♭. In particular, this changes eµ by a constant γ. The action of T ♭ on fM (A) descends from a Hamiltonian action of T ♭ on T ∗G × fM (A) is T ♭-equivariant. fM (A), and define φ(g, α) to be the projection of Ad∗(g)(α) onto (t♭ × P1)∨. Then, We will show that the restriction of µ′ to µ−1 P 1 (0) is P 1-invariant. Let (g, α, β) ∈ P 1 (0), and the map from µ−1 if u ∈ P 1, µ′(u(g, α, β)) = µ′(ug, α, Ad∗(u)β) = πT ♭(cid:0)− Ad∗(u)φ(g, α) + Ad∗(u)π−1(β)(cid:1) However, β = πP1 (Ad∗(g)(α)) lies in O 1, so φ(g, α) must lie in a coadjoint orbit containing A − γ for some γ ∈ (t♭)∨. Equation (5.6) implies that πT ♭ (−φ(g, α) + π−1(β)) = πT ♭(cid:0)− Ad∗(u)φ(g, α) + Ad∗(u)π−1(β)(cid:1) = γ. MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 49 and the moment map descends from µ′. Thus, µ′ is P 1-invariant. By Lemma 5.13, the action of T ♭ on fM (A) is Hamiltonian, It remains to show that γ = µT ♭(U g, α). By P 1-invariance, it suffices to con- sider the case where πP1 (Ad∗(g)(α)) = A1. By construction, µ(A1) = 0, so γ = − Ad∗(g)(α)t♭ = µT ♭(U g, α). We now prove that M (A) ∼= fM (A) (cid:12)−Λ T ♭. First, we show that if µT ♭ (U g, α) = −Λ, then (Qg, α) ∈ O. Let u ∈ P 1 satisfy πP1 (Ad∗(ug)α) = A1. Choosing a representative αν for α, we have Ad(ug)(αν) ∈ Aν + P. Applying πt, we see that In fact, z ∈ t ∩ P1 because the πt(Ad(ug)(αν)) = Aν + z for some z ∈ t(o). restrictions of Ad∗(ug)α and A to (t♭)∨ agree. By Proposition 2.11(3), there exists X ∈ Pr such that Ad(1 + X) Ad(ug)αν) ∈ Aν + P1, so πP(Ad∗((1 + X)ug)α) = t (A), i.e., Ad∗(g)(α) ∈ O. Thus, we have a map µ−1 π∗ T ♭ (−Λ) → M (A) given by (U g, α) 7→ (Qg, α). Lemma 5.17 shows that the fibers of the map are T ♭-orbits, and we obtain the desired isomorphism. (cid:3) Lemma 5.20. In the notation from the previous proof, the map π : O → O 1 is a T ♭-equivariant symplectic isomorphism. Proof. First, we show T ♭-equivariance. We have already observed that there is a transitive action of P 1 on O and that T ♭ stabilizes A. For any s ∈ T ♭ and u ∈ P 1, we calculate π(Ad∗(s) Ad∗(u) A) = π(Ad∗(su) A) = Ad∗(su)π( A) = πP1 (Ad∗(s) Ad∗(u) Ad∗(s−1)(π∗ = πP1 (Ad∗(s) Ad∗(u)(π∗ t (A))) t (A))) = s · (Ad∗(u)(π∗ t (A))). Next, we show that the stabilizer of A in P 1 is the same as the stabilizer of t (A). Let Aν ∈ P−r be a representative for A. In fact, Aν ∈ (t + P) ∩ (t♭)⊥. By π∗ Lemma 3.21, the stabilizer of π∗ t (A) is precisely (T (o) ∩ P 1)P r if r ≥ 2 or P 1 if r = 1 (in which case π∗ t (A) is a singleton orbit). It suffices to show that this group stabilizes A, as the stabilizer of A is a subgroup of the stabilizer of π∗ t (A). Since Aν ∈ t + P, T ∩ P 1 stabilizes A. Now, take u ∈ P r, z ∈ t♭ and X ∈ P1. We see that Ad∗(u)( A)(z + X) = A(Ad(u−1)z + Ad(u−1)X) = A(Ad(u−1)z) + A(X), so we need only check that Ad∗(u) A(z) = A(z). However, by Proposition 2.11(4), Ad∗(u) A(z) = hAd(u) Aν , ziν = hπt(Ad(u) Aν ), ziν. Proposition 2.11(3) implies that πt(Ad(u) Aν ) ≡ πt( Aν ) (mod P1). It follows that the stabilizer of A is, indeed, the same as the stabilizer of π∗ t (A); moreover, since π is a P 1-map, it follows that π is an isomorphism. Finally, we need to show that π preserves the natural symplectic form on each coadjoint orbit. In particular, since O and O 1 are P 1 orbits, it suffices by transi- tivity to show that the symplectic forms are the same at A and A1. In other words, we need to show that A([X1 + z1, X2 + z2]) = A1([X1, X2]) for zj ∈ t♭ and Xj ∈ P1. This is clear, since the restriction of A to P1 is exactly A1, and t♭ lies in the kernel of the symplectic form at A. (cid:3) 5.2. Proof of the theorem. Let V be a trivializable vector bundle on P1, and let ∇ be a meromorphic connection with singularities at {x1, . . . , xm}. We assume that ∇ has compatible framings {g1, . . . , gm} at each of the singular points and 50 CHRISTOPHER L. BREMER AND DANIEL S. SAGE t (Ai) under Pi (resp. P 1 i at xi. We define Oi ⊂ Pi (resp. O 1 that ∇ has formal type Ai ∈ t∨ i ) to be the coadjoint orbit of π∗ i ). We fix a global trivialization as in the beginning of the section; as usual, we will use this fixed trivialization to identify subgroups of GL(Vx) and GLn(Fx), etc. Definition 5.21. The principal part [∇x]pp of ∇ at x is the image of [∇x] in g∨ x by the residue-trace pairing. i ⊂ P1 To give an example, if [∇0] = M−1 dt t2 + M0 Mi ∈ GLn(C), then [∇0]pp(X) = Res(Tr((M−1 dt t + M1dt + M2tdt + . . ., with the dt t2 + M0 dt t )X)) for any X ∈ g0. We set [∇i]pp to be the principal part of ∇ at xi. It is a consequence of the duality theorem ([23, Theorem II.2]) that ∇ is uniquely determined by the collec- tion {[∇i]pp}. Moreover, the residue theorem ([23, Proposition II.6]) shows that Pi res([∇i]pp) = 0. If gi is a compatible framing for ∇ at xi, determines a unique connection with singularities only at the xi's and with principal part at xi given by Mi. πPi ((Ad∗(gi)[∇i]pp) ∈ Oi and πP1 i (Ad∗(gi)[∇i])pp) ∈ O 1 i . This follows from the observation that gi · [∇i]pp = Ad∗(gi)[∇i]pp and Proposi- tion 5.9. In particular, since gi · [∇i] is formally gauge equivalent to π∗ t (A) by an element of pi ∈ P 1 i , it follows that Ad∗(pi) Ad∗(gi)[∇i]pp = π∗ Finally, we need to define extended orbits M (A) and fM (A) in the case where A is a regular singular formal type. In particular, the corresponding uniform stratum is of the form (G, 0, β). Since π∗ t (A) is a functional on g that kills g1, we may t (A) as an element of gln(C)∨. We define (t♭)′ ⊂ gln(C)∨ to be the set think of π∗ of functionals of the form φ(X) = Tr(DX), where D ∈ t♭ is a diagonal matrix with distinct eigenvalues modulo Z. t (A). The following definition comes from Section 2 of [5]. Definition 5.22. Let A be a formal type corresponding to a stratum (G, 0, β). Define M (A) = OA, the coadjoint orbit of π∗ t (A) in g∨. Moreover, let fM (A) := {(g, α) ∈ GLn(C) × gln(C)∨ Ad∗(g)α ∈ (t♭)′} ⊂ G × g∨. Remark 5.23. This definition of M (A) coincides with the definition for r > 0 given is independent of formal type when r = 0. However, the essential results of Section 5.1 remain true in the regular singular composed with right multiplication). The moment map for T ♭ is simply (g, X) 7→ in (5.1) (where now Q = GLn(C)), but this is not true for fM (A). Indeed, fM (A) case. By [14, Theorem 26.7], fM (A) is a symplectic submanifold of T ∗ GLn(C). Moreover, T ♭ (resp. GLn(C)) acts on fM (A) by left multiplication (resp. inversion − Ad∗(g)(X), and the map (g, α) 7→ α induces an isomorphism fM (A) (cid:12)−A T ♭ ∼= Proof of Theorem 5.6. For each xi, set Mi = M (Ai), and fMi = fM (Ai). As above, a meromorphic connection ∇ on P1 is uniquely determined by the principal parts at its singular points {xi}. Moreover, any collection {Mi}, where Mi ∈ g∨ i , that also satisfies the residue condition M (A). (5.7) res(Mi) = 0 Xi MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 51 There is a map Mi → g∨ i obtained by taking (Qig, αi) to αi. Lemma 5.17 implies that this map is one-to-one, and it identifies elements of Mi with the principal part of a framed connection at xi with formal type Ai. We conclude that any element Mi satisfying (5.7) uniquely determines a connection ∇ with framed formal Mi induced by its action on global trivializations of V is the product of the left actions on Mi given in (5.2). Therefore, it follows from Proposition 5.14 that the moment map of this action is simply type Ai at xi. ofQi The action of GLn(C) onQi µ :Yi (Qigi, αi) 7→Xi res(αi). (5.7) now translates into an m-tuple lying in µ−1(0), so Mi! (cid:12)0 GLn(C). (Uigi, αi) to the data (V, ∇, g), where ∇ has principal part αi at xi and g = (Uigi). The moment map µ :Qi fMi → gln(C)∨ is defined similarly. The residue condition M ∗(A) ∼= Yi Similarly, fM ∗(A) ∼= (cid:16)Qi fMi(cid:17) (cid:12)0 GLn(C): the map µ−1(0) → fM ∗(A) takes By Lemma 5.16, GLn(C) acts freely on fMi, so the action on Qi fMi is free. It follows that fM ∗(A) is a smooth i onQi fMi commutes with the action of GLn(C), so by Lemma 5.13, there is a natural Hamiltonian action ofQi T ♭ fM ∗(A). Similarly, there is a Hamiltonian action of GLn(C) on i ) ∼= (Yi fMi) (cid:12)Qi(−Λi)Y T ♭ Moreover, Lemma 5.15 states that µ is a submersion on each factor, so µ is a submersion. Therefore, µ−1(0) is smooth. symplectic variety. . The action ofQi T ♭ Yi ( fMi (cid:12)−Λi T ♭ We now see that taking the iterated symplectic reduction of the product of local data by GLn(C) and the product of the local tori is independent of order: Finally, let Λi = Ait♭ i on i . i indeed, both are isomorphic to the symplectic reduction via the product action: i ). By Proposition 5.19, it follows that T ♭ i (Yi fMi) (cid:12)0 GLn(C)! (cid:12)Qi(−Λi)Yi Q fMi (cid:12)(0,Qi(−Λi)) (GLn(C) ×Q T ♭ fM ∗(A) (cid:12)Qi(−Λi)Yi ∼=Yi ( fMi (cid:12)−Λi T ♭ i ) (cid:12)0 GLn(C); T ♭ i ∼= M ∗(A). (cid:3) Remark 5.24. In the case m > 1 above, we only require µ to be a submersion on one M (Ai) i=1 fM (Ai). In particular, the residue map on fM (A1) ×Qm factor in Qm is submersive. Moreover, by Lemma 5.16, the action of GLn(C) on fM (A1) × M (Ai)(cid:17) (cid:12)0 GLn(C) is Qm M (Ai) is free. Therefore, M ′(A) =(cid:16) fM (A1) ×Qm smooth, and M ′(A) (cid:12)−Λ1 T ♭ 1 ∼= M (A). i=2 i=2 i=2 52 CHRISTOPHER L. BREMER AND DANIEL S. SAGE We state here a more general version of Theorem 5.6. The proof is essentially the same; however, it allows us to consider regular singular points with arbitrary monodromy. In particular, this construction includes the GLn case of the flat G- bundle constructed in [13]. Let { Oj} be a collection of 'non-resonant' adjoint orbits in gln(C); this means that the distinct eigenvalues of elements Oj do not differ by nonzero integers. Using the trace pairing, we may identify Oj with a coadjoint orbit Oj ⊂ gln(C)∨. Thus, we say that a connection ∇ on the trivial bundle V over C has residue in Oj at yj ∈ C if the principal part at yj corresponds to an element of Oj in gln(C)∨. for some X ∈ Oj. By the standard theory of regular Equivalently, [∇yj ]pp = X dt t singular point connections (see, for example, [24, Chapter II]), if a connection (V, ∇) has non-resonant residue X ∈ Oj, then (V, ∇) is formally equivalent to d + X dt t . Let B = {Oj} be a finite collection of non-resonant adjoint orbits corresponding to a collection of regular singular points {yj} ⊂ C, and let A = {Ai} be a finite collection of formal types at {xi} ⊂ C, disjoint from {yj}. Definition 5.25. Define M (A, B) to be the moduli space of connections ∇ on the trivial bundle V with the following properties: (1) (V, ∇) is compatibly framed at each xi, with formal type Ai; (2) (V, ∇) is regular singular and has residue in Oj at each yj. If A is nonempty, we define the extended moduli space M (A, B) of isomorphism classes of data (V, ∇, g), where (V, ∇) satisfy the conditions above and g = (gi) is a collection of local compatible framings at the xi's. We omit the proof of the following theorem, since it is almost identical to the proof of Theorem 5.6. We note that a similar construction is used in [4] and [3] in the case where the Ai are totally split. Theorem 5.26. (1) The moduli space M (A, B) is a symplectic reduction of the product of local data: Oj (cid:12)0 GLn(C). (2) If A is nonempty, then fM (A, B) is a symplectic manifold, and Oj (cid:12)0 GLn(C). i! . M (A, B) ∼= Yi M (Ai)! ×Yj fM (A, B) = Yi fM (Ai)! ×Yj M (A, B) ∼= fM (A, B) (cid:12)Qi(−Λi) Yi (3) T ♭ References [1] R. Abraham and J. Marsden, Foundations of Mechanics, 2nd ed., Benjamin/Cummings Pub- lishing Co., Reading, 1978. [2] D. Arinkin, "Rigid Irregular Connections on P1," Compos. Math. 146 (2010), 1323-1338. [3] P. Boalch, "G-bundles, isomonodromy, and quantum Weyl groups," Int. Math. Res. Not. 22 (2002), 1129 -- 1166. MODULI SPACES OF IRREGULAR SINGULAR CONNECTIONS 53 [4] P. Boalch, "Stokes matrices, Poisson Lie groups, and Frobenius manifolds," Invent. Math. 146 (2001), 479 -- 506. [5] P. Boalch, "Symplectic manifolds and isomonodromic deformations," Adv. Math. 163 (2001), 137 -- 205. [6] C. Bremer and D. S. Sage, "Isomonodromic deformation of connections with singularities of parahoric formal type," arXiv:1010.2292v2 [math.AG], 2010. [7] C. J. Bushnell, "Hereditary orders, Gauss sums, and supercuspidal representations of GLN ," J. Reine Agnew. Math. 375/376 (1987), 184 -- 210. [8] C. J. Bushnell and P. Kutzko, The admissible dual of GLN via compact open subgroups, Annals of Mathematics Studies, 129, Princeton University Press, Princeton, NJ, 1993. [9] P. Deligne, ´Equations diff´erentielles `a points singuliers r´eguliers, Lecture Notes in Mathe- matics, Vol. 163, Springer-Verlag, New York, 1970. [10] R. Fedorov, "Algebraic and Hamiltonian approaches to isostokes deformations," Transform. Groups 11 (2006), 137 -- 160. [11] E. Frenkel, Langlands correspondence for loop groups, Cambridge University Press, New York, 2007. [12] E. Frenkel and D. Gaitsgory, "Local geometric Langlands correspondence and affine Kac- Moody algebras," in Algebraic geometry and number theory, Progr. Math 253, Birkhauser Boston, Boston, MA, 2006, pp. 69 -- 260. [13] E. Frenkel and B. Gross, "A rigid irregular connection on the projective line," Ann. of Math. (2), 170 (2009), 1469 -- 1512. [14] V. Guillemin and S. Sternberg, Geometric Asymptotics, Mathematical Surveys, No. 14, Amer- ican Mathematical Society, Providence, 1977. [15] R. Howe and A. Moy, " Minimal K-types for GLn over a p-adic field," Ast´erisque 171/172 (1989), 257 -- 273. [16] M. Jimbo, T. Miwa, and K. Ueno, "Monodromy preserving deformations of linear ordinary differential equations with rational coefficients I," Physica D 2 (1981), 306 -- 352. [17] N. Katz, Exponential Sums and Differential Equations, Annals of Mathematical Studies 124, Princeton University Press, Princeton, NJ, 1990. [18] P. C. Kutzko, "Towards a classification of the supercuspidal representations of GLn," J. London Math. Soc. (2) 37 (1988) 265 -- 274. [19] A. H. M. Levelt, "Jordan decomposition for a class of singular differential operators," Ark. Mat. 13 (1975), 1 -- 27. [20] B. Malgrange, ´Equations Diff´erentielles `a Coefficients Polynomiaux, Progress in Mathemat- ics, Vol. 96, Birkhauser Boston, Inc., Boston, MA, 1991. [21] A. Moy and G. Prasad, "Unrefined minimal K-types for p-adic groups," Invent. Math. 116 (1994), 393-408. [22] D. S. Sage, "The geometry of fixed point varieties on affine flag manifolds," Trans. Amer. Math. Soc. 352 (2000), 2087 -- 2119. [23] J.-P. Serre, Algebraic Groups and Class Fields, Graduate Texts in Mathematics, Vol. 117, Springer-Verlag, New York, 1988. [24] W. Wasow, Asymptotic expansions for ordinary differential equations, Wiley Interscience, New York, 1976. [25] E. Witten, "Gauge theory and wild ramification," Anal. Appl. (Singap.) 6 (2008), 429 -- 501. Department of Mathematics, Louisiana State University, Baton Rouge, LA 70803 E-mail address: [email protected] E-mail address: [email protected]
1703.06805	1	1703	2017-03-20T15:36:31	Covering of elliptic curves and the kernel of the Prym map	[ "math.AG" ]	Motivated by a conjecture of Xiao, we study families of coverings of elliptic curves and their corresponding Prym map $\Phi$. More precisely, we describe the codifferential of the period map $P$ associated to $\Phi$ in terms of the residue of meromorphic $1$-forms and then we use it to give a characterization for the coverings for which the dimension of $\ker(dP)$ is the least possibile. This is useful in order to exclude the existence of non isotrivial fibrations with maximal relative irregularity and thus also in order to give counterexamples to the Xiao's conjecture mentioned above. The first counterexample to the original conjecture, due to Pirola, is then analysed in our framework.	math.AG	math	COVERING OF ELLIPTIC CURVES AND THE KERNEL OF THE PRYM MAP FILIPPO F. FAVALE AND SARA TORELLI Abstract. Motivated by a conjecture of Xiao, we study families of coverings of elliptic curves and their corresponding Prym map Φ. More precisely, we describe the codifferen- tial of the period map P associated to Φ in terms of the residue of meromorphic 1-forms and then we use it to give a characterization for the coverings for which the dimension of Ker(dP ) is the least possibile. This is useful in order to exclude the existence of non isotrivial fibrations with maximal relative irregularity and thus also in order to give counterexamples to the Xiao's conjecture mentioned above. The first counterexample to the original conjecture, due to Pirola, is then analysed in our framework. Introduction Hurwitz spaces were classically introduced and studied by Clebsh and Hurwitz (see [Cle72] and [Hur91]) as spaces parametrizing branched coverings of P1. Nowadays, the term Hurwitz space refers to a variety which parametrizes, up to equivalence, coverings π : F → E of curves with some geometric restrictions. In this article we will use a local version of Hurwitz spaces, namely a local family of coverings, whose seminal idea can be found in [Kan04]. Roughly, given a fixed covering π : F → E where E is an elliptic curve, one is able to construct a map p : F → E of curves over H, where H is a contractible open set. Then H is a parameter space for smooth coverings which share the same degree and the same ramification indices with π. Attached to a local family of coverings p : F → E with parameter space H there is the Prym map Φ, which associates to a b ∈ H the generalized Prym variety of πb = πFb : Fb → Eb, i.e. the connected component containing 0 of the kernel of the norm map Nm(πb). The Prym map is, in some sense, the analogous of the Torelli map T from Mg, the moduli space of curves of genus g, to Ag, the moduli space of principally polarized abelian varieties of dimension g. A celebrated theorem, the infinitesimal Torelli theorem, states that the differential of the Torelli map is injective outside the hyperelliptic locus of Mg and it should be interesting to have a similar theorem also for Prym maps or, at least, to their lifting P to a period domain. We will show that, in our case, i.e. when the base E is an elliptic curve, the dimension of the kernel of d oP∨ is at least 1 as a consequence of how the local families that we will use are constructed. Roughly, by composing a covering with a traslation of the base we always have coverings with the same Prym, so there is a tangent direction in the parameter space along which the Prym map is constant. Hence a question analogous to the one answered by the infinitesimal Torelli is Date: January 8, 2018. 2010 Mathematics Subject Classification. 14H40, 14H30, 14B10, 14J99. 1 Given a family of coverings with central fiber π, which conditions can we put on π in order to have that Ker(doP) has dimension 1? It is already known that an infinitesimal Torelli-like theorem for the Prym map cannot hold without restrictions as there are examples of coverings π : F → E (moreover with F non hyperelliptic) for which there are two independent directions along which d oP is 0. One of these examples, due to Pirola, will be analyzed in Section 4. This paper is devoted to the study of the Prym map Φ : H → A in the cases for which H parametrizes coverings over an elliptic curve. A further motivation to study this kind of problems comes from a conjecture about fibered surfaces. Recall that, given a fibration f : S → B of a smooth compact surface S over a smooth compact curve B, the relative irregularity qf is defined to be the difference q(S) − g(B). A modified version of a conjecture of Xiao states that, if f is not isotrivial, then (1) qf ≤(cid:24)g + 1 2 (cid:25) . The original conjecture was without the round up and has been modified after a coun- terexample of Pirola, the one that we will present in Section 4. To have an insight of what is known about the relative irregularity and about recent results about an upper bound a good reference is [BGN15]. The link between the world of non isotrivial fibrations and the one of the families of coverings we will define is broadly given as follows. The fibration f induces a surjective map alb(f ) : Alb(S) → Alb(B) = J(B) with dim(Ker(alb(f ))) = qf , which has a connected component containing 0. We shall denote it with Kf . If B0 is the open subset of B over which the fibration has smooth fibers, we denote by Fb the fiber over b ∈ B0. Via the map Fb ֒→ S we have a map JFb → Alb(S) whose image is, up to translation, exactly Kf . Dualizing we have a map K ∨ f / JF ∨ b = JFb Note that K ∨ f doesn't depend on b whereas Fb strongly depends on it. In particular we have proved that the Jacobian of every smooth fiber of a non isotrivial fibration contains a fixed abelian variety of dimension qf . Assume now that we are in an extreme case, i.e., assume that qf = g − 1. Since in this article we are only interested in non isotrivial fibrations, we will call fibration with maximal relative irregularity those with qf = g − 1. In fact, every fibration satisfies 0 ≤ qf ≤ g and the equality qf = g holds if and only if the fibration is trivial (this follows from a result of Beauville: see the appendix of [Deb82] for details). In this case dim(K ∨ f ) = qf = g − 1 and we can consider the quotient JFb/K ∨ f which will be an abelian variety of dimension g − qf = 1: an elliptic curve Eb. Φ(πb) = K ∨ f / JFb Fb pb / Eb >⑤⑤⑤⑤⑤⑤⑤⑤ πb / Eb πb Moreover, in this case K ∨ f is the connected component through the origin of the kernel of the norm map associated to the ramified covering πb : Fb → Eb, i.e. the Prym variety Φ(πb). Hence, an eventual counterexample to the modified version of the conjecture of 2 / / / / / > O O / Xiao, under the additional assumption qf = g − 1, would give a family of coverings of elliptic curves with constant Prym variety. At the moment the question Is there a non isotrivial fibration (with maximal relative irregularity qf or not) giving a counterexample to the modified Xiao's conjecture? is still completely open but, by answering precisely to our first question one should be able to construct counterexamples or to prove that, at least for the case of maximal relative irregularity, such examples cannot exist. It is worth to mention that, by the original work of Xiao (see [Xia87]), a non isotrivial fibration with maximal relative irregularity can exist only if g ≤ 7. The paper is organized as follows. In Section 1 we recall some definitions and facts about Prym varieties associated to ramified coverings and Prym maps that we are going to use extensively in what follows. In Section 2 we will extend the techniques developed in [Kan04] for coverings with simple ramification to the case of arbitrary one. The main result is this theorem Theorem (2.3). With the notations of section 2, for any ϕ ∈ Sym2(H 0(ωF )−) we have (2) doP∨(ϕ) = n Xj=1 Resaj(cid:18)m(ϕ) π∗α (cid:19) dtj + n Xk=0 m(ϕ) π∗α2 (xk)! ds. that describe the (dual of the) differential of the Period map in terms of residues of some meromorphic forms. In Section 3, given a covering π : F → E and assuming that F is not hyperelliptic, we prove Theorem 3.4, a geometric criterion on the canonical model F that is a sufficient condition in order to have dim(Ker(d oP)) = 1. Finally, in Section 4, we analyze in our framework the family that was constructed in [Pir92]. We will prove, using our framework, that the existence of the family is consistent with our Theorem as well as other interesting geometric aspects that may suggest a different way to approach, in the future, the problem of finding an answer to the second question by starting from the geometry of canonical models. Acknowledgements. The authors are both supported by INdAM - GNSAGA. The first author is supported by FIRB2012 "Moduli spaces and applications" and by IMUB during his stay at the University of Barcelona. The seminal ideas at the base of this article were born in the framework of the PRAGMATIC project of year 2016. We would like to thank Miguel ´Angel Barja, V´ıctor Gonz´alez-Alonso, Juan Carlos Naranjo and Gian Pietro Pirola for introducing us to the subject and for all the helpful remarks on the problem and the manuscript. In this section we recall some definitions that we are going to use in the following sections. 1. Some preliminaries Let F, E be two smooth curves of genus g ≥ 2 and 1 respectively and consider the covering π : F → E. One can consider the Albanese variety associated to F , which coincides with its Jacobian, because F is a curve. Namely (3) J(F ) = H 0(ωF )∨ H1(F, Z) = Alb(F ) = 3 H 1(OF ) H 1(F, Z) . This is a principally polarized abelian variety of dimension g. As E has genus 1 we have E = J(E) = Alb(E). By the universal property of Alb(F ) there is a map alb(π) such that the diagram F π (cid:9) E = JF / JE alb(π) commutes, where the map F → JF is the Albanese map of F , also called the Abel- Jacobi map. The map alb(π) is also called the norm map of π, Nmπ, and it is surjective. The generalized Prym variety associated to π : F → E (or simply Prym variety) is the connected component of Ker(alb(π)) that contains the 0, i.e. (4) P (π) = Ker(alb(π))0. P (π) is an abelian variety of dimension g − 1 with a natural polarization ΘP given by ΘJ F P via the embedding P (π) / JF. The map π : F → E induces a map trπ : H 0(ωF ) → H 0(ωE) called the trace of π (see Appendix A of [Kan04] for the definition). The trace satisfies If we define (5) trπ ◦ π∗ = Deg(π) IdH 0(ωE ) . H 0(ωF )− = Ker(trπ) we have a canonical splitting (6) H 0(ωF ) = π∗H 0(ωE) ⊕ H 0(ωF )− and we can identify the quotient H 0(ωF )/π∗H 0(ωE) with H 0(ωF )−. tangent bundle of P (π) can be described as In particular, the (7) T P (π) =(cid:18) H 0(ωF ) π∗H 0(ωE)(cid:19)∨ ⊗ OP (π) = (H 0(ωF )−)∨ ⊗ OP (π). Now we will introduce the families of coverings of elliptic curves we are interested in. Fix a smooth curve F of genus g ≥ 2 and consider a degree d covering π : F → E, where E is an elliptic curve. Denote with n R = (nj − 1)aj Xj=1 the ramification divisor and call bj the branch point corresponding to the ramification point aj, i.e. π(aj) = bj. Thus nj is the degree of π when restricted to a suitable neighborhood of aj. Fix a generator α of H 0(ωE). Choose a suitable set {∆j} of coordinate neighborhoods centered in the points bj and call wj the corresponding coordinate on E. This is not needed at the moment but observe that we can assume that α∆j = dwj. We can chose a collection of pairwise disjoint coordinate neighborhoods (Uj, zj) centered in aj in such a way that wj = πUj (zj) = znj j . 4 / / / / / / / / Denote by HE the polydisc Πn the relation j=1∆j and consider the coordinates t = (tj)n j=1 defined by We can consider, as in Section 4.1 of [Kan04], a family tj(P1, · · · , Pn) = wj(Pj). (Ψ, f ) : F → E × HE of d-sheeted branched coverings deforming π parametrized by HE such that (8) wj = ΨUj (zj, t) = znj j + tj. In this way, to each b′ ∈ HE, it is associated a covering πb′ : Fb′ → E which is a deformation of π, the central fiber. Note that (8) forces the ramification orders to remain costant and allows different branch points to move indipendently. This is what we will call in the following local family of coverings over E with central fiber π parametrized by HE. The tangent space to HE in b = (b1, . . . , bn) ∈ HE is TbHE ≃ Tbj E ≃ n Mj=1 C ∂ ∂tj , n Mj=1 where the tangent vectors on the right are evaluated in 0. Indeed, following We can also take into account the deformation of the elliptic curve. [ACG11], if one chooses c ∈ E not among the bj and considers a small coordinate neigh- borhood (N, v) of c (eventually shrinking ∆j in such a way that for all j they are disjoint from N), one can consider the associated Schiffer variation E → N of E with coordinate s. Observe that we can assume αN = dv. Taking into account also the movement of the branch points one has a family f : F → HE × N of curves of genus g that fits into the diagram (9) p / E F f HE × N / N For a choice (b′, s′) ∈ H = HE × N we have an elliptic curve Es′, the fiber of the map E → N over s′, a curve F(b′,s′) of genus g and a covering π(b′,s′) = pF(b′,s′) : F(b′,s′) → Es′. For this reason, the map p is what we will call local family of coverings with central fiber π parametrized by H or, simply, local family of coverings. The tangent space to H in (b, s) is T(b,s)H ≃ n Mj=1 C ∂ ∂tj! ⊕ C ∂ ∂s . and, clearly, containts TbHE in a natural way. We stress that, through the whole article, unless otherwise stated, we will always refer to the families of coverings constructed in this sections. If we have a family of coverings parametrized by H, for each (b, s) we can construct the Prym variety associated to the covering. Moreover, the type of polarization remains 5 / / constant. Hence we can consider the Prym map (10) H Φ / Ag−1 (b, s) ✤ / [P (π(b,s))] where Ag−1 is the moduli space of abelian varieties with polarization (which will be omitted) equal to the one of the central fiber. In the same way one has the Prym map ΦE associated to a local family of coverings over E. To avoid technical subtleties around singular points of Ag−1, we will consider the period map P : H → D (or PE : HE → D) instead of the Prym map Φ (respectively ΦE), where D is a suitable period domain for Ag−1. The interested reader is referred to [Kan04, Section 3] for technical details. Through the whole article, giving two sections s1, s2 ∈ H 0(OX (D)) we will write s1 ⊗ s2 to mean their symmetric product, i.e. 1 2 (s1 ⊗ s2 + s2 ⊗ s1) ∈ Sym2(H 0(OX(D))). If si ∈ H 0(OX (Di)), s1 · s2 will mean the evaluation of s1 ⊗ s2 in H 0(OX (D1 + D2)) under the multiplication map. 2. A direct formula for the codifferential of the Prym map In this section we will prove an explicit formula for the codifferential of the period map in terms of the residue at the ramification points of some forms. The framework is similar to the one in [Kan04] with the main difference being that we don't restrict ourselves to the case of simple ramification. First of all we introduce some notations. Fix an elliptic curve E and let π : F → E be a covering of E with F of genus g. Consider (Ψ, f ) : F → E × HE, the local family of coverings with fixed base E, central fiber π and parameter space HE constructed in Section 1. By construction, it induces a family f : F → HE with central fiber Fo = F . If we consider a minimal versal deformation f ′ : F ′ → M of F then the previous family is induced by f ′ by means of a pullback. More precisely there exists a holomorphic map hE : HE → M such that (11) F f HE / F ′ f ′ / M hE is commutative. Being f ′ a minimal versal deformation we have ToM ≃ H 1(TF ) ≃ H 0(ω⊗2 F )∨. Moreover, under this identification, if we take a tangent vector v in ToHE and evaluate dhE in v we get the Kodaira-Spencer map KSE associated to F → HE evaluated in v. We are able to prove the first important part of Theorem 2.3. 6 / / / / Proposition 2.1. Using the identifications introduced above, we have that dh∨ E : T ∨ o M → T ∨ o HE can be written as (12) dh∨ E(ϕ) = and ϕ ∈ T ∨ o M = H 0(ω⊗2 F ). γjdtj where n Xj=1 γj = 2πi Resaj(cid:16) ϕ π∗α(cid:17) Proof. For every ϕ ∈ H 0(ω⊗2 M in o, by the complex numbers γj such that F ) we have that dh∨ E(ϕ) is identified, as cotangent vector on n By construction, we can obtain these numbers simply by pairing dh∨ E(ϕ) against ∂ ∂tj : dh∨ E(ϕ) = γjdtj. Xj=1 γj = dh∨ E(ϕ)(cid:18) ∂ ∂tj(cid:19) = ϕ(cid:18)dhE(cid:18) ∂ ∂tj(cid:19)(cid:19) = ϕ(cid:18)KSE(cid:18) ∂ ∂tj(cid:19)(cid:19) . In order to develop the computation we may proceed using a description of KSE in terms of the Cech cohomology (details of this can be found in [Hor73]). To do it consider the exact sequence (13) 0 / TF dπ / π∗TE ψ / R / 0 and let δ be the coboundary map H 0(R) → H 1(TF ). Then KSE factors as δ ◦ τ = KSE where τ : TbH → H 0(R) is the characteristic map of the family (see [Hor73] for the definition and the proof of this fact). Hence we can unfold the calculation using these exact sequences. If one restricts the exact sequence (13) on Uj (or some sufficiently small subset of this coordinate neighborhood), it can be identified with (14) 0 / OUj ∂ ∂zj dπ / OUj ∂ ∂wj ψ / RUj / 0. ∂ ∂zj to njznj −1 ∂ The first map sends while the second one is simply the restriction to the ramification locus. Let U = {U0, U1, . . . , Un} where Uj for j = 1, . . . , n are the neighborhoods defined above and U0 = F \ {aj}. Let, as usual, Uα,β, be a shorthand for Uα ∩ Uβ with α < β. If η = [ηj] ∈ H 0(U, R) with η0 = 0 and ηj = pj(zj) ∂ ∂wj we have ∂wj j δ (η) = [λα,β] with λ0,j = pj(zj) njznj −1 ∂ ∂zj for j > 0 and λα,β = 0 if α, β > 0. Following [Hor73] and using Equation (8) we have (15) Hence we have τ(cid:18) ∂ ∂tj(cid:19) = [τ (j) k ] KSE(cid:18) ∂ ∂tj(cid:19) = δ(cid:18)τ(cid:18) ∂ ∂tj(cid:19)(cid:19) = [χ(j) α,β] with with 7 τ (j) k =(0 ∂ ∂wj k 6= j k = j. χ(j) α,β =( 1 nj znj −1 0 ∂ ∂zj (α, β) = (0, j) otherwise. / / / / / / / / If ϕ ∈ H 0(ω⊗2 F ) we can represent it as Cech-cocycle as [φj] where φ0 = φU0 and φj = qj(zj)dz2 j are the local expressions of ϕ in coordinates around aj. The numbers we are interested in are simply the ones obtained by considering the perfect pairing (16) H 0(ω⊗2 F ) ⊗ H 1(TF ) / H 1(ωF ) ≃ / C applied to KSE(cid:16) ∂ ∂tj(cid:17) and ϕ. Using Cech cohomology, the image in H 1(ωF ) of our product is given by the Cech class [ǫ(j) α,β] with ǫ(j) α,β =( qj(zj ) nj znj −1 dzj 0 (α, β) = (0, j) otherwise. What remains to be proven is the analogous to the calculation of [Kan04] for the case of simple ramification: roughly, one can adapt the techniques of [ACG11, pag. 14-15] to develop the last isomorphism of (16) in order to finally get γj = 2πi Res0 qj(zj)dz2 j njznj −1 dzj j = 2πi Resaj ϕ π∗α . (cid:3) Consider now the family p : F → E with central fiber π : F → E and parameter space H = HE × N as defined in Section 1. As before, we have an induced deformation f : F → H of F , its associated Kodaira-Spencer map KS and, when a minimal versal deformation f ′ : F ′ → M of F is chosen, an holomorphic map h : H → M such that (17) F f H / F ′ f ′ / M h is commutative. Again, as ToM ≃ H 1(TF ), we can identify dh with KS. We will denote by x1, . . . , xd the points of the fiber of π over the point c which, by construction, are all different. Proposition 2.2. Using the identifications introduced above, we have that dh∨ : T ∨ o M → T ∨ o H can be written for any ϕ ∈ H 0(ω⊗2 F ) = T ∨ j=1 γjdtj + γds where γj = 2πi Resaj(cid:16) ϕ o M as dh∨(ϕ) =Pn Xk=1 π∗α(cid:17) and γ = 2πi d ϕ π∗α (xk). (18) Proof. As before, by duality, dh∨(ϕ) = ϕ ◦ dh = ϕ ◦ KS. It is then clear that the formula for γj follows directly from Proposition 2.1. The one that gives γ, as it involves calculations done far from the ramification points, doesn't depend on the type of the ramifications. Hence, the one given in [Kan04] when π as only simple ramification is still valid. (cid:3) 8 / / / / Recall that we have a decomposition of H 0(ωF ) given by H 0(ωF )− ⊕ π∗H 0(ωE) where the first space is the vector space of 1-forms on F with trivial trace. This induces a decomposition on Sym2(H 0(ωF )). Unless otherwise specified, consider Sym2(H 0(ωF )−) as a subspace of Sym2(H 0(ωF )) in the natural way. Let m : Sym2(H 0(ωF )2) → H 0(ω⊗2 F ) be the multiplication map. Denote by P : H → D the period map associated to the Prym map Φ : H → Ag−1 where D is a suitable period domain. We are ready to prove Theorem 2.3. Theorem 2.3. With the notation introduced in this section, for any ϕ ∈ Sym2(H 0(ωF )−) we have (19) doP∨(ϕ) = n Xj=1 Resaj(cid:18)m(ϕ) π∗α (cid:19) dtj + d Xk=1 m(ϕ) π∗α2 (xk)! ds. Proof. Theorem 3.21 of [Kan04] expresses the codifferential of the period map calculated in ϕ ∈ Sym2(H 0(ωF )−) and paired with ∂ ∂tj as ϕ(cid:18)KS(cid:18) ∂ ∂tj(cid:19)(cid:19) without any restriction on the ramification type. with Proposition 2.2 ends the proof of the Theorem. In particular, this formula, together (cid:3) Remark 2.4. As a consequence of the last Theorem we can conclude that, if we fix E, the codifferential doP∨ E : Sym2(H 0(ωF )−) → T ∨ o HE factors as (20) H 0(ω⊗2 F ) dT ∨ Sym2 (H 0(ωF )) dh∨ E T ∨ o HE σ Sym2(H 0(ωF )−) doP∨ E where T is the Torelli map (so that m = dT ∨) and σ is the lifting of the projection of Sym2(H 0(ωF )) → Sym2(H 0(ωF )−) induced by the decomposition H 0(ωF ) = H 0(ωF )− ⊕ π∗H 0(ωE). The commutativity of the diagram is a consequence of Proposition 2.1 as, for any ϕ ∈ H 0(ωF ) ⊗ π∗H 0(ωE), we have that ϕ/π∗α is holomorphic and hence has residue zero everywhere. 3. A geometric approach via the canonical embedding In this section we will use the technical result of the previous section in order to prove that dim(Ker(d oPE)) = 1 for arbitrary ramification types and a geometric criterion to determine whether dim(Ker(d oP)) = 1 or not. First we fix some notation and facts about the canonical curves that we are going to use extensively in the following. As F has genus g ≥ 3 and is not hyperelliptic, we may identify it with its canonical model in P = PH 0(ωF )∨. This is a non-degenerate curve of degree 2g − 2, which is also projectively normal by a classical result of Max Noether (see, for example, [ACGH85]). One of the consequences of this fact is that the multiplication map mk : Symk H 0(ωF ) → H 0(ω⊗k F ) is surjective. As before we will denote m2 simply by m. We will use frequently the natural identifications H 0(OP(d)) = Symd H 0(ωF ) which enable us to identify P(Ker(md)) 9 o o ? O O o o with the space of hypersurfaces of degree d in PH 0(ωF )∨ that contain F . By abuse of notation we will simply say that an element in Symd H 0(ωF ) is an hypersurface of degree d if no confusion arises. In particular, if IF is the ideal sheaf of F in PH 0(ωF )∨, then Ker(m) = H 0(IF (2)) gives the set of all quadrics in PH 0(ωF )∨ containing the curve F , and has dimension (g−2)(g−3) . 2 Recall that the decomposition H 0(ωF ) = H 0(ωF )− ⊕ π∗H 0(ωE) where the first space is the space of forms with zero trace. Since elements in H 0(ωF ) are linear equations on PH 0(ωF )∨, all the hyperplanes defined by elements in H 0(ωF )− intersect in a single point q− of P which is a point really important in what will follows. We have also a particular hyperplane, the one defined by the subspace π∗H 0(ωE) which will be denoted by H −. More precisely, q− = P((H 0(ωF )−)⊥) and H − = P((π∗H 0(ωE))⊥) As before, we will fix a generator α of H 0(ωE) so that (21) Sym2(H 0(ωF )) = Sym2(H 0(ωF )−) ⊕(cid:0)π∗α ⊗ H 0(ωF )(cid:1) . Given a quadric Q in P we will denote by GQ ∈ Sym2(H 0(ωF )) one of its equations and by G− Q ∈ Sym2(H 0(ωF )−) and ωQ ∈ H 0(ωF ) the only elements such that GQ = G− Q + π∗α ⊗ ωQ under the decomposition (21). Finally, given a quadric Q, we will denote by Q− the cone given by the equation G− Q, i.e. the quadric such that GQ− = G− In order to prove Theorem 3.2 we will need the following result: Q− = G− Q. Lemma 3.1. We have a natural inclusion of H 0(IF (2)) in Ker(doP∨ E). Proof. Recall that, fixed a family of coverings with base E and central fiber π : F → E, by fixing a minimal versal deformation F ′ → M of F , we can construct hE : H → M like in diagram (11). As observed in Remark 2.4 we have a commutative diagram (22) 0 0 0 / Ker d oP∨ E γ / H 0(IF (2)) j ι / Sym2(H 0(ωF )−) doP ∨ E / T ∨ o HE pr dh∨ E / Sym2(H 0(ωF )) m / H 0(ω⊗2 F ) / 0 H 0(ωF ) ⊗ π∗H 0(ωE) 0 10 / / O O / / ? O O ✤ ✤ ✤ / O O O O / / / O O / ? O O O O It is easy to see that the image of pr ◦ ι lives in Ker(d oP∨ E) so we have a well defined map γ : H 0(IF (2)) → Ker(d oP∨ E). We want to prove that this map is indeed injective. This follows from the geometry of the problem. Indeed, if a quadric Q contains F , i.e. if the quadric has equation GQ = G− Q + π∗α ⊗ ωQ ∈ H 0(IF (2)), and if γ(GQ) = 0 then we have that the quadric has equation π∗α ⊗ ωQ. But this is impossible because such a quadric the union of two planes (one of which is H −) and the canonical curve is non-degenerate. Hence γ is injective. (cid:3) Theorem 3.2. Let π : F → E be a covering with F non-hyperelliptic, consider a local family of coverings with base E and parameter space HE constructed in Section 1. Let PE be the period mapping associated to the Prym map ΦE. Then dim(Ker(doPE)) = 1. Proof. First of all, observe that for dimensional reasons, one has dim(Ker(d oPE)) = 1 if and only if dim(Ker(d oP∨ E)) = g(g − 1) 2 − n + 1. From the splitting H 0(ωF ) = H 0(ωF )− ⊕ π∗H 0(ωE) we have the commutative diagram 0 / H 0(ωF ) ⊗ π∗α / Sym2(H 0(ωF )) pr Sym2(H 0(ωF )−) 0 )❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ Ψ doP ∨ E ToHE with Ψ defined by extending the formula in Theorem 2.3 to Sym2(H 0(ωF )). This can be E(H 0(ωF ) ⊗ π∗α) = {0}. In done because, as previously observed (see Remark 2.4), d oP∨ particular, we have the relation (23) dim(Ker(d oP∨ E)) = dim(Ker(Ψ)) − dim(Ker(pr)) = dim(Ker(Ψ)) − g. By definition, Ψ factors through the multiplication map m as Ψ = ¯Ψ ◦ m. The map ¯Ψ is well defined as, by Lemma 3.1, Ker(m) ⊂ Ker(Ψ). 0 / H 0(ωF ) ⊗ π∗α / Sym2(H 0(ωF )) pr Sym2(H 0(ωF )−) 0 m H 0(ω⊗2 F ) Ψ )❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ ¯Ψ doP ∨ E / ToHE Being m surjective (as F is non-hyperelliptic) we obtain the further relation (24) dim(Ker(Ψ)) = dim(Ker( ¯Ψ)) + dim(Ker(m)) = dim(Ker( ¯Ψ)) + (g − 2)(g − 3) 2 . As the divisor associated to π∗α is exactly R, the ramification divisor, we have that ωF = OF (R) and there is an exact sequence (25) 0 / ωF ·π∗α / ω⊗2 F 11 / ω⊗2 F R / 0 / / / / ) / / / / / / ) / / / / / / / which yields, denoting with V the quotient H 0(ω⊗2 F )/(H 0(ωF ) · π∗α), the exact sequences (26) 0 0 / H 0(ωF ) ·π∗α / H 0(ω⊗2 F ) ǫ / V ζ / V / H 0(ω⊗2 F R) / H 1(ωF ) / 0 / 0 Let η ∈ Ker(ǫ). We want to prove that ¯Ψ(η) = 0. This is easily proven: write η as ω · π∗α and observe that ¯Ψ(η) = ( ¯Ψ ◦ m)(ω ⊗ π∗α) = (d oP∨ E ◦pr)(ω ⊗ π∗α) = 0 because ω ⊗ π∗α ∈ Ker(pr). In particular, Ker(ǫ) ⊂ Ker(m) and we can define a map λ : V → T ∨ o HE such that ¯Ψ = λ ◦ ǫ. Moreover (27) dim(Ker( ¯Ψ)) = dim(Ker(λ)) + g. Using the second exact sequence in 26 we can also define a map µ : H 0(ω⊗2 such that µ ◦ ζ = λ. F R) → ToHE (28) H 0(IF (2)) H 0(ωF ) ⊗ π∗α Sym2(H 0(ωF )) pr Sym2(H 0(ωF )−) ·π∗α '❖❖❖❖❖❖❖❖❖❖❖ m H 0(ω⊗2 F ) Ψ ¯Ψ ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ 8rrrrrrrrrrrrr %▲▲▲▲▲▲▲▲▲▲▲▲ λ ❱❱❱❱ ǫ doP ∨ E T ∨ o HE µ V ζ / H 0(ω⊗2 F R) / H 1(ωF ) Note that we have several ways to define µ. Since ω⊗2 ω⊗2 F R = ωF (R)R the global sections of F R are just collections of meromorphic tails on the points of ramification, i.e. elements (nk−1 Xj=1 βjk dzk zj k ) ak ∈R where nk is the ramification index of the point ak. In particular, we can define µ as the map which gives the residue in the corresponding point of the meromorphic tail. This ensures that the diagram is commutative. In addition, µ is surjective (this because the image of a collection of meromorphic tails {sk}, one for each point of ramification, with β1m = δkm, generates the image), and as a consequence, ζKer(λ) is an isomorphism between Ker(λ) and Ker(µ). Hence, (29) dim Ker(d oP∨ E) = dim(Ψ) − g = dim( ¯Ψ) + dim(Ker(m)) − g = = dim(λ) + dim(Ker(m)) = dim(µ) + dim(Ker(m)) = = h0(ωF (R)R) − dim T ∨ o HE + dim(Ker(m)) = as wanted. Now we will prove the first main theorem: 12 g(g − 1) 2 − n + 1 (cid:3) / / / / / / / / _ / / t ' / / / / * / / % % % * 8 / O O / / / Theorem 3.3. Let π : F → E be a covering with F non-hyperelliptic, consider the local family of coverings with parameter space H constructed in Section 1. Let P be the period mapping associated to the Prym map Φ : H → Ag−1. Using the same notations of Theorem 2.3 we have dim(Ker(doP))) = 1 ⇐⇒ ∃β ∈ Ker(doP∨ E) Proof. First of all consider the diagrams m(β) π∗α2 (xk) 6= 0. d Xk=1 doP / TP (o)D 9sssssssss doPE ToH ToHE doP ∨ T ∨ o H T ∨ P (o)D ysssssssss doP ∨ E T ∨ o HE and observe that one always has Ker(d oPE) ⊆ Ker(d oP) Ker(d oP∨) ⊆ Ker(d oP∨ E). Moreover, the codimensions are at most 1. If one considers the exact sequences 0 0 / Ker(d oPE) / ToHE / TP (o)D / Ker(d oP) / ToH / TP (o)D / Ker(d oP∨ E)∨ / Ker(d oP∨)∨ / 0 / 0 it is clear that Ker(d oPE) = Ker(d oP) if and only if Ker(d oP∨) ( Ker(d oP∨ have E). Hence we dim(Ker(d oP))) = 1 ⇐⇒ Ker(d oP∨) ( Ker(d oP∨ E). This is true if and only there exists an element β ∈ Ker(d oP∨ vanish. This can only be possible if d oP∨(β) is not zero on ∂ taking into account the moduli of the elliptic curve. By using Theorem 2.3 we have E) on which d oP∨ doesn't ∂s , where s is the parameter and this concludes the proof. d oP∨(β) = m(β) π∗α2 (xk) d Xk=1 (cid:3) This result improves the one in [Kan04] where it is proved only for simple ramification. In the same work is proved that, for simple ramification, having the sum in Theorem 3.3 different from zero for some β ∈ Ker(d oP∨ E) is equivalent to ask that the intersection of the quadrics that contain the canonical model of F doesn't contain the point q− defined before. Unfortunately, in the case of arbitrary ramification, we are not able to prove this equivalence but only one implication. Theorem 3.4. With the same hypotesis of Theorem 3.3, if we identify F with its canonical model in PH 0(ωF )∨, then we have (30) Q =⇒ dim(Ker(doP)) = 1, q− 6∈ \F ⊂Q where Q ranges in the set of quadrics of PH 0(ωF )∨ containing F . The proof of the theorem uses some arguments developed in [Kan04] that we have sum- marized in the following Lemma. 13 / 9 ? O O o o y / / / / / / / / / / Lemma 3.5. Let Q be a quadric of PH 0(ωF )∨ containing F and denote by GQ = G− π∗α ⊗ ωQ one of its equations. Then Q + m(G− Q) π∗α2 (xk) = 0 ⇐⇒ GQ(q−) = 0 ⇐⇒ q− ∈ Q. d Xk=1 Proof. The last statement is clear by definition so we really need to prove only the first one. First of all observe that we can choose the coordinate s in such a way that α is Q = GQ − π∗α ⊗ ωQ and Q ∈ H 0(IF (2)) = Ker(m) by locally given by ds. Then, as G− hypotesis, one has m(G− Q) π∗α2 (xk) = − m(π∗α ⊗ ωQ) π∗α2 (xk) = − T rπ(ωQ) α (c). d Xk=1 But T rπ(ωQ) is an element of H 0(ωE) so it is equal to r · α for some r. Thus we have d Xk=1 d Xk=1 m(G− Q) π∗α2 (xk) = −r which is zero if and only if ωQ has trace 0, i.e. if and only if ωQ ∈ H 0(ωF )−. This happens if and only if (π∗α)⊗2 doesn't appear in the equation of Q, i.e. if and only if q− ∈ Q. (cid:3) Using Lemma 3.1 and Lemma 3.5 the proof of Theorem 3.4 is straightforward. Proof of Theorem 3.4. Assume that Q. q− 6∈ \F ⊂Q Then, there exists a quadric which cointains F but doesn't contain q−. Denote by GQ its equation. By Lemma 3.1 we know that β = γ(GQ) = G− E) and by Lemma 3.5 we have that Q ∈ Ker(d oP∨ Hence, using Theorem 3.3 we have the thesis. m(β) π∗α2 (xk) 6= 0. d Xk=1 (cid:3) Remark 3.6. In [Kan04], with different methods, it is proved that H 0(IF (2)) = Ker(doP∨ E) if the ramification is simple. This fact is exactly what allows to prove the converse impli- cation of Theorem 3.4. Remark 3.7. Notice that H 0(IF (2)) = Ker(doP∨ E) if and only if all the ramification indices are equal to 2. Indeed, denote by Rred the reduced divisor whose support equals the support of the ramification divisor. Let ¯R be R − Rred. From Riemann-Hurwitz we have 2g − 2 = deg(R) = deg(Rred) + deg( ¯R) = n + deg( ¯R). Hence, from Equation (29) one has dim Ker doP ∨ E = h0(IF (2)) + deg( ¯R). As ¯R ≥ 0 and is trivial if and only if all the ramification indices are equal to 2 the claim follows. In particular, the converse implication of (30) in Theorem 3.4 holds for coverings whose ramification indices are all equal to 2. 14 We conclude this section by proving the existence of an exact sequence which should help to measure, in a more intrinsic way, how much H 0(IF (2)) and Ker(d oP∨ E) differ. Proposition 3.8. Under the hypotesis of Theorem 3.4 there is an exact sequence (31) 0 / H 0(IF (2)) γ / Ker(doP∨ E) Ker(dh∨) H 0(ωF ) ⊗ π∗H 0(ωE) / 0. Proof. Starting from diagram (22) it is easy to see that the composition of the inclusion of H 0(ωF ) ⊗ π∗H 0(ωE) with m has image in H 0(ω⊗2 F ) but also in the kernel of dh∨. Hence there is a map ǫ : H 0(ωF ) ⊗ π∗H 0(ωE) → Ker(dh∨), which is easily proven to be injective as we have done with γ. We can also complete the diagram on the right by adding two (trivial) vertical arrows. The complete diagram looks like this (32) 0 0 0 / Ker d oP∨ E γ / H 0(IF (2)) j ι / Sym2(H 0(ωF )−) doP ∨ E / T ∨ o HE / (Ker d oPE)∨ / 0 pr dh∨ Sym2(H 0(ωF )) m / H 0(ω⊗2 F ) / 0 0 / H 0(ωF ) ⊗ π∗H 0(ωE) ǫ Ker dh∨ / Coker ǫ 0 By using the snake lemma on the central columns one obtain the wanted sequence. (cid:3) 0 0 4. An interesting family of curves In this section we review the first example, due to Pirola, of a non-trivial family of coverings of elliptic curves with 2 independent directions along which the Prym map Φ is constant. Hence the kernel of the differential of the Period map associated to Φ has dimension greater than 1. The existence of the family is proved in [Pir92] but the proof is not constructive and uses a framework different form ours. After some notations and a brief idea of how to prove the existence of this family (for details, see [Pir92]), we will prove that q− belongs to the only quadric that contains F and that for all the elements of Ker(d oP∨ E) the sum in Theorem 3.3 is 0. In order to prove the existence of such a family, let G ≃ Z3 and consider the space HG of Galois coverings π : F → E of degree 3 with ramification given by 3 points (so the number of branch points is exactly 3 and the genus of F is 4) modulo the identifications 15 / / / / / / / O O / / / / / / ? O O O O O O / / / O O / O O O O / / / ? O O O O / / / O O O O O O given by a commutative diagram like F1 ≃ /❴❴❴ F2 π1 π2 E1 ≃ /❴❴❴ E2 With this type of identification of two coverings the dimension of HG is 3. Note that, with this definition, a covering π : F → E and the covering obtained by composing π with a translation of E are equivalent: they represent the same point in HG. Fix a generator g of G and ρ, a primitive root of 1 of order 3. If V is a vector space on which G acts, we will denote by Vρk the subspace where g acts as the multiplication by ρk. As π is the quotient by the group G, the G-action on F induces several other G-actions. We will do now a small list of the one that we are going to use in this section. a) The canonical action on H 0(ωF ) via pullback: by changing, if necessary, g with g2, we have (33) H 0(ωF ) = H 0(ωF )1 ⊕ H 0(ωF )ρ ⊕ H 0(ωF )ρ2 = π∗H 0(ωE) ⊕ C2 ρ ⊕ C1 ρ2. b) A canonical G-action on H 0(ωF )− which is simply the restriction of the canonical representation on H 0(ωF ). c) An action on Sym2(H 0(ωF )), whose decomposition in irreducible subrepresentations is given by Sym2(H 0(ωF )) = C3 1 ⊕ C3 ρ ⊕ C4 ρ2. d) an action on H 0(ω⊗2 F ) using the surjectivity of m by imposing that m becomes a morphism of G-vector spaces and hence on its dual H 1(TF ). e) An action on H 0(IF (2)) as the kernel of m. f) An action on the Prym Φ(π): this is induced at level of tangent spaces (as the tangent space T0Φ(π) is H 0(ωF )−) and it is compatible with the quotient by the periods' lattice. g) An action of G on PH 0(ωF )∨ = P as every automorphism of F , seen as a canonical curve in P lifts to an automorphism of the whole space. All these actions, by construction, are compatible via the usual identification. For exam- ple, if we interpret H 0(ωF ) as the space of equations of hyperplanes of P an invariant hyperplane in P has an equation which is an eigenvector of g in H 0(ωF ). One has a Prym map Φ : HG → Ag−1 and a period map P : HG → D. We stress that, by construction, if we prove that dim(Ker(do P )) > k then, the period map P associated to the Prym map of a local family of coverings with π as central fiber will have kernel of dimension at least k + 1. The rough idea to prove that there exists a family of coverings in HG which gets contracted by Φ is to observe, as we have done in f), that the Prym map Φ factors through the inclusion of AG g−1, the space of abelian varieties of dimension g − 1 with an action of G, in Ag−1. If we denote by DG a period domain for AG g−1 we have an analogous period map P G : HG → DG. We want to get a bound on the dimension of the image d P G. Clearly, the image of d P G has dimension at most the dimension of T DG = Sym2(H 0(ωF )−)G 16 / / and the same bound holds, by construction, for the dimension of the image of d P . As Sym2(H 0(ωF )−)G is isomorphic, by b), to H 0(ωF )ρ ⊗ H 0(ωF )ρ2, we have that its dimension is 2. As T[π]HG has dimension 3 this implies that the kernel of d P has dimension at least 1 and the existence of the family is proved. Proposition 4.1. Let π : F → E with [π] ∈ HG, identify F with its canonical model and let Q be the only quadric containing F . Then q− ∈ Q and either H 0(IF (2)) ⊂ Sym2(H 0(ωF ))ρ or H 0(IF (2)) ⊂ Sym2(H 0(ωF ))ρ2. Proof. There exists only a quadric containing F because g(F ) = 4. More precisely F is the complete intersection of a quadric Q and a cubic surface C. Let GQ ∈ H 0(IF (2)) be an equation for Q. Being F invariant under the G-action introduced in g), we have that the orbit of GQ under the action given in e), is simply given by itself plus, possibly, some of its multiple by elements in C∗. The key point now is to see that H 0(IF (2))1 = 0. In order to prove this observe that, by construction, we have an exact sequence of G-vector spaces given by 0 / H 0(IF (2)) / Sym2(H 0(ωF )) / H 0(ω⊗2 F ) / 0 Hence, by taking invariant parts and dimensions we have dim(H 0(IF (2))G) = dim(Sym2(H 0(ωF ))G) − dim(H 0(ω⊗2 F )G). F )∨)G. Hence, we F )G) = 3. Using c) we have that also dim(Sym2(H 0(ωF ))G) = 3 so, as As claimed in [Pir92], we can identify T[π]HG with H 1(TF )G = (H 0(ω⊗2 have dim(H 0(ω⊗2 claimed, H 0(IF (2))1 = 0. As consequence we have either GQ ∈ Sym2(H 0(ωF ))ρ or GQ ∈ Sym2(H 0(ωF ))ρ2. Note that in both cases, as π∗H 0(ωE)2 ⊂ Sym2(H 0(ωF ))1, we have GQ(q−) = 0 so q− ∈ Q. (cid:3) Lemma 4.2. Let π : F → E with [π] ∈ HG and assume that F is not hyperelliptic. Denote by a1, a2 and a3 the 3 ramification points of π. Let L be a g1 3. Then: • L is G-invariant, i.e. g∗L = L; • h0(OF (3ai)) = 1; • If L′ is a g1 3 then L ≃ L′, i.e. there is only one g1 3 on F . Proof. Recall that every curve of genus 4 is trigonal and moreover, the number of g1 most 2. If there is only one g1 and acts on a set of two elements, it has to fix both of them. 3 is at 3 clearly it is G-invariant. If there are 2, as G has order 3 Now let's prove that h0(OF (3ai)) = 1. The Riemann-Roch formula for OF (3ai) is h0(OF (3ai)) − h1(OF (3ai)) = deg(h0(OF (3ai))) − 4 + 1 = 0 so, by Serre duality, we have From h0(OF (3ai)) = h0(ωF (−3ai)). 0 → ωF (−3ai) → ωF (−2ai) → ωF (−2ai)ai → 0 17 / / / / one has H 0(ωF (−3ai)) ≤ H 0(ωF (−2ai)). In particular, as F is not hyperelliptic we obtain that the dimension of H 0(ωF (−3ai)) is either 1 or 2. Moreover, h0(ωF (−3ai)) = 2 if and only if H 0(ωF (−3ai)) = H 0(ωF (−2ai)). But this cannot happen as the pullback η of a non- zero holomorphic form on E has a zero of multiplicity 2 exactly in the ramification points so there is at least one element in H 0(ωF (−2ai)) \ H 0(ωF (−3ai)). Hence h0(OF (3ai)) = 1 as claimed. Recall that on F there are at most two g1 3 and they are related by L ⊗ L′ = ωF = OF (2a1 + 2a2 + 2a3). Hence we will conclude by proving that L = OF (a1 + a2 + a3). Let A, B in F such that L ≃ OF (a1 + A + B). As a1 is invariant and the same holds for L, we have that OF (g(A) + g(B)) = OF (A + B). Therefore, as F is not hyperelliptic, also the equality of divisors g(A) + g(B) = A + B has to hold. Moreover, as g has order 3, it cannot exchange A and B: we have proved that A and B are ramification points. If we assume that L 6= OF (a1 + a2 + a3) there are several possibilities: A = a1 = B: This is impossible as we would have 2 = h0(L) = h0(OF (3a1)) = 1. A = a1 6= B: Assume that A = a1 and B = a2 so that L ≃ OF (2a1 + a2). Let C, D ∈ F such that OF (2a1 + a2) ≃ OF (a3 + C + D). As before, we have that C and D are ramification points and as F is not hyperelliptic the only possible option is to have C = D = a3. But then, again, we have a contradiction 2 = h0(OF (L)) = h0(OF (2a1 + a2)) = h0(OF (3a3)) = 1. A 6= a1 = B: This case is analogous to the previous one. A = B 6= a1: This case is analogous to the second one. Hence, we have proved that L = OF (a1 + a2 + a3) and thus that L ≃ L′ and there is only a g1 (cid:3) 3 on F . Proposition 4.3. Let π : F → E with [π] ∈ HG and assume that F is not hyperelliptic. Denote by Q be the only quadric containing the canonical curve F . Then Q is a quadric cone with vertex V and V 6∈ F . Moreover, the hyperplane H − is tangent to the cone and the 3 ramification points of π lie on a line on the cone. Proof. Recall that if the quadric Q containing F is smooth, then F can be seen as a curve of bidegree (3, 3) in P1 × P1 and the projections on each factor give two different g1 3. If, instead, Q is a cone (these are the only possible cases as F is non degenerate) there exists only one g1 3. Hence, by Lemma 4.2, we can conclude that Q is a cone. If V is the vertex, it is clear that V 6∈ F as, otherwise F would be singular. Now we will prove that the ramification points are on a line in the canonical model of F . By what we have seen in this section we have a decomposition of H 0(ωF ) into subrepresentations with H 0(ωF )1 = π∗H 0(ωE). We can assume, as before, that H 0(ωF )ρ has dimension 2. Denote respectively with {u0}, {u1, u2} and {u3} a basis for π∗H 0(ωE), H 0(ωF )ρ and H 0(ωF )ρ2. By abuse of notation we will write uiuj to mean ui ⊗ uj. With these coordinates, the hyperplane H − has equation u0 = 0 and q− = (1 : 0 : 0 : 0). As (34) Sym2(H 0(ωF )) = hu2 0, u1u3, u2u3i1 ⊕ hu0u1, u0u2, u2 3iρ ⊕ hu0u3, u2 1, u2 2, u1u2iρ2 18 We know by Proposition 4.1 that an equation GQ of Q is either an element of Sym2(H 0(ωF ))ρ or of Sym2(H 0(ωF ))ρ2. In the first case the generic element of Sym2(H 0(ωF ))ρ is a quadric cone and has equation u2 3 + u0(au1 + bu2) = 0. Moreover, it is easy to see that H − is tangent to the cone along the line L1 = {u3 = u0 = 0}. In the second case the generic element of Sym2(H 0(ωF ))ρ2 is a smooth quadric but it is easy to see that the generic singular element is a cone with equation u3u0 + (au1 + bu2)2 = 0. As before, H − is a plane tangent to Q along the line L2 = {u0 = au1 + bu2 = 0}. So, in both cases, as the ramification points of the canonical curve F are given exactly as H − ∩ F , they are on a line as claimed. (cid:3) Now we are going to calculate the sum in Theorem 3.3 and to see that it is zero for each element in Ker(d oP∨ E). Proposition 4.4. Let π : F → E be a Galois covering of degree 3 as before and consider a local family of coverings with central fiber π. Then, if ν(β) = one has ν(β) = 0 for all β ∈ Ker(doP∨ E). m(β) π∗α2 (xk), 3 Xk=1 Proof. Ker(d oP∨ it as E) is a subspace of Sym2(H 0(ωF )−). If β ∈ Ker(d oP∨ E) we can decompose β = β1 + βρ + βρ2 with βµ ∈ Sym2(H 0(ωF )−)µ. First of all we will prove that ν(βρ) = ν(βρ2) = 0. As c is not a branch point, we have that the fiber π−1(c) = {x1, x2, x3} over c is equal to the orbit of each of its points: π−1(c) = {x1, g(x1), g2(x1)}. Hence π∗α2 If we assume that β is in the eigenspace Sym2(H 0(ωF ))µ of g∗ then 2 Xk=0 3 m(β) π∗α2 (xk) = ν(β) = Xk=1 m(cid:16)Pd k=1(g∗)k(β)(cid:17) π∗α2 ν(β) = m(β) π∗α2 (gk(x1)) = m(β ◦ gk) (x1). 2 Xk=0 (x1) = m(cid:0)P2 k=0 µkβ(cid:1) π∗α2 (x1) = 2 Xk=0 µk! m(β) π∗α2 (x1). Hence, if µ 6= 1, we have λ(β) = 0 as claimed. Hence we have that ν(β) = ν(β1) so it is enough to prove that ker(d oP∨ E) ⊆ Sym2(H 0(ωF )−)ρ ⊕ Sym2(H 0(ωF )−)ρ2 i.e., that β1 = 0. Let a be a ramification point and consider holomorphic coordinates (U, z) centered in a and (V, w) centered in π(a) = b. Assume, moreover, that αV = dw, the relation w = z3 holds and the action of g ∈ G near a is given by z 7→ ρz for ρ 6= 1 such that ρ3 = 1. By changing ρ with ρ2 we can assume, moreover, that the decomposition of H 0(ωF ) in 19 invariant subspaces with respect to the action of G is the one given in Equation (33). Consider η ∈ H 0(ωF ). Near a we can write In particular, η ∈ H 0(ωF )G if and only if, near a we have ηU = Xj≥0 ηjzj! dz and ηjρjzj! dz. g∗ηU = ρ Xj≥0 η2+3jz2+3j! dz ηU = Xj≥0 and an analogous decomposition holds near the other ramification points. Similarly, we have ηU = Xj≥0 η3jz3j! dz and ηU = Xj≥0 η1+3jz1+3j! dz if η ∈ H 0(ωF )ρ and η ∈ H 0(ωF )ρ2 respectively. As and Sym2(H 0(ωF )−)1 = H 0(ωF )ρ ⊗ H 0(ωF )ρ2, Sym2(H 0(ωF )−)ρ = H 0(ωF )⊗2 ρ2 , Sym2(H 0(ωF )−)ρ2 = Sym2(H 0(ωF )ρ), if ϕ ∈ Sym2(H 0(ωF )−)µ we can write it in coordinate near a as for µ = 1 and as ϕU = z(cid:0)ϕ0 + ϕ1z3 + o(z5)(cid:1) dz2 ϕU =(cid:0)ϕ0 + ϕ1z3 + o(z5)(cid:1) dz2 and ϕU = z2(cid:0)ϕ0 + ϕ1z3 + o(z5)(cid:1) dz2 if µ = ρ and µ = ρ2, respectively. In the latter cases, we have that the residue of ϕ/π∗α in a is 0 as ϕ/π∗α is either holomorphic or has a pole of order 2 with coefficient of degree −1 equal to 0. Hence Sym2(H 0(ωF )−)ρ ⊕ Sym2(H 0(ωF )−)ρ2 ⊆ Ker(d oP∨ By Theorem 3.2 and using Diagram (32) we obtain dim(Ker(d oP∨ to the dimension of Sym2(H 0(ωF )−)ρ ⊕ Sym2(H 0(ωF )−)ρ2 so E). E)) = 4. This is equal Sym2(H 0(ωF )−)ρ ⊕ Sym2(H 0(ωF )−)ρ2 = Ker(d oP∨ E). In particular, β1 = 0 and ν(β) = 0 as claimed. (cid:3) References [ACG11] Arbarello E., Cornalba M., Griffiths P. A. Geometry of algebraic curves. Vol. II. With a contribution by Joseph Daniel Harris. Grundlehren der Mathematischen Wissenschaften [Fun- damental Principles of Mathematical Sciences], 268. Springer, Heidelberg, 2011. [ACGH85] Arbarello E., Cornalba M., Griffiths P. A., Harris J. Geometry of algebraic curves. Vol. I. Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 267. Springer-Verlag, New York, 1985 Albano A., Pirola G. P. Dihedral monodromy and Xiao fibrations Ann. Mat. Pura Appl. (4) 195 (2016), no. 4, 1255–1268 [AP16] 20 [BGN15] Barja M. A., Gonzalez-Alonso V., Naranjo, J. C. Xiao's conjecture for general fibred surfaces To appear in Journal fur die reine und angewandte Mathematik. [Cle72] [Deb82] [BHPV04] Barth, W.P., Hulek, K., Peters, C.A.M., Van de Ven, A. Compact complex surfaces, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathemat- ics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 4, second edn. Springer-Verlag, Berlin (2004). A. Clebsch Zur Theorie der Riemann'schen Flachen, Math. Ann. 6 (1872), 216–230 O. Debarre, In A c(cid:13)galit A c(cid:13)s num A c(cid:13)riques pour les surfaces de type g A c(cid:13)n A c(cid:13)ral (French. English summary) [Numerical inequalities for surfaces of general type] With an appendix by A. Beauville. Bull. Soc. Math. France 110 (1982), no. 3, 319–346. Horikawa E. On deformations of holomorphic maps I. J. Math. Soc. Japan 25 (1973), 372–396. Horikawa E. On deformations of holomorphic maps II. J. Math. Soc. Japan 26 (1974), 647– 667. A. Hurwitz, Uber Riemann'sche Flachen mit gegebenen Verzweigungspunkten, Math. Ann. 39 (1891),1–61. Iskovskih V. A. Fano threefolds I. Math. USSR-Izv. 11 (1977), no. 3, 485–527 (1978) Kanev V., Hurwitz spaces of triple coverings of elliptic curves and moduli spaces of abelian threefolds. Ann. Mat. Pura Appl. (4) 183 (2004), no. 3, 333–374. Miranda R. Algebraic curves and Riemann surfaces Graduate Studies in Mathematics, Vol 5, American Mathematical Society, Providence, RI, 1995 Pirola G. P. On a conjecture of Xiao J. Reine Angew. Math. 431 (1992), 75–89. Xiao G. Fibered algebraic surfaces with low slope Math. Ann. 276 (1987), 449–466. [Isk77] [Kan04] [Mir95] [Pir92] [Xia87] [Hor73] [Hor74] [Hur91] (Filippo F. Favale) Department of Mathematics, University of Trento, via Sommarive 14, I-38123 Trento, Italy E-mail address: [email protected] (Sara Torelli) Department of Mathematics, University of Pavia, Via Ferrata, 5, I-27100 Pavia, Italy E-mail address: [email protected] 21
1208.4626	2	1208	2012-09-16T09:18:40	Families of Lagrangian fibrations on hyperkaehler manifolds	[ "math.AG", "math.DG" ]	A holomorphic Lagrangian fibration on a holomorphically symplectic manifold is a holomorphic map with Lagrangian fibers. It is known that a given compact manifold admits only finitely many holomorphic symplectic structures, up to deformation. We prove that a given compact manifold with $b_2 \geq 7$ admits only finitely many deformation types of holomorphic Lagrangian fibrations. We also prove that all known hyperkahler manifolds are never Kobayashi hyperbolic.	math.AG	math	L. Kamenova, M. Verbitsky Families of Lagrangian fibrations Families of Lagrangian fibrations on hyperkahler manifolds Ljudmila Kamenova, Misha Verbitsky1 Abstract A holomorphic Lagrangian fibration on a holomorphically symplec- tic manifold is a holomorphic map with Lagrangian fibers. It is known (due to Huybrechts) that a given compact manifold admits only finitely many holomorphic symplectic structures, up to defor- mation. We prove that a given compact manifold with b2 > 7 admits only finitely many deformation types of holomorphic Lagrangian fi- brations. We also prove that all known hyperkahler manifolds are never Kobayashi hyperbolic. Contents 1 Introduction 1.1 Hyperkahler manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The Bogomolov-Beauville-Fujiki form . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 The hyperkahler SYZ conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Hyperkahler geometry: preliminary results 2.1 Teichmuller space and the moduli space . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The polarized Teichmuller space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Main results 3.1 The moduli of manifolds with Lagrangian fibrations . . . . . . . . . . . . . . . . . . 3.2 Kobayashi hyperbolicity in hyperkahler geometry . . . . . . . . . . . . . . . . . . . . 1 2 3 3 5 5 6 8 8 11 1 Introduction Irreducible compact hyperkahler manifolds, or irreducible holomorphic symplec- tic manifolds, are a natural generalization of K3 surfaces in higher dimensions. The geometry of K3 surfaces is well studied. In particular, it is known that any two K3 surfaces are deformation equivalent to each other, i.e., there is only one deformation type of K3 surfaces. A natural question to ask is whether the same is true in higher dimensions. The answer is negative due to Beauville's examples. In every possible complex dimension 2n there are at least the Hilbert scheme of n points on a K3 surface S, Hilbn(S), and the generalized Kummer varieties K n+1(A), where A is an Abelian surface. These two examples are not deformation equivalent since they have different Betti numbers. There are two more exceptional examples due to K. O'Grady in dimensions 6 and 10. It is conjectured that in every fixed dimension there are finitely many de- formation types of irreducible compact hyperkahler manifolds. It is also con- jectured that every hyperkahler manifold can be deformed to one that admits 1Partially supported by by RFBR grants 12-01-00944-, 10-01-93113-NCNIL-a, and AG Laboratory NRI-HSE, RF government grant, ag. 11.G34.31.0023. -- 1 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations a holomorphic Lagrangian fibration. It would be interesting to classify the de- formation types of tha pairs (M, L) of a hyperkahler manifold together with a Lagrangian fibration on it. In the present paper, we show that the number of deformational classes of such pairs is finite, if one fixes the smooth manifold undelying M . In [Hu2] Huybrechts proved that for a fixed compact manifold there are at most finitely many deformation types of hyperkahler structures on it. Therefore, to prove that the number of deformation classes of pairs (M, L) is finite, it would suffice to prove it when a deformational class of M is fixed. Let M π−→ X be a Lagrangian fibration. Then X is known to be projective, with H 2(X) = C, hence rk Pic(X) = 1. Therefore, the primitive ample bundle LX on X is unique (up to torsion). Denote by LM the semiample bundle π∗(LX ) on M . Clearly, c1(LM )rk M = 0; a (1,1)-class satisfying this equation is called parabolic. The Lagrangian fibration M π−→ X is uniquely determined by a class [c1(LM )] ∈ Pic(M ) which is parabolic and semiample (this is due to D. Matsushita, [Ma1]; see [Saw] for a detailed exposition of an early work on Lagrangian fibrations). Therefore, to classify the Lagrangian fibrations it would suffice to classify pairs (M, LM ), where LM is a parabolic semiample line bundle. We prove that in the Teichmuller space of hyperkahler manifolds with a fixed parabolic class the pairs admitting a Lagrangian fibration form a dense and open subset. The other main result is that the action of the monodromy group has finitely many orbits. As a corollary of these results we obtain that for a fixed compact manifold, there are only finitely many deformation types of hyperkahler structures equipped with a Lagrangian fibration. 1.1 Hyperkahler manifolds Definition 1.1: A hyperkahler manifold is a compact, Kahler, holomorphi- cally symplectic manifold. Definition 1.2: A hyperkahler manifold M is called simple if H 1(M ) = 0, H 2,0(M ) = C. Theorem 1.3: (Bogomolov's Decomposition Theorem, [Bo1], [Bes]). Any hy- perkahler manifold admits a finite covering, which is a product of a torus and several simple hyperkahler manifolds. Remark 1.4: Further on, all hyperkahler manifolds are assumed to be simple. A note on terminology. Speaking of hyperkahler manifolds, people usu- ally mean one of two different notions. One either speaks of holomorphically symplectic Kahler manifold, or of a manifold with a hyperkahler structure, that is, a triple of complex structures satisfying quaternionic relations and parallel with respect to the Levi-Civita connection. The equivalence (in compact case) -- 2 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations between these two notions is provided by the Yau's solution of Calabi's conjec- ture ([Bes]). Throughout this paper, we use the complex algebraic geometry point of view, where "hyperkahler" is synonymous with "Kahler holomorphi- cally symplectic", in lieu of the differential-geometric approach. The reader may check [Bes] for an introduction to hyperkahler geometry from the differential- geometric point of view. Notice also that we included compactness in our definition of a hyperkahler manifold. In the differential-geometric setting, one does not usually assume that the manifold is compact. 1.2 The Bogomolov-Beauville-Fujiki form Theorem 1.5: ([F]) Let η ∈ H 2(M ), and dim M = 2n, where M is hyperkahler. Then RM η2n = cq(η, η)n, for some integer quadratic form q on H 2(M ) and a constant c > 0. Definition 1.6: This form is called Bogomolov-Beauville-Fujiki form. It is defined by this relation uniquely, up to a sign. The sign is determined from the following formula (Bogomolov, Beauville; [Bea], [Hu1], 23.5) λq(η, η) = (n/2)ZX η ∧ η ∧ Ωn−1 ∧ Ω n−1 + + (1 − n)(cid:16)RX η ∧ Ωn−1 ∧ Ω n−1(cid:17) n(cid:17)(cid:16)RX η ∧ Ωn ∧ Ω n RM Ωn ∧ Ω where Ω is the holomorphic symplectic form, and λ a positive constant. Remark 1.7: The form q has signature (3, b2 − 3). It is negative definite on primitive forms, and positive definite on the space hΩ, Ω, ωi where ω is a Kahler form, as seen from the following formula µq(η1, η2) = ZX ω2n−2 ∧ η1 ∧ η2 − 2n − 2 (2n − 1)2 RX ω2n−1 ∧ η1 ·RX ω2n−1 ∧ η2 RM ω2n , µ > 0 (1.1) (see e. g. [V2], Theorem 6.1, or [Hu1], Corollary 23.9). Definition 1.8: Let [η] ∈ H 1,1(M ) be a real (1,1)-class on a hyperkahler mani- fold M . We say that [η] is parabolic if q([η], [η]) = 0. A line bundle L is called parabolic if c1(L) is parabolic. 1.3 The hyperkahler SYZ conjecture Theorem 1.9: (D. Matsushita, see [Ma1]). Let π : M −→ X be a surjective holomorphic map from a hyperkahler manifold M to X, with 0 < dim X < -- 3 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations dim M . Then dim X = 1/2 dim M , and the fibers of π are holomorphic La- grangian tori (this means that the symplectic form vanishes on the fibers).1 Definition 1.10: Such a map is called a holomorphic Lagrangian fibration. Remark 1.11: The base of π is conjectured to be rational. J.-M. Hwang ([Hw]) proved that X ∼= CP n, if it is smooth. D. Matsushita ([Ma2]) proved that it has the same rational cohomology as CP n. Remark 1.12: The base of π has a natural flat connection on the smooth locus of π. The combinatorics of this connection can be used to determine the topology of M ([KZ], [G]), Definition 1.13: Let (M, ω) be a Calabi-Yau manifold, Ω the holomorphic volume form, and Z ⊂ M a real analytic subvariety, Lagrangian with respect to ω. If Ω(cid:12)(cid:12)Z is proportional to the Riemannian volume form, Z is called special Lagrangian (SpLag). The special Lagrangian varieties were defined in [HL] by Harvey and Law- son, who proved that they minimize the Riemannian volume in their cohomology class. This implies, in particular, that their moduli are finite-dimensional. In [McL], McLean studied deformations of non-singular special Lagrangian subva- rieties and showed that they are unobstructed. In [SYZ], Strominger-Yau-Zaslow tried to explain the mirror symmetry phe- nomenon using the special Lagrangian fibrations. They conjectured that any Calabi-Yau manifold admits a Lagrangian fibration with special Lagrangian fibers. Taking its dual fibration, one obtains "the mirror dual" Calabi-Yau manifold. Definition 1.14: A line bundle is called semiample if LN is generated by its holomorphic sections, which have no common zeros. Remark 1.15: From semiampleness it obviously follows that L is nef. Indeed, let π : M −→ PH 0(LN )∗ be the standard map. Since the sections of L have no common zeros, π is holomorphic. Then L ∼= π∗O(1), and the curvature of L is a pullback of the Kahler form on CP n. However, the converse is false: a nef bundle is not necessarily semiample (see e.g. [DPS1, Example 1.7]). Remark 1.16: Let π : M −→ X be a holomorphic Lagrangian fibration, and ωX a Kahler class on X. Then η := π∗ωX is semiample and parabolic. The converse is also true, by Matsushita's theorem: if L is semiample and parabolic, L induces a Lagrangian fibration. This is the only known source of non-trivial special Lagrangian fibrations. 1Here, as elsewhere, we assume that the hyperkahler manifold M is simple. -- 4 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations Conjecture 1.17: (Hyperkahler SYZ conjecture) Let L be a parabolic nef line bundle on a hyperkahler manifold. Then L is semiample. Remark 1.18: This conjecture was stated by many people (Tyurin, Bogomolov, Hassett-Tschinkel, Huybrechts, Sawon); please see [Saw] for an interesting and historically important discussion, and [V3] for details and references. Remark 1.19: The SYZ conjecture can be seen as a hyperkahler version of the "abundance conjecture" (see e.g. [DPS2], 2.7.2). Claim 1.20: Let M be an irreducible hyperkahler manifold in one of 4 known classes, that is, a deformation of a Hilbert scheme of points on K3, a deforma- tion of generalized Kummer variety, or a deformation of one of two examples by O'Grady. Then M admits a deformation equipped with a holomorphic La- grangian fibration. Proof: When S is an elliptic K3 surface, the Hilbert scheme of points Hilbn(S) has an induced Lagrangian fibration with smooth fibers that are prod- ucts of n elliptic curves: Hilbn(S) → Symn(P1) ≃ Pn. Similarly, when A is an elliptic Abelian surface, the generalized Kummer variety K n(A) admits a Lagrangian fibration. Another construction gives Lagrangian fibrations on Hilbn(S) and on K n(A) if S contains a smooth genus n curve and if A contains a smooth genus n + 2 curve (see examples 3.6 and 3.8 in [Saw]). O'Grady's examples are deformation equivalent to Lagrangian fibrations, as follows from Corollary 1.1.10 in [R]. 2 Hyperkahler geometry: preliminary results 2.1 Teichmuller space and the moduli space Here we cite the relevant result from the deformation theory of hyperkahler manifolds. We follow [V1]. Let M be a hyperkahler manifold (compact and simple, as usual), and Comp0 be the Fr`echet manifold of all complex structures of hyperkahler type on M . The quotient Teich := Comp0 / Diff 0 of Comp0 by isotopies is a finite-dimensional complex analytic space ([Cat]). This quotient is called the Teichmuller space of M . When M is a complex curve, the quotient Comp0 / Diff 0 is the Te- ichmuller space of this curve. The mapping class group Γ = Diff + / Diff 0 acts on Teich in the usual way, and its quotient Mod is the moduli space of M . As shown in [Hu2], Teich has a finite number of connected components. Take a connected component TeichI containing a given complex structure I, and let ΓI ⊂ Γ be the set of elements of Γ fixing this component. Since Teich has only a finite number of connected components, ΓI has finite index in Γ. On the -- 5 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations other hand, as shown in [V1], the image of the group Γ is commensurable to O(cid:0)H 2(M, Z), q(cid:1). In [V1, Lemma 2.6] it was proved that any hyperkahler structure on a given simple hyperkahler manifold is also simple. Therefore, H 2,0(M, I ′) = C for all I ′ ∈ Comp. This trivial observation is a key to the following well-known definition. Definition 2.1: Let (M, I) be a hyperkahler manifold, and Teich its Teichmuller space. Consider a map Per : Teich −→ PH 2(M, C), sending J to the line H 2,0(M, J) ∈ PH 2(M, C). It is easy to see that Per maps Teich into the open subset of a quadric, defined by Per := (cid:8)l ∈ PH 2(M, C) (cid:12)(cid:12) q(l, l) = 0, q(l, l) > 0(cid:9). The map Per : Teich −→ Per is called the period map, and the set Per the period space. The following fundamental theorem is due to F. Bogomolov [Bo2]. Theorem 2.2: Let M be a simple hyperkahler manifold, and Teich its Te- ichmuller space. Then the period map Per : Teich −→ Per is a local diffeomor- phism (that is, an etale map). Moreover, it is holomorphic. Remark 2.3: Bogomolov's theorem implies that Teich is smooth. However, it is not necessarily Hausdorff (and it is non-Hausdorff even in the simplest examples). 2.2 The polarized Teichmuller space In [V4, Corollary 2.6], the following proposition was deduced from [Bou] and [DP]. Theorem 2.4: Let M be a simple hyperkahler manifold, such that all integer (1, 1)-classes satisfy q(ν, ν) > 0. Then its Kahler cone is one of the two connected components of the set K := (cid:8)ν ∈ H 1,1(M, R) (cid:12)(cid:12) q(ν, ν) > 0(cid:9). Remark 2.5: From Theorem 2.4 it follows that on a hyperkahler manifold with Pic(M ) = Z, for any rational class η ∈ H 1,1(M ) with q(η, η) > 0, either η or −η is nef. Remark 2.6: Consider an integer vector η ∈ H 2(M ) which is positive, that is, satisfies q(η, η) > 0. Denote by Teichη the set of all I ∈ Teich such that η is of type (1, 1) on (M, I). The space Teichη is a closed divisor in Teich. Indeed, by Bogomolov's theorem, the period map Per : Teich −→ Per is etale, but the image of Teichη is the set of all l ∈ Per which are orthogonal to η; this condition defines a closed divisor Cη in Per, hence Teichη = Per−1(Cη) is also a closed divisor. -- 6 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations Remark 2.7: When I ∈ Teichη is generic, Bogomolov's theorem implies that the space of rational (1, 1)-classes H 1,1(M, Q) is one-dimensional and generated by η. This is seen from the following argument. Locally around a given point I the period map Teichη −→ Per is surjective on the set Perη of all I ∈ Per for which η ∈ H 1,1(M, I). However, the Hodge-Riemann relations give Perη = (cid:8)l ∈ Per (cid:12)(cid:12) q(η, l) = 0(cid:9). Denote the set of such points of Teichη by Teichη 2.4 that, for any I ∈ Teichη gen, either η or −η is a Kahler class on (M, I). (2.1) gen. It follows from Theorem Consider a connected component Teichη,I of Teichη. Changing the sign of η if necessary, we may assume that η is Kahler on (M, I). By Kodaira's theorem about stability of Kahler classes, η is Kahler in some neighbourhood U ⊂ Teichη,I of I. Therefore, the sets and V+ := (cid:8)I ∈ Teichη V− := (cid:8)I ∈ Teichη gen. It is easy to see that Teichη gen (cid:12)(cid:12) η is Kahler on (M, I)(cid:9) gen (cid:12)(cid:12) − η is Kahler on (M, I)(cid:9) are open in Teichη gen is a complement to a union of countably many divisors in Teichη corresponding to the points I ′ ∈ Teichη with rk Pic(M, I ′) > 1. Therefore, for any connected open subset U ⊂ Teichη, the intersection U ∩ Teichη gen is represented as a disjoint union of open sets V+ ⊔ V−, every connected component of Teichη gen and of Teichη is contained in V+ or in V−. We obtained the following corollary. gen is connected. Since Teichη Corollary 2.8: Let η ∈ H 2(M ) be a positive integer vector, Teichη the corre- sponding divisor in the Teichmuller space, and Teichη,I a connected component of Teichη containing a complex structure I. Assume that η is Kahler on (M, I). Then η is Kahler for all I ′ ∈ Teichη,I which satisfy rk H 1,1(M, Q) = 1. We call the set Teichη pol of all I ∈ Teichη for which η is Kahler the polarized Teichmuller space, and η its polarization. From the above arguments it is clear that the polarized Teichmuller space Teichη pol is open and dense in Teichη. pol by the subgroup of the mapping class group It is fixing η is called the moduli of polarized hyperkahler manifolds. known (due to the general theory which goes back to Viehweg and Grothen- dieck) that Mη is Hausdorff and quasiprojective (see e.g. [Vi] and [GHS]). The quotient Mη of Teichη Remark 2.9: We conclude that there are countably many quasiprojective di- visors Mη immersed in the moduli space Mod of hyperkahler manifolds. More- over, every algebraic complex structure belongs to one of these divisors. How- ever, these divisors need not to be closed. Indeed, as proven in [AV], each of Mη is dense in Mod. -- 7 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations In [AV, Theorem 1.7], the following theorem was proven. Theorem 2.10: Let M be a compact, simple hyperkahler manifold, TeichI a connected component of its Teichmuller space, and TeichI Ψ−→ TeichI /ΓI = Mod its projection to the moduli space of complex structures. Consider a posi- tive or negative vector η ∈ H 2(M, Z), and let TeichI,η be the corresponding con- nected component of the polarized Teichmuller space. Assume that b2(M ) > 3. Then the image Ψ(TeichI,η) is dense in Mod. The proof relies on a more general proposition about lattices. Proposition 2.11: [AV, Proposition 3.2, Remark 3.12] Let V be an R-vector space equipped with a non-degenerate symmetric form of signature (s+, s−) with s+ ≥ 3 and s− ≥ 1. Consider a lattice L ⊂ V . Let Γ be a subgroup of finite index in O(L), and l ∈ L. Then Γ · Gr++(l⊥) is dense in Gr++(V ). Remark 2.12: Since the proof of this statement is symmetric in s+ and s−, the same proposition is valid if we assume that s+ ≥ 1 and s− ≥ 3. 3 Main results 3.1 The moduli of manifolds with Lagrangian fibrations Here we assume that b2(M ) > 7 as we need it for our proof of Theorem 3.4. The authors conjecture that the result is valid for smaller Betti numbers as well. Definition 3.1: Let L be a holomorphic line bundle on a hyperkahler manifold. We call L Lagrangian if it is parabolic and semiample. Definition 3.2: Let M be a hyperkahler manifold. Fix a parabolic class L ∈ H 2(M, Z). We denote by TeichL the Teichmuller space of all complex structures I of hyperkahler type on M such that L is of type (1, 1) on (M, I). Clearly, TeichL is a divisor in the whole Teichmuller space of M . The space TeichL is called the Teichmuller space of hyperkahler manifolds with parabolic class. Matsushita proves the following openness result in [Ma3, Theorem 1.1]: Theorem 3.3: Let Teich◦ Lagrangian. Then Teich◦ L is open in TeichL. L ⊂ TeichL be the set of all I ∈ TeichL for which L is The main results of the present paper are the following two theorems. Theorem 3.4: The subspace Teich◦ L ⊂ TeichL is dense and open in TeichL. -- 8 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations Proof: Fix a positive class η ∈ H 2(M, Z) and define Teich◦ L,η to be the open subset of Teich◦ L for which η is a polarization. Consider the projection Ψ to the moduli space Mod as defined in Theorem 2.10. Since Mη is quasiprojective (see [Vi]), then Ψ(Teich◦ L,η) is Zariski open, and therefore dense in Ψ(TeichL,η). Fix a negative vector L′ ∈ H 2(M, Z) such that the sublattice < L, L′ > is of rank 2. Notice that Ψ(TeichL) = {l ∈ PH 2(M, Z)q(l, l) = 0, q(l, l) > 0, q(L, l) = 0}/ΓL and Ψ(TeichL,η) = {l ∈ PH 2(M, Z)q(l, l) = 0, q(l, l) > 0, q(L, l) = 0, q(η, l) = 0}/ΓL,η. Applying Proposition 2.11 to the quotient H 2(M, Z)/ < L, L′ >, we see that Ψ(TeichL,L′,η) is dense in Ψ(TeichL,L′) for any L′. Here we needed to assume b2 > 7, because H 2(M, Z) is of signature (3, b2 − 3) and the quotient H 2(M, Z)/ < L, L′ > is of signature (2, b2 − 4). This satisfies the conditions of Proposition 2.11 since b2 − 4 > 3. However, SL′ Ψ(TeichL,L′) is dense in Ψ(TeichL), and SL′ Ψ(TeichL,L′,η) is dense in Ψ(TeichL,η). Therefore, Ψ(TeichL,η) is dense in Ψ(TeichL) and Teich◦ L is dense in TeichL. Remark 3.5: Together with Proposition 2.11, Theorem 3.4 implies that the set of manifolds with Lagrangian fibrations is dense within the deformation space of a hyperkahler manifold M , if M admits a Lagrangian fibration. Theorem 3.6: Consider the action of the monodromy group ΓI on H 2(M, Z), and let S ⊂ H 2(M, Z) be the set of all classes which are parabolic and primitive. Then there are only finitely many orbits of ΓI on S. Proof: In the proof we use Nikulin's technique of discriminant-forms described in [Ni]. Denote by Λ the lattice (H 2(M, Z), q). It is a free Z-module of finite rank together with a non-degenerate symmetric bilinear form q with values in Z. If {ei}i∈I is a basis of the lattice Λ, its discriminant is defined to be discr(Λ) = det(ei · ej). There is a canonical embedding Λ ֒→ Λ∗ = Hom(Λ, Z) using the bilinear form of Λ. The discriminant group AΛ = Λ∗/Λ is a finite Abelian group of order discr(Λ). One can extend the bilinear form to Λ∗ with values in Q and define the discriminant-bilinear form of the lattice bΛ : AΛ × AΛ → Q/Z. It is a finite non-degenerate form. A subgroup H ⊂ AΛ is isotropic if qΛH = 0, where qΛ is the quadratic form corresponding to bΛ. Given any subset K ⊂ Λ, its orthogonal complement is K ⊥ = {v ∈ Λ(v, K) = 0}. An embedding of lattices Λ1 ֒→ Λ2 is primitive if Λ2/Λ1 is a free Z-module. Take a primitive vector v ∈ Λ with q(v) = 0. We can choose a vector f ∈ Λ with minimal positive quadratic intersection α = q(v, f ). Then 0 < α 6 discr(Λ). It is implied by the following lemma: Lemma 3.7: The minimal positive intersection α divides discr(Λ). Proof: Since v is primitive, we can choose a free Z-basis {v1 = v, v2, . . . , vn} If α = min{q(v, f )f ∈ Zn}, then αZ is an ideal of Λ, where n = rk(Λ). generated by {q(v, vi), i = 1, . . . , n}. For every i = 1, . . . , n, q(v, vi) = α · ai for -- 9 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations some ai ∈ Z. Thus the matrix [q(vj , vi)] has first column divisible by α. Then det[q(vj , vi)] = discr(Λ) is divisible by α. Let K be the primitive sublattice of Λ spanned by v and f . The intersection matrix of Span(v, f ) has determinant q(v, v)q(f, f ) − q(v, f )2 = −α2 which is bounded: −discr(Λ)2 6 −α2 < 0. Since rk(K) = 2, K has at most four primitive isotropic vectors (2 rk(K) = 4). An overlattice of Λ is a lattice embedding i : Λ → Λ′ with Λ and Λ′ of the same rank, or equivalently, such that HΛ′ = Λ′/Λ is a finite Abelian group. Note that we have the inclusions: Λ ֒→ Λ′ ֒→ Λ′∗ ֒→ Λ∗. Therefore, HΛ′ ⊂ Λ′∗/Λ ⊂ Λ∗/Λ = AΛ. [Ni, Proposition 1.4.1] The correspondence Λ′ → HΛ′ de- Proposition 3.8: termines a bijection between overlattices of Λ and isotropic subgroups of AΛ. Furthermore, H ⊥ Λ′ = Λ′∗/Λ and H ⊥ Λ′ /HΛ′ = AΛ′ . Let L = K ⊥ be the orthogonal complement of K in Λ. Then K ⊕ L ⊂ Λ ⊂ K ∗ ⊕ L∗. Since det(L) is bounded, in view of Proposition 3.8, there are finitely . = K ⊕ K ⊥ because AΛ is many ways of expressing Λ as an overlattice of ΛK finite of order discr(Λ) and there are finitely many isotropic subgroups. Define the lattices Λ and Λ′ to be stably equivalent if there exists a lattice M such that Λ ⊕ M ≃ Λ′ ⊕ M . The following proposition is a reformulation of Theorem 1.1 in Chapter 9 of Cassels' book [Cas]. Proposition 3.9: There exist only a finite number of lattices stably equivalent to Λ. If we assume that there are infinitely many orbits of ΓI , this would imply that there exist infinitely many non-isomorphic pairs of lattices (K, K ⊥). Then for infinitely many of them K ⊥ would be stably equivalent to K ⊥ 1 for another K1 since there are only finitely many choices for K. This contradicts Proposition 3.9 and the result follows. Corollary 3.10: For any hyperkahler manifold, there are only finitely many orbits of ΓI on the set of all divisors TeichL with a parabolic class. Combining Corollary 3.10 and Theorem 3.4, we obtain the following result. Corollary 3.11: Let M be a hyperkahler manifold. Then there are only finitely many deformation types of Lagrangian fibrations (M, I) −→ S, for all complex structures on M . Proof: By Remark 2.7 we can assume that H 1,1(M, Q) is one-dimensional and generated by a parabolic class L. Since either L or −L is nef, we can assume L to be nef. From Theorem 3.4 it follows that for each pair (M, L) there exists a -- 10 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations unique deformation type of a fibration structure. We conclude finiteness of the deformation types of Lagrangian fibrations since there are finitely many orbits of ΓI on the set TeichL. 3.2 Kobayashi hyperbolicity in hyperkahler geometry Definition 3.12: A compact manifold M is called Kobayashi hyperbolic if any holomorphic map C −→ M is constant. For an introduction to the hyperbolic geometry, please see [L]. As an application of Theorem 3.4, we obtain the following result. Theorem 3.13: Let M be an irreducible holomorphic symplectic manifold in one of 4 known classes known, that is, a deformation of a Hilbert scheme of points on K3, a deformation of generalized Kummer variety, or a deformation of one of two examples by O'Grady. Then M is not Kobayashi hyperbolic. Proof: From Brody's lemma [L] it follows that a limit of non-hyperbolic manifolds is again non-hyperbolic. Therefore, it would suffice to find a dense set of non-hyperbolic manifolds within the moduli space. A hyperkahler man- ifold admitting a holomorphic Lagrangian fibration is non-hyperbolic, because it contains abelian subvarieties. As follows from Claim 1.20, all known types of hyperkahler manifolds admit a deformation which has a Lagrangian fibration. By Remark 3.5, such deformations are dense in the moduli. It is conjectured that all hyperkahler and Calabi-Yau manifolds are not hy- perbolic. The strongest result about non-hyperbolicity of hyperkahler manifolds so far was due to F. Campana, who proved in [Cam] that any twistor family of a hyperkahler manifold has at least one fiber which is non-hyperbolic. Acknowledgments. We are grateful to V. Nikulin for many conversations on bilinear forms, Valery Gritsenko for a helpful email, and to F. Bogomolov for his remarks. The first named author thanks the Laboratory of Algebraic Geometry and its Applications for their kind invitation and hospitality during her stay in Moscow in June 2012. We are thankful to the Simons Center for Geometry and Physics and to John Morgan for inviting the second named author and for making our work more enjoyable. Many thanks to Fr´ed´eric Campana, Simone Diverio, Klaus Hulek and Daniel Huybrechts for a discussion of non- hyperbolicity of hyperkahler manifolds. References [AV] Sasha Anan'in, Misha Verbitsky, Any component of moduli of polarized hyper- kaehler manifolds is dense in its deformation space, arXiv:1008.2480, 17 pages, 4 figures. -- 11 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations [Bea] Beauville, A. Varietes Kahleriennes dont la premi`ere classe de Chern est nulle. J. Diff. Geom. 18 (1983) 755 - 782. [Bes] Besse, A., Einstein Manifolds, Springer-Verlag, New York (1987) [Bo1] Bogomolov, F. A., On the decomposition of Kahler manifolds with trivial canonical class, Math. USSR-Sb. 22 (1974) 580 - 583. [Bo2] F. Bogomolov, Hamiltonian Kahler manifolds, Sov. Math. Dokl. 19 (1978) 1462 - 1465. [Bou] Boucksom, S., Higher dimensional Zariski decompositions, Ann. Sci. Ecole Norm. Sup. (4) 37 (2004) no. 1, 45 - 76, arXiv:math/0204336 [Cam] F. Campana, An application of twistor theory to the nonhyperbolicity of certain compact symplectic Kahler manifolds J. Reine Angew. Math., 425:1-7, 1992. [Cas] J. W. S. Cassels, Rational Quadratic Forms, Dover Publications (2008). [Cat] F. Catanese, A Superficial Working Guide to Deformations and Moduli, arXiv:1106.1368, 56 pages. [DP] Demailly, J.-P., Paun, M., Numerical characterization of the Kahler cone of a compact Kahler manifold, math.AG/0105176 also in Annals of Mathematics, 159 (2004) 1247 - 1274. [DPS1] Jean-Pierre Demailly, Thomas Peternell, Michael Schneider, Compact complex manifolds with numerically effective tangent bundles, J. Algebraic Geometry 3 (1994) 295 - 345. [DPS2] Jean-Pierre Demailly, Thomas Peternell, Michael Schneider, Pseudo-effective line bundles on compact Kahhler manifolds, International Journal of Math. 6 (2001) 689 - 741. [F] Fujiki, A. On the de Rham Cohomology Group of a Compact Kahler Symplectic Manifold, Adv. Stud. Pure Math. 10 (1987) 105-165. [GHS] Gritsenko, V., Hulek, K., Sankaran, G. K., Moduli spaces of irreducible sym- plectic manifolds, Compositio Mathematica 146 (2010) no. 2, 404 - 434, arXiv:0802.2078 [G] Mark Gross, The Strominger-Yau-Zaslow conjecture: From torus fibrations to degenerations, arXiv:0802.3407, 44 pages. [HL] R. Harvey, B. Lawson, Calibrated geometries, Acta Math. 148 (1982) 47 - 157. [Hu1] Huybrechts, Daniel, Compact hyperkahler manifolds, Calabi-Yau manifolds and related geometries, Universitext, Springer-Verlag, Berlin, 2003, Lectures from the Summer School held in Nordfjordeid, June 2001, pp. 161-225. [Hu2] Huybrechts, D., Finiteness results for hyperkahler manifolds, J. Reine Angew. Math. 558 (2003) 15 - 22, arXiv:math/0109024 [Hw] Jun-Muk Hwang, Base manifolds for fibrations of projective irreducible sym- plectic manifolds, arXiv:0711.3224, Inventiones mathematicae, vol. 174, issue 3, pp. 625 - 644. [KZ] Maxim Kontsevich, Yan Soibelman, Homological mirror symmetry and torus fibrations, arXiv:math/0011041, Symplectic geometry and mirror symmetry (Seoul, 2000), World Sci. Publishing, River Edge, NJ, 2001, pp. 203 - 263. [L] S. Lang, Introduction to complex hyperbolic spaces, Springer, New York, 1987 -- 12 -- version 2.0, September 15, 2012 L. Kamenova, M. Verbitsky Families of Lagrangian fibrations [McL] McLean, R.C. Deformations of calibrated submanifolds, Comm. Anal. Geom. 6 (1998) 705 - 747. [Ma1] D. Matsushita, On fibre space structures of a projective irreducible symplectic manifold, alg-geom/9709033, math.AG/9903045, also in Topology 38 (1999), No. 1, 79-83. Addendum, Topology 40 (2001) No. 2, 431 - 432. [Ma2] Matsushita, D., Higher direct images of Lagrangian fibrations, Amer. J. Math. 127 (2005) arXiv:math/0010283. [Ma3] D. Matsushita, On deformations of Lagrangian fibrations, arXiv: 0903.2098v1 [math.AG]. [Ni] [R] V. V. Nikulin, Integral symmetric bilinear forms and some of their applications, Math. USSR, Izv. 14 (1980) 103 - 167. A. Rapagnetta, Topological invariants of O'Grady's six dimensional irreducible symplectic variety, Math. Zeitsch., 256, No. 1 (2007) 1-34. [Saw] Sawon, J., Abelian fibred holomorphic symplectic manifolds, Turkish Jour. Math. 27 (2003) no. 1, 197 - 230, math.AG/0404362. [Ser] Serre, J.-P., Cours d'arithm´etique, Presses Universitaires de France (1970). [SYZ] A. Strominger, S.-T. Yau, and E. Zaslow, Mirror Symmetry is T -duality, Nucl. Phys. B479 (1996) 243 - 259. [V1] [V2] [V3] Verbitsky, M., A global Torelli theorem for hyperkahler manifolds, arXiv: 0908.4121, 47 pages. Verbitsky, M., Cohomology of compact hyperkahler manifolds. alg-geom elec- tronic preprint 9501001, 89 pages, LaTeX. Verbitsky, M., Hyperkahler SYZ conjecture and semipositive line bundles, arXiv:0811.0639, 21 pages, GAFA 19, No. 5 (2010) 1481-1493. [V4] Misha Verbitsky, Parabolic nef currents on hyperkaehler manifolds, arXiv:0907.4217, 22 pages. [Vi] Viehweg, E., Quasi-projective Moduli for Polarized Manifolds, Springer- Verlag, Berlin, Heidelberg, New York, 1995, Ergebnisse der Mathe- matik und ihrer Grenzgebiete, 3. Folge, Band 30, also available at http://www.uni-due.de/emat903/books.html Ljudmila Kamenova Department of Mathematics, 3-115 Stony Brook University Stony Brook, NY 11794-3651, USA Misha Verbitsky Laboratory of Algebraic Geometry, Faculty of Mathematics, National Research University HSE, 7 Vavilova Str. Moscow, Russia -- 13 -- version 2.0, September 15, 2012
1810.09338	1	1810	2018-10-22T15:02:05	On Comon's and Strassen's conjectures	[ "math.AG" ]	Comon's conjecture on the equality of the rank and the symmetric rank of a symmetric tensor, and Strassen's conjecture on the additivity of the rank of tensors are two of the most challenging and guiding problems in the area of tensor decomposition. We survey the main known results on these conjectures, and, under suitable bounds on the rank, we prove them, building on classical techniques used in the case of symmetric tensors, for mixed tensors. Finally, we improve the bound for Comon's conjecture given by flattenings by producing new equations for secant varieties of Veronese and Segre varieties.	math.AG	math	ON COMON'S AND STRASSEN'S CONJECTURES ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA Abstract. Comon's conjecture on the equality of the rank and the symmetric rank of a symmetric tensor, and Strassen's conjecture on the additivity of the rank of tensors are two of the most challenging and guiding problems in the area of tensor decomposition. We survey the main known results on these conjectures, and, under suitable bounds on the rank, we prove them, building on classical techniques used in the case of symmetric tensors, for mixed tensors. Finally, we improve the bound for Comon's conjecture given by flattenings by producing new equations for secant varieties of Veronese and Segre varieties. Contents Introduction 1. Notation 2. Comon's conjecture 3. Strassen's conjecture 4. On the rank of x0F References 1 2 4 5 7 11 Introduction Let X Ă PN be an irreducible and reduced non-degenerate variety. The rank rankXppq with respect to X of a point p P PN is the minimal integer h such that p lies in the linear span of h distinct points of X. In particular, if Y Ď X we have that rankXppq ď rankY ppq. Since the h-secant variety SechpXq of X is the subvariety of PN obtained as the closure of the union of all ph ´ 1q-planes spanned by h general points of X, for a general point p P SechpXq we have rankX ppq " h. When the ambient projective space is a space parametrizing tensors we enter the area of tensor decomposition. A tensor rank decomposition expresses a tensor as a linear combination of simpler tensors. More precisely, given a tensor T , lying in a given tensor space over a field k, a tensor rank-1 decomposition of T is an expression of the form (0.1) where the Ui's are linearly independent rank one tensors, and λi P k. The rank of T is the minimal positive integer h such that T admits such a decomposition. T " λ1U1 ` ... ` λhUh Tensor decomposition problems come out naturally in many areas of mathematics and applied sciences. For instance, in signal processing, numerical linear algebra, computer vision, numerical anal- ysis, neuroscience, graph analysis, control theory and electrical networks [KB09], [CM96], [CGLM08], [LO15], [MR13], [MR14], [BFFX17]. In pure mathematics tensor decomposition issues arise while studying the additive decompositions of a general tensor [Dol04], [DK93], [MM13], [Mas16], [RS00], [TZ11], [MMS18]. Comon's conjecture [CGLM08], which states the equality of the rank and symmetric rank of a symmetric tensor, and Strassen's conjecture on the additivity of the rank of tensors [Str73] are two of the most important and guiding problems in the area of tensor decomposition. Date: October 23, 2018. 2010 Mathematics Subject Classification. Primary 15A69, 15A72, 11P05; Secondary 14N05, 15A69. Key words and phrases. Strassen's conjecture, Comon's conjecture, Tensor decomposition, Waring decomposition. 1 2 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA More precisely, Comon's conjecture predicts that the rank of a homogeneous polynomial F P d is equal to its rank with respect to the Segre krx0, . . . , xnsd with respect to the Veronese variety V n variety S n -- pPnqd into which V n d is diagonally embedded, that is rankV n pF q " rankS npF q. Strassen's conjecture was originally stated for triple tensors and then generalized to a number of different contexts. For instance, for homogeneous polynomials it says that if F P krx0, . . . , xnsd and pF `Gq " G P kry0, . . . , ymsd are homogeneous polynomials in distinct sets of variables then rankV n`m`1 rankV n d pF q ` rankV m In Sections 2 and 3, while surveying the state of the art on Comon's and Strassen's conjectures, we push a bit forward some standard techniques, based on catalecticant matrices and more generally on flattenings, to extend some results on these conjectures, known in the setting of Veronese and Segre varieties, for Segre-Veronese and Segre-Grassmann varieties that is to the context of mixed tensors. pGq. d d d In Section 4 we introduce a method to improve a classical result on Comon's conjecture. By standard arguments involving catalecticant matrices it is not hard to prove that Comon's conjecture holds for the general polynomial in krx0, . . . , xnsd of symmetric rank h as soon as h ă`n`t d 2.2. We manage to improve this bound looking for equations for the ph ´ 1q-secant variety Sech´1pV n d q, not coming from catalecticant matrices, that are restrictions to the space of symmetric tensors of equations of the ph ´ 1q-secant variety Sech´1pS nq. We will do so by embedding the space of degree d polynomials into the space of degree d ` 1 polynomials by mapping F to x0F and then considering suitable catalecticant matrices of x0F rather than those of F itself. n , see Proposition 2 u Implementing this method in Macaulay2 we are able to prove for instance that Comon's conjecture holds for the general cubic polynomial in n ` 1 variables of rank h " n ` 1 as long as n ď 30. Note that for cubics the usual flattenings work for h ď n. Acknowledgments. The second and third named authors are members of the Gruppo Nazionale per le Strutture Algebriche, Geometriche e le loro Applicazioni of the Istituto Nazionale di Alta Matematica "F. Severi" (GNSAGA-INDAM). We thank the referees for helping us to improve the exposition. Let n " pn1, . . . , npq and d " pd1, . . . , dpq be two p-uples of positive integers. Set 1. Notation d " d1 ` ` dp, n " n1 ` ` np, and N pn, dq " pźi"1ni ` di ni Let V1, . . . , Vp be vector spaces of dimensions n1 ` 1 ď n2 ` 1 ď ď np ` 1, and consider the product The line bundle Pn " PpV 1 q PpV p q. OPnpd1, . . . , dpq " OPpV 1 qpd1q b b OPpV 1 qpdpq induces an embedding d : PpV σν n 1 q PpV prv1s, . . . , rvpsq ÞÝÑ p q ÝÑ PpSymd1 V 1 b b Symdp V p q " PN pn,dq´1, dp p s 1 b b v rvd1 where vi P Vi. We call the image SV n d " σν n d pPnq Ă PN pn,dq´1 a Segre-Veronese variety. It is a smooth variety of dimension n and degree pn1``npq! PN pn,dq´1. n1!...np! dn1 np 1 . . . d p in When p " 1, SV n d is a Veronese variety. In this case we write V n d for SV n d , and ν n 1,...,1 is a Segre variety. In this case we write S n for SV n d for the Veronese 1,...,1, embedding. When d1 " " dp " 1, SV n and σn for the Segre embedding. Note that d " σn1 where n1 " pN pn1, d1q ´ 1, . . . , N pnp, dpq ´ 1q. σν n ´ν n1 d1 ν np dp¯ , ON COMON'S AND STRASSEN'S CONJECTURES 3 Similarly, given a p-uple of k-vector spaces pV n1 1 , ..., V np p q and p-uple of positive integers d " pd1, ..., dpq we may consider the Segre-Plücker embedding σπn d : Grpd1, n1q Grpdp, npq ÝÑ PpŹd1 V n1 prH1s, . . . , rHpsq 1 b bŹdp V rH1 b b Hps ÞÝÑ np p q " PN pn,dq´1, where N pn, dq "śp i"1`ni di. We call the image SGn d " σπn d pGrpd1, n1q Grpdp, npqq Ă PN pn,dq a Segre-Grassmann variety. 1.1. Flattenings. Let V1, ..., Vp be k-vector spaces of finite dimension, and consider the tensor product V1 b ... b Vp " pVa1 b ... b Vasq b pVb1 b ... b Vbp´sq " VA b VB with A Y B " t1, ..., pu, B " Ac. Then we may interpret a tensor T P V1 b ... b Vp " VA b VB A Ñ VAc. Clearly, if the rank of T is at most r then the rank of rT is at most r as well. Indeed, a decomposition of T as a linear combination of r rank one tensors yields a linear A q Ď VAc. The as a linear map rT : V subspace of VAc, generated by the corresponding rank one tensors, containing rT pV matrix associated to the linear map rT is called an pA, Bq-flattening of T . In the case of mixed tensors we can consider the embedding Symd1 V1 b ... b Symdp Vp ãÑ VA b VB where VA " Syma1 V1 b ... b Symap Vp, VB " Symb1 V1 b ... b Symbp Vp, with di " ai ` bi for any i " 1, ..., p. In particular, if n " 1 we may interpret a tensor F P Symd1 V1 as a degree d1 homogeneous polynomial on PpV A Ñ VB is nothing but the a1-th catalecticant matrix of F , that is the matrix whose rows are the coefficient of the partial derivatives of order a1 of F . 1 q. In this case the matrix associated to the linear map rF : V Similarly, by considering the inclusion d1ľ V1 b ... b dpľ Vp ãÑ VA b VB get the so called skew-flattenings. We refer to [Lan12] for details on the subject. where VA "Źa1 V1 b ... bŹap Vp, VB "Źb1 V1 b ... bŹbp Vp, with di " ai ` bi for any i " 1, ..., p, we partial derivatives of a polynomial F P krx0, ..., xnsd are `n`s n homogeneous polynomials of degree Remark 1.2. The partial derivatives of an homogeneous polynomials are particular flattenings. The d ´ s spanning a linear space HBsF Ď Ppkrx0, ..., xnsd´sq. If F P krx0, ..., xnsd admits a decomposition as in (0.1) then F P SechpV n F P SechpV n partial derivatives of order s of F can be decomposed as linear combinations of Ld´s d q can be written as in (0.1). 1 ` ... ` λhLd If F " λ1Ld d q, and conversely a general h is a decomposition then the as well. , ..., Ld´s 1 h Therefore, the linear space ALd´s 1 , . . . , Ld´s h E contains HBsF . 1.3. Rank and border rank. Let X Ă PN be an irreducible and reduced non-degenerate variety. We define the rank rankX ppq with respect to X of a point p P PN as the minimal integer h such that there exist h points in linear general position x1, . . . , xh P X with p P xx1, . . . , xhy. Clearly, if Y Ď X we have that rankX ppq ď rankY ppq (1.4) The border rank rankXppq of p P PN with respect to X is the smallest integer r ą 0 such that p is in the Zariski closure of the set of points q P PN such that rankXpqq " r. In particular rankXppq ď rankXppq. Recall that given an irreducible and reduced non-degenerate variety X Ă PN , and a positive integer h ď N the h-secant variety SechpXq of X is the subvariety of PN obtained as the Zariski closure of the union of all ph ´ 1q-planes spanned by h general points of X. In other words rankX ppq is computed by the smallest secant variety SechpXq containing p P PN . 4 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA Now, let Y, Z be subvarieties of an irreducible projective variety X Ă PN , spanning two linear subspaces PNY :" xY y , PNZ :" xZy Ď PN . Fix two points pY P PNY , pZ P PNZ , and consider a point p P xpY , pZ y. Clearly (1.5) rankXppq ď rankY ppY q ` rankZppZ q 2. Comon's conjecture d : PpV It is natural to ask under which assumptions (1.4) is indeed an equality. Consider the Segre- Veronese embedding σν n p q " PN pn,dq´1 with V1 -- -- Vp -- V k-vector spaces of dimension n ` 1. Its composition with the diagonal embedding i : PpV q Ñ PpV d of degree d " d1 ` ` dp. Let d Ď SV n V n d in PN pn,dq´1. d be the corresponding Veronese variety. We will denote by Πn,d the linear span of V n p q is the Veronese embedding of ν n p q Ñ PpSymd1 V 1 b b Symdp V 1 q PpV 1 q PpV In the notations of Section 1.3 set X " SV n may consider its symmetric rank srkpT q :" rankV n d d and Y " V n d . For any symmetric tensor T P Πn,d we pT q as a mixed pT q and its rank rankpT q :" rankS V n d tensor. Comon's conjecture predicts that in this particular setting the inequality (1.4) is indeed an equality [CGLM08]. Conjecture 1 (Comon's). Let T be a symmetric tensor. Then rankpT q " srkpT q. Conjecture 1 has been generalized in a number of directions for complex border rank, real rank and real border rank, see [Lan12, Section 5.7.2] for a full overview. Note that when d " 2 Comon's conjecture is true. Indeed, SechpS nq is cut out by the size ph ` 1q 2 q is cut out by the size ph ` 1q ph ` 1q minors ph ` 1q minors of a general square matrix and SechpV n of a general symmetric matrix, that is SechpV n 2 q " SechpS nq X Πn,2. Conjecture 1 has been proved in several special cases. For instance, when the symmetric rank is at most two [CGLM08], when the rank is less than or equal to the order [ZHQ16], for tensors belonging to tangential varieties to Veronese varieties [BB13], for tensors in C2 b Cn b Cn [BL13], when the rank is at most the flattening rank plus one [Fri16], for the so called Coppersmith -- Winograd tensors [LM17], for symmetric tensors in C4 b C4 b C4 and also for symmetric tensors of symmetric rank at most seven in Cn b Cn b Cn [Sei18]. On the other hand, a counter-example to Comon's conjecture has recently been found by Y. Shitov [Shi17a]. The counter-example consists of a symmetric tensor T in C800 C800 C800 which can be written as a sum of 903 rank one tensors but not as a sum of 903 symmetric rank one tensors. It is important to stress that for this tensor T rank and border rank are quite different. Comon's conjecture for border ranks is still completely open [Shi17a, Problem 25]. Even though it has been recently proven false in full generality, we believe that Comon's conjecture is true for a general symmetric tensor, perhaps it is even true for those tensor for which rank T " rank T . In what follows we use simple arguments based on flattenings to give sufficient conditions for Comon's conjecture, recovering a known result, and its skew-symmetric analogue. Lemma 2.1. The tensors T P SechpSV n proper closed subset of SechpSV n variety SV n d with the Segre-Grassmann variety SGn d . d q such that dimprT pV A qq ď h´1 for a given flattening rT form a d q. Furthermore, the same result holds if we replace the Segre-Veronese Proof. Let T P SechpSV n A qq ď h ´ 1. This condition forces the pA, Bq-flattening matrix to have rank at most h ´ 1. On the other hand, by [SU00, Proposition 4.1] d q, and therefore define a proper closed subset of SechpSV n these minors do not vanish on SechpSV n d q. In the Segre-Grassmann setting we argue in the same way by using skew-flattenings. (cid:3) d q be a general point. Assume that dimprT pV Proposition 2.2. [IK99] For any integer h ă`n`t d that for any T P Uh the rank and the symmetric rank of T coincide, that is n there exists an open subset Uh Ď SecpV d 2 u nq such rankpT q " srkpT q ON COMON'S AND STRASSEN'S CONJECTURES 5 Proof. First of all, note that we always have rankpT q ď srkpT q. Furthermore, Section 1.1 yields that for A qq holds. Since T is symmetric and its catalecticant matrices are particular flattenings we get that rankpT q ě dimpHBsT q for any s ě 0. any pA, Bq-flattening rT : V A Ñ VB the inequality rankpT q ě dimprT pV d q we have srkpT q " h, and if h ă`n`s Now, for a general T P SechpV n 2 u, then Lemma 2.1 yields dimpHBsT q " h. Therefore, under these conditions we have the following chain of inequalities n , where s " t d dimpHBsT q ď rankpT q ď srkpT q " dimpHBsT q and hence rankpT q " srkpT q. (cid:3) Now, consider the Segre-Plücker embedding PpV1q . . . PpVpq Ñ PpŹd1 V1 b bŹdp Vpq " PN pn,dq´1 with V1 -- . . . -- Vp -- V k-vector spaces of dimension n`1. Its composition with the diagonal embedding i : PpV q Ñ PpV1q PpVpq is the Plücker embedding of Grpd, nq with d " d1 ` . . . ` dp. Let Grpd, nq Ď SGn d be the corresponding Grassmannian and let us denote by Πn,d its linear span in PN pn,dq´1. For any skew-symmetric tensor T P Πn,d we may consider its skew rank skrkpT q that is its rank with respect to the Grassmannian Grpd, nq Ď Πn,d, and its rank rankpT q as a mixed tensor. Playing the same game as in Proposition 2.2 we have the following. Proposition 2.3. For any integer h ă` n 2 u there exists an open subset Uh Ď SechpGrpd, nqq such that for any T P Uh the rank and the skew rank of T coincide, that is t d rankpT q " skrkpT q V Proof. As before for any tensor T we have rankpT q ď skrkpT q. For any pA, Bq-skew-flattening rT : A Ñ VB we have skrkpT q ě dimprT pV A qq. Furthermore, since rT is in particular a flattening also the inequality rankpT q ě dimprT pV Now, for a general T P SechpGrpd, nqq we have skrkpT q " h, and if h ă`n A qq, where rTs is the skew-flattening corresponding to the partition 2.1 yields skrkpT q " dimprTspV dimprTspV A qq ď rankpT q ď skrkpT q " dimprTspV s, where s " t d ps, d ´ sq of d. Therefore, we deduce that and hence rankpT q " skrkpT q. A qq holds. 2 u, Lemma A qq (cid:3) Remark 2.4. Propositions 2.2, 2.3 suggest that whenever we are able to write determinantal equations for secant varieties we are able to verify Comon's conjecture. We conclude this section suggesting a possible way to improve the range where the general Comon's conjecture holds giving a conjectural way to produce determinantal equations for some secant varieties. Set n " pn, . . . , nq, pd ` 1q-times, n1 " pn, . . . , nq, d-times, and consider the corresponding Segre varieties X :" S n, X1 :" S n1 and Veronese varieties Y " V n P Y and let Π be the linear space spanned by the polynomials of the form x0F , where F is a polynomial of degree d. This allow us to see Y1 Ď Π. Note that polynomials of the form x0Ld 1 lie in the tangent space of Y at Ld`1 d . Fix the polynomial xd`1 , and therefore rankY px0Lbdq " 2. d`1, Y1 :" V n 1 0 Hence for a polynomial F of degree d we have rankY px0F q ď 2 rankY1pF q. Our aim is to understand when the equality holds. We may mimic the same construction for the Segre varieties X and X1, and use determinantal equations for the secant varieties of X1 to give determinantal equations of the secant varieties of X In particular, as soon as d is odd and d ă n, this and henceforth conclude Comon's conjecture. produces new determinantal equations for SechpX1q and SechpY1q with 2h ă`n` d`1 would give new cases in which the general Comon's conjecture holds. Unfortunately, we are only able to successfully implement this procedure in very special cases, see Section 4. . Therefore, this n 2 Another natural problem consists in giving hypotheses under which in (1.5) equality holds. Consider 1 q " PN pn,dq´1, the triple Segre embedding σn : PpV 3 q " Pa Pb Pc Ñ PpV 1 q PpV 2 q PpV 2 bV 1 bV 3. Strassen's conjecture 6 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA and let S n be the corresponding Segre variety. Now, take complementary subspaces Pa1, Pa2 Ă Pa, Pb1, Pb2 Ă Pb, Pc1, Pc2 Ă Pc, and let S pa1,b1,c1q, S pa2,b2,c2q be the Segre varieties associated respectively to Pa1 Pb1 Pc1 and Pa2 Pb2 Pc2. In the notations of Section 1.3 set X " S n, Y " S pa1,b1,c1q and Z " S pa1,b1,c1q. Strassen's conjecture states that the additivity of the rank holds for triple tensors, or in onther words that in this setting the inequality (1.5) is indeed an equality [Str73]. Conjecture 2 (Strassen's). In the above notation let T1 P @S pa1,b1,c1qD , T2 P @S pa2,b2,c2qD be two tensors. Then rankpT1 ' T2q " rankpT1q ` rankpT2q. Even though Conjecture 2 was originally stated in the context of triple tensors that is bilinear forms, with particular attention to the complexity of matrix multiplication, a number of generalizations are immediate. For instance, we could ask the same question for higher order tensors, symmetric tensors, mixed tensors and skew-symmetric tensors. It is also natural to ask for the analogue of Conjecture 2 for border rank. This has been answered negatively [Sch81]. Conjecture 2 and its analogues have been proven when either T1 or T2 has dimension at most two, when rankpT1q can be determined by the so called substitution method [LM17], when dimpV1q " 2 both for the rank and the border rank [BGL13], when T1, T2 are symmetric that is homogeneous polynomials in disjoint sets of variables, either T1, T2 is a power, or both T1 and T2 have two variables, or either T1 or T2 has small rank [CCC15], and also for other classes of homogeneous polynomials [CCO17], [Tei15]. As for Comon's conjecture a counterexample to Strassen's conjecture has recentely been given by Y. Shitov [Shi17b]. In this case Y. Shitov proved that over any infinite field there exist tensors T1, T2 such that the inequality in Conjecture 2 is strict. In what follows we give sufficient conditions for Strassen's conjecture, recovering a known result, and for its mixed and skew-symmetric analogues. Proposition 3.1. [IK99] Let V1, V2 be k-vector spaces of dimensions n ` 1, m ` 1, and consider V " V1 ' V2. Let F P SymdpV1q Ă SymdpV q and G P SymdpV2q Ă SymdpV q be two homogeneous polynomials. If there exists an integer s ą 0 such that then srkpF ` Gq " srkpF q ` srkpGq. dimpHBsF q " srkpF q, dimpHBsGq " srkpGq Proof. Clearly, srkpF ` Gq ď srkpF q ` srkpGq holds in general. On the other hand, our hypothesis yields srkpF q ` srkpGq " dimpHBsF q ` dimpHBsF q " dimpHBsF `Gq ď srkpF ` Gq where the last inequality follows from Remark 1.2. (cid:3) Remark 3.2. The argument used in the proof of Proposition 3.1 works for F P PN pn,dq general only if for the generic rank we have t`n`d d n`1 u ď`n`t d 2 u n . For instance, when n " 3, d " 6 the generic rank is 21 while the maximal dimension of the spaces spanned by partial derivatives is 20. Proposition 3.3. Let V1, . . . , Vp and W1, . . . , Wp be k-vector spaces of dimension n1 `1, . . . , np `1 and m1 ` 1, . . . , mp ` 1 respectively. Consider Ui " Vi ' Wi for every 1 ď i ď p. Let T1 P Symd1 V1 b b Symdp Vp Ă Symd1 U1b bSymdp Up and T2 P Symd1 W1b bSymdp Wp Ă Symd1 U1b bSymdp Up be two mixed tensors. If for any i P t1, .., pu there exists a pair pai, biq with ai`bi " di and pA, Bq-flattenings rT1 : V rT2 : V A Ñ VB as in (1.1) such that A qq " rankpT1q, A qq " rankpT2q A Ñ VB, then rankpT1 ` T2q " rankpT1q ` rankpT2q. dimprT1pV dimprT2pV Proof. Clearly, rankpT1 ` T2q ď rankpT1q ` rankpT2q. On the other hand, our hypothesis yields rankpT1q ` rankpT2q " dimprT1pV A qq ` dimprT2pV where ČT1 ` T2 denotes the pA, Bq-flattening of the mixed tensor T1 ` T2. A qq " dimp ČT1 ` T2pV A qq ď rankpT1 ` T2q (cid:3) ON COMON'S AND STRASSEN'S CONJECTURES 7 Arguing as in the proof of Proposition 3.3 with skew-symmetric flattenings we have an analogous statement in the Segre-Grassmann setting. Proposition 3.4. Let V1, . . . , Vp and W1, . . . , Wp be k-vector spaces of dimension n1 ` 1, . . . , np ` 1 and m1 ` 1, . . . , mp ` 1 respectively. Consider Ui " Vi ' Wi for every 1 ď i ď p, and let T1 P be two skew-symmetric tensors with di ď mintni ` 1, mi ` 1u. Źd1 V1 b bŹdp Vp ĂŹd1 U1 b bŹdp Up and T2 PŹd1 W1 b bŹdp Wp ĂŹdp U1 b bŹdp Up rT1 : V If for any i P t1, . . . , pu there exists a pair pai, biq with ai ` bi " di and pA, Bq-skew-flattenings A Ñ VB, rT2 : V A Ñ VB as in (1.1) such that A qq " rankpT1q, A qq " rankpT2q then rankpT1 ` T2q " rankpT1q ` rankpT2q. dimprT1pV dimprT2pV 4. On the rank of x0F In this section, building on Remark 2.4, we present new cases in which Comon's conjecture holds. Recall, that for a smooth point x P X, the a-osculating space Ta xX of X at x is roughly the smaller linear subspace locally approximating X up to order a at x, and the a-osculating variety T aX of X is defined as the closure of the union of all the osculating spaces T aX " ďxPX Ta xX For any 1 ď a ď d ´ 1 the osculating space Ta d of order a at the point rLds P Vd can be written as V n rLds Ta rLds V n d "ALd´aF F P krx0, . . . , xnsaE Ď PN d is the space of homogeneous polynomials whose derivatives of order less than or rLds V n Equivalently, Ta equal to a in the direction given by the linear form L vanish. Note that dimpTa Tb rLds V n a into the osculating space Ta d for any b ď a. Moreover, for any 1 ď a ď d and rLds P V n V n d by considering d Ď Ta V n V n rLds rLds V n d q "`n`a n ´ 1 and d we can embed a copy of rLds a " tLd´aM a M P krx0, . . . , xns1u Ď Ta V n rLds V n d Remark 4.1. Let us expand the ideas in Remark 2.4. We can embed d " tx0Ld L P krx0, . . . , xns1u Ď Td V n rxd 0s V n d`1 and Remark 2.4 yields that (4.2) SechpV n d q Ď Sec2hpV n d`1q X Td rLd`1s V n d`1 This embedding extends to an embedding at the level of Segre varieties, and, in the notation of Remark 2.4, we have that SechpS n1q Ď Sec2hpS nq. Assume that for a polynomial F P SechpV n d q we have F P Sech´1pS n1q. Then x0F P Sec2h´2pS nq. Now, if we find a determinantal equation of Sec2h´2pV n d`1q coming as the restriction to Π, the space of symmetric tensors, of a determinantal equation of Sec2h´2pS nq, and not vanishing at x0F then x0F R Sec2h´2pS nq and hence F R Sech´1pS n1q proving Comon's conjecture for F . This will be the leading idea to keep in mind in what follows. The determinantal equations involved will always come from minors of suitable catalecticant matrices, that can be therefore seen as the restriction to Π of determinantal equations for the secants of the Segre coming from non symmetric flattenings. It is easy to give examples where the inequality (4.2) is strict. When n " 1 the generic rank is gd " r d`1 2 s. Then for d odd we have gd " gd´1 while for d even we have gd " gd´1 ` 1. Hence rankVd x0F ă 2 rankVd´1 F if 2 rankVd´1 F ą gd d is the rational normal curve. It is natural to ask if the inequality is indeed an equality as long as the rank is subgeneric. In the case n " 1 we have the following result. 2 , where Vd :" V 1 8 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA Proposition 4.3. Let Vd :" V 1 not exist kh ą 0 such that SechpVdq Ď Sec2h´khpVd`1q X Td Vd`1. rxd`1s d be the degree d rational normal curve. If 2h ă gd`1 then there does Proof. Clearly, it is enough to prove the statement for kh " 1. Let p P SechpVdq be a general point. Then p P@rx0Ld 1s, . . . , rx0Ld hsD with Li general linear forms. In particular Vd`1E p P H :"AT Vd`1, . . . , T rLd`1 rLd`1 s s h 1 i Note that dimpHq " 2h ´ 1. Now, assume that p is contained also in Sec2h´1pVd`1q. Then there exists a linear subspace H 1 Ă Pd`1 of dimension 2h ´ 2 passing through p intersecting Vd`1 at 2h ´ 1 points q1, . . . , qr counted with multiplicity. Let qi1, . . . , qir be the points among the qi coinciding with some of the rLd`1 s and such that the intersection multiplicity of H 1 and Vd`1 at qij is one, and qj1, . . . , qjr be the points among the qi coinciding with some of the rLd`1 s and such that the intersection multiplicity of H 1 and Vd at qjk is greater that or equal to two. Set Π :" xH, H 1y, then dimpΠq " 2h´1`2h´2´ir´2jr and Π intersects Vd`1 at 2h`p2h´1´ir´2jrq points counted with multiplicity. Consider general points b1, . . . , bs P Vd`1 with s " ir ` 2jr, and the linear space Π1 " xΠ, b1, . . . , bsy. Therefore, dimpΠ1q " 4h ´ 3 and Π1 intersects Vd`1 at 4h ´ 1 points counted with multiplicity. Since 2h ď d`3 adding enough general points to Π1 we may construct a hyperplane in Pd`1 intersecting Vd`1 at d ` 2 points counted with multiplicity, a contradiction. (cid:3) 2 i Proposition 4.3 can be applied to get results on the rank of a special class of matrices called Hankel homogeneous coordinates on Ppkrx0, x1sdq. Furthermore, consider the matrices 0 xi 1 ` . . . ` Zdxd 1 be a binary form and consider rZ0, . . . , Zds as matrices. Let F " Z0xd 0 ` . . . `` d d´iZixd´i M2n " Z0 ... Zn . . . Zn ... . . . . . . Zd ‹‚, M2n`1 " Z0 ... Zn`1 . . . Zn ... . . . . . . Zd ‹‚ It is well known that the ideal of SechpVdq is cut out by the minors of Md of size ph ` 1q ph ` 1q [LO15]. Now, consider a polynomial F P krx0, x1sd with homogeneous coordinates rZ0, . . . , Zds. Then F 1 :" x0F P krx0, x1sd`1 has homogeneous coordinates rZ 1 0, . . . , Z 1 d`1s with Z 1 i " d ` 1 ´ i d ` 1 Zi In order to determine the rank of F 1 we have to relate the rank of the matrices Z0 d d`1 Z1 ... d´n`2 d`1 Zn´1 d´n`1 d`1 Zn Z0 d d`1 Z1 ... d´n`2 d`1 Zn d´n`1 d`1 Zn`1 d d`1 Z1 . . . . . . . . . . . . d d`1 Z1 . . . . . . . . . . . . . . . . . . . . . . . . 1 d`1 Zd . . . . . . . . . . . . 1 d`1 Zd d`1´n d`1 Zn d´n d`1 Zn`1 ... 1 d`1 Zd 0 ‹‹‹‹‹‹‚ d´n d`1 Zn d´n´1 d`1 Zn`2 ... 1 d`1 Zd 0 ‹‹‹‹‹‹‚ N2n " N2n`1 " with the rank of Md. Definition 4.4. A matrix A " pAi,jq P M pa, bq such that Ai,j " Ah,k whenever i ` j " h ` k is called a Hankel matrix. ON COMON'S AND STRASSEN'S CONJECTURES 9 In particular all the matrices of the form Md and Nd considered above are Hankel matrices. Let M pa, bq be the vector space of a b matrices with coefficients in the base field k. For any h ď minta, bu let RankrpM pa, bqq Ď M pa, bq be the subvariety consisting of all matrices of rank at most h. Now, consider the map β : N ÝÑ N N given by βp2nq " pn` 1, n` 1q and βp2n` 1q " pn` 2, n` 1q. For any d ě 1 we can view the subspace Hd Ď M pβpdqq formed by matrices of the form Md as the subspace of Hankel matrices. Now, given any linear morphism f : M pa, bq Ñ M pc, dq we can ask if for some s ď mintc, du we have f pRankhpM pa, bqqq Ď RankspM pc, dqq. Corollary 4.5. Consider the linear morphism Then αdpHdq Ď Hd`1 and αdpRankhpM pβpdqq X Mdqq Ď Rank2hpM pβpd ` 1qqq X Md`1. αd : M pβpdqq ÝÑ M pβpd ` 1qq pAi,jq ÞÝÑ ´ d´pi`j´3q d`1 Ai,j¯ Proof. Since αdpAi,jq " αdpAh,kq when i ` j " h ` k we have that αdpHdq Ď Hd`1. By Proposition 4.3 RankhpM pβpdqq X Hd " SechpVdq, and by construction αdpMdq is the linear change of coordinates mapping a binary form F P krx0, x1sd to F 1 " x0F P krx0, x1sd`1. Since SechpVdq Ď Sec2hpVd`1q X Td rxd`1sVd`1, if an h h minor of a general matrix B in M pβpdqq does not vanish, under the assumption that all the ph ` 1q ph ` 1q minors of B vanish, then there is a 2h 2h minor of αdpBq that does not vanish. (cid:3) When n ě 2 we are able to determine, via Macaulay2 [Mac92] aided methods, the rank of x0F in some special cases. 4.5.1. pn, dq " p2, 2q. The variety Sec3pV 2 see for instance [LO15, Section 1.1]. With a Macaulay2 computation we prove that if F P Sec2pV 2 general then the Aronhold invariant does not vanish at x0F , hence rank x0F " 2 rank F . 3 q is the hypersurface in P9 cut out by the Aronhold invariant, 2 q is 4.5.2. pn, dq " p2, 3q. The varieties Sec5pV 2 3 q are both hypersurfaces, given respectively by the determinant of the catalecticant matrix of second partial derivatives and the Aronhold invariant [LO15, Section 1.1]. With Macaulay2 we prove that the determinant of the second catalecticant matrix does not vanish at x0F for F P Sec3pV 2 3 q general, hence rank x0F " 2 rank F . 4 q and Sec3pV 2 4.5.3. pn, dq " p3, 3q. The secant variety Sec9pV 3 cant matrix [LO15, Section 1.1] while Sec5pV 3 shows that T3 4 Ď Sec9pV 3 V 3 rx4 0s 4 q is the hypersurface cut out by the second catalecti- 3 q is the entire osculating space. A Macaulay2 computation 4 q. This proves that rank x0F ă 2 rank F , for F general. 4.5.4. pn, dq " p4, 3q. In this case Sec8pV 4 the second catalecticant matrix [LO15, Section 1.1]. Again using Macaulay2 we show that T3 Sec14pV 4 4 q is given by the determinant of 4 Ď 4 q. This proves that rank x0F ă 2 rank F , for F general. 4 and Sec14pV 4 3 q " T3 rx4 0s rx4 0s V 4 V 4 Corollary 4.6. For the osculating varieties T 3V 3 4 and T 3V 4 T 3V 3 4 Ď Sec9pV 3 4 q, T 3V 4 4 we have 4 Ď Sec14pV 4 4 q Proof. The action of P GLpn ` 1q on Pn extends naturally to an action on PN pn,dq stabilizing V n more generally the secant varieties SechpV n SechpV n conclude by applying 4.5.3 and 4.5.4. d q. Since this action is transitive on V n d and V n d Ď d . Finally, we (cid:3) d for any point rLds P V n d q if and only if Ta d we have Ta rxd 0s d that is T aV n d Ď SechV n V n d Ď SechV n rLds 4.7. Macaulay2 implementation. In the Macaulay2 file Comon-1.0.m2 we provide a function called Comon which operates as follows: - Comon takes in input three natural numbers n, d, h; 2 u - if h ă `n`t d n then the function returns that Comon's conjecture holds for the general degree d polynomial in n ` 1 variables of rank h by the usual flattenings method in Proposition 2.2. If not, and d is even then it returns that the method does not apply; 10 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA apply; - if d is odd and`n`k - if d is odd, `n`k - if d is odd,`n`k n ă 2`n`k´1 n ě 2`n`k´1 n ě 2`n`k´1 n a polynomial of the form , where k " t d`1 and 2h ´ 1 ą`n`k and 2h ´ 1 ď`n`k n n since 2h ´ 2 must be smaller than the number of order k partial derivatives; 2 u, then again it returns that the method does not n then it returns that the method does not apply n then Comon, in the spirit of Remark 4.1, produces F " hÿi"1 pai,0x0 ` ` ai,nxnqd then substitutes random rational values to the ai,j, computes the polynomial G " x0F , the catalec- ticant matrix D of order k partial derivatives of G, extracts the most up left 2h ´ 1 2h ´ 1 minor P of D, and compute the determinant detpP q of P ; - if detpP q " 0 then Comon returns that the method does not apply, otherwise it returns that Comon's conjecture holds for the general degree d polynomial in n ` 1 variables of rank h. Note that since the function random is involved Comon may return that the method does not apply even though it does. Clearly, this event is extremely unlikely. Thanks to this function we are able to prove that Comon's conjecture holds in some new cases that are not covered by Proposition 2.2. Since the case n " 1 is covered by Proposition 4.3 in the following we assume that n ě 2. Theorem 4.8. Assume n ě 2 and set h " `n`t d n . Then Comon's conjecture holds for the general degree d homogeneous polynomial in n ` 1 variables of rank h in the following cases: 2 u - d " 3 and 2 ď n ď 30; - d " 5 and 3 ď n ď 8; - d " 7 and n " 4. Proof. The proof is based on Macaualy2 computations using the function Comon exactly as shown in Example 4.9 below. (cid:3) Example 4.9. We apply the function Comon in a few interesting cases: Macaulay2, version 1.12 with packages: ConwayPolynomials, Elimination, IntegralClosure, InverseSystems, LLLBases, PrimaryDecomposition, ReesAlgebra, TangentCone i1 : loadPackage "Comon-1.0.m2"; i2 : Comon(5,3,4) Lowest rank for which the usual flattenings method does not work = 6 o2 = Comon's conjecture holds for the general degree 3 homogeneous polynomial in 6 variables of rank 4 by the usual flattenings method i3 : Comon(5,3,6) Lowest rank for which the usual flattenings method does not work = 6 o3 = Comon's conjecture holds for the general degree 3 homogeneous polynomial in 6 variables of rank 6 i4 : Comon(5,3,7) Lowest rank for which the usual flattenings method does not work = 6 o4 = The method does not apply --- The determinant vanishes i5 : Comon(5,5,21) Lowest rank for which the usual flattenings method does not work = 21 o5 = Comon's conjecture holds for the general degree 5 homogeneous polynomial in 6 variables of rank 21 i6 : Comon(4,7,35) Lowest rank for which the usual flattenings method does not work = 35 o6 = Comon's conjecture holds for the general degree 7 homogeneous polynomial in 5 variables of rank 35 ON COMON'S AND STRASSEN'S CONJECTURES 11 References [BB13] E. Ballico and A. Bernardi, Tensor ranks on tangent developable of Segre varieties, Linear Multilinear Algebra 61 (2013), no. 7, 881 -- 894. MR 3175333 [BFFX17] A. Buscarino, L. Fortuna, M. Frasca, and M.G. Xibilia, Positive-real systems under lossless transformations: [BGL13] [BL13] [CCC15] Invariants and reduced order models, Journal of the Franklin Institute 354 (2017), no. 11, 4273 -- 4288. J. Buczyński, A. Ginensky, and J. M. Landsberg, Determinantal equations for secant varieties and the Eisenbud-Koh-Stillman conjecture, J. Lond. Math. Soc. (2) 88 (2013), no. 1, 1 -- 24. MR 3092255 J. Buczyński and J. M. Landsberg, Ranks of tensors and a generalization of secant varieties, Linear Algebra Appl. 438 (2013), no. 2, 668 -- 689. MR 2996361 E. Carlini, M. V. Catalisano, and L. Chiantini, Progress on the symmetric Strassen conjecture, J. Pure Appl. Algebra 219 (2015), no. 8, 3149 -- 3157. MR 3320211 [CCO17] E. Carlini, M. V. Catalisano, and A. Oneto, Waring loci and the Strassen conjecture, Adv. Math. 314 (2017), 630 -- 662. MR 3658727 [CGLM08] P. Comon, G. Golub, L. Lim, and B. Mourrain, Symmetric tensors and symmetric tensor rank, SIAM J. [CM96] [DK93] [Dol04] [Fri16] [IK99] [KB09] [Lan12] [LM17] [LO15] Matrix Anal. Appl. 30 (2008), no. 3, 1254 -- 1279. MR 2447451 P. Comon and B. Mourrain, Decomposition of quantics in sums of powers of linear forms, Signal Processing 53 (1996), no. 2, 93 -- 107. I. V. Dolgachev and V. Kanev, Polar covariants of plane cubics and quartics, Adv. Math. 98 (1993), no. 2, 216 -- 301. MR 1213725 I. V. Dolgachev, Dual homogeneous forms and varieties of power sums, Milan J. Math. 72 (2004), 163 -- 187. MR 2099131 S. Friedland, Remarks on the symmetric rank of symmetric tensors, SIAM J. Matrix Anal. Appl. 37 (2016), no. 1, 320 -- 337. MR 3474849 Anthony Iarrobino and Vassil Kanev, Power sums, Gorenstein algebras, and determinantal loci, Lecture Notes in Mathematics, vol. 1721, Springer-Verlag, Berlin, 1999, Appendix C by Iarrobino and Steven L. Kleiman. MR 1735271 T. G. Kolda and B. W. Bader, Tensor decompositions and applications, SIAM Rev. 51 (2009), no. 3, 455 -- 500. MR 2535056 J. M. Landsberg, Tensors: geometry and applications, Graduate Studies in Mathematics, vol. 128, American Mathematical Society, Providence, RI, 2012. MR 2865915 J. M. Landsberg and M. Michałek, Abelian tensors, J. Math. Pures Appl. (9) 108 (2017), no. 3, 333 -- 371. MR 3682743 J. M. Landsberg and G. Ottaviani, New lower bounds for the border rank of matrix multiplication, Theory Comput. 11 (2015), 285 -- 298. MR 3376667 [Mac92] Macaulay2, Macaulay2 a software system devoted to supporting research in algebraic geometry and commu- [Mas16] [MM13] tative algebra, http://www.math.uiuc.edu/Macaulay2/, 1992. A. Massarenti, Generalized varieties of sums of powers, Bull. Braz. Math. Soc. (N.S.) 47 (2016), no. 3, 911 -- 934. MR 3549076 A. Massarenti and M. Mella, Birational aspects of the geometry of varieties of sums of powers, Adv. Math. 243 (2013), 187 -- 202. MR 3062744 [MMS18] A. Massarenti, M. Mella, and G. Staglianò, Effective identifiability criteria for tensors and polynomials, J. [MR13] [MR14] [RS00] [Sch81] [Sei18] [Shi17a] [Shi17b] [Str73] [SU00] [Tei15] [TZ11] , Corrigendum to "The rank of nn matrix multiplication is at least 3n Symbolic Comput. 87 (2018), 227 -- 237. MR 3744347 A. Massarenti and E. Raviolo, The rank of n n matrix multiplication is at least 3n Algebra Appl. 438 (2013), no. 11, 4500 -- 4509. MR 3034546 ´2?2n ´ 2?2n 2 ´3n" [Linear Algebra Appl. 438 (11) (2013) 4500 -- 4509] [mr3034546], Linear Algebra Appl. 445 (2014), 369 -- 371. MR 3151280 K. Ranestad and F-O. Schreyer, Varieties of sums of powers, J. Reine Angew. Math. 525 (2000), 147 -- 181. MR 1780430 A. Schönhage, Partial and total matrix multiplication, SIAM J. Comput. 10 (1981), no. 3, 434 -- 455. MR 623057 A. Seigal, Ranks and symmetric ranks of cubic surfaces, https://arxiv.org/abs/1801.05377, 2018. Y. Shitov, A counterexample to Comon's conjecture, https://arxiv.org/abs/1705.08740, 2017. 3 2 ´ 3n, Linear 2 3 2 , A counterexample to Strassen's direct sum conjecture, https://arxiv.org/abs/1712.08660, 2017. V. Strassen, Vermeidung von Divisionen, J. Reine Angew. Math. 264 (1973), 184 -- 202. MR 0521168 A. Simis and B. Ulrich, On the ideal of an embedded join, J. Algebra 226 (2000), no. 1, 1 -- 14. MR 1749874 Z. Teitler, Sufficient conditions for Strassen's additivity conjecture, Illinois J. Math. 59 (2015), no. 4, 1071 -- 1085. MR 3628301 H. Takagi and F. Zucconi, Spin curves and Scorza quartics, Math. Ann. 349 (2011), no. 3, 623 -- 645. MR 2755000 [ZHQ16] X. Zhang, Z-H. Huang, and L. Qi, Comon's conjecture, rank decomposition, and symmetric rank decomposition of symmetric tensors, SIAM J. Matrix Anal. Appl. 37 (2016), no. 4, 1719 -- 1728. MR 3574587 12 ALEX CASAROTTI, ALEX MASSARENTI, AND MASSIMILIANO MELLA Dipartimento di Matematica e Informatica, Università di Ferrara, Via Machiavelli 35, 44121 Ferrara, Italy E-mail address: [email protected] Universidade Federal Fluminense, Rua Alexandre Moura 8 - São Domingos, 24210-200 Niterói, Rio de Janeiro, Brazil E-mail address: [email protected] Dipartimento di Matematica e Informatica, Università di Ferrara, Via Machiavelli 35, 44121 Ferrara, Italy E-mail address: [email protected]
1507.03294	4	1507	2016-01-29T08:31:22	On generic vanishing for pluricanonical bundles	[ "math.AG" ]	We study cohomology support loci and higher direct images of (log) pluricanonical bundles of smooth projective varieties or log canonical pairs. We prove that the 0-th cohomology support loci of log pluricanonical bundles are finite unions of torsion translates of subtori. We also construct an example of morphism from a smooth projective variety to an abelian variety such that a higher direct image of a pluricanonical bundle to the abelian variety is not a GV-sheaf.	math.AG	math	ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES TAKAHIRO SHIBATA Abstract. We study cohomology support loci and higher direct images of (log) pluricanonical bundles of smooth projective va- rieties or log canonical pairs. We prove that the 0-th cohomol- ogy support loci of log pluricanonical bundles are finite unions of torsion translates of subtori, and we give a generalization of the generic vanishing theorem to log canonical pairs. We also construct an example of morphism from a smooth projective variety to an abelian variety such that a higher direct image of a pluricanonical bundle to the abelian variety is not a GV-sheaf. Contents 1. Introduction 2. Preliminaries 3. Cohomology support loci of pluricanonical bundles 4. Higher direct images of pluricanonical bundles References 1 4 5 13 17 1. Introduction Thoughout this paper, we always assume that all varieties are defined over the complex number field. Let X be a smooth projective variety and ∆ be a simple normal crossing divisor on X. In this paper, we prove some results of generic vanishing theory for KX + ∆ or, more generally, m(KX + ∆) for any positive integer m. In Section 3, we prove some results about the structure of coho- mology support loci for a log canonical pair (X, ∆) (for the definition of cohomology support loci, see 2.1). Originally, Simpson proved the following theorem in [Sim]: 2010 Mathematics Subject Classification. Primary 14F17, Secondary 14E30. Key words and phrases. generic vanishing, pluricanonical bundles, log canonical, cohomology support loci, GV-sheaves. 1 2 TAKAHIRO SHIBATA Theorem 1.1 (Simpson). Let X be a smooth projective variety. Then the cohomology support locus Si j(KX) = {ξ ∈ Pic0(X) hi(X, OX(KX ) ⊗ ξ) ≥ j} is a finite union of torsion translates of abelian subvarieties of Pic0(X) for any i ≥ 0 and j ≥ 1. In [ClHa], a generalization of Theorem 1.1 for Kawamata log termi- nal pairs is discussed (see [ClHa, Theorem 8.3]). We will generalize Theorem 1.1 for log canonical pairs: Theorem 1.2 (= Theorem 3.5). Let X be a smooth projective variety, ∆ be a boundary Q-divisor on X, that is, a Q-divisor whose coefficients are in [0, 1], with simple normal crossing support, f : X → A be a morphism to an abelian variety, and D be a Cartier divisor on X such that D ∼Q KX + ∆. Then the cohomology support locus j(D, f ) = {ξ ∈ Pic0(A) hi(X, OX (D) ⊗ f ∗ξ) ≥ j} Si is a finite union of torsion translates of abelian subvarieties of Pic0(A) for any i ≥ 0 and j ≥ 1. We will reduce Theorem 1.2 to the special case when ∆ is a (reduced) simple normal crossing divisor. This case was treated by Kawamata [Kaw] and was used in [CKP] for the proof of Corollary 1.4 below. In [Bud], Budur proved a result for the cohomology support loci of unitary local systems, which includes it as a special case (see Theorem 3.3 and Theorem 3.4). We will use Budur's result for the proof of Theorem 1.2. We can also prove a similar result for pluricanonical divisors: Theorem 1.3 (= Theorem 3.9). Let X be a normal projective variety, ∆ be a boundary Q-divisor on X such that KX + ∆ is Q-Cartier and (X, ∆) is log canonical, f : X → A be a morphism to an abelian variety, and D be a Cartier divisor on X such that D ∼Q m(KX + ∆) for some positive integer m. Then the cohomology support locus S0 j (D, f ) = {ξ ∈ Pic0(A) h0(X, OX (D) ⊗ f ∗ξ) ≥ j} is a finite union of torsion translates of abelian subvarieties of Pic0(A) for any j ≥ 1. Theorem 1.3 is one of the main results of this paper. Note that Theorem 1.3 states the structure of cohomology support loci of only 0- th cohomology and that (X, ∆) is an arbitrary log canonical pair. The special case when ∆ = 0 was proved by Chen and Hacon (see [ChHa, Theorem 3.2]). As a corollary, Theorem 1.3 implies the following result by Campana, Koziarz and Paun: ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 3 Corollary 1.4 ([CKP, Theorem 0.1]). Let X be a normal projective variety, ∆ be a boundary Q-divisor on X such that KX +∆ is Q-Cartier and (X, ∆) is log canonical. Assume that there exists a numerically trivial Cartier divisor ρ on X such that H 0(X, m(KX + ∆) + ρ) 6= 0 for some positive integer m. Then h0(X, m′(KX + ∆)) ≥ h0(X, m(KX + ∆) + ρ) > 0 for some suitable multiple m′ of m. Therefore, we can see Theorem 1.3 as a generalization of [CKP, The- orem 0.1]. We will also give an alternative proof of [CKP, Corollary 3.2] as an application of Theorem 1.3 (see Proposition 3.10 below). Section 4 treats the higher direct images of pluricanonical bundles to abelian varieties. In this direction, Hacon first proved the following result in [Hac]: Theorem 1.5 (Hacon). Let X be a smooth projective variety and f : X → A be a morphism to an abelian variety. Then the higher direct images Rjf∗ωX are GV-sheaves for any j. In the notations in Theorem 1.5, a coherent sheaf F on A is called a GV-sheaf on A if the cohomology support loci Si(F ) = {ξ ∈ Pic0(A) hi(A, F ⊗ ξ) 6= 0} satisfies codimPic0(A) Si(F ) ≥ i for every i > 0 (see also 2.3). Theo- rem 1.5 recovers the original generic vanishing theorem of Green and Lazarsfeld in [GrLa]. We will give some generalizations of Theorem 1.5 and the generic vanishing theorem of Green and Lazarsfeld to log canonical pairs. For the precise statements, see Proposition 4.2 and Theorem 3.7, respectively. In [PoSc], Popa and Schnell gave a similar result for pluricanonical divisors: Theorem 1.6 ([PoSc, Theorem 1.10]). Let X be a smooth projective variety and f : X → A be a morphism to an abelian variety. Then the direct image f∗(ω⊗m X ) is a GV-sheaf for any positive integer m. Theorem 1.6 is a consequence of [PoSc, Theorem 1.7]. For the proof of Theorem 1.6, the following special case of [PoSc, Theorem 1.7] is sufficient: Theorem 1.7 ([PoSc, Corollary 2.9]). Let X be a smooth projective variety and f : X → Y be a morphism to a projective variety Y of dimension n. If L is an ample and globelly generated line bundle on Y and m is a positive integer, then H i(Y, f∗ω⊗m for all i > 0 and l ≥ m(n + 1) − n. X ⊗ L⊗l) = 0 4 TAKAHIRO SHIBATA However, their proof of [PoSc, Theorem 1.7] does not work for the X . They posed the following question in higher direct images Rjf∗ω⊗m [PoSc, p. 2293]: Question 1.8. Let X be a smooth projective variety and f : X → A be a morphism to an abelian variety. Then is Rjf∗(ω⊗m X ) a GV-sheaf for any j ≥ 1 and m ≥ 2? We will give an answer of the question by constructing an example that Rjf∗(ω⊗m X ) is not a GV-sheaf for some j ≥ 1 and m ≥ 2 (see Example 4.5). The existence of such an example implies that, in the notations of Theorem 1.7, we can not obtain any effective lower bounds for positive integers l which satisfy that H i(Y, Rjf∗ω⊗m X ⊗ L⊗l) = 0 for any i > 0 (for details, see Remark 4.7). Acknowledgments. I would like to express my gratitude to Professor Osamu Fujino for his much useful advice and generous support. I would also like to thank my colleagues for discussions and the referee for many valuable comments. 2. Preliminaries In this section, we collect some basic definitions. 2.1 (Cohomology support loci). Let X be a smooth projective variety, f : X → A be a morphism to an abelian variety and F be a coherent j(F , f ) = {ξ ∈ Pic0(A) hi(X, F ⊗ f ∗ξ) ≥ j}. We sheaf on X. Set Si call Si j(F , f ) a cohomology support locus of F . This is a Zariski closed subset of the abelian variety Pic0(A). If f is the Albanese morphism, 1(F , f ) by Si(F , f ). we denote Si j(F ). We simply denote Si j(F , f ) by Si 2.2 (Fourier -- Mukai transforms). Let A be an abelian variety and A be the dual abelian variety of A. We define a functor ΦPA : Coh(A) → Coh( A) as ΦPA(F ) = p2∗(p∗ 1F ⊗ PA), where PA is the (normalized) Poincar´e bundle on A × A and p1 : A × A → A, p2 : A × A → A are the natural projections. The right derived functor RΦPA of ΦPA is called the Fourier -- Mukai transform. Let X be a smooth projective variety. Given a morphism f : X → A, we put PX = (f × id A)∗PA. We define a functor ΦPX : Coh(X) → Coh( A) as ΦPX (F ) = π2∗(π∗ 1F ⊗ PX ), where π1 : X × A → X and π2 : X × A → A are the natural projections. ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 5 2.3 (GV-sheaves). Let X be a smooth projective variety, f : X → A be a morphism to an abelian variety and F be a coherent sheaf on X. We say that F is a GV-sheaf with respect to the morphism f if codimPic0(A) Si(F , f ) ≥ i for every i > 0. This condition is equivalent to the condition that codimPic0(A) SuppRiΦPX (F ) ≥ i for every i > 0. If f = idA, then we simply call such a sheaf a GV-sheaf on A. For details, see [PaPo]. 2.4 (Q-divisors). Let X be a normal variety. A Q-divisor on X is a formal sum of finitely many prime divisors on X with rational coeffi- cients. a normal variety X, where dj are rational numbers and Dj are prime 2.5 (Operations on Q-divisors). Let D = P djDj be a Q-divisor on divisors on X. We define the round-down ⌊D⌋ of D as ⌊D⌋ =P⌊dj⌋Dj, where ⌊dj⌋ is the largest integer among all the integers that are not larger than dj. The fractional part {D} of D is defined as {D} = D − ⌊D⌋. Finally, we define D+ = Xdj >0 djDj, D− = −Xdj <0 Then we have D = D+ − D−. djDj, D=1 = Xdj =1 Dj. 2.6 (Unitary local systems). Let X be a complex manifold. A local system (of rank one) on X is a locally constant sheaf of one-dimensional C-vector spaces on X. Then there is a natural correspondence between local systems on X and characters of the fundamental group π1(X) (a character of π1(X) is a group homomorphism ρ : π1(X) → C∗). A unitary local system on X is a local system that corresponds to a character whose image is contained in U(1). We denote the set of all unitary local systems on X by UB(X). Let Picτ (X) be the subgroup of Pic(X) consisting of line bundles whose first Chern classes are torsion elements. Take L ∈ Picτ (X). L has the Chern connection ∇ : L → Ω1 X ⊗ L and Ker ∇ is a unitary local system on X. The correspondence L 7→ Ker ∇ gives an isomorphism Picτ (X) ∼= UB(X). For details, see [Kob, Chapter 1, Section 2 and Section 4]. 3. Cohomology support loci of pluricanonical bundles First, we refer to Timmerscheidt's mixed Hodge theory for unitary local systems (see [Tim]). Let X be a smooth projective variety, ∆ be a simple normal crossing divisor on X and U = X \ ∆. Take V ∈ UB(U). Then we have Deligne's canonical extension (L, ∇) of V, where L is a 6 TAKAHIRO SHIBATA holomorphic line bundle on X and ∇ is a logarithmic connection on L with poles along ∆ such that Ker ∇U = V (for details, see [EsVi, (1.3)] and [Bud, p. 226]). We have the logarithmic de Rham complex 0 → L ∇ → Ω1 X(log ∆) ⊗ L ∇ → Ω2 X (log ∆) ⊗ L ∇ → · · · . We give to the complex (Ω• X (log ∆) ⊗ L, ∇) the following filtration F : F p(Ωq Ωp X(log ∆) ⊗ L) =(0 for q < p, X (log ∆) ⊗ L for q ≥ p. Let ι : U → X be the inclusion. Rι∗V is quasi-isomorphic to Ω• X (log ∆)⊗ L (see [EsVi, (1.4)]). Therefore the filtration F induces a spectral se- quence Ep,q 1 = H q(X, Ωp X (log ∆)⊗L) = H p+q(U, V). In general, we can not expect a natural mixed Hodge structure on H p+q(U, V) when V is not defined over R. However, Timmerscheidt proved: X(log ∆)⊗L) ⇒ Hp+q(X, Ω• Theorem 3.1 ([Tim], Theorem (7.1)). In the above notations, the spectral sequence degenerates at E1. Remark 3.2. Theorem 3.1 holds for unitary local systems of arbitrary rank. Next, we refer to a result of Budur. The following statement is a special case of [Bud, Theorem 8.3]: Theorem 3.3 ([Bud]). Let X be a smooth projective variety, ∆ be a simple normal crossing divisor on X and U = X \ ∆. Then the image of the set {V ∈ UB(X) dim Grp F H p+q(U, VU ) ≥ j} in Picτ (X) is a finite union of torsion translates of abelian subvarieties for every p, q and j. Take a unitary local system V ∈ UB(X) on X and let L ∈ Picτ (X) be the corresponding line bundle. Then the canonical extension of VU is L with the Chern connection. So by Theorem 3.1, we can restate Theorem 3.3 as follows: Theorem 3.4. Let X be a smooth projective variety, ∆ be a simple normal crossing divisor on X and U = X \ ∆. Then the set {L ∈ Picτ (X) dim H q(X, Ωp X (log ∆) ⊗ L) ≥ j} is a finite union of torsion translates of abelian subvarieties for every p, q and j. ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 7 By using Theorem 3.4, we can prove Theorem 1.2 in Section 1: Theorem 3.5 (= Theorem 1.2). Let X be a smooth projective variety, ∆ be a boundary Q-divisor on X with simple normal crossing support, f : X → A be a morphism to an abelian variety and D be a Cartier divisor on X such that D ∼Q KX + ∆. Then Si j(D, f ) is a finite union of torsion translates of abelian subvarieties of Pic0(A) for every i ≥ 0 and j ≥ 1. Proof. We divide the proof into two steps. Step 1. We reduce to the case when ∆ is a (reduced ) simple normal crossing divisor. Set C = D − KX − ⌊∆⌋. Then C ∼Q {∆}, so we can take a positive integer N such that NC ∼ N{∆}. Let π : Y → X be the normalization of the cyclic cover Then it follows that OX (−kC). Spec N −1 Mk=0 N −1 π∗OY = OX(−kC + ⌊k{∆}⌋) Mk=0 N −1 Mk=0 (see [Kol, Proposition 9.8]). So π∗OY (−π∗⌊∆⌋) = OX(−kC + ⌊k{∆}⌋ − ⌊∆⌋) contains OX (−C − ⌊∆⌋) = OX (−(D − KX)) as a direct summand. Then by the standard argument (cf. the proof of [Bud, Theorem 6.1]) and Serre duality, it follows that Si j(D, f ) is a finite union of torsion translates of abelian subvarieties for every i and j if Si j(−π∗⌊∆⌋, f ◦ π) is a finite union of torsion translates of abelian subvarieties for every i and j. So we consider (Y, π∗⌊∆⌋). Note that π∗⌊∆⌋ is a reduced divisor. Take a divisor ∆Y on Y as KY + ∆Y = π∗(KX + ∆red). Then, by Hur- witz's formula, we see that ∆Y is reduced and π∗⌊∆⌋ ≤ ∆Y . Moreover, (Y, ∆Y ) is a log canonical pair since (X, ∆red) is log canonical and π is finite (see [KoMo, Proposition 5.20]). These arguments imply that (Y, π∗⌊∆⌋) is a log canonical pair. Take a log resolution µ : Y ′ → Y of (Y, π∗⌊∆⌋). Set ∆Y ′ = µ∗(KY + π∗⌊∆⌋) − KY ′, 8 then TAKAHIRO SHIBATA ∆=1 Y ′ = µ−1 ∗ (π∗⌊∆⌋) + E for some reduced µ-exceptional divisor E. Since −KY ′/Y = −(KY ′ − µ∗KY ) has no irreducible components with coefficient 1, every compo- nent of E is contained in µ∗π∗⌊∆⌋. Since E is µ-exceptional, E is in fact contained in µ∗π∗⌊∆⌋ − µ−1 ∗ π∗⌊∆⌋ − E is an effective and µ-exceptional divisor on Y ′. By the Fujino -- Kov´acs vanishing theorem (see [Kov] and [Fuj2]), Riµ∗OY ′(−∆=1 ∗ π∗⌊∆⌋. So F = µ∗π∗⌊∆⌋ − µ−1 Y ′ ) = 0 for i > 0. Therefore Rµ∗OY ′(−∆=1 Y ′ ) ∼= µ∗OY ′(−∆=1 Y ′ ) ∼= µ∗OY ′(−µ−1 ∗ π∗⌊∆⌋ − E) ∼= µ∗OY ′(−µ∗π∗⌊∆⌋ + F ) ∼= OY (−π∗⌊∆⌋). Thus we have Si j(−π∗⌊∆⌋, f ◦ π). Moreover, ∆=1 Y ′ is a simple normal crossing divisor on Y ′. So it is sufficient to give a proof in the case when ∆ is a simple normal crossing divisor. Y ′ , f ◦ π ◦ µ) = Si j(−∆=1 Step 2. Assume that ∆ is a simple normal crossing divisor. Put U = X \ ∆. By Theorem 3.4, the set {L ∈ Picτ (X) hq(X, Ωp X (log ∆) ⊗ L) ≥ j} j(KX + ∆) = Si is a finite union of torsion translates of abelian subvarieties for any p, q and j. In particular, Si j(KX + ∆, albX) is a finite union of torsion translates of abelian subvarieties for any i and j, where albX denotes the Albanese morphism of X. Considering the universality of the Albanese morphism, it follows from this that Si j(KX + ∆, f ) is also a finite union of torsion translates of abelian subvarieties. (cid:3) Remark 3.6. Note that the pair (X, ∆) in the statement of Theorem 3.5 is not an arbitrary log canonical pair. The technical reason why Theorem 3.5 for arbitrary log canonical pairs is not made is that the arguments for unitary local systems require log smoothness of (X, ∆) (i.e. the condition that X is smooth and ∆ has simple normal crossing support), and the statement for arbitrary log canonical pairs cannot easily be reduced to the log smooth case. On the other hand, Theorem 3.9 below holds for arbitrary log canonical pairs. We can also prove a generalization of the generic vanishing theorem for log canonical pairs. ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 9 Theorem 3.7. Let X be a smooth projective variety, ∆ be a boundary Q-divisor on X with simple normal crossing support, f : X → A be a morphism to an abelian variety and D be a Cartier divisor on X such that D ∼Q KX + ∆. Set l = max{dim V − dim f (V ) V = X or V is a log canonical center of (X, ∆)}. Then codimPic0(A) Si(D, f ) ≥ i − l for any i. For the definition of log canonical centers, see [Fuj1, Definition 4.6]. Proof. Let S = ⌊∆⌋ and ∆i be an irreducible component of S. Con- sider the exact sequence · · · → Rjf∗OX(D − ∆i) → Rjf∗OX (D) → Rjf∗O∆i(D∆i) → · · · . Then it follows that Rjf∗OX(D) = 0 for j > l by induction of both the dimension of X and the number of irreducible components of S. Consider the Leray spectral sequence Ep,q 2 (ξ) = H p(A, Rqf∗OX (D) ⊗ ξ) ⇒ H p+q(X, OX(D) ⊗ f ∗ξ), where ξ ∈ Pic0(A). Then it follows by the spectral sequence that Si(D, f ) ⊂ l [q=0 Si−q(Rqf∗OX(D)). Furthermore, Rqf∗OX (D) are GV-sheaves on A for all q (see Proposi- tion 4.2 below), so codim Si−q(Rqf∗OX (D)) ≥ i − q for 0 ≤ i ≤ l. Hence codim Si(D, f ) ≥ i − l. (cid:3) Remark 3.8. In the notations of Theorem 3.7, assume in addition that (X, ∆) is a Kawamata log terminal pair. Then l = dim X − dim f (X). In particular, assume ∆ = 0, then Theorem 3.7 is the original generic vanishing theorem of Green and Lazarsfeld in [GrLa]. Next, we consider the cohomology support loci of m(KX + ∆). Theorem 3.9 (= Theorem 1.3). Let X be a normal projective variety, ∆ be a boundary Q-divisor on X such that KX + ∆ is Q-Cartier and (X, ∆) is log canonical, f : X → A be a morphism to an abelian variety, and D be a Cartier divisor on X such that D ∼Q m(KX + ∆) for some positive integer m. Then S0 j (D, f ) is a finite union of torsion translates of abelian subvarieties of Pic0(A) for any j ≥ 1. 10 TAKAHIRO SHIBATA j (D, f ). This is because, once we show that S0 Proof. Fix an element ξ ∈ S0 j (D, f ). It is sufficient to show that there is a torsion translate of an abelian subvariety T ⊂ Pic0(A) such that ξ ∈ T ⊂ S0 j (D, f ) is a union of (possibly infinitely many) torsion translates of abelian subvarieties of Pic0(A), then it follows that S0 j (D, f ) is a union of finitely many of them since S0 j (D, f ) and all torsion translates of abelian subvarieties are Zariski closed subsets of Pic0(A), and there are only countably many of torsion translates of abelian subvarieties. Step 1. Take a projective birational morphism µ : X ′ → X satisfying the following conditions: (i) X ′ is a smooth projective variety; (ii) KX ′ + ∆′ = µ(KX + ∆) + E, where ∆′ is a boundary Q-divisor on X ′ and E is an effective and µ-exceptional Cartier divisor on X ′; (iii) µ∗D + f ∗ξ = V + F , where V is a free linear system and F is an effective Cartier divisor on X ′; (iv) Exc(µ) ∪ Supp∆′ ∪ SuppF is a simple normal crossing divisor on X ′. D′ = µ∗D + mE is a Cartier divisor on X ′ satisfying that D′ ∼Q mµ∗(KX + ∆) + mE = m(KX ′ + ∆′). For any η ∈ Pic0(A), h0(X ′, OX ′(D′ + µ∗f ∗η)) = h0(X, OX (D + f ∗η)). Therefore S0 j (D, f ). Furthermore, j (D′, f ◦ µ) = S0 D′ + µ∗f ∗ξ = µ∗(D + f ∗ξ) + mE = µ∗D + f ∗ξ + mE = V + F + mE. By replacing (X, ∆, D) with (X ′, ∆′, D′), we can assume that the fol- lowing conditions hold: X is smooth, D + f ∗ξ = V + F for some free linear system V and some effective Cartier divisor F , and Supp∆ ∪ SuppF is a simple normal crossing divisor. Step 2. Since ξ ∈ S0 j (D, f ), dim V ≥ j − 1. Take a general smooth divisor M ∈ V with no irreducible component contained in Supp∆ ∪ SuppF . We can take ξ0 ∈ Pic0(A) such that ξ = mξ0 in Pic0(A). Then m(KX + ∆ + f ∗ξ0) = m(KX + ∆) + f ∗ξ ∼Q D + f ∗ξ ∼ M + F. Therefore KX + ∆ + f ∗ξ0 ∼Q 1 m (M + F ). ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 11 Then we have D − F + (m − 1)f ∗ξ0 ∼Q m(KX + ∆) − F + (m − 1)f ∗ξ0 = KX + ∆ − F + (m − 1)(KX + ∆ + f ∗ξ0) ∼Q KX + ∆ − 1 m F + m − 1 m M. Set and Then and G =&(cid:18)∆ − 1 m F(cid:19)−' ≥ 0 D0 = D − F + (m − 1)f ∗ξ0 + G. D0 ∼Q KX + ∆ − 1 m F + m − 1 m − 1 m M + G, M + G m ∆ − F + 1 m is a boundary Q-divisor with simple normal crossing support. By The- orem 3.5, S0 j (D0, f ) is a finite union of torsion translates of abelian subvarieties. Moreover, h0(X, OX (D0 + f ∗ξ0)) = h0(X, OX (D − F + f ∗ξ + G)) ≥ h0(X, OX(D − F + f ∗ξ)) = dim V + 1 = j. Therefore we have ξ0 ∈ S0 j (D0, f ). Step 3. Here we will prove that S0 j (D0, f ) + (m − 1)ξ0 ⊂ S0 j (D, f ). Take α ∈ S0 j (D0, f ). Then D +f ∗(α+(m−1)ξ0) = D0 +F −G+f ∗α. F − G is obviously effective since ∆ is effective. So we have h0(X, OX(D0 + F − G + f ∗α)) ≥ h0(X, OX(D0 + f ∗α)) ≥ j. Thus α + (m − 1)ξ0 ∈ S0 j (D, f ), so it follows that S0 j (D0, f ) + (m − 1)ξ0 ⊂ S0 j (D, f ). Step 4. By Step 2, we can take an abelian subvariety B ⊂ Pic0(A) and a torsion point q ∈ Pic0(A) such that ξ0 ∈ B + q ⊂ S0 j (D0, f ). By Step 3, B +q +(m−1)ξ0 ⊂ S0 j (D, f ). Represent ξ0 ∈ B +q as ξ0 = t+q for some t ∈ B, then B + q + (m − 1)ξ0 = B + mq + (m − 1)t = B + mq. Therefore ξ = ξ0 + (m − 1)ξ0 ∈ B + q + (m − 1)ξ0 = B + mq ⊂ S0 j (D, f ). (cid:3) 12 TAKAHIRO SHIBATA Proof of Corollary 1.4. By Theorem 3.9, we may assume that ρ is a torsion element. We take a positive integer N such that N ·ρ = 0. Then h0(Nm(KX + ∆)) = h0(N(m(KX + ∆) + ρ)) ≥ h0(m(KX + ∆) + ρ). (cid:3) As another application of Theorem 3.9, we give an alternative proof of the following proposition in [CKP]: Proposition 3.10 ([CKP, Corollary 3.2]). Let X be a normal projec- tive variety, ∆ be a boundary Q-divisor on X such that KX + ∆ is Q-Cartier and (X, ∆) is log canonical, and f : X → A be a morphism to an abelian variety. Then κ(KX + ∆) ≥ κ(KX + ∆ + f ∗ξ) for every ξ ∈ Pic0(A). In [CKP], this is proved as a corollary of [CKP, Theorem 0.1] (see Corollary 1.4). First we prove the following lemma: Lemma 3.11. Let A be an abelian variety and T be the set of all torsion translates of abelian subvarieties of A. For T ∈ T , set ord(T ) = min{ord(t) t ∈ Ttor}, where Ttor denotes the set of all torsion points in T . (i) ord(mT ) ≤ ord(T ) ≤ m · ord(mT ) for any T ∈ T and m ∈ Z>0. (ii) For any p ∈ A, the set {ord(T ) T ∈ T and mp ∈ T for some m ∈ Z>0} is bounded. Proof. (i) By definition, ord(t0) = ord(T ) for some t0 ∈ T . Then ord(mT ) ≤ ord(mt0) ≤ ord(t0) = ord(T ). Take t ∈ Ttor. Then ord(T ) ≤ ord(t) ≤ m ord(mt) since m ord(mt) · t = ord(mt) · mt = 0. So ord(T ) ≤ m ord(mT ). (ii) Suppose that {ord(T ) T ∈ T and mp ∈ T for some m ∈ Z>0} is unbounded. Consider a sequence Sk = (m1, · · · , mk; T1, · · · , Tk) sat- isfying that • m1, · · · , mk ∈ Z>0, T1, · · · , Tk ∈ T , • m1p ∈ T1, m1m2p ∈ T2, · · · , m1 · · · mkp ∈ Tk, • dim T1 > dim T2 > · · · > dim Tk. By assumption, we can take mk+1 ∈ Z>0 and T ′′ mk+1p ∈ T ′′ T as T ′ k+1) > m1 · · · mk ord(Tk). We define T ′ k+1 and T ′ k = mk+1Tk. Then we have k+1 ∈ T such that k ∈ k+1, T ′ k+1 and ord(T ′′ k+1 = m1 · · · mkT ′′ ord(T ′′ k+1) m1 · · · mk k+1) ≥ ord(T ′ > m1 · · · mk ord(Tk) m1 · · · mk = ord(Tk) ≥ ord(T ′ k), ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 13 using (i). Put Tk+1 = T ′ Since ord(T ′ k), T ′ Therefore dim Tk+1 < dim T ′ k+1) > ord(T ′ k ∈ T . Then m1 · · · mk+1p ∈ Tk+1. k+1. Then it follows that Tk+1 ( T ′ k. k+1 ∩ T ′ k 6⊂ T ′ k = dim Tk. By continuing this process, we have S1, S2, · · · , Sk, Sk+1, · · · . But this process must stop because of the third condition of Sk. So we have a contradiction. (cid:3) Proof of Proposition 3.10. Let T be the set of all torsion translates of abelian subvarieties of Pic0(A). There exists a positive integer m0 such that D = m0(KX + ∆) is Cartier. Then it reduces to show that κ(D) ≥ κ(D + f ∗ξ) for any ξ ∈ Pic0(A). Set d(m) = h0(mD) and d′(m) = h0(m(D + f ∗ξ)). Then mξ ∈ S0 d′(m)(mD, f ), so by Theorem 3.9 we can take Tm ∈ T such that mξ ∈ Tm ⊂ S0 d′(m)(mD, f ). Take tm ∈ Tm such that ord(tm) = ord(Tm). It follows by Lemma 3.11 (ii) that there exists a positive integer N satisfying that ord(Tm) N for any m ∈ Z>0. Hence Ntm = 0 for any m ∈ Z>0. Then we have Therefore d′(m) = h0(m(D + f ∗ξ)) ≤ h0(mD + f ∗tm) ≤ h0(N(mD + f ∗tm)) = h0(NmD) = d(Nm). κ(D + f ∗ξ) = lim sup m ≤ lim sup m log d′(m) log m log d(Nm) log m = lim sup m log d(Nm) log Nm = κ(D). (cid:3) 4. Higher direct images of pluricanonical bundles In this section, we discuss higher direct images of pluricanonical bun- dles to abelian varieties. In [PoSc], Popa and Schnell proved the following theorem: Theorem 4.1 ([PoSc, Theorem 1.10]). Let X be a smooth projective variety and f : X → A be a morphism to an abelian variety. Then the direct image f∗(ω⊗m X ) is a GV-sheaf for any positive integer m. In addition, we can prove a log version of Theorem 1.5: 14 TAKAHIRO SHIBATA Proposition 4.2. Let X be a smooth projective variety, ∆ be a bound- ary Q-divisor on X with simple normal crossing support, f : X → A be a morphism to an abelian variety and D be a Cartier divisor on X such that D ∼Q KX + ∆. Then the higher direct images Rjf∗ωX(∆) are GV-sheaves for any j. Here we use a characterization of GV-sheaves on abelian varieties by Hacon. He proved the following criterion in [Hac]: Theorem 4.3 (Hacon). Let A be an abelian variety and F be a coher- ent sheaf on A. F is a GV-sheaf on A if and only if H k(B, φ∗F ⊗L) = 0 for k > 0, where φ : B → A is any isogeny and L is any ample line bundle on B. Proof of Propotion 4.2. Let φ : B → A be an isogeny and L be an ample line bundle on B. Set Y = X ×A B. Let ψ : Y → X and g : Y → B be the natural morphisms. Fix an integer j ≥ 0. By base change, φ∗Rjf∗ωX(∆) ∼= Rjg∗ψ∗ωX(∆) ∼= Rjg∗ωY (ψ∗∆). Applying the Ambro -- Fujino vanishing theorem (see [Amb, Theorem 3.2] and [Fuj1, Theorem 6.3]) to the log canonical pair (Y, ψ∗∆), we have H k(B, φ∗Rjf∗ωX(∆) ⊗ L) ∼= H k(B, Rjg∗ωY (ψ∗∆) ⊗ L) = 0 for k > 0. By Theorem 4.3, it follows that Rjf∗ωX(∆) is a GV-sheaf. (cid:3) In [PoSc], Popa and Schnell also asked the question whether the higher direct images Rjf∗(ω⊗m X ) are GV-sheaves or not. Note that Rjf∗ωX is a GV-sheaf for every j (see Theorem 1.5 or Proposition 4.2). Hence the question is whether Rjf∗(ω⊗m X ) is a GV-sheaf or not in the case when j ≥ 1 and m ≥ 2. Here we give an example of a higher direct image of a pluricanonical bundle which is not a GV-sheaf. Lemma 4.4. There exists an irregular smooth projective variety with big anti-canonical bundle. Proof. Let A be an abelian variety. We take an ample line bundle L on A and define a vector bundle E as the direct sum of L−1 and OA. Let π : X = PA(E) → A be the projective bundle on A associated to E. Clearly the irregularity q(X) of X is positive. The canonical bundle ωX is isomorphic to π∗(ωA ⊗det E)⊗OX (− rank E) = π∗L−1 ⊗OX (−2) (see [Laz, 7.3.A]). ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 15 We will see that ω−1 X is big. Let ξ and l be the numerical classes of OX (1) and L, respectively. Note that ξ is an effective class since H 0(X, OX(1)) = H 0(A, E) 6= 0. The numerical class of ω−1 X is equal to 2ξ + π∗l = N − 1 N 2ξ + 1 N 2ξ + π∗l, where N is a sufficiently large integer such that (1/N)2ξ + π∗l is ample. So the numerical class of ω−1 X is represented by the sum of an effective class and an ample class. Therefore ω−1 (cid:3) X is big. Example 4.5. Let X be an irregular smooth projective variety of dimension n ≥ 2 with big anti-canonical bundle and let f = albX : X → A be the Albanese morphism of X. Now we show that Rjf∗ω⊗m is not a GV-sheaf for some positive integers j and m. Let PA be the Poincar´e bundle on A × A and PX = (f × id A)∗PA. Then we have the Fourier -- Mukai transforms ΦPA : Coh(A) → Coh( A) and ΦPX : Coh(X) → Coh( A) (see 2.2). It immediately follows that ΦPX ∼= ΦPA ◦ f∗. X Now we prove the following lemma: Lemma 4.6. Let X be a smooth projective variety of dimension n and D be a big Cartier divisor on X. Then S0(mD) = {ξ ∈ Pic0(X) H 0(X, OX (mD + ξ)) 6= 0} = Pic0(X) for any sufficiently large and divisible m. Proof. Since D is big, there exist a positive integer m0, a very ample Cartier divisor H, and an effective Cartier divisor E such that m0D ∼ H + E. For any positive integer m, we have S0(mm0D) = S0(mH + mE) ⊃ S0(mH). We can take a positive integer m1 satisfying that H i(X, OX (mH + ξ)) = 0 for every ξ ∈ Pic0(X), m ≥ m1 and i > 0 (take m1 such that m1H − KX is ample). According to the notion of the Castelnuovo -- Mumford regularity, mH + ξ is 0-regular for every ξ ∈ Pic0(X) and m ≥ m1 + n, In particular, S0(mH) = Pic0(X) and so it is globally generated. for every m ≥ m1 + n. Therefore S0(mm0D) = Pic0(X) for every m ≥ m1 + n. (cid:3) By the above lemma, we can take a positive integer m such that Sn(ω⊗m X ) = −S0(ω⊗(1−m) X ) = Pic0(X). Then it follows that SuppRnΦPX (ω⊗m X ) = Pic0(X). 16 TAKAHIRO SHIBATA Consider the Grothendieck spectral sequence Ei,j 2 = RiΦPARjf∗(ω⊗m X ) ⇒ Ri+jΦPX (ω⊗m X ). Then there exists an integer i such that SuppRiΦPARn−if∗(ω⊗m X ) = Pic0(X). Note that dim f (X) > 0 since X is irregular. Therefore Rnf∗(ω⊗m X ) = 0, and so i must be positive. Note that a coherent sheaf F on A is a GV-sheaf if and only if codim SuppRjΦPAF ≥ j for j > 0 (see 2.3). Hence it follows that Rn−if∗(ω⊗m X ) is not a GV-sheaf. Remark 4.7. Let X be a smooth projective variety of dimension k, f : X → Y be a morphism to a projective variety Y of dimension n, L be an ample and globally generated line bundle on Y and m > 0, j ≥ 0 be integers. Suppose that we can take a positive integer N = N(k, n, m, j) depending only on k, n, m and j such that H i(Y, Rjf∗ω⊗m X ⊗ L⊗l) = 0 for any i > 0 and l ≥ N. Then, by the same argument as the proof of Theorem 1.6 by using Theorem 1.7 (see the proof of [PoSc, Theorem 1.10]), it follows that Rjf∗ω⊗m X is a GV-sheaf for any morphism f : X → A to an abelian variety, j ≥ 0 and m ≥ 1. But this contradicts Example 4.5. So we can not take such an integer N. Next, we consider the case when X is a surface. Example 4.5 shows X ) is not always a GV-sheaf when κ(X) = −∞. On the that Rjf∗(ω⊗m other hand, the following proposition holds: Proposition 4.8. Let X be a smooth projective surface and f : X → A be a morphism to an abelian variety. Assume that κ(X) ≥ 0. Then Rjf∗(ω⊗m X ) is a GV-sheaf for any j and m. Proof. This is obvious when dim f (X) = 0. So we may assume that dim f (X) ≥ 1. Fix a positive integer m. f∗(ω⊗m X ) is a GV-sheaf by Theorem 4.1. R2f∗(ω⊗m X ) = 0 since all fibers of f have dimension less than 2. So it is sufficient to show that R1f∗(ω⊗m X ) is a GV-sheaf. Since κ(X) ≥ 0, we can take a series of contractions of (−1)-curves ε : X → X ′ such that the canonical bundle ωX ′ of X ′ is semi-ample. Then the exceptional curves of ε are also contracted by f . Hence we obtain a morphism f ′ : X ′ → A such that f ′ ◦ ε = f . Consider the spectral sequence Ei,j 2 = Rif ′ ∗Rjε∗(ω⊗m X ) ⇒ Ri+jf∗(ω⊗m X ). Then Ei,j 2 = 0 except that (i, j) = (0, 0), (1, 0), (0, 1). So we have R1f∗(ω⊗m X ) ⊕ f ′ ∗R1ε∗(ω⊗m X ) X ). ∗R1ε∗(ω⊗m X ) ∼= R1f ′ ∼= R1f ′ ∗ε∗(ω⊗m ∗(ω⊗m X ′ ) ⊕ f ′ ON GENERIC VANISHING FOR PLURICANONICAL BUNDLES 17 Take any ample line bundle L on A. We have H k(A, R1f ′ ∗(ω⊗m X ′ ) ⊗ L) = 0 for k > 0 by the Ambro -- Fujino vanishing theorem since ωX ′ is semi-ample (see [Amb, Theorem 3.2] and [Fuj1, Theorem 6.3]). It is obvious that H k(A, f ′ since dim SuppR1ε∗(ω⊗m X ) = 0. Thus it follows that ∗R1ε∗(ω⊗m X ) ⊗ L) = 0 for k > 0 H k(A, R1f∗(ω⊗m X ) ⊗ L) = 0 for k > 0. We show by using Theorem 4.3 that R1f∗(ω⊗m X ) is a GV-sheaf. Let φ : B → A be an isogeny and L be an ample line bundle on B. Set Y = X ×A B. Let ψ : Y → X and g : Y → B be the natural morphisms. Then ψ is ´etale and κ(Y ) ≥ 0. By base change, φ∗R1f∗(ω⊗m X ) ∼= R1g∗ψ∗(ω⊗m X ) ∼= R1g∗(ω⊗m Y ). Hence H k(B, φ∗R1f∗(ω⊗m X ) ⊗ L) ∼= H k(B, R1g∗(ω⊗m Y ) ⊗ L). Here H k(B, R1g∗(ω⊗m Therefore H k(B, φ∗R1f∗(ω⊗m Theorem 4.3 that R1f∗(ω⊗m ) ⊗ L) = 0 for k > 0 by the above argument. X ) ⊗ L) = 0 for k > 0. Then it follows by X ) is a GV-sheaf. (cid:3) Y Taking the above proposition into account, we modify the question of Popa -- Schnell as follows: Question 4.9. Let X be a smooth projective variety of dimension n ≥ 3 and f : X → A be a morphism to an abelian variety. Assume that κ(X) ≥ 0. Then is Rjf∗(ω⊗m X ) a GV-sheaf for any j and m? References [Amb] F. Ambro, Quasi-log varieties, Tr. Mat. Inst. Steklova 240 (2003), 220 -- 239. [Bud] N. Budur, Unitary local systems, multipier ideals, and polynomial period- icity of Hodge numbers, Adv. Math. 221 (2009), 217-250. [ChHa] J. A. Chen, C. D. Hacon, On the irregularity of image of Iitaka fibration, Comm. in Algebra 32 (2004), no. 1, 203 -- 215. [ClHa] H. Clemens, C. D. Hacon, Deformations of the trivial line bundle and vanishing theorems, Amer. J. Math. 124 (2002), no. 4, 769 -- 815. [CKP] F. Campana, V. Koziarz, M. Paun, Numerical character of the effectivity of adjoint line bundles, Ann. Inst. Fourier (Grenoble) 62 (2012), no. 1, 107 -- 119. [EsVi] H. Esnault, E. Viehweg, Logarithmic De Rham complexes and vanishing theorems, Invent. Math. 86 (1986), 161 -- 194. [Fuj1] O. Fujino, Fundamental theorems for the log minimal model program, Publ. Res. Inst. Math. Sci. 47:3 (2011), 727 -- 789. 18 TAKAHIRO SHIBATA [Fuj2] O. Fujino, A remark on Kov´acs's vanishing theorem, Kyoto J. Math. 52 (2012), no. 4, 829 -- 832. [GrLa] M. Green, R. Lazarsfeld, Deformation theory, generic vanishing theorems, and some conjectures of Enriques, Catanese and Beauville, Invent. Math. 90 (1987), 389 -- 407. C. D. Hacon, A derived category approach to generic vanishing, J. Reine Angew. Math. 575 (2004), 173 -- 187. [Hac] [Kaw] Y. Kawamata, On the abundance theorem in the case of numerical Kodaira [Kob] [Kol] [Kov] dimension zero, Amer. J. Math. 135 (2013), no.1, 115 -- 124. S. Kobayashi, Differential geometry of complex vector bundles, Publications of the Mathematical Society of Japan, 15. J. Koll´ar, Shafarevich maps and automorphic forms, Princeton Univ. Press, 1995. S. J. Kov´acs, Du Bois pairs and vanishing theorems, Kyoto J. Math. 51 (2011), no.1, 47 -- 69. [KoMo] J. Koll´ar, S. Mori, Birational geometry of algebraic varieties, Cambridge Univ. Press, 1998. R. Lazarsfeld, Positivity in Algebraic Geometry II, Springer, 2004. [Laz] [PaPo] G. Pareschi, M. Popa, GV-sheaves, Fourier -- Mukai transform, and generic vanishing, Amer. J. Math. 133 (2011), no. 1, 235 -- 271. [PoSc] M. Popa, C. Schnell, On direct images of pluricanonical bundles, Algebra [Sim] and Number Theory 8 (2014), no. 9, 2273 -- 2295. C. Simpson, Subspaces of moduli spaces of rank one local systems, Ann. Sci. E.N.S. (4) 26 (1993), no. 3, 361 -- 401. [Tim] K. Timmerscheidt, Mixed Hodge theory for unitary local systems, J. Reine. Angew. Math. 379 (1987), 152 -- 171. Department of Mathematics, Graduate School of Science, Kyoto University, Kyoto 606-8502, Japan E-mail address: [email protected]
1607.05984	5	1607	2019-07-13T14:23:59	The global derived period map	[ "math.AG" ]	Abstract. We develop the global period map in the context of derived geometry, generalising Griffiths' classical period map as well as the infinitesimal derived period map. We begin by constructing the derived period domain which classifies Hodge filtrations and enhances the classical period domain. We analyze the monodromy action. Then we associate to any polarized smooth projective map of derived stacks a canonical morphism of derived analytic stacks from the base into the quotient of the derived period domain by monodromy. We conclude the paper by discussing a few examples and a derived Torelli problem. In the appendix we describe how to present derived analytic Artin stacks as hypergroupoids, which may be of independent interest.	math.AG	math	THE GLOBAL DERIVED PERIOD MAP CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Abstract. We develop the global period map in the context of derived geometry, generalising Griffiths' classical period map as well as the infinitesimal derived period map. We begin by constructing the derived period domain which classifies Hodge filtrations and enhances the classical period domain. We analyze the monodromy action. Then we associate to any polarized smooth projective map of derived stacks a canonical morphism of derived analytic stacks from the base into the quotient of the derived period domain by monodromy. We conclude the paper by discussing a few examples and a derived Torelli problem. In the appendix we describe how to present derived analytic Artin stacks as hypergroupoids, which may be of independent interest. Contents Introduction 1. 1.1. Outline 1.2. Notation and conventions 1.3. Acknowledgements 2. Derived geometry 2.1. Derived algebraic geometry 2.2. Derived analytic geometry following Lurie and Porta 3. The derived period domain 3.1. The classical period domain 3.2. The derived flag variety 3.3. The derived period domain I: Algebraically 3.4. The tangent space 2 4 5 5 5 5 8 11 11 13 18 24 1991 Mathematics Subject Classification. 14A20, 32G20. Key words and phrases. Derived algebraic geometry, derived analytic geometry, moduli, Hodge structures. The second author was supported by EPSRC grant EP/N015452/1 for part of this work. 1 2 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN 3.5. The derived period domain II: Analytically 3.6. Monodromy and quotient 4. The period map 4.1. The classical period map 4.2. The local derived period map I: The de Rham complex 4.3. The local derived period map II: The Hodge filtration 4.4. The global derived period map 4.5. Comparison with the classical period map 4.6. Differential of the period map 4.7. Examples Appendix A. Presenting higher analytic stacks A.1. A.2. Hypergroupoids A.3. Analytification of hypergroupoids References Introduction 26 28 31 31 32 40 46 51 52 55 57 57 57 61 63 1. Introduction We construct a global derived period map, generalising Griffiths' period map [21] and the infinitesimal derived period map of deformation theories [13, 16, 17]. We hope that this map will be a useful tool in studying derived moduli of varieties. We work over C throughout this article. Given a polarized smooth projective map f : X → S between derived Artin stacks we construct a map P from the analytification S an of the base to a quotient of a derived analytic stack U which we call the derived period domain. The underived truncation of P induces a product of the usual period maps on coarse moduli spaces; in particular a closed point s ∈ S is sent to the Hodge filtration on the fibre Xs. Moreover, P extends the infinitesimal period map. Our first result constructs the target of the derived period map: Theorem 3.22. There is a derived period domain U which is a geometric derived analytic Artin stack that extends the classical period domain. THE GLOBAL DERIVED PERIOD MAP 3 We define U as an open subspace of the analytification of a derived stack Dn(V, Q) which classifies filtrations of a complex V that satisfy the Hodge-Riemann bilinear relations with respect to a shifted bilinear form Q. (We use analytification rather than a direct construction in analytic stacks because the theory of derived analytic stacks is not yet as developed as that for derived algebraic stacks.) The main ingredients are Lurie's derived analytification functor with its universal property [35, 36], Porta's further analysis of this functor including its extension to derived Artin stacks [44], and some explicit constructions of stacks building on work of Toën and Vezzosi [66]. In order to define maps into this stack we use the universal property for derived analytification of affine schemes together with a presentation of derived (analytic) Artin stacks as simplicial derived affine schemes, resp. derived Stein spaces. This is closely related to the theory presenting stacks as hypergroupoids that was developed by Pridham [51], and that we extend to analytic stacks in the appendix. We then take the quotient of U by an arithmetic group Γ containing the fundamental group of the underlying topological space of S an. We use Deligne's work on formality [10] to show that we need not look at the action of the full simplicial loop group of the base but only the fundamental group. We can now state our main theorem, which is proved using local computations in derived algebraic geometry, a derived trace map, classical Hodge theory and some homotopy theory: f : X → S be a any polarized smooth projective Theorem 4.18. Let map between derived Artin stacks where S is connected and of finite presentation. Then there is a derived period map P : S an → U/Γ of derived analytic stacks. It then follows from the construction that P enhances the classical period map and the infinitesimal derived period map. Remark 1.1. The study of generalised period maps goes back to work in the infinitesimal setting motivated by mathematical physics. In order to study mirror symmetry for a Calabi-Yau manifold X Barannikov and Kontsevich considered the extended formal moduli space M associated to the dg Lie algebra A 0,∗(X, ∧∗TX). They construct a generalized period map M → ⊕kHk(X, C)[dim X −k] and use it to give H∗(X, ∧∗TX) the structure of a formal Frobenius algebra which is related to the Gromov-Witten invariants of the mirror partner of X [2, 3]. See also [1]. 4 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN 1.1. Outline. We begin by recalling in Section 2 some basic notions of derived algebraic and derived analytic geometry, in particular the derived analytification functor connecting them, which was defined by Lurie and further described by Porta. In Section 3 we construct the derived period domain U. After recalling the classical period domain in Section 3.1 we review the derived flag variety in Section 3.2. This derived stack classifies filtrations of a complex V. For us V will be the cohomology of a fibre of a smooth projective morphism of derived stacks. In Section 3.3 we build a geometric derived stack Dn(V, Q) classifying filtrations on a complex V equipped with a bilinear form Q, satisfying the Hodge-Riemann orthogonality relation. This is a closure of the derived period domain. We compute the tangent space in Section 3.4. In Section 3.5 we construct the derived period domain U itself using the analytification of Dn(V, Q) and the characterization of open derived substacks in terms of open substacks of the underlying underived stack. We later use the universal property of analytification to construct maps into this space. We verify U is a derived enhancement of the usual period domain. In Section 3.6 we recall Deligne's result that the derived pushforward of the constant sheaf for a family of smooth projective varieties is formal. We deduce there is no higher monodromy acting on cohomology of the fibre. We then construct the quotient of the derived period domain by the action induced by the monodromy action on V. In Section 4 we construct the derived period map. The heart of the matter is in Sections 4.2 and 4.3, where we construct the derived period map locally on contractible derived Stein spaces. As in the underived case we push forward the relative cotangent complex; the stupid filtration gives the Hodge filtration and we obtain the polarization by extending the trace pairing to derived geometry. Using the universal property of analytification we construct a map to the derived period domain. In Section 4.4 we glue the local period maps on a simplicial resolutions of S and the construction is complete. The universal example is given by the moduli stack of smooth polarized schemes. We check that our map is an enhancement of the usual period map in Section 4.5. In Section 4.6 we compute the differential and show that the derived period map extends the infinitesimal derived period map. We briefly talk about examples in Section 4.7. In Appendix A we develop the theory of hypergroupoids in derived Stein spaces as a model for derived analytic Artin stacks. THE GLOBAL DERIVED PERIOD MAP 5 1.2. Notation and conventions. We work over the field of complex numbers C throughout. Tensor products are understood over C, respectively the constant sheaf C, unless otherwise indicated. We will use the following notation: sSet is the category of simplicial sets with the classical (Quillen) model structure; sCat is the category of categories enriched over simplicial sets; dgCat is the category of differential graded categories; sAlg is the category of simplicial commutative algebras (over C). We write Map(−, −) for the mapping space in a simplicial category or in a model category, in a dg category Hom(−, −) denotes the dg enriched hom space. If there is an underlying model category structure we write RHom(−, −) for the derived hom complex. We use cohomological grading throughout. Given a derived scheme or derived analytic space X over C we denote by t(X) the underlying topological space. For a (derived) scheme V this is V(C) with the classical topology. A constant simplicial object is denoted by c, i.e. (cA)n = A for all n. For a derived Stein space T we let uT denote the derived affine scheme R Spec(O alg(T )). We write ∗ for the point, i.e. the derived scheme Spec(C) and its analytification. 1.3. Acknowledgements. We would like to thank Jon Pridham and Ian Grojnowski for introducing the first, respectively second, author to this problem, and for many helpful discussions. We are grateful to Mauro Porta for generous explanations of his work on derived analytic geometry, particularly for providing the proof of Lemma 4.13, and for many insightful discussions about this paper. In particular he pointed out a crucial mistake in the first version of this paper. Discussions of different possible solutions led to the joint work [27]. We would like to thank Barbara Fantechi, Benjamin Hennion, Daniel Huybrechts, Peter Jørgensen and Bertrand Toën for useful conversations. We also thank Serguei Barannikov for pointing out interesting references and the anonymous referees for many very helpful comments and corrections. 2. Derived geometry 2.1. Derived algebraic geometry. We will be assuming some familiarity with derived algebraic geometry, which is algebraic geometry that is locally modelled on the model category of simplicial commutative algebras instead of the category of commutative algebras. 6 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN There is a vast literature on the subject, developed by Toën-Vezzosi, Lurie and Pridham among others. For an introduction see [61, 63]. Here we just mention a few reminders and conventions. We will be studying certain derived stacks. Just like a scheme can be represented by a set-valued sheaf on commutative algebras a derived stack can be represented by a simplicial-set-valued hypersheaf on simplicial commutative algebras. (We always take stack to mean higher stack.) In fact derived stacks can be described by a model structure on presheaves of simplicial sets on simplicial commutative algebras. A representable derived stack is also called a derived affine stack, and the image of a simplicial algebra A in derived stacks is denoted R Spec(A). We will call a derived stack Artin or simply geometric if there exist certain smooth covers. Here we use geometric to mean k-geometric for some k rather than 1-geometric. The precise definition is inductive on k, beginning with representable stacks, which are (−1)-geometric. See for example Definition 1.3.3.1 in [66]. Replacing smooth by étale covers one obtains the definition of a Deligne-Mumford stack (we will call them DM stacks for short). A derived stack that is a union of geometric ones is called locally geometric. It is often interesting and consequential to show that certain derived stacks arising as moduli functors, say, are geometric. Derived stacks come with simplicial algebras of functions O and a representable map f : X → Y between derived stacks is called strong if πi(OX) ≃ πi( f −1OY) ⊗π0( f −1OY ) π0(OX) for i > 0. One fact we will use frequently is that k-geometric derived stacks are stable under homotopy pullbacks in derived stacks, Corollary 1.3.3.5 in [66]. As we work over C the normalization functor N from simplicial commutative algebras to commutative dg-algebras concentrated in non- positive degrees is a Quillen equivalence and we will freely switch between the two viewpoints. One can restrict a derived stack to an underived stack along Alg → sAlg. We denote this operation by π0 (it is often written t0 in the literature). On the other hand one can truncate a stack to a functor into sets, by applying π0 object-wise. We denote this functor by π0. This construction has the universal property of the coarse moduli space of the stack. We will often consider the double truncation π0π0 and if π0π0(X ) = X for some scheme X we will say the derived stack X is an enhancement of X. (Here we deviate a little from the literature, where one typically considers THE GLOBAL DERIVED PERIOD MAP 7 enhancements of higher stacks, i.e. a derived stack X is an enhancement of the higher stack X if π0(X ) = X.) A derived affine schemes is almost of finite presentation if the functor it represents commutes with filtered (homotopy) colimits. A derived stack is locally almost of finite presentation if it has a cover consisting of derived affine schemes almost of finite presentation. It follows from Theorem 7.4.3.18 of [38] that a derived stack X with perfect cotangent complex and such that π0(X) is finitely presented is locally almost of finite presentation. We will be considering derived stacks taking values in simplicial sets. In moduli questions these often come from functors valued in (∞, 1)- categories, and several of our constructions are clearest in terms of categories rather than simplicial sets. Our preferred models for (∞, 1)- categories are dg-categories and simplicial categories. We use the right Quillen functors DK : dgCat → sCat obtained by composing truncation, the Dold-Kan construction and the forgetful functor (see [59]), to move from dg-categories to simplicial categories. (We will sometimes use the same name for associated sheaves of dg-categories and simplicial categories). To associate a simplicial set we take the maximal subgroupoid Int(C ) and then use the functor ¯W : sCat → sSet defined in Definition 1.6 of [50]. It is weakly equivalent to taking the diagonal of the nerve. The composition of ¯W ◦ Int ◦ DK is also weakly equivalent to the maximal subgroupoid of the dg nerve in Construction 1.3.1.6 of [38]. The reader is welcome to think of either of these functors instead. Abusing notation slightly we write NW for both ¯W ◦ Int and ¯W ◦ Int ◦ DK. DK preserves homotopy limits and ¯W preserves homotopy pullbacks of simplicial categories which are homotopy groupoids, see Proposition 1.8 in [50]. Remark 2.1. To reassure the reader that the construction ¯W is the natural one we can note that if we begin with a simplicial model category M we can also consider its classifying space, which is just the nerve of the subcategory of weak equivalences. It follows from Corollary 1.10 in [50] and Proposition 8.7 in [53] that this simplicial set is weakly equivalent to ¯WM . Remark 2.2. The reader should be advised that there are different definitions of (geometric) derived stacks. We use results by Toën-Vezzosi and Lurie respectively, which a priori live in different frameworks. The framework we have used above is Toën-Vezzosi's, where we understand a derived stack as its functor of points. 8 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We will also need to refer to Lurie's approach using structured topoi. A description can be found in the next section. These are shown to give equivalent models for Deligne-Mumford stacks in [43]. It is worth noting that there is no direct approach to Artin stacks using structured topoi. The most concrete approach to derived algebraic geometry is in terms of hypergroupoids. This theory is developed by Pridham in [51], where the equivalence with Toën-Vezzosi's definitions of Artin stacks is established. This approach is based on noting that all geometric stacks can be presented as simplicial affine schemes satisfying some technical condition. It may also be extended to the analytic setting, see Appendix A. There is also some care needed as there are differences in terminology, cf. Remark 2.12 in [50], but the reader can safely ignore these differences unless she or he cares about the precise value of k for which a geometric stack is k-geometric. 2.2. Derived analytic geometry following Lurie and Porta. Since the period map is a priori holomorphic and not algebraic we have to construct its derived enhancement in the setting of derived analytic geometry. The theory of derived analytic geometry is still being developed by Lurie, Porta, and Ben-Bassat and Kremnizer. The main difficulty is that while derived algebraic geometry is modelled on simplicial algebras, derived analytic geometry should be modelled on simplicial analytic algebras, but it is not obvious what those should be. One approach, using Ind-Banach algebras, is developed in [5]. The approach we will use is Lurie's theory of structured spaces, see [36] and Sections 11 and 12 of [35], in particular as extended by Porta in his work around derived GAGA [44, 45]. Porta and Yu have also worked out derived analytic deformation theory [47]. We will use this work largely as a black box, providing us with a good theory of derived analytic DM stacks. In the following we will give some vague explanations while giving references for precise definitions and results. We recommend that the interested reader turn to the introduction of [44] for further explanations. The crucial object is a category T op(Tan) of Tan-structured topoi. We can think of an object as an ∞-topos X together with a sheaf of simplicial commutative rings O alg X and some extra structure. (In the first approximation the reader may think of X as a topological space and ignore the extra structure on OX.) THE GLOBAL DERIVED PERIOD MAP 9 Remark 2.3. Derived analytic spaces have more structure than simply a sheaf of simplicial commutative algebras because one needs to keep track of the action of holomorphic functions on subsets of Cn by post-composition. In the differentiable setting Spivak's definition of simplicial C ∞-rings is motivated by this issue, see [58]. The objects of interest to us form the subcategory dAnC of derived analytic spaces, see Definition 1.3 in [44]. They are the enhancements of complex analytic spaces. They also contain a subcategory of derived Stein spaces, which we will denote by dS tein. (It is called S tnder C in loc. cit.) Together with its analytic topology and the collection of smooth morphisms dS tein forms a geometric context in the sense of Porta-Yu [46]. That means that to define derived analytic Artin stacks one can consider the category of simplicial sheaves on dS tein and use the notion of geometric stacks in the sense of [46], see [44]. Similarly, on the algebraic side, there is a category T op(T´et) of T´et- structured topoi, and this category contains a full subcategory of geometric derived stacks equivalent to DM stacks in Toën-Vezzosi's framework. There is a forgetful functor (−)alg : T op(Tan) → T op(T´et) which corresponds to forgetting extra structure and considering just the algebraic object (X , O alg X ). Then the the following hold: • There is an analytification functor (−)an from T´et-structured topoi to Tan-structured topoi, which is right adjoint to the forgetful functor (−)alg. This is an adjunction of functors of ∞-categories. See Theorem 2.1.1 in [36]. • The analytification functor sends DM stacks locally almost of finite presentation to derived analytic spaces, see Remark 12.26 in [35]. • The analytification functor can be extended to derived Artin stacks, see [45] and also Section 3.1 of [27]. It follows from Lemma 2.35 of [46] applied to the derived context that analytification sends geometric derived algebraic stacks locally almost of finite presentation to geometric derived analytic stacks. • Analytification restricts to the usual analytification on the subcate- gory of underived schemes, see Lemma 4.4 in [45]. • The natural comparison map h : (Xan)alg → X is flat, see Corollary 5.15 in [45]. • Analytification commutes with the truncations π0 and π0, see Section 9.2 in [45] and Lemma 2.20 of [46]. 2.5 10 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN It is worth adding some explanation about analytification of Artin stacks. The analytification functor as defined by Lurie only applies to DM stacks as it depends on the notion of a structured topos. It can be turned into a functor on Artin stacks as a left Kan extension, see [45]. But as a left Kan extension the new functor does not have an automatic left adjoint. (This was pointed out to the authors by Mauro Porta.) Remark 2.4. The analytification of Artin stacks is only given by a left Kan extension when we consider derived stacks on simplicial algebras locally almost of finite presentation. This suffices for our purposes, but the theory extends to arbitrary derived stacks, see Section 3.1 of [27]. Many moduli stacks naturally occur as Artin stacks. This is true in particular for the derived period domain we study here. In joint work with Mauro Porta [27] the second author examines mapping spaces into analytified Artin stacks. Despite the lack of a left adjoint there are algebraic descriptions of these mapping spaces. However, for the purposes of this paper we will sidestep this issue by constructing maps into analytified mapping spaces directly using simplicial presentations. Next we need to recall some characterizations of sheaves on derived analytic stacks. Definition 6.3 in [45] defines a category of coherent sheaves on a derived analytic space X: First one takes the ∞-topos X associated to X and observes there is a category of O alg X -modules. The category of coherent sheaves is the subcategory of modules F whose cohomology sheaves are locally on X coherent sheaves of π0(O alg X )-modules. This is entirely analogous to the algebraic case. Let us unravel this definition. First we make the definition of a derived analytic space a little more precise. Recall the ∞-topos of a topological space T can be thought of as the (∞, 1)-category of sheaves of spaces on T . A model for its hypercompletion is given by the model category of simplicial presheaves on T (cf. Remark 6.5.0.1 in [37]). But a derived analytic space is always hypercomplete, see Lemma 3.1 of [45]. So locally we are just working with the local projective model category structure on simplicial presheaves on a topological space. The space is then equipped with a sheaf of simplicial rings O alg X . (Plus the extra structure that does not affect these definitions). Many of the derived analytic spaces in this paper will actually have an (This underlying topological space, so this description applies globally. THE GLOBAL DERIVED PERIOD MAP 11 is also true in the algebraic setting for derived schemes.) There will be a brief discussion about sheaves on stacks in Section 3.2. In general, OX -modules are OX -module objects in sheaves on an ∞-topos X , cf. Section 2.1 of [35]. As we are working over C we can model this by the model category of sheaves of chain complexes over N(O alg X ) (using the normalization functor N). There are cohomology groups defined in this setting, which just come down to the usual cohomology groups, see Definition 7.2.2.14 in [37]. There is a t-structure on Coh(X) and the heart satisfies Coh♥(X) ≃ Coh♥(π0(X)). Remark 2.5. This definition of coherent sheaves may look naive, but Porta in [44] shows that O alg X -modules are equivalent to the more natural category S p(S trloc Tan (X)/OX ). The cohomology groups can also be used to define bounded below complexes. Note however that O alg X is not necessarily bounded below. As in the underived case there is an analytification functor on sheaves, provided by pullback along the natural map hX : (Xan)alg → X, which sends bounded below coherent sheaves to bounded below coherent sheaves, see [45]. 3. The derived period domain 3.1. The classical period domain. In this section we construct the target of the period map, the (polarized) derived period domain . The derived period domain is a derived analytic stack enhancing the usual period domain. It will classify filtrations of a complex V equipped with a bilinear form Q which satisfy the Hodge-Riemann bilinear relations. For background we will recall the construction of the period domain as a subspace of a flag variety in the underived case. (Here the flag variety is the moduli of (partial) flags with prescribed dimensions in a vector space V. This variety may be described as the quotient S L(V)/P for some parabolic subgroup P.) Assume we are given a smooth projective family of varieties X → S with fibre Xs. The polarized period domain is the moduli of Hodge filtrations on Hk(Xs, C), it is a subspace of the flag variety of Hk. To be precise we are given a vector space Hk = Hk(Xs, Z) ⊗Z C with an integral structure and a bilinear form Qk(α, β) = hLn−kα, βi where L is the Lefschetz operator and h, i is the intersection form hα, βi = R α ∪ β. Note Qk is bilinear, symmetric in even degrees and anti-symmetric in odd degrees. These vector spaces are identified for all s by Ehresmann's theorem as they are diffeomorphism invariants. 12 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Remark 3.1. Qk should not be confused with the hermitian form Qk(α, ¯β). Now the polarized period domain is an open subset of a closed subset of the flag variety, given by the conditions below, which for ease of reference we call the Hodge-Riemann bilinear relations. (This is a slight abuse of language, more correctly the term is only applied to the first and third condition.) (1) F pHk = (Fk−p+1Hk)⊥ with respect to Qk. (2) Hk = F pHk ⊕ Fk−p+1Hk. (3) On H p,q prim = F pHk prim ∩ FqHk Here q = k − p and Hprim is the primitive cohomology. prim we have (−1) k(k−1) 2 ip−qQk(α, ¯α) > 0. The signature of Qk on the non-primitive parts of cohomology is easily worked out. Remark 3.2. When dealing with all cohomology groups at the same it is convenient to consider the bilinear pairing Q(α, β) = R α ∪ β instead and we will do so in future sections. The first condition become Q(F p, Fn−p+1) = 0. (The second and third condition will still be treated level-wise.) The first condition describes an algebraic subset in the flag variety. We call the closed subset given by condition (1) only the closure of the period domain. The other two conditions are (analytic) open on the filtrations satisfying the first condition. Remark 3.3. One can view the period domain in two ways: As an open subspace of a closed subspace of a flag, as above, or as a homogeneous space for the group of symmetries, roughly Aut(HR, Q)/ AutF(HR, Q) ⊂ Aut(H, Q)/ AutF(H, Q). Note that this is a quotient of real groups, which is then shown to be complex as an open subspace of a quotient of complex groups. This approach seems harder to imitate in the derived setting, which is why our starting point is the flag variety. Note that the tangent space of the period domain is given by End /F0(End), where we write End for the Lie algebra of endomorphisms of Hk compatible with the pairing, filtered in the usual way, i.e. F0(End) consists of filtered endomorphisms. See Section 1 of [8]. The image of the period map has tangent space contained in the horizontal tangent space, given by F −1(End)/F0(End), as a consequence of Griffiths transversality. THE GLOBAL DERIVED PERIOD MAP 13 To construct the period map for a family X → S one has to take the quotient of the period domain by the monodromy group π1(S ). Note that π1(S ) is a subgroup of Aut(HZ, Q). It acts on integral cohomology, hence the Z- coefficients, and it preserves the bilinear form. The fundamental group acts properly discontinuously on the period domain. 3.2. The derived flag variety. We will now start constructing the derived period domain. We will build it step by step starting from moduli stacks of perfect complexes and filtered complexes. For simplicity we will define all our stacks on the model category of non- positively graded commutative dg-algebras. We write A for an arbitrary commutative dg-algebra in non-positive degrees. We will consider its model category of dg-modules with the projective model structure (all objects are fibrant). Before we start with the constructions let us first recall the following useful fact which we will use repeatedly: Lemma 3.4. Suppose we are given an (analytic or algebraic) geometric derived stack X . Then for any open substack U of π0(X ) there is a unique geometric derived stack U with π0(U ) = U that is an open substack of X . Proof. See Proposition 2.1 in [55] for the algebraic case. The analytic case follows from Lemma 7.18 in [47]. The key point is that π0 induces an equivalence of sites between Xét and (π0X)ét, i.e. between étale maps to X and to π0(X). Open immersions are just étale maps which are monic. (cid:3) Hence open conditions can be dealt with easily by imposing open conditions on the underlying underived space. Now we recall our two main building blocks, the stacks of perfect complexes and of filtered complexes, which are locally geometric derived stacks locally almost of finite presentation. We begin with a description of quasi-coherent sheaves on derived stacks. Generally speaking, when we have a description of the ∞-category QC oh(U) of complexes of quasi-coherent sheaves for any derived affine scheme U, the most natural global definition is that QC oh(X) = holimi QC oh(Ui) where Ui is a simplicial derived affine scheme whose realization is the derived stack X. As QC oh is a stack (see for example [66]) this is well-defined. 14 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN It will be useful later on to have a very explicit model in which sheaves on a stack can be modelled by a (subcategory of a) model category. This can for example be achieved as follows. Given a cover Ui as above consider QCoh(Ui) to be the dg model category of complexes of quasi-coherent sheaves on Ui, such that its subcategory of fibrant cofibrant objects QCohc f is a model for QC oh. Any Ui → U j induces a Quillen adjunction. These data form a left Quillen presheaf i 7→ QCoh(Ui), see Appendix B of [66]. We define a category such that an object G ∈ QCoh(S ) is given by the following data: For every i there is Gi ∈ QCoh(Ui) and for every map σi j : Ui → U j there is a comparison map φi j : Gi → σ∗ i jG j. Morphisms are given by strictly compatibly families of maps. This is the category of presections with the injective model structure where cofibrations and weak equivalences are defined level-wise. Then we consider those objects G• such that all φi j are weak equivalences. These are called the homotopy cartesian sections and we consider the category of homotopy cartesian sections which are moreover fibrant and cofibrant. This dg category is a model for QC oh(S ). This result is called strictification, it seems to be part of the folklore. An early reference is [24], see also Appendix B of [66]. A detailed and accessible proof under some finiteness assumptions can be found in [57]. These finiteness assumptions are satisfied if S is of finite presentation, cf. the proof of Theorem 4.18. As pointed out by an anonymous referee, a general strictification result can be obtained from the results in [37]. We give a sketch of their argument. The key ideas are as follows: The homotopy limit of the diagram i 7→ QCoh(Ui) can be computed as a quasi-categorical limit. The ∞-functor R Spec(A) 7→ A-Mod from derived affine schemes over S considered as an ∞-category to the ∞-category of chain complexes can be turned into a cartesian fibration p : Mod → (dA f f /S )op of ∞-categories by the yoga of straightening/unstraightening. Then unravelling definitions the total space Mod of the fibration has a presentation coming from a model category. The limit of this diagram may be written as a category of cartesian sections in the ∞-categorical sense by Corollary 3.3.3.2 of [37]. Now the ∞-category of all sections can be strictified using Proposition 4.2.4.4 of [37], and cartesian sections are a left Bousfield localization of the ∞-category of all sections. Definition. Let Per f , the moduli stack of perfect complexes, be defined as follows. For any derived stack S consider a simplicial derived affine hypercover U• and let Per f (S ) be the category of fibrant cofibrant homotopy cartesian sections G• of the left Quillen presheaf QCoh such that each Gi is perfect as THE GLOBAL DERIVED PERIOD MAP 15 a O(Ui)-module. We say that an A-module M over a dg-algebra A is perfect if it is compact in the derived category of A. This is equivalent to saying M is pseudo-coherent (also called almost perfect) and of finite Tor dimension. A H0(A) is a perfect The last condition is equivalent (see [64]) to saying M ⊗L complex of H0(A)-modules, i.e. is quasi-isomorphic to a bounded complex of finite projective H0(A)-modules. In particular Per f sends Spec(A) to the dg-category of cofibrant A- modules quasi-isomorphic to strictly perfect dg modules over A. Then Per f (S ) ≔ NW(Per f (S )). (In the literature this is sometimes denoted RPer f .) Recall that when taking NW we restrict the morphisms to allow only the ones that become isomorphisms in the homotopy category. For a proof that this is a locally geometric derived Artin stack locally almost of finite presentation, see [14]. Note that this stack is equivalent to the construction of M1 in [64]. Definition. Let Filtn be the stack of filtered perfect complexes of filtration length n + 1. n (A) be the dg model category of sequences Fn → Explicitly, we let F iltmod . . . → F1 → F0 of injections of A-modules as defined in [14]. In particular, weak equivalences are given by filtered quasi-isomorphisms, i.e. morphisms inducing quasi-isomorphisms on the associated graded modules. F iltn(A) is the dg subcategory of fibrant cofibrant sequences F • such that all Fi are perfect. Remark 3.5. For later examples it is reassuring to note that it follows from Propositions 1.19 and 1.21 in [14] that any bounded filtered complexes over C is fibrant and cofibrant. HomF iltn(A)(F ∗, G∗) consists of compatible maps Fi → Gi. These maps are determined on F0, and we will sometimes write the derived hom complex as RHomF(F0, G0) if the filtrations are clear. We note that the associated simplicial category may also be defined by Dwyer-Kan localising the category of (not necessarily fibrant and cofibrant) filtrations at filtered quasi-isomorphisms. We will sometimes represent points of F iltn by filtrations which are not fibrant or cofibrant, but will always consider the derived hom space RHomF. Analogously to Per f (S ) we can define the dg category F iltn(S ) as consisting of certain fibrant cofibrant homotopy cartesian sections of the left Quillen presheaf Ui 7→ F iltmod (O(Ui)). n 16 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We then let Filtn be NW(F iltn). That this is a locally geometric derived stack locally almost of finite presentation can be deduced from Theorem 2.33 and Remark 2.34 in [14], where it is shown that ∪nFiltn and Filt1 are locally geometry. (Note that our subscript n has a different meaning from the superscript n in F iltn in loc. cit.) Remark 3.6. We note that the category F iltn(A) as constructed here and in [14] is not stable. The stable version is given by strings of n composable arrows in Per f (A) without the injectivity condition. These two constructions give rather different categories, but it follows from the proof of Theorem 4.4 of [22] that the associated homotopy categories and mapping spaces agree, and thus the two constructions give rise to the same stack. The next stack also appears in [14], but we will revisit the construction to give an explicit description of the mapping spaces. n(V)(A) = holim(F iltn(A) ⇒ Lemma 3.7. For a complex V consider F lag′ Per f (A)) where the two maps are induced by the forgetful functor π : F ∗ 7→ F0 and the constant functor V : F ∗ 7→ V. This is the dg-category of pairs (F ∗, wF) where F ∗ is an object of F iltn(A) and wF : F0 → V ⊗ A is a homotopy equivalence in Per f (A). The morphism complex is given by RHom((F ∗, wF), (G∗, wG)) = (RHomF(F0, G0) ⊕ RHom(F0, A[1]), ∆) where ∆ : ( f , h) 7→ (−d f , dh + wG ◦ f − wF ◦ δ) where δ is the identity if f has degree 0 and 0 otherwise. Proof. To show that the characterization of the homotopy limit is correct n , which has objects consisting of triples we replace F iltn by F ilt∼ (F ∗, MF, mF) where F ∗ is an object in F iltn and mF : F0 → MF is a homotopy equivalence in Per f . Morphism complexes are given by (RHomF ilt(F0, G0) ⊕ RHom(MF, MG) ⊕ RHom(F0, MG)[1], ∆) where ∆ : ( f , g, h) 7→ (−d f , dg, dh + mG ◦ f − g ◦ mF). The natural inclusion is a quasi-equivalence and the map F ilt∼ n → Per f is a fibration. This follows from the standard arguments used in the construction of the path space in dg-categories, see Section 3 of [60]. As dg-categories (with the Dwyer-Kan model structure) are right proper we can compute the homotopy limit as the limit of F iltn → Per f ← ∗, which is the category described in the Lemma. (cid:3) Lemma 3.8. The derived flag variety DFlagn(V) is defined as NW of the substack F lagn(V)′ that is given by filtrations such that the maps THE GLOBAL DERIVED PERIOD MAP 17 Hk(Fi−1 ⊗L It is geometric. A H0(A)) → Hk(Fi ⊗L A H0(A)) are injections of flat H0(A)-modules. n(V)) is the homotopy limit of Filtn ⇒ Proof. We first note that NW(F lag′ Per f as NW commutes with homotopy fibre products of simplicial categories whose homotopy categories are groupoids. As a homotopy pullback of locally geometric stacks it is locally geometric. To see the substack is locally geometric, use Proposition 2.44 in [14]. F iltn(H0(A))(F ∗ ⊗L We now want to refine this result and show that DFlagn(V) is geometric. We let k − 1 be the amplitude of the complex V and will show that DFlagn(V) is (k + 2)-geometric. We introduce the k-geometric stack Filtk n(A) which classifies filtrations F ∗ which satisfy that Exti A H0(A), F ∗ ⊗L A H0(A)) = 0 for i < −k. This is a derived geometric k-stack by Theorem 2.33 of [14], and thus is (k + 2)-geometric. Now we check that all the points in DFlagn(V) map to Filtk n(A) ⊂ Filtn(A). We need to compute Ext-groups for filtered complexes. We do this by considering the Rees construction ⊕Fiti, cf. [14]. Then Ext∗ H0(A)[t](⊕Fiti, ⊕Fiti)Gm. As Gm is reductive we can ignore it when checking vanishing of Ext-groups. A H0(A) have amplitude bounded by the amplitude of V. Moreover, All Fi ⊗L they are projective, thus there is a two-term resolution of ⊕Fiti by projective R[t]-modules (see e.g. Theorem 4.3.7 in [70]). Thus the Ext groups vanish in degrees less than k. F ilt(H0(A))(F ∗, F ∗) = Ext∗ It follows that we can construct DFlagn(V) as an open substack of the n → Per f k, where Per f k is defined analogously to homotopy fibre of Filtk Filtk (cid:3) n. Thus DFlagn(V) is (k + 2)-geometric. Moreover one can show that the derived Grassmannian DFlag1(V) is i.e. π0π0DFlag1(V) = an enhancement of the usual Grassmannian, Qki,i Gr(ki, Hi(V)), see Theorem 2.42 in [14], and similar for the derived flag varieties. The tangent complex of DFlagn(V) at (F ∗, wF) is computed by the homotopy limit of the tangent complexes: TF∗,wF DFlag(V) = cone((χ, 0) : holim(REndF(F0) ⇒ REnd(F0))[1] Here REndF(V) is the derived endomorphism space in the model category of filtrations, χ is the inclusion and 0 is the constant zero map. For this and similar cones we will write REnd(F0)/ REndF(F0) 18 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN 3.3. The derived period domain I: Algebraically. As the next step towards constructing the derived period domain we construct in this section a geometric derived algebraic stack D that classifies filtrations with a bilinear form which satisfy only the Hodge-Riemann orthogonality condition. The derived period domain U will then be the open substack of Dan determined by the second and third Hodge-Riemann condition (positivity). We fix a non-negative integer n and a complex V = VQ ⊗ C concentrated in degrees 0 to 2n with a 2n-shifted non-degenerate symmetric bilinear form. Clearly this map can be extended by multiplication to a bilinear map on V ⊗ A. Definition. Given an A-module W we denote by Sym2W the derived coinvariants of the natural Z/2Z-action on V ⊗ V and we call a map Q : Sym2W → A[k] for some integer k a shifted bilinear form. Q is called non-degenerate if the associated map q : W → W ∨[2n] is a weak equivalence. ] We want to consider the moduli stack Dn(V, Q) of filtrations F ∗ on the complex V such that Fi and Fn−i+1 are orthogonal with respect to Q. Theorem 3.9. Let (V, Q) be as above. Then there is a geometric derived stack D = Dn(V, Q) which classifies decreasing filtrations of V of length n + 1 that satisfy the Hodge-Riemann orthogonality relation with respect to Q. D enhances the product of closures of classical period domains. Here D = NW(Dn(V, Q)) where Dn(V, Q)(A) is a certain simplicial category whose homotopy category is given as follows: Objects are given by triples (F ∗, wF, QF) where • F ∗ is a filtration of perfect A-modules of length n + 1, • QF is a non-degenerate 2n-shifted bilinear form on F0 that vanishes on Fi ⊗L A Fn+1−i for all i, • wF : (F0, QF) ≃ (V ⊗ A, Q) is an isomorphism in the homotopy category of perfect complexes with shifted bilinear form, A H0(A)) are flat. such that F ∗ gives filtrations on cohomology after tensoring over A with H0 and all H j(Fi ⊗L Morphisms from (F ∗, wF, QF) to (G∗, wG, QG) are given by families of morphisms Fi → Gi in the homotopy category of F iltn(A) that are compatible with w• and Q•. Remark 3.10. In the case of interest to us V will have a real structure and Q will be real. We note that just as in the underived case the real structure only THE GLOBAL DERIVED PERIOD MAP 19 becomes relevant when considering the positivity condition for the period domain. We will construct Dn(V, Q) by adding the data of a quadratic form to F lagn(V). As a warm-up and for use in later sections we first prove the following result about a stack constructed by Vezzosi in [68]: Lemma 3.11. There is a locally geometric derived stack of perfect complexes with a 2n-shifted non-degenerate bilinear form, denoted QPer f . Moreover QPer f (A) = NW(QPer f (A)) for a simplicial category QPer f (A) whose objects are given by perfect complexes W over A with a bilinear form Q : Sym2W → A[2n] that is non-degenerate. The morphism space MapQPer f (A)((W, Q), (W ′, Q′)) is given by the homotopy fibre of the map f 7→ Q′ ◦ Sym( f ) : MapPer f (A)(W, W ′) → Map(Sym2W, A[2n]) over Q. Proof. This is the derived stack that is constructed in Section 3 of [68] and denoted by QPer f nd(2n) there. We recall the construction and prove the properties we need. We use simplicial categories as our model for (∞, 1)-categories. First we define QPer f ′ using the sheaf of simplicial categories QPer f ′ defined via the following homotopy pullback diagram. QPer f ′(A) QPer f ′(A) Per f (A)∆1 / Per f (A)∆1 ev0×ev1 Per f (A) Per f (A) Sym2 A×A[2n] Per f (A) × Per f (A) Per f (A) × Per f (A) k[Σ2] (V ⊗L A is the functor V 7→ k ⊗L Here (−)∆1 denotes the functor category from the two object category with one morphism, ∗ → ∗. Then ev0 × ev1 denotes the product of evaluation maps. Sym2 N(A) V), where the Σ2-action is given by swapping factors. To compute the homotopy limit we need to replace Per f ∆1 → Per f × Per f by a fibration. We consider the dg-category Per f ∆1,∼ which replaces morphisms in Per f ∆1, which by definition are commuting squares, by homotopy commutative squares, i.e. in Per f ∆1,∼ we have RHom( f : K → L, g : M → N) = (RHom(K, M) ⊕ RHom(L, N) ⊕ RHom(K, N)[1]), ∆ where ∆ : (x, y, h) 7→ (dx, dy, dh + y ◦ f − g ◦ x). As in Per f we only consider morphisms which become invertible in the homotopy category. Then we apply the Dold-Kan functor which preserves fibrations. / / / 20 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Unravelling definitions the limit is the category of perfect complexes with a bilinear form. The objects are given by objects W ∈ Per f (A) and Q : Sym2W → A[2n] in Per f (A)∆1. Morphisms from (W, QW) to 7→ Q′ ◦ Sym2( f ) : Map(W, U) → (U, QU) are given by the fibre of f Map(Sym2W, A[2n]) over Q. We apply NW to QPer f ′ and call the resulting derived stack QPer f ′. We know that NW respects this homotopy limit. Thus by Lemma 3.13 below QPer f ′ is a homotopy limit of locally geometric stacks and thus locally geometric. The condition that Q is non-degenerate is easily seen to be open: It is enough to check on open subsets of the underlying underived stack, and being quasi-isomorphic is an open condition on complexes. Thus the substack of non-degenerate bilinear forms QPer f (2n) ⊂ QPer f ′(2n) is also locally geometric. It is clear from these definitions that π0π0(QPer f (2n))(A) consists of isomorphism classes of perfect complexes over H0(A) with a non- degenerate quadratic form. (cid:3) Remark 3.12. The reason we have to work with simplicial categories is of course that f 7→ Sym2( f ) is not a linear map. Lemma 3.13. NW(Per f ∆1) is a locally geometric stack. Proof. We can see this by identifying our stack with the linear stack associated to the perfect complex E = H om(p∗ U )∗ on the locally 1 geometric stack Per f ×Per f . Here U is the universal complex on Per f . U , p∗ 2 Toën defines linear stacks in Section 3.3 of [63]. For a derived Artin stack X with quasi-coherent complex E we define the stack VE as a contravariant functor on derived schemes over X by associating VE(u) = MapLqcoh(S )(u∗E, OS ) to u : S → X. Now VE is locally geometric whenever E is perfect, the proof follows from Sublemma 3.9 in [64]. To show this agrees with our construction above, we consider the definition of Per f ∆1 locally. Over (V, W) : S → Per f × Per f we have: MapPer f ×Per f ((V, W), Per f ∆1) = NW (cid:16)Per f I(S )) ×Per f (S )×Per f (S ) {(V, W)}(cid:17) On the other hand VE((V, W)) = MapS (H om(V, W)∗, OS ). Both of these compute the mapping space from V to W in Per f (S ). (The characterization in terms of the fibre product can for example be found in Sections 1.2 and 2.2 of [37].) (cid:3) THE GLOBAL DERIVED PERIOD MAP 21 Remark 3.14. There is another explicit description of QPer f : as an open substack of the linear stack associated to the perfect complex H om(Sym2U , O[2n]) on Per f , following the same arguments as Lemma 3.13. To show this agrees with our construction above, consider the definition of QPer f locally. (Looking at the simplicial set, not the dg-category.) Over V : S → Per f we have: MapPer f (V, QPer f ) = Map(S , QPer f ) ×Map(S ,Per f ) {V} = (cid:16)Per f (S ) ×(Per f ×Per f )(S ) Per f ∆1(S ))(cid:17) ×Per f (S )) {V} = n(Sym2V, O)o ×Per f (S )×Per f (S ) Per f (S )∆1 And the last line is precisely the mapping space from Sym2V to O in Per f (S ), i.e. it is V(Sym2(U ))(V). Applying shift and dual does not affect the argument, hence we are done. We can use the same technique as in Lemma 3.11 to consider filtered complexes with a bilinear form: Lemma 3.15. There is a locally geometric derived stack QFiltn of perfect complexes with an (n+1)-term filtration and a non-degenerate bilinear form compatible with the filtration. To be precise QFiltn = NW(QF iltn) for a simplicial category QFiltn whose objects are filtrations of perfect A-modules Fn → Fn−1 → . . . → F0 together with a non-degenerate bilinear form Q : Sym2F0 → A[2n] such that Q vanishes on the image of Fi ⊗L The morphism space MapQPer f (A)((F ∗, QF), (G∗, QG)) is given by the fibre of the map f 7→ QG ◦Sym( f ) : MapF iltn(F ∗, G∗) → MapF ilt1(Sym2F ∗, A[2n]) over QF. A Fn+1−i for all i. Proof. We proceed like in the proof of Lemma 3.11 and consider the homotopy pullback diagram QF ilt′ QF ilt′ n(A) n(A) (1) F ilt1(A)∆1 / F ilt1(A)∆1 ev0×ev1 F iltn(A) F iltn(A) S ×(0→A[2n]) F ilt1(A) × F ilt1(A) F ilt1(A) × F ilt1(A) Here the map S sends Fn → . . . → F0 to SF → Sym2F0 where SF is the image of ⊕i(Fi ⊗L A Fn+1−i) under F0 ⊗L A F0 → Sym2F0. / / / 22 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We replace the right vertical map by a fibration in the same manner as in Lemma 3.11. Then we see that the objects of QF ilt′ n(A) are filtrations with a bilinear form that vanishes on SF, i.e. satisfies Hodge Riemann orthogonality. Similarly we write down the mapping spaces. Again we see that QFilt′ geometric stacks and thus locally geometric. We can show NW(F ilt∆1 locally geometric by an analogue of 3.13. n) is a homotopy limit of locally 1 ) is n = NW(QF ilt′ Again the substack of non-degenerate forms QFiltn is an open substack. (cid:3) Remark 3.16. A natural way to understand the orthogonality relation is to consider the map q : F0 → (F0)∨ induced by Q. The filtration F ∗ induces a filtration (F0/Fi)∨ and the orthogonality says precisely that q respects this filtration. We are now ready to construct the stack Dn(V, Q). Proof of Theorem 3.9. To define Dn(V, Q) we again repeat the construction of Lemma 3.11, this time with DFlag in place of Per f . We recall that the homotopy fibre over V of the projection map F iltk → Per f is given by F lagk(V). While it is possible to compute the homotopy limit directly, the homotopy category of the homotopy fibre product can be computed more efficiently using (the proof of) Corollary 1.12 in [50]. G Assume we are given a diagram A ← C of simplicial categories such that all their homotopy categories π0(A ) etc. are groupoids. Then (2) the cited result says we can compute π0(A ×h π0(B) π0(C ). Here the 2-fibre product of categories is defined to have as objects triples (a ∈ A , c ∈ C , ω : F(a) (cid:27) G(c)) and morphisms from (a, c, ω) to (a′, c′, ω′) are just given by morphisms f : a → a′ and g : c → c′ such that G(g) ◦ ω = ω′ ◦ F( f ). C ) as π0(A ) × F → B B 1 → Per f ∆1 over Q : Sym2V → A[2n]. We call it F lag∆1 We consider first the homotopy fibre of the map of simplicial categories F ilt∆1 1 (Q), abusing notation, and we note it is geometric by imitating the proof of Lemma 3.8. THE GLOBAL DERIVED PERIOD MAP 23 By the above, objects of the homotopy category can be written as diagrams M0 M0 f0 N0 N0 M1 / M1 M1 M1 wM / Sym2V Sym2V f1 N1 N1 N1 N1 Q A A wN where the left hand square commutes strictly and the right hand square is a commutative diagram in the homotopy category. n(V, Q) is defined as the homotopy pullback of the following diagram D ′ (suppressing A from the notation): D ′ D ′ n(V, Q) n(V, Q) F lag1(Q)∆1 / F lag1(Q)∆1 ev0×ev1 F lagn(V) F lagn(V) S ×(0→A[2n]) F lag1(Sym2V) × F lag1(A[2n]) F lag1(Sym2V) × F lag1(A[2n]) We use the same tool again to compute D ′ as the 2-fibre product. We find that its homotopy category has objects given by triples (F ∗, wF, QF) consisting of a filtration F ∗, an isomorphism wF : F0 → V ⊗ A in the homotopy category and a symmetric bilinear form QF : Sym2F0 → A compatible with Q that vanishes on S f ⊂ Sym2(F0). Morphisms are just homotopy commutative collections of object-wise morphisms. A H0(A)) → H j(Fi−1 ⊗L Then D′ ≔ NW(D ′) is geometric as a homotopy limit of geometric stacks. We define D(A) to be the subcategory of D′(A) whose objects satisfy that A H0(A)) are injections of flat H0(A)-modules. H j(Fi ⊗L These are open conditions on D′, the proof follows as in Proposition 2.41 of [14]. It remains to check that D enhances the period domain, i.e. that π0π0(D)(A) is a product of the usual period domains. To do this we assume A is an algebra (considered as a constant simplicial algebra). We observe that the objects of D(A) are formal, any (F ∗, wF, QF) is equivalent to some filtration on the complex V ⊗ A with zero differential. As the complex Fi is perfect and the cohomology groups of the Fi are flat they are projective and the complexes are formal. Thus we get a homotopy equivalence from Fn → . . . → F0 ≃ V ⊗ A to i.e. / / / / / / / / / 24 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN H(Fn) → . . . → H(F0) (cid:27) V ⊗ A, and these are the objects parametrized by the product of closed period domains. (cid:3) Remark 3.17. We note that we can equivalently define D ′(A) as the homotopy limit holim((π, V) : QF iltn(A) ⇒ QPer f (A)). The arrows are induced by the forgetful functor π : (F •, QF) 7→ (F0, QF) and the constant functor that sends any filtration to (V, Q). As homotopy limits commute this description as a homotopy fibre at (V, Q) of the projection QF ilt → QPer f of homotopy limits is the same as the homotopy pullback of homotopy fibres over V and Q that we used above. Recall that we gave a description of sheaves on a derived stack S as fibrant cofibrant homotopy cartesian sections of a left Quillen presheaf. This description works not just for Per f and F iltn but also QF iltn and indeed Dn(V, Q) There is a slight subtlety in this model: The presection U 7→ OS (U) is not fibrant, so where it appears, for example in the definition of shifted bilinear forms QF ilt, it needs to be replaced fibrantly. Of course this does not affect the homotopy category, so we will not make this explicit if there is no ambiguity. As an example we state the following global version of Theorem 3.9 to characterize maps from non-affine derived stacks to Dn(V, Q). Here by a perfect complex we mean a fibrant cofibrant homotopy cartesian section as above, and we write P for fibrant cofibrant replacement in the model category of presections. Corollary 3.18. Let S be a derived stack. Then the objects of Dn(V, Q)(S ) may be given by triples (F ∗, wF, QF) where • F ∗ is a filtration of length n + 1 of perfect complexes on S • QF : Sym2F0 → POS [2n] is a non-degenerate 2n-shifted bilinear form on F0 that vanishes on Fi ⊗L A Fn+1−i for all i, • wF : (F0, QF) ≃ (V ⊗ OS , Q) is an isomorphism in the homotopy category of complexes with the shifted bilinear forms, such that F ∗ gives filtrations on cohomology after tensoring over OS with H0(OS ) and all H j(Fi ⊗L OS H0(OS )) are flat. 3.4. The tangent space. Proposition 3.19. The tangent complex of Dn(V, Q) at (F •, wF, QF) is given by REndQF (F0) is the complex of endomorphisms of F0 that are Here REndQF antisymmetric with respect to QF, i.e. satisfy QF( f −, −) +(−1) f Q(−, f −) = (F0)/ REndF,QF the C-point (F0)[1]. THE GLOBAL DERIVED PERIOD MAP 25 (F0) are those endomorphisms that moreover respect the 0. REndF,QF filtration. Proof. We obtain this result by computing the tangent complexes of QPer f and QFiltn and appealing to the description of D′ n(V, Q) as their homotopy fibre, cf. Remark 3.17. As QPer f = holim(Per f → Per f × Per f ← Per f ∆1) and the tangent space commutes with limits we have the following pullback diagram: TQ:Sym2V→O QPer f TQ:Sym2V→O QPer f / TQPer f ∆1 TQPer f ∆1 TV Per f TV Per f (dSym2,0) TSym2V,O(Per f × Per f ) TSym2V,O(Per f × Per f ) Let us analyse the ingredients. We recall that all our stacks come from sheaves of dg-categories and then we use Corollary 3.17 from [64] which tells us that to compute the tangent space of the moduli stack of objects of a triangulated dg-category of finite type it suffices to compute the hom- spaces, i.e. TD(M ) = REnd(D)[1]. This applies since we can consider Per f and Per f ∆1 as the moduli stack of objects of a dg-category of finite type. Thus it follows that TV Per f = REnd(V)[1] TSym2V,O(Per f × Per f ) = (cid:16)REnd(Sym2V) ⊕ REnd(A)(cid:17) [1] T f :A→B(Per f ∆1) = (cid:16)REnd(A) ⊕ REnd(B) ⊕ RHom(A, B)[1], ∆(cid:17) [1] Here the last complex has differential ∆ : (x, y, h) 7→ (dx, dy, dh + gx − y f ). This follows from considering hom-spaces in the category Per f ∆1. Now to consider the map dSym2 we have to change to simplicial categories. It is clear that the induced map on derivations sends f to f ⊗ 1 + (−1) f 1 ⊗ f . Putting this together we obtain that the tangent space at (V, Q) is the homotopy kernel of the map eQ : REnd(V) → RHom(Sym2V, A[2n]) defined by eQ( f ) ≔ Q( f −, −) + (−1) f Q(−, f −). We write the kernel as REndQ(V). Over C the complexes are formal and as Q is non-degenerate eQ is surjective, so these are just endomorphisms that are anti-symmetric with respect to Q. / / / 26 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We proceed similarly for QFilt as defined in diagram (1). We use Corollary 3.17 from [64] again to see that the tangent space of F iltn(A) at F • is given by endomorphisms of F0 preserving the filtration. Then dS sends a morphism f respecting the filtration F to ( f ⊗ 1 + (−1) f 1 ⊗ f ), which respects the filtration SF ⊂ Sym2F0. Thus we find that TF∗,QF QFilt consists of filtered endomorphisms of F0 that are compatible with QF. Then T(F∗,wF ,QF)D is the homotopy equalizer of the natural inclusion and the zero map. (We use the identification of the tangent spaces to QPer f at (V ⊗ A, Q) and (F0, QF).) We conclude by noting that REndF,Q(F0) → REndQ(F0) is level-wise surjective. (cid:3) This is entirely analogous to the tangent complex of the derived flag variety, which is (REnd(V)/ REndF(V))[1]. We also note that the zeroth cohomology group consists of orthogonal endomorphisms of V modulo filtered orthogonal endomorphisms of V. This is of course the tangent space in the underived case. Remark 3.20. Equivalently we can of course write the tangent complex as the shifted cone on the map q : REnd V/ REndF V → RHom(Sym2V, A[2n])/ RHom(Sym2V/SF, A[2n]) that is induced by Q. As the tangent complex and thus the cotangent complex is perfect and as π0π0(D) is a scheme of finite presentation we deduce: Corollary 3.21. The derived period stack Dn(V, Q) is locally almost of finite presentation. 3.5. The derived period domain II: Analytically. As the period map and the second Hodge-Riemann bilinear relation are analytic by nature we need to understand the derived analytic period domain. To add the Hodge-Riemann bilinear relations we need some extra structure, so we recall that V = H(Xs, C) is equipped with a real structure and the Lefschetz operator L. We define the bilinear forms Qk on HkV by Qk(α, β) = Q(Ln−kα, β). Theorem 3.22. There is a derived analytic period domain U, an open analytic substack of Dn(V, Q)an, which is an enhancement of the period domain. Proof. We first apply the Lurie-Porta analytification functor reviewed in Section 2.2 to the geometric stack D. This is possible as D is locally almost of finite presentation by Corollary 3.21. THE GLOBAL DERIVED PERIOD MAP 27 We need to check that the underived truncation π0π0(Dan) is the closed analytic period domain, but this is immediate from Theorem 3.9 since analytification commutes with truncation. We note that orthogonality with respect to Qk follows from orthogonality with respect to Q as in the definition of Dn(V, Q) Now we construct the open substack U of Dan. Lemma 3.4 shows that it suffices to construct an open substack U ′ on the underlying underived stack. We take U ′ to be the preimage under the natural map π0(Dan) → π0π0(Dan) of the classical period domain that is defined inside π0π0(Dan) = (π0π0(D))an by the second and third Hodge-Riemann bilinear relation. Explicitly, it is the open subspace given by the conditions that FiHk ⊕ Fk−i+1Hk = Hk and the form given by (−1)k(k−1)/2ip−qQk(·, ¯·) is positive definite. (cid:3) Remark 3.23. As U is open in Dan it will have the same tangent complex. Moreover, the tangent complex is unchanged by analytification, see the discussion in Section 4.6. Let us now consider how we will construct a map into this derived period domain. The period map is naturally a map into a moduli space, however for technical reasons we have now constructed the derived period domain as an analytification of an algebraic moduli stack, rather than as an analytic moduli stack. Thus we need to consider the following question: Given the algebraic moduli stack D, what can we say about Dan? It seems too optimistic to expect an analytification of a moduli stack to always represent some corresponding analytic moduli problem. In the setting of this paper we directly construct a map to the analytified derived period domain. There are several interesting instances where the analytification of a moduli space is the corresponding analytic moduli space. For example, in the underived setting one may observe that the analytification of the Grassmannian is an analytic Grassmannian. One way to see this is that there are matching explicit constructions. In fact, something very similar holds in our situation and U turns out to be the derived analytic moduli stack of filtrations of V⊗O by perfect complexes that satisfy the Hodge-Riemann bilinear relations. We prove this result in joint work with Mauro Porta as an application of the main theorem of [27], which says that in many cases the analytification functor commutes with the mapping space functor. 28 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN 3.6. Monodromy and quotient. The period map only maps to the quotient of the period domain by monodromy, so we examine monodromy for families of algebraic varieties in the derived case, and then construct the quotient of the derived period domain by a group acting on V. Assume we are given a smooth projective map f : X → S of derived schemes, i.e. a strong map of derived schemes such that π0π0( f ) is smooth and projective. We write t(S ) for the underlying topological space of S an, i.e. π0S (C) considered in the analytic topology. The fundamental group of t(S ) acts on cohomology of the fibre as each Ri f∗ΩX/S is a local system on S . Looking instead at the complex R f∗C we see that it forms a homotopy locally constant sheaf on t(S ). We will see that R f∗Ωan A homotopy locally constant sheaf is equivalently a representation of the simplicial loop group Ωt(S ), see [26], i.e. a priori there is higher monodromy. But the degeneration of the Leray-Serre spectral sequence for families of projective varieties suggests that R f∗C is just a direct sum of local systems, i.e. there is no higher monodromy. Indeed, there is the following result due to Deligne: X/S ≃ R f∗C ⊗ O alg S an in Lemma 4.8. Theorem 3.24. Consider a triangulated category D with heart A . Let n ∈ N, V ∈ D(A ) and assume there is u : V → V[2] inducing isomorphisms: ui : Hn−i(V) ≃ Hn+i(V) for i ≥ 0. Then there is a canonical map φ from HV ≔ Pi Hi(V)[−i] to V inducing the identity on cohomology. If D is the derived category of A the existence of the isomorphism is Theorem 1.5 in [10], but the proof does not depend on this assumption, see also the beginning of [12]. The canonical map is constructed in [12]. Now we let A be the category of sheaves of abelian groups on t(S ), X be R f∗C and u be the map induced by the relative Lefschetz operator. We obtain the following (see [10] 2.6.3): The homotopy locally constant sheaf R f∗C is quasi-isomorphic to a direct sum of its cohomology sheaves, which are local systems. We also need to consider the shifted bilinear forms on R f∗C and its cohomology. We may compute R f∗(C) in such a way (for example using a Cech cover of t(X) over t(S )) that it comes with a shifted bilinear form QR induced by the pairing hα, βi = R α ∪ β given by the cup product and the trace map R : R f∗C[−2n] → R2n f∗C → C given by Verdier duality. This of course induces the pairing Q on cohomology. QR also pulls back THE GLOBAL DERIVED PERIOD MAP 29 via φ to a bilinear form on cohomology H(R f∗C) and as φ is the identity on cohomology this is again the usual pairing on cohomology. Thus we have a quasi-isomorphism φ : (H(R f∗C), Q) → (R f∗C, QR) of complexes of sheaves with shifted bilinear forms. We deduce: Corollary 3.25. The complex R f∗C with its bilinear form considered as a Ωt(S )-representation is equivalent to a representation of π1(t(S )). Note that in all this the base is allowed to be singular, the only assumption is that t(S ) is locally paracompact. When constructing the global period map we will see (cf. Lemmas 4.8 and 4.12) that this monodromy is the only obstruction to gluing the local derived period maps, i.e. the derived structure of S (which is infinitesimal) does not interfere. Remark 3.26. On the other hand Voisin shows in [69] that Deligne's decomposition does not work on the level of algebras. This suggests that extending the period map to keep track of the algebra structure on R f∗ΩX/S poses a significantly harder problem. The bilinear form is defined using the algebra structure, but we have seen that preserving the bilinear form is a much weaker condition than preserving the algebra structure. Remark 3.27. One might wonder what happens if one considers the complex R f∗ΩX/S in the case that S is underived, i.e. if one considers a classical period map taking values in complexes. Consider the Hodge filtration Fi on R f∗ΩX/S . Then we note that now all the Fi are formal, i.e. quasi-isomorphic to HFi. This follows since they have locally free cohomology. (In the derived setting this argument would not apply.) So the algebraic geometry of S is not reflected in studying Hodge structures as a complex (as opposed to a collection of cohomology groups), and the topology is not reflected beyond the fundamental group. This means for example that failures of the Torelli theorem are not resolved by considering a period map taking values in complexes. We will now connect the monodromy considerations to our construction of U. We think of V = VZ ⊗ C as the cohomology complex of a fibre Xs of our map X → S of derived schemes. (Note Xs is a classical smooth projective scheme.) Then we have a strict action of the fundamental group π1(S ) on V compatible with the form Q. In fact this action factors through the action of the universal arithmetic group Γ = Aut(VZ, Q) on V with the bilinear form Q. 30 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We will now show that the action of Γ on V induces an action on the derived period domain. Proposition 3.28. The action of Γ on V induces an action on Dn(V, Q) which restricts to an action on the derived period domain U and the homotopy quotient U/Γ exists as a geometric derived analytic stack. Proof. As V comes with a Γ-action there is a map BΓ → Per f classifying this representation and V : ∗ → Per f factors through it. Note that the simplicial set BΓ can be considered as a constant stack (i.e. we stackify the constant functor BΓ on derived schemes). We do not expect this to be geometric or locally almost of finite presentation as Γ is an infinite discrete group. As the action is compatible with Q we can consider the map γ : BΓ → QPer f classifying the pair (V, Q). Then the pullback of QFilt → QPer f along γ is a derived stack DΓ. As the homotopy pullback of DΓ → BΓ along the universal cover ∗ → BΓ is equivalent to D (see Remark 3.17) we see that DΓ is the desired quotient. Explicitly, Γ acts on Dh ≔ DΓ ×BΓ EΓ via its action on EΓ, and of course Dh ≃ D. Analytification preserves effective epimorphisms as it is defined as a left Kan extension along a continuous morphisms of sites. It also commutes with homotopy limits and thus we can consider Dan Γ as a quotient of Dan. The underlying underived stack is not necessarily geometric, but we will now construct a geometric homotopy quotient of U by Γ. As the non-degeneracy condition defining U (see Theorem 3.22) is unaffected by Γ it defines a substack U/Γ of Dan h /Γ. We also note that the open substack U ⊂ Dan corresponds to an open substack Uh of Dan h , which is preserved by the action of Γ, as this is true for the underlying underived spaces where Γ is the usual monodromy action. U/Γ is equivalent to the homotopy quotient Uh/Γ, justifying the name. Now we use the fact that the action of Γ on π0π0(Uh) (which is just the usual period domain) is properly discontinuous to show that U/Γ is geometric. Γ ≃ Dan There is an open cover U of the period domain π0π0(Uh) such that the images of any two open sets under the group action intersect only finitely many times. This gives a cover of the stack π0(Uh) and thus of the derived stack Uh with the same property. Thus Uh provides a cover of Uh/Γ that is locally a quotient by a finite group, which is enough to show U/Γ is geometric. (cid:3) THE GLOBAL DERIVED PERIOD MAP 31 4. The period map 4.1. The classical period map. In this section, which is only used as background, we recapitulate Griffiths' period map [21]. : X → S , where For a polarized family of smooth projective varieties f : S an → U/Γ, where S of finite type, there are holomorphic maps Pk U ⊂ Flag(F ∗, Hk) is the polarized period domain and Γ is the monodromy group. Here Pk(s) is FnHk ⊂ · · · ⊂ F1Hk(Xs) ⊂ Hk(Xs) (cid:27) Hk(X0). We define P(s) as the product of the Hodge filtrations on Hk(Xs) for all k. We give some details: Ω1 X/S is a vector bundle as f is assumed smooth. Then the relative de Rham complex is given by exterior powers, with a differential which is not OX-linear, hence this is not a complex of coherent sheaves. However, d is f −1OS -linear and thus we have a complex of f −1OS - modules, which pushes forward along f to a complex of OS -modules. Note that Deligne shows that the relative de Rham complex Ωan X/S is quasi- isomorphic to f −1OS an ⊗ C and uses this to show that R f∗Ωan X/S ≃ R f∗(C) ⊗ OS an, see Proposition 2.28 of [11]. This works if the base is any analytic space, and if f is just assumed smooth and separated. By Ehresmann's theorem we have diffeomorphisms Xs → X0 for every path from 0 to s, showing the cohomology sheaves are locally constant and if S an is simply connected Ri f∗Ωan X/S becomes Hi(Xs, C) ⊗ OS an. The stupid truncation F p = Ω≥p X/S is a subcomplex of sheaves of f −1OS an- modules. One then shows that the Ri f∗Ωan X/S are vector bundles (and in particular coherent sheaves). This implies that the Ri f∗F p = Ri f∗((Ωan X/S )≥p) are complexes of coherent sheaves. This can be done by trivialising the fibration locally and using Grauert's base change theorem, see [42]. The Ri f∗F p are moreover subsheaves of the Ri f∗Ωan X/S . This follows from degeneration of the Fröhlicher spectral sequence as the differentials in the Ri f∗ long exact sequence coming from Ω≥p → Ω≥p−1 → Ωp−1 vanish. Degeneration of the spectral sequence follows from a dimension count. Thus the map Pk sending S an to the filtration {Rk f∗F i} of Hk(Ωan X0 desired map to the flag variety and it factors through the period domain. ) is the Finally, we can globalize the construction by dividing out by the action of the fundamental group of S an. We use Ehresmann's theorem again to pull all our data along any path in the base in a homotopy invariant way, giving an action of the fundamental group of S an. 32 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Remark 4.1. These results are often stated for S smooth, but we need not X/S ≃ f −1OS an ⊗ C holds over an analytic space, as assume this. Deligne's Ωan do Grauert's theorem (see [20]), and Ehresmann's theorem (see Demailly's chapter in [6]). Grauert's theorem assumes that S is reduced (there is a version for a non- reduced base). But we may deal with non-reduced bases as part of our proof for the derived Poincaré Lemma, see Lemma 4.6 below. The map P is a priori not algebraic. In fact, to construct the map we needed to locally trivialize R f∗C ⊗ OS . In the analytic setting we can do this over any contractible set, so we just need S an to be locally contractible. In the algebraic setting this is typically impossible unless S is itself simply connected. Finally, let us have a look at the differential. Assume for simplicity that S is smooth. The differential of the period map dPp,k factors through the Kodaira- Spencer map B → H1(X, TX). In fact dP is the composition of Kodaira- Spencer with the map H1(X, TX) → Hom(F pHk(X), Hk(X)/F pHk) which is given by composing the natural cup product and contraction map with the natural quotient and inclusion maps: F pHk → H p,k−p(X) → H p+1,k−p−1(X) → Hk/F pHk(X). The period map satisfies Griffiths transversality, i.e. the image of the differential is contained in the subspace F p−1Hk(X)/F pHk(X). This can also X/S maps Fi to Fi−1 ⊗OS Ω1 be expressed as saying the connection on R f∗Ω∗ S . 4.2. The local derived period map I: The de Rham complex. We now define the derived version of the period map. In this section we work with derived schemes modelled on simplicial commutative algebras. We will consider a polarized smooth projective map f : X → S between geometric derived stacks. We fix the fibre Xs over some distinguished point s. This is a smooth projective variety. We assume that S is of finite presentation. Remark 4.2. We need our base stack to be locally almost of finite presentation to use analytification, and we need that it is quasi-compact to globalise our construction in Theorem 4.18. By definition the map f is smooth and projective if it representable, π0( f ) is a smooth and projective map of algebraic stacks and f is strong, i.e. THE GLOBAL DERIVED PERIOD MAP 33 πi(OX) (cid:27) πi(OS ) ⊗π0(OS ) π0(OX). Recall that smoothness of algebraic stacks can be defined locally as in Definition 1.3.6.2 of [66]. By polarized we just mean that π0( f ) is polarized. We will first consider the case that S is a quasi-separated derived scheme. We note that the definition of a derived scheme X includes that it has an underlying topological space, the same is true for its analytification. In both cases we denote the underlying topological space with the analytic topology by t(−). We will construct a map from open pieces of the analytification S an of S into the derived period domain U ⊂ Dn(V, Q)an, constructed in Theorem 3.22. Explicitly, we take V to be the complex H∗(Xs, ΩXs) with zero differentials. For Q we take the usual shifted bilinear form defined using product and trace. As we can only hope to construct the map locally we will now consider smooth maps iT : T → S an such that π0(T ) is a contractible Stein space. Locally this is always possible as every point in a complex space has a contractible Stein neighbourhood, see e.g. Theorem 3.1 in [39]. We may also pick these neighbourhoods to be relatively compact in S an (i.e. with compact closure). We write jT : XT → Xan for the pullback of Xan to T . This is in particular a smooth map. More generally, if S is a stack we pick a good hypercover T• of S an, i.e. an effective epimorphism T• → S an such such that all t(Tn) are disjoint unions of contractible spaces. This is possible by picking an affine hypercover first, analytifying and then using again that any Stein space is covered by contractible Stein spaces. Using Proposition 7.2.1.14 in [36] we may check that the map is an effective epimorphism by checking the truncated map is an epimorphism. Then we have S an ≃ hocolimI Ti by Lemma 3.6 in [45]. We will take care to perform our constructions compatibly with base change so that we can glue together the derived period map in Section 4.4. So let us consider the local situation. We will often consider the affinization of T , i.e. the derived affine scheme R Spec(O alg(T )) and denote it by uT . Abusing notation we will use f also for f an and f anXT when there is no ambiguity. Now by Theorem 3.9 one way to define a map T → U would be to have a map S → Dn(V, Q) such that its analytification satisfies the positivity conditions. Of course as long as S is algebraic this is not realistic, but we will construct a map S → QFilt as one ingredient in our construction. 34 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN In what follows we need to explicitly consider coherent sheaves on derived analytic spaces like T . Recall from Section 2.2 that these are just O alg T - modules such that the cohomology sheaves are locally coherent over π0(O alg T ). Moreover the underlying ∞-topos is the ∞-topos of t(T ). To compute derived functors we use the local model structure on presheaves. On stacks we use homotopy cartesian sections for a cover as described in Section 3.2. In particular we can apply the usual direct and inverse image functors between sheaves of complexes, on t(T ). There is extra structure on the space of functions on a derived analytic stack, but it will not affect the constructions we need. The crucial ingredient for Hodge theory is the relative de Rham complex. This is defined as the exterior algebra of the relative cotangent complex (which is a simplicial sheaf). The relative cotangent complex should be available in every good theory of derived analytic spaces and treatment of the analytic cotangent complex can now be found in [47]. For our purposes we will avoid the analytic theory and just use the analytification of the algebraic cotangent complex. (At least for Deligne-Mumford stacks this agrees with the analytic cotangent complex, see Theorem 5.20 of [47].) X/ f ∗Ω1 X/T ≔ j∗ T ((Ω1 X/S = Ω1 T h∗ XΩ1 T ; we define Ω1 T → Xan where hX is the natural map Xan → X. Thus in our set-up we may use the analytification of the relative algebraic cotangent complex. As f is smooth we think of this as the complex of Kähler differentials. So we are considering Ω1 S and restrict X/S )an) = j∗ to Xan X/S for any inclusion jT : Xan For definitions of the algebraic cotangent complex L, see [66]. We recall some basic local facts: Let f ∗ : A → B be a map of simplicial algebras. The relative cotangent complex LB/A fits into an exact sequence B ⊗L A LA → LB → LB/A. It satisfies base change, see Proposition 1.2.1.6 in [66]. For an explicit construction replace B by an algebra that is free over A, see Section 4 of [19]. In particular then each Bn is a free algebra over An. We know that LA is the simplicial A-module that is given by the Kähler differentials Ω1 in degree n. As B is free over A we have cofibrations of Bn-modules An Bn⊗An Ω1 and as the construction of Kähler differentials is functorial An in pairs of algebras we find that LB/A is the simplicial B-module that is Ω1 . If Bn Spec(B) over Spec A is smooth of dimension n then Ωn+1 ≃ 0 canonically and we may modify Ω∗ to assume this is an equality. We shall do so from now on. For a smooth morphism LB/A is perfect, see Lemma 2.2.2.5 in [66]. in degree n. We write Ωi /(Bn ⊗An Ω1 An ) = Ω1 Bn/An → Ω1 Bn for ∧iΩ1 Bn/An Bn/An THE GLOBAL DERIVED PERIOD MAP 35 We can thus define a chain complex of simplicial modules which is Ωi Bn/An in bidegree (i, n). Explicitly we have the following. Here each d is Ai-linear. d → Ω1 f ∗ 0 → B0 ↑↑ A0 ↑↑ A1 → B1 → Ω1 ↑↑↑ A2 → B2 → Ω1 ... ↑↑↑ ... d → Ω2 B0/A0 ↑↑ → Ω2 B1/A1 ↑↑↑ → Ω2 B2/A2 ... → . . . → . . . → . . . B0/A0 ↑↑ B1/A1 ↑↑↑ B2/A2 ... We denote the total complex of the associated double complex of N(A)- modules by ΩB/A. Remark 4.3. Note that we are indeed taking the exterior algebra in every degree, rather than taking the symmetric algebra in odd degrees, cf. [29]. Remark 4.4. A deeper study on differential forms in derived algebraic geometry is done in [41]. However, for a smooth morphism we do not have to worry about the subtleties dealt with in that work, cf. Section 1.2 of [41]. Lemma 4.5. The cotangent complex Ω1 X/T defined above is a coherent sheaf on Xan. Restricting to the underived setting we recover the usual cotangent complex for a smooth map, i.e. the sheaf of relative holomorphic differentials. Proof. The first claim is true as analytification preserves coherence, see [45]. Next note that holomorphic differentials are the analytification of the sheaf of Kähler differentials, and derived analytification is compatible with truncation, see Section 7 of [45]. (cid:3) For the next lemma we will consider the double chain complex of sheaves associated to the simplicial chain complex (Ω∗ X/T )•. We will denote its total complex by ΩX/T . This is a complex of N( f −1OS )-modules. When f is smooth we have the following derived Poincaré lemma: Lemma 4.6. Let T be an open subset of the analytification of a derived scheme. Then ΩX/T as constructed above is a resolution of N( f −1O alg T ). Proof. The question is local, so let us first assume X → S is a smooth map of derived affine schemes, and let iT : T → S an be an open subspace such that π0(T ) is contractible Stein and relatively compact. We consider the map A = O(S ) → B = O(X) of simplicial algebras. 36 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN B/A of N(A)-modules. We claim it is a There clearly is a map e : N(A) → Ω∗ quasi-isomorphism after base-change to O(T ), i.e. after analytification and restriction to T . We note that iT : T → S an is flat as T is relatively compact in S an, see Lemma 8.13 of [46]. Then ι = hX ◦ iT is flat since hX is flat, see Corollary 5.15 of [45]. So we aim to show that ι∗(e) is a quasi-isomorphism. We consider n, the ideal of nilpotents and elements of positive degree in N(A). By abuse of notation we will use the same name for the corresponding ideal in A. This is a simplicial ideal and A/n is reduced and underived. We also consider the ideal nB = n ⊗A B in B. Note that B ⊗A A/n is also reduced and constant (as a simplicial algebra) as we assume f smooth. The ni form an exhaustive filtration of N(A). As Ω∗ B/A is an N(A)-complex we can filter it by the ni, too. (As Ω∗ B/A⊗N(A) ni are subcomplexes.) We aim to show that the associated graded map of ι∗(e) induces a quasi-isomorphism, which shows that ι∗(e) itself is a quasi-isomorphism. As ι is flat we can equivalently show that the associated graded pieces of e become quasi- isomorphisms after applying ι∗. Write A′ for A/n and B′ for B/(n ⊗A B). Now we claim the following: • N(A′) = N(A)/n and Ω∗ B′/A′ become quasi-isomorphic when applying ι∗, • Gri nN(A) and Ω∗ B′/A′ ⊗N(A′) Gri nN(A) become quasi-isomorphic when applying ι∗, • Gri n(Ω∗ B/A) ≃ Ω∗ B′/A′ ⊗N(A′) Gri nN(A). The first claim follows from the usual underived Poincaré lemma, see for example Deligne [11], observing that Spec(B′)an → Spec(A′)an is a smooth map of analytic spaces. Then we base change to Tred and apply the exact global sections functor. The second claim follows from the first since Ω∗ pullback is compatible with tensor products. B′/A′ is flat over N(A′) and So let us turn to the third claim. Here we will have to work simplicially. We begin with i = 0 and show that Ωm B/A is Ωm A. So we can consider the natural map B′/A′. Recall that Ωm B /Ωm−1 B ⊗A Ω1 B/A/n (cid:27) Ωm Ωm B g → Ωm B′ → Ωm B′/(B′ ⊗A′ Ω1 A′) The map is clearly surjective and we claim the kernel is given by n ⊗A Ωm Ωm−1 B ⊕ B ⊗A Ω1 A. THE GLOBAL DERIVED PERIOD MAP 37 It follows from the short exact sequence n/n2 → Ω1 B + Ω1 is given by dnB + g−1(Ωi−1 So we have shown Ωm differential associated complexes of N(A′)-modules, Ω∗ B′ ⊗A′ Ω1 B/A/n (cid:27) Ωm A′) = n ⊗ Ωm B′/A′ as simplicial A′-modules. The is compatible thus we get a quasi-isomorphism with the B → Ω1 A ⊗ Ωm−1 B′ that the kernel as dn ⊂ Ω1 A. B B/A/n ≃ Ω∗ B′/A′. To consider the other associated graded pieces we observe that for any A- modules M we have (InA ⊗A M)/(In+1A ⊗A M) (cid:27) (InA/In+1A) ⊗A/I M/I. This is valid in the simplicial setting as it holds degree by degree. Again we deduce the quasi-isomorphism of complexes. Together these claims give the desired quasi-isomorphism on the associated graded modules and we have locally shown the quasi-isomorphism N( f −1O alg (cid:3) X/T of sheaves. T ) ≃ Ω∗ We could now consider the derived push-forward R f∗ΩX/T and it is not hard to deduce that R f∗C ⊗ O alg X/T . However, to define the bilinear form it is more convenient for us to forget the differentials and work with Ω# X/T , 0). Thus, instead of using Lemma 4.6 we deduce the main consequence of the Poincaré Lemma together with Hodge-to-de Rham degeneration from the classical case as follows: T ≃ R f∗Ωan X/T ≔ (Ω∗ Lemma 4.7. Let Ω# differential. Then R f∗ f −1O alg X/T = ⊕Ωi T ≃ R f∗Ω# X/T . X/T [−i] be the de Rham algebra without its X/T X/T ) (cid:27) Hk( f −1OS ) ⊗H0( f −1OS ) H0(Ωi Proof. There is a natural map induced by the inclusion f −1O alg T → Ω# and we need to check it is a quasi-isomorphism. As f is smooth, Ωi X/S is a projective OX-module [66, Definition 1.2.7.1] and in particular strong [66, Lemma 2.2.2.2], thus, after analytifying and restricting, Hk(Ωi X/T ) (cid:27) Hk(OX) ⊗H0(OX) H0(Ωi X/T ). Moreover, OX is a strong f −1OS -module and we have Hk(Ωi X/T ). We are in the analytic setting but do not need to worry about completing the tensor product as we work with finitely presented modules. The E2-term of the natural Grothendieck spectral sequence computing Rk f∗Ω# T )). As f is proper we apply X/T and f −1O alg the projection formula to the complexes Ωi 2 (cid:27) Rp f∗(H0(Ωi T ). For the projection formula one usually assumes bounded below complexes of sheaves (with the cohomological grading) and the structure sheaf is not bounded below in general, but because the fibres of our map of topological spaces are locally finite-dimensional the theorem remains true for unbounded complexes, see §6 of [56]. X/T is thus Rp f∗(H0(Ωi X/T ) ⊗ Hq( f −1O alg X/T )) ⊗ Hq(O alg to find E pq T 38 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Similarly on the left hand side we have a spectral sequence with E pq Rp f∗(H0( f −1O alg since H0( f −1O alg classical Hodge theory. 2 = T ). The natural map induces an isomorphism X/T ) is just (cid:3) T )) ⊗ Hq(O alg T ) ≃ H0(ΩX/T ) and R f∗H0(ΩX/T ) ≃ R f∗H0(Ω# Using the product on Ω# sheaf of algebras. X/T and suitable resolutions we may assume this is a Lemma 4.8. Let f : X → S be a smooth projective morphism of derived schemes and consider an open derived Stein T ⊂ S an and Ω# X/T as above. Assume moreover that t(T ) is simply connected. Then there is a quasi-isomorphism of O alg R f∗(Ω# R f∗C ⊗ O alg base change. T → T → T . We write ¯w for ¯v ◦ ϕ. These maps are compatible with smooth T -algebras ¯v : R f∗C ⊗ O alg T -modules ϕ : V ⊗ O alg X/T ) and a quasi-isomorphism of O alg Proof. By Lemma 4.7 we have a weak equivalence R f∗ f −1O alg X/T . is proper with locally finite dimensional fibres we may use the As f projection formula to find R f∗C ⊗ O alg T . We call the T composition ¯v. The tensor product here is over the constant sheaf and can be considered underived. ≃ R f∗ f −1O alg T → R f∗Ω# We may compute the natural map in the category of sheaves of algebras and thus arrange that ¯v is a homomorphism. Now R f∗C only depends on the underlying topological space, hence it is a homotopy locally constant sheaf on t(T ) by the classical result. By Corollary 3.25 this is actually formal, i.e. quasi-isomorphic to the graded local system of its cohomology sheaves. But as t(T ) is simply connected this must be a constant sheaf, which is of course quasi-isomorphic to V. Thus we let ϕ be Deligne's map from 3.24, tensored with O alg T . X/T ) → R f ′ We need to check ¯w is compatible with base change, i.e. given a smooth r : T ′ → T the quasi-isomorphism ¯w′ constructed from T ′ → S an is equivalent to r∗ ¯w under the natural transformations r∗(V ⊗ O alg T ) → V ⊗ O alg T ′ and r∗(R f∗Ω# X′/T ′). These are weak equivalences by base change for the cotangent complex and base change for derived schemes (Proposition 1.4 of [62]). The equivalence of r∗ϕ and ϕ′ is clear as everything is compatible with the Lefschetz operator. Compatibility of ¯v follows from nonabelian base change for locally compact Hausdorff spaces, see Corollary 7.3.1.18 in [37]. (cid:3) ∗(Ω# THE GLOBAL DERIVED PERIOD MAP 39 Our next ingredient is the bilinear form which we construct algebraically. It will follow directly from the existence of the trace map on f∗Ω# X/S . Lemma 4.9. Given f derived scheme there is a natural trace map R f∗(Ω# it is compatible with base change. : X → S smooth and proper and such that S is X/S ) → OS . In particular Proof. This is essentially Grothendieck-Verdier duality. First we observe that by the adjoint functor theorem under mild conditions there is an exceptional inverse image functor, i.e. a right adjoint to R f∗, denoted f !. See Section 6.4 in [34]. Secondly, we need a comparison map between differential forms and f !OS . We construct a map Ω# X/S → f !OS following the argument in the underived case given by Neeman in Construction 3.1.5 of [40]. (Note that for our purposes we do not need to check that this map is a weak equivalence.) The construction uses base change for the exceptional inverse image to formally construct a map L∆∗R∆∗OS → f !OS (where ∆ : S → S × S is the diagonal). The derived version of the necessary base change theorem is proved in Proposition 6.4.2.1 of [34]. As f has finite-dimensional fibres it is locally of finite Tor-amplitude. To complete the construction one uses the Hochschild-Kostant-Rosenberg theorem identifying Ω# X/S with L∆∗R∆∗OS . The derived HKR theorem over an affine base can be found in Corollaire 4.2 of [67]. The statement is over an underived base, but the proof works the same way over a derived ring A. To globalise from a derived affine scheme S = R Spec(A) to the case where S is a scheme, and in fact an Artin stack, we may use Proposition 1.14 of [41] which shows that global sections of the de Rham algebra may be computed as a homotopy limit of the de Rham algebras of affine pieces. It is easier to see that the construction of Hochschild homology globalises similarly: We can check directly that L∆∗R∆∗OS is compatible with base change. Compatibility of the trace map with base change follows as in Reduction 3.2.5 of [40], using again the derived base change from [34]. (cid:3) Lemma 4.10. Given f : X → S smooth proper and polarized and such that S is a derived scheme there is a natural non-degenerate shifted symmetric bilinear form QF on the perfect complex F0 = R f∗(Ω# X/S ). It reduces to the classical pairing induced by the trace in the non-derived truncation. QF is compatible with base change. 40 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN is sufficient to check that X/T , Qan Proof. We compose the trace map from Lemma 4.9 with the natural product on F0 to define QF. By construction this enhances the classical bilinear form. To check non-degeneracy of QF it the analytification is non-degenerate locally. We will compare Qan F on T ⊂ S an to Q ⊗ O alg T which is clearly non-degenerate. In other words we show that ¯w from Lemma 4.8 induces a weak equivalence (V ⊗ O alg T ) ≃ (R f∗Ω# F ). The two bilinear forms agree in the reduced case by classical results, see [54] for a comparison of trace pairing and Hodge pairing. It follows that ¯w preserves the bilinear form on the subcomplex V. Since ¯w, Qan T are all compatible with the OS -structure this gives the desired result. Finally, for base change we need to pick g : S ′ → S and compare g∗QF with QF′ where F ′ is the pushforward of the de Rham algebra on X ×S S ′. This follows from naturality of the trace map. (cid:3) F and Q ⊗ O alg T , Q ⊗ O alg 4.3. The local derived period map II: The Hodge filtration. To construct the Hodge filtration we look at the inclusion of sheaves given by the stupid truncation of the relative de Rham complex, F i ⊂ F i−1 ⊂ ΩX/T . We first consider the Hodge filtration without trivialising the de Rham complex. As our constructions are compatible with base change we may work directly over a stack. Lemma 4.11. Given f : X → S smooth proper and polarized and such that S is derived Artin stack. Then the filtered de Rham complex induces a map φ : S → QFilt. Proof. We sketch how to define the pushforward map on stacks. (See the proof of Proposition 2.1.1 in [18] for the ∞-categorical analogue.) We choose an affine hypercover {S i}i∈I of S which induces a cover {Xi = X ×S an S an is smooth. Now the pushforward maps on the Xi give a pushforward map of presections. The pushforward is a homotopy cartesian section by base change for quasi- compact and quasi-separated maps of derived schemes, see Proposition 1.4 of [62]. i } of Xan. All Xi are derived schemes as f Base change also implies that the pushforward construction is well-defined. • of S an we may find a common Given two different affine covers S • and S ′ refinement by taking a hypercover of the fibre product. Thus it is enough to check that the construction is invariant under refinements of the cover • → S • → S an. Base change provides canonical (homotopy u : S ′ THE GLOBAL DERIVED PERIOD MAP 41 coherent) comparison maps between the pullback of a pushforward on S an and the pushforward constructed on the refinement, which are moreover weak equivalences. Viewing sheaves as homotopy cartesian sections these maps give weak equivalences. The pushforward we have constructed preserves fibrations (as the adjoint preserves object-wise cofibrations), so it is Quillen. Thus we may define Fi ≔ R f∗(F i) and have a diagram Fn → Fn−1 → . . . → F0 of homotopy cartesian presections on S • which may be interpreted as complexes of sheaves on S . To see the Fi are perfect consider the pushforward R f∗Ωm X/S on S . As the pushforward of a perfect complex by a smooth proper map this is again perfect by Lemma 2.2 of [62]. We check locally that F • satisfies the conditions of Lemma 3.15. Let us assume S = R Spec(A) is affine and abuse notation by writing Fi for RΓ(Fi). It is clear that we can replace the Fi by a diagram of inclusions of cofibrant objects. Then we need to check the following: (1) All maps induce injections on H∗(− ⊗L A H0(A)), and the H∗(Fi ⊗L A H0(A)) are locally free sheaves. (2) There is a bilinear form QF. (3) The Fi satisfy orthogonality with respect to QF. The first statement follows from classical Hodge theory. Writing eS for the inclusion π0(S ) → S we can use base change to show that e∗ X/S is locally free because it agrees with R f∗e∗ π0(X)/π0(S ), which is well-known to be locally free in classical Hodge theory. We argue similarly for the injections. X/S = R f∗Ω≥p S R f∗Ω≥p XΩ≥p To define the shifted bilinear form we will be very explicit about our replacements. Some care is needed since fibrant cofibrant replacements are not strictly monoidal, and for technical reasons we have set up a strict orthogonality condition, rather than one up to homotopy. We choose an affine cover of X and the injective model category of presections as a model for sheaves of OX-modules, see the discussion before Corollary 3.18. Object-wise tensor product makes this into a symmetric monoidal model category, the pushout-product axiom holds since cofibrations are defined object-wise. We use fibrant cofibrant replacement using functorial factorization, denote this functor by P. We also consider the transfer model category structure on algebra objects, see e.g. [19], and we denote fibrant 42 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN cofibrant replacement with respect to this structure by Palg. Fibrant cofibrant objects are preserved by the forgetful functor. Next we may replace F i by a diagram of cofibrations using functorial factorization, call it Q′(Fi), and then by a diagram of cofibrations between fibrant cofibrant objects, write this as P′F i. For simplicity we write ⊗ for the derived tensor product over the structure sheaf for the rest of this section. We observe that by the monoid axiom the diagram (i, j) 7→ Q′F i ⊗ Q′F j is actually cofibrant in the projective model structure, and similarly for P′. Thus the weak equivalences of diagrams Q′(F i) ⊗ Q′(F j) → F i ⊗ F j ← Q(F i ⊗ F j) give rise to a lift Q′(F i) ⊗ Q′(F i) → Q(F i ⊗ F j) as the second map is a trivial fibration. Similarly we have functorial quasi- isomorphisms P′(F i) ⊗ P′(F j) ← Q′(F i) ⊗ Q′(F j) → P(F i ⊗ F j) where the first map is a trivial cofibration, thus there are compatible lifts ηi j : P′(F i) ⊗ P′(F j) → P(F i ⊗ F j). : PF 0 → P′(F 0). Denote We also choose a quasi-isomorphism ǫ the product on F 0 by m and then define a multiplication on P′F 0 by considering ǫ ◦ Pm ◦ η00 : P′F 0 ⊗ P′F 0 → P(F 0 ⊗ F 0) → P(F 0) → P′(F0) This product vanishes on P′F i ⊗ P′F n+1−i, as we can factor it through ηi j and m vanishes on F i ⊗ F n+1−i as Ωn+1 is 0. We push forward both P′F i and PalgF 0 and functorially replace by a diagram of cofibrations again, written as Q′. The resulting diagram is our definition of Fi. Now, using the lifting property for cofibrant diagrams as above and adjointness of f∗ there are natural and compatible maps φi j : Q′ f∗P′F i ⊗ Q′ f∗P′F j → Q( f∗P′F i ⊗ f∗P′F j) → Q f∗(P′F i ⊗ P′F j) and we can compose φ00 with Q f∗(ǫ ◦ Pm ◦ η00) and a quasi-isomorphism Q f∗(P′(F 0)) → Q′ f∗(P′(F 0)) to define a product on F0 that is zero on Fi ⊗ Fn+1−i. The construction of the bilinear form is completed by the canonical trace from Lemma 4.10. Orthogonality for the Fi is clear since the product of sections in Fi and Fn+1−i vanishes. (cid:3) To compare with R f∗C we go back to the analytic setting. Lemma 4.12. Let f : XT → T and (V, Q) be as in the previous subsection, in particular T is derived Stein. Then we have a diagram of perfect THE GLOBAL DERIVED PERIOD MAP 43 complexes over A = N(O(T )) Fn → Fn−1 → . . . → F0 ←V ⊗ A where Fi ≃ RΓ ◦ R f∗(F i), the Fi are cofibrant A-modules and the maps Fi → Fi−1 are injective. Moreover, the conditions of Theorem 3.9 are satisfied, namely: (1) The last map ¯w is a quasi-isomorphism. (2) All other maps induce injections on H∗(− ⊗L H∗(Fi ⊗L A H0(A)) are locally free sheaves. A H0(A)), and the (3) There is a bilinear form Qan F on F0 which is compatible with ¯w and Q. (4) The Fi satisfy orthogonality with respect to Qan F . In other words we get an object of Dn(V, Q)(uT ). X/T is a pullback (along hS ◦ iT ) of the perfect complex R f∗Ωm T -modules on T -modules on T and then global sections and consider the diagram Proof. We apply the push-forward R f∗ of sheaves from f −1O alg XT to O alg of RΓ ◦ R f∗(F i) for i ≥ 0. These are dg-modules over N(O(T )). It is clear that we can replace the Fi by a diagram of inclusions of cofibrant objects. We first deal with some homotopy invariant properties which will not be affected by this. To show the Fi are perfect we apply Porta's GAGA result, see Lemma 4.13 below, to show that pushforward commutes with analytification, thus R f∗Ωm X/S , and thus perfect. Now Fi is an iterated extension of perfect complexes and itself perfect as a sheaf. As it is moreover globally of finite presentation we have a perfect O(T )-module, as follows from Lemma 8.11 of [46]. For the map in the first claim we use the weak equivalence ¯w = ¯v ◦ φ from Lemma 4.8. The second statement follows from GAGA together with Lemma 4.11. We define the shifted bilinear form using the yoga of fibrant cofibrant replacement as in Lemma 4.11, letting Fi be defined as the pushforward of the analytic Hodge filtration R f∗Ω≥i As we are working with fibrant cofibrant objects we also have a quasi- isomorphism F0 → f∗PalgF 0 and we can use Lemma 4.8 to define a second quasi-isomorphism f∗PalgF 0 → R, where R is a fibrant cofibrant replacement of R f∗C⊗O alg T . Both maps are compatible up to homotopy with the multiplication map. Denote the composition by v and its homotopy inverse by ¯v. As T is contractible there is a natural quasi-isomorphism X/T . 44 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN φ : V ⊗ A → RΓ(R f∗C ⊗ O alg Lemma 4.8. T ). Let ¯w be the composition of ¯v with φ, as in It follows from the proof of Lemma 4.10 that ¯w is compatible with the shifted bilinear forms. (cid:3) We needed Porta's GAGA theorem for potentially unbounded coherent sheaves in this proof. This is a slight strengthening of Theorem 7.1 in [45]. The following proof was communicated to us by Mauro Porta. Lemma 4.13. Let f and F a coherent sheaf on X. Then the natural map (R f∗F )an → R f an is an equivalence. : X → Y be a proper morphism of derived schemes F an ∗ Proof. One shows first that if f an : Coh♥(X) → Coh+(Y) has finite cohomological dimension, i.e. there exists n such that for all i > n and all F ∈ Coh♥(X) we have Ri( f an ∗ (F )) = 0, then the result holds for unbounded coherent sheaves. ∗ To see this one follows the proof of Theorem 7.1 in [45]. As in that proof one writes F as τ≤nF → F → τ>nF . Now one observes that by the ∗ (τ>kF )an. On the result in the bounded below case (R f∗(τ>kF ))an ≃ R f an other hand the assumption of finite cohomological dimension implies that ∗ (τ≤kF )an vanish above the cohomology groups of (R f∗(τ≤kF ))an and R f an degree k + n. Varying n one sees that the natural map (R f∗F )an → R f an F an is a quasi-isomorphism. To show that f an has finite coherent cohomological dimension, for every s ∈ S an choose a Stein open neighbourhood s ∈ U ⊂ S such that ¯U is compact U ≔ U ×S an Xan can in S an. Lemma 6.2 in [46] implies that base change Xan be covered by finitely many Stein opens. Now Xan U is separated (because f is proper and thus separated), so the intersection of Stein spaces remains Stein. We can compute cohomology via Cech cohomology attached to this Stein open cover. The open cover is finite, so the complex must be finite as well and f an has finite coherent cohomological dimension. (cid:3) ∗ Remark 4.14. We also note that the Fi are coherent. This is proven by induction, using that R f∗ is exact and coherent sheaves are a stable subcategory of all sheaves. The crucial ingredient is that the R f∗Ωm X/T are coherent as pushforwards of coherent sheaves. It is shown in Theorem 6.5 of [45] that push-forward preserves bounded below coherent complexes. As Porta explained to the authors, the same arguments as in Lemma 4.13 extend the result to the unbounded case if the morphism has bounded cohomological dimension. THE GLOBAL DERIVED PERIOD MAP 45 The lemma says that any X → T gives a map from u(T ) to the algebraic moduli stack Dn(V, Q). To proceed we have to change our target to the analytification of Dn(V, Q). We then check the map factors through the substack U. Note that here we need the polarization in order to define U. Proposition 4.15. If T ⊂ S an is a simply connected derived Stein space contained in the analytification of an affine subspace of S then there is a derived period map T → U, compatible with base change. Proof. To obtain a map T → Dan it is enough to construct a homotopy n coherent map from T to the diagram QFiltan n → Per f an ← ∗, see Remark 3.17. By Lemma 4.11 there is a map φ : S → QFilt representing the filtration on R f∗ΩX/S . There is clearly a constant map υ representing V ⊗OS . We may analytify both maps, compose with the natural maps to QPer f and restrict to T to obtain maps φ, υ : T → QPer f an. To complete our construction we will exhibit a homotopy equivalence η between these maps. As there is no higher homotopy coherence in a pullback diagram this will complete the construction. To construct η we consider a simplicial hypergroupoid P• with realisation QPer f . As analytification commutes with homotopy colimits this will also give a presentation Pan • of QPer f an. We will then construct a homotopy equivalence between simplicial maps uT → P• which gives η using Lemma A.12. As the Tor dimension of V is finite we may restrict ourselves to working with a geometric substack QPer f [a,b] of QPer f . To simplify notation we will still refer to this substack as QPer f . First we construct P•. We recall that the stack of perfect complexes with finite Tor amplitude has an atlas that represents presentations of perfect complexes as iterated extensions of vector bundles, see the discussion before Lemma 3.10 in [64]. Using the same methods together with the construction from Lemma 3.11 one sees that QPer f has an atlas U representing explicit quadratic forms on presentations of perfect complexes. Then we follow the proof of Proposition 4.5 in [51] to obtain P•. In loc. cit. a simplicial affine hypercover for an arbitrary simplicial stack Z is constructed. The general construction is very delicate, but, using Pridham's notation, we are only interested in the case where the simplicial stack Z is constant and thus ν(Z) = 1. This means that given an r-atlas W of QPer f we may define a simplicial affine presentation by Pi (cid:27) W ×i ×QPer f ×i QPer f . In particular P1 represents weak equivalences between presentations of perfect complexes with non-degenerate quadratic forms. 46 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We may assume R f∗Ωan X/S is an iteration of free vector bundles, shrinking T to lie in the analytification of a suitable affine cover of S if necessary. (As the Tor-dimension is finite we only have to shrink finitely often.) Thus the complex R f∗ΩX/T on uT is represented by a map uT → W, which we may view as a map from the constant simplicial affine uT to P•. We may represent R f∗C similarly. Then the map ¯w from Lemma 4.12 is a weak equivalence between the presentations and thus provides a map Spec(O(T )) → P1, which is enough to give a homotopy between maps from the constant simplicial object c Spec(O(T )) to P•. Using Lemma A.12 we have the desired homotopy T → Pan in simplicial derived Stein spaces, giving rise to υ ≃η φ : T → QPer f an, and thus to a map T → Dn(V, Q)an Classical Hodge theory lets us check the open cohomological conditions and we see that we have a factorization P : T → U → Dn(V, Q)an. Compatibility with base change is clear for the maps to QFilt and ∗. Compatibility for η is just compatibility for ¯w from Lemma 4.8. (cid:3) • 4.4. The global derived period map. Having constructed in the previous section a period map for any small patch T of S an, we now glue them together. We first redo Lemma 4.12 for a map between derived stacks. We consider perfect complexes on S an as fibrant cofibrant homotopy cartesian sections as in the discussion preceding Corollary 3.18. : Xan → S an an analytification of a smooth Lemma 4.16. Given f projective map of derived Artin stacks and with (V, Q) as above, we have a diagram of perfect complexes on S an Fn → Fn−1 → . . . → F0 → R where R is a fibrant cofibrant model for R f∗C ⊗ O alg S an, the Fi are fibrant and cofibrant models for R f∗(F i), and the maps Fi → Fi−1 are injective. F on F0 and the conditions of Theorem Moreover, there is a bilinear form Qan 3.9 are satisfied. These data look very similar to an object in Dn(V, Q)(S ), except that we have replaced V by a sheaf on S an. Proof. As in Lemma 4.11 we construct pushforwards as homotopy cartesian sections. We choose an affine cover {S i}i∈I of S which induces a cover {Xan i give a pushforward map of presections and again it follows from base change that i } of Xan. The pushforward maps on the Xan i = Xan ×S an S an THE GLOBAL DERIVED PERIOD MAP 47 the pushforwards are homotopy cartesian. To be precise, for R f∗F i we use Proposition 1.4 of [62]. As analytification commutes with pushforward, see Lemma 4.13, base change for quasi-compact and quasi-separated maps also holds true in analytic geometry. For R f∗C we use nonabelian base change for locally compact Hausdorff spaces, see Corollary 7.3.1.18 in [37]. As in Lemma 4.11 base change also implies that the pushforward construction is well-defined. Thus we may compute Fi as in Lemma 4.12, going through the same yoga of fibrant cofibrant replacement, and obtain a filtration F ∗ of F0 and a product on F0. Similarly we compute R. We use Lemma 4.8 we obtain a map of homotopy cartesian sections ¯v : R → F 0. This is well defined as ¯v is compatible with refinement. This map is a weak equivalence as it is one locally, so we may find an inverse v as all objects are fibrant cofibrant. Using the product and the trace from Lemma 4.10 we can define Qan F compatibly with v. All the conditions may be checked locally. (cid:3) We now need to combine this with the monodromy action on cohomology. We will write V for the graded vector space V = H∗(Xs, Ω∗ ) with integral Xs structure VZ = H∗(Xs, Z) and bilinear form Q, equipped with the canonical action of Γ = Aut(VZ, Q). By abuse of notation we will also denote by V the corresponding sheaf on any space whose fundamental group maps to Γ. We first state and prove our main theorem for the case that the base S is a derived scheme. Proposition 4.17. A polarized smooth projective morphism f : X → S of derived schemes, where S is connected, quasi-separated and locally almost of finite presentation, gives rise to a derived period map P : S an → U/Γ, where U is the derived period domain for the pair (V, Q). Proof. We will choose a suitable simplicial derived Stein space T• with T• ≃ S an, cf. the discussion at the beginning of Section 4.2. Then we construct a map uT• → DΓ such that the associated map to Dan Γ factors through U/Γ, using the description in Proposition 3.28. There is a map γ : BΓ → QPer f with γ∗U = V for the universal perfect complex U . Here the constant stack BΓ is just the stackification of the constant presheaf BΓ. By the proof of Proposition 3.28 it suffices to construct maps φ : S an → QFiltan and υ : S an → BΓan together with a homotopy between the two compositions S an → QPer f an. These maps will represent R f∗(F •) and V , respectively. 48 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN We pick a good hypercover T• of S an. By this we mean T• → S an is an effective epimorphism and if we denote by (Ti)i∈I the diagram of components of T• then all t(Ti) are contractible. Next we show the monodromy induces a canonical map of derived analytic stacks S an → BΓan. We know hocolimI ∗ ≃ t(S an) by the main result of [15], thus R f∗C gives a sheaf on hocolimI ∗. By formality we may write it as a graded local system classified by a map hocolimI ∗ → BΓ and for a model hocolimJ ∗ of BΓ we may induce this by a map I → J. Thus, via the maps uTi → ∗ of derived algebraic schemes, we obtain a map uTi → BΓi of simplicial derived affine schemes. Applying Lemma A.12 and taking realizations we obtain: κ : hocolim Ti → hocolim I ∗ → hocolim J ∗ → BΓan. Note that we are comparing homotopy colimits in simplicial sets and derived stacks. But the inclusion of simplicial sets as constant stacks preserves homotopy colimits since it is a left Quillen functor into the local projective model structure on simplicial presheaves, which is a model for derived stacks. We let γ be the map BΓ → QPer f representing the Γ-representation V with its quadratic form and let υ : S an → QPer f an be the map induced by γan ◦ κ. We also define a map φ = (π ◦ φ)an : S an → QFiltan → QPer f an. Here π is just the natural forgetful map and φ is the map from Lemma 4.11. It remains to produce a homotopy η from φ to υ. We have a comparison morphisms v : R f∗Ωan X/S → R f∗C ⊗ OS an from Lemma 4.16, which we may interpret as a strict morphism of homotopy cartesian sections after pulling back to T•. On the other hand we can trivialise R f∗C ⊗ OS on T•. This gives an object in the abelian category of diagrams of chain complexes, equipped with a Lefschetz operator, so Theorem 3.24 provides a strict quasi-isomorphism of homotopy cartesian sections ϕ : H(R f∗C) ⊗ O alg . By construction there is also T• a quasi-isomorphism λ : V ⊗ O alg T • which is compatible T• with base change as transition functions for V are induced by those for H(R f∗C). → H(R f∗C) ⊗ O alg → R f∗C ⊗ O alg T• We may thus consider the composition of v and homotopy inverses of ϕ and λ as a morphism of perfect complexes over O(T•) and unravelling we obtain exactly a map p : uT• ×∆1 → P•. (Recall from the proof of Proposition 4.15 that Pi represents presentations of i + 1 perfect complexes with quadratic forms and explicit quasi-isomorphisms between them.) THE GLOBAL DERIVED PERIOD MAP 49 Thus we obtain a homotopy between two maps uT• → P• in simplicial derived affine schemes. By Lemma A.12 this gives the desired homotopy between maps S an ≃ T• → Pan • ≃ QPer f an, completing the construction of the period map. The map is independent of our choice of T•. Recall that it suffices to check it is well-defined under refinement. Thus, given a smooth map r : T ′ • → T• between good hypercovers we claim that the homotopy cartesian section • × ∆1 → P• obtained by the above construction applied to T ′ → S is p′ : uT ′ equivalent to r∗ p. The map r induces natural weak equivalences between the objects by the usual base change theorem. The morphism λ is compatible with base change by construction. The same is true for ϕ as the Lefschetz operator is compatible with r. For the morphism ¯v we refer to Lemma 4.16. Moreover, everything is functorial with respect to maps in T•. (cid:3) Theorem 4.18. Let f : X → S be a polarized smooth projective morphism of derived geometric stacks where S is connected and of finite presentation. Then there is a derived period map P : S an → U/Γ, where U is the derived period domain for the pair (V, Q). Proof. We can write S ≃ hocolim S i for a simplicial affine scheme S i, e.g. using [51]. Then S an ≃ hocolim S an i since analytification of stacks is defined as a left Kan extension. Using contractible Stein hypercovers of the S an i we may also write S an ≃ hocolim j∈J T j. The map f induces maps fi : X ×S S i → S i, and since f is smooth and projective so are the fi. Moreover they all have homeomorphic fibres, so we can fix V, Q and n, and thus the derived period domain U. We know that R f∗ f −1O alg S an as this is true locally by Lemma 4.8. It remains to show that R f∗C is quasi-isomorphic to a locally constant sheaf, i.e. it is obtained by pullback from some sheaf on BΓ. We have to be somewhat careful as we pass from the stack S an to the topological space hocolim t(Ti) and, unlike in the case of a derived scheme, sheaves on S an cannot be described as sheaves on hocolim t(T j) in general. So we first show that R f∗C is a pull-back of an infinity local system on hocolimJ t(T j) ≃ hocolimJ ∗, and then that this infinity local system is a plain (graded) local system. S an is quasi-isomorphic to R f∗C ⊗ O alg We know from Lemma 4.8 that R f∗C is quasi-isomorphic to V on every T j, thus R f∗C ⊗ O alg is just the pull back of V along u(T j) → ∗. This T j induces transition functions between the V which are quasi-isomorphisms. Thus we obtain a homotopy cartesian section C of the constant diagram indexed by J that sends every j to the model category of chain complexes. 50 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN By strictification (e.g. Theorem 1 in [26]) the category of these homotopy cartesian sections is equivalent to the homotopy limit of the constant diagram J → dgCat that sends every j to the dg-category Ch of (fibrant cofibrant) chain complexes. Now we recall that there is a co-action of simplicial sets on any model category, written (K, D) 7→ D K, see Chapter 16 of [23]. This is a Quillen bifunctor, and thus holimJ Ch ≃ ChhocolimJ ∗ ≃ ChhocolimJ t(T j). (Here we use that all the t(T j) are contractible.) Thus C can be considered as an element of Chhocolim t(T j), which is the category of infinity local systems on hocolim t(T j), see [25]. (To apply the result of [26] we need to consider a diagram indexed by a direct category, without infinite ascending chains of morphisms. By Lemma 3.10 of [51] the simplicial derived scheme S • is determined by a finite truncation. Moreover, as S is quasi-compact it is enough to consider finitely many S i. For each i we can choose the hypercovers T (i) to be bounded. Restricting to non-degenerates we may thus assume that the diagram J is in fact finite.) j of the Stein spaces S an i Now C can equivalently be considered as a homotopy locally constant sheaf on hocolim t(T j), by Theorem 12 of [26]. (The considerations above make sure that hocolim t(T j) is equivalent to a homotopy colimit of a finite diagram of points, and thus satisfies the conditions of that theorem.) Thus C lives in the bounded derived category of an abelian category, namely sheaves of abelian groups on hocolim t(T j). We can apply Theorem 3.24, since throughout all the equivalences C kept its Lefschetz operator. We deduce that C is a direct sum of its cohomology groups. So it is in fact a local system, and given by a representation V of the fundamental group of hocolim t(T j), which we may view as a sheaf pulled back via a map κ : hocolim T j → hocolim J ∗ → BΓan as in Proposition 4.17. We note that all automorphisms preserve the extra structure on V, so the classifying map of V still factors through Γ = Aut(VZ, Q). Thus R f∗C ⊗ O alg ≃ (γ ◦ κ)∗U T• in the homotopy category. is weakly equivalent to V ⊗ O alg T• To complete the proof we proceed as in Proposition 4.17. We recall that Lemma 4.11 provides a map from S to QFilt, and there is a natural map ¯w : V ⊗ O alg T• → H(R f∗C) ⊗ O alg T• → R f∗( f −1O alg T• ) → R f∗ΩX/T•. The map is a quasi-isomorphism as we can check locally by Lemma 4.8. We may consider a homotopy inverse and obtain w : F0 → V ⊗ O alg as a T• strictly compatible map of homotopy cartesian sections. This gives a map THE GLOBAL DERIVED PERIOD MAP 51 uT• × ∆1 → P• and we conclude by Lemma A.12 that this is the desired homotopy. We observe that the map to DΓ we have constructed factors through U/Γ by classical Hodge theory. The proof that this map is well-defined is just as in Proposition 4.17 as we have considered compatibility with smooth base change throughout. (cid:3) The following example is crucial: Example 1. The universal example is given by taking S to be the moduli stack of polarized schemes described in Example 3.39 of [50], or rather any quasi-compact component M of its substack with smooth fibres . This is a derived 1-geometric stack and has a universal family X → M. It follows from [50] that M is locally almost of finite presentation, so by Theorem 4.18 we have a derived period map M → UΓ. 4.5. Comparison with the classical period map. We now need to check that our construction recovers the usual period map in the underived setting. This is covered by the following theorem. : X → S a polarized smooth projective map Theorem 4.19. Consider f of derived schemes and denote the product of the classical period maps associated to π0π0( f ) by P : π0π0(S an) → π0π0(U). Then P = π0π0P. Proof. It is enough to check this locally, so let us replace S an by T and assume T is derived Stein. We may replace the target by Dan, and we know from Theorem 3.9 that π0π0Dan is the product of the usual closures of the period domains. Then we consider the map π0(P) : π0T → π0π0Dan, where P is the derived period map. We recall that the period map classifies the Hodge filtration on V, i.e. it is given by a map π0(T ) to the (underived) analytic flag variety. We are now in the underived and non-stacky situation where the technical difficulties we were dealing with before disappear. In particular it follows from inspecting the explicit construction that the analytic flag variety is the analytification of the flag variety Fl (which is indeed a smooth variety rather than an Artin stack). Moreover the period map π0(T ) → Flan arises via an honest adjunction from π0(T )alg → Fl. We may construct the flag variety as a pull-back analogously to the derived version. The period map is then defined by θ : π0(T ) → π0π0(QFilt) and a natural isomorphism ǫ between π ◦ θ and the canonical map V : π0(T ) → π0π0(QPer f ). It is clear that π0π0( φ) (cid:27) θ as it is the analytification of the 52 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN algebraic classifying map. To compare π0π0(η) and ǫ we note that both η and ǫ arise via adjunction from maps classifying corresponding modules on uT respectively π0(uT ). Thus we see that π0π0(P) recovers the classical period map. (cid:3) Together with the next section this shows that the map we construct deserves to be called the derived period map. This also shows that the underlying underived map of P is smooth if S is smooth, but it is clear that P is not strong, so it is not smooth in the sense of derived analytic geometry. 4.6. Differential of the period map. In this section we will compute the differential of our period map P by identifying it with the differential of the infinitesimal period map of derived deformation theories considered in [13, 16, 17]. We will not define derived deformation theory here, the reader unfamiliar with it may skip ahead to the concrete description of dP in Corollary 4.22 and its consequences. We recall that in [17] for a smooth projective manifold X a period map is defined which goes from the derived deformation functor associated to the Kodaira-Spencer L∞-algebra KS X to derived deformations of an L∞-algebra E associated to the de Rham complex of X. To be precise E is given by the dg-Lie algebra REnd(RΓ(Ω∗ X))[−1], We will denote this map by PFMM : RDe fKS X → RDe fE. Then Definition 3.41 of [13] presents a geometric version Pin f of this, which sends deformations of X to the derived flag variety. On objects it sends a family X → R Spec(B) over a dg-Artin algebra B to the Hodge filtration on the derived de Rham complex. X))/ REndF(RΓ(Ω∗ The following diagram of derived deformation functors is established in [13]: RDe fKS X RDe fKS X PF MM RDe fE / RDe fE ≃ ≃ RDe fX RDe fX Pin f DFlag(RΓ(X, Ω∗ DFlag(RΓ(X, Ω∗ X)) X)) The left vertical map can be considered the inverse Kodaira-Spencer map. Deformation functors, just like stacks, have tangent complexes, and the tangent of the map PFMM has a very concrete description, see Corollary / / / THE GLOBAL DERIVED PERIOD MAP 53 4.22 below. We will now show this agrees (in a reasonable sense) with the differential of our derived period map P. Before we can make precise what we mean by agreement, let us briefly recall the tangent space in a general setting. The tangent space of a functor X at a point x : C → X is the functor that associates to any shifted C- module M the space Tx(M) = X(C ⊕ M) ×X(C),x ∗, where C ⊕ M is the square-zero extension of C by M. If the functor F is homotopy-preserving and homotopy homogeneous then the tangent space is an abelian group and the homotopy groups satisfy πiTx(M) (cid:27) πi+1Tx(M[−1]), see for example Section 1 of [48]. The tangent complex as in Section 3.4 thus has i-th cohomology given by π0Tx(C[−i])). The theory of deformations in derived analytic geometry is considered in detail in [47]. Here we will only need that square zero extensions agree in the derived algebraic and the derived analytic setting, in particular that R Spec(C ⊕ M) ≃ u((R Spec(C ⊕ M))an). This follows from Proposition 8.2 of [45], where an equivalence between derived Artin rings and (suitably defined) derived analytic Artin rings is established using the functor (−)alg. As the underlying schemes are, points (−)alg and u agree in this case. Now let us prepare a comparison between dP and dPin f . We write B for R Spec(C ⊕ M). Then to every family X → S and map B → S we can associate the deformation τ∗X → B and Pin f then defines an element in RDe fE(B), i.e. a map B → DFlag. At the same time P defines an element in Dan(Ban) by postcomposition. Since P is a derived analytic map we will compare it with the analytification of Pin f . We note that the map Pin f by definition goes to the deformation space of the derived flag variety, rather than the derived period domain, i.e. it ignores the bilinear form. Thus we replace P by the composition P ′ : S → U → Dan → DFlagan, obtained by composing the period map with the inclusion U → Dan and the natural forgetful map Dan → DFlagan. We note two things: As we work infinitesimally we may ignore monodromy and assume the target of the period map is U rather than U/Γ. Secondly, we can consider the composition with the map U → DFlagan, and then we no longer need to refer to the polarization to construct the derived period map. Thus the following result analytification of the differential of the infinitesimal derived period map. the differential of P is the tells us that Proposition 4.20. Given X → S and τ : B → S the maps P ′ ◦ τan and P an in f from Ban to DFlagan agree and the correspondence is natural. 54 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Proof. We may view both P an in f and P ′ as constructed out of maps to QFiltan and ∗ with homotopies ηin f and η connecting the associated maps to QPer f an. The only difference is in the definition of the homotopies, which arise from maps win f , w : F 0 → V ⊗ OBan. The map win f is obtained by analytifying an algebraic map while w arises from the simplicial adjunction in Lemma A.12. However, as the natural map u(Ban) → B is an equivalence the two constructions agree. (cid:3) Proposition 4.21. The tangent space of Dan at any point agrees with the tangent space of D. Proof. Given the definition of the tangent space and the definition of analytification as a left Kan extension this follows immediately from the identification of derived analytic and derived algebraic Artin rings. (cid:3) In fact, by the same argument analytification does not change any tangent spaces or induced maps between them. We can thus use the computation from [13] to determine the differential of the period map. Corollary 4.22. At a point s ∈ S the differential of the derived period map dP : TsS → TF,w,QF DV,Q is obtained by composing τ∗ : TsS → T RDe fXs with the Kodaira-Spencer map T RDe fXs → T RDe fKS Xs ≃ RΓ(X, TX)[1] and the map RΓ(X, TX)[1] → (REndQF (V)/ REndF,QF (V))[1] induced by the action of TX on ΩX and the product on derived global sections. Proof. This follows from the theorem as all higher cohomology is determined by evaluating the tangent functor at shifts of C. Other than analytification the only difference to [13] is that the second map in [13] goes to (REnd(V)/ REndF(V))[1]. But this map as constructed clearly factors through (REndQF (cid:3) (V)/ REndF,QF (V))[1]. In other words, the differential of the derived period map can be computed, analogously to the underived version, by cup product and contraction. This characterization of the differential immediately gives the following corollary: Corollary 4.23. The derived period map satisfies Griffiths transversality. THE GLOBAL DERIVED PERIOD MAP 55 This says that the tangent of the derived period map lies in a subcomplex of the tangent complex of the period domain, namely endomorphisms shifting the filtration by one only. Hence we could call the image of the period map a horizontal derived substack. Remark 4.24. At this stage we would like to say that the infinitesimal derived period map is directly obtained from the global derived period map. Indeed, morally the map Pin f is the restriction of P to dg-Artin algebras. However, to make this argument precise we would need to use a moduli stack of all varieties (as opposed to polarized varieties or subschemes of a fixed variety). But the deformation functor RDe fX0 does not extend to an algebraic stack. This can be seen by considering K3-surfaces: The deformation functor is unobstructed with a 20-dimensional tangent space, but algebraic families of K3-surfaces can only be 19-dimensional. Remark 4.25. This is a good moment to note that our construction is still in some sense classical, and there should be a generalized (one could also say extended or non-commutative) version of the period map that starts from RΓ(X, ∧∗TX), which classifies A∞-deformations of Db coh(X). This appears in work by Kontsevich and Barannikov [3,32]. To take this further, one would need to understand non-commutative Hodge structures, see [31]. The reader may also consider [16] for an explicit period map for generalized deformations of a Kähler manifold and [30] for a period map for deformations of a non-commutative algebra. 4.7. Examples. We will finish by briefly talking about some examples for the derived period map. Recall that we already described the universal derived period map defined on substacks of the moduli space M of polarized schemes in Example 1. Example 2. Another class of examples is given by letting S be a subscheme of the derived Hilbert scheme DHilbY for some projective variety Y. This was constructed originally in [9] and appears in a more modern context in [33], see also Section 3.3 of [50]. DHilbY is a geometric stack locally almost of finite presentation that parametrizes derived families of subschemes of Y, to be precise it represents the functor sending S to derived Y × S -schemes which are proper and flat over S and for which the map to Y × S is a closed immersion locally almost of finite presentation. In both of these examples it is useful to observe the following: Fix a smooth scheme X0 or a smooth subscheme X0 of some projective Y. Then there is an open substack S of M respectively DHilbY containing X0 such that the 56 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN : X → S is smooth and projective. (The underived universal family f truncation of f being smooth is an open condition and the condition that f is strong is again open as it says a certain map of graded modules on π0(S ) is an isomorphism.) To apply the main theorem we just need to make sure S is quasi-compact, restricting to some substack if necessary. (Note that this is equivalent to the underlying underived space of S being quasi-compact, see Lemma 3.4.) The universal families on M and DHilbY give smooth projective families of derived schemes X → S by restriction and thus derived period maps P : S an → U/Γ. Let us now look infinitesimally at two concrete examples. We note that the tangent complex for the moduli stack M of polarized schemes at X is an extension of TX[1] by OX[1], see Example 3.39 in [50]. The tangent complex of the Hilbert scheme DHilbY at X ⊂ Y is L∗ X/Y, if X is smooth this is just the normal bundle. Thus the cases where Hi(TXs) is nonzero for i > 1 will often correspond to interesting derived information in S . Example 3. If X0 is any Calabi-Yau variety then the infinitesimal derived period map induces an injection on the cohomology groups of tangent complexes, this follows from Theorem B of [28]. We can use this to write down examples where the derived period map is a non-trivial enhancement of the usual period map. Let us for example consider an abelian surface X in a component M of the moduli stack of polarized varieties. Then X has nontrivial H2(TX), and there is a surjection from π1(TXM) to H2(TX). Thus the derived period map induces a non-trivial map in degree 1 of the tangent complex at X. We note that Theorem A of [28] says that whenever the period map induces an injection on the cohomology of tangent spaces then the deformation theory of X0 is unobstructed, as the deformation theory of a filtered complex is governed by a quasi-abelian L∞-algebra. (In fact generalized deformations of Calabi-Yau varieties are still unobstructed, see [31].) In particular the derived period map only sees unobstructed deformations of the special fibre, and the kernel of the differential of the derived period map is a natural global reduced obstruction theory. One question arising immediately from the existence of the period map is when it is injective in a suitable sense, i.e. when a moduli problem can be completely embedded in the period domain (which has a nice description in terms of linear and quadratic data). THE GLOBAL DERIVED PERIOD MAP 57 One possible way of making this precise would be to ask when π0π0(P) is an immersion and dP is injective on homotopy groups. Remark 4.26. We note that this is not the definition of a closed immersion in the sense of [66] (which does not constrain the homotopy groups of the domain), nor is it a strong map (which constrains the homotopy groups of the domain too much). This is an example of the lack of a good derived analogue of closed immersion. This is a kind of derived Torelli problem. (The reader should note that this term is already used for the problem of determining when the period map determines the derived category of a variety.) Example 4. Now consider the case where X0 is a hypersurface of degree d in P3. We may consider it in the moduli stack M of polarized surfaces. Then derived Torelli is false if d is large enough, i.e. the map on homotopy groups of tangent complexes is not an injection. It follows from our tangent space considerations that derived deformations of X0 in degree 1 surject to H2(TX0). One can compute that H2(TX0) is nonzero (using Kodaira vanishing, Hirzebruch-Riemann-Roch and the normal exact sequence). But as H∗(X0, Ω∗) is concentrated in even degrees the infinitesimal period map must send H2(TX0) to zero. This is particularly interesting since projective hypersurfaces of degree bigger than three satisfy the local Torelli theorem, see [7], thus π0π0(P) is locally an immersion. Appendix A. Presenting higher analytic stacks A.1. Introduction. In this appendix we show that Pridham's framework of presenting higher stacks as hypergroupoids developed in [51] can be used to provide models for derived analytic Artin stacks. For an overview of the general theory see [49]. While some of this discussion is not strictly necessary and only Lemma A.12 is needed in the body of the paper we feel that this discussion provides an interesting and useful viewpoint. A.2. Hypergroupoids. In [51] Pridham develops the theory of hyper- groupoids as presentations of higher stacks. To talk about hypergroupoids we need to fix a model category S with a nice subcategory A and classes C and ǫ of morphisms in Ho(A ) and Ho(S ) respectively, satisfying some conditions we will detail below. 58 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Example 5. The motivating example is when S is the model category of stacks on derived affine schemes over some ground ring k with the étale topology. A consists of the essential image of the Yoneda embedding, ǫ is the class of local surjections in Ho(S ) and C the class of smooth maps in Ho(A ) which are moreover in ǫ. We write this quadruple as (Aalg, Salg, Calg, ǫalg) We write sA for A ∆op. Definition. A map X → Y in sA is a relative (n, C)-hypergroupoid over Y if it is a Reedy fibration and the homotopy partial matching maps Xm → Mh Λm k (X) ×h Mh Λm k (Y) Ym are in C for all m, k, and weak equivalences if m > n. A relative hypergroupoid over the final object is simply called a (n, C)-hypergroupoid. Definition. A map X → Y in sA is a trivial relative (n, C)-hypergroupoid over Y if it is a Reedy fibration and the homotopy matching maps Xm → Mh ∂∆m(X) ×h Mh ∂∆m (Y) Ym are in C for all m, and weak equivalences if m ≥ n. We will fix n in this section and simply speak of hypergroupoids when C is clear from the context. op −(X) : sSetop → A is defined by → A , see Section Here the matching objects functor Mh forming the right Kan extension of the functor X : ∆ 1.1.1 of [51]. It is the derived version of HomsSet(−, X). Now the geometric realization functor − from sA to S sends hypergroupoids to geometric stacks, see Proposition 4.1 of [51]. (This is just another name for the homotopy colimit over ∆op.) For simplicity we will always compose − with stackification (i.e. fibrant replacement in S ) without mentioning it in the notation. Theorem 3.3 in [49] says that in the example above the relative category of (n, Calg)-hypergroupoids with weak equivalences given by trivial hypergroupoids gives a model for the ∞-category of strongly quasi-compact n-geometric derived Artin stacks. Remark A.1. Recall that a relative category [4] is just a category with a collection of weak equivalences satisfying some very basic conditions. Nevertheless, the model category of relative categories is Quillen equivalent to other models of ∞-categories (i.e. simplicial categories, quasi-categories, Complete Segal Spaces). THE GLOBAL DERIVED PERIOD MAP 59 We will denote the level-wise hom-space for sA by RHom, but it is important to note that we cannot compute hom-spaces level-wise. Instead we define RHom# sA (X, Y) = RHompro(sA )( X, Y) where X → X is a T -projective relative T -cocell where T is the class of trivial relative hypergroupoids. A relative T -cocell is just a transfinite composition of pullbacks of maps in T . For detailed definitions see Section 3.2 of [51]. To understand this definition it helps to note that π0(RHom#(X, Y) = limX′∈T (X) π0 RHom(X′, Y), where the limit is over trivial relative hypergroupoids over X. Then if Y is a hypergroupoid we have that RHom# see Theorem 4.10 in [51]. sA (X, Y) ≃ MapS (X, Y), We want to apply this framework to study analytic stacks. Thus we need a new choice of (A , S , C, ǫ). Recall that dS tein is the full subcategory of dAn given by those derived analytic spaces whose underlying underived analytic space is Stein. This is constructed as a quasi-category in [45]. We need A to be a pseudo-model category. To be precise we need a model category S , a class of morphisms WA in A and a fully faithful ι : A → S such that • ι(WA ) = WS ∩ ι(A ). Here WS are the weak equivalences in S . • A closed under weak equivalences in S , i.e. x ≃ ι(y) implies x is in the image of A . • A is closed under homotopy pullbacks in S Consider the quasi-category dS tein and recall that simplicial categories and quasi-categories are equivalent models of ∞-categories. We apply C, the left adjoint of the coherent nerve N, to obtain a simplicial category C[dS tein]. We consider sSetC[dS tein]), the simplicial category of functors from C[dS tein] to the simplicial model category of simplicial sets, sSet. We equip this category with the injective model structure. There is a Grothendieck topology τ on dS tein induced by étale morphisms of Tan- structured topoi (see Definition 2.3.1 of [36]). We localize sSetC[dS tein] at homotopy τ-hypercovers to obtain the model category of stacks, which we call San. As τ is (hyper-)subcanonical, see Corollary 3.6 in [45], we have a fully faithful embedding y : C[dS tein] → San. We let Aan be the closure of y(C[dS tein]) under weak equivalences and define WA by the first condition above. 60 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Thus the first two conditions are satisfied. For the third condition we need to know that dS tein is closed under homotopy pullbacks. By Section 12 of [35] dAn has (homotopy) pullbacks. Now note that π0 commutes with limits and the fibre products of Stein spaces are Stein spaces, see Example 51 (b) in [52]. Remark A.2. For the reader's peace of mind we note that working with simplicial presheaves as simplicial functors is equivalent to working with presheaves defined in the quasi-categorical setting as in [46]. This follows from Proposition 4.2.4.4 of [37]. Setting U = A = sSet we have N((sSetC[dS tein])◦) ≃ Fun(dS tein, N(sSet◦)) where N is the coherent nerve. Here (−)◦ restricts to fibrant cofibrant objects in a model category. Of course N(sSet◦) is the infinity category of simplicial sets. In other words the simplicial category (sSetC[dS tein])◦ models the quasi-category of functors from dS tein to simplicial sets. This is true for the injective or projective model structure. The reason we are working not with quasi-categories but simplicially with sSetC[dS tein] is that we rely on explicit arguments involving simplicial diagrams in Theorem A.3 below. Having made these observations we will abuse notation and use dS tein for C[dS tein] from now on. Next, given the pseudo-model category Aan ⊂ San we need a class of ǫ- morphisms in San, functioning as covers, closed under composition and homotopy pullback. They need to satisfy Properties 1.7 of [51]. We just imitate the algebraic definition and let ǫan be the class of local surjections, i.e. maps such that the associated map of simplicial sheaves is surjective on π0. Next we need a class Can of morphisms in Ho(Aan) containing isomorphisms and closed under composition and homotopy pullback, and satisfying Properties 1.8 of [51]. We define Can to consist of smooth maps which are also in ǫan. Theorem A.3. (A , San, Can, ǫan) as above satisfies Properties 1.7 and 1.8 of [51]. Proof. We just follow the proof of Proposition 1.19 of [51]. 1.8 is true for any simplicial site. For (1) see Proposition 3.1.4 in [65], (2) holds by definition and (3) is clear as smooth maps are defined locally. For 1.7 we use dS tein in place of the category T in loc. cit. and then consider simplicial objects in I ≔ San = sPrτ,in j(dS tein), which THE GLOBAL DERIVED PERIOD MAP 61 is a category of simplicial presheaves with the local injective model structure on a simplicial site, thus all the computations involved are entirely unchanged. (cid:3) Remark A.4. The last ingredient we need is Assumption 3.20 of [51]. However, we can always satisfy the condition by choosing two universes, see Remark 3.21 in loc. cit. Note that we may replace objects in sAan by objects in dS tein∆op and talk about genuine simplicial derived Stein spaces. This is shown in Section 4.4 of [51], the main input being the proof of Lemma 1.3.2.9 of [66], which goes through in the analytic setting. We can now consider the category of derived analytic stacks as modelled by hypergroupoids. As all the results in [51] hold for analytic stacks we can deduce the following: Theorem A.5. The relative category of (n, Can)-hypergroupoids, with weak equivalences given by relative trivial hypergroupoids, is a model for the ∞-category of strongly quasi-compact n-geometric derived analytic Artin stacks. Here we note that derived Artin stacks are the subcategory of derived analytic stacks San, given by the usual representability conditions, cf. Definition 2.8 of [46]. there is a difference in definition between a Remark A.6. Note that geometric context in the sense of [46], like (dS tein, τ, P), and a HA-context in the sense of [66], which is cited in [51]. But the definition of geometricity is the same in both contexts. A.3. Analytification of hypergroupoids. The derived analytification functor an : X 7→ Xan between structured ∞-topoi sends derived affine schemes to Tan-affines, see Proposition 2.3.18 of [36]. In particular derived affine schemes locally almost of finite presentation are sent to derived Stein spaces. (Analytification maps T´et-schemes to Tan-schemes, and the truncation is sent to a Stein space.) We write dA f f l f p for derived affine schemes locally almost of finite presentation. This is just the opposite of the category of homotopically finitely presented simplicial algebras. We now want to consider a partial left adjoint to this functor and define u = R Spec(O(−)) on dS tein. This is an affinized forgetful or algebraization functor. It is clear that u sends affines to affines. 62 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN Lemma A.7. There is a natural weak equivalence MapdS tein(T, Y an) ≃ MapdA f f (uT, Y) for a derived Stein spaces T and a derived affine scheme Y locally almost of finite presentation. Proof. We consider the functor R Spec(Γ(−)) ◦ (−)alg between the ∞- categories T op(Tan) and dA f f . This extends the functor u we are considering. Moreover, as a composition of right adjoints it has a left adjoint: dA f f → T op(Tan) given by inclusion followed by analytification. Now we note that an sends dA f f l f p to dS tein and conclude. (cid:3) Remark A.8. There is a slightly more concrete way of seeing this weak equivalence: The correspondence is clear if Y = A1, as in this case both sides are just global functions on T . Then we can extend to all derived affine schemes by observing that an preserves limits, and derived affine schemes are generated under limits by A1. Remark A.9. In fact we can extend Map(T, Y an) ≃ Map(uT, Y) to the case where T is a derived analytic space which is a colimit of derived Stein spaces. We can thus extend to simplicial derived Stein spaces, but note that u is not going to preserve hypergroupoids. As we may represent derived algebraic and analytic stacks by simplicial derived affine schemes resp. simplicial derived Stein spaces the following lemma is useful. We can then define a level-wise analytification an for simplicial derived affines. Lemma A.10. Level-wise analytification sends (trivial) relative hypergroupoids in dA f f l f p to (trivial) relative hypergroupoids. Proof. By Lemma A.7 an preserves homotopy limits in dA f f l f p. Moreover it sends smooth maps to smooth maps and preserves local surjections. It follows that it preserves relative hypergroupoids. (cid:3) Moreover, level-wise analytification agrees with the Lurie-Porta definition that we quoted in Section 2.2. This also shows it is well-defined under weak equivalences of hypergroupoids. Lemma A.11. Let Y be a hypergroupoid in dA f f l f p. Then an(Y)San = (YSalg)an. Proof. This just says that analytification commutes with geometric realization. This is clear as analytification of Artin stacks is defined as a left Kan extension. (cid:3) THE GLOBAL DERIVED PERIOD MAP 63 However, we note that there is no well-defined algebraization functor from analytic hypergroupoids to algebraic hypergroupoids, contrary to claims in previous versions of this paper. For a systematic way to get around this difficulty, see [27]. We can now state a way of producing maps into analytifications. Lemma A.12. For a simplicial derived Stein space T• and a derived alge- braic stack Y locally almost of finite presentation with simplicial presenta- tion Y• there is a natural map MapsAalg(uT•, Y•) → MapsAan(T•, an(Y•)) ≃ MapdAnS t(T•, Y an). Proof. The first arrow is immediate from Lemma A.7 and applying the realization functor. The second map comes from Lemma A.11. (cid:3) Remark A.13. The map in the lemma is not well-defined on the level of derived stacks and it appears that this shortcoming cannot be overcome. This means that whenever we want to apply this lemma for some cover T• of a derived analytic stack, we need to check by hand that the map we construct does not depend on our choice of T•. References [1] S. Barannikov, Generalized periods and mirror symmetry in dimensions n> 3, arXiv preprint math/9903124, (1999). [2] S. Barannikov, Quantum periods - I. Semi-infinite variations of Hodge structures, ArXiv e-prints, (2000), math/0006193. [3] S. Barannikov and M. Kontsevich, Frobenius manifolds and formality of Lie algebras of polyvector fields, Int. Math. Res. Not., 1998 (1998), pp. 201 -- 215. [4] C. Barwick and D. M. Kan, Relative categories: another model for the homotopy theory of homotopy theories, Indagationes Mathematicae, 23 (2012), pp. 42 -- 68. [5] O. Ben-Bassat and K. Kremnizer, Non-Archimedean analytic geometry as relative algebraic geometry, ArXiv e-prints, (2013), 1312.0338, to appear in Ann. Fac. Sci. Toulouse Math. [6] J. Bertin, Introduction to Hodge Theory, Panoramas et synthèses, American Mathematical Society, 2002. [7] J. Carlson, S. Muller-Stach, and C. Peters, Period mappings and period domains, vol. 85 of Cambridge Studies in Advanced Mathematics, Cambridge University Press, 2003. [8] E. H. Cattani and A. G. Kaplan, Horizontal sl 2-orbits in flag domains, Math. Ann., 235 (1978), pp. 17 -- 35. [9] I. Ciocan-Fontanine and M. Kapranov, Derived Hilbert schemes, J. Amer. Math. Soc., 15 (2002), pp. 787 -- 815. [10] P. Deligne, Théorème de Lefschetz et critères de dégénérescence de suites spectrales, Inst. Hautes Études Sci. Publ. Math., 35 (1968), pp. 259 -- 278. [11] , Équations différentielles à points singuliers réguliers, Lecture Notes in Mathematics, Vol. 163, Springer, 1970. 64 [12] CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN , Décompositions dans la catégorie dérivée, in Motives (Seattle, WA, 1991), vol. 55 of Proc. Sympos. Pure Math., Amer. Math. Soc., 1994, pp. 115 -- 128. [13] C. Di Natale, A period map for global derived stacks, ArXiv e-prints, (2014), 1407.5906. [14] , Derived moduli of complexes and derived Grassmannians, Appl. Cat. Struct, (2016). [15] D. Dugger and D. C. Isaksen, Topological hypercovers and A1-realizations, Math. Z., 246 (2004), pp. 667 -- 689. [16] D. Fiorenza and M. Manetti, A period map for generalized deformations, J. Noncomm. Geom., 3 (2009), pp. 579 -- 597. [17] D. Fiorenza and E. Martinengo, A short note on infinity-groupoids and the period map for projective manifolds, ArXiv e-prints, (2009), 0911.3845. [18] D. Gaitsgory, Notes on Geometric Langlands: Quasi-coherent sheaves on stacks. Available from the author's website at math.harvard.edu/ gaitsgde/GL/QCohtext.pdf, 2011. [19] P. Goerss and K. Schemmerhorn, Model categories and simplicial methods, in Contemporary Mathematics, Amer. Math. Soc., 2007. [20] H. Grauert, Ein theorem der analytischen garbentheorie und die modulräume komplexer strukturen, Inst. Hautes Études Sci. Publ. Math., 5 (1960), pp. 5 -- 64. [21] P. A. Griffiths, Periods of integrals on algebraic manifolds: Summary of main results and discussion of open problems, Bull. Amer. Math. Soc., 76 (1970), pp. 228 -- 296. [22] O. Gwilliam and D. Pavlov, Enhancing the filtered derived category, ArXiv e-prints, (2016), 1602.01515. [23] P. S. Hirschhorn, Model categories and their localizations, vol. 99 of Mathematical Surveys and Monographs, American Mathematical Society, 2003. [24] A. Hirschowitz and C. Simpson, Descente pour les n-champs, Arxiv e-prints, (2001), math/9807049. [25] J. V. S. Holstein, Morita cohomology, Math. Proc. Camb. Phil. Soc., 158 (2015), pp. 1 -- 26. [26] , Morita cohomology and homotopy locally constant sheaves, Math. Proc. Camb. Phil. Soc., 158 (2015), pp. 27 -- 35. [27] J. V. S. Holstein and M. Porta, Analytification of mapping stacks, Arxiv preprint arXiv:1812.09300, (2018). in preparation. [28] D. Iacono and M. Manetti, An algebraic proof of Bogomolov-Tian-Todorov theorem, in Deformation spaces, Springer, 2010, pp. 113 -- 133. [29] L. Illusie, Complexe cotangent et déformations I, vol. 239, Springer, 2006. [30] I. Iwanari, Period mappings for noncommutative algebras, ArXiv e-prints, (2016), 1604.08283. [31] L. Katzarkov, M. Kontsevich, and T. Pantev, Hodge theoretic aspects of mirror symmetry, ArXiv e-prints, (2008), 0806.0107. [32] M. Kontsevich, Deformation quantization of poisson manifolds, Lett. Math. Phys., 66 (2003), pp. 157 -- 216. [33] J. Lurie, Derived algebraic geometry. Available http://www.math.harvard.edu/∼lurie/papers/DAG.pdf. , Special algebraic geometry. Available http://www.math.harvard.edu/ lurie/papers/SAG-rootfile.pdf. at at [34] [35] , Derived algebraic geometry IX: Closed immersions, preprint available from author's website as http://math. harvard. edu/∼lurie/papers/DAG-IX. pdf, (2011). THE GLOBAL DERIVED PERIOD MAP 65 , Derived algebraic geometry V: Structured spaces, preprint available from author's website as http://math. harvard. edu/∼lurie/papers/DAG-V. pdf, (2011). , Higher topos theory, Annals of Mathematics Studies, 170 (2011). , algebra. Available from the Higher author's website as [36] [37] [38] www.math.harvard.edu/∼lurie/papers/higheralgebra.pdf, 2012. [39] N. Mihalache, Special neighbourhoods of subsets in complex spaces, Mathematische Zeitschrift, 221 (1996), pp. 49 -- 58. [40] A. Neeman, Grothendieck duality made simple, ArXiv e-prints, (2018), 1806.03293. [41] T. Pantev, B. Toen, M. Vaqui´e, and G. Vezzosi, Shifted symplectic structures, Inst. Hautes Études Sci. Publ. Math., 117 (2013), pp. 271 -- 328. [42] C. Peters and J. Steenbrink, Mixed Hodge structures, Ergebnisse der Mathematik und ihrer Grenzgebiete, Springer, 2008. [43] M. Porta, Comparison results for derived Deligne-Mumford stacks, ArXiv e-prints, (2015), 1507.00573. [44] [45] , Derived complex analytic geometry II: Square-zero extensions, ArXiv e-prints, (2015), 1507.06602. , GAGA theorems in derived complex geometry, Journal of Algebraic Geometry, 28 (2019), pp. 519 -- 565. [46] M. Porta and T. Y. Yu, Higher analytic stacks and GAGA theorems, Advances in Mathematics, 302 (2016), pp. 351 -- 409. [47] , Representability theorem in derived analytic geometry, ArXiv e-prints, (2017), 1704.01683. [48] J. P. Pridham, Representability of derived stacks, ArXiv e-prints, (2010), 1011.2189. , Notes characterising higher and derived stacks, ArXiv e-prints, (2011), [49] 1105.4853. [50] [51] , Derived moduli of schemes and sheaves, Journal of K-theory: K-theory and its Applications to Algebra, Geometry, and Topology, 10 (2012), pp. 41 -- 85. , Presenting higher stacks as simplicial schemes, Adv. Math., 238 (2013), pp. 184 -- 245. [52] R. Remmert, Local theory of complex spaces, in Several Complex Variables VII, Springer, 1994, pp. 7 -- 96. [53] C. Rezk, A model for the homotopy theory of homotopy theory, Trans. Amer. Math. Soc, 353 (2001), pp. 973 -- 1007. [54] P. Sastry and Y. L. L. Tong, The Grothendieck trace and the de Rham integral, Canadian Mathematical Bulletin, 46 (2003), pp. 429 -- 440. [55] T. Schurg, B. Toen, and G. Vezzosi, Derived algebraic geometry, determinants of perfect complexes, and applications to obstruction theories for maps and complexes, J. Reine Angew. Math., 2015 (2015), pp. 1 -- 40. [56] N. Spaltenstein, Resolutions of unbounded complexes, Compositio Math., 65 (1988), pp. 121 -- 154. [57] M. Spitzweck, Homotopy limits of model categories over inverse index categories, J. Pure Appl. Algebra, 214 (2010), pp. 769 -- 777. [58] D. I. Spivak, Derived smooth manifolds, Duke Math. J., 153 (2010), pp. 55 -- 128. [59] G. Tabuada, Differential graded versus simplicial categories, Topology Appl., 157 (2010), pp. 563 -- 593, 0711.3845. [60] , Homotopy theory of dg-categories via localizing pairs and Drinfeld's dg quotient, Homology, Homotopy Appl., 12 (2010), pp. 187 -- 219. 66 CARMELO DI NATALE AND JULIAN V. S. HOLSTEIN [61] B. Toen, Higher and derived stacks: a global overview., in Algebraic geometry -- Seattle 2005. Part 1, vol. 80 of Proc. Sympos. Pure Math, Amer. Math. Soc., 2009, pp. 435 -- 487. [62] [63] , Proper local complete intersection morphisms preserve perfect complexes, ArXiv e-prints, (2012), 1210.2827. , Derived algebraic geometry, EMS Surveys in Mathematics, 1 (2014), pp. 153 -- 240. [64] B. Toen and M. Vaquie, Moduli of objects in dg-categories, Ann. Sci. École Norm. Sup., 40 (2007), pp. 387 -- 444. [65] B. Toen and G. Vezzosi, Homotopical algebraic geometry I: Topos theory, Advances in mathematics, 193 (2005), pp. 257 -- 372. [66] [67] , Homotopical algebraic geometry II: Geometric stacks and applications, Mem. Amer. Math. Soc., 193 (2008). , Algebres simpliciales S1-equivariantes et theorie de Rham, Arxiv preprint arXiv:0904.3256, (2009), arXiv:0904.3256v1. [68] G. Vezzosi, Quadratic forms and Clifford algebras on derived stacks, Adv. Math., 301 (2016), pp. 161 -- 203. [69] C. Voisin, Chow rings and decomposition theorems for K3 surfaces and Calabi -- Yau hypersurfaces, Geometry & Topology, 16 (2012), pp. 433 -- 473. [70] C. Weibel, An introduction to homological algebra, Cambridge University Press, 1995. Carmelo Di Natale, King's College London E-mail address: [email protected] Julian V. S. Holstein, St John's College, University of Cambridge E-mail address: [email protected]
1705.07790	1	1705	2017-05-22T15:01:39	Ulrich bundles on smooth projective varieties of minimal degree	[ "math.AG" ]	We classify the Ulrich vector bundles of arbitrary rank on smooth projective varieties of minimal degree. In the process, we prove the stability of the sheaves of relative differentials on rational scrolls.	math.AG	math	ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS Abstract. We classify the Ulrich vector bundles of arbitrary rank on smooth pro- jective varieties of minimal degree. In the process, we prove the stability of the sheaves of relative differentials on rational scrolls. 1. Introduction There has been increasing interest on the classification of arithmetically Cohen- Macaulay (for short aCM) sheaves on various projective varieties, which is impor- tant in a sense that the aCM sheaves are considered to give a measurement of complexity of the underlying space. A special type of aCM sheaves, called the Ulrich sheaves, are the ones achieving the maximum possible minimal number of generators. It is conjectured in [12] that any variety supports an Ulrich sheaf. But the conjecture has been checked only for a few varieties, e.g. in case of surfaces, del Pezzo surfaces, rational normal scrolls, rational aCM surfaces in P4, ruled sur- faces and so on; see [8, 15, 16, 2]. Although there are some occasions where the classification problem of Ulrich bundles of special type is done as in [7, 9, 10], the completion of classification problem is difficult in usual. In this article we pay our attention to the smooth projective varieties of min- imal degree; smooth quadric hypersurfaces, the Veronese surface in P5 and ra- tional normal scrolls S = S(a0, . . . , an) associated to rank n + 1 bundle ⊕n i=0OP1 (ai ) on P1. We choose suitable full exceptional collections of those varieties to apply the Beilinson spectral sequence and characterize the Ulrich bundles supporting the varieties. In [15] the representation type of S is determined by considering a certain type of family of Ulrich sheaves and it motivates the study in this article. In case of the Veronese surface and smooth quadric hypersurfaces, the classical result that the only indecomposable Ulrich bundles are Ω1 P2 (3) and the spinor bundles, respectively, can be recovered; see Examples 2.6 and 2.7. The main result of this article is the characterization of Ulrich bundles on smooth rational normal scrolls as the bundles admitting a special type of filtration; see Theorem 3.7. As a direct consequence, we observe that the moduli spaces of Ulrich bundles are zero-dimensional; see remark 3.9. Since any non-linearly normal smooth variety 2010 Mathematics Subject Classification. Primary: 14J60; Secondary: 13C14, 14F05. Key words and phrases. Ulrich bundle, rational normal scroll, Beilinson spectral theorem. The first author was partly supported by a UEFISCDI grant. The second author is supported by Basic Science Research Program 2015-037157 through NRF funded by MEST and the National Re- search Foundation of Korea(KRF) 2016R1A5A1008055 grant funded by the Korea government(MSIP). The third author is supported by the framework of PRIN 2010/11 'Geometria delle variet`a algebriche', cofinanced by MIUR and GNSAGA of INDAM (Italy). The forth author is supported by a FY2015 JSPS Postdoctoral Fellowship. 1 2 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS of almost minimal degree is given as the image of a linear projection of a smooth variety of minimal degree (and the projection is an isomorphism in this case), the classification of Ulrich bundles on the non-linearly normal smooth varieties of almost minimal degree can be also given; see [19]. Indeed, being Ulrich only depends on the chosen polarization (X, L), not on taking the embedding on the complete linear system H 0(X, L). Here we summarize the structure of this article. In Section 2 we introduce the definition of Ulrich sheaves and several notions in derived category of coherent sheaves to understand the Beilinson spectral sequence. Then we investigate the indecomposable Ulrich bundles on the Veronese surface and smooth quadric hy- persurfaces. In Section 3 we collect several technical results about several bundles on rational normal scrolls; see Lemmas 3.2 and 3.4. Then we fix a certain full exceptional collection of the derived categories in Example 3.6, based on which we apply the Beilinson spectral sequence to characterize the Ulrich bundles in Theorem 3.7. 2. Preliminaries Throughout the article our base field is the filed of complex numbers C. Definition 2.1. A coherent sheaf E on a projective variety X with a fixed ample line bundle OX (1) is called arithmetically Cohen-Macaulay (for short, aCM) if it is locally Cohen-Macaulay and H i(E(t)) = 0 for all t ∈ Z and i = 1, . . . , dim(X) − 1. Definition 2.2. For an initialized coherent sheaf E on X, i.e. h0(E(−1)) = 0 but h0(E) , 0, we say that E is an Ulrich sheaf if it is aCM and h0(E) = deg(X)rank(E). Given a smooth projective variety X, let Db(X) be the the bounded derived category of coherent sheaves on X. An object E ∈ Db(X) is called exceptional if Ext•(E, E) = C. A set of exceptional objects hE0, . . . , Eni is called an exceptional collection if Ext•(Ei , Ej ) = 0 for i > j. An exceptional collection is said to be full when Ext•(Ei , A) = 0 for all i implies A = 0, or equivalently when Ext•(A, Ei ) = 0 does the same. Definition 2.3. Let E be an exceptional object in Db(X). Then there are functors LE and RE fitting in distinguished triangles LE(T ) −→ Ext•(E, T ) ⊗ E −→ T −→ LE(T )[1] RE(T )[−1] −→ T −→ Ext•(T , E)∗ ⊗ E −→ RE(T ) The functors LE and RE are called respectively the left and right mutation functor. The collections given by E∨ i = LE0 ∨Ei = REn LE1 . . . LEn−i−1 En−i ; REn−1 . . . REn−i+1 En−i , are again full and exceptional and are called the right and left dual collections. The dual collections are characterized by the following property; see [14, Section 2.6]. (1) Extk(∨Ei , Ej ) = Extk(Ei , E∨ j ) =( C 0 if i + j = n and i = k otherwise ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE 3 Theorem 2.4 (Beilinson spectral sequence). Let X be a smooth projective variety and with a full exceptional collection hE0, . . . , Eni of objects for Db(X). Then for any A in Db(X) there is a spectral sequence with the E1-term E p,q 1 = Mr+s=q Extn+r (En−p, A) ⊗ Hs(E∨ p ) which is functorial in A and converges to Hp+q(A). The statement and proof of Theorem 2.4 can be found both in [21, Corollary 3.3.2], in [14, Section 2.7.3] and in [4, Theorem 2.1.14]. Let us assume next that the full exceptional collection hE0, . . . , Eni contains only i [−ki ] with Ei a vector bundle for each i, and more- n i consists of coherent sheaves. Then the pure objects of type Ei = E ∗ over the right dual collection hE∨ Beilinson spectral sequence is much simpler since 0 , . . . , E∨ E p,q 1 = Extn+q(En−p, A) ⊗ E∨ p = H n+q+kn−p (En−p ⊗ A) ⊗ E∨ p . Note however that the grading in this spectral sequence applied for the pro- jective space is slightly different from the grading of the usual Beilison spectral sequence, due to the existence of shifts by n in the index p, q. Indeed, the E1-terms of the usual spectral sequence are H q(A(p)) ⊗ Ω−p(−p) which are zero for positive p. To restore the order, one needs to change slightly the gradings of the spectral sequence from Theorem 2.4. If we replace, in the expression Eu,v 1 = Extv(E−u , A) ⊗ E∨ n+u = H v+k−u (E−u ⊗ A) ⊗ F−u u = −n + p and v = n + q so that the fourth quadrant is mapped to the second quadrant, we obtain the following version of the Beilinson spectral sequence: Theorem 2.5. Let X be a smooth projective variety with a full exceptional collection hE0, . . . , Eni where Ei = E ∗ i [−ki ] with each Ei a vector bundle and (k0, . . . , kn) ∈ Z⊕n+1 such that there exists a sequence hFn = Fn, . . . , F0 = F0i of vector bundles satisfying (2) Extk(Ei , Fj ) = H k+ki (Ei ⊗ Fj ) =( C 0 if i = j = k otherwise i.e. the collection hFn, . . . , F0i labelled in the reverse order is the right dual collection of hE0, . . . , Eni. Then for any coherent sheaf A on X there is a spectral sequence in the square −n ≤ p ≤ 0, 0 ≤ q ≤ n with the E1-term E p,q 1 = Extq(E−p, A) ⊗ F−p = H q+k−p (E−p ⊗ A) ⊗ F−p which is functorial in A and converges to (3) E p,q ∞ =( A 0 if p + q = 0 otherwise. Example 2.6. The Veronese surface V ⊂ P5 is an arithmetically Cohen-Macaulay embedding which is not arithmetically Gorenstein. The study of vector bundles on V is trivial if we view V as P2 embedded with OP2 (2). We consider the full ex- ceptional collection hE0, E1, E2i = hOP2 , OP2 (1), OP2 (2)i and then its dual in reverse order E• consists of the following E2 = OP2 (−2) , E1 = OP2 (−1) , E0 = OP2 . 4 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS Define Fk = Fk to be the vector bundle satisfying Condition (2) in Theorem 2.5 with respect to the prescribed collection hE0, E1, E2i, that is, F2 = OP2 (−1) , F1 = Ω1 P2 (1) , F0 = OP2 . In other words, the collection hF2, F1, F0i is the right dual collection of hE0, E1, E2i. Let A be a vector bundle on V such that A(2) is Ulrich. Then we construct a Beilinson complex, quasi-isomorphic to A, by calculating H i (A ⊗ Ej ) ⊗ Fj with i, j ∈ {0, 1, 2}. We may assume due to [12, Proposition 2.1] that H i (A(−2)) = H i (A) = 0 for all i and then we can easily construct the Beilinson table as below; we put the collection F• on the top row and the collection E• on the bottom row OP2 (−1) Ω1 P2 (1) OP2 0 0 0 a 0 0 OP2 (−2) OP2 (−1) OP2 0 0 0 We notice that the only nonzero entry a := h1(A(−1)) appears on the diagonal. So the table is very simple and we can deduce that Ω1 P2 (3) is the only indecomposable Ulrich bundle on V . Example 2.7. Let Q4 be the smooth quadric hypersurface of P5. It has two distinct spinor bundles Σ1 and Σ2 of rank two given by the exact sequences for i ∈ {1, 2}. As in Example 2.6 we may consider the full exceptional collection 0 −→ Σ∨ i → O⊕4 Q4 → Σ3−i −→ 0 hE0, . . . , E5i = hOQ4, OQ4 (1), OQ4 , Σ∨ 1 (3), Σ∨ 2 (3)[1], OQ4 (3)[1]i, i.e. we have E5 = OQ4 (−3) , E4 = Σ2(−3) , E3 = Σ1(−3), E2 = OQ4 (−2) , E1 = OQ4 (−1) , E0 = OQ4 and (k0, . . . , k5) = (0, 0, 0, 0, −1, −1). Then its right dual collection hF5, . . . , F0i consists of the following vector bundles Fi = Fi for each i: F5 = OQ4 (−1) , F4 = Σ1(−1) , F3 = Σ2(−1), F2 = Ω2 P5 (2)Q4 , F1 = Ω1 P5 (1)Q4 , F0 = OQ4 , due to the characterization in (2). Let A be a vector bundle on Q4 such that A(1) is Ulrich. As in Example 2.6 we construct a Beilinson complex, quasi-isomorphic to A, by calculating H i+kj (A ⊗ Ej ) ⊗ Fj with i, j ∈ {0, . . . 5}. Again we may assume due to [12, Proposition 2.1] that H i (A(−3)) = H i (A(−2)) = H i (A(−1)) = H i (A) = 0 for all i. Moreover there are no nonzero maps from Σ1(−1) to Σ2(−1). Thus we easily can construct the Beilinson table; we put the collection F• on the top row and the collection E• on the bottom row. ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE 5 OQ4 (−1) Σ1(−1) Σ2(−1) Ω2 P5 (2)Q4 0 0 0 0 0 0 0 b 0 0 0 0 0 0 a 0 0 0 0 0 0 0 0 0 OQ4 (−3) Σ2(−3) Σ1(−3) OQ4 (−2) Ω1 P5 (1)Q4 OQ4 0 0 0 0 0 0 OQ4 (−1) OQ4 0 0 0 0 0 0 We notice that the only nonzero entries a := h3(A ⊗ Σ1(−3)) and b := h3(A ⊗ Σ2(−3)) appear on the diagonal. So the table is very simple and we can deduce that Σ1 and Σ2 are the only indecomposable Ulrich bundles on V . Similarly for smooth quadric hypersurfaces of any (even or odd) dimension we can obtain a table with nonzero entries only on the diagonal, proving that the indecomposable Ulrich bundles are the spinor bundles; see [3]. In the next section we prove that for any rational normal scroll we can obtain a table with nonzero entries only on the diagonal; we cannot expect such a behavior for varieties of almost minimal degree. Let us see an example of a Del Pezzo threefold in the following. Example 2.8. The Veronese threefold V ⊂ P9 is Del Pezzo of degree 8. The study of vector bundles on V is trivial if we view V as P3 embedded with OP3 (2). We consider the full exceptional collection hE0, . . . , E3i = hOP3 , OP3 (1), OP3 (2), OP3 (3)i and then the collection E• in reverse order is given by E3 = OP3 (−3) , E2 = OP2 (−2) , E1 = OP3 (−1) , E0 = OP3 and the right dual collection is given by F3 = OP3 (−1) , F2 = Ω2 P3 (2) , F1 = Ω1 P3 (1) , F0 = OP3 . Now let A be a vector bundle on V such that A(2) is Ulrich. Then we construct a Beilinson complex, quasi-isomorphic to A, by calculating H i (A ⊗ Ej ) ⊗ Fj with i, j ∈ {0, 1, 2, 3}. We may assume due to [12, Proposition 2.1] that H i (A(−4)) = H i (A(−2)) = H i (A) = 0 for all i and then we can easily construct the Beilinson table as below; we put the collection F• on the top row and the collection E• on the bottom row P3 (2) Ω1 OP3 (−1) Ω2 P3 (1) OP3 0 0 0 0 0 0 0 0 a 0 0 0 OP3 (−3) OP3 (−2) OP3 (−1) OP3 0 b 0 0 We notice that the only nonzero entries are a := h1(A(−1)) which is on the diagonal and b := h1(A(−3)) which is not on the diagonal. Remark 2.9. Although we have a nonzero entry off the diagonal in Example 2.8, we may still use the Beilinson table to describe the Ulrich bundles on V . Indeed, 6 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS any Ulrich bundle E on V fits into the following exact sequence 0 −→ OP3 (1)⊕b M −→ Ω1 P3 (3)⊕a −→ E −→ 0. By [5, Lemma 2.4(iii)] we also get a = b and so the map M = (sij ) is given as an (a × a)-matrix with entries sij ∈ H 0(Ω1 P3 (2)) (cid:27) C⊕4. In particular, there is no Ulrich line bundle on V and we get that the null-correlation bundles on P3 twisted by 2 are the only rank two Ulrich bundles on V . 3. Classification on rational normal scroll Let S = S(a0, . . . , an) be a smooth rational normal scroll, the image of P(E) via the morphism defined by OP(E)(1), where E (cid:27) ⊕n i=0OP1 (ai ) is a vector bundle of rank n + 1 on P1 with 0 < a0 ≤ . . . ≤ an. Letting π : P(E) → P1 be the projection, we may denote by H and F, the hyperplane section corresponding to OP(E)(1) and the fibre corresponding to π∗OP1 (1), respectively. Then we have Pic(S) (cid:27) ZhH, Fi and i=0 ai is the degree of S. We will simply denote OS (aH + bF) by OS (a + b, a), in particular, OS (F) = OS (1, 0). From now on we fix an ample line bundle on S to be OS (H) = OS (1, 1). ωS (cid:27) OS (−(n + 1)H + (c − 2)F), where c :=Pn For the computational purpose, we use the following lemma. Lemma 3.1 ([11], see also [15]). For any i = 0, . . . , n + 1, we have (i) H i(S, OS (a + b, a)) (cid:27) H i(P1, SymaE ⊗ OP1(b)) if a ≥ 0; (ii) H i(S, OS (a + b, a)) = 0 if −n − 1 < a < 0; (iii) H i(S, OS (a + b, a)) (cid:27) H n+1−i (P1, Sym−a−n−1E ⊗ OP1 (c − b − 2)) if a ≤ −n − 1. Recall the dual of the relative Euler exact sequence of S: (4) 0 −→ Ω1 S P1 (1, 1) −→ B := ⊕n i=0OS (ai , 0) −→ OS (1, 1) −→ 0, and so we have ωS P1 (cid:27) OS (−(n + 1)H + cF) (cid:27) OS (c − n − 1, −n − 1). The long exact sequence of exterior powers associated to (4) is (5) 0 −→OS (c − n, −n) −→ ∧nB(−n + 1, −n + 1) dn−1 −→ ∧n−1 B(−n + 2, −n + 2) dn−2 −→ · · · d1 −→ B −→ OS (1, 1) −→ 0. Now (5) splits into (6) 0 −→ Ωi S P1 (i, i) −→ ∧i B −→ Ωi−1 S P1 (i − 1, i − 1) −→ 0 for each i = 1, . . . , n, and we have Im(di ⊗ OS (i − 1, i − 1)) (cid:27) Ωi S P1(i, i) ⊂ ∧i B. Lemma 3.2. The bundle Ωi OS (1, 1) for each i = 0, . . . , n. S P1 (i, i + 1) is Ulrich with slope c − 1 with respect to H = Proof. Consider the twist of the following truncation of (5) by OS (i − 1, i) (7) 0 −→ OS (c − n, −n) −→ ∧n B(−n + 1, −n + 1) −→ · · · −→ ∧i+1B(−i, −i) −→ Ωi S P1 −→ 0. ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE 7 Applying Lemma 3.1 to (7), we obtain that for any k ∈ Z and j ∈ {1, · · · , n} 0 =hn−i+j (OS (c − n + i + k − 1, c − n + i + k)) =hn−i−1+j (∧nB(k − n + i, k − n + i + 1)) = · · · = hj (∧i+1B(k − 1, k)), S P1 (k + i − 1, k + i)) = 0. Thus the sheaf Ωi and it implies that hj (Ωi and hence Ωi to show S P1(i − 1, i) is aCM S P1(i, i +1) is also aCM. Now we may apply the same argument above H 0(∧i+1B(0, 1)) (cid:27) H 0(Ωi S P1 (i, i + 1)) whose dimension is c×(cid:0)n 1) is Ulrich. i(cid:1) = deg(S)×rank(Ωi S P1 (i, i+1)), and so the sheaf Ωi S P1(i, i+ The slope computation is straightforward. (cid:3) Remark 3.3. In practice, instead of working with the Ulrich bundles Ωi S P1 (i, i +1), we shall work with their twists by OS (−H) i.e. with Ωi S P1 (i − 1, i). The reason will become clear in the main Theorem. The spectral sequence will be much simpler if we use bundles with no sections. The bundles Ωi S P1 (i, i + 1) will appear to be the building blocks for all the Ulrich bundles on S. Lemma 3.4. For any i, j ∈ {0, . . . , n} with i , j, we have (8) Hom(Ωi S P1 (i, i), Ω j S P1 (j, j)) = 0. Proof. We prove the vanishing for i > j first. Denote by S P1 ⊗ (Ωi j S P1 (j, j)) (cid:27) Ω H := Hom(Ωi S P1 (i, i), Ω j S P1)∨(j − i, j − i). Similarly to the projective space case in [17, Section 0.1], the Koszul complex pro- vides us with a natural sheaf morphism α : ∧i−j B ∨ −→ H. Indeed, by dualizing (6) we obtain a short exact sequence: 0 −→ (Ω i−j−1 S P1 )∨(j − i − 1, j − i − 1) −→ ∧i−j B ∨ −→ (Ω i−j S P1 )∨(j − i, j − i) −→ 0 and α is obtained from the surjection ∧i−j B ∨ −→ (Ω S P1 )∨(i − j, i − j) and from the j natural map Ωi S P1 (j, j). For each point x ∈ P1, denote by Fx the fibre over x. After restricting to Fx and taking global sections, the map α induces the following usual natural isomorphism over the n–dimensional projective space; see [17, Section 0.1]. S P1 )∨(j − i, j − i) −→ Ω S P1(i, i) ⊗ (Ω i−j i−j (9) H 0(Fx, (∧i−j B ∨)Fx ) (cid:27) HomFx (Ωi Fx (i), Ω j Fx (j)) (cid:27) H 0(Fx, HFx ). It follows by the Grauert theorem that π∗(∧i−j B ∨) and π∗(H) are vector bun- dles of the same rank. The morphism α induces a vector bundle morphism from π∗(∧i−j B ∨) to π∗(H) which is, by (9) and Grauert, an isomorphism on each fibre (of the vector bundle), and hence it is an isomorphism. Therefore, we obtain an isomorphism H 0(P1, π∗(∧i−j B ∨)) (cid:27) H 0(P1, π∗(H)) 8 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS and, since ∧i−j B ∨ has no nontrivial global sections, it follows that H has no nonzero global sections. In the case i < j, we use the well-known fact that on the projective space Pn there are no nontrivial morphisms from Ωi Pn (i) to Ω j Pn (j) for i < j. Then HomFx (Ωi Fx (i), Ω j Fx (j) = 0 and, by Grauert's Theorem, it follows that π∗(H) = 0 and hence Hom(Ωi S P1 (i, i), Ω j S P1 (j, j)) = 0. (cid:3) Remark 3.5. Although the sheaf Ωi S P1 (i, i) is stable, as we shall see, it is not possi- ble to use in the proof of Lemma 3.4 the fact that there is no nontrivial morphism from a stable sheaf to another stable sheaf with smaller slope, because the slope of Ωi S P1 (i, i) with respect to the ample line bundle OS (1, 1) is zero for any i. Now we introduce suitable full exceptional collection; [20, Corollary 2.7], see also [1]. Example 3.6. Consider a function σ : Z → Z⊕2 sending i to the pair of integers (σ1, σ2) such that σ1 + σ2 = i and σ1 − σ2 ∈ {0, 1}. For each integer i ∈ {0, . . . , 2n + 1}, define E ′ i = OS (σ(−i + 1)) and ki = σ2(−i + 1) to have a collection hE ′ 2n+1[k2n+1], . . . , E ′ 0[k0]i. For example, we have E ′ full for Db(S). Apply the left mutation to the pair (E ′ obtain another full exceptional collection 2n+1[k2n+1] = OS (−n, −n)[−n]. Note that this collection is 0[k0]) = (OS , OS (1, 0)) to 1[k1], E ′ (10) hE2n+1[k2n+1], . . . , E0[k0]i, where E1 = OS (−1, 0) and E0 = OS . For example, when n = 3, the full exceptional collection E•[k•] consists of the following: E7[k7] = OS (−3, −3)[−3], E6[k6] = OS (−2, −3)[−3], E5[k5] = OS (−2, −2)[−2], E4[k4] = OS (−1, −2)[−2], E3[k3] = OS (−1, −1)[−1], E2[k2] = OS (0, −1)[−1], E1[k1] = OS (−1, 0) , E0[k0] = OS . Indeed, the collection above is obtained by taking the dual in reverse order of the full exceptional collection E• = hE0, . . . , E2n+1i; E• = hOS , OS (1, 0), OS (0, 1)[1], OS (1, 1)[1], OS (1, 2)[2], . . . , OS (n, n)[n]i, which is also obtained by applying a right mutation to the first pair in the follow- ing full collection hOS (−1, 0), OS , OS (0, 1)[1], OS (1, 1)[1], OS (1, 2)[2], . . . , OS (n, n)[n]i. Now let hF2n+1, . . . , F0i be the right dual collection of E•. Indeed it consists of the following vector bundles Fi = Fi for each i due to (2) F2n+1 = OS (c − 3, −1) , F2n = OS (c − 2, −1), F2n−1 = Ωn−1 S P1(n − 3, n − 1) , F2n−2 = Ωn−1 F3 = Ω1 S P1(0, 1) , F1 = OS (−1, 0) , F0 = OS . S P1(−1, 1) , F2 = Ω1 S P1(n − 2, n − 1), . . . , ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE 9 For example, for i = 2n + 1, we have H •(E2n+1[k2n+1] ⊗ F2n+1) (cid:27) H 2n+1(ωS [−n]) (cid:27) C. Theorem 3.7. Let V be a vector bundle on S. Then V is Ulrich if and only if V admits a filtration 0 = B0 ⊂ B1 ⊂ · · · ⊂ Bn+1 = V with Bi+1/Bi (cid:27) Ωi Ωi S P1 (i, i + 1)⊕ai for some ai ∈ N and each i. Moreover, the bundles S P1(i, i +1)⊕ai associated to this filtration on V coincides, up to a permutation of its factors, with the graded sheaf of the Jordan- Holder filtration of V . S P1 are stable and the graded sheafLn Ωi i=0 Proof. We have verified that each of the bundles Ωi converse implication is clear. S P1 (i, i + 1) is Ulrich and so the Now we prove the direct implication. The idea is to consider the Beilinson type spectral sequence associated to A := V (−1, −1) and identify the members of the graded sheaf associated to the induced filtration as the sheaves mentioned in the statement. We assume due to [12, Proposition 2.1] that H i (A(−j, −j)) = 0 for all 0 ≤ i ≤ n + 1 and 0 ≤ j ≤ n and consider the full exceptional collection E• and left dual collection F• in Exam- ple 3.6. We construct a Beilinson complex, quasi-isomorphic to A, by calculating H i+kj (A ⊗ Ej ) ⊗ Fj with i, j ∈ {1, . . . , 2n + 2} to get the following table. Here we use several vanishing in the intermediate cohomology of A, A(−1, −1), · · · , A(−n, −n) to- gether with vanishing of cohomology H 0 and H n+1: F2n+1 0 0 0 0 ... 0 0 0 ... 0 0 0 E2n+1 F2n 0 H n H n−1 H n−2 ... 0 0 0 ... 0 0 0 E2n F2n−1 F2n−2 0 0 0 0 ... 0 0 0 ... 0 0 0 0 0 H n H n−1 ... H 1 0 0 ... 0 0 0 E2n−1 E2n−2 F2 0 0 0 0 ... H n F1 . . . 0 . . . 0 . . . 0 . . . 0 . . . ... ... . . . 0 . . . H n−1 H n . . . H n−2 H n−1 ... . . . . . . . . . . . . ... H 2 H 1 0 E1 ... H 1 0 0 E2 F0 0 0 0 0 ... 0 0 0 ... 0 0 0 E0 Lemma 3.8. For any i, j ∈ {1} ∪ {2, 4, . . . , 2n} with i > j, we have Hom(Fi , Fj ) = 0. Proof. By Lemma 3.4, it is enough to consider the case when i = 2n and in this case we have Hom(F2n, Fj ) (cid:27) H 0(Fj (2 − c, 1)). If j = 1, then it is isomorphic to H 0(OS (1 − c, 1)) (cid:27) H 0(P1, E ⊗ OP1 (−c)) by Lemma 3.1 and it is trivial. For j > 1, we need to show that H 0(Ωi S P1 (i + 1 − c, i + 1)) = 0. Consider the twist of (7) by 10 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS OS (i + 1 − c, i + 1) and apply Lemma 3.1 to get that hn−i (OS (i + 1 − n, i + 1 − n) = hn−i−1(∧nB ⊗ OS (i + 2 − c − n, i + 2 − n)) = · · · = h0(∧i+1B ⊗ OS (1 − c, 1)) = 0 and it implies the vanishing we want. (cid:3) By Lemma 3.8, we may conclude that all the entries off the diagonal must be zero and thus we get F2n+1 F2n F2n−1 F2n−2 0 0 0 0 ... 0 0 0 ... 0 0 0 E2n+1 0 b 0 0 ... 0 0 0 ... 0 0 0 E2n 0 0 0 0 ... 0 0 0 ... 0 0 0 0 0 0 H n−1 ... 0 0 0 ... 0 0 0 E2n−1 E2n−2 . . . F2 F1 F0 0 . . . 0 0 . . . 0 0 . . . 0 0 . . . 0 ... ... ... 0 . . . 0 0 . . . 0 0 . . . 0 ... ... ... . . . H 1 0 0 0 . . . . . . 0 0 E0 E2 . . . 0 0 0 0 ... 0 0 0 ... 0 a 0 E1 where a := h1(A(−1, 0)) and b := hn(A(−n + 1, −n)). This yields to the desired filtra- tion. dles Ωi Ωi S P1 for all i. Assume that Ωi In order to verify the last assertion, it suffices to check the stability of the bun- S P1 was not stable for some i. Then, since S P1(i, i +1) is Ulrich and so semistable, it will contain a stable Ulrich subbundle, say F , with the same slope c − 1. Applying the first part of the result, we obtain a filtration on F whose terms have all the same slope c − 1, and, in particular, from j the stability, the bundle F must be isomorphic to Ω S P1 (j, j + 1) for some j. There- S P1(i, i + 1). Now S P1(i, i + 1), which implies the fore we obtain an injective morphism from Ω it follows from Lemma 3.4 that i = j and so F (cid:27) Ωi stability of Ωi j S P1 (j, j + 1) to Ωi S P1. (cid:3) Remark 3.9. A consequence of the main result is that the moduli spaces of Ulrich bundles are zero-dimensional. This follows from the fact that when we construct moduli spaces of semistable bundles, we work with S–equivalence classes. Remark 3.10. It follows directly from Theorem 3.7 that, for n ≥ 2, the only stable Ulrich bundles of rank ≤ n − 1 are OS (0, 1) and OS (c − 1, 0). In particular, if n ≥ 3, there are no stable Ulrich bundles of rank two on S and all the strictly semistable Ulrich bundles of rank two are extensions In the case n = 2 the only stable Ulrich bundle is Ω1 0 −→ OS (0, 1) −→ V −→ OS (c − 1, 0) −→ 0. S P1 (1, 2). ULRICH BUNDLES ON SMOOTH PROJECTIVE VARIETIES OF MINIMAL DEGREE 11 Remark 3.11. In surface case, the previous result is compatible with [2, Theorem 2.1]. Indeed, for the bundle E = OP1 (β) ⊕ OP1 (β − e) on P1 with β > e > 0, the only Ulrich line bundles are OS (c − 1, 0) = OS ((2β − e − 1)F) and OS (0, 1) = OS (C0 + (β − 1)F), where C0 is the minimal section. All the other Ulrich bundles arise from extension of the form 0 −→ OS (0, 1)⊕a −→ V −→ OS (c − 1, 0)⊕b −→ 0 and hence, there are no stable Ulrich bundles of rank ≥ 2. This was previously proven in [13]. Remark 3.12. For n > 2 we may compute Ext1 S (Ω1 S P1 (0, 1), OS (−1, 0)) = H 1(Ω1 S P1 (1, 1)∨), and from the dual of (4) we get h1(Ω1 S P1 (1, 1)∨) = h1(OS (−ai , 0)) = 0 if and only if ai = 1 for any i, so in this case the filtration can be simpler. Now take (a0, . . . , an) = (1, . . . , 1), i.e. S = S(1, . . . , 1) (cid:27) P1 × Pn. Then it is embedded into P2n+1 by the complete linear system OS (1, 1) as a Segre variety. Then we have Ω1 Pn . Using the notation above, we get Pn and so Ωi (cid:27) OP1 ⊠ Ω1 (cid:27) OP1 ⊠ Ωi S P1 Ext1 S (Ωi S P1 (i − 1, i), Ω S P1 S P1 (j − 1, j)) = H 1(OP1 (j − i) ⊠ Ω j j Pn(j) ⊗ Ωi Pn (i)∨) which is zero for each i ≤ j + 1 and equals H 0(OP1 (i − j − 2))∨ ⊗ ∧i−j Cn+1 for i ≥ j + 2. Example 3.13. On S = P1 × P2, a bundle V is Ulrich if and only if it admits a filtration 0 = B0 ⊂ B1 ⊂ B2 ⊂ B3 = V with B1 (cid:27) OS (0, 1)⊕a0 , B2/B1 (cid:27) OP1 (1) ⊠ Ω1 P2 (2)⊕a1 , B3/B2 (cid:27) OS (3, 0)⊕a2 for some a0, a1, a2 ∈ N. So the only stable Ulrich bundle of rank ≥ 2 is V (cid:27) OP1 (1) ⊠ Ω1 P2 (2) and all the other indecomposable Ulrich bundles arise from the extensions (11) 0 −→ OS (0, 1)⊕a0 −→ V −→ OS (3, 0)⊕a2 −→ 0. This case has been studied in more details in [6], where all the aCM bundles are classified. Example 3.14. On S = P1 × P3, a bundle V is Ulrich if and only if it admits a filtration 0 = B0 ⊂ B1 ⊂ B2 ⊂ B3 ⊂ B4 = V with B1 (cid:27) OS (0, 1)⊕a0 , B2/B1 (cid:27) OP1 (1) ⊠ Ω1 B3/B2 (cid:27) OP1 (2) ⊠ Ω2 P3 (3)⊕a2 , B4/B3 (cid:27) OS (4, 0)⊕a3 P3 (2)⊕a1 , for some a0, a1, a2, a3 ∈ N. So all the Ulrich bundles arise from the extensions 0 −→ OS (0, 1)⊕a0 ⊕ OP1(1) ⊠ Ω1 P3 (2)⊕a1 −→ V −→ OP1 (2) ⊠ Ω2 P3 (3)⊕a2 ⊕ OS (4, 0)⊕a3 −→ 0. 12 M. APRODU, S. HUH, F. MALASPINA AND J. PONS-LLOPIS References [1] M. Aprodu and V. Brınzanescu, Beilinson type spectral sequences on scrolls, Moduli spaces and vector bundles, London Math. Soc. Lecture Note Series, 359, Cambridge Univ. Press, Cambridge, (2009) 426–436. [2] M. Aprodu, L. Costa and R. M. Mir ´o-Roig, Ulrich bundles on ruled surfaces, arXiv:1609.08340v1 [math.AG], to appear in J. Pure Appl. Alg. [3] V. Ancona and G. Ottaviani, Some applications of Beilinson's theorem to projective spaces and quadrics, Forum Math. 3 (1991), 157–176. [4] C. Bohning, Derived categories of coherent sheaves on rational homogeneous manifold. Doc. Math. 11 (2006), 261–331. [5] M. Casanellas and R. Hartshorne (with an appendix by F. Geiss and F. -O. Schreyer), Stable Ulrich bundles, Int. J. Math. 23 (2012), no. 8, 1250083. [6] G. Casnati, D. Faenzi, F. Malaspina and G. Sanna, Smooth projective varieties with only a finite number of ACM bundles which are not Ulrich, in preparation. [7] I. Coskun, L. Costa, J. Huizenga, R. M. Mir ´o-Roig and M. Woolf, Ulrich Schur bundles on flag varieties, J. Algebra 474 (2017), 49–96. [8] E. Coskun, R. S. Kulkarni and Y. Mustopa, The geometry of Ulrich bundles on del Pezzo surfaces, J. Algebra 375 (2013), 280–301. [9] L. Costa and R. M. Mir ´o-Roig, GL(V )-invariant Ulrich bundles on Grassmannian, Math. Ann. 361 (2014), no. 1–2, 443–457. [10] L. Costa and R. M. Mir ´o-Roig, Homogeneous ACM bundles on a Grassmannian, Adv. Math. 289 (2016), 95–113. [11] D. Eisenbud, J. Harris, On varieties of minimal degree (a centennial account). Algebraic geome- try, Bowdoin, 1985 (Brunswick, Maine, 1985). Proceedings of Symposia in Pure Mathematics 46 (1987), no. 1, 3–13 [12] D. Eisenbud, F.- O. Schreyer, Resultants and Chow forms via exterior syzygies, J. Amer. Math. Soc. 16 (2003), no. 3, 537–579. [13] D. Faenzi and F. Malaspina, Surfaces of minimal degree of tame representation type and mutations of Cohen-Macaulay modules, Adv. Math. 310 (2017), 663–695. [14] A. Gorodentsev and S. A. Kuleshov, Helix theory, Mosc. Math. J. 4(2) (2004), 377–440. [15] R. M. Mir ´o-Roig, The representation type of rational normal scrolls, Rend. Circ. Mat. Palermo 62 (2013), 153–164. [16] R. M. Mir ´o-Roig and J. Pons-Llopis, Representation type of rational aCM surfaces X ⊂ P4, Alg. Rep. Theory 16 (2013), 1135–1157. [17] M. S. Narasimhan and G. Trautmann, Compactification of MP3 (0, 2) and Poncelet pairs of conics, Pacific J. Math. 145 (1990), no. 2, 255–365. [18] C. Okonek, M. Schneider, and H. Spindler, Vector bundles on complex projective spaces, Modern Birkhauser Classics, Birkhauser/Springer Basel AG, Basel, 2011, Corrected reprint of the 1988 edition, With an appendix by S. I. Gelfand. [19] E. Park, Smooth varieties of almost minimal degree, J. Algebra 314 (2007), 185–208. [20] D. O. Orlov, Projective bundle, monoidal transformations, and derived categories of coherent sheaves, Russian Acad. Sci. Izv. Math. 41 (1993), 133–141. [21] A. N. Rudakov, Helices and vector bundles, Seminaire Rudakov, volume 148. Cambridge University Press, 1990. Faculty of Mathematics and Computer Science, University of Bucharest, 14 Academiei Street, 010014 Bucharest, Romania & "Simion Stoilow" Institute of Mathematics of the Romanian Acad- emy, P. O. Box 1-764, 014700 Bucharest, Romania E-mail address: [email protected] & [email protected] Sungkyunkwan University, Suwon 440-746, Korea E-mail address: [email protected] Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy E-mail address: [email protected] Department of Mathematics, Kyoto University, Kyoto, Japan E-mail address: [email protected]
1609.03446	2	1609	2016-10-16T20:53:28	Foundations of Boij-S\"oderberg Theory for Grassmannians	[ "math.AG", "math.AC" ]	Boij-S\"oderberg theory characterizes syzygies of graded modules and sheaves on projective space. This paper continues earlier work with S. Sam, extending the theory to the setting of $GL_k$-equivariant modules and sheaves on Grassmannians. Algebraically, we study modules over a polynomial ring in $k n$ variables, thought of as the entries of a $k \times n$ matrix. We give equivariant analogues of two important features of the ordinary theory: the Herzog-K\"uhl equations and the pairing between Betti and cohomology tables. As a necessary step, we also extend previous results, concerning the base case of square matrices, to cover complexes other than free resolutions. Our statements specialize to those of ordinary Boij-S\"oderberg theory when $k=1$. Our proof of the equivariant pairing gives a new proof in the graded setting: it relies on finding perfect matchings on certain graphs associated to Betti tables. Finally, we give preliminary results on $2 \times 3$ matrices, exhibiting certain classes of extremal rays on the cone of Betti tables.	math.AG	math	FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS NIC FORD AND JAKE LEVINSON Abstract. Boij-Soderberg theory characterizes syzygies of graded modules and sheaves on projective space. This paper continues earlier work with S. Sam [FLS16], extending the theory to the setting of GLk-equivariant modules and sheaves on Grassmannians. Algebraically, we study modules over a polynomial ring in kn variables, thought of as the entries of a k n matrix. We give equivariant analogues of two important features of the ordinary theory: the Herzog-Kuhl equations and the pairing between Betti and cohomology tables. As a nec- essary step, we also extend the result of [FLS16], concerning the base case of square matrices, to cover complexes other than free resolutions. Our statements specialize to those of ordinary Boij-Soderberg theory when k " 1. Our proof of the equivariant pairing gives a new proof in the graded setting: it relies on finding perfect matchings on certain graphs associated to Betti tables. Finally, we give preliminary results on 2 3 matrices, exhibiting certain classes of extremal rays on the cone of Betti tables. 6 1 0 2 t c O 6 1 ] . G A h t a m [ 2 v 6 4 4 3 0 . 9 0 6 1 : v i X r a 1. Introduction 1.1. Boij-Soderberg theory. Let R " Crx1, . . . , xns be a polynomial ring and M a graded, finitely-generated R-module. The Betti table of M is the collection of numbers βijpM q :" # degree-j generators of the i-th syzygy module of M. That is, by definition, the minimal free resolution of M has the form M Ð F0 Ð F1 Ð Ð Fn Ð 0, with Fi "àj Rr´jsβij . These numbers encode much of the structure of M , such as its dimension and whether or not it is Cohen-Macaulay. They also describe geometrical properties of the associated sheaf on PpCnq. Boij-Soderberg theory was initially concerned with describing which sets of numbers βij arise as Betti tables of modules. The key early observation was that it is easier to determine which tables arise up to scalar multiple, and as such, the goal was to characterize the Boij- Soderberg cone BSn of positive scalar multiples of Betti tables [BS08, ES09, EFW11]. More recent work has focused on modules over multigraded and toric rings [EE12], and some homogeneous coordinate rings [BBEG12, GS15, KS15], as well as more detailed homological questions [NS13, BZEKS13, EES13]. A good survey of the field is [Flø12]. 1 FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 2 In each of these cases, an important feature of the theory is a duality between Betti [ES09, EE12] For a tables and cohomology tables of sheaves on an associated variety. coherent sheaf E on PpCnq, the cohomology table is the collection of numbers γijpEq :" dimC H ipEp´jqq, giving all the sheaf cohomology of all the twists of E. For graded modules and projective spaces, the duality takes the form of a bilinear pairing of the cones of Betti and cohomology tables, and it produces a point in the simplest Boij- Soderberg cone BS1. The inequalities defining BS1 therefore pull back to nonnegative bilinear pairings between Betti and cohomology tables, and these pulled-back inequalities fully characterize the two cones. In particular, the Boij-Soderberg cone (in the graded setting) is rational polyhedral, and its extremal rays and supporting hyperplanes are explicitly known. The extremal rays correspond to pure Betti tables. These are the simplest possible tables, having only one nonzero entry in each column (that is, for each i, only one βij is nonzero). Similarly, the supporting hyperplanes come from pairing with the cohomology of vector bundles with so-called supernatural cohomology . Much less is known about the Boij-Soderberg cone in other settings. 1.2. Grassmannian Boij-Soderberg theory. The goal of this paper is to continue ear- lier work of the authors, joint with S. Sam [FLS16], on extending the theory to the setting of Grassmannians Grpk, Cnq. On the geometric side, we will be interested in the cohomol- ogy of coherent sheaves on Grpk, Cnq. On the algebraic side, we consider the polynomial ring in kn variables (k ď n), Rk,n " C"xij : 1 ď i ď k 1 ď j ď nı, thought of as the entries of a k n matrix. The group GLk acts on Rk,n, and we are interested in (finitely-generated) equivariant modules M , i.e., those with a compatible GLk action. Aside from the inherent interest of understanding sheaf cohomology and syzygies on Grassmannians, there is hope that this setting might avoid some obstacles faced in other extensions of Boij-Soderberg theory, e.g. to products of projective spaces. For example, in the 'base case' of square matrices pn " kq, the 'irrelevant ideal' is the principal ideal gener- ated by the determinant, and the Boij-Soderberg cone has an especially elegant structure (see below, Section 1.3.2). We define equivariant Betti tables βpM q using the representation theory of GLk. Let SλpCkq denote the irreducible GLk representation of weight λ, where Sλ is the Schur functor. There is a corresponding free module, namely SλpCkq bC Rk,n, and every equivariant free module is a direct sum of these. Then βpM q is the collection of numbers βi,λpM q :" # copies of SλpCkq in the generators of the i-th syzygy module of M. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 3 Thus, by definition, the minimal equivariant free resolution of M has the form M Ð F0 Ð Ð Fn Ð 0, with Fi "àλ SλpCkqβiλ b Rk,n. Next, for E a coherent sheaf on Grpk, Cnq, we will define the GL-cohomology table γpEq, generalizing the usual cohomology table: γi,λpEq :" dim H ipE b SλpSqq, where S is the tautological vector bundle on Grpk, Cnq of rank k. We write BTk,n :"Ài,λ Q for the space of abstract Betti tables. Similarly, we write CTk,n :"Àiśλ Q for the space of abstract GL-cohomology tables. Remark 1.1. The case k " 1 reduces to the ordinary Boij-Soderberg theory, since an action of GL1 is formally equivalent to a grading; the module Rr´js is just SpjqpCq b R. Note also that S " Op´1q on projective space. The initial questions of Boij-Soderberg theory concerned finite-length graded modules M , i.e. those annihilated by a power of the homogeneous maximal ideal, and more generally Cohen-Macaulay modules. Similarly, we restrict our focus (for now!) on the following class of modules, which specializes to finite-length modules when k " 1: Condition 1.2 (The modules of interest). We consider finitely-generated Cohen-Macaulay modules M such that aannpM q " Pk, the ideal of maximal minors of the k n matrix. Viewing SpecpRk,nq " HompCk, Cnq as the affine variety of k n matrices, this means M is set-theoretically supported on the locus of rank-deficient matrices. That is, the sheaf associated to M on Grpk, Cnq is zero. For this reason, we refer to Pk as the irrelevant ideal for this setting. The Cohen-Macaulayness assumption means that so its minimal free resolution has length n ´ k ` 1. pdimpM q " dimpRPk q " n ´ k ` 1, Definition 1.3. We define the equivariant Boij-Soderberg cone BSk,n Ă BTk,n as the positive linear span of Betti tables βpM q, where M satisfies the assumptions of Condition 1.2. We define the Eisenbud-Schreyer cone ESk,n Ă CTk,n as the positive linear span of GL-cohomology tables of all coherent sheaves E on Grpk, Cnq. We wish to understand the cones BSk,n and ESk,n generated by equivariant Betti tables of and GL-cohomology tables. Remark 1.4 (Multiplicities and ranks). The irreducible representations SλpCkq need not be one-dimensional. As such, the corresponding free modules need not have rank 1. We will write a tilde rβ to denote the rank of the λ-isotypic component (rather than its multiplicity), and likewise write ĂBTk,n and ĄBSk,n for the spaces of rank Betti tables. Of course, we may switch between them by rescaling each entry βi,λ by dimpSλpCkqq. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 4 1.3. Results of this paper. We will generalize two important results from the existing theory on graded modules: the Herzog-Kuhl equations and the pairing between Betti and cohomology tables. Along the way, we also extend our existing result on equivariant modules for the square matrices. 1.3.1. Equivariant Herzog-Kuhl equations. In the graded setting, the Herzog-Kuhl equa- tions are n linear conditions satisfied by the Betti tables of finite-length modules M . They say, essentially, that the Hilbert polynomial of M vanishes identically, i.e., each of its coefficients is zero. We give the following equivariant analogue. Theorem 1.5. Let M be an equivariant Rk,n-module with Betti table βpM q. Assume (twisting up if necessary) that M is generated in positive degree. There is a system of `n fit inside a k pn ´ kq rectangle, called the equivariant Herzog-Kuhl equations. The following are equivalent: k linear conditions on βpM q, indexed by partitions µ ě 0 that (i) βpM q satisfies the equivariant Herzog-Kuhl equations. (ii) M is annihilated by a power of the ideal Pk of maximal minors. (iii) The sheaf associated to M on Grpk, Cnq vanishes. In particular, the hypotheses of Condition 1.2 are equivalent to the equivariant Herzog-Kuhl equations, together with the conditions βi,λ " 0 for all i ą n ´ k ` 1. We state the equations in Section 3.2, using the combinatorics of standard Young tableaux. We also give an interpretation in terms of equivariant K-theory, namely that the class of M lies in the kernel of the map of K-theory rings K GLkpSpecpRk,nqq Ñ KpGrpk, Cnqq induced by restriction (to the locus of full-rank matrices) and descent. 1.3.2. The Boij-Soderberg cone for square matrices. As in the ordinary theory, we expect the smallest case n " k to play an important role. It serves as the base case of the theory and the target of the equivariant Boij-Soderberg pairing (Section 1.3.3). Rank tables rβ turn out to be more significant here, so we will state results in terms of the cone ĄBSk,k. For square matrices, the modules of interest are Cohen-Macaulay and haveaannpM q " pdetq, so they have projective dimension 1. The cone ĄBSk,k is fully understood: Theorem 1.6 ([FLS16, Theorem 1.2]). The cone ĄBSk,k is rational polyhedral. Its sup- These rays correspond to pure tables with Ąβ0,λ " Ąβ1,µ " 1 and all other entries zero. Up porting hyperplanes are indexed by order ideals in the extended Young's lattice Y of GLk-representations. Its extremal rays are indexed by comparable pairs λ Ĺ µ from Y. to scaling, these tables come from free resolutions of the form M Ð SλpCkq'c0 b R Ð SµpCkq'c1 b R Ð 0, with all generators in type λ and all syzygies in type µ. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 5 We will need a slightly more general result for the purposes of the equivariant Boij- k,n, is the positive linear span of (rank) Betti tables of bounded minimal complexes F‚ of equivariant free modules, such that F‚ is exact away from the locus of rank-deficient matrices. Soderberg pairing, a derived analog to ĄBSk,n. Definition 1.7. The derived Boij-Soderberg cone, denotedĄBSD In this definition, we assume only that the homology modules M have aannpM q Ě Pk, not that equality holds. We also do not assume Cohen-Macaulayness. Thus, ĄBSD includes, for example, homological shifts of elements of ĄBSk,k, and Betti tables of longer tables, written rβ"λ iÐÝ µı, for i P Z and λ Ĺ µ. These are the tables with Ąβi,λ " Čβi`1,µ " Theorem 1.8. The coneĄBSD shifts of those of ĄBSk,k, spanned by the tables rβ"λ indexed by tuples p. . . , S´1, S1, S3, . . .q of convex subsets Si Ď Y, one chosen for every other spot along the complex. iÐÝÝ µı. The supporting hyperplanes are complexes. The simplest tables in the derived cone are homologically shifted pure k,k is rational polyhedral. Its extremal rays are the homological 1 and all other entries zero. We show: k,k The key idea in the above theorem is that these Betti tables are characterized by certain perfect matchings. This idea is also crucial in our construction of the pairing between Betti and cohomology tables, so we discuss it now. We introduce a graph-theoretic model of a rank Betti table (in the case of free resolutions, this construction is implicit in [FLS16, Lemma 3.6]). Definition 1.9 (Betti graphs). Let rβ P ĂBTk,k have nonnegative integer entries. The Betti graph Gprβq is defined as follows: ‚ The vertex set contains Ąβi,λ vertices labeled pi, λq, for each pi, λq, ‚ The edge set contains, for each i, all possible edges pi, λq Ð pi ` 1, µq with λ Ĺ µ. Note that this graph is bipartite: every edge connects an even-indexed and an odd-indexed vertex. Recall that a perfect matching on a graph G is a subset of its edges, such that every vertex of G appears on exactly one chosen edge. A perfect matching on Gprβq is equivalent to a decomposition of rβ as a positive integer combination of homologically-shifted pure tables: an edge pi, λq Ð pi ` 1, µq corresponds to a pure summand rβ"λ iÐÝ µı. Thus, an equivalent characterization of ĄBSD Theorem 1.10. Let rβ P BTk,k have nonnegative integer entries. Then rβ P ĄBSD only if Gprβq has a perfect matching. k,k if and k,k is: FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 6 Our proof proceeds by exhibiting this perfect matching using homological algebra. The k,k then follow from Hall's Matching Theorem; see Section supporting hyperplanes of ĄBSD 4.2 for the precise statement. 1.3.3. The pairing between Betti tables and cohomology tables. We now turn to the Boij- Soderberg pairing. This will be a bilinear pairing between abstract Betti tables β and cohomology tables γ, satisfying certain nonnegativity properties when restricted to realiz- able tables. Definition 1.11. Let β P BTk,n and γ P CTk,n be an abstract Betti table and GL- cohomology table. The equivariant Boij-Soderberg pairing is given by (1.1) rΦ : BTk,n CTk,n Ñ ĂBTk,k, ÞÑ rΦpβ, γq, with rΦ the (derived) rank Betti table with entries Ąϕi,λpβ, γq " ÿp´q"i βp,λ γq,λ. pβ, γq (1.2) In this definition, recall that the homological index of a complex decreases under the boundary map. first quadrant of the plane, whose pp, qq-entry is the collection of numbers βp,λ γq,λ for all λ. Only finitely-many of these are nonzero. The line p ´ q " i is an upwards-sloping Here is how to read the definition of rΦ. (See Example 1.14 below.) Form a grid in the diagonal through this grid, and Ąϕi,λ is the sum of the λ terms along this diagonal. cohomology table γ, and produces a rank Betti table rΦ. Intuitively, the entries of γ are dimensions of certain vector spaces (from sheaf cohomology), which, we will see, arise with multiplicities given by β in a certain spectral sequence. In particular, the quantities in (1.2) are again dimensions of vector spaces -- that is, they give a rank table. Remark 1.12. We emphasize that the pairing takes a multiplicity Betti table β and a The final main result of this paper is the nonnegativity of the pairing: k,k. The same is true with BSk,n replaced by BSD Theorem 1.13 (Pairing the equivariant cones). The pairing rΦ restricts to a map of cones, k,k (which we give explicitly) pull back to nonnegative bilinear pairings of Betti and cohomology tables, and the Betti graph BSk,n ESk,n ÑĄBSD In particular, the defining inequalities of the cone ĄBSD of rΦpβ, γq has a perfect matching. We think of this as a reduction to the base case of square Our proof proceeds by constructing a perfect matching on rΦpβ, γq, not by passing to ac- matrices (k " n). A geometric consequence is that each equivariant Betti table induces many interesting linear inequalities constraining sheaf cohomology on Grpk, Cnq. tual modules over Rk,k. It would be interesting to see a 'categorified' form of the pairing, in k,n on the source. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 7 the style of Eisenbud-Erman [EE12]. Such a pairing would construct, from a complex F‚ of Rk,n-modules and a sheaf E, a module (or complex) over Rk,k. Theorem 1.13 would follow from showing that this module is supported along the determinant locus (or that the com- plex is exact away from the determinant locus). The authors welcome any communication or ideas in this direction. Example 1.14. Let us pair the following tables for k " 2, n " 3. Both are realizable; the cohomology table is for the sheaf E " Op1q ' Op´1q. (1.3) βp,λ 1 0 2 4 ´ ´ ´ 1 ´ ´ 9 ´ ´ 3 3 ´ ´ 1 ´ ´ 1 γq,λ 1 2 0 1 ´ 3 ´ 3 ´ 1 ´ ´ ´ ´ ´ ´ 3 ´ ´ ´ 1 We arrange the pairwise products in a first-quadrant grid. The sums along the diagonals tp ´ q " iu result in the rank Betti table rΦ: ´ ´ 1 (1.4) q 4 3 3 12 9 p ´ Ąϕi,λ ´1 0 1 12 ´ 4 ´ 3 ´ ´ ´ 9 ´ ´ 3 ´ 1 ´ unique (this is not true in general): ´1 ÐÝÝÝ ´1 ÐÝÝÝ rΦ " 3 rβ" k,k. The decomposition of rΦ into pure tables happens to be Finally, we check that rΦ PĄBSD ı ` rβ" ÐÝÝ ı ` 3 rβ" This corresponds to an essentially-unique perfect matching on GprΦq. 1.4. Structure of the paper. Section 2 contains background on algebra and representa- tion theory. Sections 3, 4 and 5 respectively establish the equivariant Herzog-Kuhl equa- tions, the results on square matrices, and the pairing of Betti and cohomology tables. Finally, Section 6 has some preliminary results and examples for the case k " 2, n " 3. 0 ı ` 9 rβ" 0 ÐÝÝ ı . O O / / FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 8 1.5. Acknowledgments. This work has benefited substantially from conversations and collaboration with many people: Daniel Erman (who introduced the second author to Boij-Soderberg theory), Steven Sam, David Speyer, Greg Muller and Maria Gillespie. We are particularly grateful to David Speyer for discussions surrounding Theorem 1.5 (the equivariant Herzog-Kuhl equations), and to both Greg Muller and David Speyer for several productive discussions around the proof of the numerical pairing, Theorem 1.13. Finally, computations of free resolutions and Betti tables in Macaulay2 [GS], and facets and rays of cones in Sage [Dev16] (via SageMathCloud) have been, and continue to be, instrumental. 2. Background 2.1. Spaces of interest. Throughout, let V, W be fixed C-vector spaces of dimensions k and n, with k ď n. We set X " HompV, W q, Rk,n " SympHompV, W qq -- C"xij : 1 ď i ď k 1 ď j ď nı, so X " SpecpRk,nq, the affine variety of k n matrices, and Rk,n is the polynomial ring whose variables are the entries of the matrix. We also consider the subvarieties of full-rank and rank-deficient matrices, U " EmbpV, W q " tT : kerpT q " 0u, Xk´1 " X ´ U, which are open and closed, respectively. The locus Xk´1 is integral and has codimension n ´ k ` 1. Its prime ideal Pk is generated by the `n k maximal minors of the k n matrix pxijq. Each of the spaces X, Rk,n, U and Xk´1 has an action of GLpV q and GLpW q; we will primarily care about the GLpV q action. 2.2. GL-Representation theory. A good introduction to these notions is [Ful96]. The irreducible algebraic representations of GLpV q are indexed by weakly-decreasing integer sequences λ " pλ1 ě ě λkq, where k " dimpV q. We write SλpV q for the corresponding representation, and dλpkq for its dimension. We call Sλ a Schur functor . If λ has all nonnegative parts, we write λ ě 0 and say λ is a partition. In this case, SλpV q is functorial for linear transformations V Ñ W . If λ has negative parts, Sλ is only functorial for isomorphisms V „ÝÑ W . We often represent partitions by their Young diagrams: λ " p3, 1q ÐÑ λ " . We partially order partitions and integer sequences by containment: λ Ď µ if λi ď µi for all i. We write Y for the poset of all partitions with this ordering, called Young's lattice. We write Y for the set of all weakly-decreasing integer sequences; we call it the extended FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 9 Young's lattice. Schur functors include symmetric and exterior powers: λ " d" ðñ SλpV q "ľd pV q, λ " dhkkikkj ðñ SλpV q " SymdpV q. We'll write detpV q for the one-dimensional representation ŹdimpV qpV q " S1kpV q. We may always twist a representation by powers of the determinant: detpV qba b Sλ1,...,λkpV q " Sλ1`a,...,λk`apV q for any integer a P Z. This operation is invertible and can sometimes be used to reduce to considering the case when λ is a partition. 2.3. Equivariant rings and modules. If R is a C-algebra with an action of GLpV q, and S is any GLpV q-representation, then S bC R is an equivariant free R-module; it has the universal property HomGLpV q,RpS bC R, M q -- HomGLpV qpS, M q for all equivariant R-modules M . The basic examples will be the modules SλpV q b R. Let R " Rk,n be the polynomial ring defined above. Its structure as a GLpV q GLpW q representation is known as the Cauchy identity : Rk,n " Sym‚pHompV, W qq -- àλě0 SλpV q b SλpW q. Note that the prime ideal Pk and the maximal ideal m " pxijq of the zero matrix are GLpV q- and GLpW q-equivariant. Let M be a finitely-generated GLpV q-equivariant R-module. The module Tori RpR{m, M q naturally has the structure of a finite-dimensional GLpV q-representation. We define the equivariant Betti number βi,λpM q as the multiplicity of the Schur functor SλpV q in this Tor module, i.e. Tori RpR{m, M q -- àλ SλpV q'βi,λpM q (as GLpV q-representations). By semisimplicity of GLpV q-representations, any minimal free resolution of M can be made equivariant, so we may instead define βi,λ as the multiplicity of the equivariant free module SλpV q b R in the i-th step of an equivariant minimal free resolution of M : M Ð F0 Ð F1 Ð Ð Fd Ð 0, where Fi "àλ SλpV qβi,λpM q b R. All other notation on Betti tables is as defined in Section 1.2. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 10 3. The equivariant Herzog-Kuhl equations In this section we derive the equivariant analogue of the Herzog-Kuhl equations. This will be a system of linear conditions on the entries of an equivariant Betti table. It will detect when the resolved module M is supported only along the locus of rank-deficient matrices. 3.1. K-theory rings. For background on equivariant K-theory, we refer to the original paper by Thomason [Tho87]; a more recent discussion is [Mer05]. Excision in equivariant K-theory ([Tho87, Theorem 2.7]) gives the right-exact sequence of abelian groups K GLpV qpXk´1q iÝÑ K GLpV qpHompV, W qq j ÝÑ K GLpV qpU q Ñ 0. The pullback j, induced by the open inclusion j : U ãÑ X, is a map of rings. The pushforward i, induced by the closed embedding i : Xk´1 ãÑ X, is only a map of abelian groups. Its image is the ideal I generated by the classes of modules supported along the rank-deficient locus Xk´1. We do not attempt to describe the first term. For the second term, we have ([Tho87, Theorem 4.1] or [Mer05, Example 2 and Corollary 12]) K GLpV qpHompV, W qq -- Zrx 1 , . . . , x k sSk , the ring of symmetric Laurent polynomials in k variables (essentially the representation ring of GLpV q). Here, the class of the equivariant R-module SλpV q bC R is identified with the Schur polynomial sλpt1, . . . , tkq. If M is a finitely-generated R-module, its equivariant minimal free resolution expresses the K-class rM s as a finite alternating sum of Schur polynomials. In other words, the equivariant Betti table determines the K-class: rM s "ÿi,λ p´1qiβi,λpM qsλptq. An equivalent approach is to write M -- àλ SλpV qcλpM q as a GLpV q-representation, and define the equivariant Hilbert series of M , HM ptq "ÿλ śk " cλpM qsλptq f ptq i"1p1 ´ tiqn for some symmetric function f ptq. Then f ptq is the K-theory class of M . (If we forget the GLpV q action and remember only the grading of M , we recover the usual Hilbert series.) To see that these definitions agree, note that the second definition is additive in short exact sequences, hence is well-defined on K-classes. Replacing M by its equivariant minimal free resolution, it suffices to consider indecomposable free modules M " SλpV q bC R. Then the Cauchy identity shows HM ptq " sλptq. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 11 It will be convenient in this section to restrict to modules M generated in positive degree, i.e. βi,λpM q ‰ 0 implies λ ě 0. In this case, the class of M is a polynomial, not a Laurent polynomial. We write K GLpV q pHompV, W qq for this subring. ` Finally, we have for the third term (cf. [Tho87, Proposition 6.2]) K GLpV qpU q -- KpU {GLpV qq " KpGrpk, W qq, because the action of GLpV q is free on U . The structure of this ring is well-known from [KK90] or [Buc02]). We will only need to know K-theoretic Schubert calculus (e.g. the following: ators, indexed by partitions µ fitting inside a k pn ´ kq rectangle. These correspond to the classes rOµs of structure sheaves of Schubert varieties. It is easy to check that K GLpV q pHompV, W qq Ñ K0pGrpk, W qq is also surjective (because, e.g., matrix Schubert varieties are generated in positive degree). it is a free abelian group with an additive basis consisting of `n k gener- ` 3.2. Modules on the rank-deficient locus and the equivariant Herzog-Kuhl equa- tions. From the surjection K GLpV q pHompV, W qq Ñ K0pGrpk, W qq, we see that the ideal ` I 1 :" I X K GLpV q ` pHompV, W qq, cutting out the ideal, indexed appropriately by partitions. That is, given a K-class written in the Schur basis, k. We wish to find exactly this many linear equations as a linear subspace, has co-rank `n f " ÿλě0 aλsλ P K GLpV q ` pHompV, W qq, we wish to have coefficients bλµ for each µ Ď , such that We will then apply these equations in the case where f is the class of a module M , and aλbλµ " 0 for all µ Ď . f P I 1 if and only if ÿλě0 aλ "ÿi p´1qiβi,λpM q comes from the equivariant Betti table of M . Our approach is to prove the following: Theorem 3.1. We have I 1 " spanC sλp1 ´ tq : λ Ę (. We will prove Theorem 3.1 in the next section. Here is how it leads to the desired equations. Let bλµ be the change-of-basis coefficients defined by sending ti ÞÑ 1 ´ ti. So, by definition, Note that we have, equivalently, sλp1 ´ tq "ÿµ sλptq "ÿµ bλµsµptq. bλµsµp1 ´ tq. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 12 Thus f "ÿλ aλsλptq "ÿλ,µ aλbλµsµp1 ´ tq. The polynomials sµp1 ´ tq for all µ ě 0 form an additive basis for the K GLpV q Thus, f P I 1 if and only if the coefficient of sµp1 ´ tq is 0 for all µ Ď . That is, ` pHompV, W qq. 0 "ÿλ aλbλµ for all µ Ď . The following description of bλµ is due to Stanley. Recall that, if µ Ď λ are partitions, the skew shape λ{µ is the Young diagram of λ with the squares of µ deleted. A standard Young tableau is a filling of a (possibly skew) shape by the numbers 1, 2, . . . , t (with t boxes in all), such that the rows increase from left to right, and the columns increase from top to bottom. We write f σ for the number of standard Young tableaux of shape σ. Proposition 3.2. [Sta99] If µ Ę λ then bλµ " 0. If µ Ď λ, then bλ{µ " p´1qµ f λ{µf µ f λ λ µ dλpkq dµpkq . An equivalent formulation is bλµ " p´1qµ f λ{µ λ{µ! źpi,jqPλ{µ pk ` j ´ iq. Corollary 3.3 (Equivariant Herzog-Kuhl equations). Let M be an equivariant R-module with equivariant Betti table βi,λ. Assume M is generated in positive degree. The set-theoretic support of M is contained in the rank-deficient locus if and only if: Note that βi,λ is the multiplicity of the λ-isotypic component of the resolution of M (in cohomological degree i), whereas βi,λdλpkq " Ąβi,λ is the rank of this isotypic component. Proof. (ñ): The only thing to note is that, for simplicity, we have rescaled the µ-indexed equation by p´1qµdµpkq. (ð): If the equations are satisfied, then M maps to the trivial K-theory class on Grpk, Cnq. Since the Grassmannian is projective, it follows that the sheaf associated to M is zero. This implies the support restriction. (cid:3) The coefficient in Equation (3.1) has the following interpretation. Consider a uniformly- random filling T of the shape λ by the numbers 1, . . . , λ. Say that T splits along µ \ λ{µ if the numbers 1, . . . , µ lie in the subshape µ. Then: (3.2) f λ{µf µ f λ λ µ " ProbpT splits along µ \ λ{µ T is standardq ProbpT splits along µ \ λ{µq . (3.1) For each µ Ď : ÿi,λĚµ p´1qi βi,λdλpkq loooomoooon p"Ąβi,λq f λ{µf µ f λ λ µ " 0. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 13 3.3. Proof of Theorem 3.1. First, we recall the following fact about K-theory of Grass- mannians: Proposition 3.4 ([FS12], page 21). The following identity holds of formal power series over KpGrpk, W qq: ÿp "ľ pS‰up ÿq "ľ qQ‰uq " p1 ` uqn. It is essentially a consequence of the tautological exact sequence of vector bundles 0 Ñ S Ñ W Ñ Q Ñ 0. We rearrange this as ÿq "ľ qQ‰uq " p1 ` uqn 1 řp"Ź pS‰up " p1 ` uqn ÿp ` ´ 1qprSymppSq‰up. The key observation is that the left-hand side is a polynomial in u of degree n ´ k, since Q has rank n ´ k. Thus, the coefficient fℓ of uℓ of the right-hand side vanishes for ℓ ě n ´ k. In other words, viewing fℓ as a symmetric polynomial, we have fℓ P I 1 for ℓ ą n ´ k. We compute the coefficient fℓ. Recall that rSymppSqs " hp, the p-th homogeneous symmetric polynomial. We have ÿℓ fℓuℓ " p1 ` uqnÿp p´1qphpup " " nÿq"0 8ÿℓ"0 up`qp´1qphpn q 8ÿp"0 p´1qphp n ℓ ´ p, ℓÿp"ℓ´n uℓ so our desired coefficients are fℓ " p´1qphp n ℓ ´ p, ℓÿp"ℓ´n where in the last two lines we use the convention hp " 0 for p ă 0. We next show: Lemma 3.5. We have I 1 Ě phn´k`1p1 ´ tq, . . . , hnp1 ´ tqq. Proof. Equivalently, we change basis t ÞÑ 1 ´ t, calling the (new) ideal J, and we show J Ě phn´k`1, . . . , hnq. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 14 We consider the elements fn´k`ip1 ´ tq P J for i " 1, . . . , k. fn´k`ip1 ´ tq " n´k`iÿp"´k`i p´1qphpp1 ´ tq n ´ k ` i ´ p. n Since i ď k we have " p´1qphpp1 ´ tq n´k`iÿp"0 n ´ k ` i ´ p. n We apply the second formula from Proposition 3.2. Note that all terms are single-row partitions, λ " ppq and µ " psq, with s ď p, so f λ{µ " 1 and the change of basis is: bλµ " p´1qs f λ{µ λ{µ! pk ` sq pk ` p ´ 1q " p´1qsk ` p ´ 1 k ` s ´ 1. Hence, fn´k`ip1 ´ tq " " n´k`iÿp"0 n´k`iÿs"0 p´1qp`s pÿs"0 n´k`iÿp"s p´1qshs n n ´ k ` i ´ pk ` p ´ 1 p´1qp k ` s ´ 1hs. n ´ k ` i ´ pk ` p ´ 1 k ` s ´ 1. n We reindex, sending p ÞÑ n ´ k ` i ´ p, and reverse the order of the inner sum: " p´1qn´k`i n´k`iÿs"0 p´1qshs n´k`i´sÿp"0 p´1qpn pn ` i ´ p ´ 1 k ` s ´ 1 . The terms hs for s ď n ´ k. First, we show that all the lower terms hs, with s ď n ´ k, vanish. For these terms, we view the large binomial coefficient as a polynomial function of p. It has degree k ` s ´ 1, with zeroes at p " pn ´ k ` i ´ sq ` 1, . . . , n ` i ´ 1, so we may freely include these terms in the inner sum. It is convenient to extend the inner sum only as far as p " n, obtaining p´1qpn nÿp"0 pn ´ i ´ p ´ 1 k ` s ´ 1 . Recall from the theory of finite differences that p´1qpd dÿp"0 pgppq " 0 FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 15 whenever g is a polynomial of degree ă d. Since the above sum has degree k ` s ´ 1 ď n ´ 1, it vanishes. Thus, dropping the lower terms, we are left with (3.3) fn´k`ip1 ´ tq " p´1qi iÿs"1 p´1qshn´k`s p´1qpn i´sÿp"0 pn ` i ´ p ´ 1 n ` s ´ 1 . Showing hn´k`i P J for i " 1, . . . , k. From equation (3.3), we see directly that the coefficient of hn´k`i in fn´k`ip1 ´ tq is 1. This is the leading coefficient, so the claim follows by induction on i. (cid:3) Corollary 3.6. We have Proof. The equality of ideals J " phi : i ą n ´ kq " spanC sλ : λ Ę (. follows from Newton's identities and induction. The equality phn´k`1, . . . , hnq " phi : i ą n ´ kq phi : i ą n ´ kq " spanC sλ : λ Ę ( follows from the Pieri rule (for Ď) and the Jacobi-Trudi formula (for Ě). See [Ful96] for these identities. This shows J contains this linear span. But then quotienting by J leaves k classes. This is already the rank of KpGrpk, W qq, so we must have equality. (cid:3) at most`n Changing bases t ÞÑ 1 ´ t a final time completes the proof of Theorem 3.1. 4. Square matrices and perfect matchings Remark 4.1. In this section, rank Betti tables play a more significant role than multiplicity We now describe the Boij-Soderberg cone in the base case of square matrices; thus we set n " k for the remainder of this section. The corresponding Grassmannian is a point, tables. As such, we will state results in terms of the cones ĄBSk,k and ĄBSD so there is no dual geometric picture or cone. We will recall the description of ĄBSk,k due to [FLS16]; we then describe the derived cone ĄBSD When k " 1, the ring is just Crts, and its torsion graded modules are essentially trivial to describe. See Section 4 of [EE12] for a short, complete description of both cones. For k ą 1, however, the cones are algebraically and combinatorially interesting, although simpler than the general case. k,k. k,k. 4.1. Prior work onĄBSk,k ([FLS16]). The rank-deficient locus tdetpT q " 0u Ă HompV, W q is codimension 1. Thus, modules satisfying Condition 1.2 have free resolutions of length 1, M Ð F 0 Ð F 1 Ð 0. There is only one equivariant Herzog-Kuhl equation, labeled by the empty partition µ " ∅: (4.1) that is, rankpF 0q " rankpF 1q. ÿλ Ąβ0,λ "ÿλ Ąβ1,λ, FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 16 Algebraically, this simply says that M is a torsion module. The extremal rays and supporting hyperplanes of ĄBSk,k are as follows. Definition 4.2 (Pure tables). Fix λ, µ P Y with λ Ĺ µ. The pure table rβpλ Ĺ µq is defined by setting and all other entries 0. Ąβ0,λ " Ąβ1,µ " 1 It is nontrivial to show that each pure table rβpλ Ĺ µq is realizable up to scalar multiple ([FLS16, Theorem 4.1]). Any such table generates an extremal ray of ĄBSk,k. It is, by contrast, easy to establish the following inequalities on ĄBSk,k. Definition 4.3 (Antichain inequalities). Let S Ď Y be a downwards-closed set. Let Γ " tλ : λ Ĺ µ for some µ P Su. For any rank Betti table pĄβi,λq, the antichain inequality (for S) is then: (4.2) ÿλPΓĄβ0,λ ě ÿλPS Ąβ1,λ. (The terminology of 'antichains' is due to [FLS16], where the inequality (4.2) is stated in terms of the maximal elements of S, which form an antichain in Y.) These conditions follow directly from minimality of the underlying maps of modules: the summands corresponding to S in F1 must map into those corresponding to Γ in F0. construction was implicit in [FLS16, Lemma 3.6]. It is especially simple in this case: Finally, we recall the graph-theoretic model of rβ introduced in Section 1.3.2. This Definition 4.4. The Betti graph Gprβq is the directed bipartite graph with left vertices L and right vertices R, defined as follows: ‚ The edge set contains all possible edges λ Ð µ, from R to L, for λ Ĺ µ. ‚ The set L (resp. R) contains Ąβ0,λ (resp. Ąβ1,λ) vertices labeled λ, for each λ, The Boij-Soderberg cone ĄBSk,k is characterized as follows. Theorem 4.5 ([FLS16, Theorem 3.8]). The cone ĄBSk,k is defined by the rank equation (4.1), the conditions Ąβi,λ ě 0, and the antichain inequalities (4.2). Its extremal rays are the pure tables rβpλ Ĺ µq, for all choices of λ Ĺ µ in Y. Moreover, if rβ P rBk,k has nonnegative integer entries, then rβ P ĄBSk,k if and only if the Betti graph Gprβq has a perfect matching. A perfect matching on Gprβq expresses rβ as a positive integer sum of pure tables: an edge λ Ð µ corresponds to a summand rβ " ` rβpλ Ĺ µq ` . FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 17 It is easy to see that the cone spanned by the pure tables is contained in the cone defined by the antichain inequalities. The fact that these cones agree follows from Hall's Matching Theorem for bipartite graphs: Theorem 4.6 (Hall's Matching Theorem). Let G be a bipartite graph with left vertices L and right vertices R, with L " R. Then G has a perfect matching if and only if the following holds for all subsets S Ď R (equivalently, for all subsets S Ď L): let ΓpSq be the set of vertices adjacent to S. Then ΓpSq ě S. In the antichain inequality (4.2), S corresponds to a set of vertex labels on the right- closed sets S.) k,k. We are interested in bounded free equivariant complexes 4.2. The derived cone. We now generalize Theorem 4.5 to describe the derived cone The numbers of such vertices are then the right- and left-hand-sides of the inequality. (The hand-side of the Betti graph Gprβq. The set Γ consists of the labels of vertices adjacent to S. structure of Gprβq implies easily that it suffices to consider inequalities from downwards- ĄBSD The supporting hyperplanes of ĄBSD k,k are quite complicated and we do not establish them directly. We instead generalize the descriptions in terms of extremal rays and per- fect matchings, which remain fairly simple. We then deduce the inequalities from Hall's Theorem. all of whose homology modules are torsion. Ð Fi Ð Fi`1 Ð Fi`2 Ð , Definition 4.7 (Homologically-shifted pure tables). Fix i P Z and λ, µ P Y with λ Ĺ µ. The extremal rays of ĄBSD We define the homologically-shifted pure table rβ"λ Ąβi,λ " Čβi`1,µ " 1 k,k will be homological shifts of those of ĄBSk,k: iÐÝ µı by setting and all other entries 0. The supporting hyperplanes will be defined by the following inequalities. Recall that a convex subset S of a poset P is the intersection of an upwards-closed set with a downwards-closed set. Definition 4.8 (Convexity inequalities). For each odd i, let Si Ď Y be any convex set. For each even i, define Γi " tλ : µ Ĺ λ for some µ P Si´1u Y tλ : λ Ĺ µ for some µ P Si`1u. For any rank Betti table pĄβi,λq, the convexity inequality (for the Si's) is then: (4.3) ÿi even ÿλPΓiĄβi,λ ě ÿi odd ÿλPSiĄβi,λ. (We may, if we wish, switch 'even' and 'odd' in this definition. We will see that either collection of inequalities yields the same cone.) FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 18 We recall the general definition of the Betti graph: Definition 4.9 (Betti graphs for complexes). Let rβ P rBk,k have nonnegative integer entries. The Betti graph Gprβq is defined as follows: ‚ The vertex set contains Ąβi,λ vertices labeled pi, λq, for each pi, λq, ‚ The edge set contains, for each i, all possible edges pi, λq Ð pi ` 1, µq with λ Ĺ µ. Note that this graph is bipartite: every edge connects an even-indexed and an odd-indexed vertex. Each segment Si of Definition 4.8 corresponds to a set of vertex labels in Gprβq. The set Γi then contains the labels of vertices adjacent to Si´1 and Si`1. The numbers of vertices counted this way give the right- and left-hand-sides of the inequality (4.3). Note that if the Si's were not convex, we could replace them by their convex hulls without changing the Γi's. integers. The following are equivalent: k,k; k,k. We now characterize the derived Boij-Soderberg cone ĄBSD Theorem 4.10 (The derived Boij-Soderberg cone, for square matrices). Let rβ be an ab- stract rank Betti table. Without loss of generality, assume the entries of rβ are nonnegative (i) rβ PĄBSD (ii) rβ satisfies all the convexity inequalities, together with the rank condition ÿi,λ p´1qiĄβi,λ " 0; (iii) rβ is a positive integral linear combination of homologically shifted pure tables; (iv) The Betti graph Gprβq has a perfect matching. a pure table summand for rβ. We have (iii) ñ (ii) since the conditions (ii) hold for each homologically-shifted pure table individually. Hall's Matching Theorem gives the statement (ii) ô (iv) and shows that we may exchange 'even' and 'odd' in the definition of the convexity inequalities. Homologically-shifted pure tables are realizable, hence (iii) ñ (i). We will complete the proof by exhibiting a perfect matching on any realizable Betti graph, so that (i) ñ (iv). We postpone the proof until Section 5.1.1 (Corollary 5.9). Remark 4.11. It is clear that (iv) ñ (iii): each edge of a perfect matching indicates 5. The pairing between Betti tables and cohomology tables In this section, we establish the numerical pairing between Betti tables and cohomology tables. We recall that the pairing is defined as follows (Definition 1.11): (5.1) rΦ : BTk,n CTk,n Ñ ĂBTk,k, ÞÑ rΦpβ, γq, pβ, γq FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 19 (5.2) with rΦ the (derived) rank Betti table with entries Ąϕi,λpβ, γq " ÿp´q"i βp,λ γq,λ. Recall also that the convention is that homological degree (p and i) decreases under the boundary map of the complex. The remainder of this section is devoted to the proof of the following: a pairing of cones, Theorem 5.1 (Pairing the equivariant Boij-Soderberg cones). The pairing rΦ restricts to In light of our description (Theorem 4.10) of the derived cone ĄBSD to exhibit a perfect matching on the Betti graph of rΦpβ, γq. Along the way, we will also complete the proof of Theorem 4.10 itself, showing that ĄBSD k,n ESk,n ÑĄBSD istence of such matchings (Corollary 5.9). k,k is characterized by the ex- k,k, the goal will be BSD k,k. We sketch the construction of the pairing. Sketch of Theorem 5.1. Let β " βpF ‚q be the Betti table of a minimal free equivariant complex F ‚ of finitely-generated R-modules, with R " Rk,n the coordinate ring of the k n matrices. Assume F ‚ is exact away from the locus of rank-deficient matrices, so descending F ‚ to Grpk, Cnq gives an exact sequence of vector bundles F ‚: (5.3) F ‚ "àλ SλpV qβ‚,λ b R descends to ÝÝÝÝÝÝÝÑ F ‚ "àλ SλpSqβ‚,λ, with S the tautological subbundle on Grpk, Cnq. Let γ " γpEq be the GL-cohomology table of a coherent sheaf E on Grpk, nq. Observe that E b F ‚ is again exact. We study the hypercohomology spectral sequence. Explicitly, we take the Cech reso- lution of E b F ‚, an exact double complex of sheaves. Let E‚,‚ be the result of taking global sections: a double complex of vector spaces. Note that each column is a direct sum of complexes, according to the λ summands in (5.3). Running the sequence with the horizontal maps first, it converges to 0 on the E2 page (since E b F ‚ is an exact sequence of sheaves). In particular, TotpE‚,‚q is exact. Running the sequence with the vertical maps first, we instead obtain, on the E1 page, Ep,q 1 "àλ H qpE b SλpSqqβp,λ. Observe that the λ summand has dimension βp,λpF ‚qγq,λpEq. The pi, λq coefficient produced in the Boij-Soderberg pairing, Ąϕi,λpF ‚, Eq, is the sum of this quantity along the diagonal tp ´ q " iu. That is, rΦ is akin to a Betti table for TotpE1q: Ąϕi,λ " dimC TotpE1qi,λ. We emphasize, however, that there is no actual GLk-action on TotpE1q, nor an Rk,k-module structure. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 20 Instead, we will show by homological techniques that, for a wide class of double com- plexes including E‚,‚, there is a perfect matching on a graph associated to TotpE1q; in our (cid:3) setting, this will give the desired perfect matching on the Betti graph of rΦ. The key properties of the double complex E‚,‚ constructed above are that (1) Each term Ep,q has a direct sum decomposition labeled by a poset P ; (2) The vertical maps dv are label-preserving; (3) The horizontal maps are strictly-label-decreasing (in our setting, this follows by minimality of F ‚ and functoriality of the Cech complex); By (1) and (2), the E1 page (the homology of dv) again has a direct sum decomposition labeled by P , Ep,q . We define the following graph: 1 "ÀλPP Ep,q,λ 1 Definition 5.2. The E1 graph G " GpE‚,‚q is the following directed graph: ‚ The vertex set contains dimpEp,q,λ 1 q vertices labeled pp, q, λq, for each p, q P Z and each λ P P ; ‚ The edge set includes all possible edges pp, q, λq Ñ pp1, q1, λ1q whenever λ1 ă λ and pp1, q1q " pp ´ r, q ´ r ` 1q for some r ą 0. The edges of G are shaped like the higher-order differentials of the associated spectral sequence, and they respect the strictly-decreasing-P -labels condition. We show: Theorem 5.3. Let E‚,‚ be a double complex of vector spaces satisfying (1)-(3). If TotpE‚,‚q is exact, its E1 graph has a perfect matching. We think of this theorem as a combinatorial analog of the fact that the associated spectral sequence (beginning with the homology of dv) converges to zero. We explore this idea further in Section 5.1. In our setting, we identify the vertices of the E1 graph and the Betti graph of rΦpβ, γq; for any such identification, the edges of the E1 graph become a subset of the Betti graph's edges. (We may recover the missing edges by allowing r ď 0 in Definition 5.2.) Hence, the perfect matching produced by Theorem 5.3 is valid for the Betti graph, completing the proof of Theorem 5.1. 5.1. Perfect matchings in linear and homological algebra. Our approach uses linear maps to produce perfect matchings. The starting point is the following construction: Definition 5.4. Let T : V Ñ W be a map of vector spaces, having specified bases V, W. The coefficient graph G of T is the directed bipartite graph with vertex set V \ W and edges E " tv Ñ w : T pvq has a nonzero w-coefficientu. Note that the adjacency matrix of G is T with all nonzero coefficients replaced by 1's. Proposition 5.5. For finite-dimensional vector spaces, the coefficient graph of an isomor- phism admits a perfect matching. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 21 We will say the corresponding bijection V Ø W is compatible with T , a combinatorial analog of the fact that T is an isomorphism. The proof of existence is simple, but essentially nonconstructive in practice. Here are two ways to do it: (i) All at once: since detpT q ‰ 0, some monomial term of detpT q is nonzero. This exhibits the perfect matching. (ii) By induction, using the Laplace expansion: expand detpT q along a row or column; some term aij (complementary minor) is nonzero, and so on. Similarly, if T is merely assumed to be injective or surjective, we may produce a maximal matching in this way (choose some nonvanishing maximal minor). We generalize Proposition 5.5 to the setting of homological algebra in three ways: to infinite-dimensional vector spaces, to long exact sequences, and to double complexes (mo- tivated by spectral sequences). Proposition 5.6. For vector spaces of arbitrary dimension, the coefficient graph of an isomorphism admits a perfect matching. We will not need Proposition 5.6 for our proof of Theorem 5.3, so we prove it in the appendix. 5.1.1. Long exact sequences and the proof of Theorem 4.10. We generalize to the case of long exact sequences. Let Ð Vi δÐÝ Vi`1 Ð be a long exact sequence, with Vi a fixed basis for Vi. (The vector spaces may be finite- or infinite-dimensional.) Definition 5.7. The coefficient graph G for pV‚, δq (with respect to V‚) is the directed graph with vertex setŮi Vi and an edge v Ñ v1 whenever δpvq has a nonzero v1-coefficient. Proposition 5.8. The coefficient graph of a long exact sequence has a perfect matching. Proof. Choose subsets Fi Ă Vi descending to bases of impδq Ă Vi´1, using Zorn's Lemma in the infinite case. Let Gi " Vi ´ Fi, and let Fi " spanpFiq and Gi " spanpGiq. The δÝÑ Vi ։ Gi is an isomorphism and has the same coefficients composition δ : Fi`1 ãÑ Vi`1 as δ, restricted to Fi`1 and Gi. Thus Proposition 5.5 (or 5.6 in the infinite case) yields a matching of Fi`1 with Gi. (cid:3) At this point, we complete the proof of Theorem 4.10, characterizing the derived Boij- k,k of the square matrices. Soderberg cone ĄBSD Corollary 5.9. If rβ PĄBSD Proof. Let rβ be the Betti table of a minimal free equivariant complex pF ‚, δq of R-modules, k,k, then the Betti graph Gprβq has a perfect matching. with R " Rk,k the coordinate ring of the k k matrices, and F ‚ b Rr 1 Choose, for each F i, a C-basis of each copy of SλpV q occuring in F i. Label each basis element x by the corresponding partition λ. It follows from minimality that δpxλq is an R-linear combination of basis elements labeled by partitions λ1 Ĺ λ. det s exact. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 22 Since the homology modules are torsion, F‚ b FracpRq is an exact sequence of FracpRq- vector spaces, with bases given by the xλ's chosen above. By the previous proposition, its coefficient graph has a perfect matching. This graph has the same vertices as the Betti (cid:3) graph Gprβq, and its edges are a subset of Gprβq's edges. Remark 5.10. Rather than tensoring with FracpRq, we may instead specialize to any convenient invertible k k matrix T P HompCk, Ckq, such as the identity matrix. This approach is useful for computations, since the resulting exact sequence consists of finite- dimensional C-vector spaces. 5.1.2. Double complexes and the proof of Theorem 5.1. Finally, we generalize to the setting of double complexes and spectral sequences. Let pE‚,‚, dv, dhq be a double complex of vector spaces, with differentials pointing up and to the left: q axis /❴❴❴ p axis Ep´1,q`1 dh Ep,q`1 dv Ep´1,q dv Ep,q dh We assume the squares anticommute, so the total differential is dtot " dh ` dv, and dhdv ` dvdh " 0. We will always assume the total complex TotpE‚,‚q has a finite number of columns. Note that we do not assume a basis has been specified for each E‚,‚. We recall the complexes E‚,‚ of interest: (1) Each term Ep,q has a direct sum decomposition Ep,q " àλPP Ep,q,λ, with labels λ from a poset P . (2) The vertical differential dv is graded with respect to this labeling, and (3) The horizontal differential dh is downwards-filtered. The conditions (2) and (3) mean that dvpEp,q,λq Ď Ep,q`1,λ, and dhpEp,q,λq Ď àλ1ăλ Ep´1,q,λ1 , so the vertical differential preserves the label and the horizontal differential strictly de- creases it. homology Ep,q,λ 1 We are interested in the homology of the vertical map dv. Since dv is P -graded, so is its " Hpdvqp,q,λ. We recall that the E1 graph GpE‚,‚q is defined as follows: q vertices labeled pp, q, λq, for each p, q and each ‚ The vertex set contains dimpEp,q,λ 1 λ P P ; ‚ The edge set includes all possible edges pp, q, λq Ñ pp1, q1, λ1q whenever λ1 ă λ and pp1, q1q " pp ´ r, q ´ r ` 1q for some r ą 0. o o O O ✤ ✤ ✤ ✤ / O O o o O O FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 23 The edges of G are shaped like higher-order differentials of the associated spectral sequence, i.e. they point downwards-and-leftwards, and they respect the strictly-decreasing-P -labels condition. We wish to show: Theorem 5.11. If TotpE‚,‚q is exact, the E1 graph of E‚,‚ has a perfect matching. Remark 5.12. Consider summing the E1 page along diagonals. Call the resulting complex TotpE1q. If it were exact, the matching would exist by Proposition 5.8, and in fact would only use the edges corresponding to r " 1. Since TotpE1q is not exact in general, the proof works by modifying its maps to make it exact. Explicitly, we will exhibit a quasi-isomorphism from TotpE‚,‚q to a complex with the same terms as TotpE1q, but different maps -- whose nonzero coefficients are only in the spots permitted by the E1 graph. Since TotpE‚,‚q is exact, so is the new complex, so we will be done by Proposition 5.8. Proof. First, we split all the vertical maps: for each p, q, λ, we define subspaces B, H, B Ď E (suppressing the indices) as follows. We put B " impdvq; we choose H to be linearly disjoint from B and such that B ` H " kerpdvq; then we choose B linearly disjoint from B ` H, such that B ` H ` B " E. In particular, dv maps the subspace B isomorphically to the subsequent subspace B, and the space H descends isomorphically to Hpdvq, the E1 term. The picture of a single column of the double complex looks like the following: B H B (note that dvpBq " dvpHq " 0q ... „ b❊❊❊❊❊❊❊❊❊❊❊ c●●●●●●●●● d❍❍❍❍❍❍❍❍❍ „ „ B H B H B For the horizontal map, we have dhpBq Ă B and dhpHq Ă B ` H, and the poset labels λ strictly decrease. Our goal will be to choose bases carefully, so as to match the H basis elements to one another, in successive diagonals, while decreasing the poset labels. We first choose an arbitrary basis of each H and B space. We descend the basis of B to a basis of the subsequent B using dv. Note that every basis element has a position pp, qq and a label λ. We will write xλ if we wish to emphasize that a given basis vector x has label λ. We now change basis on the entire diagonal Ei :" Àp´q"i Ep,q. We leave the H and B bases untouched, but replace all the B basis vectors, as follows. Let bλ P Bp,q,λ and let b c d FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 24 λ " d´1 b v pbλq P pBqp,q´1,λ be its 'twin'. We define bλ :" dtotpb λq " bλ ` dhpb λq. We replace bλ by bλ, formally labeling the new basis vector by pp, q, λq. We write rBp,q,λ for the span of these b's, so in particular, rBp,q,λ :" dtot`pBqp,q´1,λ. It is clear that rB, H, B collectively gives a new basis for the entire diagonal, unitri- general, a linear combination of rB, H, B elements in all positions down-and-left of p, q, angular in the old basis. Notice also that the old basis element bλ P Bp,q,λ becomes, in with leading term bλ: bλ " bλ `ÿią0 xp´i,q´i, with xp´i,q´i P àλ1Ĺλ Ep´i,q´i,λ1 . The lower terms have strictly smaller labels λ1 Ĺ λ. (In fact, slightly more is true: if a label λ1 occurs in the i-th term, the poset P contains a chain of length ě i from λ1 to λ.) We now inspect the coefficients of pTotpE‚,‚q, dtotq in the new basis. We have dtotpb λq " bλ, dtotpbλq " 0 ` " d2 totpb λq, so the B elements map one-by-one onto the rB elements, with the same λ labels; the latter Next, for a basis element hλ P H, the coefficients change but remain 'filtered'. If dtotphλq elements then map to 0. included (in the old basis) some nonzero term t bµ, then in the new basis we have dtotphλq " dhphλq " ` t `bµ ´ dhpb µq ` . Since t is nonzero, we have µ Ĺ λ; and the additional terms coming from dhpb µq all have labels µ1 Ĺ µ. Thus all labels occurring in dtotphλq in the new basis are, again, strictly smaller than λ. We note that dtotphλq is a linear combination of rB, H, B elements down- and-left of hλ along the subsequent diagonal: rB, H Ho t❥❥❥❥❥❥❥❥❥ w♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣♣ rB, H, B . . . rB, H, B Finally, we observe that the spaces rB ` B collectively span a subcomplex of TotpEq, so we have a short exact sequence of complexes 0 Ñ TotprB ` Bq Ñ TotpEq Ñ TotpHq Ñ 0. o t w FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 25 By construction, TotprB ` Bq is exact, so TotpEq Ñ TotpHq is a quasi-isomorphism. Note that TotpHq and TotpE1q have "the same" terms, but different maps, as desired. Since TotpEq is exact, so is TotpHq. The desired matching therefore exists by Proposition 5.8. (cid:3) Remark 5.13. Our initial attempts to establish the Boij-Soderberg pairing (Theorems 5.1 and 5.3) used the higher differentials on the E1, E2, . . . pages, rather than the E0 page as above -- aiming to systematize "chasing cohomology of the underlying sheaves". The following example shows that such an approach fails on general double complexes. Example 5.14 (A cautionary example). Consider the following double complex. Each partition denotes a single basis vector with that label. f ' ' The vertical map dv preserves labels and the horizontal map dh decreases labels. The unlabeled arrows correspond to coefficients of 1, and the map f is given by f p q " , ´ . f´ ¯ " Note that the rows are exact, so the total complex is exact as well, and the spectral sequence abuts to zero. The only nonzero higher differentials are on the E1 and E3 pages. These pages, and (for contrast) the complex H constructed in Theorem 5.3, are as follows. E1 : 0 E3 : 0 H : t t t 0 t t 0 t 0 t yt 0 0 u❧❧❧❧❧❧❧❧❧❧❧ w♣♣♣♣♣♣♣♣♣ 0 All the arrows are coefficients of 1. differentials gives a valid matching (the E3 arrow violates the P -filtered condition). In particular, no combination of the E1 and E3 In contrast, H finds the (unique) valid matching Ð , Ð (. 6. Preliminary results on 2 3 matrices While a complete picture of the general case still seems quite far away, we have some partial results and suggestive examples in the particular case of 2 3 matrices, which suggest some features of the general picture. In this setting, the modules of interest (as in Condition 1.2) are Cohen-Macaulay of codimension 2. We will assume all modules are generated in positive degree. There are o o o o O O o o O O o o y u o o w FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 26 three equivariant Herzog-Kuhl equations, corresponding to µ " ∅, be simplified to , and , which may µ " ∅ : µ " : µ " : 0 "ÿi,λ 0 "ÿi,λ 0 "ÿi,λ p´1qiβi,λdλp2q, p´1qiβi,λdλp2q pλ1 ` λ2q, p´1qiβi,λdλp2q 1 2 pλ1 ` 1qλ2. All Betti tables in this section will be multiplicity Betti tables. We remark that dλp2q " 1 ` λ1 ´ λ2, dλp3q " 1 2 p1 ` λ1 ´ λ2qp1 ` λ2 ´ λ3qp2 ` λ1 ´ λ3q. 6.1. Simple and pure resolutions. The basic observation is the following. Since there are three Herzog-Kuhl equations, if we allow exactly four entries in our Betti table to be nonzero, in general we expect the equations to pick out one dimension's worth of valid tables. That is, the resulting table will be unique up to scalar multiple. Any realizable table of this form is automatically an extremal ray of BS2,3. In fact, this observation underpins the characterization of pure tables for graded modules, where every choice of increasing degree sequence results in a unique table (up to scaling). The hope might be that, by analogy with both the graded case and the square-matrix case, these tables are all realizable, and form a complete set of extremal rays of BS2,3. To that end, we will call these pure tables. For some choices of entries, the equations will be redundant, and allowing nonzero entries in only three positions will suffice. We will call the result a simple table. In this case, each column of the table has exactly one entry, like the extremal tables in the graded and square-matrix cases. (Three entries are required or the resolution will be too short.) Remark 6.1. In this section, most of the modules of interest are bi-equivariant for the actions of both GLpV q and GLpW q on R " SympHompV, W qq. So, for brevity, we write λ b µ :" SλpV q b SµpW q b R. Note that this module has rank dλp2q dµp3q, and therefore has multiplicity dµp3q if we remember only the action of GLpV q. Example 6.2. Here are all the pure tables using partitions of size between 0 and 3. 0 1 2 0 1 2 0 1 2 ∅ 3 ∅ 8 ∅ 1 3 1 6 1 3 1 FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 27 0 1 2 0 1 2 0 1 2 ∅ 2 ∅ 2 1 6 8 3 ∅ 3 1 9 3 All of these tables except the last are simple, and each one is realizable. (The third is the Betti table of M " R{P2, the quotient by the ideal of maximal minors, which is resolved by the Eagon-Northcott complex.) In the graded and square-matrix cases, it is always possible [EFW11, FLS16] to realize any pure table by a resolution that is equivariant with respect to the actions of both GLpV q and GLpW q. In fact, the same is true in these examples. The five simple tables above are realized, respectively, by resolutions of the following forms (the fourth actually gives 3 times the fourth table): ∅ b ∅ b Ð Ð ∅ b ∅ Ð ∅ b b Ð Ð b b b b b The last table is realised as follows: b Ð b À b ' b b b b b Ð b Ð Ð Ð Ð Ð One thing that stands out in the above example is that the biequivariant resolutions for each of the simple tables has, in each step, a single irreducible pGL2 GL3q-representation. This does not appear to be an accident: we can construct a large class of such tables using a technique similar to [FLS16, Theorem 4.6]. The constructed tables are as follows. At each step of the resolution, the partition shape changes by the addition of a "border strip," that is, a shape like or (i.e., a connected shape not containing q. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 28 Moreover, the second border strip is always adjacent and to the right of the first. Examples of such triples pλ, µ, νq are: 1 1 1 2 2 1 1 1 1 1 2 2 1 1 2 2 1 1 2 2 2 2 2 Here λ consists of the empty squares, µ contains the additional squares marked 1, and ν contains the squares marked 1, 2. Proposition 6.3. Let β be a simple table with entries λ Ĺ µ Ĺ ν. If µ{λ and ν{µ are adjacent, successive border strips, then β is realizable by a biequivariant resolution with an irreducible GL2 GL3 representation at each step. We omit the proof. We can also rule out almost all other possible triples pλ, µ, νq: Proposition 6.4. Let β be a simple table with entries λ Ĺ µ Ĺ ν. Then ν2 ď λ1 ` 1. That is, the shapes µ{λ and ν{µ are contained in the border strip formed by the squares along the outer edge of λ: ? ? ? ? ? ? ? They need not be connected or adjacent, though computations have suggested that almost all are. We omit the proof, which follows from considering positive integer solutions to the equivariant Herzog-Kuhl equations. For tables that are pure but not simple, the story is currently much less complete, although we can realize many of them. Also, whether or not these tables are all realizable, we don't know if the cone has other extremal rays not of this form. Still, we have some data that at least suggests some helpful patterns. First, note that in every non-simple pure table, one column has two nonzero entries. We will say the table is diamond-shaped when the middle column has two nonzero entries and Y-shaped otherwise. In several cases, we can realize Y-shaped tables using extensions. Example 6.5 (Y-shaped tables via extensions). Consider the following Y-shaped table: 0 1 2 1 3 1 1 Using the same notation as in the previous examples, we were able to realize this table by the following resolution: b ∅ Ð b Ð ´ ' ¯ b It's possible to build this resolution out of two simple resolutions in the following way. Consider the following two simple resolutions, where we now explicitly name the modules FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 29 being resolved: M Ð b ∅ N hÐÝ b f ÐÝ g ÐÝ b b eÐÝ Ð b b . Note that the second term in the resolution of M is the same as the first term in the resolution of N , and that all the other terms appear in exactly the desired positions for the resolution we wish to construct. We define E :" cokerpf gq. By minimality, E has the desired generators; it suffices to check that E has the correct Tori for i ě 1. The key fact is that there is a short exact sequence 0 Ñ N Ñ E Ñ M Ñ 0, so we can extract information about the Betti table of E from the long exact sequence in Tor. This follows essentially from computing that h e " 0, that is, the unique generator of b (which has rank 1) maps into the image of g. Thinking of Ext1pM, N q the module as obtained by applying Homp´, N q to the resolution of M , the map h therefore induces an extension class, which is E above. Abusing notation, in the equation below, λ b µ denotes the (finite-dimensional) repre- sentation SλpV q b SµpW q, not the corresponding free R-module. The long exact sequence in Tor is: 0 Ð b ∅ Ð E b R{m Ð b sÐÝ b Ð Tor1pE, R{mq Ð Ð b 0 ÐÝ b Ð Tor2pE, R{mq Ð b Ð 0. Since we have already determined Tor0, the map s above is surjective, hence an isomorphism (since its source and target are the same irreducible representation). On the second line, the indicated map is zero because its source and target are distinct irreducible representations. In this particular case, one can check by hand that h e " 0 (in fact, we just computed the minimal free resolution of cokerpf gq using Macaulay2 and checked that it had the desired form). It is not immediately clear why something similar should happen in general. Still, this cancellation seems important, and indeed shows up for many other Y-shaped pure tables. We end by discussing two examples that demonstrate phenomena that do not appear in the square-matrix or graded cases. Example 6.6 (A slightly different extension). Consider the following table: 1 2 0 9 24 8 3 FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 30 A slight modification to the procedure of Example 6.5 realizes this table. We start with the following two resolutions: M Ð N Ð b b Ð Ð b b Ð Ð b b . Note that N is the same module as in Example 6.5. This time, the V sides line up (they for V ), but the W sides do not. So, we tensor the resolution of M both have a copy of with ∅ b and the resolution of N with ∅ b . Then they become composable, with the potential for cancellation. Continuing as in Example 6.5 (and assuming the requisite terms cancel, which they do) we get a resolution of the form E Ð b b Ð b b Ð b À b , which gives exactly the table above. In this case, not all of the GLpW q-representations are irreducible (though the ones on the final term are). Example 6.7 ("Stably-realizable" Betti tables). Consider the following table: 0 1 2 ∅ 1 3 8 3 Despite satisfying the Herzog-Kuhl equations, and resembling our previous unobjectionable examples, this table is not realizable. This follows from the numerical pairing: when paired with the trivial vector bundle, the result is the following table, which clearly has no perfect matching of the appropriate type: ∅ -1 0 1 9 8 Consider, however, tensoring the table with S1V . (That is, write the Betti table that would result from tensoring such a resolution with S1V .) The result is as follows: FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 31 0 1 2 1 3 3 8 8 3 3 This table in fact is realizable. It is a linear combination of the following four realizable tables: A 0 1 2 1 2 2 B 0 1 2 C 3 1 6 1 0 1 2 1 1 3 1 D 0 1 2 1 5 1 Specifically, the large table above is 2 A ` 1 2 B ` 3 3 2 C ` D. This example suggests that it might be easier to study "stably-realizable" Betti tables, that is, Betti tables that become realizable after tensoring with some GLpV q-representation. This idea is still being explored. Appendix A. The proof of Proposition 5.6 Proposition A.1. Let V, W be vector spaces of arbitrary dimension, with specified bases V, W. Let T : V Ñ W be an isomorphism. Then the coefficient graph of T has a perfect matching. Proof. We first reduce to V, W countable. This step is due to David Lampert [Lam]. Let b P V be arbitrary. Then T pbq involves only finitely-many basis elements, say B1 Ă W. For each s P B1, T ´1psq only involves finitely-many basis elements; let A2 Ă V contain these new elements, together with b. Repeat this construction, building two sequences of coordinate subspaces pbq Ă A2 Ă A3 Ă Ă V, B1 Ă B2 Ă B3 Ă Ă W such that, for each i, Bi Ď T pAiq Ď Bi`1. Let A8, B8 be the union; it follows that T restricts to an isomorphism of countable-dimensional spaces T : A8 Ñ B8, and T as a whole splits as a direct sum of such isomorphisms. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 32 We now build the matching inductively. Fix a basis vector v P V and write T pvq " ř aiwi, and assume every wi in the sum has ai ‰ 0. Equivalently, v "ÿ aiT ´1pwiq, so some T ´1pwiq contributes a nonzero v-coefficient. Fix one such w; we match v Ø w. Note that this choice is compatible with both T and T ´1. Let C " spanpVztvuq, D " spanpWztwuq. It is now easy to show that C ãÑ V Ñ W ։ D is an isomorphism (with the same coefficients as T , but with v and w removed). We now build the matching: we alternate between V and W , always choosing the first unmatched basis vector on each side to ensure that every basis vector gets matched (note that the construction is symmetric with respect to T and T ´1). (cid:3) References [BS08] [BBEG12] Christine Berkesch, Jesse Burke, Daniel Erman, and Courtney Gibbons. The cone of Betti diagrams over a hypersurface ring of low embedding dimension. J. Pure Appl. Algebra, 216(10):2256 -- 2268, 2012. arXiv:1109.5198v2. Mats Boij and Jonas Soderberg. Graded Betti numbers of Cohen-Macaulay modules and the multiplicity conjecture. J. Lond. Math. Soc. (2), 78(1):85 -- 106, 2008. arXiv:math/0611081v2. Anders Skovsted Buch. A Littlewood-Richardson rule for the K-theory of Grassmannians. Acta Math., 189(1):37 -- 78, 2002. [Buc02] [Dev16] [EE12] [BZEKS13] Christine Berkesch Zamaere, Daniel Erman, Manoj Kummini, and Steven V. Sam. Tensor complexes: multilinear free resolutions constructed from higher tensors. J. Eur. Math. Soc. (JEMS), 15(6):2257 -- 2295, 2013. arXiv:1101.4604v5. The Sage Developers. SageMath, http://www.sagemath.org. D. Eisenbud and D. Erman. Categorified duality in Boij-Soderberg Theory and invariants of free complexes. ArXiv e-prints, May 2012. arXiv:1205.0449v2. David Eisenbud, Daniel Erman, and Frank-Olaf Schreyer. Filtering free resolutions. Compos. Math., 149(5):754 -- 772, 2013. arXiv:1001.0585v3. the Sage Mathematics Software System, [EES13] 2016. [EFW11] David Eisenbud, Gunnar Fløystad, and Jerzy Weyman. The existence of equivariant pure free [ES09] [Flø12] [FLS16] [FS12] [Ful96] [GS] [GS15] resolutions. Ann. Inst. Fourier (Grenoble), 61(3):905 -- 926, 2011. arXiv:0709.1529v5. David Eisenbud and Frank-Olaf Schreyer. Betti numbers of graded modules and cohomology of vector bundles. J. Amer. Math. Soc., 22(3):859 -- 888, 2009. arXiv:0712.1843v3. Gunnar Fløystad. Boij -- Soderberg theory: Introduction and survey. In Progress in Commutative Algebra 1, Combinatorics and homology, Proceedings in mathematics, pages 1 -- 54. du Gruyter, 2012. arXiv:1106.0381v2. N. Ford, J. Levinson, and S. V Sam. Towards Boij-Soderberg theory for Grassmannians: the case of square matrices. ArXiv e-prints, August 2016. arXiv:1608.04058. Alex Fink and David E. Speyer. K-classes for matroids and equivariant localization. Duke Math. J., 161(14):2699 -- 2723, 11 2012. arXiv:1004.2403v2. William Fulton. Young Tableaux. Cambridge University Press, 1996. Cambridge Books Online. Daniel R. Grayson and Michael E. Stillman. Macaulay2, a software system for research in algebraic geometry. Available at http://www.math.uiuc.edu/Macaulay2/. Iulia Gheorghita and Steven V Sam. The cone of Betti tables over three non-collinear points in the plane. J. Commut. Algebra, to appear, 2015. arXiv:1501.00207v1. FOUNDATIONS OF BOIJ-S ODERBERG THEORY FOR GRASSMANNIANS 33 [KK90] [KS15] [Lam] [Mer05] [NS13] [Sta99] [Tho87] isomorphism of infinite-dimensional vector David (http://mathoverflow.net/users/59248/david-lampert). Bertram Kostant and Shrawan Kumar. T -equivariant K-theory of generalized flag varieties. J. Differential Geom., 32(2):549 -- 603, 1990. Manoj Kummini and Steven V Sam. The cone of Betti tables over a rational normal curve. In Commutative Algebra and Noncommutative Algebraic Geometry, volume 68 of Math. Sci. Res. Inst. Publ., pages 251 -- 264. Cambridge Univ. Press, Cambridge, 2015. arXiv:1301.7005v2. Lampert, tion modeling URL:http://mathoverflow.net/q/243071 (version: 2016-06-27). Alexander S. Merkurjev. Equivariant K-theory. In Handbook of K-theory. Vol. 1, 2, pages 925 -- 954. Springer, Berlin, 2005. Uwe Nagel and Stephen Sturgeon. Combinatorial interpretations of some Boij-Soderberg de- compositions. J. Algebra, 381:54 -- 72, 2013. arXiv:1203.6515. Richard P. Stanley. Enumerative combinatorics. Volume 2. Cambridge studies in advanced mathematics. Cambridge university press, Cambridge, New York, 1999. Errata et addenda : p. 583-585. R. W. Thomason. Algebraic K-theory of group scheme actions. In Algebraic topology and al- gebraic K-theory (Princeton, N.J., 1983), volume 113 of Ann. of Math. Stud., pages 539 -- 563. Princeton Univ. Press, Princeton, NJ, 1987. Bijec- spaces. MathOverflow.
0908.4037	2	0908	2012-02-06T05:31:49	Unitary representations of the fundamental group of Orbifolds	[ "math.AG" ]	There is a well known bijective correspondence between isomorphism classes of polystable vector bundles $E$ with $c_i(E)=0$ for $i\geq 1$ on a smooth complex projective variety and equivalence classes of unitary representations of the fundamental group of the variety. We show that this bijective correspondence extends to smooth orbifolds.	math.AG	math	UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP OF ORBIFOLDS INDRANIL BISWAS AND AMIT HOGADI Abstract. There is a well known bijective correspondence between the iso- morphism classes of polystable vector bundles E with ci(E) = 0 for i ≥ 1 on a smooth complex projective variety and the equivalence classes of unitary representations of the fundamental group of the variety. We show that this bijective correspondence extends to smooth orbifolds. 1. Introduction Let Y /C be a connected smooth projective curve, and let E −→ Y be a polystable vector bundle of degree zero. A celebrated theorem of Narasimhan and Seshadri, [NS], says that E is necessarily given by a unitary representation of the fundamental group of Y . Let X/C be a smooth projective variety of dimension n. Fix an ample line bundle L on X in order to define the degree of a torsionfree coherent sheaf on X. The following generalization of the Narasimhan -- Seshadri theorem holds. Theorem 1.1 ([Do], [MR]). Let E be a vector bundle on X such that c1(E) = 0 and c2(E) · c1(L)n−2 = 0. Then E is polystable if and only if it is given by a unitary representation of the (topological) fundamental group of X. Some clarification on Theorem 1.1 is necessary. By [Do, p. 231, Proposition 1], a stable vector bundle on X admits a Hermitian -- Yang -- Mills connection. Since a polystable vector bundle E on X is a direct sum of stable vector bundles of same degree/rank quotient, the Hermitian -- Yang -- Mills connection on the stable direct summands together produce a Hermitian -- Yang -- Mills connection on E. If c1(E) = 0 and c2(E) · c1(L)n−2 = 0, then any Hermitian -- Yang -- Mills connection on E is flat [Ko, p. 115, Theorem 4.11]. All Chern classes in Theorem 1.1 are topological, taking values in rational cohomological classes. Our aim here is to generalize the above theorem to projective orbifolds. By an orbifold over a field k we always mean a smooth separated Deligne -- Mumford stack which is of finite type over k and whose isotropy groups at all generic points is trivial. Thus an orbifold has a dense open substack which is a scheme. We 2000 Mathematics Subject Classification. 14F05, 14J60. Key words and phrases. Orbifolds, polystable bundle, fundamental group, unitary representation. 1 2 I. BISWAS AND A. HOGADI call a Deligne -- Mumford stack projective if it has a coarse moduli space which is a projective variety. Theorem 1.2. Let X/C be an irreducible projective orbifold of dimension n, and let L be an ample line bundle on X. Let E −→ X be a vector bundle on X such that c1(E) = 0 and c2(E) · c1(L)n−2 = 0. Then E is polystable (with respect to L) if and only if it is obtained by a unitary representation of πtop 1 (X, x) for a (equivalently, any) closed point x ∈ X. Here πtop 1 (X, x) denotes the topological fundamental group of the underlying complex analytical stack. See (4.1) for definition of ample line bundles on a stack. All Chern classes in Theorem 1.2 are topological, taking values in rational cohomological classes. Theorem 1.2 follows quite easily from Theorem 1.1 in the special case when X is a global quotient of a smooth variety by a finite group. The main work needed to prove Theorem 1.2 involves deducing the general case from this special case. 2. Rigidification Let k be a field. Throughout this section we work with Deligne -- Mumford stacks over k, and assume that all these stacks are of finite type and separated over k and are quotient stacks, meaning, can be expressed as quotient of an affine scheme by the action of a linear algebraic group. As shown in [Kr, Proposition 5.2], such stacks, Zariski locally, can be expressed as quotient of an affine variety by a finite group. The inertia stack of a stack Y , viewed as a sheaf of groups on the big ´etale site of Y , will be denoted by IY . Recall that the notion of rigidification of a Deligne -- Mumford stack is a process to get rid of given stabilizers in a "minimal" manner; see [ACV, Section 5.1]. Throughout this paper, we fix the following definition of rigidification. It should be clarified that this definition differs from the one in [ACV, Section 5.1]. Definition 2.1. Let f : Y −→ X be a 1 -- morphism of Deligne -- Mumford stacks over k. This f is called a rigidification if for any atlas U −→ X, where U is a scheme, the projection Y ×X U −→ U is a coarse moduli space. Since the morphism to the coarse moduli space of a separated Deligne -- Mumford stack is always proper, it follows that a rigidification morphism is a proper mor- phism. As a direct consequence of the above definition, it follows that every mor- phism from a Deligne Mumford stack to its coarse moduli space is a rigidification. Proposition 2.5 provides additional examples of rigidification. Lemma 2.2. Let g : Y1 −→ Y2 be a 1 -- morphism of Deligne -- Mumford stacks over k. Then for i = 1, 2, there exists (i) affine schemes Ui and linear algebraic groups Gi/k acting on Ui, (ii) a group homomorphism φ : G1 −→ G2, and UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 3 (iii) a k -- morphism f : U1 −→ U2 which is equivariant with respect to φ, such that Yi = [Ui/Gi], and the following diagram is commutative U1 f U2 Y1 = [U1/G1] / [U2/G2] = Y2 Proof. Take any i ∈ [1 , 2]. Since Yi is a quotient stack, we can write Yi = [Vi/Hi], where Vi is an affine scheme and Hi is a linear algebraic group acting on Vi. Now we let Then clearly The projection eV1 = V1 ×Y2 V2 = V1 ×Y1 (V2 ×Y2 Y1) . [eV1/(H1 × H2)] = [V1/H1] . is affine, because it is a base extension of the affine morphism V2 −→ Y2. eV1 −→ V1 (cid:3) Since V1 is an affine variety, and the projection eV1 −→ V1 is affine, we conclude that eV1 is also an affine variety. Also, we have a natural map which is equivariant for the actions of H1 × H2 on eV1 and V2, with H1 × H2 acting [eV1/(H1 × H2)] = [V1/H1], the proof is completed by setting G1 = H1 × H2, G2 = H2, U1 = eV1 and U2 = V2. on V2 through the projection H1 × H2 −→ H2. Now, because of the identification h : eV1 −→ V2 The following lemma is a generalization of the universal property of coarse moduli spaces. Lemma 2.3. Let g : Y1 −→ Y2 be as in Lemma 2.2. Assume, that for every point p, the map induced by f from the isotropy group of Y1 at p to that of Y2 at f (p) is trivial. Then f factors through the coarse moduli space of Y1. Proof. Let Ui, Gi, φ, f be as in Lemma 2.2. Let K = Kernel(φ). The assumption that the map induced by f on all isotropy groups is trivial is equivalent to saying that the stabilizer of every point in U1 is contained in K. We claim that the geometric quotient V = U1//K exists as an algebraic space. In order to prove this, first consider the stack [U1/K]. It is a Deligne Mumford stack, because the isotropy groups of the action of G1 (and hence of K) on U1 are finite ´etale over k. Thus V , which is nothing but the coarse moduli space of [U1/K], exists as an algebraic space by a theorem of Keel and Mori [KM, Theorem 1.1]. Next, we will show that the action of G1/K on V is free. To prove this, take ex ∈ V and g ∈ G1/K such that g(ex) = ex. Choose a point x ∈ U1 lying over / / / 4 I. BISWAS AND A. HOGADI ex, and also fix a lift g ∈ G1 of g. Since V is a geometric quotient of U1 by K, points of V correspond to orbits in U1 for the action of K. Thus, there exists an element a ∈ K such that gx = ax . Alternatively, a−1g is in the stabilizer of the point x and hence is contained in K. Thus g ∈ K, which implies that g = eK. This proves the above assertion that the action of G1/K on V is free. Consequently, the quotient stack [V /(G1/K)] is actually an algebraic space which is also the geometric quotient of U1 by G1, and hence [V /(G1/K)] is the coarse moduli space of Y1. Since the map f : U1 −→ U2 intertwines the given action of K on U1 and the trivial action of K on U2, it factors through V . Thus we have an induced map V −→ U2 which is equivariant with respect to the action of G1/K on V and the action of G2 on U2, using the homomorphism φ in Lemma 2.2. This in turn induces a map [V /(G1/K)] −→ [U2/G2]. But as explained above, [V /(G1/K)] is the coarse moduli space of Y1. This finishes the proof. (cid:3) Lemma 2.4. Let f : Y −→ X be a 1 -- morphism of separated Deligne -- Mumford stacks over k. Let Σ denote the set of all points in Y where the induced map on isotropy groups is an isomorphism. Then Σ is an open subset of Y . Proof. Since the Deligne -- Mumford stacks are separated, the inertia stacks IY and IX are finite over Y and X respectively. Thus the kernel and cokernel of the induced map IY −→ f −1(IX) are finite over X. Hence the set of all points where both the kernel and cokernel are trivial is an open subset of Y . (cid:3) Proposition 2.5. Let U/k be a k-scheme, and let G be a finite group acting on U. Let K be a normal subgroup of G, and let V denote the geometric quotient of U by K. Then the following two hold: (i) The natural map f : [U/G] −→ [V /(G/K)] is a rigidification. (ii) If U is spectrum of a local ring, then any rigidification of [U/G] is of this type for some normal subgroup K of G. Proof. (i) Since V −→ [V /(G/K)] is ´etale, and [U/K] −→ V is a coarse moduli space, in order to prove the statement it is enough to show that the following diagram is Cartesian: [U/K] g [U/G] f ′ f V g′ / [V /(G/K)] We first recall the description of the 1-morphism f . The k-groupoid underlying the stack [U/G] is the category of all triples (cid:16)Z, P h−→ Z, P φ −→ U(cid:17) , / / / UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 5 where Z is a k scheme, h is a principal G-bundle and φ : P −→ U is a G- equivariant morphism. Given such a triple, one has the corresponding induced map φ′ : P ′ = P/K −→ V , where P/K −→ Z is now a principal G/K-bundle. One thus gets a triple (Z, P ′ h′ → Z, φ′) which defines an object in the groupoid underlying [V /(G/K)]. This assignment (Z, P, φ) −→ (Z, P ′, φ′) defines the 1-morphism f . To prove that the above diagram is Cartesian, we need to exhibit a 1-isomorphism By construction of fiber product of groupoids, the groupoid underlying [U/K] −→ [U/G] ×[V /(G/K)] V . is the category of all [U/G] ×[V /(G/K)] V (Z, P h−→ Z, φ, Z γ −→ V, θ) , where (Z, P h−→ Z, φ) is an object, say α, of groupoid underlying [U/G], and γ defines an object, say β, of the groupoid underlying V , and θ is an isomorphism f (α) −→ g′(β). Note that by above discussion, f (α) is nothing but the object defined by the triple (Z, P ′, φ′) , where P ′ is the principal G/K-bundle P/K −→ Z, and φ′ : P ′ −→ V is the induced morphism. Moreover, g′(β) is nothing but the triple where Z × G/K −→ Z is the trivial principal G/K-bundle, and (Z, Z × (G/K),eγ) , is the G/K equivariant map given by γ : Z × G/K −→ V (z, g) 7−→ g · γ(z) . Thus giving the isomorphism is equivalent to giving a trivialization (or a section) of P ′ −→ Z and imposing θ : (Z, P ′, φ′) −→ (Z, Z × (G/K),eγ) the condition that the induced map φ′ coincides with eγ. Note that γ : Z −→ V could have been any arbitrary k-morphism, to start with. Thus, the category underlying the groupoid [U/G] ×[V /(G/K)] V coincides with the category of all quadruples (Z, P h−→ Z, φ, s) , 6 I. BISWAS AND A. HOGADI where Z, P and φ are as above, and s is a trivialization of the principal G/K- bundle h′ : P/K −→ Z. But giving a trivialization of h′ is equivalent to giving a reduction of the structure group of h to K. Thus the above category of quadruples is equivalent to the category of all triples (Z, Q q −→ Z, ψ) , where q : Q −→ Z is a principal K-bundle and ψ : Q −→ U is a K-equivariant morphism. One now sees that this is also the groupoid underlying [U/K]. (ii) Let U be a spectrum of a local ring, and let G be a finite group acting on U. Define Y := [U/G]. Let f : Y −→ X be any rigidification map. Let K be the kernel of the homomorphism Ip −→ If (p), where Ip and If (p) are the isotropy groups for the unique closed point p of Y and f (p) respectively. Consider the G-invariant map g : U −→ X induced by f . Let V be the geometric quotient of U by K. We will first show that g factors through V . Indeed, the induced map from the stack [U/K] −→ X is trivial on all isotropy groups, and hence by Lemma 2.3, it factors through the coarse moduli space, which is V . Since g is G-invariant, the induced map V −→ X is G/K invariant, and hence it gives us a map [V /(G/K)] −→ X . Clearly, this map is also a rigidification, but it is also an isomorphism on the isotropy group at the closed point of the domain. Hence by Lemma 2.4, it is an isomorphism on the isotropy group at all points. Therefore, this map must be an isomorphism. (cid:3) Lemma 2.6. Let f : Y −→ X be a rigidification, and let g : Y −→ Z be any 1-morphism. Suppose g, factors through f , meaning there exists a 1-morphism h : X −→ Z and a 2-isomorphism α : h ◦ f ⇒ g. Then the pair (h, α) does not have any 2-automorphisms, i.e., for any θ : h ⇒ h, the commutativity of the diagram h ◦ f α / g θ◦f h ◦ f α / g implies that θ is identity. / / UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 7 Proof. For an algebraically closed field K over k, let XK denote the category of K-points of X. Thus f , g and h induce functors gK / / ZK =③ ③ ③ ③ hK YK fK XK ③ ③ ③ ③ and a natural equivalence αK : hK ◦fK ⇒ gK. In order to prove that θ is identity, it is enough to show that θK is identity for every algebraically closed field K over k. However, from the definition of rigidification it follows that the functors fK is full and essentially surjective. It is then a simple exercise in category theory to show that θK is identity. (cid:3) Let f : Y −→ X be a rigidification. Let IY (respectively, IX ) be the inertia sheaf of Y (respectively, X), and define K f := Ker (IY −→ f −1(IX)) . Then we claim that f has the following universal property (which is a general- ization of Lemma 2.3): Proposition 2.7. Given any 1-morphism g : Y −→ Z such that K f is contained in the kernel of IY −→ g−1(IZ), there exists a morphism h : X −→ Z, and a 2-isomorphism α : h ◦ f ⇒ g. Further (h, α) is uniquely determined up to a unique 2-isomorphism (see (2.6)). Proof. Uniqueness of the 2-isomorphism follows from Lemma 2.6. Because of the uniqueness claimed, it is enough to prove the proposition Zariski locally around a point p ∈ Y . Thus by Lemma 2.2 we may reduce to the case where Y = [U1/G1], Z = [U2/G2], and g is induced by a map h : U1 −→ U2 which is equivariant with respect to a given group homomorphism φ : G1 −→ G2. Moreover, we may assume that G1 is precisely the isotropy group at the point p, and K f ⊂ Ker (φ) is a normal subgroup of G1. By Proposition 2.5(ii), by further taking a smaller Zariski neighborhood of p, we may assume that the rigidification f : Y → X is of the form [V /(G1/K f )] where V is the geometric quotient of U1 by K f . Note that although Proposition 2.5(ii) is stated in the case where U1 is spectrum of a local ring, we use standard limiting argument to draw conclusion about a small enough Zariski neighborhood of U1 (and hence also of Y ). Since K f ⊂ Ker (φ), h induces a map from V → U2 which is equivariant w.r.t. G1/K f action on V and G2 action on U2. Thus it induces a unique map [V /(G1/K f )] −→ Z. This proves the claim. (cid:3) We have the following corollary of Proposition 2.7. Corollary 2.8. A rigidification f : Y −→ X is determined uniquely by Ker (IY −→ f −1(IX )) . = 8 I. BISWAS AND A. HOGADI 3. Rigidification of Complex Deligne -- Mumford stacks We continue using the notation and assumptions of the previous section, but now restrict ourselves to the case k = C. Although our main interest is algebraic stacks, we will have to consider complex analytic Deligne -- Mumford stacks over C which arise as infinite covering spaces of algebraic stacks. These complex analytic stacks will have the property that of being, locally in complex analytic topology, quotient of an analytic variety by a finite group. In this section we only consider complex analytic stacks which have this additional property. We have the analogous definition and results for complex analytic Deligne -- Mumford stacks. Definition 3.1. Let Y be a complex analytic Deligne -- Mumford stack. Then a rigidification of Y is a 1-morphism f : Y −→ X, which locally on X in analytic topology, is a coarse moduli space. Proof of the following universal property is similar to that of Proposition 2.7, thanks to the assumption that all complex analytic stacks considered here are locally quotient spaces for finite groups actions. Proposition 3.2. Let f : Y −→ X be a rigidification of complex analytic stack Y . Let K f be the kernel of the homomorphism IY −→ f −1(IX ). Then given any 1-morphism g : Y −→ Z such that K f is in the kernel of IY −→ g−1(IZ) , the 1-morphism g factors uniquely (up to a 2-morphism) through f . In particular the rigidification f is itself uniquely determined by the sheaf K f . Lemma 3.3. Let Y be a Deligne -- Mumford stack over C, and let h : Z −→ Y be a (possibly infinite) covering map. Let f : Y −→ X be a rigidification, and K f := Ker (IY −→ f −1(IX)) . Assume that h−1(K f ) lies in the image of the homomorphism IZ −→ h−1(IY ). Then there exists a unique (up to isomorphism) covering map X ′ −→ X such that Z ∼= X ′ ×X Y . Proof. Since h is representable, the homomorphism IZ −→ h−1(IY ) is injective. We first claim that there exists a rigidification g : Z −→ Z ′ such that h−1(K f ) = Ker (IZ −→ g−1(IZ ′)) . By uniqueness in Proposition 3.2, we may construct the rigidification analytically locally on Z. Due to Proposition 2.5, we may assume that Y = [U/G], where G is a finite group acting on an affine variety U, and there exists a normal subgroup H ⊂ G such that X = [V /(G/H)], where V is the geometric quotient of U by H. Note that since the map h is representable, eU = U ×Y Z −→ U is representable, and hence eU is a analytic space. Moreover, it has an induced action of G. Now UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 9 it is easy to see that the Z ′ = [eV /(G/H)], where eV is the geometric quotient of eU by H, satisfies the claim. Finally, X ′ = Z ′ satisfies the condition in the lemma. (cid:3) Let Y be a complex analytic Deligne -- Mumford stack over C, and let y be a C-point of Y . Let πtop 1 (Y, y) denote the topological fundamental group of Y . Let Gy be the isotropy group of the stack Y at the point y. The stack [Spec (C)/Gy] will be denoted by BGy; it is well known that [Spec (C)/Gy] is the classifying space of the group Gy. Since y is a complex point, the residual gerbe at y is neutral (as C is algebraically closed!). Thus there exists a canonical morphism ηy : Spec (C)/[Gy] −→ Y . This induces a map on the fundamental groups πtop 1 (BGy, p) −→ πtop 1 (Y, y) , where p is the unique (up to a 2-isomorphism) complex point of BGy. However, We thus get a canonical map πtop 1 (BGy, p) = Gy . φy : Gy −→ πtop 1 (Y, y) (see also ([No, Section 7]). Moreover, if y′ is any other complex point of Y , then composing φy′ with a choice of an inner automorphism gives us a map πtop 1 (Y, y) −→ πtop 1 (Y, y′) φyy′ : Gy′ −→ πtop 1 (Y, y) which is well defined up to an inner automorphism of πtop 1 (Y, y). We now set up some more notation. Let f : Y −→ X be a rigidification of Deligne-Mumford stacks over C. Consider K f defined as in the statement of Lemma 3.3. For any C-point p of Y , let K f p ⊂ Gp be the stalk of K f at p. Fix a point y ∈ Y (C), and let x = f (y). Let N top(f ) denote the normal subgroup of 1 (Y, y) generated by all φyy′(K f πtop y′), where y′ runs over C-points of Y . Note that since φyy′ is well defined up to conjugation, the subgroup N top(f ) is well defined. Lemma 3.4. Notation is as above. Let h : Z −→ Y be a connected (not nec- essarily finite) Galois ´etale cover with Galois group G. Let z ∈ Z be a C-point which lies over y. Then the following conditions are equivalent. (1) N top(f ) is contained in the kernel of πtop (2) h−1(K f ) lies in the image of IZ −→ h−1(IY ). 1 (Y, y) −→ G. Proof. Let z′ be any C-point of Z, and let y′ = h(z′). As explained above, we have a natural 1-morphism BGy′ −→ Y . 10 I. BISWAS AND A. HOGADI Since h : Z −→ Y is a Galois ´etale cover with Galois group G, it follows that Zy′ = Z ×Y BGy′ → BGy′ is also a Galois ´etale cover (not necessarily connected). Here, by a non-connected Galois ´etale cover, we mean a cover such that the group of deck transformations acts transitively on all complex points. Since all connected Galois ´etale covers of BGy′ are of the form BH for a normal subgroup H ⊂ Gy′, we see that there exists a normal subgroup H ⊂ Gy′ such that Zy′ is a disjoint union of copies of BH. Statement (1) in the lemma is equivalent to the assertion that K f y′ is contained in the kernel of the composite map Gy′ = πtop 1 (BGy′, y′) φyy′ −→ πtop 1 (Y, y) −→ G for all such z′. But the kernel of the above composite map is precisely the subgroup H, men- tioned above. Thus K f y′ ⊂ H. Note that every point p of Z lying over y′ gives rise to a connected component of Z ×Y BGy′. Moreover, this connected component is nothing but BGp, where Gp is the isotropy group of Z at p. Since Z ×Y BGy′ is disjoint union of copies of BH, we see that the isotropy group at every point p of Z lying over y′ contains H and hence also contains K f y′. Since this holds for all points y′ of Y , it is equivalent to the statement 2 in the lemma. (cid:3) The following theorem is influenced by [No], and is a straightforward general- ization of results in [No]. Theorem 3.5. Let f : Y −→ X be a proper morphism of Deligne -- Mumford stacks over C. Let y be a C-point of Y , and let x = f (y). Assume that there exists a dense open substack U ⊂ X such that the codimension of X\U is at least two, and the induced morphism is a rigidification. Then the following sequence of groups is exact f U : f −1(U) −→ U 1 −→ N top(f ) −→ πtop 1 (Y, y) −→ πtop 1 (X, x) −→ 1 . Proof. Let us first consider the case where f is a rigidification. Let Et(Y ) de- note the category of covering spaces of Y . Let eC denote the full subcategory of objects Z h−→ Y in Et(Y ) such that h−1(K f ) is contained in the image of the homomorphism IZ −→ h−1(IY ), where K f is the kernel of the homomorphism IY −→ f −1(IX). By definition of N top(f ) and in view of Lemma 3.4, we need to If X ′ −→ X is an ´etale covering, then it is easy to see that X ′ ×X Y −→ Y is in show that the category eC is equivalent to the category Et(X) of coverings of X. eC. Conversely, let Z −→ Y be an ´etale covering in eC. Then by Lemma 3.3 there is an ´etale covering X ′ −→ X such that Z = X ′ ×X Y . Now consider the general case where f : Y −→ X is a rigidification outside a closed substack of codimension at least two. As above, let Et(Y ) denote the UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 11 all coverings Z −→ Y such that h−1(K f ) is contained in the image of IY −→ f −1(IX). Let U ⊂ X be the open subset of X whose complement has codimension category of all coverings of Y , and let eC be the full subcategory consisting of at least two and such that f −1(U) −→ U is a rigidification. Let eCU ⊂ Et(f −1(U)) denote the full subcategory consisting of all coverings Z −→ f −1(U) with the property that h−1(Kf −1(U )) ⊂ Image(IZ −→ If −1(U )) (just like eC ⊂ Et(Y )). We need to show that eC is equivalent to Et(X). It is clear that for any covering X ′ −→ X in Et(X), the fiber product X ′ ×X Y −→ Y is an object of eC. To complete the proof we need to show that any h : Z −→ Y in eC comes from an object in Et(X). Let hU : ZU −→ f −1(U) denote the restriction of h to f −1(U). Since f −1(U) −→ U is a rigidification, by the above special case we know that there exists an object X ′ U ×U f −1(U). Moreover, since the complement of U ⊂ X has codimension at least two, the morphism X ′ U −→ U extends to a covering X ′ −→ X. Now to finish the proof we observe that the two covering spaces Z −→ Y and X ′ ×X Y −→ Y agree on the dense open subset f −1(U) and hence are isomorphic. (cid:3) U −→ U in Et(U) such that ZU = X ′ Theorem 3.6. Let X/C be any quasi-projective orbifold. Then there exists a proper 1 -- morphism φ : Y −→ X such that (1) Y is an orbifold which is a finite global quotient, and (2) there exists a dense open subset V ⊂ X such that φ−1(V ) −→ V is a rigidification morphism, and the complement X \ V has codimension at least two. Proof. Let q : X −→ X ′ be the coarse moduli space of X. Let D ⊂ X ′ be a (reduced) divisor such that q is an isomorphism over X ′\D. Since X ′ is normal, there exists an open subset V ⊂ X ′ such that (1) D ∩ V is smooth, and (2) X ′\V has codimension at least two. Let m be a positive integer such that for every point p of X, the order of the isotropy subgroup at p divides m. Let L be a sufficiently ample line bundle on X ′, and let D′ be the zero locus of a section of Lm ⊗ OX ′(−D) such that (D + D′) ∩ V is smooth. Note that since X ′ is not necessarily smooth, the divisor D may not be Cartier. Hence the reflexive sheaf OX ′(−D) may not be a line bundle. We replace D by D + D′ without loss of generality and assume that there exists a line bundle L on X ′ such that Lm ∼= OX ′(D). Let T ⊂ L denote the cyclic covering of X ′ defined by the section D of Lm. There is a natural action of G := Z/m on T given by the action of Gm on L; by the choice of the integer m, we have a rational map T 99K X defined over all codimension one points of X. 12 I. BISWAS AND A. HOGADI Let Γ be the closure of the graph of the rational map T 99K X. We note that Γ is a Deligne -- Mumford stack (possibly singular). Since the algorithm for resolution of singularities in [Wl, p. 782, Theorem 1.0.3] commutes with ´etale base change, it also applies to Deligne -- Mumford stacks. We thus use [Wl, p. 782, Theorem 1.0.3] and functorially resolve the singularities of Γ to obtain a smooth Deligne -- Mumford stack T ′ and a proper birational map T ′ −→ Γ. By functoriality of the resolution, there is a natural action of G on T ′ such that the map T ′ −→ Γ is G -- equivariant. Also, by construction, the induced G -- equivariant map T ′ −→ X is a morphism. Thus we have an induced morphism φ : Y −→ X, where Y = [T ′/G]. It is now straight -- forward to check that the map φ satisfies the required conditions in the theorem. (cid:3) 4. Polystable vector bundles on orbifolds In this section we prove Theorem 1.2. Throughout this section we work over the field k = C. Definition 4.1. Let Z be a projective Deligne -- Mumford stack over C. A line bundle L on Z is called ample if some power of L descends to an ample line bundle on the coarse moduli space of Z. (See also [MO, 2.1].) Given a very ample line bundle L0 on a normal projective variety M0, the degree degreeL0(F0) of a torsionfree coherent sheaf F0 on M0 with respect to L0 is defined to be the degree of the restriction of F0 to the general complete intersection curve obtained by intersection hyperplanes on M0 from the complete linear system for L0. Fix an ample line bundle L on a projective Deligne -- Mumford stack Z. Let Z0 be the coarse moduli space of Z. The degree of a torsionfree coherent sheaf F on Z with respect to L is defined to be degree(F ) = degreeL(F ) := 1 md−1n degreeLm((det F )⊗n) , where d = dim Z0, the integer m is such that Lm descends to a very ample line bundle on Z0, and n is such that (det F )⊗n descends to Z0. It is easy to see that degree(F ) is independent of the choices of m and n, but it depends on L. A torsionfree coherent sheaf F on Z is called stable (respectively, semistable) if for all coherent subsheaves F ′ ⊂ F with 0 < rank(F ′) < rank(F ), degree(F ′) rank(F ′) < degree(F ) rank(F ) (respectively, degree(F ′) rank(F ′) ≤ degree(F ) rank(F ) ) . A semistable sheaf is called polystable if it is a direct sum of stable sheaves. 4.1. Normalization. Let X be an integral variety and Spec (L) −→ X be a dominant morphism, where L is a field which is a finite extension of the function field of X. Then one has the notion of normalization of X in L. One can easily extend this notion to the case when L is replaced by a product of fields where the UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 13 morphism is now dominant when restricted to each of the component fields. Now, if eX denotes the normalization of X in L, and U −→ X is an ´etale morphism, then U ×X Spec (L) is a product of fields with the map U ×X Spec (L) −→ U being dominant on each component of the domain. Let eU denote the normaliza- tion of U in U ×X Spec (L). Then the following diagram is Cartesian: eU U eX / X In other words, normalization commutes with ´etale (or even smooth) base change. Thus, one can extend the notion to Deligne -- Mumford stacks. In other words, if X is a Deligne -- Mumford stack, and Spec (L) −→ X is a dominant morphism R) in U ×X Spec (L) (respectively, R ×X Spec (L)). Since R ⇒ U is an ´etale from a field, one can define eX, the normalization of X in L. Concretely, to define eX, one first chooses an ´etale atlas U −→ X. Define R := U ×X U, and denote by eU (respectively, eR) the normalization of U (respectively, groupoid, so is eR ⇒ eU . We leave these details to the reader, since they are simple exercise dealing with normalization of varieties. One then defines eX as the quotient [eU /eR]. This construction can be seen more clearly when X is quotient on eU , and hence one defines eX as [eU/G]. of a scheme U by a finite group G. In this case one simply has a natural G action 4.2. Proof of Theorem 1.2. We first establish some ingredients of the proof. Lemma 4.2. Let X/C be a normal orbifold, and let K be its function field. Let f : Y −→ X be a dominant finite morphism, with Y normal, such that the corresponding function field extension is Galois. Let G be this Galois group acting on Y . Let V be a vector bundle on X, and let W be a G -- invariant reflexive subsheaf of f ∗V . Then there is a closed subset S ⊂ Y of codimension at least two such that W Y \S is a pullback of a unique subsheaf of V X\f (S). Proof. First assume that X is a variety. Outside a closed subset S ⊂ Y of codimension at least two, the map f is flat, and W is locally free. The isotropies for the action of G on Y act trivially on the fibers of f ∗V , and hence on the fibers of W . Therefore, W descends outside f (S) to a unique subbundle (see [Ne, Theorem 1.2] for descent of sheaves). In the general case, let U −→ X be an atlas. From the uniqueness it is enough to prove by replacing X and Y by U and Y ×X U respectively, and also replacing the sheaves by the corresponding pullbacks. Then it is reduced to the above case. (cid:3) / / / 14 I. BISWAS AND A. HOGADI Lemma 4.3. Let X/C be a projective normal orbifold, and let L be an ample line bundle on X. Let f : Y −→ X be a dominant finite morphism, where Y /C is a normal projective variety. Let E be a vector bundle on X. Then the following two hold: (i) The vector bundle E is semistable with respect to L if and only if f ∗E is semistable with respect to f ∗(L). (ii) If E is polystable with respect to L, then f ∗E is polystable with respect to f ∗(L). Proof. After one understands the notion of normalization of a Deligne -- Mumford stack in a field (see § 4.1), the proof is similar to the case when X is a variety instead of an orbifold. If f ∗E is semistable with respect to f ∗L, then clearly E is semistable with respect to L. To prove the converse, assume that E is semistable with respect to L. We may replace Y by its normalization in the Galois closure of the function field of Y over that of X. If f ∗E is not semistable, take the first term F in the Harder -- Narasimhan filtration of f ∗E. The descent of F (see Lemma 4.2) contradicts semistability of E. Therefore, f ∗E is semistable. Now assume that E is polystable with respect to L. Hence f ∗E is semistable with respect to f ∗L (by part (i)). Consider the socle F ⊂ f ∗E; it is the unique maximal polystable subsheaf with degree(F )/rank(F ) = degree(f ∗E)/rank(f ∗E) (see [HL, p. 23, Lemma 1.5.5]). From the uniqueness of F it follows that F is a pullback of a subsheaf F ′ of E outside codimension two (Lemma 4.2). Since E is polystable, F ′ ⊂ E has a direct summand F ′′. If F 6= f ∗E, the direct sum of F with the socle of f ∗F ′′ contradicts the maximality of F . Hence F = f ∗E, implying that f ∗E is polystable. (cid:3) Let X and L be as above. Let d be the dimension of X. All Chern classes considered here are topological, taking values in rational cohomology classes. Proposition 4.4. Let f : Y −→ X be any dominant finite morphism of pro- jective orbifolds. Fix an ample line bundle L on X. So f ∗L is ample on Y . Let E be a polystable vector bundle on X with respect to L such that c1(E) = 0, and c2(E) · c1(L)d−2 = 0. Then f ∗E is a polystable vector bundle on Y with respect to f ∗L. Also, c1(f ∗E) = 0, and c2(f ∗E) · c1(f ∗L)d−2 = 0. Proof. Clearly, c1(f ∗E) = 0, and c2(f ∗E) · c1(f ∗L)d−2 = 0, because these condi- tions hold E and L. By [KV] we can find a diagram h g Y ′ f ′ X ′ Y f / X where Y ′ and X ′ are smooth projective varieties, and the horizontal morphisms are finite and dominant. By Lemma 4.3, the pullback g∗E is polystable, and also, / / / UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 15 c1(g∗E) = 0, and c2(g∗E) · c1(g∗L)d−2 = 0; note that g∗L is ample because g is finite. Thus, g∗E is given by a unitary representation of the fundamental group of X ′ (see Theorem 1.1 and the paragraph following it). Therefore, f ′∗g∗E is also given by a unitary representation of the fundamental group of Y ′, and hence it is polystable. Thus by Lemma 4.3, the pullback f ∗E is polystable. (cid:3) Remark 4.5. We first observe that proving Theorem 1.2 is equivalent to showing that if E −→ X is polystable, and πX : UX −→ X is the universal covering space of X, then (1) π∗ (2) π∗ X(E) is trivial, and X(E) admits a πtop 1 (X, x) invariant flat unitary metric. Lemma 4.6. Let X be a finite global quotient. Then Theorem 1.2 holds for X. Proof. Take E −→ X as in Theorem 1.2. First assume that E is given by a unitary representation ρ of πtop 1 (X, x). To prove that E is polystable, first note that for any vector bundle F on X given by a unitary representation of πtop 1 (X, x), we have c1(F ) = 0 = c2(F ). Also, any unitary representation is a direct sum of irreducible unitary representations. Therefore, it is enough to prove polystability under the assumption that ρ is irreducible. Assume that ρ is irreducible. By hypothesis, there exists a smooth projective variety Y and a finite group G acting on Y such that X = [Y /G]. Let f : Y −→ X be the natural map. Since f ∗E is given by the unitary representation f ∗ρ, it follows that f ∗E is polystable with respect to any polarization on Y , in particular with respect to f ∗L [Ko, pp. 177 -- 178, Theorem 8.3]. We note that the vector bundle E is polystable because f ∗E is so (see Lemma 4.3). To prove that E is stable, let F ⊂ E be a nonzero subsheaf such that rank(F ) < rank(E), and degree(F ) = 0. Then degree(f ∗F ) = 0. This implies that f ∗F generates a subbundle of f ∗E which is given by some some subrepre- sentation of f ∗ρ (see the last paragraph of page 178 in the proof of [Ko, Theorem 8.3]). Since f ∗F is a pullback from X, this contradicts the irreducibility of ρ. Hence E is stable. To prove the converse, assume that E −→ X is a polystable vector bundle with c1(E) = 0 = c2(E) · c1(L)n−2. Take Y as above such that X = [Y /G]. Then f ∗E is a polystable vector bundle on Y with c1(f ∗E) = 0 and c2(f ∗E)·c1(f ∗L)n−2 = 0, where f is the quotient map. Let πY : UY −→ Y denote the universal covering space of Y . Then π∗ on UY with a πtop UY is also the covering space for X. Moreover, if x = f (y), then πtop Y f ∗E is a trivial vector bundle 1 (Y, y) -- invariant unitary flat metric, say h−, −i. Note that 1 (Y, y) is 16 I. BISWAS AND A. HOGADI naturally a finite index subgroup of πtop representatives. It is then clear that the "averaged" metric 1 (X, x). Let {gi} denote its finite coset (v1, v2) := X i g∗ i hv1, v2i is a πtop 1 (X, x) invariant unitary flat metric on UY . (cid:3) We now reduce the general case of Theorem 1.2 to the above special case using Theorem 3.6. Proof of Theorem 1.2. As in Lemma 4.6, if E is given by a unitary representation of πtop 1 (X, x), then E is polystable, and c1(E) = 0 and c2(E) · c1(L)n−2 = 0. Assume that E is polystable, and c1(E) = 0 and c2(E) · c1(L)n−2 = 0. Let φ : Y −→ X be a morphism as guaranteed by Theorem 3.6. Fix a point y0 of Y . Then, since Y is a finite global quotient, by Lemma 4.6 and Proposition 4.4, the pullback φ∗E is given by a unitary representation of πtop 1 (Y, y0). However, for any point y ∈ Y , the kernel of Gy −→ Gφ(y) acts trivially on the fiber of φ∗(E) at y. Thus this kernel is also contained in the kernel of the unitary representation. Then by Theorem 3.5, the unitary representation of πtop 1 (Y, y0) which defines φ∗E actually factors through πtop 1 (X, φ(y0)). This completes the proof of Theorem 1.2. (cid:3) As a consequence of Theorem 1.2, we obtain the following generalization of Theorem 1.1 to certain types of singular varieties. Corollary 4.7. Let X/C be a projective variety with at worst quotient singular- ities. Assume X is the coarse moduli space of a projective orbifold eX/C (e.g. X has isolated singular points). Then Theorem 1.1 holds for X. Proof. Let L be an ample line bundle on X and E be a polystable vector bundle satisfying c1(E) = 0 and c2(E) · c1(L)n−2 = 0, where n is the dimension of X. (see Lemma 4.3), by Theorem 1.2, the vector bundle π∗E is given by a unitary representation of πtop the vector bundle arising from this representation is a pull back from X, it is Let π : eX −→ X be the given morphism. Since the pullback π∗E is polystable 1 ( eX, x) (for some geometric point x of eX). However, since clear that for every point y of eX, the isotropy group Gy at y lies in the kernel of this representation. Thus by using [No, p. 90, § 8.1] this descends to give a unitary representation of πtop (cid:3) 1 (X, x), which defines the vector bundle E. References [ACV] D. Abramovich, A. Corti and A. Vistoli, Twisted bundles and admissible covers. [Do] [HL] Comm. Alg. 31 (2003), 3547 -- 3618. S.K. Donaldson, Infinite determinants, stable bundles and curvature. Duke Math. Jour. 54 (1987), 231 -- 247. D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves. Aspects of Mathematics, E31. Friedr. Vieweg & Sohn, Braunschweig, 1997. UNITARY REPRESENTATIONS OF THE FUNDAMENTAL GROUP 17 [KM] [Ko] [Kr] [KV] [MO] [MR] [NS] [Ne] [No] [Wl] S. Keel and S. Mori, Quotients by groupoids. Ann. of Math. 145 (1997), 193 -- 213. S. Kobayashi, Differential Geometry of Complex Vector Bundles. Publications of the Math. Society of Japan 15, Iwanami Shoten Publishers, Tokyo, and Princeton Uni- versity Press, Princeton, NJ, 1987. A. Kresch, On the geometry of Deligne -- Mumford stacks. Algebraic geometry -- Seattle 2005. Part 1, 259 -- 271, Proc. Sympos. Pure Math., 80, Amer. Math. Soc., Providence, RI, 2009. A. Kresch and A. Vistoli, On coverings of Deligne -- Mumford stacks and surjectivity of the Brauer map. Bull. London Math. Soc. 36 (2004), 188 -- 192. K. Matsuki and M. Olsson, Kawamata-Viehweg vanishing as Kodaira vanishing for stacks. Math. Res. Lett. 12 (2005), 207 -- 217. V. Mehta and A. Ramanathan, Restriction of stable sheaves and representations of the fundamental group. Invent. Math. 77 (1984), 163 -- 172. M.S. Narasimhan and C.S. Seshadri, Stable and unitary vector bundles on a compact Riemann surface. Ann. of Math. 82 (1965), 540 -- 567. T. Nevins, Descent of coherent sheaves and complexes to geometric invariant theory quotients. Jour. Alg. 320 (2008), 2481 -- 2495. B. Noohi, Fundamental groups of algebraic stacks. Jour. Inst. Math. Jussieu 3 (2004), 69 -- 103. J. W lodarczyk, Simple Hironaka resolution in characteristic zero. Jour. Amer. Math. Soc. 18 (2005), 779 -- 822. School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road, Bombay 400005, India
1508.06426	1	1508	2015-08-26T09:48:05	Integral and adelic aspects of the Mumford-Tate conjecture	[ "math.AG", "math.NT" ]	Let $Y$ be an abelian variety over a subfield $k \subset \mathbb{C}$ that is of finite type over $\mathbb{Q}$. We prove that if the Mumford-Tate conjecture for $Y$ is true, then also some refined integral and adelic conjectures due to Serre are true for $Y$. In particular, if a certain Hodge-maximality condition is satisfied, we obtain an adelic open image theorem for the Galois representation on the (full) Tate module of $Y$. Our second main result is an (unconditional) adelic open image theorem for K3 surfaces. The proofs of these results rely on the study of a natural representation of the fundamental group of a Shimura variety.	math.AG	math	Integral and adelic aspects of the Mumford-Tate conjecture by Anna Cadoret and Ben Moonen Abstract. Let Y be an abelian variety over a subfield k ⊂ C that is of finite type over Q. We prove that if the Mumford-Tate conjecture for Y is true, then also some refined integral and adelic conjectures due to Serre are true for Y . In particular, if a certain Hodge-maximality condition is satisfied, we obtain an adelic open image theorem for the Galois representation on the (full) Tate module of Y . Our second main result is an (unconditional) adelic open image theorem for K3 surfaces. The proofs of these results rely on the study of a natural representation of the fundamental group of a Shimura variety. AMS 2010 Mathematics Subject Classification: 11G18, 14G35 5 1 0 2 g u A 6 2 ] . G A h t a m [ 1 v 6 2 4 6 0 . 8 0 5 1 : v i X r a Introduction This paper has two main results, one about abelian varieties, the other about K3 surfaces. The underlying mechanism is the same in both cases. To explain the results, let Y be either an abelian variety or a K3 surface over a subfield k ⊂ C that is finitely generated over Q. Write H = H1(cid:0)Y (C), Z(cid:1) in the first case and H = H 2(cid:0)Y (C), Z(cid:1)(1) if Y is a K3 surface. Let GB ⊂ GL(H) be the Mumford-Tate group. We may identify H ⊗ Z with the étale cohomology of Y with Z-coefficients (H1 or H 2(1)); this gives us a Galois representation It is known that, possibly after replacing k with a finite extension, the image of ρY is contained in GB(Z). From now on, we assume this is the case. If ℓ is a prime number, let ρY,ℓ : Gal(¯k/k) → ρY : Gal(¯k/k) → GL(H)(cid:0)Z(cid:1) . GL(H)(cid:0)Zℓ(cid:1) be the ℓ-primary component of ρY . In the case of an abelian variety, our main result says that the usual Mumford-Tate conjecture implies an integral refinement of it. If moreover a certain maximality condition is satisfied (see Sec- tion 2), we obtain an adelic open image result for ρY . These refined statements were conjectured by Serre. The precise result is as follows. Theorem A. Let Y be an abelian variety over k for which the Mumford-Tate conjecture is true. of ρY,ℓ contains the commutator subgroup of GB(Zℓ), as well as the integral homotheties Z∗ (i) The index (cid:2)GB(Zℓ) : Im(ρY,ℓ)(cid:3) is bounded when ℓ varies. Moreover, for almost all ℓ the image (ii) If the Hodge structure H1(cid:0)Y (C), Q(cid:1) is Hodge-maximal (see Def. 2.3), the image of ρY is an open subgroup of GB(Af ). ℓ · id. By a result of Larsen and Pink (see [11], Thm. 4.3), if the Mumford-Tate conjecture for an abelian variety is true for one prime number ℓ, it is true for all ℓ. The assumption that the Mumford-Tate conjecture for Y is true is therefore unambiguous. Let us also note that Hodge-maximality is a necessary condition for the image of ρY to be open in GB(Af ); see Remark 2.6. 1 In the case of a K3 surface, the Mumford-Tate conjecture is known (due to Tankeev [22] and, independently, André [1]), and we prove that the Hodge-maximality assumption is always satisfied. In this case we obtain the following adelic open image theorem. Theorem B. If Y is a K3 surface, the image of ρY is an open subgroup of GB(Af ). The proofs of these theorems rely on the fact that the moduli spaces of abelian varieties and K3 surfaces are (essentially) Shimura varieties. In Section 3 we recall how to associate to a Shimura datum (G, X) and a neat compact open subgroup K0 ⊂ G(Af ) a representation φ : π1(S0) → K0 ⊂ G(Af ), where S0 ⊂ ShK0(G, X) (over some number field F ) is a geometrically irreducible component. This reflects the fact that we have a tower of finite étale covers ShK(G, X) → ShK0(G, X) indexed by the open subgroups K ⊂ K0. For ℓ a prime number, let φℓ : π1(S0) → K0,ℓ be the ℓ-primary component of φ. The main technical result we prove, as Corollary 3.7, is that, under a mild assumption on the datum (G, X), the image of the adelic representation φ is “big” in precisely the sense as in the conclusions of the main theorems. The proof of this result only involves abstract theory of Shimura varieties; it relies on Deligne’s group-theoretic description of the reciprocity law that gives the Galois action on the geometric connected components of Sh(G, X). To understand how this result leads to our main theorems, we have to link the representation φ to the Galois representation ρY on the cohomology of Y . Taking G to be the Mumford-Tate group of Y , we obtain a Shimura variety that has an interpretation as a moduli space of abelian varieties or K3 surfaces with additional structures. If y ∈ S0(k) is the point corresponding to Y , we obtain a homomorphism σy : π1(y) → π1(S0) (with π1(y) ∼= Gal(¯k/k)), and the composition φ ◦ σy is isomorphic to the Galois representation ρY . Using a classical result of Bogomolov, the assumption that the Mumford-Tate conjecture for Y is true, for some prime number ℓ, implies that the image of φℓ ◦ σy is open in the image of φℓ. Points y ∈ S0(k) for which this holds are said to be ℓ-Galois-generic with respect to φ. For points on a Shimura variety of abelian type, a result of the first author and Kret says that being ℓ-Galois-generic for some ℓ implies something that is a priori much stronger, namely that the image of the adelic representation φ ◦ σy is open in the image of φ. (See [6], Thm. 1.1. This result is a consequence of the open adelic image theorem for abelian schemes proven by the first author in [5].) Our main results are obtained by combining this with our result on the image of φ. The first two sections of the paper are of a preliminary nature. We recall some conjectures due to Serre that refine the Mumford-Tate conjecture. Also we discuss the notion of (Hodge-)maximality, which is a necessary condition for an adelic open image theorem to hold. Section 3 forms the core of the paper. We define the representation φ associated with a Shimura variety; further we state and prove the main result, Theorem 3.6 and its Corollary 3.7, about the image of φ. In Section 4 we briefly recall various notions of Galois-genericity, and we state the result of Cadoret-Kret that we need. In Section 5, which is devoted to abelian varieties, we prove Theorem A. Also we give examples of abelian varieties for which the H1 is not Hodge-maximal. These examples suggest that Hodge-maximality depends in a rather subtle way on the structure of the Mumford-Tate group. Finally, in Section 6 we discuss K3 surfaces. We prove that the H 2(1) of a K3 surface is always Hodge-maximal, and we deduce Theorem B. Acknowledgement. We thank Akio Tamagawa for his interest and for helpful discussions. 2 1. An integral variant of the Mumford-Tate conjecture 1.1 Let Y be a smooth proper scheme of finite type over a subfield k of C that is finitely generated over Q. Fix integers i and n. Let H = H i(cid:0)Y (C), Z(cid:1)(n)/(torsion), which is a polarizable Hodge structure of weight i − 2n. We denote by GB ⊂ GL(H) the Mumford-Tate group. By this we mean that the generic fibre GB,Q ⊂ GL(HQ) is the Mumford-Tate group of HQ in the usual sense, and that GB is the Zariski closure of GB,Q inside GL(H). If the context requires it, we include Y in the notation, writing GB,Y , etc. For a prime number ℓ, let Hℓ = H i(cid:0)Y¯k, Zℓ(cid:1)(n)/(torsion), which is a free Zℓ-module of finite rank on which we have a continuous Galois representation ρℓ : Gal(¯k/k) → GL(Hℓ) . We denote by Gℓ ⊂ GL(Hℓ) the Zariski closure of the image of ρℓ. The generic fibre Gℓ,Qℓ ⊂ GL(Hℓ,Qℓ) is the Zariski closure of the image of the Galois representation on Hℓ,Qℓ. We define G0 ℓ ⊂ Gℓ to be the Zariski closure of the identity component (Gℓ,Qℓ)0. If we replace k by a finitely generated extension, Gℓ may become smaller, but its identity compo- nent G0 ℓ does not change. By a result of Serre (see [20] or [10], Prop. 6.14), there exists a finite field extension k ⊂ kconn in C (depending on Y , i and n) such that for every field K that contains kconn and every prime number ℓ, the generic fibre of Gℓ,YK is connected. Via the comparison isomorphism H ⊗ Zℓ GL(Hℓ). ∼ −→ Hℓ, we may view GB ⊗ Zℓ as a subgroup scheme of 1.2 Mumford-Tate Conjecture. With notation as above, GB ⊗ Zℓ = G0 GL(Hℓ). ℓ as subgroup schemes of Note that, though we have stated the conjecture using group schemes over Zℓ, the Mumford-Tate conjecture in this form is equivalent to the conjecture that GB ⊗ Qℓ equals G0 as algebraic subgroups of GL(Hℓ,Qℓ), which is the Mumford-Tate conjecture as it is usually stated. As Hℓ is Hodge-Tate, it follows from a result of Bogomolov [2] (with some extensions due to Serre; see also [20]) that the image of ρℓ is open in Gℓ(Qℓ). Hence the Mumford-Tate conjecture is equivalent to the assertion that Im(ρℓ) is an open subgroup of GB(Zℓ) (assuming k = kconn). Further note that the Mumford-Tate conjecture depends, a priori, on ℓ and also on the chosen complex embedding of k. ℓ,Qℓ The following strengthening of the Mumford-Tate conjecture was proposed by Serre; see Conjec- ture C.3.7 in [19] and cf. [21]. 1.3 Integral Mumford-Tate Conjecture (Serre). Retain the notation of 1.1, and assume k = kconn. Then for all ℓ the image Im(ρℓ) is contained in GB(Zℓ) as an open subgroup, and the index (cid:2)GB(Zℓ) : Im(ρℓ)(cid:3) is bounded when ℓ varies. Further, for almost all ℓ the image of ρℓ contains the commutator subgroup of GB(Zℓ) and all homotheties of the form ci−2n · id, for c ∈ Z∗ ℓ . Compared with the usual Mumford-Tate conjecture, the main point in the above conjecture is that it should be possible to bound the index of Im(ρℓ) in GB(Zℓ) by a constant independent of ℓ. 3 2. Maximality, and an adelic form of the Mumford-Tate conjecture 2.1 Definition. Let M be a connected algebraic group over a field k of characteristic 0. Let k ⊂ F be a field extension, S an algebraic group over F , and h : S → MF a homomorphism. Then we say that h is maximal if there is no non-trivial isogeny of connected k-groups M ′ → M such that h lifts to a homomorphism S → M ′ F . Note that if F is algebraically closed, maximality of h only depends on its M (F )-conjugacy class. 2.2 The following remarks closely follow [27], 0.2. Let M be a connected reductive group over a subfield k ⊂ C. Let C be a conjugacy class of complex cocharacters µ : Gm,C → MC. Let π1(M ) denote the fundamental group of M as defined by Borovoi in [3]. This is a finitely generated Z- module with a continuous action of Γ = Gal(¯k/k). If (X ∗, R, X∗, R∨) is the root datum of M¯k and Q(R∨) = hR∨i ⊂ X∗ is the coroot lattice, π1(M ) ∼= X∗/Q(R∨). The conjugacy class C of complex cocharacters corresponds to an orbit C ⊂ X∗ under the Weyl group W . As the induced W -action on π1(M ) is trivial, any two elements in C have the same image in π1(M ); call it [C ] ∈ π1(M ). If M ′ is a connected reductive k-group and f : M ′ → M is an isogeny, the map induced by f identifies π1(M ′) with a Z[Γ]-submodule of finite index in π1(M ). Conversely, every such submodule comes from an isogeny of connected k-groups, which is unique up to isomorphism over M. A conjugacy class C as above lifts to M ′ if and only if [C ] ∈ π1(M ′). We shall usually be in a situation where the Z[Γ]-submodule spanned by [C ] has finite index in π1(M ). (See below.) In this case, there is a uniquely determined maximal isogeny M ′ → M of connected k-groups such that the µ ∈ C lift to complex cocharacters of M ′. Further, the cocharacters µ ∈ C are maximal in the sense of Definition 2.1 if and only if [C ] generates π1(M ) as a Z[Γ]-module. 2.3 Definition. Let V be a Q-Hodge structure, given by the homomorphism h : S → GL(V )R. Let M ⊂ GL(V ) be the Mumford-Tate group. Then V is said to be Hodge-maximal if h : S → MR is maximal in the sense of Definition 2.1. Hodge-maximality of V is equivalent to the condition that the associated cocharacter µ : Gm,C → MC is maximal. This allows us to apply 2.2, taking k = Q and Γ = Gal(Q/Q). If C is the M (C)- conjugacy class of µ, the assumption that M is the Mumford-Tate group of V implies that the Z[Γ]- submodule of X∗(M ) generated by C has finite index. Hence also the Z[Γ]-submodule of π1(M ) generated by [C ] has finite index. By what was explained in 2.2, V is Hodge-maximal if and only if Z[Γ] · [C ] = π1(M ). 2.4 Retaining the notation and assumptions of 1.1, let H =Qℓ Hℓ, where the product is taken over all prime numbers ℓ. We then have a continuous Galois representation ρ : Gal(¯k/k) → GL( H) whose ℓ-primary component is the representation ρℓ defined in 1.1. The comparison isomorphism between singular and étale cohomology gives an isomorphism H ∼= H ⊗ Z; via this we may view ρ as a representation taking values in GL(H)(cid:0)Z(cid:1). ture C.3.8 in [19]. The following adelic version of the Mumford-Tate conjecture was proposed by Serre; see Conjec- 4 2.5 Adelic Mumford-Tate Conjecture (Serre). With notation as in 1.1, suppose that k = kconn and that the Hodge structure H is Hodge-maximal. Then Im(ρ) is an open subgroup of GB(Af ). It follows from a result of Wintenberger that the Hodge-maximality of H is essential. 2.6 Remark. (For simplicity we shall assume here that the ground field k is a number field.) Indeed, suppose there exists a non-trivial isogeny of Q-groups M ′ → GB with M ′ connected, such that h : S → GB,R lifts to a homomorphism S → M ′ R. By [26], Théorème 2.1.7, and possibly after replacing the ground field k with a finite extension, the ℓ-adic representations ρℓ lift to Galois representations with values in M ′. On the other hand, it follows from [15], Proposition 6.4, that the image of M ′(Af ) → GB(Af ) is not open in GB(Af ); hence Im(ρ) cannot be open in GB(Af ). 2.7 Let (G, X) be a Shimura datum such that G is the generic Mumford-Tate group on X. By definition, this means that there exist points h ∈ X for which there is no proper subgroup G′ ⊂ G such that h : S → GR factors through G′ R. The locus of points h for which this holds forms a subset X Hgen ⊂ X called the Hodge-generic locus. Similar to the definition in 2.3, we say that (G, X) is maximal if there is no non-trivial isogeny of Shimura data f : (G′, X ′) → (G, X). (Note that G′ is necessarily connected, as it is part of a Shimura datum.) Clearly, if (G, X) is maximal then all h : S → GR in X Hgen are maximal in the sense of Definition 2.1. Conversely, if some h ∈ X Hgen is maximal then (G, X) is maximal. (If we have f as above, f (X ′) ⊂ X is a union of connected components but need not be the whole X; however, changing f by an inner automorphism of G we can always ensure that some given h ∈ X lies in f (X ′).) To each h ∈ X corresponds a complex cocharacter µh of G, and the µh thus obtained all lie in a single G(C)-conjugacy class C (G, X). It follows from the previous remarks that (G, X) is maximal if and only if the associated class (cid:2)C (G, X)(cid:3) generates π1(G) as a Z[Γ]-module. (i) Let f : (G1, X1) → (G2, X2) be a morphism of Shimura data with f : G1 → G2 2.8 Remarks. surjective. If G1 is the generic Mumford-Tate group on X1 then G2 is the generic Mumford-Tate group on X2. If f : G1 → G2 is an isogeny then also the converse is true. (ii) If in (i) Ker(f ) is semisimple then also maximality is preserved: if (G1, X1) is maximal, so is (G2, X2). Indeed, in this case π1(G2) is a quotient of π1(G1) in such a way that (cid:2)C (G2, X2)(cid:3) is the image of (cid:2)C (G1, X1)(cid:3). (iii) Given a Shimura datum (G, X), it follows from the remarks in 2.2 that, up to isomorphism, there exists a unique isogeny of Shimura data f : ( G, X) → (G, X) such that ( G, X) is maximal. By (i), G is the generic Mumford-Tate group on X if and only if G is the generic Mumford-Tate group on X. 3. Adelic representations associated with Shimura varieties 3.1 Let (G, X) be a Shimura datum. Throughout we assume that G is the generic Mumford-Tate group on X. (See 2.7.) In this case, conditions (2.1.1.1–5) of [8] are satisfied and Z(Q) is discrete in Z(Af ). (Cf. [8], 2.1.11; for details see also [24], Lemma 5.13.) If K ⊂ G(Af ) is a compact open subgroup, we have ShK (G, X)(cid:0)C(cid:1) = G(Q)\X × G(Af )/K. For h ∈ X and γK ∈ G(Af )/K, let [h, γK] denote the corresponding C-valued point of ShK(G, X). Let K0 ⊂ G(Af ) be a neat compact open subgroup. If K ⊂ K0 is an open subgroup, the induced morphism on Shimura varieties ShK,K0 : ShK (G, X) → ShK0(G, X) is finite étale. If, moreover, K is normal in K0, this morphism is Galois with group K0/K. 5 Choose a point h0 ∈ X. Let S0,C be the irreducible component of ShK0(G, X)C that contains the point [h0, eK0]. Let F be the field of definition of this component, which is a finite extension of the reflex field E(G, X). To simplify notation, we write ShK for ShK(G, X)F . By construction, we have a geometrically irreducible component S0 ⊂ ShK0. For K an open normal subgroup of K0, let SK ⊂ ShK be the inverse image of S0 ⊂ ShK0 under the transition morphism ShK,K0. Then SK → S0 is étale Galois with group K0/K. Let ¯sK = [h0, eK] ∈ SK(C). The system of points ¯s = (¯sK ) thus obtained is compatible in the sense that ShK2,K1(¯sK2) = ¯sK1 for K2 ⊂ K1 ⊂ K0. We abbreviate ¯sK0 to ¯s0. With this choice of base points, SK → S0 corresponds to a homomorphism φK : π1(S0, ¯s0) → K0/K. If K2 ⊂ K1 are open normal subgroups of K0, the homomorphism φK1 equals the composition of φK2 : π1(S0, ¯s0) → K0/K2 and the canonical map K0/K2 → K0/K1. We may therefore pass to the limit; as the intersection of all open normal subgroups K ⊂ K0 is trivial, this gives a continuous homomorphism (3.1.1) φ¯s : π1(S0, ¯s0) → K0 . 0 ⊂ G′(Af ) be neat compact open subgroups with f (K0) ⊂ K ′ (i) The homomorphism φ is functorial in the following sense. Let f : (G, X) → 3.2 Remarks. (G′, X ′) be a morphism of Shimura data. On reflex fields we have E(G′, X ′) ⊂ E = E(G, X). Let K0 ⊂ G(Af ) and K ′ 0. Choose h0 ∈ X and let 0 = f (h0) ∈ X ′. As in 3.1, this gives rise to geometrically irreducible components S0 ⊂ ShK0(G, X)F h′ and S′ 0 give rise to compatible (G′, X ′)E systems of base points ¯s = (¯sK) and ¯s′ = (¯s′ restricts to a morphism S0 → S′ (G′, X ′)F ′, and it is easy to see that EF ′ ⊂ F . Further, h0 and h′ K ′). The morphism Sh(f ) : ShK0(G, X) → ShK ′ 0. We then have a commutative diagram 0,F over F with ¯s0 7→ ¯s′ 0 ⊂ ShK ′ 0 0 π1(S0, ¯s0) Sh(f )∗ π1(S′ 0,F , ¯s′ 0) ⊂ π1(S′ 0, ¯s′ 0) φ¯s K0 f φ′ ¯s′ K ′ 0 (ii) The homomorphism φ is essentially independent of the choice of h0 ∈ X and the resulting system of base points ¯s. If we choose another point h′ 0 ∈ X that lies in the same connected component as h0, this gives rise to a different collection of base points ¯s′. There is a canonically determined 0). For α in this class, the homomorphisms conjugacy class of isomorphisms α : π1(S0, ¯s0) φ¯s and φ¯s′ ◦ α differ by an inner automorphism of K0. −→ π1(S0, ¯s′ ∼ If h′ 0 lies in a different connected component of X, there exists an inner automorphism α = Inn(g) of (G, X) such that α(h0) and h′ 0 lie in the same component of X. By functoriality together with the previous case, it follows that the associated representations φ¯s and φ¯s′ are conjugate when we restrict to suitable subgroups of finite index in the respective π1’s. In view of the above remarks, we shall from now on omit the base point ¯s0 from the notation, unless it plays a role in the discussion. 3.3 Our main goal in this section is to describe the image of the representation φ = φ¯s defined in 3.1. In order to do this, we need to recall some definitions and results from the theory of Shimura varieties. For proofs of the stated results we refer to [8], Section 2. Throughout, (G, X) is a Shimura datum as in 3.1. Let G(R)+ ⊂ G(R) be the subgroup of elements that are mapped into the identity component Gad(R)+ ⊂ Gad(R) (for the Euclidean topology) under the adjoint map. 6 Let γ : G → Gder denote the simply connected cover of the derived group of G. Then G(Q)γ G(A) is a closed normal subgroup of G(A), and we define π(G) = G(A)/G(Q)γ G(A). Next define ¯π0π(G) = π0π(G)/π0G(R)+, where π0 means the group of connected components. This ¯π0π(G) is an abelian profinite group. Define G(Q)+ = G(Q) ∩ G(R)+, and let G(Q)− homomorphism G(Af ) → ¯π0π(G) induces an isomorphism G(Af )/G(Q)− + risk of confusion we identify the two groups. + denote its closure inside G(Af ). The natural −→ ¯π0π(G). If there is no ∼ The group G(Af ) acts on the Shimura variety Sh(G, X) from the right. This action makes the set π0(cid:0)Sh(G, X)C(cid:1) a torsor under ¯π0π(G). 3.4 Let E = E(G, X) be the reflex field and Eab its maximal abelian extension. As π0(cid:0)Sh(G, X)C(cid:1) is a torsor under ¯π0π(G), which is abelian, the action of Gal(E/E) on π0(cid:0)Sh(G, X)C(cid:1) gives rise to a well-determined homomorphism (3.4.1) rec : Gal(Eab/E) → ¯π0π(G) ∼= G(Af )/G(Q)− + , called the reciprocity homomorphism. Let q : G(Af ) → ¯π0π(G) be the canonical map. For K ⊂ G(Af ) a compact open subgroup, we have an induced action of ¯π0π(G) on the set of irreducible components of ShK (G, X)C. All these components have the same stabilizer in ¯π0π(G), namely q(K). Let recK : Gal(Eab/E) → ¯π0π(G)/q(K) denote the reciprocity map modulo q(K). 3.5 Proposition. Retain the notation and assumptions of 3.1. Then the image of the homomorphism φ : π1(S0) → K0 is the subgroup q−1(cid:0)Im(rec)(cid:1) ∩ K0 of K0. Proof. Since F is defined to be the field of definition of the irreducible component S0,C ⊂ ShK0(G, X)C, we have as subgroups of Gal(Eab/E). For K a normal open subgroup of K0, the set of geometric irreducible components of SK is a torsor under q(K0)/q(K) ⊂ ¯π0π(G)/q(K), and the irreducible components of SK (over F ) correspond to the orbits under the action of Gal(Eab/F ) via recK. Hence the image of Gal(Eab/F ) = rec−1(cid:0)q(K0)(cid:1) φK : π1(S0) → K0/K is the inverse image under q : K0/K → q(K0)/q(K) of recK(cid:0)Gal(Eab/F )(cid:1). The latter group is the image of(cid:0)Im(rec) ∩ q(K0)(cid:1) in q(K0)/q(K), and so we find that Im(φK ) is the image of q−1(cid:0)Im(rec)(cid:1) ∩ K0 in K0/K. The proposition follows by passing to the limit. 3.6 Theorem. With assumptions as in 3.1, the cokernel of the reciprocity map (3.4.1) has finite exponent, and it is a finite discrete group (i.e., Im(rec) ⊂ ¯π0π(G) is an open subgroup) if (G, X) is maximal. Before we start discussing the proof, let us give the main corollary of this result. 3.7 Corollary. With assumptions as in 3.1, consider the homomorphism φ : π1(S0) → K0, and for a prime number ℓ, let φℓ : π1(S0) → K0,ℓ be its ℓ-primary component. Fix a free Z-module H of finite rank and a closed embedding i : G ֒→ GL(H ⊗ Q), and let G ⊂ GL(H) be the Zariski closure of G in GL(H). (i) There exists a positive integer N , depending only on G and X, such that (cid:2)G (Zℓ) : Im(φℓ)(cid:3) ≤ N for all ℓ. 7 (ii) For almost all ℓ the image of φℓ contains the commutator subgroup of G (Zℓ). (iii) If (G, X) is maximal in the sense defined in 2.7, Im(φ) is an open subgroup of G(Af ). Proof. In view of Proposition 3.5, (iii) of the corollary is immediate from the second assertion in the theorem. For (ii) we only have to note that Im(φ) contains the commutator subgroup of K0 and that K0,ℓ = G (Zℓ) for almost all ℓ. It remains to deduce (i) from the theorem. With the notation of 3.4, q−1(cid:0)Im(rec)(cid:1) is a normal subgroup of G(Af ), with profinite abelian quotient G(Af )/q−1(cid:0)Im(rec)(cid:1) ∼= Coker(rec). If m is the exponent of Coker(rec), it follows that K0,ℓ/Im(φℓ) is a compact abelian ℓ-adic analytic group that is killed by m. Hence it is finite. It follows that Im(φℓ) has finite index in G (Zℓ) for all ℓ, and so it suffices to prove (i) for all ℓ sufficiently large. Choose a multiple M of m such that G is reductive over Z[1/M ] and K0,ℓ = G (Zℓ) for all ℓ > M. be the simply (Fℓ) is the normal subgroup Fix an ℓ > M, and let G0 = G ⊗ Fℓ denote the characteristic ℓ fibre of G . Let G0 → G der connected cover of the derived subgroup. The image of G0(Fℓ) → G der G0(Fℓ)+ ⊳ G0(Fℓ) that is generated by the ℓ-Sylow subgroups. 0 0 Still with ℓ > M, the image of φℓ contains the subgroup of G (Zℓ) generated by the ℓ-Sylow groups, and hence contains all elements g ∈ G (Zℓ) whose reduction modulo ℓ lies in G0(Fℓ)+. It therefore suffices to bound the m-torsion in G0(Fℓ)/G0(Fℓ)+ by a constant independent of ℓ. As a first step, let µ be the kernel of G0 → G der ; this is a group of multiplicative type whose rank µ only depends on G. (If R is the absolute rank of G, it is known that µ has rank at most 2R.) The (Fℓ)/G0(Fℓ)+ injects into H 1(Fℓ, µ), which is a quotient of µ(Fℓ) (cohomology of procyclic quotient G der 0 0 0 0 groups). In particular, (cid:2)G der G0(Fℓ)/G der is a subgroup of G ab by m on G ab most rm. (Fℓ) by a constant independent of ℓ. Writing G ab (Fℓ) : G0(Fℓ)+(cid:3) divides µ. It therefore suffices to bound the m-torsion in (Fℓ) 0 (Fℓ). Further, if r is the rank of the torus Gab, the kernel of multiplication 0 (Fℓ) has cardinality at is a finite group scheme of rank rm; hence the m-torsion in G ab , the group G0(Fℓ)/G der 0 = G0/G der 0 0 0 We now turn to the proof of Theorem 3.6. In 3.9 and 3.11 we first prove the result in two special cases; the general case is then deduced in 3.13. To prove that Coker(rec) has finite exponent (resp., is finite), we use Deligne’s group-theoretic description of the reciprocity map. (See [8], Théorème 2.6.3.) If E = E(G, X) is the reflex field, we simply write E∗ for the Q-torus ResE/Q Gm,E. Class field theory gives an isomorphism π0π(E∗) ∼ −→ Gal(Eab/E), which we normalize as in [8], 0.8. We start with a general remark that is useful to us. 3.8 Remark. Let f : G1 → G2 be a surjective homomorphism of reductive Q-groups. Factor f as G1 ։ G′ 2 ψ −→ G2 , where G′ 2 = G1/Ker(f )0. We make the simplifying assumption that Ker(ψ) (which is the finite étale that the image of ψ(A) : G′ By [15], Proposition 6.5, G1(A) → G′ group scheme π0(cid:0)Ker(f )(cid:1)) is commutative, as this is the only case we need. then H 1(cid:0)Spec(A)ét, Ker(ψ)(cid:1), and hence also the cokernel of ψ(A), is killed by M. 2(A) has open image. The commutativity of Ker(ψ) implies 2(A) → G2(A) is a normal subgroup. Further, if M is the rank of Ker(ψ) 3.9 The case of a torus. Suppose G = T is a torus, in which case X = {h} is a singleton. By definition of the reflex field E, the corresponding cocharacter µ : Gm,C → TC is defined over E; so we 8 have a homomorphism µ : Gm,E → TE. Restricting scalars to Q and composing with the norm map then gives a homomorphism of algebraic tori ν : E∗ Res(µ) −−−−→ ResE/QTE NormE/Q −−−−−−→ T . Via the class field isomorphism, the reciprocity map (3.4.1) is the inverse of the composition π0π(E∗) π0π(ν) −−−−→ π0π(T ) pr −−→ ¯π0π(T ) . We apply what was explained in 2.2, taking M = T . In this case π1(T ) is just the cocharacter group X∗(T ). With Γ = Gal(Q/Q), the assumption that T is the Mumford-Tate group of h means that Z[Γ] · µ has finite index in X∗(T ). Further, (T, {h}) is maximal if and only if X∗(T ) is generated by µ as a Z[Γ]-module. On the other hand, by definition of the reflex field E, the stabilizer of µ ∈ X∗(T ) in Γ is precisely the subgroup ΓE = Gal(Q/E) ⊂ Γ. If a ⊂ Z[Γ] is the left ideal generated by the augmentation ideal of Z[ΓE], we have X∗(E∗) ∼= Z[Γ]/a as Galois modules, and X∗(ν) : X∗(E∗) → X∗(T ) is the map given by (γ mod a) 7→ γ · µ. From the preceding remarks it therefore follows that ν, as a homomorphism of algebraic tori, is surjective, and that Ker(ν) is connected if (T, {h}) is Hodge-maximal. The assertion of 3.6 now follows from Remark 3.8. 3.10 Let f : (G1, X1) → (G2, X2) be a morphism of Shimura data. Let Ei (i = 1, 2) be the reflex field of (Gi, Xi), and denote by Eab its maximal abelian extension. Then E1 is a finite extension of E2 i in C and we have a commutative diagram Gal(Eab 1 /E1) rec(G1,X1) ¯π0π(G1) ∼= G1(Af )/G1(Q)− + (3.10.1) ¯f Gal(Eab 2 /E2) rec(G2,X2) ¯π0π(G2) ∼= G2(Af )/G2(Q)− + in which ¯f denotes the map induced by f . The image of the left vertical map is a subgroup of finite index in Gal(Eab 2 /E2). 3.11 The case when Gder is simply connected. Next we treat the case when the derived group Gder is simply connected. Let Gab = G/Gder and let p : G → Gab be the canonical map. Then hab = p ◦ h is independent of h ∈ X, and (Gab, {hab}) is a Shimura datum. By Remark 2.8(ii), if (G, X) is maximal, so is (Gab, X ab). We apply 3.10 to the morphism p : (G, X) → (Gab, {hab}). By [7], Théorème 2.4, the right vertical map in the diagram is surjective with finite kernel. (Deligne’s result says that p induces an isomorphism on the groups that we denote by π0π; the groups ¯π0π are quotients of these by finite subgroups.) The theorem for (G, X) therefore follows from the result for (Gab, X ab), which was proven in 3.9. 3.12 Lemma. Let (G, X) be a Shimura datum as in 3.1. Then there exists a Shimura datum ( G, X) and a morphism f : ( G, X) → (G, X) such that (a) the group G is the generic Mumford-Tate group on X; (b) the homomorphism f : G → G is surjective, and the induced f der : Gder → Gder is the simply connected cover of Gder. 9 Moreover, if (G, X) is maximal, we can choose ( G, X) such that it is maximal, too, and such that the kernel of f : G → G is connected. Proof. By [13], Application 3.4, there exists a morphism of Shimura data f1 : (G1, X) → (G, X) such 1 → Gder is the simply connected cover. that f1 : G1 → G is surjective, Ker(f1) is a torus, and f der Let G ⊂ G1 be the generic Mumford-Tate group on X, and let f : G → G be the restriction of f1 to G. The assumption that G is the generic Mumford-Tate group on X implies that f is surjective, and as G is normal in G1, this implies that Gder = Gder 1 . So ( G, X) and f satisfy (a) and (b). Next assume (G, X) is maximal. We claim that for any f : ( G, X) → (G, X) such that (a) and (b) hold, Ker(f ) is connected (hence a torus). Indeed, let (G2, X2) be the quotient of ( G, X) by Ker(f )0. We then have an induced morphism of Shimura data (G2, X2) → (G, X). The map G2 → G is an : Gder 1 isogeny with kernel the group scheme π0(cid:0)Ker(f )(cid:1) of connected components of Ker(f ). By maximality of (G, X) this implies that π0(cid:0)Ker(f )(cid:1) is trivial, which proves the claim. The Shimura datum ( G, X) we have obtained need not be maximal. By Remark 2.8(iii), there exists an isogeny ( G, X) → ( G, X) with ( G, X) maximal, and by 2.8(i) G is the generic Mumford-Tate group on X. As G → G is an isogeny and Gder is simply connected, we have Gder ∼ −→ Gder, and as we have already seen, the maximality of (G, X) implies that the kernel of G → G is connected. 3.13 To complete the proof of Theorem 3.6, let (G, X) be the Shimura datum considered in the assertion, and take f : ( G, X) → (G, X) as in Lemma 3.12. As shown in 3.11, the theorem is true for ( G, X). In view of diagram (3.10.1), it suffices to show that the induced map G(Af ) → G(Af ) has cokernel of finite exponent, and that the image of this map is open if (G, X) is maximal. This follows from Remark 3.8. 4. Galois-generic points 4.1 Let S be a geometrically connected scheme of finite type over a field k. Assume given a continuous representation ψ : π1(S) → G(Af ), where G is an algebraic group over Q. (Throughout, we omit the choice of a geometric base point of S from the notation.) If ℓ is a prime number, we denote by ψℓ : π1(S) → G(Qℓ) the ℓ-primary component of ψ. If y is a point of S, we have a homomorphism σy : π1(y) → π1(S), well-determined up to conjuga- tion. (Recall that π1(y) is the absolute Galois group of the residue field k(y).) 4.2 Definition. is open in the image of ψ. (i) A point y ∈ S is said to be Galois-generic with respect to ψ if the image of ψ ◦ σy (ii) A point y ∈ S is said to be ℓ-Galois-generic with respect to ψ if the image of ψℓ ◦ σy is open in the image of ψℓ. If it is clear which ψ we mean, we omit the phrase “with respect to ψ”. Clearly, if a point y is Galois-generic, it is ℓ-Galois generic for all ℓ. The following theorem by the first author and Kret (see [6], Theorem 1.1) shows that for the representation associated with a Shimura variety of abelian type, the converse holds. 4.3 Theorem. Let (G, X) be a Shimura datum of abelian type, K0 ⊂ G(Af ) a neat compact open subgroup, h0 ∈ X a base point. Let φ¯s : π1(S0) → K0 be the associated representation, as defined in 3.1. If a point y ∈ S is ℓ-Galois-generic for some prime number ℓ then y is Galois-generic. 10 5. Application to abelian varieties 5.1 For g ≥ 1, equip Z2g with the standard symplectic form, and let GSp2g be the reductive group over Z of symplectic similitudes of Z2g. Let H± g be the set of homomorphisms h : S → GSp2g,R that define a Hodge structure of type (−1, 0) + (0, −1) on Z2g for which ±2πi · ψ is a polarization. The real group GSp2g(R) acts transitively on H± g ) is a Shimura datum with reflex field Q. g , and the pair (GSp2g, H± Let K(3) ⊂ GSp2g(Z) be the subgroup of elements that reduce to the identity modulo 3. Then ShK(3)(GSp2g, H± g ) is isomorphic (over Q) to the moduli space Ag,3 of g-dimensional principally polar- ized abelian varieties with a (Jacobi) level 3 structure. See for instance [7], Section 4. In what follows we identify the two schemes. is finitely generated over Q. Assume that all 3-torsion points of B are k-rational. Let (B, λ) be a principally polarized abelian variety of dimension g over a subfield k ⊂ C that (This implies −→ Z2g. This gives a Hodge structure on Z2g; let h0 ∈ H± g be the corresponding point. As in section 3.1, [h0, eK(3)] defines a C-valued point ¯t0 ∈ Ag,3(C). The corresponding level 3 structure on (BC, λ) is defined over k, which means that ¯t0 comes from a k-valued point t0 ∈ Ag,3(k) by composing it with the given embedding k ֒→ C. that Q(ζ3) ⊂ k.) Choose a similitude i : H1(cid:0)B(C), Z(cid:1) ∼ Let A0 ⊂ Ag,3 ⊗ Q(ζ3) be the irreducible component such that t0 ∈ A0(k). This component is geometrically irreducible. The construction of 3.1 gives a representation φ¯t : π1(A0, ¯t0) → K(3) ⊂ GL2g(Z) . (Here, as in 3.1, ¯t refers to the compatible system of base points (¯tK) obtained from h0.) The point t0 gives rise to a homomorphism σt0 : Gal(¯k/k) → π1(A0, ¯t0) such that the composition with the projection is canonically defined, not only up to conjugation. π1(A0, ¯t0) → Gal(cid:0)Q/Q(ζ3)(cid:1) is the natural homomorphism Gal(¯k/k) → Gal(cid:0)Q/Q(ζ3)(cid:1). Note that σt0 On the other hand, if we let H = H1(cid:0)B(C), Z(cid:1) then we may identify the full Tate module B[n](cid:0)¯k(cid:1) (limit over all positive integers n, partially ordered by divisibility) with H ⊗ Z. Via H = lim ←−n the chosen similitude i we obtain an isomorphism H gives a representation ∼ −→ Z2g. The natural Galois action on H therefore The following result is an immediate consequence of the definitions and the modular interpretation g ). See also [24], Remark 2.8, as well as the next section, where we discuss the analogous of Sh(GSp2g, H± (but slightly more involved) case of K3 surfaces. ρB : Gal(¯k/k) → GL2g(cid:0)Z(cid:1) . 5.2 Proposition. The representations φ¯t ◦ σt0 and ρB are the same. The main result of this section is that the usual form of the Mumford-Tate conjecture implies the integral and adelic versions of the Mumford-Tate conjecture as formulated in 1.3 and 2.5. 5.3 Theorem. Let B be an abelian variety over a subfield k ⊂ C that is finitely generated over Q. Assume that for some prime number ℓ the Mumford-Tate conjecture for B is true. Then the integral and adelic Mumford-Tate conjectures for B are true as well. Note that in this case the last part of Conjecture 1.3 says that the image of ρB contains an open subgroup of Z∗ · id. This is in fact a result proven by Wintenberger [28]. (The result is stated in loc. cit. only for k a number field; this implies the same result over finitely generated fields, as we can 11 specialize B to an abelian variety over a number field in such a way that the Mumford-Tate group does not change.) Proof. In proving the theorem, we may replace the ground field k with a finite extension and B with an isogenous abelian variety. Hence we may assume that B admits a principal polarization and that all 3-torsion points of B are k-rational. This puts us in the situation of 5.1. We retain the notation and choices introduced there. Via the chosen similitude H ∼ −→ Z2g we may view the Mumford-Tate group GB of B as an algebraic subgroup of GSp2g,Q. We take its Zariski closure GB ⊂ GSp2g as integral model. (There is no need to introduce new notation for this integral form.) With h0 ∈ H± g be the GB(R)-orbit of h0. The pair (GB, X) is a Shimura datum, and by construction we have a morphism f : (GB, X) → (GSp2g, H± and, with E the reflex field of (GB, X), let Sh(f ) : ShK0(GB, X) → Ag,3 ⊗ E be the morphism induced by f . g ). Let K0 = f −1(cid:0)K(3)(cid:1), which is a neat compact open subgroup of GB(Af ), g as in 5.1, let X ⊂ H± Let ¯s0 = [h0, eK0], which is a C-valued point of ShK0(GB, X) whose image under Sh(f ) is ¯t0. As in 3.1, let S0,C ⊂ ShK0(GB, X)C be the irreducible component containing ¯s0, let F ⊂ C be its field of definition, and let S0 ⊂ ShK0 be the geometrically irreducible component thus obtained. Possibly after replacing k with a finite extension, ¯s0 comes from a point s0 ∈ S0(k), whose image in Ag,3(k) is the point t0 that corresponds to (B, λ) equipped with a suitable level 3 structure. This point s0 gives rise to a homomorphism σs0 : Gal(¯k/k) → π1(S0, ¯s0). With φ¯s : π1(S0, ¯s0) → K0 ⊂ GB(Z) the homomorphism (3.1.1), it follows from the functoriality explained in Remark 3.2 together with Proposition 5.2 that φ¯s ◦ σs0 is the Galois representation on the (full) Tate module of B. Let ℓ be a prime number and φℓ : π1(S0, ¯s0) → GB(Zℓ) the ℓ-primary component of φ. If the Mumford-Tate conjecture for B is true at ℓ, s0 is ℓ-Galois-generic with respect to the representation φ. (See the remark after 1.3.) By Theorem 4.3, s0 is then Galois-generic, and in view of the description of Im(φ) given in Corollary 3.7, this implies that Conjecture 1.3 and, if the Hodge structure H is maximal, Conjecture 2.5 are true for B. (i) There are many special classes of abelian varieties for which the Mumford-Tate 5.4 Remarks. conjecture is known. For a sample of such results, see for instance [14], Section 5, or the more recent [12] and the references contained therein. On the other hand, already for abelian varieties of dimension 4 the Mumford-Tate conjecture remains open. (ii) When we restrict our attention to the image of ρ intersected with the derived subgroup of GB, a result related to our Theorem 5.3 was obtained by Hui and Larsen; see [9], Theorem 4.2. Their work is based on a very different approach. 5.5 Example. As a first application we recover the result given in [21], 11.11. If B is a g-dimensional abelian variety with g odd (or g = 2, or g = 4) and End(B¯k) = Z, it is known that the Mumford-Tate conjecture for B is true and that the Mumford-Tate group is the full GSp2g. (See [14], Theorem 5.14, g ) is maximal; the for a more general result.) conclusion therefore is that in this case the image of the representation ρB is open in GSp2g(Af ). It is easily seen that the Shimura datum (GSp2g, H± Next we want to give examples of abelian varieties B over finitely generated subfields of C for which the Hodge structure H1(cid:0)B(C), Q(cid:1) is not Hodge-maximal. Such examples of course also give us Shimura data of Hodge type that are not maximal. The first examples we discuss are of CM type; after that, we discuss an example in which the Mumford-Tate group is the almost direct product of 12 the homotheties and a semisimple group. 5.6 Example. For our first construction, we start with a totally real field E0 of degree g over Q. Let σ1, . . . , σg be the complex embeddings of E0. Let k be an imaginary quadratic field. Then E = k · E0 is a CM field. Fix an embedding α : k → C, and let τi (i = 1, . . . , g) be the complex embedding of E that extends σi and such that τik = α. Thus, T =(cid:8)τ1, . . . , τg, ¯τ1, . . . , ¯τg(cid:9) is the set of complex embeddings of E. Consider the CM type Φ on E given by Φ = {τ1, ¯τ2, . . . , ¯τg} . The pair (E, Φ) gives rise to an isogeny class of g-dimensional complex abelian varieties B, determined by the rule that H1(cid:0)B(C), Q(cid:1) ∼= E as a Q-vector space, with Hodge decomposition of H1(cid:0)B(C), C(cid:1) ∼= ⊕τ ∈T C(τ ) given by the rule that C(τ ) is of type (−1, 0) if τ ∈ Φ and of type (0, −1) otherwise. If g > 1 then Φ is a primitive CM type; in this case B is simple. As any abelian variety of CM type, B is defined over a number field, and by a result of Pohlmann [16] the Mumford-Tate conjecture is true for B. As in Section 3, if F is a number field we simply write F ∗ for the torus ResF/Q Gm,F . Let 0 be the norm homomorphism, and let U ⊂ E∗ be the subtorus given by U = Norm : E∗ → E∗ Norm−1(Q∗). The cocharacter group X∗(E∗) is the free Z-module on the set T . The cocharacter group of U is given by X∗(U ) =( g Xi=1 aiτi + g Xi=1 bi¯τi ∈ X∗(E∗)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ai + bi is independent of i) . The elements fi = τi − ¯τi (i = 1, . . . , g) together with fg+1 =Pg The cocharacter µ : Gm,C → E∗ corresponding to the Hodge structure H1(cid:0)B(C), Q(cid:1) is given by µ = τ1 + ¯τ2 + · · · + ¯τg = f1 + fg+1. The Galois conjugates of µ are the elements fi + fg+1 for i = 1, . . . , g together with their complex conjugates f1 + · · · + fi + · · · + fg + fg+1, for i = 1, . . . , g. These are cocharacters in X∗(U ), and for g > 2 they span a submodule of index g − 2 in X∗(U ). The conclusion, i=1 ¯τi form a basis for X∗(U ). therefore, is that U is the Mumford-Tate group of B if g > 2, and that H1(cid:0)B(C), Q(cid:1) is not Hodge- maximal if g > 3. In this last case, the image of the adelic Galois representation is therefore not open in the adelic points of the Mumford-Tate group. 5.7 Example. For our final example, we consider a Shimura datum (G, X) such that (a) G is an inner form of a split group; (b) π1(G) is non-cyclic. Note that (b) holds if Gab has dimension at least 2, or if dim(Gab) = 1 and Gder is not simply- connected. (In the latter case this follows using [3], Cor. 1.7.) If (a) holds, Gal(Q/Q) acts trivially on π1(G) (see [3], Lemma 1.8), and by what was explained in 2.2 and 2.7, we conclude that any (G, X) satisfying (a) and (b) is non-maximal. To obtain a concrete example, let D be a quaternion algebra over Q that is non-split at infinity, i.e., D ⊗Q R is Hamilton’s quaternion algebra H. The canonical involution ∗ on D is then a positive involution. Let n = 2r be an even positive integer with n ≥ 6, and consider a free (left-) D-module V of rank n equipped with a (−1)-hermitian form Ψ of discriminant 1 and Witt index r. Let G′ = UD(V, Ψ) be the corresponding unitary group, which we view as an algebraic subgroup of GLQ(V ), and let G ⊂ GLQ(V ) be the algebraic group generated by G′ together with the homotheties Gm · id. The 13 group G′ ⊗ R is isomorphic to the identity component of U∗ n(H) (which in some literature is denoted by SO∗(2n)), and there is a unique G(R)-conjugacy class of homomorphisms h : S → GR that make the pair (G, X) a Shimura datum of PEL type. (Cf. [8], Section 1.3.) The corresponding Shimura variety parametrizes polarized abelian varieties of dimension 2n with an action by (an order in) D, which is of Albert Type III. As G′ is a Q-simple group, G is the generic Mumford-Tate group on X. Our assumption that Ψ has trivial discriminant implies that the index of G′ is 1Dn; see [23], Table II, pages 56–57. This means that G′ (and hence also G) is an inner form of the split form, i.e., condition (a) is satisfied. On the other hand, dim(Gab) = 1 and Gder is not simply-connected, so also (b) is satisfied. (In fact, π1(G) ∼= Z × (Z/2Z).) We conclude that (G, X) is not maximal. If B is a complex abelian variety that corresponds to a Hodge-generic point of the Shimura variety defined by (G, X), the Hodge structure H1(cid:0)B(C), Q(cid:1) is not Hodge-maximal. 6. Application to K3 surfaces 6.1 Let L be a number field, G a connected reductive group over L, and let M = ResL/Q G. Then MC ∼= Qσ∈Σ Gσ, where Σ is the set of complex embeddings of L and Gσ = G ⊗L,σ C. With Γ = Gal(Q/Q) and ΓL = Gal(Q/L), the fundamental group π1(M ) is the Γ-module obtained from the ΓL-module π1(G) by induction. Let µ be a complex cocharacter of M. As in 2.2, its conjugacy class C defines an element [C ] ∈ π1(M ). Let W ⊂ π1(M ) be the Z[Γ]-submodule generated by [C ]. Suppose there is a unique τ ∈ Σ such that the projection µτ of µ onto the factor Gτ is non-trivial. View L as a subfield of C via τ ; then µτ is a complex cocharacter of G. Its conjugacy class Cτ defines an element [Cτ ] in π1(G). Let Wτ ⊂ π1(G) be the Z[ΓL]-submodule that it generates. In this situation we have W = Z[Γ] ⊗Z[ΓL] Wτ as submodule of π1(M ) = Z[Γ] ⊗Z[ΓL] π1(G). Consequently, µ is maximal as a cocharacter of M if and only if µτ is maximal as a complex cocharacter of G. 6.2 Proposition. Let V be a polarizable Q-Hodge structure of K3 type, by which we mean that V is of type (−1, 1) + (0, 0) + (1, −1) with Hodge numbers 1–n–1 for some n. Then V is Hodge-maximal. Proof. Without loss of generality we may assume that V is simple as a Hodge structure, which in this case means that there are no non-zero Hodge classes in V . (By definition, Hodge-maximality only depends on the Mumford-Tate group M and the defining homomorphism h : S → MR; these do not change if we replace V with V ⊕ Q(0).) Let L = EndQHS(V ) be the endomorphism algebra of V as a Q-Hodge structure, and choose a polarization form ψ : V × V → Q. As shown by Zarhin in [29], L is a field which is either totally real or a CM field. First suppose L is totally real. By [25], Lemma 3.2, dimL(V ) ≥ 3. By [29], Theorem 2.2.1, M = ResL/Q SOL(V, Ψ), where Ψ : V × V → L is the unique symmetric L-bilinear form on V such that traceL/Q ◦ Ψ = ψ. In particular, M is semisimple. If dimL(V ) is odd, M is simply connected and there is nothing to prove. Next suppose dimL(V ) = 2l is even. Let Σ be the set of complex embeddings of L. Write G = SOL(V, Ψ) and let µ : Gm,C → MC ∼= Qσ∈Σ Gσ be the cocharacter that gives the Hodge structure. There is a unique τ ∈ Σ such that µτ 6= 1, so we are in the situation of 6.1. The root system of GC is of type Dl, and we follow the notation of [4], Planche IV. Note that the root system in this case is self-dual; further, the calculation that follows goes through without changes if l = 2. With respect to the basis ε1, . . . , εl for Rl = X∗(G) ⊗ R, we have X∗(G) = Zl, and the coroot lattice Q(R∨) consists of the vectors (m1, . . . , ml) ∈ Zl for which P mj is even. On the other hand, 14 the cocharacter µτ corresponds to the vector (1, 0, . . . , 0); its image in π1(G) = X∗(G)/Q(R∨) ∼= Z/2Z is therefore the non-trivial class. By what was explained in 2.2 this implies the assertion. Next suppose L is a CM field. Let L0 ⊂ L be the totally real subfield. There is a unique symmetric hermitian form Ψ : V × V → L (with respect to complex conjugation on L) such that traceL/Q ◦ Ψ = ψ, and by [29], Theorem 2.3.1, M = ResL0/Q UL(V, Ψ). Write G = UL(V, Ψ), and let Σ be the set of complex embeddings of L0. As in the totally real case there is a unique τ ∈ Σ such that the cocharacter µ is non-trivial on the factor Gτ . If n = dimL(V ), we have GC ∼= GLn in such a way that µτ is conjugate to the cocharacter Gm → GLn given by z 7→ diag(z, 1, . . . , 1). It is straightforward to check that the corresponding class in π1(GLn) ∼= Z is a generator, and again by 2.2 and 6.1 this implies the assertion. In the proposition it is essential that we work with a Hodge structure of weight 0. 6.3 Remark. As is well-known, if Y is a complex K3 surface, the Hodge structure H = H 2 general, Hodge-maximal; but H(1) = H ⊗Q(1) is. For instance, if EndQHS(H) = Q, the Mumford-Tate group of H is the group GO(H, φ) of orthogonal similitudes, where φ is a polarization form. We have a non-trivial isogeny CSpin(H, φ) → GO(H, φ), such that the homomorphism h : S → GO(H, φ)R that defines the Hodge structure on H lifts to a homomorphism h : S → CSpin(H, φ)R. Cf. [26], 2.2.3–4. By contrast, the Mumford-Tate group of H(1) is the special orthogonal group SO(H, φ). We can still lift to CSpin(H, φ), but the homomorphism CSpin(H, φ) → SO(H, φ) is not an isogeny. prim(cid:0)Y (C), Q(cid:1) is not, in 6.4 As a preparation for the main result of this section, we need to recall some facts about the moduli of polarized K3 surfaces. We closely follow Rizov [17], [18]. 8 8 Fix a natural number d. Let (L0, ψ) be the quadratic lattice U ⊕3 ⊕ E⊕2 (with U the hyperbolic lattice). With {e1, f1} the standard basis of the first copy of U, let (L2d, ψ2d) be the sublattice he1 + df1i ⊕ U ⊕2 ⊕ E⊕2 of L0. In what follows we write SO for the Z-group scheme SO(L2d, ψ2d). For n ≥ 1, let K(n) ⊂ SO(Af ) be the compact open subgroup of elements in SO(Z) that reduce to the identity modulo n. If K is an open subgroup of K(n) for some n ≥ 3, Rizov defines in [17], Section 6, a moduli stack F2d,K over Q of K3 surfaces with a primitive polarization of degree 2d and a level K structure. (In fact, Rizov does this over open parts of Spec(Z), but for our purposes it suffices to work over Q.) By [18], Cor. 2.4.3, F2d,K is a scheme. If (Y, λ) is a K3 surface over a field k of characteristic 0 equipped with a primitive polarization of degree 2d, a level K(n)-structure on (Y, λ) is an isometry H 2 −→ L2d/nL2d. (See [17], Example 5.1.3.) prim(Y¯k, Z/nZ)(cid:0)1(cid:1) ∼ The construction of 3.1 has an analogue in this setting. Let F0,C be an irreducible component 2d,K(3) ⊗ C, and let F ⊂ C be its field of definition, so that we have a geometrically irreducible of F component F0 ⊂ F 2d,K(3), which for K normal in K(3) is Galois with group K(3)/K. Let FK ⊂ F2d,K ⊗ F be the inverse image of F0. Suppose we are given a compatible collection ¯y = (¯yK) of geometric points of the FK, for K open and normal in K(3). We write ¯y0 for ¯yK(3). We then have and associated homomorphism 2d,K(3) ⊗F . For K ⊂ K(3) we have an étale morphism F K,K(3) : F2d,K → F (6.4.1) Φ¯y : π1(F0, ¯y0) → K(3) ⊂ SO(Z) . 6.5 With SO as above, let Ω± be the space of homomorphisms h : S → SOR that give L2d ⊗ Q a Hodge structure of type (−1, 1) + (0, 0) + (1, −1) with Hodge numbers 1–19–1, such that ±ψ2d is a polarization. The group SO(R) acts transitively on Ω±, and the pair (SOQ, Ω±) is a Shimura datum of abelian type with reflex field Q. On of the main results of [18] (loc. cit., Thm. 3.9.1) is that for an open subgroup K ⊂ K(3) there 15 is an étale morphism of Q-schemes (6.5.1) jK : F2d,K → ShK(SOQ, Ω±) in such a way that for K2 ⊂ K1 the diagram (6.5.2) F2d,K2 FK2,K1 F2d,K1 jK2 jK1 ShK2(SOQ, Ω±) ShK2,K1 ShK1(SOQ, Ω±) is cartesian. The image of jK is the complement of a divisor (ibid., 3.10(B)). 6.6 Theorem. Let Y be a K3 surface over a subfield k ⊂ C that is finitely generated over Q. Let H = H 2(cid:0)Y (C), Z(cid:1)(1), and let GB ⊂ GL(H) be the Mumford-Tate group. Let ρY : Gal(¯k/k) → GL(H)(cid:0)Z(cid:1) be the Galois representation on H = H 2(cid:0)Y¯k, Z(cid:1)(1), which we identify with H ⊗ Z via the comparison isomorphism between singular and étale cohomology. Then the image of ρY has a subgroup of finite index which is an open subgroup of GB(Z). Proof. The proof is very similar to that of Theorem 5.3. The main difference is that for K3 surfaces the Mumford-Tate conjecture is known, due to results of Tankeev [22] and André [1], and that by Proposition 6.2 the Hodge structure on H is always Hodge-maximal. We retain the notation introduced in 6.4 and 6.5. Choose a primitive polarization λ on Y , say of degree 2d. Further choose an isometry i : H −→ L2d. These choices give us a compatible system ¯y = (¯yK ) of points ¯yK ∈ F2d,K (C), where K runs through the set of open subgroups of K(3). Possibly after replacing k with a finite extension in C, we may assume that k = kconn and that ¯y0 = ¯yK(3) comes from a k-valued point y0 ∈ F 2d,K(3)(k) by composing it with the embedding k ֒→ C. Of course, y0 is just the moduli point of (Y, λ) equipped with a suitable level 3 structure. ∼ Via the chosen isometry i the Hodge structure on HQ defines a point h0 ∈ Ω±. Let ¯t = (¯tK) be the system of C-valued points [h0, eK] of ShK (SO, Ω±), and abbreviate ¯tK(3) to ¯t0. The construction of the period map (6.5.1) is such that jK (¯yK) = ¯tK for all K ⊂ K(3). Let t0 = jK(3)(y0), which is a k-valued point of ShK(3)(SO, Ω±). Let F0 ⊂ F 2d,K(3) ⊗ k and S0 ⊂ ShK(3)(SO, Ω±) ⊗ k be the irreducible components containing ¯y0 and ¯t0, respectively; as they are smooth over k and have a k-rational point, these components are geometrically irreducible. By construction, j0 = jK(3) restricts to an étale morphism j0 : F0 → S0 over k. Consider the homomorphism φ¯t : π1(S0, ¯t0) → K(3) as in 3.1. We also have the homomorphism Φ¯y : π1(F0, ¯t0) → K(3) of (6.4.1). (In both cases we have now extended the base field to k.) The fact that the diagrams (6.5.2) are Cartesian implies that Φ¯y = φ¯t ◦ j0,∗. Let Hprim ⊂ H be the primitive integral cohomology, and identify the primitive étale cohomology with Z-coefficients Hprim ⊂ H with Hprim ⊗ Z. Via the chosen isometry i, the Galois action on Hprim is a representation ρY,prim : Gal(¯k/k) → SO(Z). Note that the Galois action ρY on H leaves Hprim stable and is trivial on the complement; hence the image of ρY is the same as the image of ρY,prim. On the other hand, the k-rational point y0 gives rise to a section σy0 of the homomorphism π1(F0, ¯y0) → Gal(¯k/k) induced by the structural morphism F0 → Spec(k). The composition Φ¯y ◦ σy0 : Gal(¯k/k) → K(3) ⊂ SO(Z) is the same as ρY,prim. The composition j0,∗ ◦ σy0 : Gal(¯k/k) → π1(S0, ¯t0) is the section σt0 given by the point t0 ∈ S0(k). It follows that φ¯t ◦ σt0 : Gal(¯k/k) → SO(Z) is the same as ρY,prim. 16 The rest is the same as in the proof of Theorem 5.3. Let GB ⊂ GSp2g be the Mumford-Tate group, and let X ⊂ Ω± be the GB(R)-orbit of h0. The pair (GB, X) is a Shimura datum and we Sh(f ) : ShK0(GB, X) → ShK(3)(SO, Ω±) ⊗ E be the morphism induced by f . We have a compatible have a morphism f : (GB, X) → (SO, Ω±). Let K0 = f −1(cid:0)K(3)(cid:1), let E be the reflex field, and let system ¯s of points ¯sK = [h0, eK] ∈ ShK (GB, X)(cid:0)C(cid:1) with Sh(f )(cid:0)¯s(cid:1) = ¯t. The point ¯s0 = ¯sK0 comes from a k-valued point s0, and if S0 ⊂ ShK0(GB, X)k is the irreducible component in which it lies, s0 gives a section σs0 : Gal(¯k/k) → π1(S0, ¯s0). Finally, if φ¯s : π1(S0, ¯s0) → K0 ⊂ GB(Z) is the repre- sentation (3.1.1), it follows from the functoriality explained in Remark 3.2 that φ¯s ◦ σs0 is the same as ρY,prim. By the Mumford-Tate conjecture, s0 is ℓ-Galois-generic with respect to φ¯s for every ℓ. By Theo- rem 4.3 it follows that s0 is Galois-generic, and by Corollary 3.7(iii), the theorem follows. References [1] Y. André, On the Shafarevich and Tate conjectures for hyper-Kähler varieties. Math. Ann. 305 (1996), no. 2, 205–248. [2] F. Bogomolov, Sur l’algébricité des représentations ℓ-adiques. C. R. Acad. Sci. Paris Sér. A-B 290 (1980), no. 15, A701–A703. [3] M. Borovoi, Abelian Galois cohomology of reductive groups. Mem. Amer. Math. Soc. 132 (1998), no. 626. [4] N. Bourbaki, Éléments de mathématique. Groupes et Algèbres de Lie, Chap. 4, 5 et 6. Masson, 1981. [5] A. Cadoret, An adelic open image theorem for Abelian schemes. To appear in Int. Math. Res. Not. doi:10.1093/imrn/rnu259 [6] A. Cadoret, A. Kret, Galois-generic points on connected Shimura varieties – I, Adelic versus ℓ-adic. Preprint, 2015. [7] P. Deligne, Travaux de Shimura. Sém. Bourbaki (1970/71), Exp. No. 389. In: Lecture Notes in Math. 244, Springer, Berlin, 1971; pp. 123–165. [8] P. Deligne, Variétés de Shimura: interprétation modulaire, et techniques de construction de modèles canon- iques. In: Automorphic forms, representations and L-functions (Proc. Symp. Pure Math. 33, Part 2), pp. 247–289. A.M.S., 1979. [9] C. Hui, M. Larsen, Adelic openness without the Mumford-Tate conjecture. Preprint, arXiv:1312.3812. [10] M. Larsen, R. Pink, On ℓ-independence of algebraic monodromy groups in compatible systems of represen- tations. Invent. Math. 107 (1992), no. 3, 603–636. [11] M. Larsen, R. Pink, Abelian varieties, ℓ-adic representations, and ℓ-independence. Math. Ann. 302 (1995), no. 3, 561–579. [12] D. Lombardo, On the ℓ-adic Galois representations attached to nonsimple abelian varieties. Preprint, arXiv:1402.1478 [13] J. Milne, K.-y. Shih, Conjugates of Shimura varieties. In: Hodge cycles, motives, and Shimura varieties. Lecture Notes in Math. 900, Springer, 1982; pp. 280–356. [14] R. Pink, ℓ-adic algebraic monodromy groups, cocharacters, and the Mumford-Tate conjecture. J. reine angew. Math. 495 (1998), 187–237. [15] V. Platonov, A. Rapinchuk, Algebraic groups and number theory. Pure and Applied Mathematics 139. Academic Press, Inc., Boston, MA, 1994. [16] H. Pohlmann, Algebraic cycles on abelian varieties of complex multiplication type. Ann. of Math. (2) 88 (1968), 161–180. [17] J. Rizov, Moduli stacks of polarized K3 surfaces in mixed characteristic. Serdica Math. J. 32 (2006), no. 2–3, 131–178. [18] J. Rizov, Kuga-Satake abelian varieties of K3 surfaces in mixed characteristic. J. reine angew. Math. 648 (2010), 13–67. 17 [19] J.-P. Serre, Représentations ℓ-adiques. In: Algebraic number theory (Kyoto Internat. Sympos., Res. Inst. Math. Sci., Univ. Kyoto, Kyoto, 1976), pp. 177–193. Japan Soc. Promotion Sci., Tokyo, 1977. [=OEu- vres 112.] [20] J.-P. Serre, Lettre à Ken Ribet du 1/1/1981. In: OEuvres, Volume IV, number 133. [21] J.-P. Serre, Propriétés conjecturales des groupes de Galois motiviques et des représentations ℓ-adiques. In: Motives (Proc. Symp. Pure Math. 55, Part 1), pp. 377–400. A.M.S., 1994. [=OEuvres 161.] [22] S. Tankeev, Surfaces of K3 type over number fields and the Mumford-Tate conjecture. I, II. Math. USSR- Izv. 37 (1991), no. 1, 191–208, and Izv. Math. 59 (1995), no. 3, 619–646. [23] J. Tits, Classification of algebraic semisimple groups. In: Algebraic Groups and Discontinuous Subgroups (Proc. Sympos. Pure Math. 9), pp. 33–62. A.M.S., 1966. [24] E. Ullmo, A. Yafaev, Mumford-Tate and generalised Shafarevich conjectures. Ann. Math. Qué. 37 (2013), no. 2, 255–284. [25] B. van Geemen, Real multiplication on K3 surfaces and Kuga-Satake varieties. Michigan Math. J. 56 (2008), no. 2, 375–399. [26] J.-P. Wintenberger, Relèvement selon une isogénie de systèmes ℓ-adiques de représentations galoisiennes associés aux motifs. Invent. Math. 120 (1995), no. 2, 215–240. [27] J.-P. Wintenberger, Une extension de la théorie de la multiplication complexe. J. reine angew. Math. 552 (2002), 1–14. [28] J.-P. Wintenberger, Démonstration d’une conjecture de Lang dans des cas particuliers. J. reine angew. Math. 553 (2002), 1–16. [29] Yu. Zarhin, Hodge groups of K3 surfaces. J. reine angew. Math. 341 (1983), 193–220. [email protected] Centre de Mathématiques Laurent Schwartz – École Polytechnique, 91128 Palaiseau, France [email protected] Radboud University Nijmegen, IMAPP, Nijmegen, The Netherlands 18
1609.00910	1	1609	2016-09-04T09:45:10	Endomorphisms of projective bundles over a certain class of varieties	[ "math.AG" ]	Let $B$ be a simply-connected projective variety such that the first cohomology groups of all line bundles on $B$ are zero. Let $E$ be a vector bundle over $B$ and $X={\mathbb P} (E)$. It is easily seen that a power of any endomorphism of $X$ takes fibers to fibers. We prove that if $X$ admits an endomorphism which is of degree greater than one on the fibers then $E$ splits into a direct sum of line bundles.	math.AG	math	Endomorphisms of projective bundles over a certain class of varieties Ekaterina Amerik, Alexandra Kuznetsova September 24, 2018 Introduction During the last 20 years, the question which smooth projective varieties have endomorphisms of degree greater than one (which we shall sometimes simply call "endomorphisms", as opposed to automorphisms) has attracted some attention for both geometric and dynamical reasons (see e.g. [ARV], [Bea], [N], [NZ] - this is only a beginning of the list). Though in this generality it is still far from being solved, there is a number of partial results suggesting that varieties with such endomorphisms generally come from two obvious cases (tori and toric varieties) by means of simple geometric constructions such as taking a product with another smooth projective variety or taking a quotient by a finite freely acting group. For instance, Nakayama proved in the beginning of 2000's that a rational smooth projective surface with endomorphisms must be toric. Around the same time, one of the authors of the present note has considered the case of a projective bundle X over a projective base B, p : X → B, and proved that if X has an endomorphism commuting with the projection onto the base, then X must be a quotient of a product B′ × Pr by a finite freely acting group. A simple remark on endomorphisms of projective bundles X = P(E), where E is a vector bundle ([A], p. 18) is that a power of any f : X → X sends fibers to fibers and thus must be over an endomorphism of the base; so if by any chance we know that all endomorphisms of B are of finite order - for instance when B is of general type - then this result describes the situation completely. The argument (the "only if" part, the "if" part being rather standard) pro- ceeds as follows. One considers the space of all morphisms Rm(P(V )) from Pn = P(V ) to itself given by degree m polynomials (well-known to be an affine variety) and its quotient M by P GL(V ) (that is, the spectrum of the ring of the invariants). It turns out that for m big enough, P GL(V ) acts with finite stabilizers, so M is the geometric quotient (i.e. actually parameterizes the orbits of the action). Now let X = P(E) be a projective bundle over B. An endomor- phism f of X over B naturally induces a morphism from B to M . Its image must be a point since B is projective and M is affine. Let t be a lift of this point to Rm(P(V )). Over a suitably fine open covering (Uα) of B we have fα = hα · t, where hα is in P GLn+1(OUα ). Denote by gαβ the transition functions of our 1 projective bundle, it follows that h−1 α gαβhβ ∈ Stab(t), in other words, by chang- ing the trivialization we make the transition functons of X constant with values in a finite group, q. e. d.. In general, for an endomorphism f of P(E) we may suppose that f is over an endomorphism g of the base; there are then two cases to be treated: the case where f induces isomorphisms of fibers (considered as exceptional; when X = P(E) it means that g∗E is a shift of E by a line bundle) and the case where the degree of f on the fibers is greater than one. In [A], only the rank-two case (that of projective line bundles) was considered. It was established that either X is a finite quotient of a product or E has a subbundle. This last statement has been pursued further to yield that E must split into a direct sum of line bundles after a finite, not necessarily ´etale, base change ([A], theorem 2); from a different point of view, one can restrict to a specific class of bases to obtain a stronger statement. For instance, if B satisfies the condition H 1(B, L) = 0 for any line bundle L, then having a subbundle is equivalent to splitting for rank-two bundles. It therefore follows from the results of [A] that if B is simply connected and H 1(B, L) = 0 for any line bundle L on B, then an X with endomorphisms of degree greater than one on fibers must be the projectivization of a split rank-two bundle. The purpose of the present note is to prove this result in the case of arbitrary rank projective bundles over such specific bases. Theorem 1. Let B be a simply-connected projective variety such that for any line bundle L its first cohomology H 1(B, L) = 0. Let E be a vector bundle of rank n + 1 on B. If there exists a fiberwise endomorphism P(E) π B φ Φ P(E) π / B (1) of degree greater than one on the fibers, then E splits into a direct sum of line bundles: E = Li (2) n+1 ⊕ i=0 What we show is in fact slightly more general, as in [A]. Theorem 2. Let B be as in the previous theorem and E and F vector bundles of rank n + 1 on B. If there exists a morphism φ : P(E) → P(F ) (3) over B which is of degree greater than one, then E and F both split into a direct sum of line bundles. Obviously, theorem 1 follows from this statement: consider the endomor- phism φ as a morphism P(E) → P(Φ∗(E)), and apply the theorem 2. 2 / / / In the ideal situation, one would like to prove the statement of Theorem 1 for an arbitrary toric base B. The reason is that the projectivization of a vector bundle over a toric base is itself toric if and only if the bundle is split ([D]). This would therefore strongly support the principle that varieties with endomorphisms are closely related to toric varieties or tori. However few toric bases (e.g. Pn, n ≥ 2, or products of such) actually satisfy the cohomology vanishing condition as above; so more work is needed to obtain such a result. It is certainly related to the fact that we never make use of a condition like F = g∗E in Theorem 2. 1 Reduction to invariant theory Let V and W be vector spaces of dimension n + 1. Denote by Rm(P(V ), P(W )) the set of all morphisms between P(V ) and P(W ) given by homogeneous poly- nomials of degree m without a common zero except at (0, 0, . . . , 0): y0 = f0(x0,x1, . . . , xn) y1 = f1(x0,x1, . . . , xn) · · · yn = fn(x0,x1, . . . , xn) (4) This is an affine variety, indeed the complement to the hypersurface defined by the resultant of the fi in the projective space P(Hom(W ∗, SmV ∗)), with the action of P GL(V ) × P GL(W ) given by ((g, h) · f )(x) = h−1(f (g(x))). (5) The quotient M of Rm(P(V ), P(W )) by this action (i.e. the spectrum of the ring of invariants), in contrast with the case of the action of P GL(V ) when V = W ([A]) is not a geometric quotient: indeed some points have infinite stabilizers, and all the adherent orbits give the same point on the quotient. Let us denote by M0 the "bad subset" of M (by definition it consists of points corresponding to orbits not separated by the invariants). When some fiber of a vector bundle E over B is identified with V and that of F with W , a morphism of projective bundles P(E) → P(F ) over a base B gives, in the same way as in [A], a map from B to M , which must be constant as soon as B is projective. If the image point is not in M0 we conclude as before that P(E) and P(F ) trivialize after a finite unramified base change. If B is simply-connected, this yields that these are already trivial on B, and in particular they split into a direct sum of line bundles. So the interesting case is when the image point lands in M0. In this situation, we strive to deduce some informaion about the geometry of our morphism. We aim to show that E and F have subbundles E′ and F ′ such that the inverse image of P(F ′) is P(E′) and that the map f induces a morphism on the quotients. This shall enable us to 3 conclude by induction in the case when the cohomological condition on B is satisfied. Let us also remark that replacing our original endomorphism φ of P(E) by a power, we may assume that m is greater than the rank n + 1 of the vector bundles E and F , as we shall for the computations in the next section. 2 Unstable morphisms In this section we consider two vector spaces V and W of dimension n + 1 and a morphism f between their projectivisations of degree d = mn. First of all assume f is stabilized by an infinite subgroup Stab(f ) in P GL(V ) × P GL(W ). Recall from [A]: Lemma 1 ([A], Lemma 1.2). If m > n + 1, then a unipotent element u of P GL(V ) × P GL(W ) does not stabilize any element of Rm(P(V ), P(W )). By this lemma the subgroup Stab(f ) ⊂ P GL(V ) × P GL(W ) consists of semisimple elements. Take any of these elements and consider the minimal subgroup in the stabilizer that contains this element. The connected component of the unity of this subgroup is an algebraic torus or trivial. If it is trivial for any element in Stab(f ), then the stabilizer is discrete and therefore is finite. If Stab(f ) is infinite, it contains a subgroup isomorphic to Gm. Lifting its action on P(V ) and P(W ) to an action on V and W we assume that it is given by gb,c : Gm → GL(V ) × GL(W ) gb,c(λ) = (diag(λc0 , λc1, . . . , λcn ); diag(λb0 , λb1 , . . . , λbn )) (6) in appropriate coordinates on V and W . In these coordinates, let the morphism f be given by (f0, f1, . . . , fn) with y0 = f0(x0, . . . , xn) = X I=m y1 = f1(x0, . . . , xn) = X I=m a0,I xI a1,I xI · · · yn = fn(x0, . . . , xn) = X I=m an,I xI (7) Here I = (i0, i1, . . . , in) is a multiindex and I = i0 + i1 + · · · + in. Applying an element of the diagonal group in gb,c(λ) ∈ Stab(f ), we get the following formulae for gb,c · f : yj = X λhc,Ii−bj aj,I xI (8) 4 Here h−, −i denotes the scalar product between multiindexes: F (I) := hc, Ii = n X j=0 cjij (9) Since gb,c stabilizes f there exists a constant C, such that for any j, I with aj,I 6= 0 hc, Ii − bj = C (10) Consider the n + 1) -- dimensional lattice Λ ∼= Zn+1 ⊂ Rn+1 = Λ ⊗ R. Denote by pi ∈ Λ the vertex corresponding to the i−th base vector (0, . . . , 0, m, 0, . . . , 0). For any subset {pi1, . . . , pik } denote by ∆(pi1 , . . . , pik ) ⊂ Λ the simplex of di- mension k − 1 with vertexes pi1 , . . . , pik . Set ∆ = ∆(p0, p1, . . . , pn) ⊂ Λ (11) i2 p2 = (0, 0, m) ∆ p0 = (m, 0, 0) i0 p1 = (0, m, 0) i1 Figure 1: The simplex ∆ in the case n = 2 Equations (10) define n + 1 hyperplanes in Rn (not necessarily distinct). Let us denote them by Πj . Now let us consider the Newton polyhedron of fj: N P (fj) := Conv{I ∈ Λ aj,I 6= 0} (12) and prove some easy facts about Newton polyhedra of the morphism f . Proposition 1. If f has infinite stabilizer then N P (fj) ⊂ Πj ∩ ∆. 5 Proof. As the degree of fj equals m, the polyhedron N P (fj) lies in the simplex ∆. By the previous calculation we see that if gb,c stabilizes f , then (10) holds and consequently the multi-indices of the monomials of fj lies in the hyperplane Πj. Lemma 2. If f is a morphism of projective spaces then every vertex of ∆ is contained in one of the hyperplanes Πj. Proof. Assume the vertex p0 = (m, 0, . . . , 0) does not lie in any Πj. Conse- quently no polynomial fj contains the monomial xm 0 . Then all fj vanish at the point (1 : 0 : · · · : 0) ∈ P(V ), so f is not a morphism. Lemma 3. Each hyperplane Πj contains some vertex of ∆. Moreover a hy- perplane repeated exactly k + 1 times (i.e. corresponding to the polynomials f0, . . . , fk, up to renumbering) contains exactly k + 1 vertices of ∆. Proof. Since all the hyperplanes Πj are parallel, if they contain a common vertex they coincide. There is a natural partition of the set H(f ) of equations H1 ⊔ H2 ⊔ · · · ⊔ Hl = H(f ) (13) where a subset Hi consists of equations corresponding to the same hyperplane Πi, as well as of the set of vertices V (∆) = V1 ⊔ V2 ⊔ · · · ⊔ Vl (14) where Vi consists of vertices lying in Πj. Since V (∆) = n + 1 = H(f ) it follows that either the statement of the lemma is true or k + 1 = Vi > Hi = s + 1 for some i. Assume Vi > Hi. The polynomials fi indexed by Hi contain monomi- als depending only on the variables indexed by Vi, but the others do not: up to renumbering, fs+1, . . . , fn are zero as soon as xk+1 = . . . xn = 0. Then f0, . . . , fs define a regular map of the subspace of P(V ) given by the vanishing of xk+1, . . . xn to the subspace of P(W ) given by the vanishing of ys+1, . . . , yn, but this is impossible since the dimension of the source would then be greater than that of the target. From these assertions we deduce the following statement. Proposition 2. Let f be a morphism between P(V ) and P(W ) with infinite stabilizer in P GL(V ) × P GL(W ). There exist V ′ ⊂ V (∆) and H ′ ⊂ H(f ) such that V ′ = H ′ < n + 1 and N P (fj) ⊂ ∆(V ′) (15) for any fj ∈ H ′. 6 Π0 Π1 Π2 Π0 = Π1 Π2 Figure 2: Two types of Newton polyhedra of f0, f1 and f2 in the case n = 2. Proof. Let us recall the function F from (9). Denote M ′ = max{F (pi)}, where pi runs through the set of vertices of ∆. Set H ′ = {fj F Πj = M ′} (16) As F is not constant on ∆, ∅ ( H ′ ( H(f ). Denote by V ′ the set of vertices on the hyperplane corresponding to the equations in H ′. By the previous lemma V ′ = H ′. Obviously, Πj ∩ ∆ = ∆(V ′) and so the polynomials fj ∈ H ′ depend only on the variables corresponding to the vertices in V ′. So far, we have discussed the morphisms of projective spaces with infinite stabilizer in P GL(V ) × P GL(W ). But our goal is to study the morphisms f with non-closed orbits under the group action. By a generalization of the Hilbert -- Mumford criterion ([Bir] Theorem 4.2), we reach the boundary of the orbit (P GL(V ) × P GL(W )) · f while acting on f by one-parameter subgroups gb,c(Gm) as in (6). As earlier, the map gb,c(λ) · f is given by the equations (8). Let us introduce a new notation Kj = min{Iaj,I 6=0}{hc, Ii − bj} (17) Set K = minj{Kj}. Then we can describe the limit of gb,c(λ) · f when λ goes to zero. Lemma 4. Denote ¯f = limλ→0(gb,c(λ) · f ), then ¯fj(x0, . . . , xn) = X aj,I xI hc,Ii−bj=K and the original map was of type: fj = X aj,I xI + X aj,I xI hc,Ii−b0=K hc,Ii−b0>K (18) (19) 7 The proof is a straightforward calculation. Obviously, the group Gb,c stabilize the morphism ¯f , so ¯f has infinite stabi- lizer and in Proposition 2 we have a description of its Newton polyhedron. Now consider the set of half-spaces Π+ j = {I ∈ Λ ⊗ R hc, Ii − bj ≥ K ⊂ Rn} (20) Lemma 4 implies that N P (fj) = Π+ j ∩ ∆. From the proof of Proposition 2 we see that there is always a hyperplane Πj intersecting our simplex ∆ by a face and such that the rest of the simplex is below Πj. Thus the following holds. Proposition 3. If f is an unstable morphism between P(V ) and P(W ), then there are nonempty sets V ′ ⊂ V (∆) and H ′ ⊂ H(f ) such that V ′ = H ′ < n+1 and N P (fj) ⊂ ∆(V ′) (21) for any fj ∈ H ′. Π0 Π1 Π2 Π0 Π1 Π2 Π0 Π1 Π2 N P (f0) N P (f1) N P (f2) Figure 3: Here are Newton polyhedra of fi in the case n = 2. On each picture we highlight N P (fi) with a blue colour. Proof. Actually, consider the set H ′ from the previous lemma. As for any Πj ∈ H ′, the restriction of function F to Πj equals M ′ = max{F (pi)}, then ∆ ⊂ {I F (I) ≤ M ′} (22) Therefore for any Πj ∈ H ′ the half-space Π+ j also intersects ∆ by ∆(V ′). In the language of equations this means that the first s + 1 equations depend only on the first s + 1 variables. 3 Proof of the theorem From the previous section we deduce a useful corollary about morphisms be- tween projective bundles: 8 Corollary 1. Assume φ : P(E) → P(F ) is a morphism over the base B of degree d > 1, such that its restriction to a fiber corresponds to an unstable orbit in Rm(P(V ), P(W )). Then there are subbundles E0 ( E and F0 ( F , such that φ−1(P(F0)) = P(E0) (23) and 0 < rk(E0) = rk(F0) < rk(E) = rk(F ). Proof. By the results in the previous section, in any fiber of P(F ) there are coordinates in which for any 0 ≤ j ≤ s yj = fj(x0, . . . , xs) (24) We claim that the preimage of the subspace H = {y0 = · · · = ys = 0} is the subspace {x0 = · · · = xs = 0}. Indeed the last subspace is certainly contained in the preimage of the first one. If there is another point P = (p0 : . . . ps : ps+1 : · · · : pn) in that preimage, consider the projective subspace generated by P and the last n − s base vectors: its dimension is n − s, so it must have nonempty intersection with the subvariety given by the equations fs+1(x0, . . . , xn) = · · · = fn(x0, . . . , xn) = 0 (25) which has dimension at least s. Any point in this intersection must be an indeterminacy point of f , a contradiction. These subspaces H fit together in a subbundle F0 ( F . The same happen to their preimages, giving a subbundle φ−1(F0) = E0 ( E. To complete the proof of the theorem let us consider a linear mapping in- duced by the morphism φ: φ∗ : F ∗ → SmE∗ (26) As we have shown we have subbundles E0 and F0, such that the following diagram commutes: 0 0 / (F/F0)∗ (φ/φ0)∗ F ∗ φ∗ / F ∗ 0 φ∗ 0 / (SmE/SmE0)∗ / SmE∗ / SmE∗ 0 / 0 / 0 Consider the bundle (SmE/SmE0)∗ and write (SmE/SmE0)∗ ∼= ⊕m−1 i=0 SiE∗ 0 ⊗ Sm−i(E/E0)∗ In particular there is a projection (SmE/SmE0)∗ pr0−−→ Sm(E/E0)∗ (27) (28) (29) and pr0 ◦ (φ/φ0)∗ = ψ∗ induces a map between projective bundles P(E/E0) and P(F/F0) given by degree m polynomials. In fact this map is regular, that 9 / / / / / / / / / is, a morphism. To check this one observes that one may view (x0 : · · · : xs) and (y0 : · · · : ys) from corollary 1 as coordinates on the projectivization of the quotients, and the map of these projectivizations is then given by f0, . . . fs. To say that this map has no indeterminacy point (p0 : · · · : ps) is the same as to say that the preimage of P(F0) from corollary 1 contains nothing but P(E0). Proof of the Theorem 2. We argue by induction on n + 1 = rkE. If rkE = 1 then E is already linear, so the base of induction is trivial. Suppose now, that for all ranks less then n + 1 the statement is true. The restriction of the morphism φ to a fiber gives us an element in Rm(P(V ), P(W ))/ (P GL(V ) × P GL(W )). If this element corresponds to a stable orbit in Rm(P(V ), P(W )), then the argument in the proof of theorem 1 in [A] proves that after a finite ´etale base change both P(E) and P(F ) trivialize. As the variety B is simply-connected, there are no nontrivial ´etale base changes, so both P(E) and P(F ) are trivial and hence split. If we get an unstable orbit, then by corollary 1 the bundles E and F sit in short exact sequences: 0 → E0 → E → E/E0 → 0 0 → F0 → F → F/F0 → 0 (30) and there are morphisms given by polynomials of the same degree m > 1 be- tween the projectivisations of bundles E0, F0, E/E0 and F/F0, namely φ0 : P(E0) → P(F0) ψ : P(E/E0) → P(F/F0) (31) By the inductive assumption all these bundles must split into direct sums of line bundles. Since for any line bundle L on B, its first cohomology H 1(B, L) = 0, we see that Ext1(E/E0, E0) = Ext1(F/F0, F0) = 0 (32) So the extensions are trivial too. Consequently E and F split into a direct sum of line bundles. Acknowledgements: This paper has been prepared within the framework of a subsidy granted to the NRU HSE Laboratory of Algebraic Geometry by the Government of the Russian Federation for the implementation of the Global Competitiveness Program. The first-named author was partially supported by the Young Russian Mathematics award. References [A] Amerik, E.: On endomorphisms of projective bundles, Manuscripta Math. 111 (2003), no. 1, 17 -- 28 (2003) 10 [ARV] Amerik, E.; Rovinsky, M.; Van de Ven, A. A boundedness theorem for morphisms between threefolds. Ann. Inst. Fourier (Grenoble) 49 (1999), no. 2, 405 -- 415. [Bea] Beauville, A. Endomorphisms of hypersurfaces and other manifolds. In- ternat. Math. Res. Notices 2001, no. 1, 53 -- 58. [Bir] Birkes, D.: Orbits of linear algebraic groups. Annals of Math. 93 (1971), 459 -- 475 [D] Druel, S. Structures de contact sur les vari´et´es toriques. Math. Ann. 313 (1999), no. 3, 429 -- 435. [N] Nakayama, N. Ruled surfaces with non-trivial surjective endomorphisms. Kyushu J. Math. 56 (2002), no. 2, 433 -- 446. [NZ] Nakayama, Noboru; Zhang, De-Qi Building blocks of ´etale endomorphisms of complex projective manifolds. Proc. Lond. Math. Soc. (3) 99 (2009), no. 3, 725 -- 756 National Research University Higher School of Economics Laboratory of Algebraic Geometry and Applications 6 Usacheva str., Moscow, Russia 11
0712.2641	3	0712	2011-08-30T17:04:04	Chern class formulas for $G_2$ Schubert loci	[ "math.AG" ]	We define degeneracy loci for vector bundles with structure group $G_2$, and give formulas for their cohomology (or Chow) classes in terms of the Chern classes of the bundles involved. When the base is a point, such formulas are part of the theory for rational homogeneous spaces developed by Bernstein-Gelfand-Gelfand and Demazure. This has been extended to the setting of general algebraic geometry by Giambelli-Thom-Porteous, Kempf-Laksov, and Fulton in classical types; the present work carries out the analogous program in type $G_2$. We include explicit descriptions of the $G_2$ flag variety and its Schubert varieties, and several computations, including one that answers a question of W. Graham. In appendices, we collect some facts from representation theory and compute the Chow rings of quadric bundles, clarifying a previous computation of Edidin and Graham.	math.AG	math	CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI DAVE ANDERSON Abstract. We define degeneracy loci for vector bundles with structure group G2, and give formulas for their cohomology (or Chow) classes in terms of the Chern classes of the bundles involved. When the base is a point, such formulas are part of the theory for rational homogeneous spaces devel- oped by Bernstein -- Gelfand -- Gelfand and Demazure. This has been extended to the setting of general algebraic geometry by Giambelli -- Thom -- Porteous, Kempf -- Laksov, and Fulton in classical types; the present work carries out the analogous program in type G2. We include explicit descriptions of the G2 flag variety and its Schubert varieties, and several computations, including one that answers a question of W. Graham. In appendices, we collect some facts from representation theory and com- pute the Chow rings of quadric bundles, correcting an error in [Ed-Gr]. Contents Introduction 1. 2. Overview 3. Octonions and compatible forms 4. Topology of G2 flags 5. Cohomology of flag bundles 6. Divided difference operators and Chern class formulas 7. Variations Appendix A. Lie theory Appendix B. References Integral Chow rings of quadric bundles 1 4 9 15 18 21 23 27 32 34 1. Introduction Let V be an n-dimensional vector space. The flag variety F l(V ) parametrizes all complete flags in V , i.e., saturated chains of subspaces E• = (E1 ⊂ E2 ⊂ · · · ⊂ En = V ) (with dim Ei = i). Fixing a flag F• allows one to define Schubert varieties in F l(V ) as the loci of flags satisfying certain incidence conditions with F•; there is one such Schubert variety for each permutation of {1, . . . , n}. This Date: February 1, 2010. Key words and phrases. degeneracy locus, equivariant cohomology, flag variety, Schubert variety, Schubert polynomial, exceptional Lie group, octonions. This work was partially supported by NSF Grants DMS-0502170 and DMS-0902967. 1 2 DAVE ANDERSON generalizes naturally to the case where V is a vector bundle and F• is a flag of subbundles. Here one has a flag bundle Fl(V ) over the base variety, whose fibers are flag varieties, with Schubert loci defined similarly by incidence conditions. Formulas for the cohomology classes of these Schubert loci, as polynomials in the Chern classes of the bundles involved, include the classical Thom -- Porteous -- Giambelli and Kempf -- Laksov formulas (see [Fu1]). The above situation is "type A," in the sense that F l(V ) is isomorphic to the homogeneous space SLn/B (with B the subgroup of upper-triangular matrices). There are straightforward generalizations to the other classical types (B, C, D): here the vector bundle V is equipped with a symplectic or nondegenerate symmetric bilinear form, and the flags are required to be isotropic with respect to the given form. Schubert loci are defined as before, with one for each element of the corresponding Weyl group. The problem of finding formulas for their cohomology classes has been studied by Harris -- Tu [Ha-Tu], J´ozefiak -- Lascoux -- Pragacz [J´o-La-Pr], and Fulton [Fu2, Fu3], among others. One is naturally led to consider the analogous problem in the five remain- In exceptional types, however, it is not so obvious how the ing Lie types. Lie-theoretic geometry of G/B generalizes to the setting of vector bundles in algebraic geometry. The primary goal of this article is to carry this out for type G2. To give a better idea of the difference between classical and exceptional types, let us describe the classical problem in slightly more detail. The flag bundles are the universal cases of general degeneracy locus problems in algebraic geometry. Specifically, let V be a vector bundle of rank n on a variety X, and let ϕ : V ⊗ V → k be a symplectic or nondegenerate symmetric bilinear form (or the zero form). If E• and F• are general flags of isotropic subbundles of V , the problem is to find formulas in H ∗X for the degeneracy locus Dw = {x ∈ X dim(Fp(x) ∩ Eq(x)) ≥ rw w0(q, p)}, in terms of the Chern classes of the line bundles Eq/Eq−1 and Fp/Fp−1, for all p and q. (Here w is an element of the Weyl group, considered as a permutation via an embedding in the symmetric group Sn; w0 is the longest element, correspond- ing to the permutation n n − 1 · · · 1; and rw(q, p) = #{i ≤ q w(i) ≤ p} is a nonnegative integer depending on w, p, and q.) Such formulas have a wide range of applications: for example, they appear in the theory of special divisors and variation of Hodge structure on curves in algebraic geometry [Ha-Tu, Pa-Pr], and they are used to study singularities of smooth maps in differential geom- etry (work of Feh´er and Rim´anyi, e.g., [Fe-Ri]). They are also of interest in combinatorics (e.g., work of Lascoux -- Schutzenberger, Fomin -- Kirillov, Pragacz, Kresch -- Tamvakis). See [Fu-Pr] for a more detailed account of the history. In this article, we pose and solve the corresponding problem in type G2: Let V → X be a vector bundle of rank 7, equipped with a nondegener- ate alternating trilinear form γ :V3 V → L, for a line bundle L. Let CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 3 E• and F• be general flags of γ-isotropic subbundles of V , and let Dw = {x ∈ X dim(Fp(x) ∩ Eq(x)) ≥ rw w0(q, p)}, where w is an element of the Weyl group for G2 (the dihedral group with 12 elements). Find a formula for [Dw] in H ∗X, in terms of the Chern classes of the bundles involved. The meaning of "nondegenerate" and "γ-isotropic" will be explained below (§§2.1 -- 2.2), as will the precise definition of Dw (§2.5). In order to establish the relation between group theory and geometry, we give descriptions of the G2 flag variety and its Schubert subvarieties which appear to be new, although they will not surprise the experts (§4, §A.4). This is done in such a way as to make the transition to flag bundles natural. We then give presentations of the cohomology rings of these flag bundles, including ones with integer coefficients (Theorem 5.4). Finally, we prove formulas for the classes of Schubert varieties in flag bundles (§6); the formulas themselves are given in [An1, Appendix D.2]. We also discuss alternative formulas, answer a question of William Graham about the integrality of a certain rational cohomology class, and prove a result giving restrictions on candidates for "G2 Schubert polynomials" (§7). We also need a result on the integral cohomology of quadric bundles, which were studied in [Ed-Gr]. Appendix B corrects a small error in that article. Various constructions of exceptional-type flag varieties have been given using techniques from algebra and representation theory; those appearing in [La-Ma], [Il-Ma], and [Ga] have a similar flavor to the one presented here. A key feature of our description is that the data parametrized by the G2 flag variety naturally determine a complete flag in a 7-dimensional vector space, much as isotropic flags in classical types determine complete flags by taking orthogonal complements. The fundamental facts that make this work are Proposition 2.2 and its cousins, Corollary 3.12 and Propositions 3.15 and 3.16. Formulas for degeneracy loci are closely related to Giambelli formulas for equivariant classes of Schubert varieties in the equivariant cohomology of the corresponding flag variety. We will usually use the language of degeneracy loci, but we discuss the connection with equivariant cohomology in §2.6. In brief, the two perspectives are equivalent when det V and L are trivial line bundles. Another notion of degeneracy loci is often useful, where one is given a map of vector bundles ϕ : E → F on X, possibly possessing some kind of symmetry, and one is interested in the locus where ϕ drops rank. This is the situation considered in [Ha-Tu], for example, with F = E∗ and symmetric or skew-symmetric maps. We investigate the G2 analogue of this problem in [An2]. When the base X is a point, so V is a vector space and the flag bundle is just the flag variety G/B, most of the results have been known for some time; essentially everything can be done using the general tools of Lie theory. For example, a presentation of H ∗(G/B, Z) was given by Bott and Samelson [Bo-Sa], and (different) formulas for Schubert classes in H ∗(G/B, Q) appear in [BGG]. Since this article also aims to present a concrete, unified perspective on the G2 flag variety, accessible to general algebraic geometers, we wish to emphasize 4 DAVE ANDERSON geometry over Lie theory: we are describing a geometric situation from which type-G2 groups arise naturally. Reflecting this perspective, we postpone the Lie- and representation-theoretic arguments to Appendix A. We shall use some of the notation and results of this appendix throughout the article, though, so the reader less familiar with Lie theory is advised to skim at least §A.1, §A.3, and §A.5. Notation and conventions. Unless otherwise indicated, the base field k will have characteristic not 2 and be algebraically closed (although a quadratic extension of the prime field usually suffices). When char(k) = 2, several of our defini- tions and results about forms and octonions break down. However, most of the other main results hold in arbitrary characteristic, including the descrip- tion of the G2 flag variety and its cohomology, the degeneracy locus formulas, and the parametrizations of Schubert cells; see [An1, Chapter 6] for details in characteristic 2. Angle brackets denote the span of enclosed vectors: hx, y, zi := span{x, y, z}. For a vector bundle V on X and a point x ∈ X, V (x) denotes the fiber over x. If X → Y is a morphism and V is a vector bundle on Y , we will often write V for the vector bundle pulled back to X. If V is a vector space and E is a subspace, [E] denotes the corresponding point in an appropriate Grassmannian. We generally use the notation and language of (singular) cohomology, but this should be read as Chow cohomology for ground fields other than C. (Since the varieties whose cohomology we compute are rational homogeneous spaces or fibered in homogeneous spaces, the distinction is not significant.) Acknowledgements. This work is part of my Ph. D. thesis, and it is a pleasure to thank William Fulton for his encouragement in this project and careful readings of earlier drafts. Conversations and correspondence with many people have ben- efitted me; in particular, I would like to thank Robert Bryant, Skip Garibaldi, William Graham, Sam Payne, and Ravi Vakil. Thanks also to an anonymous referee for comments on the manuscript. 2. Overview We begin with an overview of our description of the G2 flag variety and statements of the main results. Proofs and details are given in later sections. 2.1. Compatible forms. Let V be a k-vector space. Let β be a nondegenerate symmetric bilinear form on V , and let γ be an alternating trilinear form, i.e., γ : V3 V → k. Write v 7→ v† for the isomorphism V → V ∗ defined by β, and ϕ 7→ ϕ† for the inverse map V ∗ → V . (Explcitly, these are defined by v†(u) = β(v, u) and ϕ(u) = (ϕ†, u) for any u ∈ V .) Our constructions are based on the following definitions: Definition 2.1. Call the forms γ and β compatible if (2.1) 2 γ(u, v, γ(u, v, ·)†) = β(u, u)β(v, v) − β(u, v)2 CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 5 for all u, v ∈ V . An alternating trilinear form γ :V3 V → k is nondegenerate if there exists a compatible nondegenerate symmetric bilinear form on V . The meaning of the strange-looking relation (2.1) will be explained in §3; see Proposition 3.3. (The factor of 2 is due to our convention that a quadratic norm and corresponding bilinear form are related by β(u, u) = 2 N (u).) A pair of compatible forms is equivalent to a composition algebra structure on k ⊕ V (see §3). Since a composition algebra must have dimension 1, 2, 4, or 8 over k (by Hurwitz's theorem), it follows that nondegenerate trilinear forms exist only when V has dimension 1, 3, or 7. In each case, there is an open dense GL(V )- orbit in V3 V ∗ consisting of nondegenerate forms. When dim V = 1, the only alternating trilinear form is zero, and any nonzero bilinear form is compatible with it. When dim V = 3, an alternating trilinear form is a scalar multiple of the determinant, and given a nondegenerate bilinear form, it is easy to show that there is a unique compatible trilinear form up to sign. When dim V = 7, it is less obvious that V3 V ∗ has an open GL(V )-orbit, especially if char(k) = 3, but it is still true (Proposition A.1). The choice of γ determines β uniquely up to scalar -- in fact, up to a cube root of unity (see Proposition A.3). Associated to any alternating trilinear form γ on a seven-dimensional vector space V , there is a canonical map Bγ : Sym2 V → V7 V ∗, determining (up to scalar) a bilinear form βγ. We will give the formula for char(k) 6= 3 here. Following Bryant [Br], we define Bγ by (2.2) Bγ(u, v) = − γ(u, ·, ·) ∧ γ(v, ·, ·) ∧ γ, 1 3 where γ(u, ·, ·) : V2 V → k is obtained by contracting γ with u. Choosing an isomorphismV7 V ∗ ∼= k yields a symmetric bilinear form βγ. If βγ is nondegen- erate, then a scalar multiple of it is compatible with the trilinear form γ; thus γ is nondegenerate if and only if βγ is nondegenerate. The form βγ is defined in characteristic 3, as well, and the statement still holds (see Lemma 3.9 and its proof). 2.2. Isotropic spaces. For the rest of this section, assume dim V = 7. Given a nondegenerate trilinear form γ on V , say a subspace F of dimension at least 2 is γ-isotropic if γ(u, v, ·) ≡ 0 for all u, v ∈ F . (That is, the map F ⊗ F → V ∗ induced by γ is zero.) Say a vector or a 1-dimensional subspace is γ-isotropic if it is contained in a 2-dimensional γ-isotropic space. If β is a compatible bilinear form, every γ-isotropic subspace is also β-isotropic (Lemma 3.5); as usual, this means β restricts to zero on the subspace. Since β is nondegenerate, a maximal β-isotropic subspace has dimension 3. Proposition 2.2. For any (nonzero) isotropic vector u ∈ V , the space Eu = {v hu, vi is γ-isotropic} is three-dimensional and β-isotropic. Moreover, every two-dimensional γ-isotropic subspace of Eu contains u. 6 DAVE ANDERSON The proof is given at the end of §3.2. The proposition implies that a maximal γ-isotropic subspace has dimension 2, and motivates the central definition: Definition 2.3. A γ-isotropic flag (or G2 flag) in V is a chain F1 ⊂ F2 ⊂ V of γ-isotropic subspaces, of dimensions 1 and 2. The variety parametrizing γ- isotropic flags is called the γ-isotropic flag variety (or G2 flag variety), and denoted F lγ(V ). The γ-isotropic flag variety is a smooth, six-dimensional projective variety (Proposition 4.1). See §A.4 for its description as a homogeneous space. Proposition 2.2 shows that a γ-isotropic flag has a unique extension to a com- plete flag in V : set F3 = Eu for u spanning F1, and let F7−i be the orthogonal space F ⊥ i , with respect to a compatible form β. (Since a compatible form is unique up to scalar, this is independent of the choice of β.) This defines a closed immersion F lγ(V ) ֒→ F lβ(V ) ⊂ F l(V ), where F lβ(V ) and F l(V ) are the (classical) type B and type A flag varieties, respectively. From the definition, there is a tautological sequence of vector bundles on F lγ(V ), S1 ⊂ S2 ⊂ V, and this extends to a complete γ-isotropic flag of bundles S1 ⊂ S2 ⊂ S3 ⊂ S4 ⊂ S5 ⊂ S6 ⊂ V by the proposition. Similarly, there are universal quotient bundles Qi = V /S7−i. 2.3. Bundles. Now let V → X be a vector bundle of rank 7, and let L be a line bundle on X. An alternating trilinear form γ :V3 V → L is nondegenerate if it is locally nondegenerate on fibers. Equivalently, we may define the Bryant form Bγ : Sym2 V → det V ∗ ⊗ L⊗3 by Equation (2.2), and γ is nondegenerate if and only if Bγ is (so Bγ defines an isomorphism V ∼= V ∗ ⊗ det V ∗ ⊗ L⊗3). A subbundle F of V is γ-isotropic if each fiber F (x) is γ-isotropic in V (x); for F of rank 2, this is equivalent to requiring that the induced map F ⊗F → V ∗ ⊗L 1 ⊗ V ∗ ⊗ L) be zero. If F1 ⊂ V is γ-isotropic, the bundle EF1 = ker(V → F ∗ (If u is a vector in a fiber F1(x), then has rank 3 and is isotropic for Bγ. EF1(x) = Eu, in the notation of §2.2.) Given a nondegenerate form γ on V , there is a γ-isotropic flag bundle Flγ(V ) → X, with fibers F lγ(V (x)). This comes with universal γ-isotropic subbundles Si and quotient bundles Qi, as before. 2.4. Chern class formulas. In the setup of §2.3, one has Schubert loci Ωw ⊆ Flγ(V ) indexed by the Weyl group. There is an embedding of W = W (G2) in the symmetric group S7 such that the permutation corresponding to w ∈ W is determined by its first two values. We identify w with this pair of integers, so CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 7 w = w(1) w(2); see §A.3 for more on the Weyl group. As in classical types, we set (2.3) rw(q, p) = #{i ≤ q w(i) ≤ p}. Given a fixed γ-isotropic flag F• on X, the Schubert loci are defined by Ωw = {x ∈ Flγ(V ) rk(Fp → Qq) ≤ rw(q, p) for 1 ≤ p ≤ 7, 1 ≤ q ≤ 2}. These are locally trivial fiber bundles, whose fibers are Schubert varieties in F lγ(V (x)). The G2 divided difference operators ∂s and ∂t act on Λ[x1, x2], for any ring Λ, by (2.4) (2.5) ∂s(f ) = ∂t(f ) = f (x1, x2) − f (x2, x1) x1 − x2 ; f (x1, x2) − f (x1, x1 − x2) −x1 + 2x2 . If w ∈ W has reduced word w = s1 · s2 · · · sℓ (where si is the simple reflection s or t), then define ∂w to be the composition ∂s1 ◦ · · · ◦ ∂sℓ. This is independent of the choice of word; see §A.5. (As mentioned in §A.3, each w ∈ W (G2) has a unique reduced word, with the exception of w0, so independence of choice is actually lack of choice in this case.) These formulas also define operators on H ∗Flγ(V ). (See §6.) Let V be a vector bundle of rank 7 on X equipped with a nondegenerate form γ : V3 V → kX , and assume det V is trivial. Let F1 ⊂ F2 ⊂ · · · ⊂ V be a complete γ-isotropic flag in V . Set y1 = c1(F1), y2 = c1(F2/F1). Let Flγ(V ) → X be the flag bundle, and set x1 = −c1(S1) and x2 = −c1(S2/S1), where S1 ⊂ S2 ⊂ V are the tautological bundles. Theorem 2.4. We have where Gw = ∂w0 w−1Gw0, and [Ωw] = Gw(x; y), Gw0(x; y) = (x3 1 2 ×(x2 2 + x1 y1 y2 − y2 1 y2 + y1 y2 2) 1 y1 + x1 y2 1 − 2 x2 1 + x1 y1 + y1 y2 − y2 1 − x1 y2 2)(x2 − x1 − y2). in H ∗(Flγ(V ), Z). (Here w0 is the longest element of the Weyl group.) The proof is given in §6, along with a discussion of alternative formulas, includ- ing ones where γ takes values in M ⊗3 for an arbitrary line bundle M . 2.5. Degeneracy loci. Returning to the problem posed in the introduction, let V be a rank 7 vector bundle on a variety X, with nondegenerate form γ and two (complete) γ-isotropic flags of subbundles F• and E•. The first flag, F•, allows us to define Schubert loci in the flag bundle Flγ(V ) as in §2.4. The second flag, 8 DAVE ANDERSON E•, determines a section s of Flγ(V ) → X, and we define degeneracy loci as scheme-theoretic inverse images under s: Dw = s−1Ωw ⊂ X. When X is Cohen-Macaulay and Dw has expected codimension (equal to the length of w; see §A.3), we have (2.6) [Dw] = s∗[Ωw] = Gw(x; y) in H ∗X, where xi = −c1(Ei/Ei−1) and yi = c1(Fi/Fi−1). More generally, this polynomial defines a class supported on Dw, even without assumptions on the singularities of X or the genericity of the flags F• and E•; see [Fu1] or [Fu-Pr, App. A] for the intersection-theoretic details. 2.6. Equivariant cohomology. Now return to the case where V is a 7-dimensional vector space. One can choose a basis f1, . . . , f7 such that Fi = hf1, . . . , fii forms a complete γ-isotropic flag in V , and let T = (k∗)2 act on V ∼= k7 by (z1, z2) 7→ diag(z1, z2, z1z−1 2 , 1, z−1 1 z2, z−1 2 , z−1 1 ). Write t1 and t2 for the corresponding weights. Then T preserves γ and acts on F lγ(V ). The total equivariant Chern class of V is cT (V ) = (1 − t2 2)(1 − (t1 − t2)2), so we have 1)(1 − t2 H ∗ T (F lγ(V ), Z[ 1 2 ]) = Z[ 1 2 ][x1, x2, t1, t2]/(r2, r4, r6), 2, (t1 − t2)2). A pre- with the relations r2i = ei(x2 sentation with Z coefficients can be deduced from Theorem 5.4; see Remark 5.5. 2, (x1 − x2)2) − ei(t2 1, x2 1, t2 Theorem 2.4 yields an equivariant Giambelli formula: [Ωw]T = Gw(x; t) in H ∗ T F lγ. In fact, this formula holds with integer coefficients: the Schubert classes form a basis for H ∗ T (F lγ, Z) over Z[t1, t2], so in particular there is no torsion, and H ∗ T (F lγ, Z) includes in H ∗ T (F lγ, Z[ 1 2 ]). The equivariant geometry of F lγ is closely related to the degeneracy loci In the setup of §2.5, assume V problem; we briefly describe the connection. has trivial determinant and γ has values in the trivial bundle, so the structure group is G = G2. The data of two γ-isotropic flags in V gives a map to the classifying space BB ×BG BB, where B ⊂ G is a Borel subgroup, and there are universal degeneracy loci Ωw in this space. On the other hand, there is an isomorphism BB ×BG BB ∼= EB ×B (G/B), carrying Ωw to EB ×B Ωw. Since T (F lγ) = H ∗(EB ×B (G/B)), and [Ωw]T = [EB ×B Ωw], a Giambelli formula H ∗ for [Ωw]T is equivalent to a degeneracy locus formula for this situation. One may then use equivariant localization to verify a given formula; this is essentially the approach taken in [Gr2]. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 9 2.7. Other types. It is reasonable to hope for a similar degeneracy locus story in some of the remaining exceptional types. Groups of type F4 and E6 are closely related to Albert algebras, and bundle versions of these algebras have been defined and studied over some one-dimensional bases [Pu]. Concrete realizations of the flag varieties have been given for types F4 [La-Ma], E6 [Il-Ma], and E7 [Ga]. Part of the challenge is to produce a complete flag from one of these realizations, and this seems to become more difficult as the dimension of the minimal irreducible representation increases with respect to the rank. 3. Octonions and compatible forms Any description of G2 geometry is bound to be related to octonion algebras, since the simple group of type G2 may be realized as the automorphism group of an octonion algebra; see Proposition 3.2 below. For an entertaining and wide- ranging tour of the octonions (also known as the Cayley numbers or octaves), see [Ba]. The basic linear-algebraic data can be defined as in §2, without reference to octonions, but the octonionic description is equivalent and sometimes more concrete. In this section, we collect the basic facts about octonions that we will use, and establish their relationship with the notion of compatible forms introduced in §2.1. Most of the statements hold over an arbitrary field, but we will continue to assume k is algebraically closed of characteristic not 2. While studying holonomy groups of Riemannian manifolds, Bryant proved several related facts about octonions and representations of (real forms of) G2. In particular, he gives a way of producing a compatible bilinear form associated to a given trilinear form; we will use a version of this construction for forms on vector bundles. See [Br] or [Ha] for a discussion of the role of G2 in differential geometry. As far as I am aware, the results in §§3.2 -- 3.3 have not appeared in the litera- ture in this form, although related ideas about trilinear forms on a 7-dimensional vector space can be found in [Br, §2]. 3.1. Standard facts. Here we list some well-known facts about composition algebras, referring to [Sp-Ve, §1] for proofs of any non-obvious assertions. Definition 3.1. A composition algebra is a k-vector space C with a nonde- generate quadratic norm N : C → k and an algebra structure m : C ⊗ C → C, with identity e, such that N (uv) = N (u)N (v). Denote by β′ the symmetric bilinear form associated to N , defined by β′(u, v) = N (u + v) − N (u) − N (v). (Notice that β′(u, u) = 2N (u).) Since N (u) = N (eu) = N (e)N (u) for all u ∈ C, it follows that N (e) = 1 and β′(e, e) = 2. The possible dimensions for C are 1, 2, 4, and 8. A composition algebra of dimension 4 is called a quaternion algebra, and one of dimension 8 is an octonion algebra; octonion algebras are neither associative nor commutative. 10 DAVE ANDERSON If there is a nonzero vector u ∈ C with N (u) = 0, then C is split. (Otherwise C is a normed division algebra.) Any two split composition algebras of the same dimension are isomorphic. Over an algebraically closed field, C is always split, so in this case there is only one composition algebra in each possible dimension, up to isomorphism. Define conjugation on C by u = β′(u, e)e − u. Every element u ∈ C satisfies a quadratic minimal equation (3.1) so (3.2) u2 − β′(u, e)u + N (u)e = 0, uu = uu = N (u)e. Write V = e⊥ ⊂ C for the imaginary subspace. For u ∈ V , u = −u, so u2 = −N (u)e, that is, N (u) = − 1 2 β′(u2, e). For u, v ∈ V , we have β′(u, v)e = N (u + v)e − N (u)e − N (v)e (3.3) = −uv − vu. Although C may not be associative, we always have u(uv) = (uu)v = N (u)v and (uv)v = u(vv) = N (v)u for any u, v ∈ C. Also, for u, v, w ∈ C we have (3.4) β′(uv, w) = β′(v, uw) = β′(u, wv). A nonzero element u ∈ C is a zerodivisor if there is a nonzero v such that uv = 0. We have 0 = u(uv) = (uu)v = N (u)v, so (3.5) u is a zerodivisor iff N (u) = 0. The relevance to G2 geometry comes from the following: Proposition 3.2 ([Sp-Ve, §2]). Let C be an octonion algebra over any field k. Then the group G = Aut(C) of algebra automorphisms of C is a simple group of type G2, defined over k. In fact, G ⊂ SO(V, β) ⊂ SO(C, β′), where V = e⊥. If char(k) 6= 2, G acts irreducibly on V . (cid:3) 3.2. Forms. The algebra structure on C corresponds to a trilinear form γ′ : C ⊗ C ⊗ C → k, using β′ to identify C with C ∗. Specifically, we have (3.6) γ′(u, v, w) = β′(uv, w). Restricting γ′ to V , we get an alternating form which we will denote by γ. (This follows from (3.4) and the fact that u = −u for u ∈ V .) Also, β′ restricts to a nondegenerate form β on V , defining an isomorphism V → V ∗. The multiplication map m : C ⊗ C → C, with C = k ⊕ V , is characterized by (3.7) (3.8) (3.9) m(u, v) = − 1 2 β(u, v)e + γ(u, v, ·)† for u, v ∈ V ; m(u, e) = m(e, u) = u m(e, e) = e. for u ∈ V ; CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 11 Conversely, given a trilinear form γ ∈ V3 V ∗ and a nondegenerate bilinear form β ∈ Sym2 V ∗, extend β orthogonally to C = k ⊕ V and define a multipli- cation m according to formulas (3.7) -- (3.9) above. Proposition 3.3. This multiplication makes C into a composition algebra with norm N (u) = 1 2 β′(u, u) if and only if γ and β are compatible, in the sense of Definition 2.1. Proof. This is a simple computation: For u, v ∈ V , we have β′(uv, uv) N (uv) = = = 1 2 1 2 1 4 β′(γ(u, v, ·)†, γ(u, v, ·)†) β′(− 1 2 β(u, v)e, − β(u, v)β(u, v) + 1 2 β(u, v)e) + 1 2 1 γ(u, v, γ(u, v, ·)†), 2 and N (u)N (v) = 1 4 β(u, u)β(v, v). (cid:3) Remark 3.4. Similar characterizations of octonionic multiplication have been given, usually in terms of a cross product on V . See [Br, §2] or [Ha, §6]. Lemma 3.5. Suppose γ and β are compatible forms on V , defining a composi- tion algebra structure on C = k ⊕ V . Then L ⊂ V is γ-isotropic iff uv = 0 in C for all u, v ∈ L. In particular, any γ-isotropic subspace is also β-isotropic. Proof. Let γ′ and β′ be the forms corresponding to the algebra structure. One implication is trivial: If uv = 0 for all u, v ∈ L, then β′(uv, ·) = γ′(u, v, ·) ≡ 0 on C, so γ(u, v, ·) ≡ 0 on V and L is γ-isotropic. Conversely, suppose L is γ-isotropic. First we show L is β-isotropic. Given any u ∈ L, choose a nonzero v ∈ u⊥ ∩ L. Since L is γ-isotropic, γ(u, v, ·)† = 0, so u and v are zerodivisors: uv = − 1 2 β(u, v) e + γ(u, v, ·)† = 0. Therefore N (u) = N (v) = 0, so N and β are zero on L. By (3.7), this also implies uv = 0 for all u, v ∈ L. (cid:3) Finally, it will be convenient to use certain bases for C and V . We need a well-known lemma: Lemma 3.6 ([Sp-Ve, (1.6.3)]). There are elements a, b, c ∈ C such that e, a, b, ab, c, ac, bc, (ab)c forms an orthogonal basis for C. Such a triple is called a basic triple for C. In fact, given any a ∈ V = e⊥ with N (a) = 1, we can choose b and c so that a, b, c is an orthonormal basic triple; similarly, if a and b are orthonormal vectors generating a quaternion subalgebra, we can find c so that a, b, c is an orthonormal basic triple. 12 DAVE ANDERSON If a, b, c are an orthonormal basic triple, let {e0 = e, e1, . . . , e7} be the cor- responding basis (in the same order as in Lemma 3.6). This is a standard orthonormal basis for C. With respect to the basis {e1, . . . , e7} for the imag- inary octonions V , we have β(ep, eq) = 2 δpq, and (3.10) γ = 2 (e∗ p ∧ e∗ 123 + e∗ q ∧ e∗ 257 − e∗ r. (Here e∗ 145 − e∗ 167 − e∗ p is the map eq 7→ δpq.) 246 − e∗ 347 − e∗ 356), where e∗ pqr = e∗ Remark 3.7. Note that for p > 0, e2 p = −e. This standard orthonormal basis is analogous to the standard basis "1, i, j, k" for the quaternions. Conventions for defining the octonionic product in terms of a standard basis vary widely in the literature, though -- Coxeter [Co, p. 562] calculates 480 possible variations! A choice of convention corresponds to a labelling and orientation of the Fano arrangement of 7 points and 7 lines; the one we use agrees with that of [Fu-Ha, p. 363]. (Coincidentally, our choice of γ very nearly agrees with the one used in [Br, §2]: there the signs of e∗ 356 are positive, and the common factor of 2 is absent.) 347 and e∗ We will most often use a different basis. Define (3.11) 2 (e1 + i e2) 2 (e5 + i e6) 2 (e4 + i e7) f1 = 1 f2 = 1 f3 = 1 f4 = i e3 f5 = − 1 f6 = − 1 f7 = − 1 2 (e4 − i e7) 2 (e5 − i e6) 2 (e1 − i e2), and call this the standard γ-isotropic basis for V . (Here i is a fixed square root of −1 in k.) With respect to this basis, the bilinear form is given by (3.12) β(fp, f8−q) = −δpq, for p 6= 4 or q 6= 4; β(f4, f4) = −2. The trilinear form is given by (3.13) (As above, f ∗ γ = f ∗ 147 + f ∗ p denotes fq 7→ δpq.) 246 + f ∗ 345 − f ∗ 237 − f ∗ 156. Example 3.8. We can use the expression (3.13) to compute the octonionic product f2 f3. By (3.7) -- (3.9), this is f2 f3 = − 1 2 β(f2, f3) e + γ(f2, f3, ·)† = γ(f2, f3, ·)†. Since γ(f2, f3, fj) = −δ7,j = β(f1, fj), we see γ(f2, f3, ·)† = f1. Therefore f2 f3 = f1. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 13 We use computations in the f basis to prove another characterization of nondegenerate forms. 1 3 (u, v) 7→ − γ(u, ·, ·) ∧ γ(v, ·, ·) ∧ γ is nondegenerate. conclusion holds.) bilinear form defined as in (2.2) (for char(k) 6= 3), by composing (In fact, βγ is also defined if char(k) = 3, and the same Lemma 3.9. Let γ :V3 V → k be a trilinear form, and let βγ be a symmetric with an isomorphism V7 V ∗ ∼= k. Then γ is nondegenerate if and only if βγ Proof. Let U ⊂V3 V ∗ be the set of nondegenerate forms, and let U ′ ⊂V3 V ∗ First suppose γ is nondegenerate. Since U is a GL(V )-orbit in V3 V ∗, we 1234567 to identify V7 V ∗ with k, we find βγ = β, i.e., may choose a basis {fj} so that γ has the expression (3.13). Computing in this basis, and using f ∗ βγ(fp, f8−q) = −δpq for p, q 6= 4, and βγ(f4, f4) = −2. Indeed, we have be the set of forms γ such that βγ is nondegenerate; we want to show U = U ′. (By Proposition A.1, U is open and dense.) γ(f1, ·, ·) ∧ γ(f7, ·, ·) ∧ γ = (f ∗ = 3f ∗ 47 − f ∗ 1234567. 56) ∧ (f ∗ 14 − f ∗ 23) ∧ γ γ and βγ are compatible forms. (For an arbitrary choice of isomorphism, βγ is a scalar multiple of a compatible form.) The others are similar. In particular, with this choice of isomorphismV7 V ∗ ∼= k, a rank 7 free Z-module, fix a basis f1, . . . , f7, and let γZ :V3 VZ → Z be given To see this works in characteristic 3, one can avoid division by 3. Let VZ be by (3.13). The same computation shows that γZ(fp, ·, ·) ∧ γZ(f8−q, ·, ·) ∧ γZ = 3 δpq f ∗ 1234567 for p, q 6= 4, and γZ(f4, ·, ·) ∧ γZ(f4, ·, ·) ∧ γZ = 6 f ∗ 1234567, so one can define βγ over Z. (For nondegeneracy, one still needs char(k) 6= 2 here.) For the converse, note that the terms in the compatibility relation (2.1) make sense for all γ in U ′, since here γ(u, v, ·)† is well-defined. We have seen that the relation holds on the dense open subset U ⊂ U ′, so it must hold on all of U ′. Therefore every γ in U ′ has a compatible bilinear form, i.e., γ is in U . (cid:3) The following two lemmas prove Proposition 2.2: Lemma 3.10. If u ∈ V is a nonzero isotropic vector, then Eu = {v ∈ V uv = 0} = {v ∈ V γ(u, v, ·) ≡ 0} is a three-dimensional β-isotropic subspace. 14 DAVE ANDERSON Proof. By definition, Eu consists of zero-divisors, so it is β-isotropic by (3.5). Since β is nondegenerate on V , we know dim Eu ≤ 3. In fact, it is enough to observe that G = Aut(C) acts transitively on the set of isotropic vectors (up to scalar); this follows from Proposition A.5. Thus for any u, we can find g ∈ G such that g · u = λf1 for some λ 6= 0. Clearly g · Eu = Eg·u = Ef1, and one checks that f1f2 = f1f3 = 0. (cid:3) Lemma 3.11. Let u ∈ V be a nonzero isotropic vector, and let v, w ∈ Eu be such that {u, v, w} is a basis. Then vw = λu for some nonzero λ ∈ k. Proof. First note that vw = −wv, since −vw − wv = β(v, w)e = 0. If {u, v′, w′} is another basis, with v′ = a1u + a2v + a3w and w′ = b1u + b2v + b3w, then a2b3 − a3b2 6= 0, so v′w′ = (a2b3)vw + (a3b2)wv = (a2b3 − a3b2)vw is a nonzero multiple of vw. Now it suffices to check this for the standard γ-isotropic basis, and indeed, we computed f2f3 = f1 in Example 3.8. (cid:3) Corollary 3.12. Let V = L1 ⊕ · · · ⊕ L7 be a splitting into one-dimensional subspaces such that L1 is γ-isotropic, and L1 ⊕ L2 ⊕ L3 = Eu for a generator u ∈ L1. Then the map V ⊗V → V ∗ ∼= V induced by γ restricts to a G-equivariant isomorphism L2 ⊗ L3 (cid:3) ∼ −→ L1. Finally, the following lemma is verified by a straightforward computation: Lemma 3.13. Let T = (k∗)2 act on V via the matrix diag(z1, z2, z1z−1 2 , 1, z−1 1 z2, z−1 2 , z−1 1 ) (in the f -basis). Then T preserves the forms β and γ of (3.12) and (3.13). (cid:3) The corresponding weights for this torus action are {t1, t2, t1 − t2, 0, t2 − t1, −t2, −t1}. 3.3. Octonion bundles. Let X be a variety over k. The notion of composition algebra can be globalized: Definition 3.14. A composition algebra bundle over X is a vector bun- dle C → X, equipped with a nondegenerate quadratic norm N : C → kX , a multiplication m : C ⊗ C → C, and an identity section e : kX → C, such that N respects composition. (Equivalently, for each x ∈ X, the fiber C(x) is a composition algebra over k.) Since char(k) 6= 2, there is a corresponding nondegenerate bilinear form β′ on C. We will also allow composition algebras whose norm takes values in a m line bundle M ⊗2; here the multiplication is C ⊗ C −→ C ⊗ M , and the identity e −→ C. Here a little care is required in the definition. The composition is M m −→ C ⊗ M should be the identity, and the other composition C ⊗ M id⊗e −−−→ C ⊗ C CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 15 (m ◦ (e ⊗ id)) should be the canonical isomorphism. The compatibility between m and N is encoded in the commutativity of the following diagram: C ⊗ C m ✲ C ⊗ M N ⊗ N ✲ ✛ M ⊗4. N ⊗ (N ◦ e) The norm of e is the quadratic map M → M ⊗2 corresponding to M ⊗2 β′ −→ M ⊗2. Replacing C with eC = C ⊗ M ∗, one obtains a composition algebra whose norm Many of the properties of composition algebras discussed above have straight- takes values in the trivial bundle. forward generalizations to bundles; we mention a few without giving proofs. Using β′ to identify C with C ∗ ⊗ M ⊗2, the multiplication map corresponds to a trilinear form γ′ : C ⊗ C ⊗ C → M ⊗3. The imaginary subbundle V is the orthogonal complement to e in C, so C = M ⊕ V . The bilinear form β′ restricts to a nondegenerate form β on V , and γ′ restricts to an alternating form octonion bundles and reducing to the local case: from the forms β and γ on V , and there is an analogue of Proposition 3.3. The analogues of Proposition 2.2 and Corollary 3.12 can be proved using γ : V3 V → M ⊗3. As before, the multiplication on M ⊕ V can be recovered Proposition 3.15. Let γ :V3 V → M ⊗3 and β : Sym2 V → M ⊗2 be (locally) compatible forms. Let F1 ⊂ V be a γ-isotropic line bundle, and let ϕ : V → 1 ⊗ V ∗ ⊗ M ⊗3 be the map defined by γ. Then the bundle F ∗ has rank 3 and is β-isotropic. (cid:3) EF1 = ker(ϕ) Proposition 3.16. Let V be as in Proposition 3.15, and suppose there is a splitting V = L1 ⊕ · · · ⊕ L7 into line bundles such that L1 is γ-isotropic, and L1 ⊕ L2 ⊕ L3 = EL1. Then the map V ⊗ V → V ∗ ∼= V ⊗ M induced by γ and β restricts to an isomorphism L2 ⊗ L3 ∼ −→ L1 ⊗ M . (cid:3) Remark 3.17. Composition algebras may defined over an arbitrary base scheme X; in fact, as with Azumaya algebras, one is mainly interested in cases where X is defined over a non-algebraically closed field or a Dedekind ring. Petersson has classified such composition algebra bundles in the case where X is a curve of genus zero [Pe]. Since then, some work has been done over other one-dimensional bases, but the theory remains largely undeveloped. 4. Topology of G2 flags There are two "γ-isotropic Grassmannians" parametrizing γ-isotropic sub- spaces of dimensions 1 or 2, which we write as Q or G, respectively; thus F lγ embeds in Q×G. Since γ-isotropic vectors are just those v such that β(v, v) = 0, Q is the smooth 5-dimensional quadric hypersurface in P(V ). 16 DAVE ANDERSON Proposition 4.1. The γ-isotropic flag variety is a smooth, 6-dimensional pro- jective variety. Moreover, both projections F lγ → Q and F lγ → G are P1- bundles. Proof. The quadric Q comes with a tautological line bundle S1 ⊂ VQ. By Proposition 2.2, the form γ also equips Q with a rank-3 bundle S3 ⊂ VQ, with fiber S3([u]) = Eu, the space swept out by all γ-isotropic 2-spaces containing u. Thus S1 ⊂ S3, and from the definitions we have F lγ(V ) = P(S3/S1) → Q. (We use the convention that P(E) parametrizes lines in the vector bundle E.) Similarly, if S2 is the tautological bundle on G, we have F lγ(V ) = P(S2) → G. (cid:3) This also shows that G is smooth of dimension 5. Remark 4.2. The definition of F lγ(V ) can be reformulated as follows. Let F l = F l(1, 2; V ) be the two-step partial flag variety. The nondegenerate form γ is also a section of the trivial vector bundleV3 V ∗ on F l. By restriction it gives a section of the rank 5 vector bundleV2 S∗ 5, where S1 ⊂ S2 ⊂ V is the tautological flag on F l and Q5 = V /S2. Then F lγ ⊂ F l is defined by the vanishing of this section. 2 ⊗ Q∗ Remark 4.3. Projectively, F lγ parametrizes data (p ∈ ℓ), where ℓ is a γ- isotropic line in Q, and p ∈ ℓ is a point. Thus Proposition 2.2 says that the union of such ℓ through a fixed p is a P2 in Q, and conversely, given such a P2 one can recover p (as the intersection of any two γ-isotropic lines in the P2). 4.1. Fixed points. Let {f1, f2, . . . , f7} be the standard γ-isotropic basis for V , and let T = (k∗)2 act as in Lemma 3.13, via the matrix diag(z1, z2, z1z−1 Write e(i j) for the two-step flag hfii ⊂ hfi, fji. 2 , 1, z−1 1 z2, z−1 2 , z−1 1 ). Proposition 4.4. This action of T defines an action on F lγ(V ), with 12 fixed points: e(1 2), e(1 3), e(2 1), e(2 5), e(3 1), e(3 6), e(5 2), e(5 7), e(6 3), e(6 7), e(7 5), e(7 6). Proof. Since T preserves β, it acts on Q, fixing the 6 points [f1], [f2], [f3], [f5], [f6], [f7]. Since T preserves γ, it acts on F lγ, and the projection F lγ → Q is T -equivariant. The T -fixed points of F lγ lie in the fibers over the fixed points of Q. Since each of these 6 fibers is a P1 with nontrivial T -action, there must be 2 · 6 = 12 fixed points. To see the fixed points are as claimed, note that the bundle S3 on Q is equivariant, and the fibers S3(x) = Ex at each of the fixed points are as follows: Ef1 = hf1, f2, f3i Ef2 = hf2, f1, f5i Ef3 = hf3, f1, f6i Ef5 = hf5, f2, f7i Ef6 = hf6, f3, f7i Ef7 = hf7, f5, f6i. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 17 Indeed, one simply checks that in each triple, the (octonionic) product of the first vector with either the second or the third is zero. (Alternatively, one can com- pute directly using the form (3.13).) Now the T -fixed lines in S3([fi])/S1([fi]) are [fj], where fj is the second or third vector in the triple beginning with fi. Thus the 12 points are e(i j), where fi is the first vector and fj is the second or third vector in one of the above triples. (cid:3) In general, the T -fixed points of a flag variety are indexed by the correspond- ing Weyl group W , which for type G2 is the dihedral group with 12 elements. We will write elements as w = w(1) w(2), for w(1) and w(2) such that e(w(1) w(2)) is a T -fixed point, as in Proposition 4.4. We fix two simple reflections generating W , s = 2 1 and t = 1 3. See §A.3 for more details on the Weyl group and its embedding in S7. 4.2. Schubert varieties. Fix a (complete) γ-isotropic flag F• in V . Each T - fixed point is the center of a Schubert cell, which is defined by X o w = {E• ∈ F lγ dim(Fp ∩ Eq) = rw(q, p) for 1 ≤ q ≤ 2, 1 ≤ p ≤ 7}, where rw(q, p) = #{i ≤ q w(i) ≤ p}, just as in the classical types. Also as in classical types, these can be parametrized by matrices, where Ei is the span of the first i rows. For example, the big cell is 7 6 = X a X o Y Z S T f b c d e 1 1 0 ! ∼= A6, where lowercase variables are free, and X, Y, Z, S, T are given by X = −ae − bd − c2 Y = −a − bf + cd − cef Z = −cf − d2 + def S = c + de − e2f T = −d + ef. (These equations can be obtained by octonionic multiplication; considering the two row vectors as imaginary octonions, the condition is that their product be zero. In fact, X, Y, Z are already determined by β-isotropicity.) Parametriza- tions of the other 11 cells are given in [An1, Appendix D.1]. The Schubert varieties Xw are the closures of the Schubert cells; equiva- lently, Xw = {E• ∈ F lγ dim(Fp ∩ Eq) ≥ rw(q, p) for 1 ≤ q ≤ 2, 1 ≤ p ≤ 7}. From the parametrizations of cells, we see dim Xw = ℓ(w). To get Schubert varieties with codimension ℓ(w), define Ωw = Xw w0. These can also be described using the tautological quotient bundles: Ωw = {x ∈ F lγ rk(Fp(x) → Qq(x)) ≤ rw(q, p)}. 18 DAVE ANDERSON Schubert varieties in Q and G are defined by the same conditions. (Note that w and w s define the same varieties in G, and w and w t define the same variety in Q. Write w for the corresponding equivalence class.) With the exception of X1 2, all Schubert varieties in F lγ are inverse images of Schubert varieties in Q or G: Proposition 4.5. Let p : F lγ → Q and q : F lγ → G be the projections. Then Xw = p−1Xw if w(1) < w(2) (except when w = 1 2), and Xw = q−1Xw if w(1) > w(2). The proof is immediate from the definitions. For instance, Xtst = X36 is a P2 it parametrizes all 1-dimensional subspaces of a fixed isotropic 3-space. in Q: Its inverse image in F lγ is p−1Xtst = Xtst = Ωsts. 5. Cohomology of flag bundles 5.1. Compatible forms on bundles. Let V be a rank 7 vector bundle on a variety X, equipped with a nondegenerate form γ : V3 V → L, and let Bγ : Sym2 V → det V ∗ ⊗ L⊗3 be the Bryant form (§2.1, §2.3). Assume there is a line bundle M such that (5.1) det V ∗ ⊗ L⊗3 ∼= M ⊗2. (For example, this holds if V has a maximal Bγ-isotropic subbundle F , for then we can take M = F ⊥/F . There exist Zariski-locally trivial bundles V without this property, though -- see [Ed-Gr, p. 293].) Lemma 5.1. In this setup, L ∼= M ⊗3 ⊗ T , for some line bundle T such that T ⊗3 is trivial. If L has a cube root, then T is trivial and M ∼= det V ⊗ (L∗)⊗2. The proof is straightforward; see [An1, Lemma 3.2.1] for details. From now on, we will assume V has a maximal Bγ-isotropic subbundle F = F3 ⊂ V . We also assume L has a cube root on X, so L ∼= M ⊗3. (By a theorem of Totaro, one can always assume this so long as 3-torsion is ignored in Chow groups (or cohomology); see [Fu2]. In the case at hand, Lemma 5.1 gives a direct reason.) 5.2. A splitting principle. For the next three subsections, we assume the line bundle M is trivial; this implies det V is also trivial. The case for general M will be described in §5.5. In this context, the relevant version of the splitting principle is the following: Lemma 5.2. Assume V is equipped with a nondegenerate trilinear form γ : V3 V → kX . There is a map f : Z → X such that f ∗ : H ∗X → H ∗Z is injective, and f ∗V ∼= L1 ⊕ L2 ⊕ · · · ⊕ L7, with Ei = L1 ⊕ · · · ⊕ Li forming a complete γ-isotropic flag in f ∗V . The proof is given in [An1, Lemma 3.2.2]. Given such a splitting, we can use β to identify L8−i with L∗ i , and Proposition 3.16 implies L3 ∼= L1 ⊗ L∗ (5.2) 2. Thus V ∼= L1 ⊕ L2 ⊕ (L1 ⊗ L∗ 2) ⊕ kX ⊕ (L∗ 1 ⊗ L2) ⊕ L∗ 2 ⊕ L∗ 1. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 19 Alternatively, using only a γ-isotropic flag of subbundles F1 ⊂ F2 ⊂ V , we have V ∼= F2 ⊕ (F1 ⊗ (F2/F1)∗) ⊕ kX ⊕ (F ∗ 1 ⊗ (F2/F1)) ⊕ F ∗ 2 . Since V is recovered from the data of L1 and L2, the universal base for V (with the assumed restrictions) is BGL1 × BGL1. 5.3. Chern classes. We continue to assume the line bundle M is trivial, and let F1 ⊂ F2 ⊂ F3 ⊂ V be a γ-isotropic flag in V . It follows from (5.2) that (5.3) c(V ) = (1 − y2 1)(1 − y2 2)(1 − (y1 − y2)2), where yi = c1(Li), and also that c1(F3) = 2 c1(F1). Let Q(V ) → X be the quadric bundle, with its tautological bundles S1 ⊂ S3 ⊂ V . Set x1 = −c1(S1) and α = [P(F3)] in H ∗Q(V ). The classes 1, x1, x2 form a basis for H ∗Q(V ) over H ∗X; see Appendix B. 1, α, x1 α, x2 1 α Lemma 5.3. We have c1(S3) = −2 x1, c2(S3) = 2 x2 and 1 + c2(F3) − 2 c1(F1)2. Proof. The expression for c1(S3) follows from (5.2). We have V /F ⊥ 3 V /S⊥ 3 3, so c(V ) = c(F3) · c(F ∗ 3 ) = c(S3) · c(S∗ 3 ). In particular, ∼= S∗ ∼= F ∗ 3 and c2(V ) = 2 c2(F3) − c1(F3)2 = 2 c2(S3) − c1(S3)2 = 2(c2(F3) − 2 c1(F1)2) = 2(c2(S3) − 2 x2 1). Up to 2-torsion, then, the formula for c2(S3) holds. Since the classifying space for this setup is BGL1 × BGL1, and there is no torsion in its cohomology, it follows that the formula also holds with integer coefficients. (cid:3) 5.4. Presentations. Using the fact that Flγ(V ) is a P1-bundle over a quadric bundle, we can give a presentation of its integral cohomology. First recall the presentation for H ∗Q(V ) (Theorem B.1). We continue to assume M is trivial, and hence also det V . Fix F1 ⊂ F3 ⊂ V as before, and let S1 ⊂ S3 ⊂ V be the tautological bundles on Q(V ). Let x1 = −c1(S1) and α = [P(F3)] in H ∗Q(V ). Then H ∗(Q(V ), Z) = (H ∗X)[x1, α]/I, where I is generated by 2α = x3 α2 = (c3(V /F3) + c1(V /F3) x2 1 − c1(F3) x2 1 + c2(F3) x1 − c3(F3), 1) α. Theorem 5.4. With notation as above, we have Flγ(V ) = P(S3/S1) → Q(V ) → X. Let x2 = −c1(S2/S1) be the hyperplane class for this P1-bundle. Then H ∗(Flγ(V ), Z) = (H ∗X)[x1, x2, α]/J, 20 DAVE ANDERSON where J is generated by the three relations (5.4) (5.5) (5.6) 2α = x3 α2 = (c3(V /F3) + c1(V /F3) x2 1 − c1(F3) x2 1 + c2(F3) x1 − c3(F3), 1) α, x2 1 + x2 2 − x1x2 = 2 c1(F1)2 − c2(F3). In fact, α is the Schubert class [Ωsts], defined in §6 below. Proof. Since Flγ(V ) = P(S3/S1) → Q(V ), we have H ∗Flγ = (H ∗Q)[x2]/(x2 2 + c1(S3/S1) x2 + c2(S3/S1)). One easily checks c1(S3/S1) = −x1, and 1 = x2 c2(S3/S1) = c2(S3) − x2 1 + c2(F3) − 2 c1(F1)2 by Lemma 5.3. This gives the third relation, and the first two relations come from the relations on H ∗Q. Finally, it is not hard to see that the 12 elements 1, x1, x2 1, α, x1 α, x2 1 α, x2, x1 x2, x2 1 x2, x2 α, x1 x2 α, x2 1 x2 α form a basis for the ring on the RHS over H ∗X, and we know they form a basis for H ∗Flγ over H ∗X. (cid:3) Remark 5.5. To obtain a presentation for H ∗ −cT 1 (Si/Si−1), ci(F3) = (−1)ici(V /F3) = ei(t1, t2, t1 − t2), and c1(F1) = t1. If we take coefficients in Z[ 1 2 ], the cohomology ring has a simpler presentation T (F lγ, Z), set α = [Ωsts]T , xi = similar to that for classical groups: Proposition 5.6. Suppose V has a splitting as in (5.2), with M trivial. Let Λ = H ∗X. Then H ∗(Flγ(V ), Z[ 1 r2i = ei(x2 1, x2 2 ]) ∼= Λ[x1, x2]/(r2, r4, r6), where 2, (y1 − y2)2). 2, (x1 − x2)2) − ei(y2 1, y2 Proof. The relations must hold, by (5.3). Monomials in x1 and x2 are global classes on Flγ that restrict to give a basis for the cohomology of each fiber, so the claim follows from the Leray -- Hirsch theorem. (cid:3) Taking X to be a point, these presentations specialize to give well-known presentations of H ∗F lγ (cf. [Bo-Sa]): Corollary 5.7. Let F lγ be the γ-isotropic flag variety, and let p : F lγ → Q be the projection to the quadric. Set α = [Ωsts] ∈ H ∗(F lγ, Z). Then we have H ∗(F lγ, Z) = Z[x1, x2, α]/(x2 1 + x2 2 − x1x2, 2 α − x3 1, α2), and H ∗(F lγ, Z[ 1 2 ]) = Z[ 1 = Z[ 1 2 ][x1, x2]/(ei(x2 2 ][x1, x2]/(x2 1, x2 1 + x2 2, (x1 − x2)2))i=1,2,3 2 − x1x2, x6 1). CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 21 5.5. Twisting. Now we allow γ to take values in L ∼= M ⊗3 for an arbitrary line bundle M on X, so det V ∼= M ⊗7 and the corresponding bilinear form has values in M ⊗2. The splitting principle (Lemma 5.2) holds as stated for γ : V3 V → M ⊗3. The compatible bilinear form β now identifies L8−i with L∗ have L3 ∼= L1 ⊗ L∗ i ⊗ M ⊗2, and we 2 ⊗ M . Thus V ∼= L1 ⊕ L2 ⊕ (L1 ⊗ L∗ 2 ⊗ M ) ⊕ M (5.7) ⊕(L∗ 1 ⊗ L2 ⊗ M ) ⊕ (L∗ 2 ⊗ M ⊗2) ⊕ (L∗ 1 ⊗ M ⊗2). Since V is recovered from the data of L1, L2, and M , the universal base for V is (BGL1)3. This space has no torsion in cohomology; it follows that we may deduce integral formulas using rational coefficients. As described in [Fu2], this situation reduces to the case where L is trivial. If Let eV = V ⊗ M ∗, so γ : V3 V → L determines a form eγ : V3eV → kX. V = L1 ⊕ · · · ⊕ L7 is a γ-isotropic splitting as in Lemma 5.2, we have eV = eL1 ⊕ · · · ⊕eL7, where eLi = Li ⊗ M ∗. Thus 1)(1 −ey2 where v = c1(M ),eyi = yi−v, soey3 =ey1−ey2 = y1−y2. Note that y1−y2 = y3−v, A rank 2 subbundle E ⊂ V is γ-isotropic if and only if eE = E ⊗ M ∗ ⊂ eV is eγ-isotropic (a map is zero iff it is zero after twisting by a line bundle), so we have an isomorphism Flγ(V ) ∼= Fleγ(eV ), and the tautological subbundles are related by eSi = Si ⊗ M ∗. Therefore exi = −c1(eSi/eSi−1) = xi + v. The presentation for H ∗Flγ(V ) is obtained from Proposition 5.6 by replacing yi with yi − v and xi with xi + v. since using γ and β there is an isomorphism L2 ⊗ L3 ∼= L1 ⊗ M . c(eV ) = (1 −ey2 2)(1 −ey2 3), 6. Divided difference operators and Chern class formulas For now, assume γ takes values in the trivial bundle. Given V → X with a (complete) γ-isotropic flag of subbundles F•, Schubert loci Ωw ⊂ Flγ(V ) are defined by rank conditions as in §2.4. (These are the same conditions as in §4.2 when X is a point.) As usual, there are two steps to producing formulas for these Schubert loci: first find a formula for the most degenerate locus (the case w = w0), and then apply divided difference operators to obtain formulas for all w ≤ w0. Theorem 6.1 and Lemma 6.3 prove Theorem 2.4. Theorem 6.1. Assume M is trivial, and let F1 ⊂ F2 ⊂ F3 ⊂ V be a γ-isotropic flag. Then [Ωw0] ∈ H ∗Flγ(V ) is given by [Ωw0] = (x3 1 2 ×(x2 1 + c2(F3) x1 − c3(F3)) 1 − c1(F3) x2 1 + c1(F1) x1 + c2(F3) − c1(F1)2)(x2 − x1 − c1(F3/F1)). 22 DAVE ANDERSON Setting y1 = c1(F1) and y2 = c1(F2/F1), we have c(F3) = (1 + y1)(1 + y2)(1 + y1 − y2), so this formula becomes [Ωw0] = Gw0(x; y), where Gw0(x; y) = 1 − x1 y2 2 + x1 y1 y2 − y2 1 y2 + y1 y2 2) (x3 1 2 ×(x2 1 y1 + x1 y2 1 − 2 x2 1 + x1 y1 + y1 y2 − y2 2)(x2 − x1 − y2). Proof. Let p : Flγ = P(S3/S1) → Q be the projection. The locus where S1 = F1 is p−1P(F1), so its class is p∗[P(F1)]. On P(F1) ⊂ Q, we have S1 = F1 and S3 = F3; thus on p−1P(F1), the locus where S2 = F2 is defined by the vanishing of the composed map F2/F1 = F2/S1 → S3/S1 → S3/S2. This class is given by c1((F2/F1)∗ ⊗ S3/S2) = x2 − x1 − c1(F2/F1), so pushing forward by the inclusion p−1P(F1) ֒→ Flγ, we have [Ωw0] = p∗[P(F1)] · (x2 − x1 − c1(F2/F1)). To determine [P(F1)] in H ∗Q, we first find the class in H ∗P(F3) and then push forward. By [Fu4, Ex. 3.2.17], this is x2 1 + c1(F3/F1) x1 + c2(F3/F1), and pushing forward is multiplication by α = [P(F3)]. Using the relation given in §5.4, we have [P(F1)] = α · (x2 1 + c1(F1) x1 + c2(F3) − c1(F1)2) 1 − c1(F3) x2 1 + c2(F3) x1 − c3(F3))(x2 (x3 = 1 2 1 + c1(F1) x1 + c2(F3) − c1(F1)2). (cid:3) Recall that the divided difference operators for G2 are defined as in §2.4, using the formulas (2.4) and (2.5) for the operators ∂s and ∂t corresponding to simple reflections. These operators may be constructed geometrically, using a correspondence as described in [Fu1]. Let Q(V ) and G(V ) be the quadric bundle and bundle of γ-isotropic 2-planes in V , respectively, and set Zs = Flγ(V )×G(V ) Flγ(V ) and Zt = Flγ(V ) ×Q(V ) Flγ(V ), with projections ps : Zs → Flγ and i pt i : Zt → Flγ. The proofs of the following two lemmas are the same as in classical types; see [An1, §4.1] for details. Lemma 6.2. As maps H ∗Flγ → H ∗Flγ, ∂s = (ps ∂t = (pt 1)∗ ◦ (ps 1)∗ ◦ (pt 2)∗ and 2)∗. Lemma 6.3. We have and 0 ∂s[Ωw] = ( [Ωw s] ∂t[Ωw] = ( [Ωw t] 0 if ℓ(w s) < ℓ(w); otherwise; if ℓ(w t) < ℓ(w); otherwise. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 23 By making the substitutions xi 7→ xi + v and yi 7→ yi − v, we obtain formulas for the more general case, where γ has values in M ⊗3 for arbitrary M . Theorem 6.4. Let γ : V3 V → M ⊗3 be a nondegenerate form, with a γ- isotropic flag F• ⊂ V . Let v = c1(M ). Let ∂s be defined as above, and let ∂t be given by (6.1) Then ∂t(f ) = f (x1, x2) − f (x1, x1 − x2 − v) −x1 + 2x2 + v . where Gw = ∂w0 w−1Gw0, and [Ωw] = Gw(x; y; v), Gw0(x; y; v) = 1 2 (x3 1 − x1 y2 1 y1 + x1 y2 1 − 2 x2 1 v − 7 x1 y1 v + x1 y2 v + 2 y2 +5 x2 +8 x1 v2 − 6 y1 v2 + 2 y2 v2 + 4 v3) 2 + x1 y1 y2 − y2 1 y2 + y1 y2 2 1 v + y1 y2 v − 2 y2 2 v ×(x2 1 + x1 y1 + y1 y2 − y2 2 + x1 v + y2 v)(x2 − x1 − y2 + v). 7. Variations Any formula for the class of a degeneracy locus depends on a choice of repre- sentative modulo the ideal defining the cohomology ring; here we discuss some alternative formulas. In type A, the Schubert polynomials of Lascoux and Schutzenberger are generally accepted as the best polynomial representatives for Schubert classes and degeneracy loci: they have many remarkable geometric and combinatorial (and aesthetic) properties. In other classical types, several choices have been proposed -- see, e.g., [Bi-Ha, La-Pr, Kr-Ta, Fo-Ki, Fu2] -- but Fomin and Kirillov [Fo-Ki] gave examples showing that no choice can satisfy all the properties possessed by the type A polynomials, if one insists on having polynomials in the Chern roots of the tautological bundles. We suggest that an investigation of alternative G2 formulas could shed some light on the problem for classical types, and vice versa. On one hand, possibilities for G2 formulas impose some limitations on what one might hope to find for general Lie types; on the other hand, one can use known formulas for type B3, together with the embedding of F lγ in F lβ, to obtain new G2 formulas. We explore the latter point of view at the end of this section, using the polynomials of [Bi-Ha]. Proposition 7.1 (cf. [Gr1]). Let (2 x1 − x2 − y1 + 2 y2)(2 x1 − x2 − y1 − y2)(x1 − 2 x2 + y1 + y2) 2 + 2 x3 2 − 2 y3 1 + 3 y2 1y2 + 3 y1y2 2 − 2 y3 2). 1 54 ×(2 x3 eGw0(x; y) = Then [Ωw0] = eGw0(x; y) in H ∗Flγ(V ). 1 − 3 x2 1x2 − 3 x1x2 24 DAVE ANDERSON Proof. Up to a change of variables, this is proved in [Gr1]. (To recover Graham's notation, set (7.1) ξ1 = 1 ξ2 = 1 ξ3 = − 1 3 (2x1 − x2), 3 (−x1 + 2x2), 3 (x1 + x2), η1 = − 1 η2 = − 1 η3 = 1 3 (2y1 − y2), 3 (−y1 + 2y2), 3 (y1 + y2), and replace ξ, η with x, y.) (cid:3) η1η2η3). This led him to suggest that 1 2 (ξ1−η2)(ξ1−η3)(ξ2−η3)(ξ1ξ2ξ3+ 2 (ξ1ξ2ξ3 + η1η2η3) might be an integral Remark 7.2. In Graham's notation, eGw0 = − 27 class. In fact, only 27 times this class is integral: Taking [Ωw]T = eGw(x; t) = ∂w0w−1eGw0(x; t), we compute 27(cid:0)3[Ωtst]T + 3(t1 + t2)[Ωst]T + (t1 + t2)(2t1 − t2)[Ωt]T(cid:1) 1 2 in H ∗ (In fact, the two sides are equal as polynomials, not just as classes.) Since the equivariant Schubert classes [Ωw]T form a basis for H ∗ T (pt, Z) = Z[t1, t2], the right-hand side cannot be integral. T (F lγ, Q); here the t's are related to the η's as in (7.1). T (F lγ, Z) over H ∗ (ξ1ξ2ξ3 + η1η2η3) = − 1 It is interesting to note that the integral class − 27 2 (ξ1ξ2ξ3 + η1η2η3) is positive in the sense of [Gr2, Theorem 3.2]: the coefficients in its Schubert expansion are nonnegative combinations of monomials in the positive roots. It is therefore natural to ask whether this is the equivariant class of a T -invariant subvariety of F lγ. In fact, it is the class of a T -equivariant embedding of SL3/B.1 Remark 7.3. Graham's polynomial yields a simpler formula for the case where γ takes values in the trivial bundle, but det V = M is not necessarily trivial. (In this case, recall that M ⊗3 is trivial.) Making the substitutions xi 7→ xi + v and yi 7→ yi − v, with 3 v = 0, we obtain [Ωw0] = 1 54 ×(2 x3 (2 x1 − x2 − y1 + 2 y2)(2 x1 − x2 − y1 − y2)(x1 − 2 x2 + y1 + y2) 2 + v3). 1x2 − 3 x1x2 1y2 + 3 y1y2 1 − 3 x2 2 + 2 x3 1 + 3 y2 2 − 2 y3 2 − 2 y3 There is a more transparent choice of polynomial representative for [Ωw0] ∈ H ∗F lγ (i.e., the case where the base is a point): The class of a point in the 5-dimensional quadric Q is 1 1. Since F lγ is a P1 bundle over Q, and x2 is the Chern class of the universal quotient bundle, the class of a point in F lγ is [Ωw0] = 1 2 x5 Starting from Gw0 = 1 1x2, one can compute polynomials Gw for Schubert classes [Ωw] using divided difference operators. The resulting formulas are dis- played in Table 1. 2 x5 2 x5 1x2. 1This embedding projects to a P2 ⊂ G. It is different from the embeddings of SL3/B corresponding to the inlcusion of Lie algebras sl3 ⊂ g2, which project to P2's in Q. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 25 w w0 ststs tstst tsts stst sts tst ts st s t id 2)x1x2 2x1 + x3 2)x2 1 2 + x1x3 2)x1 2 + 2x3 2 2 + x4 2) 1x2 1 + x2 1x2 + x2 Gw 1 2 x5 1 2 x5 1 1 2 (x3 1 + x2x2 1 1 − 3x1x2 + 3x2 2 (4x2 1 1x2 1 + x3 2 (x4 1 1 − 3x1x2 + 3x2 2 (4x2 2x3 2 x1x2 3x2 2x2 x1 x1 + x2 1. 1 + 1 1 − 2x1x2 + 2x2 2 1 − x1x2 + 2x2 2 2 x2 1x2 + 1 Table 1. Schubert polynomials for F lγ. The polynomials Gw computed from Gw0 = 1 1x2 have negative coefficients. In fact, it is impossible to find a system of positive polynomials using divided difference operators. In this respect, the problem of "G2 Schubert polynomials" is worse than the situation for types B and C: they cannot even satisfy two of Fomin-Kirillov's conditions [Fo-Ki].2 Specifically, we have the following: 2 x5 Proposition 7.4. Let {Pw w ∈ W } be a set of homogeneous polynomials in the variables x1 and x2, with deg Pw = ℓ(w). Suppose ∂sPw = ( Pws when ℓ(w s) < ℓ(w); when ℓ(w s) > ℓ(w) 0 and ∂tPw = ( Pwt when ℓ(w t) < ℓ(w); when ℓ(w t) > ℓ(w). 0 Then for some w, Pw has both positive and negative coefficients. 2To be precise, the conditions we consider are [Fo-Ki, (3)] and a stronger version of [Fo-Ki, (1)]. 26 DAVE ANDERSON Proof. One just calculates, starting from Pid = 1, and finds that the positivity requirement leaves no choice in the polynomials up to degree 4: Pw0 = ? Pststs =? Ptsts =? Psts = 1 2 x3 1 Pts = x2 1 Ps = x1 Ptstst =? Pstst = 1 2 x2 1x2 2 Ptst = 1 2 (x2 1x2 + x1x2 2) Pst = 1 2 (x2 1 + x1x2 + x2 2) Pt = x1 + x2 Pid = 1. 1 +bx3 However, no degree 4 polynomial P = Ptsts satisfies all the hypotheses. (Indeed, if P = ax4 2, then ∂tP = 0 implies d = −2e and b+c+d+e = 0, hence d = e = b = c = 0. On the other hand, ∂sP = 1 2) requires a = e and b − d = 1 (cid:3) 2 , which is inconsistent with b = c = d = e = 0.) 1x2 +· · ·+ex4 2 (x2 1x2 + x1x2 In spite of this, one might look for polynomials which are positive in some other set of variables. One natural choice is to use x2 and x3 = x1 − x2; in fact, the polynomials computed from Gw0 = 1 2 (x2 + x3)5x2 are positive in these variables. 1x2 = 1 2 x5 Finally, we consider an approach to G2 Schubert polynomials using the poly- nomials BBH w defined for type B flag varieties by Billey and Haiman [Bi-Ha]. The Billey -- Haiman polynomials involve a plethystic substitution, and the geometric meaning of the new variables is not altogether clear. The change of variables has considerable advantages, though: the resulting polynomials are canonically defined, and they satisfy the conditions of [Fo-Ki] for the classical types. Using the embedding of F lγ(V ) in F lβ(V ), one can restrict the Billey -- Haiman polynomials for type B3 to obtain classes in H ∗F lγ. In fact, the polynomial BBH 2 3 1 restricts to the class [Ωw0], and applying the G2 divided difference op- erators yields the appealing formulas displayed in Table 2; note that they are positive, have integral coefficients, and involve combinations of symmetric func- tions and type A Schubert polynomials.3 However, they do not multiply exactly 3One of the other five degree 6 type B3 polynomials also restricts to [Ωw0 ], namely BBH 2 1 3, but the resulting formulas are not positive. The remaining four polynomials restrict to zero in H ∗F lγ. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 27 as Schubert classes, as one sees by comparing GBH difference lies in the ideal defining H ∗F lγ.) s · GBH t with GBH ts + GBH st . (The To interpret these formulas, set z1 = x2 − x1 and z2 = −x2. For a strict partition λ, Pλ = Pλ(eX) is the Schur P -function in a new set of variables eX = (ex1,ex2, . . .), related to Z = (z1, z2, . . .) by the substitution pk(eX) = −pk(Z)/2. One can compute the table using an analogue of [Bi-Ha, Corollary 4.5] to expand ∂sPλ. w GBH w P42 + P32z1 w0 ststs P32 tstst tsts stst sts tst ts st s t id 2P41 + (P4 + 2P31)z1 + P3z2 1 P4 + 2P31 + P3(z1 + z2) 2P31 + (P3 + 2P21)z1 + P2z2 1 P3 + 2P21 + P2(z1 + z2) 3P3 + 4P21 + 4P2z1 + 2P2(z1 + z2) + 4P1z2 4P2 + 4P1(z1 + z2) + (z1 + z2)2 3P2 + 2P1z1 + 2P1(z1 + z2) + z2 2P1 + (z1 + z2) 4P1 + z1 + 2(z1 + z2) 1 1 + z1z2 1 + 2P1z1z2 + z3 1 + z2 1z2 Table 2. Type G2 "Billey -- Haiman" polynomials, computed from GBH w0 = BBH 2 3 1. Appendix A. Lie theory In this appendix, we recall general facts about representation theory and ho- mogeneous spaces for linear algebraic groups, and apply them show the above description of the G2 flag variety agrees with the Lie-theoretic one. Proposi- tions A.1 and A.3 are the basic representation-theoretic facts relating G2 to compatible forms; their proofs are given in [An1, Appendix A]. Proposition A.5 identifies the γ-isotropic flag variety F lγ with the homogeneous space G2/B. All the remaining facts are standard, and can be found in e.g. [Fu-Ha], [Hu1], [Hu2], [De]. A.1. General facts. Let G be a simple linear algebraic group, fix a maximal torus and Borel subgroup T ⊂ B ⊂ G, and let W = N (T )/T be the Weyl group. Let R, R+, and ∆ be the corresponding roots, positive roots, and simple roots, respectively. For α ∈ ∆, let sα ∈ W be the corresponding simple reflection, and also write sα ∈ N (T ) for a choice of lift; nothing in what follows will depend on 28 DAVE ANDERSON the choice. For a subset S ⊂ ∆, let PS be the parabolic subgroup generated by B and {sα α ∈ S}. (Such parabolic subgroups are called standard.) Writebı = ∆ r {si}, so Pbı is the maximal parabolic in which the ith simple root is omitted. ∼= Gr(2, 5).) Write g, b, t, p, for the corresponding Lie (For example, SL5/Pb2 algebras. The length of an element w ∈ W is the least number ℓ = ℓ(w) such that w = s1 · · · sℓ (with sj = sαj for some αj ∈ ∆); such a minimal expression for w is called a reduced expression. Let w0 be the (unique) longest element of W . The Bruhat order on W is defined by setting v ≤ w if there are reduced expressions v = sβ1 · · · sβℓ(v) and w = sα1 · · · sαℓ(w) such that the β's are among the α's. For each w ∈ W , there is a Schubert cell X o w = BwB/B in G/B, of dimension ℓ(w). The Schubert varieties Xw are the closures of cells, and Xv ⊆ Xw iff v ≤ w. The irreducible representations of G are indexed by dominant weights; write Vλ for the representation corresponding to the dominant weight λ. In charac- teristic 0, if pλ ∈ P(Vλ) is the point corresponding to a highest weight vector, then G · pλ is the unique closed orbit, and is identified with G/PS(λ), where S(λ) is the set of simple roots orthogonal to λ with respect to a W -invariant inner product. In positive characteristic, G/PS(λ) can still be embedded in P(V ) for some representation with highest weight λ, but V need not be irreducible. (See [Hu2, §31] for these facts about representations in arbitrary characteristic.) A.2. Representation theory of G2. The root system of type G2 has simple roots α1 and α2 (with α2 the long root), and positive roots α1, α2, α1+α2, 2α1+ α2, 3α1 + α2, 3α1 + 2α2. The lattice of abstract weights is the same as the root lattice (cf. [Hu2, §A.9]); it follows that up to isomorphism, there is only one simple group of type G2 (over the algebraically closed field k). From now on, let G denote this group, and fix T ⊂ B ⊂ G corresponding to the root data. By Proposition 3.2, G ∼= Aut(C), where C is the unique octonion algebra over k. Let V = e⊥ ⊆ C be the imaginary subspace. The dominant Weyl chamber for this choice of positive roots is the cone spanned by α4 and α6; denote these fundamental weights by ω1 and ω2, re- spectively. One checks that V has highest weight ω1, and is irreducible for char(k) 6= 2, so V = Vω1 is the minimal irreducible representation, called the standard representation of G.4 The adjoint representation g has highest weight ω2. (This is irreducible if char(k) = 0, but not if char(k) = 3.) Over any field, one has g ⊆V2 V . Let γ be the alternating trilinear form on V ⊂ C induced by the multi- plication, let {f1, . . . , f7} be the standard γ-isotropic basis (3.11). From the description of G as the automorphisms of C, it is clear that G preserves γ. In fact, the converse is almost true: 4If char(k) = 2, the representation V = e⊥ ⊂ C contains an invariant subspace spanned by e. In this case, the irreducible representation Vω1 = V /(k · e) is 6-dimensional [Sp-Ve, §2.3]. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 29 Proposition A.1. Choose a basis {f1, . . . , f7} for V , and let γ ∈ V3 V ∗ be given by γ = f ∗ 156 − f ∗ 237, 147 + f ∗ 246 + f ∗ 345 − f ∗ as in (3.13). Let G(γ) ⊂ GL(V ) be the stabilizer of γ under the natural action, and let SG(γ) = G(γ) ∩ SL(V ). Then SG(γ) is simple of type G2, and G(γ) = µ3 × SG(γ). Moreover, the orbit GL(V ) · γ is open inV3 V ∗. For k = C, this is well known; see [Br, §2] or [Fu-Ha, §22]. For arbitrary fields, see [An1, Propositions 6.1.4 and A.2.2], and compare [As, (3.4)]. The proof of this proposition also shows the following: Corollary A.2. Let V , γ, and SG(γ) be as in Proposition A.1. Then SG(γ) acts irreducibly on V . (cid:3) Note that w0 ∈ W acts on the weight lattice by multiplication by −1. This implies that every irreducible representation of G is isomorphic to its dual. Using Schur's lemma, there is a unique (up to scalar) G-invariant bilinear form on each irreducible representation [Hu2, §31.6]. In particular, we have the following: Proposition A.3. Let V be a 7-dimensional vector space, with nondegenerate trilinear form γ :V3 V → k. Then γ determines a compatible form β uniquely up to scaling by a cube root of unity. (cid:3) Remark A.4. In characteristic 0, the description of G2 (or g2) as the stabilizer of a generic alternating trilinear form is due to Engel, who also found an invariant symmetric bilinear form. For a history of some of the early constructions of G2, see [Ag]. A.3. The Weyl group. The Weyl group of type G2 is the dihedral group with 12 elements. Let α1 and α2 be the simple roots, and let s = sα1 and t = sα2 be the corresponding simple reflections generating W = W (G2). Thus W has a presentation hs, t s2 = t2 = (st)6 = 1i. With the exception of w0, each element of W (G2) has a unique reduced expression. The Hasse diagram for Bruhat order is as follows: w0 = 7 6 (tststs = ststst) (ststs) 7 5 (tsts) 6 3 (sts) 5 2 (ts) 3 1 (s) 2 1 6 7 (tstst) 5 7 (stst) 3 6 (tst) 2 5 (st) 1 3 (t) id = 1 2 The indexing w = w(1) w(2), for 1 ≤ w(1), w(2) ≤ 7, arises as follows. There is an embedding W (G2) ֒→ W (A6) = S7, given by s 7→ τ12τ35τ67 and t 7→ τ23τ56, 30 DAVE ANDERSON where τij is the permutation transposing i and j. (This also factors through W (B3).) Thus each w is identified with a permutation w(1) w(2) · · · w(7), and in fact, the full permutation is determined by w(1) w(2). This inclusion of Weyl groups corresponds to the inclusion G2 ֒→ SL7 deter- mined by the basis {f1, . . . , f7} for V = Vω1 and the trilinear form γ of (3.13), together with the inclusion of tori (z1, z2) 7→ (z1, z2, z1z−1 2 , z−1 1 ). Thus a natural way to extend w ∈ W to a full permutation is as follows. Given w(1) w(2), let w(3) be the number such that Efw(1) = hfw(1), fw(2), fw(3)i as in §4.1. Then define w(4), . . . , w(7) by requiring w(i) + w(8 − i) = 8. For example, 6 3 extends to 6 3 7 4 1 5 2. Note that (w · w0)(i) = 8 − w(i). 2 , 1, z−1 1 z2, z−1 All this can be summarized in the following diagram: 6 6 3 6 7 7 6 7 7 5 3 3 6 3 1 4 5 7 5 2 5 1 3 1 t 1 2 2 1 s 2 5 2 A.4. Homogeneous spaces. We can now identify the homogeneous spaces for G2. We take G = Aut(C) for an octonion algebra C, as above, and let β and γ be the corresponding compatible forms on the imaginary subspace V ⊂ C. From the root data, one sees dim G = 14, dim B = 8, dim Pb1 = dim Pb2 = 9, and dim T = 2. Thus dim G/B = 6 and dim G/Pb1 = dim G/Pb2 = 5. Proposition A.5. Let F lγ, Q, and G be as in §4. Then Q ∼= G/Pb1, G ∼= G/Pb2, and F lγ ∼= G/B. Proof. The homogeneous spaces G/Pb1 and G/Pb2 are the closed orbits in P(V ) and P(g), respectively. Since G preserves β, G/Pb1 must be contained in the quadric hypersurface Q ⊂ P(V ), but dim G/Pb1 = 5, so it is all of Q. To see G/Pb2 = G, note that G/Pb2 ⊂ P(g) ⊂ P(V2 V ), so G/Pb2 ⊂ Gr(2, 7). Since G preserves γ, we must have G/Pb2 ⊆ G; thus it will suffice to show G is irreducible and 5-dimensional. For this, consider F lγ = {(p, ℓ) p ∈ ℓ} ⊂ Q × G, and notice that the first projection identifies F lγ with the P1-bundle P(S3/S1) → Q, so F lγ is smooth and irreducible of dimension 6. On the other hand, the second projection is obviously a P1-bundle. Finally, since F lγ is a 6-dimensional G-invariant subvariety of G/Pb1 × G/Pb2, (cid:3) it follows that F lγ = G/B. Remark A.6. A similar description of G/Pb2, among others, can be found in [La-Ma]. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 31 Proposition A.7. Let i : G ֒→ G′ be an inclusion of semisimple algebraic groups, and let B ⊂ G and B′ ⊂ G′ be Borel subgroups with i(B) ⊂ B′. Also denote by i the induced inclusions of flag varieties G/B ֒→ G′/B′ and Weyl groups W ֒→ W ′. Then for each w ∈ W , the Schubert cells are related by BwB/B = (B′i(w)B′/B′) ∩ (G/B). More generally, let P ⊂ G and P ′ ⊂ G′ be parabolic subgroups such that P = P ′ ∩ G. Then the same conclusion holds for G/P ֒→ G′/P ′, that is, BwP/P = (B′i(w)P ′/P ′) ∩ (G/P ) for all w ∈ W . A.5. The Borel map and divided differences. Let M ⊂ t∗ be the weight lattice. For general G/B, there is a Borel map c : Sym∗ M → H ∗(G/B) induced by the Chern class map c1 : M → H 2(G/B), where M ⊂ t∗ is the weight lattice. More precisely, this map is defined as follows. Identify M with the character group of B, and associate to χ ∈ M the line bundle Lχ = G ×B C. Then c1(χ) is defined to be c1(Lχ). In fact, c1 is an isomorphism, and this induces an action of W in the evident way: for w ∈ W and x = c1(χ) ∈ H 2(G/B), define w · x = c1(w · χ). (See [BGG, De].) The Borel map becomes surjective after extending scalars to Q, and defines an isomorphism H ∗(G/B, Q) ∼= Sym∗ MQ/I, where I = (Sym∗ MQ)W + is the ideal of positive-degree Weyl group invariants. For a simple root α, define the divided difference operator ∂α on H ∗(G/B) by (A.1) ∂α(f ) = f − sα · f α . These act on Schubert classes as follows [De]: (A.2) ∂α[Ωw] = ( [Ωw sα] when ℓ(w sα) < ℓ(w); when ℓ(w sα) > ℓ(w). 0 In particular, [Ωsα] can be identified with the weight at the intersection of the hyperplanes orthogonal to α and the (affine) hyperplane bisecting α. In the case of G2 flags, we know [Ωs] = x1 and [Ωt] = x1 + x2. Looking at the root diagram, then, we see x1 = α4 and x2 = α3. Therefore α1 = x1 − x2, α2 = −x1 + 2x2, and s · x1 = x2, s · x2 = x1, t · x1 = x1, t · x2 = x1 − x2. With these substitutions, the operators of (A.1) agree with those defined in §6 ((2.4) and (2.5)). 32 DAVE ANDERSON Appendix B. Integral Chow rings of quadric bundles In this appendix, we consider schemes over an arbitrary field k, and use the language of Chow rings rather than cohomology. We prove the following fact about odd-rank quadric bundles: Theorem B.1. Let V be a vector bundle of rank 2n + 1 on a scheme X, and suppose V is equipped with a nondegenerate quadratic form. Assume there is p a maximal (rank n) isotropic subbundle F ⊂ V . Let Q −→ X be the quadric bundle of isotropic lines in V , let h ∈ A∗Q be the hyperplane class (restricted from H = c1(O(1)) ∈ A∗P(V )), and let f = [P(F )] ∈ A∗Q. Then A∗Q = A∗X[h, f ]/I, where the ideal I is generated by the two relations (B.1) (B.2) 2f = hn − c1(F ) hn−1 + · · · + (−1)ncn(F ), f 2 = (cn(V /F ) + cn−2(V /F ) h2 + · · · ) f. (Here h and f have degrees 1 and n, respectively.) A similar presentation for even-rank quadrics was first given by Edidin and Graham [Ed-Gr, Theorem 7]; in fact, the second of the two relations is the same as theirs. Our purpose here is to correct a small error in the statement of the second half of their theorem (which concerned odd-rank quadrics). Before giving the proof, we recall two basic formulas for Chern classes. Let L be a line bundle. For a vector bundle E of rank n, we have (cf. [Fu4, Ex. 3.2.2]) (B.3) Also, if cn(E ⊗ L) = nXi=0 ci(E) c1(L)n−i. 0 → L → E → E′ → 0 is an exact sequence of vector bundles, then inverting the Whitney formula gives (B.4) ck(E′) = ck(E) − ck−1(E) c1(L) + · · · + (−1)kc1(L)k. Proof. The classes h, h2, . . . , hn−1, f, f h, . . . , f hn−1 form a basis of A∗Q as an A∗X-module, since they form a basis when restricted to a fiber. It is easy to see that these elements also form a basis of the ring A∗X[h, f ]/I. Therefore it suffices to establish that the relations generating I hold in A∗Q. Let i : Q ֒→ P(V ) be the inclusion of the quadric in the projective bundle. By [Fu4, Ex. 3.2.17], we have n+1Xi=0 in A∗P(V ), where ci = ci(V /F ). (Following the common abuse of notation, we have written ci for p∗ci.) On the other hand, Q ⊂ P(V ) is cut out by a section i∗f = [P(F )] = ci H n+1−i CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 33 of OP(V )(2), so [Q] = 2 H in A∗P(V ). Therefore i∗i∗f = 2 h f , and we have (B.5) 2 h f = hn+1 + c1 hn + · · · + cn+1. (Up to this point, we are repeating the argument of [Ed-Gr].) To prove the first relation, expand hn in the given basis: (B.6) with ak ∈ AkX. Our goal is to show a0 = 2, and ak = (−1)k+1ck(F ) for k > 0. That a0 = 2 can be seen by restricting to a fiber: the Chow ring of an hn = a0 f + a1 hn−1 + · · · + an, odd-dimensional quadric in projective space is given by Z[h, f ]/(hn − 2f, f 2). Multiplying (B.6) by h and expanding in the basis, we have hn+1 = 2 h f + 2 a1 f + (a2 + a2 1) hn−1 + · · · + (an + a1 an−1) h + a1 an. On the other hand, if we rearrange and expand (B.5), we obtain hn+1 = 2 h f − 2 c1 f − (c2 + c1 a1)hn−1 − · · · − (cn + c1 an−1) h − (cn+1 + c1 an). Comparing coefficients, we have 2 a1 = −2 c1; ak = −ck − ak−1(a1 + c1) (2 ≤ k ≤ n); a1 an = −cn+1 − c1 a1. From the first of these equations, we see a1 + c1 = τ, for some τ ∈ A1X such that 2 τ = 0. (Note that τ = 0 only if cn+1(V /F ) = 0, which need not be true in general.) The remaining equations give (B.7) ak = −ck + ck−1 τ − ck−2 τ 2 + · · · − (−1)kτ k (1 ≤ k ≤ n), and −cn+1 = an τ . (Of course, the signs on powers of τ make no difference, but we will include them as a visual aid.) We claim τ = c1(F ⊥/F ). This can be proved in the universal case. Specifying the maximal isotropic subbundle F ⊂ V reduces the structure group from O2n+1 to a parabolic subgroup whose Levi factor is GLn × Z/2Z, so the universal base is (an affine bundle over) BG = BGLn × BZ/2Z. Every such maximal isotropic classifying space BG. More precisely, one should use Totaro's algebraic model Now A∗(BGLn × BZ/2Z) ∼= Z[c1, . . . , cn, t]/(2t), so there is only one nonzero model, one may have to replace X by an affine bundle or Chow envelope, as in [Gr1, p. 486]. subbundle F ⊂ V on X is pulled back from a universal subbundle eF ⊂ eV on the for BG; to ensure V is pulled back from the corresponding eV on the algebraic 2-torsion class of degree 1, namely t. Since t = c1(eF ⊥/eF ), it pulls back to explained in [To], as is its computation. To see that t = c1(eF ⊥/eF ), note that GLn × Gm ⊂ GL2n+1, corresponding to the splitting eV ∼= eF ⊕ (eF ⊥/eF ) ⊕ eF ∗.) c1(F ⊥/F ), so the claim is proved. (The meaning of the Chow ring of BG is the inclusion GLn × Z/2Z ⊂ O2n+1 ⊂ GL2n+1 also factors as GLn × Z/2Z ⊂ 34 DAVE ANDERSON Using the exact sequence 0 → F ⊥/F → V /F → V /F ⊥ → 0 and Formula (B.4), Equation (B.7) implies ak = −ck(V /F ⊥). Since V /F ⊥ ∼= F ∗, we obtain ak = (−1)k+1ck(F ), as desired. The second relation is proved by the argument given in [Ed-Gr]. Let j : P(F ) ֒→ Q be the inclusion, and let NP(F )/Q be the normal bundle. By the self- intersection formula, j∗cn(NP(F )/Q) = f 2. On the other hand, using NQ/P(V ) = O(2) and NP(F )/P(V ) = V /F ⊗ O(1), and tensoring with O(−1), we have 0 → O(1) → V /F → NP(F )/Q ⊗ O(−1) → 0 on P(F ); thus NP(F )/Q = ((V /F )/O(1)) ⊗ O(1). By Formulas (B.3) and (B.4), we have cn(NP(F )/Q) = cn(V /F ) + cn−2(V /F ) h2 + · · · . The relation (B.2) follows after applying j∗. (cid:3) [Ag] [An1] [An2] [As] [Ba] [BGG] [Bi-Ha] [Bo-Sa] [Br] [Co] [De] [Ed-Gr] [Fe-Ri] [Fo-Ki] [Fu1] [Fu2] [Fu3] References I. Agricola, "Old and new on the exceptional group G2," Notices Amer. Math. Soc. 55 (2008), no. 8, 922 -- 929. D. Anderson, Degeneracy loci and G2 flags, Ph. D. thesis, University of Michigan, 2009, http://www.lsa.math.umich.edu/∼dandersn/papers/thesis.pdf. D. Anderson, "Degeneracy of triality-symmetric morphisms," arXiv:0901.1347. M. Aschbacher, "Chevalley groups of type G2 as the group of a trilinear form," J. Algebra 109 (1987), 193 -- 259. J. Baez, "The octonions," Bull. Amer. Math. Soc. 39 (2002), 145 -- 205. I. Bernstein, I. Gelfand, and S. Gelfand, "Schubert cells and cohomology of the spaces G/P ," Russian Math. Surveys 28 (1973), 1 -- 26. S. Billey and M. Haiman, "Schubert polynomials for the classical groups," J. Amer. Math. Soc. 8, no. 2 (1995), 443 -- 482. R. Bott and H. Samelson, "The integral cohomology ring of G/T ," Proc. Nat. Acad. Sci. 7 (1955), 490 -- 493. R. L. Bryant, "Metrics with exceptional holonomy," Ann. of Math. 126 (1987), no. 3, 525 -- 576. H. S. M. Coxeter, "Integral Cayley numbers," Duke Math. J. 13 (1946), 561 -- 578. M. Demazure, "D´esingularisation des vari´et´es de Schubert g´en´eralis´ees," Ann. Sci. cole Norm. Sup. 7 (1974), 53 -- 88. D. Edidin and W. Graham, "Characteristic classes and quadric bundles," Duke Math. J. 78 (1995), 277 -- 299. L. Feh´er and R. Rim´anyi, "Schur and Scubert polynomials as Thom polynomials -- cohomology of moduli spaces," Cent. European J. Math. 4 (2003), 418 -- 434. S. Fomin and A. N. Kirillov, "Combinatorial Bn-analogues of Schubert polynomi- als," Trans. Amer. Math. Soc. 348 (1996), no. 9, 3591 -- 3620. W. Fulton, "Flags, Schubert polynomials, degeneracy loci, and determinantal for- mulas," Duke Math. J. 65 (1992), no. 3, 381 -- 420. W. Fulton, "Schubert varieties in flag bundles for the classical groups," in Proceed- ings of the Hirzebruch 65 Conference on Algebraic Geometry, 241 -- 262, Israel Math. Conf. Proc., 9, 1996. W. Fulton, "Determinantal formulas for orthogonal and symplectic degeneracy loci," J. Differential Geom. 43 (1996), no. 2, 276 -- 290. CHERN CLASS FORMULAS FOR G2 SCHUBERT LOCI 35 [Fu4] [Fu-Ha] W. Fulton and J. Harris, Representation Theory: A First Course, Springer-Verlag, W. Fulton, Intersection Theory, second edition, Springer, 1998. 1991. [Fu-Pr] W. Fulton and P. Pragacz, Schubert Varieties and Degeneracy Loci, Springer, 1998. S. Garibaldi, "Structurable algebras and groups of type E6 and E7," J. Algebra [Ga] 236 no. 2 (2001), 651 -- 691. W. Graham, "The class of the diagonal in flag bundles," J. Differential Geometry 45 (1997), 471 -- 487. W. Graham, "Positivity in equivariant Schubert calculus," Duke Math. J. 109 (2001), 599 -- 614. [Gr2] [Gr1] [Gri-Ram] S. Griffeth and A. Ram, "Affine Hecke algebras and the Schubert calculus," Euro- [Ha-Tu] [Ha] [Hu1] [Hu2] [Il-Ma] pean J. of Combin. 25 (2004), no. 8, 1263 -- 1283. J. Harris and L. Tu, "On symmetric and skew-symmetric determinantal varieties," Topology 23 (1984) No. 1, 71 -- 84. F. R. Harvey, Spinors and Calibrations, Academic Press, Inc., Boston, 1990. J. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer, 1972. J. Humphreys, Linear Algebraic Groups, Springer, 1981. A. Iliev and L. Manivel, "The Chow ring of the Cayley plane," Compos. Math. 141 (2005), no. 1, 146 -- 160. [Kn-Mi] [Kr-Ta] [La-Ma] [J´o-La-Pr] T. J´ozefiak, A. Lascoux, and P. Pragacz, "Classes of determinantal varieties asso- ciated with symmetric and skew-symmetric matrices," Izv. Akad. Nauk SSSR Ser. Mat. 45 (1981), no. 3, 662 -- 673. A. Knutson and E. Miller, "Grobner geometry of Schubert polynomials," Ann. Math., 161 (2005), no. 3, 1245 -- 1318. A. Kresch and H. Tamvakis, "Double Schubert polynomials and degeneracy loci for the classical groups," Ann. Inst. Fourier (Grenoble) 52 (2002), no. 6, 1681 -- 1727. J. M. Landsberg and L. Manivel, "The sextonions and E7 1 (2006), no. 1, 143 -- 179. A. Lascoux and P. Pragacz, "Operator calculus for eQ-polynomials and Schubert polynomials," Adv. in Math. 140 (1998), 1 -- 43. I. G. Macdonald, Notes on Schubert Polynomials, Publ. LACIM 6, Univ. de Qu´ebec `a Montr´eal, Montr´eal, 1991. A. Parusinski and P. Pragacz, "Chern-Schwartz-MacPherson classes and the Euler characteristic of degeneracy loci and special divisors," J. Amer. Math. Soc. 8 (1995), 793 -- 817. H. P. Petersson, "Composition algebras over algebraic curves of genus zero," Trans. Amer. Math. Soc. 337 (1993), 473 -- 493. S. Pumplun, "Albert algebras over curves of genus zero and one," arXiv:0709.2308. T. A. Springer and F. D. Veldkamp, Octonions, Jordan Algebras, and Exceptional Groups, Springer, 2000. B. Totaro, "The Chow ring of a classifying space," Proc. Sympos. Pure Math. 67 (1999), AMS, 249 -- 281. ," Adv. Math. 201 [Pu] [Sp-Ve] [To] 2 [La-Pr] [Ma] [Pa-Pr] [Pe] Department of Mathematics, University of Michigan, Ann Arbor, MI 48109 E-mail address: [email protected]
1212.0991	1	1212	2012-12-05T10:51:41	The Bertini involution	[ "math.AG" ]	We summarize and extend E. Moody's results on the explicit equations related to the Bertini involution.	math.AG	math	THE BERTINI INVOLUTION ALEX DEGTYAREV Abstract. We summarize and extend E. Moody's results on the explicit equa- tions related to the Bertini involution. These notes are the result of my attempt to understand E. Moody's paper [1]. I correct a few misprints in [1] and take the computation a bit further. I express my admiration to Ethel I. Moody, who managed to perform this tedious computation in the pre-Maple era. A Maple implementation of most equations is found at http://www.fen.bilkent.edu.tr/~degt/papers/Bertini.zip. This text is not intended as an 'official' publication; it is distributed in the hope that it may be useful. It can be cited by its arXiv location. Whenever possible, I try to keep the original notation of [1]. 1.1. The results of [1]. Consider the pencil of cubics 1. The Bertini involution (1.1) where λw(x) + µw′(x) = 0, w(x) = x2 3(a1x1 + a2x2) + x3(b1x2 1 + b2x1x2 + b3x2 2) + (c1x2 1x2 + c2x1x2 2) and similar for w′, so that the coordinate vertices are amongst the basepoints of the pencil. The point (0 : 0 : 1) will play a special role. The curve of the pencil passing through a point y is given by (1.2) Clearly, W3(x) := w(x)w′(y) − w′(x)w(y) = 0. W3(x) = x2 3(A1x1 + A2x2) + x3(B1x2 1 + B2x1x2 + B3x2 2) + (C1x2 1x2 + C2x1x2 2), where Ai(y) := aiw′(y) − a′ iw(y) and similar for Bi, Ci. The tangent to (1.2) at (0 : 0 : 1) meets the curve again at r = (r1 : r2 : r3), where (1.3) r1 = A2r′ 1, r2 = −A1r′ 1, 2 − B2A1A2 + B3A2 1, 1 := B1A2 r′ r3 = A1A2r′ 3, r′ 3 := A2C1 − A1C2. The locus of these points is (1.4) γ4(y) := y1A1 + y2A2 = 0. 2000 Mathematics Subject Classification. 14E07. Key words and phrases. Bertini involution, Geiser involution. These notes are to be extended should there be any interesting development. They will be available at http://www.fen.bilkent.edu.tr/~degt/papers/papers.htm and on the arXiv. 1 2 ALEX DEGTYAREV Apart from the basepoints, the locus (1.4) meets (1.2) at a single point r. The line (ry) meets (1.2) at a third point z, and the Bertini involution can be defined as the map y 7→ z. Let κ := a1b′ C5(y) := A2[B1 + κy1y2 φ6(y) := A1C2 + y3C5(y), 1 − a′ 3]y2 + [A1 − κy2 1y3]y2[A2y3 + B3y2]y1 + κB3y1y3, 1b1 and ψ6(y) := A2C1 + y3C5(y). (Following [1], we use [e]u do indicate that e has a common factor u and this factor has been removed.) In these notations, the Bertini involution is (1.5) z1 = φ6[A2 2φ6 + B3r′ 1]y1 , z2 = ψ6[A2 1ψ6 + B1r′ 1]y2, z3 = ψ6φ6C5. Apart from the basepoint (0 : 0 : 1) of the pencil, the fixed point locus of this involution is the curve (1.6) K(y) := ψ6[A1y3 + B1y1]y2 − φ6[A2y3 + B3y2]y1 = 0. Remark 1.7. The expressions for r, C5, and K found in [1] contain a number of misprints. The corrections suggested are verified by the identities in §1.2 below, as well as by (2.2) and (2.3). 1.2. Further observations. The expression for the Bertini involution, see (3.1), is obtained by substituting z = lr + my and solving W3(lr + my) = 0, see (1.2), in l : m. (This equation is linear since W3(r) = W3(y) = 0.) Note that {ψ6 = 0} and {φ6 = 0} are the curves contracted to the basepoints (1 : 0 : 0) and (0 : 1 : 0), respectively. Hence, they can also be found from the identities y3r1 − y1r3 = A2γ4φ6, y2r3 − y3r2 = A1γ4ψ6. Besides, one has y1r2 − y2r1 = −r′ 1γ4. A point y is fixed by (3.1) if and only if the tangent at y to the member (1.1) of the pencil passing through y meets the curve again at r. In [1], the equation (1.6) of the fixed point locus is obtained by eliminating λ : µ from (1.1) and the polar conic to (1.1) with respect to r (after the substitution x 7→ y). Alternatively, K can be found as the common factor of y3z1 − y1z3 and y2z3 − y3z2, using the identities y3z1 − y1z3 = −φ6KA2, y2z3 − y3z2 = −ψ6KA1. Note that the rightmost factors are just two particular members of the pencil. 2. The map P2 99K Σ2 From now on, we assume that the distinguished basepoint (0 : 0 : 1) is simple. 2.1. The anti-bicanonical map. Let Y be the plane P2 blown up at all basepoints (including infinitely near) of the pencil other than (0 : 0 : 1). It is a (nodal, in general) Del Pezzo surface of degree 1, and the anti-bicanonical linear system maps Y to a quadric cone in P3. According to [1], the proper transforms of the sextics {φ6 = 0} and {ψ6 = 0} are in −2KY . Hence, the space of sections H 0(Y ; −2KY ) is generated by φ6 (or ψ6) and w2, ww′, w′2, and the map y 7→ ¯z ∈ P3 is given by ¯z0 = φ6(y), ¯z1 = w2(y), ¯z2 = w(y)w′(y), ¯z3 = w′2(y). Its image is the cone ¯z1 ¯z3 = ¯z 2 2. The passage to the affine coordinates ¯x := ¯z1/¯z2, ¯y := ¯z0/¯z2 blows up the vertex and maps the cone to the Hirzebruch surface Σ2 THE BERTINI INVOLUTION 3 with the exceptional section E of self-intersection (−2) (the exceptional divisor over the vertex). The composed rational map P2 99K Σ2 is (2.1) ¯x = w(y)/w′(y), ¯y = φ6(y)/w′2(y). Alternatively, Y with the remaining basepoint (0 : 0 : 1) blown up is a rational Jacobian elliptic surface: the elliptic pencil is (1.1) and the distinguished section is the exceptional divisor over (0 : 0 : 1). The Bertini involution becomes the fiberwise multiplication by (−1), and the quotient blows down to the Hirzebruch surface Σ2. The quotient map is generically two-to-one; its ramification locus is the union of the exceptional section E ⊂ Σ2 and a certain proper trigonal curve, viz. the image of {K = 0}. The pull-backs of the fibers of Σ2 are the anti-canonical curves in Y (i.q. the members of the original pencil (1.1) of cubics), and the pull-backs of the proper (i.e., disjoint from E) sections of Σ2 are the anti-bicanonical curves other than those representable in the form {α1w2 + α2ww′ + α3w′2 = 0}. 2.2. The ramification locus. Since {ψ6 = 0} is the pull-back of a section of Σ2, there must be a relation (after the substitution x 7→ y) of the form (2.2) ψ6 = φ6 + S2(w, w′), where S2 is a certain homogeneous polynomial of degree 2, see below. The curve {φ6 = 0} is contracted by the Bertini involution. Hence, the pull-back in Y of its image {¯y = 0} splits into two components (sections of the elliptic pencil), of which one is contracted by the blow down map Y → P2. It follows that the free term R2 3 in equation (2.3) below is indeed a perfect square. Since {K = 0} is the pull-back of the ramification locus (other than E), which is a proper trigonal curve, there must be a relation (2.3) K 2 = −4φ3 6 + φ2 6P2(w, w′) + φ6Q4(w, w′) + R2 3(w, w′), where P2, Q4, and R3 are certain homogeneous polynomials of degree 2, 4, and 3, respectively. Let S2(t1, t2) = P2 etc. The coefficients pi, qi, ri, si are found by a direct computation. They reduce to a remarkably simple form: i=0 siti 1t2−i 2 s0 = a2c1 − a1c2, r0 = −a1b2c2 + a1b3c1 + a2b1c2, q0 = 4(a1c2 − b1b3)s0 + 2b2r0, p0 = b2 2 − 4a2c1 − 4b1b3 + 8a1c2 and pi = (−1)i{p0}i, qi = (−1)i{q0}i, ri = (−1)i{r0}i, si = (−1)i{s0}i, where { · }m is defined as follows: {e}m is the sum of (cid:0) n factors with their primed versions. (For example, one has {a1c2}1 = a1c′ {b2 to homogeneous polynomials by linearity. if e is a degree n monomial in a1, . . . , c2, then m(cid:1) monomials, each obtained from e by replacing m of its n 1c2, 1c1.) This definition extends 2, and {a1b1c1}2 = a1b′ 2}1 = 2b2b′ 2 + a′ 1 + a′ 1b′ 1c′ 1b1c′ 1 + a′ Warning 2.4. The operation { · }m is used only to shorten the notation. As with the derivative, this operation should be performed before any substitution of any particular values of the coefficients (see, e.g., the substitution a1 = a2 = 0 in §3). 4 ALEX DEGTYAREV Remark 2.5. Observe that S2(w, w′) remains unchanged under the transformation ai ↔ a′ i. The same holds for P2(w, w′) and Q4(w, w′), whereas R3(w, w′) changes sign. i, wi ↔ w′ i, ci ↔ c′ i, bi ↔ b′ Problem 2.6. The symmetry in Remark 2.5 is easily explained by interchanging w and w′. However, is there a geometric explanation for the 'regular' behaviour of the other coefficients? Summarizing, we see that the map P2 99K Σ2 given by (2.1) takes the sextics {φ6 = 0} and {ψ6 = 0} to the sections {¯y = 0} and {¯y = −S2(¯x)}, respectively. The map is generically two-to-one (the deck translation being the Bertini involution), and its ramification locus in Σ2 is the union of E and the trigonal curve −4¯y3 + ¯y2P2(¯x) + ¯yQ4(¯x) + R2 3(¯x) = 0. As usual, we treat the homogeneous bivariate polynomials S2, P2, Q4, and R3 as univariate ones via S2(¯x) := S2(¯x, 1) = P2 Problem 2.7. Can one express coefficients si in terms of pi, qi, and ri? In other words, does a choice of the ramification locus in Σ2 and one of the sections select automatically the other section? i=0 si ¯xi etc. 2.3. Other sextics contracted by the involution. The basepoint (0 : 0 : 1) plays a special role in the definition of the Bertini involution. The other basepoints are not special. In particular, any other basepoint (u1 : u2 : u3) gives rise to a sextic {ψu 6 = 0} contracted to this point and to a splitting section of Σ2 whose pull-back this sextic is. Assuming that u1 6= 0 and normalizing the coordinates as (1, u2, u3), we have (2.8) where Su ψu 6 = φ6 + Su 2 (w, w′), 2 is a homogeneous polynomial of degree 2 whose coefficients su 0 = s0 + (a2c2u2 + (a2b2 − a1b3)u3) + a2b3u2u3 + a2 su 2u2 3, i are su i = (−1)i{su 0 }i. As above, the image of {ψu 6 = 0} in Σ2 is the section {¯y = −Su 2 (¯x)}. These equations are easily obtained by changing the coordinates and placing the basepoint in question to (1 : 0 : 0). 3. The Geiser involution Now, we assume that the pencil has exactly one basepoint infinitely near to the distinguished point (0 : 0 : 1). In other words, all members of (1.1) have a common tangent at (0 : 0 : 1) and, hence, exactly one of them has a double point at (0 : 0 : 1). We can assume that this singular member is {w(x) = 0}, thus letting a1 = a2 = 0. The resulting special case of the Bertini involution is the Geiser involution, see [1]. 3.1. The involution. Most formulas in §1.1 simplify dramatically. One obviously has A1 = −a′ 2w(y); hence, γ4 = wγ1, see (1.4), where 1w(y) and A2 = −a′ (3.1) γ1 := −(a′ 1y1 + a′ 2y2) is the defining polynomial of the common tangent to the members of the pencil at the distinguished basepoint (0 : 0 : 1). (Here and below, w without an argument stands for w(y).) Next, there are splittings, see (1.3), 1 = w2r′ r′ 1, 3 = wr′ r′ 3, ri = w3ri, i = 1, 2, 3, THE BERTINI INVOLUTION 5 with r1 = −a′ 2r′ 1, r2 = a′ 1r′ 1, r3 = a′ 1a′ 2r′ 3, 1 = a′2 r′ Furthermore, one has 2 B1 − a′ 1a′ 2B2 + a′2 1 B3, r′ 3 = a′ 1C2 − a′ 2C1. φ6 = wφ3, ψ6 = wψ3, C5 = w C, where 1b1y1y2 3]y2 + a′ 1[a′ 2y3(w − b1y2 1y3) − B3y2]y1y2, C(y) := −a′ φ3(y) := −a′ ψ3(y) := −a′ 2[B1 − a′ 1C2 + y3 C, 2C1 + y3 C. Finally, after reducing the common factor w3 in (3.1), the Geiser involution takes the form z1 = φ3[a′2 2 wφ3 + B3r′ 1]y1, z2 = ψ3[a′2 1 wψ3 + B1r′ 1]y2, z3 = ψ3φ3 C. The loci contracted to the basepoints (1 : 0 : 0) and (0 : 1 : 0) are the cubics {ψ6 = 0} and {φ6 = 0}, respectively, and the fixed point locus is the sextic K(y) := ψ3[−a′ 1wy3 + B1y1]y2 − φ3[−a′ 2wy3 + B3y2]y1 = 0. One has K = w K, see (1.6). The identities of §1.2 turn into y3r1 − y1r3 = a′ 2γ1φ3, y2r3 − y3r2 = a′ 1γ1ψ3, y1r2 − y2r1 = −r′ 1γ1, y3z1 − y1z3 = a′ 2φ3 K, y2z3 − y3z2 = a′ 1ψ3 K. 3.2. The double covering P2 99K P2. Let Z be the plane P2 blown up at the basepoints (including infinitely near) of the pencil other than (0 : 0 : 1) and the infinitely near one. It is a (nodal, in general) Del Pezzo surface of degree 2, and the anti-canonical linear system maps Z to P2. This map is generically two-to-one, its deck translation is the Geiser involution, and its ramification locus is a quartic curve in P2. The anti-canonical system is the web of cubics passing through the seven points blown up; the space H 0(Z; −KZ) is generated by any of φ3 or ψ3 and 99K P2, y 7→ ¯z, is given by by w and w′. Hence, the corresponding rational map P2 ¯z0 = φ3(y), ¯z1 = w(y), ¯z2 = w′(y). It is straightforward that q0 = r0 = s0 = 0. Hence, there are splittings Q4(t, t′) = t Q3(t, t′), R3(t, t′) = t R2(t, t′), S2(t, t′) = t S1(t, t′) and relations (2.2) and (2.3) turn into K 2 = −4φ3 3w + φ2 ψ3 = φ3 + S1(w, w′), 3P2(w, w′) + φ3 Q3(w, w′) + R2 2(w, w′). Thus, the ramification locus is the quartic 4¯z 3 0 ¯z1 = ¯z 2 (3.2) In the coordinates chosen, the lines {¯z0 = 0} and {¯z0 + S1(¯z1, ¯z2) = 0} are double tangents (in the generalized sense) to this quartic. 0P2(¯z1, ¯z2) + ¯z0 Q3(¯z1, ¯z2) + R2 2(¯z1, ¯z2). Similarly, given another basepoint (1 : u2 : u3), one has Su 2 (t, t′) = t Su 1 (t, t′) and the cubic {ψu 3 = 0} singular at this point is given by (cf. (2.8)) ψu 3 = φ3 + Su 1 (w, w′). 6 ALEX DEGTYAREV Remark 3.3. The coefficients of the polynomials S1, Q3, and R2 are the same as those of S2, Q4, and R3, respectively, see §2.2, with an obvious shift by one. It is worth emphasizing that the brace operation { · }i should be evaluated before the substitution a1 = a2 = 0. 3.3. A few further observations. Unlike the general case considered in §2, now, the fixed point locus { K = 0} does pass through (0 : 0 : 1). In fact, since this curve is the branch set, it follows that { K = 0} is also the locus of the singular points of the singular members of the web of cubics defined by the seven non-distinguished basepoints. This curve has a double point at each of the seven basepoints. In particular, it is an anti-bicanonical curve in Z. Under the natural identification of the web and the dual plane (P2), the locus of the singular members themselves is the curve dual to (3.2), including the lines through the singular points of (3.2). As another observation, note that {ψ3 = 0} and {φ3 = 0} are special members of the web, viz. those singular at (1 : 0 : 0) and (0 : 1 : 0), respectively. As above, these cubics are contracted by the involution to the corresponding basepoints. References 1. Ethel I. Moody, Notes on the Bertini involution, Bull. Amer. Math. Soc. 49 (1943), 433 -- 436. MR 0008163 (4,253c) Bilkent University, Department of Mathematics, 06800 Ankara, Turkey E-mail address: [email protected]
1607.07193	1	1607	2016-07-25T09:57:57	Semiample and k-ample vector bundles	[ "math.AG" ]	We show that tensor products of semiample vector bundles are semiample. For k-ampleness in the sens of Sommese, we show that over compact complex manifolds tensor products of semiample and k-ample vector bundles are k-ample, and the sum of k-ample vector bundles is k-ample. In particular results of Sommese on k-ampleness are strenghtened.	math.AG	math	SEMIAMPLE AND k-AMPLE VECTOR BUNDLES F. LAYTIMI AND W. NAHM Abstract. We show that tensor products of semiample vector bundles are semiample. For k-ampleness in the sense of Sommese, we show that over compact complex manifolds tensor products of semiample and k-ample vector bundles are k-ample, and the sum of k-ample vector bundles is k-ample. In particular, results of Sommese on k-ampleness are strenghtened. 1. Tensor products of semiample vector bundles Definition 1.1. Let X be a complex space. A line bundle L on X is said to be semiample if for some r > 0, Lr is generated by sections. A vector bundle E is semiample if OPE(1) is semiample. Notation 1.2. Since we often need to use the phrase "generated by sections", we will abbreviate it by "gbs". We start by giving several interpretations of the statement: OPE(r) is gbs. 1) Let π : PE → X be the projection. The fibre of π over a point x ∈ X is given by the 1-codimensional subspaces V of Ex, where Ex is the fibre of E over x. Points of PE will be denoted by pairs (x, V ). Let V = v⊥ for v ∈ E ∗ x, and let v be the corresponding element of (Ex/V )∗. A section s ∈ H 0(PE, OPE(r)) generates the fibre of OPE(r) over (x, V ) iff where h , i denotes the pairing of dual vector spaces. hs(x, V ), vri 6= 0, 2) Equivalently, let s be the section in H 0(X, SrE) which corresponds to s by the natural isomorphism H 0(PE, OPE(r)) ≃ H 0(X, SrE). Then s generates the fibre of OPE(r) over (x, v⊥) iff hs(x), v⊗ri 6= 0. 1991 Mathematics Subject Classification. 14F17. 1 3) Given x ∈ X, the value s(x) of each section s ∈ H 0(X, SrE) x to C. can be regarded as a homogeneous polynomial map of degree r from E ∗ If the fibre of OPE(r) over v⊥ is not generated by any global section, then these polynomials have a common zero at v. If OPE(r) is gbs, then 0 is the only common zero. 4) Since v⊗r annihilates the kernel of the natural map E ⊗r ։ SrE, the fibre of OPE(r) over (x, v⊥) is generated by a section iff there is a section s ∈ H 0(X, E ⊗r) such that hs(x), v⊗ri 6= 0. If OPE(r) is gbs, then OPE(kr) is gbs for every positive integer k. All these facts will be used in the following without reference. Theorem 1.3. Direct sum and tensor product of semiample vectors bundles on a complex space X are semiample. This result is implied by the following two lemmata. Lemma 1.4. Let X be a complex space and E, F vector bundles on X. Suppose that OPE(r) and OPF (r) are gbs (see Notation 1.2) for some r > 0. Then OP(E⊕F )(r) is gbs. Proof: For each x ∈ X and each non-zero u ∈ (E ∗ x ) we have to find a section in H 0(X, Sr(E ⊕ F )) which is not annihilated by u⊗r. Consider u = uE +uF , where uE, uF are the components in E ∗ x resp. We may assume that uE 6= 0. Since OPE(r) is gbs, there is a section in H 0(X, SrE) which is not annihilated by u⊗r E . The image of this section under the natural injection H 0(X, SrE) ֒→ H 0(X, Sr(E ⊕ F )) is not annihilated by u⊗r. x ⊕ F ∗ x, F ∗ Lemma 1.5. Let X be a complex space and E, F vector bundles on X. Suppose that OPE(r) and OPF (r) are gbs (see Notation 1.2) for some r > 0. Then there exists a positive integer n depending only on r and e = rkE such that OP(E⊗F )(nr) is gbs. The proof of this lemma needs some preparations. We will use the terminology of decorated oriented graphs (see Feynman diagrams in physics). Decoration means that to each vertex and to each arrow another object is associated. The set of arrows of a graph γ will be denoted by A(γ). Any arrow a is said to have a head h(a) and a tail t(a). A vertex which is the head of exactly n arrows is said to have indegree n, a vertex which is the tail of exactly n arrows is said to have outdegree n. The bidegree of a vertex is written as (indegree, outdegree). 2 Definition 1.6. Let V, W be vector spaces. Let i = 1, . . . , n, vi ∈ SrV , wi ∈ SrW , u ∈ (V ⊗ W )∗. Then ΓV,W (n, r, v1, . . . , vn, w1, . . . , wn, u) is the set of all decorated oriented graphs with 2n fixed vertices αi, βi, such that for each i the vertex αi has bidegree (0, r) and is decorated by v(αi) = vi, the vertex βi has bidegree (r, 0) and is decorated by w(βi) = wi, and such that all rn arrows are decorated by u. We will omit the indices V, W of ΓV,W , when no ambiguity arises. Definition 1.7. Let γ ∈ Γ(n, 1, v1, . . . , vn, w1, . . . , wn, u). The value γ of γ is defined by γ = Ya∈A(γ) u(cid:0)v(h(a)) ⊗ w(t(a))(cid:1). Definition 1.8. Let γ ∈ Γ(n, r, v1, . . . , vn, w1, . . . , wn, u). Then the value γ of γ is defined by the following two conditions: Firstly, γ depends linearly on all vi and wi. Secondly, if vi = v⊗r for i = 1, . . . , n, vi ∈ V , wi ∈ W let the expanded graph γex ∈ Γ(rn, 1, v1, . . . , v1 , . . . , wn, . . . , wn , . . . , vn, . . . , vn , wi = w⊗r , w1, . . . , w1 , u) i i {z } {z } {z } r times r times r times be a decorated oriented graph with vertices αl i, i = 1, . . . , n, l = 1, . . . , r, such that there is a bijection ξ : A(γex) → A(γ) with h(a) = αi if h(ξ(a)) = αl i be decorated by vi, βl i by wi, for i = 1, . . . , n, l = 1, . . . , r. Then i, t(a) = βi if t(ξ(a)) = βl i. Moreover, let αl i, βl r times {z } γ = γex. For later use, note that γex has a symmetry group of order (r!)2ns−1 metries of γ. γ , where sγ is the order of the group of vertex preserving sym- Proposition 1.9. There is a function ν : N×N → N with the following property. Consider finite dimensional vector spaces V, W with d = dimV and subspaces A ⊂ SrV , B ⊂ SrW , such that the corresponding spaces of polynomial maps have 0 as only common zero. Let n = ν(r, d). Then for any non-zero u ∈ (V ⊗ W )∗ there is a positive integer m with m ≤ n, elements vi ∈ A, wi ∈ B for i = 1, . . . , m and a decorated directed graph γ ∈ ΓV,W (m, r, v1, . . . , vm, w1, . . . , wm, u) such that γ 6= 0. Proof: We will construct a function ν which is far from optimal, but sufficient for our purpose. We first show that it suffices to prove the proposition for the case that the map u ∈ Hom(V, W ∗) corresponding to u is bijective. Otherwise, 3 let V ′ = V /ker(u), W ′ = W/ker(u∗), let πV , πW be the corresponding projections πV : SrV ։ SrV ′, πW : SrW ։ SrW ′ and let u′ be the element of (V ′ ⊗ W ′)∗ induced by u. If γ ∈ Γ(m, r, V1, . . . , Vm, W1, . . . , Wm, u), and γ ′ ∈ Γ(m, r, πV (V1), . . . , πV (Vm), πW (W1), . . . , πW (Wm), u′) have the same underlying undecorated graph, then γ ′ = γ. If u is bijective, we can identify W with V ∗ and write u(v ⊗ w) = hv, wi for v ∈ V , w ∈ W . Let SV be the symmetric algebra over V . There are natural multiplication maps and contraction maps which restrict to mr : SrV → End(SV ), ir : SrW → End(SV ) N : SrV → Hom(SN −rV, SN V ), mr ir N : SrW → Hom(SN V, SN −rV ), with SN V = 0 for negative N. For r = 1 the maps can be character- ized by multilinearity and the properties m1(v′)v⊗N = πS(v′ ⊗ v⊗N ), i1(w)v⊗N = Nhw, viv⊗(N −1) for v′, v ∈ V , w ∈ W , where πS is the pro- jection of the tensor algebra of V to the symmetric algebra SV . For general r they are characterized by multilinearity and the properties mr(v⊗r) = m1(v)r, ir(w⊗r) = i1(w)r. For r = 1 let P be a product with the factors m1(vi), i1(wi), taken in any order, where vi ∈ V , wi ∈ W , i = 1, . . . , m. For any non-negative integer N, the product P restricts to a map PN ∈ End(SN V ). For the trace of PN one finds tr(PN ) = Xγ∈Γm cγγ, where and Γm = ΓV,W (m, 1, v1, . . . , vm, w1, . . . , wm, h, i) d + m − 1 (cid:19), cγ = (cid:18)d + ρ + N − 1 where ρ is the cardinality of the set of arrows a of γ such that the factor m1(v(h(a))) lies to the right of the factor i1(w(t(a))) in P . For ρ > 0 this follows by induction on ρ from i1(w)m1(v) = m1(v)i1(w) + hw, vi and for ρ = 0 from cyclic invariance of the trace and induction on N. 4 For n = 0 and N = 0 the statement is obvious. The specific form of cγ is not important in the following. We have given it because of its unexpected simplicity. For arbitrary r the calculation can be reduced to the case r = 1 as in the definition of γ. Let P be a product with the factors mr(vi), ir( wi), taken in any order, where vi ∈ SrV , wi ∈ SrW , i = 1, . . . , m. For any N, P restricts to a map PN ∈ End(SN V ). To calculate the trace of PN it is sufficient to consider the case vi = v⊗r for i = 1, . . . , n, vi ∈ V , wi ∈ W . One finds , wi = w⊗r i i tr(PN ) = Xγ∈Γr m cγγ, where and Γr m = ΓV,W (m, r, v1, . . . , vm, w1, . . . , wm, h, i) cγ = (r!)2ms−1 d + rm − 1 (cid:19). γ (cid:18)d + rρ + N − 1 Here ρ is the cardinality of the set of arrows a of γ such that the factor mr(v(h(a))) lies to the right of the factor ir( w(t(a))) in P , and sγ is the order of the group of vertex preserving symmetries of γ. N ( w1) · · · mr r be the subalgebra of End(SN V ) generated by the elements Let LN N (v)ir {mr N ( w)v ∈ A, w ∈ B}. This algebra is spanned by products N (v1)ir PN = mr N ( wm), where vi ∈ A, wi ∈ B for i = 1, . . . , m. We have seen that the trace of PN is given by a linear com- bination of numbers γ, where γ ∈ Γ(m, r, v1, . . . , vm, w1, . . . , wm, h, i). If those traces all vanish, then LN is a nilalgebra. In particular, there r is a non-trivial subspace V0 ⊂ SN V such that N (vm)ir mr N (v)ir N ( w)V0 = 0 for all v ∈ A, w ∈ B. Since the kernel of mr N (v) vanishes for v 6= 0, this means that ir N ( w)V0 = 0 for all w ∈ B. Since the homogeneous polynomial maps in B have 0 as only common zero, one can apply a theorem of Macaulay ([1] Sec. 6, or [2] p. 85 theorem 4.48). According to this theorem the ideal in SW generated by B contains SN W if N ( w)V0 = 0 for all w ∈ SN W and yields a N ≥ rd, which implies iN contradiction. Thus LN is not nil for N ≥ rd. r Let LN m be the subspace of LN N (v)ir r generated by products of length ≤ m N ( w). For each m one has either m+1) > dim(LN of elements of the form mr dim(LN Let D(N) = (dim SN V )2. Since D(N) ≥ dimLN In particular, Lrd D(N ) = LN r . D(rd) contains elements of non-vanishing trace. Thus it r , one has LN m) or LN m = LN r . 5 suffices to take for ν(r, d) the least common multiple of all integers less or equal to D(rd). 2. Proof of Lemma 1.5 For each x ∈ X and each non-zero u ∈ (Ex × Fx)∗ we have to find a section in H 0(X, (E ⊗ F )⊗nr) which is not annihilated by u⊗nr. Let IE : H 0(X, SrE)⊗n → H 0(X, E ⊗rn), IF : H 0(X, SrF )⊗n → H 0(X, F ⊗rn), J : H 0(X, E ⊗rn) ⊗ H 0(X, F ⊗rn) → H 0(X, (E ⊗ F )⊗rn) be the canonical morphisms. The permutation group S(rn) acts on E ⊗rn in the standard way, which yields a map Σ : S(rn) × H 0(X, E ⊗rn) → H 0(X, E ⊗rn). Let Φ : S(rn) × H 0(X, SrE)⊗n ⊗ H 0(X, SrF )⊗n → H 0(X, (E ⊗ F )⊗nr) be defined by Φ = J ◦(Σ⊗Id)◦(Id×IE⊗IF ). Though Φ factors through the coset map S(rn) → S(rn)/S(r)×n, it yields sufficiently many sec- tions for our purpose. For x ∈ X, σ ∈ S(rn), si ∈ H 0(X, SrE), ti ∈ H 0(X, SrF ), there is an element γ ∈ ΓEx,Fx(n, r, s1(x), . . . , sn(x), t1(x), . . . , tn(x), u), such that (2.1) hΦ(σ, s1 ⊗ · · · ⊗ sn ⊗ t1 ⊗ · · · ⊗ tn)x, u⊗nri = γ. Conversely, for each γ ∈ Γ(n, r, s1(x), . . . , sN (x), t1(x), . . . , tn(x), u) one can find a permutation σ ∈ S(rn) satisfying the equation(2.1). By Proposition 1.9 there exist sections si, ti and γ ∈ Γ(n, r, s1(x), . . . , sN (x), t1(x), . . . , tn(x), u) such that γ 6= 0. The image of the corresponding section Φ(σ, s1 ⊗ · · · ⊗ sn ⊗ t1 ⊗ · · · ⊗ tn) in H 0(X, S ⊗nr(E ⊗ F )) yields a section in H 0(P(E ⊗ F ), OP(E⊗F )(nr)) which generates the fibre over (x, u⊥). Remark 2.1. The arguments in sections 1 and 2 remain valid when C is replaced by any algebraically closed field of characteristic 0 . Thus tensor products of semiample vector bundles on algebraic varieties over such a field are semiample, too. 6 3. Direct sums and tensor products of k-ample vector bundles The following definition was introduced by Sommese [3]. Definition 3.1. A line bundle L on a compact complex space X is k-ample if 1) L is semiample 2) the fibers of the corresponding morphism φ : X → PH 0(X, Lr)∗ have dimensions less or equal to k. A vector bundle E is said to be k-ample if OPE(1) is k-ample. Note that 0-ample is the same as ample, and k-ample implies (k +1)- ample. Moreover (dim PE)-ample is the same as semiample. Sommese proved in ([3] p. 235, Corollary (1.10)) that direct sums and tensor products of k-ample vector bundles are k-ample, if these vector bundles are generated by sections. The aim of the present section is to remove the latter restriction, moreover for the product it is sufficient that one factor is k-ample and the other is semiample. Theorem 3.2. E ⊕ F is k-ample if and only if E, F are k-ample. Proof: Since the quotient bundle of a k-ample bundle is k-ample, the 'only if' part is obvious. For the 'if' part we use criterion (1.7.3) of proposition (1.7) of Sommese [3]. First note that OPE(r1) gbs (see Notation 1.2) and OPF (r2) gbs imply that OPE(r) and OPF (r) are gbs whenever r is a common multiple of r1, r2. By Lemma 1.4 this implies that OP(E⊕F )(r) is gbs. Now assume that there is a holomorphic finite to one map φ : Z → X of a compact analytic space Z to X, such that dimZ = k + 1 and that there is a surjective map φ∗(E ⊕ F ) ։ Q with a trivial bundle Q. In particular there is a non-trivial section of φ∗(E ∗ ⊕ F ∗). We may assume that the component s : Z → φ∗E ∗ of this section is non-trivial. Since OPE(r) is gbs, for any z ∈ Z such that s(z) 6= 0 there is a section σ : X → SrE such that hφ∗σ(z), s(z)⊗ri 6= 0. In particular, hφ∗σ, sri yields a non-trivial section of the trivial line bundle over Z. Such a section must be constant, which implies that s cannot vanish anywhere. Consequently it yields a trivial quotient bundle of φ∗E, contrary to the assumption that E is k-ample. For the main theorem of this section we need two lemmata. We first reformulate criterion (1.7.4), proposition (1.7) of Sommese [3]. Lemma 3.3. Let E be semiample. Then E being k-ample is equivalent to the following condition: Given any coherent sheaf G on X, there 7 exists an integer N(G) such that for n > N(G) and j > k one has H j(X, G ⊗ E ⊗n) = 0. Proof: In its original form, criterion (1.7.4) uses the symmetric prod- uct SnE instead of E ⊗n. Since the symmetric product is a direct sum- mand of E ⊗n, the new criterion implies the old. To show the converse, note that E ⊗n is isomorphic to a direct sum of direct summands of Sn(E ⊕rkE). Moreover by theorem 3.2 E ⊕rkE being k-ample is equiva- lent to E being k-ample. Lemma 3.4. Let F be a semiample vector bundle on a smooth complex space X. Then there are positive integers r, L and nl for l = 0, 1, . . ., and graded locally free algebraic sheaves F1, . . . , FL of finite rank with the following properties. The graded sheaf SF = ⊕n∈NSnF has a reso- lution 0 → FL → . . . → F0 → SF → 0, where each Fi has a filtration Fi = F 0 i ⊃ . . . i ⊃ F 1 i ⊃ F 2 compatible with the grading such that for l = 0, 1, . . . where [s] denotes the shift in degree by s. i /F l+1 F l i ≃ Fi[−lr]⊕nl, Proof: Choose r such that OPF (r) is gbs. Let Γ be the (isomorphic) image of H 0(PF, OPF (r)) in H 0(X, SrF ). The elements of SΓ act nat- urally on SF by multiplication, in particular on 1 ∈ S0F. Let F be the smallest subsheaf of SF which contains the subsheaf H 0(X, SrF )1 of SrF and admits the multiplicative action of SF . By Macaulay's theorem we have SnF ⊂ F if n > N, N = (r − 1)rk(F ). By induction on s one has Ssr+nF ⊂ SsΓSnF for s ∈ N if r + n > N. The direct summands of SF can be written in the form Ssr+nF with n ≤ N. Thus the multiplication map SΓ ⊗ (⊕N n=0SnF ) → SF is surjective. This implies that SF is a sheaf of finitely generated (SΓ ⊗ OX )-modules. Since SΓ ⊗ OX is a sheaf of regular local rings, SF has local minimal resolutions of finite length L which are unique up to isomorphism. Gluing the corresponding free (SΓ ⊗ OU )-modules over open sets U ⊂ X by these isomorphisms, one obtains a resolution of SF by sheaves of locally free (SΓ ⊗ OX)-modules Fi of finite rank. Let F l i = SlΓFi 8 for i = 0, . . . , L, and Fi = Fi/F 1 i /F l+1 F l i . This implies i ≃ SlΓ ⊗ Fi for l = 0, 1, . . .. Thus the statement of the proposition is true when nl is the dimension of the vector space SlΓ. Theorem 3.5. If E is a k-ample and F a semiample vector bundle on a compact complex manifold X, then E ⊗ F is k-ample. Proof: The sheaf (E ⊗ F )⊗n is isomorphic to a direct sum of direct summands of E ⊗n ⊗ SF ′, where F ′ = F ⊕rkF . Thus it suffices to show that H j(X, G ⊗ SF ′ ⊗ E ⊗n) vanishes for j > k and sufficiently large n. The notations of lemmata 3.3 and 3.4 will be adapted to F ′ instead of F by adding primes. Let n > N(G ⊗ F ′ i) for i = 1, . . . , L′. Then H j(X, G⊗F ′l i ⊗E ⊗n) = 0 for each l and j > k. Since the filtration i ⊗ E ⊗n) = 0 for of F ′ j > k, i = 1, . . . , L′, and consequently H j(X, G ⊗ SF ′ ⊗ E ⊗n) = 0. i is finite at each degree this implies H j(X, G ⊗ F ′ i/F ′l+1 Remark 3.6. Theorem 3.5 implies that there is no need for the gbs conditions in lemma (1.11.4) and proposition (1.13) of [3] when X is smooth. The lemma and the proposition can be strengthened as follows: Lemma 3.7. Let E be a k-ample vector bundle on a compact complex manifold X, Gr(s, E) be the bundle of s-codimensional subspaces of the fibres of E and ξ(s, E) the tautological line bundle of Gr(s, E). Then ξ(s, E) is k-ample. In the proof of this lemma, k-ampleness of E only enters through the k-ampleness of ΛsE, which is a quotient bundle of E ⊗s. Proposition 3.8. For i = 1, . . . , n let Ei be ki-ample vector bundles of rank ri on a projective manifold X. H p(X, ∧qT ∗ X ⊗ ∧s1E1 ⊗ · · · ⊗ ∧snEn) = 0 if p + q > dimX +Pi si(ri − si) + min{k1, . . . , kn}. In the proof, k-ampleness of the Ei only enters through the k-ampleness of ∧s1E1 ⊗ · · · ⊗ ∧snEn, which is a quotient bundle of E ⊗s1 1 ⊗ · · · ⊗ E ⊗sn See also [5], where lemma 2.1 can easily be generalized to k-ample. Finally we note two corollaries to theorems 3.2 and 3.5 n . Corollary 3.9. If E is a k1-ample vector bundle and F is k2-ample vector bundle on a complex manifold, then E ⊕ F is max{k1, k2}-ample and E ⊗ F is min{k1, k2}-ample. 9 Corollary 3.10. One can extend the dominance ampleness theorem; theorem 3.7 in [4] to an analogous dominance k-ampleness theorem. Indeed, the proof of the dominance ampleness theorem uses no other property of ampleness but the fact that direct summands, direct sums and tensors products of ample vector bundles are ample. References [1] F.S. Macaulay, The algebraic geometry of modular systems, Cambridge Univ. Press, 1916. [2] B. Shiffmann, A.J. Sommese, Vanishing theorems on complex manifolds, Progress in mathematics Vol. 56(1985). [3] A.J. Sommese, Submanifold of Abelian Varieties,, Math.Ann.233(1978), 229- 256. [4] F.Laytimi,W. Nahm, A generalization of Le Potier vanishing theorem, Manuscripta Math.113(2004),no 2, 165-189. [5] F.Laytimi,W. Nahm, A vanishing theorem, Nagoya Math. J. Vol. 180 (2005) 35-43. F. L.: Math´ematiques - bat. M2, Universit´e Lille 1, F-59655 Vil- leneuve d'Ascq Cedex, France E-mail address: [email protected] W. N.: Dublin Institute for Advanced Studies, 10 Burlington Road, Dublin 4, Ireland E-mail address: [email protected] 10
1904.00298	1	1904	2019-03-30T21:57:26	Irreducibility of analytic arc-sections of hypersurface singularities	[ "math.AG" ]	We explore the existence of irreducible and reducible arc-sections in an irreducible hypersurface singularity germ along finite projections. In particular we provide examples of irreducible isolated hypersurface singularities for which no irreducible arc-sections exist, and show that reducible ones always exist. Moreover, we give an algorithm to check if a given projection allows irreducible arc-sections, and find them if they exist.	math.AG	math	Irreducibility of analytic arc-sections of hypersurface singularities. Miguel Angel Marco-Buzunariz ∗, Maria Pe Pereira† April 2, 2019 Abstract We explore the existence of irreducible and reducible arc-sections in an irreducible hypersurface singularity germ along finite projections. In particular we provide examples of irreducible isolated hypersurface singularities for which no irreducible arc-sections exist, and show that reducible ones always exist. Moreover, we give an algorithm to check if a given projection allows irreducible arc-sections, and find them if they exist. 1 Introduction Given a hypersurface germ singularity (X, ¯0) ⊆ (Cn+1, ¯0) defined by the equa- tion F (x1, . . . , xn+1) = 0, and an injective arc γ : (C, 0) → (Cn, ¯0), we say that the plane curve defined by F (γ(t), xn+1) = 0 is an arc-section of (X, ¯0). Consider, for example, a suspension singularity given by an equation of the form zm = f (x, y) with f (x, y) an irreducible germ of order d with g.c.d.(m, d) = 1. Given any generic line L in the XY -plane parametrized as (at, bt), we consider the plane curve germ given by zm = f (at, bt). It is easy to see that this plane branch is irreducible since the associated Newton polygon consists on a unique bounded segment with no integer points in its interior. Note that, in the case ord(f ) > m, we have that the projection direction is not contained in the tangent cone of the surface at the origin viewed as a set in CP2. We wonder if this is the general case. That is: Question. Let (X, ¯0) be an irreducible hypersurface germ defined in Cn+1 by the equation F (x1, ..., xn+1) = 0. Do there exist a linear projection π : Cn+1 → Cn, say (x1, ..., xn+1) 7→ (x1, .., xn) with π (X,¯0) finite and an injective arc γ : (C, 0) → (Cn, ¯0) such that F (γ(t), xn+1) = 0 is an irreducible plane curve germ? This was privately formulated by A. Ploski and E. Garc´ıa Barroso to the second author. We study it here for the first time, developping some tools to study variations of the question in the surface case. ∗Partially supported by MTM2016-76868-C2-2-P and Grupo "Investigaci´on en Educaci´on Matem´atica" of Gobierno de Arag´on/Fondo Social Europeo. †Partially supported by MTM2017-89420-P, MTM2016-76868-C2-1-P and ERC Consolida- tor Grant NMST. 1 We give a negative answer to the question in Proposition 4.2, by showing a family of counterexamples. Furthermore, in Theorem 5.2 we show that arcs where the curve is reducible always exist. Section 7 gives a method to determine if arcs that produce irreducible sections do exist or not (and find them in case they do). Note that the existence of irreducible arc-sections in a hypersurface (X, ¯0) over arcs not tangent to the discriminant of a projection implies that (X, ¯0) is irreducible. The article is structured as follows. Sections 2 and 3 fix the notation. In Section 4 we give necessary conditions on the tangent cone of (X, ¯0) for irre- ducible arc-sections to exist (see Lemma 4.1). We also give examples where they do never exist for any projection (see Example 4.4). These are also examples of irreducible hypersurfaces (X, ¯0) whose intersection with any smooth surface is reducible. In Section 5 we see that reducible arc-sections always exist for every finite projection πX,¯0 , and moreover, for every finite projection πX,¯0, there are arc- sections with as many components as the degree of πX,¯0. In subsection 5.1, we see that, apart from the cases where the tangent cone of (X, ¯0) is a union of lines, the generic arc is never irreducible. In this case, we see that the arc with irreducible arc-section has to be tangent to the discrimi- nant. In Section 6, we give an alternative description of the tangent cone of the discriminant in many cases which can be usefull in order to simplify the search for irreducible arc-sections. In Section 7, we give an algorithm to determine whether irreducible arc- sections over an arc not contained in the discriminant exists or not; and find them in case they do. 2 Notation and general setting Let (X, ¯0) be a hypersurface germ defined by F (x1, ..., xn+1) = 0. Let m be the multiplicity of (X, ¯0). Let π : Cn+1 → Cn be a linear projection such that π(X,¯0) is finite (see Remark 3.4). After a linear change of coordinates we can assume that π is given by (x1, ..., xn+1) 7→ (x1, .., xn). We say the projection is transverse (with respect to (X, ¯0)) if it is in a direction not contained in the tangent cone of (X, ¯0). It is non-transverse in the other case. Let ∆ denote the discriminant of π X . Note that the restriction πX\π−1(∆) is a regular covering of certain degree d. Then, given a loop δ in Cn \ ∆ we have the monodromy of the covering along δ which gives an element of Sym(d), the permutation group of d points. Since πX,¯0 is finite, we can assume F (x1, ..., xn+1) is a Weierstrass polyno- mial in xn+1 of degree d. Moreover, if the projection is transverse then d = m. In the other case, we have d > m. We denote by CX,¯0 the tangent cone of (X, ¯0) seen, either as a conic subva- riety germ of (Cn+1, ¯0) or a subvariety of CPn. 2 Given π : Cn+1 → Cn as above, we denote by pπ the point in CPn that gives the projection direction. We denote by π : CPn → CPn−1 the projection from pπ. Given a subspace V of CPn, we denote by < V > the linear subspace of Cn+1 that it generates. 3 Arc-sections. In this section we fix notation and definitions regarding arcs and arc-sections. Definition 3.1. An arc in (Cn, ¯0) is a holomorphic mapping germ γ : (C, 0) → (Cn, ¯0). A representative of an arc γ is the restriction of γ to some closed disk D ⊂ C centered at 0 ∈ C contained in the domain of γ. A smooth arc is an arc γ such that γ′(0) 6= ¯0. We say that an arc γ(t) is transversal to a subvariety Y of (Cn, ¯0) at t0 ∈ C if γ′(t0) 6= ¯0 and γ′(t0) is not contained in the tangent space of Y at γ(t0). We say that γ(t) is transversal to Y if it is transversal at any point. In general we will only consider injective representatives γD (which always exist for generically 1:1 arcs). In particular we will assume that D is small enough so that (γD)−1(¯0) = {0}. Then, note that γ∂D is a simple loop in Cn \ {¯0}. Lemma 3.2. Let (X, ¯0), F (x1, ..., xn+1) ∈ C{x1, ..., xn+1} and π be as above. Given an arc α : (C, 0) → (Cn, ¯0) whose image is not completely contained in the discriminant ∆, the following are equivalent: 1. F (α(t), xn+1) = 0 defines an irreducible germ at (0, 0) ∈ D × C 2. π−1(α(D)) ∩ X is an irreducible germ at ¯0 ∈ Cn+1 for an injective repre- sentative α : D → Cn such that α(D) ∩ ∆ = {¯0} 3. the monodromy along α(∂D) is transitive (with αD a representative as in (2)), that is, it gives a d-cycle of Sym(d). Proof. To see (1) ⇔ (2) we observe that the mapping C2 → Cn+1 (t, xn+1) 7→ (α(t), xn+1) is injective restricted to D × C since αD is injective. Then, its restriction to the germ F (α(t), xn+1) = 0 is a homeomorphism onto π−1(α(D)) ∩ X. Then, one of them is irreducible if and only if so is the other. To see (2) ⇔ (3) we note that the singular locus of π−1(α(D)) ∩ X is just the origin. Then, π−1(α(D)) ∩ X is irreducible if and only if it remains connected after removing the origin. The map π π−1(α(D\{0}))∩X onto α(D \ {0}) is a cover over a space that can be retracted to the simple closed loop α(∂D). So π−1(α(D \ {0})) ∩ X is connected if and only if the monodromy along α(∂D) is transitive. Note that equivalence of (1) and (2) is also true if the arc is contained in the discriminant. Last lemma justifies the following definition and allows us to think geomet- rically about the problem. 3 Definition 3.3. Let π : Cn+1 → Cn be a linear projection such that π(X,¯0) is finite. An arc-section of (X, ¯0) is a section of (X, ¯0) of the form π−1(α(D)) ∩ X for certain injective arc representative α : D → Cn that is either contained in the discriminant ∆ of πX or α(D) ∩ ∆ = {¯0}. So, our main question is about the existence or non-existence of irreducible and reducible arc-sections for irreducible hypersurface germs. Note that if α(D) is contained in ∆, then there are embedded components in the arc-section. So, for arcs in Cn whose image is contained in the discriminant, we ask whether they give irreducible arc-sections considered with its reduced structure. Remark 3.4. As we said, we always assume πX,¯0 is finite/proper, which is equivalent to π−1(¯0) = {¯0}. If the projection is not proper, then an arc-section either has π−1(¯0) as irreducible component or is a (cylindrical) surface. Note that a non-trasverse linear projection, that is a projection in a direction con- tained in the tangent cone of (X, ¯0), is always proper in case this line of the tangent cone is not contained in X. 4 Looking at the tangent cone of an arc-section. Here we will see some necessary conditions on the tangent cone of (X, ¯0) for the existence of irreducible arc-sections. Lemma 4.1. Let π : Cn+1 → Cn be a linear projection such that π(X,¯0) is finite. Let L be a line in CPn containing the projection direction. Assume the intersection (with reduced structure) of L and CX,¯0 in CPn is a union of isolated points. Consider any arc in (Cn, ¯0) with tangent at the origin in the direction given by a point of L. Then the reduced tangent cone of the corresponding arc- section is given by the intersection CX,¯0 ∩ L. In particular, if the intersection is 2 or more different points then the arc section is reducible. Proof. Note first that the tangent space of a set of the form π−1(α(D)) is the plane generated by the projection direction of π and the tangent of the arc α. Viewed in CPn, it is the line joining the corresponding points. After the blowing up of the origin of Cn+1, we get the strict transform of the set π−1(α(D)) and the hypersurface (X, ¯0). Their common intersection with the exceptional divisor is a set of isolated points by hypothesis. Around any of these points we have the intersection of a hypersurface and a surface that is not contained in it (because the projection is proper) in a smooth space. Then, the intersection is a (not necessarily irreducible) curve. Hence, for any point of CX,¯0 ∩ L we have at least one component of the arc section. Note that the only case in which the hypothesis of the previous proposition are not satisfied is when L is contained in CX,¯0. Note also that, if CX,¯0 intersects every line in more that one point, no irreducible arc-sections can exist. This happens, for instance, if CX,¯0 is a curve with degree bigger than three and no hyperflexes. Then, we can summarize our conclusion in two propositions: 4 Proposition 4.2. There exist irreducible hypersurface germs (X, ¯0) with all its arc-sections being reducible. Moreover, if we apply any analytic change of coordinates, the resulting germs also satisfy this property. Corollary 4.3. There exist irreducible hypersurface germs (X, ¯0) such that the intersection with any smooth surface germ (Y, ¯0) is reducible. Example 4.4. The smooth quartic given by x4 + y4 + x2z2 + yz3 + z4 = 0 does not have hyper-flexes, and then it is an explicit example. Note that the set of smooth quartics without hyperflexes is Zariski open in the set of quartics. In this example, the equation considered in affine coordinates gives the germ in (C3, ¯0), and the same equation considered as homogenous coordinates in CP2 gives its tangent cone. Example 4.5. Another example is any surface with smooth irreducible tangent cone of degree greater or equal than 4 and no hyperflexes. For every degree this is a Zariski open set of the set of surfaces of that degree. Other examples are given by surfaces whose tangent cone is the union of two non-tangent conics. Proposition 4.6. A necessary condition for the existence of an irreducible arc- section in (X, ¯0) is that there exists a line L, which may be contained in one irreducible component of CX,¯0, that meets any component of CX,¯0 that does not contain it (if any) in exactly one point. This point would be an intersection point of all the irreducible components of CX,¯0, in case there are more than one. When this necessary condition is achieved, that is, in case there exists such a line L, then the starting point to look for irreducible arcs sections is the following: (i) consider a projection π in the direction of a point pπ ∈ L. Since we ask the projection to be proper, we have to require the line < pπ >⊂ Cn+1 to not be contanined in (X, ¯0); (ii) once the projection is fixed, the only arcs that may lead to irreducible arc sections must have tangent cone equal to π(L′) with L′ some line as in Corollary 4.6 passing through the projection point. In Section 6, we give an alternative description of the possible tangent cones in some cases. Remark 4.7. We can easily summarize, in the surface case, the cases in which such a line L can exist. There are 5 cases for CX,¯0 with reduced structure, that are (a) a line, (b) a union of lines meeting in a common point p, (c) a conic, (d) an irreducible surface of degree d > 2 that intersects some tangent line with multiplicity d and (e) a reducible surface with a common point p of all its components (and such that there is a line through p that is a tangent line of all the components of degree d ≥ 2 of maximal intersection multiplicity). We can saee that (d) is never the case if CX,¯0 is a generic variety of degree greater or equal than 4. We retake the examples in the introduction, that cover the cases (a)-(b) in the previous remark: Example 4.8. For the surface zd = f (x, y) with ord(f ) = m > d with gcd(d, m) = 1. The tangent cone is z = 0. Then, we can expect irreducible arc sections for 5 any projection transverse to z = 0 and in any direction. In particular for the projection in the direction of z one can check easily that (at, bt) gives irreducible arc section whenever the initial form of f evaluated at (a, b) doesn't vanish. Example 4.9. In the case of zd = f (x, y) with ord(f ) = m < d and gcd(d, m) = 1, we have that the tangent cone can be in general given by a union of lines (in CP2). So, we can expect irreducible arc sections in any direction for a non- transverse projection. Then, for example, for the projection to the XY-plane, for generic (a, b), it is easy to see that (at, bt) give irreducible arc sections. Looking at the possibilities (a)-(e) for the case of surfaces in Remark 4.7, we see that the existence of irreducible arc-sections for the generic projection may only hold in the cases (a)-(c) and cases in (e) where the tangent cone is a product of tangent cones as in (a) − (c) (as for example CX,¯0 being a union of a conic and some lines through a given point). A natural question would be wether the existance of an irreducible arc-section in these cases is invariant by changing the generic projection or even by analytical change of coordinates. 5 Existence of reducible arc-sections with d com- ponents Let (X, ¯0) be a surface germ in (C3, ¯0) and let π : C3 → C2 be a projection so that πX is finite. Definition 5.1. We define transversal monodromies as follows: (a) we say that the monodromy of a loop δ parametrizing the boundary of a small disk transversal to a component of ∆ is the transversal monodromy associated to the component of ∆ it meets. (b) Given any exceptional component Ei of a composition of blow ups over the origin of (C2, 0), we can consider any arc γ in C2 with transversal lifting through Ei not meeting any other component of the total transform of ∆. Given D such that γ(D) ∩ ∆ = {¯0}, the monodromy of γ∂D is called the transverse monodromy associated to the divisor Ei. Take an embedded resolution of (C2, ∆). Let ∆∗ be the total transform of ∆. Take a normal crossing point in ∆∗. It has two local branches, F1 and F2. Let σ1 and σ2 be the permutations associated to its transversal monodromies. It is well known that meridians around two smooth curves as F1 and F2 meeting transversaly (that is, loops bounding small transversal discs to Fi) commute. Since the monodromy is a group morphism, σ1 and σ2 must commute too. Now, if we make an extra blowup at this point, a new divisor F3 will appear, whose permutation corresponding to the transversal monodromy will be the composition of σ1 and σ2 (the order doesn't really matter, since they commute). By successive blowing ups we can obtain divisors whose corresponding transver- sal permutation is σn 2 for every positive integers n, m. When n and m are multiples of the order of σ1 and σ2, we will obtain a divisor with trivial permu- tation. 1 · σm So we have proven the following: 6 Theorem 5.2. Let π : C3 → C2 be a linear projection such that π(X,¯0) is finite. There always exist an arc γ for which the curve (X, ¯0) ∩ π−1(γ(D)) has as many irreducible components as the degree of the cover πX . Proof. Just take γ to be the contraction of a smooth arc transverse to a divisor with trivial permutation. Remark 5.3. A divisor with trivial permutation can always be obtained by blowing up a generic point in any component of the exceptional divisor and then keep blowing up the point at the normal crossing of the divisors that appears. In that case the sequence of permutations that appear are consecutive powers of the original one. This implies that there are arcs with totally reducible arc-sections with arbi- trary tangent cone. Example 5.4. For the case zd = f (x, y), if x = 0 is a generic line for f (x, y) = 0 (i.e. not tangent to it), for the projection to the XY - plane, the arc (td+1, td) gives a reducible arc-section. In this case this is easy to check since zd − f (td+1, td) = 0 has tangent cone zd − tkd = 0 with k the multiplicity of f (x, y) = 0, which factorizes in d factors. 5.1 The generic arc-section Let (X, ¯0) be a surface germ in (C3, ¯0). For a given projection π : C3 → C2 with πX,¯0 finite we will say that a generic arc is a smooth arc not tangent to the discriminant ∆ of πX,¯0. All generic arcs give the same associated monodromy. Moreover, this is the case for any projection in a direction of the open set Ω ⊂ Grass1(C3) of directions that are Zariski equisingularity. This is because we can deform one generic arc to the other. Then, one is transitive if and only if all of them are. Given a singular arc transverse to the discriminant, then its monodromy is a power of the monodromy of the generic smooth arc by arguments in Section 5. Then, if a singular arc gives rise to an irreducible arc section, its monodromy must be transtive and since it is the power of the monodromy of the generic arc, the monodromy of the generic arc should be also transitive. That is: Corollary 5.5. If an arc not tangent to ∆ gives an irreducible arc-section, then all smooth arcs not tangent to the discriminant do. Corollary 5.6. If the generic arc gives an irreducible arc section, then either the tangent cone CX,¯0 is a line (with certain multiplicity) or the tangent cone is a union of lines and we are projecting from the singular point of CX,¯0. Proof. Looking at the cases (a) − (e) in Remark 4.7 we see that these two cases are the only ones that allow irreducible arc sections in any generic direction. Corollary 5.7. If CX,¯0 is not a line or a union of lines and there is an irre- ducible arc-section over an arc γ, then the arc γ is tangent to ∆. 7 6 Discriminant of the tangent cone coincides with the tangent cone of the discriminant After Corollary 5.7 we are interested in the tangent cone of ∆ in order to know the tangent line of an arc with irreducible-arc section. Let (X, ¯0) be a surface in (C3, 0). We want to prove that when projecting in the direction of a point in a line L as in the hypothesis of Lemma 4.1, we can know in advance the only possible tangent line for an arc with irreducible arc-section. Following [1], we consider the Nash modification eX that is the closure in C3 × Grass2(C3) ≈ C3 × CP2 of {(x, TxX) : x ∈ X}. The projection of eX onto the first factor is called the Nash blowup N : eX → X and the one onto the second, the gauss map eG : eX → CP2. We recall that N −1(0) ⊂ {0} × CP2 is the union of CX,¯0, that denotes the dual curve of the tangent cone CX,¯0 ⊂ CP2, and some lines that correspond to pencils of planes passing through the exceptional tangent lines. Every point in N −1(0) ⊂ {0} × CP2 corresponds to a hyperplane in (C3, 0), limit of tangent hyperplanes to (X, ¯0). In particular, a point in the dual of the tangent cone, represents the plane in C3 induced by the tangent line to its dual point in CX,¯0 ⊂ CP2. Proposition 6.1. Assume the projection direction is not an exceptional tan- gent. Then, the tangent of the discriminant ∆ coincides with the discriminant of πCX,¯0 . Proof. The projection direction pπ gives a line Tπ in CP2 ≈ CP2 that corre- sponds to the pencil of hyperplanes that contains the direction. The set N (G−1(Tπ)) gives the polar curve Γ of the projection. Since Tπ is not the pencil of an exceptional tangent, we have that N (G−1(Tπ)) has one dimensional component in (X, ¯0). This is because Tπ is not the pencil of an exceptional tangent line and because G−1(Tπ) is defined by 1 equation in a surface eX and G−1(Tπ) is not contained in N −1(0). It is immediate that whenever we have a point in the discriminant of ¯π, that is, whenever there exists q ∈ CX,¯0 such that qpπ is a line tangent to CX,¯0 at q, we have an intersection point of Tπ ∩ CX,¯0, that is a limit tangent hyperplane to (X, ¯0) that contains < pπ >. And reciprocally, an intersection point of Tπ ∩ CX,¯0 gives a line tangent to CX,¯0 passing by pπ. Finally, the tangent line at the origin of every branch of the discriminant π(Γ) is not zero and is given by limits of tangent lines out of the origin. These lines are the image by π of the tangent lines of the polar curve, which are contained in the hyperplanes corresponding to points of Tπ. Then, since the tangent line is not zero, it has to be in the discriminant or π. 8 Corollary 6.2. Assume all the points on CX,¯0 whose tangent line is as in Lemma 4.1, are not exceptional tangent lines. Assume CX,¯0 is not a line or a union of lines. Then, any arc with irreducible arc-section has as tangent line a point in the discriminant of πCX,¯0 . Examples where the hypothesis of Corollary 6.2 are satisfied are any surface with CX,¯0 smooth, since it is known that there is no exceptional tangents in this case (see (2.2.1) in [1]). 7 Algorithm to find an irreducible arc-section. Let (X, ¯0) be a surface germ in (C3, ¯0). Based on the previous sections, we get a general method to determine if a given surface germ projection π : (X, ¯0) → (C2, 0) admits irreducible arc sections or not. First of all, we find the discriminant ∆. The following steps are: 1. Compute an embedded resolution of singularities of ∆, φ : ( eC2, E) → (C2, ¯0). Applying the same sequence of blowups, we get a map eφ : ( eX, ¯E) → (X, ¯0). 2. For each normal crossing along the total transform of ∆, that is, for every intersection of two branches B1, B2 in φ−1(∆), get two small smooth disks D1, D2 such that: • Di is centered in a point of Bi, is transversal to it, and touches no other component of φ−1(∆) • There exists a point q in the intersection of the boundaries ∂D1 ∩∂D2 • Both disks are inside a small polydisc centered in the normal crossing point, that meets no other component of φ−1(∆). 3. Consider π ◦ eφ restricted to the preimage of ∂D1 and ∂D2 as in 2, and compute the permutations P1 and P2 of the corresponding monodromy based in q. 4. The projection π allows an irreducible arc section if and only if for some P1, P2 as before, the group generated by them contains a transitive ele- ment. Example 7.1. Consider the surface germ given by z4 − 4xz + 3y2 = 0, projected along the z direction. Note that it is an isolated singularity, and hence it must be irreducible. Its discriminant is given by the equation (y3 − x2)(y3 + x2) = 0. We compute a resolution of singularities by the sequence of blowups in fig- ure 1. We get three exceptional divisors and two branches. Let's focus for instance in the normal crossing at the origin of the last chart. The two discs B1, B2 can be taken as x = 1 2 respectively, so the base point q would be ( 1 2 ). The composition of the blow-up transformations applied to the equation of X is z4 − 4(x2y3)z + 3(xy2)2 = 0. 2 and kxk ≤ 1 2 , 1 2 , kyk ≤ 1 2 , y = 1 9 (y3 − x2)(y3 + x2) (x, y) 7→ (xy, x) y4(y − x2)(y + x2) (x, y) 7→ (x, xy) y4x6(y − x)(y + x) (x, y) 7→ (xy, y) (x, y) 7→ (x, xy) y12x6(1 − x)(1 + x) y4x12(y − 1)(y + 1) Figure 1: Sequence of blowups to resolve the singularities of the discriminant. 10 2 , t 2 ) and substitute, obtaining z4 − 1 To compute the permutation around the boundary of the disk, we parametrize the disk as ( 1 64 t4. This is a squarefree homogenous equation in two variables, so it factors as the product of four linear forms. That is, the arc section corresponding to this arc is a union of four lines, and the corresponding monodromy has a trivial permutation. 8 t3z + 3 The other disk can be parametrized as ( t equation z4 − 1 boundary of the unit disk is given by the braid shown in Fig 2. 2 ) and after composing, we get the 64 t2 = 0. The braid monodromy of this curve around the 8 t2z + 3 2 , 1 Figure 2: The braid (σ0σ2σ1)2 This braid can be expressed in the Artin presentation (see for instance [2]) as (σ0σ2σ1)2. Its permutation is (1, 4)(2, 3). In this case, the group generated by these two permutations only has two elements, none of which is transitive. We can repeat these computations with the rest of the crossings. In this case, one of the permutations will always be trivial (since they correspond to the first divisor we have computed). Similar computations show that the rest of the permutations we obtain are (3, 4), (3, 4) and (1, 4, 2) respectively. None of them are transitive and the group generated by them and the trivial permutation will not contain any transitive element. That is: in this case, there is no irreducible arc section. In fact, not every possible branch must be checked. It is enough to study the branches ∆i of ∆ whose tangent line is contained in a line L as in Lemma 4.1. Moreover, we have a necessary condition: Lemma 7.2. Consider an arc that can be deformed to a branch ∆i of the discriminant without intersecting the discriminant outside ∆i, and whose cor- responding arc-section is irreducible, then a parametrizaton of ∆i also has an irreducible arc-section. Proof. Let γi : (D, 0) → (C2, 0) be such an arc, and δ : (D, 0) → (C2, 0) be a parametrization of ∆i. Take a continuous deformation T : (D, 0) × I → (C2, 0) such that T (D,0)×{0}= γi, T (D,1)×{0}= δi, and T (∂D × [0, 1)) ∩ ∆ = ∅. The restriction of this deformation to the boundary of D gives a continuous deformation from the closure of the braid corresponding to γi to the closure of the braid corresponding to ∆i. Since the arc section of γi is irreducible, its braid is transitive, or equivalently, its closure is a knot. Since the deformation of the arcs does not intersect the discriminant, the different steps of the deformation (except for the last one) are knots isotopic to the original one. So, the closure of the braid corresponding to ∆i is the image of a knot by a continuous function, and hence, it must have also only one connected component. Hence the braid corresponding to ∆i must also be transitive. 11 This phenomenon happens, for instance, with the arcs that are parallel to the branch ∆i in the chart centered at the normal crossing between ∆i and the rest of the exceptional divisor. So if we are interested in finding irreducible arc sections, in the previous method we can skip the cases around branches of the discriminant whose parametrization doesn't give an irreducible arc section. 7.1 Sections over arcs contained in the discriminant We finish with an example where the behaviour over a parametrization of a branch of the discriminant changes under a small change of the projection. Example 7.3. Consider the surface germ given by z3 − (x − y)(x + y)(x − 2y)(x + 2y). If we project in the z direction, the discriminant is formed by the four lines x = y, x = −y, x = 2y and x = −2y. Over a generic arc such as x = 0 we get an irreducible arc-section (with equation z3 − y4). Over each point of the discriminant, there is only one point of the surface. So, if we take an arc that parametrizes one component of the discriminant, the arc section that we get is a multiple line. Now consider a line that doesn't go through the origin, say x = ǫ for some small ǫ > 0. It meets transversally the discriminant in four points (one for each component) (ǫ, ǫ), (ǫ, −ǫ), (ǫ, 2ǫ), (ǫ, −2ǫ). Over each of these four points, there is only one preimage. It is easy to check that the local equation for the arc section over the arc that parametrizes the line x = ǫ centered at any of these four points is a vertical flex. Figure 3: A rotation causes the discriminant to unfold from lines to cusps. Now we make a small deformation of the projection (e.g. make the change of variables y = y + δz for some small δ > 0). Then the four lines of the 12 discriminant get deformed into 4 cusps. Let's see the arc section for an arc that parametrizes one of them. To do so, consider again the section over the line x = ǫ. It is the same small change of variables aplied to the previous curve with four vertical flexes. If we rotate slightly a vertical flex, it splits in two different simple vertical tangencies. That means that over each point of the discriminant, there will be two points of the surface: one with multiplicity 2 and other with multiplicity 1. Hence, the arc section corresponding to a parametrization of one of the cusps will contain a component with multiplicity 2 and another with multiplicity 1. That is, it will be reducible. Note that both projections are non-transversal to the surface and have dif- ferent discriminants but the generic arc section is irreducible in both cases. If we apply the previous algorithm to this example, we can find a smooth arc whose arc section has 3 components, just by making it have contact 3 with one of the lines of the discriminant. An explicit equation could be (t − t3, t + t3); then the equation for the arc section is −12t10 − 40t8 − 12t6 + z3, which is analytically equivalent to −12t6 + z3. Clearly this factorizes as the product of three components. References [1] Le D ung Tr´ang and Bernard Teissier. Sur la g´eom´etrie des surfaces com- plexes. I. Tangentes exceptionnelles. Amer. J. Math., 101(2):420 -- 452, 1979. [2] E. Artin. Theory of braids. Ann. of Math. (2), 48:101 -- 126, 1947. 13
1307.2316	1	1307	2013-07-09T01:28:04	Transformations of polar Grassmannians preserving certain intersecting relations	[ "math.AG", "math.CO" ]	Let $\Pi$ be a polar space of rank $n\ge 3$. Denote by ${\mathcal G}_{k}(\Pi)$ the polar Grassmannian formed by singular subspaces of $\Pi$ whose projective dimension is equal to $k$. Suppose that $k$ is an integer not greater than $n-2$ and consider the relation ${\mathfrak R}_{i,j}$, $0\le i\le j\le k+1$ formed by all pairs $(X,Y)\in {\mathcal G}_{k}(\Pi)\times {\mathcal G}_{k}(\Pi)$ such that $\dim_{p}(X^{\perp}\cap Y)=k-i$ and $\dim_{p} (X\cap Y)=k-j$ ($X^{\perp}$ consists of all points of $\Pi$ collinear to every point of $X$). We show that every bijective transformation of ${\mathcal G}_{k}(\Pi)$ preserving ${\mathfrak R}_{1,1}$ is induced by an automorphism of $\Pi$ and the same holds for the relation ${\mathfrak R}_{0,t}$ if $n\ge 2t\ge 4$ and $k=n-t-1$. In the case when $\Pi$ is a finite classical polar space, we establish that the valencies of ${\mathfrak R}_{i,j}$ and ${\mathfrak R}_{i',j'}$ are distinct if $(i,j)\ne (i',j')$.	math.AG	math	Transformations of polar Grassmannians preserving certain intersecting relations Wen Liua,b Mark Pankovc Kaishun Wanga ∗ a Sch. Math. Sci. & Lab. Math. Com. Sys., Beijing Normal University, Beijing, 100875, China b Ins. Math. & Inf., Hebei Normal University, Shijiazhuang, 050024, China c Department of Mathematics and Computer Science, University of Warmia and Mazury Olsztyn, Poland Abstract Let Π be a polar space of rank n ≥ 3. Denote by Gk(Π) the polar Grassmannian formed by singular subspaces of Π whose projective dimension is equal to k. Suppose that k is an integer not greater than n − 2 and consider the relation Ri,j, 0 ≤ i ≤ j ≤ k + 1 formed by all pairs (X, Y ) ∈ Gk(Π) × Gk(Π) such that dimp(X ⊥ ∩ Y ) = k − i and dimp(X ∩ Y ) = k − j (X ⊥ consists of all points of Π collinear to every point of X). We show that every bijective transformation of Gk(Π) preserving R1,1 is induced by an automorphism of Π and the same holds for the relation R0,t if n ≥ 2t ≥ 4 and k = n − t − 1. In the case when Π is a finite classical polar space, we establish that the valencies of Ri,j and Ri′ ,j ′ are distinct if (i, j) 6= (i′, j ′). 2010 MSC: 51A50; 51E20 Key words: Polar space; polar Grassmannian; intersecting relation; automorphism 1 Introduction Let V be an n-dimensional vector space V (over a division ring). Denote by Gk(V ) the Grassmannian formed by k-dimensional subspaces of V . Suppose that 1 < k ≤ n − k. For every integer i satisfying 1 ≤ i ≤ k we define the following relation Ri := {(X, Y ) ∈ Gk(V ) × Gk(V ) : diml(X ∩ Y ) = k − i}. We write diml for the linear dimension (the dimension of vector spaces and their subspaces), since we want to distinguish it from the projective dimension (the dimension of projective spaces and their subspaces) which will be denoted by dimp. Note that (X, Y ) ∈ Ri if and only if the distance between X and Y in the Grassmann graph Γk(V ) = (Gk(V ), R1) is equal to i. The same relations can be defined on dual polar spaces. Let Π be a polar space of rank n. Denote by Gk(Π) the polar Grassmannian formed by singular subspaces of Π whose projective dimension is equal to k. The associated dual polar space is formed by maximal singular subspaces, i.e. singular subspaces of dimension n − 1. For any integer i satisfying 1 ≤ i ≤ n we define Ri := {(X, Y ) ∈ Gn−1(Π) × Gn−1(Π) : dimp(X ∩ Y ) = n − 1 − i}. ∗Corresponding author. Email addresses: [email protected] (Kaishun Wang) [email protected] (Wen Liu), [email protected] (Mark Pankov), 1 As above, we have (X, Y ) ∈ Ri if and only if the distance between X and Y in the dual polar graph Γn−1(Π) = (Gn−1(Π), R1) is equal to i. By [4], every automorphism of the Grassmann graph Γk(V ) is induced by a semilinear automorphism of V or a semilinear isomorphism of V to the dual vector space V ∗ (the second possibility can be realized only in the case when n = 2k). Similarly, every automorphism of the dual polar graph Γn−1(Π) is induced by an automorphism of the polar space Π. The latter was proved by Chow [4] for classical polar spaces only, but Chow's method works in the general case [20, Section 4.6]. Some results closely related to these statements were obtained [7, 8, 9, 10, 11, 12, 16, 21, 22] and we refer [20] for a survey. Every bijective transformation of Gk(V ) preserving Rk is an automorphism of Γk(V ) [2, 9]. For the relation Ri with 1 < i < k the same is not proved. However, all bijective transformations of Gk(V ) preserving R1 ∪ · · · ∪ Rm are automorphisms of Γk(V ) for every integer m < k [17]. This is a generalization of the previous result; indeed, if m = k − 1 then the considered above transformations preserve Rk. The same statement is proved for some dual polar spaces [15]. Results of similar nature were established for other objects [1, 5, 6, 13, 14]. Now suppose that k is an integer not greater than n − 2 and consider the relation Ri,j, 0 ≤ i ≤ j ≤ k + 1 formed by all pairs satisfying the following conditions (X, Y ) ∈ Gk(Π) × Gk(Π) dimp(X ⊥ ∩ Y ) = k − i and dimp(X ∩ Y ) = k − j (X ⊥ consists of all points of Π collinear to every point of X). All automorphisms of the polar Grassmann graph Γk(Π) = (Gk(Π), R0,1) are described in [20, Section 4.6]. They are induced by automorphisms of Π, except the case when n = 4, k = 1 and our polar space is of type D (every (n − 2)-dimensional singular subspace is contained in precisely two maximal singular subspaces). Also, every automorphism of so-called weak Grassmann graph Γw k (Π) = (Gk(Π), R0,1 ∪ R1,1) is induced by an automorphism of Π [19]. We show that every bijective transformation of Gk(Π) preserving R1,1 is induced by an automorphism of Π (Theorem 1). Our second result (Theorem 2) states that the same holds for the relation R0,t if n ≥ 2t ≥ 4 and k = n − t − 1. The latter relation is equivalent to the fact that two elements with the relation R0,t span a maximal singular subspace. Note that for finite symplectic and hermitian polar spaces, the first result under some conditions was proved in [18, 27]. For a finite classical polar space, the valency of every Ri,j was given by Stanton [23]. We establish that the valencies of Ri,j and Ri′,j ′ are distinct if (i, j) 6= (i′, j ′) (Proposition 3). 2 Polar spaces We recall some basic properties of polar spaces and refer [3, 20, 25] for their proofs. Let P be a non-empty set whose elements are called points and L be a family formed by proper subsets of P called lines. Two distinct points joined by a line are said to be collinear. 2 Let Π = (P, L) be a partial linear space, i.e. each line contains at least two points and for any distinct collinear points p, q ∈ P there is precisely one line containing them, this line is denoted by pq. We say that S ⊂ P is a subspace of Π if for any distinct collinear points p, q ∈ S the line pq is contained in S. A singular subspace is a subspace where any two distinct points are collinear. Note that the empty set and a single point are singular subspaces. Using Zorn lemma, we show that every singular subspace is contained in a certain maximal singular subspace. From this moment we suppose that Π is a polar space. This means that the following axioms hold: (P1) each line contains at least three points, (P2) there is no point collinear to all points, (P3) if p ∈ P and L ∈ L then p is collinear to precisely one point or all points of the line L, (P4) every flag formed by singular subspaces is finite. If there is a maximal singular subspace of Π containing more than one line then all maximal singular subspaces of Π are projective spaces of the same finite dimension ≥ 2. We say that the rank of Π is n if this dimension is equal to n − 1. In the case when the rank of Π is not less than 4, every maximal singular subspace M can be identified with the projective space associated to a certain n-dimensional vector space V (over a division ring). Then every non-empty singular subspace S ⊂ M will be identified with the corresponding subspace of the vector space V . The collinearity relation on Π is denoted by ⊥. We write p ⊥ q if p, q ∈ P are collinear points and p 6⊥ q otherwise. If X, Y ⊂ P then X ⊥ Y means that every point of X is collinear to all points of Y . For every subset X ⊂ P satisfying X ⊥ X the minimal singular subspace containing X is called spanned by X and denoted by hXi. For every subset X ⊂ P we denote by X ⊥ the subspace of Π formed by all points collinear to all points of X. Fact 1 Let X be a subset of P satisfying X ⊥ X and spanning a maximal singular subspace M . Then p ⊥ X implies that p ∈ M . Fact 2 If M is a maximal singular subspace of Π then for every point p ∈ P such that p 6∈ M we have dimp(p⊥ ∩ M ) = n − 2. Fact 3 For every singular subspace S there are maximal singular subspaces M and N such that S = M ∩ N . 3 Results Let Π = (P, L) be a polar space of rank n ≥ 3. Recall that the polar Grassmannian formed by all k-dimensional singular subspaces of Π is denoted by Gk(Π). Suppose that k ≤ n − 2. We consider the relation Ri,j, 0 ≤ i ≤ j ≤ k + 1 is formed by all pairs (X, Y ) ∈ Gk(Π) × Gk(Π) dimp(X ⊥ ∩ Y ) = k − i and dimp(X ∩ Y ) = k − j. satisfying Since the equality dimp(X ⊥ ∩ Y ) = dimp(Y ⊥ ∩ X) 3 holds for any pair X, Y ∈ Gk(Π), this relation is symmetric. Every automorphism of Π (a bijective transformation of P preserving L) induces a transformation of Gk(Π) which preserves all Ri,j. First, we determine all automorphisms of the graph Γ′ k(Π) = (Gk(Π), R1,1). Theorem 1 Every automorphism of Γ′ k(Π) is induced by an automorphism of Π. Remark 1 For k = 0 this statement is trivial. The edges of Γ′ 0(Π) are pairs of non-collinear points of Π and every automorphism of this graph is an automorphism of the collinearity graph Γ0(Π). It is well-known that the class of automorphisms of Γ0(Π) coincides with the class of automorphisms of Π. In the case when k ∈ {1, . . . , n−3}, the distance between S, U ∈ Gk(Π) in the Grassmann graph Γk(Π) is equal to 2 if and only if (S, U ) belongs to R1,1 ∪ R0,2. The distance between S, U ∈ Gn−2(Π) in Γn−2(Π) is equal to 2 if and only if (S, U ) ∈ R1,1 (if k = n − 2 then R0,2 is empty). Theorem 1 gives the following. Corollary 1 Let f be a bijective transformation of Gn−2(Π) satisfying the following con- the distance between S, U ∈ Gn−2(Π) in Γn−2(Π) is equal to 2 if and only if the dition: distance between f (S) and f (U ) in Γn−2(Π) is equal to 2. Then f is an automorphism of Γn−2(Π). Suppose that k = n−t−1, where t is an integer satisfying n ≥ 2t ≥ 4. Then (S, U ) ∈ R0,t is equivalent to the fact that S and U span a maximal singular subspace of Π. Our second result describes all automorphisms of the graph Theorem 2 Every automorphism of Γ′′ n−t−1(Π) is induced by an automorphism of Π. Γ′′ k(Π) = (Gk(Π), R0,t). 4 Cliques From this moment we suppose that k ∈ {1, . . . , n − 2}. For every singular subspace N such that dimp N < k we denote by [N ik the set of all elements of Gk(Π) containing N . This subset is said to be a big star if N ∈ Gk−1(Π). If N and M are singular subspaces satisfying N ⊂ M and dimp N < k < dimp M then we denote by [N, M ]k the set of all S ∈ Gk(Π) such that N ⊂ S ⊂ M . This subset is called a star if N ∈ Gk−1(Π) and M ∈ Gn−1(Π). In the case when N = ∅, we write hM ]k instead of [N, M ]k. We say that hM ]k is a top if M ∈ Gk+1(Π). All maximal cliques of the Grassmann graph and the weak Grassmann graph Γk(Π) = (Gk(Π), R0,1) Γw k (Π) = (Gk(Π), R0,1 ∪ R1,1) are known [20, Subsection 4.5.1 and Subsection 4.6.2]. Every maximal clique of Γw k (Π) is a big star or a top. Every maximal clique of Γk(Π) is a star or a top. In the case when k = n − 2, every star is contained in a certain top and all maximal cliques of Γk(Π) are tops. 4 Proposition 1 Every clique of Γ′ k(Π) is contained in a big star. If C is a clique of Γ′ k(Π) then it is a clique of Γw Proof. star or a top. Since any two distinct elements of a top are non-adjacent vertices of Γ′ C is a subset in a big star. k (Π). Hence C is contained in a big k(Π), ✷ Let N ∈ Gk−1(Π). For every M ∈ Gk+1(Π) containing N the subset [N, M ]k is said to be a line. The big star [N ik together with all lines defined above is a polar space of rank n − k [20, Lemma 4.4]. This polar space will be denoted by ΠN . Lemma 1 Let N ∈ Gk−1(Π). Two distinct elements of the big star [N ik are adjacent vertices of the graph Γ′ k(Π) if and only if they are non-collinear points of the polar space ΠN . Proof. Easy verification. ✷ Proposition 2 Suppose that k = n − t − 1, where t is an integer satisfying n ≥ 2t ≥ 4. If S and U are adjacent vertices of Γ′′ k(Π), then M := hS ∪ U i is a maximal singular subspace of Π and every clique of Γ′′ k(Π) containing S and U is a subset in hM ]k. It is clear that S ⊥ U and M is a maximal singular subspace. Let C be a clique of k(Π) containing S and U . Then for every A ∈ C we have A ⊥ S and A ⊥ U . By Fact 1, ✷ Proof. Γ′′ this implies that A ⊂ M . Hence C is contained in hM ]k. 5 Proof of Theorem 1 Lemma 2 Let p and q be non-collinear points of Π and let t be a point of Π collinear to at least one of the points p, q. Then there exists a point of Π non-collinear to p, q, t. Proof. Suppose that the statement fails and every point of Π is collinear to at least one of the points p, q, t, in other words, P = p⊥ ∪ q⊥ ∪ t⊥. (1) First we show that every maximal singular subspace of Π contains at least one of the points p, q, t. Let M be a maximal singular subspace of Π. If each of the points p, q, t does not belong to M then p⊥ ∩ M, q⊥ ∩ M, t⊥ ∩ M are (n − 2)-dimensional subspaces of M (Fact 2) and (1) implies that M is the union of these subspaces. The latter is impossible, since a projective space cannot be presented as the union of three hyperplanes. By our assumption, t is collinear to at least one of the points p and q. Suppose that q ⊥ t. Since p 6⊥ q, the line qt contains the unique point s collinear to p. One of the following possibilities is realized: (1) s = t, (2) s 6= t. 5 In the case (1), we take any point v on the line qt different from q and t. It is clear that p 6⊥ v. By Fact 3, there exist maximal singular subspaces M and N such that M ∩ N = {v}. Since p 6⊥ v, they do not contain p. Then one of these subspaces contains q and the other contains t. So, each of these subspaces contains two distinct points of the line qt. This means that this line is contained in M ∩ N which is impossible. In the case (2), we take any point w on the line ps different from p and s. As in the previous case, we consider maximal singular subspaces M and N such that M ∩ N = {w}. One of these subspaces contains q or t. Then at least one of the points q, t is collinear to w. Since q and t both are collinear to s and w is on the line ps, one of the points q, t is collinear to all points of the line ps. Thus p is collinear to q or t which is impossible. ✷ Lemma 3 Suppose that N ∈ Gk−1(Π). Let P, Q ∈ [N ik be adjacent vertices of Γ′ let T ∈ [N ik be a vertex of Γ′ there exists S ∈ [N ik adjacent to P, Q, T in Γ′ k(Π) and k(Π) non-adjacent to at least one of the vertices P, Q. Then k(Π). Proof. By Lemma 1, P and Q are non-collinear points of ΠN and T is a point of ΠN collinear to at least one of the points P, Q. We apply Lemma 2 to the polar space ΠN and get the claim. ✷ Let f be an automorphism of Γ′ Show that f transfers big stars to subsets of big stars. We take any N ∈ Gk−1(Π). Let k(Π) contained in the big star [N ik. Then f (P ) and f (Q) k(Π). P and Q be adjacent vertices of Γ′ are adjacent vertices of Γ′ k(Π) contained in the big star [N ′ik, where N ′ = f (P ) ∩ f (Q). We assert that f (T ) ∈ [N ′ik for every T ∈ [N ik and prove this statement in several steps. k(Π) adjacent to both P and Q. k(Π) which, by Proposition 1, is contained in a (i). First, we consider the case when T is a vertex of Γ′ Then f (P ), f (Q), f (T ) form a clique in Γ′ certain big star [N ′′ik. We have which gives the claim. N ′ = f (P ) ∩ f (Q) = N ′′ (ii). Consider the case when T is a vertex of Γ′ k(Π) adjacent to precisely one of the vertices P, Q. Suppose that T is adjacent to P . Lemma 3 implies the existence of a vertex S ∈ [N ik in the graph Γ′ k(Π) adjacent to P, Q, T . By (i), f (S) belongs to [N ′ik. Then f (P ) ∩ f (S) = N ′ and f (P ), f (S), f (T ) form a clique of Γ′ k(Π). As in (i), we show that f (T ) ∈ [N ′ik. (iii). Suppose that T is a vertex of Γ′ k(Π) non-adjacent to both P and Q. As above, we consider S ∈ [N ik which is a vertex of Γ′ k(Π) adjacent to P, Q, T and obtain that f (S) ∈ [N ′ik. We apply the arguments from (ii) to P, S, T and establish that f (T ) ∈ [N ′ik. So, f transfers big stars to subsets of big stars. The same arguments show that f −1 sends big stars to subsets of big stars. This means that f and f −1 both map big stars to big stars, i.e. there exists a bijective transformation g of Gk−1(Π) such that f ([N ik) = [g(N )ik for every N ∈ Gk−1(Π). Let U ∈ Gk(Π). Then g transfers the top hU ]k−1 to the top hf (U )]k−1. Indeed, we have N ∈ hU ]k−1 ⇔ U ∈ [N ik ⇔ f (U ) ∈ [g(N )ik ⇔ g(N ) ∈ hf (U )]k−1. 6 Similarly, g−1 sends hU ]k−1 to the top hf −1(U )]k−1. Therefore, g and g−1 both transfer tops to tops. Since for any two adjacent vertices of the Grassmann graph Γk−1(Π) there is a top containing them, g is an automorphism of Γk−1(Π). In some special cases, there are automorphisms of Γk−1(Π) which are not induced by automorphisms of Π. However, every automorphism of Γk−1(Π) transferring tops to tops is induced by an automorphism of Π [20, Section 4.6.1]. Thus g is induced by an automorphism of Π. An easy verification shows that this automorphism also induces f . 6 Proof of Theorem 2 In this section we suppose that k = n − t − 1, where t is an integer satisfying n ≥ 2t ≥ 4. Lemma 4 Let S, U be adjacent vertices of Γ′′ subspace spanned by S and U . If T ∈ hM ]k is a vertex of Γ′′ one of the vertices S, U , then there exists a vertex Q of Γ′′ k(Π). Let also M be the maximal singular k(Π) non-adjacent to at least k(Π) adjacent to S, U, T . Proof. We suppose that T is non-adjacent to S (the case when T is non-adjacent to U is similar). First of all, we show that the general case can be reduced to the case when N := S ∩ U and T are disjoint. Indeed, if the projective dimension of W := N ∩ T is equal to w ≥ 0 then S, U, T can be naturally identified with vertices of the graph Γ′′ k−w−1(ΠW ) which is isomorphic to the subgraph of Γ′′ k(Π) induced on [W ik. So, we suppose that N and T are disjoint. Since n ≥ 4, we identify M with the projective space associated to an n-dimensional vector space V . Every non-empty singular subspace of M will be identified with the corresponding subspace of the vector space V . We set m := diml N = n − 2t and suppose that diml(S ∩ T ) = i, diml(U ∩ T ) = j. Note that i > m, since S and T are non-adjacent vertices of Γ′′ k(Π). First we consider the case when i + j = n − t, i.e. T is spanned by S ∩ T and U ∩ T . Since diml(S/N ) = diml(U/N ) = t, we can choose vectors x1, . . . , xt ∈ S \ N and y1, . . . , yt ∈ U \ N such that S = N + hx1, . . . , xti, U = N + hy1, . . . , yti, T = hx1, . . . , xii + hyt−j+1, . . . , yti. The vectors x1 + y1, . . . , xt + yt are linearly independent. We define Q := N + hx1 + y1, . . . , xt + yti. An easy verification shows that S + Q = U + Q = T + Q = M which implies that Q is a vertex of Γ′′ k(Π) adjacent to S, U, T . Now suppose that l := n − t − (i + j) > 0. Then t = i + j − m + l. Since i > m, we have i + j − m > 0. We choose linearly independent vectors x1, . . . , xi+j−m, x′ 1, . . . , x′ l ∈ S \ N 7 and linearly independent vectors y1, . . . , yi+j−m, y′ 1, . . . , y′ l ∈ U \ N such that T = hx1, . . . , xii + hyi−m+1, . . . , yi+j−mi + hx′ 1 + y′ 1, . . . , x′ l + y′ li. We denote by Q the subspace spanned by N and the vectors x1 + y′ 1, x′ 1 + y′ 2, . . . , x′ l−1 + y′ l, x′ l + y1, x2 + y2, . . . , xi+j−m + yi+j−m. Using the equalities S = N + hx1, . . . , xi+j−m, x′ U = N + hy1, . . . , yi+j−m, y′ li, 1, . . . , x′ 1, . . . , y′ li, we establish that S + Q = U + Q = M . To complete the proof we need to check that T + Q = M . The conditions x1, . . . , xi, yi−m+1, . . . , yi+j−m ∈ T and x2 + y2, . . . , xi+j−m + yi+j−m ∈ Q imply that xp ∈ T + Q for every p and yp ∈ T + Q if p ≥ 2. Since x1 ∈ T and x1 + y′ we have y′ that all x′ that y1 ∈ T + Q. Therefore, Q + T coincides with S + U = M . 1 ∈ Q, 1 ∈ T + Q. Step by step, we establish l ∈ T + Q and x′ l + y1 ∈ Q guarantee ✷ 1 ∈ T implies that x′ q belong to T + Q. The conditions x′ 1 ∈ T + Q. Then x′ q and y′ 1 + y′ Let f be an automorphism of Γ′′ k(Π). Show that for every M ∈ Gn−1(Π) there exists M ′ ∈ Gn−1(Π) such that f (hM ]k) ⊂ hM ′]k. We choose S, U ∈ hM ]k such that M is spanned by S ∪ U . Then S and U are adjacent vertices of Γ′′ k(Π) and the same holds for f (S) and f (U ). Hence M ′ := hf (S) ∪ f (U )i is a maximal singular subspace. We assert that f (T ) ∈ hM ′]k for every T ∈ hM ]k and prove this statement in several steps. (i). If T ∈ hM ]k is a vertex of Γ′′ k(Π) adjacent to S and U then f (S), f (U ), f (T ) form a clique in Γ′′ k(Π) and, by Proposition 2, we have f (T ) ∈ hM ′]k. (ii). Now, let T ∈ hM ]k be a vertex of Γ′′ k(Π) adjacent to precisely one of the vertices U and S. Suppose that T is adjacent to U and non-adjacent to S. Let Q ∈ hM ]k be a vertex of Γ′′ k(Π) adjacent to S, U, T (Lemma 4). By (i), f (Q) belongs to hM ′]k. Then We apply (i) to U, Q, T and establish that f (T ) belongs to hM ′]k. hf (U ) ∪ f (Q)i = M ′. (2) (iii). Consider the case when T is a vertex of Γ′′ k(Π) non-adjacent to both S and U . k(Π) adjacent to S, U, T . Then As above, we consider Q ∈ hM ]k which is a vertex of Γ′′ f (Q) ∈ hM ′]k and (2) holds. We apply (ii) to U, Q, T and obtain that f (T ) ∈ hM ′]k. We apply the above arguments to f −1 and establish the existence of a bijective trans- formation g of Gn−1(Π) such that for every M ∈ Gn−1(Π). f (hM ]k) = hg(M )]k 8 Let U be a singular subspace of Π such that k < dimp U < n − 1. Then there exist M1, M2 ∈ Gn−1(Π) satisfying U = M1 ∩ M2. We have hU ]k = hM1]k ∩ hM2]k, f (hU ]k) = hg(M1)]k ∩ hg(M2)]k = hg(M1) ∩ g(M2)]k. We set and get an extension of g to a transformation of g(U ) := g(M1) ∩ g(M2) Gk+1(Π) ∪ · · · ∪ Gn−1(Π) (3) such that f (hU ]k) = hg(U )]k for every U belonging to (3). We apply the same arguments to the pair g−1, f −1 and show that g is bijective. It is clear that f is inclusion preserving, i.e. S ⊂ U ⇐⇒ f (S) ⊂ f (U ) for any S, U belonging to (3). The latter guarantees that f sends Gi(Π) to itself for every integer i satisfying k < i ≤ n− 1. Therefore, f and f −1 both send tops to tops which implies that f is induced by an automorphism of Π. 7 Valency of the relation for finite classical polar spaces Let V be an n-dimensional vector space over the finite field Fq together with a non-degenerate symplectic, hermitian or symmetric form of Witt index d ≥ 2. For every m ∈ {1, . . . , d} we denote by Nm the set of all m-dimensional totally isotropic subspaces. The associated polar space Π is defined as follows: the point set is N1 and the lines are defined by elements of N2 (the line corresponding to S ∈ N2 consists of all elements of N1 contained in S). This is a polar space of rank d and every Gi(Π) can be naturally identified with Ni+1. For more information, see [24, 26]. For 0 ≤ i ≤ j ≤ m, denote by ni,j the valency of the relation {(S, U ) ∈ Nm × Nm : diml(S⊥ ∩ U ) = m − i, diml(S ∩ U ) = m − j}, (4) where S⊥ is the orthogonal complement to S. If m < d then (4) coincides with the relation Ri,j. In the case when m = d, we have S⊥ ∩ U = S ∩ U for any S, U ∈ Nm and (4) is the relation Ri defined on the dual polar space. The valency ni,j is computed in [23] as follows: ni,j = qj2+i(n−2m−2j+ 3 2 i+ 1 2 −µ−ν)(cid:20)m j (cid:21)(cid:20)j i(cid:21) j−i−1 Ys=0 n 2 (q −m−µ−s − 1)(q n 2 −m−ν−s + 1)(qs+1 − 1)−1, where µ = 1 hermitian and symmetric cases, respectively. 2 n − d and ν is a number such that µ + ν equals 0, 1 2 , 1 in the symplectic, Proposition 3 All valencies ni,j are pairwise distinct. 9 Lemma 5 Let f (x) be a polynomial of degree at least 1 over the rational number field Q. (i) If f (x) has the following two factorizations: f (x) = (xr1 − 1)i1(xr2 − 1)i2 · · · (xrl − 1)il = (xt1 − 1)j1 (xt2 − 1)j2 · · · (xth − 1)jh , where r1 > · · · > rl and t1 > · · · > th, then h = l, rs = ts and is = js, s = 1, . . . , l. (ii) If f (x) has the following two factorizations: f (x) = (xr + 1)(xr1 − 1)i1 · · · (xrl − 1)il = (xt1 − 1)j1 · · · (xth − 1)jh , then there exists an s ∈ {1, . . . , h} such that ts = 2r. (iii) If f (x) has the following two factorizations: f (x) = (xk1 + 1) · · · (xkf + 1)(xr1 − 1)i1 · · · (xrl − 1)il = (xt1 − 1)j1 · · · (xth − 1)jh , where k1, . . . , kf are distinct odd integers, then there exist s1, . . . sf ∈ {1, . . . , h} such that ts1 = 2k1, . . . , tsf = 2kf . Proof. (i) Consider the term of second minimum degree of f (x), we have (−1)i1+···+il−1ilxrl = (−1)j1+···+jh−1jhxth , which implies that jh = il and th = rl. So (xr1 − 1)i1 · · · (xrl−1 − 1)il−1 = (xt1 − 1)j1 · · · (xth−1 − 1)jh−1 , By induction, (i) holds. (ii) The polynomial (xr − 1)f (x) has the following two factorizations: (xr − 1)f (x) = (x2r − 1)(xr1 − 1)i1 · · · (xrl − 1)il = (xr − 1)(xt1 − 1)j1 · · · (xth − 1)jh . So (ii) holds by (i). The proof of (iii) is similar to that of (ii), and will be omitted. ✷ Proof of Proposition 3: Suppose na,b = na′,b′. Since q is a prime power, we obtain b2 + a(n − 2m − 2b + 3 2 a + 1 2 − µ − ν) = b′2 + a′(n − 2m − 2b′ + 3 2 a′ + 1 2 − µ − ν), (5) and b−a−1 b′−a′−1 (cid:2)m b(cid:3)(cid:2)b a(cid:3) b′(cid:3)(cid:2)b′ = (cid:2)m a′(cid:3) Qs=0 Qs=0 (q n 2 −m−µ−s − 1)(q n 2 −m−ν−s + 1)(qs+1 − 1)−1 (q n 2 −m−µ−s − 1)(q n 2 −m−ν−s + 1)(qs+1 − 1)−1. Simplifying (5), we have (b − b′)(2b + 2b′ − 4a′) = (a′ − a)(2l + 3a′ + 3a − 4b), where l = n − 2m + 1 2 − µ − ν. Write (6) (7) fi,j(x) = gi,j(x) = j−i−1 m−j Qs=0 Qs=1 (x2s − 1) (xn−2m−2µ−2s − 1)(xn−2m−2ν−2s + 1), (x2s − 1) i Qs=1 (x2s − 1)2. j−i Qs=1 10 The equality (6) implies that fa,b(q)ga′,b′(q) = fa′,b′(q)ga,b(q) for all prime powers q, so fa,b(x)ga′,b′ (x) = fa′,b′(x)ga,b(x). Computing the degree of this polynomial, we have (b − b′)(2l − 5b − 5b′ + m + 8a′) = (a′ − a)(−2l − 5a − 5a′ + 8b). Equalities (7) and (9) imply that (b − b′)(2l + 2m − b − b′) = (a′ − a)(2l + a + a′). (8) (9) (10) Suppose b 6= b′. Without loss of generality, we assume that b > b′. Since 2l+2m−b−b′ > 0 and 2l + a + a′ > 0, by (10) one gets a′ > a and b − a − b′ + a′ ≥ 2. (11) Write k(x) = h(x) = b−a b−a−1 Qs=b′−a′+1 Qs=b′−a′ (x2s − 1)2, (xn−2m−2µ−2s − 1)(xn−2m−2ν−2s + 1) m−b′ Qs=m−b+1 (x2s − 1) a′ Qs=a+1 (x2s − 1). The equation (8) implies that k(x) = h(x). Set h1(x) = x2n−4m−4ν−4b′+4a′ − 1, h2(x) = x2n−4m−4ν−4b+4a+4 − 1. By Lemma 5 there exist s1, s2 ∈ {b′ − a′ + 1, . . . , b − a} such that h1(x) = x2s1 − 1 and h2(x) = x2s2 − 1. It follows that s1 = n − 2m − 2ν − 2b′ + 2a′, s2 = n − 2m − 2ν − 2b + 2a + 2. By (11) we have s1 − s2 = 2b − 2a − 2b′ + 2a′ − 2 > b − a − (b′ − a′ + 1), a contradiction. So b = b′ and a = a′, as desired. ✷ Remark 2 It is well known that Nm has a structure of a symmetric association scheme, each relation of which has form (4). By Theorem 1 every automorphism of this scheme is induced by an automorphism of the associated polar space Π. Acknowledgement W. Liu was supported by NSFC(11271004). K. Wang was supported by NSFC(11271047) and the Fundamental Research Funds for the Central University of China. References [1] P. Abramenko, H. Van Maldeghem, On opposition in spherical buildings and twin buildings, Ann. Comb. 4 (2000), 125 -- 137. [2] A. Blunck, H. Havlicek, On bijections that preserve complementarity of subspaces, Discrete Math. 301 (2005), 46 -- 56. [3] F. Buekenhout, A. M. Cohen, Diagram geometry, Springer 2013. 11 [4] W.L. Chow, On the geometry of algebraic homogeneous spaces, Ann. of Math. 50 (1949), 32 -- 67. [5] E. Govaert, H. Van Maldeghem, Distance-preserving maps in generalized polygons. Part I: Maps on flags, Beitrage Algebra Geom. 43 (2002), 89 -- 110. [6] E. Govaert, H. Van Maldeghem, Distance-preserving maps in generalized polygons. Part II: Maps on points and/or lines, Beitrage Algebra Geom. 43 (2002), 303 -- 324. [7] H. Havlicek, On isomorphisms of Grassmann spaces, Mitt. Math. Ges. Hamburg. 14 (1995), 117 -- 120. [8] H. Havlicek, Chow's theorem for linear spaces, Discrete Math. 208/209 (1999), 319 -- 324. [9] H. Havlicek, M. Pankov, Transformations on the product of Grassmann spaces. Demon- stratio Math. 38 (2005), 675 -- 688. [10] W.-l. Huang, Adjacency preserving transformations of Grassmann spaces, Abh. Math. Sem. Univ. Hamburg 68 (1998),65 -- 77. [11] W.-l. Huang, Adjacency preserving mappings of invariant subspaces of a null system, Proc. Amer. Math. Soc. 128 (2000), 2451 -- 2455. [12] W.-l. Huang, Characterization of the transformation group of the space of a null sys- tem, Results Math. 40 (2001), 226 -- 232. [13] W.-l. Huang, H. Havlicek, Diameter preserving surjections in the geometry of matrices, Linear Algebra Appl. 429 (2008), 376 -- 386. [14] W.-l. Huang, Bounded distance preserving surjections in the geometry of matrices, Linear Algebra Appl. 433 (2010), 1973-1987. [15] W.-l. Huang, Bounded distance preserving surjections in the projective geometry of matrices, Linear Algebra Appl. 435 (2011), 175 -- 185. [16] A. Kreuzer, On isomorphisms of Grassmann spaces, Aequationes Math. 56 (1998), 243 -- 250. [17] M. H. Lim, Surjections on Grassmannians preserving pairs of elements with bounded distance. Linear Algebra Appl. 432 (2010), 1703 -- 1707. [18] W. Liu, C. Ma, K. Wang, Full automorphism group of generalized unitary graphs, Linear Algebra Appl. 437 (2012) 684 -- 691. [19] M. Pankov, K. Prazmovski, M. Zynel, Geometry of polar Grassmann spaces, Demon- stratio Math., 39 (2006), 625-637. [20] M. Pankov, Grassmannians of classical buildings, Algebra and Discrete Math. 2, World Scientific, Singapore, 2010. [21] M. Pankov, Metric characterization of apartments in dual polar spaces, J. Combin. Theory Ser. A 118 (2011), 1313 -- 1321. [22] M. Pankov, Embeddings of Grassmann graphs, Linear Algebra Appl. 436 (2012), 3413 -- 3424. [23] D. Stanton, Some q-Krawtchouk polynomials on Chevalley groups, Amer. J. Math. 102 (1980), 625 -- 662. 12 [24] D.E. Taylor, The geometry of the classical groups, Heldermann Verlag, Berlin, 1992. [25] J. Ueberberg, Foundations of incidence geometry. Projective and polar spaces, Springer 2011. [26] Z. Wan, Geometry of classical groups over finite fields, Science Press, Bejing, New York, 2002. [27] L. Zeng, Z. Chai, R. Feng, C. Ma, Full automorphism group of the generalized sym- plectic graph, Sci. China Math., doi: 10.1007/s11425-013-4651-8. 13
1011.3236	3	1011	2012-07-27T17:29:38	Phylogenetic invariants for group-based models	[ "math.AG" ]	In this paper we investigate properties of algebraic varieties representing group-based phylogenetic models. We propose a method of generating many phylogenetic invariants. We prove that we obtain all invariants for any tree for the binary Jukes-Cantor model. We conjecture that our method can give all phylogenetic invariants for any tree. We show that for 3-Kimura our conjecture is equivalent to the conjecture of Sturmfels and Sullivant. This, combined with the results of Sturmfels and Sullivant, would make it possible to determine all phylogenetic invariants for any tree for 3-Kimura model, and also other phylogenetic models. Next we give the (first) example of a non-normal general group-based model for an abelian group. Following Kubjas we also determine some invariants of group-based models showing that the associated varieties do not have to be deformation equivalent.	math.AG	math	PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS MARIA DONTEN-BURY AND MATEUSZ MICHA LEK Abstract. In this paper we investigate properties of algebraic varieties representing group- based phylogenetic models. We propose a method of generating many phylogenetic invari- ants. We prove that we obtain all invariants for any tree for the two-state Jukes-Cantor model. We conjecture that for a large class of models our method can give all phylogenetic invariants for any tree. We show that for 3-Kimura our conjecture is equivalent to the con- jecture of Sturmfels and Sullivant [22, Conjecture 2]. This, combined with the results in [22], would make it possible to determine all phylogenetic invariants for any tree for 3-Kimura model, and also other phylogenetic models. Next we give the (first) examples of non-normal varieties associated to general group-based model for an abelian group. Following Kubjas [17] we prove that for many group-based models varieties associated to trees with the same number of leaves do not have to be deformation equivalent. 1. Introduction Phylogenetics is a science that tries to reconstruct the history of evolution. It is strongly connected with many branches of mathematics including algebraic geometry. To each possi- ble history of evolution, represented by a tree, one can associate an algebraic variety, whose special points correspond to possible probability distributions on the DNA states of the living species. For a detailed introduction the reader is advised to look in [20] and for an algebraic point of view in [12]. From the point of view of applications, one is interested in computing phylogenetic in- variants that are polynomials defining the variety. It is very hard to find them in general, however for some special models of evolution much progress has been made. In this paper we are mainly dealing with a large class of equivariant models [11] -- so called general group- based models. The varieties associated to these models have a natural torus action with a dense orbit (see [15], [13], [18]). The most influential paper in this area is [22], where the authors gave the description of the generators of the ideal, assuming that the ideal of the claw tree is known. Unfortunately not much is known1 on the ideals of the claw trees, apart from the case of the two-state Jukes-Cantor model [8]. In particular we do not even know if the degree in which they are generated is bounded while the number of leaves grows to infinity (Conjecture 1 and 2 in [22]). The description of the ideal of the claw tree is also the main missing ingredient in the description of the ideal of any tree for equivariant models [11, p. 17]. 2010 Mathematics Subject Classification. 52B20, 13P25. Key words and phrases. phylogenetic tree, group-based model, phylogenetic invariant. The first author is supported by a grant of Polish MNiSzW (N N201 611 240). The second author is supported by a grant of Polish MNiSzW (N N201 413 539). 1After the submission of this paper a result concerning set-theoretic generation in some bounded degree was presented in [10]. 1 2 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK In this paper we propose a method of finding the ideals of the claw trees using a geometric approach -- cf. Section 3.2. We conjecture that the varieties associated to large claw trees are scheme-theoretic intersections of varieties associated to trees of smaller valency. This would enable generating the ideals recursively. We prove the conjecture for the Jukes-Cantor model 3.11. An interesting fact is that we can show that our conjecture is equivalent to the one made by Sturmfels and Sullivant for the 3-Kimura model, Proposition 3.14. For any general group -- based model the phylogenetic invariants of degree d for a claw tree with inner vertex of degree n can be explicitly derived from the phylogenetic invariants for trees with vertices of degree at most n − 1 for d << n. For the details see Section 3. We also investigate geometric properties of algebraic varieties representing phylogenetic models. In particular we give an example of a model associated to an abelian group that gives a non-normal variety. The variety associated to any group containing Z6 or Z2 × Z2 × Z2 or Z4 × Z2 or Z8 and the claw tree with three leaves is not normal (Computation 4.1 and Proposition 4.2). The variety associated to any abelian group of cardinality at most 5 or to Z7 and any trivalent tree is normal. The variety associated to the group Z2 and any tree is normal (Proposition 3.12). The results on normality are strongly connected to deformation problems. It is well- known that algebraic varieties representing trivalent trees with the same number of leaves are deformation equivalent for the binary Jukes-Cantor model. The original geometric proof can be found in [6] and a new, more combinatorial one, in [16]. A new result of Kaie Kubjas shows that this is not true for the 3-Kimura model [17]. Our results are as follows. The varieties associated to two different trivalent trees with the same number of leaves -- the caterpillar and snowflake -- and one of the groups Z3, Z4, Z5, Z7 or Z2 × Z2 have different Hilbert polynomials (Computation 4.4). The idea of the proof is to calculate the number of integer points in nP , where P is the polytope associated to the algebraic model of a phylogenetic tree. This task is much easier for normal varieties (in this case we obtain Hilbert-Ehrhart polynomial of the algebraic model). One of the tools that we use is a program that computes the polytope defining a toric vari- ety for a given tree and a group (see section 2.1). Our program, implementing the algorithm described in [18], can be found at http://www.mimuw.edu.pl/~marysia/polytopes (with a detailed instruction and specification of the input and output data format). It can be used to print the vertices of the polytope associated to a tree given by the user in an input file and one of the groups (with small numbers of elements) defined in the source code, but also by a slight modification of a source code one can obtain polytopes associated to models with other groups action. Acknowledgements The authors would like to thank very much Jaros law Wi´sniewski for introducing them to the topic and encouraging them to study it further. We thank Winfried Bruns for the help with computations. We would also like to thank Elizabeth Allman and Laurent Manivel for useful remarks. The first author is supported by a grant of Polish MNiSzW (N N201 611 240). The second author is supported by a grant of Polish MNiSzW (N N201 413 539). PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 3 2. General group-based models identify basis elements of W and states of investigated random variables. By a model we To a tree T , a vector space W with a distinguished basis and a subspace fW ⊂ End(W ) one can associate an algebraic variety X(T,fW ). People familiar with phylogenetics should mean a choice of fW . The construction of the variety X(T,fW ) is well-known for specialist and not necessary in general to understand the main ideas of the paper. It can be found for example in [12]. Thus we focus on a class of general group-based models and describe the construction in this setting. All the trees that we consider are rooted and we assume that the edges are directed away from the root. Definition 2.1 (General group-based model). Let W be the regular representation of a finite, G invariant endomorphisms of W . abelian group G. For a general group-based model we define the subspace fW := End(W )G as as the Cavender-Farris-Neyman model. The elements of fW represented as matrices are of Example 2.2. For the group Z2 we obtain the two-state Jukes-Cantor model also known the form: (cid:20) a b b a (cid:21) . For the group Z2 × Z2 we obtain the 3-Kimura model. The elements of fW represented as matrices are of the form:   c d a b b a d c c d a b d c b a   . AsfW is defined by the group G we will write X(T, G) instead of X(T,fW ). In this case we can easily describe the affine cone over the variety X(T, G). It is a spectrum of the semigroup algebra, with a semigroup generated by the lattice points of a polytope. The construction of the polytope depends on the tree and the group. It is purely combinatorial and very easy -- see Algorithm 1. The description of the defining polytope is also given in Theorem 2.5. Sometimes we do not specify the model and denote the associated variety X(T ). The fact that to a general group-based models one can associate a (not necessary normal) toric variety was observed by many authors [13], [24], [22]. To describe the defining polytope we need the following definitions. Definition 2.3 (Group based flows, sockets). A group based flow is a function n : E → G, where E is the set of edges of the tree. Moreover we require that for any inner vertex v of the tree, e0 an incoming edge and e1, . . . , ek outgoing edges we have n(e0) = n(e1) + · · · + n(ek). A socket is a function s : L → G, where L is the set of leaves of the tree, that is edges adjacent to vertices of degree one. Moreover we require that the sum Pl∈L s(l) is equal to the neutral element of the group. Let us explain the terminology. A group based flow assigns elements of a group to edges, with a condition that for each inner vertex the sum of elements associated to incoming edges equals the sum of those associated to outgoing edges. If the group was equal to R this would 4 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK be the condition for a flow, where leaves are sources and sinks. We use the terminology of group based flows, as this is a direct generalization to any abelian group of the notion introduced in [6] for Z2. As all the flows that we use are group based, we write flow, meaning a group based flow. Example 2.4. Consider the group G = Z3 and the following tree: ◦ e2 e1 ⑧⑧⑧⑧⑧⑧⑧⑧ ❇❇❇❇❇❇❇❇ e5 e4 e3 ⑧⑧⑧⑧⑧⑧⑧⑧ Here e2, e3, e4 and e5 are leaves. An example of a socket is an association e2 → 1, e3 → 1, e4 → 2, e5 → 2. We can make a flow using the same association and extending it by e1 → 2. By restricting a flow to leaves we get a socket and this map is in fact a bijection. Let us present a combinatorial description of the polytope that represents this toric variety. Theorem 2.5 ([18]). Let P be a polytope representing a general group-based model. All integral points of P are vertices. There is a natural bijection between vertices of P and flows. The polytope P is a subpolytope of a lattice with basis elements indexed by pairs (e, g) where e is an edge of a tree and g a group element. The vertex of P associated to a flow n is a sum of all basis elements indexed by such pairs (e, g) that satisfy n(e) = g. (cid:3) The generating binomials of a toric ideal associated to a polytope P correspond to integral relations between integer points of this polytope -- see for example [21], [9], [14]. Hence in our situation phylogenetic invariants correspond to relations between flows. In the abelian case each such relation can be described in the following way. We number all edges of a tree from 1 to e. The flows are specific e-tuples of group elements. For example for the claw tree these are e-tuples of group elements summing up to the neutral element. Each relation of degree d between the flows is encoded as a pair of matrices with d columns and e rows with entries that are group elements. We require that each column represents a flow. Moreover the rows of both matrices are the same up to permutation. For a general group-based model this is a purely combinatorial description of all phylogenetic invariants for any tree [22]. This construction can be generalized to other, so called G-models [18], [5]. Example 2.6. Consider the binary Jukes-Cantor model, that is the model corresponding to the group Z2 and the following tree. (2.1) ❆❆❆❆❆❆❆❆ ⑥⑥⑥⑥⑥⑥⑥ v1 ✂✂✂✂✂✂✂✂ ❁❁❁❁❁❁❁❁ PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 5 The leaves adjacent to v1 will be represented by first two rows. The third row corresponds to the inner edge. An example of a relation is given by a pair of matrices:   1 0 0 1 1 1 1 0 0 1 0 0   ,   0 1 1 0 1 1 1 0 0 1 0 0   . The numbers 0 and 1 are treated as elements of Z2. Due to the definition of the socket the third row has to be the sum of both the first two and last three rows. 2.1. Algorithm of finding the polytope. The first step of all our computations is passing from an abstract model to the lattice polytope associated to it. The following algorithm for computing the associated polytope was proposed in [18, Sect. 4]. Let E be the set of edges of the tree and N be the set of its inner vertices. Algorithm 1. (1) Orient the edges of the tree from the root. (2) For each inner vertex choose one outgoing edge. (3) Make a bijection b : G → B ⊂ ZG, where B is the standard basis of ZG. (4) Consider all possible associations of elements of G with not-chosen edges (there are GE−N such associations). (5) For each such associations, make a full association by assigning an element of G to each chosen edge in such a way that the (signed) sum of elements around each inner vertex gives a neutral element in G. (6) For each full association output the vertex of the polytope: (b(ge)e∈E), where ge is the element of the group associated to edge e. There are two non obvious points in the implementation. One is step 2 of the algorithm: making a choice of an outgoing edge from each vertex (the tree is rooted and the edges are directed from the root). It is much easier to choose incoming edge for each vertex except the root, as this choice is almost canonical (depends only on the rooting), so does not have to be stored in the memory. By precomputing the group operations and storing the result we obtain the complexity O(NGE−N ), as predicted in [18]. The current version of the program operates on the abelian groups defined in the source. As a result we have a fast program which takes a tree in a simple text format as an input and allows to choose one of the groups from the library. It computes the list of vertices of a polytope associated to the input model and outputs it to a file. It also enables the user to work with this polytope, given as an object of an inner class of the program, in the further computations. For example, it simplified significantly the programming necessary to perform the computations of Hilbert-Ehrhart polynomials, described in 4.2. 3. Phylogenetic invariants In this section we investigate the most important objects of phylogenetic algebraic geom- etry -- ideals of phylogenetic invariants. The main problem in this area is to give an effective description of the whole ideal of the variety associated to a given model on a tree. Our task is to find an efficient way to compute generators of these ideals. 6 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK We suggest a way of obtaining all phylogenetic invariants of a claw tree of a general group- based model -- more precisely we conjecture that our invariants generate the whole ideal of the variety. These, together with the results of [22] could provide an algorithm listing all generators of the ideal of phylogenetic invariants for any tree and for any general group-based model. 3.1. Inspirations. The inspirations for our method were the conjectures made by Sturmfels and Sullivant in [22]. They are still open but, as we will see, they strongly support our ideas. In particular, we will prove in Proposition 3.14 that our algorithm works for trees with more than 8 leaves for the 3-Kimura model if we assume that the weaker conjecture made in [22] holds. First we introduce some notation. Let Kn,1 be a claw tree with n leaves. Let G be a finite abelian group. Let φ(G, n) = d be the least natural number such that the ideal associated to Kn,1 for the general group-based model G is generated in degree d. The phylogenetic complexity of the group G is defined as φ(G) = supnφ(G, n). Based on numerical results Sturmfels and Sullivant suggested the following conjecture: Conjecture 3.1. For any abelian group G we have φ(G) ≤ G. This conjecture was separately stated for the 3-Kimura model, that is for G = Z2 × Z2. Still very little is known about the function φ apart from the case of the binary Jukes- Cantor model (see also [8]): Proposition 3.2 (Sturmfels, Sullivant). In case of the binary Jukes-Cantor model φ(Z2) = 2. (cid:3) There are also some computational results -- to the table in [22] presenting the computa- tions made by Sturmfels and Sullivant a few cases can be added. Computation 3.3. Using 4ti2 software [2] we obtained the following: • φ(6, Z3) = 3, • φ(4, Z5) = 4, • φ(3, Z8) = 8, • φ(3, Z2 × Z2 × Z2) = 8, • φ(3, Z4 × Z2) = 8. For the 3-Kimura model we do not even know whether the function φ is bounded. As we will see later, this conjecture is strongly related to the one stated in the next section. 3.2. A method for obtaining phylogenetic invariants. We propose a method that is inspired by the geometry of the varieties we consider. First we have to introduce some notation. Definition 3.4 (contraction, prolongation). We say that a tree T1 is obtained by a contrac- tion of an edge e of the tree T2 if the vertices of T1 correspond to vertices of T2 with the two vertices adjacent to e identified. Notice that there is a bijection between edges of T2 different from e and edges of T1. In such a situation we say that T2 is a prolongation of T1. PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 7 T1 = ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ✄✄✄✄✄✄✄✄ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ ❀❀❀❀❀❀❀❀ = T2 Remark 3.5. Note that these definitions are not the same as the definitions of flattenings introduced in [3] and further studied in [11]. Assume that we are in an abelian case. Using Algorithm 1 and Definition 2.3 one can see that vertices of the polytope correspond to sockets. On the other had it is well known that vertices of the polytope correspond to coordinates of the ambient space of the variety associated to the polytope. In this setting the variety X(T1) associated to the tree T1 is in a natural way a subvariety of X(T2), for any model. Notice that we can identify sockets of both varieties, as we may identify their leaves, so both varieties are contained in Ps, where s is the number of sockets. The natural inclusion corresponds to the projection of character lattices: we forget all the coordinates corresponding to the edge joining the vertices v1 and v2. Now the following conjecture is natural: Conjecture 3.6. The variety X(Kn,1) is equal to the (scheme theoretic) intersection of all the varieties X(Ti), where Ti is a prolongation of Kn,1 that has only two inner vertices, both of them of valency at least three. As X(Kn,1) is a subvariety of X(Ti) for any prolongation Ti one inclusion is obvious. Note also that the valency condition is made, because otherwise the conjecture would be obvious -- one of the varieties that we intersect would be equal to X(Kn,1) (contraction of a vertex of degree 2 does not change the corresponding variety). All Ti have strictly smaller maximal valency than Kn,1, so if the conjecture holds then we can inductively use Theorem 23 of Sturmfels and Sullivant [22] (see also Theorem 12 [23]) to obtain all phylogenetic invariants for a given model for any tree of any valency, knowing just the ideal of the K3,1. In such a case the ideal of X(Kn,1) is just the sum of ideals of trees with smaller valency. More precisely, if 3.6 holds then the degree in which the ideals of claw trees are generated cannot grow when the number of leaves gets bigger. This means that φ(G) = φ(G, 3) which can be computed in many cases. In particular, Conjecture 3.6 implies all cases of Conjecture 3.1 in which we can compute φ(G, 3) -- this includes the most interesting 3-Kimura model. Definition 3.7 (tripod). The tree K3,1 will be called a tripod. Remark 3.8. A reader familiar with phylogenetics may observe that varieties X(T1) and X(T2) are naturally contained in the same ambient space for any model, even if it does not give rise to toric varieties. Thus Conjecture 3.6 can help to compute the ideals of claw trees for a large class of phylogenetic models. Of course one may argue that Conjecture 3.6 above is too strong to be true. We prove it for the binary Jukes-Cantor model in Proposition 3.11. We will also consider two modifications of this conjecture to weaker conjectures that can still have a lot of applications. The first modification just states that Conjecture 3.6 holds for n large enough. o o / / o o / / o o / / o o / / 8 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK Proposition 3.9. Conjecture 3.6 holds for n large enough if and only if the function φ is bounded. Proof. One implication is obvious. Suppose that 3.6 holds for n > n0. We choose such d that the ideals associated to Kl,1 are generated in degree m for l ≤ n0. Using 3.6 and the results of [22] we can describe the ideal associated to Kn,1 as the sum of ideals generated in degree m. It follows that this ideal is also generated in degree m, so the function φ is bounded by m. For the other implication let us assume that φ(n) ≤ m. Let us consider any binomial B that is in the ideal of the claw tree and is of degree less or equal to m. We prove that B belongs to the ideal of some prolongation of a tree T , which is in fact more than the statement of Conjecture 3.6. Such a binomial can be described as a linear relation between (at most m) vertices of the polytope of this variety. Each vertex is given by an association of orbits of characters to edges such that there exist representatives of orbits that sum up to a trivial character. Let us fix such representatives, so that each vertex is given by n characters summing up to a trivial character. Now the binomial B can be presented as a pair of matrices A1 and A2 with characters as entries. Each column of the matrices is a vertex of the polytope. The matrices have at most m columns and exactly n rows. Let us consider the matrix A = A1 − A2, that is entries of the matrix A are characters that are differences of entries of A1 and A2. We can subdivide the first column of A into groups of at most H elements summing up to a trivial character. Then inductively we can subdivide the rows into groups of at most Hi elements summing up to a trivial character in each column up to the i-th one. For n > Hm + 1 we can find a set S of rows of A such that the characters sum up to a trivial character in each column restricted to S, such that both the cardinality of S and of its complement are greater then 1. Note that the sums of the entries lying in a chosen column and in the rows in S are the same in A1 and A2. Therefore, adding to both matrices an extra row whose entries are equal to the sum of the entries in the subset S gives a representation of a binomial B on a prolongation of T . (cid:3) In particular, this means that if Conjecture 3.1 of Sturmfels and Sullivant holds for the 3-Kimura model, then Conjecture 3.6 also holds for this model for n > 257. Later we will significantly improve this estimation. For the second modification of Conjecture 3.6 let us recall a few facts on toric varieties. Let T1 and T2 be two tori with lattices of characters given respectively by M1 and M2. Assume that both of them are contained in a third torus T with the character lattice M. The inclusions give natural isomorphisms M1 ≃ M/K1 and M2 ≃ M/K2, where K1 and K2 are torsion free lattices corresponding to characters that are trivial when restricted respectively to T1 and T2. The ideal of each torus (inside the big torus) is generated by binomials corresponding to such trivial characters. The points of T are given by semigroup morphisms M → C∗. The points of Ti are those morphisms that associate 1 to each character from Ki. We see that the points of the intersection T1 ∩ T2 are those morphisms M → C∗ that associate 1 to each character from the lattice K1 + K2. Of course the (possibly reducible) intersection Y is generated by the ideal corresponding to K1 + K2. This lattice may be not saturated, but Y contains a distinguished torus T ′, that is one of its connected components. If K ′ is the saturation of the lattice K1 + K2 then the characters of T ′ are given by the lattice PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 9 M/K ′. Let Xi be the toric variety that is the closure of Ti, and X ′ be the closure of T ′. We call the toric variety X ′ the toric intersection of X1 and X2. In the setting of 3.6 we conjecture the following: Conjecture 3.10. The toric variety X(T ) is the toric intersection of all the toric varieties X(Ti). This conjecture differs from the previous one by the fact that we allow the intersection to be reducible, with one distinguished irreducible component equal to X(T ). We state this conjecture, because it can be checked using only the tori. As the biologically meaningful points are contained in the torus (see [7]), this conjecture is of much importance for applica- tions. Moreover, it is quite easy to check it for trees with small enough number of leaves using computer programs. To explain it properly, let us consider the following general setting. Assume that the tori Ti are associated to polytopes Pi and that T is just the torus of the projective space Pn ⊇ Ti. Let Ai be a matrix whose columns represent vertices of the polytope Pi. The characters trivial on Ti or respectively binomials generating the ideal of Ti are exactly represented by integer vectors in the kernel of Ai. The characters trivial on the intersection are given by integer vectors in ker A1 + ker A2. Note that the ideal of the toric intersection T ′ of the tori Ti in T is generated by binomials corresponding to characters trivial on T ′, that is by the saturation of ker A1 + ker A2. These binomials define a toric variety in Pn. This variety is contained in the intersection (in fact it is a toric component) of the toric varieties that are the closures of Ti. The equality may not hold however, as the intersection might be reducible. In Conjecture 3.10 we have to compare two tori, one contained in the other. To do this, it is enough to compare their dimension, that is the rank of the character lattice. Let us note that the dimension of the intersection T1 ∩ T2 is given by n minus the dimension (as a vector space) of ker A1 + ker A2, as it is equal to the rank of the lattice Zn ∩ (ker A1 + ker A2). To compute this dimension it is enough to compute the ranks of matrices A1, A2 and B, where B is a matrix obtained by putting A1 under A2 (that is, ker B = ker A1 ∩ ker A2). This can be done very easily using GAP ([1]). The results obtained for small trees will be used in the following section. 3.3. Main Results. To support Conjecture 3.6 let us consider the case of binary Jukes- Cantor model. This model is well understood -- see for example [6], [8], [22]. In particular the quadratic Grobner basis was explicitly constructed for any tree in [8, Proposition 3]. Now we can prove the following: Proposition 3.11. Conjecture 3.6 holds for the binary Jukes-Cantor model. Proof. We use the same notation as in the proof of Proposition 3.9. From 3.2 we know that φ(Z2) = 2. Let us consider any binomial of degree 2 for a claw tree with n leaves. This is given by a pair of matrices A1, A2 with 2 columns each. Let A = A1 − A2, where the difference uses the group law. We construct a subset S of the set of rows which gives a prolongation of the tree. By permuting columns of A2 we may assume that the entries in the first row of the matrix A are trivial. Let A′ be the matrix obtained by deleting the first row of A. If we have a row 00 in A′ then we are done, so assume there are only 01, 10 and 11. Notice that 01 and 10 cannot occur at all, as A1 and A2 would not have the same rows up to permutation. For n > 3 we can take twice 11 as a strict subset of the set of rows, summing up to zero in each column. (cid:3) 10 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK From the proof above it follows that in fact to obtain the variety of the claw tree for the binary Jukes-Cantor model it is enough to intersect three varieties corresponding just to three subdivisions. This subdivisions correspond to S containing exactly first and second row or first and third, or second and third row. The following fact is an easy consequence of the results of [8] and [22]. However, due to lack of references we include it. For the trivalent trees in was proved in [6, Appendix]. Proposition 3.12. The toric variety associated to any tree and the binary Jukes-Cantor model is projectively normal for any tree. Proof. By [18, Lemma 5.1] it is enough to consider claw trees. Now one can apply [19, Proposition 1.6], where the assumptions are satisfied due to [8, Proposition 3]. (cid:3) In general we conjecture the following. Conjecture 3.13. Consider a finite abelian group G. Suppose that the variety associated to the tripod for the general group based model is projectively normal. Than the variety associated to any tree is projectively normal. Now we prove the following conditional result for the 3-Kimura model: Proposition 3.14. If Conjecture 3.1 of Sturmfels and Sullivant holds then Conjecture 3.6 holds for n > 8. Proof. We use the same notation as in the previous proof, but instead of considering the matrix A (corresponding to a chosen binomial) with k columns and entries from Z2 × Z2 we assume that it has 2k columns and entries from Z2. Let us note that the number of 1 in each row both in even and odd columns has to be even. This follows from the fact that rows of A are differences of rows that were equal up to permutation. This means that both projections from Z2 × Z2 → Z2 gave rows that were equal up to permutation. The difference of such vectors always has an even number of 1. Once again we may assume that the entries in the first row of A are trivial characters, that is they are equal to zero. Let A′ be the matrix obtained by deleting the first row of A. For each subset of rows of A′ we may consider a vector of length equal to the number of columns of A′, whose entries are given by sums of characters from the subset. Note that this vector always has an even number of 1 both in even and odd columns. Because we assume Conjecture 3.1, the matrix A′ has at most 8 columns. By Dirichlet's principle, if n > 8 then we can find two subsets of rows of A′ that are not complements of each other, such that their sum vector is the same. If we take a symmetric difference of these subsets, we obtain a strict, nonempty set S of rows of A′, summing up in each column to a trivial character. We add the first row of A to S or its complement, so that both sets have more than one element. Thus we obtain a subdivision of the set of rows of A such that the given binomial is in the ideal of the tree corresponding to this division. (cid:3) For n ≤ 8 we checked, using the computer programs Polymake, 4ti2, Macaulay2 and GAP, that the toric intersection of the tori of subdivisions gives the torus of the claw tree. We used the linear algebra described in the previous section. This proves that if Conjecture 3.1 holds for 3-Kimura model, then Conjecture 3.10 holds. Moreover, in all the checked cases it was enough to consider just two subdivisions. To summarize, we know that for 3-Kimura model Conjecture 3.6 implies both Conjectures 3.10 and 3.1 and moreover Conjecture 3.1 implies 3.10 and for n > 8 also Conjecture 3.6. PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 11 3.4. Example of an application. Let us present the method of finding phylogenetic in- variants on an example. For simplicity consider the Jukes-Cantor model. Suppose that one is interested in phylogenetic invariants for K4,1. This example is well-known and phylogenetic invariants can be found using many different methods. We have chosen this example as the number of phylogenetic invariants is small enough to be included in the paper. Suppose that we already know the phylogenetic invariants for trivalent trees. For other group-based models, if this step is not clear, we refer to [22]. Consider a prolongation of K4,1: 1 2 ❃❃❃❃❃❃❃❃ ❃❃❃❃❃❃❃ 3 4 There are 8 variables given respectively by: q0000, q0011, q0101, q0110, q1001, q1010, q1100, q1111. For the prolongation the ideal is generated by two relations: q0000q1111 − q1100q0011, q1010q0101 − q1001q0110. We may consider a different prolongation obtained by interchanging the leaves numbered 2 and 3. We obtain the following relations: q0000q1111 − q1010q0101, q1010q0101 − q1001q0110. These four (if fact one is redundant and it is enough to consider three of them) generate the ideal for K4,1. 4. Computational results 4.1. Example of a non-normal general group-based model. Knowing that the pro- jective variety associated to a general group-based model is toric, it is natural to ask whether it is normal. It is a very important property, as a lot of theorems in toric geometry work just for normal varieties. We check it investigating the polytope P associated to a model: it is normal if for any natural number n any point in the polytope nP is a sum of n points of P . Computations described in [18] have shown that for trivalent trees for the groups Z2, Z3, Z4 and Z2 × Z2 the associated varieties are projectively normal. However for the 2-Kimura model the associated variety is not normal. We are interested in the question whether all models for abelian (or at least cyclic) groups are normal. Now, using our implementation of Algorithm 1 and Normaliz (see [4]) we are able to check normality for a few more models. It is well known that toric fiber products of polytopes associated to two trees T1 and T2 gives a polytope associated to the tree obtained by identifying one leaf of T1 and one of T2 (see [22], [23]). Note that any trivalent tree can be obtained from a tripod (a tree with one inner vertex and three leaves) by a series of such gluing procedures. Hence, because of Lemma 5.1 from [18], if we check normality for the chosen group and the tripod, we know whether all algebraic varieties for this group and any trivalent tree are normal. Using our program we can obtain the set of vertices of the polytope related to the investigated group 12 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK and the tripod. Finally, we apply Normaliz [4] to check the normality of this polytope (in the lattice generated by its vertices). Thus we obtain Computation 4.1. The polytope associated with the tripod and one of the groups G = Z6, Z8, Z2 × Z2 × Z2, Z4 × Z2 is not normal. Hence the algebraic variety representing this model is not normal. In particular, the class of general group-based models contains non-normal models. We believe it can be difficult to characterize the class of groups for which the associated varieties are normal, or even to determine a big (infinite) class of normal, toric general group-based models. On the other hand one has the following result: Proposition 4.2. Let T be a phylogenetic tree and let G1 be a subgroup of an abelian group G2. If the variety corresponding to the tree T and group G1 is not normal then the variety corresponding to the tree T and group G2 is also not normal. Proof. Let Mi be a lattice whose basis is indexed by pairs of an edge of a tree and an element of the group Gi. The inclusion G1 ⊆ G2 gives us a natural injective morphism f : M1 → M2. Let Pi ⊂ Mi be the polytope associated to the model for the tree T and group Gi. Let Mi ⊂ Mi be a sublattice spanned by vertices of the polytope Pi. As P1 is not normal in the lattice spanned by its vertices, there exists a point x ∈ nP1 ∩ M1, that is not a sum of n vertices of the polytope P1. Let us consider y = f (x). The vertices of P1 are mapped to vertices of P2. We see that y ∈ nP2 ∩ M2. If P2 was normal in M2 we would be able to write y =Pn i=1 qi with qi ∈ P2. Let us notice that each point in the image f (M1) has zero on each entry of the coordinates indexed by any edge and any element of the group g ∈ G2 \ G1. In particular y has zero on these entries. As all entries of all vertices of P2 are nonnegative, this proves that all entries indexed by any edge and any element of the group g ∈ G2 \ G1 are zero for qi. However, we see that vertices of P2 that have all non-zero entries on coordinates indexed by pairs of an edge and an element g ∈ G1 are in the image of P1. Hence qi = f (pi) for pi ∈ P1. We see (cid:3) that x =P pi, which is impossible. In particular we see that all abelian groups G such that G is divisible by 6 or 8 give rise to non-normal models. Let P be the polytope associated to the tripod and the group Z6. We have already seen that P is not normal, hence the associated affine variety is not normal. One would be also interested if the associated projective variety is normal or, equivalently, if the polytope is very ample. By direct computation for (any) cone associated to a vertex of a polytope P we obtain the following result. Computation 4.3. The polytope P associated to the tripod and the group Z6 is not very ample. Hence the associated projective toric variety is not normal. 4.2. Hilbert-Ehrhart polynomials. The binary Jukes-Cantor model (for trivalent trees) has an interesting property, stated and proved in [6]: an elementary mutation of a tree gives a deformation of the associated varieties (see Construction 3.23). This implies that binary Jukes-Cantor models of trivalent trees with the same number of leaves are deformation equivalent (Theorem 3.26 in [6]). As it was not obvious what to expect for other models, we computed Hilbert-Ehrhart polynomials, which are invariants of deformation, in some simple cases. PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 13 4.2.1. Numerical results. We checked models for two different trees with six leaves (this is the least number of leaves for which there are non-isomorphic trees, exactly two), the snowflake and the 3-caterpillar. The most interesting ones were the cases of the biologically meaningful 2-Kimura and 3-Kimura models. The value of the Ehrhart polynomial of a polytope P for a natural number n is the number of lattice points in nP . Thus one way to determine the Hilbert-Ehrhart polynomial of a toric variety is to compute numbers of lattice points in some multiples of the associated polytope. Even if it is not possible to get enough data to determine the polynomials (eg. because the numbers are too big), sometimes we can say that polynomials for two models are not equal, because their values for some n are different. Before we completed our computations, Kubjas computed numbers of lattice points in the third dilations of the polytopes for 3-Kimura model on the snowflake and the 3-caterpillar with 6 leaves and got 69248000 and 69324800 points respectively (see [17]). Thus she proved that varieties associated with these models are not deformation equivalent. Our computations confirm her results as for the 3-Kimura model and also give the following Computation 4.4. The varieties associated with 2-Kimura models for the snowflake and the 3-caterpillar trees have different Ehrhart polynomials. In the second dilations of the polytopes there are 56992 lattice points for the snowflake and 57024 for the 3-caterpillar. Also the pairs of varieties associated with general group-based models for the snowflake and the 3-caterpillar trees and (1) G = Z3, (2) G = Z4, (3) G = Z5, (4) G = Z7 have different Hilbert-Ehrhart polynomials and therefore are not deformation equivalent. (For these groups the associated varieties are normal, which can be checked using Polymake.) The precise results of the computations are presented in the Appendix. In the cases of (1) G = Z8, (2) G = Z2 × Z2 × Z2, (3) G = Z9 the varieties have different Hilbert functions. As we know that the models for Z8 and Z2 × Z2 × Z2 are not normal and for Z9 we could not check the normality, these results do not imply that there is no deformation equivalence in these cases. 4.2.2. Technical details. The first attempt to compute numbers of lattice points in dilations of a polytope was the direct method: constructing the list of lattice points in nP by adding vertices of P to lattice points in (n − 1)P and reducing repeated entries. This algorithm is not very efficient, but (after adding a few technical upgrades to the implementation) we were able to confirm Kubjas' results [17]. However, this method does not work for non-normal polytopes. As we planned to investigate 2-Kimura model, we had to implement another algorithm. The second idea is to compute inductively the relative Hilbert polynomials, i.e. number of points in the n-th dilation of the polytope intersected with the fiber of the projection onto the group of coordinates that correspond to a given leaf. Our approach is quite similar to the methods used in [17] and [23]. 14 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK First we compute two functions for the tripod. Let P ⊂ Z3m ∼= Zm × Zm × Zm be the polytope associated to a tripod. Let pri : Z3m ∼= Zm × Zm × Zm → Zm be the projection onto the i-th group of coordinates. We distinguish one edge of the tripod corresponding to the third group of coordinates in the lattice. Let f be a function such that f (a) for a = (a1, . . . , am) ∈ Zm is the number of lattice points in (a1 + · · · + am)P that project to a by pr3. We compute f (a) for sufficiently many values of a to proceed with the algorithm. Example 4.5. The polytope P for the binary Jukes-Cantor model has the following vertices: v1 = (0, 1, 0, 1, 0, 1), v2 = (0, 1, 1, 0, 1, 0), v3 = (1, 0, 0, 1, 1, 0), v4 = (1, 0, 1, 0, 0, 1). These are the only integral points in P . In this case f (1, 0) = 2 because there are exactly two points, (1, 0, 0, 1, 1, 0) and (0, 1, 1, 0, 1, 0), that are in 1P = P and project to (1, 0) via the third projection. The function f will be our base for induction. Next, we need to compute the number of points in the fiber of a projection onto two distinguished leaves. Let g be a function such that g(a, b) for (a, b) = (a1, . . . , am, b1, . . . , bm) ∈ Zm × Zm is the number of lattice points in (a1 + · · · + am)P that project to a by pr3 and to b by pr2. We compute g(a, b) for sufficiently many pairs (a, b) to proceed with the algorithm. Let T be a tree with a corresponding polytope P and a distinguished leaf l. Let h be a function such that h(a) for a = (a1, . . . , am) ∈ Zm is equal to the number of points in the fiber of the projection corresponding to leaf l of (a1 + · · · + am)P onto a. We construct a new tree T ′ by attaching a tripod to a chosen leaf of T . We call T ′ a join of T and the tripod. The chosen leaf of T ′ will be one of the leaves of the attached tripod. As proved in [6], [22], [18], [23] (depending on the model), the polytope associated to a join of two trees is a fiber product of the polytopes associated to these trees. Thus we can calculate the function h′ for T ′ by a following rule: h′(a) = Pb g(a, b)h(b), where the sum is taken over all b ∈ Zm such that g(a, b) 6= 0. This allows us to compute inductively the relative Hilbert polynomial. The last tripod could be attached in the same way. Then one obtains the Hilbert function from relative Hilbert functions simply by summing up over all possible projections. However, it is better to do the last step in a different way. Suppose that as before we are given a tree T with a distinguished leaf l and a corresponding relative Hilbert function h. We compute the Hilbert function of the tree T ′ that is a join of the tree T and a tripod using the equality h′(n) =Pa f (a)h(a), where a = (a1, . . . , am) and P ai = n. The function f is the basis for induction introduced above. Thus, decomposing the snowflake and the 3-caterpillar trees to joins of tripods, we can in- ductively compute (a few small values of) the corresponding Hilbert functions. This method works also for non-normal models, if only the Hilbert function for the tripod can be computed. In particular, for 2-Kimura model the computations turned out to be possible, because its polytope for the tripod is quite well understood (see [18], 5.4), at least to describe fully its second dilation. This way we obtained the results of 4.4. PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 15 Appendix Here we present the precise results of the computations of Hilbert-Ehrhart polynomials for a few models, stated in 4.4. For each of the first groups we considered the numbers of lattice points in consecutive dilations are given. For the groups Z8, Z2 × Z2 × Z2 and Z9 we computed only the Hilbert function and, as the first two are not normal and for the last one we could not check the normality, we do not know if it is equal to the Hilbert-Ehrhart polynomial. Models for G = Z3. dilation snowflake 3-caterpillar 243 243 21627 21627 904069 903187 21496023 21451311 331976637 330935625 3662146270 3647265274 30920349834 30770591364 210269891871 209116329075 1196661601837 1189466778457 5831112858273 5868930577941 25205348411361 25377886917819 3-caterpillar 1024 1024 396928 396928 69324800 69248000 5990170739 5977866515 291864710144 291069470720 8967198289920 8995715702784 1024 396928 69248000 6122557220 310273545216 10009786400352 10062179606880 3-caterpillar 1024 396928 69324800 6138552524 311525688320 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5 6 1 2 3 4 5 6 Models for G = Z2 × Z2 (3-Kimura). dilation snowflake Models for G = Z4. dilation snowflake Models for G = Z5. dilation snowflake 3-caterpillar 3125 3834375 2230596875 3125 3834375 2229584375 640338121875 642089603125 1 2 3 4 16 MARIA DONTEN-BURY AND MATEUSZ MICHA LEK Models for G = Z7. In this case the first three dilations of the polytopes have the same number of points. The numbers of points in fourth dilations were too big to obtain precise results. Hence we computed only the numbers of points mod 64, which is sufficient to prove that the Hilbert-Ehrhart polynomials are different. Models for G = Z8. dilation snowflake Models for G = Z2 × Z2 × Z2. dilation snowflake 1 2 3 4 1 2 3 1 2 3 dilation snowflake 3-caterpillar 16807 117195211 16807 117195211 423913952448 423913952448 ≡ 54 mod 64 ≡ 14 mod 64 32768 454397952 3375180251136 3375013036032 3-caterpillar 32768 454397952 32768 454397952 3375180251136 3375013036032 3-caterpillar 32768 454397952 Models for G = Z9. dilation snowflake 59049 1499667453 20938605820263 20937202945056 3-caterpillar 59049 1499667453 1 2 3 PHYLOGENETIC INVARIANTS FOR GROUP-BASED MODELS 17 References [1] GAP -- Groups, Algorithms, and Programming, Version 4.4.10. The GAP Group, (http://www.gap- system.org), 2007. [2] 4ti2 team. 4ti2 -- a software package for algebraic, geometric and combinatorial problems on linear spaces. www.4ti2.de. [3] E. S. Allman and J. A. Rhodes. Phylogenetic ideals and varieties for the general Markov model. Advances in Applied Mathematics, 40(2):127 -- 148, 2008. [4] W. Bruns, B. Ichim, and C. Soger. Normaliz. http://www.mathematik.uni-osnabrueck.de/normaliz. [5] W. Buczy´nska, M. Donten, and J. A. Wi´sniewski. Isotropic models of evolution with symmetries. Con- temporary Mathematics, 496:111 -- 132, 2009. [6] W. Buczy´nska and J. A. Wi´sniewski. On geometry of binary symmetric models of phylogenetic trees. J. Eur. Math. Soc., 9(3):609 -- 635, 2007. [7] M. Casanellas and J. Fernandez-Sanchez. Geometry of the Kimura 3-parameter model. Advances in Applied Mathematics, 41(3):265 -- 292, 2008. [8] J. Chifman and S. Petrovi´c. Toric ideals of phylogenetic invariants for the general group-based model on claw trees K1,n. Proceedings of the 2nd international conference on Algebraic biology, pages 307 -- 321, 2007. [9] D. Cox, J. Little, and H. Schenck. Toric varieties, volume 124 of Graduate Studies in Mathematics. American Mathematical Society, 2011. [10] J. Draisma and R. H. Eggermont. Finiteness results for abelian tree models. arXiv:1207.1282v1 [math.AG], 2012. [11] J. Draisma and J. Kuttler. On the ideals of equivariant tree models. Mathematische Annalen, 344(3):619 -- 644, 2009. [12] N. Eriksson, K. Ranestad, B. Sturmfels, and S. Sullivant. Phylogenetic algebraic geometry. Projective Varieties with Unexpected Properties; Siena, Italy, pages 237 -- 256, 2004. [13] S. N. Evans and T. P. Speed. Invariants of some probability models used in phylogenetic inference. Ann. Statist., 21(1):355 -- 377, 1993. [14] W. Fulton. Introduction to Toric Varieties, volume 131 of Annals of Mathematics Studies. Princeton University Press, 1993. [15] M. D. Hendy. The relationship between simple evolutionary tree models and observable sequence data. Systematic Zoology, 38:310 -- 321, 1989. [16] N. Ilten. Deformations of Rational Varieties with Codimension-One Torus Action. Doctoral Thesis, FU Berlin, 2010. [17] K. Kubjas. Hilbert polynomial of the Kimura 3-parameter model. arXiv:1007.3164v1 [math.AC], 2010. [18] M. Micha lek. Geometry of phylogenetic group-based models. Journal of Algebra, 339:339 -- 356, 2011. [19] H. Ohsugi and T. Hibi. Toric ideals generated by quadratic binomials. Journal of Algebra, 218:509 -- 527, 1999. [20] L. Pachter and B. Sturmfels. Algebraic Statistics for Computational Biology. Cambridge University Press, 2005. [21] B. Sturmfels. Groebner bases and convex polytopes, volume 8 of University Lecture Series. American Mathematical Society, 1996. [22] B. Sturmfels and S. Sullivant. Toric ideals of phylogenetic invariants. J. Comput. Biology, 12:204 -- 228, 2005. [23] S. Sullivant. Toric fiber products. J. Algebra, 316:560 -- 577, 2007. [24] L. A. Sz´ekely, M. A. Steel, and P. L. Erdos. Fourier calculus on evolutionary trees. Appl. Math., 14(2):200 -- 210, 1993.
1808.00311	2	1808	2018-11-10T21:16:00	Four dimensional Fano quiver flag zero loci (with an appendix by T. Coates, E. Kalashnikov, and A. Kasprzyk)	[ "math.AG" ]	Quiver flag zero loci are subvarieties of quiver flag varieties cut out by sections of homogeneous vector bundles. We prove the Abelian/non-Abelian Correspondence in this context: this allows us to compute genus zero Gromov--Witten invariants of quiver flag zero loci. We determine the ample cone of a quiver flag variety, disproving a conjecture of Craw. In the Appendices, which are joint work with Tom Coates and Alexander Kasprzyk, we use these results to find four-dimensional Fano manifolds that occur as quiver flag zero loci in ambient spaces of dimension up to 8, and compute their quantum periods. In this way we find at least 141 new four-dimensional Fano manifolds.	math.AG	math	FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI ELANA KALASHNIKOV Abstract. Quiver flag zero loci are subvarieties of quiver flag varieties cut out by sections of homogeneous vector bundles. We prove the Abelian/non-Abelian Correspondence in this context: this allows us to compute genus zero Gromov -- Witten invariants of quiver flag zero loci. We determine the ample cone of a quiver flag variety, disproving a conjecture of Craw. In the Appendices, which are joint work with Tom Coates and Alexander Kasprzyk, we use these results to find four-dimensional Fano manifolds that occur as quiver flag zero loci in ambient spaces of dimension up to 8, and compute their quantum periods. In this way we find at least 141 new four-dimensional Fano manifolds. 1. Introduction In this paper, we consider quiver flag zero loci: smooth projective algebraic varieties which occur as zero loci of sections of homogeneous vector bundles on quiver flag varieties. We prove the Abelian/non-Abelian Correspondence of Ciocan-Fontanine -- Kim -- Sabbah in this context, which allows us to compute genus zero Gromov -- Witten invariants of quiver flag zero loci1. We also determine the ample cone of a quiver flag variety, disproving a conjecture of Craw. Our primary motivation for these results is as follows. There has been much recent interest in the possibility of classifying Fano manifolds using Mirror Symmetry. It is conjectured that, under Mirror Symmetry, n-dimensional Fano manifolds should correspond to certain very special Laurent polynomials in n variables [8]. This conjecture has been established in dimensions up to three [9], where the classification of Fano manifolds is known [20 -- 22, 29 -- 33]. Little is known about the classification of four-dimensional Fano manifolds, but there is strong evidence that the conjecture holds for four-dimensional toric complete intersections [12]. Not every Fano manifold is a toric complete intersection, but the constructions in [9] show that every Fano manifold of dimension at most three is either a toric complete intersection or a quiver flag zero locus. One might hope, therefore, that most four-dimensional Fano manifolds are either toric complete intersections or quiver flag zero loci. In the Appendices, which are joint work with Tom Coates and Alexander Kasprzyk, we use the structure theory developed here to find four-dimensional Fano manifolds that occur as quiver flag zero loci in ambient spaces of dimension up to 8, and compute their quantum periods. 141 of these quantum periods were previously unknown. Thus we find at least 141 new four-dimensional Fano manifolds. This computation is described in Appendix A. The quantum periods, and quiver flag zero loci that give rise to them, are recorded in Appendix B. Figure 1 overleaf shows the distribution of degree and Euler number for the four-dimensional quiver flag zero loci that we found, and for four-dimensional Fano toric complete intersections. Key words and phrases. Fano manifolds, mirror symmetry, quantum differential equations, Picard -- Fuchs equa- tions, quiver flag varieties. 1Another proof of this, using different methods, has recently been given by Rachel Webb [37]. 1 2 E L A N A K A L A S H N K O V I Figure 1. Degrees and Euler numbers for four-dimensional Fano quiver flag zero loci and toric complete intersections; cf. [12, Figure 5]. Quiver flag zero loci that are not toric complete intersections are highlighted in red. Quiver Flag Zero Loci02004006008000200400600800DegreeEuler NumberToric Complete Intersections02004006008000200400600800DegreeEuler NumberFOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 3 2. Quiver flag varieties Quiver flag varieties, which were introduced by Craw [13], are generalizations of Grassmannians and type A flag varieties. Like flag varieties, they are GIT quotients and fine moduli spaces. We begin by recalling Craw's construction. A quiver flag variety M (Q, r) is determined by a quiver Q and a dimension vector r. The quiver Q is assumed to be finite and acyclic, with a unique source. Let Q0 = {0, 1, . . . , ρ} denote the set of vertices of Q and let Q1 denote the set of arrows. Without loss of generality, after reordering the vertices if necessary, we may assume that 0 ∈ Q0 is the unique source and that the number nij of arrows from vertex i to vertex j is zero unless i < j. Write s, t : Q1 → Q0 for the source and target maps, so that an arrow a ∈ Q1 goes from s(a) to t(a). The dimension vector r = (r0, . . . , rρ) lies in Nρ+1, and we insist that r0 = 1. M (Q, r) is defined to be the moduli space of θ-stable representations of the quiver Q with dimension vector r. 2.1. Quiver flag varieties as GIT quotients. Consider the vector space Rep(Q, r) = Hom(Crs(a) , Crt(a)) (cid:77) a∈Q1 and the action of GL(r) :=(cid:81)ρ 1, we may identify G with (cid:81)ρ ρ(cid:89) θ(g) = det(gi), Rep(Q, r), set si =(cid:80) i=0 GL(ri) on Rep(Q, r) by change of basis. The diagonal copy of GL(1) in GL(r) acts trivially, but the quotient G := GL(r)/ GL(1) acts effectively; since r0 = i=1 GL(ri). The quiver flag variety M (Q, r) is the GIT quotient Rep(Q, r)//θ G, where the stability condition θ is the character of G given by g = (g1, . . . , gρ) ∈ ρ(cid:89) GL(ri). For the stability condition θ, all semistable points are stable. To identify the θ-stable points in i=1 i=1 a∈Q1,t(a)=i rs(a) and write ρ(cid:77) Rep(Q, r) = Hom(Csi, Cri). Then w = (wi)ρ i=1 is θ-stable if and only if wi is surjective for all i. Example 2.1. Consider the quiver Q given by i=1 so that ρ = 1, n01 = n, and the dimension vector r = (1, r). Then Rep(Q, r) = Hom(Cn, Cr), and the θ-stable points are surjections Cn → Cr. The group G acts by change of basis, and therefore M (Q, r) = Gr(n, r), the Grassmannian of r-dimensional quotients of Cn. More generally, the quiver gives the flag of quotients Fl(n, a, b, . . . , c). In this case G ∼=(cid:81)ρ toric quiver flag varieties. Quiver flag varieties are non-Abelian GIT quotients unless the dimension vector r = (1, 1, . . . , 1). i=1 GL1(C) is Abelian, and M (Q; r) is a toric variety. We call such M (Q, r) 2.2. Quiver flag varieties as moduli spaces. To give a morphism to M (Q, r) from a scheme B is the same as to give: • globally generated vector bundles Wi → B, i ∈ Q0, of rank ri such that W0 = OB; and • morphisms Ws(a) → Wt(a), a ∈ Q1 satisfying the θ-stability condition up to isomorphism. Thus M (Q, r) carries universal bundles Wi, i ∈ Q0. It is also a Mori Dream Space. The GIT description gives an isomorphism between the Picard group of M (Q, r) and the character group χ(G) ∼= Zρ of G. When tensored with Q, the fact that this is a Mori Dream space implies that this isomorphism induces an isomorphism of wall and chamber structures given by the Mori structure (on the effective cone) and the GIT structure (on χ(G) ⊗ Q); in particular, the 1rn1anb...cj . Then there are maps M (Q, r) = Yρ → Yρ−1 → ··· → Y1 → Y0 = Spec C, ∼= Gr(Fi, ri), where Fi is the locally free sheaf (cid:77) a∈Q1,t(a)=i Fi = W (i−1) s(a) induced by isomorphisms Yi sion (cid:80)ρ 4 ELANA KALASHNIKOV GIT chamber containing θ is the ample cone of M (Q, r). The Picard group is generated by the determinant line bundles det(Wi), i ∈ Q0. 2.3. Quiver flag varieties as towers of Grassmannian bundles. Generalizing Example 2.1, all quiver flag varieties are towers of Grassmannian bundles [13, Theorem 3.3]. For 0 ≤ i ≤ ρ, let Q(i) be the subquiver of Q obtained by removing the vertices j ∈ Q0, j > i, and all arrows attached to them. Let r(i) = (1, r1, . . . , ri), and write Yi = M (Q(i), r(i)). Denote the universal bundle Wj → Yi by W (i) of rank si on Yi−1. This makes clear that M (Q, r) is a smooth projective variety of dimen- i=1 ri(si − ri), and that Wi is the pullback to Yρ of the tautological quotient bundle over Gr(Fi, ri). Thus Wi is globally generated, and det(Wi) is nef. Furthermore the anticanonical line bundle of M (Q, r) is (cid:79) a∈Q1 det(Wt(a))rs(a) ⊗ det(Ws(a))−rt(a) . i :=(cid:80) In particular, M (Q, r) is Fano if si > s(cid:48) a∈Q1,s(a)=i rt(a). This condition is not if and only if. 2.4. Quiver flag zero loci. We have expressed the quiver flag variety M (Q, r) as the quotient by G of the semistable locus Rep(Q, r)ss ⊂ Rep(Q, r). A representation E of G, therefore, defines a vector bundle EG → M (Q, r) with fiber E; here EG = E ×G Rep(Q, r)ss. In the second half of this paper, we will study subvarieties of quiver flag varieties cut out by regular sections of such bundles. We refer to such subvarieties as quiver flag zero loci, and such bundles as homogeneous bundles. Quiver flag varieties are also natural ambient spaces via the multi-graded Plucker embedding ( [13], [14]). The representation theory of G =(cid:81)ρ i=1 GL(ri) is well-understood, and we can use this to write down the bundles EG explicitly. Irreducible polynomial representations of GL(r) are indexed by partitions (or Young diagrams) of length at most r. The irreducible representation corresponding to a partition α is the Schur power SαCr of the standard representation of GL(r) [17, chapter 8]. For example, if α is the partition (k) then SαCr = Symk Cr, the kth symmetric power, and if α is the partition (1, 1, . . . , 1) of length k then SαCr =(cid:86)k Cr, the kth exterior power. Irreducible representation indexed by (α1, . . . , αρ) is (cid:78)ρ M (Q, r) is(cid:78)ρ polynomial representations of G are therefore indexed by tuples (α1, . . . , αρ) of partitions, where αi has length at most ri. The tautological bundles on a quiver flag variety are homogenous: if E = Cri is the standard representation of the ith factor of G, then Wi = EG. More generally, the i=1 SαiCri , and the corresponding vector bundle on i=1 SαiWi. Example 2.2. Consider the vector bundle Sym2 W1 on Gr(8, 3). By duality -- which sends a quotient C8 → V → 0 to a subspace 0 → V ∗ → (C8)∗ -- this is equivalent to considering the 1 on the Grassmannian of 3-dimensional subspaces of (C8)∗, where S1 is the vector bundle Sym2 S∗ tautological sub-bundle. A generic symmetric 2-form ω on (C8)∗ determines a regular section of 1 , which vanishes at a point V ∗ if and only if the restriction of ω to V ∗ is identically zero. Sym2 S∗ So the associated quiver flag zero locus is the orthogonal Grassmannian OGr(3, 8). 2.5. The Euler sequence. Quiver flag varieties, like both Grassmannians and toric varieties, have an Euler sequence. Proposition 2.3. Let X = M (Q, r) be a quiver flag variety, and for a ∈ Q1, denote Wa := s(a) ⊗ Wt(a). There is a short exact sequence W ∗ Wi ⊗ W ∗ 0 → ρ(cid:77) i → (cid:77) Wa → TX → 0. i=1 a∈Q1 FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 5 Proof. We proceed by induction on the Picard rank ρ of X. If ρ = 1 then this is the usual Euler sequence for the Grassmannian. Suppose that the proposition holds for quiver flag varieties of Picard rank ρ − 1, ρ > 1. Then the fibration π : Gr(π∗Fρ, rρ) → Yρ−1 from §22.3 above gives a short exact sequence where S is the kernel of the projection π∗Fρ → Wρ. Note that ρ ⊗ Wρ → S∗ ⊗ Wρ → 0 0 → Wρ ⊗ W ∗ ρ → π∗F∗ π∗F∗ ρ ⊗ Wρ = Wa and that TX = TYρ−1 ⊕ S∗ ⊗ Wρ. (cid:77) a∈Q1,t(a)=ρ As S∗ ⊗ Wρ is the relative tangent bundle to π, the proposition follows by induction. (cid:3) If X is a quiver flag zero locus cut out of the quiver flag variety M (Q, r) by a regular section of the homogeneous vector bundle E then there is an exact sequence 0 → TX → TM (Q,r)X → E → 0. Thus TX is the K-theoretic difference of homogeneous vector bundles. 3. Quiver flag varieties as subvarieties There are three well-known constructions of flag varieties: as GIT quotients, as homogenous spaces, and as subvarieties of products of Grassmannians. Craw's construction gives quiver flag varieties as GIT quotients. General quiver flag varieties are not homogenous spaces, so the sec- ond construction does not generalize. In this section we generalize the third construction of flag varieties, exhibiting quiver flag varieties as subvarieties of products of Grassmannians. It is this description that will allow us to prove the Abelian/non-Abelian correspondence for quiver flag varieties. Proposition 3.1. Let M (Q, r) be a quiver flag variety with ρ > 1. Let si = dim H 0(M (Q, r), Wi). Then M (Q, r) is cut out of Y =(cid:81)ρ i=1 Gr(si, ri) by a canonical section of E = a∈Q1,s(a)(cid:54)=0 s(a) ⊗ Qt(a) S∗ (cid:77) (cid:77) Let Fi =(cid:76) where Si and Qi are the tautological sub-bundle and quotient bundle on the ith factor of Y . Proof. Recall that si is the dimension of the space of paths from 0 to vertex i (Corollary 3.5, [13]). Thus t(a)=i Qs(a). Combining the canonical surjections Css(a) ⊗O → Qs(a) gives a surjection Csi ⊗ O → Fi that fits into the exact sequence Css(a) ⊕ Cn0i. a∈Q1,t(a)=i,s(a)(cid:54)=0 Csi = 0 → (cid:77) Ss(a) → C si ⊗ O → Fi → 0. Thus and it follows that E =(cid:76)ρ t(a)=i,s(a)(cid:54)=0 (Csi∗ ⊗ O)/F ∗ (cid:77) i , (Csi∗ ⊗ O)/F ∗ i ). ∼= i t(a)=i,s(a)(cid:54)=0 i=2 Hom(Q∗ S∗ s(a) i ⊂(cid:76) Consider the section s of E given by the compositions Q∗ i . The section s vanishes at quotients (V1, . . . , Vρ) if and only if V ∗ s(a); dually, the zero locus is where there is a surjection Fi → Qi for each i. Now it isn't hard to see that Z(s) is M (Q, r). M (Q, r) parametrizes surjections C si → Vi that factor as i → Csi∗ ⊗ O → (Csi∗ ⊗ O)/F ∗ t(a)=i V ∗ Since the Wi are globally generated, there is a unique map C si → (cid:77) t(a)=i Vs(a) → Vi. M (Q, r) → Y = Gr( si, ri) ρ(cid:89) i=1 6 ELANA KALASHNIKOV such that Qi → Y pulls back to Wi → M (Q, r). The discussion above shows that the image of this map lies in the zero locus of s. Similarly, the universal property of M (Q, r) gives rise to a unique map Z(s) → M (Q, r) such that Wi → M (Q, r) pulls back to Qi → Z(s). Because M (Q, r) is a fine moduli space, the composition of these maps M (Q, r) → Z(s) → M (Q, r) must be the identity. Similarly, the composition Z(s) → M (Q, r) → Z(s) is the identity, and we conclude that (cid:3) Z(s) and M (Q, r) are canonically isomorphic. vector bundle E(cid:48) on(cid:81)ρ a generic section of the vector bundle F := E(cid:48) ⊕(cid:76) Suppose that X is a quiver flag zero locus cut out of M (Q, r) by a regular section of a homoge- neous vector bundle E. Since M (Q, r) and the product of Grassmannians described above are both GIT quotients by the same group G, the representation of G that determines E also determines a i=1 Gr( si, ri). We see that X is deformation equivalent to the zero locus of s(a) ⊗ Qt(a). Although the product of Grassmannians is a quiver flag variety, this is not generally an additional model of X as a quiver s(a) ⊗ Qt(a) in F does not in general come from a representation flag zero locus, as the summand S∗ of G. a∈Q1,s(a)(cid:54)=0 S∗ Remark 3.2. Suppose α is a non-negative Schur partition. Then [35] shows that Sα(Q) is globally generated on Gr(n, r). The above construction allows us to generalise this to quiver flag varieties: Sα(Wi) is globally generated on M (Q, r). 4. Equivalences of quiver flag zero loci The representation of a given variety X as a quiver flag zero locus, if it exists, is far from unique. In this section we describe various methods of passing between different representations of the same quiver flag zero locus. This is important in practice, because our systematic search for four- dimensional quiver flag zero loci described in the Appendix finds a given variety in many different representations. Furthermore, geometric invariants of a quiver flag zero locus X can be much easier to compute in some representations than in others. The observations in this section allow us to compute invariants of four-dimensional Fano quiver flag zero loci using only a few representations, where the computation is relatively cheap, rather than doing the same computation many times and using representations where the computation is expensive. 4.1. Dualising. As we saw in the previous section, a quiver flag zero locus X given by (M (Q, r), E) can be thought of as a zero locus in a product of Grassmannians Y . Unlike general quiver flag varieties, Grassmannians come in canonically isomorphic dual pairs: The isomorphism interchanges the tautological quotient bundle Q with S∗, where S is the tauto- logical sub-bundle. One can then dualize some or none of the Grassmannian factors in Y , to get different models of X. Depending on the representations in E, after dualizing, E may still be a homogenous vector bundle, or the direct sum of a homogeneous vector bundle with bundles of the i ⊗ Wj. If this is the case, one can then undo the product representation process to obtain form S∗ another model (M (Q(cid:48), r(cid:48)), E(cid:48) Example 4.1. Consider X given by the quiver G) of X. and bundle ∧2W2; here and below the vertex numbering is indicated in blue. Then writing it as a product: with bundle ∧2W2 ⊕ S∗ 1 ⊗ W2 and dualizing the first factor, we get 1rn1n-rn103182210318228FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 7 with bundle W1 ⊗ W2 ⊕ ∧2W2, which is a quiver flag zero locus. 4.2. Removing arrows. Example 4.2. Recall that Gr(n, r) is the quiver flag zero locus given by (Gr(n + 1, r), W1). This is because the space of sections of W1 is Cn+1, where the image of the section corresponding to v ∈ Cn+1 at the point φ : Cn+1 → W in Gr(n + 1, r) is φ(v). This section vanishes precisely when v ∈ ker φ, so we can consider its zero locus to be Gr(Cn+1/(cid:104)v(cid:105), r) ∼= Gr(n, r). The restriction of W1 to this zero locus Gr(n, r) is W1, and the restriction of the tautological sub-bundle S is S⊕OGr(n,r). This example generalises. Let M (Q, r) be a quiver flag variety. A choice of arrow i → j in Q i ⊗ Wj, and the zero locus of this section is M (Q(cid:48), r), where determines a canonical section of W ∗ Q(cid:48) is the quiver obtained from Q by removing one arrow from i → j. Example 4.3. Similarly, Gr(n, r) is the zero locus of a section of S∗, the dual of the tautological sub-bundle, on Gr(n + 1, r + 1). The exact sequence 0 → W ∗ 1 → (Cn+1)∗ → S∗ → 0 shows that a global section of S∗ is given by a linear map ψ : Cn+1 → C. The image of the section corresponding to ψ at the point s ∈ S is ψ(s), where we evaluate ψ on s via the tautological inclusion S → Cn+1. Splitting Cn+1 = Cn⊕C and choosing ψ to be projection to the second factor shows that ψ vanishes precisely when S ⊂ Cn, that is, precisely along Gr(n, r). The restriction of S to this zero locus Gr(n, r) is S, and the restriction of W1 is W1 ⊕ OGr(n,r). 4.3. Grafting. Let Q be a quiver and S ⊂ {1, . . . , ρ}. We say that a vertex i is graftable for S if: • ri = 1 and 0 < i < ρ; • there is a path between vertex i and vertex j for all j ∈ S; • if we remove all of the arrows from i to vertices in S, we get a disconnected quiver. Example 4.4. In the quiver below, vertex 1 is not graftable for {2}. If we removed the arrow from vertex 0 to vertex 2, then vertex 1 would be graftable for {2}. Proposition 4.5. Let Q be a quiver and let i be a vertex of Q that is graftable for S. Let Q(cid:48) be the quiver obtained from Q by replacing each arrow i → j, where j ∈ S, by an arrow 0 → j. Then M (Q, r) = M (Q(cid:48), r) for any dimension vector r. Proof. Recall from §22.2 the moduli problem represented by M (Q, r) and note, since i is graftable for S, that Wi is a line bundle. Setting (cid:40) Wj (cid:55)→ i ⊗ Wj W ∗ Wj j ∈ S j (cid:54)∈ S transforms the moduli problem for M (Q, r) into that for M (Q(cid:48), r). Example 4.6. Consider the quiver flag zero locus X given by the quiver in (a) below, with bundle (cid:3) W1 ⊗ W3 ⊕ W ⊕2 1 ⊕ det W1. W1 ⊗ W3 ⊕ det W1 Writing X inside a product of Grassmannians gives W1 ⊗ W3 ⊕ W ⊕2 (b), with arrow bundle S∗ 2 ⊗ W1. Removing the two copies of W1 using Example 4.2 gives 1 ⊕ det W1 on the quiver in on the quiver in (c), with arrow bundle S∗ det S∗ 1 , and taking care to absorb the extra factors of W3 and S∗ 2 ⊗ W1. Applying Example 4.3 to remove det W1 = 2 which arise from W1 ⊗ W3 1 = S∗ 1051822810112228 ELANA KALASHNIKOV and the arrow bundle, we see that X is given by W1 ⊗ W3 on the quiver in (d), with arrow 2 ⊗ W1. Dualising at vertices 1 and 2 now gives the quiver in (e), with arrow bundle bundle S∗ 1 ⊗ W2 ⊕ S∗ 1 ⊗ W3. Finally, undoing the product representation exhibits X as the quiver flag S∗ variety for the quiver in (f ). (a) (d) (b) (e) (c) (f ) 5. The ample cone We now discuss how to compute the ample cone of a quiver flag variety. This is essential if one wants to search systematically for quiver flag zero loci that are Fano. In [13], Craw gives a conjecture that would in particular solve this problem, by relating a quiver flag variety M (Q, r) to a toric quiver flag variety. We give a counterexample to this conjecture, and determine the ample cone of M (Q, r) in terms of the combinatorics of the quiver: this is Theorem 5.12 below. Our method also involves a toric quiver flag variety: the Abelianization of M (Q, r). 5.1. The multi-graded Plucker embedding. Given a quiver flag variety M (Q, r), Craw defines a multi-graded analogue of the Plucker embedding: p : M (Q, r) (cid:44)→ M (Q(cid:48), 1) Here Q(cid:48) is the quiver with the same vertices as Q but with the number of arrows i → j given by dim H 0(cid:0)Hom(cid:0) det(Wi), det(Wj)(cid:1)(cid:1) /S with 1 = (1, . . . , 1). where S is spanned by maps which factor through maps to det(Wk) with i < k < j ( [13], Example 2.9 in [14]). This induces an isomorphism p∗ : P ic(X) ⊗ R → Pic(X) ⊗ R that sends det(W (cid:48) i ) (cid:55)→ det(Wi). In [13], it is conjectured that this induces a surjection of Cox rings Cox(M (Q(cid:48), 1)) → Cox(M (Q, r)). This would give information about the Mori wall and chamber structure of M (Q, r). In particular, by the proof of Theorem 2.8 of [26], a surjection of Cox rings together with an isomorphism of Picard groups (which we have here) implies an isomorphism of effective cones. We provide a counterexample to the conjecture. To do this, we exploit the fact that quiver flag varieties are Mori Dream Spaces, and so the Mori wall and chamber structure on NE1(M (Q, r) ⊂ Pic(M (Q, r) coincides with the GIT wall and chamber structure. This gives GIT characterizations for effective divisors, ample divisors, nef divisors, and the walls. Theorem 5.1. V = CN with stability condition ω ∈ χ(G) = Hom(G, C∗). Identifying Pic(X) ∼= χ(G), we have [16] Let X be a Mori Dream Space obtained as a GIT quotient of G acting on that: • v ∈ χ(G) is ample if V s(v) = V ss(v) = V s(ω). • v is on a wall if V ss(v) (cid:54)= V s(v). • v ∈ NE1(X) if V ss (cid:54)= ∅. When combined with King's characterisation [24] of the stable and semistable points for the GIT problem defining M (Q, r), this determines the ample cone of any given quiver flag variety. In Theorem 5.12 below we make this effective, characterising the ample cone in terms of the combinatorics of Q, but this is already enough to see a counterexample to the conjecture. 10628238511051862823810516628238104155272371011522723710115222232FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 9 Example 5.2. Consider the quiver Q and dimension vector r as in (a). The target M (Q(cid:48), 1) of the multi-graded Plucker embedding has the quiver Q(cid:48) shown in (b). (a) (b) representation R = (Ri)i∈Q0 is semi-stable with respect to χ = (χi)ρ i) ≥ 0. i=0 χi dimC(Ri) = 0; and In this case, M (Q(cid:48), 1) is a product of projective spaces and so the effective cone coincides with the nef cone, which is just the closure of the positive orthant. The ample cone of M (Q, r) is indeed the positive orthant, as we will see later. However, the effective cone is strictly larger. We will i=0 Gl(ri): a use King's characterisation of semi-stable points with respect to a character χ of (cid:81)ρ • (cid:80)ρ • for any subrepresentation R(cid:48) of R,(cid:80)ρ Consider the character χ = (−1, 3) of G, which we lift to a character of (cid:81)ρ (cid:20)0 i=0 Gl(ri) by taking χ = (−3,−1, 3). We will show that there exists a representation R = (R0, R1, R2) which is semi- stable with respect to χ. The maps in the representation are given by a triple (A, B, C) ∈ M at(3× 5) × M at(2 × 3) × M at(2 × 3). Suppose that i=0 χi dimC(R(cid:48) i=0 if and only if (cid:21) (cid:21) A has full rank, B = 0 1 0 0 , C = 0 0 0 0 1 , (cid:20)1 0 and that R(cid:48) is a subrepresentation with dimensions a, b, c. We want to show that −3a− b + 3c ≥ 0. If a = 1 then b = 3, as otherwise the image of A is not contained in R(cid:48) 1. Similarly, this implies that c = 2. So suppose that a = 0. The maps B and C have no common kernel, so b > 0 implies c > 0, and −b + 3c ≥ 0 as b ≤ 3. Therefore R is a semi-stable point for χ, and as quiver flag varieties are Mori Dream Spaces, χ is in the effective cone. 5.2. Abelianization. We consider now the toric quiver flag variety associated to a given quiver flag variety M (Q, r) which arises from the corresponding Abelian quotient. Let T ⊂ G be the diagonal maximal torus. Then the action of G on Rep(Q, r) induces an action of T on Rep(Q, r), and the inclusion i : χ(G) (cid:44)→ χ(T ) allows us to interpret the special character θ as a stability condition for the action of T on Rep(Q, r). The Abelian quotient is then Rep(Q, r)//θT . Let us see that Rep(Q, r)//θT is a toric quiver flag variety. Let λ = (λ1, . . . , λρ) denote an element of i=1(C∗)ri, where λj = (λj1, . . . , λjrj ). The action of λ on the (i, j) entry x of the rt(a)×rs(a) matrix which is the a ∈ Q1 component of an element in Rep(Q, r) is T =(cid:81)ρ x (cid:55)→ λ−1 s(a)ixλt(a)j. Hence this is the same as the group action on the quiver Qab with vertices 0 = {vij : 0 ≤ i ≤ ρ, 1 ≤ j ≤ ri} Qab and the number of arrows between vij and vik is the number of arrows in the original quiver between vertices i and k. Hence Rep(Q, r)//θT = M (Qab, 1). We call Qab the Abelianized quiver. Example 5.3. Let Q be the quiver Then Qab is 103152221011101245102132132410 ELANA KALASHNIKOV Martin [28] has studied the relationship between the cohomology of Abelian and non-Abelian quotients. We state his result specialized to quiver flag varieties, then extend this to a comparison of the ample cones. To simplify notation, denote MQ = M (Q, r), MQab = M (Qab, (1, . . . , 1)) and V = Rep(Q, r) = Rep(Qab, (1, . . . , 1)). For v ∈ χ(G), let V s v (T ) denote the T -stable points of V and V s v (G) denote the G-stable points, dropping the subscript if it is clear from context. It is easy to i=1 Sri. Let π : V s(G)/T → V s(G)/G be the projection. The Weyl group acts on the cohomology of M (Qab, 1), and also on the Picard group, by permuting the Wvi1, . . . , Wviri see that V s(G) ⊂ V s(T ). The Weyl group W of (G, T ) is(cid:81)ρ . It is well-known (see e.g. Atiyah -- Bott [3]) that π∗ : H∗(V s(G)/T )W ∼= H∗(MQ). Theorem 5.4. [28] There is a graded surjective ring homomorphism φ : H∗(MQab, C)W → H∗(V s(G)/T, C) π∗−→ H∗(MQ, C) where the first map is given by the restriction V s(T )/T → V s(G)/T . The kernel is the annihilator of e =(cid:81)ρ (cid:81) i=1 1≤j,k≤ri c1(W ∗ vij ⊗ Wvik ). Remark 5.5. This means that any class σ ∈ H∗(MQ) can be lifted (non-uniquely) to a class σ ∈ H∗(MQab). Moreover, e ∩ σ is uniquely determined by σ. Corollary 5.6. Let E be a representation of G and hence T . We can form the vector bundles EG = (V s(G) × E)/G → MQ, G = (V s(G) × E)/T → V s(G)/T, E(cid:48) ET = (V s(T ) × E)/T → MQab, where in all cases the group acts diagonally. Then φ(ci(ET )) = ci(EG). Proof. Let f be the inclusion V s(G)/T → V s(T )/T . Clearly f∗(ET ) = E(cid:48) restriction of ET . Considering the square G as E(cid:48) G is just the G = (V s(G) × E)/T E(cid:48) EG = (V s(G) × E)/G V s(G)/T π / V s(G)/G, we see that π∗(EG) = E(cid:48) π∗(ci(EG)). The result now follows from Martin's theorem (Theorem 5.4). G. Then we have that f∗(ET ) = π∗(EG), and so in particular f∗(ci(ET ) = (cid:3) The corollary shows that the restriction of Martin's isomorphism to degree 2 is just i : c1(Wi) (cid:55)→ ri(cid:88) c1(Wvij ). j=1 In particular we have that i(ωMQ) = ωMQab . Proposition 5.7. Let Amp(Q), Amp(Qab) denote the ample cones of MQ and MQab respectively. Then i(Amp(Q)) = Amp(Qab)W . Proof. Let v be a character for G, denoting its image under i : χ(G) (cid:44)→ χ(T ) as v as well. First v (T ). To see this, suppose v ∈ V is semi-stable for G, v. Let λ : C∗ → T be note that V ss a one-parameter subgroup of T such that limt→0 λ(t) · v exists. By inclusion, λ is a one-parameter v (G) ⊂ V ss 101112131415161718/ / / FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 11 subgroup of G, and so (cid:104)v, λ(cid:105) ≥ 0 by semi-stability of v. Hence v ∈ V ss v ∈ NE1(MQ), then V ss v (T ) (cid:54)= ∅, and hence v ∈ NE1(MQab )W . v (G) (cid:54)= ∅, so V ss v (T ). It follows that, if Ciocan-Fontanine -- Kim -- Sabbah use duality to construct a projection [6] p : NE1(MQab) → NE1(MQ). Suppose that α ∈ Amp(Q). Then for any C ∈ NE1(MQab), i(α) · C = α · p(C) ≥ 0. So i(α) ∈ Amp(Qab)W . Let Wall(G) ⊂ Pic(MQ) denote the union of all GIT walls given by the G action, and similarly for Wall(T ). Recall that v ∈ Wall(G) if and only if it has a non-empty strictly semi-stable locus. Suppose v ∈ Wall(G), with v in the strictly semi-stable locus. That is, there exists a non-trivial λ : C∗ → G such that limt→0 λ(t)·v exists and (cid:104)v, λ(cid:105) = 0. Now we don't necessarily have Im(λ) ⊂ T , but the image is in some maximal torus, and hence there exists g ∈ G such that Im(λ) ⊂ g−1T g. Consider λ(cid:48) = gλg−1. Then λ(cid:48)(C∗) ⊂ T. Since g · v is in the orbit of v under G, it is semi-stable with respect to G, and hence with respect to T . In fact, it is strictly semi-stable with respect to T , since limt→0 λ(cid:48)(t)g · v = limt→0 gλ(t) · v exists, and (cid:104)v, λ(cid:48)(cid:105) = (cid:104)v, λ(cid:105) = 0. So as a character of T , v has a non-empty strictly semi-stable locus, and we have shown that i(Wall(G)) ⊂ Wall(T )W . This means that the boundary of i(Amp(Q)) is not contained in Amp(Qab)W . Since both are full dimensional cones in the W invariant subspace, the inclusion i(Amp(Q)) ⊂ Amp(Qab)W is in fact (cid:3) an equality. Example 5.8. Consider again the example The Abelianization of this quiver is Walls are generated by collections of divisors that generate cones of codimension 1. We then intersect them with the Weyl invariant subspace, generated by (1, 1, 1, 0, 0) and (0, 0, 0, 1, 1). In this subspace, the walls are generated by This is consistent with the previous example. (1, 1, 1, 0, 0), (0, 0, 0, 1, 1), (−2,−2,−2, 3, 3). 5.3. The toric case. As a prelude to determining the ample cone of a general quiver flag variety, we first consider the toric case. Recall that a smooth projective toric variety (or orbifold) can be obtained as a GIT quotient of CN by an r-dimensional torus. Definition 5.9. The GIT data for a toric variety is an r-dimensional torus K with cocharacter lattice L = Hom(C∗, K), and N characters D1, . . . , DN ∈ L∨, together with a stability condition w ∈ L∨ ⊗ R. These linear data give a toric variety (or Deligne -- Mumford stack) as the quotient of an open subset Uw ⊂ CN by K, where K acts on CN via the map K → (C∗)N defined by the Di. Uw is defined as (z1, . . . , zN ) ∈ CN (cid:12)(cid:12)(cid:12) w ∈ Cone(Di : zi (cid:54)= 0) (cid:110) (cid:111) that is, its elements can have zeroes at zi, i ∈ I, only if w is in the cone generated by Di, i (cid:54)∈ I. Assume that all cones given by subsets of the divisors that contain w are full dimensional, as is the case for toric quiver flag varieties. Then the ample cone is the intersection of all of these. , 1031522210125115135152142222212 ELANA KALASHNIKOV In [15], the GIT data for a quiver flag variety is detailed; we present it slightly differently. The torus is K = (C∗)ρ. Let e1, . . . , eρ be standard basis of L∨ = Zρ and e0 = 0. Then each a ∈ Q1 gives a weight Da = −es(a) + et(a). The stability condition is 1 = (1, 1, . . . , 1). Identify L∨ ∼= Pic M (Q, 1). Then Da = Wa := W ∗ A minimal full dimensional cone for a toric quiver flag variety is given by ρ linearly independent Dai, ai ∈ Q1. Therefore for each vertex i with 1 ≤ i ≤ ρ, we need an arrow ai with either s(a) = i or t(a) = i, and these arrows should be distinct. For the positive span of these divisors to contain 1 requires that Dai has t(ai) = i. Fix such a set S = {a1, . . . , aρ}, and denote the corresponding cone by CS. As mentioned, the ample cone is the intersection of such cones CS. The set S determines a path from 0 to i for each i, given by concatenating (backwards) ai with as(ai) and so on; let us write fij = 1 if aj is in the path from 0 to i, and 0 otherwise. Then s(a) ⊗ Wt(a). ρ(cid:88) j=1 ei = fijDaj . the jth column of AS is (cid:80)ρ This gives us a straightforward way to compute the cone CS. Let BS be the matrix with columns given by the Dai, and let AS = B−1 S . The columns of AS are given by the aforementioned paths: i=1 fijei. If c ∈ Amp(Q), then ASc ∈ AS Amp(Q) ⊂ ASCS. Since ASDai = ei, this means that ASc is in the positive orthant. Proposition 5.10. Let M (Q, 1) be a toric quiver flag variety. Let c ∈ Amp(Q), c = (c1, . . . , cρ), be an ample class, and suppose that vertex i of the quiver Q satisfies the following condition: for all j ∈ Q0 such that j > i, there is a path from 0 to j not passing through i. Then ci > 0. Proof. Choose a collection S of arrows aj ∈ Q1 such that the span of the associated divisors Daj contains the stability condition 1, and such that the associated path from 0 to j for any j > i does not pass through i. Then the (i, i) entry of AS is 1 and all other entries of the ith row are zero. (cid:3) As ASc is in the positive orthant, ci > 0. Corollary 5.11. Let M (Q, r) be a quiver flag variety, not necessarily toric. If c = (c1, . . . , cρ) ∈ Amp(Q) and rj > 1, then cj > 0. Proof. Consider the Abelianized quiver. For any vertex v ∈ Qab 0 , we can always choose a path from the origin to v that does not pass through vj1: if there is an arrow between vj1 and v, then there is an arrow between vj2 and v, so any path through vj1 can be rerouted through vj2. Then (cid:3) we obtain that the j1 entry of i(c) is positive -- but this is just cj. 5.4. The ample cone of a quiver flag variety. Let M (Q, r) be a quiver flag variety and Q(cid:48) be the associated Abelianized quiver. For each i ∈ {1, . . . , ρ}, define Ti := {j ∈ Q0 all paths from 0 to j pass through i}, but pretending that a path can always circumvent a vertex j when rj > 1. This is motivated by the path structure of the Abelianized quiver. Note that i ∈ Ti, and that if ri = 1 then Ti = {i}. Theorem 5.12. The nef cone of M (Q, r) is given by the following inequalities. Suppose that a = (a1, . . . , aρ) ∈ Pic(MQ). Then a is nef if and only if (1) i = 1, 2, . . . , ρ. (cid:88) rjaj ≥ 0 j∈Ti Proof. We have already shown that the Weyl invariant part of the nef cone of MQ(cid:48) := M (Q(cid:48), 1) is the image of the nef cone of MQ := M (Q, r) under the natural map π : Pic(MQ) → Pic(MQ(cid:48)). Label the vertices of Q(cid:48) as vij, i ∈ {0, . . . , ρ}, j ∈ {1, . . . , ri}, and index elements of Pic(MQ(cid:48)) as (bij). The inequalities defining the ample cone of MQ(cid:48) are given by a choice of arrow Aij ∈ 1, t(Aij) = vij for each vij. This determines a path Pij from 0 → vij for each vertex vij. For Q(cid:48) each vij the associated inequality is: (2) (cid:88) vij∈Pkl bkl ≥ 0. Suppose that a is nef. First we show that a satisfies the inequalities (1). We have shown that aj ≥ 0 if rj > 1, so it suffices to consider i such that ri = 1. For each such i, take a choice of FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 13 arrows such that if vkl ∈ Pi1, k ∈ Ti. Moreover, choose arrows such that s(Aik) = s(Ail) for all k, l. Then this gives the inequalities (1), after restricting to the Weyl invariant locus. Therefore, if C is the cone defined by (1), we have shown that Nef(MQ) ⊂ C. Suppose now that a ∈ C and take a choice of arrows Akl. Write π(a) = (aij). We prove that the inequalities 2 are satisfied starting at vρρ. For ρ, the inequality is aρρ ≥ 0, which is certainly satisfied. Suppose the (ij + 1), (ij + 2), . . . , (ρρ) inequalities are satisfied. The inequality we want to establish for (ij) is rkak + X = aij + akl + Γ ≥ 0, rl(cid:88) (cid:88) (cid:88) k∈Ti−{i} l=1 (cid:33) (cid:32) (cid:88) (cid:88) ai + where k∈Ti−{i} vkl∈Pst ast ≥ 0, and (cid:3) As a ∈ C it suffices to show that Γ ≥ 0. By the induction hypothesis akl +(cid:80) s(Akl)=vij ,k(cid:54)∈Ti vkl∈Pst akl + Γ = ast . therefore Γ ≥ 0. This shows that π(a) satisfies (2). 5.5. Nef line bundles are globally generated. We conclude this section by proving that nef line bundles on quiver flag varieties are globally generated. This is well-known for toric varieties. This result will be important for us because in order to use the Abelian/non-Abelian Correspondence to compute the quantum periods of quiver flag zero loci, we need to know that the bundles involved are convex. Convexity is a difficult condition to understand geometrically, but it is implied by global generation. Let M (Q, r) be a quiver flag variety and Q(cid:48) be the associated Abelianized quiver. For each i ∈ {1, . . . , ρ}, let Ti be as above, and define: • Hi := {j ∈ Q0 j (cid:54)= 0 and all paths from 0 to i pass through j}; • h(i) := max Hi; h(cid:48)(i) := min Hi; and • Ri := {j h(cid:48)(j) = i}; again pretending that a path can always circumvent a vertex j when rj > 1. Note that i ∈ Hi. Observe too that the Ri give a partition of the vertices of Q, and that if j, k ∈ Ti, then j, k ∈ Rh(cid:48)(i). Proposition 5.13. Let L be a nef line bundle on M (Q, r). Then L is globally generated. i is given by the character (cid:81) χai Let fi be any homogenous polynomial of degree di = (cid:80) Proof. Let L be a nef line bundle on M (Q, r) given by (a1, . . . , aρ). A section of L is a G- equivariant section of the trivial line bundle on Rep(Q, r), where the action of G on the line bundle i . A point of Rep(Q, r) is given by (φa)a∈Q1 , φa : Crs(a) → Crt(a) , where G acts by change of basis. A choice of path i → j on the quiver gives an equivariant map Rep(Q, r) → Hom(Cri , Crj ) where G acts on the image by g · φ = gjφg−1 . If ri = rj = 1, such maps can be composed. We will show that L is globally generated. It suffices to show this in the case that {j aj (cid:54)= 0} ⊂ Ri for some i. This is because if j ∈ Ri, the inequalities in Theorem 5.12 which involve j only involve ak for k ∈ Ri. So if L = L1 ⊗ ··· ⊗ Lρ such that Li only has non-zero powers of det(Wj), j ∈ Ri, then L is nef if and only if all the Li are nef. So suppose that {j aj (cid:54)= 0} ⊂ Ri. If ri > 1, then Ri = {i} and L = det(Wi)⊗ai, which is known to be globally generated. So we further assume that ri = 1. rkak ≥ 0 in the maps given by paths 0 → i. That is, fi is a G-equivariant map Rep(Q, r) → C, where G acts on the image with i . For j in Ri with rj = 1 and j (cid:54)= i, let fj be any homogenous polynomial of degree character χdi rkak ≥ 0 in the maps given by paths h(j) → j (note that by definition, rh(j) = 1). That −dj is, fj is a G-equivariant map Rep(Q, r) → C, where G acts on the image with character χ h(j)χdj j . For j in Ri with rj > 1, let fj be a homogenous polynomial of degree ak ≥ 0 in the minors of the matrix whose columns are given by the paths h(j) → j. That is, fj is a G-equivariant map Rep(Q, r) → C, where G acts on the image with character χ j . For any x ∈ Rep(Q, r) which is semi-stable, and for any j ∈ Ri, there exists such an fj with fj(x) (cid:54)= 0, by construction of h(j). fj : Rep(Q, r) → C, and take the section s to be a sum of such s(cid:48). Then s is a Define s(cid:48) :=(cid:81) −rj aj h(j) χaj (cid:80) k∈Tj k∈Ti j∈Ri (cid:89) j∈Ri 14 ELANA KALASHNIKOV G-equivariant map Rep(Q, r) → C, where G acts on the image with character −rj aj h(j) χaj χ j . χbj j = χdi −dj h(j)χdj χ j i · (cid:89) (cid:88) j∈Ri,j(cid:54)=i,rj =1 rjaj − (cid:88) · (cid:89) (cid:88) k∈Ri,k(cid:54)=i,h(k)=i j∈Tk bi = j∈Ti j∈Ri,j(cid:54)=i,rj >1 rkak. We need to check that bj = aj for all j. This is obvious for j ∈ Ri with rj > 1. For j = i, This simplifies to ai in view of the fact that for all j ∈ Ti, there is a unique k such that h(k) = i and j ∈ Tk. The check for j ∈ Ri with rj = 1 is similar. Therefore s gives a well-defined section of L. For any x ∈ Rep(Q, r) semi-stable, there exists an s such that s(x) (cid:54)= 0, so L is globally generated. (cid:3) 6. The Abelian/non-Abelian Correspondence The main theoretical result of this paper, Theorem 6.3 below, proves the Abelian/non-Abelian Correspondence with bundles [6, Conjecture 6.1.1] for quiver flag zero loci. This determines all genus-zero Gromov -- Witten invariants, and hence the quantum cohomology, of quiver flag varieties, as well as all genus-zero Gromov -- Witten invariants of quiver flag zero loci involving cohomology classes that come from the ambient space. In particular, it determines the quantum period of a quiver flag varieties or quiver flag zero locus X with c1(T X) ≥ 0. 6.1. A brief review of Gromov -- Witten theory. We give a very brief review of Gromov -- Witten theory, mainly to fix notation, See [6,9] for more details and references. Let Y be a smooth projective variety. Given n ∈ Z≥0 and β ∈ H2(Y ), let M0,n(Y, β) be the moduli space of genus zero stable maps to Y of class β, and with n marked points [25]. While this space may be highly singular and have components of different dimensions, it has a virtual fundamental class [M0,n(Y, β)]virt of the expected dimension [4, 27]. There are natural evaluation maps evi : M0,n(Y, β) → Y taking a class of a stable map f : C → Y to f (xi), where xi ∈ C is the ith marked point. There is also a line bundle Li → M0,n(Y, β) whose fiber at f : C → Y is the cotangent space to C at xi. The first Chern class of this line bundle is denoted ψi. Define: (3) (cid:104)τa1(α1), . . . , τan (αn)(cid:105)n,β = (cid:90) n(cid:89) ev∗ i (αi)ψai i [M0,n(Y,β)]virt i=1 where the integral on the right-hand side denotes cap product with the virtual fundamental class. If ai = 0 for all i, this is called a (genus zero) Gromov -- Witten invariant and the τ notation is omitted; otherwise it is called a descendant invariant. It is deformation invariant. the element of Q[H2(Y )] representing β ∈ H2(Y ). Write N (Y ) for the Novikov ring of Y : We consider a generating function for descendant invariants called the J-function. Write qβ for (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) cβ ∈ C, for each d ≥ 0 there are only finitely many  . β such that ω · β ≤ d and cβ (cid:54)= 0 cβqβ β∈NE1(Y )  (cid:88) T ∈ H∗(Y ) (cid:55)→ 1 + T z−1 + z−1(cid:88) Here ω is the Kahler class on Y . The J-function assigns an element of H∗(Y ) ⊗ N (Y )[[z−1]] to every element of H∗(Y ), as follows. Let φ1, . . . , φN be a homogenous basis of H∗(Y ), and let φ1, . . . , φN be the Poincar´e dual basis. Then the J-function maps (cid:104)(cid:104)φi/(z − ψ)(cid:105)(cid:105)φi. N (Y ) = (4) Here 1 is the unit class in H 0(Y ), and (5) (cid:104)(cid:104)φi/(z − ψ)(cid:105)(cid:105) = (cid:88) ∞(cid:88) ∞(cid:88) n=0 a=0 qβ i 1 n!za+1(cid:104)τa(φi), T, . . . , T(cid:105)n+1,β. β∈NE1(Y ) The small J-function is the restriction of the J-function to H 0(Y ) ⊕ H 2(Y ); closed forms for the small J-function of toric complete intersections and toric varieties are known [18]. The quantum period GY (t) is the component of J(0) along 1 ∈ H•(Y ) after the substitutions z (cid:55)→ 1, qβ (cid:55)→ FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 15 t(cid:104)−KY ,β(cid:105). This is a power series in t. The quantum period satisfies an important differential equation called the quantum differential equation. Let X ⊂ Y be the zero locus of a generic section of a convex vector bundle E → Y and let e denote the total Chern class, which evaluates on a vector bundle F of rank r as e(F ) = λr + λr−1c1(F ) + ··· + λcr−1(F ) + cr(F ). The notation here indicates that one can consider e(F ) as the C∗-equivariant Euler class of F , with respect to the canonical action of C∗ on F which is trivial on the base of F and scales all fibers. In this interpretation, λ ∈ H•C∗ (pt) is the equivariant parameter. The twisted J-function Je,E is defined exactly as the J-function (4), but replacing the virtual fundamental class which occurs there (via equations 5 and 3) by [M0,n(Y, β)]virt ∩ e(E0,n,β), where E0,n,β is π∗(ev∗ n+1(E)), π : M0,n+1(Y, β) → M0,n(Y, β) is the universal curve, and evn+1 : M0,n(Y, β) → Y is the evaluation map. E0,n,β is a vector bundle over M0,n(Y, β), because E is convex. Functoriality for the virtual fundamental class [23] implies that j∗Je,E(T )(cid:12)(cid:12)λ=0 = JX (j∗T ) where j : X → Y is the embedding [7, Theorem 1.1]. Thus one can compute the quantum period of X from the twisted J-function. We will use this to compute the quantum period of Fano fourfolds which are quiver flag zero loci. The Abelian/non-Abelian correspondence is a conjecture [6] relating the J-functions (and more broadly, the quantum cohomology Frobenius manifolds) of GIT quotients V //G and V //T , where T ⊂ G is the maximal torus. It also extends to considering zero loci of representation theoretic bundles, by relating the associated twisted J-functions. As the Abelianization V //T of a quiver flag variety V //G is a toric quiver flag variety, the Abelian/non-Abelian correspondence conjectures a closed form for the J-functions of Fano quiver flag zero loci. Ciocan-Fontanine -- Kim -- Sabbah proved the Abelian/non-Abelian correspondence (with bundles) when V //G is a flag manifold [6]. We will build on this to prove the conjectures when V //G is a quiver flag variety. 6.2. The I-Function. We give the J-function in the way usual in the literature: first, by defining a cohomology-valued hypergeometric function called the I-function (which should be understood as a mirror object, but we omit this perspective here), then relating the J-function to the I-function. We follow the construction given by [6] in our special case. Let X be a quiver flag zero locus given by (Q, EG), and write MQ = M (Q, r) for the ambient quiver flag variety. Let (Qab, ET ) be the associated Abelianized quiver and bundle, MQab = M (Qab, (1, . . . , 1)). Assume, moreover, that ET splits into nef line bundles; this implies that both ET and EG are convex. To define the I-function, we need to relate the Novikov rings of MQ and MQab. Let Pic Q (respectively Pic Qab) denote the Picard group of MQ (respectively of MQab), and similarly for the cones of effective curves and effective divisors. The isomorphism Pic Q → (Pic Qab)W gives a projection p : NE1(MQab ) → NE1(MQ). In the bases dual to det(W1), . . . , det(Wρ) of Pic MQ and Wij, 1 ≤ i ≤ ρ, 1 ≤ j ≤ ri of Pic MQab , this is r1(cid:88) p : (d1,1, . . . , d1,r1, d2,1, . . . , dρ,rρ) (cid:55)→ ( rρ(cid:88) d1i, . . . , dρi). For β = (d1, . . . , dρ), define ρ(cid:88) (β) = i=1 i=1 di(ri − 1). Then, following [6, equation 3.2.1], the induced map of Novikov rings N (MQab) → N (MQ) sends i=1 β (cid:55)→ (−1)(β)qβ q where β = p( β). We write β → β if and only if β ∈ NE1(MQab ) and p( β) = β. MQab giving the split bundle ET . Given d ∈ NE1(MQab) define For a representation theoretic bundle EG of rank r on MQ, let D1, . . . , Dr be the divisors on (cid:81)r (cid:81)r i=1 (cid:81) (cid:81) i=1 IEG( d) = m≤(cid:104) d,Di(cid:105)(Di + mz) m≤0(Di + mz) . 16 ELANA KALASHNIKOV Notice that all but finitely many factors cancel here. theoretic bundle, in the sense that there exists AG, BG such that If E is K-theoretically a representation 0 → AG → BG → E → 0 is an exact sequence, we define (6) IE( d) = (cid:81)ρ IBG( d) IAG( d) . (cid:81) (cid:81)ρ i=1 (cid:81) (cid:81) (cid:81) (cid:81) (cid:81) (cid:81) 1 Example 6.1. The Euler sequence from Proposition 2.3 shows that for the tangent bundle TMQ ITMQ ( d) = a∈Qab m≤0(Da + mz) i=1 j(cid:54)=k a∈Qab 1 m≤(cid:104) d,Da(cid:105)(Da + mz) (cid:81) m≤(cid:104) d,Dij−Dik(cid:105)(Dij − Dik + mz) j(cid:54)=k m≤0(Dij − Dik + mz) . Here Da is the divisor on MQab corresponding to the arrow a ∈ Qab corresponding to the tautological bundle Wij for vertex ij. 1 , and Dij is the divisor Example 6.2. If X is a quiver flag zero locus in MQ defined by the bundle EG, then the adjunction formula implies that ITX ( d) = ITMQ ( d)IEG( d). Define the I-function of X ⊂ MQ to be (cid:88) (cid:88) (cid:81)ρ d→d i=1 IX,MQ = d∈NE1(MQ) (cid:81) (cid:81) (cid:81) (cid:81) 1 a∈Qab m≤0(Da + mz) a∈Qab 1 m≤(cid:104) d,Da(cid:105)(Da + mz) Since (7) ITX ( d) = (−1)(d)qdITX ( d). (cid:81) (cid:81)ρ (cid:81) (cid:81) j(cid:54)=k i=1 (cid:81) m≤(cid:104) d,Dij−Dik(cid:105)(Dij − Dik + mz) j(cid:54)=k m≤0(Dij − Dik + mz) is invariant under the action of the Weyl group on the Dij, by viewing these as Chern roots of the tautological bundles Wi we can express it as a function in the Chern classes of the Wi. We can therefore regard ITX ( d) as an element of H•(MQ, C). Thus the I-function is an element of H•(MQ, C) ⊗ N (MQ) ⊗ C((z−1)). If X is Fano then (8) where O(z−2) denotes terms of the form αzk with k ≤ −2 and C ∈ H 0(MQ, C) ⊗ N (MQ); furthermore C vanishes if the Fano index of X is greater than 1. Theorem 6.3. Let X be a Fano quiver flag zero locus given by (Q, EG), and let j : X → MQ be the embedding of X into the ambient quiver flag variety. Then IX,MQ(z) = 1 + z−1C + O(z−2) JX (0) = e−c/zj∗IX,MQ where c = j∗C. Remark 6.4. Via the Divisor Equation and the String Equation [34, §1.2], Theorem 6.3 deter- mines JX (τ ) for τ ∈ H 0(X) ⊕ H 2(X). 6.3. Proof of Theorem 6.3. Givental has defined [11,19] a Lagrangian cone LX in the symplectic vector space HX := H∗(X, C)⊗N (X)⊗C((z−1)) that encodes all genus-zero Gromov -- Witten invari- ants of X. Note that JX (T ) ∈ HX for all T . The J-function has the property that (−z)JX (T,−z) is the unique element of LX of the form −z + T + O(z−1) and this, together with the expression (8) for the I-function and the String Equation JX (T + c, z) = ec/zJX (T, z) shows that Theorem 6.3 follows immediately from Theorem 6.5 below. Theorem 6.5 is stronger: it does not require the hypothesis that the quiver flag zero locus X be Fano. Theorem 6.5. Let X be a quiver flag zero locus given by (Q, EG), and let j : X → MQ be the embedding of X into the ambient quiver flag variety. Then (−z)j∗IX,MQ(−z) ∈ LX . t(a)=i Qs(a). V is globally generated and hence convex. It is not representation theoretic, but it is K-theoretically: the sequence 0 → F ∗ i ⊗ Qi → (C si)∗ ⊗ Qi → (C si)∗ ⊗ Qi/F ∗ i → 0 is exact. Let i : MQ → Y denote the inclusion. a representation theoretic vector bundle E(cid:48) locus X is cut out of Y by a regular section of V (cid:48) = V ⊕ E(cid:48) Both Y and MQ are GIT quotients by the same group; we can therefore canonically identify GMQ is EG. Our quiver flag zero G on Y such that E(cid:48) G. Note that ITMQ ( d)IV ( d) = ITY ( d)IV (cid:48)( d). The I-function IX,MQ defined by considering X as a quiver flag zero locus in MQ with the bundle EG then coincides with the pullback i∗IX,Y of the I-function defined by considering X as a quiver flag zero locus in Y with the bundle V (cid:48). It therefore suffices to prove that (−z)(i ◦ j)∗IX,Y (−z) ∈ LX . FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 17 constructed a vector bundle V on Y such that MQ is cut out of Y by a regular section of V : i=1(P si−1)ri its Abelianization. In §3 we Proof. Let Y =(cid:81)ρ i=1 Gr( si, ri), and denote by Y ab =(cid:81)ρ where si is the number of paths from 0 to i and Fi =(cid:76) ρ(cid:77) V = Qi ⊗ Csi∗/F ∗ i=2 i We consider a C∗-equivariant counterpart of the I-function, defined as follows. For a represen- tation theoretic bundle WG on Y , let D1, . . . , Dr be the divisors on Y ab giving the split bundle WT , and for d ∈ NE1(Y ab) set C∗ WG m≤(cid:104) d,Di(cid:105)(λ + Di + mz) (cid:81)r (cid:81)r (cid:81) (cid:81) ( d) = i=1 I . i=1 m≤0(λ + Di + mz) We extend this definition to bundles on Y -- such as V (cid:48) -- that are only K-theoretically representation theoretic in the same way as (6). Recalling that ITY ( d) = we define (cid:81)ρ (cid:81) (cid:81)ρ i=1 (cid:81) (cid:81) (cid:81) m≤(cid:104) d,Dij−Dik(cid:105)(Dij − Dik + mz) m≤0(Dij − Dik + mz) j(cid:54)=k i=1 j(cid:54)=k C∗ I X,Y (z) = (cid:88) (cid:88) d∈NE1(Y ) d→d (cid:81) (cid:81)ρ (cid:81)ri (cid:81) (cid:81)ri (cid:81)ρ (−1)(d)qdITY ( d) I C∗ V (cid:48) ( d). j=1 j=1 i=1 i=1 m≤0(Dij + mz)si m≤(cid:104) d,Dij(cid:105)(Dij + mz)si , C∗ X,Y . In view of [7, Theorem 1.1], it therefore suffices to prove that The I-function IX,Y can be obtained by setting λ = 0 in I X,Y (−z) ∈ Le,V (cid:48) C∗ (−z)I where Le,V (cid:48) is the Givental cone for the Gromov -- Witten theory of Y twisted by the total Chern class e and the bundle V (cid:48). If V (cid:48) were a representation theoretic bundle, this would follow immediately from the work of In fact V (cid:48) is only K- Ciocan-Fontanine -- Kim -- Sabbah: see the proof of Theorem 6.1.2 in [6]. theoretically representation theoretic, but their argument can be adjusted almost without change to this situation. Suppose that AG and BG are homogenous vector bundles, and that 0 → AG → BG → V → 0 is exact. Then we can also consider an exact sequence 0 → AT → BT → F → 0 on the Abelianization, and define VT := F. Using the notation of the proof of [6, Theorem 6.1.2], the point is that (cid:52)(V )(cid:52)(AG) = (cid:52)(BG) Here, (cid:52)(V ) is the twisting operator that appears in the Quantum Lefschetz theorem [11]. We can then follow the same argument for After Abelianizing, we obtain (cid:52)(BT )/(cid:52)(AT ) = (cid:52)(F ), and conclude that (cid:52)(BG)/(cid:52)(AG) (−z)I X,Y (−z) ∈ Le,V (cid:48) C∗ 18 ELANA KALASHNIKOV as claimed. This completes the proof. (cid:3) FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 19 Appendix A. Computations T. Coates, E. Kalashnikov, A. Kasprzyk In this appendix, we describe the computer search for four dimensional Fano quiver flag zero loci with codimension at most four. Code to perform this and similar analyses, using the computational algebra system Magma [5], is available at the repository [1]. A database of Fano quiver flag varieties, which was produced as part of the calculation, is available at the repository [2]. A.1. Classifying Quiver Flag Varieties. The first step is to find all Fano quiver flag varieties of dimension at most 8. A non-negative integer matrix A = [ai,j]0≤i,j≤ρ and a dimension vector r ∈ Zρ+1 >0 determine a vertex-labelled directed multi-graph: the ρ + 1 vertices are labelled by the ri, and the adjacency matrix for the graph is A. Such a graph, if it is acyclic with a unique source, and the label of the source is 1, also determines a quiver flag variety. Two (A, r) pairs can determine the same graph and hence the same quiver flag varieties. Definition A.1. A pair (A, r) determining a quiver flag variety is in normal form if r is increasing and, under all permutations of the ρ + 1 indices that preserve r, the columns of A are lex minimal. Two pairs in normal form determine the same quiver flag variety (and hence the same graph) if and only if they are equal. Recall that quiver flag varieties are towers of Grassmannians (see §2.3), and that the ith step in the tower is given by the relative Grassmannian Gr(Fi, ri), where Fi is a vector bundle of rank si. Using this construction it is easy to see that if si = ri then this quiver flag variety is equivalent to the quiver flag variety Q with vertex i removed, and one arrow k → j for every path of the form k → i → j. Therefore we can assume that si > ri, and hence that every vertex contributes strictly positively to the dimension of the quiver flag variety. With this constraint, there are only finitely many quiver flag varieties with dimension at most 8, and each such has at most 9 vertices. The algorithm to build all quiver flag varieties with dimension at most 8 is as follows. Start with the set S of all Grassmannians of dimension at most 8. Given an element of S of dimension less than 8, add one extra labelled vertex and extra arrows into this vertex, in all possible ways such that the dimension of the resulting quiver flag variety is at most 8. Put these in normal form and include them in S. Repeat until there are no remaining elements of S of dimension less than 8. In this way we obtain all quiver flag varieties of dimension at most 8. We then compute the ample cone and anti-canonical bundle for each, and discard any which are not Fano. We find 223044 Fano quiver flag varieties of dimension at most 8; 223017 of dimension 4 ≤ d ≤ 8. Of these 50617 (respectively 50612) are non-toric quiver flag varieties. d 1 2 3 4 1 2 3 4 5 6 7 8 1 3 2 2 8 11 48 3 17 44 262 2 27 118 4 41 264 903 2 54 498 2484 4 74 872 5852 r 5 6 7 8 231 1647 7005 23268 1202 10618 54478 6541 69574 36880 Table 1. The number of Fano quiver flag varieties by dimension d and Picard rank r Remark A.2. In our codebase we define new Magma intrinsics QuiverFlagVariety(A,r), which creates a quiver flag variety from an adjacency matrix A and dimension vector r, and which creates a Fano quiver flag variety, in normal form, from its ID [1, 2]. We assign IDs to Fano quiver flag varieties of dimension at most 8, in the range {1 . . . 223044}, by placing them in QuiverFlagVarietyFano(id), 20 ELANA KALASHNIKOV normal form and then ordering them first by dimension, then by Picard rank, then lexicographically by dimension vector, then lexicographically by the columns of the adjacency matrix. We also define Magma intrinsics NefCone(Q), MoriCone(Q), PicardLattice(Q), and CanonicalClass(Q) that compute the nef cone, Mori cone, Picard lattice and canonical class of a quiver flag variety Q, and an intrinsic PeriodSequence(Q,l) that computes the first l + 1 terms of the Taylor expansion of the regularised quantum period of Q. See §A.5 for more details. A.2. The Class of Vector Bundles that We Consider. We consider all bundles E on a given quiver flag variety that: • are direct sums of bundles of the form (9) L ⊗ Sα1(W1) ⊗ ··· ⊗ Sαρ (Wρ) where each Sαi is a positive Schur power and L is a nef line bundle; and • have rank c, where c is four less than the dimension of the ambient quiver flag variety. Remark 3.2 shows that positive Schur powers Sα(Wi) are globally generated, and Proposition 5.13 shows that nef line bundles are globally generated. Since the tensor product of globally generated vector bundles is globally generated, the first condition ensures that E is globally generated. In particular, therefore, the zero locus X of a generic section of E is smooth. The second condition ensures that the zero locus X, if non-empty, is a fourfold. Global generation also implies that the bundle E is convex, which allows us to compute the quantum period of X as described in §6.1. sequence of non-negative integers, and write L = (cid:78)ρ Consider a summand as in (9). We can represent the partition αi as a length ri decreasing j=1(det Wj)aj where aj may be negative. Therefore each such summand is determined a length ρ sequence of generalised partitions: the partition (with possibly negative entries) corresponding to index i is αi + (ai, . . . , ai). Remark A.3. In our codebase we define a new Magma intrinsic QuiverFlagBundle(Q,[A1,...,Ak]) which creates a bundle of the above form, on the quiver flag variety Q, from a sequence of gen- eralised partitions (A1, . . . , Ak). We also define an intrinsic FirstChernClass(E) that computes the first Chern class of such a bundle E; intrinsics Degree(E) and EulerNumber(E) that com- pute the degree and Euler number2 of the zero locus X of a generic section of E; and intrinsics HilbertCoefficients(E,l) and PeriodSequence(E,l) that compute the first l + 1 terms of, re- spectively, the Hilbert series of X and the Taylor expansion of the regularised quantum period of X. See §A.5. A.3. Classifying Quiver Flag Bundles. In this step, we describe the algorithm for determining all bundles on a given quiver flag variety that determine a smooth four-dimensional Fano quiver flag zero locus. A vector bundle as above is determined by a tuple (A, r, P ), where A is an adjacency matrix, r is a dimension vector r, and P = (P1, . . . , Pk) is a sequence where each Pi is a length-ρ sequence of generalised partitions such that the jth partition in each Pi is of length rj. Note that we regard the summands (9) in our vector bundles as unordered; also, as discussed above, different pairs (A, r) can determine the same quiver flag variety. We therefore say that a tuple (A, r, P ) is in normal form if the pair (A, r) is in normal form, P is in lex order, and under all permutations of the vertices preserving these conditions, the sequence P is lex minimal; we work throughout with tuples in normal form. Given a Fano quiver flag variety M (Q, r) of dimension 4 + c, c ≤ 4, with anti-canonical class −KQ and nef cone Nef(Q), we search for all bundles E such that • E is a direct sum of bundles of the form (9); • rank(E) = c; • −KQ − c1(E) ∈ Amp(Q). The last condition ensures that the associated quiver flag zero locus X, if non-empty, is Fano. We proceed as follows. We first find all possible summands that can occur; that is, all irreducible vector bundles E of the form (9) such that rank(E) ≤ c and −KQ − c1(E) ∈ Amp(Q). Let Irr(Q) be the set of all such bundles. Write Irr(Q) = Irr(Q)1 (cid:116) Irr(Q)2, where Irr(Q)1 contains vector bundles of rank strictly larger than 1, and Irr(Q)2 contains only line bundles. We then search for two vector 2This is the Euler characteristic of X as a topological space. FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 21 bundles E1, E2 such that Ei is a direct sum of bundles from Irr(Q)i and that E = E1 ⊕ E2 satisfies the conditions above. For each x ∈ Nef(Q) such that −KQ − x is ample, we find all possible ways to write x as (10) x = ai l(cid:88) i=1 m(cid:88) y = bj where the ai are (possibly repeated) elements of a Hilbert basis for Nef(Q). There are only finitely many decompositions (10); finding them efficiently is a knapsack-type problem that has already been solved [12]. For each c ≤ c and each partition of the ai into at most c/2 groups S1, . . . , Ss, we find all possible choices of F1, . . . , Fs ∈ Irr1 such that c1(Fi) = aj rank(F1) + ··· + rank(Fs) = c. Set E1 = F1 ⊕ ··· ⊕ Fs. Then for each y ∈ Nef(Q) such that −KQ − x − y is ample, we again find all ways of writing (cid:88) j∈Si as a sum of Hilbert basis elements. Each partition of the bj into c − c groups gives a choice of nef line bundles L1, . . . , Lc−c ∈ Irr2(Q), and we set E2 = ⊕Lj. Remark A.4. Treating the higher rank summands Irr2 and line bundles Irr1 separately here is not logically necessary, but it makes a huge practical difference to the speed of the search. j=1 A.4. Classifying Quiver Flag Zero Loci. For each of the Fano quiver flag varieties Q of di- mension between 4 and 8, found in §A.1, we use the algorithm described in §A.3 to find all bundles on Q of the form described in §A.2. This produces 10788446 bundles. Each such bundle E deter- mines a quiver flag zero locus X that is either empty or a smooth Fano fourfold. We discard any varieties that are empty or disconnected, and for the remainder compute the first fifteen terms of the Taylor expansion of the regularised quantum period of X, using Theorem 6.3. For many of the quiver flag zero loci that we find this computation is extremely expensive, so in practice it is essential to use the equivalences described in §4 to replace such quiver flag zero loci by equivalent and more tractable models. The number of equivalence classes is far smaller than the number of quiver flag zero loci that we found, and so this replaces roughly 10 million calculations, many of which are hard, by around half a million calculations, almost all of which are easy. In this way we find 749 period sequences. We record these period sequences, together with the construction, Euler number, and degree for a representative quiver flag zero locus, in Appendix B below. 141 of the period sequences that we find are new. Thus we find at least 141 new four-dimensional Fano manifolds3. Remark A.5. A computationally cheap sufficient condition for a quiver flag zero locus to be empty arises as follows. If W is the tautological quotient bundle on Gr(n, r), where 2r − 1 > n, then a generic global section of ∧2W or Sym2 W has an empty zero locus. Thus if i is a vertex in a quiver Q such that all arrows into i are from the source, and 2ri − 1 > n0i = si, then there are no global sections of ∧2Wi or Sym2 Wi with non-empty zero locus: to see this, apply Proposition 3.1 to Q. A.5. Cohomological Computations for Quiver Flag Zero Loci. In this section we describe how we compute the degree, Euler characteristic, Hilbert series, and Taylor expansion of the regularised quantum period for quiver flag varieties and quiver flag zero loci. This relies on Martin's integration formula [28] and Theorem 6.3. Let V be a smooth projective variety with an action of G on V , let T be a maximal torus in G, and consider the GIT quotients V //G and V //T determined by a character of G. Let π : V ss(G)/T → V //G be the projection and i : V ss(G)/T → V ss(T )/T = V //T be the inclusion. Let λ∈Roots(G) c1(Lλ), where Lλ is the line bundle on V //T associated W be the Weyl group, and e =(cid:81) to the character λ. 3To be precise: we find at least 141 four-dimensional Fano manifolds for which the regularised quantum period was not previously known. The regularised quantum period of a Fano manifold X is expected to completely determine X. See [10, 12] for known quantum periods. 22 ELANA KALASHNIKOV Theorem A.6 (Martin's Integration Formula, [28]). For any a ∈ H∗(V //G, C) and any a ∈ H∗(V //T, C) satisfying π∗(a) = i∗(a)(cid:90) (cid:90) a = 1 W V //G V //T a ∪ e. (cid:90) (cid:90) If a ∈ H∗(V //G, C) and a ∈ H∗(V //T, C) satisfy π∗(a) = i∗(a) then we say that a is a lift of a. In our case the Abelianization V //T is a smooth toric variety, and the cohomology rings of such varieties, being Stanley -- Reisner rings, are easy to work with computationally [5, 36]. For example, we can use this to compute the number of components h0(X,OX ) of a Fano quiver flag zero locus X. By Kodaira vanishing, h0(X,OX ) = χ(X), and applying the Hirzebruch -- Riemann -- Roch theorem gives (11) χ(OX ) = ch(OX ) ∪ T d(TX ) = T d(TX ). X X We need to find a lift of the Todd class of TX . Writing TX as a K-theoretic quotient of representation theoretic bundles via the Euler sequence, as in the proof of Theorem 6.3, gives the lift that we seek; we then use Martin's formula to reduce the integral (11) to an integral in the cohomology ring of the Abelianization. The same approach allows us to compute the first two terms χ(X,−KX ), χ(X,−2KX ) of the Hilbert series of X -- which determine the entire Hilbert series, since X is a Fano fourfold -- as well as the degree and Euler characteristic of X. To compute the first few Taylor coefficients of the quantum period of X, we combine this approach with the explicit formula in Theorem 6.3. FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 23 Appendix B. Regularized Quantum Periods for Quiver Flag Zero Loci T. Coates, E. Kalashnikov, A. Kasprzyk B.1. The Table of Representatives. As described in Appendix A, we divided the 4-dimensional quiver flag zero loci X that we found into 749 buckets, according to the first 15 terms of the Taylor expansion of the regularised quantum period of X. We refer to these Taylor coefficients as the period sequence. Table 3 below gives, for each of the 749 period sequence buckets, a representative quiver flag zero locus X as well as the degree and Euler number of X. (In some cases we do not know that all the quiver flag zero loci in a bucket are isomorphic, but we checked that they all have the same degree, Euler number, and Hilbert series.) The quiver flag zero locus X is represented by the adjacency matrix and dimension vector of its ambient quiver flag variety Y = M (Q, r), together with the sequence of generalised partitions that determine a vector bundle E → Y such that X is the zero locus of a generic section of E. The generalised partitions are written as Young diagrams, with: • ∅ representing the empty Young diagram; • a filled Young diagram, such as , representing the dual to the vector bundle represented by the unfilled Young diagram . Filled Young diagrams that occur always represent line bundles. The entries in Table 3 give representatives of each period sequence bucket that are chosen so as to make the computation of geometric data (the period sequence etc.) straightforward4. Even though the Table is constructed by considering all four-dimensional Fano manifolds that occur as quiver flag zero loci in codimension up to four, in a four cases there is no tractable representative as a quiver flag zero locus of low codimension. In these cases the Table contains a representative as a quiver flag zero locus in higher codimension; the reader who prefers models in lower-dimensional ambient spaces should consult Table 2. Period ID Adjacency matrix Dimension vector Generalized partitions 73 144 439 552 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 3 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 1 0 0 0 1 1 0 1 0 0 1 3 0 0 0 0 2 1 0 0 0 5 0 0 5 0 0 1 2 2 2 2 (∅, , , ∅) (∅, , ∅) 1 1 2 2 2 1 4 4 1 2 4 (cid:18) (cid:19) (cid:18) , ∅ (cid:18) , , (cid:19) (cid:16) , (cid:18) , ∅(cid:17) , , ∅ (cid:19) , (cid:19) , ∅ (cid:18) (cid:19) , ∅(cid:17) , , ∅ (cid:16) , Table 2. Representatives for certain Period IDs in codimension at most four Remark B.1. The data in Tables 3 and 4 can also be found, in machine readable form, in the ancillary files that accompany this paper. B.2. The Table of Period Sequences. Table 4 records the first 8 terms of the period sequence, α0, α1, . . . , α7, for each of the 749 period sequence buckets. It also records, where they exist, the names of known four-dimensional Fano manifolds which have the same first fifteen terms of the period sequence. Notation is as follows: • Pn denotes n-dimensional complex projective space; • Qn denotes a quadric hypersurface in Pn+1; • FI4 4They are chosen to minimize the quantity(cid:80)ρ k is the kth four-dimensional Fano manifold of index 3, as in [10, §5]; i=0 r2 i , which is a rough proxy for the complexity of the Chow ring of the Abelianization. 24 ELANA KALASHNIKOV k is the kth four-dimensional Fano manifold of index 2 and Picard rank 1, as in [10, §6.1]; k is the kth four-dimensional Fano manifold of index 2 and Picard rank at least 2, as in [10, §6.2]; k is the kth four-dimensional toric Fano manifold, as in [10, §7]; • V 4 • MW4 • BØS4 • Strk are the Strangeway fourfolds described in [10, §8]; • CKPk is the kth four-dimensional toric complete intersection, as in [12]; • S2 • V 3 • B3 • MM3 ρ -- k denotes the kth entry in the Mori -- Mukai list of three-dimensional Fano manifolds of Picard rank ρ [29 -- 33]. We use the ordering as in [9], which agrees with the original papers of Mori -- Mukai except when ρ = 4. k denotes the del Pezzo surface of degree k; k denotes the three-dimensional Fano manifold of Picard rank 1, Fano index 1, and degree k; k denotes the three-dimensional Fano manifold of Picard rank 1, Fano index 2, and degree 8k; Remark B.2. It appears from Table 4 as if the period sequences with IDs 72 and 73 might coin- cide. This is not the case. The coefficients α8, α9, and α10 in these cases are: Period ID α8 α9 α10 72 73 32830 32830 212520 227640 1190952 1190952 Remark B.3. 590 of the period sequences that we find coincide with period sequences for toric complete intersections, at least for the first 15 terms. 579 of these are realised by quiver flag zero loci that are also toric complete intersections. For the remaining 11 cases -- period sequences with IDs 17, 48, 73, 144, 145, 158, 191, 204, 256, 280, and 282 -- there is no model as a toric complete intersection that is also a quiver flag zero locus in codimension at most four. In four of these cases -- with IDs 17, 48, 144, and 256 -- the toric complete intersection period sequence is realised by a smooth four-dimensional toric variety. Remark B.4. An earlier version of this paper omitted two of the period sequences that we find below, due to erroneous hand calculations in special cases. In this version all computations are performed in software, in a uniform way; we believe that this makes them more likely to be correct. Acknowledgements. EK was supported by the Natural Sciences and Engineering Research Council of Canada, and by the EPSRC Centre for Doctoral Training in Geometry and Number Theory at the Interface, grant number EP/L015234/1. TC was supported by ERC Consolidator Grant number 682602 and EPSRC Programme Grant EP/N03189X/1. AK was supported by EP- SRC Fellowship grant EP/N022513/1. The computations that underpin this work were performed on the Imperial College HPC cluster. We thank Andy Thomas, Matt Harvey, and the Research Computing Service team at Imperial for invaluable technical assistance. Table 3: Certain 4-dimensional Fano manifolds with Fano index 1 that arise as quiver flag zero loci Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 1 2 3 4 5 6 7 8 9 10 11 12 13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 0 0 0 3 6 0 1 0 1 2 0 1 1 0 2 1 0 0 1 0 0 1 1 2 0 1 3 0 0 1 0 1 1 1 0 0 1 0 0 3 5 0 0 5 0 0 3 0 0 3 0 0 2 1 0 0 2 0 0 1 4 0 0 3 1 0 2 0 0 0 3 0 0 0 5 0 0 2 0 0 0 3 0 0 0 2 0 0 0 2 0 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 ( , ),( , ∅) ( ) ( , ) ( , ∅, ∅) (∅, ) 1 1 1 ( , ) 1 1 1 1 1 1 1 1 ( , , ∅) 1 1 1 ( , ∅),( , ∅) 625 512 512 431 513 513 459 417 486 432 405 384 351 5 8 6 9 9 9 12 13 12 9 12 13 9 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 2 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 14 15 16 17 18 19 20 21 22 23 24 25 3 0 0 0 0 3 0 0 0 2 1 0 0 3 0 0 0 0 4 0 0 0 2 0 1 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 5 0 0 0 0 1 1 3 0 0 0 0 0 0 1 1 0 0 2 0 0 0 1 0 0 0 1 0 0 0 2 0 0 3 5 0 1 0 1 3 0 0 2 1 0 3 0 0 0 5 0 0 0 0 0 0 1 1 6 0 0 0 0 0 1 5 0 0 0 1 0 0 1 5 0 0 4 0 0 1 1 1 ( , ),( , ) 1 1 2 ( , ) 1 1 1 1 ( , ∅, ∅) 1 1 1 2 (∅, , ∅),(∅, , ∅),(∅, , ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ∅) 1 1 1 (∅, ),( , ∅) 1 1 1 2 (∅, , ∅),( , ∅, ∅) 1 1 1 2 1 1 1 1 1 1 1 1 2 1 2 2 (∅, (∅, , , ),( , ∅, ∅) ),( , ∅, ∅) ),( , ( , (cid:16) (cid:17) (cid:16) (cid:17) (cid:16) (cid:16) , ∅(cid:17) , , ) , ∅(cid:17) 486 433 401 406 322 378 358 347 330 297 405 325 9 9 13 13 18 10 13 13 14 13 6 10 Continued on next page. 2 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 26 27 28 29 30 31 32 33 34 35 36 37 38 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 7 0 1 0 1 5 0 0 1 0 0 1 2 0 1 5 0 0 6 0 1 0 1 1 0 1 0 0 4 1 0 0 1 0 0 0 3 0 0 0 0 1 6 0 0 4 0 0 1 0 0 1 5 0 0 5 0 0 5 0 0 4 0 0 4 0 0 2 0 0 0 1 0 0 1 0 0 0 0 1 5 0 0 0 2 0 1 0 4 0 0 0 1 1 1 1 2 (∅, ∅, , ∅),(∅, , ∅, ∅),( , ∅, ∅, ∅) 1 1 ( ),( ) 1 1 1 (∅, ),( , ) 1 1 2 ( , ),( , ) 1 1 1 1 (∅, ),( , ∅, , ) ) ),( , (cid:17) ( , (cid:16) , ( ) ( , ) 1 1 1 1 1 4 1 1 1 1 1 1 1 1 1 1 1 ( , ),( , ) 1 1 1 1 1 1 1 1 ( , ∅, ),( , , ) 290 324 292 273 261 244 225 243 211 544 512 464 431 18 12 14 9 17 16 5 27 29 8 8 12 11 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 2 7 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 39 40 41 42 43 44 45 46 47 48 49 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 2 0 1 1 0 0 1 1 0 0 1 2 0 1 1 0 0 1 0 0 4 0 0 0 4 0 0 1 1 3 0 0 0 2 0 0 0 5 0 0 4 0 0 1 0 0 1 1 1 0 0 4 0 0 1 1 0 0 4 0 0 1 1 0 0 1 0 0 1 5 0 0 0 2 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 0 1 2 0 2 0 1 0 1 1 1 (∅, ) 1 1 1 ( , ) ( , ∅, ) (∅, , ) ( , ∅) 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 ( , ),( , ∅) 1 1 1 1 ( , ∅, ),( , ∅, ) 1 1 1 1 1 1 1 2 (∅, ∅, ),(∅, ∅, ) 1 1 1 1 480 416 400 383 350 480 432 496 432 433 432 8 10 12 13 12 12 9 12 12 13 12 Continued on next page. 2 8 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 50 51 52 53 54 55 56 57 58 59 2 0 0 0 0 3 0 0 0 0 1 0 0 1 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 1 0 0 0 1 1 0 0 0 2 1 0 0 1 1 0 0 1 1 0 0 1 0 0 0 0 2 1 0 0 0 1 1 0 0 1 1 0 0 1 0 2 0 0 0 1 0 0 0 1 0 0 0 3 1 1 0 0 3 0 0 0 2 1 0 0 0 0 0 0 2 1 0 0 1 0 2 0 0 1 1 1 0 0 0 3 0 0 0 1 0 1 0 0 4 0 0 0 4 0 0 0 2 0 1 0 3 0 0 0 4 0 0 0 0 1 1 0 0 0 3 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 ( , ∅, ∅, ∅) 1 1 1 1 (∅, , ∅),( , , ) 1 1 1 1 (∅, ∅, ) 1 1 1 1 1 1 1 1 (∅, , ) (∅, , ) 1 1 1 1 1 (∅, , , ),( , ∅, , ) 1 1 1 1 1 (∅, , ∅, ) 1 1 1 1 ( , , ∅) 432 400 384 378 464 416 384 384 357 336 16 12 16 12 16 12 13 16 17 14 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 2 9 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 60 61 62 63 64 65 66 67 68 69 70 1 0 0 1 0 4 0 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 2 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 1 1 0 0 4 0 0 0 3 0 0 0 3 0 0 4 0 0 0 4 1 0 0 1 0 0 3 0 0 0 3 1 0 2 0 0 0 0 2 3 0 0 0 4 0 0 1 0 0 1 1 0 0 1 4 0 0 1 1 0 0 3 0 0 0 5 0 0 0 1 0 2 4 0 0 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 (∅, ),( , ∅) 1 1 1 1 ( , ∅, ∅),( , ∅, ∅) 1 1 1 1 ( , ∅, ∅),( , , ) 1 1 1 ( , ),( , ) 1 1 1 1 ( , ∅, ),( , ∅, ∅) 1 1 1 1 1 1 1 (∅, ),( , ) 1 1 1 1 (∅, ),(∅, , ∅) , 1 1 1 ( , ) 357 336 324 336 303 270 480 432 432 368 352 16 13 12 12 13 9 12 12 8 13 12 Continued on next page. 3 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 71 72 73 74 75 76 77 78 79 2 0 0 0 0 3 0 0 0 0 4 0 0 0 0 3 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 3 3 0 0 0 0 0 0 0 3 0 0 0 3 1 0 0 1 0 0 0 0 2 0 0 0 0 1 0 0 0 3 0 1 0 0 1 0 0 0 1 1 3 0 0 0 0 0 2 0 2 1 0 0 1 2 1 0 0 0 0 0 1 1 0 4 0 0 0 1 1 0 0 1 0 1 0 0 1 1 0 0 0 0 0 1 0 1 2 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 0 3 0 0 0 0 4 0 0 0 1 1 1 1 1 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ) 1 1 1 2 2 (∅, ∅, , ∅),(∅, ∅, , ∅),( , ∅, ∅, ) 1 1 1 1 1 1 1 1 (∅, (∅, , , ),(∅, , ∅) ),( , ∅, ) 1 1 2 2 (∅, , ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 2 ( , , ∅, ∅),( , , ∅, ),( , ∅, ∅, ∅) 1 1 1 1 (∅, ),( , ∅, , ) 448 389 369 352 368 337 347 331 335 16 16 17 13 12 13 16 17 15 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 3 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 80 81 82 83 84 85 86 87 88 89 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 0 0 0 2 0 0 1 1 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 1 1 0 0 0 1 1 0 0 0 3 3 0 0 0 1 0 0 1 3 0 0 0 2 0 0 1 0 0 0 0 2 1 0 0 1 0 0 0 1 1 2 0 0 1 1 0 0 0 1 0 1 0 2 3 0 0 0 5 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 0 5 0 0 0 2 0 0 0 1 4 0 0 0 3 0 0 0 0 3 0 0 0 2 0 0 1 0 3 0 0 0 0 3 0 0 0 0 1 1 1 1 ( , ∅, ∅),( , ∅, ∅) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 (∅, , ) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 1 1 (∅, , , ),( , ∅, ∅, ∅) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 1 1 1 1 1 (∅, (∅, , , ),( , ∅, ∅) , ∅),( , ∅, ∅, ∅) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 305 368 352 336 352 346 310 299 289 304 13 12 12 14 16 12 17 15 18 13 Continued on next page. 3 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 90 91 92 93 94 95 96 97 98 99 4 0 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 1 2 0 0 0 0 2 0 0 0 2 0 0 0 3 0 0 0 2 0 0 1 1 0 0 0 3 1 0 2 0 0 0 3 1 0 0 0 1 3 0 0 0 1 0 0 0 1 2 0 0 1 1 0 0 1 2 0 0 1 1 0 0 1 1 0 0 1 4 0 0 1 1 0 0 1 0 1 0 0 2 0 1 0 1 0 0 0 1 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 1 1 1 1 1 (∅, ∅, , ),( , ∅, ∅, ∅) 1 1 1 1 ( , ∅, ∅),( , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ),( , , ∅, ) 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 ( , ∅, ),( , ∅, ) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 ( , ∅, ) 1 1 1 1 ( , ∅, ),( , , ∅) 1 1 1 ( , ∅) 1 1 1 1 (∅, ∅, ),( , ∅, ) 309 273 299 282 288 282 266 249 216 480 18 15 17 14 8 18 16 17 16 16 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 3 3 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 100 101 102 103 104 105 106 107 108 109 3 0 0 0 1 1 0 0 0 3 0 0 0 0 4 0 0 0 3 0 1 0 3 0 0 0 1 3 0 0 0 0 3 0 0 0 0 2 0 0 1 0 2 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 0 1 0 0 1 0 0 0 0 3 0 0 0 3 0 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 1 2 1 0 0 4 0 0 5 0 0 1 0 0 1 0 0 0 0 3 0 2 0 0 1 3 0 0 0 0 0 0 0 2 1 0 0 0 1 5 0 0 ) 1 1 1 1 1 2 2 1 1 2 (∅, (cid:16)∅, (cid:16)∅, , (cid:17) (cid:17) ),( , ∅, (cid:16) , ∅(cid:17) (cid:17) (cid:16)∅, , , ,( , ∅) 1 1 1 1 1 (∅, , ∅, ),( , ∅, ∅, ) 1 1 1 1 2 (∅, , , ),( , ∅, , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 1 1 (∅, , , ),( , ∅, , ) , ∅),( , ∅, ∅, ∅) 1 1 1 1 1 1 1 2 (∅, (cid:16)∅, , (cid:17) (cid:16)∅, , (cid:17) (cid:16)∅, , (cid:17) 384 352 320 362 305 352 304 304 283 272 13 9 10 17 18 16 14 17 18 8 Continued on next page. 3 4 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 110 111 112 113 114 115 116 117 118 119 120 121 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 1 0 1 2 0 1 3 0 1 1 0 0 1 1 0 1 0 0 0 1 1 0 0 1 0 0 0 2 2 0 1 1 0 0 1 1 0 0 1 0 1 0 3 6 0 0 5 0 0 4 0 0 1 1 0 0 4 0 0 1 0 0 1 4 0 0 0 0 1 0 3 5 0 0 2 1 0 0 2 0 0 1 2 0 0 0 5 0 0 0 4 0 0 0 2 0 1 0 3 0 0 0 4 0 0 0 4 0 0 0 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 1 1 (∅, ),(∅, ) (∅, ),( , ) 1 1 1 ( , ),( , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 2 1 1 2 2 ( , (cid:16)∅, ) , ∅(cid:17) (cid:16)∅, , , ∅(cid:17) ,( , ∅, ∅) 1 1 1 1 (∅, , ),( , , ∅) 1 1 1 1 (∅, , ∅),(∅, , ∅) 1 1 1 (∅, ),( , ∅) 1 1 1 1 (∅, ),(∅, , , ) 1 1 1 1 (∅, , ∅),( , ∅, ) 256 320 304 272 282 272 257 256 256 260 229 230 17 0 12 15 16 10 15 18 15 16 19 20 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 3 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 122 123 124 125 126 127 128 129 130 131 132 133 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 3 0 1 5 0 0 0 0 0 1 1 0 1 1 0 1 2 0 1 1 0 0 1 1 0 0 1 2 0 1 0 0 6 1 0 0 1 2 0 0 1 4 0 0 4 0 0 0 0 0 2 4 0 0 5 0 0 4 0 0 1 1 0 0 1 0 0 1 4 0 0 4 0 0 1 0 0 1 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 1 1 1 1 (∅, , ),( , , ∅) 1 1 1 ( , ),( 1 1 2 ( , ∅),( , , ∅) (cid:16) ), (cid:17) , 1 1 1 2 1 1 2 1 1 1 (∅, (cid:16)∅, , ∅),(∅, (cid:17) (cid:16) , , , ∅),(∅, , ∅),( , ∅, ∅) (cid:17) (∅, ) 1 1 1 ( , ) 1 1 1 1 (∅, , ) 1 1 1 1 ( , , ) 1 1 1 ( , ) 1 1 1 ( , ),( , ∅),( , ∅),( , ∅) 1 1 1 1 213 196 240 211 224 240 224 202 180 163 192 464 19 16 13 15 13 -12 20 24 32 31 18 12 Continued on next page. 3 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 134 135 136 137 138 139 140 141 142 143 144 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 3 5 0 0 3 0 0 1 0 0 1 0 0 4 0 0 4 0 0 0 3 3 0 0 0 0 0 0 2 3 0 0 0 0 2 0 0 0 0 1 0 0 1 4 0 0 3 1 0 2 0 0 1 4 0 0 4 0 0 0 1 0 2 3 1 0 0 0 0 0 3 3 0 0 0 0 3 0 0 0 0 1 0 0 2 3 0 0 0 3 0 0 0 0 0 1 0 1 2 0 1 0 0 2 0 0 0 0 3 0 0 0 0 4 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 (∅, ∅, (cid:16)∅, ),(∅, , ∅, ∅) , ,( , ∅),( , ) (cid:17) (∅, ) 1 1 1 1 (∅, , ∅) 1 1 1 1 1 1 1 1 1 1 1 1 2 ( , ∅),( , ∅) (∅, ),( , ) ),( , ∅, ∅) (∅, (cid:16)∅, , (cid:17) (cid:16)∅, , (cid:17) 1 1 1 1 1 (∅, ∅, ∅, ),( , , ∅, ) 1 1 1 2 2 (∅, ∅, , ),( , ∅, ∅, ) 448 384 384 352 320 320 432 400 352 400 369 12 16 8 10 16 12 12 12 10 16 17 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 3 7 3 8 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 145 146 147 148 149 150 151 152 153 154 3 0 0 0 0 2 0 0 0 0 1 0 1 0 0 3 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 2 0 0 0 0 3 0 0 0 1 1 0 0 0 2 0 0 0 0 2 0 0 0 0 3 0 0 1 1 1 0 0 1 0 0 0 1 0 1 0 1 0 0 0 0 3 0 0 0 1 2 3 0 0 0 0 0 0 0 3 0 0 0 0 3 0 1 0 1 0 3 0 0 0 3 0 0 0 0 4 0 0 4 0 0 0 0 0 1 0 1 0 1 0 0 1 1 0 1 0 0 4 0 0 0 0 0 1 0 1 3 0 0 0 0 4 0 0 0 2 0 1 0 4 0 0 0 1 1 2 2 (∅, , ∅),(∅, , ∅) 1 1 1 1 1 (∅, ∅, ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, , ∅, ),(∅, , ∅, ∅) 1 1 1 1 1 ( , ∅, ∅, ∅) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 ( , ∅, ) 1 1 1 2 ( , , ), (cid:16)∅, ∅, (cid:17) ,( , ∅, ∅) 353 368 367 351 352 352 331 326 319 320 14 16 15 15 13 16 17 16 13 12 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 155 156 157 158 159 160 161 162 163 164 2 0 0 0 0 0 0 0 1 0 0 2 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 0 0 0 1 0 0 0 3 1 0 1 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 0 2 0 0 0 3 0 0 0 3 0 0 0 0 0 3 1 0 0 0 1 0 1 0 1 5 0 0 0 1 0 1 0 1 0 2 1 0 0 1 0 0 0 0 2 0 1 0 2 0 1 0 2 0 0 1 0 0 1 1 0 1 0 0 4 0 0 0 4 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 0 3 0 0 0 3 0 0 0 1 1 1 1 1 1 (∅, ∅, ∅, , ),(∅, ∅, ∅, , ∅),( , ∅, , ∅, ) 310 21 1 1 1 1 1 ( , ∅, , ∅) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 2 2 ( , ) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 1 (∅, ∅, ),( , ∅, ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 (∅, ∅, ),( , , ) 303 304 274 288 256 416 384 378 368 17 16 16 15 12 16 16 15 12 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 3 9 4 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 165 166 167 168 169 170 171 172 173 174 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 4 2 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 0 3 0 0 0 0 2 0 0 0 0 2 3 0 0 0 0 0 0 0 3 0 0 0 3 1 1 0 0 3 0 0 0 0 0 0 2 0 0 0 0 3 0 0 0 1 1 0 0 0 0 3 0 0 0 0 3 1 0 0 1 0 1 0 0 2 3 0 0 0 3 0 0 0 4 0 0 0 0 1 0 1 0 0 0 1 0 0 1 0 0 1 0 1 0 0 1 0 1 2 1 0 0 1 0 1 0 0 0 0 0 2 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 2 0 0 0 0 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 ( , ∅, ),( , , ) 1 1 1 ( , ∅) 1 1 1 1 1 (∅, ∅, ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, ∅, ),( , , ∅, ) 1 1 1 1 1 (∅, ∅, , ∅),( , , ∅, ) 1 1 1 1 1 (∅, ∅, , ∅),( , ∅, ∅, ∅) (cid:16)∅, , ∅(cid:17) (cid:16)∅, , , ∅(cid:17) 1 1 2 2 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 336 335 272 352 335 336 320 314 298 309 12 13 16 16 16 16 17 16 13 19 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 175 176 177 178 179 180 181 182 183 184 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 0 2 0 0 0 1 0 0 0 1 1 0 0 0 0 2 0 0 0 3 0 0 3 0 0 0 0 0 2 0 0 1 1 0 0 0 4 3 0 0 0 1 0 0 0 0 1 4 0 0 0 3 0 0 0 0 0 0 0 0 3 0 1 0 2 5 0 0 0 0 0 0 0 2 3 0 0 0 3 0 0 0 3 0 1 0 1 0 0 0 1 0 4 0 0 0 1 0 1 0 0 0 1 0 0 2 3 0 0 0 2 0 1 0 3 0 0 0 0 3 0 0 0 2 0 0 0 0 0 1 0 1 0 0 0 3 0 0 0 0 2 0 0 0 0 2 0 0 1 0 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 1 ( , ∅, , ∅, ) (cid:16)∅, ∅, (cid:17) ,(∅, , ∅),(∅, , ∅),( , , ∅) 1 1 1 2 1 1 1 1 1 (∅, , ∅, ),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 2 (∅, , ),( , , ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ∅) 1 1 1 1 ( , ∅, ) 288 304 325 304 304 304 304 273 293 272 15 14 20 12 16 17 16 15 16 13 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 4 1 4 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 185 186 187 188 189 190 191 192 193 194 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 1 1 0 0 0 0 3 0 0 0 1 0 0 0 0 3 0 0 0 1 3 0 0 0 0 0 0 0 3 0 0 0 3 0 0 0 0 2 0 0 0 1 1 3 0 0 0 0 0 0 1 1 4 0 0 0 0 1 0 0 1 3 0 0 0 3 0 0 0 0 1 0 0 1 0 0 0 4 1 0 0 0 1 1 0 0 0 1 3 0 1 0 0 1 0 0 1 4 0 0 0 1 0 0 0 1 3 0 1 0 1 1 1 0 0 1 0 0 1 4 0 0 0 2 0 0 0 1 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 1 1 ( , ∅, ∅, ),( , , ∅, ) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, ∅, ∅) (cid:16)∅, ∅, (cid:17) ,(∅, , ∅),(∅, (cid:16)∅, , ∅), (cid:17) , 1 1 1 2 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) (cid:16)∅, ∅, (cid:16)∅, ∅, (cid:17) (cid:17) 1 1 1 2 1 1 1 2 (cid:16)∅, ∅, (cid:16)∅, ∅, , , ,( , ∅, ∅) (cid:17) (cid:17) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 1 1 1 1 (∅, ),(∅, , ∅),( , ∅, ∅),( , , , ) 1 1 1 1 1 (∅, ∅, , ∅),( , , ∅, ) 277 252 277 272 262 256 241 256 256 252 17 16 17 12 19 14 17 18 14 22 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector (cid:16) (cid:16) (cid:17) , , , (cid:17) Generalized partitions ,(∅, ) 1 1 2 1 1 1 1 (∅, , ∅),(∅, , ∅) 1 1 1 1 ( , ∅, ) 1 1 1 1 1 (∅, ∅, ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ),( , , ∅, ) Degree Euler Number 288 304 255 330 288 10 12 19 20 17 195 196 197 198 199 200 201 202 203 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 2 1 0 0 1 0 0 0 0 2 0 0 0 0 2 0 0 0 0 1 1 0 0 0 0 1 0 0 0 1 0 0 0 0 3 0 5 0 0 0 0 3 3 0 0 0 1 0 0 0 1 1 0 0 0 1 2 0 0 0 0 0 0 0 0 0 1 4 0 0 0 0 1 0 0 2 4 0 0 0 1 0 2 0 1 0 1 0 0 3 0 0 0 0 2 0 0 0 0 0 0 0 3 0 0 1 0 1 0 1 1 0 0 0 1 0 1 0 0 0 4 0 0 0 0 2 0 0 1 0 1 0 0 1 0 0 4 0 0 0 0 2 0 0 0 0 1 1 1 1 1 1 ( , ∅, ∅, ∅, ∅) 299 20 1 1 1 1 2 (∅, , , ),(∅, , ∅, ∅),( , ∅, ∅, ∅) (cid:16)∅, ∅, (cid:17) (cid:16)∅, , , ∅(cid:17) ,( , ∅, ∅) 1 1 2 2 1 1 1 1 1 (∅, , ∅, ),(∅, , , ) 278 267 271 16 13 19 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 4 3 Continued from previous page. Period ID Adjacency matrix Dimension vector (cid:16)∅, , ∅, (cid:17) Generalized partitions Degree Euler Number ,( , ∅, , ) 257 19 1 1 1 1 2 204 205 206 207 208 209 210 211 212 0 0 0 3 0 0 2 0 0 1 0 2 0 0 1 0 0 2 0 0 1 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 1 1 0 0 0 0 3 0 0 0 1 2 0 0 0 0 2 0 0 0 1 1 0 0 1 2 0 0 0 0 1 0 0 0 1 0 1 0 0 0 1 2 0 0 0 0 0 0 0 0 3 2 0 0 0 0 1 0 0 0 1 0 0 0 1 3 0 0 0 3 0 0 0 0 2 0 0 0 0 0 3 0 0 0 0 2 0 1 0 0 0 1 1 0 1 1 0 1 0 0 2 0 1 0 0 5 0 0 0 2 0 1 0 1 1 1 1 1 1 (∅, ∅, , ∅, ),(∅, ∅, , , ),( , , ∅, ∅, ) 283 21 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 1 1 1 1 1 1 (∅, ∅, (∅, ∅, , , ),( , ∅, ∅, ∅) ),(∅, ∅, , ),(∅, , ∅, ∅),( , ∅, ∅, ∅) 1 1 1 1 1 ( , ∅, ∅, ∅),( , ∅, ∅, ∅) 1 1 1 1 1 1 1 1 2 (∅, ∅, (cid:16)∅, ∅, ),( , ∅, , (cid:17) (cid:16)∅, ∅, , , (cid:17) ) ,(∅, , ∅),( , ∅, ∅) 1 1 1 2 ( , ∅, ) 277 262 257 257 256 241 235 18 18 17 17 18 15 15 Continued on next page. 4 4 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 213 214 215 216 217 218 219 220 221 222 223 3 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 1 1 0 0 1 2 0 0 0 0 3 0 0 1 0 0 2 2 3 0 1 3 0 0 0 0 0 2 0 1 0 2 0 1 0 2 1 0 0 0 1 2 0 0 1 3 0 0 0 0 0 0 1 2 5 0 0 3 0 0 0 4 0 0 4 1 0 0 1 0 4 4 0 0 0 4 0 0 0 2 0 0 0 1 3 0 0 0 4 0 0 0 0 1 0 0 1 3 0 0 0 2 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 (∅, , ∅),(∅, , ∅) 1 1 1 1 1 (∅, , ∅),( , ∅, , , ) 1 1 1 1 ( , , ∅) 1 1 1 1 (∅, ),( , ∅, ∅),( , ∅, ∅) , 1 1 1 1 1 1 1 3 (∅, (cid:16)∅, , ∅, ),( , (cid:17) (cid:16)∅, , , ∅, (cid:16)∅, , (cid:17) ) (cid:17) ,( , ∅) 1 1 1 1 ( , ∅, ∅),( , ∅, ∅),( , ∅, ∅) 1 1 1 1 1 1 (∅, ),( , ∅) (∅, ),( , ) 1 1 1 1 (∅, , ∅),(∅, , ) 256 240 235 219 225 229 224 195 288 240 256 18 16 19 19 16 21 9 18 16 16 16 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 4 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions (cid:16)∅, , ∅(cid:17) (cid:16)∅, , , ∅(cid:17) ,( , ∅, ) 1 1 2 2 224 225 226 227 228 229 230 231 232 233 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 1 0 0 1 3 0 0 0 0 0 0 0 1 0 0 1 1 0 0 0 0 2 5 0 0 3 0 0 0 0 0 0 1 0 0 0 0 2 1 0 0 1 3 0 0 0 5 0 0 0 0 0 0 0 1 4 0 0 0 0 0 0 0 4 4 0 0 5 0 0 0 2 1 0 0 0 0 0 1 1 4 0 0 0 3 0 1 0 2 1 0 0 0 0 3 0 0 4 0 0 0 0 0 1 0 2 2 0 1 0 4 0 0 0 0 1 0 0 2 4 0 0 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 1 (∅, ∅, ),( , , ∅) 1 1 1 2 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 2 1 1 1 2 (∅, (∅, (∅, , , , ),(∅, , ) (cid:16)∅, , ∅), , , (cid:17) ,(∅, , ∅, ∅),( , ∅, ∅, ∅) ),(∅, ),( , ∅, ∅) , (∅, ∅, ),(∅, , ∅, ∅),(∅, , ∅, ∅),( , ∅, , , ) (cid:16)∅, (cid:17) ,( , ),( , ∅) (cid:16) (cid:17) , ∅, ),( , ∅, ∅), , (∅, (cid:16)∅, ∅, (cid:17) ,(∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 4 6 E L A N A K A L A S H N K O V I Degree Euler Number 251 230 225 236 220 241 240 225 230 235 14 18 18 19 18 18 8 17 14 19 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 234 235 236 237 238 239 240 241 242 243 3 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 0 0 1 0 0 0 2 2 0 0 1 5 0 0 1 0 0 1 1 0 0 1 0 0 0 0 2 1 0 1 0 2 0 0 1 0 0 0 1 1 3 0 0 0 0 0 0 3 2 0 0 1 4 0 0 4 0 0 0 3 0 0 0 0 0 0 1 1 5 0 0 0 2 1 0 0 3 0 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 2 0 1 0 3 0 1 0 0 1 0 0 3 4 0 0 0 3 0 0 0 1 1 1 1 1 1 1 2 2 (∅, (cid:16)∅, , ∅, ∅),( , ∅, , ∅) (cid:17) , ,( , , ∅) 1 1 1 1 (∅, , ∅),( , , ) 1 1 1 1 1 1 2 (∅, (cid:16)∅, (cid:17) , ∅),( , ∅, ∅) ,( , ∅),( , ) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , , ) (cid:16)∅, , (cid:16) (cid:17) , , ∅, (cid:17) 1 1 1 3 1 1 1 1 (∅, , ),( , , ∅) 215 199 208 200 192 266 224 235 224 203 21 16 18 24 0 20 19 20 16 20 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 4 7 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions (cid:16)∅, ∅, (cid:17) (cid:16)∅, , , (cid:17) 1 1 1 2 244 245 246 247 248 249 250 251 252 253 254 255 4 0 0 0 4 0 0 0 5 0 0 0 5 0 0 0 2 0 1 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 2 3 0 0 0 0 7 0 0 0 2 0 0 5 0 0 0 4 1 0 0 1 0 0 0 4 5 0 0 0 0 3 0 0 4 1 1 0 0 0 0 0 3 3 1 0 3 0 0 0 0 0 2 5 0 0 1 0 0 1 3 0 0 0 0 1 0 2 3 1 0 5 0 0 4 0 0 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 ( , ) 1 1 1 ( , ),( , ∅),( , ∅),( , ) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 (∅, ),( , ),( , ∅),( , ∅) 1 1 1 1 (∅, ),( , ∅, ),( , ∅, ∅),( , ∅, ∅) , 1 1 1 1 ( , , ) , ∅),( , ∅, ∅),( , ∅, ∅) (cid:17) 1 1 1 1 1 1 2 1 1 1 1 1 1 (∅, (cid:16)∅, , (cid:17) ),(∅, (cid:16)∅, , ( , ∅),( , ( , ∅),( , ) ) ,( , ) 4 8 E L A N A K A L A S H N K O V I Degree Euler Number 203 208 192 208 208 176 193 176 177 177 160 144 17 16 30 16 20 16 19 33 21 17 30 23 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 256 257 258 259 260 261 262 263 264 265 0 1 1 1 0 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 2 0 0 3 0 0 4 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 0 3 3 0 0 0 3 0 0 0 0 0 0 0 4 5 0 0 4 0 0 0 0 0 3 0 0 0 0 2 0 0 0 0 3 1 0 0 1 1 0 0 1 0 1 0 1 3 0 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 0 0 0 1 0 1 4 0 0 0 4 0 0 0 4 0 0 0 1 1 1 1 2 (∅, ∅, ∅, ),(∅, ∅, ∅, ) 1 1 1 1 (∅, ),(∅, , ∅) , 1 1 1 (∅, ),( , ∅) 1 1 1 ( , ),( , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 (∅, ∅, ),( , ∅, ∅, ∅) , 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 1 2 2 (∅, (cid:16)∅, ∅, (cid:17) ),( , ∅, ∅) , (cid:16)∅, ,(∅, , ∅), 385 384 320 256 384 336 320 304 288 273 17 12 10 16 16 16 16 18 13 14 Continued on next page. , ∅(cid:17) - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 4 9 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 266 267 268 269 270 271 272 273 274 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 2 0 0 0 2 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 2 1 1 0 0 0 0 0 0 3 0 0 0 2 1 0 0 0 1 0 0 0 0 1 1 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 3 4 0 0 0 0 0 1 0 1 5 0 0 0 2 0 0 0 0 0 3 0 0 0 0 1 0 1 0 0 0 1 0 0 1 1 0 1 0 0 4 0 0 0 0 0 0 0 3 0 3 0 0 0 0 3 0 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 1 1 (∅, , , ),( , ∅, ∅, ∅) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 1 1 (∅, ∅, ∅, , ),(∅, ∅, , ∅),( , , ∅, ∅, , ) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅) 1 1 1 1 1 (∅, , , ),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 250 336 320 320 304 293 288 282 288 19 20 12 20 16 18 16 18 16 Continued on next page. 5 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 275 276 277 278 279 280 281 282 283 284 0 0 0 0 3 0 3 0 0 0 0 0 2 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 0 0 1 0 0 0 0 0 2 0 0 0 0 2 0 0 0 3 2 0 1 0 0 0 0 3 0 0 4 0 0 0 0 3 0 0 0 2 0 1 0 1 2 1 0 0 1 0 0 0 1 0 0 0 0 0 3 0 1 0 2 4 0 0 0 1 0 0 1 4 0 0 0 0 1 0 2 0 0 0 3 5 0 0 0 3 0 0 0 2 0 0 0 0 0 0 1 0 0 2 3 0 0 0 1 0 0 0 1 2 0 1 0 3 0 0 0 4 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 (∅, , ∅) 1 1 1 1 1 1 (∅, ∅, ∅, ∅, ),(∅, ∅, , , ),( , , ∅, ∅, ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , , ) 1 1 1 1 1 2 2 1 1 1 1 1 1 1 1 2 (∅, (cid:16)∅, (∅, (cid:16)∅, , ∅),( , , ) (cid:17) ,( , ∅) , , ),( , ∅, ∅, ∅) (cid:17) , ,( , , ∅),( , , ) 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 272 286 278 272 256 228 256 225 224 324 14 16 21 17 14 20 18 19 19 18 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 5 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 285 286 287 288 289 290 291 292 293 2 0 0 0 0 0 0 0 0 0 3 0 2 0 0 0 0 0 1 0 0 1 0 3 0 0 0 0 4 0 0 0 0 3 0 0 0 0 3 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 3 0 0 0 0 0 1 1 0 0 0 0 2 0 0 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 0 1 1 0 0 0 0 0 2 0 1 0 1 0 5 0 0 2 0 0 0 0 0 0 0 0 3 1 0 0 0 1 0 0 0 0 0 4 0 1 0 1 0 1 0 0 0 1 4 0 0 0 0 3 0 0 0 0 1 0 1 0 0 0 0 1 0 1 2 0 0 0 0 0 1 0 1 0 0 4 0 0 0 1 1 1 1 1 1 (∅, , ∅, ∅, ∅) 293 21 1 1 1 1 1 (∅, ∅, , ),(∅, , , ) 1 1 1 1 1 1 1 2 (∅, ∅, ),(∅, , ∅, ∅) , (cid:16) (cid:17) , ,( , ∅), (cid:16)∅, (cid:17) (cid:16)∅, , (cid:17) 1 1 1 1 1 (∅, ∅, ∅, ),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 288 267 240 278 272 16 17 13 20 16 1 1 1 1 1 1 (∅, ∅, , , ∅),(∅, , ∅, , ),( , ∅, ∅, , ∅) 272 21 1 1 1 1 1 (∅, , , ),( , , ∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 261 260 20 16 Continued on next page. 5 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 241 244 246 246 240 239 14 13 18 17 19 13 22 18 18 16 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 5 3 1 1 1 1 1 (∅, ∅, ∅, ),(∅, ∅, , ),(∅, , ∅, ∅),( , ∅, , ) 236 1 1 1 1 (∅, , ∅),( , ∅, ∅),( , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅),( , ∅, ∅, ∅),( , ∅, , ) (cid:16)∅, (cid:17) (cid:16)∅, , , (cid:17) , ,(∅, ∅, ) 1 1 1 2 226 230 209 (cid:16)∅, ∅, (cid:17) (cid:16)∅, , , ∅(cid:17) ,( , , ∅) 1 1 2 2 1 1 1 2 ( , ∅, ∅),( , ∅, ∅),( , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅) (cid:16)∅, ∅, ∅, (cid:17) ,(∅, , ∅, ∅),( , , ∅) , 1 1 1 1 2 1 1 1 1 1 (∅, , ∅),( , ∅, , , ) 1 1 1 1 (∅, , ∅),(∅, , ∅) 294 295 296 297 298 299 300 301 302 303 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 1 0 0 0 0 2 0 0 0 3 0 0 0 0 2 0 0 1 2 0 0 0 0 3 0 0 3 0 1 0 0 1 3 0 0 0 0 1 0 0 1 0 0 0 0 1 0 0 0 0 2 0 1 0 2 0 0 0 0 3 4 0 0 0 0 0 0 0 3 3 0 0 0 4 0 0 0 2 0 1 0 1 0 0 0 1 0 0 2 0 0 3 0 0 0 0 3 0 0 0 0 0 1 0 2 3 0 0 0 0 0 1 0 2 0 1 3 0 4 0 0 0 0 4 0 0 0 0 3 0 0 0 0 4 0 0 0 0 3 0 0 0 0 5 4 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 304 305 306 307 308 309 310 311 312 313 2 0 0 0 0 3 0 0 0 0 2 0 0 0 0 0 3 0 0 0 0 3 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 1 1 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 5 0 0 0 2 0 0 0 3 0 0 0 1 1 0 0 0 0 3 2 1 0 0 3 0 0 0 0 0 0 1 1 0 0 0 0 1 1 1 0 0 0 1 5 0 0 0 0 0 4 4 0 0 0 0 1 0 0 3 1 1 0 0 0 3 0 0 0 5 0 0 0 1 0 0 0 0 1 3 0 0 0 0 2 0 1 0 0 3 0 0 0 1 1 1 1 (∅, , ),( , ∅, ) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ),( , ∅, ∅) 1 1 1 1 1 1 ( , , ∅, ∅, ) 1 1 1 1 1 (∅, ∅, , ),( , , ∅, ) 1 1 1 1 1 (∅, ∅, , ∅),( , ∅, ∅, ∅) 1 1 1 (∅, ),( , ),( , ∅),( , ) 1 1 1 1 (∅, ),(∅, , , ) 256 246 245 238 220 251 230 250 224 240 8 18 19 22 16 21 18 22 16 12 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 230 218 208 214 1 1 1 2 2 ( , (cid:16)∅, ∅, ∅, , ∅, ∅), (cid:17) (cid:16)∅, ∅, ∅, , (cid:17) ,( , ∅, ∅, ∅) 209 18 14 20 20 15 20 20 18 9 22 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 5 5 209 193 202 198 202 314 315 316 317 318 319 320 321 322 323 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 0 0 1 0 0 0 0 2 3 0 0 0 0 0 0 1 4 0 0 0 0 0 0 0 3 0 0 0 1 1 3 0 0 0 0 0 0 3 0 1 0 0 2 0 2 0 0 0 0 0 0 0 3 5 0 0 0 3 0 0 0 4 0 0 0 0 1 0 0 1 3 0 0 0 0 3 0 1 0 4 0 0 0 1 0 0 0 1 3 0 0 0 0 0 0 2 0 1 2 1 0 0 2 0 1 0 3 0 0 0 0 1 0 0 2 3 0 0 0 0 3 0 0 0 0 2 0 0 1 0 2 0 0 0 0 2 0 0 0 0 1 1 1 1 1 (∅, ∅, , ∅),( , ∅, , ) , (cid:17) (cid:16)∅, ), , 1 1 1 2 1 1 1 2 (∅, , ∅),(∅, (cid:16) , ∅, (cid:17) 1 1 1 2 (∅, , ∅),( , ∅, ∅),( , , ) 1 1 1 1 1 (∅, ∅, ),(∅, , , , ) 5 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 324 325 326 327 328 329 330 331 332 333 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 1 0 0 1 2 0 0 0 0 0 3 0 0 0 0 0 2 0 0 0 0 2 0 0 0 4 0 0 1 1 0 0 0 0 2 0 0 0 0 2 3 0 0 0 1 1 0 1 0 1 0 1 1 0 0 1 0 0 0 0 4 0 0 0 0 3 0 1 0 2 3 0 0 0 0 0 0 0 3 0 0 0 0 3 3 0 0 0 3 0 0 0 0 4 0 0 4 0 0 0 0 1 1 0 0 0 1 1 0 0 4 0 0 0 4 0 0 0 0 0 1 0 1 0 0 1 0 1 2 0 0 0 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 1 ( , ∅, ∅),( , , ∅) 1 1 1 1 1 1 2 2 (∅, (cid:16) , , ∅) , ∅(cid:17) (cid:16) , , ∅(cid:17) (cid:16) , , , ∅, (cid:17) 1 1 1 2 (∅, , ),( , , ) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 1 1 1 1 (∅, ∅, ),(∅, , ),( , ∅, ∅),( , , ) 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),( , , ∅, ) 178 186 209 204 214 209 204 188 194 188 20 20 17 15 20 19 20 19 19 22 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 334 335 336 337 338 339 340 341 342 343 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 2 0 0 1 0 0 0 4 0 0 1 1 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 1 0 0 0 1 0 0 1 1 2 1 0 0 2 1 0 0 1 0 0 1 3 0 0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 1 2 0 0 0 1 0 0 0 1 3 0 0 0 4 0 0 0 3 0 0 0 5 0 0 0 4 0 0 0 4 0 0 0 0 3 0 0 0 0 0 1 0 0 2 4 0 0 0 4 0 0 0 4 0 0 0 1 0 0 1 0 1 0 0 1 0 2 0 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 (∅, ∅, ),( , ∅, ),( , ∅, ∅),( , , ) 1 1 1 1 (∅, , ),( , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 2 2 ( , , ∅, ∅),( , ∅, ∅, ∅),( , ∅, ∅, ∅),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ) 1 1 1 2 1 1 1 2 1 1 1 2 (∅, , ∅), (cid:16)∅, ∅, (cid:16)∅, ∅, (cid:17) , ∅, (cid:16) (cid:17) (cid:17) , ,( , (cid:16)∅, , ∅) (cid:17) , ,( , ∅, ∅),( , ∅, ∅) 244 208 198 173 199 199 214 178 188 183 24 22 18 23 15 13 24 19 16 17 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 5 7 Continued from previous page. Period ID Adjacency matrix Dimension vector 1 1 2 2 Generalized partitions Degree Euler Number (cid:16) (cid:17) , , ,(∅, , ∅) 344 345 346 347 348 349 350 351 352 353 354 3 0 0 0 4 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 4 0 0 0 0 3 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 2 5 0 0 0 1 0 0 1 2 0 0 1 1 0 0 1 4 0 0 0 0 0 0 2 0 0 0 4 0 0 0 4 0 0 3 0 4 0 0 1 0 0 2 0 3 0 0 3 1 0 0 2 0 0 1 3 0 0 1 2 1 0 0 0 0 0 2 0 1 0 3 0 1 0 3 5 0 0 1 0 0 1 0 1 1 1 1 (∅, , ),( , , ) 1 1 2 2 (∅, , ) 1 1 1 1 (∅, ),(∅, , , ∅) 1 1 1 1 ( , , ∅),( , , ∅) 1 1 1 1 1 2 3 (∅, (cid:16)∅, , ∅),( , , ∅) (cid:16)∅, , (cid:17) (cid:17) (cid:17) (cid:16)∅, , 1 1 1 1 1 (∅, , , ),( , ∅, , ∅) 1 1 1 1 1 1 1 1 1 1 1 , , (∅, (∅, (∅, ),(∅, , ∅),( , ∅, ∅),( , , ) ),(∅, , ),( , ∅, ∅),( , ∅, ∅) ),( , ) 5 8 E L A N A K A L A S H N K O V I 178 177 150 176 166 166 177 187 172 178 176 17 21 18 20 22 20 18 24 20 22 16 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 355 356 357 358 359 360 361 362 363 364 365 366 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 1 0 0 1 0 0 0 3 0 0 0 3 0 0 0 2 0 0 0 3 4 0 0 0 0 4 0 0 4 3 0 0 6 0 0 0 0 3 4 0 0 2 1 0 0 1 0 0 1 1 0 0 1 1 0 0 2 1 0 0 1 3 1 0 4 0 0 4 0 0 0 5 0 0 1 0 1 3 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 1 1 1 ( , ),( , ) 1 1 1 1 (∅, , ) 1 1 1 1 (∅, , ),( , ∅, ) 1 1 1 1 ( , ∅, ),( , , ) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 2 1 2 ( , ∅, ∅),( , (cid:16)∅, (cid:17) (cid:16)∅, , , ∅) (cid:17) (cid:16) , (cid:17) , ( , ),( , ∅) (∅, (cid:16) (cid:16) (cid:17) , , , ),( (cid:17) (cid:16) (cid:16) (cid:17) (cid:16) (cid:17) , , , (cid:17) , ,( , ∅) ) (cid:17) (cid:16) (cid:16) (cid:17) , , 1 1 1 1 (∅, ∅, ),( , , ) 160 149 162 151 160 145 129 112 384 320 224 336 32 34 19 25 24 33 31 52 16 8 12 20 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 5 9 6 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 367 368 369 370 371 372 373 374 375 376 3 0 0 0 0 1 0 0 1 0 3 0 0 0 0 2 0 0 0 0 0 0 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 3 0 0 0 0 2 0 0 0 0 3 0 0 0 0 0 2 3 0 0 0 0 0 0 3 7 0 0 0 0 0 3 0 0 0 4 0 0 0 0 3 1 1 0 0 0 0 0 0 2 0 1 0 0 1 1 0 0 0 1 0 1 0 0 1 0 1 0 2 2 1 0 2 1 0 0 1 1 0 0 1 0 1 0 0 4 0 0 0 4 0 0 0 0 1 0 0 0 1 2 0 0 0 0 0 4 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ),( , , ∅, ) 1 1 1 1 1 1 (∅, ∅, ∅, ),(∅, ∅, , , , ),(∅, , ∅, , ) (cid:16)∅, ∅, (cid:17) (cid:16)∅, , , ∅(cid:17) (cid:16)∅, , , ∅(cid:17) ,( , ∅, ∅) 1 1 2 2 1 1 1 1 (∅, , ),( , ∅, ∅) 1 1 2 ( , ),( , ) 1 1 1 1 (∅, , ),( , , ) 1 1 1 1 ( , ∅, ),( , , ∅) 304 282 304 294 288 272 256 193 208 186 16 20 20 25 20 12 16 21 20 33 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 377 378 379 380 381 382 383 384 385 3 0 0 0 0 2 0 0 0 0 2 0 0 0 0 0 4 0 0 0 0 2 0 0 1 0 0 1 0 0 0 1 0 0 2 0 0 0 1 0 0 1 0 0 0 1 0 0 1 1 0 1 0 0 2 0 0 0 0 0 1 0 0 0 1 3 0 0 0 0 2 0 0 0 0 0 4 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 0 0 2 0 0 0 3 0 0 0 3 1 0 1 0 0 0 0 3 0 0 0 0 4 0 0 0 1 1 0 0 1 0 0 1 0 0 0 0 2 0 0 0 0 4 0 0 0 0 4 0 0 0 4 5 0 0 1 1 1 1 1 (∅, , ∅, ∅),(∅, , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 288 282 24 22 1 1 1 1 1 1 (∅, ∅, ∅, ∅, ) 272 20 1 1 1 1 1 (∅, ∅, ∅, ),(∅, , , ) 1 1 1 1 1 (∅, ∅, ),(∅, , ∅, ∅) , 1 1 1 1 1 1 (∅, , ∅, , ∅),(∅, , ∅, , ),(∅, , ∅, ∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),( , , ) 1 1 1 1 1 2 2 (∅, (cid:16)∅, , (cid:17) ),(∅, (cid:16)∅, , (cid:17) (cid:16) , , ∅),( , ∅, ∅),( , ∅, ∅) (cid:16) , ∅(cid:17) , , ∅(cid:17) 256 256 246 240 224 193 22 16 21 16 13 15 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 6 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 386 387 388 389 390 391 392 393 394 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 2 0 0 0 1 2 0 0 0 0 0 0 2 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 4 0 0 0 0 0 0 0 2 0 2 0 0 5 1 0 0 0 0 0 3 1 0 0 0 1 0 0 0 0 0 3 0 1 0 0 1 0 1 0 0 1 4 0 0 0 3 0 0 0 0 5 0 0 0 0 0 0 0 4 2 0 0 0 0 0 2 0 1 0 0 2 0 0 0 1 2 0 0 0 1 2 0 0 1 0 3 0 0 0 0 2 0 0 0 0 0 2 0 0 0 0 3 0 0 0 0 3 0 0 0 0 0 0 0 0 4 0 1 1 1 1 (∅, ∅, ),(∅, , ∅),( , ∅, ∅) 1 1 1 1 1 1 1 1 2 (∅, ∅, ∅, ),(∅, (cid:16)∅, ∅, (cid:17) (cid:16)∅, ∅, , , (cid:17) , ) (cid:16)∅, ∅, , (cid:17) ,( , ∅, ∅) 1 1 1 (∅, ),(∅, ) 1 1 1 1 1 (∅, ∅, , ),(∅, ∅, , ∅),(∅, , ∅, ∅),(∅, , , ) 252 256 240 192 240 20 20 12 20 16 1 1 1 1 1 1 (∅, ∅, , , ),(∅, , ∅, ),( , ∅, ∅, , ∅) , 230 22 1 1 1 1 1 (∅, , , ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ),(∅, , , ) 219 220 214 20 22 19 Continued on next page. 6 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 395 396 397 398 399 400 401 402 403 404 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 3 0 0 0 3 3 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 1 1 0 0 0 0 1 2 0 0 0 0 1 0 3 0 0 0 0 3 0 0 0 3 4 1 0 0 0 0 0 4 0 0 0 0 3 0 0 0 1 1 3 0 0 0 0 0 0 2 0 0 1 0 1 3 0 0 0 0 0 0 0 3 3 0 0 0 0 1 1 0 0 0 0 1 0 1 0 1 0 0 1 4 0 0 0 0 3 0 0 0 0 4 0 0 3 0 0 0 0 2 0 0 0 0 3 0 0 0 0 4 0 0 0 0 4 0 0 0 0 1 1 1 1 (∅, , ∅),(∅, , ) 208 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ),( , ∅, ∅, ∅),( , ∅, , ) 204 1 1 1 1 1 1 2 ( , ∅, ∅),( , , ) (cid:16)∅, (cid:17) ,(∅, ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ) 1 1 1 1 1 (∅, ∅, ∅, ),( , , , ) 1 1 1 1 1 (∅, ∅, ∅, ),(∅, , ∅, ∅) 1 1 1 1 (∅, ∅, ),( , , ∅) 1 1 1 1 2 ( , ∅, , ∅), (cid:16) (cid:17) , ∅, 1 1 2 2 , (cid:16)∅, ∅, ∅, , ∅(cid:17) (cid:16) , (cid:16)∅, , ∅), , , (cid:17) (cid:17) (cid:16) , ,(∅, ∅, , ∅(cid:17) , 192 164 230 224 224 204 204 193 14 18 18 20 22 12 22 20 18 17 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 6 3 6 4 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 405 406 407 408 409 410 411 412 413 414 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 3 0 0 0 6 0 0 0 0 0 0 0 2 1 0 0 1 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 1 1 0 0 0 0 2 0 0 0 3 0 2 0 1 2 1 0 1 0 0 1 1 0 2 0 0 0 0 0 0 0 3 0 0 0 1 2 0 0 0 0 3 3 0 0 0 0 0 0 0 3 4 0 0 0 0 3 0 0 2 0 0 0 0 0 4 0 1 0 1 0 1 0 0 0 1 0 0 1 0 1 1 0 1 0 0 4 0 0 0 0 0 3 0 0 0 0 3 0 2 0 0 0 0 0 3 0 0 0 0 3 0 0 0 0 2 0 0 0 0 4 0 0 0 0 1 1 1 1 1 1 2 2 (cid:16) , 1 1 2 ( , ), , , ) , ∅(cid:17) ,(∅, , ∅) (∅, ∅, ),( , (cid:16) (cid:17) , , ∅(cid:17) (cid:16) , 1 1 1 1 1 1 (∅, ∅, ∅, , ),(∅, , ∅, , ),( , , ∅, ∅, ) 1 1 1 1 (∅, ∅, ),( , ∅, ) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ∅) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , ∅, ∅) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , , ) (cid:16)∅, (cid:17) , ,(∅, ∅, ),( , , ∅) 1 1 1 2 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ∅),(∅, , , ),( , ∅, ∅, ∅) 192 183 161 209 182 197 203 188 188 188 22 16 31 23 22 24 19 21 14 20 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 415 416 417 418 419 420 421 422 423 424 3 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 2 0 0 0 0 2 0 0 0 2 1 0 1 1 0 0 0 3 0 0 0 1 0 0 0 1 0 0 0 1 0 0 3 0 0 1 0 3 0 0 0 0 3 0 0 0 0 3 0 0 0 3 3 0 0 0 0 2 0 1 0 2 0 0 4 0 0 0 0 2 0 0 3 0 0 0 4 0 0 0 0 0 1 0 1 0 0 1 0 1 4 0 0 0 3 0 0 0 3 0 0 0 5 0 0 0 4 0 0 0 5 0 0 0 3 0 1 0 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ),( , ∅, ∅) 1 1 1 1 1 (∅, , , ),( , ∅, ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),( , ∅, , ) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 1 ( , ∅, ),( , , ∅) 1 1 1 1 1 1 1 2 1 1 1 2 1 1 1 2 1 1 1 2 (∅, , ∅),(∅, , ∅) (cid:16)∅, ∅, (cid:17) (cid:16)∅, ∅, , (∅, , ∅),(∅, , ), , , (cid:16) (cid:16)∅, ∅, , (cid:16)∅, ∅, (cid:17) (cid:16)∅, ∅, (cid:17) , (cid:17) ,(∅, , ∅) , , ∅, (cid:16)∅, ∅, (cid:17) (cid:17) (cid:16) (cid:16)∅, ∅, (cid:17) (cid:17) , ,( , (cid:17) (cid:16) ,( , ∅, ∅) (cid:17) , ∅), , , 184 182 182 176 162 161 183 172 167 161 24 24 22 20 21 22 15 18 19 20 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 6 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 425 426 427 428 429 430 431 432 433 434 435 2 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 1 0 0 0 3 0 0 0 2 0 0 0 0 2 0 0 0 1 3 0 0 0 0 0 3 0 0 1 1 2 0 0 0 0 0 0 3 6 0 0 5 0 0 0 2 0 1 0 2 0 0 0 0 0 0 2 0 4 0 0 3 1 0 0 1 0 2 3 0 0 0 2 0 0 1 0 1 0 3 3 0 0 0 4 0 0 0 3 0 0 0 0 4 0 0 0 5 0 0 0 3 0 0 0 3 1 0 0 4 0 0 0 1 1 1 1 2 3 1 1 1 1 1 1 1 2 (∅, ),( , (cid:16) ) (cid:17) ( , ∅), , (∅, ∅, (cid:16) , , ),(∅, , ) ,( , ∅, ∅),( , ∅, ∅),( , ∅, ∅) (cid:17) 1 1 1 1 1 (∅, , ∅),( , ∅, ∅, ∅) , 1 1 1 2 1 2 2 ( , (cid:16)∅, , ∅), (cid:17) (cid:16) (cid:17) (cid:16)∅, ∅, , (cid:17) , , ,(∅, , ∅) ,(∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),( , ∅, ∅) 1 1 1 1 ( , , ) 1 1 1 1 1 1 1 1 (∅, (∅, , , ),(∅, , ) ),(∅, , ∅),(∅, , ∅),( , ∅, ∅) 160 147 192 163 167 162 148 156 141 161 151 30 19 4 17 24 18 20 20 33 24 22 Continued on next page. 6 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 436 437 438 439 440 441 442 443 444 445 446 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 3 0 0 3 0 5 0 0 0 0 0 3 1 0 2 0 0 2 5 0 0 0 0 2 0 0 0 2 0 0 0 2 1 0 0 1 0 1 0 2 5 0 0 0 6 0 0 0 1 0 1 5 0 0 6 0 0 0 1 0 2 1 0 0 1 1 1 0 0 5 0 0 0 3 0 0 0 0 0 3 0 4 0 0 0 4 0 0 0 5 0 0 0 5 0 0 0 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 1 (∅, , ),( , ∅, ) 1 1 1 2 1 1 4 (cid:18) (cid:19) (cid:18) (∅, ∅, ),( , ∅, ∅, , , (cid:19) , ) (cid:18) ∅, 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 ( , ),( , ) 1 1 1 1 2 ),(∅, ) (∅, (cid:16) (cid:17) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 1 (∅, ∅, ),(∅, , ) (cid:19) (cid:18) , ∅, (cid:19) ,( , ) 151 148 131 141 130 129 256 192 288 240 218 25 22 25 19 26 41 0 16 24 8 20 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 6 7 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ) 1 1 1 1 1 (∅, , , ),(∅, , ∅, ) 1 1 1 1 1 1 1 2 (∅, ∅, (cid:16)∅, ∅, (cid:17) ),(∅, ∅, ),( , ∅, ),( , , ) ,( , ∅, ∅),( , ∅, ∅),( , ∅, ∅) 1 1 1 1 (∅, , ∅),( , ∅, ∅) 1 1 1 1 1 (∅, , , ),(∅, , , ) 1 1 1 1 (∅, ),(∅, , , ) 1 1 1 1 1 (∅, ∅, ∅, ),(∅, ∅, , ∅),(∅, , , ),( , , ∅, ) 204 1 1 1 1 1 1 1 1 (∅, (∅, , , ),(∅, , ) ),( , ∅, ) 176 270 447 448 449 450 451 452 453 454 455 456 3 0 0 0 0 2 0 0 0 0 2 0 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 0 2 0 0 0 3 0 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 0 2 0 1 0 0 2 0 0 0 0 4 1 0 0 1 0 2 0 0 2 1 0 0 0 0 0 0 3 2 0 0 1 0 0 0 0 3 2 1 0 0 0 0 0 4 1 0 0 0 1 0 1 0 0 1 6 0 0 0 4 0 0 0 3 0 0 0 0 1 0 0 2 4 0 0 0 0 0 0 0 3 4 0 0 0 3 0 0 0 6 8 E L A N A K A L A S H N K O V I 252 240 188 224 180 224 208 30 20 8 12 36 16 4 22 24 21 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 457 458 459 460 461 462 463 464 465 466 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 0 0 0 3 0 0 0 0 2 3 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 1 1 0 0 1 0 0 0 0 4 0 0 0 3 0 1 0 2 0 1 0 2 0 0 0 0 3 0 2 0 1 1 0 0 0 1 0 0 0 0 0 3 2 0 0 0 5 0 0 0 0 0 1 1 0 1 0 3 4 0 0 0 3 0 0 0 2 0 1 0 0 0 3 0 0 2 0 0 0 0 0 0 0 0 0 3 5 0 0 0 2 0 0 0 0 2 0 0 0 0 0 0 0 4 0 2 0 0 0 0 2 0 0 0 0 0 4 0 0 0 0 2 0 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),( , ∅, ) 1 1 1 1 (∅, ),(∅, , , ) 1 1 1 1 1 1 1 2 2 (∅, ∅, , ∅),( , (cid:16) , ∅(cid:17) (cid:16) , , , ) , ∅, , ∅(cid:17) (cid:16) , , ∅(cid:17) , ,( , ∅, ∅) 1 1 1 1 1 1 (∅, ∅, ∅, ),(∅, ∅, , , , ),( , , ∅, , ∅) 1 1 1 1 1 (∅, ∅, , ∅),(∅, ∅, , ∅),(∅, , ∅, ∅),( , , ∅, ) 1 1 1 1 1 2 3 (∅, ∅, ),( , ∅, (cid:16)∅, (cid:17) (cid:16)∅, , (cid:17) (cid:16) , ) , ∅(cid:17) (cid:16) , , ∅(cid:17) 1 1 1 1 1 ( , , ∅, ),( , ∅, ∅, ) 1 1 1 1 (∅, ),(∅, , , ) 208 206 202 208 198 188 172 162 188 185 24 13 28 12 24 19 20 14 24 17 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 6 9 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 467 468 469 470 471 472 473 474 475 476 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 4 2 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 2 1 0 1 3 0 0 0 0 0 0 2 1 0 1 1 0 1 0 3 0 1 0 3 2 0 0 1 0 0 0 0 3 0 1 0 0 2 0 2 0 0 4 0 0 5 0 0 0 0 0 0 3 3 0 0 0 4 0 0 0 3 0 0 0 3 1 0 0 0 1 1 0 1 2 0 0 0 0 4 0 0 0 0 0 0 0 4 3 0 0 0 2 0 0 0 0 2 0 0 1 0 3 0 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),( , , 1 1 1 1 (∅, ),(∅, , ∅),(∅, , ∅),( , , ∅, ) ) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , ∅, ∅) (cid:16)∅, ∅, (cid:17) (cid:16) (cid:17) (cid:16)∅, , 1 1 1 2 1 2 2 1 2 2 (cid:17) ,(∅, , ∅),( , ∅, ∅),( , ∅, ∅) (cid:16)∅, , (cid:16) (cid:17) , , (cid:17) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ∅),(∅, , , ),( , ∅, , ) 1 1 1 1 (∅, , ),( , ∅, ∅) 172 182 172 182 172 162 146 165 162 156 22 22 24 20 22 18 20 6 22 20 Continued on next page. 7 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 477 478 479 480 481 482 483 484 485 486 2 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 2 0 0 0 1 0 1 1 0 0 0 3 0 0 0 2 0 0 0 3 0 2 0 1 0 0 0 3 0 0 0 3 0 0 0 3 1 0 0 1 1 0 1 0 1 2 0 0 0 0 1 0 4 1 0 0 1 0 0 0 3 0 4 0 0 0 1 0 4 1 1 0 1 0 1 0 1 2 0 0 0 0 0 4 0 0 4 0 0 0 3 0 0 0 5 0 0 0 2 0 0 0 0 3 0 0 0 3 0 0 0 4 0 0 0 1 1 1 1 1 ( , , ∅, ),( , , ∅, ) (cid:16) (cid:17) (cid:16) , , ∅(cid:17) (cid:16) , , (cid:16) , ∅(cid:17) , , , ∅(cid:17) , , ∅, 1 1 2 3 1 1 1 1 ( , ∅, ),( , ∅, ) 1 1 1 1 (∅, ),(∅, , ∅),(∅, , ∅),( , , , ) ),( , , ∅) 1 1 1 1 1 1 1 1 2 1 2 2 (∅, ∅, (cid:16)∅, ∅, (cid:17) (cid:16)∅, , (cid:16) , (cid:17) , ∅(cid:17) ,( , , , ) (cid:16) , (cid:16) , ∅(cid:17) , , ∅(cid:17) 1 1 1 1 (∅, ),(∅, , ∅),(∅, , , ),( , ∅, ∅) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 (∅, ∅, ),( , , ) 167 162 146 146 150 162 146 146 144 140 22 16 24 22 32 15 16 20 34 36 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 7 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 487 488 489 490 491 492 493 494 495 496 497 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 2 0 0 0 3 0 0 0 2 0 0 0 2 0 0 0 3 2 0 0 0 1 0 2 2 0 2 5 0 0 0 0 0 2 2 0 0 1 2 0 0 1 0 0 0 4 2 0 0 2 2 0 0 1 0 1 0 2 2 1 0 0 5 0 0 4 0 0 0 1 0 0 2 3 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 3 0 1 0 2 0 0 0 0 2 0 0 1 0 1 1 1 1 ( , , ∅) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 (∅, , ∅),(∅, , ∅),( , ∅, ∅),( , , ) 1 1 1 1 (∅, , ∅),( , , ) 1 1 1 1 (∅, , ),( , , ∅) 1 1 1 1 (∅, , ∅),( , , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 1 1 1 1 2 (∅, ),( , ∅) ( , (cid:16) (cid:17) ),( (cid:16) , , ∅) (cid:17) 1 1 1 1 1 (∅, ∅, ∅, ),(∅, , ∅, ) 133 142 136 130 130 125 140 128 98 160 216 35 20 22 26 34 34 24 40 52 28 36 Continued on next page. 7 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 498 499 500 501 502 503 504 505 506 507 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 2 0 0 0 1 0 0 0 2 0 0 0 0 2 0 0 0 0 2 0 0 0 3 2 0 0 1 0 0 0 0 2 0 0 0 0 2 0 0 0 3 6 0 0 4 0 0 0 0 0 0 3 0 0 0 0 2 0 1 0 0 1 0 0 0 4 4 0 0 0 3 0 0 0 0 3 0 0 0 1 3 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 0 2 0 0 0 0 4 0 0 0 3 0 0 0 0 2 0 0 0 0 3 0 0 0 0 3 0 0 1 0 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),( , ∅, , ) 1 1 1 1 ( , ∅, ∅),( , , ) 1 1 1 1 1 1 2 (∅, (cid:16)∅, ∅, (cid:17) ),( , ∅) ,(∅, , ∅),(∅, (cid:16) , ∅), , ∅, (cid:17) 1 1 1 1 (∅, , ∅),( , ∅, ) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅) , 1 1 1 1 1 (∅, ∅, ∅, ),(∅, , ∅, ∅) 1 1 1 1 (∅, ∅, ),(∅, , ),( , ∅, ∅),( , , ) 224 240 208 192 144 192 176 192 172 168 16 28 16 12 36 8 9 12 20 24 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 7 3 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number (cid:17) (cid:16) , , , (cid:17) (cid:17) (cid:16) , (cid:16) (cid:16) (cid:17) (cid:16) (cid:16) , ∅(cid:17) , , , , (cid:17) , 1 1 2 1 2 2 1 1 1 1 1 (∅, ∅, , ),( , , ∅, ) 1 1 1 1 1 1 1 2 (∅, ∅, (cid:16)∅, , ),(∅, (cid:17) (cid:16)∅, , , ) , (cid:17) ,( , , ∅),( , , ∅) 1 1 1 1 (∅, ∅, ),(∅, , ∅),( , , ),( , , ) 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , ) 1 1 1 1 1 1 1 2 2 1 1 2 (∅, (cid:16)∅, (cid:16)∅, , ),( , , ∅) , , ∅(cid:17) (cid:16)∅, (cid:17) (cid:16)∅, ,( , , , (cid:17) 1 1 1 1 (∅, , ),( , ∅, ∅) , ) , ∅(cid:17) , (cid:16)∅, , ∅(cid:17) , ,( , , ∅, ∅) 141 131 134 508 509 510 511 512 513 514 515 516 517 518 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 1 0 1 0 0 0 0 2 0 0 0 2 0 0 2 1 0 0 0 3 0 0 0 0 2 0 0 0 2 0 0 2 0 0 1 0 1 0 0 0 2 0 5 0 4 0 0 0 0 0 0 3 3 0 0 0 2 0 0 0 0 0 0 4 0 0 0 0 3 0 0 0 4 2 0 0 0 0 5 0 0 1 0 0 1 3 0 0 0 0 3 0 0 0 0 0 0 3 0 0 0 0 1 0 1 3 0 1 0 0 0 4 0 4 0 0 0 0 1 0 0 1 3 0 0 0 0 0 2 0 1 5 0 0 0 7 4 E L A N A K A L A S H N K O V I 160 130 176 176 152 152 156 144 13 31 20 4 16 22 24 24 20 18 38 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions (cid:16)∅, (cid:17) (cid:16) , , (cid:17) , , ,( , , ∅),( , ∅, ∅) 1 1 1 2 519 520 521 522 523 524 525 526 527 528 529 2 0 0 0 0 0 0 4 0 0 1 0 0 2 4 0 0 0 5 0 0 0 5 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 0 0 0 0 2 0 0 0 1 0 0 0 3 0 0 0 3 0 0 0 1 0 0 0 3 4 0 0 0 3 0 0 0 0 3 0 0 1 2 0 0 0 0 0 0 0 3 4 0 0 0 0 0 0 3 0 0 0 3 4 0 0 0 0 1 0 1 4 0 0 0 4 1 0 5 0 0 5 0 0 1 1 1 1 1 (∅, , , ),(∅, , , ) 1 1 1 2 (∅, , ∅),(∅, (cid:16)∅, , ∅), (cid:17) (cid:16) , , , ∅, (cid:17) 1 1 1 1 (∅, ∅, ),(∅, , ∅),( , ∅, ∅),( , , ) (∅, , ∅),(∅, , ∅),( , ∅, ∅),( , ∅, ∅) 1 1 1 1 1 1 1 2 (cid:16)∅, ∅, (cid:17) ,(∅, , ∅),(∅, , ∅),( , , ∅) 1 1 1 1 (∅, , ),( , ∅, ∅) , ∅), , ∅, (cid:16) (cid:16)∅, (cid:17) , (cid:17) (cid:16)∅, (cid:17) , ,( , (cid:17) , (cid:16)∅, , ) (cid:17) 1 1 1 2 1 1 2 1 1 1 1 1 2 (∅, (cid:16)∅, ( , (cid:16)∅, ),( , ) (cid:17) ,( , ∅) Degree Euler Number 141 146 136 136 131 136 124 116 114 112 110 21 25 22 20 22 8 27 25 23 68 23 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 7 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 530 531 532 533 534 535 536 537 538 539 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 0 2 0 0 0 3 0 0 0 3 0 0 0 2 0 0 0 0 2 0 0 0 0 2 5 0 0 0 0 0 3 0 0 0 0 3 2 1 0 0 1 0 0 1 1 0 0 1 1 1 0 0 0 0 0 5 0 0 0 0 2 0 0 0 0 3 4 0 0 0 0 0 1 0 1 4 0 0 0 5 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 0 0 1 0 0 1 3 0 0 0 0 3 0 0 0 0 3 0 0 0 0 1 1 1 (∅, ),( , ) 192 -12 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 (∅, , , ),( , ∅, ∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 1 1 1 (∅, (∅, , , ),( , ∅, ),( , ∅, ) ) 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),(∅, , ) 1 1 1 1 1 (∅, ∅, , ),( , ∅, , ) 1 1 1 1 1 (∅, ∅, , ∅),(∅, , ∅, ) 240 210 180 141 176 154 192 192 162 28 35 40 -3 -4 30 16 16 19 Continued on next page. 7 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 540 541 542 543 544 545 546 547 548 549 4 0 0 0 0 1 0 0 1 2 0 0 0 0 4 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 0 0 1 0 0 2 2 0 0 0 0 2 0 1 1 0 2 0 0 1 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 0 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 0 2 0 0 0 3 0 0 0 0 2 3 0 1 0 0 0 0 2 1 0 1 2 0 0 0 0 0 0 0 4 2 0 0 0 0 2 0 0 1 3 1 0 0 1 0 0 1 0 0 0 0 2 4 0 0 0 0 0 0 3 5 0 0 1 1 1 1 (∅, ∅, ),( , ∅, ∅) 1 1 1 1 1 (∅, , , ),(∅, , ∅, ) 1 1 1 1 1 (∅, ∅, ∅, ),( , ∅, ∅, ∅) 1 1 1 1 1 1 1 1 (∅, (∅, , , ),( , , ) ),(∅, , ∅) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 2 ( , ),( , ) 1 1 1 1 1 1 1 2 (∅, ∅, , ∅),(∅, (cid:16)∅, (cid:17) (cid:16)∅, , , (cid:17) (cid:16)∅, , , (cid:17) ) ,( , ∅) 156 156 146 125 134 160 155 116 146 126 24 28 24 29 34 24 24 21 23 16 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 7 7 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 550 551 552 553 554 555 556 557 558 559 3 0 0 0 3 0 0 0 0 3 0 0 0 4 0 0 0 3 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 2 5 0 0 2 0 0 0 4 0 0 0 0 0 1 0 0 0 3 0 0 0 3 1 0 1 0 0 1 0 1 0 2 0 0 0 0 2 5 0 0 2 1 0 1 4 0 0 0 2 0 0 0 1 0 2 0 2 0 2 4 0 0 5 0 0 1 1 1 1 (∅, , ),( , , ) 3 0 0 0 0 1 1 1 1 1 1 1 2 (∅, (cid:16)∅, , (cid:17) (cid:16)∅, , (cid:17) , ∅),( , ∅, , ∅) ,( , ),( , ) 1 1 1 1 (∅, ),(∅, , , ∅) 1 1 2 1 1 1 2 ) ( , (cid:16)∅, ∅, (cid:17) (cid:16)∅, , , (cid:17) 1 1 1 1 (∅, ),( , ∅, ∅) , ),(∅, 1 1 1 1 1 2 2 1 1 2 (∅, (cid:16)∅, (cid:16)∅, , (cid:17) (cid:17) (cid:16) (cid:16)∅, , , , , ∅(cid:17) (cid:16) (cid:17) , ) , (cid:17) 130 126 116 120 90 120 119 113 114 110 23 27 19 26 18 32 34 45 32 30 Continued on next page. 7 8 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 560 561 562 563 564 565 566 567 568 569 570 571 4 0 0 0 3 0 1 0 4 0 0 0 4 0 0 0 3 0 0 0 0 1 0 0 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 2 3 0 0 0 0 0 2 0 0 0 3 0 0 0 3 0 0 1 1 0 2 8 0 1 0 1 0 0 0 0 2 2 0 1 0 0 0 4 3 0 0 0 3 1 0 2 0 0 1 1 0 0 1 0 1 0 2 5 0 0 4 0 0 7 0 0 0 0 0 0 3 6 0 0 1 1 1 1 (∅, ),(∅, , ∅),( , ∅, ∅),( , , , ) 1 1 1 1 1 1 2 ),( , , ∅) (∅, ∅, (cid:16)∅, (cid:17) 1 1 1 1 (∅, ),( , ∅, ∅) , 1 1 1 1 ( , ∅, ),( , , ) 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 (∅, (cid:16)∅, , (cid:17) ),( , (cid:16)∅, , , ∅) (cid:17) ,( , ∅) ( ( , ) ),( ),( ) (∅, ),(∅, ),( , ) 1 1 1 1 1 (∅, ∅, , ),(∅, , ∅, ) 1 1 1 (∅, ),( , ),( , ) 110 110 99 113 103 98 94 81 128 124 180 114 34 34 33 40 63 55 36 101 48 36 42 42 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 7 9 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 572 573 574 575 576 577 578 579 580 581 0 0 0 2 0 2 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 0 2 0 0 0 0 2 0 0 0 3 0 0 2 0 0 0 3 0 0 0 3 0 0 0 0 2 0 0 0 5 0 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 0 3 0 0 0 4 4 0 0 0 1 0 2 0 0 0 4 4 0 0 0 0 4 0 0 0 0 1 0 0 1 5 0 0 0 5 0 0 0 0 1 0 0 2 4 0 0 0 3 0 0 0 4 0 0 0 1 1 1 1 1 1 1 1 1 (∅, (∅, , , ),( , ∅, , , ) ),(∅, , ∅),(∅, , ∅),(∅, , ∅) 1 1 1 1 1 (∅, , , ),(∅, , ∅, ) 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ∅),(∅, , ∅) 1 1 1 1 (∅, ∅, ),(∅, , ∅),(∅, 1 1 1 1 1 (∅, , ∅, ∅),(∅, , , 1 1 1 1 (∅, ∅, ),(∅, (cid:16) ),(∅, (cid:17) , , 1 1 2 ( , ∅),( , ∅), ),( , ∅, ∅) , ) , ∅),( , , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 (∅, ∅, ),(∅, , ),( , , ),( , , ) 176 160 146 160 136 140 132 101 144 120 12 0 16 4 24 26 22 31 -8 36 Continued on next page. 8 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions (cid:16)∅, (cid:17) ,( , ∅) (cid:16)∅, (cid:16) , (cid:17) (cid:17) (cid:16)∅, (cid:16) , , , (cid:17) (cid:17) , 1 1 3 1 2 2 582 583 584 585 586 587 588 589 590 591 592 593 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 4 0 0 0 0 0 2 0 0 0 3 3 0 1 0 0 0 2 0 0 0 3 3 0 0 2 0 1 0 0 2 0 0 0 0 2 1 0 1 0 5 0 0 4 0 0 0 0 0 4 0 0 0 4 5 0 0 0 0 0 4 0 0 0 4 5 0 0 6 0 0 2 0 0 0 2 0 0 1 2 0 0 0 5 0 0 0 4 0 0 0 5 0 0 0 4 0 0 0 0 0 5 0 4 0 0 0 0 0 5 0 1 1 1 1 (∅, , ∅),(∅, , ∅),(∅, , ∅),( , ∅, ∅) 1 1 1 1 (∅, ),(∅, , ),( , ∅, ∅),( , ∅, ∅) , 1 1 1 ( , ),( , ),( , ) 1 1 1 1 (∅, ∅, ),(∅, , ∅),(∅, , ∅),( , , ) 1 1 1 1 1 2 2 1 1 1 1 1 1 2 1 1 1 1 1 1 1 2 ),(∅, (∅, (cid:16) , (cid:17) (cid:16) , , , ),( , ∅, ∅),( , , , ) (cid:17) ),(∅, ),( , ∅) (∅, (cid:16)∅, (∅, ∅, (cid:16)∅, , , (cid:17) (cid:16)∅, , ),( , (cid:17) (cid:16)∅, , (cid:17) (cid:16)∅, , (cid:17) (cid:16)∅, , , , ) , (cid:17) (cid:16)∅, , , (cid:17) (cid:16) (cid:17) , , , (cid:17) , , Degree Euler Number 116 106 115 116 99 120 110 85 120 160 134 130 16 21 27 22 51 24 26 28 24 8 36 20 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 8 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 594 595 596 597 598 599 600 601 602 603 604 2 0 0 0 0 3 0 0 0 1 3 0 0 0 6 0 0 0 3 0 1 0 4 0 0 0 3 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 3 3 0 1 2 0 1 0 0 0 2 0 0 0 2 3 0 1 0 0 0 2 0 0 0 2 3 0 0 0 0 1 0 0 0 0 2 2 1 0 0 5 0 0 4 0 0 0 0 0 3 3 0 0 0 4 0 0 0 0 0 5 3 0 0 0 3 1 0 5 0 0 1 1 1 1 1 (∅, , , ),( , ∅, , ) 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 1 1 2 (∅, ( , ),( , (cid:16) ), , , ∅) ),( (cid:17) 1 1 1 1 (∅, ∅, ),(∅, , ∅),(∅, , ∅),( , ∅, ∅) 1 1 1 1 1 1 2 (∅, (cid:16)∅, , (cid:17) ),( , ∅, ) ,( , ),( , ∅) 1 1 1 1 (∅, ),(∅, , ∅),(∅, , ∅),( , , , ) 1 1 1 1 1 2 2 1 1 3 (∅, (cid:16)∅, (cid:16)∅, , (cid:17) (cid:17) , , (cid:16)∅, (cid:16)∅, (cid:16)∅, , (cid:17) (cid:17) ),( , ∅, ∅) (cid:17) (cid:16)∅, (cid:16) , , (cid:17) (cid:17) ,( , ∅), 125 112 104 100 110 104 100 100 99 100 95 28 56 40 32 26 37 24 36 40 22 31 Continued on next page. 8 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 605 606 607 608 609 610 611 612 613 614 615 616 3 0 0 0 0 3 0 0 0 3 0 0 0 0 4 0 0 0 4 0 0 0 3 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 4 0 0 0 0 0 0 2 0 0 3 3 0 0 0 0 0 2 0 0 0 2 6 0 0 0 0 0 3 1 0 1 0 0 0 2 5 0 0 0 0 0 3 4 1 0 0 0 0 0 2 5 0 0 4 1 0 0 0 0 5 0 0 0 5 3 0 0 0 0 0 3 5 0 0 0 0 0 4 4 0 0 1 1 1 1 ( , , ),( , , ) 1 1 1 (∅, ),( , ),( , ) 1 1 1 1 1 (∅, , ∅, ),( , ∅, , ) 1 1 1 1 1 1 (∅, ),(∅, (∅, ),(∅, ) ) 1 1 1 1 (∅, ∅, ),(∅, , ),(∅, , ),( , , ) 1 1 1 1 (∅, ),(∅, , ∅),(∅, , , ),( , ∅, ) 90 84 150 216 132 120 116 45 52 49 24 12 40 20 1 1 2 ( , ∅),( , ) 113 -11 1 1 1 1 1 1 4 1 1 1 1 1 1 2 (∅, (cid:16) , ),( , , ) , (cid:17) (∅, ∅, (cid:16)∅, (cid:17) ),(∅, , ),(∅, , ∅),( , , ) ,( , ),( , ∅) - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 102 86 100 96 21 34 34 8 Continued on next page. 8 3 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 8 4 E L A N A K A L A S H N K O V I 90 192 128 144 112 114 105 95 85 90 83 80 25 32 4 4 24 31 25 29 42 32 72 36 Continued on next page. (cid:17) , , ) (cid:16)∅, (cid:17) ,( , ) 1 1 2 1 1 1 1 (∅, ∅, ),(∅, ∅, ) 1 1 1 ( , ),( , ∅),( , ),( , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 (∅, ),( , ) 3 0 0 0 0 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 2 1 1 2 1 1 1 1 1 1 3 (∅, (cid:16)∅, (∅, ( , (cid:16)∅, (∅, (cid:16) , , , ∅),( , ∅, ∅, ∅) (cid:17) (cid:16)∅, , , (cid:17) (cid:16)∅, , , (cid:17) (cid:16) , , ),(∅, , , ∅),(∅, , ∅),( , ∅, ), (cid:16) (cid:16) (cid:17) , , , (cid:16) (cid:17) (cid:17) , (cid:17) , ,( , ∅) , (cid:17) ),( , , ) ,( , ∅) 617 618 619 620 621 622 623 624 625 626 627 628 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 2 0 0 7 0 0 0 2 0 0 3 0 0 0 0 2 0 0 2 0 0 0 0 2 5 0 0 2 0 1 0 0 0 3 3 0 1 5 0 0 0 1 0 1 3 0 0 0 0 0 3 5 0 0 0 0 0 0 2 2 0 0 0 0 0 0 4 4 0 0 4 0 0 0 1 0 3 4 0 0 5 0 0 0 4 0 0 0 3 0 0 0 0 0 0 5 0 5 0 0 0 2 0 0 0 Continued from previous page. Period ID Adjacency matrix Dimension vector (cid:16)∅, (cid:16) (cid:17) , Generalized partitions (cid:16) , , (cid:17) (cid:17) , ) 629 630 631 632 633 634 635 636 637 638 639 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 7 0 1 0 1 1 0 1 0 0 6 0 0 0 3 2 0 1 0 0 0 2 0 0 2 0 0 0 0 2 0 0 0 2 5 0 0 6 0 0 6 0 0 4 0 0 0 1 0 2 5 0 0 0 0 0 3 2 0 0 0 0 1 0 2 0 1 0 3 5 0 0 0 4 0 0 0 0 0 5 0 4 0 0 0 3 0 0 0 1 2 2 1 1 ( ),( 1 1 1 1 1 1 1 1 1 (∅, (∅, (∅, ),( , ) ),( , ) ),( , ),( , ),( , ) 1 1 1 1 (∅, ∅, ),(∅, ∅, ),(∅, , ),( , , ) 1 1 1 ( , ),( , ) 1 1 1 1 1 1 1 2 (∅, (cid:16)∅, , ∅),(∅, , (cid:17) (cid:16)∅, , , ) , (cid:17) (cid:16)∅, , (cid:16) (cid:17) , , , , (cid:17) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 1 (∅, ),(∅, , , ) Degree Euler Number 80 96 93 92 192 168 83 128 100 104 100 35 90 51 78 32 40 77 -8 32 40 26 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 8 5 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 640 641 642 643 644 645 646 647 648 649 650 651 4 0 0 0 3 0 0 0 5 0 0 0 3 0 0 0 2 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 3 3 0 0 2 0 1 0 0 0 2 2 0 1 0 0 0 2 3 0 1 3 0 1 1 0 0 1 0 0 0 2 0 0 5 0 1 0 2 0 0 0 3 3 1 0 5 0 0 0 0 0 4 5 0 0 0 1 0 3 4 0 0 4 0 0 4 1 0 0 0 0 0 3 5 0 0 1 1 1 1 (∅, , ∅),(∅, , ) 1 1 1 1 1 1 2 1 1 1 ( , (cid:16)∅, , (cid:17) , ),( , (cid:16)∅, , ∅) (cid:17) (∅, ),( , ∅) 94 84 82 90 1 1 1 1 (∅, ∅, ),(∅, ∅, ),(∅, , ),( , , ) 144 1 1 1 (∅, ),( , ) 1 1 1 1 (∅, , ∅),(∅, , ) 1 1 1 ( , ),( , ∅) 1 1 1 ( , ),( , ) 1 1 1 1 (∅, , ∅),(∅, , ∅) 1 1 1 1 (∅, ),( , ∅, , ) 1 1 1 (∅, ),( , ∅),( , ∅),( , ) 82 89 74 68 108 86 80 32 54 54 12 48 84 40 68 93 20 21 40 Continued on next page. 8 6 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 652 653 654 655 656 657 658 659 660 661 662 663 4 0 0 0 5 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 3 0 0 3 0 0 0 0 0 2 0 0 0 2 0 0 0 2 0 0 5 0 0 6 0 0 0 2 0 0 0 2 3 0 0 3 0 0 0 0 0 3 0 5 0 3 2 0 0 0 0 4 0 1 0 4 0 0 0 6 5 0 0 4 0 0 0 0 0 3 0 0 0 3 0 5 0 5 0 0 1 1 1 1 1 1 2 1 1 1 (∅, (cid:16) , , ),( , (cid:17) (cid:16) , , , ) (cid:17) (cid:16) , (cid:17) (cid:16) , , (cid:17) , (∅, ),(∅, ) 1 1 1 1 (∅, ∅, ),(∅, ∅, ),(∅, , ),( , , ) 1 1 1 1 (∅, , ),(∅, , ) 1 1 1 1 (∅, , ∅),(∅, , ),(∅, , ),( , ∅, ∅) 1 1 1 1 1 1 ( , ∅),( , ∅),( , ∅),( , ∅) ( , ∅),( , ∅),( , ∅),( , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 1 1 2 1 1 3 (∅, (cid:16) (cid:16)∅, , (cid:17) 80 75 68 120 128 104 70 70 84 74 74 40 42 64 56 0 12 60 52 58 64 38 , ∅),( , (cid:17) , ∅) , (cid:16) (cid:17) ,(∅, ) , ,( , ∅) 112 -12 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 8 7 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions (cid:17) (cid:16)∅, (cid:16) , (cid:17) (cid:16) (cid:17) , , (∅, ),(∅, ),( , ∅) 1 1 2 1 1 2 1 1 1 664 665 666 667 668 669 670 671 672 673 674 675 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 4 0 0 2 0 0 0 0 0 0 1 3 0 0 3 0 0 7 0 0 0 0 0 2 0 0 6 0 0 0 2 0 0 5 4 0 0 4 0 0 4 1 0 2 0 0 0 0 6 0 0 0 5 0 5 1 0 0 3 0 0 0 0 2 4 0 0 0 1 0 1 5 0 0 3 1 1 0 3 0 0 0 5 0 0 0 5 0 0 0 1 1 1 1 (∅, ∅, ),(∅, ∅, ) 1 1 1 2 1 1 2 1 1 1 1 1 2 ( (cid:16) , , , (cid:17) , (cid:16) (cid:17) (cid:16) , , , ),( , , ∅),( , ∅, ∅) (cid:17) (cid:16)∅, , (cid:17) (∅, ),(∅, ),( , ) (∅, ),( , ) 1 1 1 1 (∅, ∅, ),( , ∅, 1 1 1 (∅, ),( , ∅),( ) , ),( , ) 1 1 1 1 (∅, ∅, ),(∅, , ) 1 1 1 (∅, ),(∅, ),( , ),( , ) 8 8 E L A N A K A L A S H N K O V I Degree Euler Number 69 64 88 84 89 70 64 55 112 80 88 96 68 77 40 32 -3 49 56 59 -8 24 32 64 Continued on next page. Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 676 677 678 679 680 681 682 683 684 685 686 687 688 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 5 0 0 7 3 0 0 3 0 0 0 0 0 2 0 0 6 6 0 1 0 1 0 0 0 2 3 0 0 2 0 1 0 0 0 2 0 1 0 4 5 0 0 3 0 0 0 4 0 4 1 0 0 0 0 4 4 0 0 5 0 0 0 0 0 3 5 0 0 4 0 0 0 0 0 2 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 1 1 1 1 (∅, , ),(∅, , ) 1 1 1 1 1 1 1 1 2 1 1 1 (∅, ),( , ),( , ∅),( , ) ( , ∅),( , ∅),( , ),( , ) (cid:16) , (cid:17) (∅, ),( , ) 1 1 1 1 (∅, ),(∅, , , ∅) 1 1 1 ( , ),( , ∅),( , ∅),( , ) 1 1 ( ) 1 1 1 ( , ) 1 1 1 1 1 1 2 (∅, (cid:16)∅, , ∅),( , ∅, ∅) (cid:17) (cid:16)∅, , (cid:17) ,( , ∅) 1 1 1 ( , ) 1 1 1 1 (∅, ),( , ∅, , ) 74 60 60 56 54 96 55 64 61 68 96 52 65 42 68 48 61 94 -24 85 188 149 62 -20 168 37 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 8 9 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 689 690 691 692 693 694 695 696 697 698 699 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 1 0 0 4 3 0 0 0 0 5 0 0 7 0 0 0 2 0 0 3 0 0 4 0 0 4 0 0 0 3 2 0 0 3 0 0 4 0 0 3 1 0 5 0 0 3 0 0 0 0 0 3 5 0 0 4 0 0 4 0 0 0 1 0 1 5 0 0 4 0 0 0 4 0 0 0 1 1 2 ( , ) 1 1 1 1 1 1 1 1 1 (∅, (∅, (∅, ),( , ) ) ),( , ),( , ),( , ) 1 1 1 ( , ),( , ),( , ),( , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 (∅, ),( , ∅) 1 1 1 ( , ),( , ) 1 1 1 (∅, ),( , ) 1 1 1 1 1 1 2 (∅, ∅, (cid:16)∅, (cid:17) (cid:16) , , ),( , , ) (cid:17) 51 162 99 64 56 144 96 80 144 126 60 51 27 27 72 40 36 -9 35 36 45 40 Continued on next page. 9 0 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 700 701 702 703 704 705 706 707 708 709 710 711 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 7 0 0 0 2 0 0 0 2 2 0 2 0 0 0 2 0 0 0 2 0 0 4 0 0 5 3 0 1 0 0 0 2 0 1 0 2 0 0 0 4 3 0 0 0 0 0 3 3 0 0 0 3 0 0 0 0 0 2 0 1 0 1 4 0 0 5 0 0 4 0 0 0 1 0 4 4 0 0 0 3 0 0 0 4 0 0 0 3 0 1 0 5 0 0 0 5 0 0 0 2 0 0 0 1 1 1 1 (∅, ∅, ),(∅, , ) 108 54 1 1 1 1 (∅, , ),( , , ) 1 1 1 ( , ),( , ∅),( , ),( , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 1 (∅, ∅, ),( , , ) 1 1 1 ( , ∅) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 1 (∅, ∅, ),(∅, , ∅) 1 1 1 ( , ),( , ) 1 1 1 1 1 1 (∅, ),(∅, ),( , ),( , ) (∅, ),( , ∅) 1 1 1 1 (∅, , ∅),(∅, , ) 54 50 90 81 51 96 78 46 72 66 63 105 86 63 36 135 -12 0 93 72 84 48 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 9 1 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 712 713 714 715 716 717 718 719 720 721 722 723 724 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 6 3 0 0 6 0 0 0 3 3 0 0 0 0 7 0 0 0 2 0 0 0 2 0 0 4 0 0 0 2 3 0 0 0 0 3 4 1 0 4 0 0 0 5 0 5 0 0 5 0 0 3 0 0 0 0 0 3 0 0 0 3 4 0 0 0 0 0 2 5 0 0 5 0 0 4 0 0 0 4 0 0 0 5 0 0 0 1 1 1 1 1 1 1 1 2 1 4 1 1 1 1 1 2 1 1 1 (∅, ),( , ) ),( , ),( , ∅),( (cid:17) (cid:16) , (cid:17) (cid:16) , , (∅, (cid:16) (cid:18) (cid:19) , (cid:17) , , ) ,( , ∅) ( , (cid:16)∅, ),( , ) (cid:17) (cid:16)∅, , (cid:17) (cid:16) ,( , ∅), (cid:17) , ( , ),( , ∅),( , ∅),( , ) 1 1 1 1 1 1 1 1 (∅, (∅, , , ),(∅, , ) ),( , , ) 1 1 1 ( , ∅),( , ∅) 1 1 1 1 1 1 2 1 1 1 (∅, (cid:16)∅, , ∅),( , ∅, ) (cid:17) (cid:16)∅, , (cid:17) ,( , ∅), (cid:16) (cid:17) , (∅, ),( , ∅) 48 60 80 42 41 50 45 80 62 40 53 45 40 99 12 -32 73 109 52 63 -28 44 152 80 94 144 Continued on next page. 9 2 E L A N A K A L A S H N K O V I Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 725 726 727 728 729 730 731 732 733 734 735 736 737 738 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 3 0 0 4 0 0 4 0 0 3 6 0 0 0 7 0 0 8 0 0 8 6 0 0 0 4 0 0 2 2 0 0 0 0 2 3 1 0 5 0 0 4 0 0 4 0 0 5 0 0 3 0 0 2 0 0 2 0 0 4 0 0 6 0 0 5 1 0 6 0 0 1 1 1 ( , ) 1 1 1 1 1 1 1 1 1 1 1 1 1 2 1 1 1 1 1 1 1 1 1 1 2 (∅, (∅, ),( , ) ),( , ) (∅, ),( ( , ∅),( , (cid:16) (cid:17) (cid:16) (cid:17) , , ) , ) (cid:17) (cid:16) (∅, ),( , ),( , ),( , ) ( , ∅),( , ),( , ),( , ) ( , (cid:16) (cid:17) ),( (cid:16) (cid:17) , , (cid:16) (cid:17) , ),( (cid:16) , ),( , ) , (cid:17) 1 1 1 ( , ),( , ) 1 1 1 1 1 1 1 1 1 (∅, ),(∅, (∅, ),(∅, ) ) (∅, ),( , ) 38 42 54 44 35 33 64 40 36 28 26 48 30 27 191 27 81 116 155 90 -48 72 92 140 251 -72 114 99 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 9 3 Continued from previous page. Period ID Adjacency matrix Dimension vector Generalized partitions Degree Euler Number 739 740 741 742 743 744 745 746 747 748 749 6 0 0 5 0 0 6 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 5 0 9 0 5 0 0 0 3 0 0 2 0 0 3 8 0 7 0 7 0 6 0 1 1 1 1 2 1 1 1 2 1 1 1 1 1 1 ),( , (cid:16) ),( ),( (∅, (cid:16) (cid:16) (cid:17) ( , (cid:17) (cid:16) , ) ),( ) (cid:17) (cid:17) (∅, (∅, ),( , ) ),( , ∅) 1 1 1 ( , ),( , ) 1 1 1 1 1 1 1 1 ( ( ( ( ),( ),( ) ),( ),( ) ) ) 26 20 16 15 32 20 17 12 9 8 5 186 176 224 318 -112 212 293 324 369 552 825 9 4 E L A N A K A L A S H N K O V I Table 4: Some regularized period sequences obtained from 4-dimensional Fano manifolds that arise as quiver flag zero loci. Period ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 Name 124, P4, CKP1 115, CKP2 BØS4 BØS4 CKP3, Q4 CKP4 BØS4 118, CKP8 BØS4 47, CKP10 BØS4 94, CKP11 BØS4 37, CKP12 CKP13, BØS4 74 CKP14 CKP15, BØS4 86 CKP16 CKP18 P2 × P2, CKP20, FI4 6, BØS4 123 114, CKP21 BØS4 CKP23, BØS4 46 CKP25, BØS4 32 CKP26 CKP29 CKP30 FI4 5 CKP33, FI4 4 CKP34 CKP35 CKP36 Str1 FI4 CKP39 3, CKP37 α0 α1 α2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 α3 0 0 0 0 6 6 6 6 6 6 6 6 6 12 12 12 12 12 12 12 12 12 12 18 18 18 24 24 24 24 24 30 36 36 α4 0 24 48 48 0 24 24 24 48 48 48 48 72 0 0 0 24 24 48 48 48 72 96 0 48 48 0 48 48 96 120 120 0 144 α5 120 120 0 120 120 0 120 120 0 0 120 120 120 0 120 120 0 240 0 0 120 120 120 0 0 120 0 0 120 120 120 240 0 120 α6 0 0 0 0 90 90 90 90 90 90 90 90 90 900 540 900 900 900 540 900 540 540 540 1710 1710 2430 3240 3240 3600 3240 3240 5850 8100 8100 α7 0 0 0 0 0 2520 1260 2520 2520 3780 2520 3780 5040 0 0 0 3780 5040 7560 7560 7560 10080 15120 0 11340 11340 0 15120 15120 30240 40320 50400 0 75600 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 9 5 Continued from previous page. Period ID Name α0 α1 α2 122, 121 104 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 α3 0 0 0 0 0 0 0 0 0 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 12 12 α4 6 30 30 30 54 54 54 54 54 6 6 30 30 30 30 30 30 30 30 54 54 54 54 54 54 54 54 54 78 78 102 6 6 α5 120 0 120 120 0 0 120 120 240 180 240 60 60 120 120 180 180 180 240 60 60 60 180 180 180 240 240 360 180 360 600 120 180 α6 20 740 380 740 740 1100 740 1100 1460 110 110 470 830 470 830 470 470 830 830 830 830 1190 830 1190 1190 1190 1190 1550 1190 1910 2990 920 920 9 6 E L A N A K A L A S H N K O V I α7 2520 0 2520 2520 0 0 2520 2520 5040 2940 3780 2940 2940 3780 2520 4200 5460 5460 5040 2940 4200 4200 5460 6720 7980 6300 7560 8820 7980 11340 17640 840 1680 Continued on next page. 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 106, CKP79 111, CKP76 CKP47, BØS4 P1 × P3, CKP51, BØS4 MW4 14 BØS4 109, CKP54 CKP55, BØS4 CKP60, MW4 13 CKP61, MW4 12 CKP64 CKP65 CKP67 BØS4 CKP78 BØS4 CKP80, BØS4 45 CKP81, BØS4 41 P1×MM3 BØS4 82, CKP84 CKP85, BØS4 113 BØS4 92, CKP86 CKP89 BØS4 CKP92 CKP93 CKP96, BØS4 91 CKP98 CKP99 BØS4 CKP101 CKP102 CKP103 CKP104 CKP107 CKP109 BØS4 112, CKP110, P2 × S2 81, CKP100 8 2 -- 33, BØS4 110, CKP83 52, CKP91 Continued from previous page. Period ID 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 Name CKP111, P1 × Q3 CKP113 CKP114 BØS4 BØS4 BØS4 CKP119 P1 × MM3 60, CKP116 88, CKP117 35, CKP118 2 -- 30, CKP120 BØS4 93, CKP121 85, CKP127 CKP122 CKP123 CKP124 CKP125 CKP126 BØS4 CKP128 CKP130 CKP131 CKP132 CKP134 CKP135 CKP136 CKP137 CKP138 CKP139 CKP141 CKP142 CKP144 CKP145 BØS4 CKP147 51, CKP146 α0 α1 α2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 α3 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 18 18 18 α4 6 6 6 30 30 30 30 30 30 30 30 30 30 54 54 54 54 54 54 54 54 78 78 78 78 78 102 102 102 126 198 6 6 6 α5 240 300 360 120 180 180 180 240 300 300 300 360 420 120 120 180 240 240 300 360 420 240 300 300 360 480 240 480 480 720 1200 180 240 300 α6 560 920 560 920 1280 1280 1640 1280 920 1280 1640 1280 1280 1280 1640 1640 1280 1280 2000 1640 2000 2000 2000 2720 2000 2360 2000 2720 2720 3800 6320 1370 1730 1730 α7 2520 4200 5040 4620 5460 5460 5460 5040 9240 7980 7980 7560 11760 8400 8400 9240 9660 10080 11760 12600 15540 14700 14280 16800 15120 17640 18480 20160 22680 30240 52920 1260 2100 2940 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 9 7 Continued from previous page. Period ID Name α0 α1 α2 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 CKP148 CKP151, BØS4 73 CKP152 CKP153 CKP154 CKP155 CKP158 CKP159 CKP160 CKP161 CKP162 CKP163 CKP164 CKP165 CKP166 CKP167 CKP168 Str2 CKP169 CKP170 CKP171 CKP172 CKP174 CKP175 BØS4 BØS4 P1 × MM3 2 -- 35 43, CKP181 117, CKP183, MW4 17, 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 4 4 α3 18 18 18 18 18 18 18 18 18 24 24 24 24 24 24 24 24 24 24 24 24 24 30 30 30 36 36 36 36 36 42 0 0 α4 6 30 30 54 54 78 78 102 102 6 6 6 54 54 54 54 102 102 126 150 174 246 54 78 126 6 6 150 294 438 150 36 60 α5 420 240 360 180 240 300 360 300 360 240 360 540 360 360 480 540 420 480 660 720 960 1440 600 960 540 360 720 840 1680 2640 900 120 0 α6 1730 2090 2450 2090 2810 2450 3170 3170 3890 3260 3260 3260 3980 4340 4700 4340 4700 4700 5780 6140 7220 9740 6590 7670 7670 8120 8120 11000 15320 20360 14690 400 1480 9 8 E L A N A K A L A S H N K O V I α7 5460 7140 9660 11340 13440 18900 21000 26460 28560 1680 3360 6720 18480 18480 21000 21840 35280 35280 49560 55440 70560 105840 26040 46200 56700 2520 8400 86520 178920 287280 99540 5040 0 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 α3 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 CKP185, BØS4 36 MW4 11 MW4 10, CKP186 CKP187 CKP189, MW4 7 BØS4 P1, P1 × MM3 CKP197 120, P1×P1×P2, P2×P1× 2 -- 34, CKP195 3 -- 30, CKP200, BØS4 89 3 -- 26, CKP204, 103 P1×MM3 CKP201, BØS4 34 CKP203 P1 × MM3 BØS4 CKP205, BØS4 CKP206, BØS4 44 CKP207 CKP208 CKP209 CKP214 CKP215 102 29, CKP217 BØS4 CKP218 CKP219 CKP220 CKP222 CKP224 P1×MM3 8 × S2 8 , BØS4 S2 CKP227, BØS4 CKP228, P1 × MM3 CKP230 CKP231 3 -- 31, CKP225, BØS4 72 83, CKP226 101, P2 × S2 7 2 -- 31 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 0 0 0 0 0 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 12 12 12 12 12 12 α4 60 84 84 84 108 36 36 36 60 60 60 60 60 60 60 60 60 84 84 84 84 84 84 108 108 132 36 36 36 36 36 36 α5 120 0 0 240 0 240 300 360 180 180 240 240 240 240 300 300 360 240 240 300 360 360 420 240 300 660 360 360 360 420 480 480 α6 1480 2200 2560 2560 3280 490 490 490 1570 1570 1210 1570 1570 1930 1210 1570 1570 2650 2650 2290 2650 3010 2650 3370 3370 4810 940 1300 1300 940 1300 1300 α7 5040 0 0 10080 0 7560 9240 12600 5460 6720 10080 8820 9660 9660 11760 10500 13860 10080 12180 13020 15120 17220 16800 13860 15540 30660 8400 8400 9660 11760 13440 14700 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 9 9 Continued from previous page. Period ID Name α0 α1 α2 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 3 -- 25, CKP233 P1 × MM3 BØS4 108 CKP239 CKP240, P1 × MM3 CKP243 CKP244 CKP236, 3 -- 23 CKP246 CKP247 CKP248 BØS4 80, CKP251 CKP252 CKP255 P1 × MM3 3 -- 19, CKP256 CKP258 CKP259 CKP260 CKP262 CKP265 CKP266 CKP267 CKP268 CKP269 CKP270 2 -- 27 CKP277, P1 × MM3 CKP279 CKP280, BØS4 53 CKP282 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 α3 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 18 18 18 18 18 α4 36 60 60 60 60 60 60 60 60 84 84 84 84 84 84 84 84 84 84 84 108 108 108 108 108 108 132 132 36 60 60 84 84 α5 720 360 360 420 480 480 480 540 600 360 420 420 420 480 480 480 480 600 600 720 540 600 600 600 720 720 600 720 720 600 840 480 540 α6 940 2020 2380 2020 2020 2380 2740 2020 2020 3100 2740 2740 3100 2740 3100 3100 3460 3460 3820 3100 4180 3820 4180 4900 4180 4900 4900 5260 2110 2830 3910 3190 3910 1 0 0 E L A N A K A L A S H N K O V I α7 25200 10920 13440 14280 17220 18480 18480 19320 23520 15960 17640 19320 18900 19740 20160 23520 22260 28560 27300 32760 27720 28560 31080 32340 38640 37380 33600 37800 21000 19740 32340 20580 25200 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 CKP283, BØS4 84 CKP284 CKP285 CKP286 CKP287 CKP288 CKP289 CKP290 CKP291 CKP292 CKP293 CKP294 CKP295 CKP296 CKP297 CKP298 CKP299 CKP300 CKP301 CKP302 CKP303 CKP304 CKP305 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 α3 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 24 24 24 24 24 24 24 24 24 24 24 24 24 α4 84 84 84 84 84 108 108 108 108 108 108 108 108 132 132 132 132 156 156 180 180 36 36 84 84 84 84 108 108 132 132 132 132 156 α5 600 600 720 720 780 600 600 660 720 780 780 840 960 780 840 960 960 840 1020 1020 1080 720 1080 720 840 1080 1140 960 1080 840 840 840 1020 960 α6 3550 4270 3910 4630 4270 4270 4630 4990 4990 4990 5350 4990 6070 5350 5710 7150 7510 7150 7870 7870 9310 3640 3640 5800 5800 5800 5800 6880 6520 6880 7240 7960 7600 7960 α7 25620 26880 32340 32340 34020 30660 31920 34020 38220 39060 40320 44940 49980 42840 48720 55020 57540 56280 63000 66780 77700 16800 33600 31920 38220 48720 51660 49560 58800 53340 53760 54600 60480 63420 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 1 0 1 Continued from previous page. Period ID Name α0 α1 α2 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 CKP306 CKP307 CKP308 CKP309 CKP310 CKP311 CKP312 CKP313 CKP314 CKP315 CKP316 CKP318 CKP319 CKP320 CKP321 16, CKP325 8 , CKP328, P1 × 2 -- 32, MW4 CKP322 CKP323 CKP324, BØS4 38 P1×MM3 CKP326, MW4 8 MW4 5, CKP327 P1 × P1 × S2 3 -- 28, BØS4 MM3 CKP330, P1 × MM3 CKP332 CKP334 CKP335 107 3 -- 24 CKP336 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 6 6 6 6 6 6 6 6 6 6 6 α3 24 24 24 24 24 30 30 30 30 30 30 30 36 36 36 36 42 42 42 48 60 60 0 0 0 0 6 6 6 6 6 6 6 α4 156 180 204 228 276 84 84 132 132 156 204 228 36 84 156 324 156 180 252 180 204 564 90 114 138 186 90 114 114 138 138 138 186 α5 1080 1440 1260 1440 1680 840 1200 960 1140 1320 1440 1440 1800 1440 1200 2160 1680 2040 2040 1920 2640 4140 120 0 0 0 300 300 360 300 360 420 360 α6 9040 11560 10480 12280 13720 6610 8050 8770 9490 11650 12730 12370 8500 10660 12820 20740 16510 19390 21190 22000 33340 49900 1860 3300 4740 7980 1950 3390 3390 4830 4830 4830 8070 1 0 2 E L A N A K A L A S H N K O V I α7 72240 99120 95760 110880 137760 36540 54600 61740 70560 92400 113820 116340 58800 64680 90720 223440 119700 155400 196980 156240 231840 648480 7560 0 0 0 13020 14280 18480 15540 21000 24360 24780 Continued on next page. Continued from previous page. Period ID 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 Name 8 × S2 7 , CKP340 2 -- 29 90, CKP345 3 -- 20 3 -- 17, CKP351 79, S2 BØS4 CKP341, P1 × MM3 P1×MM3 4 -- 10, BØS4 CKP346, P1 × MM3 CKP347 CKP348 CKP349 P1 × MM3 CKP352 CKP354 CKP355 CKP356 CKP357 CKP359 CKP360 CKP361 CKP363 P2 × S2 CKP367 CKP368, P1 × MM3 CKP369 6 , BØS4 99, CKP365 3 -- 18 CKP372 CKP373, P1 × MM3 CKP374 CKP376 CKP377 3 -- 16 CKP378 CKP380 CKP381 α0 α1 α2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 α3 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 α4 90 90 114 114 114 114 114 138 138 138 138 138 138 162 162 186 186 90 114 114 114 114 138 138 138 138 138 138 138 162 162 162 162 α5 540 600 540 600 660 660 720 600 600 600 660 660 780 600 720 720 900 720 780 840 960 1140 780 900 900 960 960 1020 1080 960 960 1080 1080 α6 2760 2400 3840 3840 3840 4200 4200 5280 5280 5640 5280 5640 5640 7080 7080 8520 8880 3570 5010 4650 5010 4650 6090 6090 6090 5730 7170 6450 7890 7530 7890 8250 8970 α7 21420 26040 23940 28560 32760 32760 36960 31080 33600 35280 35280 36540 45360 38640 44520 51240 63000 28980 34860 38220 47040 61740 39900 46620 47460 52080 56700 57960 66780 58380 58380 65940 71820 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 1 0 3 Continued from previous page. Period ID Name α0 α1 α2 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 CKP382 CKP383 CKP384 2 -- 25 CKP388, P1 × MM3 CKP392 CKP393 CKP394 CKP395 CKP397 CKP399 CKP400 CKP401 P1 × MM3 CKP403 CKP404 CKP405 CKP406 2 -- 24, CKP402 CKP407 CKP408 CKP409 CKP410 CKP411 CKP412 CKP413 CKP414 CKP415 CKP416 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 α3 18 18 18 18 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 30 30 30 30 30 30 30 30 α4 186 186 186 210 114 138 138 138 138 162 162 186 186 186 186 186 210 210 210 210 210 210 234 234 282 282 162 186 210 210 234 234 282 282 α5 1080 1140 1140 1320 1200 1080 1260 1320 1440 1140 1320 1200 1200 1260 1320 1560 1440 1500 1500 1620 1800 1800 1800 1920 1920 2280 1680 1800 1620 1740 1680 1980 1980 2160 α6 8970 8970 9690 11850 5820 7980 7980 9060 7980 8700 9780 9780 10860 10140 10500 12660 11940 12660 12660 13020 13380 15180 16620 16980 19140 21300 11670 13470 13470 14550 15270 16710 18150 19950 1 0 4 E L A N A K A L A S H N K O V I α7 69720 74340 76440 96180 57120 57960 67620 78120 82320 67200 80640 76440 82320 78120 85680 110880 107520 107940 110460 115500 133980 138600 146160 153720 169260 199080 103320 122220 116340 126420 133560 159600 168420 186480 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 CKP417 CKP418 CKP419 CKP420 CKP421 CKP422 CKP423 CKP424 CKP425 CKP426 CKP427 CKP428 CKP429 CKP430 CKP432 CKP433 CKP435 CKP436 CKP437 CKP438 CKP439 MW4 P1×P1×P1×P1, P1×MM3 MW4 V 4 14 18, BØS4 9, P1 × B3 119, CKP440, 3 -- 27 5 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 8 8 8 α3 36 36 36 36 36 36 36 36 36 36 36 36 36 36 36 36 42 42 42 48 48 48 48 54 60 60 66 84 120 0 0 0 α4 186 186 186 186 210 210 234 234 258 258 282 306 330 330 330 378 162 306 306 282 282 426 522 378 354 474 474 714 1146 168 192 288 α5 1560 1920 2040 2520 1800 2100 1800 2520 2280 2340 2520 2280 2640 2880 3240 3480 2760 2460 2820 2760 2760 3360 4800 3480 4080 3960 4860 6840 11280 0 0 0 α6 12480 15360 15720 16080 16440 16800 16080 19680 20400 21120 22920 23280 27600 27600 30480 34080 17610 24090 26970 27420 28500 37860 51180 38670 44520 45600 57930 96360 192300 5120 6920 15200 α7 97440 131880 138600 180600 136920 154980 137760 201600 191520 196980 224280 221760 274680 278040 312480 352800 178920 229320 270060 253680 257040 406560 595560 392700 441840 503160 637560 1211280 2817360 0 0 0 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 1 0 5 Continued from previous page. Period ID Name α0 α1 α2 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 BØS4 97, 4 -- 11, P1 × MM3 CKP442, P1 × P1 × S2 7 P1 × MM3 3 -- 21, CKP443 CKP444 P1 × MM3 4 -- 9, CKP445 7 × S2 CKP446, S2 CKP447, P1 × MM3 P1 × MM3 CKP448 2 -- 26 4 -- 8 7 , BØS4 75 78, CKP451 6 , BØS4 CKP449 CKP450 8 × S2 S2 CKP452 CKP454, P1 × MM3 CKP455 CKP456, P1 × MM3 CKP457 CKP458 CKP459 4 -- 7 3 -- 15 4 -- 5, CKP461 2 -- 22 3 -- 13, CKP463 CKP460 P1 × MM3 P1 × MM3 CKP462 P1 × MM3 CKP464 CKP465 CKP466 CKP467 CKP468 CKP469 CKP470 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 α3 6 6 6 12 12 12 12 12 12 12 12 18 18 18 18 18 18 18 18 18 24 24 24 24 24 24 24 24 24 24 24 24 24 α4 168 192 216 168 168 192 192 216 216 288 360 168 168 192 216 216 216 216 240 288 168 216 216 216 240 240 240 264 264 264 288 312 360 α5 360 360 360 720 720 720 780 840 1440 1080 1200 1020 1080 1080 1080 1140 1200 1260 1380 1560 1440 1440 1560 2160 1560 1560 1740 1680 1740 1920 1920 2160 2160 α6 5210 7010 8810 5660 6020 7460 7460 9620 8540 16100 23300 6830 6470 8270 10070 10070 10430 11150 13310 16910 8360 10880 11240 11240 13040 13400 13760 15200 15920 17360 18440 22040 25640 1 0 6 E L A N A K A L A S H N K O V I α7 19740 21000 22260 39480 39480 42000 47880 57960 126000 92400 117600 54600 59220 63000 69300 72660 79380 87360 105000 136500 78960 89040 100800 168000 105840 110460 126420 126840 133980 154560 159600 194880 218400 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 CKP471 P1 × MM3 CKP473 CKP474 3 -- 11, CKP472 CKP475 CKP476 CKP477 CKP478 CKP479 CKP480 CKP481 CKP482 CKP483 CKP484 CKP485 CKP486 CKP487 CKP488 CKP489, P1 × MM3 2 -- 18 CKP490 CKP491 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 α3 30 30 30 30 30 30 30 30 36 36 36 36 36 36 36 36 36 36 36 36 36 36 42 42 42 42 48 48 48 48 48 48 48 48 α4 216 264 264 264 288 288 312 336 216 264 264 288 312 312 312 336 360 360 360 360 408 432 312 360 408 456 264 336 360 384 408 432 432 504 α5 1800 1980 1980 2160 2100 2220 2580 2520 3600 2280 2880 2700 2520 2760 2760 2760 2760 2940 3060 3300 3360 3780 3000 3540 3480 4320 4320 4680 3360 3960 3960 4140 4320 4800 α6 13490 16370 16730 17090 19250 20330 23930 25010 16100 19340 20420 21500 22940 24380 25100 26180 27260 28340 29780 31220 35180 39500 27350 32750 36350 44270 27440 35000 31040 37160 38960 42560 44720 51560 α7 116340 142800 147000 165900 171360 185640 229320 233940 294000 173880 232680 216720 210420 239820 243600 251160 261240 277200 295680 320040 358680 413280 263340 346080 368340 479220 366240 467040 295680 400680 410760 454440 487200 572040 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 1 0 7 Continued from previous page. Period ID Name α0 α1 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 CKP492 CKP493 CKP494 CKP495 CKP496 4 , CKP500, MW4 6 98, CKP501, 6 2 -- 23, CKP502 7 × S2 6 , BØS4 76 4 -- 4 2 -- 21 5 -- 3, BØS4 CKP497 CKP498 P1 × B3 V 4 12 P1×MM3 P1 × P1 × S2 P1 × MM3 CKP503 CKP504, S2 CKP505, P1 × MM3 CKP506 P1 × MM3 CKP507 P1 × MM3 CKP509, P1 × MM3 CKP510 CKP511 CKP512, P2 × S2 CKP513 CKP514 CKP515 P1 × MM3 CKP516 CKP517 CKP518 2 -- 20 5 CKP520 3 -- 12, CKP508 2 -- 19 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 α2 8 8 8 8 8 8 8 8 8 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 α3 48 54 54 60 60 60 66 72 84 0 0 12 12 12 18 24 24 24 24 30 30 30 30 36 36 36 36 36 36 36 36 36 42 α4 504 360 480 360 552 672 456 792 408 318 438 270 318 366 270 318 318 342 462 342 342 366 462 270 366 366 366 390 390 414 414 486 414 α5 4920 4200 5160 5160 5280 7200 6000 8460 8040 0 0 840 960 960 1320 1800 2400 1920 2640 2340 2520 2520 2760 2160 2760 2880 3000 2940 3000 3180 3480 3720 3480 α6 53000 39770 53810 45260 60740 83060 61550 104120 78740 15220 28900 11080 15760 20800 12610 17380 18460 19900 35740 21070 21430 24670 35110 15040 25840 28000 26200 27640 28720 31960 33400 42400 33850 1 0 8 E L A N A K A L A S H N K O V I α7 613200 406980 608580 514080 685440 1032360 699300 1339800 887040 0 0 55440 74760 82320 91560 135240 215040 154560 287280 186060 208740 221760 290640 134400 235200 271740 260400 255360 273000 306600 351960 420000 334320 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 CKP521 CKP522 CKP523 CKP524 CKP525 CKP526 CKP527 CKP528 CKP529 CKP530 CKP531 CKP532 CKP533 CKP534 CKP535 CKP536 CKP538 CKP539 CKP540 CKP541 CKP542 CKP543 CKP544 CKP545 V 4 10 6 × S2 S2 P1 × MM3 8 × S2 S2 6 , CKP546, BØS4 77 4 -- 3, CKP547 5 , CKP548 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 12 12 12 12 α3 42 42 48 48 48 48 48 48 48 48 48 54 54 54 54 60 60 60 60 60 66 66 72 72 72 78 78 84 96 168 0 24 24 36 α4 414 462 414 414 462 486 486 486 510 534 558 486 534 582 606 510 582 582 654 654 750 846 558 726 846 750 846 750 702 1566 684 396 444 396 α5 3840 4080 4080 4320 4200 4320 4680 5400 5160 4920 5280 4920 5100 5580 6180 6120 5760 6360 6840 7080 7920 10080 7200 8280 9360 8700 10140 8520 10560 23040 0 2160 2160 2820 α6 38530 41770 36460 38260 41140 44740 47260 49420 56260 53020 57700 49150 54550 63910 68230 59680 61480 68680 77680 77320 93970 125290 74620 97660 113860 107110 124030 102880 124660 402940 58800 23160 26760 24060 α7 407820 436800 387240 425040 425880 465360 519960 631680 632100 578760 646800 534240 594300 711060 805140 714000 683760 807240 924840 945840 1156680 1619940 890400 1212120 1475880 1328880 1643880 1244040 1538880 6002640 0 186480 191520 219240 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 1 0 9 Continued from previous page. Period ID Name α0 α1 α2 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 3 -- 10, CKP549 P1 × MM3 CKP550 CKP551 P1 × MM3 CKP552 CKP553, P1 × MM3 CKP554 CKP555 2 -- 17 3 -- 7 CKP556 CKP557, P1 × MM3 2 -- 16 CKP558 CKP559 CKP560 CKP561 CKP562 CKP563 CKP564 CKP565 CKP566 P1 × B3 3 , MW4 P1 × P1 × S2 CKP568 7 × S2 S2 5 , CKP569 3, CKP567 5 , P1 × MM3 6 -- 1, 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 14 14 14 α3 36 36 36 42 42 48 48 48 48 54 60 60 60 60 60 60 66 66 72 72 72 72 72 78 84 90 96 96 108 120 0 30 36 α4 492 492 540 540 540 564 588 588 636 732 636 636 684 684 708 780 804 828 708 756 780 828 876 876 876 1116 1140 1356 756 1284 690 546 546 α5 3360 3540 5400 4140 4560 4680 5040 5040 5940 7680 6000 6120 6840 6840 6840 8400 8400 8880 9120 8580 8340 9000 9600 10440 9960 13860 14400 17640 16320 17700 0 2760 3480 α6 35220 38460 41700 43230 49710 48000 54480 55200 68880 84810 64020 63300 76620 77340 77700 101460 100830 108030 93000 97320 97320 108480 118200 125490 122700 184350 193080 245640 155100 253200 50900 33350 37040 1 1 0 E L A N A K A L A S H N K O V I α7 319200 371700 705600 423360 528360 486360 577920 588000 780780 1136520 698460 693000 893760 898800 893760 1254960 1237740 1369620 1254960 1209180 1178520 1354920 1501920 1649340 1540560 2553600 2721600 3609480 2494800 3671640 0 246540 330540 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 2 -- 15 CKP570 CKP571 CKP572, P1 × MM3 CKP573 CKP574, P1 × MM3 4 -- 1 P1 × MM3 3 -- 8, CKP575 CKP576 CKP577 CKP578 CKP579 CKP580 CKP581 CKP582 CKP583 CKP584 CKP585 CKP586 CKP587 CKP588 CKP589 CKP590 CKP591 CKP592 CKP593 CKP594 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 α3 36 36 36 36 48 54 60 60 66 72 72 72 78 78 78 84 84 84 96 96 96 96 102 102 102 108 108 120 120 138 144 144 156 180 α4 690 690 714 858 690 690 786 786 834 882 1002 1026 834 906 1146 930 1074 1074 1170 1194 1266 1434 1242 1338 1530 1218 1314 1506 1554 2106 1506 1506 2226 2082 α5 3960 5760 4320 4560 5280 5700 7140 7320 8160 9240 10800 10560 8880 9600 12780 10320 12600 12720 14040 15360 15240 18600 15720 17280 19800 17400 18240 20640 20520 30120 21480 24480 33000 33480 α6 57200 59000 59720 83840 59540 61070 82760 84920 95450 109940 138020 136220 98870 112190 174830 122720 161600 163040 184820 201740 207860 276620 211910 237830 284990 224600 245120 296420 298940 474530 311900 349700 534200 560480 α7 468720 821520 519120 637560 594720 631260 933240 981120 1126440 1355760 1807680 1733760 1177260 1363320 2377620 1529640 2187360 2202060 2526720 2897160 2918160 4253760 2994600 3452400 4270980 3334800 3690960 4484760 4515000 7913220 4544400 5456640 9067800 9276960 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 1 1 1 Continued from previous page. Period ID Name α0 α1 α2 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 5 , CKP598 CKP595 CKP596, V 4 8 CKP597 6 × S2 S2 CKP599 CKP600, P1 × MM3 P1 × MM3 CKP602 CKP603, P1 × MM3 CKP604 CKP605 CKP606 3 -- 6 2 -- 12, CKP601 2 -- 13 P1 × MM3 CKP608 2 -- 11, CKP607 CKP609 CKP610 CKP611 CKP612 CKP613 CKP614 P1 × MM3 CKP615 2 -- 14 CKP616 CKP617 CKP618 CKP619 CKP620 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 18 18 18 18 18 18 18 18 18 18 α3 288 0 24 42 60 66 72 78 84 84 90 90 96 108 108 108 108 114 114 114 120 126 204 48 90 102 108 114 120 120 120 132 138 α4 2994 1296 1296 720 1344 936 1056 1080 1104 1152 1176 1200 1632 1248 1488 1488 1632 1488 1488 1512 1488 1752 3264 1494 1302 1398 1542 1542 1878 1878 2022 1926 2166 α5 58440 0 4320 4920 11520 8280 9840 11040 11400 12600 12900 13440 19320 15600 18600 18960 20640 19440 19740 23640 19440 23940 52680 9120 13260 16200 18180 19200 23400 25200 26160 25800 30240 α6 1220900 160000 163240 58390 192940 97630 122920 138490 137860 162700 164590 175750 302200 188260 261700 267460 309220 268990 275470 303190 268360 355570 952180 206820 168570 212670 249480 262890 351180 379980 421020 388800 478170 1 1 2 E L A N A K A L A S H N K O V I α7 21414960 0 840000 567840 2347800 1086540 1428000 1725780 1685040 2132760 2139060 2332680 4447800 2538480 3797640 3922800 4640160 3973620 4077780 5285700 3894240 5509980 18086880 1864800 2089080 2919420 3486420 3780420 5323080 6032880 6607440 6041280 7777560 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 CKP621 CKP622 CKP623 CKP624 5 × S2 CKP625, S2 CKP626, P2 × S2 CKP627 CKP628 CKP629 5 4 CKP630 Str3 CKP631 8 × S2 4 3 -- 3, CKP635 CKP633, S2 CKP634 P1 × MM3 CKP636 CKP637 CKP638 CKP639 CKP640, V 4 6 CKP641 CKP642 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 18 18 18 18 18 18 18 20 20 20 20 20 20 20 20 20 20 20 22 22 22 22 22 22 22 22 22 22 22 22 24 24 24 α3 144 156 156 156 174 192 228 60 102 120 120 126 144 156 156 168 168 198 102 120 132 144 162 168 186 186 192 246 264 264 0 36 72 α4 2118 2190 2310 2358 2454 2862 2934 1140 1188 1668 1860 1908 2148 2340 2700 2580 2580 3228 1434 1914 2058 2394 2490 2634 3090 3354 3258 4290 4122 4554 3240 3240 3288 α5 30960 32760 33240 34920 38880 46440 55320 9120 11760 21120 23280 24480 31800 34080 41040 38400 39600 52260 13740 23280 24360 34200 34260 38040 47880 52980 51720 74280 77880 82200 0 10800 21600 α6 481860 513720 537120 564120 636030 802980 969840 121700 123050 303320 342200 361010 505280 540740 697700 629120 648920 925130 160510 347980 345280 557140 531490 613660 824530 960970 914620 1433830 1476220 1613740 672000 680100 720600 α7 7971600 8536080 8919960 9502920 11007780 14515200 18061680 1377600 1391880 4519200 5115600 5470920 8329440 8942640 12503400 10709160 11239200 17075100 1881180 5206320 4867800 9241680 8504160 10263120 14728980 17852100 16742880 28650720 29789760 33027120 0 3528000 7101360 Continued on next page. - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 1 1 3 Continued from previous page. Name 7 -- 1, P1 × P1 × S2 4 , P1 × MM3 CKP643 7 × S2 S2 CKP645 CKP646, P1 × MM3 4 , CKP644 2 -- 9 CKP647 CKP648 CKP649 CKP650 6 × S2 4 CKP651 CKP652, S2 CKP653 CKP654 CKP655 CKP656 CKP657 CKP658 CKP659 CKP660 4 , CKP662 2 -- 10, CKP663 CKP661 5 × S2 S2 P1 × MM3 CKP664 CKP665 CKP666 CKP667 CKP668 P1 × V 3 14 Period ID 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 α0 α1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 α2 24 24 24 24 24 24 24 24 24 24 26 26 26 26 26 26 28 28 28 28 28 28 30 30 30 30 30 32 32 32 34 34 34 α3 96 102 144 174 186 192 192 234 264 264 72 108 216 246 288 396 240 258 288 306 342 432 126 216 240 300 372 318 384 384 312 390 498 α4 1704 1704 3480 2784 3144 3048 3192 3648 4632 5352 3534 1998 4302 4302 5166 6222 3996 4764 5484 5580 6540 9660 2658 3858 4338 6690 7314 6144 7728 8112 5910 8694 10278 α5 14400 15720 46920 37680 47280 45840 48120 60780 83040 101040 22320 19080 72480 72120 102960 151080 62400 82200 100800 104100 129540 210240 27720 54000 66960 124920 153720 113280 157800 167520 97920 179520 245040 α6 193920 205530 909600 578490 804390 757680 816000 1060350 1611960 2040360 787580 270440 1371500 1339550 2038580 3168440 1067680 1573390 2038960 2099350 2770570 5004640 439590 891660 1182900 2778600 3385200 2304770 3492320 3766640 1820140 4180750 5923330 1 1 4 E L A N A K A L A S H N K O V I α7 2150400 2452380 16450560 9059820 14118720 13077120 14306040 19603500 32664240 43219680 7514640 3435600 27676320 25814460 44530080 74446680 19007520 31316460 42887040 44273880 61901700 126134400 6247500 14726880 21408240 61790400 79195200 48799800 80806320 88438560 34520640 100127580 153543600 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 CKP669 CKP670 CKP671 2 -- 7 CKP672, P1 × MM3 CKP673 CKP674 CKP675, S2 CKP676 CKP677 CKP678 4 × S2 4 2 -- 6, CKP680 CKP679 P1 × MM3 CKP681 V 4 4 , CKP682 CKP683 CKP684 P1 × V 3 12 CKP685 CKP686 3 , CKP687 P2 × S2 CKP688 CKP689 CKP690 CKP692, S2 CKP693 CKP694 P1 × P1 × S2 MM3 CKP696, S2 8 -- 1 8 × S2 3 3 , CKP695, P1 × 7 × S2 3 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 36 36 36 36 36 36 38 38 38 40 44 44 44 44 44 46 46 48 48 48 50 50 52 52 54 54 54 54 56 56 56 58 58 α3 336 360 396 456 552 768 348 384 396 192 516 636 696 744 888 528 714 0 216 660 600 792 696 1044 498 528 744 888 498 528 600 492 498 α4 6708 7188 7572 9876 12852 18996 6954 8106 8010 4776 11580 15804 17388 18396 23052 11826 18618 15120 15408 15552 13758 21078 17412 29124 9882 11178 19194 24378 10536 11832 14424 11214 11214 α5 119520 134400 143160 214680 304080 500640 117840 156480 150600 59520 248880 393480 445680 492360 649200 238560 496560 0 151320 367320 288480 635760 424440 874080 162000 207720 481680 677520 171900 217920 317100 178440 181800 α6 2419200 2795400 2921580 5072760 7828200 14713200 2268560 3390500 3136160 1120000 5903540 10666340 12371480 14028200 19904120 5341780 14203810 7392000 7959000 9396300 6659420 18069260 11365000 28285540 2938770 4427820 13279500 20447820 3240110 4748600 7961600 3502120 3561250 α7 50507520 60459840 62324640 125137740 210966000 450203040 44336040 76130880 67735080 19138560 145945800 301939680 359059680 419215440 635293680 122340960 428469300 0 117482400 251895000 157802400 600739440 317604000 956113200 54057780 98491680 381906000 644873040 60897480 106293600 207233040 65938320 68151720 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I Continued on next page. 1 1 5 Continued from previous page. Period ID Name α0 α1 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 3 , CKP697 5 × S2 3 6 × S2 S2 CKP698 CKP699 CKP700, S2 CKP701 CKP702 P1 × MM3 CKP704 CKP705 4 × S2 S2 CKP707 CKP708 CKP709 CKP710 P1 × V 3 10 2 -- 5, CKP703 3 , CKP706 CKP711 2 -- 4, CKP713 CKP712 P1 × MM3 CKP714 CKP715 CKP716 CKP717 CKP718 CKP719 CKP720, S2 CKP721 CKP722 3 × S2 3 CKP723, P1 × V 3 CKP724 8 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 α2 58 60 60 60 64 66 66 68 68 68 74 78 78 78 80 80 84 84 90 90 92 92 92 102 102 102 102 104 108 128 138 150 154 168 α3 888 504 1068 1212 522 852 1356 816 852 1320 588 1140 1176 1680 1212 1320 1932 2148 1788 2040 1518 1626 2112 1950 2274 2688 3408 2472 984 2976 4650 4866 3840 4752 α4 23694 11916 30156 35916 13392 21510 47574 21012 22308 43236 17550 32706 34002 60066 36240 38688 69636 77316 59886 76014 47172 51492 83820 67002 84330 106410 146250 97944 37260 120960 222918 241002 159486 219624 α5 632400 195120 893280 1134480 225720 504000 1614240 465960 520680 1421040 319560 877320 937080 2142720 1020360 1078320 2622480 3051480 2032920 2873160 1357680 1574580 3281280 2266320 3207480 4495680 6695280 3940320 848880 4959840 11448480 12623040 6504960 10383840 α6 18393340 3962040 28423860 38512860 4887190 13009080 58420920 11662880 13640900 51100520 7862600 26208960 28577940 82424580 31974020 32604200 106446900 128188740 74950920 117404820 42774050 52448150 141863600 83881470 132223890 203447460 334814340 171825080 26609400 221633120 632940330 711272850 284808340 531501360 1 1 6 E L A N A K A L A S H N K O V I α7 559525680 78104880 948659040 1368087000 102194400 347891040 2223985680 297392760 368091360 1914785040 185440080 814453920 909170640 3324124440 1043489160 1016215200 4526098920 5649930720 2894154480 5023514160 1385508600 1816414320 6368328960 3245543280 5710371660 9658434240 17506424880 7840793520 804368880 10369947840 36647730000 42024975300 12889551360 28511659680 Continued on next page. Continued from previous page. Period ID Name α0 α1 α2 α3 α4 α5 α6 α7 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 CKP725 CKP726 CKP727, P1 × V 3 CKP728 CKP729 CKP730 6 CKP731 4 , CKP732 P1 × V 3 CKP733 CKP734 CKP735 CKP736 CKP737 CKP738 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 184 224 272 398 420 444 468 540 1040 1386 1946 1992 2136 2664 6804 12816 99000 5688 9312 13560 17616 19992 22404 24852 37632 105984 166284 215808 227472 262896 466368 2040912 5435904 130800000 286008 580704 952176 1221810 1488708 1771596 2065764 3836268 15564048 28575342 35318526 38459880 48275736 115475112 852143652 3188239632 233995275000 14876160 38555520 73148160 85572960 114603120 146305440 180367920 420664320 2472668160 5322513240 5981882880 6796332000 9412519800 31137505920 389608626240 2051802731520 462392774925120 982577026659240000 2197113382189414080000 49505030400 206084027520 516454715280 493612489440 824518956240 1221757064640 1685867765400 6131551910400 75673543680000 223880895211680 200205416839680 252711084477600 435882277192320 2746619333498880 98894833331535360 1032164932439531520 837897160 2752140320 5996559080 6386359700 9497959800 13047797460 17014559940 49565795760 422070022400 1065056580360 1074550170260 1282447706160 1975803279600 9021039724800 191430924575040 1419118168838400 - I F O U R D M E N S I O N A L F A N O I Q U V E R F L A G Z E R O L O C I 1 1 7 118 ELANA KALASHNIKOV References [1] Code repository. bitbucket.org/fanosearch/magma-core, 2018. [2] Code and data repository. bitbucket.org/fanosearch/db, 2018. [3] M. F. Atiyah and R. Bott. The moment map and equivariant cohomology. Topology, 23(1):1 -- 28, 1984. [4] K. Behrend and B. Fantechi. The intrinsic normal cone. Invent. Math., 128(1):45 -- 88, 1997. [5] W. Bosma, J. Cannon, and C. Playoust. The Magma algebra system. I. The user language. J. Symbolic Comput., 24(3-4):235 -- 265, 1997. Computational algebra and number theory (London, 1993). [6] I. Ciocan-Fontanine, B. Kim, and C. Sabbah. The abelian/nonabelian correspondence and Frobenius manifolds. Invent. Math., 171(2):301 -- 343, 2008. [7] T. Coates. The quantum Lefschetz principle for vector bundles as a map between Givental cones. arXiv:1405.2893 [math.AG], 2014. [8] T. Coates, A. Corti, S. Galkin, V. Golyshev, and A. Kasprzyk. Mirror symmetry and Fano manifolds. In European Congress of Mathematics, pages 285 -- 300. Eur. Math. Soc., Zurich, 2013. [9] T. Coates, A. Corti, S. Galkin, and A. Kasprzyk. Quantum periods for 3-dimensional Fano manifolds. Geom. Topol., 20(1):103 -- 256, 2016. [10] T. Coates, S. Galkin, A. Kasprzyk, and A. Strangeway. Quantum periods for certain four-dimensional Fano manifolds. Experimental Mathematics, 0(0):1 -- 39, 2018. [11] T. Coates and A. Givental. Quantum Riemann-Roch, Lefschetz and Serre. Ann. of Math. (2), 165(1):15 -- 53, 2007. [12] T. Coates, A. Kasprzyk, and T. Prince. Four-dimensional Fano toric complete intersections. Proc. A., 471(2175):20140704, 14, 2015. [13] A. Craw. Quiver flag varieties and multigraded linear series. Duke Math. J., 156(3):469 -- 500, 2011. [14] A. Craw, Y. Ito, and J. Karmazyn. Multigraded linear series and recollement. Mathematische Zeitschrift, 289(1):535 -- 565, Jun 2018. [15] A. Craw and G. G. Smith. Projective toric varieties as fine moduli spaces of quiver representations. American Journal of Mathematics, 130(6):1509 -- 1534, 2008. [16] I. V. Dolgachev and Y. Hu. Variation of geometric invariant theory quotients. Inst. Hautes ´Etudes Sci. Publ. Math., (87):5 -- 56, 1998. With an appendix by Nicolas Ressayre. [17] W. Fulton. Young tableaux, volume 35 of London Mathematical Society Student Texts. Cambridge University Press, Cambridge, 1997. With applications to representation theory and geometry. [18] A. Givental. A mirror theorem for toric complete intersections. In Topological field theory, primitive forms and related topics (Kyoto, 1996), volume 160 of Progr. Math., pages 141 -- 175. Birkhauser Boston, Boston, MA, 1998. [19] A. B. Givental. Symplectic geometry of Frobenius structures. In Frobenius manifolds, Aspects Math., E36, pages 91 -- 112. Friedr. Vieweg, Wiesbaden, 2004. [20] V. A. Iskovskih. Fano threefolds. I. Izv. Akad. Nauk SSSR Ser. Mat., 41(3):516 -- 562, 717, 1977. [21] V. A. Iskovskih. Fano threefolds. II. Izv. Akad. Nauk SSSR Ser. Mat., 42(3):506 -- 549, 1978. [22] V. A. Iskovskih. Anticanonical models of three-dimensional algebraic varieties. In Current problems in mathe- matics, Vol. 12 (Russian), pages 59 -- 157, 239 (loose errata). VINITI, Moscow, 1979. [23] B. Kim, A. Kresch, and T. Pantev. Functoriality in intersection theory and a conjecture of Cox, Katz, and Lee. J. Pure Appl. Algebra, 179(1-2):127 -- 136, 2003. [24] A. D. King. Moduli of representations of finite-dimensional algebras. Quart. J. Math. Oxford Ser. (2), 45(180):515 -- 530, 1994. [25] M. Kontsevich. Enumeration of rational curves via torus actions. In The moduli space of curves (Texel Island, 1994), volume 129 of Progr. Math., pages 335 -- 368. Birkhauser Boston, Boston, MA, 1995. [26] J. Levitt. On embeddings of Mori dream spaces. Geom. Dedicata, 170:281 -- 288, 2014. [27] J. Li and G. Tian. Virtual moduli cycles and Gromov-Witten invariants of algebraic varieties. J. Amer. Math. Soc., 11(1):119 -- 174, 1998. [math.SG], 2000. [28] S. Martin. Symplectic quotients by a nonabelian group and by its maximal torus. arXiv:math/0001002 [29] S. Mori and S. Mukai. Classification of Fano 3-folds with B2 ≥ 2. Manuscripta Math., 36(2):147 -- 162, 1981/82. [30] S. Mori and S. Mukai. On Fano 3-folds with B2 ≥ 2. In Algebraic varieties and analytic varieties (Tokyo, [31] S. Mori and S. Mukai. Classification of Fano 3-folds with B2 ≥ 2. I. In Algebraic and topological theories [32] S. Mori and S. Mukai. Erratum: "Classification of Fano 3-folds with B2 ≥ 2". Manuscripta Math., 110(3):407, 1981), volume 1 of Adv. Stud. Pure Math., pages 101 -- 129. North-Holland, Amsterdam, 1983. (Kinosaki, 1984), pages 496 -- 545. Kinokuniya, Tokyo, 1986. 2003. [33] S. Mori and S. Mukai. Extremal rays and Fano 3-folds. In The Fano Conference, pages 37 -- 50. Univ. Torino, Turin, 2004. [34] R. Pandharipande. Rational curves on hypersurfaces (after A. Givental). Ast´erisque, (252):Exp. No. 848, 5, 307 -- 340, 1998. S´eminaire Bourbaki. Vol. 1997/98. [35] D. M. Snow. On the ampleness of homogeneous vector bundles. Transactions of the American Mathematical Society, 294(2):585 -- 594, 1986. [36] The Sage Developers. SageMath, the Sage Mathematics Software System (Version 8.4.0), 2018. http://www.sagemath.org. [37] R. Webb. The Abelian-Nonabelian correspondence for I-functions. arXiv:1804.07786 [math.AG], 2018. FOUR-DIMENSIONAL FANO QUIVER FLAG ZERO LOCI 119 Department of Mathematics, Imperial College London, 180 Queen's Gate, London SW7 2AZ, UK E-mail address: [email protected]
1710.05902	2	1710	2019-01-31T06:10:19	Trace and Kunneth formulas for singularity categories and applications	[ "math.AG", "math.CT", "math.KT" ]	We present an $\ell$-adic trace formula for saturated and admissible dg-categories over a base monoidal dg-category. Moreover, we prove K\"unneth formulas for dg-category of singularities, and for inertia-invariant vanishing cycles. As an application, we prove a version of Bloch's Conductor Conjecture (stated by Spencer Bloch in 1985), under the additional hypothesis that the monodromy action of the inertia group is unipotent.	math.AG	math	Trace and Kunneth formulas for singularity categories and applications Bertrand Toen∗ and Gabriele Vezzosi January 2019 Abstract We present an ℓ-adic trace formula for saturated and admissible dg-categories over a base monoidal dg-category. Moreover, we prove Kunneth formulas for dg-category of singularities, and for inertia- invariant vanishing cycles. As an application, we prove a version of Bloch's Conductor Conjecture (stated by Spencer Bloch in 1985), under the additional hypothesis that the monodromy action of the inertia group is unipotent. Contents 1 Introduction 2 A non-commutative trace formalism 2.1 ∞-Categories of dg-categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The ℓ-adic realization of dg-categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 The ℓ-adic Chern character . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Trace formula for dg-categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Invariant vanishing cycles 3.1 Trivializing the Tate twist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Reminders on actions of the inertia group . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 . . . . . . . . . . . . . . . . . . . . . . . . 3.4 A Kunneth theorem for invariant vanishing cycles . . . . . . . . . . . . . . . . . . . . . Invariant vanishing cycles and dg-categories 4 Kunneth formula for dg-categories of singularities 4.1 The monoidal dg-category B and its action . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Kunneth formula for dg-categories of singularities . . . . . . . . . . . . . . . . . . . . . 4.3 Saturatedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 6 6 8 11 12 16 17 17 20 23 28 28 32 36 ∗Partially supported by ERC-2016-ADG-741501 and ANR-11-LABX-0040-CIMI within the program ANR-11-IDEX- 0002-02. 1 5 Application to Bloch's conductor formula with unipotent monodromy 5.1 Bloch's Conductor Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 An analog of Bloch's Conjecture for unipotent monodromy . . . . . . . . . . . . . . . . A Localizations of monoidal dg-categories 1 Introduction 37 38 39 44 Motivated by applications to arithmetic geometry, the first one being presented in this paper, we de- velop methods of derived and non-commutative geometry on a mixed-characteristic base. Though both derived and non-commutative geometry have been mainly used so far over a base field (possibly of positive characteristic), their flexibility is definitely wider, as we try to demonstrate in this paper. More precisely, we prove and investigate a trace formula for dg-categories, and Kunneth formulas for dg- categories of singularities and for inertia-invariant vanishing cycles, and we give an application of these results to a categorical version of Bloch's Conductor Conjecture (BCC) in the presence of unipotent monodromy. We are convinced that the tools developed in this paper will have other applications to arithmetic geometry, some of which are already under investigation by the authors. Another seemingly more flexible tool, that of integration maps from K-theory of dg-categories, will be developed in a forthcoming paper and applied to more general cases of the original BCC. In this paper, we present four main results. As a first step, in Section 2, we prove a quite general trace formula for dg-categories (or non-commutative schemes). This is done by using the non-commutative ℓ-adic realization functor rℓ recently introduced in [BRTV] as a functor from dg-categories over an excel- lent discrete valuation ring A to ℓ-adic sheaves on Spec A. Via rℓ, we first introduce a ℓ-adic version of the Chern character for dg-categories over A, as a lax-monoidal natural transformation Chℓ : HK → rℓ, where HK is the non-connective, homotopy invariant K-theory functor on dg-categories, and rℓ is the Eilenberg-Mac Lane construction applied to the derived global sections of the ℓ-adic realization functor rℓ. We prove the following result (Theorem 2.4.9 in the paper). Theorem A (Trace formula for dg-categories). Let B be a monoidal dg-category over A. For T a dg-category which is a (left) B-module, satisfying both a version of smoothness and properness over B and an admissibility property1 with respect to rℓ, a trace formula holds Chℓ([HH(T /B; f )]) = trrℓ(B)(rℓ(f )) (1) for any endomorphism f : T → T of B-modules. Here [HH(T /B; f )] is the class in HK0(HH(B/A)) induced by the endomorphism f , and HK0(HH(B/A)) is the 0-th K-theory of Hochschild homology of B over A. The r.h.s. trrℓ(B)(rℓ(f )) is the trace2 of the endo- morphism rℓ(f ), and the trace formula (1) is an equality in π0(rℓ(HH(B/A))) ≃ H 0(S´et, rℓ(HH(B/A))). Note that the extra generality of working over B (instead of over A, or over an E∞- algebra over A) is actually needed for applications, as shown in Section 5: the ℓ-adic realization of a dg-category is 1The ℓ-adic realization functor is only lax-monoidal, and we say that T is rℓ-admissible over B if rℓ behaves like a symmetric monoidal functor for T over B (see Definition 2.4.8). 2Or rather its image under the canonical map α : H 0(S´et, HH(rℓ(B)/rℓ(A))) → H 0(S´et, rℓ(HH(B/A))). 2 it is a module over Qℓ(β) := ⊕n∈ZQℓ(2n), so it has no chance of being smooth always 2-periodic, i.e. and proper over A itself, unless it is trivial; in particular, no reasonable trace formula is available over A. Our second main result (Section 3) is a Kunneth type formula for inertia-invariant vanishing cycles. Here we push forward the investigation begun in [BRTV] about the relation between vanishing cycles and the ℓ-adic realization of the dg-category of singularities. For X and Y regular schemes, endowed with a flat, proper and generically smooth map to the strictly henselian excellent trait S = Spec A, we define an inertia-invariant convolution product (E ⊛ F )I, for E an ℓ-adic sheaf on the special fiber Xs, and F an ℓ-adic sheaf on the special fiber Ys (I := Gal( ¯K/K) denoting the inertia group in this situation). If we denote by νX (respectively, νY ) the complex of vanishing cycles for X/S (respectively, Y /S), we prove the following result (Theorem 3.4.2 in the paper). Theorem B (Kunneth formula for inertia-invariant vanishing cycles). We have equivalences (νX ⊛ νY )I ≃ rℓ(Sing(X ×S Y )) ≃ Cofib(ηX×SY : Qℓ(β) −→ ωX×S Y (β)). where Sing(X ×S Y ) = Cohb(X ×S Y )/Perf(X ×S Y ) is the dg-category of singularities of the fiber product X ×S Y , and ηX×S Y is the (2-periodized) ℓ-adic fundamental class. This is our Kunneth formula for inertia-invariant vanishing cycles, and it seems to be a new result in the theory of vanishing cycles, especially in the mixed characteristic case. Note that it might also be viewed as a Thom-Sebastiani formula for inertia-invariant vanishing cycles but we would like to stress that it is not a consequence of the usual Thom-Sebastiani formula for vanishing cycles (see [Il2]), and that it holds only if inertia invariants are taken into account in defining the convolution (E ⊛ F )I3. We also discuss appropriate conditions ensuring that (νX ⊛νY )I is equivalent to (νX ⊠νY )I (Corollary 3.4.5). The third main result of this paper is a Kunneth type formula for dg-categories of singularities (Section 4). For X and Y regular schemes, endowed with a flat, proper and generically smooth map to the strictly henselian excellent trait S = Spec A, we may consider the dg-categories of singularities Sing(Xs) := Cohb(Xs)/Perf(Xs) and Sing(Ys) := Cohb(Xs)/Perf(Xs) of the corresponding special fibers. Consider B := Sing(s ×S s) (s being the closed point in S, and s ×S s being the derived fiber product); then B is a monoidal dg-category for the convolution product coming from the derived groupoid struc- ture of s ×S s. Moreover, B acts on both Sing(Xs) and Sing(Ys), in such a way that the tensor product Sing(Xs)o ⊗B Sing(Ys) makes sense as a dg-category over A (Proposition 4.1.7). Our Kunneth formula for dg-categories of singularities is then the following result (Theorem 4.2.1 in the paper). Theorem C (Kunneth formula for dg-categories of singularities). There is a canonical equiv- alence Sing(Xs)o ⊗B Sing(Ys) ≃ Sing(X ×S Y ) as dg-categories over A. it is false if we replace Sing by Cohb. Also Note that this result is peculiar to singularity categories: notice that B is defined as the convolution dg-category of a derived groupoid, and is an object in derived algebraic geometry that cannot be described within classical algebraic geometry. Finally, we prove that 3In particular, only (E ⊛ F )I, and not (E ⊛ F ), makes sense in our context. 3 Sing(Xs) is smooth and proper (i.e. saturated) over B, for any X regular scheme, endowed with a flat, proper and generically smooth map to S. This fact is well-known in characteristic zero (see, for instance, [Pr]) but is, in our opinion, a deep and surprising result in our general setting. Smoothness and properness over B is one half of the properties needed to apply the trace formula to Sing(Xs) over B. Note that, without further hypothesis, it is not true that Sing(Xs) is also admissible (with respect to rℓ). Our fourth and final main result is a categorical version of Bloch's Conductor formula for unipotent monodromy (section 5). In his seminal 1987 paper [Bl], S. Bloch introduced what is now called Bloch's intersection number [∆X , ∆X]S, for a flat, proper map of schemes X −→ S where X is regular, and S is a henselian trait S. This number can be defined as the degree of the localized top Chern class of the coherent sheaf Ω1 X/S and measures the "relative" singularities of X over S. In the same paper, Bloch introduced his famous conductor formula, which can be seen as a conjectural computation of the Bloch's intersection number in terms of the arithmetic geometry of X/S. It reads as follows. Bloch's Conductor Conjecture (BCC) We have an equality [∆X , ∆X]S = χ(X¯k) − χ(X ¯K) − Sw(X ¯K), where X¯k and X ¯K denotes the special and generic geometric fibers of X over S, χ(−) denotes Qℓ-adic Euler characteristic, and Sw(−) is the Swan conductor. In [Bl] the above formula is proven in relative dimension 1. Further results implying special cases of BCC have been obtained since then by Kato-Saito [Ka-Sa] and others, the most recent one being a full proof in the geometric case by T. Saito [Sai] (see section §5 for a more detailed discussion about the status of the art about BCC). In the mixed characteristic case, the conjecture is open in general outside the cases covered in [Ka-Sa]. In particular, for isolated singularities the above conjecture already appeared in Deligne's expos´e [SGA7-I, Exp. XVI], and remains open. In Section 5 we propose a first step towards a new understanding of Bloch's conductor formula using the results developed in the previous sections. We start by defining an analog of Bloch's intersection number, that we call the categorical Bloch's intersection class (Definition 5.2.1), and denote it by [∆X , ∆X]cat S . It is an element in H 0(S´et, rℓ(HH(B/A))), where B = Sing(s ×S s) (as in Section 4, see above), and it is basically defined as an intersection class in the setting of non-commutative algebraic geometry. The precise comparison with the original Bloch's number is not covered in this work and will appear in a forthcoming paper. It is easy to see that there are canonical inclusions Z ֒→ Qℓ ֒→ H 0(S´et, HH(rℓ(B)/rℓ(A))), and for any λ ∈ Z, we will write λ∧ its image under the canonical map α : H 0(S´et, HH(rℓ(B)/rℓ(A))) → H 0(S´et, rℓ(HH(B/A))). Our fourth main result is then the following theorem (Theorem 5.2.2 in the paper). Theorem D (Categorical BCC for unipotent monodromy) With the notations of BCC, if the monodromy action of the inertia on H ∗(X ¯K, Qℓ) is unipotent, then we have an equality [∆X , ∆X]cat S = χ(X¯k)∧ − χ(X ¯K)∧. 4 Since unipotent monodromy action implies tame action, the Swan conductor vanishes for unipotent monodromy, so that Theorem D is completely analogous to BCC under this hypothesis. We believe the main ideas in our proof of Theorem D are new in the subject, and might be also useful to answer other related questions in algebraic and arithmetic geometry. The key point in the proof of Theorem D is that it is a direct consequence of the trace formula for dg-categories (Theorem A). By the main result of [BRTV], the ℓ-adic realization of the dg-category of singularities Sing(Xs) of the special fiber is the inertia invariant part of vanishing cohomology of X/S (suitably 2-periodized). Moreover, our Kunneth formulas for inertia-invariant vanishing cycles (Section 3), and for dg-categories of sin- gularities (Section 4), imply that Sing(Xs) is smooth and proper over B. Finally, again our Kunneth formulas together with the hypothesis of unipotent monodromy show that Sing(Xs) is also rℓ-admissible, so that we may apply our trace formula to the identity endomorphism of Sing(Xs) over B: Theorem D is then an immediate corollary of the trace formula applied to the identity map of Sing(Xs). We conclude this introduction by remarking that the hypothesis of unipotent inertia action in Theo- rem D is, of course, a bit restrictive: unipotent monodromy implies tame monodromy and our theorem does not deal with the interesting arithmetic aspects encoded by the Swan conductor. However we are convinced that we can go beyond the unipotent case, and prove some new cases of BCC con- jecture (possibly the whole unrestricted conjecture), by considering Sing(X ×S S′) as a module over B′ := Sing(S′ ×S S′), where S′ = Spec A′ → S is a totally ramified Galois extension of strictly henselian traits such that the inertia for S′ acts unipotently on H ∗((X ×S S′) ¯K ′, Qℓ). Here K ′ is the fraction field of A′, and the existence of such an extension S′ → S is guaranteed by Grothendieck local monodromy theorem ([SGA7-I, Exp. I, Th´eor`eme 1.2]). This strategy for BCC, as well as the comparison between the categorical and the classical Bloch numbers, will be investigated in a future work. Acknowledgments. We thank A. Abb`es, A. Blanc, M. Robalo and T. Saito, for various helpful conversations on the topics of this paper. We also thank all the french spiders and the south-italy sea waves in the 2017 summer nights, for keeping the authors company while they were, separately, writing down a first version of this manuscript. This work is part of a project that has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation Programme (Grant Agreement No 741501). Notations. • Throughout the text, A will denote a (discrete) commutative noetherian ring. When needed, A will be required to satisfy further properties (such as being excellent, local and henselian) that will be made precise in due course. • L(A) will denote the A-linear dg-category of (fibrant and) cofibrant A-dg-modules localized with respect to quasi-isomorphisms. • dgCatA will denote the Morita ∞-category of dg-categories over A (see §2.1). • Top denotes the ∞-category of spaces (obtained, e.g. as the coherent nerve of the Dwyer-Kan localization of the category of simplicial sets along weak homotopy equivalences). Sp denotes the ∞-category of spectra. 5 • If R is a an associative and unital monoid in a symmetric monoidal ∞-category C, we will write Mod(R) or ModC(R) for the ∞-category of left R-modules in C ([Lu-HA]). 2 A non-commutative trace formalism 2.1 ∞-Categories of dg-categories We denote by A a commutative ring. We remind here some basic facts about the ∞-category of dg- categories, its monoidal structure and its theory of monoids and modules. We consider the category dgCatA of small A-linear dg-categories and A-linear dg-functors. We remind that an A-linear dg-functor T −→ T ′ is a Morita equivalence if the induced functor of the corresponding derived categories of dg-modules f ∗ : D(T ′) −→ D(T ) is an equivalence of categories (see [To1] for details). The ∞-category of dg-categories over S is defined to be the localisation of dgCatA along these Morita equivalences, and will be denoted by dgCatS or dgCatA. Being the ∞-category associated to a model category, dgCatA is a presentable ∞-category. As in [To1, § 4], the tensor product of A-linear dg-categories can be derived to a symmetric monoidal structure on the ∞-category dgCatA. This symmetric monoidal structure moreover distributes over colimits making dgCatA into a presentable symmetric monoidal ∞-category. We have a notion of rigid, or, equivalently, dualizable, object in dgCatA. It is a well known fact that dualizable objects in dgCatA are precisely smooth and proper dg-categories over A (see [To2, Prop. 2.5]). The compact objects in dgCatA are the dg-categories of finite type over A in the sense of [To-Va]. We denote their full sub-∞-category by dgCatf t S is preserved by the monoidal structure, and moreover any dg-category is a filtered colimit of dg-categories of finite type. We thus have a natural equivalence of symmetric monoidal ∞-categories A ⊂ dgCatA. The full sub-category dgCatf t dgCatA ≃ Ind(dgCatf t S ). We will from time to time have to work in a bigger ∞-category, denoted by dgCATA, that contains dgCatA as a non-full sub-∞-category. By [To2], we have a symmetric monoidal ∞-category dgCatlp A of presentable dg-categories over A. We define dgCATA the full sub-∞-category of dgCatlp A consisting of all compactly generated dg-categories. The ∞-category dgCatA can be identified with the non-full sub-∞-category of dgCATA which consists of compact objects preserving dg-functors. This provides a faithful embedding of symmetric monoidal ∞-categories dgCatA ֒→ dgCATA. At the level of objects, this embedding sends a small dg-category T to the compactly generated dg- category T of dg-modules over T o. An equivalent description of dgCATA is as the ∞-category of small dg-categories together with the mapping spaces given by the classifying space of all bi-dg-modules be- tween small dg-categories. Definition 2.1.1 A monoidal A-dg-category is a unital and associative monoid in the symmetric monoidal ∞-category dgCatA. A module over a monoidal A-dg-category B will, by definition, mean a left B-module in dgCatA in the sense of [Lu-HA], and the ∞-category of (left) B-modules will be denoted by dgCatB. 6 For a B-module T , we have a morphism µ : B ⊗A T → T in dgCatA, that will be simply denoted by (b, x) 7→ b ⊗ x. For a monoidal A-dg-category B, we will denote by B⊗-op the monoidal A-dg-category where the monoid structure is the opposite to the one of B, i.e. b ⊗op b′ := b′ ⊗ b. Note that B⊗-op should not be confused with Bo (which is still a monoidal A-dg-category), where the "arrows" and not the monoid structure have been reversed, i.e. Bo(b, b′) := B(b′, b). By definition, a right B-module is a (left) B⊗-op-module. The ∞-category of right B-modules will be denoted by dgCatB⊗-op, or simply by dgCatB. If B is a monoidal A-dg-category, then B⊗-op ⊗A B is again a monoidal A-dg-category, and B can be considered either as a left B⊗-op ⊗A B (denoted by BL), or as a right B⊗-op ⊗A B-module (denoted by BR). For T a B-module, and T ′ a right B-module, then T ′ ⊗A T is naturally a right B⊗-op ⊗A B-module, and we define which is an object in dgCatA. T ′ ⊗B T := (T ′ ⊗A T ) ⊗B⊗-op⊗AB BL Let B be a monoidal A-dg-category. We can consider the symmetric monoidal embedding dgCatA ֒→ dgCATA, so that the image bB of B is a monoid in dgCATA. The ∞-category of bB-modules in dgCATA is denoted by dgCATB, and its objects are called big B-modules. The natural ∞-functor dgCatB −→ dgCATB is faithful and its image consists of all big B-modules bT such that the morphism bB ⊗AbT −→ bT is a small morphism. It is known (see [To2]) that the symmetric monoidal ∞-category dgCATA is rigid, that for any bT its dual is given by cT o, and that the evaluation and coevaluation morphisms are defined by T considered as T o ⊗A T -module. This formally implies that if bT is a big B-module, then its dual cT o is naturally a right big B-module. We thus have two big morphisms These morphisms also provide a third big morphism µ : bB ⊗AbT −→ bT , µo : cT o ⊗A bB −→ cT o. h : cT ob⊗AbT −→ bB. The big morphism h is obtained by duality from the right adjoint to µ. We now make the following definitions. µ∗ : bT −→ bB ⊗AbT Definition 2.1.2 Let B be a monoidal dg-category, and T a B-module. We say that: 1. T is cotensored (over B) if the big morphism µo defined above is a small morphism (i.e. a morphism in dgCatA); 2. T is proper or enriched (over B) if the big morphism h defined above is a small morphism (i.e. a morphism in dgCatA). We can make the above definition more explicit as follows. Let B and T be as above; for two objects b ∈ B and x ∈ T , we can consider the dg-functor xb : T o −→ L(A) 7 sending y ∈ T to T (b ⊗ y), x) where (b, x) 7→ b ⊗ x : B ⊗A T −→ T is the B-module structure on T . Then, T is cotensored over B if and only if for all b and x the above dg-module xb : T o −→ L(A) is compact in the derived category D(T o) of all T o-dg-modules. When T is furthermore assumed to be triangulated, then this is equivalent to ask for the dg-module to be representable by an object xb ∈ T . In a similar manner, we can phrase properness of T over B by saying that, for any x ∈ T and y ∈ T , the dg-module Bo −→ L(A) sending b to T (b ⊗ x, y) is compact (or representable, if B is furthermore assumed to be triangulated). Remark 2.1.3 It is important to notice that, by definition, when T is cotensored, then the big right B-module cT o is in fact a small right B-module. Equivalently, when T is cotensored, then T o is naturally a right B-module inside dgCatA, with the right module structure given by the morphism in dgCatA sending (x, b) ∈ T o ⊗A B to the cotensor xb ∈ T o. µo : T o ⊗A B −→ T o We end this section by recalling some facts about existence of tensor products of modules over monoidal dg-categories. As a general fact, since dgCatA is a presentable symmetric monoidal ∞- category, for any monoidal dg-category B there exists a tensor product ∞-functor ⊗B : dgCatB × dgCatB −→ dgCatA, sending a left B-module T and a right B-module T ′ to T ′ ⊗B T (see [Lu-HA]). Now, if T is a B-module which is also cotensored (over B) in the sense of Definition 2.1.2, we have that T o is a right B-module, and we can thus form T o ⊗B T ∈ dgCatA. When T is not cotensored, the object T ⊗B T o does not exist anymore. However, we can always consider the presentable dg-categories bT and cT o as left and right modules over bB, respectively, and their tensor product cT ob⊗ bB bT now only makes sense as a presentable dg-category which has no reason to be compactly generated, in general. Of course, when T is cotensored, this presentable dg-category is compactly generated and we have \T o ⊗B T ≃ cT ob⊗ bB bT . Remark 2.1.4 The following, easy observation, will be useful in the sequel. Let B be a monoidal dg-category, and assume that B is generated, as a triangulated dg-category, by its unit object 1 ∈ B. Then, any big B-module is small (i.e. in the image of dgCatB −→ dgCATB), and also cotensored. 2.2 The ℓ-adic realization of dg-categories We denote by SHS the stable A1-homotopy ∞-category of schemes over S (see [Vo, Def. 5.7] and [Ro, § 2]). It is a presentable symmetric monoidal ∞-category whose monoidal structure will be denoted by ∧S. Homotopy invariant algebraic K-theory defines an E∞-ring object in SHS that we denote by BUS (a more standard notation is KGLS). We denote by ModSHS (BUS) the ∞-category of BUS-modules 8 in SHS. It is a presentable symmetric monoidal ∞-category whose monoidal structure will be denoted by ∧BUS . As proved in [BRTV], there exists a lax symmetric monoidal ∞-functor M− : dgCatS −→ ModSHS (BUS), which will be denoted by T 7→ MT (while it is denoted by T 7→ M∨ S(T ) in [BRTV]). The precise construction of the ∞-functor M− is rather involved and uses in an essential manner the theory of non- commutative motives of [Ro] as well as the comparison with the stable homotopy theory of schemes. Intuitively, the ∞-functor M− sends a dg-category T to the homotopy invariant K-theory functor S′ 7→ HK(S′ ⊗S T ). To be more precise, there is an obvious forgetful ∞-functor U : ModSHS (BUS) −→ Fun∞(Smop S , Sp), to the ∞-category of presheaves of spectra on the category SmS of smooth S-schemes. For a given dg-category T over S, the presheaf U(MT ) is defined by sending a smooth S-scheme S′ = Spec A′ → Spec A = S to HK(A′ ⊗A T ), the homotopy invariant non-connective K-theory spectrum of A′ ⊗A T (see [Ro, 4.2.3]). The ∞-functor M− satisfies some basic properties which we recall here. 1. The ∞-functor M− is a localizing invariant, i.e. for any short exact sequence T0 ֒→ T −→ T /T0 of dg-categories over A, the induced sequence MT0 / MT / MT /T0 exhibits MT0 has the fiber of the morphism MT → MT /T0 in Mod(BUS). 2. The natural morphism BUS −→ MA, induced by the lax monoidal structure of M−, is an equiva- lence of BUS-modules. 3. The ∞-functor T 7→ MT commutes with filtered colimits. 4. For any quasi-compact and quasi-separated scheme X, and any morphism p : X −→ S, we have a natural equivalence of BUS-modules MPerf(X) ≃ p∗(BUX), where p∗ : ModSHX (BUX) −→ ModSHS (BUS) is the direct image of BU-modules, and Perf(X) is the dg-category of perfect complexes on X. We now let ℓ be a prime number invertible in A. We denote by Shct Qℓ(S) the ∞-category of con- structible Qℓ-adic complexes on the ´etale site S´et of S. It is a symmetric monoidal ∞-category, and we denote by ShQℓ(S) := Ind(Shct Qℓ(S)) its completion under filtered colimits (see [Ga-Lu, Def. 4.3.26]). According to [Ro, Cor. 2.3.9], there exists an ℓ-adic realization ∞-functor Rℓ : SHS −→ ShQℓ(S). 9 / / By construction, Rℓ is a symmetric monoidal ∞-functor sending a smooth scheme p : X −→ S to p!p!(Qℓ), or, in other words, to the relative ℓ-adic homology of X over S. We let T := Qℓ[2](1), and we consider the E∞-ring object in ShQℓ(S) Qℓ(β) := ⊕n∈ZT⊗n. In this notation, β stands for T, and Qℓ(β) for the algebra of Laurent polynomials in β, so we might as well write Qℓ(β) = Qℓ[β, β−1]. As shown in [BRTV], there exists a canonical equivalence Rℓ(BUS) ≃ Qℓ(β) of E∞-ring objects in ShQℓ(S), that is induced by the Chern character from algebraic K-theory to ´etale cohomology. We thus obtain a well-defined symmetric monoidal ∞-functor Rℓ : ModSHS (BUS) −→ ModShQℓ (S)(Qℓ(β)), from BUS-modules in SHS to Qℓ(β)-modules in ShQℓ(S). By pre-composing with the functor T 7→ MT , we obtain a lax monoidal ∞-functor rℓ := Rℓ ◦ M− : dgCatS −→ ModShQℓ (S)(Qℓ(β)). Definition 2.2.1 The ∞-functor defined above is called the ℓ-adic realization functor for dg-categories over S. rℓ : dgCatS −→ ModShQℓ (S)(Qℓ(β)) From the standard properties of the functor T 7→ MT , recalled above, we obtain the following properties for the ℓ-adic realization functor T 7→ rℓ(T ). 1. The ∞-functor rℓ is a localizing invariant, i.e. for any short exact sequence T0 ֒→ T −→ T /T0 of dg-categories over A, the induced sequence rℓ(T0) / rℓ(T ) / rℓ(T /T0) is a fibration sequence in Mod(Qℓ(β)). 2. The natural morphism Qℓ(β) −→ rℓ(A), induced by the lax monoidal structure, is an equivalence in ModShQℓ (S)(Qℓ(β)). 3. The ∞-functor rℓ commutes with filtered colimits. 4. For any separated morphism of finite type p : X −→ S, we have a natural morphism of Qℓ(β)- modules rℓ(Perf(X)) −→ p∗(Qℓ(β)), where p∗ : ModShQℓ (S)(Qℓ,X(β)) −→ ModShQℓ (S)(Qℓ(β)) is induced by the direct image Shct Shct Qℓ equivalence. (S) of constructible Qℓ-complexes. If either p is proper, or A is a field, this morphism is an Qℓ(X) −→ 10 / / Remark 2.2.2 The ℓ-adic realization functor of Definition 2.2.1 can be easily "sheafified", as follows. For X a separated, excellent scheme, and T a sheaf of dg-categories on XZar (i.e. T ∈ dgCatX := lim dgCatU • where U • is the Cech-nerve of any Zariski open cover U of X), by taking the limit along the Cech-nerve of a Zariski open cover of X, we define M− X : dgCatX → ModSHX (BUX) and Rℓ,X : ModSHX (BUX) −→ ModShQℓ (X)(Qℓ,X(β)). The composite rℓ,X := Rℓ,X ◦ M− X : dgCatX −→ ModShQℓ (X)(Qℓ,X(β)) is called the ℓ-adic realization functor for dg-categories over X. Obviously we have rℓ = rℓ,S, for rℓ as in Definition 2.2.1. Moreover, for X as above, any separated morphism of finite type p : X −→ S induces a commutative diagram4 dgCatX rℓ,X / ModShQℓ (X)(Qℓ,X(β)) p∗ p∗ dgCatS rℓ / ModShQℓ (S)(Qℓ(β)) 2.3 The ℓ-adic Chern character As explained in [BRTV], there is a symmetric monoidal ∞-category SHnc over A. As an ∞-category it is the full sub-∞-category of ∞-functors of (co)presheaves of spectra A of non-commutative motives dgCatf t A −→ Sp, satisfying Nisnevich descent and A1-homotopy invariance. We consider Γ : ShQℓ(S) −→ dgQℓ, the global section ∞-functor, taking an ℓ-adic complex on S´et to (Qℓ, −) : dgQℓ −→ its hyper-cohomology. Composing this with the Dold-Kan construction RMapdgQℓ Sp we obtain an ∞-functor − : ShQℓ(S) −→ Sp, which computes hyper-cohomology of S´et with ℓ-adic coefficients, i.e. natural isomorphisms for any E ∈ ShQℓ(S), we have H i(S´et, E) ≃ π−i(E) , i ∈ Z. By what we have seen in our last paragraph, the composite functor T 7→ rℓ(T ) provides a (co)presheaf of spectra dgCatf t A −→ Sp, satisfying Nisnevich descent and A1-homotopy invariance. It thus defines an object rℓ ∈ SHnc A . The fact that rℓ is lax symmetric monoidal implies moreover that rℓ is endowed with a natural structure of a E∞-ring object in SHnc A . Each T ∈ dgCatf t A defines a corepresentable object hT ∈ SHnc A , characterized by the (∞-)functorial equivalence RMapSHnc A (hT , F ) ≃ F (T ), 4Note that the functor p∗ : dgCatX → dgCatS consists simply in viewing a dg-category over X as an A−dg-category via the morphism p. 11 / / for any F ∈ SHnc that we have a natural equivalence of spectra A . The existence of hT is a formal statement, however the main theorem of [Ro] implies RMapSHnc A (hT , hA) ≃ HK(T ), where HK(T ) stands for non-connective homotopy invariant algebraic K-theory of the dg-category T . In other words, T 7→ HK(T ) defines an object in SHnc A which is isomorphic to hB. By Yoneda lemma, we thus obtain an equivalence of spaces RMaplax−⊗(HK, rℓ) ≃ RMapE∞−Sp(S, rℓ(A)) ≃ ∗. In other words, there exists a unique (up to a contractible space of choices) lax symmetric monoidal natural transformation HK −→ rℓ, between lax monoidal ∞-functors from dgCatf t usual, by passing to Ind-completion dgCatA ≃ Ind(dgCatf t A ). A to Sp. We extend this to all dg-categories over A, as Definition 2.3.1 The natural transformation defined above is called the ℓ-adic Chern character. It is denoted by Chℓ : HK(−) −→ rℓ(−). Remark 2.3.2 Definition 2.3.1 contains a built-in, formal Grothendieck-Riemann-Roch formula. In- deed, for any B-linear dg-functor f : T −→ T ′, the square of spectra HK(T ) Chℓ,T rℓ(T ) f! f! HK(T ′) Chℓ,T ′ / rℓ(T ′) commutes up to a natural equivalence. 2.4 Trace formula for dg-categories Let C⊗ be a symmetric monoidal ∞-category ([To-Ve-1], [Lu-HA, Definition 2.0.0.7]). Hypothesis 2.4.1 The underlying ∞-category C has small sifted colimits, and the tensor product pre- serves small colimits in each variable. Definition 2.4.2 Let C⊗ be a symmetric monoidal ∞-category satisfying Hypothesis 2.4.1. We denote by Alg(C) the (∞, 2)-category of algebras in C⊗ denoted by Alg(1)(C⊗) in [Lu-COB, Definition 4.1.11]. Informally, one can describe Alg(C) as the (∞, 2)-category with: • objects: associative unital monoids (=: E1-algebras) in C. • MapAlg(C)(B, B′) := BimodB′,B(C⊗), the ∞-category of (B′, B)-bimodules. 12 / / / • The composition of 1-morphisms (i.e. of bimodules) is given by tensor product. • The composition of 2-morphisms (i.e. of morphisms between bimodules) is the usual composition. Definition 2.4.3 Let B be an algebra in C and X a left B-module. Let us identify X with a 1- morphism X : 1C → B in Alg(C). A right B-dual of X is defined as a right adjoint Y : B → 1C to X. Unraveling the definition, we get that a right dual of X is a left B⊗-op-module Y , the unit u of ad- junction (or coevaluation) is a map coev : 1C → Y ⊗B X in C, the counit v of adjunction (or evaluation) is a map ev : X ⊗ Y → B of (B, B)-bimodules; u and v satisfy usual compatibilities. Note that, if a right B-dual of X exists then it is "unique" (i.e. unique up to a contractible space of choices). If the right B-module Y is the right B-dual of the left B-module X, then we can define the trace of any map f : X → X of left B-modules, as follows. Recall that we have a coevaluation map in C coev : 1C → Y ⊗B X in C and an evaluation map of (B, B)-bimodules ev : X ⊗ Y → B. Consider the graph Γf defined as the composite 1C coev / / Y ⊗B X id⊗f / Y ⊗B X. We now elaborate on the evaluation map. Observe that • B ∈ C has a left B ⊗ B⊗-op-module structure that we will denote by BL. • B ∈ C has a right B ⊗ B⊗-op-module structure (i.e. a left B⊗-op ⊗ B-module), that we will denote by BR. • ev : X ⊗ Y → BL is a map of left B ⊗ B⊗-op- modules. • the composite ev′ : Y ⊗ X σ / X ⊗ Y ev / / BR is a map of left (B⊗-op ⊗ B)-modules Apply (−) ⊗B⊗B⊗-op BL to the composite ev′ : Y ⊗ X σ / X ⊗ Y ev / / BR to get evHH : (Y ⊗ X) ⊗B⊗B⊗-op BL / BR ⊗B⊗B⊗-op BL =: HHC(B) . Now observe that (Y ⊗ X) ⊗B⊗B⊗-op BL ≃ Y ⊗B X in C. Note that, by definition, HHC(B) is (only) an object in C, called the Hochschild homology object of B. Definition 2.4.4 The non-commutative trace of f : X → X over B is defined as the composite T rB(f ) : 1C Γf / Y ⊗B X ≃ (Y ⊗ X) ⊗B⊗B⊗-op BL evHH / / BR ⊗B⊗B⊗-op BL =: HHC(B) . T rB(f ) is a morphism in C. 13 / / / / / Remark 2.4.5 Let B ∈ CAlg(C⊗), and let us still denote by B its image via the canonical map CAlg(C⊗) → AlgE1(C⊗). In this case, ModB(C⊗) is a symmetric monoidal ∞-category, and if X ∈ ModB(C⊗) is a dualizable object (in the usual sense), then its (left and right) dual in ModB(C⊗) is also a right-dual of X according to Definition 2.4.3. In this case, any f : X → X in ModB(C⊗), has therefore two possible traces, a non-commutative one (as in Definition 2.4.4) which is a morphism in C, and a more standard, commutative one T rB(f ) : 1C −→ HHC(B) T rc B(f ) : B −→ B which is a morphism on ModB(C⊗). The two traces are related by the following commutative diagram 1C uB B T rB(f ) T rc B (f ) HHC(B) / B a where a : HHC(B) → B is the canonical augmentation (which exists since B is commutative), and uB : 1C → B is the unit map of the algebra B in C. The case of dg-categories. Let us specialize the previous discussion to the case C⊗ = dgCatA. Let B be a monoidal dg-category, i.e. an associative and unital monoid in the symmetric monoidal ∞-category dgCatA. Proposition 2.4.6 For any B-module T which is cotensored in the sense of Definition 2.1.2, the big B-module bT ∈ dgCATB has a right dual in the symmetric monoidal ∞-category dgCATA whose underlying big dg-category is cT o. Proof. This is very similar to the argument used in [To2, Prop. 2.5 (1)]. We consider cT o, and we define evaluation and coevaluation maps as follows. The big morphism h introduced right before Definition 2.1.2 whose domain is naturally a bB-bimodule, can be canonically lifted to a morphism of bimodules. We choose this as our evaluation morphism. h : cT ob⊗AbT −→ bB, The coevaluation is then obtained by duality. We start by the diagonal bimodule sending (x, y) to T (y, x). This morphism naturally descends to (T o ⊗B T )o, providing a dg-functor T : (T o ⊗A T )o −→ L(A) = bA. Note that T o is naturally a right B-module since T is assumed to be cotensored (and note that, otherwise, (T o ⊗B T )o −→ bA. T o ⊗B T would not make sense). This dg-functor is an object in \T o ⊗B T ≃ cT ob⊗AbT , and thus defines a coevaluation morphism bA −→ cT ob⊗AbT . 14 / / / These two evaluation and coevaluation morphisms satisfy the required triangular identities, and thus ✷ make cT o a right dual to bT . According to the previous Proposition, any cotensored B-module T has a big right dual, so it comes equipped with big evaluation and coevaluation maps. Definition 2.4.7 For a monoidal dg-category B, and a B-module T ∈ dgCatB, we say that T is saturated over B if 1. T is cotensored (over B), and 2. the evaluation and coevaluation morphisms are small (i.e. are maps in dgCatA). In particular, if T saturated over B, and f : T → T is a morphism in dgCatB, then the trace T rB(f : T → T ) : A → HH(B/A) = BR ⊗B⊗-op⊗AB BL is also small, i.e. it is a morphism inside dgCatA. Since our ℓ-adic realization functor rℓ is only lax-monoidal, in order to establish our trace formula for an endomorphism f : T → T of a saturated B-module, we need to restrict to those saturated B-modules on which rℓ is in fact symmetric monoidal. Definition 2.4.8 A saturated T ∈ dgCatB is called ℓ⊗-admissible if the canonical map rℓ(T op) ⊗rℓ(B) rℓ(T ) → rℓ(T op ⊗B T ) is an equivalence in ShQℓ(S). The trace T rB(f ) is a map A → HH(B/A), hence it induces a map in Sp K(T rB(f )) : K(A) → K(HH(B/A)) which is actually a map of K(A)-modules (in spectra), since K is lax-monoidal. Hence it corresponds to an element denoted as Therefore, its image by the ℓ-adic Chern character Chℓ,0 := π0(Chℓ) [HH(T /B, f )] ≡ trB(f ) ∈ K0(HH(B/A)). Chℓ,0 : K0(HH(B/A)) → HomD(rℓ(A))(rℓ(A), rℓ(HH(B/A))) ≃ H 0(S´et, rℓ(HH(B/A))), is an element Chℓ,0([HH(T /B; f )]) ∈ H 0(S´et, rℓ(HH(B/A))). On the other hand, the trace T rrℓ(B)(rℓ(f )) of rℓ(f ) over rℓ(B) is, by definition, a morphism rℓ(A) ≃ Qℓ(β) → HH(rℓ(B)/rℓ(A)) in Modrℓ(A)(ShQℓ(S)). We may further compose this with the canonical map HH(rℓ(B)/rℓ(A)) → rℓ(HH(B/A)) 15 (given by lax-monoidality of rℓ(−)), to get a map rℓ(A) → rℓ(HH(B/A)) in Modrℓ(A)(ShQℓ(S)). This is the same thing as an element denoted as trrℓ(B)(rℓ(f )) ∈ π0(rℓ(HH(B/A))) ≃ H 0(S´et, rℓ(HH(B/A))). Theorem 2.4.9 Let B a monoidal dg-category over A, T ∈ dgCatB a saturated and ℓ⊗-admissible B-module, and f : T → T map in dgCatB. Then, we have an equality in H 0(S´et, rℓ(HH(B/A)). Chℓ,0([HH(T /B, f )]) = trrℓ(B)(rℓ(f )) Proof. This is a formal consequence of uniqueness of right duals, and of the resulting fact that traces are preserved by symmetric or lax symmetric monoidal ∞-functors under our admissibility condition. The key statement is the following lemma, left as an exercise to the reader, and applied to the case where F is our ℓ-adic realization functor. Lemma 2.4.10 Let F : C −→ D be a lax symmetric monoidal ∞-functor between presentable symmetric monoidal ∞-categories. Let B be a monoid in C, M a left B-module, and f : M −→ M a morphism of B-modules. We assume that M has a right dual M o, and that the natural morphism F (M o) ⊗F (B) F (M) −→ F (M o ⊗B M) is an equivalence. Then, F (M) has a right dual, and we have F (T r(f )) = i(T r(F (f ))) as elements in π0(HomD(1, F (HH(B))), where i is induced by the natural morphism HH(F (B)) −→ F (HH(B)). ✷ 3 Invariant vanishing cycles This section gathers general results about inertia-invariant vanishing cycles (I-vanishing cycles, for short), their relations with dg-categories of singularities, and their behaviour under products. These results are partially taken from [BRTV], and the only original result is Proposition 3.4.2 that can be seen as a version of Thom-Sebastiani formula in the mixed-characteristic setting. All along this section, A will be a strictly henselian excellent dvr with fraction field K = Frac(A), and perfect (hence algebraically closed) residue field k. We let S = Spec A. All schemes over S are assumed to be separated and of finite type over S. We denoted by i : s := Spec k −→ S the closed point of S, and j : η := Spec K −→ S its generic point. For an S-scheme X, we denote by Xs := X ×S s its special fiber, and Xη = X ×S η its generic fiber. Accordingly, we write X¯η := X ×S Spec K sp for the geometric generic fiber. 16 3.1 Trivializing the Tate twist We let ℓ be a prime invertible in k, and we denote by p the characteristic exponent of k. As k is algebraically closed, we may, and will, choose once for all a group isomorphism µ∞(k) ≃ µ∞(K) ≃ (Q/Z)[p−1] between the group of roots of unity in k and the prime-to-p part of Q/Z. Equivalently, we have chosen a given group isomorphism lim (n,p)=1 µn(k) ≃ Z′, where Z′ := lim(n,p)=1 Z/n. In particular, we have selected a topological generator of lim(n,p)=1 µn(k), corresponding to the image of 1 ∈ Z inside Z′. The choice of the isomorphism above also provides a specific induced isomorphism Qℓ(1) ≃ Qℓ of Qℓ-sheaves on S, where (1) denotes, as usual, the Tate twist. By taking tensor powers of this isomorphism, we get various induced isomorphisms Qℓ(i) ≃ Qℓ for all i ∈ Z. We remind that the absolute Galois group I of K (which coincides with the inertia group in our case) sits in an extension of pro-finite groups5 1 / P / I / It ≃ Z′ / 1 , where P is a pro-p-group (the wild inertia subgroup). For any continuous finite dimensional Qℓ- representation V of I, the group P acts by a finite quotient GV on V . Moreover, the Galois cohomology of V can then be explicitly identified with the two-terms complex V G 1−T / V G where T is the action of the chosen topological generator of It. This easily implies that for any Qℓ- representation V of I, the natural pairing on Galois cohomology H i(I, V ) ⊗ H 1−i(I, V ∨) −→ H 1(I, Qℓ) ≃ Qℓ is non-degenerate. In other words, if we denote by V I the complex of cohomology of I with coefficients in V , we have a natural quasi-isomorphism (V I)∨ ≃ (V ∨)I[1]. 3.2 Reminders on actions of the inertia group Let X −→ S be an S-scheme (separated and of finite type, according to our conventions). We recall from [SGA7-II, Exp. XIII, 1.2] that we can associate to X a vanishing topos (X/S)ν et which is defined as (a 2-)fiber product of toposes (X/S)ν et := (Xs)∼ et ×s∼ et η∼ et. Since S is strictly henselian, s∼ a product of topos. The topos η∼ et is in fact the punctual topos, and the fiber product above is in fact et is equivalent to the topos of sets with continuous action of I = 5Note that the tame inertia quotient It is canonically isomorphic to Z′(1), and it becomes isomorphic to Z′ through our choice. 17 / / / / / Gal(K sp/k), where K sp denotes a seprable closure of K. Morally, (X/S)ν on Xs, endowed with a continuous action of I (see [SGA7-II, Exp. XIII, 1.2.4]). et is the topos of ´etale sheaves As explained in [BRTV] we have an ℓ-adic ∞-category D((X/S)ν et, Zℓ). Morally speaking, objects of this ∞-category consist of the data of an object E ∈ D( ¯Xs, Zℓ) = ShZℓ( ¯Xs) together with a continuous action of I. We say that such an object is constructible if E is a constructible object in D( ¯Xs, Zℓ) = ShZℓ( ¯Xs), and we denote by Dc((X/S)ν et, Zℓ) the full sub-∞-category of constructible objects. Definition 3.2.1 The ∞-category of ind-constructible I-equivariant ℓ-adic complexes on Xs is defined by DI ic(Xs, Qℓ) := Ind(Dc((X/S)ν et, Zℓ) ⊗Zℓ Qℓ). The full sub-∞-category of constructible objects is DI c(Xs, Qℓ) := Dc((X/S)ν et, Zℓ) ⊗Zℓ Qℓ. Note that since we have chosen trivialisations of the Tate twists, Qℓ(β) is identified with Qℓ[β, β−1] where β is a free variable in degree 2. This is a graded algebra object in DI ic(Xs, Qℓ), and we define ic(Xs, Qℓ(β)) as the ∞-category of Qℓ(β)-modules in DI the ∞-category DI ic(Xs, Qℓ), or equivalently, the 2-periodic ∞-category of ind-constructible Qℓ-adic complexes on (X/S)ν et. Definition 3.2.2 An object E ∈ DI ic(Xs, Qℓ(β)) is constructible if it belongs to the thick triangulated sub-∞-category generated by objects of the form E0(β) = E0 ⊗Qℓ Qℓ(β) for E0 a constructible object in DI ic(Xs, Qℓ). The full sub-∞-category of DI ic(Xs, Qℓ(β)) consisting of constructible objects is denoted DI c(Xs, Qℓ(β)). Similarly, for any S-scheme X, we define Dc(X, Qℓ(β)) as the full sub-∞-category of objects in Dic(X, Qℓ(β)) generated by E0(β) for E0 constructible. Note that strictly speaking an object of DI underlying object in Dic(Xs, Qℓ) is 2-periodic. c(Xs, Qℓ(β)) is not constructible in the usual sense, as its The topos (X/S)ν ∞-functor denoted by et comes with a natural projection (X/S)ν et −→ (Xs)∼ et whose direct image is an (−)I : DI ic(Xs, Qℓ) −→ Dic(Xs, Qℓ) called the I-invariants functor. This ∞-functor preserves constructibility. The ∞-categories DI ic(Xs, Qℓ) and Dic(Xs, Qℓ) carries natural symmetric monoidal structures and the ∞-functor (−)I comes equipped with a natural lax symmetric monoidal structure (being induced by the direct image of a morphism of toposes). Moreover, (−)I is the right adjoint of the symmetric monoidal ∞-functor U : Dic(Xs, Qℓ) −→ DI ic(Xs, Qℓ) the struc- ture of a Dic(Xs, Qℓ)-module via Dic(Xs, Qℓ) × DI ic(Xs, Qℓ) : (E, F ) 7→ U(E) ⊗ F . This bi-functor distributes over colimits, thus by the adjoint theorem, we get an enrichment of DI ic(Xs, Qℓ) over Dic(Xs, Qℓ). Note that DI ic(Xs, Qℓ) endowing objects in Dic(Xs, Qℓ) with the trivial action of I. This gives DI ic(Xs, Qℓ) is also enriched over itself. ic(Xs, Qℓ) → DI It is important to notice that the I-invariants functor commutes with base change in the following sense. Let f : Spec k −→ Xs be a geometric point. The morphism f defines a geometric point of (Xs)∼ et and thus induces a geometric morphism of toposes et −→ (X/S)ν η∼ et. 18 We thus have an inverse image functor f ∗ : DI c(Xs, Qℓ) −→ Dc(η, Qℓ) = Dc(I, Qℓ). where Dc(I, Qℓ) is the ∞-category of finite dimensional complexes of ℓ-adic representations of I. As usual, the square of ∞-functors DI c(Xs, Qℓ) (−)I Dc(Xs, Qℓ) = Shct Qℓ(Xs) f ∗ Dc(I, Qℓ) f ∗ / Dc(Qℓ) (−)I comes equipped with a natural transformation f ∗((−)I) ⇒ (f ∗(−))I. It can be checked that this natural transformation is always an equivalence. geometric point x in Xs, we have a natural equivalence of ℓ-adic complexes (E)I E ∈ DI c(Xs, Qℓ). In particular, for any x ≃ (Ex)I, for any The dualizing complex ω of the scheme Xs can be used in order to obtain a dualizing object in c(Xs, Qℓ) endowed with the trivial I-action. We c(Xs, Qℓ) as follows. We consider ω as an object in DI DI then have an equivalence of ∞-categories DI : DI c(Xs, Qℓ) −→ DI c(Xs, Qℓ)op sending E to RHom(E, ω), where RHom denotes the natural enrichment of DI c(Xs, Qℓ) over itself. We obviously have a canonical biduality equivalence D2 I ≃ id. The duality functor DI is compatible with the usual Grothendieck duality functor D for the scheme Xs up to a shift, as explained by the following lemma. Lemma 3.2.3 For any object E ∈ DI c(Xs, Qℓ), there is a functorial equivalence in Dc(Xs, Qℓ) d : D(EI)[−1] ≃ (DI(E))I. Proof. Taking I-invariants is a lax monoidal ∞-functor, so we have a natural map EI ⊗ DI(E)I −→ (E ⊗ DI(E))I, that can be composed with the evaluation morphism E ⊗ DI(E) −→ ω to obtain EI ⊗ DI(E)I −→ ωI. As the action of I on ω is trivial, we have a canonical equivalence ωI ≃ ω⊗QI ℓ ≃ ω⊕ω[−1]. By projection on the second factor we get a pairing EI ⊗ DI(E)I −→ ω[−1], and thus a map DI(E)I −→ D(EI)[−1]. We claim that the above morphism is an equivalence in Dc(Xs, Qℓ). For this it is enough to check that the above morphism is a stalkwise equivalence. Now, the stalk of the above morphism at a geometric point x of Xs can be written as (E(x)∨)I −→ (E(x)I)∨[−1] 19 / / / x(X, E) ∈ Dc(Qℓ) is the local cohomology of E at x, and (−)∨ is now the standard lin- where E(x) := H ∗ ear duality over Qℓ. The result now follows from the following well-known duality for Qℓ-representations of I: for any finite dimensional Ql-representation V of I, the fundamental class in H 1(I, Qℓ) ≃ Qℓ, in- duces a canonical isomorphism of Galois cohomologies H ∗(I, V ∨) ≃ H 1−∗(I, V )∨. ✷ 3.3 Invariant vanishing cycles and dg-categories From [SGA7-II, Exp. XIII] and [BRTV, 4.1], the vanishing cycles construction provides an ∞-functor φ : Dc(X, Qℓ) −→ DI c(Xs, Qℓ). Applied to the constant sheaf Qℓ, we get this way an object denoted by νX/S (or simply νX if S is clear) in DI c(Xs, Qℓ). Definition 3.3.1 The I-invariant vanishing cycles of X relative to S (or I-vanishing cycles, for short) is the object νI X := (νX )I ∈ Dc(Xs, Qℓ). There are several possible descriptions of invariant vanishing cycles. First of all, by its very definition, X := (ψX)I by means of an exact triangle in Dc(Xs, Qℓ) νI X is related to the I-invariant nearby cycles ψI QI ℓ / ψI X / νI X . (2) Another description, in terms of local cohomology, is the following. We let U = XK be the open complement of Xs inside X, and jX : U ֒→ X and iX : Xs ֒→ X the corresponding immersions. Then, the I-vanishing cycles enters in an exact triangle in Dc(Xs, Qℓ) νI X / Qℓ / i! X (Qℓ)[2]. (3) Triangle (2) follows from the octahedral axiom applied to the triangles (1) and Qℓ −→ QI ℓ ≃ Qℓ ⊕ Qℓ[−1] −→ Qℓ[−1], taking also into account the triangle i! X Qℓ −→ Qℓ −→ i∗ X (jX)∗j∗ X Qℓ ≃ ψI X . We get one more description of νI X ⊗ Qℓ(β)) using the ℓ-adic realization of the dg-category of singularities studied in [BRTV], at least when X is a regular scheme with smooth generic fiber. Let Sing(Xs) = Cohb(Xs)/Perf(Xs) be the dg-category of singularities of the scheme Xs. This dg-category is naturally linear over the dg-categories Perf(Xs) and Perf(X), and thus we can take its ℓ-adic realization rℓ,X(Sing(Xs)) over X (see Remark 2.2.2) which is a Qℓ,X(β)-module in Dic(X, Qℓ) X (or rather, of νI X (β) := νI 20 / / / / supported on Xs, hence can be identified with a Qℓ,Xs(β)-module in Dic(Xs, Qℓ). When X is a regular scheme and XK is smooth over K, we have from [BRTV] a canonical equivalence in Dc(Xs, Qℓ(β)) 6 νI X(β)[1] ≃ rℓ,X(Sing(Xs)) (4) where νI X(β) stands for νI X ⊗ Qℓ(β). We conclude this section with another description of νI X(β), see equivalence (6), this time in terms of sheaves of singularities. We need a preliminary result. Lemma 3.3.2 We assume that S is excellent. Let p : X → S be a separated morphism of finite type. We write rℓ,X : SHX → ShQℓ(X), for the ℓ-adic realization over X, as in Remark 2.2.2.Then, we have 1. rℓ,X(Perf(X)) ≃ Qℓ,X(β) in Dic(X, Qℓ). 2. rℓ,X(Cohb(X)) ≃ ωX(β) in Dic(X, Qℓ), where ωX ≃ p!(Qℓ) is the ℓ-adic dualizing complex of X 7. 3. There exists a canonical map ηX : Qℓ,X (β) −→ ωX(β) in Dc(X, Qℓ(β)), called the 2-periodic ℓ-adic fundamental class of X. Proof. First of all, separated finite type morphisms of noetherian schemes are compactifiable (by Nagata's theorem), thus we can assume that p : X → S is proper. X is M∨ X (T )), we first construct a map αmot : M∨ (1) follows immediately from [BRTV, Prop. 3.9 and formula (3.7.13))]. In order to prove (2) we first produce a map α : rℓ(Cohb(X)) → ωX(β). In the notations of [BRTV, §3] (note that [BRTV]'s notation for MT X in SH(X), whose ´etale ℓ-adic realization will be α. Since p is proper, αmot is the same thing, by adjunction, as a map S(Cohb(X))), where Cohb(X) p∗(M∨ is viewed as a dg-category over S, via p. If Y is smooth over S, we have by [Pr, Prop. B.4.1], an equivalence of S-dg-categories X(Cohb(X))) → BUS in SH(S). Now, p∗(M∨ X(Cohb(X)) → p!(BUS) =: ωmot X(Cohb(X))) is just M∨ Cohb(X) ⊗S Cohb(Y ) ≃ Cohb(X ×S Y ) (5) S(Cohb(X)) ∈ SH(S) is the ∞-functor Y 7→ KH(Cohb(X ×S Y )), and Through this identification, M∨ KH(Cohb(X ×S Y )) is equivalent to the G-theory spectrum G(X ×S Y ) of X ×S Y , by A1-invariance of G-theory. Since S is regular, BUS ≃ GS := G(−/S) canonically in SH(S), and we can take the S(Cohb(X))) → BUS ≃ GS to be the push forward p∗ on G-theories G(X ×S −) → G(−/S). map M∨ This gives us a map αmot : M∨ X . Now observe that by [BRTV, formula (3.7.13)], the ´etale ℓ-adic realization of p!(BUS) is canonically equivalent to p!(Qℓ(β)) ≃ ωX(β) (since ´etale ℓ-adic realization commutes with six operations, [BRTV, Rmk. 3.23]). Therefore, we get our map α : rℓ(Cohb(X)) → ωX(β). Checking that α is an equivalence is a local statement, i.e. it is enough to show that if j : V = Spec, A ֒→ X is an open affine subscheme, then j∗(α) is an equivalence. Now, j∗rℓ(Cohb(X)) ≃ rℓ,V (j∗Cohb(X)) ≃ rℓ,V (Cohb(V )) (where rℓ,V denotes the ℓ-adic realization over V ), and j∗ωX ≃ j!ωX ≃ ωV , so j∗α identifies with a map rℓ,V (Cohb(V )) → ωV (β). Since V is affine and of finite type over S, we can choose a closed immersion i : V ֒→ V ′, with V ′ affine and smooth (hence X(Cohb(X)) → p!(BUS) =: ωmot 6Strictly speaking, in [BRTV] the equivalence (4) is proved only after push-forward to S but the very same proof shows also the equivalence (4). 7Note that since S is excellent, X is excellent so that ωX exists by a theorem of Gabber ([ILO, Exp. XVII, Th. 0.2]). 21 regular) over S. Let h : V ′ \ V ֒→ V ′ be the complementary open immersion. Since V ′ and V ′ \ V are regular, by Quillen localization and the properties of the nc realization functor M∨ (see [BRTV]), we get a cofiber sequence M∨ V ′(Cohb(V )/V ′) → BUX ′ → h∗BUV ′\V , where the notation Cohb(V )/V ′ means that Cohb(V ) is viewed as a dg-catgeory over V ′, via i. In other V (Cohb(V )). If we apply i∗ to this cofiber sequence, and compare what words, M∨ we obtain to the application of i∗ to the standard localization sequence V ′(Cohb(V )/V ′) ≃ i∗M∨ i∗i!BUV ′ → BUV ′ → h∗h∗BUV ′ = h∗BUV ′\V , we finally get, after ´etale ℓ-adic realization, that ωV (β) ≃ i!Qℓ(β) ≃ rℓ,V (Cohb(V )). This implies that j∗α is also an equivalence. By (1) and (2), the map in (3) is finally obtained by applying rℓ,X to the inclusion Perf(X) → Cohb(X). ✷ Remark 3.3.3 Note that Lemma 3.3.2 applies, in particular, to give a 2-periodic ℓ-adic fundamental class map ηU : Qℓ(β) −→ ωU (β) for any open subscheme U ֒→ X over S, whenever X is proper over S. Definition 3.3.4 Let X/S be a separated and finite type S-scheme. The sheaf of singularities of X is defined to be the cofiber of the 2-periodic ℓ-adic fundamental class morphism ηX (Lemma 3.3.2 (3)) ωo X := Cofib(ηX : Qℓ(β) −→ ωX(β)). Remark 3.3.5 By construction of ηX in Lemma 3.3.2, ωo X is then the ℓ-adic realization rℓ,X(Sing(X)) of the dg-category Sing(X), considered as a dg-category over X, and this is true without any regularity hypothesis on X. Remark 3.3.6 If p : X → S is a proper lci map from a derived scheme X , we can still define a 2- periodic ℓ-adic fundamental class map ηX , as in Lemma 3.3.2. This can be done by observing that the pushforward on G-theories along the inclusion of the truncation t0X → X is an equivalence, and that p being lci we have a natural inclusion Perf(X ) → Cohb(X ). We further observe that in this case, while the ∞-category Dc(X , Qℓ(β)) only depends on the reduced subscheme (t0X )red, and the same is true for the objects Qℓ,X (β) and ωX (β), in contrast, the morphism ηX does depend on the derived structure on X , and thus it is not a purely topological invariant. Let us come back to X a regular scheme, proper over S. We have a canonical equivalence in Dc(Xs, Qℓ(β)) νI X (β)[1] ≃ ωo Xs. (6) This is a reformulation of the equivalence (4), in view of Lemma 3.3.2. Remark 3.3.7 When X is not regular anymore, but still proper and lci8 over S, there is nonetheless a natural morphism νI Xs, constructed as follows. Consider again the triangle (3) X (β)[1] −→ ωo 8Note that a morphism of finite type between regular schemes is lci, since we can check that its relative cotangent complex has perfect amplitude in [−1, 0]. νI X / Qℓ / i! X (Qℓ)[2]. 22 / / On X, we do have the 2-periodic ℓ-adic fundamental class ηX : Qℓ(β) −→ ωX(β), and by taking its !-pullback by i! X(Qℓ)(β) −→ ωXs(β). This produces a sequence of morphisms X , we get a morphism i! νI X (β) / Qℓ(β) / i! X (Qℓ)[2](β) = i! X (Qℓ)(β) / ωXs(β). The resulting composite morphism Qℓ(β) → ωXs(β) is the 2-periodic ℓ-adic fundamental class of Xs. Moreover, by construction, the composition νI / ωXs(β) is canonically the zero map, and this induces the natural morphism / Qℓ(β) X (β) αX : νI X (β)[1] −→ ωo Xs we were looking for. Summing up, the morphism αX always exists for any proper, lci scheme X over S, and is an equivalence whenever X is regular. 3.4 A Kunneth theorem for invariant vanishing cycles In this section, we consider two separated, finite type S-schemes X and Y , such that XK and YK are smooth over K, and both X and Y are regular and connected. For simplicity9 we also assume that X and Y are flat over S. We set Z := X ×S Y , and consider this as a scheme over S. We have Zs ≃ Xs ×s Ys, and the ∞-category DI c(Zs, Qℓ) comes equipped with pull-back functors p∗ : DI c(Xs, Qℓ) −→ DI c(Zs, Qℓ) ←− DI c(Ys, Qℓ) : q∗. By taking their tensor product, we get an external product functor ⊠ := p∗(−) ⊗ q∗(−) : DI c(Xs, Qℓ) × Dc(Ys, Qℓ) −→ DI c(Zs, Qℓ). For two objects E ∈ DI c(Xs, Qℓ) and F ∈ DI c(Ys, Qℓ), we can define the Kunneth morphism in Dc(Zs, Qℓ) k : (E ⊠ F )I[−1] −→ EI ⊠ F I as follows. Since Grothendieck duality on Zs is compatible with external products, in order to define k it is enough to define its dual D(EI) ⊠ D(F I) −→ D((E ⊠ F )I)[1]. By Lemma 3.2.3, the datum of such a morphism is equivalent to that of a morphism DI(E)I ⊠ DI(F )I −→ DI(E ⊠ F )I. We now define k as the map induced by the composite DI(E)I ⊠ DI(F )I µ(−)I / (DI(E) ⊠ DI(F ))I (µDI )I / DI(E ⊠ F )I where µ(−)I is the lax monoidal structure on (−)I, and µDI the one on DI 10. 9In the non-flat case, the fiber product of X and Y over S, to be considered below, should be replaced by the derived fiber product. 10Note that µDI is in fact an equivalence. 23 / / / / / / / Definition 3.4.1 With the above notations, the I-invariant convolution of the two objects E ∈ DI and F ∈ DI definition it sits in a triangle c(Xs, Qℓ) c(Ys, Qℓ) is defined to be the cone of the Kunneth morphism, and denoted by (E ⊛ F )I. By (E ⊠ F )I[−1] k / EI ⊠ F I / (E ⊛ F )I. The main result of this section is the following proposition, relating the I-invariant convolution of vanishing cycles on X and Y to the dualizing complex of Z. It can be also considered as a computation of the ℓ-adic realization of the ∞-category Sing(Z) of singularities of Z. Note that, as X and Y are generically smooth over S, so is Z, and thus the 2-periodic ℓ-adic fundamental class map ηZ : Qℓ(β) −→ ωZ(β) of Lemma 3.3.2 is an equivalence over the generic fiber. Therefore, ωo Z is supported on Zs, so that it can (and will) be considered canonically as an object in Dc(Zs, Qℓ). Theorem 3.4.2 With the above notations and assumptions, there is a canonical equivalence in Dc(Zs, Qℓ(β)). ωo Z ≃ (νX ⊛ νY )I(β) Proof. The proof of this theorem will combine various exact triangles together with an application of Gabber's Kunneth formula for nearby cycles. To start with, the vanishing cycles νZ of Z sits in an exact triangle in DI c(Zs, Qℓ) Qℓ / ψZ / νZ , where ψZ = ψZ(Qℓ) is the complex of nearby cycles of Z over S. According to [Be-Be, Lemma 5.1.1] or [Il1, 4.7]), we have a natural equivalence in DI c(Zs, Qℓ), induced by external product ψZ ≃ ψX ⊠ ψY . The object νZ then becomes the cone of the tensor product of the two morphisms in DI c(Zs, Qℓ) Qℓ −→ p∗(ψX ) Qℓ −→ q∗(ψY ) where p and q are the two projections from Z down to X and Y , respectively. Now, cones of tensor products are computed via the following well known lemma. Lemma 3.4.3 Let C be a stable symmetric monoidal ∞-category, and u : x → y v : x′ → y′ two morphisms. Let C(u) be the cone of u, C(v) be the cone of v, and C(u ⊗ v) the cone of the tensor product u ⊗ v : x ⊗ x′ → y ⊗ y′. Then, there exists a natural exact triangle C(u) ⊗ x′ L x ⊗ C(v) / C(u ⊗ v) / C(u) ⊗ C(v). 24 / / / / / / Proof of lemma. Factor u ⊗ v as x ⊗ x′ u⊗id triangles / y ⊗ x′ id⊗v / y ⊗ y′ , and apply the octahedral axiom to the x ⊗ x′ u⊗id / y ⊗ x′ f / C(u) ⊗ x′ y ⊗ x′ id⊗v / y ⊗ y′ / y ⊗ C(v) d′ [1] / y ⊗ x′[1] , to get a triangle C(u) ⊗ x′ / C(u ⊗ v) / y ⊗ C(v) θ [1] / C(u) ⊗ x′[1] together with the compatibility θ = f [1] ◦ d′. Now observe that θ ◦ (u ⊗ idC(v)) = 0, and apply the octahedral axiom to the triangles x ⊗ C(v) u⊗id / y ⊗ C(v) / C(u) ⊗ C(v) , y ⊗ C(v) θ / C(u) ⊗ x′[1] f / C(u ⊗ v)[1] to conclude. ♦ Lemma 3.4.3 implies the existence of a natural exact triangle in DI c(Zs, Qℓ) which, by taking I-invariants, yields an exact triangle in Dc(Zs, Qℓ) νX ⊞ νY / νZ / νX ⊠ νY , (T1) νI X ⊞ νI Y / νI Z / (νX ⊠ νY )I. Lemma 3.4.4 Let k be an algebraically closed field, s := Spec k, pX : X → s, pY : Y → s be proper morphisms of schemes, and p1 : Z := X ×s Y → X, p2 : Z := X ×s Y → Y the natural projections. If ωZ, ωX, ωY denote the Qℓ-adic dualizing complexes of Z, X, and Y , respectively, there is a canonical equivalence a : p∗ 1ωX ⊗ p∗ 2ωY −→ ωZ. Proof of lemma. We first exhibit the map a. We denote simply by HomT (−, −) the derived internal hom in Dc(T, Qℓ) (so that D := HomZ(−, ωZ) is the Qℓ-adic duality on Z). By adjunction, giving a map a is the same thing as giving a map ωX → (p1)∗HomZ(p∗ 2ωY ). Since pX is proper, by [SGA4-III, Exp XVIII, 3.1.12], we have a canonical equivalence 2ωY , p! (p1)∗HomZ(p∗ 2ωY , p! 2ωY ) −→ p! X(pY )∗HomY (ωY , ωY ). Therefore, we are left to define a map ωX ≃ p! XQℓ → p! X(pY )∗HomY (ωY , ωY ), and we take p! One can then prove that a is an equivalence, by checking it stalkwise. X(α) for this map, where α is the adjoint to the canonical map Qℓ ≃ p∗ Y Qℓ → HomY (ωY , ωY ). ✷ 25 / / / / / / / / / / / / / / / / / / By Lemma 3.4.4, the dualizing complex ωZs = (Zs → s)!Qℓ of Zs ≃ Xs ×s Ys is canonically equivalent to ωXs ⊠ ωYs, and, through this equivalence, the virtual fundamental class of Zs ηZs : Qℓ(β) −→ ωZs(β) ≃ ωXs ⊠ ωYs(β) is simply given by the external tensor product of the virtual fundamental classes of Xs and Ys. By Lemma 3.4.3 we thus get a second exact triangle in Dc(Zs, Qℓ(β)) (T2) ωo Xs ⊞ ωo Ys / ωo Zs / ωo Xs ⊠ ωo Ys. There is a morphism from the triangle (T1) to the triangle (T2) which is defined using the natural morphism αZ : νI Z(β)[1] −→ ωo Zs introduced in Remark 3.3.7. In fact, Z is proper and lci over S (since X/S is flat and lci11, and being lci is stable under flat base change and composition), and the map αZ is defined for any proper, lci scheme Z over S, being an equivalence when Z is regular with smooth generic fiber. Using the compatible maps αX, αY and αZ we get a commutative square νI X(β)[1] ⊞ νI Y (β)[1] / νI Z(β)[1] αZ ⊕αY ωo Xs ⊞ ωo Ys αZ / ωo Zs. This produces a morphism from triangle (T 1) (tensored by Qℓ(β)[1]) to triangle (T 2) νI X (β)[1] ⊞ νI Y (β)[1] νI Z(β)[1] (νX ⊠ νY )I(β)[1] (7) ωo Xs ⊞ ωo Ys αZ / ωo Zs / ωo Xs ⊠ ωo Ys. Since X and Y are regular with smooth generic fibers, the maps αX and αY are equivalences, therefore the leftmost vertical morphism is also an equivalence. Thus the right hand square is a cartesian square. Now, the rightmost vertical morphism can be written, again using the equivalences αX and αY , as (νX ⊠ νY )I(β)[1] −→ (νI X(β)[1]) ⊠Qℓ(β) (νI Y (β)[1]) ≃ (νI X ⊠ νI Y )[2](β) This morphism is the Kunneth map k of Definition 3.4.1 tensored by Qℓ[2](β) ≃ Qℓ(β), and thus its cone is (νX ⊛ νY )I(β). In order to finish the proof of the proposition it then remains to show that the cone of the middle vertical morphism in (7) can be canonically identified with ωo Z. αZ : νI Z(β)[1] −→ ωo Zs 11Again, note that a morphism of finite type between regular schemes is lci, since we can check that its relative cotangent complex has perfect amplitude in [−1, 0]. 26 / / / / / / / / / / For this, we remind the exact triangle (3) tensored by Qℓ(β) Qℓ(β) / i! Z(Qℓ(β)) / νI Z(β)[1]. Using the fundamental class ηZ : Qℓ(β) → ωZ(β), we get a morphism of triangles Qℓ(β) id i! Z(Qℓ(β)) νI Z(β)[1] i! Z (ηZ ) Qℓ(β) / i! Z(ωZ(β)) / ωo Zs. The right hand square is thus cartesian, so that the cone of the vertical morphism on the right is canonically identified with the cone of the vertical morphism in the middle. By definition, this cone is i! Z) is canonically equivalent to ωo ✷ Z). Since ZK is smooth over K, the ℓ-adic complex ωo Z, and we conclude. Z(ωo Z is supported on Zs, and thus i! Z(ωo Corollary 3.4.5 We keep the same notations and assumptions as in Proposition 3.4.2, and we further assume one of the following conditions: 1. the I-action on νX and on νY is tame, or 2. the reduced scheme (Xs)red is smooth over k. Then, there is a canonical equivalence in Dc(Zs, Qℓ(β)). ωo Z ≃ (νX ⊠ νY )I(β) Proof. It is enough to prove that under any one of the two assumptions, (νX ⊛ νY )I is canonically equivalent to (νX ⊠ νY )I. If the scheme (Xs)red is smooth over k, then we have νI X (β) = 0. Indeed, triangle (3) can be then re-written νI X / Qℓ / Qℓ[2n + 2] where n is the dimension of Xs. By tensoring by Qℓ(β), we get a triangle νI X(β) / Qℓ(β) b / Qℓ[2n + 2](β) ≃ Qℓ(β) where b is an equivalence. Therefore, νI implies that (νX ⊛ νY )I ≃ (νX ⊠ νY )I. X(β) = 0, and, by definition of I-invariant convolution, this Assume now that the action of I on νX and νY is tame. This means that the action of I factors through the natural quotient I −→ It, where It is the tame inertia group, which is canonically isomorphic to Z′, the prime-to-p part of the profinite completion of Z. As we have chosen a topological generator T of It (see Section 3.1), the actions of I are then completely caracterized by the automorphisms T on 27 / / / / / / / / / / / / νX and νY . Moreover, νI similarly for νI Y . From this it is easy to see that the Kunneth map X is then naturally equivalent to the homotopy fiber of (1 − T ) : νX → νX , and fits in an exact triangle (νX ⊠ νY )I[−1] −→ νI X ⊠ νI Y (νX ⊠ νY )I[−1] / νI X ⊠ νI Y / (νX ⊠ νY )I where the second morphism is induced by the lax monoidal structure on (−)I. We conclude that there is a natural equivalence (νX ⊛ νY )I ≃ (νX ⊠ νY )I. ✷ 4 Kunneth formula for dg-categories of singularities 4.1 The monoidal dg-category B and its action We keep our standing assumptions: A is an excellent strictly henselian dvr with perfect residue field k and fraction field K. We let S = Spec, A and s = Spec k, as usual, and choose an uniformizer π of A. S s (derived fiber product), considered as a derived scheme over S. The derived scheme G has a canonical structure of groupoid in derived schemes acting on s. The composition in the groupoid G induces a convolution monoidal structure on the dg-category of coherent complexes on G We let G := s ×h More explicitly, we have a map of derived schemes ⊙ : Cohb(G) ⊗A Cohb(G) −→ Cohb(G). G ×s G q / G, defined as the projection on the first and third components s ×S s ×S s → s ×S s. We then define ⊙ by the formula E ⊙ F := q∗(E ⊠s F ) for two coherent complexes E and F on G. More generally, if X −→ S is any scheme, with special fiber Xs (possibly a derived scheme, by taking the derived fiber at s), the groupoid G acts naturally on Xs via the natural projection qX : G ×s Xs ≃ (s ×S s) ×s (s ×S X) ≃ s ×S Xs −→ Xs. This defines an external action by E ⊙ M := (qX )∗(E ⊠s M). ⊙ : Cohb(G) ⊗A Cohb(Xs) −→ Cohb(Xs) The homotopy coherences issues for the above ⊙-structures can be handled using the fact that the construction Y 7→ Cohb(Y ) is in fact a symmetric lax monoidal ∞-functor from a certain ∞-category of correspondences between derived schemes to the ∞-category of dg-categories over A. As a result, Cohb(G) is endowed with a natural structure of a monoid in the symmetric monoidal ∞-category dgCatA, and that, for any X/S, Cohb(Xs) is naturally a module over Cohb(G) in dgCatA. However, for our purposes it will be easier and more efficient to provide explicit models for both Cohb(G) and its action on Cohb(Xs). This will be done locally in the Zariski topology in a similar spirit to [BRTV, Section 2]; the global construction will then be obtained by a rather straightforward gluing procedure. 28 / / / 4.1.1 The monoidal dg-categories B+ and B Let KA be the Koszul commutative A-dg-algebra of A with respect to π KA : A π / / A sitting in degrees [−1, 0]. The canonical generator of KA in degree −1 will be denoted by h. In the same way, we define the commutative A-dg-algebra K 2 A := KA ⊗A KA which is the Koszul dg-algebra of A with respect to the sequence (π, π). As a commutative graded A-algebra, K 2 A is SymA(A2[1]), and it is endowed with the unique multiplicative differential sending the two generators h and h′ in degree −1 to π (and hh′ to π · h′ − π · h). Moreover, K 2 map are the two natural inclusions KA −→ K 2 KA. The composition (or coproduct) in this Hopf algebroid structure is given by A has a canonical structure of Hopf algebroid over KA, in which the source and target A → A, whereas the unit is given by the multiplication K 2 ∆ := id ⊗ 1 ⊗ id : KA ⊗A KA = K 2 A = KA ⊗A KA ⊗A KA. A −→ K 2 A ⊗KA K 2 Finally, the antipode is the automorphism of K 2 Hopf algebroid endows the dg-category Mod(K 2 monoidal structure ⊙. It is explicitely given for two object E and F , by the formula A exchanging the two factors KA. This structure of A), of K 2(A)-dg-modules, with a unital and associative E ⊙ F := E ⊗KA F where the KA ⊗A KA ⊗A KA-module on the rhs is considered as a K 2 A module via the map ∆. The unit of this monoidal structure is the object KA, viewed as a K 2 A → KA. It is not hard to see that ⊙ preserves cofibrant K 2 A) into a monoidal model category in the sense of [Ho, Ch. 4]. Note however that the unit KA is not cofibrant in this model structure. A-dg-modules; more generally it makes Mod(K 2 A-module by the multiplication K 2 Definition 4.1.1 The monoidal dg-category B+ cofibrant dg-modules over K 2 with the monoidal structure ⊙ described above. A), the dg-category of all A which are perfect over A, together with the unit object KA. It is endowed str is defined to be Modc(K 2 By Appendix A, the localization of the dg-category B+ str along all quasi-isomorphisms, defines a monoidal dg-category. Definition 4.1.2 The monoidal dg-category B+ is defined to be the localization W −1 eq (B+ str), where Weq is the set of quasi-isomorphisms. symmetric monoidal ∞-category dgCatA. It is naturally a unital and associative monoid in the Remark 4.1.3 Note that B+ defined above is a model for Cohb(G), for our derived groupoid G = s×S s above. A is quasi-isomorphic to the normalization of the simplicial algebra k ⊗L A k. Now, since G → S is a closed immersion, a quasi-coherent complex E on G is coherent Indeed, the commutative dga K 2 29 iff its direct image on S is coherent, hence perfect, S being regular. In particular, we have that Cohb(G) is equivalent to the dg-category of all cofibrant K 2 A-dg-modules which are perfect over A The latter dg-category is also naturally equivalent to the localization of B+ str along quasi-isomorphisms12 We now introduce the weak monoidal dg-category B, defined as a further localization of B+. This will be our main "base monoid" for the module dg-categories we will be interested in. Definition 4.1.4 The weak monoidal dg-category B is defined to be the localization where W is the set of morphisms in B+ whose cone is perfect as a K 2 A-dg-module. B := LW (B+), are perfect over K 2 str whose cones str directly, as As B+ it itself defined as a localization of B+ str, if Wpe denotes the morphisms in B+ A, we have Weq ⊆ Wpe, then B can also be realized as localization of B+ B ≃ LWpeB+ str. Since the monoidal structure ⊙ is compatible with Wpe, this presentation implies that B comes equipped with a natural structure of an associative and unital monoid in dgCatA. Note, moreover, that B comes equipped with a natural morphism of monoids B+ −→ B given by the localization map. 4.1.2 The local actions Let now X = Spec R be a regular scheme flat over A. As done above for the monoid structure on B+, we will define a strict model for Cohb(Xs), together with a strict model for the Cohb(G)-action on Cohb(Xs). In order to do this, let KR be the Koszul dg-algebra of R with respect to π, which comes equipped with a natural map KA −→ KR of cdga's over A. We consider Modc(KR), the dg-category of all cofibrant KR-dg-modules which are perfect as R-modules (note that R is regular, and see Remark 4.1.3). The same argument as in Remark 4.1.3 then shows that this dg-category is naturally equivalent to Cohb(Xs). Moreover, Modc(KR) has a structure of a B+ str-module dg-category defined as follows. For E ∈ B+ str, and M ∈ Modc(KR), we can define E ⊙ M := E ⊗KA M, where, in the rhs, we have used the "right" KA-dg-module structure on E, i.e. the one induced by the composition KA ∼ / / A ⊗A KA id⊗u / KA ⊗A KA , u : A → KA being the canonical map. As E is either the unit or it is cofibrant over K 2 A (and thus cofibrant over KA), E ⊗KA M is again a cofibrant KR-module, and again perfect over R, i.e. E ⊙ M ∈ Modc(KR). By localization along quasi-isomorphisms, (see Proposition 4.1.5 for details) we 12We leave to the reader to check that adding the unit objet KA does not change the localization. 30 / obtain that Cohb(Xs) carries a natural B+-module structure as an object in the symmetric monoidal ∞-category dgCatA. We now apply a similar argument in order to define a B-action on Sing(Xs). Let again X = Spec R be a regular scheme over A, and consider Modc(KR) as a B+ str-module dg-category as above. Let WR,pe be the set of morphisms in Modc(KR) whose cones are perfect dg-modules over KR. By localization we then get a B-module structure on LWR,peModc(KR). Note that the localization LWR,peModc(KR) is a model for the dg-category Sing(Xs), which therefore comes equipped with the structure of a B-module in dgCatA. We gather the details of above constructions in the following Proposition 4.1.5 Let X = Spec R be a regular scheme, flat over S = Spec A, and Xs its special fiber. Then there is a canonical B+-module structure (resp., B-module structure) on Cohb(Xs) (resp., on Sing(Xs)), inside dgCatA. eq B+ str ⊗A W −1 A. We have W −1B∗ str ⊗A T str → W −1 str ⊗A T str → W −1 eq B+ T,eqT str ≃ T in dgCatA. Analogously, W −1 str ⊗A T str → T str (defined in 4.1.2) sends Weq ⊗ id ∪ id ⊗ WT,eq to WT,eq. Proof. This is an easy application of the localization results presented in Appendix A. We first treat the case of B+ and T := Cohb(Xs). If T str := Modc(Xr) and WT,eq denotes the quasi- isomorphisms in T str, we have W −1 str ≃ B+ in dgCatA. In order to apply the localization result of Appendix A, we need to prove that the tensor product ⊙ : B+ If L, L′ ∈ B+ str, : L → L′, w : E → E′ are quasi-isomorphisms, then w′ ⊙ idE is again a E, E′ ∈ T str, and w′ quasi-isomorphism (because L, and L′ are cofibrant over K 2 A, hence over KA, and thus w′ is in fact a homotopy equivalence), and the same is true for idL ⊙ w (since L is cofibrant over KA). Therefore, ⊙ does send Weq ⊗ id ∪ id ⊗ WT,eq to WT,eq, and there is an induced canonical map (Weq ⊗ id ∪ T,eqT str ≃ Cohb(Xs). By composing this with the natural equivalence id ⊗ WT,eq)−1B+ (Weq ⊗id∪id⊗WT,eq)−1B+ T,eqT str (Appendix A), we finally get our B+-module structure on Cohb(Xs) inside dgCatA. We now treat the case of B and T = Sing(Xs). Here, we consider the pairs (T str = Modc(Xr), WT ) where str, W ), where W are the maps in B+ WT are the maps in T str whose cones are perfect over KR, and (B∗ str str ≃ B, and W −1 whose cones are perfect over K 2 T T str ≃ Sing(Xs), and we need to prove that both W ⊙ id and id ⊙ WT are contained in WT . Let u : L → L′ ∈ W and v : E → E′ ∈ WT , and C(−) denote the cone construction. We have C(idL ⊙ v) ≃ L ⊗KA C(v) and C(u ⊙ idE) ≃ C(u) ⊗KA E. By hypothesis, L is perfect over A hence over KA (since KA is perfect over A), and C(v) is perfect over KA, since X → S is lci (as a map of finite type between regular schemes), and thus KA → KR is derived lci (recall that X/S is flat so that Xs is also the derived fiber), and pushforward along a lci map preserves perfect complexes. So, C(idL ⊙v) ∈ WT . On the other hand, the "right-hand" map KA → K 2 A (with respect to which L and L′ are viewed as KA-dg-modules in the definition of ⊙) is derived lci, hence C(u) is perfect over KA, being perfect over K 2 A by hypothesis. Moreover, since s → S is a closed immersion, E is perfect over KA iff (X → S)∗E is perfect (= coherent, S being regular) over S; but (Xs → X)∗E is perfect by hypothesis, and pushforward along X → S preserves perfect complexes, since X/S is lci. Therefore C(u ⊙ idE) ∈ WT , and we deduce that ⊙ does send W ⊗ id ∪ id ⊗ WT to WT . This gives us an induced canonical map (W ⊗ id ∪ id ⊗ WT )−1B+ T T str ≃ Sing(Xs). By composing this with the natural equivalence (W ⊗ id ∪ id ⊗ WT )−1B+ T T str (Appendix A), we finally get our B-module structure on Sing(Xs) inside dgCatA. str ⊗A T str → W −1 str ⊗A T str → W −1B+ str ⊗A W −1 31 ✷ Lemma 4.1.6 The natural morphism Cohb(Xs)o ⊗B+ B −→ Sing(Xs) is an equivalence of B-modules. Proof. This is a reformulation of [BRTV, Proposition 2.43]. ✷ 4.1.3 The global actions We now let X be a regular scheme, not necessarily affine anymore. We have by Zariski descent Cohb(Xs) ≃ lim Spec R ⊂ X Modc(KR), where the limit is taken over all affine opens Spec R of X. The right hand side of the above equivalence is a limit of dg-categories underlying B+-module structures (Proposition 4.1.5). As the forgetful functor from B+-modules to dg-categories reflects limits, this endows Cohb(Xs) with a unique structure of B+-module. In the same way, we have Zariski descent for Sing(Xs) in the sense that Sing(Xs) ≃ lim Spec R⊂X LWR,pe(Modc(KR)). The right hand side of the above equivalence is a limit of dg-categories underlying B-module structures (Proposition 4.1.5). As the forgetful functor from B-modules to dg-categories reflects limits, this makes the dg-category Sing(Xs) into a B-module in a natural way. An important property of these B+ and B-module structures is given in the following proposition. Proposition 4.1.7 Let X be a flat regular scheme over S. 1. The B+-module structure on Cohb(Xs) is cotensored. 2. The B-module structure on Sing(Xs) is cotensored. Proof. This follows from Remark 2.1.4 since the monoidal triangulated dg-categories B+ and B are both generated by their unit objects. ✷ 4.2 Kunneth formula for dg-categories of singularities In the previous section we have seen that, for any regular scheme X over S, the dg-category Sing(Xs) are equipped with a natural B-module structure. In this section we compute tensor products of dg- categories of singularities over B. 32 From a general point of view, let T ∈ dgCatA be a B-module, and assume that T is also co-tensored over B (Def. 4.1.7). Then T o has a natural structure of a B⊗−op-module given by co-tensorisation. By Proposition 4.1.7, we may take T = Sing(Xs), so that T o is a B⊗−op-module. In particular, if we have another regular scheme Y , we are entitled to consider the tensor product Sing(Xs)o ⊗B Sing(Ys), which is a well defined object in dgCatA. We further assume, for simplicity, that X and Y are also flat over S. The main result of this section is the following proposition, which is a categorical counterpart of our Kunneth formula for vanishing cycles (Proposition 3.4.2). Theorem 4.2.1 Let X and Y be two regular schemes, flat over S, with smooth generic geometric fibers. There is a natural equivalence in dgCatA Sing(Xs)o ⊗B Sing(Ys) ≃ Sing(X ×S Y ). Proof. As B is a localization of B+, the natural ∞-functor on enriched dg-categories (which is lax symmetric monoidal) dgCatB −→ dgCatB+ is fully faithful. Moreover, it commutes with tensor products in the following sense: for two objects T and T ′ in dgCatB, there is a natural equivalence T o ⊗B T ′ ≃ T o ⊗B+ T ′. Therefore, in order to prove the theorem it enough to construct an equivalence of dg-categories over A Sing(Xs)o ⊗B+ Sing(Ys) ≃ Sing(X ×S Y ). Now, we claim that the result is local on Z := X ×S Y . Indeed, we have two prestacks of dg-categories on the small Zariski site Zzar: U ×S V 7→ Sing(Us)o ⊗B Sing(Vs) U ×S V 7→ Sing(U ×S V ) for any two affine opens U ⊂ X and V ⊂ Y . These two prestacks are stacks of dg-categories, and thus we have equivalences in dgCatA Sing(Xs)o ⊗B Sing(Ys) ≃ lim U ⊂X,V ⊂Y Sing(Us)o ⊗B Sing(Vs) Sing(X ×S Y ) ≃ lim U ⊂X,V ⊂Y Sing(U ×S V ). (8) (9) The stack property (9) is proved in [BRTV, 2.3] (where it is moreover shown that this is a stack for the h-topology). The stack property (8) is then a consequence of the same descent argument for dg-categories of singularities. Indeed, we have the following lemma. Lemma 4.2.2 Let Z be an S-scheme, and F be a stack of OZ-linear dg-categories. Assume that F is a B⊗−op-module stack13, and let T0 be a B-module dg-category. Then, the prestack F ⊗B T0 of dg-categories of ZZar, sending W ⊂ Z to F (W ) ⊗B T0 is a stack. 13I.e., F is a stack on ZZar with values in the ∞-category of B⊗−op-modules in dgCatA. 33 Proof of Lemma 4.2.2. This is an application of the main result of [To2]. We denote, as usual, by bT := RHom(T, bA) the (non-small) dg-category of all T o-dg-modules. By the main result of [To2], the dg-category \(F (W ) ⊗B T0) lim W ⊂Z is compactly generated, and its dg-category of compact objects is equivalent in dgCatA to limW ⊂Z F (W )⊗B T0. Moreover, we have where b⊗ is the symmetric monoidal structure on presentable dg-categories. As b⊗ is rigid when restricted to compactly generated dg-categories, we have that b⊗ bB distributes over limits on both factors, and we have \(F (W ) ⊗B T0) ≃ \F (W )b⊗ bB bT0, Passing to the sub-dg-categories of compact objects, we find that \(F (Z) ⊗B T0) ≃ lim W ⊂Z \F (W )b⊗ bB bT0. F (Z) ⊗B T0 ≃ lim W ⊂Z F (W ) ⊗B T0 which is the statement of the lemma. ✸ Lemma 4.2.2 immediately implies the stack property (8). We are thus reduced to the case where X and Y are both affine, and we have to produce an equivalence that is compatible with Zariski localization on both X and Y . Sing(Xs)o ⊗B Sing(Ys) ≃ Sing(X ×S Y ) In this case, we start by the following. Lemma 4.2.3 There is a natural equivalence of dg-categories over A Cohb(Xs)o ⊗B+ Cohb(Ys) ≃ Cohb Zs(Z), where Z := X ×S Y , and the right hand side is the dg-category of coherent complexes on Z with cohomology supported on the special fiber Zs. This equivalence is furthermore functorial in X and Y . Proof of lemma 4.2.3. We use the strict models introduced in our last section. Let X := Spec B and Y := Spec C, and KB, KC the Koszul dg-algebras of B and C with respect to the element π. As in our previous section, we have the Hopf dg-algebroid K 2 A and its monoidal dg-category of modules Modc(K 2 A) acts on both Modc(KB) and Modc(KC), the dg-categories of cofibrant KB (resp. KC) dg-modules which are perfect over B (resp. over C). A). Now, Modc(K 2 We define a dg-functor θ : (Modc(KB))o ⊗Modc(K 2 A) Modc(KC) −→ Modc(B ⊗A C), where Modc(B ⊗A C) is the category of perfect B ⊗A C-dg-modules. This dg-functor sends a pair of objects (E, F ) to the object D(E) ⊗KA F , where D(E) = HomKB(E, KB) is the KB-linear dual of E. After localization with respect to quasi-isomorphisms, we get a well defined morphism in dgCatA In order to finish the proof, we have to check the following two conditions: θ : Cohb(Xs)o ⊗B+ Cohb(Ys) −→ Cohb(Z). 34 1. The image of θ generates (by shifts, sums, cones and retracts) the full sub-dg-category Cohb Zs(Z); 2. The dg-functor θ above is fully faithful. Now, on the level of objects the dg-functor θ sends a pair (E, F ), of coherent complexes on Xs and Ys, to the coherent complex on Z j∗(D(E) ⊠k F ), where j : Zs ֒→ Z is the closed embedding, ⊠k denotes the external product on Zs = Xs ×s Ys, and D(E) denotes the Grothendieck dual of E on the Gorenstein scheme Xs. It is known that coherent complexes of the form E ⊠k F generate Cohb(Zs). As the coherent complexes of the form j∗(G), for G ∈ Cohb(Zs), clearly generate the dg-category Cohb Zs(Z), we see that condition (1) above is satisfied. It now remains to show that θ is fully faithful. Given two pairs of objects (E, F ), (E′, F ′) ∈ Cohb(Xs) × Cohb(Ys), we have the morphism induced by θ RHom(E′, E) ⊗B+(1) RHom(F, F ′) −→ RHom(j∗(D(E) ⊠ F ), j∗(D(E′) ⊠ F ′)), where B+(1) denotes the algebra of endomorphism of the unit in B+. As we have already seen, B+ ≃ k[u] as an E1-algebra. As Xs and Ys are Gorenstein schemes, the structure sheaf O is a dualizing complex, and E 7→ D(E) is an (anti)equivalence. We thus have RHom(E′, E) ≃ RHom(D(E), D(E′)), and the above morphism can thus be written in the form RHom(E, E′) ⊗k[u] RHom(F, F ′) −→ RHom(j∗(E ⊠ F ), j∗(E′ ⊠ F ′)), where it is simply induced by the direct image functor j∗. Both the source and the target of the above morphism enter in a distinguished triangle. On the left hand side, for any two k[u]-dg-modules M and N, we have a triangle of A-dg-modules (M ⊗k N)[−2] / M ⊗k N / M ⊗k[u] N, where the morphism on the right is the natural projection. This exact triangle follows from the isomor- phism M ⊗k[u] N ≃ (M ⊗k N) ⊗k[u]⊗kk[u] k[u], and from the exact triangle of k[u] bi-dg-modules (k[u] ⊗k k[u])[−2] a / k[u] ⊗k k[u] m / / k[u], where a is given by multiplication by (u ⊗ 1 − 1 ⊗ u), and m is the product map. On the right hand side, we have, by adjunction, RHom(j∗(E ⊠ F ), j∗(E′ ⊠ F ′)) ≃ RHom(j∗j∗(E ⊠ F ), E′ ⊠ F ′). The adjunction map j∗j∗ → id, provides an exact triangle of coherent sheaves on Zs E ⊠k F [1] / j∗j∗(E ⊠k F ) / E ⊠k F. The coboundary map of this triangle E ⊠k F −→ E ⊠k F [2] is precisely given by the action of k[u]. We thus obtain another exact triangle (RHom(E, F ) ⊗k RHom(E′, F ′))[−2] / RHom(E, F ) ⊗k RHom(E′, F ′) 35 / / / / / / / / / RHom(j∗(E ⊠ F ), j∗(E′ ⊠ F ′)). By inspection, the morphism θ is compatible with these two triangles, and provides an equivalence RHom(E, E′) ⊗k[u] RHom(F, F ′) −→ RHom(j∗(E ⊠ F ), j∗(E′ ⊠ F ′)). Therefore θ is fully faithful. ✸ Since X/S and Y /S are generically smooth, also Z = X ×S Y is generically smooth over S. Therefore Sing(Z) ≃ Cohb Zs(Z)/Perf Zs(Z). Hence, in order to finish the proof of Proposition 4.2.1, we are left to prove that the image of Perf Zs(Z) under a quasi-inverse of θ is generated by Cohb(Xs)o ⊗B+ Perf(Ys) and Perf(Xs)o ⊗B+ Cohb(Ys). Now, the dg-category Cohb(Xs)o ⊗B+ Perf(Ys) is generated by the image of the natural dg-functor Cohb(Xs)o ⊗A Perf(Ys) −→ Cohb(Xs)o ⊗B+ Perf(Ys), and the dg-category Perf(Xs)o ⊗B+ Cohb(Ys) is generated by the image of the natural dg-functor Perf(Xs)o ⊗A Cohb(Ys) −→ Perf(Xs)o ⊗B+ Cohb(Ys). Therefore, what we have to prove is that Perf Zs(Z) is generated by objects of the form E ⊠A F , for E ∈ Cohb(Xs), F ∈ Cohb(Ys), one of them being perfect. We first notice that, if E or F is perfect, indeed, E ⊠A F is a perfect complex on Z. To see this, we may localize on Z, and write X = Spec B and Y = Spec C. The objects E and F then correspond to bounded coherent complexes over Bs = B ⊗A k and Cs = C ⊗A k respectively. Assume that E is perfect (the complementary case being totally analogous). By d´evissage we can assume that E = Bs, so that E ⊠A F ≃ (B ⊠A F ) ⊗A k. In other words, if we write j : Ys ֒→ Y and i : Zs ֒→ Z for the closed embeddings, and p : Z −→ Y for the second projection, we have E ⊠A F ≃ i∗i∗p∗(j∗(F )). But j∗(F ) is perfect on Y (because Y is regular), so p∗j∗(F ) is perfect on Z. Moreover, since i is an lci morphism, i∗i∗ preserves perfect complexes on Z, and we conclude that, indeed, E ⊠A F is a perfect complex on Z. Finally, the fact that the image of Cohb(Xs)o⊗B+ Perf(Ys) and Perf(Xs)o⊗B+ Cohb(Ys) inside Cohb Zs(Z) generates the whole dg-category Perf Zs(Z) follows from the fact that the image of Perf(Xs)o ⊗A Perf(Ys) inside Perf(Z) already generates Perf Zs(Z). Indeed, for two perfect complexes E on Xs and F on Ys, the image of E ⊠A F in Perf(Z) is of the form i∗(E ⊠k F ) ⊕ i∗(E ⊠k F )[1]. As objects of the form E ⊠k F generate Perf(Zs) we are done, and Proposition 4.2.1 is proved. ✷ 4.3 Saturatedness As a consequence of the Kunneth formula for dg-categories of singularities we prove the following result. Proposition 4.3.1 Let X be a regular and flat S-scheme. 36 / 1. If X is proper over S, then the dg-category Cohb(Xs) is proper over B+. 2. If X is proper over S, then the dg-category Sing(Xs) is saturated over B. Proof. (1) We have to show that the big morphism h : \Cohb(Xs)o ⊗A Cohb(Xs) −→ cB+ ≃ \Cohb(G). is small. Here G = s ×S s, and the morphism h is obtained as follows. The derived scheme G is a derived groupoid acting on Xs by means of the natural projection on the last two factors The projection on the first and third factors provides another morphism µ : G ×s Xs = s ×S s ×S X −→ Xs. p : G ×s Xs −→ Xs. Thus, the morphisms p and µ together define a morphism of derived schemes q : G ×s Xs −→ Xs ×S Xs. Finally, we denote by r : G ×s Xs −→ G the natural projection. Now, for two coherent complexes E and F on Xs, we first form the external Hom complex HomA(E, F ) which is a coherent complex on Xs ×S Xs, and we have h(E, F ) ≃ r∗(q∗HomA(E, F )). This is a quasi-coherent complex on G. Both q and r are local complete intersection morphisms of derived schemes, and moreover r is proper. This implies that q∗ and r∗ preserve coherent complexes, and thus that h(E, F ) is coherent on G. (2) We have Sing(Xs) ≃ Cohb(Xs) ⊗B+ B, thus (1) implies that Sing(Xs) is proper over B. To prove it is smooth we need to prove that the coevaluation big morphism A −→ Sing(Xs)o ⊗B Sing(Xs) is a small morphism. Using our Kunneth formula for dg-category of singularities (Proposition 4.2.1) this morphism corresponds to the data of an ind-object in Sing(X ×S X). This object is the structure sheaf of the diagonal ∆X inside X ×S X which is an object in Sing(X ×S X). This shows that the coevaluation morphism is a small morphism and thus that Sing(Xs) is indeed saturated over B. ✷ Remark 4.3.2 Proposition 4.3.1 (2) remains true if we only suppose that the singular locus of Xs is proper over s = Spec k. 5 Application to Bloch's conductor formula with unipotent monodromy In this Section we first recall Bloch's Conductor Conjecture and the current state of the art for it, then we prove a version of Bloch's conductor under the hypothesis that the monodromy action is unipotent, where Bloch's number is replaced by a categorical Bloch class (see Definition 5.2.1). We are convinced that Bloch's number always agree with the categorical Bloch class but we will not prove (nor use) this fact in this paper. We are aware of a proof of this comparison in the geometric case (i.e. when the inclusion s → S has a retraction) but we will defer this to another paper, where we hope to give the comparison also when S has mixed characteristic. 37 5.1 Bloch's Conductor Conjecture Our base scheme is a discrete valuation ring S = Spec A, with perfect residue field k, and fraction field K. Let p : X −→ S be proper and flat morphism of finite type, and of relative dimension n. We assume that the generic fiber XK is smooth over K, and that X is a regular scheme. We write ¯K for the separable closure of K (inside a fixed algebraic closure). In his 1985 paper [Bl], Bloch formulated the following conductor formula conjecture which is a kind of vast arithmetic generalization of Gauss-Bonnet formula, where an intersection theoretic (coherent) term, the Bloch's number, is conjectured to be equal to an arithmetic (´etale) term, the Artin conductor. We address the reader to [Bl] and [Ka-Sa] for more detailed definitions of the various objects involved in the statement. Conjecture 5.1.1 [Bloch's conductor Conjecture] Under the above hypotheses on p : X → S, we have an equality [∆X , ∆X]S = χ(X¯k, ℓ) − χ(X ¯K, ℓ) − Sw(X ¯K), where χ(Y, ℓ) denotes the Qℓ-adic Euler characteristic of a variety Y , for ℓ prime to the characteristic of k, Sw(X ¯K) is the Swan conductor of the Gal( ¯K/K)-representation H ∗(X ¯K, Qℓ), and [∆X , ∆X]S is Bloch's number of X/S, i.e. the degree in CH0(k) ≃ Z of Bloch's localised self-intersection (∆X , ∆X)S ∈ CH0(Xk) of the diagonal in X. The (negative of the) rhs is called the Artin conductor of X/S, and denoted by Art(X/S). Conjecture 5.1.1 is known to hold in several special cases that we recall below: 1. When k is of characteristic zero, Conjecture 5.1.1 follows from the work of [Kap]. When further- more Xs has only isolated singularities the conductor formula was known as the Milnor formula stating that the dimension of the space of vanishing cycles equals the dimension of the Jacobian ring. 2. When S is equicharacteristic, Conjecture 5.1.1 has been proved recently in [Sai], based on Beilin- son's theory of singular support of ℓ-adic sheaves. The special subcase of isolated singularities already appeared in [SGA7-I, Exp. XVI]. 3. When X is semi-stable over S, i.e. the reduced special fiber (Xs)red ⊂ X is a normal crossing divisor, Conjecture 5.1.1 was proved in [Ka-Sa]. In view of the above list of results, one of the major open cases is that of isolated singularities in mixed characteristic, which is the conjecture appearing in Deligne's expos´e [SGA7-I, Exp. XVI]. It is classical and easy to see that Conjecture 5.1.1 can be reduced to the case where S is excellent and strictly henselian (so that k algebraically closed)14. We will thus assume from now on that k is algebraically closed. 14We first reduce to the complete dvr (hence henselian and excellent) case by proper base-change, and then we further reduce to the excellent, strictly henselian case. 38 5.2 An analog of Bloch's Conjecture for unipotent monodromy We start by introducing a categorical variant of Bloch's number [∆X .∆X]S defined in terms of dg- categories of singularities. The dg-category Sing(Xs) of singularities of the special fiber comes equipped with its canonical B-module structure described in Proposition 4.1.7. As X is proper over S, we know by Proposition 4.3.1 that Sing(Xs) is saturated over B. We are thus entitled to take the trace (Definition 2.4.4) of the identity of Sing(Xs), which is a morphism A −→ HH(B/A) in dgCatA. This trace morphism is then, by definition, determined by a perfect HH(B/A)o-dg-module HH(Sing(Xs)/B), and thus provides us with a class in K-theory [HH(Sing(Xs)/B)] ∈ K0(HH(B/A)). The Chern character natural transformation (applied to HH(B/A), see Definition 2.3.1) of this class is an element Chℓ,HH(B/A)([HH(Sing(Xs)/B)]) in π0rℓ(HH(B/A)) = H 0(S´et, rℓ(HH(B/A))). Definition 5.2.1 With the above notations and hypotheses on X/S, the categorical Bloch class of X/S is defined as [∆X , ∆X]cat S := Chℓ([HH(Sing(Xs)/B]) ∈ H 0(S´et, rℓ(HH(B/A))). Notice that though B is just an associative algebra in dgCatA (an E2-algebra over A), its ℓ-adic realization rℓ(B) is in fact a commutative monoid in ShQℓ(S), naturally equivalent to i∗(Qℓ(β)⊕Qℓ(β)[1]) where i : s → S is the inclusion of the closed point. Therefore there is a canonical augmentation a : HH(rℓ(B)/rℓ(A)) → rℓ(B), hence an induced augmentation a : H 0(S´et, HH(rℓ(B)/rℓ(A))) = π0HH(rℓ(B)/rℓ(A)) −→ π0rℓ(B) = H 0(S´et, rℓ(B)) ≃ Qℓ that is a left inverse to σ : Qℓ ≃ H 0(S´et, rℓ(B)) → H 0(S´et, HH(rℓ(B)/rℓ(A))) (induced by rℓ(B) → HH(rℓ(B)/rℓ(A))). For λ ∈ Qℓ, we will simply write λ∧ for the image of λ via the composition Qℓ ≃ H 0(S´et, rℓ(B)) σ / H 0(S´et, HH(rℓ(B)/rℓ(A))) α / / H 0(S´et, rℓ(HH(B/A))). We are now ready to prove our version of Bloch's Conductor Conjecture for unipotent monodromy.. Theorem 5.2.2 Let X/S be as in Conjecture 5.1.1, and assume further that the inertia subgroup I := Gal( ¯K/K unr) ⊆ Gal( ¯K/K) acts unipotently on H ∗(X ¯K, Qℓ). Then we have an equality in H 0(S´et, rℓ(HH(B/A))). [∆X, ∆X ]cat S = χ(X¯k, ℓ)∧ − χ(X ¯K, ℓ)∧ 39 / Remark 5.2.3 Since we have not proved that the map α : H 0(S´et, HH(rℓ(B)/rℓ(A))) → H 0(S´et, rℓ(HH(B/A))) is injective, the equality in Theorem 5.2.2 is not a priori an equality of integers (and it might even be the trivial equality 0 = 0 of classes in H 0(S´et, rℓ(HH(B/A)))). However, we conjecture that the canonical map u : B → HH(B/A) is in fact an A1-homotopy equivalence; this would imply that rℓ(u) : rℓ(B) → rℓ(HH(B/A)) is an equivalence in ShQℓ(S), so that H 0(S´et, rℓ(HH(B/A))) ≃ H 0(S´et, rℓ(B)) ≃ Qℓ. In particular α would be injective, and the equality in Theorem 5.2.2 would indeed be an equality of integers. Remark 5.2.4 Recall that the inertia group I sits in an extension of pro-finite groups 1 / P / I / It / 1 , where It is the tame inertia quotient, and the wild inertia P is a pro-p-subgroup. For any continuous finite dimensional Qℓ-representation V of I, the group P acts through a finite quotient on V (see []). Therefore, if I is supposed to act unipotently on V , then P acts necessarily trivially, i.e. the I-action factors through a It-action, i.e., by definition, the I action is tame. By definition, the Swan conductor [Ka-Sa, §6.1]) vanishes for a tame I-representation V . As a consequence, granting SwI(V )(see e.g. Remark 5.2.3, under the hypothesis of unipotent action of I on H ∗(X ¯K, Qℓ), Conjecture 5.1.1 becomes Theorem 5.2.2 if Bloch's number [∆X , ∆X ]S is replaced by the categorical Bloch class [∆X , ∆X]cat S . Though we currently know a proof only in the geometric case, we are actually convinced that the categorical Bloch class is always, i.e. regardless any hypothesis on the monodromy action and on the mixed or equal characteristic property of S, an integer equal to Bloch's intersection number [∆X , ∆X]S appearing in Conjecture 5.1.1. This fact will not be used in this paper and will be more closely investigated in a future one. Proof of Thm. 5.2.2. We may suppose S strictly henselian, so that I = Gal( ¯K/K). We want to apply our trace formula (Theorem 2.4.9) to T = Sing(Xs) and f = id. In order to do this, we need to check that the conditions of being saturated and ℓ⊗-admissible over B are met by such T . By Proposition 4.3.1, we know that T is saturated over B. Let us now show that T is also ℓ⊗-admissible over B, i.e. that the canonical map ϕ : rℓ(T o) ⊗rℓ(B) rℓ(T ) −→ rℓ(T o ⊗B T ) is an equivalence. First of all we notice that, since both the source and the target of ϕ are ℓ-adic complexes on S, supported at s, it is enough to show that i∗(ϕ) is an equivalence, i : s → S being the canonical inclusion (note that i∗i∗ is equivalent to the identity functor). Let us first elaborate on the target of i∗(ϕ). By Kunneth for dg-categories of singularities, we have the canonical equivalence of Proposition 4.2.1 Moreover, since a unipotent action of I is also tame, by Cor. 3.4.5 we have T o ⊗B T ≃ Sing(X ×S X). rℓ(Sing(X ×S X)) ≃ (pZs)∗(νX ⊠ νX )I(β), 40 / / / / where pZs : Zs = Xs ×s Xs → S is the composite Zs cycles for X/S on Xs. Therefore qZs / / s i / S , and, as usual, νX denotes vanishing i∗rℓ(Sing(X ×S X)) ≃ (qZs)∗(νX ⊠ νX )I(β) in ShQℓ(s). Since qZs = qXs ×s qXs, where qXs : Xs → s is the canonical map, we have i∗rℓ(Sing(X ×S X)) ≃ (qZs)∗(νX ⊠ νX)I(β) ≃ ((qXs)∗(νX ) ⊗Qℓ (qXs)∗(νX ))I(β) (10) in ShQℓ(s), that we may rewrite as i∗rℓ(Sing(X ×S X)) ≃ (H(Xs, νX) ⊗Qℓ H(Xs, νX))I(β) (11) where we have introduced the hypercohomology H(Xs, E) := (qXs)∗(E), for E ∈ ShQℓ(Xs). Let us now look more carefully at the source of the map i∗(ϕ). By the main theorem of [BRTV] (see also formula (4)), we have where pXs : Xs → S is the composite Xs qXs / / s i / S . Therefore, rℓ(T ) ≃ (pXs)∗(νX[−1]I)(β) i∗(rℓ(T o) ⊗rℓ(B) rℓ(T )) ≃ i∗(rℓ(T o)) ⊗i∗rℓ(B) i∗rℓ(T ) ≃ ((qXs)∗(νX [−1])I ⊗QI ℓ qXs)∗(νX [−1])I)(β) (12) in ShQℓ(s), that we may rewrite as15 i∗(rℓ(T o) ⊗rℓ(B) rℓ(T )) ≃ (H(Xs, νX[−1])I ⊗QI ℓ H(Xs, νX[−1])I)(β). (13) By recalling that E(β) ≃ E[−2](β), for any E ∈ ShQℓ(Xs), and by combining (11) and (13), we see that i∗(ϕ) : i∗(rℓ(T o) ⊗rℓ(B) rℓ(T )) −→ i∗rℓ(T o ⊗B T ) is then equivalent to ψ[−2](β), where ψ is the lax-monoidal structure morphism for the functor (−)I, applied to the pair of ℓ-adic complexes (H(Xs, νX ), H(Xs, νX)), ψ : H(Xs, νX)I ⊗QI ℓ H(Xs, νX)I −→ (H(Xs, νX) ⊗Qℓ H(Xs, νX))I. (14) The fact that the morphism ψ is an equivalence is a consequence of the following general lemma. c(Spec ¯K, Qℓ) consist- Lemma 5.2.5 Let Duni(I, Qℓ) be the full sub-∞-category of Dc(Spec K, Qℓ) ≃ DI ing of all objects E for which the action of I on each H i(E) is unipotent. Then the invariant ∞-functor induces an equivalence of symmetric monoidal ∞-categories (−)I : Duni(I, Qℓ) ≃ Dperf(Qℓ[ǫ1]), where Qℓ[ǫ1] is the free commutative Qℓ-dg-algebra generated by ǫ1 in degree 1, and Dperf(Qℓ[ǫ1]) is its ∞-category of perfect dg-modules. 15Note that rℓ(T ) ≃ rℓ(T o) because the K-theory of a dg-category is canonically isomorphic to the K-theory of the opposite dg-category. 41 / / Proof of lemma 5.2.5. This is a well known fact. The ∞-functor (−)I : Dc(Spec K, Qℓ) −→ Dc(Qℓ) is lax symmetric monoidal, so it induces a lax monoidal ∞-functor (−)I : Dc(Spec K, Qℓ) −→ D(QI ℓ). It is easy to see that QI ℓ is canonically equivalent to Qℓ[ǫ1], and the choice of such an equivalence only depends on the choice of a generator of H 1(I, Qℓ) ≃ Qℓ. We thus have an induced lax symmetric monoidal ∞-functor (−)I : Dc(Spec K, Qℓ) −→ D(Qℓ[ǫ1]). The above ∞-functor is in fact symmetric monoidal, as it preserves unit objects, and moreover Dc(Spec K, Qℓ) is generated by the unit object Qℓ. The lemma is then a direct consequence of the following fact: compactly generated stable ∞-category C, then the ∞-functor let x ∈ C be a compact object in a C(x, −) : C −→ ModSp(End(x)) induces an equivalence of stable ∞-categories hxi ≃ PerfSp(End(x)), where hxi ⊂ C denotes the full stable ∞-category generated by the object x, and End(x) denotes the ♦ ring spectrum of endormophisms of x. The above lemma implies that the map ψ[−2] : H(Xs, νX [−1])I ⊗QI ℓ H(Xs, νX [−1])I −→ (H(Xs, νX[−1]) ⊗Qℓ H(Xs, νX[−1]))I is an equivalence, and thus, as observed above, the same is true for i∗(ϕ), and hence for the admissibility map ϕ : rℓ(T o) ⊗rℓ(B) rℓ(T ) −→ rℓ(T o ⊗B T ), so that T is indeed ℓ⊗-admissible over B as we wanted. Now that we have checked all the conditions for the trace formula of Theorem 2.4.9 to hold in our case, we get an equality Chℓ([HH(T /B, id)]) = trrℓ(B)(id : rℓ(T ) → rℓ(T )) (15) in H 0(S´et, rℓ(HH(B/A))). Recall that trrℓ(B)(id : rℓ(T ) → rℓ(T )) := α(T rrℓ(B)(id : rℓ(T ) → rℓ(T ))), where α : H 0(S´et, HH(rℓ(B)/rℓ(A))) → H 0(S´et, rℓ(HH(B/A))) is the canonical map. We are left to identify the two sides of (15). The left hand side is, by definition, our categorical Bloch number [∆X , ∆X]cat S . 42 Let us now investigate the r.h.s. of (15): we need to show that T rrℓ(B)(rℓ(idT )) = χ(X¯k, ℓ)−χ(X ¯K, ℓ). As already noticed, rℓ(B) is a commutative monoid in ShQℓ(S), naturally equivalent to i∗(Qℓ(β) ⊕ Qℓ(β)[1]) where i : s → S is the inclusion of the closed point. Therefore there is a canonical augmenta- tion a : HH(rℓ(B)/rℓ(A)) → rℓ(B), hence an induced augmentation a : H 0(S´et, HH(rℓ(B)/rℓ(A))) = π0HH(rℓ(B)/rℓ(A)) −→ π0rℓ(B) = H 0(S´et, rℓ(B)). Consider the diagram H 0(S´et, HH(rℓ(B)/rℓ(A))) α / H 0(S´et, rℓ(HH(B/A))) a σ 4❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ can Qℓ ≃ H 0(S´et, rℓ(B)) where can is the map induced by the canonical map u : B → HH(B/A), and σ is the map induced by rℓ(u), so that a ◦ σ = id. By compatibility between the non-commutative trace and the commutative trace (Remark 2.4.5) a(T rrℓ(B)(rℓ(idT )) = T rc rℓ(B)(rℓ(idT )) . Since rℓ(B) ≃ i∗(Qℓ(β) ⊕ Qℓ(β)[1]), we have K0(rℓ(B)) ≃ Z and the following commutative diagram Qℓ ≃ H 0(S´et, rℓ(B)) K0(rℓ(B)) a σ / H 0(S´et, HH(rℓ(B)/rℓ(A))) T r(id−) T rc(id−) 1❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞ 4❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ k❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ iso s s s s s s s s s s s s s s s s s s s s Z / H 0(S´et, rℓ(B)) 9s s s s s s where the maps with source Z are the unique maps of commutative rings, and a ◦ T r(id−) = T rc(id−) again by the functorial compatibility of non-commutative and commutative traces. By considering the class [rℓ(T )] in K0(rℓ(B)), and using that a ◦ σ = id, we deduce from the previous commutative diagram the equality of ℓ-adic numbers σ(T rc rℓ(B)(rℓ(idT )) = T rrℓ(B)(rℓ(idT )). (16) ℓ is such that K0(QI Now, unfolding the definition of T rrℓ(B), and using Lemma 5.2.5, the ℓ-adic number (16) can be described as follows. The dg-algebra QI ℓ) ≃ Z. Viewing Z inside Qℓ, this isomorphism is induced by sending the class of dg-module E to the trace of the identity inside HH0(QI ℓ) ≃ Qℓ. Using the functoriality of the trace for the morphism of commutative dg-algebras QI ℓ(β), we see that (16) is simply the trace of the identity on H(Xs, νX[−1]), as an object in Duni(I, Qℓ). This trace is easy to compute, as it equals 1 on the unit object Qℓ. Since the unit object generates Duni(I, Qℓ), we have that the trace of the identity on any object E ∈ Duni(I, Qℓ) equals the Euler characteristic of the underlying complex of Qℓ-vector spaces. ℓ −→ QI We thus have shown that (16) can be rewritten as T rrℓ(B)(rℓ(idT )) = X (−1)i dimQℓ H i−1(Xs, νX). i 43 / O O 4 / / 4 1 k a a ` ` 9 However, by proper base change, the complex H(Xs, νX) appears in an exact triangle H(X¯k, Qℓ) / H(X ¯K, Qℓ) / H(Xs, νX) and thus we have the equality T rrℓ(B)(rℓ(idT )) = χ(X¯k, ℓ) − χ(X ¯K, ℓ) in Z ֒→ Qℓ ֒→ H 0(S´et, HH(rℓ(B)/rℓ(A))). Therefore as required. ✷ [∆X , ∆X]cat S = χ(X¯k, ℓ)∧ − χ(X ¯K, ℓ)∧, A Localizations of monoidal dg-categories In this Appendix we remind some basic facts about localizations of dg-categories as introduced in [To1]. The purpose of the section is to explain the multiplicative properties of the localization construction. In particular, we explain how localization of strict monoidal dg-categories gives rise to monoids in dgCatA, and thus to monoidal dg-categories in the sense of our Definition 2.1.1. Let T be a dg-category over A, together with W a set of morphisms in Z 0(T ), the underlying category of T (this is the category of 0-cycles in T , i.e. Z 0(T )(x, y) := Z 0(T (x, y))). For the sake of brevity, we will just say that W is a set of maps in T . In other words, we allow W not to be strictly speaking a subset of the set of morphisms in T , but just a set together with a map W → Mor(T ) from W to the set of morphisms in T . Recall that a localization of T with respect to W , is a dg-category LW T together with a morphism in dgCatA l : T −→ LW T such that, for any U ∈ dgCatA, map induced by l on mapping spaces Map(LW T, U) −→ Map(T, U) is fully faithful and its image consists of all T → U sending W to equivalences in U (i.e. the induced functor [T ] → [U] sends elements of W to isomorphisms in [U]). As explained in [To1], localization always exists, and are unique up to a contractible space of choices (because they represents an obvious ∞-functor). We will describe here a model for (T, W ) 7→ LW T which will have nice properties with respect to tensor products of dg-categories. For this, let dgcatW,c be the category of pairs (T, W ), where T is a dg-category with cofibrant hom's over A, and W a set of maps in T . Morphisms (T, W ) −→ (T ′, W ′) in dgcatW,c A are dg-functors T −→ T ′ sending W to W ′. A We fix once for all a factorization ∆1 A j / I p 1 A, / ∆ 44 / / / / A is the A-linearisation of the category ∆1 that with j a cofibration and p a trivial fibration. Here ∆1 1 classifies morphisms, and ∆ A is the linearisation of the category that classifies isomorphisms. For an object (T, W ) ∈ dgcatW,c A we define W −1T by the following cocartesian diagram in dg-categories `W ∆1 A `W I T / W −1T, where `W ∆1 Lemma A.0.1 The canonical morphism l : T −→ W −1T defined above is a localization of T along W . A −→ T is the canonical dg-functor corresponding to the set W of morphisms in T . Proof. According to [To1], the localization of T can be construced as the homotopy push-out of dg-categories `W ∆1 A T `W A / LW T. The lemma then follows from the observation that when T has cofibrant hom's, then the push-out diagram defining W −1T is in fact a homotopy push-out diagram. ✷ The construction (T, W ) −→ W −1T clearly defines a functor dgcatW,c A −→ dgcatc A to dgcatc from dgcatW,c comes equipped with a natural symmetric colax monoidal structure. Indeed, dgcatW,c monoial category, where the tensor product is given by A, the category of dg-categories with cofibrant hom's. Moreover, this functor A is a symmetric A (T, W ) ⊗ (T ′, W ′) := (T ⊗A T ′, W ⊗ id ∪ id ⊗ W ′). We have a natural map T ⊗A T ′ −→ (W −1T ) ⊗A ((W ′)−1T ′), which by construction has a canonical extension (W ⊗ id ∪ id ⊗ W ′)−1(T ⊗A T ′) −→ (W −1T ) ⊗A ((W ′)−1T ′). A The unit in dgcatW,c is (A, ∅), which provides an canonical isomorphism (∅)−1A ≃ A. These data endow the functor (T, W ) 7→ W −1T with a symmetric colax monoidal structure. By composing with the canonical symmetric monoidal ∞-functor dgcatc A −→ dgCatA, we get a symmetric colax monoidal ∞-functor dgcatW,c A −→ dgCatA, which sends (T, W ) to W −1T . By [To3, Ex. 4.3.3], this colax symmetric monoidal ∞-functor is in fact monoidal. We thus have a symmetric monoidal localization ∞-functor W −1(−) : dgcatW,c A −→ dgCatA. 45 / / / / / / As a result, if T is a (strict) monoid in dgcatW,c A , then W −1T carries a canonical structure of a monoid in dgCatA. This applies particularly to stict monoidal dg-categories endowed with a compatible notion of equivalences. By MacLane coherence theorem, any such a structure can be turned into a strict monoid in dgcatW,c A , and by localization into a monoid in dgCatA. In other words, the localization of a monoidal dg-categorie along a set of maps W that is compatible with the monoidal structure, is a monoid in dgCatA. The same is true for dg-categories which are modules over a given monoidal dg-category. References [SGA4-III] Artin, Michael; Alexandre Grothendieck; Jean-Louis Verdier, eds. (1972). S´eminaire de G´eom´etrie Alg´ebrique du Bois Marie - 1963-64 - Th´eorie des topos et cohomologie ´etale des sch´emas - (SGA 4) - vol. 3. Lecture notes in mathematics 305. Berlin; New York: Springer-Verlag. pp. vi+640. ISBN 978-3-540-06118-2. MR 0354654. doi:10.1007/BFb0070714 [Be-Be] A. A. Beilinson, J. Bernstein, A proof of Jantzen conjectures, in Gel'fand, Sergei; Gindikin, Simon, I. M. Gel'fand Seminar, Adv. Soviet Math., 16, Providence, R.I.: American Mathematical Society, 1993, pp. 1-50, ISBN 978-0-8218-4118-1 [Bla] A. Blanc, Topological K-theory of complex noncommutative spaces. Compos. Math. 152 (2016), no. 3, 489-555. [BRTV] A. Blanc, M. Robalo, B. Toen, G. Vezzosi, Motivic realizations of singularity categories and vanishing cycles, Journal de l'´Ecole Polytechnique, Tome 5 (2018), p. 651-747. [Bl] S. Bloch, Cycles on arithmetic schemes and Euler characteristics of curves, Alg. Geometry, Bow- doin, 1985, 421-450, Proc. Symp. Pure Math. 46, Part 2, A.M.S., Providence, RI (1987). [SGA7-II] P. Deligne, N. Katz, eds. S´eminaire de G´eom´etrie Alg´ebrique du Bois Marie - 1967-69 - Groupes de monodromie en g´eom´etrie alg´ebrique - (SGA 7) - vol. 2. Lecture notes in mathematics 340. Berlin; New York: Springer-Verlag. pp. x+438 (1973). [Ga-Lu] D. Gaitsgory, J. Lurie, Weil's conjecture over function fields, Preprint available at http://www.math.harvard.edu/∼lurie/papers/tamagawa.pdf [SGA7-I] A. Grothendieck, S´eminaire de G´eom´etrie Alg´ebrique du Bois Marie - 1967-69 - Groupes de monodromie en g´eom´etrie alg´ebrique - (SGA 7) - vol. 1. Lecture notes in mathematics 288. Berlin; New York: Springer-Verlag. viii+523 (1972). [Ho] M. Hovey, Model Categories, Mathematical Surveys and Monographs, Volume 63, AMS (1999). [Il1] L. Illusie, Autour du th´eor`eme de monodromie locale, in "P´eriodes p-adiques, S´eminaire de Bures, 1988", Ast´erisque 223, 1994, 9-57. [Il2] L. Illusie, Around the Thom-Sebastiani theorem, with an appendix by Weizhe Zheng. Manuscripta Math. 152 (2017), no. 1-2, 61-125. 46 [ILO] L. Illusie, Y. Laszlo, F. Orgogozo, Travaux de Gabber sur l'uniformisation locale et la cohomologie e´tale des sch´emas quasi-excellents. Ast´erisque 363-364, SMF, 2014. [Ka-Sa] K. Kato and T. Saito, On the conductor formula of Bloch, Publ. Math. IHES 100 (2005), 5-151. [Kap] M. Kapranov, On DG-modules over the de Rham complex and the vanishing cycles functor. Algebraic geometry (Chicago, IL, 1989), 57-86, Lecture Notes in Math., 1479, Springer, Berlin, 1991. [Lu-COB] J. Lurie, On the classification of topological field theories, Current developments in mathe- matics, 2008, Int. Press, Somerville, MA, 2009, pp. 129-280. MR2555928 (2010k:57064) [Lu-HA] J. Lurie, Higher Algebra, http://www.math.harvard.edu/∼lurie/papers/HA.pdf, April 2016 ver- sion. [Or] F. Orgogozo, Conjecture de Bloch et nombres de Milnor, Annales de l'Institut Fourier, 53 no. 6, 2003, 1739-1754 [Pr] A. Preygel, Thom-Sebastiani & Duality for Matrix Factorizations. Thesis. Preprint arXiv:1101.5834. [Ro] M. Robalo, K-Theory and the bridge from Motives to non-commutative Motives, Advances in Math. Volume 269, 10 January 2015, Pages 399-550. [Sai] T. Saito, Characteristic cycles and the conductor of direct image. Preprint arXiv:1704.04832. [To1] B. Toen, DG-categories and derived Morita theory, Invent. Math. 167 (2007), no. 3, 615-667. [To2] B. Toen, Derived Azumaya algebras and generators for twisted derived categories. Invent. Math. 189 (2012), no. 3, 581-652. [To3] B. Toen, Lectures on dg-categories. Topics in algebraic and topological K-theory. 243-302, Lecture Notes in Math., 2008, Springer, Berlin, 2011. [To-Va] B. Toen, M. Vaqui´e, Moduli of objects in dg-categories. Ann. Sci. ´Ecole Norm. Sup. (4) 40 (2007), no. 3, 387-444. [To-Ve-1] B. Toen, G. Vezzosi, Caract`eres de Chern, traces ´equivariantes et g´eom´etrie alg´ebrique d´eriv´ee. Selecta Math. (N.S.) 21 (2015), no. 2, 449-554. [To-Ve-2] B. Toen, G. Vezzosi, G´eom´etrie non-commutative, formule des traces et conducteur de Bloch, S´eminaires et Congr`es, 31, Soc. Math. de France (2018), Actes du 1er congr`es national de la SMF - C. Lecouvrey, ed., p. 77-107 (2018), p. 1-23 [Vo] V. Voevodsky, A1-homotopy theory, In Proceedings of the International Congress of Mathemati- cians, Vol. I (Berlin, 1998), 579-604 (electronic), 1998. Bertrand Toen, IMT, CNRS, Universit´e de Toulouse, [email protected] Gabriele Vezzosi, DIMAI, Universit`a di Firenze, [email protected] 47
0902.4668	3	0902	2012-10-23T17:52:58	Weak Landau-Ginzburg models for smooth Fano threefolds	[ "math.AG" ]	The paper is joined with arXiv:0911.5428 and improved. We prove that Landau-Ginzburg models for all 17 smooth Fano threefolds with Picard rank 1 can be represented as Laurent polynomials in 3 variables exhibiting them case by case. We check that these Landau-Ginzburg models can be compactified to open Calabi-Yau varieties. In the spirit of L. Katzarkov's program we prove that numbers of irreducible components of the central fibers of compactifications of these pencils are dimensions of intermediate Jacobians of Fano varieties plus 1. In particular these numbers do not depend on compactifications. We state most of known methods of finding Landau-Ginzburg models in terms of Laurent polynomials. We discuss Laurent polynomial representation of Landau-Ginzburg models of Fano varieties and state some problems related to it.	math.AG	math	Weak Landau -- Ginzburg models for smooth Fano threefolds VICTOR PRZYJALKOWSKI Abstract. We prove that Landau -- Ginzburg models for all 17 smooth Fano threefolds with Picard rank 1 can be represented as Laurent polynomials in 3 variables exhibiting them case by case. We check that these Landau -- Ginzburg models can be compactified to open Calabi -- Yau varieties. In the spirit of L. Katzarkov's program we prove that numbers of irreducible components of the central fibers of compactifications of these pencils are dimensions of intermediate Jacobians of Fano varieties plus 1. In particular these numbers do not depend on compactifications. We state most of known methods of finding Landau -- Ginzburg models in terms of Laurent polynomials. We discuss Laurent polynomial representation of Landau -- Ginzburg models of Fano varieties and state some problems related to it. 1. Introduction Mirror Symmetry conjectures relate symplectic properties of a variety X with algebro- geometric ones for its mirror symmetry pair -- a variety Y (or one-parametric family of Calabi -- Yau varieties Y → A1) and vice-versa, relate algebro-geometric properties of X with symplectic ones of Y . Homological Mirror Symmetry (see [Ko94]) treats mirror correspondence in terms of derived categories. It associate two categories with each variety or family. Given a symplectic form on X (in our considerations it is an anticanonical form ωX ) one can associate a so called Fukaya category F uk (X) with X whose objects are Lagrangian submanifolds with respect to the symplectic form. The relative version of this category, a Fukaya -- Seidel category F S(Y ) can be associated with Y . Algebraic side of the picture is presented by a derived category of coherent sheaves Db(X) for X and a derived category of singularities Db sing(Y ) for Y -- a direct sum of categories over all fibers whose objects are complexes of coherent sheaves modulo perfect complexes. Homological Mirror Symmetry conjecture for Fano varieties predicts that for any Fano manifold X there exists a so called Landau -- Ginzburg model Y → A1 such that their categories are cross-equivalent: F uk (X) ≃ Db sing(Y ) and Db(X) ≃ F S(Y ). Homological Mirror Symmetry conjecture is very powerful but unfortunately it is very hard to prove it for particular mirror pairs. So the natural first step is checking coincidence of some invariants of categories discussed above. A natural invariant of a category C is its Hochschild cohomology HH ∗(C). For Fukaya category Hochschild cohomology is nothing but Quantum cohomology (see Subsection 2.1). Scaling the symplectic form one can vary Fukaya categories. In this way one obtain a so called non-commutative Hodge structure. In the similar way one can associate a non-commutative Hodge structure with Db sing(Y ) (more precisely, with each singular fiber of Y → A1). For definitions and constructions of these structures see [KKP08]. Key words and phrases. Weak Landau -- Ginzburg model, Fano threefold, toric degeneration, interme- diate Jacobian. The work was partially supported by grants NSF FRG DMS-0854977, NSF DMS-0854977, NSF DMS- 0901330, grants FWF P 24572-N25 and FWF P20778, RFFI grants 11-01-00336-a, 11-01-00185-a, and 12-01-31012, grants MK−1192.2012.1, NSh−5139.2012.1, and AG Laboratory GU-HSE, RF government grant, ag. 11 11.G34.31.0023. 1 The coincidence of non-commutative Hodge structures is called Mirror Symmetry con- jecture of variations of Hodge structures. It enables one to translate the mirror corre- spondence for Fano varieties to a quantitative level. We formulate it in the following way. In a lot of cases one can assume that Y is a complex torus (see Compactification principle 32). In this case the complex-valued function can be represented by a Laurent polynomial, which, under some assumptions, is called weak Landau -- Ginzburg model for X. Thus it turns out that the problem of finding a weak Landau -- Ginzburg model for X can be reduced to finding a certain Laurent polynomial. The particular series combinato- rially constructed by this polynomial (the so called constant terms series) should be equal to the so called constant term of regularized I-series for X constructed by geometrical data (numbers of rational curves lying on X). For more detailed background see [Prz08]. There are 17 families of smooth Fano threefolds with Picard rank 1, see [Isk77]. Each of them is determined by its index and its anticanonical degree. In this paper we find weak Landau -- Ginzburg models (some of them are known but had not been written down) for all 17 families. It turns out that these models are Laurent polynomials in 3 variables that have Calabi -- Yau compactifications to families of K3 surfaces (Theorem 18). Which numerical invariants of a Fano variety can be reconstructed from its Landau -- Ginzburg model (or from a weak one) and how? C. van Enckevort and D. van Straten ([vEvS06]) suggest to extract characteristic numbers of a general anticanonical section of Fano variety by writing down a monodromy of the dual family in a specific basis. L. Katzarkov's recent idea (see say [KKP08], [ILP11], [KP12]) is to relate the Hodge type of a Fano variety to a structure of a cental fiber of dual Landau -- Ginzburg model and to a sheaf of vanishing cycles for this fiber. Theorem 22 says that numbers of irreducible components (without multiplicities) of central fibers of Calabi -- Yau compactifications of weak Landau -- Ginzburg models for Fano threefolds we found are dimensions of their intermediate Jacobians plus 1. Actually, under natural assumptions this number does not depend on particular weak Landau -- Ginzburg model, see Subsection 6.2 and [DKLP] (this statement is not clear for Fano threefolds of Picard rank greater than 1, see [DKLP] for this case). For rational Fano threefolds this theorem appears in [AAK12] via constructing Landau -- Ginzburg models for these varieties. For an explanation of phenomenon of Theorem 22 see Remark 24. This paper is the first step in studying weak Landau -- Ginzburg models for Fano three- folds. We refer to [ILP11], [DKLP], [CKP12a], [CKP12b] where certain properties of some weak Landau -- Ginzburg models and their relations to Homological Mirror Symmetry are studied. We write down a table with weak Landau -- Ginzburg models for Fano threefolds here for convenience. N in the table stays for the number of variety with respect to lexicographic order (index, degree), I stays for the index of a variety and deg stays for its anticanonical degree. Later we prove that the polynomials from the table are actually weak Landau -- Ginzburg models for corresponding Fano varieties (Theorem 18). We also observe most known methods of finding weak Landau -- Ginzburg models (Section 3) and discuss some problems related to them (Section 6). Polynomials in the table are not unique weak Landau -- Ginzburg models for Picard rank 1 Fano threefolds (cf. Subsection 6.2). More examples see, for instance, in [CKP12a], [CKP12b]. 2 N I deg h12 ♯ comp. Description Weak LG model 1 2 3 4 1 1 1 1 2 4 6 8 52 53 30 20 14 31 21 15 5 1 10 10 11 6 7 8 9 1 1 1 1 12 14 16 18 10 1 22 7 5 3 2 0 8 6 4 3 1 11 2 8 · 1 21 22 12 2 8 · 2 10 11 13 2 8 · 3 14 2 8 · 4 15 2 8 · 5 16 3 27 · 2 17 4 64 5 2 0 0 0 6 3 1 1 1 Sextic double solid X2 (dou- ble cover of P3 ramified over smooth sextic). The general element of the family is quartic X4. Smooth complete intersec- tion of quadric and cubic X6. Smooth complete intersec- tion of three quadrics X8. The general element of the family is X10, a section of G(2, 5) by 2 hyperplanes and quadric in Plucker em- bedding. Variety X12. Variety X14, a section of G(2, 6) by 5 hyperplanes in Plucker embedding. Variety X16. Variety X18. Variety X22. Double Veronese cone V1 (double cover of the cone over the Veronese surface branched in a smooth cu- bic). Quartic double solid V2 (double cover of P3 ramified over smooth quartic). Smooth cubic V3. Smooth intersection of two quadrics V4. Variety V5, a section of G(2, 5) by 3 hyperplanes in Plucker embedding. Smooth quadric Q. P3. (x+y+z+1)6 xyz (x+y+z+1)4 xyz (x+1)2(y+z+1)3 xyz (x+1)2(y+1)2(z+1)2 xyz (1+x+y+z+xy+xz+yz)2 xyz (x+z+1)(x+y+z+1)(z+1)(y+z) xyz (x+y+z+1)2 + (x+y+z+1)(y+z+1)(z+1)2 x xyz (x+y+z+1)(x+1)(y+1)(z+1) xyz (x+y+z)(x+xz+xy+xyz+z+y+yz) xyz xy z + y + 1 x + 1 z + x y + z + 1 z + x + y + 1 x + z xy + z z + 4 y + z xy (x+y+1)6 xy2z + z (x+y+1)4 xyz + z (x+y+1)3 xyz + z (x+1)2(y+1)2 xyz + z x + y + z + 1 x + 1 y + 1 z + xyz (x+1)2 xyz + y + z x + y + z + 1 xyz Table 1: Weak Landau -- Ginzburg models for Fano three- folds. 3 2. Mirror Symmetry of variations of Hodge structures We consider smooth projective varieties over C. For any such variety X we denote In this paper Calabi -- Yau varieties are varieties with trivial H2(X, Z)/tors by H2(X). canonical class. 2.1. Regularized quantum D-modules. Let X be a smooth Fano variety1. To it one can associate a set of Gromov -- Witten invariants of genus 0. These invariants are numbers counting rational curves lying on X. Consider γ1, . . . , γm ∈ H ∗(X, Z), k1, . . . , km ∈ Z≥0, m ∈ Z+, and β ∈ H2(X). The (m-pointed) genus 0 Gromov -- Witten invariant with descendants that correspond to this data (see [Ma99], VI -- 2.1) is denoted by hτk1γ1, . . . , τkmγmiβ. Given these invariants (more precisely, prime three-pointed ones, i.e. those with m = 3 and k1 = k2 = k3 = 0) one can define a (small) quantum cohomology ring. This ring is the deformation of the ordinary cohomology ring. Definition 1 (see [Ma99], Definition 0.0.2). Consider a Novikov ring Λ -- the ring of poly- nomials over C in formal variables tβ, β ∈ H2(X), with natural relations tβ1tβ2 = tβ1+β2. Fix an effective basis ∆ ⊂ H ∗(X, Z). The quantum cohomology ring is a vector space QH ∗(X) = H ∗(X, Z) ⊗ Λ with quantum multiplication -- the bilinear map given by ⋆ : QH ∗(X) × QH ∗(X) → QH ∗(X) γ1 ⋆ γ2 = Xγ∈∆, β∈H2(X) tβhγ1, γ2, γ∨iβγ for any γ1, γ2 ∈ H ∗(X), where γ∨ is the Poincar´e dual class to γ (we identify elements γ ∈ H ∗(X) and γ ⊗ 1 ∈ QH ∗(X)). Notice that QH ∗(X) is graded by deg tβ = −KX · β and the constant term of γ1 ⋆ γ2 (with respect to t) is γ1 · γ2, so QH ∗(X) indeed is a deformation of H ∗(X, C). Let H = −KX and let QH ∗ H (X) be the minimal subring of QH ∗(X) containing H. Assume it is generated over Λ by the linear space H ∗ H(X) ⊗ Λ. The variety X is called quantum minimal if dimΛ QH ∗ H(X) = dim X +1. Examples of quantum minimal varieties are complete intersections in (weighted) projective spaces or Fano threefolds with Picard rank 1. H(X), i.e. QH ∗ H = dimC H ∗ H = H ∗ Next we describe the construction of regularized quantum D-module (or, equivalently, Dubrovin's second structural connection). More precisely, this D-module contains an essential submodule corresponding to H ∗ H(X). As we need only this essential part, we give the definition of this submodule; one should replace H ∗ H(X) by H ∗(X) in the definition to get the definition of the whole module. For more particular definition of this D-module for quantum minimal case see [GS07], [Prz07], and [Prz08]. Consider a torus T = Spec B, where B = C[t, t−1]. Let HQ be the trivial vector bundle H(X). Let S = H 0(HQ) and let ⋆ : S × S → S be the quantum H(X) ⊗ C[t, t−1]). Let ∂t . Consider a (flat) connection ∇ on HQ defined on the sections over T with fiber H ∗ multiplication (we can consider it as an operation on S ∼= QH ∗ D = B[ ∂ γ ∈ H ∗ ∂t ] and D = t ∂ H(X) as (cid:18)∇(γ), t ∂ ∂t(cid:19) = KV ⋆ γ 1This assumption can be weakened; we are interested in the case of smooth Fano varieties, so we give definitions in this particular case. 4 (the pairing is the natural pairing between differential forms and vector fields). This connection provides the structure of a D-module for S by D(γ) = (∇(γ), D). Let Q be this D-module. It is not regular in general, so we need to "regularize" it to obtain the regular one. Let E = D/D(t ∂ ∂t − t) be the exponential D-module. Define the regularization of Q as Qreg = µ∗(Q ⊠ E), where µ : T × T → T is the multiplication In other words, Qreg is a convolution with the and ⊠ is the external tensor product. anticanonical exponential D-module. It can be represented by a differential operator, which is divisible by D on the left: Qreg ∼= D/D(DLX). The differential operator LX is called the regularized quantum differential operator. If X is quantum minimal, then LX is said to be of type DN, see [Go05], 2.10, and these are studied in [GS07]. For N = 3 this operator is given explicitly in [Go05], Example 2.11, in terms of structural constants of quantum multiplication by the anticanonical class (two-pointed Gromov -- Witten invariants). Thus, there is an operator of type D3 associated to every smooth Fano threefold with Picard group Z. For all smooth Fano threefolds with Picard rank 1 the operators of type D3 are known (see, for instance, [Go05], 5.8). Let H 0 be the class in H 0(X, Z) dual to the fundamental class of a quantum minimal variety X. Consider a series I X H 0 = 1 +Xβ hτ−KX ·β−2H 0iβ · t−KX ·β, where the sum is taken over all β ∈ H2(X) such that −KX · β ≥ 2. For quantum minimal variety X this series is a unique analytic solution of the equation LX I = 0 of type H 0 = 1 + a1t + a2t2 + . . . ∈ C[[t]], I X ai ∈ C (see [Prz07], Corollary 2.2.6, and references therein). Definition 2. This series is called the fundamental term of the regularized I-series of X. 2.2. Weak TLG = Gn resented by Laurent polynomial: f = f (x1, x−1 1 term (i. e. the coefficient at x0 Landau -- Ginzburg i=1 Spec C[x±1 m = Qn 1 · . . . · x0 i models. Consider torus ] and a function f on it. This function can be rep- n ). Let φf (i) be the constant . . . , xn, x−1 a n) of f i, and put Φf = ∞Xi=0 φf (i) · ti ∈ C[[t]]. Definition 3. The series Φf is called the constant terms series of f . Definition 4. Let X be a smooth n-dimensional Fano variety and let I X H 0 ∈ C[[t]] be its fundamental term of regularized I-series. The Laurent polynomial f ∈ C[Zn] is called a very weak Landau -- Ginzburg model for X if (up to a shift f 7→ f + α, α ∈ C) Φf = I X H 0. A very weak Landau -- Ginzburg model f ∈ C[Zn] is called a weak one if there is a fiberwise compactification of a family f : (C∗)n → C whose total space is (an open) smooth Calabi -- Yau variety. Such compactification is called a Calabi -- Yau compactification. Remark 5. If the total space of a family and its base are smooth then the general fiber is smooth. Remark 6. There is a slightly different definition of weak Landau -- Ginzburg model in the literature (see say [Prz08]). By this definition a weak Landau -- Ginzburg model is a very weak one whose general fiber is birational to Calabi -- Yau variety while our definition says that (strengthened) this property holds for all fibers (even reducible ones). However in 5 practice these conditions are equivalent: we do not know natural examples when these definitions differ. The meaning of the definition is the following (see [BS85], 10 or [Be83], pp. 50 -- 52). Consider a pencil TLG → B = P[u : v] \ (0 : 1) with fibers Yα = {1 − αf = 0}, α ∈ C \ {0} ∪ {∞}. The following proposition is a mathematical folklore (see [Prz08] for the proof). Proposition 7. Assume the Newton polytope of f ∈ C[Zn] contains 0 in its interior. Let t ∈ B be the local coordinate around (0 : 1). Then there is a fiberwise (n − 1)-form ωt ∈ Ωn−1 TLG/B and a (locally defined ) fiberwise (n − 1)-cycle ∆t such that Φf =Z∆t ωt. This means that Φf is a solution of the Picard -- Fuchs equation for the pencil {Yt}. Remark 8. Let P Ff = P Ff (t, ∂ ∂t) be a Picard -- Fuchs operator of {Yt}. Denote the order of P Ff by m and denote the degree with respect to t by r. Let Y be a semistable compactification of {Yt} (so we have the map ef : Y → P1; denote it for simplicity by f ). Denote the dimension of the transcendental part of Rn−1f! ZY by mf (for an algorithm for computing it see [DH86]), and denote the number of singularities of f counted with multiplicities by rf . Then m ≤ mf and r ≤ rf . So we can write a differential operator of bounded order by t and D as an operator with indeterminant coefficients. As Φf annihilates it, we get a system of infinite number of linear equations. To check that LX = P Ff we need to solve this system (it has a unique solution, up to scaling, so we need to solve a finite system of linear equations). However in practice it is enough to compare the first few coefficients of the expansion of Φf and I X H 0 determines LX . So if these terms coincide with the first few terms of Φf then the differential operator vanishing Φf is LX (up to high order coefficients) which means that P Ff = LX . H 0. Indeed, we know that the first few terms of I X Question 9. Let us be given a polytope ∆. Can one find (effectively) a number s = s(∆) such that for any polynomial f whose Newton polytope is ∆, the first s coefficients of Φf determine the other ones (so in order to prove that I X H 0 it is enough to check coincidence of the first s coefficients of both series). In other words, is it true that the linear system of equations on coefficients of a Picard -- Fuchs operator for Laurent polynomial with given Newton polytope for the first s terms is nondegenerate? H 0 = Φf for any I X 3. Methods for finding weak Landau -- Ginzburg models We observe here some methods for finding very weak Landau -- Ginzburg models for some Fano varieties, in particular for complete intersections in projective spaces and Grassmannians, and for varieties admitting small toric degenerations. They are weak ones in practice and usually it is not complicated to prove this. However we do not know a general method of proving this in some cases. In the next section we find weak Landau -- Ginzburg models for Fano threefolds with Picard number 1. 3.1. Small toric degenerations. We start from description of mirrors for Fano vari- eties admitting the so called small toric degenerations. This description was suggested in [BCFKS97] and [Ba97]. Assume that a smooth Fano variety X admits a degeneration to a terminal Gorenstein i ) be the set of integral generators toric variety Y . Let {v1, . . . , vn} ⊂ Zk, vi = (v1 i , . . . , vk 6 fX = Qr where k0 = N −P ki. i=1(xi,1 + . . . + xi,ki−1 + 1)ki Q xi,j ·Q yi + y1 + . . . + yk0 ∈ C[{x±1 jl , y±1 s }], of rays of the fan of Y . Denote xvi = xv1 Landau -- Ginzburg model for X (up to a shift f → f + α, α ∈ C) is 1 · . . . · xvk k ∈ C[x±1 1 , . . . , x±1 i i k ]. Then the weak nXi=1 xvi. Problem 10. Prove this. In a case of high index one can use Quantum Lefschetz -- type arguments (applied to hyperplane section) and Givental's formula for I-series of smooth toric varieties (cf. [Ga07], Proposition 1.7.15) for proving that these polynomials are very weak Landau -- Ginzburg models. For proving that corresponding families have Calabi -- Yau compactifications one should check that singularities of elements of families either admit a crepant resolution or "come from the ambient toric variety". This is enough for the proof, because these elements are anticanonical sections of ambient toric variety. For references see [Ba93], in particular see Theorem 4.1.9. 3.2. Complete intersections. The suggestions for Landau -- Ginzburg models for com- plete intersections in projective spaces were given by Hori and Vafa in [HV00]. In terms of Laurent polynomials their suggestions can be stated in the following way. The weak Landau -- Ginzburg model for smooth complete intersection X of r hypersurfaces of degrees k1, . . . , kr in PN is (up to the shift f → f + α, α ∈ C) Proposition 11. The polynomial fX is a weak Landau -- Ginzburg model for X. Proof. According to the well-known Givental's formula for constant term of I-series of X (up to the shift), I X H 0 = Qr j=0(kji)! (i!)N +1 tk0i. ∞Xi=0 Qr j=0(kj i)! (i!)N +1 X is One can check that the constant term of f n if n = k0i and 0 in the other case. Consider the compactification of the pencil corresponding to fX given by the natural embedding (C∗)N −r ֒→ P(y0 : . . . : yk0) × P(x1,1 : . . . : x1,k1) × . . . × P(xr,1 : . . . : xr,kr). This compactification is (singular) relative hypersurface of multidegree (k0 + 1, k1, . . . , kr) in Pk0 × Pk1−1 × . . . Pkr−1 and hence has trivial canonical class. It is easy to check that singularities of the compactified pencil are "purely canonical", which means that it admits a crepant resolution. Indeed, they are products of du Val singularities and linear spaces of codimension 2 or products of ordinary double points and linear spaces of codimen- sion 3 (away from intersections of components of singularities). If we blow up any cDV singularity of codimension 2 we get singularities of the same type again (singularities "in- tersect transversally"). Blowing up singularities one by one and taking small resolutions we get a crepant resolution. Hence the total space of the resolution of the compactifica- tion is a Calabi -- Yau variety, so the Laurent polynomial we consider is actually a weak Landau -- Ginzburg model. (cid:3) Remark 12. The same can be done for smooth complete intersections of Cartier divisors in weighted projective spaces. That is one can define Hori -- Vafa-type Laurent polynomials for them similar to ones for usual intersections complete defined above and prove that they are very weak Landau -- Ginzburg models ([Prz05] and [Prz10], Theorem 9). For Fano index one case they are known to be weak ones (see Remark 19). In general case this is not proven yet. 7 3.3. Grassmannians. Weak Landau -- Ginzburg models for Grassmannians were sug- gested in [EHX97], B 25. Later in [BCFKS97], using the construction of small toric degenerations of Grassmannians (see references therein), these Landau -- Ginzburg models were obtained via small toric degenerations technics. The suggested model for G(k, N) is X −1 ab (Xa+1,b + Xa,b+1) + X −1 N −k,k ∈ C[{Xab, X −1 ab }] X1,1 + X1≤a≤N −k, 1≤b≤k (the variables are Xa,b, 1 ≤ a ≤ N −k, 1 ≤ b ≤ k; for a > N −k or b > k we put Xab = 0). Problem 13. Prove that the polynomial written down above is actually weak (or at least very weak) Landau -- Ginzburg model for G(k, N). The problem is combinatorial: to solve it one should find the coefficients of constant terms series for this polynomial and compare this series with the constant term of regularized I-series for G(k, N) found in [BCK03] (see also [BCFKS98], Conjecture 5.2.3). For methods for proving the Calabi -- Yau condition see Problem 10. Remark 14. In the similar ways one can write down very weak Landau -- Ginzburg models for complete flag manifolds ([Gi96]) and partial flag manifolds ([BCFKS98]). For general flag varieties G/P see [Re07]. For minuscule varieties see [BG]. 3.4. Complete intersections in Grassmannians. In this subsection we describe the suggestions for weak Landau -- Ginzburg models of complete intersections in Grassman- nians. The idea for writing them down is the particular case of the method for com- plete intersections in varieties admitting small toric degenerations. More precisely, let Y = X ∩ Y1 ∩ . . . ∩ Ys be a complete intersection in a variety admitting a degeneration to terminal Gorenstein toric variety X. Let {v1, . . . , vn} ⊂ Zk be the set of integral generators of rays of the fan of X as before. Let D(vi)'s be divisors corresponding to vi's. Let {pl j) as cohomological classes. Then the Landau -- Ginzburg model for Y is (conjecturally) ri}, l = 1, . . . , s, be subsets of {v1, . . . , vn} such that Yl =Pj D(pl 1, . . . , pl with potential Pn 1 + . . . + xpl {xpl i=1 xvi (see [Ba97]). ri = 1} ∈ C[x±1 1 , . . . , x±1 k ], l = 1, . . . , s, We describe this procedure for complete intersections in Grassmannian G(m, r) fol- lowing [BCFKS97] (and changing the notation for simplicity)2. Consider the following Laurent polynomials in variables Xab, 1 ≤ a ≤ r − m, 1 ≤ b ≤ m. X11, X1i X1,i−1 + X2i X2,i−1 + . . . + Xr−m,i Xr−m,i−1 , i = 2, . . . , m, Xj+1,1 Xj1 + Xj+1,2 Xj2 + . . . + , j = 1, . . . , r − m − 1, Xj+1,m Xjm 1 . Xr−m,m Given any of these polynomials consider the set of rays of the fan of X associated with summands of the polynomial. The sum of the boundary divisors associated with these rays is equivalent (in cohomology) to a Picard group generator for X. The cohomological class of a hypersurface in Grassmannian is given by its degree. This means that one can find s sums of boundary divisors such that each divisor is contained in at most one sum and ith sum is equivalent to Yi. This gives us a Landau -- Ginzburg model for Y . 2It was suggested for Calabi -- Yau complete intersections but it works for Fano varieties in absolutely the same way (as usual, modulo shift f → f + α, α ∈ C). 8 Problem 15. In all cases we consider there are (birational) changes of variables for the Landau -- Ginzburg models for Fano complete intersections in G(m, r) obtained in the way described above, such that after these changes Landau -- Ginzburg models are functions on a complex torus (that is, Laurent polynomials). More precisely, one can express one variable in terms of others from each equation, put them in the Laurent polynomial of Grassmannian and make a linear change of variables to make denominators monomial. Prove this in the general case. It does not follow from this procedure that the Laurent polynomial obtained in this way is a very weak Landau -- Ginzburg model for a complete intersection in G(m, r). In practice it is even a weak one; one can check this in each particular case. Problem 16. Prove this in the general case (cf. Problems 10 and 13). Example 17. Consider a Fano threefold X14. By definition it is the section of G(2, 6) cut out by five hyperplanes. The Landau -- Ginzburg model is the variety (cid:26)X11 = 1, X21 X11 + X22 X12 = 1, X31 X21 + X32 X22 = 1, X41 X31 + X42 X32 = 1, 1 X42 = 1(cid:27) ⊂ C[{Xij, X −1 ij }], 1 ≤ i ≤ 4, 1 ≤ j ≤ 2, with potential X11 + X21 + X12 X11 + X22 X12 + X31 + X22 X21 + X32 X22 + X41 + X32 X31 + X42 X32 + X42 X41 + 1 X42 . Denote X12 = a, X22 = b, X32 = c. Then X21 = a − b a , X31 = (a − b)(b − c) ab , X41 = (c − 1)(a − b)(b − c) abc . So the potential is 5 + a + ab a − b + abc (a − b)(b − c) + abc (a − b)(b − c)(c − 1) = 5 + a2 a − b + abc2 (a − b)(b − c)(c − 1) . Denote x = a − b, y = b − c, z = c − 1. Then a = x + y + z + 1, b = y + z + 1, c = z + 1 and we get the potential f = 5 + (x + y + z + 1)2 x + (x + y + z + 1)(y + z + 1)(z + 1)2 xyz ∈ C[x, x−1, y, y−1, z, z−1]. The constant term of the regularized I-series for X14 (shifted by 4) is I X14 H 0 = 1 + 4t + 48t2 + 760t3 + 13840t4 + 273504t5 + 5703096t6 + . . . (see [Prz04]). It is easy to see that the constant terms series for f14 = f − 5 equals I X14 H 0 up to more then 16 coefficients. This means that they are equal (see Remark 8). The fiberwise compactification of the pencil f = λ, λ ∈ C given by the natural map Spec C[x, x−1, y, y−1, z, z−1] → P(x : y : z : w) gives a family of quartic surfaces. Singu- larities of this family are ordinary double points or du Val along lines (see the proof of Proposition 11). So the family admit a Calabi -- Yau compactification. Therefore the Laurent polynomial we obtain is a weak Landau -- Ginzburg model for X14. 9 4. Fano threefolds of Picard rank 1 In this section we study Laurent polynomials from Table 1 and prove that they are weak Landau -- Ginzburg models for corresponding Fano varieties. First we describe how these weak Landau -- Ginzburg models (we call them standard 3) are obtained. Varieties 1, 11, 12 are hypersurfaces in weighted projective spaces of degree 6 in P(1 : 1 : 1 : 1 : 3), of degree 6 in P(1 : 1 : 1 : 2 : 3), and of degree 4 in P(1 : 1 : 1 : 1 : 2) respectively. Their weak Landau -- Ginzburg models can be found by Hori -- Vafa procedure similar to procedure for complete intersections in projective spaces described in Subsec- tion 3.2 (see Remarks 19 and 12). Varieties 2, 3, 4, 13, 14, 16, 17 are complete intersections, so their weak Landau -- Ginzburg models can be found using Proposition 11. Varieties 5, 7, 15 are complete intersections in Grassmannians, so the corresponding polynomials can be obtained using the method described in Problem 15. The polynomial for X14 is studied in Example 17. There is another way to obtain the same polynomials for V5 and X10. Indeed, V5 has a small toric degeneration (this is proved by S. Galkin in his Thesis [Ga07]), so its weak Landau -- Ginzburg model is given by the corresponding polytope (see 3.1). According to V. Golyshev (see [Go05]), the Landau -- Ginzburg model for X10 is a quotient of the model for V5 (see [ILP11] and [DKLP] for the proof). Taking invariants of the quotient and changing coordinates one can get a weak Landau -- Ginzburg model for X10; the form we write down is convenient for calculations. The polynomial for 6-th variety X12 is found in [BP84]. We change coordinates a bit to get the convenient form as written. Finally, polynomials for varieties 8, 9, 10 are found in [Prz08]. There is a misprint in the polynomial for X16 in the journal version of [Prz08]; it is corrected in the preprint on arXiv. It is remarkable that some of these polynomials were found under the assumption that there are Gorenstein toric degenerations of corresponding varieties. Later S. Galkin in his Thesis ([Ga07]) proved that there is a terminal Gorenstein toric degeneration of X22, so the corresponding polynomial can be obtained using a method from Subsection 3.1. Theorem 18. Standard polynomials are weak Landau -- Ginzburg models for Fano three- folds with Picard rank 1. Proof. Direct computations show that these polynomials are very weak Landau -- Ginzburg models (see Remark 8). Straightforward compactifications of pencils corre- sponding to all polynomials except for polynomials for V1 and X2 give relative quartics in P3 or (for complete intersections) relative anticanonical sections in products of projective spaces. Singularities of obtained total spaces are du Val singularities along lines and or- dinary double points on each step of their minimal resolutions (cf. Proposition 11). This gives Calabi -- Yau compactifications of families. Consider a pencil (x + y + 1)6 xy2z + z = λ, λ ∈ C, associated with V1. Compactify this family to a family of surfaces in A3: (x + y + 1)6 = (λ − z)xy2z. Then changing the variables a = x + y + 1, we get a6 = (λ − z)(a − y − 1)y2z. 3This term is local for the paper. Weak Landau -- Ginzburg models from Table 1, from our point of view, are not better then others. 10 Consider this family as a family of surfaces lying in (C∗)3 with coordinates a, y, z (in other words let us divide by a, y, z). Changing the variables b = y/a, c = yz/a, we get a family a4 = (λb − c)(a − ab − 1)c. Compactify it to a family in P3. We get a family of quartics with du Val singularities along lines (see the proof of Proposition 11). After a resolution of these singularities one get a family whose total space is Calabi -- Yau and the initial family of hypersurfaces in torus is embedded to this resolution. Finally consider a pencil for X2. We have a family (x + y + z + 1)6 = λxyz. In birational coordinates a = x + y + z + 1, b = x/(x + y + z + 1), c = y/(x + y + z + 1) we get a4 = λbc(a − ab − ac − 1). As before one can check that the initial family can be compactified to a Calabi -- Yau threefold. (cid:3) Remark 19. Let us prove that the general element of a standard weak Landau -- Ginzburg model for X2 can be compactified to a K3 surface in another, more conceptual way (suggested by V. Golyshev). Remember that a Hori -- Vafa mirror for a hypersurface of degree d in P(w0 : . . . : wn) is (cid:26) yw0 · . . . · ywn n = 1 0 y0 + . . . + yk = 1, where w0 + . . . + wk = d, with the potential f = y0 + . . . + yn. For X2, the hypersurface of degree 6 in P(1 : 1 : 1 : 1 : 3), we have (up to a shift f → f −1) the variety (cid:26) y0y1y2y3y3 4 = 1 y1 + y2 + y3 + y4 = 1 with the potential Taking change of variables f = y0. y1 = x x + y + z + t , y2 = y x + y + z + t , y3 = z x + y + z + t , y4 = t x + y + z + t (where x, y, z, t are projective coordinates) we get the Landau -- Ginzburg model So, in local chart, say t 6= 0, we finally get the weak Landau -- Ginzburg model y0xyzt3 = (x + y + z + t)6, f = y0. f ′ 2 = (x + y + z + 1)6 xyz . The general element of the pencil corresponding to f ′ of the initial Hori -- Vafa model. Inverse the potential: u = 1/f . Then we get the pencil 2 is birational to the general element y1y2y3y3 4 = u, y1 + y2 + y3 + y4 = 1. This model is exactly the Landau -- Ginzburg model for P(1 : 1 : 1 : 3) (see [CG06], (2)). So, by Theorem 1.15 in [CG06], a general element of the pencil we are interested in is birational to a K3 surface. 11 5. Reconstructing Hodge numbers The following fact can be proved immediately by constructing Hironaka's house. Fact 20. Let X and Y are two birational smooth Calabi -- Yau varieties. Then they are birational in codimension 1. By definition of a flop they differ by flops. Corollary 21. Any two Calabi -- Yau compactifications of weak Landau -- Ginzburg model differ by flops. Theorem 22. Let X be a smooth Fano threefold with Picard rank 1. Let f : (C∗)3 → C be its standard weak Landau -- Ginzburg model. Let kX be the number of irreducible components of the central fiber of a Calabi -- Yau compactification of f . Then kX = h12(X) + 1. Proof. The proof is given by direct calculations of Calabi -- Yau compactifications considered in Theorem 18 in all 17 cases, one by one (see Examples 29 and 30 and [IKP11], [KP12]). By Corollary 21 numbers of components of central fibers of all Calabi -- Yau compactifications of f are the same. (cid:3) Remark 23. As X is a Fano threefold and Pic (X) = Z, the Hodge diamond of X is the following. 0 0 0 0 h12(X) 0 1 1 1 1 0 h12(X) 0 0 0 0 Thus Theorem 22 enables one to reconstruct all Hodge numbers of Picard rank 1 Fano threefolds. Remark 24. According to [GKR12], the phenomenon appeared in Theorem 22 can be explained (at least for complete intersections) as follows. Consider a perverse sheaf F of vanishing cycles to the central fiber of f . If Homological Mirror Symmetry holds for its fiberwise compactification, then kX can be computed via spectral sequence from [GKR12] for H 1(F ) and H 3(F ). Hence by [GKR12] H 1(F ) and H 3(F ) are isomorphic as Hodge structures to H 12(X) and H 21(X). From this point of view the statement of Theorem 22 holds for any Landau -- Ginzburg model (cf. Corollary 44). Thus Theorem 22 confirms Homological Mirror Symmetry conjecture for Picard rank 1 Fano threefolds and their (compactified) standard Landau -- Ginzburg models. One can see on examples that this phenomenon holds in higher Picard rank case. Question 25. Direct computations, even in Picard rank 1 case as in Theorem 22, can be very complicated (usually for big Hodge numbers). So doing them for all Picard ranks is a huge technical problem. Is it possible to do it (or at least prove Theorem 22) in more conceptual way? Remark 26. An analog of Theorem 22 in higher dimensions, due to [GKR12], is the following. Let us have an n-dimensional Fano variety X. Let k be a number of components of a central fiber of its (toric) Landau -- Ginzburg model of the same dimension. Then Homological Mirror Symmetry involves that k = h1,n−1(X) + 1. Problem 27. Prove this for toric Landau -- Ginzburg models of Fano varieties of dimension greater then 3. 12 Example 28. Consider a general cubic fourfold. Its Hodge structure is a sum of a part generated by a hyperplane section and a primitive Hodge structure in dimension 4 -- weight 2 Hodge structure of dimensions (1, 20, 1). In particular h13 = 1 and h22 = 21. The central fiber of Calabi -- Yau compactification of Hori -- Vafa type toric Landau -- Ginzburg model has 2 components intersecting by particular Picard rank 20 K3 surface (see [KP09]). Example 29. Consider the variety X16. Its standard weak Landau -- Ginzburg model is (x + 1)(y + 1)(z + 1)(x + y + 1) xyz . The compactification in a projective space is a family of quartics {(x + t)(y + t)(z + t)(x + y + z + t) = λxyzt} ⊂ A[λ] × P[x : y : z : t]. There are 4 components of the central fiber λ = 0. Singularities are the disjoint union of 9 "horizontal" lines. All of them are products of du Val singularities of type A1 and affine lines in the neighborhood of a general point. After blowing them up we get three ordinary double points in the central fiber. So finally we get no new components of the central fiber and kX16 = 4 = h12(X16) + 1. Example 30. Consider the variety X18. Its standard weak Landau -- Ginzburg model is (x + y + z)(x + xz + xy + xyz + z + y + yz) xyz . The compactification in a projective space is a family of quartics {(x + y + z)(xt2 + xzt + xyt + xyz + zt3 + yt2 + yzt) = λxyzt} ⊂ A[λ] × P[x : y : z : t]. There are 2 components of the central fiber λ = 0. Singularities are 3 "horizontal" lines globally of type A1 along lines, 3 "horizontal" lines globally of type A2 along lines, and one "horizontal" line ℓ which is, away from the central fiber, of type A1 along a line. The intersection of two components of the central fiber is a plane cubic with one node; this node lies on ℓ. Blowing ℓ up we get one more "vertical" line of singularities globally of type A1 along a line. Two components of the fiber over 0 intersect now at the union of two lines (one of them is a singularity of our threefold); these lines intersect at two points. Blowing the remaining singularity up we get three surfaces over 0. Each two of them intersect by a rational curve, and three such lines intersect at two points. So finally we get kX18 = 3 = h12(X18) + 1. Remark 31. More complicated cases of compactifications are described in detail in [IKP11], [KP12]. 6. Properties, examples, problems In the previous sections we considered weak Landau -- Ginzburg models for Picard rank 1 Fano threefolds. However for given variety such weak Landau -- Ginzburg model is not unique. In this section we discuss how to choose "correct" ones and what depends on a particular choice. We consider two approaches: "global to local" one claiming which weak Landau -- Ginzburg models are correct and "local to global" one saying that for Picard rank one Fano threefolds nothing depend on a particular choice. The second approach is studied and discussed in [DKLP]. 13 6.1. Global to local. Let us come back to the initial definition of dual Landau -- Ginzburg model for a Fano variety X of dimension n. Remind that a Laurent polynomial f in n variables is called a very weak Landau -- Ginzburg model for X if I X In other words, period condition for f is satisfied. However this condition is not enough to "feel geometry" of X. To get stronger definition one can use the following principle. H 0 = Φf . Principle 32 (Compactification principle). There exists a fiberwise compactification of the family of fibers of "good" very weak Landau -- Ginzburg model (defined up to flops) satisfying (B-side of ) Homological Mirror Symmetry conjecture. In particular this means that it should exist a fiberwise compactification to an (open) smooth Calabi -- Yau variety -- family of compact Calabi -- Yau varieties. This condition is strong enough: say if f (x1, . . . , xn) is a very weak Landau -- Ginzburg model for X then for big enough k Laurent polynomial f (xk 1, . . . , xn) is a very weak Landau -- Ginzburg model for X but compactification principle fails for it. Thus a necessary condition for very weak Landau -- Ginzburg model to satisfy the compactification principle is a Calabi -- Yau condition saying that there should exist a fiberwise compactification whose total space is a smooth Calabi -- Yau variety. However these two conditions are not enough. Example 33. A 4-dimensional cubic (see [KP09]) has the following two weak Landau -- Ginzburg models: (x + y + 1)3 xyzw + z + w and (cid:18)x1 + x2 + 1 x1x2(cid:19)(cid:18)y1 + y2 + 1 y1y2(cid:19) . The Calabi -- Yau compactification of the second one has complicated central fiber with many components and wrong invariants. Example 34. Let X be a complete intersection of two quadrics in P5. Consider the following weak Landau -- Ginzburg model for X: 1 1 (cid:18)x + x(cid:19)(cid:18)y + y(cid:19)(cid:18)z + 1 z(cid:19) . The number of components over zero of its compactification in P3 × A1 is 30 while h12(X) = 2. Remark 35. Coordinates of singular fibers of weak Landau -- Ginzburg model are deter- mined by its Picard -- Fuchs equation. Example 34 shows that even for Picard rank 1 Fano threefolds the number of components of fiber over zero is not determined by an equation. The last condition we want to put on Laurent polynomials is related to toric degener- ations (cf. Subsection 3.1). Consider a Fano variety X and its weak Landau -- Ginzburg model f . We say that f satisfies toric condition if there is an embedded degeneration X T to a toric variety T whose fan polytope (the convex hull of integral genera- tors of rays of T 's fan) coincides with the Newton polytope (the convex hull of non-zero coefficients) of f . In this case f is called a toric Landau -- Ginzburg model. One can easily check that weak Landau -- Ginzburg models from Examples 33 and 34 are not toric: degrees of Fano varieties differ from degrees of toric varieties associated with Laurent polynomials. From this point of view we state the following conjecture. Conjecture 36 (Strong version of Mirror Symmetry of variations of Hodge structures conjecture). Any smooth Fano variety has a toric Landau -- Ginzburg model. 14 Corollary 37. Any smooth Fano variety has a toric degeneration. By Theorem 18, Proposition 11, and [ILP11] this conjecture holds for Picard rank 1 Fano threefolds and complete intersections. So we hope that the following picture holds. Optimistic picture 38. Toric degenerations of smooth Fano varieties are in 1-to-1 cor- respondence with toric Landau -- Ginzburg models. Compactification principle holds for them. Question 39. Is the opposite to the second part of optimistic picture holds? That is it true that all Landau -- Ginzburg models (from Homological Mirror Symmetry point of view) of the same dimension as an initial Fano variety are compactifications of toric ones? In particular, is it true that all of them are rational? These questions are treated in [DKLP]. Question 40. Should we put some conditions on toric varieties in the optimistic picture? Remark 41. Toric varieties can have several smoothings to different smooth Fano varieties, see [CCG+], [CKP12a], and [CKP12b]. 6.2. Local to global. Another way of determining which weak Landau -- Ginzburg models are "correct" is known only for threefolds. We sketch it here for Picard rank 1 Fano varieties. For more details see [DKLP]. As we mentioned, Homological Mirror Symmetry conjecture expects that fibers of Landau -- Ginzburg model for Fano variety are Calabi -- Yau varieties. More precisely, they are expected to be mirror dual to anticanonical sections of a Fano variety. For the threefold case this duality is essentially the classical Dolgachev -- Nikulin duality of K3 surfaces. Let H be a hyperbolic lattice, Z ⊕ Z with intersection form (cid:18) 0 1 1 0 (cid:19) . intersection lattice on the The is K = H ⊕ H ⊕ H ⊕ E8(−1) ⊕ E8(−1). Consider a family UL of K3 surfaces whose lattice of algebraic cycles contains L ⊂ K (and coincides with L for general K3 surface in the family). Consider a lattice M ′ = L⊥, the orthogonal to L in K. Let M ′ = H ⊕ M. second cohomology on any K3 surface Definition 42. The family of K3 surfaces UM is called the Dolgachev -- Nikulin dual family to UL. Consider a principally polarized family of anticanonical sections of a Fano threefold X of index i and degree (−KX )3 = i3k. It is nothing but ULn with Ln = h2ni, 2n = ik. The lattice Ln is a sublattice of H. Using this embedding to one of the H-summands of K we can see that its Dolgachev -- Nikulin dual lattice is Mn = H ⊕ E8(−1) ⊕ E8(−1) + h−2ni. The surfaces with Picard lattices Mn are Shioda -- Inose surfaces. They are resolutions of quotients of specific K3 surfaces S by Nikulin involution, the one keeping transcendental lattice TS; it interchanges two copies of E8(−1). Another description of Shioda -- Inose surfaces is Kummer ones going back to products of elliptic curves with n-isogenic ones. Mn-polarized Shioda -- Inose surfaces form an 1-dimensional irreducible family. It turns out that fibers of standard toric Landau -- Ginzburg models can be compactified to Shioda -- Inose surfaces dual to anticanonical sections of Fano threefolds: Theorem 43 ([DKLP]). Let X be a Picard rank 1 Fano threefold of index i and (−KX )3 = i3k. Then fibers of Calabi -- Yau compactifications of standard toric Landau -- Ginzburg model for X are Shioda -- Inose surfaces with Picard lattice Mik/2. 15 We call (toric) weak Landau -- Ginzburg models with Shioda -- Inose condition good. Thus compactifications of good Landau -- Ginzburg models are, modulo coverings and the standard action of P SL(2, C) on the base, the unique families of corresponding Shioda -- Inose surfaces. More precisely, they are index-to-one coverings of the moduli spaces. Corollary 44 (cf. Corollary 21). A compactification (to a smooth Calabi -- Yau variety) of good weak Landau -- Ginzburg model is unique up to flops. This means that if Homological Mirror Symmetry for Picard rank 1 Fano threefolds holds then Landau -- Ginzburg models for it are, up to flops, compactifications of the stan- dard ones. Moreover, all other good Landau -- Ginzburg models are birational over the base A1 to the standard Landau -- Ginzburg models. 6.3. Toric degenerations and Laurent polynomials. There are several ways how to find (toric) weak Landau -- Ginzburg models for given Fano variety X. Some of them are described in Section 3. However in some cases these methods do not work, say, if an easy geometric description for X is not known. In this case one can hope to "guess" weak Landau -- Ginzburg model via toric degenerations in the following way. First, one should guess a toric variety with the same numerical invariants as X -- degree, Picard rank, etc. Then (this is not necessary) one should prove that this toric variety is a degeneration of X. Thus, by optimistic picture, there should be a toric Landau -- Ginzburg model whose Newton polytope is a fan polytope of the toric variety we find. The problem is to reconstruct a Laurent polynomial with prescribed constant terms series from its Newton polytope ("put coefficients on integral points of the polytope"). This problem is not solved yet. However there are several approaches for doing it, mostly in canonical and up to threefold case. Binomial principle says that on vertices of polytope we should put 1's and on the i'th (from any end) integral point of an edge of integral length n is (cid:0)n for a lot of cases (in other words, for toric varieties with cDV singularities that is, ones whose integral points of a fan polytope (except for the origin) lie on edges). i(cid:1). This principle works Conjecture 45 (Prokhorov, [Pr05]). Fano threefolds with cDV singularities are smooth- able. However this is not always the case, and we need to consider worse singularities. For them (for canonical4 threefold case) Minkowski principle, suggested in [CCG+], can be applied. This principle is a generalization of binomial principle in the following way. Remind that a segment of integral length n is a Minkowski sum of n indecomposable integral poly- topes -- segments of length 1. A unique polynomial in one variable x, having coefficient 1 at vertices of this segment is x + 1. Binomial principle says that coefficients at length n segment are given by a product of n polynomials x + 1 corresponding to each Minkowski summand. This description can be generalized to two-dimensional case, that is, to facets of 3-dimensional polytope. Consider each facet and its Minkowski decomposition to in- decomposable summands. Assume that all of these summands are of type Ak, k ≥ 0 -- Newton polytopes of polynomials (x + 1)k + y; if this is not the case the principle can't be applied. Then coefficients for the initial facet are given by product of the polynomials corresponding to summands. The idea of the Minkowski decomposition of facets is that such decompositions vanish some obstructions for deformations of toric varieties (see [Alt94]). This means, in particu- lar, that different Minkowski decompositions of facets give different Laurent polynomials corresponding to different smooth Fano varieties having degenerations to the same toric 4Canonicity means that the origin is a unique integral point in the interior of the fan polytope. 16 variety (see Remark 41). The other corollary is that toric varieties for which Minkowski principle can be applied tend to be smoothable in the anticanonical embedding. From the other hand non-Minkowski Laurent polynomials tend to "correspond" (after defining mirror correspondence for them) to toric degenerations in non-anticanonical embedding or even non-smoothable varieties. Example 46. Toric Landau -- Ginzburg model (from a point of view of [CG06] or [AKO04]) for P(1 : a1 : . . . : an) is x1 + . . . + xn + . 1 x a1 1 ·...·xan n Example 47 ([CKP12a], [CKP12b]). Consider P(1 : 1 : 2 : 4). Non-Minkowski toric Landau -- Ginzburg model for P3, a Laurent polynomial (x + 1)2 xyz + y z + z, corresponds to smoothing P(1 : 1 : 2 : 4) to P3 as quadrics in P(1 : 1 : 1 : 1 : 2) (see [KP12], Example 2.13). Another non-Minkowski polynomial, x + y + z + 1/xy2z4, is a Landau -- Ginzburg model for P(1 : 1 : 2 : 4) itself. Question 48. Is it true that for toric Landau -- Ginzburg models for smooth Fano varieties associated with toric degenerations in anticanonical embedding Minkowski principle holds? What about "if and only if "? it is not enough to consider Gorenstein5 toric Landau -- Remark 49. Unfortunately, Ginzburg models for all Fano threefolds. For example sextic double solid has no Gorenstein toric Landau -- Ginzburg model as there is no integral polytope of volume 2 3 containing only one integral point in the interior. For similar reasons there is no Gorenstein toric Landau -- Ginzburg model for double Veronese cone. Standard toric Landau -- Ginzburg models for both of these varieties are not canonical. All the remaining Picard rank 1 Fano threefolds have Gorenstein toric Landau -- Ginzburg models. Among 105 smooth Fano threefolds 98 ones can have Gorenstein toric Landau -- Ginzburg models. All of them have at least very weak ones (see [CCG+]). For more details and examples see [CKP12a]. 3! = 1 Example 50. There are 5 possible Gorenstein toric degenerations for P3 (see [Ka08]). Three of them give toric Landau -- Ginzburg models. There are 5 possible Gorenstein toric degenerations for quadric in P4 (see [Ka08]). Four of them give toric Landau -- Ginzburg models. Remark 51. There is an infinite number of (non-Gorenstein) toric weak Landau -- Ginzburg models for P3. Say, ones whose Newton polytopes are fan polytopes of P(a2, b2, c2, abc), where (a, b, c)'s are Markov triples, that is, triples of natural numbers whose greatest common divisor is 1 and a2 + b2 + c2 = 3abc (cf. [HP05] and [GU]). Due to Prokhorov's remark P3 does degenerate to such projective spaces (this follows from [HP05], Theorem 1.1). The author is grateful to I. Cheltsov, S. Galkin, V. Golyshev, A. Iliev, L. Katzarkov, V. Lazic, V. Nikulin, D. Orlov, K. Shramov, and A. Wilson for helpful comments and im- portant remarks, and to referee for advises on reorganization of the paper. References [AAK12] M. Abouzaid, D. Auroux, L. Katzarkov, Lagrangian fibrations on blowups of toric varieties and mirror symmetry for hypersurfaces, submitted to Ann. Math., preprint arXiv:1207.0042. 5 Toric variety is Gorenstein if a polytope dual to its fan polytope is integral. In particular, Gorenstein toric varieties are canonical. 17 [Alt94] K. Altmann, The versal Deformation of an isolated toric Gorenstein Singularity, Invent. Math., 128(3):443 -- 479, 1997, preprint arXiv:alg-geom/9403004. [AKO04] D. Auroux, L. Katzarkov, D. Orlov, Mirror symmetry for weighted projective planes and their noncommutative deformations, Ann. of Math. (2) 167 (2008), no. 3, 867 -- 943, preprint arXiv:math/0404281. [AKO05] D. Auroux, L. Katzarkov, D. Orlov, Mirror symmetry for Del Pezzo surfaces: Vanishing cycles and coherent sheaves, Inv. Math. 166, No. 3 (2006), 537 -- 582, preprint arXiv:math.AG/0506166. [Ba93] V. V. Batyrev, Dual polyhedra and mirror symmetry for Calabi -- Yau hypersurfaces in toric vari- eties, J. Algebraic Geom. 3 (1994), no. 3, 493 -- 535, preprint arXiv:alg-geom/931000. [Ba97] V. V. Batyrev, Toric Degenerations of Fano Varieties and Constructing Mirror Manifolds, Collino, Alberto (ed.) et al., The Fano conference. Papers of the conference, organized to commemorate the 50th anniversary of the death of Gino Fano (1871 -- 1952), Torino, Italy, September 29 -- October 5, 2002. Torino: Universita di Torino, Dipartimento di Matematica. 109 -- 122 (2004), preprint arXiv:alg-geom/9712034. [Be83] F. Beukers, Irrationality of π2, periods of an elliptic curve and Γ1(5), Approximations diophanti- ennes et nombres transcendants, Colloq. Luminy/Fr. 1982, Prog. Math. 31, 47 -- 66. [BP84] F. Beukers, C. A. M. Peters, A family of K3 surfaces and ζ(3), J. Reine Angew. Math. 351, 42 -- 54 (1984). [BCFKS97] V. V. Batyrev, I. Ciocan-Fontanine, B. Kim, D. van Straten, Conifold transitions and mirror symmetry for Calabi -- Yau complete intersections in Grassmannians, Nucl. Phys., B 514, No.3, 640 -- 666 (1998), preprint arXiv:alg-geom/9710022. [BCFKS98] V. V. Batyrev, I. Ciocan-Fontanine, B. Kim, D. van Straten, Mirror Symmetry and Toric Degenerations of Partial Flag Manifolds, Acta Math. 184, No. 1 (2000), 1 -- 39, preprint arXiv:math.AG/9803108. [BCK03] A. Bertram, I. Ciocan-Fontanine, B. Kim, Two Proofs of a Conjecture of Hori and Vafa, Duke Math. J. 126, No. 1, 101 -- 136 (2005), preprint arXiv:math.AG/0304403. [BS85] F. Beukers, J. Stienstra, On the Picard -- Fuchs equation and the formal Brauer group of certain elliptic K3 surfaces, Math. Ann, 271 (1985), 269 -- 304. [BG] A. Bondal, S. Galkin, Mirror symmetry for minuscule varieties, in preparation. [CKP12a] I. Cheltsov, L. Katzarkov, V. Przyjalkowski, Birational geometry via moduli spaces, to appear in "Birational geometry, rational curves, and arithmetic -- Simons symposium 2012". [CKP12b] I. Cheltsov, L. Katzarkov, V. Przyjalkowski, Projecting Fanos in the mirror, in preparation. [CG06] A. Corti, V. Golyshev, Hypergeometric Equations and Weighted Projective Spaces, Sci. China, Math. 54, No. 8, 1577-1590 (2011), preprint arXiv:math/0607016. [CCG+] T. Coates, A. Corti, S. Galkin, V. Golyshev, A. Kasprzyk. Fano varieties and extremal Laurent polynomials. A collaborative research blog, http://coates.ma.ic.ac.uk/fanosearch/. [DH86] V. I. Danilov, A. G. Khovanskij, Newton polyhedra and an algorithm for computing Hodge-Deligne numbers, Math. USSR, Izv. 29, 279 -- 298 (1987). [DKLP] C. Doran, L. Katzarkov, J. Lewis, V. Przyjalkowski, Modularity of Fano threefolds, in prepara- tion. [EHX97] T. Eguchi, K. Hori, C.-Sh. Xiong, Gravitational quantum cohomology, Int. J. Mod. Phys. A 12, No. 9, 1743 -- 1782 (1997), preprint arXiv:hep-th/9605225. [HP05] P. Hacking, Yu. Prokhorov, Smoothable del Pezzo surfaces with quotient singularities, Com- pos. Math. 146, No. 1, 169 -- 192 (2010), see Degenerations of del Pezzo surfaces I, preprint arXiv:math/0509529. [HV00] H. Hori, C. Vafa, Mirror symmetry, preprint arXiv:hep-th/0002222. [Ga07] S. Galkin, Thesis, http://www.mi.ras.ru/$\sim$galkin/papers/disser.pdf (in russian), Small toric degenerations of Fano 3-folds, http://www.mi.ras.ru/$\sim$galkin/work/3a.pdf, preprint (2008). [GU] S. Galkin, A. Usnich, Mutations of potentials, http://member.ipmu.jp/sergey.galkin/papers/ipmu-10-0100.pdf, preprint. [Gi96] A. Givental, Stationary phase integrals, quantum Toda lattices, flag manifolds and the mirror conjecture, Topics in singularity theory, 103 -- 115, Amer. Math. Soc. Transl. Ser. 2, 180, Amer. Math. Soc., Providence, RI, 1997, preprint arXiv:alg-geom/9612001. [Go05] V. Golyshev, Classification problems and mirror duality, Young, Nicholas (ed.), Surveys in geom- etry and number theory. Reports on contemporary Russian mathematics. Cambridge: Cambridge University Press. LMS Lecture Note Ser. 338, 88 -- 121 (2007), preprint arXiv:math.AG/0510287. [GS07] V. Golyshev, J. Stienstra, Fuchsian equations of type DN, Comm. Num. Th. Phys. 1 (2007), no. 2, 323 -- 346, preprint arXiv:math/0701936. 18 [GKR12] M. Gross, L. Katzarkov, H. Ruddat, Towards Mirror Symmetry for Varieties of General Type, submitted to JAMS, preprint arXiv:1202.4042. [IKP11] A. Iliev, L. Katzarkov, V. Przyjalkowski, Double solids, categories and non-rationality, to appear in PEMS, Shokurov's volume, 2012, preprint arXiv:1102.2130. [ILP11] N. Ilten, J. Lewis, V. Przyjalkowski. Toric Degenerations of Fano Threefolds Giving Weak Landau -- Ginzburg Models, to appear in Journal of Algebra, ArXiv: 1102.4664. [Isk77] V. A. Iskovskikh, Fano treefolds I, II, Math USSR, Izv. 11 (1977-1978), 485 -- 527, 469 -- 506. [Ka08] A. Kasprzyk, Canonical toric Fano threefolds, preprint arXiv:0806.2604. [KKP08] L. Katzarkov, M. Kontsevich, T. Pantev, Hodge theoretic aspects of mirror symmetry, Proceed- ings of Symposia in Pure Mathematics 78, 87 -- 174 (2008), preprint arXiv:0806.0107. [KP09] L. Katzarkov, V. Przyjalkowski, Generalized Homological Mirror Symmetry and cubics, Proc. Steklov Inst. Math., vol. 264, 2009, 87 -- 95. [KP12] L. Katzarkov, V. Przyjalkowski, Landau -- Ginzburg models -- old and new. Akbulut, Selman (ed.) et al., Proceedings of the 18th Gokova geometry -- topology conference. Somerville, MA: International Press; Gokova: Gokova Geometry-Topology Conferences, 97 -- 124 (2012). [Ko94] M. Kontsevich, Homological algebra of mirror symmetry, Proc. International Congress of Matem- aticians (Zurich 1994), Birkhauzer, Basel, 1995, pp. 120 -- 139, preprint (1994), alg-geom/9411018. [Ma99] Yu. Manin, Frobenius manifolds, quantum cohomology, and moduli spaces, Colloquium Publica- tions. American Mathematical Society (AMS). 47. Providence, RI: American Mathematical Society (AMS) (1999). [Pr05] Yu. Prokhorov, The degree of Fano threefolds with canonical Gorenstein singularities, Sbornik: Mathematics, 196 (2005), 77 -- 114, preprint math/0610026. [Prz04] V. Przyjalkowski, Gromov -- Witten invariants of Fano threefolds of genera 6 and 8, Sb. Math. 198 (2007), no. 3-4, 433 -- 446, preprint math.AG/0410327. [Prz05] V. Przyjalkowski, Quantum cohomology of smooth complete intersections in weighted projec- tive spaces and singular toric varieties, Sb. Math. 198 (2007), No. 9 -- 10, 1325 -- 1340, preprint math.AG/0507232. [Prz07] V. Przyjalkowski, Minimal Gromov -- Witten ring, Izv. Math, Vol. 72, No. 6 (2008), 203 -- 226, preprint arXiv:0710.4084. [Prz08] V. Przyjalkowski. On Landau -- Ginzburg models for Fano varieties, Comm. Num. Th. Phys., Vol. 1, No. 4, 713 -- 728 (2008), preprint arXiv:0707.3758. [Prz10] V. Przyjalkowski. Hori -- Vafa mirror models for complete intersections in weighted projective spaces and weak Landau -- Ginzburg models, Cent. Eur. J. Math. 9, No. 5, 972 -- 977 (2011), preprint arXiv:1003.5200. [Re07] K. Rietsch, A mirror symmetric solution to the quantum Toda lattice, Commun. Math. Phys. 309, No. 1, 23-49 (2012), preprint arXiv:0705.3202. [vEvS06] C. van Enckevort, D. van Straten, Monodromy calculations of fourth order equations of Calabi -- Yau type, Yui, Noriko (ed.) et al., Mirror symmetry V. Proceedings of the BIRS workshop on Calabi- Yau varieties and mirror symmetry, December 6 -- 11, 2003. Providence, RI: American Mathematical Society (AMS); Somerville, MA: International Press. AMS/IP Studies in Advanced Mathematics 38, 539-559 (2006), preprint arXiv:math/0412539. Steklov Mathematical Institute, 8 Gubkina street, Moscow 119991, Russia E-mail address: [email protected], [email protected] 19
1606.02046	3	1606	2018-04-09T13:48:50	Global generation and very ampleness for adjoint linear series	[ "math.AG", "math.CV" ]	Let $X$ be a smooth projective variety over an algebraically closed field $\mathbb{K}$ with arbitrary characteristic. Suppose $L$ is an ample and globally generated line bundle. By Castelnuovo--Mumford regularity, we show that $K_X \otimes L^{\otimes \dim X} \otimes A$ is globally generated and $K_X \otimes L^{\otimes (\dim X+1)} \otimes A$ is very ample, provided the line bundle $A$ is nef but not numerically trivial. On complex projective varieties, by investigating Kawamata-Viehweg-Nadel type vanishing theorems for vector bundles, we also obtain the global generation for adjoint vector bundles. In particular, for a holomorphic submersion $f:X\longrightarrow Y$ with $L$ ample and globally generated, and $A$ nef but not numerically trivial, we prove the global generation of $ f_*(K_{X/Y})^{\otimes s}\otimes K_Y \otimes L^{\otimes \dim Y} \otimes A$ for any positive integer $s$.	math.AG	math	Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang Abstract. Let X be a smooth projective variety over an algebraically closed field K with arbitrary characteristic. Suppose L is an ample and globally generated line bundle. By Castelnuovo–Mumford regularity, we show that KX ⊗ L⊗ dim X ⊗ A is globally generated and KX ⊗ L⊗(dim X+1) ⊗ A is very ample, provided the line bundle A is nef but not numerically trivial. On complex projective varieties, by investigating Kawamata-Viehweg-Nadel type vanishing theorems for vector bun- dles, we also obtain the global generation for adjoint vector bundles. In partic- ular, for a holomorphic submersion f : X → Y with L ample and globally gen- erated, and A nef but not numerically trivial, we prove the global generation of f∗(KX/Y )⊗s ⊗ A for any positive integer s. ⊗ KY ⊗ L⊗ dim Y 8 1 0 2 r p A 9 ] . G A h t a m [ 3 v 6 4 0 2 0 . 6 0 6 1 : v i X r a Contents Introduction 1. 2. Fujita Conjecture type theorems on projective varieties over arbitrary fields 3. Vanishing theorems for vector bundles on compact Kahler manifolds 4. Fujita Conjecture type theorems on complex projective varieties References 1 4 6 12 15 1. Introduction In [20], Fujita proposed the following conjecture Conjecture. Let X be a smooth projective variety and L be an ample line bundle. Then (1) KX ⊗ L⊗(dim X+1) is globally generated; (2) KX ⊗ L⊗(dim X+2) is very ample. Fujita's conjecture is a deceptively simple open question in classical algebraic geometry. Up to dimension 4, the global generation conjecture has been proved ([47, 13, 31]). Recently, Fei Ye and Zhixian Zhu proved it in dimension 5 ([56]). Also, there are many other "Fujita Conjecture type" theorems have been proved, and we refer the reader to [9, 34, 1, 50, 51, 28, 29, 52, 23, 2, 32, 44, 3, 45, 48, 55, 18, 19]) and the references therein. In this paper, we prove Fujita Conjecture type theorems by using analytic methods in complex geometry when the twisted line (resp. vector) bundle is nef (resp. Nakano semi- positive). 1 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series 1.1. Fujita Conjecture type theorems on projective varieties over arbitrary fields. Let K be an algebraically closed field with arbitrary characteristic. By using characteristic p methods, Keeler proved in [32, Theorem 1.1] the following interesting result. Theorem 1.1 (Keeler). Let X be a smooth projective variety over K with dimension n. Suppose L is an ample and globally generated line bundle, and A is an ample line bundle. Then (1) KX ⊗ L⊗n ⊗ A is globally generated; (2) KX ⊗ L⊗(n+1) ⊗ A is very ample. In the case K = C, Angehrn and Siu proved the very ampleness part of Theorem 1.1 by analytic methods ([1, Lemma 11.1]). In [48, Corollary 4.8], Schwede generalized the global generation part of Theorem 1.1 to the case when A is nef and big, and ch(K) = p > 0. The first result of our paper deals with a more general case when A is nef but not numerically trivial. Theorem 1.2. Let X be a smooth projective variety over K with dimension n. Suppose L is an ample and globally generated line bundle. If A is a nef but not numerically trivial line bundle, then (1) KX ⊗ L⊗n ⊗ A is globally generated; (2) KX ⊗ L⊗(n+1) ⊗ A is very ample. As far as the authors know, this is the greatest generality in which Fujita Conjecture type theorem has been proved in any characteristic. Theorem 1.2 is also optimal in the sense that one can not drop the non-triviality condition on A, which can be seen from the example (X, L) = (Pn,O(1)). One may want to know the limit case when A is indeed a trivial line bundle. For the global generation part, Smith proved in [52, Theorem 2] that, this example is the only exceptional case, i.e., if (X, L) 6= (Pn,O(1)), then KX ⊗ L⊗n is globally generated. Her proof relies on the "tight closure" methods in the frame of commutative algebra. 1.2. Fujita Conjecture type theorems on complex projective varieties. In this sub- section, we focus on the cases in complex geometry, i.e. X is defined over the complex number field C. At first, we obtain a general version of the global generation part in Theorem 1.2: Theorem 1.3. Let X be a compact Kahler manifold of dimension n and L be an ample and globally generated line bundle. Suppose (A, e−2ϕ) be a pseudo-effective line bundle and I(ϕ) is the multiplier ideal sheaf. If the numerical dimension of (A, ϕ) is not zero, i.e. nd(A, ϕ) 6= 0, then (1.1) is globally generated. KX ⊗ L⊗n ⊗ A ⊗ I(ϕ) A key ingredient in the proof of Theorem 1.3 relies on vanishing theorems for pseudo-effective line bundles ([26, Corollary 1.7], [11, Theorem 0.15], [25, Corollary 3.2] or a weaker version [5, Theorem 1.3]). By using analytic methods, we also investigate the globally generated property for adjoint vector bundles. 2 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang Theorem 1.4. Let X be a compact Kahler manifold of dimension n and L be an ample and globally generated line bundle. Let (E, h) be a Hermitian holomorphic vector bundle with Nakano semi-positive curvature. Suppose A is a nef but not numerically trivial line bundle, then the adjoint vector bundle (1.2) is globally generated. KX ⊗ L⊗n ⊗ (E ⊗ A) Theorem 1.4 is derived from the following vanishing theorem for vector bundles, building on ideas in [26], [10] and [5]. Theorem 1.5. Let X be a complex projective variety with dim X = n. If (A, e−2ϕ) is a pseudo-effective line bundle and (E, h) a Nakano semi-positive vector bundle, then (1.3) H q(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) = 0 for q > n − nd(A, ϕ). Remark. According to [27], it is not hard to see that Theorem 1.5 is also true on compact Kahler manifolds. Hence, there is a version of Theorem 1.4 for pseudo-effective line bundle (A, e−2ϕ) (e.g. Theorem 4.3). For simplicity, we only formulate applications for nef line bundles (see Theorem 4.4 for general cases). As an application of Theorem 1.4, we obtain the following result in pure algebraic language. Theorem 1.6. Let f : X → Y be a holomorphic submersion between two complex projective varieties and dim Y = n. Suppose L → Y is an ample and globally generated line bundle, and A → Y is a nef but not numerically trivial line bundle. Then f∗(KX/Y )⊗s ⊗ KY ⊗ L⊗n ⊗ A (1.4) is globally generated for any s ≥ 1. As a special case of Theorem 1.6, we obtain the following well-known result of Koll´ar ([33, Theorem 3.5, Theorem 3.6]): Corollary 1.7 (Koll´ar). Let f : X → Y be a holomorphic submersion between two smooth projective varieties and dimC Y = n. Suppose L → Y is an ample and globally generated line bundle, then is globally generated for s ≥ 1. f∗(KX/Y )⊗s ⊗ KY ⊗ L⊗(n+1) As another application of Theorem 1.4, we also get global generation of pluricanonical adjoint bundles of canonically polarized families. Theorem 1.8. Let f : X → S be a holomorphic family of canonically polarized compact Kahler manifolds effectively parameterized by a smooth projective variety S with dimension n. Suppose L → S is an ample and globally generated line bundle, and A → S is a nef line bundle. Then (1.5) is globally generated for s > 1. f∗(K ⊗s X ) ⊗ L⊗n ⊗ A 3 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series 2. Fujita Conjecture type theorems on projective varieties over arbitrary fields In this section, we investigate Fujita Conjecture type theorems on projective varieties over algebraically closed fields and prove Theorem 1.2. Let K be an algebraically closed field with arbitrary characteristic, and X be a smooth projective variety over K. Firstly we introduce the theory of Castelnuovo–Mumford regularity. 2.1. Castelnuovo–Mumford regularity. Suppose L is an ample and globally generated line bundle over X. Definition 2.1. A coherent sheaf F on X is m-regular with respect to L if (2.1) q > 0. for H q(X,F ⊗ L⊗(m−q)) = 0 The following results are well-known (e.g. [35, Section 1.8], or [14, Section 5.2]), and for the sake of completeness we include a proof here. Lemma 2.2 (Mumford). Let F be a 0-regular coherent sheaf on X with respect to L, then F is generated by its global sections. Proof. Suppose dim X = n. We shall use standard hyperplane induction method to prove it. For simplicity, we write L = OX (1). Since the coherent sheaf F has finitely many associated points, we can choose s ∈ H 0(X, L) such that the corresponding divisor B does not contain any associated point of F . Hence, for any i ≥ 0, we have the exact sequence (2.2) 0 → F (−i − 1) ⊗s→ F (−i) → FB(−i) → 0. By using the associated long exact sequence (2.3) ··· → H i(X,F (−i)) → H i(X,FB(−i)) → H i+1(X,F (−i − 1)) → ··· and 0-regularity of F , we see H i(X,FB(−i)) = 0 for i > 0, i.e. FB is 0-regular with respect to L. Similarly, for any q ≥ 0, we have (2.4) ··· → H q(X,F (−q)) → H q(X,F (1)(−q)) → H q(X,FB(1)(−q)) → ··· . We show F (1) is 0-regular if F is 0-regular. Indeed, we have FB is 0-regular and by hyperplane induction hypothesis, FB(1) is 0-regular. By (2.4), F (1) is 0-regular. Hence, we know F (k) is 0-regular for all k ≥ 0. In the commutative diagram H 0(X,F (k)) ⊗ H 0(X,O(1)) rk⊗1 / H 0(X,FB(k)) ⊗ H 0(X,OX (1)) ιX ιB H 0(X,F (k)) ⊗s / H 0(X,F (k + 1)) rk+1 / H 0(X,FB(k + 1)), rk⊗1 and rk+1 are surjective according to long exact sequence associated to (2.2). By diagram chasing, it is obvious that ιX is surjective if and only if ιB is surjective. Note that in the above commutative diagram we can replace B by intersections of suitable divisors in L. Hence, we can show ιX is surjective by induction on dim B. When dim B = 0, it is easy to see ιB is 4 / / / Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang surjective. Hence, by induction, we know ιX is surjective. For any x ∈ X and large N , in the commutative diagram H 0(X,F ) ⊗ H 0(X,O(1))⊗N ι / H 0(X,F (N )) 1⊗ev1 ev2 H 0(X,F ) ⊗ (O(1)x)⊗N f / (F (N ))x, ι is surjective since ιX is surjective. On the other hand, 1 ⊗ ev1 and ev2 are both surjective and so f is surjective, and we deduce H 0(X,F ) → Fx is surjective. 2.2. The proof of Theorem 1.2. Before giving the proof of Theorem 1.2, we present a more general result. (cid:3) Theorem 2.3. Let X be a smooth projective variety over K with dimension n. Suppose L is an ample and globally generated line bundle. If A is a line bundle such that L ⊗ A is ample and the Kodaira-Iitaka dimension κ(A∗) = −∞, then (1) KX ⊗ L⊗n ⊗ A is globally generated; (2) KX ⊗ L⊗(n+1) ⊗ A is very ample. ∗ (KX ) ⊗ L⊗n ⊗ A is 0-regular for Proof. Suppose ch(K) = p > 0. We shall show that F k all k ≥ k0 where F : X → X is the absolute Frobenius morphism. When 0 < q < n, i.e. n − q ≥ 1, L⊗(n−q) ⊗ A is ample since both L and L ⊗ A are ample. Hence, by the Serre vanishing theorem, for each q with 0 < q < n, there exists a positive constant kq = k(q) > 0 such that (2.5) H q(X, KX ⊗ L⊗pk(n−q) ⊗ A⊗pk ) = 0, for k ≥ kq. By the projection formula, one has ∗ (KX ⊗ L⊗pk(n−q) ⊗ A⊗pk F k (2.6) ) = F k ∗ (KX ) ⊗ L⊗(n−q) ⊗ A. Since the Frobenius morphism F is a finite morphism, we get H q(X, F k ∗ (KX ) ⊗ L⊗(n−q) ⊗ A) ∼= H q(X, F k ∗ (KX ⊗ L⊗pk(n−q) ⊗ A⊗pk )) ∼= H q(X, KX ⊗ L⊗pk(n−q) ⊗ A⊗pk ) = 0. When q = n, we want to show (2.7) By the projection formula again, we have F k H n(X, F k ∗ (KX) ⊗ A) = 0. ∗ (KX ) ⊗ A = F k ∗ (KX ) ⊗ A) ∼= H n(cid:16)X, F k ∗ (cid:16)KX ⊗ A⊗pk(cid:17) and ∗ (KX ⊗ A⊗pk )(cid:17) H n(X, F k ∼= H n(X, KX ⊗ A⊗pk ∼= H 0(X, (A⊗pk )∗)∗. ) Since the Kodaira-Iitaka dimension κ(A∗) = −∞, we have H 0(X, (A∗)⊗ℓ) = 0 for any ℓ > 0. Hence we get (2.7). By Definition 2.1, F k ∗ (KX ) ⊗ A ⊗ L⊗n is 0-regular for all k ≥ k0 where 5 / / Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series k0 = max{k1,··· , kn−1}. According to Lemma 2.2, F k Thanks to [32, Lemma 3.3], KX ⊗ L⊗n ⊗ A is a quotient of ∗ (KX )⊗ L⊗n ⊗ A is globally generated. F k ∗ (KX ) ⊗ A ⊗ L⊗n for all k ≥ k0 and so KX ⊗ L⊗n ⊗ A is globally generated. It is well-known that KX ⊗ L⊗(n+1) ⊗ A is very ample if and only if for every x ∈ X, ∗ (KX ) ⊗ L⊗n ⊗ A is 0-regular with mx ⊗ KX ⊗ L⊗(n+1) ⊗ A is globally generated. Since F k ∗ (KX ) ⊗ L⊗(n+1) ⊗ A is also respect to L, it is proved in [32] that for every x ∈ X, mx ⊗ F k 0-regular and so it is globally generated. By [32, Lemma 3.3], mx ⊗ KX ⊗ L⊗(n+1) ⊗ A is a quotient of mx⊗ F k ∗ (KX )⊗ L⊗(n+1)⊗ A. Hence mx⊗ KX ⊗ L⊗(n+1)⊗ A is globally generated. When ch(K) = 0, L⊗k ⊗ A is ample for every k > 0. Hence by Kodaira vanishing theorem, we have H q(X, KX ⊗ L⊗(n−q) ⊗ A) = 0 (2.8) for 0 < q < n. Since κ(A∗) = −∞, we also have (2.9) Hence, by Lemma 2.2 again, we know KX ⊗ L⊗n⊗ A is 0-regular with respect to L and hence it is globally generated. The very ampleness of KX ⊗ L⊗(n+1)⊗ A can be proved similarly. (cid:3) Lemma 2.4. Let A be a nef but not numerically trivial line bundle over a smooth projective variety X over K. Then the Kodaira-Iitaka dimension κ(A∗) = −∞, i.e. H 0(X, (A∗)⊗ℓ) = 0 for all ℓ > 0. H n(X, KX ⊗ A) ∼= H 0(X, A∗) = 0. Proof. We also use A to denote the divisor class of A. Since A is nef but not numerically trivial, it is well-known that (e.g. [6, Section 3.8]) that there exists an ample divisor H, such that A · H n−1 > 0. We show κ(A∗) = −∞. Suppose H 0(X,O(−ℓA)) 6= 0 for some ℓ > 0. Then −ℓA is effective. Let −ℓA =P νiDi where Di are irreducible divisors and νi ≥ 0. Since −ℓA is not numerically trivial, there is at least one νi > 0. By Nakai-Moishezon criterion for ampleness, for ample divisor H in X, we have H n−1 · (−ℓA) > 0, i.e. H n−1 · A < 0 which contradicts A · H n−1 > 0. The proof of Theorem 1.2. It follows from Lemma 2.4 and Theorem 2.3. (cid:3) Remark 2.5. Theorem 2.3 is more general than Theorem 1.2. In Theorem 2.3, A can be certain numerically trivial line bundles. It is easy to see that, if A is a non-torsion point in P ic0(X), Theorem 2.3 also works. For instance, let E = C/(Z ⊕ √−1Z) be an elliptic curve. Suppose A = O(P − Q) where P is a rational point and Q is an irrational point on E. Then A is numerically trivial and κ(A∗) = −∞. Indeed, (A∗)⊗ℓ has no nonzero section for any ℓ > 0. Otherwise, the divisor ℓ(P − Q) is linearly equivalent to the zero divisor, which is absurd. 3. Vanishing theorems for vector bundles on compact Kahler manifolds In this section, we investigate various vanishing theorems for vector bundles, which are the key ingredients in the proof of Fujita Conjecture type theorems. In particular, we give the proof of Theorem 1.5. 6 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang Let E be a holomorphic vector bundle over a compact complex manifold X and h be a smooth Hermitian metric on E. There exists a unique connection ∇ which is compatible with the Hermitian metric h and the complex structure on E. It is called the Chern connection of i=1 be local holomorphic coordinates on X and {eα}r (E, h). Let {zi}n α=1 be a local frame of E. The curvature tensor ΘE ∈ Γ(X, Λ2T ∗X ⊗ E ∗ ⊗ E) has components (3.1) ∂2hαβ ∂zi∂zj + hγδ ∂hαδ ∂zi ∂hγβ ∂zj . Rijαβ = − Here and henceforth we sometimes adopt the Einstein convention for summation. Definition 3.1. A Hermitian holomorphic vector bundle (E, h) is called Nakano positive (resp. Nakano semi-positive) if Rijαβuiαujβ > 0 (resp. ≥ 0) for nonzero vector u = (uiα) ∈ Cnr. Let's describe some elementary properties on positive vector bundles. Lemma 3.2 (Nakano vanishing theorem). Let X be a compact Kahler manifold. Suppose (E, h) → X is a Hermitian holomorphic vector bundle with Nakano positive curvature, then (3.2) H q(X, KX ⊗ E) = 0, q ≥ 1. Lemma 3.3. Let (X, ωg) be a compact Kahler manifold. (1) Let (E, h) be a Nakano positive vector bundle and A a nef line bundle. Then E ⊗ A admits a Hermitian metric with Nakano positive curvature. (2) Let (E, h) be a Nakano semi-positive vector bundle and A be an ample line bundle. (3) Let (E, hE) and (eE, h Then E ⊗ A admits a Hermitian metric with Nakano positive curvature. eh) is also Nakano semi-positive. eE) be two Nakano semi-positive vector bundles, then (E⊗ eE, h⊗ √−1ΘE(u(x), u(x)) ≥ 2εu(x)2 Proof. (1). For the fixed Kahler metric ωg on X, there exists a constant ε > 0 such that for any u ∈ Γ(X, T 1,0X ⊗ E). On the other hand, by analytic definition of nefness (e.g. [10]), there exists a smooth metric h0 on the nef line bundle A such that g⊗h √−1ΘA ≥ −εωg. √−1ΘE⊗A(u ⊗ v, u ⊗ v) ≥(cid:0)√−1ΘE(u, u) − εu2 (3.3) The curvature of h ⊗ h0 on E ⊗ A is ΘE⊗A = ΘE · idA + idE · ΘA. Hence, for any u ∈ Γ(X, T 1,0X ⊗ E) and v ∈ Γ(X, A) (3.4) Therefore, E ⊗ A is Nakano positive. The proof of (2) is similar to that of (1). (3). By using curvature formula of h ⊗eh on E ⊗ eE, ΘE⊗A = ΘE · id eE + idE · Θ for any (local) vectors u ∈ Γ(X, T 1,0X ⊗ E ⊗ eE) with the form u = uiαAdzi ⊗ eα ⊗ eA in the local holomorphic frames {zi, eα, eA} of {X, E, eE}, we obtain g⊗h(cid:1)v2 h0 ≥ εu2 g⊗hv2 h0. ΘE⊗ eE(u, u) = ΘE (3.5) eE, eE AB + Θ eE ijABuiαAujβB · hE αβ. ijαβuiαAujβB · h 7 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series It is nonnegative and one can see that by choosing normal coordinates for hE and h fixed point. eE at a (cid:3) We need the following fundamental result in [12, Proposition 1.16]. Lemma 3.4. Let E → X be a nef vector bundle over a compact complex manifold X. Suppose σ ∈ H 0(X, E ∗) is a nonzero section, then σ does not vanish anywhere. By refining the Bochner technique, we obtain the following vanishing theorem for vector bundles with "degenerate" curvature tensors. Theorem 3.5. Let (X, ω) be a compact Kahler manifold of dimension n. Let (E, h) be a holomorphic vector bundle and A be a line bundle. Suppose either (1) E is nef and the second Ricci curvature trωΘE is semi-positive, and A is semi-ample but non-trivial; or (2) the second Ricci curvature trωΘE is strictly positive and A is nef. Then (3.6) H 0(X, E ∗ ⊗ A∗) = 0. Proof. (1). Since A is semi-ample, Ak is generated by its global sections for large k. Hence, there is an induced smooth Hermitian metric hA on A such that the curvature ΘA is semi- positive, i.e. √−1ΘA = −√−1∂∂ log hA ≥ 0. On the other hand, since A is not trivial, for the fixed Kahler metric ω on X, the scalar curvature function trωΘA is non-negative, but not identically zero. Indeed, if it is identically zero, we deduce that ΘA is identically zero, and so A is trivial. The curvature tensor of E ⊗ A can be written as ΘE⊗A = ΘE ⊗ idA + idE ⊗ ΘA. In order to prove (3.5), we argue by contradiction. Suppose H 0(X, E ∗ ⊗ A∗) 6= 0, i.e. there exists a nonzero section σ ∈ H 0(X, E ∗ ⊗ A∗). By Lemma 3.4, σ is nowhere vanishing. By using standard Bochner identity over E ∗ ⊗ A∗, (3.7) , Λω], ∆∂ = ∆∂ + [√−1ΘE ∗⊗A∗ we obtain (3.8) Note that ΘE ∗ (3.9) 0 = k∇σk2 − (trωΘE ∗⊗A∗ σ, σ). = −(ΘE)T as (1, 1)-form valued r × r matrices. Hence, 0 = k∇σk2 +(cid:16)(cid:16)(cid:2)trωΘE(cid:3)T ⊗ idA + idE ⊗ trωΘA(cid:17) σ, σ(cid:17) . By using local holomorphic frames {zi}, {e1,··· , er}, {e} of X, E and A respectively, we write ω = √−1gijdzi ∧ dzj, ΘE = Rβ dzi ∧ dzj ⊗ eα ⊗ eβ, ΘA = RA ij dzi ∧ dzj, σ = σαeα ⊗ e. ijα 8 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang We obtain (3.10) ZX hA(cid:16)gijRijβασασβ + gijRA ijhαβσασβ(cid:17) ωn = 0. By assumption, the (transposed) second Ricci curvature (cid:2)trωΘE(cid:3)T = Xi,j gijRijβα  is Hermitian semi-positive as a (r × r) matrix and so (3.11) gijRijβασασβ + gijRA ij hαβσασβ ≡ 0. It implies Since σ is nowhere vanishing, we obtain hαβσασβ > 0 at each point. Therefore gijRA ij (hαβσασβ) ≡ 0. trωΘA = gijRA ij ≡ 0. This is a contradiction. (2). For nef A, we use similar ideas as described in the first part of Lemma 3.3. Since trωΘE is strictly positive, there exists ε > 0 such that (trωΘE)T (u(x), u(x)) ≥ (n + 1)εu(x)2 h for any u ∈ Γ(X, E) since X is compact. On the other hand, since A is nef, there exists a smooth metric h0 on the nef line bundle A such that √−1ΘA ≥ −εωg. Hence, for any u ∈ Γ(X, E) and v ∈ Γ(X, A) (cid:16)(cid:2)trωΘE(cid:3)T ⊗ idA + idE ⊗ trωΘA(cid:17) (u ⊗ v, u ⊗ v) ≥ (cid:0)(trωΘE)T (u, u) − nεu2 h(cid:1)v2 h0 ≥ εu2 hv2 h0. Therefore in (3.11), σ ≡ 0, i.e. H 0(X, E ∗ ⊗ A∗) = 0. Remark 3.6. The semi-positivity of the second Ricci curvature trωΘE can be replaced by the semi-stability of E with respect to ω following Donaldson-Uhlenbeck-Yau's theorem. Moreover, the Griffiths (or Nakano, or dual Nakano) semi-positivity of E can also imply the semi-positivity of the second Ricci curvature trωΘE. (cid:3) The following Kawamata-Viehweg-Nadel type vanishing theorem for a semi-positive vector bundle twisted by a big line bundle is essentially known to experts (e.g., [7, Theorem 4.2.4], [10, Theorem 5.11], [15, Theorem 1.2], [16, Theorem 1.1], [46, Theorem 1.1], [17], [39]), although the statement is not written down precisely. For the sake of completeness, we include a short sketch here, following the approach in [10, Theorem 5.11] for line bundles. Lemma 3.7. Let (X, ω) be a Kahler weakly pseudo-convex manifold, and let A be a holo- morphic line bundle over X equipped with a (possibly) singular Hermitian metric h = e−2ϕ. Assume that √−1ΘA ≥ εω 9 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series for some continuous positive function ε on X. If (E, hE) is a Nakano semi-positive vector bundle, then hE · e−2ϕ and ∂ E⊗A E⊗A E⊗A H q(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) = 0 (3.12) for all q ≥ 1. Proof. Let L q be the sheaf of germs of (n, q)-forms u with values in E⊗A and with measurable ug⊗hE · e−2ϕ are locally integrable. The coefficients such that both u2 ∂ ) which is a fine resolution of the sheaf O(KX ⊗ E ⊗ A) ⊗ I(ϕ), i.e. we have the following exact sequence (3.13) 0 → O(KX ⊗ E ⊗ A) ⊗ I(ϕ) → L 0 → L 1 → ··· → L n → 0. operator defines a complex of sheaves (L •, ∂ Indeed, it follows from a vector bundle version of Hormander L2-estimate ([10, Corollary 5.3]) since the vector bundle E⊗A has a singular metric which is Nakano positive in the sense of cur- rent. By using the L2 estimate again (e.g. [10, Theorem 5.11]), one can show H q(Γ(X, L •)) = 0 for q ≥ 1 and we obtain the desired vanishing cohomologies. (cid:3) Next, we introduce two different concepts on numerical dimension for nef and pseudo-effective line bundles. Definition 3.8. Let N be a nef line bundle over a compact Kahler manifold X with dimC X = n. The numerical dimension ν(N ) of N is defined as ([10, Definition 6.20]) 1(N ) 6= 0 ∈ H 2k(X, R).} ν(N ) = max{k = 0,··· , n ck (3.14) Definition 3.9. Let (A, e−2ϕ) be a pseudo-effective line bundle over a compact Kahler mani- fold X of dimension n. The numerical dimension nd(A, ϕ) of A is defined in [5, Definition 3.1] as the largest number such that the cohomological product h(√−1∂∂ϕ)ki 6= 0. Note that, in general these two definitions do not coincide even for nef line bundles ([5, Remark 7], or [11, Remark 4.3.5]). For a discussion of the relationship between various defi- nitions of numerical dimensions, we refer to the paper [11, Section 4.3]. Based on their solution to Demailly's strong openness conjecture ([25]), Guan-Zhou achieved in [26, Corollary 1.7] (see also [25, Corollary 3.2]) a celebrated Kawamata-Viehweg-Nadel type vanishing theorem which generalizes a theorem in [5, Theorem 1.3] (see also Demailly's survey paper [11] for a variety of vanishing theorems). Lemma 3.10. Let (A, e−2ϕ) be a pseudo-effective line bundle over a compact Kahler manifold X of dimension n. Then for any q > n − nd(A, ϕ), (3.15) H q(X, KX ⊗ A ⊗ I(ϕ)) = 0. We shall use the ideas in the proof of Lemma 3.10 (e.g. [5] and [25]) and Lemma 3.7 to prove Theorem 1.5. The proof of Theorem 1.5. Suppose nd(A, ϕ) = n. By using Guan-Zhou's solution to Demailly's strong openness conjecture ([25]) and the construction in [5, Lemma 5.5], there exists a singular metric e−2ϕ1 on the pseudo-effective line bundle A such that I(ϕ1) = I(ϕ) 10 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang and it is curvature current √−1∂∂ϕ1 > cω for some smooth positive (1, 1) form ω and constant c > 0. By applying Lemma 3.7 to (X, E, A,I(ϕ1)), we get (3.16) q > 0. for H q(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) = 0, Now we assume that nd(A, ϕ) < n. We use similar ideas as in [10, Theorem 6.25], [5, Proposition 5.6]. Let B be a very ample divisor such that B⊗A is ample and I(ϕB ) = I(ϕ)B ([17, Theorem 1.10]). We consider the exact sequence 0 → OX (−B) → OX → (iB)∗OB → 0. By tensoring with KX ⊗ E ⊗ A ⊗ B ⊗ I(ϕ) and using adjunction formula, one gets ··· → H q(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) → H q(X, KX ⊗ E ⊗ A ⊗ B ⊗ I(ϕ)) → H q(B, KB ⊗ (E ⊗ A)B ⊗ I(ϕB)) → H q+1(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) → ··· . Since E is Nakano semi-positive, and A ⊗ B is ample, by Lemma 3.7, we have (3.17) H q(X, KX ⊗ E ⊗ A ⊗ B ⊗ I(ϕ)) = 0 q > 0. for Therefore H q(B, KB ⊗ (E ⊗ A)B ⊗ I(ϕB)) ∼= H q+1(X, KX ⊗ E ⊗ A ⊗ I(ϕ)) (3.18) for every 0 < q < n. Moreover, EB is also Nakano semi-positive over B. Hence the induction hypothesis implies that the cohomology group on B on the right hand side of (3.18) is zero when q > n − nd(A, ϕ). (cid:3) Similarly, we have the following variant of Theorem 1.5 (see also [10, Theorem 6.17] for the line bundle case.) Proposition 3.11. Let X be a smooth projective variety. Let A be a line bundle over X such that some positive multiple mA can be written as mA = N + D where N is a nef line bundle and D is an effective divisor. If (E, h) is a Nakano semi-positive vector bundle, then (3.19) H q(X, KX ⊗ E ⊗ A ⊗ I(m−1D)) = 0 for q > n − ν(N ), where ν(N ) is the numerical dimension of the nef line bundle N . As a special case, one has Corollary 3.12. If (E, hE) is a Nanako semi-positive vector bundle and A is a nef line bundle, then (3.20) H q(X, E ∗ ⊗ A∗) = 0 if q < ν(A). In particular, if in addition, A is not numerically trivial, then (3.21) H 0(X, E ∗ ⊗ A∗) = 0. Proof. We can take m = 1, D = 0 and A = N in Proposition 3.11. By Serre duality, we obtain (3.20). Hence, when ν(A) ≥ 1, or equivalently, ν(A) 6= 0, H 0(X, E ∗ ⊗ A∗) = 0. It is well-known that for a nef line bundle A, ν(A) = 0 if and only if A is numerically trivial. (cid:3) 11 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series As an application of Theorem 1.5, we obtain general vanishing theorems on the relative setting which generalize Koll´ar's vanishing theorems. Theorem 3.13. Let f : X → Y be a holomorphic submersion between two smooth complex projective varieties and dimC Y = n. Let (A, e−2ϕ) be a pseudo-effective line bundle over X. If both E1 and E2 are Nakano semi-positive vector bundles, then for any s ≥ 1 and q > n − nd(A, ϕ), we have H q(cid:0)Y, f∗(KX/Y ⊗ E1)⊗s ⊗ KY ⊗ E2 ⊗ A ⊗ I(ϕ)(cid:1) = 0, as long as f∗(KX/Y ⊗ E1) is locally free. Proof. By using Theorem 1.5, we only need to show that f∗(KX/Y ⊗ E1)⊗s ⊗ E2 is Nakano semi-positive for any s ≥ 1. Indeed by a result of [41] (see also [38]), if f∗(KX/Y ⊗E1) is locally free, then f∗(KX/Y ⊗ E1) has a Nakano semi-positive metric. By part (3) of Lemma 3.3, we deduce is Nakano semi-positive for any s ≥ 1. Remark 3.14. Theorem 3.13 also holds when X and Y are compact Kahler manifolds. f∗(KX/Y ⊗ E1)⊗s ⊗ E2 (cid:3) 4. Fujita Conjecture type theorems on complex projective varieties In this section, we derive Fujita Conjecture type theorems on complex projective varieties and prove Theorem 1.3, Theorem 1.4, Theorem 1.6 and Theorem 1.8. The proof of Theorem 1.3. We show the coherent sheaf KX ⊗ L⊗n ⊗ A⊗ I(ϕ) is 0-regular, and so by Lemma 2.2, it is globally generated. When 0 < q < n, L⊗(n−q) ⊗ A is indeed a big line bundle, and we can apply Nadel vanishing theorem ([42] or [11, Theorem 0.3] or Lemma 3.10), (4.1) H q(X, KX ⊗ L⊗(n−q) ⊗ A ⊗ I(ϕ)) = 0. When q = n, we need to show (4.2) which follows from Lemma 3.10 and the assumption nd(A, ϕ) ≥ 1. H n(X, KX ⊗ A ⊗ I(ϕ)) = 0, (cid:3) The proof of Theorem 1.4. By Castelnuovo-Mumford regularity (e.g. Lemma 2.2), we only need to prove KX ⊗ L⊗n ⊗ (E ⊗ A) is 0-regular with respect to L. Hence, it suffices to show (4.3) q > 0. H q(X, KX ⊗ L⊗(n−q) ⊗ (E ⊗ A)) = 0 for all For 0 < q < n, we claim that the vector bundle L⊗(n−q)⊗(E⊗A) has a smooth metric with strictly positive curvature in the sense of Nakano, and by Nakano vanishing theorem( Lemma 3.2), we have the desired vanishing cohomologies. Indeed, since n − q ≥ 1, L⊗(n−q) ⊗ A is ample. By Lemma 3.3, E ⊗ L⊗(n−q) ⊗ A is strictly positive in the sense of Nakano. 12 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang When q = n, we need to show H n(X, KX ⊗ E⊗ A) = 0 or equivalently H 0(X, E ∗⊗ A∗) = 0 when E is Nakano semi-positive and A is nef but not numerically trivial. This is assured by Corollary 3.12. The case when q > n is obvious and we complete the proof of Theorem 1.4. (cid:3) We have the following variant of Theorem 1.4. Theorem 4.1. Let X be a compact Kahler manifold and L be an ample and globally gener- ated line bundle. Let (E, h) be a Hermitian holomorphic vector bundle with Nakano positive curvature. Suppose A to be a nef line bundle, then the vector bundle (4.4) is globally generated. KX ⊗ L⊗n ⊗ E ⊗ A Proof. We use similar ideas as described in the proof of Theorem 1.4. Suppose E is Nakano positive and A is nef, then by Lemma 3.3, the vector bundle E⊗A admits a smooth Hermitian metric whose curvature is strictly positive in the sense of Nakano. In particular, the second Ricci curvature trωΘE⊗A is strictly positive. We obtain the vanishing cohomology in Theorem 3.5. Finally, one can follow the steps in the proof of Theorem 1.4. (cid:3) It is proved in [4, Theorem 1.2] that if E is a semi-ample (resp. ample) vector bundle, then E ⊗ det E is Nakano semi-positive (resp. Nakano positive). Hence, by Theorem 1.4 and Theorem 4.1, we get Corollary 4.2. Let X be a smooth complex projective variety and L be an ample and globally generated line bundle. Let E be a vector bundle and A be a line bundle. Suppose either (1) E is semi-ample, A is nef but not numerically trivial; or (2) E is ample and A is nef. Then the vector bundle (4.5) is globally generated. KX ⊗ L⊗n ⊗ (E ⊗ det E) ⊗ A Thanks to Theorem 1.5, one can also get the following variant of Theorem 1.4. Theorem 4.3. Let (X, ω) be a compact Kahler manifold and L → X be an ample and globally generated line bundle. Suppose (A, e−2ϕ) is a pseudo-effective line bundle and I(ϕ) is the multiplier ideal sheaf. Let E be a Nakano semi-positive vector bundle. If the numerical dimension nd(A, ϕ) 6= 0, then KX ⊗ L⊗n ⊗ E ⊗ A ⊗ I(ϕ) is globally generated. The proof of Theorem 1.6. By the assumption, the direct image sheaf of the relative canonical line bundle f∗(KX/Y ) is indeed a holomorphic vector bundle. In the literatures, [54, it is known that the vector bundle f∗(KX/Y ) is weakly positive in suitable sense (e.g. Theorem III], [24, Corollary 5], [33, Corollary 3.7]). Here, we use a recent fact [4, Theorem 1.2] 13 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series of Berndtsson that f∗(KX/Y ) is actually semi-positive in the sense of Nakano. Let E = f∗(KX/Y )⊗s and so (4.6) f∗(KX/Y )⊗s ⊗ KY ⊗ L⊗n ⊗ A = KY ⊗ L⊗n ⊗ (E ⊗ A). According to Lemma 3.3, E is also semi-positive in the sense of Nakano. By Theorem 1.4, KY ⊗ L⊗n ⊗ (E ⊗ A) is globally generated as long as A is nef but not numerically trivial. The proof of Theorem 1.6 is completed. (cid:3) As an application of Theorem 3.13, we have the following slightly general version of The- orem 1.6 . Theorem 4.4. Let f : X → Y be a holomorphic submersion between two smooth complex projective varieties and dimC Y = n. Suppose that L → Y is an ample and globally generated line bundle, and (A, e−2ϕ) → Y be a pseudo-effective line bundle with nd(A, ϕ) 6= 0. Suppose both E1 and E2 are Nakano semi-positive, then for any s ≥ 1 (4.7) f∗(KX/Y ⊗ E1)⊗s ⊗ KY ⊗ L⊗n ⊗ E2 ⊗ A ⊗ I(ϕ) is globally generated as long as f∗(KX/Y ⊗ E1) is locally free. The proof of Theorem 1.8. When the family X → S is effectively parameterized, Schu- macher proved in [49, Theorem 1] that the naturally induced Hermitian metric on the relative canonical line bundle KX /S is strictly positive. By [49, Corollary 2] or [4, Theorem 1.2], we know (4.8) f∗(K ⊗s X/S) is strictly Nakano positive for all s > 1. One can write (4.9) f∗(K ⊗s X ) ⊗ L⊗n ⊗ A = f∗(K ⊗s X/S) ⊗ KS ⊗ L⊗n ⊗ (K ⊗(s−1) S ⊗ A) We first observe that the canonical bundle KS is nef. Indeed, by a recent result in [53, Theorem 1], the compact complex base S is actually (Kobayashi) hyperbolic. Hence, it contains no rational curve. By using Mori's cone theorem [40], we deduce KS is nef since it is a projective manifold without rational curve. The global generation of the vector bundle f∗(K ⊗s X/S) is Nakano positive and K ⊗(s−1) (cid:3) ⊗ A) follows from Theorem 4.1 since f∗(K ⊗s X/S) ⊗ KS ⊗ L⊗n ⊗ (K ⊗(s−1) ⊗ A is nef. S S Remark 4.5. On smooth complex projective varieties, we can also derive globally genera- tion for symmetric powers and wedge powers of vector bundles by using the corresponding vanishing theorems (e.g. [38]). Acknowledgements. This work was partially supported by China's Recruitment Program of Global Experts and NSFC 11688101. It was carried out when the authors attended the seminar on "Fujita con- jecture and related topics " held at Math Institute of Chinese Academy of Sciences, and the authors would like to thank all members of this seminar for stimulating discussions. In par- ticular, the authors wish to thank Yifei Chen, Baohua Fu and Xiaotao Sun for some useful 14 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang suggestions. The second author would like to thank Professor S.-T. Yau for his advice, sup- port and encouragement. He also thanks Kefeng Liu, Valentino Tosatti and Xiangyu Zhou for helpful discussions. The authors would also like to thank the anonymous referees whose comments and suggestions helped improve and clarify the paper. References [1] Angehrn, Urban; Siu, Yum-Tong. Effective freeness and point separation for adjoint bundles. Invent. Math. 122 (1995), no. 2, 291–308. [2] Arapura, D. Frobenius amplitude and strong vanishing theorems for vector bundles. With an appendix by Dennis S. Keeler. Duke Math. J. 121 (2004), no. 2, 231–267. [3] Bayer, A.; Bertram, A.; Macri, E.; Toda, Y. Bridgeland stability conditions of threefolds II: An application to Fujita's conjecture. J. Algebraic Geom. 23 (2014), no. 4, 693–710. [4] Berndtsson, B. Curvature of vector bundles associated to holomorphic fibrations. Ann. of Math. (2) 169 (2009), no. 2, 531–560. [5] Cao, Junyan. Numerical dimension and a Kawamata-Viehweg-Nadel-type vanishing theorem on compact Kahler manifolds. Compos. Math. 150 (2014), no. 11, 1869–1902. [6] Debarre O. Higher-dimensional algebraic geometry. Springer Science& business Media, 2013. [7] de Cataldo, M. A. Singular Hermitian metrics on vector bundles. J. Reine Angew. Math. 502 (1998), 93–122. [8] Deligne, Pierre; Illusie, Luc. Rel`evements modulo p2 et d´ecomposition du complexe de de Rham. Invent. Math. 89 (1987), no. 2, 247–270. [9] Demailly, J.-P. A numerical criterion for very ample line bundles. J. Differential Geom. 37 (1993), no. 2, 323–374. [10] Demailly, J.-P. Analytic Methods in Algebraic Geometry, Higher Education Press, Surveys of Modern Mathematics, Vol. 1, 2010 [11] Demailly, J.-P. On the cohomology of pseudo-effective line bundles. arXiv:1401.5432. (v.2. Jan 2015) [12] Demailly, J.-P.; Peternell, T.; Schneider, M. Compact complex manifolds with numerically ef- fective tangent bundles. J. Algebraic Geom. 3 (1994), no. 2, 295–345. [13] Ein, L.; Lazarsfeld, R. Global generation of pluricanonical and adjoint linear series on smooth projective threefolds. J. Amer. Math. Soc. 6 (1993), no. 4, 875–903. [14] Fantechi, B.; Gottsche, L.; Illusie, L.; Kleiman, S.; Nitsure, N.; Vistoli, A. Fundamental al- gebraic geometry. Grothendieck's FGA explained. Mathematical Surveys and Monographs, 123. American Mathematical Society, Providence, RI, 2005. [15] Fujino, O. A transcendental approach to Kollar's injectivity theorem. Osaka J. Math. 49(2012), no. 3, 833–852. [16] Fujino, O. A transcendental approach to Kollar's injectivity theorem II. J. Reine Angew. Math. 681 (2013), 149–174. [17] Fujino, O.; Matsumura, S.-M. Injectivity theorem for pseudo-effective line bundles and its ap- plications. arXiv:1605.02284 [18] Fujino, O. Effective base-point-free theorem for semi-log canonical surfaces. arXiv:1509.07268 (v.3, Jan 2016) [19] Fujino, O.; Matsumura, S. Injectivity theorem for pseudo-effective line bundles and its applica- tions. arXiv:1605.02284. (v.1, May 2016) [20] Fujita, T. Contribution to Birational Geometry of Algebraic Varieties: Open Problems; the 23rd International Symposium, Division of Mathematics, the Taniguchi Foundation; August 22–27, 1988, Katata. 1988. 15 Xiaoyu Su and Xiaokui Yang Global generation and very ampleness for adjoint linear series [21] Hacon, C. On the log canonical inversion of adjunction. Proc. Edinb. Math. Soc. (2), 57(1):139– 143, 2014. [22] Hartshorne, Robin Algebraic geometry. Graduate Texts in Mathematics, No. 52. Springer-Verlag, New York-Heidelberg, 1977. xvi+496 pp. ISBN: 0-387-90244-9 [23] Heier, G. Effective freeness of adjoint line bundles. Doc. Math., 7, 31–42, 2002. [24] Griffiths, Phillip; Tu, Loring. Curvature properties of the Hodge bundles. Topics in transcenden- tal algebraic geometry (Princeton, N.J., 1981/1982), 29–49, Ann. of Math. Stud., 106, Princeton Univ. Press, Princeton, NJ, 1984. [25] Guan, Qi'an; Zhou, Xiangyu. A proof of Demailly's strong openness conjecture. Ann. of Math. (2) 182 (2015), no. 2, 605–616. [26] Guan, Qi'an; Zhou, Xiangyu. Strong openness conjecture and related problems for plurisubhar- monic functions. arXiv:1401.7158. (v.1 Jan. 2014) [27] Guan, Qi'an; Zhou, Xiangyu. Priviate communication. [28] Helmke, S. On Fujita's conjecture. Duke Math. J. 88 (1997), no. 2, 201–216. [29] Helmke, S. On global generation of adjoint linear systems. Math. Ann., 313(4):635–652, 1999. [30] Hindry, M.; Silverman, J. Diophantine geometry: An introduction, Graduate Texts in Mathe- matics, No. 201. Springer-Verlag, New York, 2000. [31] Kawamata, Y. On Fujita's freeness conjecture for 3-folds and 4-folds. Math. Ann. 308 (1997), no. 3, 491–505. [32] Keeler, Dennis S. Fujita's conjecture and Frobenius amplitude. Amer. J. Math. 130 (2008), no. 5, 1327–1336. [33] Koll´ar, J. Higher direct images of dualizing sheaves. I, Ann. of Math. (2) 123 (1986) 11–42. [34] Koll´ar, J. Effective base point freeness. Math. Ann., 296(4):595–605, 1993. [35] Lazarsfeld, R. Positivity in algebraic geometry. I, II. Springer-Verlag, Berlin, 2004. [36] Liu, Kefeng; Sun, Xiaofeng; Yang, Xiaokui. Positivity and vanishing theorems for ample vector bundles. J. Algebraic Geom. 22 (2013), no. 2, 303–331. [37] Liu, K.-F.; Yang, X.-K. Curvatures of direct image sheaves of vector bundles and applications I. J. Differential Geom. 98 (2014), 117–145. [38] Liu, K.-F.; Yang, X.-K. Effective vanishing theorems for ample and globally generated vector bundles. Comm. Anal. Geom. 23 (2015), no.4, 797–818. [39] Matsumura, S-M. A vanishing theorem of Kollar–Ohsawa type. arXiv:1511.04222. To appear in Math. Ann. [40] Mori, S. Threefolds whose canonical bundles are not numerically effective. Ann. of Math. (2) 116 (1982), no. 1, 133–176. [41] Mourougane, Ch.; Takayama, S. Hodge metrics and the curvature of higher direct images. Ann. Sci. ´Ec. Norm. Sup´er. (4) 41 (2008) no. 6, 905–924. [42] Nadel, Alan. Multiplier ideal sheaves and Kahler-Einstein metrics of positive scalar curvature. Ann. of Math. (2) 132 (1990), no. 3, 549–596. [43] Nakano, S. On complex analytic vector bundles, J. Math. Soc. Japan, 7 (1955), 1–12. [44] Pareschi, G.; Popa, M. M-regularity and the Fourier-Mukai transform. Pure Appl. Math. Q. 4 (2008), no. 3, part 2, 587–611. [45] Popa, M.; Schnell, C. On direct images of pluricanonical bundles. Algebra Number Theory 8 (2014), no. 9, 2273–2295. [46] Paufi, H. The Nakano vanishing theorem and a vanishing theorem of Demailly-Nadel type for holomorphic vector bundles. arXiv:1212.4417. (v.1, Dec. 2012) [47] Reider, Igor. Vector bundles of rank 2 and linear systems on algebraic surfaces. Ann. of Math. (2) 127 (1988), no. 2, 309–316. 16 Global generation and very ampleness for adjoint linear series Xiaoyu Su and Xiaokui Yang [48] Schwede, Karl. A canonical linear system associated to adjoint divisors in characteristic p > 0. J. Reine Angew. Math. 696 (2014), 69–87. [49] Schumacher, G. Positivity of relative canonical bundles and applications. Invent. Math. 190(2012), No. 1, 1–56. [50] Siu, Yum-Tong. Effective very ampleness. Invent. Math. 124 (1996), no. 1-3, 563–571. [51] Smith, Karen E. Fujita's freeness conjecture in terms of local cohomology. J. Algebraic Geom. 6 (1997), no. 3, 417–429. [52] Smith, Karen E. A tight closure proof of Fujita's freeness conjecture for very ample line bundles. Math. Ann. 317 (2000), no. 2, 285–293. [53] To, W.-K.; Yeung, S.-K. Finsler metrics and Kobayashi hyperbolicity of the moduli spaces of canonically polarized manifolds. Ann. of Math. (2) 181 (2015), no. 2, 547–586. [54] Viehweg, E. Weak positivity and the additivity of the Kodaira dimension for certain fibre spaces, In: Algebraic Varieties and Analytic Varieties, Advanced Studies in Pure Math.1 (1983) 329–353. [55] Ye, Fei; Zhu, Zhixian, Global generation of adjoint line bundles on projective 5-folds. Manuscripta Math. 153 (2017), no. 3-4, 545–562. [56] Ye, Fei; Zhu, Zhixian, On Fujita's freeness conjecture in dimension 5. arXiv:1511.09154. (v.1 Nov. 2015) Address of Xiaoyu Su: Academy of Mathematics and Systems Science, Chinese Academy of Sci- ences, Beijing, 100190, China. E-mail address: [email protected] Address of Xiaokui Yang: Morningside Center of Mathematics, Institute of Mathematics; Hua Loo-Keng Center for Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, 100190, China. E-mail address: [email protected] 17
1511.03259	6	1511	2017-09-01T12:54:24	A non-archimedean Ax-Lindemann theorem	[ "math.AG", "math.LO", "math.NT" ]	We prove a statement of Ax-Lindemann type for the uniformization of products of Mumford curves whose associated fundamental groups are non-abelian Schottky subgroups of $\mathop{\rm PGL}(2,\bar{\mathbf Q_p})$ contained in $\mathop{\rm PGL}(2,\bar{\mathbf Q})$. In particular, we characterize bi-algebraic irreducible subvarieties of the uniformization.	math.AG	math	A non-archimedean Ax-Lindemann theorem Antoine Chambert-Loir Univ. Paris Diderot, Sorbonne Paris Cité, Institut de Mathématiques de Jussieu-Paris Rive Gauche, UMR 7586, F-75013, Paris, France E-mail: [email protected] François Loeser Sorbonne Universités, UPMC Univ Paris 06, UMR 7586 CNRS, Institut Mathématique de Jussieu-Paris Rive Gauche, F-75005, Paris, France E-mail: [email protected] À Daniel Bertrand, en témoignage d'amitié 7 1 0 2 p e S 1 ] . G A h t a m [ 6 v 9 5 2 3 0 . 1 1 5 1 : v i X r a 1. Introduction theorem states classical Lindemann -- Weierstrass if alge- 1.1. -- The braic numbers α1, . . . , αn are Q-linearly independent, then their exponen- tials exp(α1), . . . , exp(αn) are algebraically independent over Q. More gen- erally, if α1, . . . , αn are any Q-linearly independent complex numbers, no longer assumed to be algebraic, Schanuel's conjecture predicts that the field Q(α1, . . . , αn, exp(α1), . . . , exp(αn)) has transcendence degree at least n over Q. In [2], Ax established power series and differential field versions of Schanuel's conjecture. In particular the part of Ax's results corresponding to the Linde- mann -- Weierstrass theorem can be recasted into geometrical terms as follows: that Theorem 1.2 (Exponential Ax-Lindemann). -- Let exp : Cn → (C×)n be the morphism (z1, . . . , zn) 7→ (exp(z1), . . . , exp(zn)). Let V be an irreducible algebraic subvariety of (C×)n and let W be an irreducible component of a maximal algebraic subvariety of exp−1(V ). Then W is geodesic, that is, W is defined by a finite family of equations of the form Pn i=1 aizi = b with a1, . . . , an ∈ Q and b ∈ C. In the breakthrough paper [26], Pila succeeded in providing an unconditional proof of the André-Oort conjecture for products of modular curves. One of his main ingredients was to prove a hyperbolic version of the above Ax-Lindemann theorem, which we now state in a simplified version. Let h denote the complex upper half-plane and j : h → C the elliptic modular function. By an algebraic subvariety of hn, we shall mean the trace in hn of an algebraic subvariety of Cn. An algebraic subvariety of hn is said to be geodesic if 2 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER it can be defined by equations of the form zi = ci and zk = gkℓzℓ, with ci ∈ C and gkℓ ∈ GL(2, Q)+. Theorem 1.3 (Hyperbolic Ax-Lindemann). -- Let j : hn → Cn be the mor- phism (z1, . . . , zn) 7→ (j(z1), . . . , j(zn)). Let V be an irreducible algebraic subvariety of Cn and let W be an irreducible component of a maximal algebraic subvariety of j −1(V ). Then W is geodesic. Pila's method to prove this Ax-Lindemann theorem is quite different from the differential approach of Ax. It follows a strategy initiated by Pila and Zannier in their new proof of the Manin-Mumford conjecture for abelian varieties [31]; that approach makes crucial use of the bound on the number of rational points of bounded height in the transcendental part of sets definable in an o-minimal structure obtained by Pila and Wilkie in [30]. Recently, still using the Pila and Zannier strategy, Klingler, Ullmo and Yafaev have succeeded in proving a very general form of the hyperbolic Ax-Lindemann theorem valid for any arithmetic variety ([20], see also [36] for the compact case). 1.4. -- In the recent paper [10], Cluckers, Comte and Loeser established a non- archimedean analogue of the Pila-Wilkie theorem of [30] in its block version of [25]. The purpose of this paper is to use this result to prove a version of Ax-Lindemann for products of algebraic curves admitting a non-archimedean uniformization and whose corresponding Schottky group is "arithmetic" and has rank at least 2 (theorem 2.7). In particular, this theorem applies for products of Shimura curves admitting a p-adic uniformization à la Cherednik-Drinfeld (see section 3). The basic strategy we use is strongly inspired by that of Pila [26] (see also [28]), though some new ideas are required in order to adapt it to the non-archimedean setting. Similarly as in Pila's approach one starts by working on some neighbor- hood of the boundary of our space (which, instead of a product of Poincaré upper half-planes, is a product of open subsets of the Berkovich projective line). Analytic continuation and monodromy arguments are replaced by more algebraic ones and explicit matrix computations by group theory considerations. We also take advan- tage of the fact that Schottky groups are free and of the geometric description of their fundamental domains. Compared with Pila's proof, where parabolic elements are used in a crucial way, one main difficulty of the nonarchimedean situation lies in the fact that all nontrivial elements of a Schottky groups are hyperbolic. To conclude, let us note that there are cases where p-adic analogues of theorems in transcendental number theory seem to require other methods than those used to prove their complex counterparts. For instance, it is still an open problem to prove a p-adic analogue, for values of the p-adic exponential function, of the classical Lindemann-Weierstrass theorem. Since his first works (see, for example, [5]), Daniel Bertrand has shown deep insight into p-adic transcendental number theory, and disseminated his vision within the mathematical community. We are pleased to dedicate this paper to him. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 3 Acknowledgements. -- The research leading to this paper was initiated during the 2014 MSRI program "Model Theory, Arithmetic Geometry and Number The- ory". We would like to thank the MSRI for its congenial atmosphere and hospital- ity. This research was partially supported by ANR-13-BS01-0006 (Valcomo) and by ANR-15-CE40-0008 (Défigéo). The second author was partially supported by the European Research Council under the European Community's Seventh Framework Programme (FP7/2007-2013)/ERC Grant Agreement nr. 246903 NMNAG and the Institut Universitaire de France. During the early stage of this project, the second author benefited from stimulat- ing discussions with Barry Mazur at MSRI, to whom we express our heartfelt thanks. We are also grateful to Daniel Bertrand for continuous support and encouragement. The comments of Yves André, Jean-François Boutot, Zoé Chatzidakis, Antoine Ducros, Florent Martin and Jonathan Pila, helped us to improve this paper, as well as the numerous remarks of a diligent referee; we thank them warmly. 2. Statement of the theorem 2.1. Non-archimedean analytic spaces. -- Given a complete non-archimedean valued field F , we shall consider in this paper F -analytic spaces in the sense of Berkovich [3, 4]. However, the statements, and essentially the proofs, can be carried on mutatis mutandis in the rigid analytic setting. In this context, there is a notion of irreducible component (see [17], or [13] for the rigid analytic version). If V is an algebraic variety over F , we denote by V an the corresponding F -analytic space. There is a canonical topological embedding of V (F ) in V an, and its image is closed if F is locally compact. If F ′ is a complete non-archimedean extension of F , we denote by XF ′ the F ′- analytic space deduced from an F -analytic space X by base change to F ′. 2.2. Schottky groups. -- Let p be a prime number; we denote by Cp the com- pletion of an algebraic closure of Qp and let F be a finite extension of Qp contained in Cp. The group PGL(2, F ) acts by homographies on the F -analytic projective line Pan 1 . In the next paragraphs, we recall from [19] a few definitions concerning Schottky groups in PGL(2, F ), their limit sets and the associated uniformizations of algebraic curves. One says that a discrete subgroup Γ of PGL(2, F ) is a Schottky group if it is finitely generated, and if no element (6= id) has finite order [19, I, (1.6)]. If Γ is a Schottky group, then Γ is free; moreover, any discrete finitely generated subgroup of PGL(2, F ) possesses a normal subgroup of finite index which is a Schottky group, [19, I, (3.1)]. 4 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER We say that Γ is arithmetic if its elements can be represented by matrices whose coefficients lie in a number field. In this case, it follows from the hypothesis that Γ is finitely generated that there exists a number field K ⊂ F such that Γ ⊂ PGL(2, K). 2.3. Limit sets. -- Let Γ be a Schottky subgroup of PGL(2, F ). Its limit set is the set LΓ of all points in P1(Cp) of the form limn(γn · x), where (γn) is a sequence of distinct elements of Γ and x ∈ P1(Cp), [19, I, (1.3)]. By [19, I, (1.6)], the limit set LΓ is a compact subset of P1(F ). If the rank of Γ is at least 2, then LΓ is a perfect (that is, closed and without isolated points) subset of P1(F ), see [19, I, (1.6.3) and (1.7.2)]. Let ΩΓ = (P1)an LΓ; it is a Γ-invariant open set of Pan 1 . By lemma 5.4 below, it is geometrically irreducible. 2.4. Quotients. -- Let us assume that Γ is a Schottky group and let g be its rank. From the explicit description of the action of the group Γ given by [19, I, §4] and recalled in §6.5 below, see also [3, p. 86], it follows that the group Γ acts freely on ΩΓ, and the quotient space ΩΓ/Γ admits a unique structure of an F -analytic space such that the projection pΓ : ΩΓ → ΩΓ/Γ is both a topological covering and a local isomorphism. Moreover, ΩΓ/Γ is the F -analytic space associated with a smooth, geometrically connected, projective F -curve XΓ of genus g ([19, III, (2.2)] and [3, theorem 4.4.1, p. 86]), canonically determined by the gaga theorem in this context, [3, theorem 3.4.12, p. 68]. 2.5. -- Let us now consider a finite family (Γi)16i6n of Schottky subgroups of PGL(2, F ) of rank > 2. Let us set Ω = Qn i=1 XΓi, and let p : Ω → X an be the morphism deduced from the morphisms pΓi : ΩΓi → X an Γi . i=1 ΩΓi and X = Qn 2.6. Flat subvarieties. -- Let K be a complete extension of F and let W be a closed analytic subspace of ΩK. The following terminology is borrowed from the analogous notions in the differ- ential geometry of hermitian symmetric domains. We say that W is irreducible algebraic if there exists a K-algebraic subvariety Y 1 )K such that W is an irreducible component of the analytic space ΩK ∩ Y an. 1 )K; it is irreducible of (Pn In this case, one can take for Y the Zariski closure of W in (Pn and satisfies dim(Y ) = dim(W ) (see [17], proposition 4.22). We say that W is flat if it can be defined by equations of the following form: (1) zi = c, for some i ∈ {1, . . . , n} and c ∈ ΩΓi(K); (2) zj = g · zi, for some pair (i, j) of distinct elements of {1, . . . , n} and g ∈ PGL(2, F ). Assume that W is flat and let Y be the subvariety of (Pn 1 )K defined by equations of this form which define W on ΩK. There exists a subset I of {1, . . . , n} such that the projection qI : Pn 1 given by the coordinates in I induces an isomorphism of Y to (PI 1)K. This implies that qI induces an isomorphism from W to Qi∈I Ωi,K. In particular, W is irreducible, even geometrically irreducible, hence is irreducible 1 → PI A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 5 algebraic. Conversely, we observe that if W is geometrically irreducible and if there exists a complete extension L of K such that WL is flat, then W is flat. We say that W is geodesic if, moreover, the elements g in (2) can be taken such that gΓig−1 and Γj are commensurable (i.e., their intersection has finite index in both of them). Here is the main result of this paper. Theorem 2.7 (Non-archimedean Ax-Lindemann theorem) Let F be a finite extension of Qp and let (Γi)16i6n be a finite family of arithmetic i=1 ΩΓi and i=1 XΓi, and let p : Ω → X an be the morphism deduced from the morphisms Schottky subgroups of PGL(2, F ) of ranks > 2. As above, let us set Ω = Qn X = Qn pΓi : ΩΓi → X an Γi . Let V be an irreducible algebraic subvariety of X and let W be an irreducible algebraic subvariety of Ω, maximal among those contained in p−1(V an). Then every irreducible component of WCp is flat. The proof of this theorem is given in section 8; it follows the strategy of Pila -- Zannier. In the archimedean setting, this strategy relies crucially on a theorem of Pila -- Wilkie about rational points on definable sets; we recall in section 4 the non-archimedean analogue of this theorem, due to Cluckers, Comte and Loeser (see [10]), which is used here. It is at this point that we need the assumption that the group Γ be arithmetic. This restriction is inherent to Pila -- Zannier's strategy and we do not know whether it can be bypassed. In section 6, we recall the reader a few more facts on p-adic Schottky groups and p-adic uniformization, essentially borrowed from the book [19]. In a final section, we prove a characterization (theorem 9.2) of geodesic subva- rieties of Ω as the geometrically irreducible algebraic subvarieties whose projection to X is algebraic ("bi-algebraic subvarieties"), in analogy with what happens in the context of Ax's theorem or of Shimura varieties. 3. The example of Shimura curves We begin by recalling the definition of Shimura curves and their p-adic uniformiza- tion. The litterature is unfortunately rather scattered; we refer to [6] for more detail, as well as to chapter 0 of [9]. 3.1. Complex Shimura curves. -- Let B be a quaternion division algebra with center Q; we assume that it is indefinite, namely B ⊗Q R ≃ M2(R). Let then OB be a maximal order of B, that is a maximal subring of B which is isomorphic to Z4 as a Z-module. Let H be the algebraic group of units of OB, modulo center, considered as a Z-group scheme. For every field R containing Q, one has H(R) = (B⊗QR)×/Z((B⊗QR)×); in particular, the group H(R) is isomorphic to PGL(2, R), 6 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER and we fix such an isomorphism. Then the group H(R) acts by homographies on the double Poincaré upper half-plane h± = C R. Let also ∆ be a congruence subgroup of H(Z); recall that this means that there exists an integer N > 1 such that ∆ contains the kernel of the canonical morphism H(Z) → H(Z/NZ). We assume that ∆ has been chosen small enough so that the stabilizer of every point of h± is trivial. The quotient h±/∆ has a natural structure of a compact Riemann surface and the projection p : h± → h±/∆ is an étale covering. This curve parameterizes triples (V, ι, ν), where V is a complex two dimensional abelian variety, ι : OB → End(V ) is a faithful action of OB on V and ν is a level structure "of type ∆" on V . When ∆ is the kernel of H(Z) to H(Z/NZ), for some integer N > 1, such a level structure corresponds to an equivariant isomorphism of VN , the subgroup of N-torsion of V , with OB/N OB. By [35], it admits a canonical structure of an algebraic curve S which can be defined over a number field E in C. 3.2. p-adic uniformization of Shimura curves. -- Let p be a prime number at which B ramifies, which means that B ⊗Q Qp is a division algebra. Let also F be the completion of the field E at a place dividing p; we denote by Cp the p-adic completion of an algebraic closure of F . We still denote by S the F -curve deduced from an E-model of the complex curve S. Let Ω = (P1)an F P1(Qp) be the extension of scalars to F of Drinfeld's upper half plane. According to the theorem of Cherednik-Drinfeld ([8, 16]; see also [6] for a detailed exposition), and up to replacing F by a finite unramified extension, the F -analytic curve San admits a "p-adic uniformization" which takes the form of a surjective analytic morphism j : Ω → San, which identifies San with the quotient of Ω by the action of a subgroup Γ of PGL(2, Qp). Up to replacing ∆ by a smaller congruence subgroup, which replaces S by a finite (possibly ramified) covering, we may also assume that Γ is a p-adic Schottky subgroup acting freely on Ω, and that j is topologically étale. Then the morphism j : Ω → San is the universal cover of San. Let us describe this subgroup. Let A be the quaternion division algebra over Q with the same invariants as B, except for those invariants at p and ∞ which are switched. In particular, A ⊗Q R is Hamilton's quaternion algebra, while A ⊗Q Qp ≃ M2(Qp). Let G be the algebraic group of units of A, modulo center; in particular, G(Qp) ≃ PGL(2, Qp). As explained in [6], the discrete subgroup Γ is the intersection of G(Q) with a compact open subgroup of G(Af), the adelic group associated with G where the place at ∞ is omitted. Lemma 3.3. -- The group Γ is conjugate to an arithmetic Schottky subgroup in PGL(2, Qp), its rank is at least 2, and its limit set is equal to P1(Qp). A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 7 Proof. -- The group Γ is a discrete subgroup of PGL(2, Qp) hence its limit set In other words, the Drinfeld upper half- LΓ is a Γ-invariant subset of P1(Qp). plane Ω = Pan LΓ. By the theory of 1 Mumford curves and Schottky groups, see [19], the analytic curve (Pan LΓ)/Γ is 1 algebraic, and admits the analytic curve San = Ω/Γ as an open subset. According to the Cherednik-Drinfeld theorem, the curve San is projective. This implies that Ω = Pan 1 P1(Qp) is an open subset of ΩΓ = Pan 1 LΓ, hence LΓ = P1(Qp). After base change to Qp, the algebraic Q-group G becomes isomorphic to PGL(2)Qp. Consequently, there exists a finite algebraic extension K of Q, con- tained in Qp, such that GK ≃ PGL(2)K. By such an isomorphism, G(Q) is mapped into PGL(2, K); this implies that the group Γ is conjugate to an arithmetic group. Since Γ is a Schottky group, it is free. Since it is non-abelian, its rank is at least 2. By this lemma, the following result is a special case of our main theorem 2.7. Theorem 3.4. -- Let F be a finite extension of Qp, let Ω = (P1)an F P1(Qp) and let j : Ωn → San be the Cherednik -- Drinfeld uniformization of a product of Shimura curves. Let V be an irreducible algebraic subvariety of S and let W ⊂ Ωn a maximal irreducible algebraic subvariety of j −1(V an). Then every irreducible component of WCp is flat. 3.5. -- By the same arguments, one can show that our main theorem 2.7 also applies to the uniformizations of Shimura curves associated with quaternion division algebras over totally real fields, as considered by Cherednik [8] and Boutot -- Zink [7]. 3.6. -- As suggested by J. Pila and explained to us by Y. André, theorem 3.4 can also be deduced from its complex analogue, which is a particular case of [36]. The crucial ingredient is a deep theorem of André ([1], III, 4.7.4) stating that the p-adic uniformization and the complex uniformization of Shimura curves satisfy the same non-linear differential equation. His proof relies on a delicate description of the Gauss-Manin equation in terms of convergent crystals and on the tempered funda- mental group introduced by him. From that point on, one can apply Seidenberg's embedding theorem [34] in differential algebra to prove that both the complex and non-archimedean Ax-Lindemann theorems are equivalent to a single statement in differential algebra, in the original spirit of Ax's paper [2]. 4. Definability -- A p-adic Pila-Wilkie theorem 4.1. -- There are two distinct notions of p-adic analytic geometry, one is "naïve", and the other one is rigid analytic. (Regarding rigid analytic geometry, we shall work in the framework defined by Berkovich.) These two notions give rise to three 8 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER classes of sets, and we shall use them all in this paper. Let F be a finite extension of Qp. a) Semialgebraic and subanalytic subsets of Qn p are defined by Denef and van den Dries in [14]; see also [10, p. 26]. Replacing Qp by a finite extension F , this leads to an analogous notion of F -semialgebraic, or F -subanalytic, subset of F n. Considering affine charts, one then defines F -semialgebraic or F -subanalytic subsets of V (F ), for every (quasi- projective, say) algebraic variety V defined over F . On the other hand, the Weil restriction functor assigns to V an algebraic vari- ety W defined over Qp together with a canonical identification V (F ) → W (Qp); we shall say that a subset of V (F ) is Qp-semialgebraic, resp. Qp-subanalytic, if its image in W (Qp) is Qp-semi-algebraic, resp. Qp-subanalytic. Observe that F -semi- algebraic subsets of V (F ) are Qp-semi-algebraic, and that F -subanalytic subsets of V (F ) are Qp-subanalytic. Recall that a F -subanalytic subset S is said to be smooth of dimension d at a point x if it possesses a neighborhood U which is isomorphic to the unit ball of F d; then S is smooth of dimension d at every point of U. b) In [21], Lipshitz defined a notion of rigid subanalytic subset of Cn p . We shall use in this paper the variant ([22], definition 2.1.1) where the coefficients of all polynomials and power series involved belong to F ; we will call them rigid F - subanalytic. The notion extends to subsets of V (Cp), where V is an algebraic variety defined over F . These classes of sets are stable under boolean operations and projections (corol- lary 4.3 of [23]), admit cell decompositions (theorem 7.4 of [11]), a natural notion of dimension (in fact, they are b-minimal in the sense of [12]), as well as a natural notion of smoothness. Lemma 4.2. -- Let F be a finite extension of Qp contained in Cp and let V be an algebraic variety over F . Let Z be a rigid F -subanalytic subset of V (Cp). Then Z(F ) = Z ∩ V (F ) is an F -subanalytic subset of V (F ). Proof. -- We may assume that V = An. Then Z can be defined by a quantifier- free formula of the above-mentioned variant of Lipshitz's analytic language, and our claim follows from the very definition of this language. 4.3. -- A block in Qn p is either empty, or a singleton, or a smooth subanalytic subset of pure dimension d > 0 which is contained in a smooth semialgebraic subset of dimension d. p × Qt A family of blocks in Qn p × Qs p is a subanalytic subset W such that there exists an p, there p such that the fibers Wσ and Zτ are smooth of the same dimension, integer t > 0 and a semialgebraic set Z ⊂ Qn exists τ ∈ Qt and Wσ ⊂ Zτ . (In particular, the sets Wσ, for σ ∈ Qs p such that for every σ ∈ Qs p, are blocks in Qn p .) Let F be a finite extension of Qp. Considering Weil restriction, we deduce from these notions the definition of a block in F n, or of a family of blocks in F n × Qt p. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 9 4.4. -- Let H be the standard height function on Q; for x ∈ Q, written as a fraction a/b in lowest terms, one has H(x) = max(a , b). We also write H for the height function on Qn defined by H(x1, . . . , xn) = maxi(H(xi)). Viewing GL(d, Q) as a subspace of Qd2 , it defines a height function on GL(d, Q). There exists a strictly positive real number c such that H(gg′) 6 cH(g)H(g′) for every g, g′ ∈ GL(d, Q), and H(g−1) ≪ H(g)c for every g ∈ GL(d, Q). When d = 2 and g ∈ SL(2, Q), one even has H(g−1) = H(g). Let g ∈ GL(d, Q). If g is diagonal, then H(gn) = H(g)n for every n ∈ Z. More generally, if g is semisimple, we have upper and lower bounds H(g)n ≪ H(gn) ≪ H(g)n, for every n ∈ Z. By abuse of language, if G is a linear algebraic Q-group, we implicitly choose an embedding in some linear group, which furnishes a height function H on G(Q). The actual choice of this height function depends on the chosen embedding, but any other height function H ′ is equivalent, in the sense that there is a strictly positive real number c such that H(x)1/c ≪ H ′(x) ≪ H(x)c for every x ∈ G(Q). 4.5. -- Let Z be a subset of F n and let K be finite extension of Q contained in F . We write Z(K) = Z ∩ K n (K-rational points of Z). For every real number T , we define Z(K; T ) = {x ∈ Z(K) ; H(x) 6 T }; for every integer D, we also define Z(D; T ) to be the set of points x ∈ Z(F ) such that [Q(xi) : Q] 6 D for every i ∈ {1, . . . , n} and H(x) 6 T . These are finite sets. We say that Z has many K-rational points if there exist strictly positive real numbers c, α such that Card (cid:16)Z(K; T )(cid:17) > cT α for all T large enough. This notion only depends on the equivalence class of the height. 4.6. -- In [10], Cluckers, Comte and Loeser established a p-adic analogue of a theorem of Pila-Wilkie [30] concerning the rational points of a definable set. We will use the following variant of [10, Theorem 4.2.3]. Theorem 4.7. -- Let F be a finite extension of Qp and let K be a finite extension of Q, contained in F . Let Z ⊂ F n be a Qp-subanalytic subset. Let ε > 0. There exist s ∈ N, c ∈ R and a family of blocks W ⊂ Z × Qs p satisfying the following property: for every T > 1, there exists a subset ST ⊂ Qs p of cardinality < cT ε such that Z(K; T ) ⊂ Sσ∈ST Wσ. Proof. -- Let d = [F : Qp]. By Krasner's lemma, there exists an algebraic num- ber e ∈ F of degree d such that F = Qp(e). Then the basis (1, e, . . . , ed−1) defines a Qp-linear bijection ψ : Qd p be its p inverse. ∼−→ F , (x1, . . . , xd) 7→ P xiei−1. Let ϕ : F ≃ Qd By construction, if K is a number field contained in F and x ∈ K d, then ψ(x) ∈ K(e); in particular, [Q(ψ(x)) : Q] 6 d[Q(x) : Q]. Conversely, if x ∈ K, then the coordinates of ϕ(x) in Qd p belong to the Galois closure K(e)′ of the compositum K · 10 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Q(e) hence are algebraic numbers, of degrees 6 D = [K(e)′ : Q]. In other words, ϕ and ψ induce bijections at the level of algebraic points. Since these maps are linear, there exists a positive real number a > 0 such that a−1H(x) 6 H(ϕ(x)) 6 aH(x) for every x ∈ K. We deduce from ϕ a Qp-linear isomorphism ϕ : F n → Qnd p . In particular, Z ′ = ϕ(Z) is a subanalytic subset of Qnd p . The morphism ϕ maps algebraic points of given degree to algebraic points of uniformly bounded degree, and there exists a positive real number a > 0 such that a−1H(x) 6 H(ϕ(x)) 6 aH(x) for every x ∈ Z(K). The definition of a family of blocks that we have adopted here is slightly stronger than the one used in Theorem 4.2.3 of [10]. However, all proofs go over without any modification, so that there exists a family of blocks W ′ ⊂ Z ′ × Qs p such that p of cardinality < cT ε such that for any T > 1, there exists a subset ST ⊂ Qs Z ′(D; T ) ⊂ Sσ∈ST W ′ p be the map (x, y) 7→ (ϕ(x), y) and let W = ψ−1(W ′) ⊂ F n × Qs p. By definition, W is a family of blocks in Z. Moreover, for any T > 1, one has σ. Let ψ : F n × Qs p → Qnd p × Qs Z(F ; T ) ⊂ ψ−1(Z ′(D; aT )) ⊂ [ σ∈SaT ϕ−1(W ′ σ) = [ σ∈SaT Wσ. Since Card(SaT ) 6 caεT ε, the family of blocks W satisfies the requirements of the theorem. 5. Zariski closures and analytic functions 5.1. -- Let F be a complete non-archimedean valued field. Let V be an F -scheme of finite type. One says that a subset K of V an is sparse if there exist a set T and a subset Z of V an × T such that for every t ∈ T , Zt = {x ∈ V an ; (x, t) ∈ Z} is a Zariski-closed subset of V an with empty interior, and K = St∈T Zt. Lemma 5.2. -- A sparse set has empty interior. Proof. -- Let us say that a point x ∈ V an is maximally Abhyankar if the rational rank of the value group of H (x) is equal to dimx(V an). If V is irreducible, then maximally Abhyankar points are dense in V an; moreover, each of them is Zariski dense. Let K be a sparse set in V an; write K = St Zt as above. Let us argue by contradiction and let U be a non-empty subset of V an contained in K. By what precedes, there exists a maximally Abhyankar point x ∈ U. Let t ∈ T be such that x ∈ Zt. Then Zt contains the Zariski closure of x in V an, so that Zt contains an irreducible component of V an, contradicting the definition of a sparse set. Lemma 5.3. -- Let F ′ be an algebraically closed complete extension of F , let F ′ → V an be the base change morphism. Let K be a closed sparse subset of V an q : V an and let K ′ = q−1(K). Then K ′ is sparse. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 11 F ′ shows that K ′ is sparse as well. Proof. -- Indeed, if K = St∈T Z an St∈T (Zt)an Lemma 5.4. -- Let us assume that K is sparse, and let C ⊂ V be a geometrically irreducible curve such that C an 6⊂ K. Then C an K is connected. is a description of the sparse set K, then K ′ = t Proof. -- Using lemma 5.3, we reduce to the case where F is algebraically closed; we may moreover assume that C is reduced. Let K = St∈T Z an t be a description of K as above. Up to adding the singular locus of C to K, we may assume that C is smooth. By assumption, for every t ∈ T , C 6⊂ Z an t ∩ C an consists of rigid points of C an, hence K ∩ C an consists of rigid points of C an. In the topological description of smooth geometrically irreducible analytic curves as real graphs ([3], chapter 4), their rigid points are endpoints, so that C an (K ∩ C an) is connected as well. ; consequently, Z an t Proposition 5.5. -- Let F be a complete non-archimedean valued field. Let V be an F -scheme of finite type which is geometrically connected (resp. geometrically ir- reducible) and let K be a closed sparse subset of V an. Then V an K is a geometrically connected (resp. geometrically irreducible) analytic space. The particular case K = ∅ implies the "gaga"-type consequence that if V is geometrically connected (resp. geometrically irreducible), then so is V an. Proof. -- Using lemma 5.3, we reduce to the case where F is algebraically closed. By assumption, V is connected. Let us prove that V an K is connected. Let x, y ∈ V an K. Let F ′ an algebraically closed complete valued field containing F ′ → V an be the both H (x) and H (y), and view x, y as elements of V (F ′); let q : V an base change morphism and let K ′ = q−1(K); by lemma 5.3, this is a sparse subset of V an F ′ . By [24, p. 56], there exists an irreducible curve C ⊂ VF ′ which passes through x and y. Then C an is connected. One has C 6⊂ K ′, by definition of K ′; it follows from lemma 5.4 that C an (K ′ ∩ C an) is connected. Consequently, x and y belong to the same component of V an F ′ K ′, hence their images in V an K belong to the same connected component. This proves that V an K is connected. Let us now assume that V is geometrically irreducible. The normalization mor- phism p : W → V is finite, and W is geometrically connected. Since p−1(K) is a sparse subset of W an, it follows from the first part of the lemma that W an p−1(K) is geometrically connected. Since W an is the normalization of V an ([18], lemma 2.7.15), then W an p−1(K) = p−1(V an K) is the normalization of V an K. By theorem 5.17 of [17], this implies that V an K is geometrically irreducible. Corol lary 5.6. -- Let F be a complete valued field, let V be an F -scheme of finite type and let K be a closed sparse subset of V an. The set of irreducible compo- nents of V an K is finite. If V is equidimensional, then each of them has dimen- sion dim(V ). Proof. -- We may assume that V is irreducible. Let Ω = V an K. Let E be the completion of an algebraic closure of F . By proposition 5.5, ΩE ∩ Z an is irreducible, 12 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER for every irreducible component Z of VE, and the family of these intersections is the family of irreducible components of ΩE. The finiteness statement then follows from [17, lemme 4.25], while the one about dimension follows from proposition 4.22 of [17]. Corol lary 5.7. -- Let F be a complete valued field, let V be an irreducible F - scheme of finite type and let K be a closed sparse subset of V an. Let W be an irreducible component of V an K. If W is geometrically irreducible, then V is geometrically irreducible as well, one has W = V an K, and W is topologically dense in V an. j j E (lemma 5.3). Consequently, Lj = V an , for every j. By proposition 5.5, Wj = V an Proof. -- Let E be a complete algebraically closed extension of F , let V1, . . . , Vn be the irreducible components of VE; let L be the preimage of K in VE; it is a closed sparse subset of V an j ∩ L is a closed sparse subset of V an Lj is geomet- rically irreducible. The automorphism group Aut(E/F ) acts transitively on the set {V1, . . . , Vn} of irreducible components of VE, hence on the set {W1, . . . , Wn} of irreducible components of V an L. Since VE is geometrically irreducible, there E exists an index j such that WE = Wj; then Aut(E/F ) fixes Wj, so that n = 1 and j = 1. This proves that V is geometrically irreducible. By proposition 5.5, one has W = V an K. By lemma 5.2, W is topologically dense in V an. Proposition 5.8. -- Let F be a finite extension of Qp. Let A be an affine scheme of finite type over F and let Ω ⊂ Aan be the complement of a closed sparse subset. Let X be a closed analytic subspace of Ω. Let V be a Qp-semi-algebraic subset of A(F ), contained in X(F ), and let W be its Zariski closure in A. One has W an ∩ Ω ⊂ X. Proof. -- The following proof is inspired by that of lemma 4.1 in [29]. We argue by noetherian induction on W , assuming that if W ′ is the Zariski closure of a Qp-semi-algebraic subset V ′ of A(F ) contained in X(F ), and if W ′ ( W , then (W ′)an ∩ Ω ⊂ X. First assume that W is not irreducible. Then any irreducible component W ′ of W is the Zariski closure in A of V ∩ W ′(F ), a Qp-semi-algebraic subset of A(F ); by induction, (W ′)an ∩ Ω ⊂ X, so that W an ∩ Ω ⊂ X. Let K = Aan Ω; by assumption, K is closed and sparse; let K = S San We may thus assume that W is irreducible; since its subset W (F ) of F -rational points contains V , it is Zariski-dense in W , so that W is geometrically irreducible. t be a presentation of K, where for every t, St is a Zariski-closed subset with empty interior of A. Since W is irreducible and not contained in St, W ∩St is a strict Zariski-closed subset of W . Consequently, W an ∩ K is a sparse subset of W an. By proposition 5.5, W an ∩ Ω is thus a geometrically irreducible analytic space. Let R be the Weil restriction functor from F to Qp. By definition, A(F ) is iden- tified with R(A)(Qp) and we shall write R(V ) for the image of V inside R(A)(Qp). Let then Z be the Zariski closure of R(V ) inside R(A). A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 13 Let Z ′ be an irreducible component of Z. Then Z ′ ∩ R(V ) is a semi-algebraic subset of R(A), of the form R(V ′), for a unique Qp-semi-algebraic subset V ′ of V . When Z ′ varies, the corresponding subsets V ′ cover V ; we may thus choose Z ′ such that V ′ is Zariski dense in W . Replacing V by V ′, we may assume that Z is irreducible; then it is geometrically irreducible, because its set of Qp-points is Zariski dense. Since V is Qp-semialgebraic, the subset R(V ) of R(A)(Qp) is semialgebraic, hence the dimension of Z coincides with the dimension of V as a Qp-semialgebraic subset of A(F ). Consequently, dimZar(Z) = dim(Z(Qp)) = dim(R(V )). Since W is a Zariski closed subset of A containing V , the subscheme R(W ) is Zariski closed in R(A) and contains R(V ), so that Z ⊂ R(W ). By Weil restriction, the inclusion Z → R(W ) corresponds to a morphism g : ZF → W . Let x ∈ A(F ) and let x ∈ R(A)(Qp) be the corresponding point; if x ∈ V , then x ∈ R(V ) ⊂ Z(Qp), hence x ∈ ZF (F ). By the definition of the Weil restriction functor, one has g(x) = x. In particular, the image of ZF (F ) under g contains V , hence g is dominant, by definition of W . The morphism g induces an analytic morphism gan : Z an F → W an ⊂ Aan. The F ; since F is geometrically irreducible, corollary 5.6 implies that (gan)−1(W an ∩ Ω) is ge- F ). Let Y = (gan)−1(W an ∩ X); it is a inverse image of W an ∩ Ω is the complement of a closed sparse subset of Z an Z an ometrically irreducible, of dimension dim(Z an Zariski closed analytic subset of (gan)−1(W an ∩ Ω). Let us admit for a moment that dim(Y ) = dim(ZF ) and let us conclude that W an ∩ Ω ⊂ X. Since dim(Z an F ) = dim(ZF ) = dim((gan)−1(W an ∩ Ω)), we see that Y = (gan)−1(W an ∩ X) = (gan)−1(W an ∩ Ω). The morphism g : ZF → W being dominant, its image contains a non-empty open subset W ′ of W . Since W is geometrically irreducible, (W ′)an is dense in W an; in particular, the image of gan meets any non-empty open subset of W an. Since (gan)−1(W an ∩ (Ω X)) is empty, by the preceding equality, this implies that W an ∩ (Ω X) is empty, hence W an ∩ Ω = W an ∩ X. It remains to prove the equality dim(Y ) = dim(ZF ). Let us consider a semi-algebraic cell decomposition of R(A)(Qp) which is adapted to R(V ), Z(Qp), Zsing(Qp), and to their singular loci: a finite partition of R(A)(Qp) into "open cells" such that these Qp-semi-algebraic subsets are unions of cells ([15], see also [12]). Let C be a cell of dimension dim(R(V )) which is contained in R(V ). Since dim(Zsing(Qp)) 6 dim(Zsing) < dim(Z) = dim(R(V )), the cell C is disjoint from Zsing(Qp). By definition of a cellular decomposition, C is open in R(V ) and in (Z Zsing)(Qp). Let C be the subset of V corresponding to C. Since the identification of C with C provided by the Weil restriction functor is a homeomorphism which respects the singular loci, C is an open subset of V . Let x be a point of C and let x be the corresponding point of C. By what precedes, R(V ), Z(Qp) and Z are smooth at x, so that Tx(R(V )) = Tx(Z(Qp)) = 14 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Tx(Z). In particular, these three Qp-vector spaces have the same dimension, equal to dim(Tx(V )) = dim(V ). Since g(x) = x ∈ X, one has x ∈ Y ; more generally, C ⊂ Y . The tangent space Tx(Y ) of Y at x is an F -vector subspace of Tx(ZF ) = (Tx(Z))F which contains Tx( C) = Tx(Z). Consequently, Tx(Y ) = Tx(ZF ). This implies that the analytic space Y has dimension dim(ZF ), and concludes the proof. 6. Complements on p-adic Schottky groups and uniformization Let F be a finite extension of Qp. Unless precised otherwise, analytic spaces are F -analytic spaces. 6.1. -- Let a ∈ F and r ∈ R>0; as usual, we let B(a, r) and E(a, r) be the subsets of (A1)an of points x such that T (x) − a < r and T (x) − a 6 r respectively. The subspace B(a, r) is called a bounded open disk; we say that E(a, r) is the corresponding bounded closed disk. If B is a bounded open disk, we write B+ for the corresponding bounded closed disk. We will say that such a disk is strict if its radius r belongs to F ×Q. To these disks, we also add the unbounded open disks Pan 1 E(a, r) and the B(a, r). An unbounded disk is said to be strict if its unbounded closed disks Pan 1 complementary disk is strict. The image by an homography γ ∈ PGL(2, F ) of an open (resp. closed, resp. strict) disk is again an open (resp. closed, resp. strict) disk. 6.2. -- We endow P1(Cp) with the distance given by δ(x, y) = x − y max(1, x) max(1, y) for x, y ∈ Cp -- it is invariant under the action of PGL(2, OCp). Moreover, an elementary calculation shows that every element g ∈ PGL(2, Cp) is Lipschitz for this distance (see also Thm 1.1.1 of [33]). 6.3. -- Let Γ be a Schottky group in PGL(2, F ), let LΓ ⊂ P1(F ) be its limit set and ΩΓ = Pan LΓ. For any rigid point x ∈ ΩΓ, let δΓ(x) be the δ-distance of x 1 to LΓ. For every γ ∈ PGL(2, F ), there exists a real number c > 1 such that c−1δΓ(z) 6 δΓ(γ · z) 6 cδΓ(z) for every rigid point z ∈ ΩΓ. Lemma 6.4. -- Let G be a compact subset of ΩΓ. There exists a strictly positive real number c such that δΓ(x) > c for every rigid point x ∈ G. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 15 Proof. -- Arguing by contradiction, we assume that there exists a sequence (xn) of rigid points of G such that δΓ(xn) → 0. For every n, let ξn ∈ LΓ such that δΓ(xn) = δ(xn, ξn); it exists since LΓ is compact. Extracting a subsequence if necessary, we assume that the sequence (ξn) converges to a point ξ of LΓ. Then δ(xn, ξ) → 0. This implies that the sequence (xn) converges to ξ in the Berkovich space Pan 1 . Since G is compact, one has ξ ∈ G, a contradiction. 6.5. -- Let Γ be a Schottky subgroup of PGL(2, F ). Let us assume that the point at infinity ∞ does not belong to its limit set LΓ. Then, by [19, I, (4.3)], the group Γ admits a basis (γ1, . . . , γg) and a good fundamental domain FΓ with respect to this basis, in the following sense: (1) There exists a finite family (B1, . . . , Bg, C1, . . . , Cg) of strict bounded open 1 , . . . , B+ g , C + 1 , . . . , C + g are pairwise Bi) = C + i and γi(Pan 1 B+ i ) = Ci for disks in Pan (S Bi ∪ S Ci); (2) The corresponding bounded closed disks B+ 1 such that FΓ = Pan 1 disjoint; Let then F◦ (3) The elements γ1, . . . , γg satisfy γi(Pan 1 i (cid:17); i ∪ S C + (cid:16)S B+ Γ = Pan 1 every i ∈ {1, . . . , g}. With this notation, let W = Pan 1 taining F, stable under each γi. γi(Pan 1 Bi) = C + S Bi; this is an affinoid domain of Pan con- Bi, hence γiW ⊂ Indeed, one has W ⊂ Pan 1 1 i , hence the claim since C + j is disjoint from each Bi. Moreover, the following properties are satisfied: (4) One has Sγ∈Γ γ · FΓ = P1 LΓ; (5) For γ ∈ Γ, one has FΓ ∩ γ · FΓ 6= ∅ if and only if γ ∈ {id, γ±1 (6) For every γ ∈ Γ {id}, one has F◦ In this context, we identify an element γ of Γ with a reduced word in the letters 1 , . . . , γ± For every γ ∈ Γ {id}, [19, I, §4, p. 29] define a bounded open disk B(γ), equal i ), according to whether the last letter ; in any case, one has γ · ∞ ∈ B(γ). g } and denote its length by ℓΓ(γ). either to γ · (Pan 1 of the reduced word representing γ is γi or γ−1 Moreover, they prove: i ) or to γ · (Pan 1 Γ ∩ γ · FΓ = ∅. 1 , . . . , γ±1 g }; {γ± B+ C + i (7) B(γ ′) ⊂ B(γ) if and only if γ is an initial subword of γ ′; (8) For every integer n, one has Pan 1 [ ℓΓ(γ)<n γ · F = [ ℓΓ(γ)=n B(γ); (9) There exists a real number c > 1 such that for every γ, the radius of the disk B(γ) is ≪ c−ℓΓ(γ); (10) The intersection of every decreasing sequence of open disks (B(γn)), where ℓΓ(γn) = n, is reduced to a limit point of Γ, and every limit point can be obtained in this way. 16 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Proposition 6.6. -- Let Γ be a Schottky group in PGL(2, F ) and let G be a com- pact analytic domain of ΩΓ. There exist positive real numbers a, b such that for every γ ∈ Γ and every rigid point x ∈ γ · G, one has ℓΓ(γ) 6 a − b log(δΓ(x)). Proof. -- To prove this proposition, we may extend the scalars to a finite extension of F and henceforth assume that the limit set LΓ is not equal to P1(F ). Placing a point of P1(F ) LΓ at infinity, §6.5 furnishes a basis (γ1, . . . , γg) and a good funda- mental domain F with respect to this basis of the form F = Pan i=1 Ci). 1 Let b and c > 1 be positive real numbers such that the diameter of B(γ) is bounded by bc−ℓΓ(γ), for every γ ∈ Γ {id}. i=1 Bi ∪ Sg (Sg Let x ∈ ΩΓ and let γ ∈ Γ be such that x ∈ γ · F. Let ξ ∈ LΓ(x) such that δΓ(x) = δ(x, ξ). Since the disk B(γ) contains both x and ξ, one has δΓ(x) 6 bc−ℓΓ(γ), that is, ℓΓ(γ) 6 1 log(c) (− log(δΓ(x)) + log(b)) , since log(c) > 0. This proves the proposition in the particular case where G = F. Let us now prove the general case. Let a be a real number such that δγ(x) > a > 0 for every rigid point of G (lemma 6.4). The preceding inequality shows that there exists a finite subset S of Γ such that G meets γ · F if and only if γ ∈ S. It then follows from property (8) that G is contained in the finite union Ss∈S s · F. To conclude the proof, we observe that if x ∈ γ · G, then there exists s ∈ S such that x ∈ γs · F. The proposition then follows from the particular case already treated and from the inequality ℓΓ(γ) 6 ℓΓ(γs) + ℓΓ(s). Corol lary 6.7. -- Let G and G′ be compact analytic domains of ΩΓ. The set of γ ∈ Γ such that γ · G ∩ G′ 6= ∅ is finite. Proof. -- Let S be this set. For γ ∈ S, the intersection γ · G ∩ G′ is a non-empty affinoid domain of Pan 1 , hence it contains a rigid point xγ. With a and b as in the statement of proposition 6.6, one has ℓΓ(γ) 6 a − b log(δΓ(xγ)). Since xγ ∈ G′, δΓ(xγ) is bounded from below, by lemma 6.4. This shows that ℓΓ(γ) is bounded above, when γ runs over S. Proposition 6.8. -- Let Γ be a Schottky group in PGL(2, F ) and let g be its rank. Let ξ ∈ LΓ and let U be an open neighborhood of ξ in Pan 1 . There exist an open neighborhood U ′ of ξ, contained in U, a basis γ1, . . . , γg of Γ, an affinoid domain F ⊂ ΩΓ such that the following properties hold: (1) One has F ⊂ U ′; (2) For every i, one has γi(U ′) ⊂ U ′; (3) One has Sγ∈Γ γF = ΩΓ. Such an affinoid domain will called a fundamental set. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 17 (Sg i=1 Bi ∪ Sg Proof. -- We first treat the case where LΓ 6= P1(F ). Placing a point of P1(F ) LΓ at infinity, §6.5 furnishes a basis (γ1, . . . , γg) and a good fundamental domain F with respect to this basis of the form F = Pan 1 i=1 Ci). By (10), for every integer n > 1, there is an element γ ∈ Γ of length n such that ξ ∈ B(γ); if n is large enough, one has B(γ)+ ⊂ U, because the diameter of B(γ)+ tends to 0 when n = ℓΓ(γ) tends to ∞. Since γ · F ⊂ B(γ)+, this implies that γ · F ⊂ U. Up to changing the basis (γ1, . . . , γg) into (γ−1 g ), and exchanging Bi and Ci for every i, we may assume that the last letter of γ is γs, for some s ∈ Sg {1, . . . , g}. Set W = Pan i=1 Bi; recall that W is an affinoid domain of Pan 1 1 containing F and stable under γ1, . . . , γg. By definition, one has 1 , . . . , γ−1 B(γ)+ = γ · (Pan 1 Bs) ⊃ γ · W, since W ⊂ Pan 1 Bs. Let us now set F′ = γ · F, W ′ = γ · W and γ ′ construction, F′ and W ′ are affinoid domains of Pan U, the translates of F′ under Γ cover ΩΓ, and W ′ is stable under the basis (γ ′ of Γ. i = γγiγ−1 for i ∈ {1, . . . , g}. By 1 such that F′ ⊂ W ′ ⊂ B(γ)+ ⊂ 1, . . . , γ ′ g) This almost proves (1 -- 3), except that W ′ is affinoid and not open. To conclude the construction, one sets U ′ to be the interior of W ′, and one redoes the construction starting from U ′ instead of U. The second paragraph of the proof shows that there exists γ ′ ∈ Γ such that γ ′ · F′ is contained in U ′. The affinoid affinoid γ ′ · F′, the open subset U ′ and the basis (γ ′ g) satisfy the requirements of the proposition. 1, . . . , γ ′ Let us now treat the case where LΓ = P1(F ). Let F ′ be a finite extension of F of degree > 1. The preceding construction can be applied starting with a point of P1(F ′) LΓ and furnishes an open neighborhood V ′ of ξ in (Pan 1 )F ′, contained in UF ′, a basis (γ1, . . . , γg) of Γ and an affinoid domain F′ of ΩΓ,F ′ satisfying properties (1 -- 3). The images U ′ of V ′ and F of F′ by the projection (Pan 1 satisfy the required properties. 1 )F ′ → Pan Lemma 6.9. -- Let Γ be an arithmetic Schottky group in PGL(2, F ) and let H be a height function on PGL(2, Q). There exists a positive real number c such that H(γ) 6 cℓΓ(γ)+1, for every γ ∈ Γ. Proof. -- Let (γ1, . . . , γg) be a basis of Γ as above. Let c1 be a positive real number such that H(hh′) 6 c1H(h)H(h′) for every h, h′ ∈ PGL(2, Q). Let c = c1 sup(H(id), H(γ1), . . . , H(γg)). One proves by induction on ℓΓ(γ) that c1H(γ) 6 sup(c1H(γ± 1 ), . . . , c1H(γ± g ))ℓΓ(γ)c1H(id) 6 c1cℓΓ(γ)+1 for every γ ∈ Γ, as was to be shown. Lemma 6.10. -- Let Γ be a Schottky subgroup of PGL(2, F ) and let ∆ be a subset of P1(F ) of cardinality 2. Let K be a number field contained in F . The stabilizer of ∆ inside Γ does not have many K-rational points. 18 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Proof. -- Let S be this stabilizer; we may assume that S 6= {id}. Let g ∈ S {id}. Then g is hyperbolic (see [19], page 7, line 2), hence has exactly two rational fixed points in P1(F ). Up to a change of projective coordinates, we may thus assume that ∆ = {0, ∞}. Then every element h of S is of the form z 7→ λ(h)z, for some unique element λ(h) ∈ K ×; moreover, unless h = id, any such h is hyperbolic hence is represented by a matrix having two eigenvalues with distinct absolute values, so that λ(h) 6= 1. Let us choose h ∈ S {id} such that λ(h) is > 1 and minimal. By euclidean division, one has S = hhi. Then S ∩ PGL(2, K) is generated by an element of the form ha, for some a ∈ Z. Since ha is semisimple, we have H(ha)n ≪ H(han) ≪ H(ha)n, for every n ∈ Z (see §4.4). This shows that S ∩ PGL(2, K) does not have many rational points. In §8, we will need the following lemma. Lemma 6.11. -- Let r be a positive real number and let f ∈ Cp[[z]] be a power series which converges on the closed disk E(0, r). Let L1 and L2 be closed subsets of Cp such that f −1(L2) ⊂ L1; for every x ∈ Cp, let δ(x; L1) and δ(x; L2) be the distances of x to L1 and L2 respectively. Then there exists real numbers m > 0, c > 0, and s such that 0 < s < r and such that δ(f (x); L2) > cδ(x; L1)m for every x ∈ E(0, s). Proof. -- Write f = P cnzn. We may assume that there exists a ∈ C× p such that r = a; composing f with homographies which map E(0, r) to E(0, 1) and f (E(0, r)) into the disk E(0, 1), we assume that r = 1 and that cn 6 1 for all n. (Recall from §6.2 that homographies are Lipschitz for the distance δ.) Let us first treat the case where f (0) 6∈ L2. Then there exists a real number s > 0 such that E(f (0), s) ∩ L2 = ∅. For every x ∈ E(0, 1) such that x < s, one has f (x) − f (0) < s, hence δ(f (x); L2) > s. It suffices to set m = 0 and c = s. We now assume that f (0) ∈ L2, hence 0 ∈ L1. Let m = ord0(f − f (0)). Since f ′(z) = Pn>m ncnzn−1, there exists a real number s such that 0 < s 6 1 and such that f ′(z) = mcm zm−1 provided z 6 s. Moreover, (cid:12)(cid:12)(cid:12) 6 1 for every n > 0 and any z ∈ E(0, 1). Considering the Taylor expansion f (n)(z)/n!(cid:12)(cid:12)(cid:12) f (y) = X n>0 1 n! f (n)(x)(y − x)n, we then see that there exists a real number s′ such that f (E(x, u)) = E(f (x), f ′(x) u) for every real number u such that 0 < u 6 s′ and every x ∈ E(0, 1) such that 0 < x 6 s. If u < δ(x; L1), then E(x, u)∩L1 = ∅, hence E(f (x), f ′(x) u)∩L2 = ∅; consequently, δ(f (x); L2) > f ′(x) δ(x; L1). Since 0 ∈ L1, one has x > δ(x; L1). Consequently, δ(f (x); L2) > mcm xm−1 δ(x; L1) > mcm δ(x; L1)m. This concludes the proof. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 19 7. Automorphisms of curves The following result is already present in [27]. For the clarity of exposition, we isolate it as a lemma. Lemma 7.1. -- Let k be an algebraically closed field of characteristic zero, let B be a smooth connected projective k-curve, let f : B → P1 be a nonconstant morphism and let Rf ⊂ B be the ramification locus of f (the set of points of B at which f is not étale) and let ∆f = f (Rf ) be its discriminant locus. Assume that there exist automorphisms g ∈ Aut(P1) and h ∈ Aut(B) such that f ◦ h = g ◦ f . Assume that g has infinite order. Then B is isomorphic to P1, and one of the following cases holds: -- The morphism f is an isomorphism (and ∆f = ∅); -- One has Card(Rf ) = 2, and g(∆f ) = ∆f . Proof. -- By construction, f induces a finite étale covering of P1 ∆f . Let b ∈ Rf ; one has df (b) = 0, hence d(f ◦ h)(b) = d(g ◦ f )(b) = 0. Since h is an automorphism of B, this implies that df (h(b)) = 0, hence h(b) ∈ Rf . We thus have h(Rf ) ⊂ Rf , hence h(Rf ) = Rf , because h is an isomorphism. Consequently, g(∆f ) = ∆f , so that some power of g fixes ∆f pointwise. Since the identity is the only homography that fixes 3 points and g has infinite order, this implies that Card(∆f ) 6 2. If Card(∆f ) 6 1, then P1 ∆f is simply connected, hence f is an isomorphism (and ∆f = ∅). Otherwise, one has Card(∆f ) = 2. Let n = deg(f ). Up to a change of projective coordinates in P1, we may assume that ∆f = {0, ∞}. Then g is a homothety, because it leaves ∆f invariant and has infinite order (otherwise, it would be of the form g(z) = a/z). Since all finite étale coverings of P1 ∆f are of Kummer type (equivalently, π1(P1 ∆f ) = Z), one has B ≃ P1 and the morphism f is conjugate to the morphism z 7→ zn from P1 to itself. We then remark that h is a homography of infinite order. Indeed, if he = idB, then f = ge ◦ f , hence ge = id since f is surjective, hence e = 0, since g has infinite order. As above, the formula h(Rf ) = Rf then implies that Card(Rf ) 6 2. On the other hand, Card(Rf ) > Card(∆f ) = 2, hence Card(Rf ) = 2. Proposition 7.2. -- Let k be a field of characteristic zero. Let B be an integral k-curve in Pn 1 possessing a smooth k-rational point. Let ΓB be the stabilizer of B in (Aut(P1))n and let Γ1 ⊂ Aut(P1) be its image under the first projection. Assume that Γ1 contains an element of infinite order. Then, one of the following cases holds: (1) The morphism p1B is constant; (2) The morphism p1B is an isomorphism and the components of its inverse are either constant or homographies; (3) There is a subset of P1(k) of cardinality 2 which is invariant under every element of Γ1. 20 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Proof. -- Assume that p1B is not constant. Let ν : B′ → B be the normalization 1 = p1 ◦ ν : B′ → P1. Let g = (g1, . . . , gn) be an element of ΓB; of B and let p′ There exists a unique automorphism h of B′ that lifts g; then p′ 1 ◦ h = g1 ◦ p′ 1. Since the curve B has smooth rational points, the curve B′ is geometrically inte- gral. Choosing g such that g1 has infinite order, the preceding lemma implies that Card(Rp′ ) ∈ {0, 2}. 1 Let us first assume that Card(Rp′ ) = 2. Then Card(∆p′ 1 ) = 2 as well. Moreover, 1 the relation p′ 1 ◦ h = g1 ◦ p′ 1 implies that g1(∆p′ ) ⊂ ∆p′ 1 , so that case (3) holds. 1 Let us now assume that Card(Rp′ 1) = 0 and fix g such that g1 has infinite order. By the preceding lemma, p′ 1 is an isomorphism; this implies that p1B is an isomorphism as well. Let f be its inverse and let f1, . . . , fn be its components; assume that case (2) does not hold, that is, for some j, the rational map fj is neither constant, nor a homography; its ramification locus Rj is non-empty. Since g1 has infinite order, the relation gj ◦ fj = fj ◦ g1 implies that gj has infinite order as well. By the preceding lemma, one has Card(Rj) = 2. Let then g′ = (g′ n) be any element of ΓB. The relation g′ 1(Rj) ⊂ Rj, so that case (3) holds. 1 implies that g′ j ◦ fj = fj ◦ g′ 1, . . . , g′ We will reduce the proof of theorem 2.7 to its following variant: 8. Proof of theorem 2.7 Proposition 8.1. -- Let F be a finite extension of Qp and let (Γi)16i6n be a finite family of arithmetic Schottky subgroups of PGL(2, F ) of ranks > 2. As above, let us set Ω = Qn i=1 XΓi, and let p : Ω → X an be the morphism deduced from the morphisms pΓi : ΩΓi → X an Γi . i=1 ΩΓi and X = Qn Let V be an irreducible algebraic subvariety of X and let W be an irreducible If W is algebraic subvariety of Ω, maximal among those contained in p−1(V an). geometrically irreducible, then it is flat. Lemma 8.2. -- Proposition 8.1 implies theorem 2.7. Proof. -- Let Y be the Zariski closure of W in Pn 1 ; by assumption, W is an ir- reducible component of Y an ∩ Ω. Let W0 be an irreducible component of WCp. By [17, théorème 7.16, (v)], there exists a finite extension F ′ of F , contained in Cp, and an irreducible component W ′ of WF ′ such that W0 = W ′ Cp. Then W ′ is geometrically irreducible, as well as its Zariski closure Y ′. By proposition 5.5, Ω ∩ Y ′ is geometrically irreducible; the inclusion W ′ ⊂ Ω ∩ Y ′ and the inequality dim(W ′) = dim(W0) = dim(W ) = dim(Y ) > dim(Y ′) imply that W ′ = Ω ∩ Y ′; in particular, W ′ is irreducible algebraic and is contained in p−1(V an F ′ ). Let us show that it is maximal. Let W ′ 1 ⊂ ΩF ′ be an irreducible algebraic subvariety contained in p−1(V an 1. The 1 )F is Zariski closed, because F ′ is a finite extension of F , and image Y1 of Y ′ 1 )F ′ be the Zariski closure of W ′ F ′ ) such that W ′ ( W ′ 1 in (Pn 1 ⊂ (Pn 1; let Y ′ A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 21 Y ′ 1 ⊂ (Y1)F ′. Moreover, Y ⊂ Y1. There exists a unique irreducible component W1 of Ω ∩ Y1 that contains W , and W ′ 1 is an irreducible component of W1,F ′. Neces- sarily, W1 is contained in p−1(V an), because W ′ F ′ ); this contradicts the maximality of W . 1 ⊂ p−1(V an Applying proposition 8.1 to W ′, we conclude that W ′ is flat. Consequently, W0 = Cp is flat, as was to be shown. W ′ 8.3. -- To prove proposition 8.1, we argue by induction and assume that it holds if there are less that n factors. Let W be an irreducible algebraic subvariety of Ω, maximal among those contained in p−1(V an) and geometrically irreducible. Let Y be an irreducible subvariety of Pn 1 such that W is an irreducible component of Y an ∩ Ω. By corollary 5.7, Y is geometrically irreducible, W = Y an ∩Ω and W is topologically dense in Y . The proof that W is flat requires intermediate steps and will be concluded in proposition 8.11. A crucial step will consist in proving that the stabilizer of W inside Γ has many points of bounded heights (proposition 8.10). To that aim, we define in section 8.7 an F -subanalytic subset R of PGL(2, F )n. The definition, close to that of a similar set in [26, 28], guarantees the following important property (lemma 8.8): if B is a small enough subset of R then, for every g ∈ B, the translate (g · Y an) ∩ Ω is contained in p−1(V an), and is independent of g. At this point, the maximality of W is invoked. The existence of such blocks is established by applying the p-adic Pila-Wilkie theorem of [10]. We thus prove that R has many rational points (lemma 8.9); these points are constructed using the action of the Schottky groups in a neighborhood of a boundary point ξ, applying material recalled in section 6. The construction of such a point ξ, performed in lemma 8.5, is actually the starting point of the proof. The actual statement of proposition 8.10 furnishes elements in Γ of a precise form. Using proposition 7.2, we will finally conclude the proof of proposition 8.1. 8.4. -- By assumption, W = Y an ∩Ω; consequently, the jth projection qj : (P1)n → P1 is constant on Y if and only if it is constant on W , if and only if the jth projection from X to Xj is constant on V , and in this case, its image is an F -rational point of P1, because W is geometrically irreducible. Deleting these constant factors, we thus assume that there does not exist j ∈ {1, . . . , n} such that the jth projection qj : (P1)n → P1 is constant on Y . Consequently, qjY : Y → P1 is surjective, for every j; in particular, Y an meets q−1 j (LΓj ). Let m = dim(Y ); by what precedes, we have m > 0, and Y an 6⊂ Ω. Lemma 8.5. -- Up to reordering the coordinates, there exists a smooth rigid point 1 )an such that the following ξ ∈ Y an and a connected open neighborhood U of ξ in (Pn properties hold: (1) The first component q1(ξ) of ξ belongs to the limit set LΓ1 of Γ1; 22 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER (2) Letting J = {1, . . . , m}, the projection qJ : Pn 1 → PJ 1 induces a finite étale morphism from U ∩ Y an to its image in (PJ 1 )an; (3) For every j ∈ {1, . . . , n} and every point y ∈ U ∩ Y an such that qj(y) ∈ LΓj , one has q1(y) ∈ LΓ1. Proof. -- For every subset V of Y an, let us define a relation (cid:22)V on {1, . . . , n} as follows: i (cid:22)V j if and only if, for every y ∈ V such that qi(y) ∈ LΓi, one has qj(y) ∈ LΓj . This is a preordering relation. If U ⊂ V ⊂ Y an and i (cid:22)V j, then i (cid:22)U j. We will define a decreasing sequence (V0, V1, . . . , Vn) of non-empty open subsets of Y an and a sequence (j0, j1, . . . , jn) of elements of {1, . . . , n}, such that for every k, qjk(Vk) meets LΓjk and 1, . . . , k (cid:22)Vk jk. We start with V0 = Y an. We have reduced ourselves to the case where qj(Y an) = P1 for every j; in particular, qj(Y an) meets LΓj ; we may take j0 = 1. Let k > 0 be such that V0, V1, . . . , Vk and j0, j1, . . . , jk are defined. If k + 1 (cid:22)Vk jk, we set Vk+1 = Vk and jk+1 = jk. Otherwise, one has k + 1 6(cid:22)Vk jk, hence there exists y ∈ Vk such that qk+1(y) ∈ LΓk+1 and qjk(y) 6∈ LΓjk ); this is an open neighborhood of y in Vk such that qjk+1(Vk+1) meets LΓjk+1 . By , so that jk (cid:22)Vk+1 k + 1. construction, no element z of Vk+1 satisfies qjk(z) ∈ LΓjk We then set jk+1 = k + 1. . Let Vk+1 = Vk ∩ (qjk)−1(ΩΓjk Let V = Vn and i = jn; let y ∈ V such that qi(y) ∈ LΓi. Let Z be the dense open subscheme of Y consisting of smooth points at which dqi does not vanish. Then Z an is open and dense in Y an, and V ∩ Z an is open and dense in V , hence qi(V ∩ Z an) is dense in qi(V ). Since LΓi has no isolated points, we may assume that y ∈ Z an. Rigid points are dense in q−1 i (qi(y)) ∩ V ∩ Z an; there exists a rigid point ξ in (qi)−1(qi(y)) ∩ V ∩ Z an. Since qi(y) is a rigid point, the point ξ is a rigid point of V ∩ Z an (and not only of its fiber of qi). Moreover, qi(ξ) = qi(y) ∈ LΓi. such that the projection qJ from V to (PJ dim(V ) = m. Consequently, there exists an open neighborhood U of ξ in (Pn such that qJ induces a finite étale morphism from U ∩ Y an to its image in (PJ Since dqi does not vanish at ξ, there exists a subset J of {1, . . . , n} containing i 1 )an is finite étale at ξ. One has Card(J) = 1 )an 1 )an. Reordering the coordinates, we may assume that i = 1 and J = {1, . . . , m}, hence the lemma. 8.6. -- Choose ξ, J = {1, . . . , m} and U as in the previous lemma; we may even assume that U is of the form U1 × · · · × Un, where, for each i, Ui is an open neigh- borhood of qi(ξ) in Pan 1 . Let F ′ be a finite extension of F such that ξ ∈ Y (F ′). Since W is geometrically irreducible, WF ′ is an irreducible algebraic subvariety of Ω. It is also maximal. Note that the flatness of WF ′ implies the flatness of W . Replacing F by F ′, we thus may assume that ξ ∈ Y (F ); then qJ induces a local isomorphism at ξ. Let ϕ = (ϕ1, . . . , ϕn) : O → Y an ∩ U be an analytic section of qJ Y an∩U , defined on an open neighborhood O of qJ (ξ); we may assume that O = U1 × · · · × Um. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 23 By condition (3) of lemma 8.5 one has q1(ϕ−1 j (LΓj )) ⊂ LΓ1 for every j ∈ {1, . . . , n}. 8.7. -- Let G be the Q-algebraic group PGL(2)n, and let G0 be the algebraic subgroup of G defined by (8.7.1) (g1, . . . , gn) ∈ G0 ⇔ g2 = · · · = gm = 1. We denote by q1, . . . , qn the projections of G to PGL(2). For every compact analytic domain F of Ω, we define a subset RF of G0(F ) by (8.7.2) g ∈ RF ⇔ dim(g · Y an ∩ F ∩ p−1(V an)) = m. Lemma 8.8. -- Let F be an affinoid domain of Ω. (1) The set RF is an F -subanalytic subset of G0(F ). (2) For every g ∈ RF, one has (g · Y an) ∩ Ω ⊂ p−1(V an). (3) Let M ⊂ RF be a subset whose Zariski closure is irreducible; for every g, h ∈ M, one has g · Y = h · Y . Proof. -- (1) The sets V and Y are algebraic over F , hence V (Cp) and Y (Cp) are rigid F -subanalytic. Since F is affinoid, the morphism pF defines a rigid F - subanalytic map from F(Cp) to V (Cp), so that (F ∩ p−1(V an))(Cp) is a rigid F - subanalytic set. Consequently, taking Cp-points, (g · Y an ∩ F ∩ p−1(V an))g furnishes a rigid F -subanalytic family of rigid F -subanalytic subsets of Ω(Cp), parameterized by G0(Cp). By b-minimality, the set of points g ∈ G0(Cp) such that dim(g · Y an ∩ F ∩ p−1(V an)) = m is a rigid F -subanalytic subset of G0(Cp). It then follows from lemma 4.2 that RF is an F -subanalytic subset of G0(F ). (2) Let g ∈ RF and let us prove that (g · Y an) ∩ Ω ⊂ p−1(V an). Since g · Y an is irreducible and g · Y an ∩ F has dimension m = dim(g · Y an), this intersection is Zariski dense in g · Y an. Moreover, there exists a finite extension F ′ of F such that F ′ ∩ F(F ′) is Zariski dense in YF ′ (it suffices that g · Y an ∩ F admits a smooth g · Y an F ′-point), so that the Zariski closure of g · Y an ∩ F(F ′) in (Pn 1 )F ′ is equal to g · YF ′. Moreover, g · Y (F ′) ∩ F(F ′) is F ′-semialgebraic, hence proposition 5.8 implies that g · Y an F ′ ). Since p is defined over F and g ∈ G(F ), this implies that (g · Y an) ∩ Ω ⊂ p−1(V an). F ′ ∩ ΩF ′ ⊂ p−1 F ′ (V an (3) As a subset, (M · Y an) ∩ Ω is contained in p−1(V an). By proposition 5.8, its Zariski closure Y ′ satisfies (Y ′)an ∩ Ω ⊂ p−1(V an) as well. Since Y and the Zariski closure of M are geometrically irreducible, Y ′ is geometrically irreducible. Let g ∈ M; then Y an ⊂ g−1M · Y an ⊂ g−1 · (Y ′)an, hence W ⊂ g−1 · (Y ′)an ∩ Ω. By maximality of W , one has W = g−1 · (Y ′)an ∩ Ω. This implies g · Y = Y ′, thus g · Y = h · Y for every g, h ∈ M. We return to the context of §8.6. In particular, ξ is a point of Y (F ) such that q1(ξ) ∈ LΓ1, and the restriction to Y of the projection to the first m coordinates is étale at ξ, with a local analytic section ϕ defined on U1 × · · · × Um. Lemma 8.9. -- There exist a real number c > 0, fundamental sets Fi ⊂ ΩΓi and a subset Υ of RF ∩ Γ, where F = Q Fi, such that 24 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER (1) For all T large enough, one has Card(ΥT ) > T c, where ΥT denotes the set of all γ ∈ Υ such that H(γ) 6 T ; (2) The projection q1 is injective on Υ; (3) For all j ∈ {1, . . . , n} such that qj(ξ) 6∈ LΓj , one has Card(qj(Υ)) = 1. We recall that there exists a number field K contained in F such that Γ ⊂ PGL(2, K)n, and H is induced by a fixed height function on PGL(2, Q)n. In par- ticular, lemma 8.9 implies that the subset RF of PGL(2, F )n has many K-rational points, in the sense of section 4.5. Proof. -- Let q be the genus of XΓ1; by proposition 6.8, there exists a basis α1, . . . , αq of Γ1, an open neighborhood U ′ 1 of q1(ξ), which is contained in U1 and stable under the action of α1, . . . , αq, and a fundamental set F1 for Γ1 contained in U ′ 1. For simplicity of notation, we now assume that U1 = U ′ 1. We have introduced in §8.6 a local analytic section ϕ = (ϕ1, . . . , ϕn) : U1 × · · · × Um → Y an ∩ U1 × · · · × Un of the projection qJ : Y → PJ 1 , where J = {1, . . . , m}. Let j ∈ {1, . . . , n} be such that qj(ξ) 6∈ LΓj . Then qj(ξ) has a compact analytic neighborhood U ′ j contained in ΩΓj . Shrinking U1, . . . , Um if necessary, we assume that the image of ϕj is contained in U ′ j for every such j. Let a′ = (a1, . . . , an) ∈ W be a rigid point that belongs to the image of ϕ and such that a1 ∈ F1. Let a = (a1, . . . , am); we have a′ = ϕ(a). For j ∈ {2, . . . , n}, we also choose a fundamental set Fj that contains aj. We claim that we can complete any element γ1 ∈ F1 which is a positive word γ1 in α1, . . . , αq to an element γ ∈ Γ such that γ−1 ∈ RF and H(γ) ≪ cℓΓ1 (γ1), for some real number c. Let us now prove the asserted claim. For any positive word γ1 in α1, . . . , αq, one has γ1 · a1 ∈ U1; in particular, we can consider the point a(γ1) = (γ1 · a1, a2, . . . , am) of U1 × · · · × Um and its image ϕ(a(γ1)) under the section ϕ. By §6.3, there exists a real number c1 > 1 such that δ(αj ·a1; LΓ1) > c−1 1 δ(a1; LΓ1), uniformly in a1. By induction on the length ℓΓ1(γ1) of the positive word γ1, this implies the inequality (8.9.1) δ(γ1 · a1; LΓ1) > c −ℓΓ1 (γ1) 1 . We first set γ2 = · · · = γm = 1 Let then j > m. Let ψj : U1 → Uj be the analytic map defined by ψj(x) = ϕj(x, a2, . . . , am). By construction (lemma 8.5), if ψj(x) = ϕj(x, a2, . . . , am) ∈ LΓj , one has x = q1(x, a2, . . . , am) ∈ LΓ1. In other words, one has ψ−1 j (LΓj ) ⊂ LΓ1. Applying lemma 6.11 to ψj, we obtain an inequality of the form δ(ϕj(x, a2, . . . , am); LΓj ) ≫ δ(x; LΓ1)k, for some integer k > 0, for all x ∈ U1. In particular, (8.9.2) δ(ϕj(a(γ1)); LΓj ) ≫ δ(γ1 · a1; LΓ1)k. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 25 By proposition 6.8, there exists γj ∈ Γj such that ϕj(a(γ1)) ∈ γj · Fj. By propo- sition 6.6 and lemma 6.9, one has (8.9.3) H(γj) ≪ δ(ϕj(a(γ1)); LΓj )−κ, where κ is a positive real number, independent of γ1. By equations (8.9.1), (8.9.2) and (8.9.3), we thus have (8.9.4) H(γj) ≪ δ(γ1 · a1; LΓj )−kκ ≪ c ℓΓ1 (γ1)kκ 1 . Let c = ckκ 1 . Let γ = (γ1, . . . , γn) ∈ Γ. By what precedes, H(γ) ≪ cℓΓ1 (γ1). Moreover, ϕj(a(γ1)) ∈ γj · Fj for every j: this follows from the fact that aj ∈ Fj if j 6 m, and from the construction of γj if j > m. Let us prove γ−1 ∈ RF. One has W ⊂ p−1(V an) by assumption; since γ ∈ Γ, this implies γ−1 · W ⊂ p−1(V an). Consequently, γ−1 · Y an ∩ F ∩ p−1(V an) ⊃ γ−1 · W ∩ F ∩ p−1(V an) = γ−1 · W ∩ F. The analytic morphism U1 × . . . Um → W, (x1, . . . , xm) 7→ ϕ(γ1 · x1, x2, . . . , xm) is an immersion and maps the point a = (a1, . . . , am) to the point ϕ(a(γ1)) ∈ γ · F. Since a is a rigid point, this morphism maps a neighborhood of a into γ · F, so that dim(W ∩ γ · F) > m. This proves γ−1 ∈ RF. Applying lemma 6.9 to estimate H(γ1), we thus have shown the existence of a positive real number c such that for every positive word γ1 in α1, . . . , αq, there exists an element γ = (γ1, . . . , γn) completing γ1 such that H(γ) ≪ cℓΓ1 (γ1) and γ−1 ∈ RF ∩ Γ. Let Υ′ be the set of all such elements γ−1, where γ1 ranges over positive words in α1, . . . , αq. It is a subset of RF ∩ Γ. By construction, the projection q1 is injective on Υ′. Moreover, since the number of positive words of length ℓ in α1, . . . , αq is qℓ, T is bounded from below by qlog(T )/ log(c) = T log(q)/ log(c), and the the cardinality of Υ′ exponent of T is strictly positive, since q > 2. Finally, let j be such that qj(ξ) 6∈ LΓj . By construction, ϕj(a(γ1)) ∈ γjFj, hence γjFj meets U ′ j. By corollary 6.7, the set Sj of such elements γj in Γj is finite. It follows that there is a subset Υ of Υ′ that satisfies the conclusion of the proposition. Proposition 8.10. -- Let G′ such that gj = id if qj(ξ) 6∈ LΓj . Both the stabilizer of W inside G′ image in Γ1 under the first projection have many rational points. 0 be the subgroup of G0 consisting of elements (gj) 0 ∩ Γ and its Proof. -- Let c, Υ, Fi, F = Q Fi, R = RF be as given by lemma 8.9; let T0 > 1 be such that Card(ΥT ) > T c for T > T0. Let K be a number field contained in F such that all groups Γj are contained in PGL(2, K); the points of R ∩ Γ are K-rational points. Recall that for every real number T , we denote by R(K; T ) the set of K-rational points of R of height 6 T . One has ΥT = Υ ∩ R(K; T ). 26 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER Since R is F -subanalytic (lemma 8.8), it is also Qp-subanalytic and we may apply the p-adic Pila-Wilkie theorem of [10], as stated in theorem 4.7. Let thus s ∈ N, d ∈ R, ε > 0, and B ⊂ R × Qs p be a family of blocks such that for every T > 1, p of cardinality < dT ε such that R(K; T ) ⊂ Sσ∈ΣT Bσ. there exists a subset ΣT ∈ Qs Let also t ∈ N and Z ⊂ G0(F ) × Qt p be a semi-algebraic subset such that for every σ ∈ Qs p such that Bσ ⊂ Zτ and dim(Bσ) = dim(Zτ ). Let finally r be an upper bound for the number of irreducible components of the Zariski closure of the sets Zτ , for τ ∈ Qt p. p, there exists τ ∈ Qt Let T > T0. Since ΥT ⊂ R(K; T ), by the pigeonhole principle, there exists σ ∈ ΣT such that Card(ΥT ∩ Bσ) > Card(ΥT ) Card(ΣT ) > 1 d T c−ε. Moreover, the Zariski closure of Bσ in PGL(2)n F has at most r irreducible com- ponents. Consequently, we may choose such an irreducible component M whose trace M on Bσ satisfies Card(ΥT ∩ M) > 1 dr T c−ε. (Observe that M is indeed the Zariski closure of M.) Let g ∈ ΥT ∩ M. Since the Zariski closure of M is irreducible and M ⊂ RF, it follows from lemma 8.8 that the stabilizer of W inside G0 ∩ Γ contains g−1M, hence g−1(ΥT ∩ M). By construction, the image of g−1(ΥT ∩ M) under the projection of index j is {id} if qj(ξ) 6∈ LΓj . This shows in particular that the stabilizer of W 0 ∩ Γ contains g−1(ΥT ∩ M). This set contains > T c−ε/dr points, and their inside G′ heights are ≪ T 2; the same holds for its image by the first projection, since this projection is injective on g−1(Υ ∩ M). We thus have shown that the stabilizer of W inside G′ 0 ∩ Γ has many rational points, as well as its image under the first projection. This concludes the proof of the proposition. Proposition 8.11. -- The subvariety W is flat. Proof. -- We have constructed in §8.6 an analytic map ϕ : U1 × · · · × Um → Y , which is a local section of the projection to the m first coordinates. Let a ∈ Qm i=2 (ΩΓi ∩ Ui); let us denote by Wa the fiber of W over a under the 1 ; let us define Ya similarly. When a varies, the number of i=2 Pan projection to Qm irreducible components of Ya is uniformly bounded. Let ψa : (U1)H (a) → Y an a be the analytic morphism deduced from ϕ. We claim that the components of ψa are either constant or homographies. Let g ∈ G0 ∩ Γ be an element such that g · W = W , g1 6= id, and gj = id if qj(ξ) 6∈ LΓj (proposition 8.10). Since g · W = W , one has g · Y = Y , hence A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 27 g · Wa = Wa and g · Ya = Ya. The element g induces a commutative diagram Ya ψa (P1)H (a) g g1 Ya ψa (P1)H (a), H (a). Let Y ′ where the section ψa is analytic and defined over the open subset (U1)H (a) of (P1)an a be the irreducible component of Ya that contains ψa(ξ1); it is geometrically irreducible. Recall that g1 has infinite order; replacing g1 and g by some fixed power, we may thus assume that g · Y ′ a = Y ′ a. By proposition 7.2, either Y ′ a → (P1)H (a) is an isomorphism and the components of its inverse are constant or homographies, or there exists a subset ∆ of P1(H (a)) such that Card(∆) = 2 and g1(∆) = ∆ for every element g = (g1, . . . , gn) ∈ G′ 0 ∩ Γ such that g · W = W and g · Y ′ a. Let us assume that we are in the latter case. Using that Γ1 ⊂ PGL(2, F ), we see that ∆ ⊂ P1(F ). By lemma 6.10, the projection to Γ1 of the stabilizer of W inside G′ 0 ∩ Γ has few rational points, contradicting proposition 8.10. a = Y ′ We thus have shown that the components of the analytic map ψa are either con- stant or given by homographies. Let j ∈ {m + 1, . . . , n}. Let us first assume that qj(ξ) ∈ ΩΓj . Then gj = id, whence the relation ψa,j = ψa,j ◦ g1. Since g1 6= id, this implies that ψa,j is constant, i.e., ϕj does not depend on the coordinate x1. Since U is reduced, the morphism ϕj is deduced by pull-back of an analytic map θj : Qm value qj(ξ), the section ψa,j cannot be constant. It is thus a homography τj,a. Let us then assume that qj(ξ) ∈ LΓj . Since the jth component of ϕ takes the i=2 Ui → Pan 1 . A priori, one has τj,a ∈ PGL(2, H (a)) for every a. However, by condition (3) of j (LΓj ) ⊂ LΓ1. The limit sets LΓ1 and LΓj are contained in j,a maps an infinite subset of P1(F ) lemma 8.5, one has ϕ−1 P1(F ) and have no isolated points, so that τ −1 into P1(F ); this implies that τj,a ∈ PGL(2, F ). Observe that for x ∈ U1 ∩P1(F ), one has τj,a ·x = ψa,j(x) = ϕ(x, a). In particular, the assignment a 7→ τj,a is induced by an analytic morphism. Since it takes its values in PGL(2, F ), it is constant. Let J ′ and J ′′ be the set of all j ∈ {m + 1, . . . , n} such that qj(ξ) belongs to LΓj and to ΩΓj respectively. Let Ω′ = ΩΓ1 × Qj∈J ′ ΩΓj and Ω′′ = Qm i=2 ΩΓi × Qj∈J ′′ ΩΓj ; similarly, write X ′ = X1 × Qj∈J ′ Xj and X ′′ = Qm i=2 Xi × Qj∈J ′′ Xj, and decompose the projection p : Ω → X as (p′, p′′), where p′ : Ω′ → X ′ and p′′ : Ω′′ → X ′′ are the natural projections. Let Z ′ be the graph in (P1 × Qj∈J ′ P1)an of (τj)j∈J ′ and Z ′′ ⊂ (Qm i=2 P1 × Qj∈J ′′ P1)an be the graph of (θj)j∈J ′′. Let Y ′ and Y ′′ be the Zariski closure of Z ′ and Z ′′, let W ′ and W ′′ be their traces in Ω′ and Ω′′, and let V ′ and V ′′ be the Zariski closures of p′(Z ′) and p′′(Z ′′). It is clear that Y ′ = Z ′ is the curve in P1 × Qj∈J ′ P1 28 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER (with coordinates x1 and xj for j ∈ J ′) given by the equations xj = τj(x1), and W ′ is its trace on Ω′. In particular, W ′ is flat. By construction, Z ′ × Z ′′ is a subspace of Y an which meets W in a Zariski dense subset of itself; hence Y = Y ′ × Y ′′ and W = Ω ∩ Y an = W ′ × W ′′. Moreover, p(W ) = p′(W ′) × p′′(W ′′) ⊂ V , hence V ′ × V ′′ ⊂ V . Consequently, W ′′ is a maximal algebraic irreducible subset of (p′′)−1((V ′′)an). By induction, W ′′ is flat. Consequently, W = W ′ × W ′′ is flat, as was to be shown. 9. A characterization of geodesic subvarieties 9.1. -- Let F be a finite extension of Qp and let (Γi)16i6n be a finite family of arithmetic Schottky subgroups of ranks > 2 in PGL(2, F ) Let us set Ω = Qn i=1 ΩΓi, X = Qn i=1 XΓi, and let p : Ω → X an be the morphism deduced from the morphisms pΓi : ΩΓi → X an Γi . Theorem 9.2. -- Let W be a Zariski closed subvariety of Ω, geometrically irre- ducible. Then the following properties are equivalent: (i) The variety W is geodesic ; (ii) Its projection p(W ) is algebraic; (iii) The dimension of the Zariski closure of p(W ) in X is equal to dim(W ). Proof. -- Let us assume that W is geodesic and let us show that p(W ) is algebraic. We may assume that no projection pΓi is constant on W . Define a relation ∼ on {1, . . . , n} given by i ∼ j if there exists g ∈ PGL(2, F ) (necessarily unique) such that gΓig−1 and Γj are commensurable and zj = g · zi for every z ∈ W . This is an equivalence relation. Fix an element j in each equivalence class; for i such that i ∼ j, we may replace Γi by its conjugate gΓig−1 and assume that zj = zi on W . This shows that W and Ω decompose as a product indexed by the set of equivalence classes of the following particular situation: all the subgroups Γi are commensurable, and W is the diagonal of Ω. It thus suffices to treat this particular case. i Let Γ0 = Ti Γi and X0 be the algebraic curve associated with ΩΓ0/Γ0. Then, for every i, the morphism fi : W → X an deduced from f = pW factors as the i composition of the uniformization p0 : ΩΓ0 → X an 0 → . By gaga ([3], corollary 3.5.2; [32], appendix), a finite analytic morphism of X an algebraic curves is algebraic; consequently, there exists a finite morphism qi : X0 → Xi such that fi = qan i ◦ p0. Then p(W ) is the image of X0 by the finite morphism q = (q1, . . . , qn) : X0 → X, hence is algebraic. This shows that (i) implies (ii). Since it is clear that (ii) implies (iii), it remains to prove that (iii) implies (i). 0 and of a finite morphism X an Let us assume now that the dimension of the Zariski closure V of p(W ) in X is equal to the dimension of W . By construction, W is a maximal irreducible algebraic subvariety of p−1(V an). By proposition 8.1, W is flat. A similar analysis as in the proof of the first implication shows that there is a partition of the indices {1, . . . , n} A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 29 under which W decomposes as a product of flat curves and points. Since it suffices to prove that each of these curves is geodesic, we may assume that W is a flat curve, of the form W = {(z, g2 · z, . . . , gn · z)} ∩ Ω, where g2, . . . , gn ∈ PGL(2, F ). Let us first assume that n = 2. Let then g ∈ PGL(2, F ) be such that W = {(z, g · z)} ∩ Ω and let us prove that Γ2 and gΓ1g−1 are commensurable, a property which is equivalent to the finiteness of both orbit sets Γ2\Γ2gΓ1 and Γ1\Γ1g−1Γ2. Let us argue by contradiction and assume that Γ2\Γ2gΓ1 is infinite. (The other finiteness is analogous, or follows by symmetry.) Fix a rigid point z ∈ ΩΓ1. Let A ⊂ Γ1 be a set such that gA is a set of representatives of Γ2\Γ2gΓ1; by assumption, A is infinite. Since Γ\W ⊂ V an, the algebraic variety V contains the infinite set of points p(a · z, g · az) = (p1(z), p2(ga · z)), for a ∈ A, hence it contains its Zariski closure {p1(z)} × X2. Since this holds for every z ∈ W , we deduce that V contains X1 × X2, contradicting the assumption that dim(W ) = 1. Let us now return to the general case. To prove that W is geodesic, it suffices to establish that the subgroups Γj and gjΓ1g−1 are commensurable, for every j ∈ {2, . . . , n}. Up to renumbering the indices, it suffices to treat the case j = 2. Let Ω′ = ΩΓ1 × ΩΓ2, let p′ : Ω′ → X ′ = X1 × X2 be the uniformization map, and denote by π the projections from Ω to Ω′ and from X to X ′. Let W ′ = π(W ) and V ′ = π(V ). By Chevalley's theorem, V ′ is an algebraic curve in X ′. Obviously, W ′ is a flat curve contained in (p′)−1((V ′)an), hence is an maximal irreducible algebraic subset of (p′)−1((V ′)an) ∩ Ω′. By the case n = 2, the Schottky groups Γ2 and g2Γ1g−1 are 2 commensurable, as was to be shown. This concludes the proof of theorem 9.2. j Corol lary 9.3. -- Let V be an irreducible curve in X. Then every irreducible algebraic subvariety of ΩCp which is maximal among those contained in p−1(V an Cp) is geodesic. Proof. -- Let W0 be a irreducible algebraic subvariety of ΩCp, maximal among those contained in p−1(V an let us prove that W0 is geodesic. We may as- sume that dim(W0) > 0. Since p is surjective and has discrete fibers, one has dim(p−1(V an Cp), hence dim(W0) = 1, so that W0 is an irreducible component of p−1(V an)Cp. By theorem 7.16 of [17], there exists a finite extension E of F and an irreducible component W of p−1(V an)E such that W0 = WCp. Cp)) = dim(V an Cp); By theorem 9.2, W is geodesic. Consequently, W0 is geodesic. Remark 9.4. -- This corollary suggests that the main results of the paper extend to maximal algebraic irreducible subvarieties of p−1(V an)Cp, without assuming that they are defined over a finite extension of F . 30 ANTOINE CHAMBERT-LOIR & FRANÇOIS LOESER References [1] Y. André -- Period mappings and differential equations. From C to Cp, MSJ Mem- oirs, vol. 12, Mathematical Society of Japan, Tokyo, 2003, Tôhoku-Hokkaidô lectures in arithmetic geometry, with appendices by F. Kato and N. Tsuzuki. [2] J. Ax -- "On Schanuel's conjectures", Ann. of Math. (2) 93 (1971), p. 252 -- 268. [3] V. G. Berkovich -- Spectral theory and analytic geometry over non-Archimedean fields, Mathematical Surveys and Monographs, vol. 33, American Mathematical So- ciety, Providence, RI, 1990. [4] , "Étale cohomology for non-Archimedean analytic spaces", Publ. Math. Inst. Hautes Études Sci. (1993), no. 78, p. 5 -- 161. [5] D. Bertrand -- "Séries d'Eisenstein et transcendance", Bull. Soc. Math. France 104 (1976), no. 3, p. 309 -- 321. [6] J.-F. Boutot & H. Carayol -- "Uniformisation p-adique des courbes de Shimura: les théorèmes de Čerednik et de Drinfel'd", Astérisque, no. 196-197, p. 45 -- 158, Astérisque 196-197, Soc. Math. France, 1991. [7] J.-F. Boutot & T. Zink -- "The p-adic uniformization of Shimura curves", https://www.math.uni-bielefeld.de/~zink/p-adicuni.ps, 1995. [8] I. V. Čerednik -- "Towers of algebraic curves that can be uniformized by discrete subgroups of PGL2(kw) × E", Mat. Sb. (N.S.) 99(141) (1976), no. 2, p. 211 -- 247, 296. [9] P. L. Clark -- "Rational points on Atkin-Lehner quotients of Shimura curves", Ph.D. Thesis, Harvard University, 2003, http://alpha.math.uga.edu/~pete/thesis.pdf. [10] R. Cluckers, G. Comte & F. Loeser -- "Non-archimedean Yomdin-Gromov parametrizations and points of bounded height", Forum Math. Pi (2015), no. 3, e5 (60 p.). [11] R. Cluckers, L. Lipshitz & Z. Robinson -- "Analytic cell decomposition and analytic motivic integration", Ann. Sci. École Norm. Sup. 39 (2006), no. 4, p. 535 -- 568. [12] R. Cluckers & F. Loeser -- "b-minimality", J. Math. Log. 7 (2007), no. 2, p. 195 -- 227. [13] B. Conrad -- "Irreducible components of rigid spaces", Ann. Inst. Fourier (Grenoble) 49 (1999), no. 2, p. 473 -- 541. [14] J. Denef & L. van den Dries -- "p-adic and real subanalytic sets", Ann. of Math. 128 (1988), no. 1, p. 79 -- 138. [15] J. Denef -- "p-adic semi-algebraic sets and cell decomposition", J. Reine Angew. Math. 369 (1986), p. 154 -- 166. [16] V. G. Drinfel'd -- "Coverings of p-adic symmetric domains", Funkcional. Anal. i Priložen. 10 (1976), no. 2, p. 29 -- 40. [17] A. Ducros -- "Les espaces de Berkovich sont excellents", Ann. Inst. Fourier (Greno- ble) 59 (2009), no. 4, p. 1443 -- 1552. [18] [19] L. Gerritzen & M. van der Put -- Schottky groups and Mumford curves, Lecture , "Families of Berkovich spaces", 2016, arXiv:1107.4259v5. Notes in Mathematics, vol. 817, Springer, Berlin, 1980. [20] B. Klingler, E. Ullmo & A. Yafaev -- "The hyperbolic Ax-Lindemann- Weierstrass conjecture", Publ. Math. Inst. Hautes Études Sci. 123 (2016), p. 333 -- 360, arXiv:1307.3965. A NON-ARCHIMEDEAN AX-LINDEMANN THEOREM 31 [21] L. Lipshitz -- "Rigid subanalytic sets", Amer. J. Math. 115 (1993), no. 1, p. 77 -- 108. [22] L. Lipshitz & Z. Robinson -- "Model completeness and subanalytic sets", Astérisque (2000), no. 264, p. 109 -- 126. , "Quasi-affinoid varieties", Astérisque (2000), no. 264, p. 127 -- 149. [23] [24] D. Mumford -- Abelian varieties, Oxford Univ. Press, 1974. [25] J. Pila -- "On the algebraic points of a definable set", Selecta Math. (N.S.) 15 (2009), no. 1, p. 151 -- 170. [26] [27] [28] , "O-minimality and the André-Oort conjecture for Cn", Ann. of Math. 173 (2011), no. 3, p. 1779 -- 1840. , "Modular Ax-Lindemann-Weierstrass with derivatives", Notre Dame J. Form. Log. 54 (2013), no. 3-4, p. 553 -- 565. , "Functional transcendence via o-minimality", in O-Minimality and Diophan- tine Geometry (G. O. Jones & A. J. Wilkie, eds.), London Mathematical Society Lecture Note Series, vol. 421, Cambridge University Press, 2015, p. 66 -- 99. [29] J. Pila & J. Tsimerman -- "The André-Oort conjecture for the moduli space of abelian surfaces", Compos. Math. 149 (2013), no. 2, p. 204 -- 216. [30] J. Pila & A. J. Wilkie -- "The rational points of a definable set", Duke Math. J. 133 (2006), no. 3, p. 591 -- 616. [31] J. Pila & U. Zannier -- "Rational points in periodic analytic sets and the Manin- Mumford conjecture", Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. 19 (2008), no. 2, p. 149 -- 162. [32] J. Poineau -- "Raccord sur les espaces de Berkovich", Algebra Number Theory 4 (2010), no. 3, p. 297 -- 334. [33] R. S. Rumely -- Capacity theory on algebraic curves, Lecture Notes in Math., vol. 1378, Springer-Verlag, Berlin, 1989. [34] A. Seidenberg -- "Abstract differential algebra and the analytic case", Proc. Amer. Math. Soc. 9 (1958), p. 159 -- 164. [35] G. Shimura -- "On the zeta-functions of the algebraic curves uniformized by certain automorphic functions", J. Math. Soc. Japan 13 (1961), p. 275 -- 331. [36] E. Ullmo & A. Yafaev -- "Hyperbolic Ax-Lindemann theorem in the cocompact case", Duke Math. J. 163 (2014), no. 2, p. 433 -- 463.
1903.02107	3	1903	2019-03-13T04:47:01	The Bogomolov-Tian-Todorov Theorem Of Cyclic $A_\infty$-Algebras	[ "math.AG", "math.QA" ]	Let $A$ be a finite-dimensional smooth unital cyclic $A_\infty$-algebra. Assume furthermore that $A$ satisfies the Hodge-to-de-Rham degeneration property. In this short note, we prove the non-commutative analogue of the Bogomolov-Tian-Todorov theorem: the deformation functor associated with the differential graded Lie algebra of Hochschild cochains of $A$ is smooth. Furthermore, the deformation functor associated with the DGLA of cyclic Hochschild cochains of $A$ is also smooth.	math.AG	math	THE BOGOMOLOV-TIAN-TODOROV THEOREM OF CYCLIC A∞-ALGEBRAS JUNWU TU Abstract. Let A be a finite-dimensional smooth unital cyclic A∞-algebra. Assume furthermore that A satisfies the Hodge-to-de-Rham degeneration prop- erty. In this short note, we prove the non-commutative analogue of the Bogomolov-Tian-Todorov theorem: the deformation functor associated with the differential graded Lie algebra of Hochschild cochains of A is smooth. Furthermore, the deformation functor associated with the DGLA of cyclic Hochschild cochains of A is also smooth. 1. The non-commutative Bogomolov-Tian-Todorov Theorem Let X be a Calabi-Yau manifold, i.e. a compact complex manifold with trivial canonical bundle. It is a classical result of Bogomolov-Tian-Todorov [11][12] that the formal deformation functor associated with the differential graded Lie algebra (DGLA) (cid:77) gX := A0,q(ΛpTX ) p,q In or- of the Dolbeault resolution of holomorphic poly-vector fields is smooth. der to prove this, the key observation was the existence of a BV operator ∆ : Λ∗TX→Λ∗−1TX , which "trivializes" the Lie bracket by the Tian-Todorov identity (1) [α, β] = ∆(α ∧ β) − ∆(α) ∧ β − (−1)αα ∧ ∆(β). With the above formula, the smoothness of the deformation functor follows easily from the classical ∂∂-Lemma in Hodge theory. Following Kontsevich-Soibelman [8] and Katzarkov-Kontsevich-Pantev [7], one can formulate the compactness, smoothness, and the Calabi-Yau property purely in terms of the differential graded category of coherent sheaves on X. Thus, a natural question is whether the analogues of the Bogomolov-Tian-Todorov's theorem holds for any smooth and proper Calabi-Yau categories. This question might have been a folklore theorem for experts in the field. The purpose of this note is to fill some of the missing details in the literature. A large class of dg categories of interests is compactly generated by a single object. For this reason, instead of considering formal deformations of dg cat- egories (whatever that means), we shall consider deformations of A∞-algebras which is also much more tractable. To state the non-commutative version of the Bogomolov-Tian-Todorov theorem precisely, we first fix some notations and con- ventions. Throughout the paper, we use the homological degree of chain complexes. If A is a chain complex, its suspension is denoted by sA with (sA)n := An−1. For Institute of Mathematical Sciences, ShanghaiTech University, Shanghai, China, 201210. E- mail: tujw.at.shanghaitech.edu.cn. 1 a unital A∞-algebra A, denote by C−∗(A) (C∗(A)) its reduced Hochschild cochain complex (chain complex respectively). The minus sign is due to that we use homo- logical degree. Let A be a cyclic unital A∞-algebra, denote by C λ(A) ⊂ C−∗(A) the sub-complex consisting of cyclic cochains with respect to the pairing on A. Theorem 1.1. Let A be a Z/2Z-graded, finite-dimensional smooth unital cyclic A∞-algebra. Assume furthermore that A satisfies the Hodge-to-de-Rham degenera- tion property. Then we have (A.) The deformation functor DefsC−∗(A) associated with the DGLA sC−∗(A) of Hochschild cochains is smooth. (B.) The deformation functor DefsCλ(A) associated with the DGLA sC λ(A) of cyclic Hochschild cochains is also smooth. (C.) The natural transformation f : DefsCλ(A)→DefsC−∗(A) associated with the canonical inclusion map sC λ(A) (cid:44)→ sC−∗(A) is smooth. In particular, every deformation of the A∞ structure of A lifts to a deformation of the cyclic A∞ structure of A. Remark 1.1. The assumption of the Hodge-to-de-Rham degeneration property au- tomatically holds for any Z-graded smooth and proper A∞-algebra by Kaledin [6]. In the general Z/2Z-graded case, this remains an open conjecture by [8] [7]. Part (A.) of the above Theorem was proved by Isamu Iwanari [5] with a different method. 2. BV∞-algebra structure on C−∗(A) To prove the above Theorem 1.1, one follows the same idea as in the proof of Bogomolov-Tian-Todorov's Theorem. However, the key identity Equation 1 fails to hold. It only holds up to homotopy. Also, the cup product on C−∗(A) is only commutative up to homotopy. Thus it is natural to work with a homotopy version of the underlying algebraic structures. In this section, we first exhibit a homotopy BV algebra structure, or BV∞- algebra structure on C−∗(A). The definition of BV∞-algebras used in this paper is from the article [3]. In fact, it was argued in Loc. Cit. that combining a TCFT structure defined by [1] [8] and the formality of the operad BV, one easily deduces the existence of a BV∞-algebra structure on C−∗(A) with A as in Theorem 1.1. However, to make such structure useful in order to deduce Theorem 1.1, one needs to say a bit more about this BV∞ structure. For example, its underlying Lie∞ algebra is in fact given by the differential graded Lie algebra(cid:0)C−∗(A), δ, [−,−]G (cid:1). For this purpose, we need to use a construction of Tamarkin in his proof of the Deligne's conjecture [10]. We introduce the following notations: • Lie -- The Lie operad. • Lie∞ -- The homotopy Lie operad. • E2 -- The operad whose representation gives Gerstenhaber algebras. • G∞ -- The homotopy E2 operad. • BV -- The operad whose representation gives BV algebras. • BV∞ -- The homotopy BV operad. • B∞ -- The brace operad [4, Section 5.2]. • F -- The operad defined by Tamarkin in [10, Section 6]. • C comb ∗ (F D) -- The operad of black-and-white ribbon trees defined by Kontsevich- Soibelman [8, Section 11.6], see also Wahl-Westerland [14, Section 2]. This operad gives a combinatorial model for the framed little disk operad. 2 • X -- A cofibrant replacement of C comb (F D). ∗ For an operad O, denote by O {1} its shifted version so that an O {1}-algebra structure on a chain complex A is equivalent to an O-algebra structure on sA. The endomorphism operad of a chain complex is denoted by End(−). The starting point to construct a BV∞ structure on C−∗(A) is that the operad C comb ∗ (F D) naturally acts on C−∗(A): (2) C comb ∗ (F D)→End(cid:0)C−∗(A)(cid:1). We refer to Kontsevich-Soibelman [8] and Wahl-Westerland [14] for details of this action. Here we illustrate this action with a few examples. Indeed, the following black-and-white ribbon tree ∗ inside C comb under the isomorphism C−∗(A) ∼=(cid:0)C∗(A)(cid:1)∨ (F D)(1) gives the pull-back of the Connes operator B : C∗(A)→C∗(A) . Denote it by ∆ : C−∗(A)→C−∗(A). (F D)(2). For exam- There are also binary operators associated with trees in C comb ple, consider the following two black-and-white ribbon trees: ∗ T = R = The graph T gives the familiar cup product on C−∗(A), while the graph R gives the first brace operator ◦ on C−∗(A) whose commutator (after shift) is the Gerstenhaber Lie bracket [−,−]G. Lemma 2.1. There is a morphism of operad h : B∞→C comb ∗ (F D) defined by h(mk) := Tk, k ≥ 2 h(m1,k) := Wk, k ≥ 1, h(mp,k) := 0, p ≥ 2. Here Tk and Wk are given by the following black-and-white ribbon trees: Tk := , Wk := ··· ··· Proof. This is a straight-forward check. In [10], Tamarkin constructed morphisms s : G∞→F and t : F→B∞. (cid:3) 3 Lemma 2.2. There exists a commutative diagram: F t B∞ h X C comb ∗ (F D) Proof. This follows the lifting property since X→C comb while F by construction is cofibrant. The framed little disk operad is known to be formal with cohomology the BV operad, which implies that BV ∼= C comb (F D) in the homotopy category of differen- tial graded operads. Since the operad X is cofibrant, and BV is fibrant (as any dg operad is fibrant), we obtain a morphism X→BV such that the roof diagram (F D) is a trivial fibration, (cid:3) ∗ ∗ X C comb ∗ (F D) BV represents an isomorphism in the homotopy category of differential graded operads. Lemma 2.3. The following diagram is commutative: Lie∞ {1} G∞ F X BV∞ BV Proof. It is clear that the left composition factors as Lie∞ {1}→Lie{1}→BV. For the right composition, consider the following composition Lie∞ {1}→F→X→C comb ∗ (F D). By Lemma 2.2 above, it is equal to Lie∞ {1}→F→B∞→C comb ∗ (F D), which by [10] can be factored as Lie∞ {1}→Lie{1}→B∞→C comb ∗ (F D). This shows that both compositions vanish on generators lk of Lie∞ {1} with k ≥ 3. (cid:3) For k = 2, this is a direct check by definition. 4 By definition of X , the right vertical map in the above diagram is a trivial fibration since BV is minimal. The left vertical map is a cofibration. Thus by the lifting property, we obtain a map BV∞→X : Lie∞ {1} G∞ F X BV∞ BV (3) ∗ We define a BV∞-algebra structure on C−∗(A) via the composition: (F D)→End(cid:0)C−∗(A)(cid:1). is given by the differential graded Lie algebra(cid:0)sC−∗(A), δ, [−,−]G Corollary 2.1. Let C−∗(A) be endowed with the BV∞-algebra structure defined by the Equation 3. Then its underlying Lie∞ structure on the suspension of C−∗(A) ρ : BV∞→X→C comb (cid:1). Proof. This follows from Lemma 2.2 and Tamarkin's commutative diagram: Lie∞ {1} G∞ Lie{1} F B∞ Consider a subset of generating operators of a BV∞ structure given by (cid:8)ld kk ≥ 1, d ≥ 0(cid:9) (cid:3) which label a basis of the convolution between the Koszul dual cooperad of Lie and that of the operad generated by the circle operator ∆. We denote the sub-operad k in BV∞ by qLie∞ {1}. The notation is because that a qLie∞ {1}- generated by ld algebra structure on a chain complex V is equivalent to an Lie∞ structure on sV [[u]] (with u a degree 2 formal variable), which may be thought of as a "quantum" Lie∞ {1} structure. Corollary 2.2. The induced qLie∞ {1} structure on C−∗(A) is of the form (4) 2 = [−,−]G, l0 k(α1 ··· αk) = 0,∀α1,··· , αk ∈ Ker∆, ∀k ≥ 2, k = 0 ∀ k ≥ 3, (5) (6) l0 1 = δ, l0 l1 1 = ∆, l1 k = 0, ∀ d ≥ 2. ld Proof. Property (3) is proven in the previous Corollary 2.1. To prove Property (4) k is equal to 2d + 2k − 3. But, and (5), observe that the degree of the operator ld (F D)(k) is equal to 2k − 1, the top dimensional chains in the chain complex C comb ∗ 5 which implies the vanishing in (5). The operator l1 graph 1 = ∆, since the black-and-white (cid:0)C comb (F D)(1)(cid:1) ∼= H1(S1). For k ≥ 2, since the degree of the operator l1 is the unique degree one graph representing the fundamental class in the homology group H1 is 2k − 1 which is top dimensional in the chain complex C∗comb∗(F D)(k). Thus l1 k is a linear combination of operators coming from black-and-white ribbon trees with k white vertices, and of degree 2k − 1. One can show (easy combinatorics) that any such tree must contain at least one white vertex of the form: ∗ k ··· If we input any α ∈ Ker∆ at this type of white vertex, the result gives zero by (cid:3) definition of the action of ribbon trees on C−∗(A) (by Equation 2). 3. Proof of Theorem 1.1 As explained in the previous section, via the composition in Equation 3, we have a BV∞ structure on C−∗(A). Thus, we may form its bar-cobar resolution: ΩBC−∗(A) ∼= C−∗(A), implies that ι is a morphism of Lie∞-algebras. 6 which yields a differential graded BV algebra homotopy equivalent to C−∗(A). At this point, we use the following theorem due to Katzarkov-Kontsevich-Pantev [7] and Terilla [13]. sequence associated with the complex(cid:0)S[[u]], d + u∆(cid:1) endowed with the u-filtration is degenerate at the first page. Then both the DGLA's(cid:0)S, d, [−,−](cid:1) and(cid:0)S[[u]], d + u∆, [−,−](cid:1) are homotopy abelian. Theorem 3.1. Let S be a differential graded BV algebra. Assume that the spectral Since the degeneration of the spectral sequence is a homotopy invariant property (see [2]), we may use the above theorem to deduce the homotopy abelian property of the DGLA sC−∗(A). Note that here it is essential that the BV∞ structure on C−∗(A) extends the DGLA structure of sC−∗(A) by Corollary 2.1. Similarly, we may also restrict the BV∞ structure on C−∗(A) to the sub-operad qLie∞ {1}. Then theorem above implies that the following Lie∞-algebra (cid:0)sC−∗(A)[[u]], δ + u∆, [−,−]G + u · l1 2, u · l1 is homotopy abelian. Here the structure maps l1 k are as in Corollary 2.2. The following lemma then finishes the proof of (B.) in Theorem 1.1, using the homotopy invariance of deformation functors. k (k ≥ 3)(cid:1) Lemma 3.1. The canonical inclusion map ι : sC λ(A) (cid:44)→ sC−∗(A)[[u]] is a quasi-isomorphism of Lie∞-algebras. Proof. This inclusion is a quasi-isomorphism is a classical result, see for example [9]. By Corollary 2.2, the higher brackets l1 k vanish on elements inside Ker∆ while we certainly have that Im ι ⊂ Ker∆. This shows that k(ια1,··· , ιαk) = 0, ∀α1,··· , αk ∈ sC λ(A), l1 (cid:3) Denote by π : sC−∗(A)[[u]]→sC−∗(A) the projection map defined by setting u = 0. Part (C.) of Theorem 1.1 easily follows from (A.) and (B.). Indeed, since both functors are smooth, it suffices to check that the inclusion map π ◦ ι : sC λ(A) (cid:44)→ sC−∗(A) induces a surjection on the tangent space of the associated deformation functors. This is clear as ι is a quasi-isomorphism, and π is a surjective on cohomology by the Hodge-to-de-Rham degeneration assumption. References [1] Costello, K. Topological conformal field theories and Calabi-Yau categories. Adv. Math. 210 (2007), no. 1, 165-214. [2] Dotsenko, V.; Shadrin, S.; Vallette, B. De Rham cohomology and homotopy Frobenius man- ifolds. J. Eur. Math. Soc. (JEMS) 17 (2015), no. 3, 535-547. [3] Galvez-Carrillo, I.; Tonks, A.; Vallette, B. Homotopy Batalin-Vilkovisky algebras, J. Non- commut. Geom. 6 (2012), no. 3, 539-602. [4] Getzler, E.; Jones., J. D. S. Operads, Homotopy algebra, and iterated integrals for double loop spaces, hep-th/9403055. [5] Iwanari, I. Period mappings for noncommutative algebras. arXiv:1604.08283. [6] Kaledin, D. Non-commutative Hodge-to-de-Rham degeneration via the method of Deligne- Illusie. Pure Appl. Math. Q. 4 (2008), no. 3, Special Issue: In honor of Fedor Bogomolov. Part 2, 785-875. [7] Katzarkov, L.; Kontsevich, M.; Pantev, T., Hodge theoretic aspects of mirror symmetry. From Hodge theory to integrability and TQFT tt*-geometry, 87-174, Proc. Sympos. Pure Math., 78, Amer. Math. Soc., Providence, RI, 2008. [8] Kontsevich, M., Soibelman, Y., Notes on A-infinity algebras, A-infinity categories and non- commutative geometry. I, Homological mirror symmetry, 153-219, Lecture Notes in Phys., 757, Springer, Berlin, 2009. [9] Loday, J.-L. Cyclic homology. Appendix E by Maria O. Ronco. Grundlehren der Mathe- matischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 301. Springer- Verlag, Berlin, 1992. xviii+454 pp. ISBN: 3-540-53339-7. [10] Tamarkin, D. Another proof of M. Kontsevich formality arXiv:math/9803025. theorem for Rn. [11] Tian, G. Smoothness of the universal deformation space of compact Calabi-Yau manifolds and its Petersson-Weil metric. Mathematical aspects of string theory (San Diego, Calif., 1986), 629-646, Adv. Ser. Math. Phys., 1, World Sci. Publishing, Singapore, 1987. [12] Todorov, A. The Weil-Petersson geometry of the moduli space of SU(n ≥ 3) (Calabi-Yau) manifolds. I, Comm. Math. Phys., 1989, vol. 126, pp. 325-346. [13] Terilla, J. Smoothness theorem for differential BV algebras. J. Topol. 1 (2008), no. 3, 693-702. [14] Wahl, N.; Westerland, C. Hochschild homology of structured algebras. Adv. Math. 288 (2016), 240-307. 7
1304.5248	1	1304	2013-04-18T20:33:51	Gorenstein in codimension 4 - the general structure theory	[ "math.AG", "math.AC" ]	I describe the projective resolution of a codimension 4 Gorenstein ideal, aiming to extend Buchsbaum and Eisenbud's famous result in codimension 3. The main result is a structure theorem stating that the ideal is determined by its (k+1) x 2k matrix of first syzygies, viewed as a morphism from the ambient regular space to the Spin-Hom variety SpH_k in Mat(k+1,2k). This is a general result encapsulating some theoretical aspects of the problem, but, as it stands, is still some way from tractable applications.	math.AG	math	Gorenstein in codimension 4 -- the general structure theory Miles Reid Abstract . G A h t a m [ 1 v 8 4 2 5 . 4 0 3 1 : v i X r a I describe the projective resolution of a codimension 4 Gorenstein ideal, aiming to extend Buchsbaum and Eisenbud's famous result in codimension 3. The main result is a structure theorem stating that the ideal is determined by its (k+1)×2k matrix of first syzygies, viewed as a morphism from the ambient regular space to the Spin-Hom variety SpHk ⊂ Mat(k + 1, 2k). This is a general result encapsulating some theoretical aspects of the problem, but, as it stands, is still some way from tractable applications. This paper introduces the Spin-Hom varieties SpHk ⊂ Mat(k + 1, 2k) for k ≥ 3, that I define as almost homogeneous spaces under the group GL(k + 1)× O(2k) (see 2.4). These serve as key varieties for the (k + 1)× 2k first syzygy matrixes of codimension 4 Gorenstein ideals I in a polynomial ring S plus appropriate presentation data; the correspondence takes I to its matrix of first syzygies. Such ideals I are parametrised by an open sub- scheme of SpHk(S) = Mor(Spec S, SpHk). The open condition comes from the Buchsbaum -- Eisenbud exactness criterion "What makes a complex ex- act?" [BE1]: the classifying map α : Spec S → SpHk must hit the degeneracy locus of SpHk in codimension ≥ 4. The map α has Cramer-spinor coordinates Li and σJ in standard repre- sentations kk+1 and k2k−1 of GL(k + 1) and Pin(2k) (see 3.3), and the k × k minors of M1(I) are in the product ideal I · Sym2({σJ}). The spinors them- selves should also be in I, so that the k×k minors of M1(I) are in I 3; this goes some way towards explaining the mechanism that makes the syzygy matrix M1(I) "drop rank by 3 at one go" -- it has rank k outside V (I) = Spec(S/I) and ≤ k − 3 on V (I). 1 The results here are not yet applicable in any satisfactory way, and raise almost as many questions as they answer. While Gorenstein codimension 4 ideals are subject to a structure theorem, that I believe to be the correct codimension 4 generalisation of the famous Buchsbaum -- Eisenbud theorem in codimension 3 [BE2], I do not say that this makes them tractable. Thanks I am grateful to Chen Jungkai for inviting me to AGEA, and to him and his colleagues at University of Taiwan for generous hospitality. My visit was funded by Korean Government WCU Grant R33-2008-000-10101-0, which also partly funded my work over the last 4 years, and I am very grateful to Lee Yongnam for setting up the grant and administering every aspect of it. I wish to thank Fabrizio Catanese, Eduardo Dias, Sasha Kuznetsov and Liu Wenfei for contributing corrections, questions and stimulating discussion. I owe a particular debt of gratitude to Alessio Corti for detailed suggestions that have helped me improve the layout and contents of the paper. Website See www.warwick.ac.uk/staff/Miles.Reid/codim4 for material ac- companying this paper. 1 Introduction Gorenstein rings are important, appearing throughout algebra, algebraic geometry and singularity theory. A common source is Zariski's standard construction of graded ring over a polarised variety X, L: the graded ring R(X, L) = Ln≥0 H 0(X, nL) is a Gorenstein ring under natural and fairly mild conditions (cohomology vanishing plus KX = kXL for some kX ∈ Z, see for example [GW]). Knowing how to construct R(X, L) by generators and relations gives precise answer to questions on embedding X ֒→ Pn and determining the equations of the image. 1.1 Background and the Buchsbaum -- Eisenbud result I work over a field k containing 1 2 (such as k = C, but see 4.5 for the more general case). Let S = k[x1, . . . , xn] be a positively graded polynomial ring with wt xi = ai, and R = S/IR a quotient of S that is a Gorenstein ring. Equivalently, Spec R ⊂ Spec S = An k is a Gorenstein graded scheme. By the Auslander -- Buchsbaum form of the Hilbert syzygies theorem, R has a 2 minimal free graded resolution P• of the form 0 ← P0 ← P1 ← · · · ← Pc ← 0 ↓ R (1.1) where P0 = S → R = S/IR is the quotient map, and P1 → S gives a minimum set of generators of the ideal IR. Here the length c of the resolution equals n − depth R, and each Pi is a graded free module of rank bi. I write Pi = biS (as an abbreviation for S ⊕bi), or Pi = Lbi j=1 S(−dij) if I need to keep track of the gradings. The condition depth R = dim R that the depth is maximal characterises the Cohen -- Macaulay case, and then c = codim R = codim(Spec R ⊂ Spec S). If in addition Pc is a free module of rank 1, so that Pc ∼= S(−α) with α the adjunction number, then R is a Gorenstein ring of canonical weight κR = α −P ai; for my purposes, one can take this to be the definition of Gorenstein. Duality makes the resolution (1.1) symmetric: the dual complex (P•)∨ = S(R, ωS), which is iso- HomS(P•, Pc) resolves the dualising module ωR = Extc morphic to R (or, as a graded module, to R(κR) with κR = α −P ai), so that P• ∼= (P•)∨. In particular the Betti numbers bi satisfy the symmetry bc−i = bi, or Pc−i = HomS(Pi, Pc) ∼= bi Mj=1 S(−α + dij), where Pi = bi Mj=1 S(−dij). The Buchsbaum -- Eisenbud symmetriser trick [BE2] adds precision to this (this is where the assumption 1 2 ∈ S comes into play): There is a symmetric perfect pairing S2(P•) → Pc inducing the duality P• ∼= (P•)∨. The idea is to pass from P• as a resolution of R to the complex P• ⊗ P• (the total complex of the double complex) as a resolution of R ⊗S R (left derived tensor product), then to replace P• ⊗ P• by its symmetrised version S2(P•). In the double complex P• ⊗ P•, one decorates the arrows by signs ±1 to make each rectangle anticommute (to get d2 = 0). The symmetrised complex S2(P•) then involves replacing the arrows by half the sum or differ- ences of symmetrically placed arrows. (This provides lots of opportunities for confusion about signs!) 3 For details, see [BE2]. The conclusion is that P• has a ±-symmetric bilinear form that induces perfect pairings Pi ⊗ Pc−i → Pc = S for each i, compatible with the differentials. The Buchsbaum -- Eisenbud structure theorem in codimension 3 is a simple consequence of this symmetry, and a model for what I try to do in this paper. Namely, in codimension 3 we have 0 ← P0 ← P1 ← P2 ← P3 ← 0, (1.2) with P0 = S, P3 ∼= S, P2 = Hom(P1, P3) ∼= P ∨ 1 , and the matrix M defining the map P1 ← P2 is skew (that is, antisymmetric). If I set P1 = nS then the respective ranks of the differentials in (1.2) are 1, n − 1 and 1; since M is skew, his rank must be even, so that n = 2ν + 1. Moreover, the kernel and cokernel are given by the Pfaffians of M, by the skew version of Cramer's rule. Generalising the Buchsbaum -- Eisenbud Theorem to codimension 4 has been a notoriously elusive problem since the 1970s. 1.2 Main aim This paper starts by describing the shape of the resolution of a codimension 4 Gorenstein ring by analogy with (1.2). The first syzygy matrix M1 : P1 ← P2 is a (k + 1)×2k matrix whose k + 1 rows generically span a maximal isotropic space of the symmetric quadratic form on P2. The ideal IR is generated by the entries of the map L : P0 ← P1, which is determined by the linear algebra of quadratic forms as the linear relation that must hold between the k + 1 rows of M1. This is all uncomplicated stuff, deduced directly from the symmetry trick of [BE2]. It leads to the definition of the Spin-Hom varieties SpHk in the space of (k + 1) × 2k matrixes (see Section 2.4). The first syzygy matrix M1 is then an S-valued point of SpHk, or a morphism α : Spec S → SpHk. The converse is more subtle, and is the main point of the paper. By construction, SpHk supports a short complex P1 ← P2 ← P3 of free modules with a certain universal property. If we were allowed to restrict to a smooth open subscheme S0 of SpHk meeting the degeneracy locus SpHdgn in codi- mension 4, the reflexive hull of the cokernel of M1 and the kernel of M2 would provide a complex P• that resolves a sheaf of Gorenstein codimension 4 ideals in S0. (This follows by the main proof below). k 4 Unfortunately, this is only an adequate description of codimension 4 Gorenstein ideals in the uninteresting case of complete intersection ideals. Any other case necessarily involves smaller strata of SpHk, where SpHk is singular. Thus to cover every codimension 4 Gorenstein ring, I am forced into the logically subtle situation of a universal construction whose universal space does not itself support the type of object I am trying to classify, namely a Gorenstein codimension 4 ideal. See 4.3 for further discussion of this point. Main Theorem 2.5 gives the universal construction. To paraphrase: for a polynomial ring S graded in positive degrees, there is a 1-to-1 correspondence between: (1) Gorenstein codimension 4 graded ideals I ⊂ S and (2) graded morphisms α : Spec S → SpHk for which α−1(SpHdgn k mension ≥ 4 in Spec S. ) has codi- I should say at once that this is intended as a theoretical structure result. It has the glaring weakness that it does not so far make any tractable predictions even in model cases (see 4.7 for a discussion). But it is possibly better than no structure result at all. 1.3 Contents of the paper Section 2.1 describes the shape of the free resolution and its symmetry, fol- lowing the above introductory discussion. Section 2.4 defines the Spin-Hom variety SpHk ⊂ Mat(k + 1, 2k), to serve as my universal space. The defini- tion takes the form of a quasihomogeneous space for the complex Lie group G = GL(k + 1) × O(2k) or its spin double cover GL(k + 1) × Pin(2k). More explicitly, define SpHk as the closure of the G-orbit SpH0 k = G · M0 of the typical matrix M0 = ( Ik 0 0 0 ) under the given action of G = GL(k + 1)× O(2k) on Mat(k + 1, 2k). k. Once these definitions are in place, Section 2.5 states the main theorem, and proves it based on the exactness criterion of [BE1]. is the complement SpHk \ SpH0 The degeneracy locus SpHdgn k The Spin-Hom varieties SpHk have a rich structure arising from repre- sentation theory. A matrix M1 ∈ SpH0 k can be viewed as an isomorphism between a k-dimensional space in kk+1 and a maximal isotropic space for ϕ in k2k. This displays SpH0 k as a principal GL(k) bundle over Pk×OGr(k, 2k). Section 3 discusses the properties of the SpHk in more detail, notably their 5 symmetry under the maximal torus and Weyl group. The spinor and non- spinor sets correspond to the two different spinor components OGr(k, 2k) and OGr′(k, 2k) of the maximal isotropic Grassmannian. I introduce the Cramer-spinor coordinates σJ in 3.3; the main point is that, for a spinor subset J ∪ J c, the (k + 1) × k submatrix of M1 ∈ SpHk formed by those columns has top wedge factoring as (L1, . . . , Lk+1)· σ2 J where L : P0 ← P1 is the vector of equations (see Lemma 3.3.2). Ensuring that the appropriate square root σJ is defined as an element σJ ∈ S involves the point that, whereas the spinor bundle defines a 2-torsion Weil divisor on the affine orthogonal Grassmannian a OGr(k, 2k) ⊂Vk k2k (the affine cone over OGr(k, 2k) in Plucker space) and on SpHk, its birational transform under the classifying maps α : Spec S → SpHk of Theorem 2.5 is the trivial bundle on Spec S. and define k → kk+1 ⊗ k2k−1. At the same time, they an equivariant morphism SpH0 vanish on the nonspin variety SpH′ k, corresponding to the other component OGr′(k, 2k) of the Grassmannian of maximal isotropic subspaces; this has nonspinor coordinates, that vanish on SpHk. Between them, these give set theoretic equations for SpHk and its degeneracy locus. The spinor coordinates vanish on the degeneracy locus SpHdgn k The final Section 4 discusses a number of issues with my construction and some open problems and challenges for the future. 2 The main result For a codimension 4 Gorenstein ideal I with k + 1 generators, the module P2 of first syzygies is a 2k dimensional orthogonal space with a nondegenerate (symmetric) quadratic form ϕ. The k + 1 rows of the first syzygy matrix M1(R) span an isotropic subspace in P2 with respect to ϕ. Since the maxi- mal isotropic subspaces are k-dimensional, this implies a linear dependence relation (L1, . . . , Lk+1) that bases coker M1 and thus provides the generators of I. A first draft of this idea was sketched in [Ki], 10.2. 2.1 The free resolution Let S = k[x1, . . . , xN ] be the polynomial ring over an algebraically closed field k of characteristic 6= 2, graded in positive degrees. Let IR be a homo- geneous ideal with quotient R = S/IR that is Gorenstein of codimension 4; 6 equivalently, IR defines a codimension 4 Gorenstein graded subscheme V (IR) = Spec R ⊂ AN k = Spec S. Suppose that IR has k + 1 generators L1, . . . , Lk+1. It follows from the Auslander -- Buchsbaum form of the Hilbert syzygies theorem and the sym- metriser trick of Buchsbaum -- Eisenbud [BE2] that the free resolution of R is 0 ← P0 ← P1 ← P2 ← P3 ← P4 ← 0, (2.1) where P0 = S, P4 ∼= S, P3 = Hom(P1, P4) ∼= P ∨ 1 ; and moreover, P2 has a nondegenerate symmetric bilinear form ϕ : S2P2 → P4 compatible with the complex P•, so that P2 → P1 is dual to P3 → P2 under ϕ. The simple cases of 2.3, Examples 2.1 -- 2.3 give a sanity check (just in case you are sceptical about the symmetry of ϕ). A choice of basis of P2 gives ϕ the standard block form1 ( 0 I I 0 ). Then the first syzygy matrix in (2.1) is M1(R) = (A B), where the two blocks are (k + 1) × k matrixes satisfying (A B) ( 0 I I 0 ) t(A B) = 0, (2.2) that is, A tB + B tA = 0, or A tB is skew. I call this a (k + 1) × 2k resolution (meaning that the defining ideal IR has k + 1 generators yoked by 2k first syzygies). The number of equations in (2.2) is(cid:0)k+2 k = 8, the variety defined by (2.2) involves (cid:0)k+2 2 (cid:1). For example, in the typical case 2 (cid:1) = 45 quadratic equations in 2k(k + 1) = 144 variables. The scheme Vk defined by (2.2) appears in the literature as the variety of complexes. However it is not really the right object -- it breaks into 2 irreducible components for spinor reasons, and it is better to study just one, which is my SpHk. 2.2 The general fibre Let ξ ∈ Spec S = AN be a point outside V (IR) = Spec R with residue field K = k(ξ) (for example, a k-valued point, with K = k, or the generic point, 1In the graded case this is trivial because ϕ is homogeneous of degree 0, so is basically a nondegenerate quadratic form on a vector space V2 with P2 = V2 ⊗ S. See the discussion in 4.5 for the more general case. 7 with K = Frac S). Evaluating (2.1) at ξ gives the exact sequence of vector spaces 0 ← V0 ← V1 ← V2 ← V3 ← V4 ← 0 (2.3) over K, where V0 = K, V4 ∼= K, V1 = (k + 1)K, V3 = Hom(V1, V4) ∼= V ∨ 1 , and V2 = 2kK with the nondegenerate quadratic form ϕ = ( 0 I I 0 ). Over K, the maps in (2.3) can be written as the matrixes (cid:0)0 . . . 0 1(cid:1)(cid:18)Ik 0 0 0 0(cid:19)(cid:18) 0 Ik 0(cid:19) 0... 1! . 0 (2.4) This data determines a fibre bundle over AN \ V (IR) with the exact complex (2.3) as fibre, and structure group the orthogonal group of the complex, which I take to be GL(k + 1) × O(2k) or its double cover GL(k + 1) × Pin(2k). 2.3 Simple examples . −x4 . x2 . . −x4 x3 (A B) =  Example 2.1 A codimension 4 complete intersection has L = (x1, x2, x3, x4) and Koszul syzygy matrix −x4 . . x1 x3 −x2 . −x3 x1 . x2 −x1 . . . . In this choice, A = M1,2,3 has rank 3 and V3 A = x2 4 · (x1, . . . , x4). See 3.3 for spinors. A spinor subset J ∪ J c has an odd number i of columns from A and the complementary 3 − i columns from B. For example, columns 1, 5, 6 give a 4 × 3 matrix with V3 M1,5,6 = x2 1 · (x1, x2, x3, x4). Example 2.2 Another easy case is that of a hypersurface section h = 0 in a codimension 3 ideal given by the Pfaffians Pf i of a (2l + 1) × (2l + 1) skew matrix M. The syzygy matrix is   . (2.5) (A B) = −hI2l+1 Pf 1 . . . Pf 2l+1 M 0 . . . 0! . (2.6) One sees that a spinor σJ corresponding to 2l + 1 − 2i columns from A and a complementary 2i from B is of the form hl−i times a diagonal 2i × 2i Pfaffian of M. Thus the top wedge of the left-hand block A of (2.6) equals σ2 · (h, Pf 1, . . . , Pf 2l+1) where σ = hl. 8 Example 2.3 The extrasymmetric matrix a b c   d e f e f g h h i −λa −λb −λc   M = (2.7) with a single multiplier λ is the simplest case of a Tom unprojection (see [TJ], Section 9 for details). Let I be the ideal generated by the 4 × 4 Pfaffians of M. The diagonal entries d, g, i of the 3 × 3 symmetric top right block are all unprojection variables; thus i appears linearly in 4 equations of the form i · (a, d, e, g) = · · · , and eliminating it projects to the codimension 3 Gorenstein ring defined by the Pfaffians of the top left 5 × 5 block. If λ ∈ S is a perfect square, I is the ideal of Segre(P2 × P2) ⊂ P8 up to a coordinate change, but the Galois symmetry √λ 7→ −√λ swaps the two factors. See [TJ], Section 9 for more details, and for several more families of examples; in any of these cases, writing out the resolution matrixes (A B) with the stated isotropy property makes a demanding but rewarding exercise for the dedicated student. By extrasymmetry, out of the 15 entries of M, 9 are independent and 6 repeats. His 4 × 4 Pfaffians follow a similar pattern. I write the 9 generators of the ideal I of Pfaffians as the vector L = (cid:2)λac + eh − f g, −λab − dh + ef, λa2 + dg − e2, ah − bg + ce, −af + be − cd, λb2 + di − f 2, λbc + ei − f h, λc2 + gi − h2, ai − bh + cf(cid:3) 9 Its matrix of first syzygies M1 is the transpose of . . . . . . . . (2.8) e g h −λa . d e f . λa . a b −a . c −b −c . −d −e −f −e −g −h −h . . λc . −λb f −h . . f . i . . i . h . . . . . −b . −c . c . . i . . . . . . . . i . . . −h . . . . . . . . . . . . . . . g −e d . e −d . . −h h −f . . −c c −b −a . . −a . . . . −a . . λc −g −λb . . −λc . i f h . f −λc λb . e −d −λa b −h f . d e g M1 is of block form (A B) with two 9×8 blocks, and one checks that LM1 = 0, and M1 is isotropic for the standard quadratic form J = ( 0 I I 0 ), so its kernel is M2 =(cid:0) tB tA(cid:1). The focus in (2.8) is on i as an unprojection variable, multiplying d, e, g, a. One recognises its Tom3 matrix as the top 5 × 5 block, and the Koszul syzygy matrix of d, e, g, a as Submatrix([6, 7, 8, 14, 15, 16], [6, 7, 8, 9]); compare [KM]. For some of the spinors (see Section 3), consider the 8 × 9 submatrixes formed by 4 out of the first 5 rows of (2.8), and the complementary 4 rows from the last 8. One calculates their maximal minors with a mild effort: V8 M1,2,3,4,13,14,15,16 = a2(af − be + cd)2 · L, V8 M1,2,3,5,12,14,15,16 = a2(ah − bg + ce)2 · L, V8 M1,2,4,5,11,14,15,16 = a2(−λa2 − dg + e2)2 · L, V8 M1,3,4,5,10,14,15,16 = a2(−λab − dh + ef )2 · L, V8 M2,3,4,5,9,14,15,16 = a2(−λac − eh + f g)2 · L. (2.9) The factor a comes from the 3 × 3 diagonal block at the bottom right, and the varying factors are the 4 × 4 Pfaffians of the first 5 × 5 block. Compare 4.4 for a sample Koszul syzygy. 10 Exercise 2.4 Apply column and isotropic row operations to put the vari- able f down a main diagonal of B; check that this puts the complementary A in the form of a skew 8 × 8 matrix and a row of zeros. Hint: order the rows as 15, 16, 12, 11, 6, 2, 1, 5, 7, 8, 4, 3, 14, 10, 9, 13 and the columns as 1, 2,−3, 4, 5,−7, 8, 9, 6. (See the website for the easy code.) Do the same for either variable e, h, and the same for any of a, b, c (involving the multiplier λ). Thus the isotropy condition tMJM can be thought of as many skew sym- metries. These examples provide useful sanity checks, with everything given by transparent calculations; it is reassuring to be able to verify the symmetry of the bilinear form on P2 asserted in Proposition 1, the shape of A tB in (2.2), which parity of J gives nonzero spinors σJ , and other minor issues of this nature. I have written out the matrixes, spinors, Koszul syzygies etc. in a small number of more complicated explicit examples (see the website). It should be possible to treat fairly general Tom and Jerry constructions in the same style, although so far I do not know how to use this to predict anything useful. The motivation for this paper came in large part from continuing attempts to understand Horikawa surfaces and Duncan Dicks' 1988 thesis [Di], [R1]. 2.4 Definition of the Spin-Hom variety SpHk Define the Spin-Hom variety SpHk ⊂ Mat(k + 1, 2k) as the closure under G = GL(k + 1) × O(2k) of the orbit of M 0 = (cid:0) Ik 0 0 0(cid:1), the second matrix in (2.4). It consists of isotropic homomorphisms V1 ← V2, in other words matrixes M1 whose k + 1 rows are isotropic and mutually orthogonal vectors in V2 w.r.t. the quadratic form ϕ, and span a subspace that is in the given component of maximal isotropic subspaces if it is k-dimensional. In more detail, write SpH0 k = G· M 0 ⊂ Mat(k + 1, 2k) for the orbit, SpHk for its closure, and SpHdgn k for the degeneracy locus, consisting of matrixes of rank < k. Section 3 discusses several further properties of SpHk and its degeneracy locus SpHdgn k = SpHk \ SpH0 . k 11 2.5 The Main Theorem Assume that S is a polynomial ring graded in positive degrees. Let I be a homogeneous ideal defining a codimension 4 Gorenstein subscheme X = V (I) ⊂ Spec S. Then a choice of minimal generators of I (made up of k + 1 elements, say) and of the first syzygies between these defines a morphism α : Spec S → SpHk such that α−1(SpHdgn) has the same support as X, and hence codimension 4 in Spec S. Conversely, let α : Spec S → SpHk ⊂ Mat(k + 1, 2k) be a morphism for which α−1(SpHdgn) has codimension ≥ 4 in Spec S. Assume that α is graded, that is, equivariant for a positively graded action of Gm on SpHk ⊂ Mat(k + 1, 2k). Let M1 = (A B) be the matrix image of α (the matrix entries of M1 or the coordinates of α are elements of S). Then by construction M1 and J tM1 define the two middle morphisms of a complex. I assert that this extends to a complex 0 ← P0 L←− P1 M1←−− P2 J tM1←−−− P3 tL←− P4 ← 0. (2.10) in which P0, P4 ∼= S, the complex is exact except at P0, and the image of L = (L1, . . . , Lk+1) generates the ideal of a Gorenstein codimension 4 subscheme X ⊂ Spec S. 2.6 Proof The first part follows from what I have already said. The converse follows by a straightforward application of the exactness criterion of [BE1]. The complex P• of (2.10) comes directly from M1. Namely, define P0 as the reflexive hull of coker{M1 : P1 ← P2} (that is, double dual); it has rank 1 because M1 has generic rank k. A graded reflexive module of rank 1 over a graded regular ring is free (this is the same as saying that a Weil divisor on a nonsingular variety is Cartier), so P0 ∼= S. Given P3 ∼= P ∨ 1 , the generically surjective map S ∼= P0 ← P1 is dual to an inclusion S ֒→ P3 that maps to the kernel of P2 ← P3. The key point is to prove exactness of the complex P0 ϕ1←− P1 ϕ2←− P2 ϕ3←− P3 ϕ4←− P4 ← 0, where I write ϕ1 = (L1, . . . , Lk+1), ϕ2 = M1, etc. to agree with [BE1]. The modules and homomorphisms P0, ϕ1, P1, ϕ2, P2, ϕ3, P3, ϕ4, P4 of this complex 12 have respective ranks 1, 1, k + 1, k, 2k, k, k + 1, 1, 1, which accords with an exact sequence of vector spaces, as in (2.3 -- 2.4); this is Part (1) of the criterion of [BE1], Theorem 1. The second condition Part (2) requires the matrixes of ϕi to have maximal nonzero minors generating an ideal I(ϕi) that contains a regular sequence of length i. However, P• is exact outside the degeneracy locus, that is, at any point ξ ∈ Spec S for which α(ξ) /∈ SpHdgn , and by assumption, the locus of such points has codimension ≥ 4. Thus the maximal minors of each ϕi generate an ideal defining a subscheme of codimension ≥ 4. In a Cohen -- Macaulay ring, an ideal defining a subscheme of codimension ≥ i has height ≥ i. Q.E.D. k 3 Properties of SpHk and its spinors This section introduces the spinors as sections of the spinor line bundle S on SpHk. The nonspinors vanish on SpHk and cut it out in Vk set theoretically. The spinors vanish on the other component SpH′ k and cut out set theoretically the degeneracy locus SpHdgn in SpHk. k The easy bit is to say that a spinor is the square root of a determinant on Vk ⊂ Mat(k + 1, 2k) that vanishes to even order on a divisor of SpHk because it is locally the square of a Pfaffian. The ratio of two spinors is a rational function on SpHk. The tricky point is that the spinors are sections of the spinor bundle S on SpHk that is defined as a Pin(2k) equivariant bundle, so not described by any particularly straightforward linear or multilinear algebra. As every- one knows, the spinor bundle S on OGr(k, 2k) is the ample generator of Pic(OGr(k, 2k)), with the property that S ⊗2 is the restriction of the Plucker bundle O(1) on Gr(k, 2k). On the affine orthogonal Grassmannian in Plucker space a Gr(k, 2k) ⊂ Vk k2k, it corresponds to a 2-torsion Weil divisor class. I write out a transparent treatment of the first example in 3.2. I need to argue that the spinors pulled back to my regular ambient Spec S by the appropriate birational transform are elements of S (that is, poly- nomials), rather than just sections of a spinor line bundle. The reason that I expect to be able to do this is because I have done many calculations like the Tom unprojection of 2.3, Example 2.3, and it always works. In the final analysis, I win for the banal reason that the ambient space Spec S has no 2-torsion Weil divisors in its class group (because S is factorial), so that the 13 birational transform of the spinor bundle S to Spec S = AN is trivial. The Cramer-spinor coordinates of the syzygy matrix M1 = (A B) have the potential to clarify many points about Gorenstein codimension 4: the generic rank of M1 is k, but it drops to k − 3 on Spec R; its k × k minors are in I 3 R. There also seems to be a possible explanation of the difference seen in examples between k even and odd in terms of the well known differences between the Weyl groups Dk (compare 3.1.3). 3.1 Symmetry View GL(k+1) as acting on the first syzygy matrix M1(R) by row operations, and O(2k) as column operations preserving the orthogonal structure ϕ, or the matrix ( 0 I m and Weyl group Ak = Sk+1 of the first factor GL(k + 1) act in the obvious way by scaling and permuting the rows of M1. I 0 ). The maximal torus Gk+1 I need some standard notions for the symmetry of O(2k) and its spinors. For further details, see Fulton and Harris [FH], esp. Chapter 20 and [CR], Section 4. Write V2 = k2k for the 2k dimensional vector space with basis e1, . . . , ek and dual basis f1, . . . , fk, making the quadratic form ϕ = ( 0 I I 0 ). Write U = U k = he1, . . . , eki, so that V2 = U ⊕ U ∨. The orthogonal Grass- mannian OGr(k, 2k) is defined as the variety of k-dimensional isotropic sub- spaces that intersect U in even codimension, that is, in a subspace of dimen- sion ≡ k modulo 2. 3.1.1 The Dk symmetry of OGr(k, 2k) and SpHk i I describe the Dk Weyl group symmetry of the columns in this notation (compare [CR], Section 4). The maximal torus Gk m of O(2k) multiplies ei by λi and fi by λ−1 , and acts likewise on the columns of M1 = (A B). The Weyl group Dk acts on the ei, fi and on the columns of M1 = (A B) by permutations, as follows: the subgroup Sk permutes the ei simultaneously with the fi; and the rest of Dk swaps evenly many of the ei with their corresponding fi, thus taking U = he1, . . . , eki to another coordinate k-plane in OGr(k, 2k). Exercise: The younger reader may enjoy checking that the k − 1 permutations si = (i, i + 1) = (eiei+1)(fifi+1) together with sk = (ekfk+1)(ek+1fk) are involutions satisfying the standard Coxeter relations of type Dk, especially (sk−1sk)2 = 1 and (sk−2sk)3 = 1. 14 3.1.2 Spinor and nonspinor subsets The spinor sets J ∪ J c index the spinors σJ (introduced in 3.3). Let {ei, fi} be the standard basis of k2k with form ϕ = ( 0 I I 0 ). There are 2k choices of maximal isotropic subspaces of k2k based by a subset of this basis; each is based by a subset J of {e1, . . . , ek} together with the complementary subset J c of {f1, . . . , fk}. The spinor subsets are those for which #J has the same parity as k, or in other words, the complement #J c is even; the nonspinor subsets are those for which #J has the parity of k−1. The spinor set indexes a basis σJ of the spinor space of OGr(k, 2k), and similarly, the nonspinor set indexes the nonspinors σ′ J ′ of his dark twin OGr′(k, 2k). The standard affine piece of OGr(k, 2k) consists of k-dimensional spaces based by k vectors that one writes as a matrix (I A) with A a skew k × k matrix. The spinor coordinates of (I A) are the 2i × 2i diagonal Pfaffians of A for 0 ≤ i ≤ [ k 2 ]. They correspond in an obvious way to the spinor sets just defined and they are the spinors apart from the quibble about taking an overall square root and what bundle they belong to. 3.1.3 Even versus odd The distinction between k even or odd is crucial for anything to do with O(2k), Dk, spinors, Clifford algebras, etc. The spinor and nonspinor sets correspond to taking a subset J of {e1, . . . , ek} and the complementary set J c of {f1, . . . , fk}. The 2k choices correspond to the vertices of a k-cube. When k is even this is a bipartite graph; the spinors and nonspinors form the two parts. By contrast, for odd k, both spinors and nonspinors are indexed by the vertices of the k-cube divided by the antipodal involution ([CR], Section 4 writes out the case k = 5 in detail). For simplicity, I assume that k is even in most of what follows; the com- mon case in applications that I really care about is k = 8. Then J = ∅ and J c = {1, . . . , k} is a spinor set, and the affine pieces represented by (I X) and (Y I) (with skew X or Y ) are in the same component of OGr(k, 2k). The odd case involves related tricks, but with some notable differences of detail (compare [CR], Section 4). 3.1.4 The other component OGr′ and SpH′ k I write OGr′(k, 2k) for the other component of the maximal isotropic Grass- mannian, consisting of subspaces meeting U in odd codimension. Swapping 15 oddly many of the ei and fi interchanges OGr and OGr′. Likewise, SpH′ the closure of the G-orbit of the matrix M ′ corresponding pair of columns of M0. k is 0 obtained by interchanging one Claim 3.1 Write Vk for the scheme defined by (2.2) (that is, the "variety of complexes"). It has two irreducible components Vk = SpHk ∪ SpH′ k containing matrixes of maximal rank k. The two components are generically reduced and intersect in the degenerate locus SpHdgn . (But one expects Vk to have embedded primes at its smaller strata, as in the discussion around (3.5).) k This follows from the properties of spinor minors ∆J discussed in Exer- cise 3.2.1: the ∆J are k × k minors defined as polynomials on Vk, and vanish on SpH′ k but are nonzero on a dense open subset of SpHk. 3.2 A first introduction to OGr(k, 2k) and its spinors The lines on the quadric surface provide the simplest calculation, and already have lots to teach us about OGr(2, 4) and OGr(k, 2k): the conditions for the 2 × 4 matrix (3.1) N =(cid:18)a b c d x y z t(cid:19) to be isotropic for ( 0 I I 0 ) are ax + by = 0, az + bt + cx + dy = 0, cz + dt = 0. (3.2) Three equations (3.2) generate an ideal IW defining a codimension 3 complete intersection W ⊂ A8 that breaks up into two components Σ⊔Σ′, correspond- ing to the two pencils of lines on the quadric surface: the two affine pieces of OGr(2, 4) that consist of matrixes row equivalent to (I A) or (A I), with A a skew matrix, have one of the spinor minors ∆1 = ad − bc or ∆2 = xt − yz nonzero, and dx − bz = at − cy = 0 and dy − bt = −(az − cx) (3.3) on them. This follows because all the products of ∆1, ∆2 with the nonspinors minors dx− bz, at− cy are in IW , as one checks readily. Thus if ∆1 6= 0 (say), I can multiply by the adjoint of the first block to get (cid:18) d −b −c a (cid:19)(cid:18)a b x y c d z t(cid:19) =(cid:18)∆1 0 ∆1 az − cx at − cy(cid:19) dx − bz dy − bt 0 (3.4) 16 where the second block is skew. Note that (3.5) If ∆1 6= 0, the relations (3.2) imply that we are in Σ. The ideal of Σ is obtained from (3.2) allowing cancellation of ∆1; in other words IΣ = [IW : ∆1] is the colon ideal with either of the spinor minors ∆1 or ∆2. ∆1 ·(cid:0)∆1∆2 − (az − cx)2(cid:1) ∈ IW . The second block in (3.4) is only skew mod IW after cancelling one of a, b, . . . , t; similarly ∆1∆2 − (az − cx)2 /∈ IW , so that (3.5) involves cancelling ∆1. Thus a geometric description of Σ, Σ′ ⊂ Mat(k, 2k) should usually lead to ideals with embedded primes at their intersection or its smaller strata. Now by relation (3.5), the Plucker embedding takes OGr(2, 4) to the conic XZ = Y 2, with X = ∆1 = ad − bc, Y = az − cx, Z = ∆2 = xt − yz. This is (P1,O(2)) parametrised by u2, uv, v2 where u, v base H 0(P1,O(1)). Thus X = u2, Y = uv and Z = v2 on OGr(2, 4); the spinors are u and v. The ratio u : v equals X : Y = Y : Z. Each of ∆1 and ∆2 vanishes on a double divisor, but the quantities u = √∆1, v = √∆2 are not themselves polynomial. The conclusion is that the minors ∆1 and ∆2 are spinor squares, that is, squares of sections u, v of a line bundle S, the spinor bundle on OGr(2, 4). If we view OGr(2, 4) as a subvariety of Gr(2, 4), only S ⊗2 extends to the Plucker line bundle O(1). Embedding OGr(2, 4) in the Plucker space P(V2 C4) and taking the affine cone gives the affine spinor variety a OGr(2, 4) as the cone over the conic, and S with its sections u, v as the ruling. cones in linearly disjoint vector subspaces of the Plucker space V2 C4, and the spinor bundle on the union has a divisor class that is a 2-torsion Weil divisor on each component. This picture is of course the orbifold quotient of ±1 acting on two planes A2 meeting transversally in A4. In fact a OGr(2, 4) and his dark twin a OGr′ are two ordinary quadric 3.2.1 Exercise Generalise the above baby calculation to the subvariety Wk ⊂ Mat(k, 2k) of matrixes (A X) whose k rows span an isotropic space for ( 0 I I 0 ), or in equations, the k × k product A tX is skew. Assume k is even. (1) Wk ⊂ Mat(k, 2k) is a complete intersection subvariety of codimension (cid:0)k+1 2 (cid:1). [Hint: Just a dimension count.] (2) Wk breaks up into two irreducible components Σ∪Σ′, where Σ contains the space spanned by (I X) with X skew, or more generally, by the span 17 of the columns J ∪ J c for J a spinor set; its nondegenerate points form a principal GL(k) bundle over the two components OGr⊔ OGr′ of the maximal isotropic Grassmannian. (3) For J a spinor set, the k×k spinor minor ∆J of (A X) (the determinant of the submatrix formed by the columns J ∪ J c) is a polynomial on Mat(k × 2k) that vanishes on Σ′, and vanishes along a double divisor of Σ, that is, twice a prime Weil divisor DJ . (4) The Weil divisors DJ1 and DJ2 corresponding to two spinor sets J1 and J2 are linearly equivalent. [Hint: First suppose that J1 is obtained from J by exactly two transpositions, say (e1f2)(e2f1), and argue as in (3.5) to prove that σJ σJ1 restricted to Σ is the square of either minor obtained by just one of the transpositions.] 3.2.2 Spinors on OGr(k, 2k) The orthogonal Grassmann variety OGr(k, 2k) has a spinor embedding into P(k2k−1), of which the usual Plucker embedding OGr(k, 2k) ⊂ Gr(k, 2k) ֒→ P(cid:16) k ^ k2k(cid:17) is the Veronese square. The space of spinors k2k−1 is a representation of the spin double cover Pin(2k) → O(2k). A point W ∈ OGr(k, 2k) is a k-dimensional subspace W k ⊂ k2k isotropic for ( 0 I I 0 ) and intersecting U = he1, . . . , eni in even codimension. I can write a basis as the rows of a k × 2k matrix NW . If I view W as a point of Gr(k, 2k), its Plucker coordinates are all the k × k minors of NW . There are (cid:0)2k k(cid:1) of these (that is, 12870 if k = 8), a fraction of which vanish OGr(k, 2k), as the determinant of a skew matrix of odd size. The finer embedding of OGr(k, 2k) is by spinors. The spinors σJ are sections of the spinor line bundle S, 2k−1 of them (which is 128 if k = 8, about 1/100 of the number of Plucker minors). Each comes by taking a k× k submatrix formed by a spinor subset of columns of NW (in other words, restricting to an isotropic coordinate subspace of k2k in the specified com- ponent OGr(k, 2k)), taking its 2κ × 2κ minor (where κ =(cid:2) k 2(cid:3)) and factoring it as the perfect square of a section of S. The only general reason for a 2κ × 2κ minor to be a perfect square is that the submatrix is skew in some 18 basis; in fact, as in (3.4), after taking one fixed square root of a determinant, and making a change of basis, the maximal isotropic space can be written as (I X) with X skew, and the spinors are all the Pfaffians of X. 3.3 Cramer-spinor coordinates on SpHk 3.3.1 Geometric interpretation A point of the open orbit SpH0 k ⊂ SpHk is a matrix M of rank k; it defines an isomorphism from a k-dimensional subspace of V1 (the column span of M) to its row span, a maximal isotropic subspace of V2 in the specified component OGr(k, 2k). Therefore the nondegenerate orbit SpH0 k ⊂ SpHk has a morphism to P(V ∨ 1 ) × OGr(k, 2k) that makes it a principal GL(k) bundle. The product P(V ∨ 1 )× OGr(k, 2k) is a projective homogeneous space under G = GL(k + 1) × Pin(2k) It embeds naturally in the projectivisation of kk+1⊗k2k−1, with the second factor the space of spinors. This is the representation of G with highest weight vector v = (0, . . . , 0, 1) ⊗ (1, 0, . . . , 0). The composite SpH0 k → P(V ∨ 1 ) × OGr(k, 2k) ֒→ P(kk+1 ⊗ k2k−1 ) (3.6) takes the typical matrix M0 (or equivalently, the complex (2.4)) to v. The Cramer-spinor coordinates of α ∈ SpHk(S) are the bihomogeneous coordinates under the composite map (3.6). 3.3.2 Spinors as polynomials The spinors σJ occur naturally as sections of the spinor line bundle S on OGr(k, 2k), and so have well defined pullbacks to SpH0 k or to any scheme T with a morphism α : T → SpH0 k. For σJ to be well defined in H 0(OT ), the pullback of the spinor line bundle to T must be trivial. Lemma 3.2 Let α ∈ Mor(Spec S, SpHk) = SpHk(S) be a classifying map as in Theorem 2.5 and write M1 ∈ Mat(S, k + 1, 2k) for its matrix (with entries in S). Then for a spinor set J ∪ J c (as in 3.1.2), the (k + 1) × k submatrix NJ of M1 with columns J ∪ J c has ^k where L = (L1, . . . , Lk+1) generates the cokernel of M1, and σJ ∈ S. NJ = L · σ2 J , (3.7) 19 3.4 Proof A classifying map α ∈ SpHk(S) as in Theorem 2.5 restricts to a morphism α from the nondegenerate locus Spec S \ V (IR) to SpH0 k; on the complement of V (IR), the matrix M1 has rank k, and its kth wedge defines the composite morphism to the product Pk × Gr(k, 2k) in its Segre embedding: Spec S \ V (IR) → SpH0 k → Pk × OGr(k, 2k) ֒→ Pk × Gr(k, 2k) ⊂ P(cid:16)kk+1 ⊗^k V 2k(cid:17) . (3.8) J already on the level of Pk × OGr(k, 2k). The entries of Vk NJ are k + 1 coordinates of this morphism, and are of the form Li · σ2 Note that Spec S \ V (IR) is the complement in Spec S = AN of a subset of codimension ≥ 4 so has trivial Pic. Each maximal minor of NJ splits as Li times a polynomial that vanishes on a divisor that is a double (because it is the pullback of the square of a spinor); therefore the polynomial is a perfect square in S. QED The following statement is the remaining basic issue that I am currently unable to settle in general. Conjecture 3.3 Under the assumptions of Lemma 3.3.2, σJ ∈ IR. This is clear when R is reduced, that is, IR is a radical ideal. Indeed if σJ is a unit at some generic point ξ ∈ V (IR) = Spec R, then (3.7) implies that IR is generated at ξ by the k × k minors of the (k + 1) × k matrix NJ ; these equations define a codimension 2 subscheme of Spec S, which is a contradiction. This case is sufficient for applications to construction of ordinary varieties, but not of course to Artinian subschemes of A4. The conjecture also holds under the assumption that IR is generically a codimension 4 complete intersection. Indeed, the resolution of IR near any generic point ξ ∈ V (IR) is then the 4 × 6 Koszul resolution of the complete intersection direct sum some nonminimal stuff that just add invertible square matrix blocks. Then both the Li and the σJ are locally given by Example 2.1. At present, the thing that seems to make the conjecture hard is that the definition of the σJ and the methods currently available for getting formulas for them consists of working on the nondegenerate locus of SpHk: choose a block diagonal form and take the Pfaffian of a skew complement, . . . This is just not applicable at points σ ∈ V (IR). 20 The conjecture could possibly be treated by a more direct understanding of the spin morphism Spec S → k2k defined by spinors and nonspinors, not passing via the square root of the Plucker morphism as I do implicitly in Lemma 1 by taking Vk. 4 Final remarks, open problems 3k2+k 4.1 Birational structure and dimension of SpHk A general M = (A B) ∈ SpHk has k + 1 rows that span a maximal isotropic space U ∈ OGr(k, 2k) and 2k columns that span a k-dimensional vector subspace of kk+1, that I can view as a point of Pk; thus SpH0 k is a principal GL(k) bundle over Pk×OGr(k, 2k). In particular, dim SpHk = k2 +k +(cid:0)k 2(cid:1) = The tangent space to SpHk at the general point M0 =(cid:0) Ik 0 0 0(cid:1) is calculated by writing an infinitely near matrix as M0 +(cid:16) A′ ak+1 bk+1(cid:17); here the blocks A′ k are k × k matrixes, and ak+1 and bk+1 are 1 × k rows. Then the tangent space to Vk defined by A tB = 0 is the affine subspace obtained by setting B′ k to be skew and bk+1 = 0. Therefore SpHk has codimension k and B′ k B ′ . 2 k (cid:0)k+1 2 (cid:1) + k and dimension 2k(k + 1) −(cid:0)k+1 It is interesting to observe that equations (2.2) express SpHk ∪ SpH′ an almost complete intersection. Namely, (2.2) is a set of (cid:0)k+1 2 (cid:1) equations in (cid:0)k+1 2 (cid:1) − 1. A2k(k+1) vanishing on a variety of dimension 3k2+k 2 (cid:1) − k = 3k2+k , that is, of codimension k as . 2 2 4.2 Intermediate rank The Spin-Hom variety SpHk certainly contains degenerate matrixes M1 of rank k − 1 or k − 2, but any morphism Spec S → SpHk that hits one of these must hit the degeneracy locus in codimension ≤ 3, so does not correspond to anything I need here. The following claim must be true, but I am not sure where it fits in the logical development. Claim 4.1 Every point P ∈ SpHk corresponds to a matrix M1 = (A B) of If a morphism α : Spec S → SpHk takes ξ to a matrix M1 of rank ≤ k. rank k + 1 − i for i = 1, 2, 3, 4 then α−1(SpHdgn ) has codimension ≤ i in a neighbourhood of ξ. In other words, a morphism α that is regular in the k 21 sense of my requirement never hits matrixes M1 of rank intermediate between k and k − 3; and if α is regular then α−1(SpHdgn ) has codimension exactly 4. k 4.3 The degeneracy locus as universal subscheme k The proof in 2.6 doesn't work for SpHk itself in a neighbourhood of a point of SpHdgn , because taking the reflexive hull, and asserting that P0 is locally free works only over a regular scheme. Moreover, it is not just the proof that goes wrong. I don't know what happens over the strata of SpHdgn k where M1 drops rank by only 1 or 2. We discuss the speculative hope that SpHdgn k ⊂ SpHk has a description as a kind of universal codimension 4 subscheme, with the inclusions enjoying some kind of Gorenstein adjunction properties. But if this is to be possible at all, we must first discard uninteresting components of SpHdgn corresponding to matrixes of intermediate rank k − 1 or k − 2. It is possible that there is some universal blowup of some big open in SpHk that supports a Gorenstein codimension 4 subscheme and would be a uni- versal space in a more conventional sense. Or, as the referee suggests, there might be a more basic sense in which appropriate codimension 4 components Γ of the degeneracy locus are universal Gorenstein embeddings, meaning that the adjunction calculation ωΓ = Ext4 OSpH(OΓ, ωSpH) for the dualising sheaf is locally free and commutes with regular pullbacks. k 4.4 Koszul syzygies Expressing the generators of I as a function of the entries of the syzygy matrix is essentially given by the map V2 P1 → P2 that writes the Koszul syzygies as linear combinations of the minimal syzygies. The Li are certainly linear combinations of the entries of M1. More precisely, since the 2k columns of M1 provide a minimal basis for the syzygies, they cover in particular the Koszul syzygies Li · Lj − Lj · Li ≡ 0. This means that for every i 6= j there is column vector vij with entries in S such that M1vij = (. . . , Lj, . . . , Li, . . . ) is the column vector with Lj in the ith place and Li in the jth and 0 elsewhere. For example, referring to Example 2.3, you might enjoy the little exercise in linear algebra of finding the vector v = (−λc, λb, 0, 0, 0, d, e, g, 0, 0, 0, 0, 0, 0, 0, 0) for which v tM1 = (−λab − dh + ef,−λac − eh + f g, 0, 0, 0, 0, 0, 0, 0), 22 where tM1 is the matrix of (2.8), and similarly for 35 other values of i, j. 4.5 More general ambient ring S I restrict to the case of ideals in a graded polynomial ring over a field of characteristic 6= 2 in the belief that progress in this case will surely be followed by the more general case of a regular local ring. Then P2 is still a free module, with a perfect symmetric bilinear form S2(P2) → P4, with respect to which P1 ← P2 is the dual of P2 ← P3. This can be put in the form ( 0 I I 0 ) over the residue field k0 = S/mS of S if we assume that k(S) is algebraically closed and contains 1 2; we can do the same over S itself if we assume that S is complete (to use Hensel's Lemma). At some point if we feel the need for general regular rings, we can probably live with a perfect quadratic form ϕ and the dualities it provides, without the need for the normal form ( 0 I I 0 ). 4.6 More general rings and modules Beyond the narrow question of Gorenstein codimension 4, one could ask for the structure of any free resolution of an S-module M or S-algebra R. As in 2.2, one can say exactly what the general fibre is, and think of the complex P• as a fibre bundle over S \ Supp M with some product of linear groups as structure group. If we are doing R-algebras, the complex P• also has a symmetric bilinear structure, that reduces the structure group. My point is that if we eventually succeed in making some progress with Gorenstein codimension 4 rings, we might hope to also get some ideas about Cohen -- Macaulay codimension 3 and Gorenstein codimension 5. For example, in vague terms, there is a fairly clear strategy how to find a key variety for the resolution complexes of Gorenstein codimension 5 ideals, by analogy with my Main Theorem 2.5. In this case, the resolution has the shape 0 ← P0 ← P1 ← P2 ← P3 ← P4 ← P5 ← 0, (4.1) with P0 = S, P1 = (a + 1)S, P2 = (a + b)S and P3, . . . , P5 their duals. The complex is determined by two syzygy matrixes M1 ∈ Mat(a + 1, a + b) of generic rank a defining P1 ← P2 and a symmetric (a+b)×(a+b) matrix M2 of generic rank b defining P2 ← P3 = P ∨ 2 , constrained by the complex condition 0 0 ), M2 =(cid:0) 0 0 0 Ib(cid:1), M1M2 = 0. The "general fibre" is given by the pair M1 = ( Ia 0 the appropriate key variety is its closed orbit under GL(a + 1) × GL(a + b). 23 The maximal nonzero minors of M1 and M2 define a map to a highest weight orbit in a a b Hom(cid:16) ^ P2, ^ P1(cid:17) × Sym2(cid:16) ^ P2(cid:17). 4.7 Difficulties with applications I expand what the introduction said about the theory currently not being ap- plicable. We now possess hundreds of constructions of codimension 4 Goren- stein varieties, for example, the Fano 3-folds of [TJ], but their treatment (for example, as Kustin -- Miller unprojections) has almost nothing to do with the structure theory developed here. My Main Theorem 2.5 does not as it stands construct anything, because it does not say how to produce mor- phisms α : Spec S → SpHk, or predict their properties. The point that must be understood is not the key variety SpHk itself, but rather the space of morphisms Mor(Spec S, SpHk), which may be intractable or infinitely com- plicated (in the sense of Vakil's Murphy's law [Va]); there are a number of basic questions here that I do not yet understand. Even given α, we do not really know how to write out the equations (L1, . . . , Lk+1), other than by the implicit procedure of taking hcfs of k × k minors. One hopes for a simple formula for the defining relations Li as a function of the first syzygy matrix M1 = (A B). Instead, one gets the vector (L1, . . . , Lk+1) by taking out the highest common factor from Vk MI for any spinor subset I, asserting that it is a perfect square σ2 that as it stands this is only implicitly a formula for the Li. J . The disadvantage is 4.8 Obstructed constructions One reason that Mor(S, SpHk) is complicated is that the target is big and singular and needs many equations. However, there are also contexts in which S-valued points of much simpler varieties already give families of Gorenstein codimension 4 ideals that are obstructed in interesting ways. b1 b2 b3 b4 ) with entries in a regular ring S, the Given a 2× 4 matrix A = ( a1 a2 a3 a4 6 equations V2 A = 0 define a Cohen -- Macaulay codimension 3 subvariety V ⊂ Spec S. An elephant X ∈ −KV is then a Gorenstein subvariety of codimension 4 with a 9× 16 resolution. If we are in the "generic" case with 8 independent indeterminate entries, V is the affine cone over Segre(P1 × P3), and X is a cone over a divisor of bidegree (k, k + 2) in Segre(P1 × P3). 24 Although X ⊂ V is a divisor, if we are obliged to treat it by equations in the ambient space Spec S, it needs 3 equations in "rolling factors format". The general case of this is contained in Dicks' thesis [Di], [R1]: choose two vectors m1, m2, m3, m4 and n1, n2, n3, n4, and assume that the identity holds as an equality in the ambient ring S. Then the 3 equations X aini ≡X bimi X aimi =X bimi ≡X aini =X bini = 0 define a hypersurface X ⊂ V that is an elephant X ∈ −KV and thus a Gorenstein subvariety with 9 × 16 resolution. The problem in setting up the data defining X is then to find solutions in S of (4.2). In other words, these are S-valued points of the affine quadric cone Q16, or morphisms Spec S → Q16. How to map a regular ambient space to the quadratic cone Q16 is a small foretaste of the more general problem of the classifying map Spec S → SpHk. This case is discussed further in [Ki], Example 10.8, which in particular writes out explicitly the relation between (4.3) and the classifying map Spec S → SpHk of Theorem 2.5. There are many quite different families of solutions to this problem, de- pending on what assumptions we make about the graded ring S, and how general we take the matrix A to be; different solutions have a number of important applications to construction and moduli of algebraic varieties, in- cluding my treatment of the Horikawa quintic n-folds. Another illustration of the phenomenon arises in a recent preprint of Catanese, Liu and Pignatelli [CLP]. Take the 5 × 5 skew matrix (4.2) (4.3) (4.4) z1 v u z2 D y m25 l m35 m45 M =    with entries in a regular ring S0, and suppose that v, u, z2, D forms a regular sequence in S. Assume that the identity z1m45 − ym35 + lm25 ≡ av + bu + cz2 + dD (4.5) holds as an equality in S0. The identity (4.5) puts the Pfaffian Pf 23.45 in the ideal (v, u, z2, D); the other 4 Pfaffians are in the same ideal for the trivial reason that every term involves one entry from the top row of M. 25 This is a new way of setting up the data for a Kustin -- Miller unprojection: write Y ⊂ Spec S0 for the codimension 3 Gorenstein subscheme defined by the Pfaffians of M. It contains the codimension 4 complete intersection V (v, u, z2, D) as a codimension 1 subscheme, and unprojecting V in Y adjoins an unprojection variable x2 having 4 linear equations x2 · (v, u, z2, D) = · · · , giving a codimension 4 Gorenstein ring with 9 × 16 resolution. The problem of how to fix (4.5) as an identity in S0 is again a question of the S0-valued points of a quadric cone, this time a quadric Q14 of rank 14. [CLP], Proposition 5.13 find two different families of solutions, and exploit this to give a local description of the moduli of their surfaces. At first sight this looks a bit like a Jerry15 unprojection. In fact one of the families of [CLP] (the one with c0 = Bx = 0) can easily be massaged to a conventional Jerry15 having a double Jerry structure (compare [TJ], 9.2), but this does not seem possible for the more interesting family in [CLP] with Dx = (l/c0)Bx. Question Do these theoretical calculations contain the results of [Di], [CLP] and the like? Answer Absolutely not. They may provide a framework that can produce examples, or simplify and organise the construction of examples. To get complete moduli spaces, it is almost always essential to use other methods, notably infinitesimal deformation calculations or geometric constructions. Question The fact that S can have various gradings seems to add to the complexity of the space Mor(S, SpHk), doesn't it? Answer That may not be the right interpretation -- we could perhaps think that Mor(S, SpHk) (or even the same just for Mor(S, Q2k) into a quadric of rank 2k ≥ 4) is infinite dimensional and infinitely complicated, so subject to Murphy's law [Va], but that when we cut it down to graded in given degrees, it becomes finitely determined, breaking up into a number of finite dimensional families that may be a bit singular, but can be studied with success in favourable cases. 4.9 Problem session 4.9.1 Computing project It is a little project in computer algebra to write an algorithm to put the projective resolution (2.1) in symmetric form. This might just be a straight- 26 forward implementation of the Buchsbaum -- Eisenbud symmetrised complex S2P• outlined in Section 1. Any old computer algebra package can do syzy- gies, but as far as I know, none knows about the symmetry in the Gorenstein case. We now have very many substantial working constructions of codimen- sion 4 Gorenstein varieties. We know in principle that the matrix of first syzygies can be written out in the (A B) form of (2.8), but as things stand, it takes a few hours or days of pleasurable puzzling to do any particular case. 4.9.2 Linear subvarieties What are the linear subvarieties of SpHk? The linear question may be tractable, and may provide a partial answer to the quest for an explicit structure result. The Spin-Hom variety SpHk is defined near a general point by quadratic equations, so its linear subspaces can be studied by the tangent-cone con- struction by analogy with the linear subspaces of quadrics, Segre products or Grassmannians: the tangent plane TP at P ∈ V intersects V in a cone, so that linear subspaces of V through P correspond to linear subspaces in the base of the cone. Now choose a point of the projected variety and continue. Presumably at each stage there are a finite number of strata of the variety in which to choose our point P , giving a finite number of types of Π up to symmetry. I believe that the two famous cases of the Segre models of P2× P2 and P1 × P1 × P1 are maximal linear space of SpH8. It is possible that this method can be used to understand more general morphisms Spec S → SpHk from the regular space Spec S. In this context, it is very suggestive that Tom and Jerry [TJ] are given in terms of linear subspaces of Gr(2, 5). In this case, the intersection with a tangent space is a cone over P1 × P2, so it is clear how to construct all linear subspaces of Gr(2, 5), and equally clear that there are two different families, and how they differ. 4.9.3 Breaking the Ak and Dk symmetry Experience shows that the bulk constructions of Gorenstein codimension 4 ideals do not have the symmetry of the Buchsbaum -- Eisenbud Pfaffians in codimension 3. The equations and syzygies invariably divide up into subsets that one is supposed to treat inhomogeneously. For example, in the 9 × 16 27 unprojection cases, the defining equations split into two sets, the 5 Pfaffian equations of the variety in codimension 3 not involving the unprojection variable s, and the 4 unprojection equations that are linear in s. The columns of the syzygy matrix (A B) are governed by the algebraic group Spin(2k) of type Dk, whereas its rows are governed by GL(k + 1) of type Ak. The common bulk constructions of Gorenstein codimension 4 ideals seem to to accommodate the Ak symmetry of the rows of M1 and the Dk symmetry of its columns by somehow breaking both to make them compatible. This arises if you try to write the 128 spinor coordinates σJ as linear combinations of the 9 relations (L1, . . . , Lk+1), so relating something to do with the columns of M1 to its rows. This symmetry breaking and its effect is fairly transparent in 2.3, Example 2.2, (2.6). Example 2.3 is more typical. (This case comes with three different Tom projections, so may be more amenable.) Of the 128 spinors σJ , it turns out that 14 are zero, 62 are of the form a monomial times one of the relations Li (as in (2.9)), and the remainder are more complicated (probably always a sum of two such products). Mapping this out creates a correspondence from spinor sets to relations, so from the rows of M1 to its columns; there is obviously a systematic structure going on here, and nailing it down is an intriguing puzzle. How this plays out more generally for Kustin -- Miller unprojection [KM], [PR] and its special cases Tom and Jerry [TJ] is an interesting challenge. 4.9.4 Open problems To be useful, a structure theory should make some predictions. I hope that the methods of this paper will eventually be applicable to start dealing with issues such as the following: • k = 3. A 4 × 6 resolution is a Koszul complex. • k = 4. There are no almost complete intersection Gorenstein ideals. Equivalently, a 5 × 8 resolution is nonminimal: if X is Gorenstein codimension 4 and (L1, . . . , L5) generate IX then the first syzygy matrix M1 has a unit entry, making one of the Li redundant. This is a well known theorem of Kunz [K], but I want to deduce it by my methods. • k = 5. Is it true that a 6 × 10 resolution is a hypersurface in a 5 × 5 Pfaffian as in 2.3, Example 2.2? 28 The same question for more general odd k: are hypersurfaces in a codimension 3 Gorenstein varieties the only cases? Is this even true for all the known examples in the literature? This might relate to my even versus odd remark in 3.1.3. • k = 6. I would like to know whether every case of 7 × 12 resolution is the known Kustin -- Miller unprojection from a codimension 4 complete intersection divisor in a codimension 3 complete intersection. • k = 8. As everyone knows, the main case is 9×16. How do we apply the theory to add anything useful to the huge number of known examples? There are hints that something along these lines may eventually be pos- sible, but it is not in place yet. References [TJ] Gavin Brown, Michael Kerber and Miles Reid, Fano 3-folds in codi- mension 4, Tom and Jerry. Part I, Compositio Math. 148 (2012) 1171 -- 1194 [BE1] David Buchsbaum and David Eisenbud, What makes a complex ex- act? J. Algebra 25 (1973) 259 -- 268 [BE2] David Buchsbaum and David Eisenbud, Algebra structures for finite free resolutions, and some structure theorems for ideals of codimen- sion 3, Amer. J. Math. 99 (1977) 447 -- 485 [CLP] Fabrizio Catanese, LIU Wenfei, Roberto Pignatelli, The moduli space of even surfaces of general type with K 2 = 8, pg = 4 and q = 0, arXiv:1209.0034, 29 pp. [CR] A. Corti and M. Reid, Weighted Grassmannians, in Algebraic geom- etry, de Gruyter, Berlin, 2002, pp. 141 -- 163 [Di] Duncan Dicks, Surfaces with pg = 3, K 2 = 4 and extension- deformation theory, 1988, Warwick PhD thesis, 125 + vi pp. (avail- able from my website) [FH] W. Fulton and J. Harris, Representation theory. A first course, Grad. Texts in Math. 129, Springer-Verlag, New York 1991 29 [GW] GOTO Shiro and WATANABE Keiichi, On graded rings. I, J. Math. Soc. Japan 30 (1978) 179 -- 213 [K] Ernst Kunz, Almost complete intersections are not Gorenstein rings, J. Algebra 28 (1974) 111 -- 115 [KM] A. Kustin and M. Miller, Constructing big Gorenstein ideals from small ones, J. Algebra 85 (1983) 303 -- 322 [PR] Stavros Argyrios Papadakis and Miles Reid, Kustin -- Miller unpro- jection without complexes, J. Algebraic Geom. 13 (2004) 563 -- 577, Preprint math.AG/0011094 [Ki] M. Reid, Graded rings and birational geometry, in Proc. of algebraic geometry symposium (Kinosaki, Oct 2000), K. Ohno (Ed.), 1 -- 72, get from www.warwick.ac.uk/staff/Miles.Reid/3folds [R1] M. Reid, Surfaces with pg = 3, K 2 = 4 according to E. Horikawa and D. Dicks, in Proceedings of Algebraic geometry mini-symposium (Tokyo, Dec 1989), 1 -- 22 [Va] Ravi Vakil, Murphy's law in algebraic geometry: badly-behaved de- formation spaces, Invent. Math. 164 (2006) 569 -- 590 Miles Reid, Mathematics Institute, University of Warwick, Coventry CV4 7AL, England e-mail: [email protected] 30
1905.03386	1	1905	2019-05-08T22:53:01	Singular fibres of very general Lagrangian fibrations	[ "math.AG" ]	Let $\pi:X\rightarrow\mathbb{P}^n$ be a (holomorphic) Lagrangian fibration that is very general in the moduli space of Lagrangian fibrations. We conjecture that the singular fibres in codimension one must be semistable degenerations of abelian varieties. We prove a partial result towards this conjecture, and describe an example that provides further evidence.	math.AG	math	Singular fibres of very general Lagrangian fibrations∗ Justin Sawon May, 2019 Abstract Let π : X → Pn be a (holomorphic) Lagrangian fibration that is very general in the moduli space of Lagrangian fibrations. We conjecture that the singular fibres in codimension one must be semistable degenerations of abelian varieties. We prove a partial result towards this conjecture, and describe an example that provides further evidence. 1 Introduction In [25] the author proved that there are finitely many deformation classes of (holomorphic) Lagrangian fibrations π : X → Pn satisfying certain natural conditions. One of these conditions was that the singular fibres in codimension one should be semistable degenerations of abelian varieties. We will describe explicitly what this means shortly. At this stage, let us just point out that semistable degenerations are 'nice' in the following sense: if ¯An denotes a toroidal compactification of the moduli space An of abelian varieties, then the boundary of ¯An parametrizes semistable degenerations. The singular fibres of Lagrangian fibrations in codimension one were studied by Ma- tushita [13, 15] and Hwang and Oguiso [9, 10]. Their classifications include many fibres that are not semistable, because they consider all Lagrangian fibrations, not just very general ones. Matsushita [14, 16] proved that inside the moduli space M of deformations of X as a complex manifold, there is a hypersurface H parametrizing Lagrangian fibrations; by very general we mean that X corresponds to a very general point of H, i.e., contained in the complement of countably many Zariski closed subsets. Any Lagrangian fibration can be deformed slightly so that it becomes very general. Hwang and Oguiso's results imply that if a singular fibre Xt in codimension one is semistable then it must be a rank-one semistable degeneration. Explicitly, this means that the normalization of Xt must be a P1-bundle, or disjoint union of P1-bundles, over an (n − 1)-dimensional abelian variety A. Write this as Xt = k ai=1 Yi. Moreover, each Yi must be a P1-bundle over A given by projectivizing a topologically trivial rank-two bundle, Yi ∼= P(OA ⊕ L) → A ∗2010 Mathematics Subject Classification. 14D06, 14K10, 53C26. 1 2 1 INTRODUCTION where L ∈ Pic0A. This implies that Yi has distinguished zero and infinity sections, (Yi)0 and (Yi)∞. The fibre Xt itself is obtained by gluing zero sections to infinity sections, so that (Yi)∞ ∼= A is identified with (Yi+1)0 ∼= A (with Yk+1 denoting Y1). In general these identifications A ∼= (Yi)∞ → (Yi+1)0 ∼= A do not take 0 to 0, so Xt itself is not a fibration over A. Figures 1 and 2 show examples with k = 1 and k = 3, respectively. Figure 1: Rank-one semistable degeneration s∞ P1 s0 A P1 P1 P1 Figure 2: Reducible semistable degeneration We expect the following behaviour: Conjecture 1 Let π : X → Pn be a very general Lagrangian fibration, and let t be a general point of the hypersurface ∆ ⊂ Pn parametrizing singular fibres. Then Xt is a semistable degeneration of abelian varieties. 3 For K3 surfaces, the conjecture states that a very general elliptic K3 surface has only singular fibres of type Ik in Kodaira's classification. This is certainly true; indeed, a very general elliptic K3 surface will have only I1 singular fibres (rational nodal curves), as can be shown by explicitly finding such a surface and showing that it deforms in a 19-dimensional family. The challenge is to find a proof of this fact that can be extended to higher dimensional Lagrangian fibrations. Although we do not achieve this, we do find a simple proof of a weaker property of elliptic K3 surfaces that generalizes to give: Theorem 2 Let π : X → Pn be a very general Lagrangian fibration, and let t be a general point of the hypersurface ∆ ⊂ Pn parametrizing singular fibres. Then Xt is either 1. an ´etale quotient of an n-dimensional abelian variety (such a fibre is singular because it will have multiplicity greater than one), 2. a semistable degeneration of abelian varieties, or 3. a fibration over an (n − 1)-dimensional abelian variety A by singular elliptic curves of Kodaira type II (a cuspidal rational curve), III (two rational curves meeting at a tacnode), or IV (three rational curves meeting at a point), up to an ´etale cover. Of course, we believe that only the second case is possible. Theorem 2 has been obtained independently by C. Lehn (Theorem 5.7 of [12]) as a corollary of his generalization of Voisin's results on deformations of Lagrangian submanifolds to deformations of Lagrangian normal crossing subvarieties. Our proof is more direct: a standard argument shows that a very general Lagrangian fibration must have Picard number ρ(X) = 1, and this places restrictions on the structure of the discriminant hypersurface π−1(∆) from which the theorem follows. Note that it is not true that the singular fibres in codimension two (or higher) must be semistable. For example, let π : X → Pn be a Beauville-Mukai system [3, 17], i.e., the compactified relative Jacobian of a complete linear system of curves on a K3 surface. There will be cuspidal curves in codimension two in this linear system, and their compactified Jacobians will not be semistable; see [24] for details. Moreover, we can make this Lagrangian fibration very general by deforming it in a 20-dimensional family, inside the 21-dimensional moduli space of deformations of X, while preserving the existence of these non-semistable singular fibres in codimension two. In Section 2 we recall some facts about singular fibres of elliptic K3 surfaces. In Section 3 we extend these results to higher dimensional Lagrangian fibrations, thereby proving our main result, Theorem 2. Section 4 contains a number of illustrative examples. The author would like to thank Concettina Galati, Andreas Knutsen, and Christian Lehn for helpful conversations, and the Erwin Schrodinger Institute for hospitality. The author gratefully acknowledges support from the NSF, grant numbers DMS-1206309 and DMS-1555206. 2 K3 surfaces We begin by recalling Kodaira's classification of singular fibres of elliptic surfaces. Theorem 3 (see Barth et al. [4]) A non-multiple singular fibre of a minimal elliptic sur- face must be one of the following types: • Ik, k ≥ 1, a cycle of k rational curves with intersection matrix given by the Cartan matrix of the affine Dynkin diagram of type Ak−1 (with I1 a nodal rational curve), 4 2 K3 SURFACES • II, a cuspidal rational curve, • III, two rational curves meeting at a tacnode, • IV , three rational curves meeting at a single point, • I ∗ k , k ≥ 0, k + 5 rational curves with intersection matrix given by Dk+4, • II ∗, nine rational curves with intersection matrix given by E8, • III ∗, eight rational curves with intersection matrix given by E7, • IV ∗, seven rational curves with intersection matrix given by E6. A multiple singular fibre must be of type mI0, a smooth elliptic curve with multiplicity m ≥ 2, or of type mIk with k ≥ 1, i.e., type Ik with multiplicity m ≥ 2. Let S → P1 be an elliptic K3 surface, with fibre class F ∈ H1,1(S) ∩ H2(S, Z). If we deform S so that F remains algebraic, i.e., of type (1, 1), then the resulting K3 surface will still be elliptic. This means that elliptic K3 surfaces are codimension one inside the 20-dimensional moduli space of all K3 surfaces. Moreover, a very general elliptic K3 surface will have Picard number ρ = 1; note that it is non-projective and does not admit a section (or even a multi-valued rational section). Lemma 4 Let S → P1 be a very general elliptic K3 surface. Then every singular fibre of S is reduced and irreducible. Thus by Kodaira's classification, every singular fibre is either of type I1 (nodal rational curve) or type II (cuspidal rational curve). Proof The N´eron-Severi group of a non-projective elliptic surface is spanned by the fibre class F and the irreducible components of the singular fibres. Suppose there are k singular fibres S1, . . . , Sk with m1, . . . , mk irreducible components, respectively. Obviously some linear combination of the components of Si is linearly equivalent to F , so each Si really contributes mi − 1 additional independent classes to the N´eron-Severi group. Thus the rank of the N´eron-Severi group is ρ(S) = 1 + k Xi=1 (mi − 1). Recall that the Shioda-Tate formula [26] for an elliptic surface with a section is ρ(S) = 2 + rankM W (π) + k Xi=1 (mi − 1), where M W (π) is the Mordell-Weil group of sections of π : S → P1. Our formula above is essentially a degenerate version of the Shioda-Tate formula for non-projective surfaces. Since a very general elliptic K3 surface has ρ(S) = 1, we immediately see that mi = 1 for all i, i.e., every singular fibre of S is irreducible. Multiple fibres contribute non-trivially to Kodaira's formula for the canonical bundle of an elliptic surface, but K3 surfaces have trivial canonical bundles, and thus they cannot have any multiple fibres. Thus every singular fibre of S is reduced. (cid:3) 5 Question Although it does not follow from the above arguments, very general elliptic K3 surfaces actually have only nodal rational curves as singular fibres. This can be proved by explicitly constructing such a K3 surface and then showing that there are 19-parameters de- scribing its deformations. Is there a deformation argument that eliminates cuspidal rational curves? Given an elliptic fibration over a disc with a cuspidal rational curve over 0, we can deform it to a fibration with two nodal rational curves. However, we need a global argument to show that such a deformation fits in to an elliptic K3 surface. 3 Higher dimensions We want to extend the ideas of the previous section to higher dimensional Lagrangian fibrations, i.e., fibrations π : X → Pn where X is an irreducible holomorphic symplectic manifold and the general fibre is an n-dimensional abelian variety (see [20] and the references therein). We start by describing Hwang and Oguiso's [9, 10] Kodaira-type classification of general singular fibres. By Proposition 3.1(2) of [9] the discriminant locus ∆ ⊂ Pn parametrizing singular fibres is a hypersurface. Roughly speaking, the main observation of Hwang and Oguiso is that for a general singular fibre Xt, above a general point t ∈ ∆, there is a residual (n − 1)-dimensional abelian variety present and the fibre is degenerating in the one remaining dimension. To state this precisely, let V be a component of the reduction (Xt)red of Xt, and let V be its normalization. Then the Albanese map V → Alb( V ) is a fibre bundle with fibre either P1 or an elliptic curve. The image of such a fibre in (Xt)red is called a characteristic leaf ; if two (or more) leaves meet they form a characteristic curve, and once we add in multiplicities coming from the multiplicities of the components of Xt we obtain a characteristic 1-cycle. Hwang and Oguiso gave a classification of these characteristic 1-cycles. Theorem 5 (Hwang-Oguiso [9, 10]) Let Xt be a general singular fibre with multiplicity mi m Ci m. Then all characteristic 1-cycles Pi miCi in Xt are isomorphic, and the 1-cycle Pi is either 1. a smooth elliptic curve, 2. one of the singular elliptic fibres of Kodaira's classification, as in Theorem 3, or 3. of type A∞, i.e., a 1-cycle Pi∈Z Ci consisting of a chain of infinitely many rational curves Ci, with Ci meeting Cj if and only if j = i ± 1. A singular fibre that is a product of an (n − 1)-dimensional abelian variety and a singular elliptic curve from Kodaira's classification will have characteristic 1-cycle of the same type as the singular elliptic curve. For example, the product of an (n−1)-dimensional abelian variety and an elliptic curve of type I2 will have characteristic 1-cycle of type I2; such a singular fibre will have two irreducible components. But singular fibres need not be products, and so it is possible for an irreducible singular fibre to have a reducible characteristic 1-cycle, as illustrated in Figure 3. 6 3 HIGHER DIMENSIONS Figure 3: Irreducible singular fibre with characteristic 1-cycle of type I2 Remark Unlike elliptic K3 surfaces, the singular fibres of Lagrangian fibrations can have multiplicities. Hwang and Oguiso [10] showed that various values up to and including six are possible, depending on the type of the (characteristic 1-cycle of the) singular fibre. In particular, it is possible for the (reduced) characteristic 1-cycle of a singular fibre to be a smooth elliptic curve if the fibre has multiplicity m > 1. Some examples will be given in Section 4.1. Remark Of course, a singular fibre cannot have infinitely many components. Thus a characteristic 1-cycle of type A∞ must 'wrap around' the singular fibre. An example of a reduced and irreducible singular fibre with characteristic 1-cycle of type A∞ will be given in Section 4.2. Let π : X → Pn be a Lagrangian fibration with n ≥ 2. Let F ∈ Hn,n(X) ∩ H2n(X, Z) be the fibre class and let L ∈ H1,1(X) ∩ H2(X, Z) be the pullback of a hyperplane in Pn. Inside the moduli space M of deformations of X as a complex manifold, there is a hypersurface HF parametrizing deformations such that F remains algebraic (see Voisin [28]) and there is also a hypersurface HL parametrizing deformations such that L remains algebraic. In fact, these hypersurfaces are identical and Matsushita [14, 16] proved that they parametrize deformations of X that remain Lagrangian fibrations. In a family of deformations of X, the Picard number is upper semicontinuous; Oguiso [18] proved that the Picard number jumps up on a dense countable subset. More importantly for us, a family parametrized by M′ ⊂ M of codimension k cannot contain only deformations of X with Picard numbers > k. In particular, a family parametrized by a hypersurface in M must contain deformations with Picard numbers 1 (or 0). It follows that a very general Lagrangian fibration will have Picard number ρ(X) = 1; it will also be non-projective and will not admit a (multi-valued rational) section. Our main result, Theorem 2, follows directly from the following proposition: the three cases for the characteristic 1-cycle correspond directly to the three cases for the structure of the singular fibre in Theorem 2. Proposition 6 Let π : X → Pn be a very general Lagrangian fibration, and let t be a general point of the discriminant locus ∆ ⊂ Pn. Then the (reduced) characteristic 1-cycle of the singular fibre Xt is either 1. a smooth elliptic curve, 7 2. a singular elliptic curve of Kodaira type Ik with k ≥ 1, or of type A∞, or 3. a singular elliptic curve of Kodaira type II, III, or IV . Remark In the first case, Xt must have multiplicity m = 2, 3, 4, or 6 (not 1 or it would be a smooth fibre). In the second case, we must have m = 1 for type Ik with k odd, and m = 1 or 2 for type Ik with k even and type A∞. In the third case, we must have m = 1 or 5 for type II, m = 1 or 3 for type III, and m = 1 or 2 for type IV (see Hwang and Oguiso [10]). Before presenting the proof of Proposition 6, we state a generalization of the Shioda-Tate formula to higher dimensional fibrations by abelian varieties. Theorem 7 (Oguiso [19], Theorem 1.1) Let ϕ : X → Y be a proper surjective mor- phism with rational section O, whose generic fibre A := Xη is an abelian variety defined over the field K = C(Y ) with origin O. Assume further that X and Y have only Q-factorial ra- tional singularities, ϕ is equi-dimensional in codimension one, and h1(X, OX ) = h1(Y, OY ). Write ∆ = ∪k i=1∆i for the decomposition into irreducible components of the discriminant divisor ∆ ⊂ Y , and assume that ϕ−1(∆i) ⊂ X consists of mi irreducible components. Then the Mordell-Weil group M W (ϕ), i.e., the group A(K) of K-rational points of A (equiva- lently, the group of rational sections of ϕ), is a finitely generated abelian group of rank rankM W (ϕ) = ρ(X) − ρ(Y ) − rankN S(AK) − k Xi=1 (mi − 1). Proof of Proposition 6 We apply Oguiso's theorem to a Lagrangian fibration π : X → Pn. In this case, X and Y = Pn are both smooth, Lagrangian fibrations are equi-dimensional, and h1(X, OX ) = h1(Pn, OPn) = 0. The theorem assumes the existence of a (rational) section. In [23] the author proved that there is a hypersurface in HF = HL parametrizing Lagrangian fibrations that admit sections; in other words, Lagrangian fibrations that admit sections are codimension two in M. A very general Lagrangian fibration that admits a section will therefore have Picard number ρ(X) = 2. In addition, the existence of a section implies that X is projective and hence the N´eron-Severi group N S(AK) must have rank at least one. The Shioda-Tate formula, rankM W (π) = 2 − 1 − rankN S(AK) − k Xi=1 (mi − 1), then forces rankM W (π) = 0, rankN S(AK) = 1, and mi = 1 for all i. If we consider instead a Lagrangian fibration π : X → Pn that does not admit any (multi-valued rational) section, then there is an obvious analogue of the formula in which rankM W (π) and rankN S(AK) both vanish and k ρ(X) = ρ(Pn) + (mi − 1). Xi=1 For a very general Lagrangian fibration ρ(X) = 1 and again we find that mi = 1 for all i. Summarizing, we have proved that if the discriminant divisor ∆ ⊂ Pn of a very general Lagrangian fibration π : X → Pn decomposes into irreducible components as ∆ = ∪k i=1∆i, then π−1(∆i) is irreducible for all i. Next, let t be a general point of ∆i ⊂ ∆. The statement above does not imply that the fibre Xt is irreducible. It is possible that the fibre Xt has several irreducible components 8 4 EXAMPLES Y1, . . . , Yl that are all contained in the single irreducible divisor π−1(∆i), because of 'mon- odromy' permuting the components as we move around a loop in ∆i starting and ending at t, avoiding the codimension one subset ∆i0 ⊂ ∆i parametrizing non-general singular fibres. However, we see that this monodromy representation π1(∆i\∆i0, t) −→ Syml must act transitively on the set of components of Xt. This implies that the components Yj must all have the same multiplicity. Referring to Theorem 5, we can conclude that the (reduced) characteristic 1-cycle of Xt must belong to one of the three cases of the proposition. The other possibilities in Hwang and Oguiso's classification are singular elliptic curves of Kodaira type I ∗ m with m ≥ 0, II ∗, III ∗, and IV ∗, but these all have components with different multiplicities, which is not allowed. (cid:3) 4 Examples In this section we describe the singular fibres in codimension one for several examples of Lagrangian fibrations. 4.1 Multiple fibres with smooth reduction In Example 6.2 of [10], Hwang and Oguiso described several examples where the singular fibre has smooth reduction, which is equivalent to the (reduced) characteristic 1-cycle being a smooth elliptic curve. Their examples are local Lagrangian fibrations, but the same constructions easily extend to give global compact Lagrangian fibrations. For instance, let us extend Example 6.2(v) which has singular fibres of multiplicity 2 and characteristic 1- cycle an arbitrary smooth elliptic curve E1 with equation y2 = x3 + ax + b. Let E2, E3, and E4 be arbitrary smooth elliptic curves with coordinates t, z, and s, respectively, and let p2 be a 2-torsion point on E3. Then define X to be the quotient of E1 × E2 × E3 × E4 by the fixed-point-free involution g∗ : ((x, y), t, z, s) 7−→ ((x, −y), −t, z + p2, s). The symplectic form σ := dx∧dt therefore descends to X. The projection y + dz ∧ ds on E1 × E2 × E3 × E4 is preserved by g∗ and X −→ E2/ ± 1 × E4 ∼= P1 × E4 [((x, y), t, z, s)] 7−→ (±t, s) makes X into a Lagrangian fibration. The singular fibres sit above q2 × E4, where q2 is a 2-torsion point in E2 (fixed by ±1), and they look like the hyperelliptic surface E1 × E3/((x, y), z) ∼ ((x, −y), z + p2) with multiplicity 2. This example is an isotrivial Lagrangian fibration: all the smooth fibres are isomorphic to E1 × E3. We can construct a non-isotrivial example by replacing E1 × E2 by a certain elliptic K3 surface S. Specifically, choose for S an elliptic K3 surface admitting a symplectic involution τ which acts as ±1 on each fibre and as ±1 on the base P1 = C ∪ {∞}. The only 4.2 Beauville-Mukai systems 9 fibres of S → P1 that are fixed by τ are those above 0 and ∞; we assume these are smooth. We can now modify the example above by defining X to be the quotient of S × E3 × E4 by the fixed-point-free involution g∗ : (q, z, s) 7−→ (τ (q), z + p2, s). The symplectic form σ := σS + dz ∧ ds descends to X and the projection to P1/ ± 1 × E4 makes X into a Lagrangian fibration. The singular fibres sit above {0} × E4 and {∞} × E4 and look like the hyperelliptic surfaces S0 × E3/(w0, z) ∼ (−w0, z + p2) and S∞ × E3/(w∞, z) ∼ (−w∞, z + p2) with multiplicity 2, where S0 and S∞ are the elliptic fibres of S → P1 above 0 and ∞, respectively, with coordinates w0 and w∞. Note that a theorem of Hwang [8] asserts that the base of a Lagrangian fibration of an irreducible holomorphic symplectic manifold must be isomorphic to projective space, if it is smooth. However, in the examples above X is not an irreducible holomorphic symplectic manifold; indeed, it admits an ´etale cover that is a product, E1×E2×E3×E4 and S×E3×E4, respectively. Consequently, bases that are not isomorphic to P2 can arise. 4.2 Beauville-Mukai systems Recall the construction of the Beauville-Mukai integrable system [3, 17]. Let S be a K3 surface containing a smooth genus n curve C. Then C moves in an n-dimensional linear system, C ∼= Pn; denote by C → Pn the corresponding family of curves. If S is very general in the sense that its N´eron-Severi group N S(S) is generated over Z by [C] then every curve in the family C is reduced and irreducible. This means that we can apply the Altman and d (C/Pn). Kleiman construction [1] of the compactified relative Jacobian to obtain X := Jac We can also regard X as a Mukai moduli space of stable sheaves on S with Mukai vector v = (0, [C], d + 1 − n). Here we think of an element of X as a torsion sheaf ι∗L, where ι : C ֒→ S is the embedding of a curve into the K3 surface and L is a degree d line bundle (or more generally, a rank one torsion-free sheaf) on C; then ι∗L is stable in the sense of Simpson [27]. This latter point of view shows that X admits a holomorphic symplectic structure, and X → Pn is therefore a Lagrangian fibration. Moreover, the definition of X as a Mukai moduli space makes sense even when N S(S) 6∼= Z.[C], and provided the Mukai vector v is isotropic and a general polarization of S is chosen, we will once again obtain a (smooth, compact) Lagrangian fibration X → Pn. For a very general (polarized) K3 surface, a general codimension one singular curve in the family C will have a single node. This can be proved by studying the Beauville-Mukai system and using properties of Lagrangian fibrations (see Lemma 2.4 of [24]); it is also a special case of Lemma 3.1 of Chen [5]. Let C be such a curve, with normalization C of genus n − 1, and C obtained by identifying two points p and q ∈ C. The normalization of d C of C is a P1-bundle over the Jacobian Jacd C of C, and the the compactified Jacobian Jac d compactified Jacobian Jac C itself is obtained by identifying the 0- and ∞-sections of this P1-bundle via a translation 0-section ∼= Jacd C ⊗O(p−q) −→ Jacd C ∼= ∞-section. 10 4 EXAMPLES Since the characteristic leaves are the images of the P1-fibres, the type of the characteristic 1-cycle therefore depends on whether or not O(p − q) is a torsion line bundle in Jac0 C. Lemma 8 The characteristic 1-cycle of a general singular fibre of a Beauville-Mukai system constructed from a very general K3 surface is of type A∞. Remark A local example of a Lagrangian fibration with general fibres of type A∞ was described by Matsushita (see Proposition 4.13 in [9]). Proof Consider first the genus n = 2 case, studied in detail in [22]. We can write the discriminant locus as a disjoint union ∆ = (∆\∆0) ∪ ∆0 = (∆\∆0) ∪ ∆c ∪ ∆nn, where ∆\∆0 parametrizes curves with one node, ∆c parametrizes curves with one cusp, and ∆nn parametrizes curves with two nodes. Note that ∆c is the set of 72 cusps of ∆ and ∆nn is the set of 324 nodes of ∆. Consider the normalization C → (∆\∆0) of the family of nodal curves C → (∆\∆0). Because the cuspidal curves over ∆c also have normalizations of genus one, this family extends to a family C → (∆\∆0) ∪ ∆c of smooth genus one curves. The family of line bundles O(p − q) then defines a section of the relative Jacobian Jac0( C/(∆\∆0)) over ∆\∆0. As we approach a cuspidal curve, p → q, so this section extends to a section of the relative Jacobian Jac0( C/(∆\∆0) ∪ ∆c). Moreover, this section intersects the zero section, i.e., takes the value O, precisely above ∆c, because O(p − q) = O(p − p) = O only for a cuspidal curve. In particular, the section is non-zero and not constant. It follows that for a general curve in ∆\∆0, with a single node, O(p − q) will not be torsion in Jac0 C, and therefore the compactified Jacobian Jac C will be of type A∞, i.e., the characteristic 1-cycle will consist of an infinite chain of rational curves. The same argument works for higher genus curves. For a very general K3 surface S, Galati and Knutsen proved that in the n-dimensional linear system C ∼= Pn there will exist a curve with a cusp and n − 2 nodes (Theorem 1.1 in [6]). Moreover, they showed that the n − 2 nodes may be smoothed independently, producing an (n − 2)-dimensional family of curves with single cusps. In other words, ∆\∆0 parametrizes curves with one node, and there again exists a non-empty codimension one subset ∆c ⊂ ∆ (i.e., codimension two in C) parametrizing cuspidal curves. The proof for n > 2 then proceeds as above. (cid:3) d Remark The section of Jac0( C/(∆\∆0) ∪ ∆c) in the above proof will also intersect the (multi-)section of k-torsion points, for any k ≥ 2. At these points of intersection O(p − q) d will be k-torsion and therefore the compactified Jacobian Jac C will be of type Ik, i.e., the characteristic 1-cycle will consist of k rational curves forming a cycle. Thus the Beauville- Mukai system contains singular fibres corresponding to characteristic 1-cycles of all types Ik for k ≥ 2 and A∞. There will be no singular fibres corresponding to characteristic 1-cycles of type I1, as these would require p = q, which as we saw only occurs for cuspidal curves over ∆c. 4.3 Hilbert schemes of elliptic K3 surfaces 11 4.3 Hilbert schemes of elliptic K3 surfaces Next we consider an elliptic K3 surface S → P1. Assume S → P1 admits a section, but is otherwise very general in the sense that it contains exactly 24 nodal rational curves as singular fibres, above the points p1, . . . , p24 ∈ P1. The Hilbert scheme HilbnS of n points on S is an irreducible holomorphic symplectic manifold [2], and the elliptic fibration on S induces a Lagrangian fibration HilbnS −→ SymnS −→ SymnP1 = Pn, where the first map is the Hilbert-Chow morphism. The fibre over a general point of SymnP1, given by an n-tuple {x1, . . . , xn} of distinct points on P1, is isomorphic to the product Sx1 × · · · × Sxn of the corresponding elliptic fibres of S → P1. The singular fibres occur over the hyperplanes ∆i := {{x1, . . . , xn} ∈ SymnP1 xj = pi for some j} for i = 1, . . . , 24, and over the 'big diagonal' ∆0 := {{x1, . . . , xn} ∈ SymnP1 xj = xk for some j and k}. Consider the former; moreover, without loss of generality, let {x1, . . . , xn} be a general point of ∆1 with x1 = p1, xj 6= pi for all j ≥ 2 and i, and xj 6= xk for all j and k. Then the singular fibre over {x1, . . . , xn} will be isomorphic to Sx1 × Sx2 × · · · × Sxn ∼= Sp1 × Sx2 × · · · × Sxn where Sp1 is a nodal rational curve and Sx2, . . . , Sxn are smooth elliptic curves. This is semistable and the characteristic 1-cycle is clearly of type I1. Next consider a general point {x1, . . . , xn} of ∆0; without loss of generality, assume x1 = x2, xj 6= xk otherwise, and xj 6= pi for all j and i. For simplicity, we first consider the n = 2 case. Write E for the elliptic curve Sx1. The Hilbert scheme Hilb2S is obtained from Sym2S by blowing up the diagonal; thus each point of the diagonal in Sym2S is replaced by a P1. The singular fibre over {x1 = x2} ∈ ∆0 is therefore isomorphic to the union of Sym2E and a P1-bundle over D, where D ∼= E is the diagonal in Sym2E. Now Sym2E is also a P1-bundle over E, because we have the Abel-Jacobi map Sym2E −→ Pic2E ∼= E taking a degree two divisor to its corresponding line bundle. The diagonal D is a 4-valued section of this P1-bundle; for instance, it intersects the fibre above O(2y0) ∈ Pic2E in the four points 2y1 ∈ D ⊂ Sym2E where y1 − y0 is 2-torsion in E. Instead, one could observe that the composition of the maps E ∼= D ֒→ Sym2E −→ Pic2E takes y to O(2y), i.e., is multiplication by 2. Putting everything together, we see that the 0 , i.e., a central P1 singular fibre over {x1 = x2} will have characteristic 1-cycle of type I ∗ with four other P1s attached in a D4 configuration. For n > 2 we simply take the product of the above with the smooth (n − 2)-dimensional 0 . In particular, abelian variety Sx3 × · · · × Sxn , so the characteristic 1-cycle is still of type I ∗ 12 4 EXAMPLES the singular fibres over ∆0 are never semistable. However, we will see in the next subsection that they become semistable after a small deformation of the Lagrangian fibration. Thinking of Pn as parametrizing degree n polynomials, ∆0 is precisely the hypersurface parametrizing polynomials with a repeated root. A resultant argument then show that ∆0 has degree 2(n − 1), but it seems that one should attach a multiplicity to ∆0 because it parametrizes non-semistable singular fibres. We will say more about this in the next subsection. 4.4 Deforming to a very general Lagrangian fibration We will now show that the Hilbert scheme of an elliptic K3 surface can be deformed, as a Lagrangian fibration, to a Beauville-Mukai integrable system. Under this deformation, the non-semistable singular fibres of the Hilbert scheme become semistable fibres of the Beauville-Mukai system. We first describe this in dimension 2n = 4, and then in higher dimensions 2n ≥ 6. Let S be an elliptic K3 surface admitting a section D, and assume that the N´eron- Severi group is generated by D and the fibre F . In what follows we will often abbreviate H0(S, O(D)) as H0(S, D), etc. Lemma 9 There is a natural isomorphism H0(S, 2F + D) ∼= H0(S, 2F ) ∼= C3. Thus D is the base locus of the linear system 2F + D, the movable part consists of pairs of fibres, and 2F = Sym2P1 ∼= P2 because each fibre moves in a pencil. Proof Firstly (2F + D).F = 1, (2F + D).D = 0, and (2F + D)2 = 2, so 2F + D is nef and big. By the Kodaira vanishing theorem and the fact that S has trivial canonical bundle, we deduce that H1(S, 2F + D) = 0. The short exact sequence 0 −→ O(2F ) −→ O(2F + D) −→ O(2F + D)D ∼= OD −→ 0 then yields the long exact sequence 0 → H0(S, 2F ) → H0(S, 2F + D) → H0(D, OD) ∼= C → H1(S, 2F ) → 0. The Leray spectral sequence for π : S → P1 yields the exact sequence H1(P1, O(2)) = 0 −→ H1(S, 2F ) = H1(S, π∗O(2)) −→ H0(P1, R1π∗π∗O(2)) −→ 0, which implies H1(S, 2F ) ∼= H0(P1, R1π∗π∗O(2)) = H0(P1, O(2) ⊗ R1π∗OS) = H0(P1, O) ∼= C, because R1π∗OS ∼= O(−2) for an elliptic K3 surface. The long exact sequence above now looks like 0 −→ H0(S, 2F ) −→ H0(S, 2F + D) −→ C ∼=−→ C −→ 0. 4.4 Deforming to a very general Lagrangian fibration We conclude that the first map is also an isomorphism, i.e., H0(S, 2F + D) ∼= H0(S, 2F ). Of course, the Leray spectral sequence also yields H0(S, 2F ) = H0(S, π∗O(2)) ∼= H0(P1, O(2)) ∼= C3. 13 (cid:3) The first order deformations of S are parametrized by H1(S, T ) ∼= H1(S, Ω1) ∼= H1,1(S), where the first isomorphism is induced by interior product with the holomorphic symplectic form σ, which gives an isomorphism T ∼= Ω1. In terms of periods, deforming in the direction α ∈ H1,1(S) changes σ into σt = σ + tα to first order in t. We can globalize this by adding a quadratic term and defining Then σ2 t = 0 and σt := σ + tα − t2(cid:18) α2 2σ¯σ(cid:19) ¯σ. σt ¯σt = (cid:18)1 + t2α2 2σ¯σ (cid:19)2 σ¯σ is positive for all t sufficiently close to 0, thus giving a family of K3 surfaces S → ∆ defined over a disc ∆ of some radius. Consider the deformation in the direction α = D, and assume that σ is normalized so that σ¯σ = 1. Then σt = σ + tD + t2¯σ, σt ¯σt = (1 − t2)σ¯σ, and thus our family S → ∆ is defined over a disc of radius 1. Now σt.(2F + D) = σ.(2F + D) + tD.(2F + D) + t2 ¯σ.(2F + D) = 0, so 2F + D remains of type (1, 1) on St for all t. Moreover, for very general t the N´eron-Severi group of St is generated by 2F + D, and thus 2F + D will be ample on St. The linear system 2F + D gives a relative map S/∆ −→ P2 ∆ over ∆. For very general t this map is a branched double cover of P2, while for t = 0 the image of S = S0 → P2 is a conic. The drop in dimension of the image at t = 0 occurs because the linear system 2F + D has base locus D on S = S0, whereas for t 6= 0 the curve D disappears, i.e., is no longer of type (1, 1). Next we define a relative moduli space M/∆ of sheaves on S/∆ by defining Mt to be the Mukai moduli space of stable sheaves on St with Mukai vector v = (0, [2F + D], −1). A general element of Mt will look like ι∗L, where ι : C ֒→ St is the inclusion of a curve C in the linear system 2F + D and L is a degree zero line bundle on C. (In general, the Mukai vector of ι∗L is (0, [C], d + 1 − g) for a degree d line bundle L on a curve C of genus g.) 14 4 EXAMPLES Proposition 10 The relative moduli space M/∆ is a family of Lagrangian fibrations over P2 ∆. Specifically 1. for very general t, Mt is a Beauville-Mukai integrable system, 2. for t = 0, M0 is birational to Hilb2S0 and the Lagrangian fibration is induced by the original elliptic fibration on S0 = S. Remark The relative map S/∆ → P2 Lagrangian fibration M/∆ → P2 Thus the P2 ∆ in the proposition is really dual to the earlier P2 ∆. ∆ comes from maps to 2F + D∨, whereas the relative ∆ comes from mapping a sheaf to its support in 2F + D. Proof For very general t, Mt is a Beauville-Mukai system by definition; so there is nothing to prove in part 1. When t = 0, the linear system 2F +D has D as a base locus. Thus every curve C in the linear system 2F + D looks like F1 + F2 + D, where F1 and F2 are a pair of fibres of the elliptic fibration S → P1. Assume that F1 and F2 are smooth and distinct. A degree zero line bundle L on F1 + F2 + D is given by line bundles L1, L2, and L3 on F1, F2, and D, respectively, plus isomorphisms (L1)p ∼= (L3)p at p = F1 ∩ D and (L2)q ∼= (L3)q at q = F2 ∩ D. The degrees d1, d2, and d3 of L1, L2, and L3 must sum to zero. In fact, stability of ι∗L implies that d1 = d2 = d3 = 0. The only degree zero line bundle on D ∼= P1 is the trivial line bundle, whereas degree zero line bundles on F1 and F2 are parametrized by Pic0F1 ∼= F1 and Pic0F2 ∼= F2, respectively. Therefore the fibre of M0 → 2F + D over the point F1 + F2 + D is isomorphic to F1 × F2. Similarly, the elliptic fibration S0 → P1 induces a Lagrangian fibration Hilb2S0 −→ Sym2P1 ∼= P2 whose fibre over a general point F1 +F2 ∈ 2F = Sym2P1 ∼= P2 is also isomorphic to F1 ×F2. Since these Lagrangian fibrations both admit global sections, they must be birational. (cid:3) Remark The argument easily extends to some singular fibres. For example, suppose that F1 is a nodal rational curve. Then we should allow L1 to be a rank one torsion-free sheaf F1. on F1. F1 is isomorphic to F1, as required. The case F1 = F2, even for F1 smooth, However, Pic appears to be more complicated. 0 In other words, we should replace Pic0F1 by the compactified Jacobian Pic 0 Remark For each fibre, we identified Pic0F with F . This identification is fixed by taking OF to the basepoint in F given by the intersection F ∩ D with the section D. Alternatively, we could work with the moduli space M (0, [C], 1) whose general element looks like ι∗L for a degree two line bundle L. Then stability will force L to come from degree one line bundles on F1 and F2 and a degree zero line bundle on D. The fibre over F1 + F2 + D would then be Pic1F1 × Pic1F2 ∼= F1 × F2, and this isomorphism is canonical: we don't even need a basepoint in each fibre. Corollary 11 The Hilbert scheme Hilb2S0 can be deformed to a Beauville-Mukai integrable system, as a Lagrangian fibration over P2. Proof This is achieved by a birational modification of the relative moduli space M/∆. We showed that the fibre M0 above t = 0 is birational to Hilb2S0. Birational holomorphic symplectic manifolds correspond to non-separated points in the moduli space (see Huy- brechts [7]), so we can replace M0 by Hilb2S0 in a family. (cid:3) 4.4 Deforming to a very general Lagrangian fibration 15 Assume that S is very general, so that it contains exactly 24 nodal rational curves as singular fibres. The Lagrangian fibration on Hilb2S0 has a discriminant locus consisting of 24 lines and a conic, with the lines tangent to a conic; the singular fibres above the lines are semistable, whereas those above the conic are not. A Beauville-Mukai integrable system will have a discriminant locus of degree 30 (see [21]), and general singular fibres will be semistable. Thus the corollary demonstrates how non-semistable fibres can deform to semistable fibres under a small deformation of the Lagrangian fibration. In addition, it suggests that the conic should be counted with multiplicity three, so that the degree of the discriminant locus is preserved under deformations. We now extend this example to higher dimensions. Lemma 12 For n ≥ 3 there is a natural isomorphism H0(nF + D) ∼= H0(nF ) ∼= Cn+1. Thus D is the base locus of the linear system nF + D, the movable part consists of sets of n fibres, and nF = SymnP1 ∼= Pn. Proof Firstly (nF + D).F = 1, (nF + D).D = n − 2 > 0, and (nF + D)2 = 2n − 2 > 0, so nF + D is ample. By the Kodaira vanishing theorem H1(S, nF + D) = 0. The short exact sequence 0 −→ O(nF ) −→ O(nF + D) −→ O(nF + D)D ∼= OD(n − 2) −→ 0 then yields the long exact sequence 0 → H0(S, nF ) → H0(S, nF + D) → H0(D, OD(n − 2)) ∼= Cn−1 → H1(S, nF ) → 0. The Leray spectral sequence for π : S → P1 yields the exact sequence H1(P1, O(n)) = 0 −→ H1(S, nF ) = H1(S, π∗O(n)) −→ H0(P1, R1π∗π∗O(n)) −→ 0, which implies H1(S, nF ) ∼= H0(P1, R1π∗π∗O(n)) = H0(P1, O(n) ⊗ R1π∗OS) = H0(P1, O(n − 2)) ∼= Cn−1. The long exact sequence above now looks like 0 −→ H0(S, nF ) −→ H0(S, nF + D) −→ Cn−1 ∼=−→ Cn−1 −→ 0. We conclude that the first map is also an isomorphism, i.e., H0(S, nF + D) ∼= H0(S, nF ). Thus nF + D is ample but not very ample: it has base locus D. Finally, the Leray spectral sequence also yields H0(S, nF ) = H0(S, π∗O(n)) ∼= H0(P1, O(n)) ∼= Cn+1. (cid:3) 16 4 EXAMPLES Let us deform S in the direction α = (2 − n)F + D by taking σt := σ + tα − t2(cid:18) α2 2σ¯σ(cid:19) ¯σ = σ + t((2 − n)F + D) + t2(n − 1)¯σ, where we once again assume the normalization σ¯σ = 1. Denote by S → ∆ the resulting deformation of S, over a disc ∆ of suitable radius. Then nF + D will remain of type (1, 1), because σt.(nF + D) = σ.(nF + D) + t((2 − n)F + D).(nF + D) + t2(n − 1)¯σ.(nF + D) = 0. Moreover, nF + D will be a generator of the N´eron-Severi group for very general t. The linear system nF + D gives a relative map S/∆ −→ Pn ∆. This is an embedding for very general t, but for t = 0 the image of S = S0 → Pn is a normal rational curve, because nF + D has a base locus D on S = S0. We define a relative moduli space M/∆ by defining Mt to be the Mukai moduli space of stable sheaves on St with Mukai vector v = (0, [nF + D], 1 − n). A general element will look like ι∗L for a degree zero line bundle L on a curve ι : C ֒→ St in nF + D. Proposition 13 The relative moduli space M/∆ is a family of Lagrangian fibrations over Pn ∆. Specifically 1. for very general t, Mt is a Beauville-Mukai integrable system, 2. for t = 0, M0 is birational to HilbnS0 and the Lagrangian fibration is induced by the original elliptic fibration on S0 = S. Proof For very general t, Mt is a Beauville-Mukai system by definition; so there is nothing to prove in part 1. When t = 0, the linear system nF + D has D as a base locus. Thus every curve C in the linear system nF + D looks like F1 + . . . + Fn + D, where Fi are fibres of the elliptic fibration S → P1. Assume that the Fi are smooth and distinct. A degree zero line bundle L on F1 + . . . + Fn + D is given by line bundles L1, . . . , Ln+1 on F1, . . . , Fn, and ∼= (Ln+1)pi at pi = Fi ∩ D, for 1 ≤ i ≤ n. Stability D, respectively, plus isomorphisms (Li)pi of ι∗L implies that the Li all have degree zero. Therefore the fibre of M0 → nF + D over the point F1 + . . . + Fn + D is isomorphic to Pic0F1 × · · · × Pic0Fn ∼= F1 × · · · × Fn. Similarly, the elliptic fibration S0 → P1 induces a Lagrangian fibration HilbnS0 −→ SymnP1 ∼= Pn whose fibre over a general point F1 + . . . + Fn ∈ nF = SymnP1 ∼= Pn is also isomorphic to F1 × · · · × Fn. Since these Lagrangian fibrations both admit global sections, they must be birational. (cid:3) REFERENCES 17 As before, a birational modification of the relative moduli space M/∆ now yields the following. Corollary 14 The Hilbert scheme HilbnS0 can be deformed to a Beauville-Mukai integrable system, as a Lagrangian fibration over Pn. A Beauville-Mukai integrable system over Pn will have a discriminant locus of degree 6(n + 3) (see [21]), and general singular fibres will be semistable. The discriminant locus of the Lagrangian fibration on HilbnS0 consists of 24 hyperplanes and a degree 2n − 2 hypersurface R. The latter is the image of the 'big' diagonal under the projection P1 × · · · × P1 −→ SymnP1 ∼= Pn, and its degree can be calculated by identifying Pn with the space of homogeneous polynomials f of degree n; R is then given by the discriminant of f , i.e., the resultant of f and f ′, which has degree 2n − 2. The singular fibres above the hyperplanes are semistable, whereas those over R are not. Indeed, the behaviour is similar to the n = 2 case: over a general point of R, the singular fibre will look like the product of n − 2 elliptic curves and a singular fibre over the conic in the n = 2 case. Thus we once again see how non-semistable fibres can deform to semistable fibres under a small deformation of the Lagrangian fibration. Moreover, we expect that R should again be counted with multiplicity three, so that the total degree of the discriminant locus, 24 + 3(2n − 2) = 6(n + 3), is preserved under the deformation. References [1] A. Altman and S. Kleiman, Compactifying the Picard scheme, Adv. in Math. 35 (1980), no. 1, 50 -- 112. [2] A. Beauville, Vari´et´es Kahleriennes dont la premi`ere classe de Chern est nulle, Jour. Diff. Geom. 18 (1983), 755 -- 782. [3] A. Beauville, Counting rational curves on K3 surfaces, Duke Math. J. 97 (1999), no. 1, 99 -- 108. [4] W. Barth, K. Hulek, C. Peters, and A. Van de Ven, Compact complex surfaces, 2nd ed., Ergebnisse der Mathematik (3) 4, Springer, Berlin, 2004. [5] X. Chen, Rational curves on K3 surfaces, J. Alg. Geom. 8 (1999), 245 -- 278. [6] C. Galati and A. Knutsen, On the existence of curves with Ak-singularities on K3 surfaces, Math. Re. Lett. 21 (2014), no. 5, 1069 -- 1109. [7] D. Huybrechts, Birational symplectic manifolds and their deformations, J. Differen- tial Geom. 45 (1997), no. 3, 488 -- 513. [8] J.-M. Hwang, Base manifolds for fibrations of projective irreducible symplectic man- ifolds, Invent. Math. 174 (2008), no. 3, 625 -- 644. 18 REFERENCES [9] J.-M. Hwang and K. Oguiso, Characteristic foliation on the discriminant hyper- surface of a holomorphic Lagrangian fibration, Amer. J. Math. 131 (2009), no. 4, 981 -- 1007. [10] J.-M. Hwang and K. Oguiso, Multiple fibers of holomorphic Lagrangian fibrations, Commun. Contemp. Math. 13 (2011), no. 2, 309 -- 329. [11] L. Kamenova, Hyperkahler fibrations and Hilbert schemes, MIT PhD thesis, 2006 (available from http://hdl.handle.net/1721.1/34686). [12] C. Lehn, Deformations of Lagrangian subvarieties of holomorphic symplectic mani- folds, Math. Res. Lett. 23 (2016), no. 2, 473 -- 497. [13] D. Matsushita, On singular fibres of Lagrangian fibrations over holomorphic sym- plectic manifolds, Math. Ann. 321 (2001), no. 4, 755 -- 773. [14] D. Matsushita, Higher direct images of dualizing sheaves of Lagrangian fibrations, Amer. J. Math. 127 (2005), no. 2, 243 -- 259. [15] D. Matsushita, A canonical bundle formula for projective Lagrangian fibrations, preprint arXiv:0701.0122. [16] D. Matsushita, On deformations of Lagrangian fibrations, K3 surfaces and their moduli, 237 -- 243, Progr. Math. 315, Birkhauser/Springer, Cham, 2016. [17] S. Mukai, Symplectic structure of the moduli space of simple sheaves on an abelian or K3 surface, Invent. Math. 77 (1984), 101 -- 116. [18] K. Oguiso, Local families of K3 surfaces and applications, J. Algebraic Geom. 12 (2003), no. 3, 405 -- 433. [19] K. Oguiso, Shioda-Tate formula for an abelian fibered variety and applications, J. Korean Math. Soc. 46 (2009), no. 2, 237 -- 248. [20] J. Sawon, Abelian fibred holomorphic symplectic manifolds, Turkish Jour. Math. 27 (2003), no. 1, 197 -- 230. [21] J. Sawon, On the discriminant locus of a Lagrangian fibration, Math. Ann. 341 (2008), no. 1, 201 -- 221. [22] J. Sawon, Twisted Fourier-Mukai transforms for holomorphic symplectic four-folds, Adv. Math. 218 (2008), no. 3, 828 -- 864. [23] J. Sawon, Deformations of holomorphic Lagrangian fibrations, Proc. Amer. Math. Soc. 137 (2009), no. 1, 279 -- 285. [24] J. Sawon, On Lagrangian fibrations by Jacobians I , J. Reine Angew. Math. 701 (2015), 127 -- 151. [25] J. Sawon, A finiteness theorem for Lagrangian fibrations, J. Algebraic Geom. 25 (2016), 431 -- 459. [26] T. Shioda, On elliptic modular surfaces, J. Math. Soc. Japan 24 (1972), 20 -- 59. REFERENCES 19 [27] C.T. Simpson, Moduli of representations of the fundamental group of a smooth pro- jective variety I , Publ. Math. I.H.E.S. 79 (1994), 47 -- 129. [28] C. Voisin, Sur la stabilit´e des sous-variete lagrangiennes des vari´et´e symplectiques holomorphes, Complex projective geometry, London Math. Soc. Lecture Note Ser. 179 (1992), 294 -- 303, Cambridge Univ. Press, Cambridge. Department of Mathematics University of North Carolina Chapel Hill NC 27599-3250 USA [email protected] sawon.web.unc.edu
1109.3294	1	1109	2011-09-15T09:03:12	Remarks on the Milnor conjecture over schemes	[ "math.AG", "math.KT", "math.NT" ]	The Milnor conjecture has been a driving force in the theory of quadratic forms over fields, guiding the development of the theory of cohomological invariants, ushering in the theory of motivic cohomology, and touching on questions ranging from sums of squares to the structure of absolute Galois groups. Here, we survey some recent work on generalizations of the Milnor conjecture to the context of schemes (mostly smooth varieties over fields of characteristic not 2). Surprisingly, a version of the Milnor conjecture fails to hold for certain smooth complete p-adic curves with no rational theta characteristic (this is the work of Parimala, Scharlau, and Sridharan). We explain how these examples fit into the larger context of an unramified Milnor question, offer a new approach to the question, and discuss new results in the case of curves over local fields and surfaces over finite fields.	math.AG	math	Remarks on the Milnor conjecture over schemes Asher Auel Abstract. The Milnor conjecture has been a driving force in the theory of qua- dratic forms over fields, guiding the development of the theory of coho- mological invariants, ushering in the theory of motivic cohomology, and touching on questions ranging from sums of squares to the structure of ab- solute Galois groups. Here, we survey some recent work on generalizations of the Milnor conjecture to the context of schemes (mostly smooth vari- eties over fields of characteristic 6= 2). Surprisingly, a version of the Milnor conjecture fails to hold for certain smooth complete p-adic curves with no rational theta characteristic (this is the work of Parimala, Scharlau, and Sridharan). We explain how these examples fit into the larger context of the unramified Milnor question, offer a new approach to the question, and discuss new results in the case of curves over local fields and surfaces over finite fields. The first cases of the (as of yet unnamed) Milnor conjecture were studied in Pfister's Habilitationsschrift [85] in the mid 1960s. As Pfister [86, p. 3] himself points out, "[the Milnor conjecture] stimulated research for quite some time." Indeed, it can be seen as one of the driving forces in the theory of quadratic forms since Milnor's original formulation [66] in the early 1970s. The classical cohomological invariants of quadratic forms (rank, dis- criminant, and Clifford–Hasse–Witt invariant) have a deep connection with the history and development of the subject. In particular, they are used in the classification (Hasse–Minkowski local-global theorem) of quadratic forms over local and global fields. The first "higher invariant" was de- scribed in Arason's thesis [1], [3]. The celebrated results of Merkurjev [62] and Merkurjev–Suslin [64] settled special cases of the Milnor conjecture in the early 1980s, and served as a starting point for Voevodsky's development of the theory of motivic cohomology. Other special cases were settled by Arason–Elman–Jacob [5] and Jacob–Rost [46]. Voevodsky's motivic coho- mology techniques [99] ultimately led to a complete solution of the Milnor conjecture, for which he was awarded the 2002 Fields Medal. The consideration of quadratic forms over rings (more general than fields) has its roots in the number theoretic study of lattices (i.e. quadratic forms over Z) by Gauss as well as the algebraic study of division algebras and hermitian forms (i.e. quadratic forms over algebras with involution) by 2010 Mathematics Subject Classification. Primary 11-02, 19-02; Secondary 11Exx, 19G12. 2 Asher Auel Albert. A general framework for the study of quadratic forms over rings was established by Bass [17], with the case of (semi)local rings treated in depth by Baeza [9]. Bilinear forms over Dedekind domains (i.e. unimodular lat- tices) were studied in a number theoretic context by Frohlich [35], while the consideration of quadratic forms over algebraic curves (and their function fields) was initiated by Geyer, Harder, Knebusch, Scharlau [44], [39], [53], [55]. The theory of quadratic (and bilinear) forms over schemes was devel- oped by Knebusch [54], [52], and utilized by Arason [2], Dietel [27], Parimala [76], [83], Fern´andez-Carmena [32], Sujatha [94], [82], Arason–Elman–Jacob [6], [7], and others. A theory of symmetric bilinear forms in additive and abelian categories was developed by Quebbemann–Scharlau–Schulte [87], [88]. Further enrichment came eventually from the triangulated category techniques of Balmer [10], [11], [12], and Walter [101]. This article will focus on progress in generalizing the Milnor conjecture to these contexts. These remarks grew out of a lecture at the RIMS-Camp-Style semi- nar "Galois-theoretic Arithmetic Geometry" held October 19-23, 2010, in Kyoto, Japan. The author would like to thank the organizers for their wonderful hospitality during that time. He would also like to thank Stefan Gille, Moritz Kerz, R. Parimala, and V. Suresh for many useful conversa- tions. The author acknowledges the generous support of the Max Plank Institute for Mathematics in Bonn, Germany where this article was written under excellent working conditions. This author is also partially supported by National Science Foundation grant MSPRF DMS-0903039. Conventions. A graded abelian group or ring Qn≥0 M n will be denoted by M •. If 0 ⊂ ··· ⊂ N 2 ⊂ N 1 ⊂ N 0 = M is a decreasing filtration of a ring M by ideals, denote by N •/N •+1 = Qn≥0 N n/N n+1 the associated graded ring. Denote by 2M the elements of order 2 in an abelian group M . All abelian groups will be written additively. §1. The Milnor conjecture over fields ⊗• 2 ) = Ln≥2 H n(F, µ Let F be a field of characteristic 6= 2. The total Milnor K-ring K • M(F ) = T •(F ×)/ha⊗ (1− a) : a ∈ F ×i was introduced in [66]. The total Galois co- ⊗n homology ring H •(F, µ 2 ) is canonically isomorphic, under our hypothesis on the characteristic of F , to the total Galois cohomol- ogy ring H •(F, Z/2Z) with coefficients in the trivial Galois module Z/2Z. The Witt ring W (F ) of nondegenerate quadratic forms modulo hyperbolic forms has an decreasing filtration 0 ⊂ ··· ⊂ I 1(F ) ⊂ I 0(F ) = W (F ) gen- erated by powers of the fundamental ideal I(F ) of even rank forms. The Milnor conjecture relates these three objects: Milnor K-theory, Galois co- homology, and quadratic forms. The quotient map K 1 M(F ) = F × → F ×/F ×2 ∼= H 1(X, µ2) induces ⊗• a graded ring homomorphism h• : K • 2 ) called the norm residue symbol by Bass–Tate [16]. The Pfister form map K 1 M(F ) = F × → I(F ) given by a 7→≪ a ≫=< 1,−a > induces a group homo- morphism K 1 M(F )/2 → I 1(F )/I 2(F ) (see Scharlau [91, 2 Lemma 12.10]), M(F )/2 → H •(F, µ Remarks on the Milnor conjecture over schemes 3 which extends to a surjective graded ring homomorphism s• : K • I •(F )/I •+1(F ), see Milnor [66, Thm. 4.1]. M(F )/2 → Theorem 1 (Milnor conjecture). Let F be a field of characteristic 6= 2. There exists a graded ring homomorphism e• : I •(F )/I •+1(F ) → ⊗• H •(F, µ 2 ) called the higher invariants of quadratic forms, which fits into the following diagram K • M(F )/2 h• H •(F, µ ⊗• 2 ) I •(F )/I •+1(F ) s• 6nnnnnnnnnnnn e• of isomorphisms of graded rings. Many excellent introductions to the Milnor conjecture and its proof exist in the literature. For example, see the surveys of Kahn [47], Friedlander– Rapoport–Suslin [34], Friedlander [33], Pfister [84], and Morel [68]. The conjecture breaks up naturally into three parts: the conjecture for the norm residue symbol h•, the conjecture for the Pfister form map s•, and the conjecture for the higher invariants e•. Milnor [66, Question 4.3, §6] originally made the conjecture for h• and s•, which was already known for finite, local, global, and real closed fields, see [66, Lemma 6.2]. For general fields, the conjecture for h1 follows from Hilbert's theorem 90, and for s1 and e1 by elementary arguments. The conjecture for s2 is easy, see Pfister [85]. Merkurjev [62] proved the conjecture for h2 (hence for e2 as well), with alternate proofs given by Arason [4], Merkurjev [63], and Wadsworth [100]. The conjecture for h3 was settled by Merkurjev–Suslin [64] (and independently by Rost [89]). The conjecture for e• can be divided ⊗n into two parts: to show the existence of maps en : I n(F ) → H n(X, µ 2 ) (which are a priori only defined on generators, the Pfister forms), and then to show they are surjective. The existence of e3 was proved by Arason [1], [3]. The existence of e4 was proved by Jacob–Rost [46] and independently Szyjewski [93]. Voevodsky [99] proved the conjecture for h•. Orlov–Vishik– Voevodsky [73] proved the conjecture for s•, with different proofs given by Morel [69] and Kahn–Sujatha [48]. 1.1. Classical invariants of quadratic forms The theory of quadratic forms over a general field has its genesis in Witt's famous paper [102]. Because of the assumption of characteristic 6= 2, we do not distinguish between quadratic and symmetric bilinear forms. The orthogonal sum (V, b) ⊥ (V ′, b′) = (V ⊕ V ′, b + b′) and tensor product (V, b) ⊗ (V ′, b′) = (V ⊗ V ′, b ⊗ b′) give a semiring structure on the set of isometry classes of symmetric bilinear forms over F . The hyperbolic plane is the symmetric bilinear form (H, h), where H = F 2 and h((x, y), (x′, y′)) = xy′ + x′y. The Witt ring of symmetric bilinear forms is the quotient of the Grothendieck ring of nondegenerate symmetric bilinear forms over F with / / 6 4 Asher Auel respect to ⊥ and ⊗, modulo the ideal generated by the hyperbolic plane, see Scharlau [91, Ch. 2]. The rank of a bilinear form (V, b) is the F -vector space dimension of V . Since the hyperbolic plane has rank 2, the rank modulo 2 is a well defined invariant of an element of the Witt ring, and gives rise to a surjective ring homomorphism e0 : W (F ) = I 0(F ) → Z/2Z = H 0(F, Z/2Z) whose kernel is the fundamental ideal I(F ). The signed discriminant of a non-degenerate bilinear form (V, b) is de- fined as follows. Choosing an F -vector space basis v1, . . . , vr of V , we con- sider the Gram matrix Mb of b, i.e. the matrix given by Mb = (b(vi, vj)). Then b is given by the formula b(v, w) = vtMbw, where v, w ∈ F r ∼= V . The Gram matrix of b, with respect to a different basis for V with change of basis matrix T , is T tMbT . Thus det Mb ∈ F ×, which depends on the choice of basis, is only well-defined up to squares. For a ∈ F ×, denote by (a) its class in the abelian group F ×/F ×2. The signed discriminant of (V, b) is defined as (−1)r(r−1)/2 det Mb ∈ F ×/F ×2. Introducing the sign into the signed discriminant ensures its vanishing on the ideal of hyperbolic forms, hence it descents to the Witt group. While the signed discriminant is not additive on W (F ), its restriction to I(F ) gives rise to a group homomorphism e1 : I(F ) → F ×/F ×2 ∼= H 1(F, µ2) which is easily seen to be surjective. It's then not difficult to check that its kernel coincides with the square I 2(F ) of the fundamental ideal. See Scharlau [91, §2.2] for more details. The Clifford invariant of a non-degenerate symmetric bilinear form (V, b) is defined in terms of its Clifford algebra. The Clifford algebra C(V, b) of (V, b) is the quotient of the tensor algebra T (V ) = Lr≥0 V ⊗r by the two-sided ideal generated by {v ⊗ w + w ⊗ v − b(v, w) : v, w ∈ V }. If (V, b) has rank r, then C(V, b) = C0(V, b)⊕ C1(V, b) inherits the structure of a Z/2Z-graded semisimple F -algebra of F -dimension 2r, see Scharlau [91, §9.2]. By the structure theory of the Clifford algebra, C(V, b) or C0(V, b) is a central simple F -algebra depending on whether (V, b) has even or odd rank, respectively. The Clifford invariant c(V, b) ∈ Br(F ) is then the Brauer class of C(V, b) or C0(V, b), respectively. Since the Clifford algebra and its even subalgebra carry canonical involutions of the first kind, their respective classes in the Brauer group are of order 2, see Knus [57, §IV.7.8]. While the Clifford invariant is not additive on W (F ), its restriction to I 2(F ) gives rise to a group homomorphism e2 : I 2(F ) → 2Br(F ) ∼= H 2(F, µ2) ∼= H 2(F, µ ⊗2 2 ), see Knus [57, IV Prop. 8.1.1]. Remarks on the Milnor conjecture over schemes 5 Any symmetric bilinear form (V, b) over a field of characteristic 6= 2 can be diagonalized, i.e. a basis can be chosen for V so that the Gram ma- trix Mb is diagonal. For a1, . . . , ar ∈ F ×, we write < a1, . . . , ar > for the standard symmetric bilinear form with associated diagonal Gram matrix. For a, b ∈ F ×, denote by (a, b)F the (quaternion) F -algebra generated by symbols x and y subject to the relations x2 = a, y2 = b, and xy = −yx. For example, (−1,−1)R is Hamilton's ring of quaternions. Then the dis- criminant and Clifford invariant can be conveniently calculated in terms of a diagonalization. For (V, b) ∼=< a1, . . . , ar >, we have d±(V, b) = ((−1)r(r−1)/2a1 ··· ar) ∈ F ×/F ×2 and (1) c(V, b) = α(r)(−1, a1 ··· ar)F + β(r)(−1,−1)F +X (aj, aj)F ∈ 2Br(F ) i<j where α(r) = (r − 1)(r − 2) 2 , β(r) = (r + 1)r(r − 1)(r − 2) 24 , see Lam [59], Scharlau [91, II.12.7] or Esnault–Kahn–Levine–Viehweg [30, §1]. §2. Globalization of cohomology theories Generalizations (what we will call globalizations) of the Milnor conjec- ture to the context of rings and schemes have emerged from many sources, see Parimala [75], Colliot-Th´el`ene–Parimala [21], Parimala–Sridharan [78], Monnier [67], Pardon [74], Elbaz-Vincent–Muller-Stach [28], Gille [41], and Kerz [49]. To begin with, one must ask for appropriate globalizations of the objects in the conjecture: Milnor K-theory, Galois cohomology theory, and the Witt group with its fundamental filtration. While there are many possible choices of such globalizations, we will focus on two types: global and unramified. 2.1. Global globalization Let F be a field of characteristic 6= 2. Let FieldF (resp. RingF ) be the category of fields (resp. commutative unital rings) with an F -algebra struc- ture together with F -algebra homomorphisms. Let SchF be the category of F -schemes, and SmF the category of smooth F -schemes. We will denote, by the same names, the associated (large) Zariski sites. Let Ab (resp. Ab•) be the category of abelian groups (resp. graded abelian groups), we will always consider Ab as embedded in Ab• in degree 0. Let M • : FieldF → Ab• be a functor. A globalization of M • to rings (resp. schemes) is a functor M • : RingF → Ab• (resp. contravariant functor If M • is a globalization of M • to M • : SchF → Ab•) extending M •. 6 Asher Auel rings, then we can define a globalization to schemes by taking the sheaf M• associated to the presheaf U 7→ M •(Γ(U, OU )) on SchF (always considered with the Zariski topology). "Naıve" Milnor K-theory. For a commutative unital ring R, mimicking Milnor's tensorial construction (with the additional relation that a⊗(−a) = 0, which is automatic for fields) yields a graded ring K • M(R), which should be referred to as "naıve" Milnor K-theory. This already appears in Guin [43, §3] and later studied by Elbaz-Vincent–Muller-Stach [28]. Naıve Milnor K-theory has some bad properties when R has small finite residue fields, see Kerz [50] who also provides a improved Milnor K-theory repairing these defects. Thomason [96] has shown that there exists no globalization of Milnor K-theory to smooth schemes which satisfies A1-homotopy invariance and has a functorial homomorphism to algebraic K-theory. ⊗• 2 ) on SchF . ´Etale cohomology. ´Etale cohomology provides a natural globalization of Galois cohomology to schemes. We will thus consider the functor X 7→ H • ´et(X, µ Global Witt group. For a scheme X, the global Witt group W (X) of regular symmetric bilinear forms introduced by Knebusch [52] provides a natural globalization of the Witt group to schemes. Other possible global- izations are obtained from the Witt groups of triangulated category with duality introduced by Balmer [10], [11], [12], [13]. These include: the de- rived Witt group of the bounded derived category of coherent locally free OX -modules; the coherent Witt group of the bounded derived category of quasicoherent OX -modules with coherent cohomology (assuming X has a dualizing complex, see Gille [40, §2.5], [41, §2]); the perfect Witt group of the derived category of perfect complexes of OX -modules. The global and derived Witt groups are canonically isomorphic by Balmer [12, Thm. 4.3]. All of the above Witt groups are isomorphic (though not necessarily canonically) if X is assumed to be regular. Fundamental filtration and the classical invariants. Globalizations of the classical invariants of quadratic forms are defined as follows. Let (E , q) be a regular symmetric bilinear form of rank n on X. The rank (modulo 2) of E gives rise to a functorial homomorphism e0 : W (X) → Homcont(X, Z/2Z) = H 0 ´et(X, Z/2Z), whose kernel I 1(X) is called the fundamental ideal of W (X). The signed discriminant form (det E , (−1)n(n−1)/2 det q) gives rise to a functorial homomorphism e1 : I 1(X) → H 1 ´et(X, µ2) see Knus [57, III §4.2] and Milne [65, III §4]. Alternatively, the center of the (even) Clifford OX -algebra of (E , q) defines a class in H 1 ´et(X, Z/2Z) called the Arf invariant, which coincides with the signed discriminant under the canonical morphism H 1 ´et(X, µ2) (see Knus [57, IV Prop. 4.6.3] or Parimala–Srinivas [81, §2.2]). Denote the kernel of e1 by I 2(X), ´et(X, Z/2Z) → H 1 Remarks on the Milnor conjecture over schemes 7 which is an ideal of W (X). Note that I 2(X) may not be the square of the ideal I 1(X). The Clifford OX -algebra C (E , q) gives rise to a functorial homomor- phism e2 : I 2(X) → 2Br(X) called the Clifford invariant, see Knus–Ojanguren [58] and Parimala–Srinivas [81, §2]. Denote the kernel of e2 by I 3(X), which is an ideal of W (X). As Parimala–Srinivas [81, p. 223] point out, there is no functorial map I 2(X) → H 2 ´et(X, µ2) lifting the Clifford invariant. Instead, we can work with Grothendieck–Witt groups. The rank (modulo 2) gives rise to a func- torial homomorphism ge0 : GW (X) → H 0 ´et(X, Z/2Z) with kernel denoted by GI(X). The signed discriminant gives rise to a functorial homomorphism ge1 : GI 1(X) → H 1 ´et(X, µ2) with kernel denoted by GI 2(X). The class of the Clifford OX -algebra, together with it's canonical involution (via the "involutive" Brauer group construction of Parimala–Srinivas [81, §2]), gives rise to a functorial homo- morphism ge2 : GI 2(X) → H 2 ´et(X, µ2) also see Knus–Parimala–Sridharan [56]. Denote the kernel of ge2 by GI 3(X), which is an ideal of GW (X). Lemma 2.1. Let X be a scheme satisfying 2Br(X) ∼= 2H 2 ´et(X, Gm). Then under the quotient map GW (X) → W (X), the image of the ideal GI n(X) is precisely the ideal I n(X), for n ≤ 3. Proof. For n = 1, 2 this is a consequence of the following diagram with exact rows and columns 0 0 K0(X) H K0(X) H 0 0 / GI n(X) / GI n−1(X) gen−1 H n−1 ´et (X, Z/2Z) / I n(X) / I n−1(X) en−1 / H n−1 ´et (X, Z/2Z) / 0 / 0 0 0 0 which is commutative since hyperbolic spaces have even rank and trivial signed discriminant. Here, K0(X) is the Grothendieck group of locally free / / / / / / / / / 8 Asher Auel OX -modules of finite rank and H is the hyperbolic map V 7→ H(V ) = (cid:0)V ⊕ V ∨, ((v, f ), (w, g)) 7→ f (w) + g(x)(cid:1). For n = 3, we have the formula ge2(H(V )) = c1(V , µ2) due to Esnault– Kahn–Viehweg [29, Prop. 5.5] (combined with (1)). Here c1(−, µ2) is the 1st Chern class modulo 2, defined as the first coboundary map in the long-exact sequence in ´etale cohomology ··· → Pic(X) 2−→ Pic(X) c1−→ H 2 ´et(X, µ2) → H 2 ´et(X, Gm) 2−→ H 2 ´et(X, Gm) → ··· arising from the Kummer exact sequence 1 → µ2 → Gm 2−→ Gm → 1, see Grothendieck [42]. The claim then follows by a diagram chase through 0 0 0 0 0 0 / K ′ 0(X) H / GI 3(X) / K0(X) H / GI 2(X) / I 3(X) / I 2(X) det / Pic(X)/2 / 0 ge2 e2 c1 / H 2 ´et(X, µ2) / 2Br(X) 0 0 0 where the right vertical column arises from the Kummer sequence, and K ′ 0(X) is the subgroup of K0(X) generated by locally free OX -modules whose determinant is a square. Q.E.D. Remark 2.1. The hypothesis that 2Br(X) = 2H 2 ´et(X, Gm) is satisfied if X is a quasi-compact quasi-separated scheme admitting an ample invertible sheaf by de Jong's extension [26] (see also [61, Th. 2.2.2.1]) of a result of Gabber [36]. The existence of global globalizations of the higher invariants (e.g. a globalization of the Arason invariant) remains a mystery. Esnault–Kahn– Levine–Viehweg [30] have shown that for a regular symmetric bilinear form (E , q) that represents a class in GI 3(X), the obstruction to having an Ara- son invariant in H 3 ´et(X, Z/2Z) is precisely the 2nd Chern class c2(E ) ∈ CH 2(X)/2 in the Chow group modulo 2 (note that the invariant c(E ) ∈ Pic(X)/2 of [30] is trivial if (E , q) represents a class in GI 3(X)). They also provide examples where this obstruction does not vanish. On the other hand, higher cohomological invariants always exist in unramified cohomol- ogy. / / / / / / / / / / Remarks on the Milnor conjecture over schemes 9 2.2. Unramified globalization A functorial framework for the notion of "unramified element" is estab- lished in Colliot-Th´el`ene [20, §2]. See also the survey by Zainoulline [103, §3]. Rost [90, Rem. 5.2] gives a different perspective in terms of cycle mod- ules, also see Morel [69, §2]. Assume that X has finite Krull dimension and is equidimensional over a field F . For simplicity of exposition, assume that X is integral. Denote by X (i) its set of codimension i points. Denote by LocalF the category of local F -algebras together with local F -algebra morphisms. Given a functor M • : LocalF → Ab•, call M • ur(X) = \ x∈X (1) im(cid:0)M •(OX,x) → M •(F (X))(cid:1) the group of unramified elements of M • over X. Then X 7→ M • globalization of M • to schemes. ur(X) is a Given a functor M • : SchF → Ab•, there is a natural map M •(X) → M • ur(X). If this map is injective, surjective, or bijective we say that the in- jectivity, weak purity, or purity property hold, respectively. Whether these properties hold for various functors M • and schemes X is the subject of many conjectures and open problems, see Colliot-Th´el`ene [20, §2.2] for ex- amples. Unramified Milnor K-theory. Define the unramified Milnor K-theory (resp. modulo 2) of X to be the graded ring of unramified elements K • M,ur(X) (resp. K • M,ur/2(X)) of the "naıve" Milnor K-theory (resp. modulo 2) func- tor K • M be the Zariski sheaf on SchF associated to "naıve" Milnor K-theory and K• M/2 the asso- ciate sheaf quotient, which is the Zariski sheaf associated to the presheaf M,ur(X) = Γ(X,K• U 7→ K • M) and K • M/2) when X is smooth over an infinite field (compare with the remark in §2.1) by the Bloch–Ogus–Gabber theorem for Milnor K-theory, see Colliot-Th´el`ene–Hoobler–Kahn [23, Cor. 5.1.11, §7.3(5)]. Also, see Kerz [49]. Note that the long exact sequence in Zariski cohomology yields a short exact sequence M(U )/2, see Morel [69, Lemma 2.7]. Then K • M,ur/2(X) = Γ(X,K• M/2) restricted to LocalF , see §2.1. Let K• M (resp. K • 0 → K • M,ur(X)/2 → K • M,ur/2(X) → 2H 1(X,K• M) → 0 still assuming X is smooth over an infinite field. ⊗• ⊗• ur(X, µ ´et(−, µ 2 ). Letting H• ´et(−, µ Unramified cohomology. Define the unramified ´etale cohomology (mod- ⊗• ulo 2) of X to be the graded ring of unramified elements H • 2 ) of the functor H • ´et be the Zariski sheaf on SchF associ- ated to the functor H • ur(X, Z/2Z) when X is smooth over a field of characteristic 6= 2 by the exactness of the Gersten complex for ´etale cohomology, see Bloch–Ogus [19, §2.1, Thm. 4.2]. Unramified fundamental filtration of the Witt group. Define the unram- ified Witt group of X to be the abelian group of unramified elements Wur(X) of the global Witt group functor W . Letting W be the Zariski sheaf asso- ciated to the global Witt group functor, then Wur(X) = Γ(X,W) when X 2 ), then Γ(X,H• ´et) = H • 10 Asher Auel is regular over a field of characteristic 6= 2 by Ojanguren–Panin [72] (also see Morel [69, Thm. 2.2]). Writing I n ur(X) = I n(F (X)) ∩ Wur(X), then the functors I n ur(−) are Zariski sheaves (still assuming X is regular), denoted by I n, which form a filtration of W, see Morel [69, Thm. 2.3]. Note that the long exact sequence in Zariski cohomology yields short exact sequences 0 → I n ur(X)/I n+1 ur (X) → I n/I n+1(X) → H 1(X,I n+1)′ → 0 where H 1(X,I n+1)′ = ker(cid:0)H 1(X,I n) → H 1(X,I n+1)(cid:1) and we are still assuming X is regular over a field of characteristic 6= 2. If the obstruction group H 1(X,I n+!)′ is nontrivial, then not every element of I n/I n+1(X) is represented by a quadratic form on X. If X is the spectrum of a regular (X) = I •/I •+1(X), see Morel [69, Thm. 2.12]. local ring, then I • ur(X) does not necessarily mean the nth power of Iur(X). This is true, however, when X is the spectrum of a regular local ring containing an infinite field of characteristic 6= 2, see Kerz–Muller-Stach [51, Cor. 0.5]. Remark 2.2. As before, the notation I n ur(X)/I •+1 ur 2.3. Gersten complexes Gersten complexes (Cousin complexes) exists in a very general frame- work, but for our purposes, we will only need the Gersten complex for Milnor K-theory, ´etale cohomology, and Witt groups. Gersten complex for Milnor K-theory. Let X be a regular excellent M) denote the Gersten complex for Milnor integral F -scheme. Let C(X, K n K-theory 0 / K n M(F (X)) ∂Kn M / Lx∈X(1) K n−1 M (F (x)) n−1 M ∂K / Ly∈X(2) K n−2 M (F (y)) / ··· where ∂KM is the "tame symbol" homomorphism defined in Milnor [66, Lemma 2.1]. We have that H 0(C(X, K n M,ur(X). See Rost [90, §1] or Fasel [31, Ch. 2] for more details. We will also consider the Gersten complex (C, K n M/2) for Milnor K-theory modulo 2, for which we have that H 0(C(X, K n M)) = K n M/2)) = K n M,ur/2(X). Gersten complex for ´etale cohomology. Let X be a smooth integral F - scheme, with F of characteristic 6= 2. Let C(X, H n) denote the Gersten complex for ´etale cohomology 0 ∂Hn ∂Hn−1 / Hn(F (X)) Hn−1(F (x)) / Lx∈X(1) ⊗n 2 ) and ∂H is the homomorphism induced from where H n(−) = H n(−, µ the spectral sequence associated to the coniveau filtration, see Bloch–Ogus [19]. Then we have that C(X, H n) is a resolution of H n / Ly∈X(2) Hn−2(F (y)) / ··· ur(X, µ ⊗n 2 ). / / / / / / / / Remarks on the Milnor conjecture over schemes 11 Gersten complex for Witt groups. Let X be a regular integral F -scheme of finite Krull dimension. Let C(X, W ) denote the Gersten–Witt complex 0 / W (F (X)) ∂W / Lx∈X(1) W (F (x)) ∂W / Ly∈X(2) W (F (y)) / ··· where ∂W is the homomorphism induced from the second residue map for a set of choices of local parameters, see Balmer–Walter [15]. Because of these choices, C(X, W ) is only defined up to isomorphism, though there is a canonical complex defined in terms of Witt groups of finite length modules over the local rings of points. We have that H 0(C(X, W )) = Wur(X). Fundamental filtration. The filtration of the Gersten complex for Witt groups induced by the fundamental filtration was first studied methodically by Arason–Elman–Jacob [5], see also Parimala–Sridharan [78], Gille [41], and Fasel [31, §9]. fundamental filtration as follows: The differentials of the Gersten complex for Witt groups respect the ∂I n(cid:16) M x∈X (p) I n(F (x))(cid:17) ⊂ M y∈X (p+1) I n−1(F (y)), see Fasel [31, Thm. 9.2.4] and Gille [41]. Thus for all n ≥ 0 we have complexes C(X, I n) 0 / I n(F (X)) ∂In / Lx∈X(1) I n−1(F (x)) ∂In−1 / Ly∈X(2) I n−2(F (y)) / ··· which provide a filtration of C(X, W ) in the category of complexes of abelian groups. Here we write I n(−) = W (−) for n ≤ 0. We have that H 0(C(X, I n)) = I n The canonical inclusion C(X, I n+1) → C(X, I n) respects the differen- tials, and so defines a cokernel complex C(X, I n/I n+1) ur(X). 0 / I n/I n+1(F (X)) / Lx∈X(1) I n−1/I n(F (x)) / Ly∈X(2) I n−2/I n−1(F (y)) / ··· see Fasel [31, D´ef. 9.2.10], where I n/I n+1(L) = I n(L)/I n+1(L) for a field L. We have that H 0(C(X, I n/I n+1)) = I n/I n+1(X) Unramified norm residue symbol. The norm residue symbol for fields provides a morphism of complexes hn : C(X, K n M/2) → C(X, H n), where the map on terms of degree j is hn−j. By the Milnor conjecture for fields, this is an isomorphism of complexes. Upon restriction, we have an isomor- ⊗n phism hn ur : K n 2 ). Upon further restriction, we have an injection hn ur(X, µ M,ur(X)/2 → H n Unramified Pfister form map. The Pfister form map for fields provides a morphism of complexes sn : C(X, K n M/2) → C(X, I n/I n+1), where the map on terms of degree j is sn−j. Upon restriction, we have a homomorphism sn ur : K n M,ur/2(X) → I n/I n+1(X). See Fasel [31, Thm. 10.2.6]. M,ur/2(X) → H n ur : K n ur(X, µ ⊗n 2 ). / / / / / / / / / / / / 12 Asher Auel Unramified higher cohomological invariants. By the Milnor conjecture for fields, there exists a higher cohomological invariant morphism of com- plexes en : C(X, I n) → C(X, H n), where the map on terms of degree j is en−j. Upon restriction, we have homomorphisms en ur(X) → ur ⊗n H n ur(X, µ 2 ). Furthermore, on the level of complexes, the higher cohomological invari- ⊗n 2 ) factoring through to en (X) → H n : I n ur(X, µ ur : I n ur(X)/I n+1 ur ant morphisms factors through to a morphism of complexes en : C(X, I n/I n+1) → C(X, H n), which by the Milnor conjecture over fields, is an isomorphism. ⊗n Upon restriction, we have isomorphisms en 2 ). Also see Morel [69, §2.3]. ur : I n/I n+1(X) → H n ur(X, µ 2.4. Motivic globalization There is another important globalization of Milnor K-theory and Galois cohomology, but we only briefly mention it here. Conjectured to exist by Beılinson [18] and Lichtenbaum [60], and then constructed by Voevodsky [99], motivic complexes modulo 2 give rise to Zariski and ´etale motivic cohomology groups modulo 2 H n Zar(X, Z/2Z(m)) and H n ´et(X, Z/2Z(m)). For a field F , Nesterenko–Suslin [70] and Totaro [97] establish a canon- ical isomorphism H n M(F )/2 while the work of Bloch, Gabber, and Suslin (see the survey by Geisser [37, §1.3.1]) establishes ⊗n an isomorphism H n 2 ) (for F of characteris- tic 6= 2). The natural pullback map Zar(Spec F, Z/2Z(n)) ∼= K 2 ´et(Spec F, Z/2Z(n)) ∼= H n(F, µ ε∗ : H n Zar(Spec F, Z/2Z(n)) → H n ´et(Spec F, Z/2Z(n)) induced from the change of site ε : X´et → XZar is then identified with the norm residue homomorphism. Thus H n ´et(−, Z/2Z(n)) provide motivic globalizations of the mod 2 Milnor K-theory and Galois co- homology functors, respectively. On the other hand, there does not seem to exist a direct motivic globalization of the Witt group or its fundamental filtration. Zar(−, Z/2Z(n)) and H n §3. Globalization of the Milnor conjecture Unramified. Let F be a field of characteristic 6= 2. Summarizing the results cited in §2.2–2.3, we have a commutative diagram of isomorphisms H• ´et h• ;vvvvvvvvv e• K• M/2 s• I •/I •+1 of sheaves of graded abelian groups on SmF . In particular, we have such a commutative diagram of isomorphisms on the level of global sections. What we will consider as a globalization of the Milnor conjecture - the unramified Milnor question - is a refinement of this. / / ; Remarks on the Milnor conjecture over schemes 13 (S) Let X be a smooth scheme over a field of characteristic 6= 2. Then M,ur(X)/2 → ur gives rise to a homomorphism s• (X). restricting s• ur(X)/I •+1 I • ur : K • ur Question 3.1 (Unramified Milnor question). Let X be a smooth scheme over a field of characteristic 6= 2. Consider the following diagram: K • M,ur(X)/2 ? ur(X)/I •+1 I • ur (X) i• K i• I h• ur H • ur(X, µ ⊗• 2 ) 7nnnnnnnnnnnn e• ur M/2(X) / K• s• ur / I n/I n+1(X) (1) Is the inclusion i• (2) Is the inclusion i• (3) Does the restriction of s• I surjective? K surjective? ur to K • ur(X)/I •+1 I • ur (X)? If so, is it an isomorphism? M,ur(X)/2 have image contained in Note that in degree n, Questions 3.1 (1), (2), and (3) can be rephrased in terms of the obstruction groups, respectively: does H 1(X,I n+1)′ van- ish; does 2H 1(X,Kn ur yield a map 2H 1(X,Kn From now on we shall focus mainly on the unramified Milnor question for quadratic forms (i.e. Question 3.1(1)), which was already explicitly asked by Parimala–Sridharan [78, Question Q]. M) → H 1(X,I n+1)′ and is it an isomorphism? M) vanish; and does the restriction of sn Global Grothendieck–Witt. We mention a global globalization of the Milnor conjecture for quadratic forms. Because of the conditional definition of the global cohomological invariants, we restrict ourselves to the classical invariants on Grothendieck–Witt groups defined in §2.1. Question 3.2 (Global Merkurjev question). Let X be a regular scheme with 2 invertible. For n ≤ 2, consider the homomorphisms, ´et(X, Z/2Z) gen : GI n(X)/GI n+1(X) → H n induced from the (classical) cohomological invariants on Grothendieck–Witt groups. Are they surjective? This can be viewed as a globalization of Merkurjev's theorem. Indeed, first note that the cases n = 0, 1 of Question 3.2 are easy. Next, a conse- quence of Lemma 2.1 is that ge2 : GI 2(X) → H 2 ´et(X, Z/2Z) is surjective if and only e2 : I 2(X) → 2Br(X) is surjective. This, in turn, is a consequence of a positive answer to Question 3.1(1) for any X satisfying weak purity for the Witt group (see §3.1 for examples). Motivic. The globalization of the Milnor conjecture for Milnor K-theory using Zariski and ´etale motivic cohomology modulo 2 (see §2.4) is the (n, n) modulo 2 case of the Beılinson–Lichtenbaum conjecture: for a smooth vari- ety X over a field, the canonical map H n ´et(X, Z/2Z(m)) is an isomorphism for n ≤ m. The combined work of Suslin–Voevodsky [95] Zar(X, Z/2Z(m)) → H n / / / / 7 14 Asher Auel and Geisser–Levine [38] show the Beılinson–Lichtenbaum conjecture to be a consequence of the Bloch–Kato conjecture, a proof of which has been announced by Rost and Voevodsky. the global and unramified Witt groups and cohomology. 3.1. Some purity results In this section we review some of the purity results (see §2.2) relating Purity results for Witt groups. For a survey on purity results for Witt groups, see Zainoulline [103]. Purity for the global Witt group means that the natural map W (X) → Wur(X) is an isomorphism. Theorem 2. Let X be a regular integral noetherian scheme with 2 invertible. Then purity holds for the global Witt group functor under the following hypotheses: (1) X is dimension ≤ 3, (2) X is the spectrum of a regular local ring of dimension ≤ 4, (3) X is the spectrum of a regular local ring containing a field. For part (1), the case of dimension ≤ 2 is due to Colliot-Th´el`ene–Sansuc [22, Cor. 2.5], the case of dimension 3 and X affine is due to Ojanguren– Parimala–Sridharan–Suresh [71], and for the general case (as well as (2)) see Balmer–Walter [15]. For (3), see Ojanguren–Panin [72]. As a consequence, for any scheme over which purity for the Witt group holds, the unramified Milnor question for en (with n ≤ 2) is equivalent to the analogous global Milnor question. This is especially useful for the case of curves. Purity results for ´etale cohomology. For X geometrically locally facto- rial and integral, the purity property holds for unramified cohomology in degree ≤ 1, i.e. H 0 ur(X, Z/2Z) = Z/2Z, ´et(X, Z/2Z) = H 0 and H 1 ´et(X, µ2) = H 1 ur(X, µ2) see Colliot-Th´el`ene–Sansuc [24, Cor. 3.2, Prop. 4.1]. For X smooth over a field of characteristic 6= 2, we have a canoni- cal identification 2Br(X) = H 2 ur(X, µ2) by Bloch–Ogus [19] such that the canonical map H 2 ´et(X, µ2) → H 2 ur(X, µ2) = 2Br(X) is the map arising from the Kummer exact sequence already considered in the proof of Lemma 2.1. 3.2. Positive results We now survey some of the known positive cases of the unramified Milnor question in the literature. Theorem 3 (Kerz–Muller-Stach [51, Cor. 0.8], Kerz [49, Thm. 1.2]). Let R be a local ring with infinite residue field of characteristic 6= 2. Then the unramified Milnor question (all parts of Question 3.1) has a positive answer over Spec R. Hoobler [45] had already proved this in degree 2. Remarks on the Milnor conjecture over schemes 15 The following result was communicated to us by Stefan Gille (who was inspired by Totaro [98]). Theorem 4. Let X be a proper smooth integral variety over a field of characteristic 6= 2. If X is F -rational then the unramified Milnor question (all parts of Question 3.1) has a positive answer over X. M,ur(X), H n ur(X, µ ⊗n 2 ), and I n Proof. The groups K n ur(X) are birational invariants of smooth proper F -varieties. To see this, one can use Colliot- Th´el`ene [20, Prop. 2.1.8e] and the fact that the these functors satisfy weak purity for regular local rings (see Theorem 2). Another proof uses the fact that the complexes C(X, K n M), C(X, H n), and C(X, I n) are cycle modules in the sense of Rost, see [90, Cor. 12.10]. In any case, by Colliot-Th´el`ene [20, Prop. 2.1.9] the pullback induces isomorphisms K n M,ur(Pm) (first ⊗n ⊗n proved by Milnor [66, Thm. 2.3] for P1), H n(F, µ 2 ), and ur(F ) ∼= I n M(F )/2 ∼= I n K n M,ur(X)/2 and I n (X), and the theorem fol- lows from the Milnor conjecture over fields. Q.E.D. ur(Pm, µ ur(F )(Pm) for all n ≥ 0 and m ≥ 1. In particular, K n M(F ) ∼= K n 2 ) ∼= H n (F ) ∼= I n ur(X)/I n+1 ur ur(F )/I n+1 ur The following positive results are known for low dimensional schemes. Recall the notion of cohomological dimension cd (F ) of a field (see Serre [92, I §3.1]), virtual cohomological 2-dimension vcd 2(F ) = cd 2(F (√−1)) and their 2-primary versions. Denoting by d(F ) any of these notions of dimension, note that if d(F ) ≤ k and dim X ≤ l then d(F (X)) ≤ k + l. Theorem 5 (Parimala–Sridharan [78], Monnier [67]). Let X be a smooth integral curve over a field F of characteristic 6= 2. Then the unramified Mil- nor question for quadratic forms (Question 3.1(1)) has a positive answer over X in the following cases: (1) cd 2(F ) ≤ 1, (2) vcd (F ) ≤ 1, (3) cd 2(F ) = 2 and X is affine, (4) vcd (F ) = 2 and X is affine. Proof. For (1), this follows from Parimala–Sridharan [78, Lemma 4.1] and the fact that e1 is always surjective. For (2), the case vcd (F ) = 0 (i.e. F is real closed) is contained in Monnier [67, Cor. 3.2] and the case vcd (F ) = 1 follows from a straightforward generalization to real closed fields of the results in [78, §5] for the real numbers. For (3), see [78, Lemma 4.2]. For (4), the statement follows from a generalization of [78, Thm. 6.1]. Q.E.D. We wonder whether vcd can be replaced by vcd 2 in Theorem 5. Parimala– Sridharan [78, Rem. 4] ask whether there exist affine curves (over a well- chosen field) over which the unramified Milnor question has a negative an- swer. For surfaces, there are positive results are in the case of vcd (F ) = 0. If F is algebraically closed, then the unramified Milnor question for quadratic 16 Asher Auel forms (Question 3.1(1)) has a positive answer by a direct computation, see Fern´andez-Carmena [32]. If F is real closed, one has the following result. Theorem 6 (Monnier [67, Thm. 4.5]). Let X be smooth integral surface over a real closed field F . If the number of connected components of X(F ) is ≤ 1 (i.e. in particular if X(F ) = ∅), then the unramified Milnor question for quadratic forms (Question 3.1(1)) has a positive answer over X. Examples of surfaces with many connected components over a real closed field, and over which the unramified Milnor question still has a pos- itive answer, are also given in Monnier [67]. Finally, as a consequence of [8, Cor. 3.4], the unramified Milnor ques- tion for quadratic forms (Question 3.1(1)) has a positive answer over any scheme X satisfying: 2Br(X) is generated by quaternion Azumaya algebras (i.e. indexperiod for 2-torsion classes); or 2Br(X) is generated by Azumaya algebras of degree dividing 4 (i.e. indexperiod2 for 2-torsion classes) and Pic(X) is 2-divisible. In particular, this recovers the known cases of curves over finite fields (by class field theory) and surfaces over algebraically closed fields (by de Jong [25]). §4. Negative results Alex Hahn asked if there exists a ring R over which the global Merkurjev question (Question 3.2) has a negative answer, i.e. e2 : I 2(R) → 2Br(R) is not surjective. The results of Parimala, Scharlau, and Sridharan [77], [78], [79], show that there exist smooth complete curves X (over p-adic fields F ) over which the unramified Milnor question (Question 3.1(1)) in degree 2 (and hence, by purity, the global Merkurjev question) has a negative answer. Remark 4.1. The assertion (in Gille [41, §10.7] and Pardon [74, §5]) that the unramified Milnor question (Question 3.1(1)) has a positive answer over any smooth scheme (over a field of characteristic 6= 2) is incorrect. In these texts, the distinction between the groups I n (X) and I n/I n+1(X) is not made clear. ur(X)/I n+1 ur Definition 4.1 (Parimala–Sridharan [78]). A scheme X over a field F has the extension property for quadratic forms if there exists x0 ∈ X(F ) such that every regular quadratic form on X \ {x0} extends to X. Proposition 4.1 (Parimala–Sridharan [78, Lemma 4.3]). Let F be a field of characteristic 6= 2 and with cd 2F ≤ 2 and X a smooth integral F - curve. Then the unramified Milnor question for quadratic forms (Question 3.1(1)) has a positive answer for X if and only if X has the extension property. The extension property is guaranteed when a residue theorem holds for the Witt group. The reside theorem for X = P1 is due to Milnor [66, §5]. For nonrational curves, the choice of local uniformizers inherent in defining the residue maps is eliminated by considering quadratic forms with values in the canonical bundle ωX/F . Remarks on the Milnor conjecture over schemes 17 Definition 4.2. Let X be a scheme and L an invertible OX -module. An (L -valued) symmetric bilinear form on X is a triple (E , b, L ), where E is a locally free OX -module of finite rank and b : S2E → L is an OX -module morphism. Theorem 7 (Geyer–Harder–Knebusch–Scharlau [39]). Let X be a smooth proper integral curve over a field F of characteristic 6= 2. Then there is a canonical complex (which is exact at the first two terms) 0 / W (X,ωX/F ) / W (F (X),ωF (X)/F ) ∂ωX / Lx∈X(1) W (F (x),ωF (x)/F ) TrX/F / W (F ) and thus in particular W (X, ωX/F ) has a residue theorem. Now any choice of isomorphism ϕ : N ⊗2 ∼= ωX/F , induces a group iso- morphism W (X) → W (X, ωX/F ) via (E , q) 7→ (E ⊗N , ϕ◦(q⊗idN ), ωX/F ). Thus in particular, if ωX/F is a square in Pic(X), then X has the extension property. Conversely: Theorem 8 (Parimala–Sridharan [79, Thm. 3]). Let F be a local field of characteristic 6= 2 and X a smooth integral hyperelliptic F -curve of genus ≥ 2 with X(F ) 6= ∅. Then the unramified Milnor question for quadratic forms (Question 3.1(1)) holds over X if and only if ωX/k is a square. Example 4.1 (Parimala–Sridharan [79, Rem. 3]). Let X be the smooth proper hyperelliptic curve over Q3 with affine model y2 = (x2− 3)(x4 + x3 + x2 + x + 1). One can show using [77, Thm. 2.4] that ωX/F is not a square. The point (y, x) = (√31, 2) defines a Q3-rational point of X. Hence by Theorem 8, the unramified Milnor question has a negative answer over X. Note that possible counter examples which are surfaces could be ex- tracted from the following result. Theorem 9 (Monnier [67, Thm. 4.5]). Let X be a smooth integral surface over a real closed field F . Then the unramified Milnor question for quadratic forms (Question 3.1(1)) has a positive answer over X if and only if the cokernel of the mod 2 signature homomorphism is 4-torsion. §5. Line bundle-valued quadratic forms Let X be a smooth F -scheme. Let W (X, L ) be the Witt group of L -valued symmetric bilinear forms on X and Wtot(X) = LL W (X, L ) the total Witt group, where the sum is taken over a set of representatives of Pic(X)/2. While this group is only defined up to non-canonical isomor- phism depending on our choice of representatives, none of our cohomological invariants depend on such isomorphisms, see [8, §1.2]. Furthermore, we will not consider any ring structure on this group. Thus we will not need to descend into most of the important considerations of Balmer–Calm`es [14]. Let e0 be the usual rank modulo 2 map e0 tot : Wtot(X) → Z/2Z = H 0 ur(X, Z/2Z) / / / / 18 Asher Auel and I 1 [80]) defines a surjective map tot(X) = ⊕L I 1(X, L ) its kernel. Then the signed discriminant (see e1 tot : I 1 tot(X) → H 1 ´et(X, µ2) = H 1 ur(X, µ2). Finally, denote by I 2(X, L ) ⊂ I 1(X, L ) the subgroup generated by forms of trivial Arf invariant and I 2 tot(X) = ⊕L I 2(X, L ). Then there exists a total Clifford invariant for line bundle-valued quadratic forms e2 tot : I 2 tot(X) → 2Br(X) = H 2 ur(X, µ2) defined in [8, Prop. 1.4]. The surjectivity of the total Clifford invariant can be viewed as a version of the global Merkurjev question (Question 3.2) for line bundle-valued quadratic forms. Theorem 10 ([8]). Let X be a smooth proper integral curve over a local field F of characteristic 6= 2 or a smooth proper integral surface over a finite field F of characteristic 6= 2. Then the total Clifford invariant e2 tot : I 2 tot(X) → 2Br(X) is surjective. The surjectivity of the total Clifford invariant can also be reinterpreted as the statement that while not every class in I 2/I 3(X) = H 2 ur(X) is rep- resented by a quadratic form on X, every class is represented by a line bundle-valued quadratic form on X. References [ 1 ] J. Arason, Quadratic forms and Galois-cohomology, S´eminaire de Th´eorie des Nombres, 1972–1973 (Univ. Bordeaux I, Talence), Exp. No. 22, Lab. Th´eorie des Nombres, Centre Nat. Recherche Sci., Talence, 1973, p. 5. [ 2 ] J. Arason, Der Wittring projektiver Raume, Math. Ann. 253 (1980), 205–212. [ 3 ] J´on Kr. Arason, Cohomologische invarianten quadratischer Formen, J. Alge- bra 36 (1975), no. 3, 448–491. [ 4 ] , A proof of Merkurjev's theorem, Quadratic and Hermitian forms (Hamilton, Ont., 1983), CMS Conf. Proc., vol. 4, Amer. Math. Soc., Provi- dence, RI, 1984, pp. 121–130. [ 5 ] J´on Kr. Arason, Richard Elman, and Bill Jacob, The graded Witt ring and Ga- lois cohomology. I, Quadratic and Hermitian forms (Hamilton, Ont., 1983), CMS Conf. Proc., vol. 4, Amer. Math. Soc., Providence, RI, 1984, pp. 17–50. , On the Witt ring of elliptic curves, K-theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992), Proc. Sympos. Pure Math., vol. 58, Amer. Math. Soc., Providence, RI, 1995, pp. 1–25. [ 6 ] [ 7 ] , The Witt ring of an elliptic curve over a local field, Math. Ann. 306 (1996), no. 2, 391–418. Remarks on the Milnor conjecture over schemes 19 [ 8 ] Asher Auel, Surjectivity of the total Clifford invariant and Brauer dimension, arXiv:1108.5728v1, 2011. [ 9 ] Ricardo Baeza, Quadratic forms over semilocal rings, Lecture Notes in Math- ematics, Vol. 655, Springer-Verlag, Berlin, 1978. [ 10 ] Paul Balmer, Derived Witt groups of a scheme, J. Pure Appl. Algebra 141 (1999), 101–129. [ 11 ] [ 12 ] , Triangular Witt groups. I. The 12-term exact sequence, K-Theory 19 (2000), 311–363. , Triangular Witt groups. II. From usual to derived, Math. Z. 236 (2001), 351–382. [ 13 ] Paul Balmer, Witt groups, Handbook of K-theory. Vol. 1, 2, Springer, Berlin, 2005, pp. 539–576. [ 14 ] Paul Balmer and Baptiste Calm`es, Bases of total Witt groups and lax- similitude, preprint arXiv:1104.5051v1, April 2011. [ 15 ] Paul Balmer and Charles Walter, A Gersten-Witt spectral sequence for regular schemes, Ann. Sci. ´Ecole Norm. Sup. (4) 35 (2002), no. 1, 127–152. [ 16 ] H. Bass and J. Tate, The Milnor ring of a global field, Algebraic K-theory, II: "Classical" algebraic K-theory and connections with arithmetic (Proc. Conf., Seattle, Wash., Battelle Memorial Inst., 1972), Springer, Berlin, 1973, pp. 349–446. Lecture Notes in Math., Vol. 342. [ 17 ] Hyman Bass, Algebraic K-theory, W. A. Benjamin, Inc., New York- Amsterdam, 1968. [ 18 ] A. A. Beılinson, Height pairing between algebraic cycles, K-theory, arithmetic and geometry (Moscow, 1984–1986), Lecture Notes in Math., vol. 1289, Springer, Berlin, 1987, pp. 1–25. [ 19 ] Spencer Bloch and Arthur Ogus, Gersten's conjecture and the homology of schemes, Ann. Sci. ´Ecole Norm. Sup. (4) 7 (1974), 181–201 (1975). MR 0412191 (54 #318) [ 20 ] J.-L. Colliot-Th´el`ene, Birational invariants, purity and the Gersten conjecture, K-theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992), Proc. Sympos. Pure Math., vol. 58, Amer. Math. Soc., Providence, RI, 1995, pp. 1–64. [ 21 ] J.-L. Colliot-Th´el`ene and R. Parimala, Real components of algebraic varieties and ´etale cohomology, Invent. Math. 101 (1990), no. 1, 81–99. [ 22 ] J.-L. Colliot-Th´el`ene and J.-J. Sansuc, Fibr´es quadratiques et composantes connexes r´eelles, Math. Ann. 244 (1979), no. 2, 105–134. [ 23 ] Jean-Louis Colliot-Th´el`ene, Raymond T. Hoobler, and Bruno Kahn, The Bloch-Ogus-Gabber theorem, Algebraic K-theory (Toronto, ON, 1996), Fields Inst. Commun., vol. 16, Amer. Math. Soc., Providence, RI, 1997, pp. 31–94. [ 24 ] Jean-Louis Colliot-Th´el`ene and Jean-Jacques Sansuc, Cohomologie des groupes de type multiplicatif sur les sch´emas r´eguliers, C. R. Acad. Sci. Paris S´er. A-B 287 (1978), no. 6, A449–A452. [ 25 ] A. J. de Jong, The period-index problem for the Brauer group of an algebraic surface, Duke Math. J. 123 (2004), no. 1, 71–94. [ 26 ] A.J. de Jong, A result of Gabber, preprint, 2003. [ 27 ] Gerhard Dietel, Wittringe singularer reeller Kurven. I, II, Comm. Algebra 11 (1983), no. 21, 2393–2448, 2449–2494. 20 Asher Auel [ 28 ] Philippe Elbaz-Vincent and Stefan Muller-Stach, Milnor K-theory of rings, higher Chow groups and applications, Invent. Math. 148 (2002), no. 1, 177– 206. [ 29 ] H´el`ene Esnault, Bruno Kahn, and Eckart Viehweg, Coverings with odd ram- ification and Stiefel-Whitney classes, J. reine Angew. Math. 441 (1993), 145–188. [ 30 ] H´el`ene Esnault, Bruno Kahn, Marc Levine, and Eckart Viehweg, The Arason invariant and mod 2 algebraic cycles, J. Amer. Math. Soc. 11 (1998), no. 1, 73–118. [ 31 ] Jean Fasel, Groupes de Chow-Witt, M´em. Soc. Math. Fr. (N.S.) (2008), no. 113, viii+197. [ 32 ] Fernando Fern´andez-Carmena, On the injectivity of the map of the Witt group of a scheme into the Witt group of its function field, Math. Ann. 277 (1987), no. 3, 453–468. [ 33 ] Eric M. Friedlander, Motivic complexes of Suslin and Voevodsky, Ast´erisque (1997), no. 245, Exp. No. 833, 5, 355–378. [ 34 ] Eric M. Friedlander, Michael Rapoport, and Andrei Suslin, The mathematical work of the 2002 Fields medalists, Notices Amer. Math. Soc. 50 (2003), no. 2, 212–217. [ 35 ] A. Frohlich, On the K-theory of unimodular forms over rings of algebraic integers, Quart. J. Math. Oxford Ser. (2) 22 (1971), 401–423. [ 36 ] Ofer Gabber, Some theorems on Azumaya algebras, The Brauer group (Sem., Les Plans-sur-Bex, 1980), Lecture Notes in Math., vol. 844, Springer, Berlin, 1981, pp. 129–209. [ 37 ] Thomas Geisser, Motivic cohomology, K-theory and topological cyclic homol- ogy, Handbook of K-theory. Vol. 1, 2, Springer, Berlin, 2005, pp. 193–234. [ 38 ] Thomas Geisser and Marc Levine, The Bloch-Kato conjecture and a theorem of Suslin-Voevodsky, J. Reine Angew. Math. 530 (2001), 55–103. [ 39 ] W.-D. Geyer, G. Harder, M. Knebusch, and W. Scharlau, Ein Residuensatz fur symmetrische Bilinearformen, Invent. Math. 11 (1970), 319–328. [ 40 ] Stefan Gille, On Witt groups with support, Math. Ann. 322 (2002), no. 1, 103–137. [ 41 ] , A graded Gersten-Witt complex for schemes with a dualizing complex and the Chow group, J. Pure Appl. Algebra 208 (2007), no. 2, 391–419. [ 42 ] Alexander Grothendieck, Dix expos´es sur la cohomologie des sch´emas, ch. VIII Classes de Chern et repr´esentations lin´eaires des groupes discrets, pp. 215– 305, North-Holland Publishing Company, Amsterdam; Masson & Cie, ´Editeur, Paris, 1968. [ 43 ] Daniel Guin, Homologie du groupe lin´eaire et K-th´eorie de Milnor des an- neaux, J. Algebra 123 (1989), no. 1, 27–59. [ 44 ] Gunter Harder, Halbeinfache Gruppenschemata uber vollstandigen Kurven, In- vent. Math. 6 (1968), 107–149. [ 45 ] Raymond T. Hoobler, The Merkuriev-Suslin theorem for any semi-local ring, J. Pure Appl. Algebra 207 (2006), no. 3, 537–552. [ 46 ] Bill Jacob and Markus Rost, Degree four cohomological invariants for qua- dratic forms, Invent. Math. 96 (1989), no. 3, 551–570. [ 47 ] Bruno Kahn, La conjecture de Milnor (d'apr`es V. Voevodsky), Ast´erisque (1997), no. 245, Exp. No. 834, 5, 379–418, S´eminaire Bourbaki, Vol. 1996/97. Remarks on the Milnor conjecture over schemes 21 [ 48 ] Bruno Kahn and R. Sujatha, Motivic cohomology and unramified cohomology of quadrics, J. Eur. Math. Soc. (JEMS) 2 (2000), no. 2, 145–177. [ 49 ] Moritz Kerz, The Gersten conjecture for Milnor K-theory, Invent. Math. 175 (2009), no. 1, 1–33. [ 50 ] , Milnor K-theory of local rings with finite residue fields, J. Algebraic Geom. 19 (2010), no. 1, 173–191. MR 2551760 (2010j:19006) [ 51 ] Moritz Kerz and Stefan Muller-Stach, The Milnor-Chow homomorphism re- visited, K-Theory 38 (2007), no. 1, 49–58. [ 52 ] M. Knebusch, Symmetric bilinear forms over algebraic varieties, Conference on Quadratic Forms–1976 (Kingston, Ont.) (G. Orzech, ed.), Queen's Papers in Pure and Appl. Math., no. 46, Queen's Univ., 1977, pp. 103–283. [ 53 ] M. Knebusch and W. Scharlau, Quadratische Formen und quadratische Reziprozitatsgesetze uber algebraischen Zahlkorpern, Math. Z. 121 (1971), 346–368. [ 54 ] Manfred Knebusch, Grothendieck- und Wittringe von nichtausgearteten sym- metrischen Bilinearformen, S.-B. Heidelberger Akad. Wiss. Math.-Natur. Kl. 1969/70 (1969/1970), 93–157. [ 55 ] , On algebraic curves over real closed fields. II, Math. Z. 151 (1976), no. 2, 189–205. [ 56 ] M.-A. Knus, R. Parimala, and R. Sridharan, On compositions and triality, J. Reine Angew. Math. 457 (1994), 45–70. [ 57 ] Max-Albert Knus, Quadratic and hermitian forms over rings, Springer-Verlag, Berlin, 1991. [ 58 ] Max-Albert Knus and Manuel Ojanguren, The Clifford algebra of a metabolic space, Arch. Math. (Basel) 56 (1991), no. 5, 440–445. [ 59 ] T. Y. Lam, The algebraic theory of quadratic forms, Benjamin/Cummings Publishing Co. Inc. Advanced Book Program, Reading, Mass., 1980, Revised second printing, Mathematics Lecture Note Series. [ 60 ] S. Lichtenbaum, Values of zeta-functions at nonnegative integers, Number theory, Noordwijkerhout 1983 (Noordwijkerhout, 1983), Lecture Notes in Math., vol. 1068, Springer, Berlin, 1984, pp. 127–138. [ 61 ] Max Lieblich, Moduli of twisted sheaves, Duke Math. J. 138 (2007), no. 1, 23–118. [ 62 ] A. S. Merkur'ev, On the norm residue symbol of degree 2, Dokl. Akad. Nauk SSSR 261 (1981), no. 3, 542–547. [ 63 ] Alexander Merkurjev, On the norm residue homomorphism of degree two, Pro- ceedings of the St. Petersburg Mathematical Society. Vol. XII (Providence, RI), Amer. Math. Soc. Transl. Ser. 2, vol. 219, Amer. Math. Soc., 2006, pp. 103–124. [ 64 ] A.S. Merkurjev and A.A. Suslin, The norm residue homomorphism of degree 3, Math. USSR Izv. 36 (1991), 349–368. [ 65 ] J. S. Milne, ´Etale cohomology, Princeton Mathematical Series, no. 33, Prince- ton University Press, Princeton, N.J., 1980. [ 66 ] John Milnor, Algebraic K-theory and quadratic forms, Invent. Math. 9 (1969/1970), 318–344. [ 67 ] Jean-Philippe Monnier, Unramified cohomology and quadratic forms, Math. Z. 235 (2000), no. 3, 455–478. [ 68 ] F. Morel, Voevodsky's proof of Milnor's conjecture, Bull. Amer. Math. Soc. (N.S.) 35 (1998), no. 2, 123–143. 22 Asher Auel [ 69 ] Fabien Morel, Milnor's conjecture on quadratic forms and mod 2 motivic com- plexes, Rend. Sem. Mat. Univ. Padova 114 (2005), 63–101 (2006). [ 70 ] Yu. P. Nesterenko and A. A. Suslin, Homology of the general linear group over a local ring, and Milnor's K-theory, Izv. Akad. Nauk SSSR Ser. Mat. 53 (1989), no. 1, 121–146. [ 71 ] M. Ojanguren, R. Parimala, R. Sridharan, and V. Suresh, Witt groups of the punctured spectrum of a 3-dimensional regular local ring and a purity theorem, J. London Math. Soc. (2) 59 (1999), no. 2, 521–540. [ 72 ] Manuel Ojanguren and Ivan Panin, A purity theorem for the Witt group, Ann. Sci. ´Ecole Norm. Sup. (4) 32 (1999), no. 1, 71–86. [ 73 ] D. Orlov, A. Vishik, and V. Voevodsky, An exact sequence for KM ∗ /2 with applications to quadratic forms, Ann. of Math. (2) 165 (2007), no. 1, 1–13. for Archives, [ 74 ] William regular http://www.math.uiuc.edu/K-theory/0419/, May 2000. K-theory preprint, schemes, Pardon, The filtered Gersten–Witt resolution Preprint [ 75 ] R. Parimala, Witt groups of affine three-folds, Duke Math. J. 57 (1988), no. 3, 947–954. [ 76 ] , Witt groups of conics, elliptic, and hyperelliptic curves, J. Number Theory 28 (1988), no. 1, 69–93. MR 925609 (89a:14028) [ 77 ] R. Parimala and W. Scharlau, On the canonical class of a curve and the extension property for quadratic forms, Recent advances in real algebraic geometry and quadratic forms (Berkeley, CA, 1990/1991; San Francisco, CA, 1991), Contemp. Math., vol. 155, Amer. Math. Soc., Providence, RI, 1994, pp. 339–350. [ 78 ] R. Parimala and R. Sridharan, Graded Witt ring and unramified cohomology, K-Theory 6 (1992), no. 1, 29–44. [ 79 ] , Nonsurjectivity of the Clifford invariant map, Proc. Indian Acad. Sci. Math. Sci. 104 (1994), no. 1, 49–56, K. G. Ramanathan memorial issue. [ 80 ] R. Parimala and R. Sridharan, Reduced norms and pfaffians via Brauer-Severi schemes, Contemp. Math. 155 (1994), 351–363. [ 81 ] R. Parimala and V. Srinivas, Analogues of the Brauer group for algebras with involution, Duke Math. J. 66 (1992), no. 2, 207–237. [ 82 ] R. Parimala and R. Sujatha, Witt group of hyperelliptic curves, Comment. Math. Helv. 65 (1990), no. 4, 559–580. [ 83 ] Raman Parimala, Witt groups of curves over local fields, Comm. Algebra 17 (1989), no. 11, 2857–2863. [ 84 ] A. Pfister, On the Milnor conjectures: history, influence, applications, Jahres- ber. Deutsch. Math.-Verein. 102 (2000), no. 1, 15–41. [ 85 ] Albrecht Pfister, Quadratische Formen in beliebigen Korpern, Invent. Math. 1 (1966), 116–132. MR 0200270 (34 #169) [ 86 ] , Some remarks on the historical development of the algebraic theory of quadratic forms, Quadratic and Hermitian forms (Hamilton, Ont., 1983), CMS Conf. Proc., vol. 4, Amer. Math. Soc., Providence, RI, 1984, pp. 1–16. [ 87 ] H. G. Quebbemann, R. Scharlau, W. Scharlau, and M. Schulte, Quadratis- che Formen in additiven Kategorien, Bull. Soc. Math. France Suppl. Mem. (1976), no. 48, 93–101, Colloque sur les Formes Quadratiques (Montpellier, 1975). Remarks on the Milnor conjecture over schemes 23 [ 88 ] H.-G. Quebbemann, W. Scharlau, and M. Schulte, Quadratic and Hermitian forms in additive and abelian categories, J. Algebra 59 (1979), no. 2, 264– 289. [ 89 ] Markus Rost, On Hilbert Satz 90 for k3 for degree-two extensions, preprint, May 1986. [ 90 ] Markus Rost, Chow groups with coefficients, Doc. Math. 1 (1996), No. 16, 319–393 (electronic). [ 91 ] Winfried Scharlau, Quadratic and Hermitian forms, Grundlehren der Math- ematischen Wissenschaften [Fundamental Principles of Mathematical Sci- ences], vol. 270, Springer-Verlag, Berlin, 1985. [ 92 ] Jean-Pierre Serre, Cohomologie galoisienne, 5th ed., 2nd print., Lecture Notes in Math., vol. 5, Springer-Verlag, Berlin, 1997. [ 93 ] M. Shyevski, The fifth invariant of quadratic forms, Dokl. Akad. Nauk SSSR 308 (1989), no. 3, 542–545. [ 94 ] R. Sujatha, Witt groups of real projective surfaces, Math. Ann. 288 (1990), no. 1, 89–101. [ 95 ] Andrei Suslin and Vladimir Voevodsky, Bloch-Kato conjecture and motivic cohomology with finite coefficients, The arithmetic and geometry of algebraic cycles (Banff, AB, 1998), NATO Sci. Ser. C Math. Phys. Sci., vol. 548, Kluwer Acad. Publ., Dordrecht, 2000, pp. 117–189. [ 96 ] R. W. Thomason, Le principe de scindage et l'inexistence d'une K-theorie de Milnor globale, Topology 31 (1992), no. 3, 571–588. [ 97 ] Burt Totaro, Milnor K-theory is the simplest part of algebraic K-theory, K- Theory 6 (1992), no. 2, 177–189. [ 98 ] , Non-injectivity of the map from the Witt group of a variety to the Witt group of its function field, J. Inst. Math. Jussieu 2 (2003), no. 3, 483–493. [ 99 ] Vladimir Voevodsky, Motivic cohomology with Z/2-coefficients, Publ. Math. Inst. Hautes ´Etudes Sci. (2003), no. 98, 59–104. [100] Adrian R. Wadsworth, Merkurjev's elementary proof of Merkurjev's theorem, Applications of algebraic K-theory to algebraic geometry and number the- ory, Part I, II (Boulder, Colo., 1983), Contemp. Math., vol. 55, Amer. Math. Soc., Providence, RI, 1986, pp. 741–776. [101] Charles Walter, Grothendieck-Witt groups of triangulated categories, preprint, K-theory preprint archive, 2003. [102] E. Witt, Theorie der quadratischen Formen in beliebigen Korpern., J. reine angew. Math. 176 (1936), 31–44 (German). [103] Kirill Zainoulline, Witt groups of varieties and the purity problem, Qua- dratic forms, linear algebraic groups, and cohomology, Dev. Math., vol. 18, Springer, New York, 2010, pp. 173–185. Department of Mathematics & CS Emory University 400 Dowman Drive NE W401 Atlanta, GA 30322 E-mail address: [email protected]
1309.1035	1	1309	2013-09-04T14:14:26	Cartier Crystals	[ "math.AG", "math.AC" ]	Building on our previous work "Cartier modules: finiteness results" we start in this manuscript an in depth study of the derived category of Cartier modules and the cohomological operations which are defined on them. After localizing at the sub-category of locally nilpotent objects we show that for a morphism essentially of finite type $f$ the operations $Rf_$ and $f^!$ are defined for Cartier crystals. We show that, if $f$ is of finite type (but not necessarily proper) $Rf_$ preserves coherent cohomology (up to nilpotence) and that $f^!$ has bounded cohomological dimension. In a sequel we will explain how Grothendieck-Serre Duality relates our theory of Cartier Crystals to the theory of $\tau$-crystals as developed by Pink and the second author.	math.AG	math	Cartier Crystals Manuel Blickle∗ Gebhard Bockle† July 31, 2018 Contents 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Summary of main results 1.2 Notation and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Cartier sheaves and Cartier crystals 2.1 Nilpotence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ´Etale pullback and localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 2.3 Cartier crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Cohomological operations for Cartier sheaves 3.1 Comparison of derived categories . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Pushforward . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Twisted inverse image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Support, Kashiwara equivalence, and applications 4.1 An exact triangle and Kashiwara equivalence . . . . . . . . . . . . . . . . . . . . 4.2 Finite cohomological dimension of the twisted inverse image . . . . . . . . . . . . 4.3 Stalks of Cartier crystals and the crystalline support . . . . . . . . . . . . . . . . References 1 Introduction 1 3 6 7 7 10 12 14 17 18 20 28 39 39 41 44 51 For a Noetherian scheme X over the finite field Fp and finite Fp-algebra Λ, the second author and Pink develop in [BP09] a cohomological theory of crystals over function fields and apply their theory to give a completely algebraic proof of Goss's rationality conjecture on L-functions ∗Johannes Gutenberg-Universitat Mainz (JGU), Institut fur Mathematik, Staudingerweg 9, D-55128 Mainz Ger- many, [email protected] †Universitat Heidelberg, IWR, Im Neuenheimer Feld 368, D-69120 Heidelberg, Germany, [email protected] 1 1 Introduction of families of A-motives (after Anderson). In fact [BP09] accomplishes much more in that it introduces an abelian category of τ -crystals, containing families of A-motives as a special case, and develops a cohomology theory for them. Roughly speaking a τ -sheaf on X (over Λ) is a coherent OX×Λ-module F together with a structural map τ : F −→ (σ×id)∗F where σ : X −→ X denotes the absolute q-power Frobenius on X. The τ -sheaves on X form an abelian category, from which the category of τ -crystals is obtained by localizing at the Serre subcategory of nilpotent τ -sheaves. Another crucial result in [BP09] is that, if Λ is finite, then the category of τ -crystals is equivalent to the category of constructible sheaves of Λ-modules on the ´etale site of X. This may be viewed as a vast generalization of Artin-Schreier theory and has been anticipated by Deligne in [Del77] A similar correspondence was developed in [EK04] by Emerton and Kisin. For a smooth scheme of finite type over a field of positive characteristic they consider quasi-coherent OX -modules M with a unit Frobenius semi-linear action. The unit condition means that the linearization σ∗M −→ M of σM : M −→ σ∗M , given by sending r ⊗ m to r · σM (m), is an isomorphism. This implies that σM is injective, hence M has no σ -- nilpotent elements. In addition the unit condition implies that M is closed under certain nilpotent extensions. Together, the unit condition can be regarded as a rigidification of the theory of quasi-coherent OX[σ] -- modules. On the level of the derived categories [EK04] prove a contravariant correspondence between locally finitely generated unit OX -modules and constructible ´etale Fp -- sheaves. As above they also show their correspondence with a finite coefficient ring Λ. The theories in [BP09] and [EK04] may hence be viewed as different approaches to compute ´etale cohomology of Fp-sheaves in positive characteristic. Each approach is tailored to certain applications and has its own benefits. The present paper grew out of the desire to establish a direct link between these theories for arbitrary essentially finite type coefficients Λ. The key to such a correspondence ought to be furnished by Grothendieck-Serre duality. It became quickly clear that an important intermediate category would have to be the category dual to the category of τ -crystals. The following diagram summarizes the situation {τ -- crystals} ≃ G-S-duality [BP09] ∼= {constr. Λ -- mod on Xet} ≃ [EK04] {κ -- crystals} o ⊗ωX ∼= {γ -- crystals} ∼= / {f.g. unit OX,σ -- mod} What we accomplish in this article (and its sequel [BB11b]) is a complete description of the left part of this diagram. That is, for any Noetherian σ-finite scheme X we introduce the category of κ-crystals which under Grothendieck-Serre duality corresponds to τ -crystals of [BP09]. We also define cohomological operations for κ-crystals and show in [BB11b] that these correspond, via Grothendieck-Serre duality, to the ones defined in [BP09]. The bottom row in the above diagram only holds if X is regular and was treated in a basic form already in [BB11c] and the intermediate category of γ-sheaves was thoroughly studied in [Bli08]; the full equivalence in the bottom row, including the compatibility with the cohomological operations on either side, will be treated in another sequel [BB11a] to the present paper. The maps to the top right corner are results in [BP09] and [EK04], respectively, and only yield an equivalence for finite coefficient rings Λ. Note that the horizontal equivalences are covariant equivalences on the underlying abelian categories, whereas the vertical ones are contravariant and require the passage to the derived category. Furthermore, the theory in the top left corner applies 2 / / O O o / O O 1 Introduction to any Noetherian scheme X, whereas the theory in the bottom right corner is only applicable to smooth schemes of finite type over a fixed base field. During the work on this project it became clear that the category of κ-crystals has very inter- esting properties and applications: implicitly in the work of Gabber [Gab04] and and explicitly in Anderson [And00] κ-sheaves are a central concept. The trace formula in the monograph [BP09] makes crucial use of [And00] and hence in an indirect way relies on κ-sheaves as well. This lead us to establish in [BB11c] the following important structural result for Cartier crystals in the case that Λ = Fq, cf. [BB11a, Sec 4] Theorem. Let X be a locally Noetherian scheme over Fq such that the Frobenius σ on X is a finite map. Then within the category of Cartier crystals the following holds. (a) Coherent Cartier crystals have finite length. (b) The Hom sets of coherent Cartier crystals are finite dimensional Fq-vetorspaces. By (a) and (b) coherent Cartier crystals have only finitely many sub-crystals. In [BB11c] we derive a number of applications of the above theorem to local cohomology and to Lyubeznik's theory of F -finite modules [Lyu97]. The latter is a local version of Emerton-Kisin's theory of finitely generated unit OX [σ]-modules. Other applications are visible in the work of the second author in [Bli13] and of Schwede in [Sch11]. Both works show that the theory of Cartier crystals is related to the theory of test ideals [HT04, BMS08]; the latter plays a crucial role in the birational geometry of algebraic varieties in positive characteristic. We expect that the formalism, and in particular the cohomological operations, which are developed in the present paper will be useful for the study of test ideals. For example, in some cases, one can realize the test ideal within the category of Cartier crystals as an intermediate extension functor. This functorial viewpoint on test ideals may help to give a conceptual explanation for the transformation rules of test ideals which were obtained recently [ST13, BST11]. Finally one may speculate, that Cartier Crystals play for test ideals a similar role as D-modules play for multiplier ideals in characteristic zero. In characteristic zero, jumping numbers of multiplier ideals have an interpretation as the jumps in the V filtration of a certain D-modules [BS05]. Perhaps in positive characteristic jumping numbers of test ideals have an analogous interpretation in terms of Cartier crystals, which could be related to constructible p-torsion sheaves via our duality [BB11b] and the equivalence in [BP09]. 1.1 Summary of main results Throughout we assume that our schemes are Noetherian and separated over Fq, the field of p elements. Except when explicitly stated otherwise, we assume throughout the paper that the q-power Frobenius σ is a finite map. Let Λ be an essentially finite type Fq-algebra which will serve for us as a coefficient ring. We denote C = Spec Λ. The purpose of this paper is to introduce various categories of κ-crystals and to develop a theory of cohomological operations for them. As indicated above, we define (quasi-) coherent κ-sheaves on X over Λ as (quasi-) coherent OX×C -modules M equipped with a OX×C-linear map κM : (σ × id)∗M −→ M 3 1 Introduction and denote the category of such by Cohκ(X, Λ) (respectively, QCohκ(X, Λ)). The coherent κ- sheaves form an abelian category which contains the Serre subcategory of nilpotent κ-sheaves Nilκ(X, Λ), consisting of those M such that some power of κM is zero. The category of κ- crystals Crysκ(X, Λ) is then the localization of Cohκ(X, Λ) at its subcategory Nilκ(X, Λ). In order to define functors it is necessary to work in the large ambient category of κ quasi-crystals which is defined analogously as the localization of QCohκ(X, Λ) at the locally nilpotent (i.e. inductive limits of nilpotent) κ-sheaves LNilκ(X, Λ). One of our primary goals is to develop functorial operations on Crysκ(X, Λ). These categories, and some more, are described in detail in Section 2. Duality for a finite map applied to the Frobenius endomorphism σ says, that to give a κ- structure on a quasi-coherent OX-module M , i.e. a morphism κ : (σ ×id)∗M −→ M , is equivalent to giving a morphismeκ : M −→ (σ × id)!M , where (σ × id)!M = (σ × id)−1 RH om X×C((σ × id)∗OX×Λ, M ) is viewed as an object in D+ coh(QCoh(X × Λ)). These observations suggest that the functors Rf∗ and f ! for an essentially finite type morphism f : Y −→ X known from Grothendieck-Serre duality, induce functors on an appropriate derived category of κ-crystals. The theory developed in Section 3 culminates in the following results. For the push-forward functor on κ-sheaves in Theorem 3.2.18 we shall prove: Theorem. If f : Y −→ X is an arbitrary morphism, then the usual push-forward functor R(f × id)∗ on quasi-coherent sheaves induces an exact Λ-linear functor Rf∗ : D∗(QCrysκ(Y, Λ)) −→ D∗(QCrysκ(X, Λ)) for ∗ ∈ {b, +, −, ∅}. Furthermore if f is of finite type then this restricts to a functor Rf∗ : D∗ crys(QCrysκ(Y, Λ)) −→ D∗ crys(QCrysκ(X, Λ)) . Here D∗ , Λ)) denotes the derived category of κ quasi-crystals whose cohomology is locally nil coherent (i.e. extension of locally nilpotent by coherent) with boundedness condition ∗. crys(QCrysκ( The crucial point to note in this result is that we do not assume that f is proper in the second statement. This is in stark contrast to the classical situation where f∗ generally does not preserve coherence. What we show, going back to a technique of Anderson in [And00], is that if f is of finite type, then Rf∗ preserves coherence "up to local nilpotence", i.e. that Rf∗ of a coherent κ-sheaf is an extension of a locally nilpotent by a coherent κ-sheaf. The main results for the pull-back for κ-sheaves are Theorems 3.3.22 and 4.2.1. Theorem. If f : Y −→ X is essentially of finite type, then the twisted inverse image functor (f × id)! on quasi-coherent sheaves induces an exact Λ-linear functor f ! : D+ crys(QCrysκ(X, Λ)) −→ D+ crys(QCrysκ(Y, Λ)) . of bounded cohomological dimension. In particular, f ! restricts to a functor on Db crys(QCrysκ( , Λ)). 4 1 Introduction At first glance, the boundedness of the cohomological dimension may seem unexpected. It is motivated by the duality between τ - and κ-sheaves which is treated in [BB11b], which is of finite cohomological dimension, and the fact that the corresponding functor f ∗ is exact on coherent τ -crystals. The proof of the above result is based on properties of the functor f ! from Grothendieck-Serre- Duality in its extension by Lipman [Lip09] and Nayak [Nay09]. One difficulty lies in showing that the functor f ! defined on κ-sheaves preserves nilpotence and hence induces a functor on crystals; cf. Proposition 3.3.21. We give two proofs for this crucial statement. One uses Nayak's extension of Nagata compactification to essentially finite type morphisms [Nay09], another one uses a notion of stalks for κ-crystals that we develop in Section 4.3. Finally, the boundedness of the cohomological dimension is based on the devissage theorem stated next that allows one to decompose X and Y into a disjoint union of locally closed regular subschemes. As explained above, the two functors Rf∗ and f ! we define for κ-crystals are directly induced from the corresponding functors on the underlying quasi-coherent sheaves. This occupies most of Section 2 and Section 3. This approach has the advantage that many of the compatibilities among these functors follow, often rather directly, from a corresponding well-established statement for the functors on the underlying categories of quasi-coherent sheaves. Some more work is necessary to show the following adjointness result from Corollary 3.3.24. Proposition. (a) Let f : Y −→ X be a proper morphism. Then as functors on categories D+ crys(QCrysκ( , Λ)) the functor Rf∗ is naturally left adjoint to f !. (b) If j : Y −→ X is an open immersion, then j! = j∗ is naturally left adjoint to Rj∗. For an open immersion j : U ֒−→ X and a closed complement i : Z ֒−→ X the above adjunction yields natural morphisms i∗i! −→ id and id −→ j∗j∗. In Theorem 4.1.1 we prove the following technically important result regarding their combination: Theorem. In D+ ? (QCrysκ(X, Λ)), for ? ∈ {∅, crys}, there is a natural exact triangle i∗i! −→ id −→ Rj∗j∗ −→ i∗i![1] This can be used to derive a Kashiwara-type equivalence, roughly saying that for a closed imbedding i : Y ֒−→ X, the functors i! and i∗ are inverse equivalences between the derived category of Cartier crystals on Y and the derived category of Cartier crystals on X which are supported on Y ; the precise statement can be found in Theorem 4.1.2. On the level of the underlying abelian categories and with Λ = Fp, this has been shown already in [BB11c]. The main application we give of the above triangle is its role in the proof of the boundedness for f !, where it allows a devissage argument to reduce to the case of maps between smooth schemes. We conclude the paper by studying a notion of stalks and their relation to the notion of crystalline support. For a point ix : x ֒−→ X of X and a Cartier sheaf V on X we define the stalk of V at X to be i! xV . A key point is that these stalks can be used to check nilpotence; in Corollary 4.3.3 we show that V is locally nilpotent if and only if i! xV is cohomologically nilpotent for all (not necessarily closed, if X is not of finite type over a field) points x ∈ X. This pointwise criterion for local nilpotence can in turn be used to give an alternative proof of the fact that f ! preserves cohomological nilpotence by reducing the statement to the case of a map between fields, where it can checked by hand. We then show that the closure of the set of all x ∈ X such that i! xV is not cohomologically nilpotent is in fact a suitable notion of support for the crystal 5 1 Introduction associated to V . It is the smallest closed set Z such that V X−Z is locally nilpotent, and hence zero as a crystal. 1.2 Notation and conventions For the remainder of this article, we fix the following notation: p is a prime number and q is a power pe of p. All schemes are assumed to be Noetherian and separated over the prime field Fq. All morphisms and fiber products, and all tensor products of modules and algebras, are taken over Fq, unless otherwise specified. By X, Y , etc., we denote Noetherian schemes over the field Fq. The q-power Frobenius mor- phism on X, which acts on functions by f 7→ f q, is denoted σX : X −→ X. Mostly we simply write σ for σX. Except when explicitly said otherwise, we assume throughout that our schemes satisfy the condition that the Frobenius endomorphism σX is a finite map of schemes. This is a crucial assumption in our development our theory of Cartier crystals. Only in Section 2 we occasionally drop this assumption and point out if a result or construction holds more gener- ally. As pointed out by Gabber in [Gab04, Remark 13.6], the finiteness of the Frobenius on any Noetherian separated scheme X implies that X admits a dualizing complex; a proof is given in op.cit. in the case that X is affine. Both conditions are clearly satisfied if X is essentially of finite type over a σ-finite field k, which is our prime case of interest. We also fix an Fq-algebra Λ which is essentially of finite type over Fq. By C we denote the corresponding affine scheme Spec Λ. The role of Λ will be that of a coefficient ring. Examples of particular interest are that Λ is a regular finite type Fq-algebra or that Λ is a local Artin ring. The assumptions on C imply that X × C is again Noetherian for every Noetherian scheme X over k. This is useful in dealing with coherent sheaves on X × C. Remark 1.2.1. After the predecessor to this paper [BB11c] was written the authors became aware of a notational conflict with the notion of a Cartier module in the theory of commutative formal groups which was originally introduced by Manin [Man63], cf. [Zin84]. We want to stress that our notion is different from theirs and hope that this does not cause confusion for the readers. Essentially finite type morphisms. Let us recall some classes of morphisms that in the clas- sical literature on algebraic geometry have not been considered much, and where only recently foundational results have been obtained, see [Nay09]. These are useful in the formulation and application of Grothendieck-Serre duality. Let ϕ : A −→ B be a ring homomorphism. If B is a localization of A and ϕ is the canonical localization map then ϕ is called a localizing homomorphism. If ϕ factors as A −→ C −→ B with A −→ C of finite type and C −→ B localizing, then ϕ is called essentially of finite type. Definition 1.2.2. Let f : Y −→ X be a morphism of Noetherian schemes. (a) The map f is said to be essentially of finite type if every point y ∈ Y has an affine open neighborhood V = Spec A such that f −1V −→ V is covered by finitely many affine open Ui = Spec(Bi) for which the corresponding ring homomorphisms ϕi : A −→ Bi are essentially of finite type. 6 2 Cartier sheaves and Cartier crystals (b) If in (a) each ϕi is moreover a localizing homomorphism, then one calls f is a localizing morphism. (c) If f is a localizing morphism and set-theoretically injective (or equivalently separated and injective on generic points) it is called a localizing immersion. In many respects localizing immersions behave like open immersions; any finite type localizing immersion is an open immersion. Each, localizing morphism, localizing immersion and morphism essentially of finite type is preserved under composition and base change. An important example of a localizing immersion is the natural map Spec OY,y −→ Y . A natural extension of Nagata's compactification theorem of morphisms of finite type is the following factorization theorem of Nayak [Nay09, Thm. 4.1] for essentially finite type morphisms. Theorem 1.2.3. Let f : Y −→ X be a separated essentially finite type morphism of Noetherian −−→ X where k is a localizing immersion and p is proper. schemes. Then f factors as Y k−−→ Z p It follows from [Nay09, Lem. 3.2] that the set of such factorizations forms a filtering inverse system, i.e. any two such are dominated by a third. An essentially smooth (essentially ´etale) morphism of Noetherian schemes is a separated for- mally smooth (formally ´etale) morphism that is essentially of finite type. Localizing morphisms are essentially ´etale but the latter class of morphisms is larger. 1.3 Acknowledgements We thank Joe Lipman and Suresh Nayak for answering our questions about some of the more intricate compatibilities of Grothendieck-Serre duality. Their clarifications were crucial during the final stages of the preparation of this manuscript. We are also grateful to Axel Stabler for a careful reading of an earlier version of this manuscript pointing out numerous typos and inaccuracies. Both authors were supported by the DFG funded research grant SFB/TRR45 Periods, mod- uli, and the arithmetic of algebraic varieties. The first author was supported by his Heisenberg fellowship. 2 Cartier sheaves and Cartier crystals In this section we give the definition of Cartier sheaves and Cartier crystals and set up the basic theory. For a more gentle introduction in the case that the coefficient algebra Λ is just Fp, we refer to [BB11c]. Definition 2.0.1. A Cartier sheaf on X over Λ, or shorter, a κ-sheaf on X is a pair V = (V , κV ) consisting of a quasi-coherent sheaf V on X × C and an OX×C-linear homomorphism κV : (σ × id)∗V −→ V . A homomorphism of κ-sheaves V −→ W on X is a homomorphism of the underlying quasi- coherent sheaves ϕ: V −→ W for which the following diagram commutes: (σ × id)∗V κV (σ×id)∗ϕ V ϕ (σ × id)∗W κW / W . 7 / / / 2 Cartier sheaves and Cartier crystals The sheaf underlying a κ-sheaf W will always be denoted W , unless otherwise specified. To avoid cumbersome notation we will often abbreviate κ = κW if the underlying sheaf is clear from the context. Example 2.0.2. If k is a perfect field, then the pth root morphism provides a canonical Cartier structure on k. If k is a field such that (k : kp) is finite, then k and σ!k = σ−1 Homk(σ∗k, k) are abstractly isomorphic fields. Any choice of such isomorphism equips k with the structure of a Cartier module, but, unless k is perfect, there is no canonical such choice. Example 2.0.3. The prototypical example of a Cartier sheaf (and namesake thereof) is the classical Cartier operator on the dualizing sheaf ωX = Ωn X of a smooth n-dimensional scheme X of finite type over Fp. It is induced by the Cartier isomorphism (see, for example [EV92]) H i(σ∗Ω• X ) ∼=−−→ Ωi X, which yields after composition with the projection σ∗Ωn X−→→H n(σ∗Ω• X ) the map κ = κX : σ∗ωX −→ ωX. In local coordinates x1, . . . , xn it is given by xi1 1 · · · xin n dx1 ∧ . . . ∧ dxn 7→ x i1+1 p −1 1 in +1 p · · · x n −1 dx1 ∧ . . . ∧ dxn where a non-integral exponent renders the whole expression on the right as zero. One verifies that this construction does not depend on the choice of local coordinates. An explicit computation shows that the adjoint of this mapeκ : ωX −→ σ!ωX is an isomorphism. Example 2.0.4. More generally, if X smooth over a σ-finite field k, then -- after a Cartier structure on κk : σk∗k −→ k was picked -- by the same formula as above there is a Cartier structure on ωX given in local coordinates by having κk also act on the coefficients from k. In particular, if k is perfect, there is a canonical Cartier structure on ωX for any finite type smooth scheme X over k. Even more generally, if X is normal and of finite type over k, then the Cartier structure on the smooth locus i : U ֒−→ X on ωU induces a Cartier structure on the dualizing sheaf ωX := i∗ωU canonically. Also in this case, the adjoint ωX −→ σ!ωX is an isomorphism. Definition 2.0.5. (a) A κ-sheaf V is called coherent if its underlying sheaf is coherent. (b) A κ-sheaf V is called ind-coherent if it is the union (or equivalently the filtered direct limit) of coherent κ-sheaves. The category formed by all κ-sheaves over Λ on X and with the above homomorphisms is denoted QCohκ(X, Λ). The full subcategory of all coherent κ-sheaves is denoted Cohκ(X, Λ), that of all ind-coherent κ-sheaves IndCohκ(X, Λ). We begin with a simple, but useful characterization of ind-coherent κ-sheaves. Lemma 2.0.6. A κ-sheaf V is ind-coherent if and only if for every coherent OX×Λ-submodule U ⊆ V , the κ-subsheaf of V generated by U is coherent. 8 2 Cartier sheaves and Cartier crystals Proof. If V is ind-coherent, then it is the union of its coherent κ-sub-sheaves. Since U ⊆ V is coherent, it must hence be contained in some coherent κ-sheaf. It follows that the κ-subsheaf U :=Pi κiU generated by U must also be contained in a coherent κ-sheaf which implies that U itself is coherent since X is Noetherian. Conversely, as a quasi-coherent OX×Λ-module, V is the union of its coherent submodules. If the κ-subsheaf generated by each of these is also coherent, then, clearly V is the union of coherent κ-sheaves, hence V is ind-coherent. Remark 2.0.7. There exist ind-coherent κ-sheaves which are not coherent, e.g. any infinite direct sum of non-zero coherent κ-sheaves. More interestingly, there also exist quasi-coherent κ-sheaves which are not ind-coherent: Take k = Fp, R = k[x], X = Spec R, Λ = k, and let V be the κ-sheaf on X corresponding to the R-module M = ∪n∈Nk[xp−n ] with κ the inverse of σ which is an isomorphism on M . Then V contains no non-zero coherent κ-subsheaf. Clearly, the categories Cohκ(X, Λ), QCohκ(X, Λ), and IndCohκ(X, Λ) are abelian Λ-linear categories, and, since σ is affine and hence σ∗ is exact, constructions such as kernel, cokernel, etc. are the usual ones on the underlying quasi-coherent sheaves, with the respective structural map κ obtained by functoriality. In particular, the formation of kernel, cokernel, image and co-image is preserved under the inclusions Cohκ(X, Λ) ⊂ IndCohκ(X, Λ) ⊂ QCohκ(X, Λ). From this it follows immediately, that Cohκ(X, Λ) is a Serre subcategory of QCohκ(X, Λ), i.e. it is an abelian subcategory that is closed under extensions. This statement holds for IndCohκ(X, Λ) only under the additional assumption that the Frobenius σ on X is finite: Proposition 2.0.8. Cohκ(X, Λ) is a Serre sub-category of QCohκ(X, Λ). If the Frobenius σ is finite on X, then IndCohκ(X, Λ) is also a Serre sub-category of QCohκ(X, Λ). Proof. The only point that is not obvious is that the extension of ind-coherent κ-sheaves is again ind-coherent. For this, consider a short exact sequence 0 −→ V ′ −→ V π−−→ V ′′ −→ 0 of κ-sheaves, assuming that the ends are ind-coherent. It is enough to show that given a coherent κ-subsheaf W ′′ of V ′′, there is a coherent κ-subsheaf W of V that surjects onto W ′′. Let W be any coherent subsheaf of V that surjects onto W ′′. Note that π(κ(σ × id)∗W ) = κ((σ × id)∗(π(W ))) ⊆ κ((σ × id)∗W ′′) ⊆ W ′′. Hence, π induces a surjection π : (W + κ((σ × id)∗W )) −→ W ′′. Denoting by H the kernel of this map we have κ((σ ×id)∗W ) ⊆ W +H . Now, since σ, and hence (σ × id) is a finite map by assumption, (σ × id)∗W is coherent. This implies that H is a coherent subsheaf of V ′. Since the latter is assumed to be ind-coherent, H is contained in some coherent κ-subsheaf. After enlarging H -- which does not affect the inclusion κ((σ × id)∗W ) ⊆ W + H -- we may hence assume that H itself is a coherent κ-subsheaf of V ′. Then we have κ((σ × id)∗(W + H )) = κ((σ × id)∗W ) + κ((σ × id)∗(H )) ⊆ W + H + H ⊆ W + H showing that W := W + H is a coherent κ-subsheaf of V mapping epimorphically to W ′′. 9 2 Cartier sheaves and Cartier crystals An example for the necessity of the assumption that σ is finite in the above proposition is given below in Example 2.1.6. Below it will be crucial to know that QCohκ(X, Λ) as well as IndCohκ(X, Λ) are Grothendieck categories, i.e. they form an abelian category with exact filtered direct limits and a generator. Theorem 2.0.9. QCohκ(X, Λ) is a Grothendieck category. Its subcategory of ind-coherent ob- jects IndCohκ(X, Λ) is closed under filtered direct limits in QCohκ(X, Λ) and is itself a Grothen- dieck category. Proof. The proof is almost the same as that of [BP09, Thm. 3.2.7]. Only in showing that QCohκ(X, Λ) has a generator is slightly different (in the case that σ is not finite). For any cardinal λ we say that a κ-sheaf V is λ-coherent if its underlying sheaf V is locally generated by a set of cardinality ≤ λ. Let λ be an infinite cardinal such that σ∗OX is locally generated by a set of cardinality ≤ λ. Now the proof proceeds as that of [BP09, Thm. 3.2.7] replacing ω by λ throughout. 2.1 Nilpotence For any κ-sheaf V we define the iterates κn κV ◦ (σ × id)∗κn V . Thus they are OX×Λ-linear homomorphisms V of κV by setting inductively κ0 V := id and κn+1 V := κn V : (σn × id)∗V −→ V . ∗ V :=(cid:0)(σn ×id)∗V , (σn ×id)∗κV(cid:1) is a κ-sheaf in its own right, and κn V is a homomorphism ∗ V −→ V . In particular, its image is a κ-subsheaf of V . Note that if σ is finite then Each σn of κ-sheaves σn the κ-sheaves σn ∗ V are coherent, whenever V is so. Definition 2.1.1. all, n ≫ 0. (a) A κ-sheaf V is called nilpotent if κn V vanishes for some, or equivalently (b) A κ-sheaf V is called locally nilpotent if it is a union of nilpotent κ-subsheaves. (c) By LNilκ(X, Λ) we denote the full subcategory of QCohκ(X, Λ) formed by all locally nilpo- tent κ-sheaves, by Nilκ(X, Λ) we shall denote the intersection LNilκ(X, Λ) ∩ Cohκ(X, Λ). (d) By LNilCohκ(X, Λ) we denote the full subcategory of IndCohκ(X, Λ) of objects M which contain a coherent C ⊆ M such that the quotient M /C lies in LNilκ(X, Λ). We will see shortly that LNilCohκ(X, Λ) is the Serre subcategory of IndCohκ(X, Λ) generated by Cohκ(X, Λ) and LNilκ(X, Λ). It follows easily from the noetherianess of objects in Cohκ(X, Λ) that any object in Nilκ(X, Λ) is nilpotent. However, there are nilpotent κ-sheaves which are not coherent, and hence not in Nilκ(X, Λ); for example any non-coherent OX×C -sheaf with zero structural map κ. Remark 2.1.2. It is important to note that local nilpotence for κ-sheaves does not mean that all local sections are killed by some power of κ. This is too weak. As an example consider k = Fp, X = Spec R, R = k[x] and κ : R −→ R given by xi 7→ x((i+1)/p)−1 where we define x raised to a non-integral exponent as equal to zero. Then clearly, every single element of r is killed by some power of κ, however R does not have a κ-submodule which is nilpotent. 10 2 Cartier sheaves and Cartier crystals The correct condition for local nilpotence, say of a κ-sheaf M on an affine scheme X = Spec R, is that the R ⊗ Λ-module generated by every section m ∈ M is annihilated by some power of for some e ≥ 0 we have κe((R ⊗ Λ)m) = 0. This is made precise in the following two κ, i.e. lemmata. Lemma 2.1.3. For a κ-sheaf M we define M e as the largest κ-subsheaf on which κe is zero. We let M nil be the largest locally nilpotent κ-subsheaf of M . (a) M nil =Se∈N M e. (b) M e is the sheaf (c) M nil is the sheaf U 7→ {m ∈ M (U )κe(σ × id)e ∗(OX×C · m) = 0}. U 7→ {m ∈ M (U )∃e ∈ N : κe(σ × id)e ∗(OX×C · m) = 0}. Proof. Clearly, since κ is additive, the sum of two subsheaves on which κe acts as zero also has this property, so M e is well defined. Equally clearly, M nil is equal to the union of all nilpotent subsheaves which implies the equality M nil = Se∈N M e, and this shows (a). To prove (b), let us denote the sheaf defined in (b) by N e. Then κe(N e) = 0 and hence N e ⊆ M e. The reverse inclusion can be checked (affine) locally where it is clear by definition. A similar argument applies to (c). Corollary 2.1.4. A κ-sheaf M is locally nilpotent if and only if for all coherent OX×C-subsheaves M there is a e ≥ 0 such that κe((σ × id)e ∗(M )) = 0. Since every quasi-coherent OX×C-module is the union of its coherent submodules it follows that LNilκ(X, Λ) ⊆ IndCohκ(X, Λ). Moreover one can easily show that LNilκ(X, Λ) is closed under filtered direct limits. Proposition 2.1.5. All categories in the following diagram Nilκ(X, Λ) 1❞❞❞ -❩❩❩ LNilκ(X, Λ) Cohκ(X, Λ) -❬❬❬ 1❝❝❝ LNilCohκ(X, Λ) / IndCohκ(X, Λ), (2.1) are Serre sub-categories of IndCohκ(X, Λ), and Nilκ(X, Λ), Cohκ(X, Λ) are Serre sub-categories of QCohκ(X, Λ). If the Frobenius σ on X is finite, then all categories in diagram (2.1) are Serre sub-categories of QCohκ(X, Λ). Proof. In Proposition 2.0.8 above we already observed that Cohκ(X, Λ) is a Serre subcategory of QCoh(X, Λ), and that IndCohκ(X, Λ) is a Serre subcategory of QCohκ(X, Λ) under the assumption that σ is finite. The only non-trivial part in each claim is to show that the respective sub-category is closed under extensions. For this let 0 −→ V ′ −→ V −→ V ′′ −→ 0 be a short exact sequence of κ-sheaves. We first verify that Nilκ(X, Λ) is closed under extensions in QCoh(X, Λ). For this assume that the outer terms of the above sequence lie in Nilκ(X, Λ) and 11 { - # 1 { - / # 1 2 Cartier sheaves and Cartier crystals V the middle in QCoh(X, Λ). The coherence of the middle term V is clear, since Cohκ(X, Λ) is a Serre subcategory which contains Nilκ(X, Λ). Choose n, m ∈ N such that κm V ′′ = 0. Then V is nilpotent since κm+n V ′ = 0 and κn = 0. To check that LNilκ(X, Λ) is closed under extensions in IndCohκ(X, Λ) assume that the outer terms of the sequence above are in LNilκ(X, Λ) and the middle lies in IndCohκ(X, Λ). In partic- ular, V is a union of its coherent κ-sub-sheaves W . For any such there is a short exact sequence 0 −→ W ∩ V ′ −→ W −→ W /(W ∩ V ′) −→ 0. Its outer terms lie in Nilκ(X, Λ) = LNilκ(X, Λ) ∩ Cohκ(X, Λ). By the previous case W must be nilpotent, and so V lies in LNilκ(X, Λ). Finally, if the outer terms lie in LNilCohκ(X, Λ) we have coherent C ′ ⊆ V ′ and C ′′ ⊆ V ′′ such that the both quotients are in LNilκ(X, Λ). Since V ∈ IndCohκ(X, Λ) we can find a coherent κ-subsheaf C ⊆ V surjecting onto C ′′ and containing C ′. Possibly enlarging C ′ such that there is a short exact sequence C ′ ֒−→ C −→→C ′′ we see that the quotient V /C is an extension of the locally nilpotent V ′/C ′ and V ′′/C ′′. By the preceding case it follows that V /C ∈ LNilκ(X, Λ), hence V ∈ LNilCohκ(X, Λ). Example 2.1.6. If σ is not finite, then neither IndCohκ(X, Λ), LNilκ(X, Λ), nor LNilCohκ(x, Λ) is closed under extensions in QCohκ(X, Λ). For an example take k a field with [k : kp] countably infinite (e.g. k is a countable, purely transcendental extension of Fp), X = Spec k, and Λ = k. Let {ei}i∈N be a basis of k over kp and let V = ⊕i∈Nk · gi be a countable direct sum of copies of k. We define a κ-structure on V as follows: κ(eig1) = e1gi+1 and for j ≥ 2 we set κ(eigj) =(ei+1gj 0 if i < j if i ≥ j Then for j ≥ 2, the coherent κ-subsheaf kgj is nilpotent of order j, hence V ′ = ⊕j≥2kgj is a locally nilpotent ind-coherent κ-subsheaf of V . The quotient V /V ′ is coherent and has zero structural map (since κ(V ) ⊆ V ′). However V cannot be locally nilpotent as this would require that k · g1 is annihilated by a single power of κ. But κ(k · g1) = V ′ by construction, and V ′ is clearly not annihilated by a single power of κ, this is not possible. Similarly, V cannot be ind-coherent, since any κ-subsheaf that has nonzero intersection with kg1 must contain kg1, and hence contains κ(kg1) = V ′. Since V ′ is not coherent it follows that V cannot be ind-coherent. Hence V is an example for an extension of ind-coherent locally nilpotent κ-sheaves that is itself neither ind-coherent nor locally nilpotent. Clearly, V is also not in LNilCohκ(X, Λ) even though both V ′ and V ′′ are. 2.2 ´Etale pullback and localization For an essentially ´etale map of schemes f : Y −→ X the pullback f ∗ of quasi-coherent sheaves induces a pullback functor for κ-sheaves. With this at hand we will show that all of the categories introduced above localize well, so that we will be able to check membership in each of these locally on an affine cover. This will be convenient in several cases below. As we will explain later, for an arbitrary essentially of finite type map it is the extraordinary inverse image functor f ! from [Har66] that induces a functor for Cartier sheaves. Since for an essentially ´etale map f one has f ∗ ∼= f ! this will recover the more elementary construction that follows now. We start to recall a well known lemma. 12 2 Cartier sheaves and Cartier crystals Lemma 2.2.1. Let j : U −→ X be a an essentially ´etale1 map of schemes, then U σ U j j X σ / X is a cartesian square. Furthermore, the natural transformation of functors j∗σ∗ −→ σ∗j∗ is an isomorphism. Proof. We need to show that the natural map σU/X : U −→ U ×X,σ X is an isomorphism. Since it is a bijection on the level of underlying topological spaces, we may localize to check that the map is an isomorphism. Then one reduces to the cases that j is finite and ´etale (see for example [HH90, Page 60]), j is an open immersion or that j is a localization at some multiplicative set -- both cases are straightforward. For the second statement, we apply to the unit of adjunction id −→ j∗j∗ the functor σ∗ to get σ∗ −→ σ∗j∗j∗ ∼= j∗σ∗j∗. Adjunction for j∗ and j∗ again yields the natural transformation j∗σ∗ −→ σ∗j∗, and flat base change implies that this is an isomorphism. Of course we use the fact that essentially ´etale implies flat, since both, ´etale morphisms as well as localizations are flat. Definition 2.2.2. Let j : U −→ X be an essentially ´etale map and V a κ-sheaf on X × C. We define j∗V as the κ-sheaf whose underlying sheaf is (j × id)∗V with structural morphism given by the following composition (σ × id)∗(j × id)∗V ∼= (j × id)∗(σ × id)∗V (j×id)∗κ −−−−−−→ (j × id)∗V where the first isomorphism is the one from Lemma 2.2.1. Of particular interest to us are the case where U ⊆ X is an open immersion. This is covered by the above, hence V U ×C := f ∗V is naturally a κ-sheaf on U (with coefficients in Λ). Similarly, if S is an arbitrary sheaf of multiplicative sets on X, then S−1V is naturally a κ-sheaf. In this case, the structural map κ is locally given by κ(v/s) := κ(vsq−1)/s. If x ∈ X is a point, then we denote by V x the κ-sheaf on Spec OX,x (with coefficients in Λ) whose underlying OX,x ⊗Λ-module is (j × id)∗V where j : Spec OX,x −→ X is the natural map. It follows immediately from the functoriality and exactness that pullback along essentially ´etale maps preserves (local) nilpotence and other properties of a κ-sheaf: Lemma 2.2.3. Let j : Y −→ X be essentially ´etale. Then the exact functor j∗ preserves nilpo- tence, local nilpotence, coherence, local nil-coherence and ind-coherence of κ-sheaves. Proof. Nilpotence is preserved by functoriality. Coherence is clear. Since j∗ is exact and commutes with filtered direct limits the other assertions are immediate. Here is a converse of this observation that any of these properties may be checked on an open cover. 1It is conceivable, that this lemma is valid if j is only formally ´etale. As Brian Conrad pointed out to us, for maps of Noetherian schemes formally ´etale does imply flatness (this is not true if one drops Noetherian), so the flat base change still applies. 13 / / / 2 Cartier sheaves and Cartier crystals Lemma 2.2.4. Let V be a κ-sheaf on X (Noetherian) and {Ui}i∈I a (Zariski) open covering of X. Then V is nilpotent if and only if for every Ui the restriction V Ui is nilpotent, i.e. nilpotence can be checked on Zariski coverings. Furthermore, V is locally nilpotent if and only if for all x ∈ X the stalk V x is locally nilpotent, i.e. local nilpotence can be checked on stalks (and hence on Zariski open coverings). Proof. From Lemma 2.1.3 it follows that the formation of the (locally) nilpotent part V e for e ≥ 0 and V nil commute with the restriction to an open set and with taking stalks. Now, V is nilpotent of order e (resp. locally nilpotent) if and only if for some e ≥ 0 the inclusion V e ⊆ V is equality (resp. the inclusion V nil ⊆ V is equality). From this it immediately follows that V is nilpotent of order e (resp. locally nilpotent) if and only if for all x ∈ X the stalks V x are nilpotent of order e (resp. locally nilpotent). This shows all statements except the implication that if all V Ui are nilpotent (of some order ei) then so is V . But this follows since we always assume that X is Noetherian which allows us to resort to a finite cover and to take e as the maximum of the finitely many ei. Then it follows that V is nilpotent of order e. Next we point out that membership in the categories in Diagram (2.1) can also be checked locally. Proposition 2.2.5. (a) Membership in each of the categories of Diagram (2.1), Cohκ, Nilκ, LNilκ, LNilCohκ, or IndCohκ, can be checked on Zariski open covers. (b) For V ∈ Cohκ membership in Nilκ can be checked on stalks, i.e. after pullback to the spectrum of the local rings of the points of X. Proof. For (a) note that coherence is local for Zariski coverings, which shows the statement for Cohκ. For Nilκ and LNilκ the statement follows from the preceding Lemma 2.2.4. Using the characterization in Lemma 2.0.6 that M ∈ IndCohκ(X, Λ) if and only if the κ-subsheaf generated by any coherent OX×Λ-submodule of M is also coherent, assertion (a) follows for IndCohκ(X, Λ) as well, since coherence for OX×C-modules is checked locally. It remains to show (a) for LNilCohκ. For this let U1, . . . , Un be an open affine cover of X such that for each i the restriction V Ui is in LNilCohκ. That is we have for each i a short exact sequence 0 −→ C i −→ V Ui −→ N i −→ 0 where C i is a coherent and N i a nilpotent κ-sheaf on Ui. Extend each C i to a coherent subsheaf C ′ i of V . Since V is ind-coherent, by Lemma 2.0.6 the κ-subsheaf C of V which is generated by the sum of all Ci is coherent. By (a) for LNilκ it follows that V /C is locally nilpotent and hence V ∈ LNilCohκ. For (b) observe that for coherent κ-sheaves nilpotence is the same as local nilpotence. Hence the assertion follows from Lemma 2.2.4. 2.3 Cartier crystals Our aim is to study Cartier sheaves up to nilpotence. This is made precise by the process of localization of the category of coherent Cartier sheaves Cohκ(X, Λ) at its Serre subcategory of nilpotent Cartier sheaves Nilκ(X, Λ). We follow the setup for localization as recalled in [BP09, § 2.2]. The process is very similar to the passage from the category of (bounded) complexes 14 2 Cartier sheaves and Cartier crystals of an abelian category to the derived category by inverting quasi-isomorphisms, but simpler. Namely, the Serre sub-categories of a Grothendieck category are in one to one correspondence to the saturated multiplicative systems. Starting with the Serre subcategory of nilpotent Cartier sheaves, the corresponding saturated multiplicative system consists precisely of those maps ϕ whose kernel and cokernel are locally nilpotent. We define: Definition 2.3.1. A homomorphism of κ-sheaves is called a nil-isomorphism if both its kernel and cokernel are locally nilpotent. Since a coherent κ-sheaf is nilpotent if and only it is locally nilpotent, a map of coherent κ- sheaves is a nil-isomorphism if and only if its kernel and cokernel are nilpotent (and not merely locally nilpotent). The following simple characterization of nil-isomorphisms will be useful. We omit the straightforward proof. Proposition 2.3.2. Kernel and cokernel of a homomorphism ϕ: V −→ W of κ-sheaves are nilpotent if and only if there exist n ≥ 0 and a homomorphism of κ-sheaves α making the following diagram commute: σn ∗ V σn ∗ ϕ σn ∗ W κn α =③③③③③③③③ κn V ϕ / W . In particular, if such a diagram exists, then ϕ is a nil-isomorphism. If ϕ: V −→ W is a ho- momorphism of coherent κ-sheaves then the existence of the diagram is equivalent to ϕ being a nil-isomorphism. Corollary 2.3.3. For every κ-sheaf V and for every n ≥ 0, the homomorphism κn is a nil-isomorphism. V : σn ∗ V −→ V Proof. Apply Proposition 2.3.2 to ϕ = κn V and α = idσn ∗ V . By Proposition 2.1.5 the category Nilκ(X, Λ) is a Serre subcategory of Cohκ(X, Λ). By [BP09, Prop. 2.3.2] the corresponding saturated multiplicative system is the class of nil-isomorphisms. In the remainder of this section, we will study basic properties of the associated localized category. Our first concern is existence. Recall first that QCohκ(X, Λ) is a Grothendieck category by Theorem 2.0.9. Thus it is locally small, and the same follows for any full subcategory. From [BP09, Prop. 2.3.4] we deduce that localization of Cohκ(X, Λ) with respect to nil-isomorphisms yields a well defined abelian category. Definition 2.3.4. The localization of Cohκ(X, Λ) at the multiplicative system of nil-isomorphism is denoted by Crysκ(X, Λ). We refer to the objects of Crysτ (X, Λ) as Cartier-Λ-crystals on X, or just as Cartier crystals, or κ-crystals. In order to define certain functors on the category of κ-crystals it is necessary to give a slightly different presentation of Crysκ(X, Λ) within a bigger ambient category. Recall the natural inclu- sion of subcategories Cohκ(X, Λ) ⊆ LNilCohκ(X, Λ) ⊆ IndCohκ(X, Λ) ⊆ QCohκ(X, Λ) (2.2) 15 / / = / 2 Cartier sheaves and Cartier crystals By localization at the multiplicative system of nil-isomorphisms we obtain a sequence of inclusions Crysκ(X, Λ) ⊆ LNilCrysκ(X, Λ) ⊆ IndCrysκ(X, Λ) ⊆ QCrysκ(X, Λ). (2.3) Note that in order for the localization to exist we must know that the localizing subcategory LNilκ is a Serre-subcategory. Hence in the case of QCohκ we need to assume that the Frobenius on X is finite, cf. Proposition 2.1.5. One has the following immediate consequence. Proposition 2.3.5. The inclusions in (2.3) are fully faithful and Crysκ(X, Λ) ⊆ LNilCrysκ(X, Λ) is an equivalence of categories. Proof. The full faithfulness of Crysκ(X, Λ) ⊆ LNilCrysκ(X, Λ) follows from Nilκ(X, Λ) = LNilκ(X, Λ) ∩ Cohκ(X, Λ) which is true by definition. For the other of inclusions of (2.3) this assertion is obvious. By definition it follows that the essential image of Crysκ(X, Λ) in IndCrysκ(X, Λ) is equal to LNilCrysκ(X, Λ). Hence the second claim follows. However Crysκ(X, Λ) is not closed under isomorphism in IndCrysκ(X, Λ), hence is not a Serre subcategory. But its essential image LNilCrysκ(X, Λ) is a Serre subcategory of IndCrysκ(X, Λ). If σ is finite, both are Serre subcategories of QCrysκ(X,Λ). Remark 2.3.6. Recall that a localized category possesses the same class of objects as the original one, but with different homomorphism sets. To avoid confusion one should keep in mind that the categorical properties of an object depend solely on the homomorphisms. Thus in the interest of clarity we will always speak of the κ-crystal associated to a κ-sheaf, and of the κ-sheaf underlying a κ-crystal, although these objects are really 'the same'. This rule will help to clarify whether a property refers to the category Cohκ(X, Λ) or to Crysκ(X, Λ). Based on the equivalence of Serre subcategories and saturated multiplicative systems, e.g. [BP09, Prop. 2.3.2], we say that a κ-crystal is zero if and only if its underlying κ-sheaf is nilpotent. To distinguish morphisms in the quotient category Crysκ(X, Λ) from those in Cohκ(X, Λ) we often denote them by dotted arrows / . If a morphism in Crysκ(X, Λ) is the image of a morphism in Cohκ(X, Λ), by abuse of notation we also denote it by a solid arrow. We use double arrows =⇒ to denote nil-isomorphisms. From Theorem 2.0.9 and [BP09, Prop. 2.4.3, Prop. 2.4.9] we deduce: Theorem 2.3.7. QCrysκ(X, Λ) is a Grothendieck category, and IndCrysκ(X, Λ) is a full sub- category which is closed under filtered direct limits and hence itself a Grothendieck category. Crysκ(X, Λ) is a full subcategory of both. In particular, all three are locally small abelian cate- gories. As with nil-isomorphisms between coherent κ-sheaves, there is a standard way to represent homomorphisms of crystals. Proposition 2.3.8. Any homomorphism ϕ: V suitable n by a diagram / W in Crysκ(X, Λ) can be represented for κn ⇐= σn V ∗ V −→ W . 16 / / 3 Cohomological operations for Cartier sheaves Proof. Any morphism in the localized category Crysκ(X, Λ) can be represented by a diagram −→ W in Cohκ(X, Λ) where s is a nil-isomorphism. Applying Proposition 2.3.2 to s, V we obtain a commutative diagram s⇐= V ′ ψ s V e ❇❇❇❇❇❇❇❇ ❇❇❇❇❇❇❇❇ κn V ′ α σn ∗ V ψ / W =④④④④④④④④ ψ α where κn is again a nil-isomorphism by Corollary 2.3.3. Thus ϕ is also represented by the lower edge of this diagram, which proves the assertion. 3 Cohomological operations for Cartier sheaves For left OX [σ] -- modules (called τ -sheaves in the terminology of [BP09]) the authors of [BP09] construct the following cohomological functors: (derived) pullback f ∗ for any morphism f : Y −→ X of schemes, (derived) proper push-forward f! for morphisms of finite type and (derived) tensor product one expects that the dual operations, namely f !, f∗ and a twisted tensor product ⊠ are naturally defined for right OX[σ] -- modules, i.e. for Cartier sheaves. In essence this will turn out to be the case, except that for f ! we require f to be essentially of finite type. This seems a natural condition, since already in Grothendieck-Serre duality for coherent sheaves one has such a restriction on f !. In this section we will give the constructions of f !, f∗ on QCrysκ(. . .) and describe the main properties and compatibilities of these functors; see Section 3.2 and Section 3.3. A further functor ⊠, a kind of tensor product, is postponed past the main duality theorem which is treated in [BB11b]. As we set things up, it will be clear that the functors f ! and f∗ for Cartier sheaves agree with the corresponding functors on the underlying quasi-coherent sheaves. This is convenient as we can use the full machinery of [Har66] and [Lip09] which frees us from checking many compatibilities which are already verified for f ! and f∗ in the classical context. In this sense our approach is similar to the one of Emerton and Kisin in [EK04] who also use the close relation of their functors for unit OF,X-modules to the classical ones on the underlying OX-modules. The precise relationship between our and Emerton and Kisin's theory will be explained in detail in [BB11a]. An important step to transfer the construction of the derived functors f∗ and f ! for quasi- coherent sheaves to Cartier sheaves is a basic but fundamental comparison result in Section 3.1 be- tween certain derived categories, namely that D∗(QCohκ(X, Λ)) is equivalent to D∗(QCoh(X × C))κ -- here the latter denotes the category of pairs consisting of an object V • in the derived category of quasi-coherent sheaves together with a homomorphism σ∗V • −→ V • in the derived category of quasi-coherent sheaves. For D∗(QCoh(X × C))κ it is immediate that the desired functors f ! and f∗ are just the restrictions of the corresponding functors on the derived category of quasi-coherent sheaves D∗(QCoh(X × C)) on X × C. It is also beneficial to have a direct construction of functors as derived functors on derived categories of Cartier sheaves or Cartier crystals. For τ -sheaves this is the approach taken in [BP09]. In Section 3.2 we include a treatment along these lines for the functor Rf∗ based on the use of Cech resolutions, which shows that Rf∗ is the derived functor of f∗ on D∗(QCohκ(X, Λ)). 17 k s / ] O O = 3 Cohomological operations for Cartier sheaves This very much simplifies the proof that Rf∗ preserves nilpotence as one can use a spectral sequence computation. For f ! we do not have such a description and we have to work significantly more to obtain the same result. In the remainder of this article, both of the two viewpoints represented by the categories D∗(QCohκ(X, Λ)) and D∗(QCoh(X × C))κ serve their purpose: On the one hand, the category D∗(QCoh(X × C))κ is taken under Grothendieck-Serre duality to a category which can be compared with a derived category of τ -sheaves. This will be important in [BB11a], the second part of this work, where we shall treat the subtleties of such a duality for derived categories of quasi- crystals with bounded cohomology in crystals between the κ- and τ -side. On the other hand, the category D∗(QCohκ(X, Λ)) admits a direct passage to the derived category D∗(QCrysκ(X, Λ)) of quasi-crystals, allowing us to transfer the derived functors Rf∗ and f ! to quasi-crystals as well. In addition we follow the viewpoint of [BP09] and work out much of our theory for the derived subcategory D∗(IndCohκ(X, Λ)) of D∗(QCohκ(X, Λ)) built out of ind-objects. In this setting, we construct Rf∗ directly and in [BB11a] we shall also construct f ! for morphisms f : Y −→ X with X regular. There we shall also see that the ind-categories will be technically useful in the passage from bounded κ-complexes with coherent cohomology to derived categories of κ-coherent sheaves. 3.1 Comparison of derived categories In this subsection we prove the comparison result on derived categories which was alluded to above. It is crucial to the construction of our cohomological operations (which are described below) and also crucial to the duality between κ-sheaves and the τ -sheaves of [BP09] which is treated in [BB11b]. For ∗ ∈ {b, +, −, ∅} and ? ∈ {lnil, coh, lncoh, ind, ∅}, is a homomorphism in D∗(QCoh(X × C)), such that for all i ∈ Z the induced pair κ = κV • : (σ × id)∗V • −→ V • (cid:16)H i(V •), H i(κV •) : (σ × id)∗H i(V •) ∼= H i((σ × id)∗V •) H iκ−−−−→ H i(V •)(cid:17) lies in LNilκ(X, Λ) for ? = lnil, in Cohκ(X, Λ) for ? = coh, in LNilCohκ(X, Λ) for ? = lncoh, in IndCohκ(X, Λ) for ? = ind, or without further condition if ? = ∅. A morphism from (V •, κV ) to (W •, κW ) is a homomorphism ϕ : V • −→ W • ∈ D∗ ?(QCoh(X × C)), such that the following diagram commutes in D∗(QCoh(X × C)) (σ × id)∗V • κV • (σ×id)∗ϕ (σ × id)∗W • κW • V • ϕ / W •. There is an obvious canonical inclusion D∗ ?(QCohκ(X, Λ)) ֒−→ D∗ ?(QCoh(X × C))κ (3.1) (3.2) where the former denotes the derived category of complexes of κ -- sheaves with cohomological bounded-ness condition ∗ and whose cohomology lies in ?. The main result of the present sub- section is the following comparison theorem which is essential to the construction of functors. 18 / / / 3 Cohomological operations for Cartier sheaves Theorem 3.1.1. For ∗ ∈ {∅, +, −, b} and ? ∈ {coh, lncoh, ind, ∅}, the natural inclusion (3.2) is an equivalence of categories. Proof. In this proof we denote quasi-isomorphisms by a double arrow "⇐=". We begin by showing that the inclusion (3.2) is essentially surjective. Suppose that the diagram κV • : (σ × id)∗V • ϕ0−−−→ V • 1 ψ0⇐= V •. represents an element in D∗ this object lies in the image of D∗ construct a commutative diagram ?(QCoh(X × C))κ. We need to show that, up to quasi-isomorphisms, 0 : = V •, below we shall inductively ?(QCohκ(X, Λ)). Setting V • (σ × id)∗V • 0 3 (σ × id)∗V • 1 3 (σ × id)∗V • 2 3 . . . V • 0 ϕ0 3 V • 1 ϕ1 3 V • 2 ϕ2 3 V • 3 3 . . . in Db coh(QCoh(X × C)) of complexes such that the horizontal maps are quasi-isomorphisms. Since (σ × id)∗ is exact and commutes with direct limits this map of directed systems induces on the direct limit V • ∞ a (honest) map of complexes κ∞ : (σ × id)∗V • ∞ −→ V • ∞ (3.3) which makes the following diagram commutative showing that (V∞, κ∞) is quasi-isomorphic to (V •, κV •): (σ × id)∗V • 3 (σ × id)∗V • ∞ (3.4) κV • V • κ∞ 3 V • ∞ By the construction of V • ∞ being quasi-isomorphic to V •, it satisfies the same constraints on cohomology (∅, lnil, coh, lncoh or ind). If moreover V • is concentrated in degrees [a, b], then the cohomology of V • ∞ is concentrated in this interval. Thus by standard truncation functors, we may assume that V • ∞ itself is concentrated in [a, b], and hence satisfies the same boundedness conditions as V •. We now give the rather straightforward construction of the directed system announced above. Consider the following diagram where the left portion represents our given morphism κV •. (σ × id)∗V • 0 ϕ0 (σ×id)∗ψ0 3 (σ × id)∗V • 1 V • 0 ψ0 3 V • 1 ψ1 3❴❴❴❴❴❴❴❴❴❴ ❴❴❴❴❴❴❴❴❴❴ ϕ1 V • 2 Since σ is affine, the functor (σ × id)∗ is exact and hence the top map is still a quasi-isomorphism. Now the dashed arrows exist and form a commuting square by the axiom of localization changing a left fraction into a right fraction. Iterating this procedure we obtain in the limit (3.3). 19 + + + + + + + + + + ✤ ✤ ✤ + + 3 Cohomological operations for Cartier sheaves To show full faithfulness consider a morphism W • / V • in D∗ ?(QCoh(X × C))κ. By essential surjectivity (3.4) we may assume that we have a diagram of complexes (σ × id)∗V • (σ×id)∗α 3 (σ × id)∗U • (σ×id)∗ϕ (σ × id)∗W • κV • V • α κU • 3 U • ϕ κW • W •, where all vertical maps are maps of complexes and both squares commute in the derived category ==⇒ T • D∗ commute as maps of complexes to T •. ?(QCoh(X × C)), i.e. the squares composed with a suitable quasi-isomorphism U • ϕ (σ×id)∗ϕ ⇐====== (σ × id)∗U • κU •−−−−→ U • ϕ Now observe that (σ × id)∗T • ==⇒ T • together with T • defines an object of D∗ ?(QCoh(X ×C))κ, which by essential surjectivity is quasi-isomorphic to an object in the image of D∗ ?(QCohκ(X, Λ)). Replacing (U •, κU •) by this object, we may assume that the previous diagram commutes as a diagram of complexes. This proves that the given morphism V • ?(QCohκ(X, Λ)) under the natural inclusion (3.2). An entirely similar argument shows that the image of a morphism under (3.2) is zero in Db ?(QCohκ(X, Λ)). Hence the proof of full faithfulness is complete. coh(QCoh(X × C)κ if and only if it was zero in D∗ / W • arises from a morphism in D∗ In anticipation of the duality proven in [BB11b] we now record a result for τ -sheaves of [BP09] analogous to Theorem 3.1.1 for κ-sheaeves above: By [BP09, Ch. 3] a τ -sheaf on X over Λ is a pair F = (F , τ ) consisting of a quasi-coherent OX×C-module F together with an OX×C-linear map τ : (σ × id)∗F −→ F . As in Section 2 one introduces for such pairs categories of τ -sheaves Cohτ , Nilτ , LNilτ , LNilCohτ , IndCohτ , QCohτ and, by localization, of crystals Crysτ , LNilCrysτ , IndCrysτ , QCrysτ for any argument (X, Λ). In the same way as at the beginning of this section for κ-sheaves, one can form the derived categories D∗ ?(QCoh(X, Λ))τ , where the latter denotes pairs (F •, τ ) consisting of F • ∈ D∗ ?(QCoh(X, Λ))τ and a morphism in the derived category τ : F • −→ (σ × id)∗F •. The following analog of Theorem 3.1.1 holds in this context: ?(QCohτ (X, Λ)) and D∗ Theorem 3.1.2. For ∗ ∈ {∅, +, −, b} and ? ∈ {lnil, coh, lncoh, ind, ∅}, the natural inclusion D∗ ?(QCohτ (X, Λ)) ֒−→ D∗ ?(QCoh(X × C))τ is an equivalence of categories. Proof. The proof proceeds in exactly the same way as that for Theorem 3.1.1 with one change. The basic objects in that proof where morphisms (σ × id)∗V • −→ V • while the basic objects here will be morphisms F • −→ (σ × id)∗F •. 3.2 Pushforward Throughout this subsection let f : Y −→ X be a morphism of schemes. We first define a derived functor Rf∗ on Cartier sheaves. Then we explain how to transfer this functor to Cartier crystals. Both results are rather straightforward but serve as a model for other derived functors on Cartier crystals such as f ! for f a finite morphism. Next we present a more delicate result who's proof goes back to an idea of Anderson in [And00]. It says that if f is of finite type, then Rf∗ preserves (up to equivalence of categories) the derived category of Cartier crystals. 20 / + o o + o o / 3 Cohomological operations for Cartier sheaves Definition 3.2.1. For any W ∈ QCohκ(Y, Λ) we define f∗W as the κ-sheaf on X over Λ whose underlying sheaf is (f × id)∗W and whose structural morphism is the composite κf∗W : (σ × id)∗(f × id)∗V ≃−−→ (f × id)∗(σ × id)∗V (f ×id)∗κV −−−−−−−−→ (f × id)∗V . The construction of the derived functor of f∗ is facilitated by the fact that the usual Cech complex of a κ-sheaf associated to an open covering is naturally a complex of κ-sheaves. This is simply a consequence of the ´etale pullback for κ-sheaves constructed in Definition 2.2.2. Concretely, for any finite cover U of Y and any V ∈ QCohκ(Y, Λ) we denote by C • U(V ) the Cech complex of V , cf. [Har77, III.4] or [BP09, 5.4]. It is a resolution of V in D∗(QCohκ(Y, Λ)) via the natural co-augmentation homomorphism V −→ C • U(V ). The construction of the Cech resolution is functorial in V and thus extends to complexes V •. By CU(V •) we denote the total complex of the bicomplex thus obtained2. It is a resolution of V • in D∗(QCohκ(Y, Λ)). Following [Har66, Thm. I.5.1] or [BP09, Thm. 6.4.1], one deduces: Theorem 3.2.2. For any ∗ ∈ {b, +, −, ∅} the functor f∗ possesses a right derived functor Rf∗: D∗(QCohκ(Y, Λ)) −→ D∗(QCohκ(X, Λ)). If U is a finite affine cover of X, then Rf∗ can be defined as V • 7→ f∗(cid:0) CU(V •)(cid:1). If moreover f is proper then Rf∗ restricts to a functor Rf∗ : D∗ coh(QCohκ(Y, Λ)) −→ D∗ coh(QCohκ(X, Λ)) Remark 3.2.3. The fact that we may compute Rf∗ on κ-sheaves via a Cech resolution associated to a finite cover shows that we have for any complex V • ∈ D∗ coh(QCohκ(Y, Λ)) a convergent spectral sequence of κ-sheaves The usual functor R(f × id)∗ on complexes of quasi-coherent sheaves also defines a functor Ei,j 2 = Rif∗(H jV •) =⇒ Ri+jf∗V •. R(f × id)∗ : D∗(QCoh(Y × C))κ −→ D∗(QCoh(X × C))κ , (V •, κV •) 7−→(cid:0)R(f × id)∗V •, R(f × id)∗(cid:0)κV •(cid:1)(cid:1). which agrees with Rf∗ for κ-sheaves via the equivalence of categories in Theorem 3.1.1 This means, for a κ-sheaf V on Y we may compute Rif∗V as the pair consisting of the sheaf Ri(f × id)∗V together with the map Ri(f × id)∗(τV ). This is again a consequence of the observation that the same Cech resolution computes both, Rf∗ and R(f × id)∗: Theorem 3.2.4. There is a commutative diagram D∗(QCohκ(Y, Λ)) Rf∗ / D∗(QCohκ(X, Λ)) ≃ ≃ D∗(QCoh(Y × C))κ R(f ×id)∗ / D∗(QCoh(X × C))κ with the vertical equivalences from Theorem 3.1.1. 2Since we are working on noetherian schemes and restrict to finite covers for our Cech complexes this works even for unbounded complexes 21 / / 3 Cohomological operations for Cartier sheaves To prepare the passage of a functor from a derived category of Cartier sheaves to a derived category of Cartier crystals, we make the following definition. Definition 3.2.5. Complexes in D∗ lnil(QCoh(Y, Λ))κ are called cohomologically nilpotent. A morphism of complexes V • −→ W • in D∗(QCohκ(Y, Λ)) is called a nil-quasi-isomorphism if its cone is cohomologically nilpotent. The set of all such morphisms is denoted S −1 qnil. D∗ lnil(QCohκ(Y, Λ)) is a full triangulated subcategory of D∗(QCohκ(Y, Λ)). By the general theory of such, see [Ver96], the nil-quasi-isomorphisms form a saturated multiplicative system of D∗(QCohκ(Y, Λ)) and thus the localization S −1 qnil D∗(QCohκ(Y, Λ)) is defined. It is natural to compare this localized category with the derived category obtained from the localization of QCohκ at nil-isomorphisms. Clearly one has a homomorphism of derived categories D∗(QCohκ(Y, Λ)) −→ D∗(QCrysκ(Y, Λ)) which maps all nil-quasi-isomorphism to isomorphisms. Indeed one has the following formal result: Theorem 3.2.6. For ∗ ∈ {∅, +, −, b}, the natural morphism S −1 qnil D∗(QCohκ(Y, Λ)) −→ D∗(QCrysκ(Y, Λ)) is an equivalence of categories. The analogous result holds if (QCohκ, QCrysκ) is replaced by either of (IndCohκ, IndCrysκ) or (LNilCohκ, LNilCrysκ). Proof. For quasi- and ind-objects, the first two parts of the theorem follows from [BP09, Thm. 2.6.2] since by Theorem 2.3.7 the relevant categories are Grothendieck categories and the locally nilpo- tent ones a Serre subcategory closed under filtered direct limits. To show the equivalence S −1 qnil D∗(QCohκ(Y, Λ)) −→ D∗(LNilCrysκ(Y, Λ)) claimed in the third statement, one cannot appeal directly to [BP09, Theorem 2.6.2] since LNilCohκ(Y, Λ) is not closed under infinite direct sums and hence not a Grothendieck cate- gory. Instead, for ∗ ∈ {b, −, +} the result is a special case of [Miy91, Thm. 3.2]. It is a challenging exercise (left to the reader) to check that the argument of op.cit. can be extended to ∗ = ∅ Since localization at S −1 qnil preserves cohomology up to local nilpotence, we obtain the following consequence: Corollary 3.2.7. For ∗ ∈ {∅, +, −, b}, the isomorphism of Theorem 3.2.6 restricts to an equiv- alence of categories S −1 qnil D∗ lncoh(QCohκ(Y, Λ)) −→ D∗ crys(QCrysκ(Y, Λ)), where the index crys (resp lncoh) indicates that the cohomology is in LNilCrysκ (resp. LNilCohκ). The same result holds, if we either replace (lncoh, crys) by (indcoh, indcrys) or if we replace (QCohκ, QCrysκ) by (IndCohκ, IndCrysκ). 22 3 Cohomological operations for Cartier sheaves Proof. We sketch the proof of the first statement and leave the analogous remaining statements to the reader. First one observes that the inclusion S −1 qnil D∗ lncoh(QCohκ(Y, Λ)) ֒−→ S −1 qnil D∗(QCohκ(Y, Λ)) establishes the former as a full subcategory of the latter. For this note that D∗ is a full subcategory of D∗(QCohκ(Y, Λ)). Now let lncoh(QCohκ(Y, Λ)) ϕ : U • ⇐= W • −→ V • qnil D∗(QCohκ(Y, Λ) with U •, V • ∈ D∗ represent a morphism in S −1 lncoh(QCohκ(Y, Λ). We have to show that W • is also in D∗ lncoh(QCohκ(Y, Λ). But this follows easily since U • ⇐= W • is a nil-quasi-isomorphism, i.e. its cone lies in LNilκ. The long exact cohomology sequence of the cone-triangle shows that the cohomology of W • is an extension of LNilκ and LNilCohκ, hence also lies in LNilCohκ since the latter is a Serre subcategory and hence closed under extensions by Proposition 2.1.5. Now consider the diagram S −1 qnil D∗(QCohκ(Y, Λ)) ≃ / D∗(QCrysκ(Y, Λ)) S −1 qnil D∗ lncoh(QCohκ(Y, Λ)) ≃ /❴❴❴❴❴ D∗ crys(QCrysκ(Y, Λ)) where the upwards arrows are full subcategories and the top is the equivalence of Theorem 3.2.6. Since S −1 nil LNilCohκ = LNilCrysκ it is now just an unraveling of definitions to verify that under the equivalence of the top row, the bottom two full subcategories correspond to each other, we leave this to the reader. Remark 3.2.8. In [BB11b] we shall prove a deep refinement of Corollary 3.2.7 for ∗ ∈ {b, +} and cohomology in coh on the left hand side. The relevance of Theorem 3.2.6 and its corollary for the construction of functors on derived categories of crystals is explained by the following: Proposition 3.2.9. Suppose F : D∗(QCohκ(Y, Λ)) −→ D∗(QCohκ(X, Λ)) is an exact functor of triangulated categories, e.g. a derived functor. If F preserves cohomological nilpotence, i.e., lnil(QCohκ(X, Λ)), then F maps nil-quasi-isomorphism again to such and it induces an exact functor (denoted by the same name): if F(cid:0)D∗ lnil(QCohκ(Y, Λ))(cid:1) ⊂ D∗ F : D∗(QCrysκ(Y, Λ)) −→ D∗(QCrysκ(X, Λ)). An analogous statement also holds if (QCohκ, QCrysκ) is replaced by (IndCohκ, IndCrysκ) or (LNilCohκ, Crysκ). Moreover both of the above assertions hold with additional conditions on the cohomology: If the given functor preserves cohomology in LNilCohκ, or IndCohκ, respectively, then the resulting functor preserves cohomology in crys or indcrys. 23 / ? O O / ? O O 3 Cohomological operations for Cartier sheaves Proof. Consider the composition of the given functor F with the localization functor D∗(QCohκ(X, Λ)) −→ S −1 a unique functor qnil D∗(QCohκ(X, Λ)). By the hypothesis and the universal property of localization there exists qnil D∗(QCohκ(Y, Λ)) −→ S −1 qnil D∗(QCohκ(X, Λ)) eF : S −1 whose composite with the localization D∗(QCohκ(Y, Λ)) −→ S −1 qnil D∗(QCohκ(Y, Λ)) is equal to the above composition. The first assertion now follows by an application of Theorem 3.2.6 to domain and range of the functor. The same proof applies to (IndCohκ, IndCrysκ) and (LNilCohκ, Crysκ) instead of (QCohκ, QCrysκ). For the remaining assertions one uses Corollary 3.2.7 instead of Theorem 3.2.6. The following lemma is useful for verifying the conditions of the preceding Proposition 3.2.9. Lemma 3.2.10. Suppose F = Rf is the right derived functor of a left exact functor f : QCohκ(Y, Λ) −→ QCohκ(X, Λ). Let A (resp. A ′) be a Serre subcategory of QCohκ(Y, Λ) (resp. QCohκ(X, Λ)), and let V • be a complex in QCohκ(Y, Λ). Assume that (a) each Rif = H i(Rf ) maps A into the Serre subcategory A ′, and (b) there exists a convergent spectral sequence for complexes in D+(QCohκ(Y, Λ))3 Ei,j 2 = Rif (H jV •) =⇒ Ri+jf V •. Then F(cid:0)D+ A (QCohκ(Y, Λ))(cid:1) ⊂ D+ A ′(QCohκ(X, Λ)). the conclusion holds for any of b, −, ∅ in place of +. If moreover F is of finite cohomological dimension and (b) holds for unbounded complexes, then Proof. The convergence of the spectral sequence (together with the boundedness of the coho- mological dimension of F if V • is unbounded) implies that the limit term Rnf V • is a finite successive extension of κ-subquotients of the Rif (H jV •) for i + j = n. By assumption each of these is in A , hence the result follows since A ′ is a Serre subcategory Our first application of the above lemma is to Rf∗. Corollary 3.2.11. Let f : Y −→ X be any morphism of noetherian schemes with σX finite. Then lnil(QCohκ(Y, Λ))(cid:1) ⊂ D∗ Rf∗(cid:0)D∗ If moreover f is proper, then Rf∗(cid:0)D∗ lnil(QCohκ(X, Λ)). lncoh(QCohκ(Y, Λ))(cid:1) ⊂ D∗ Proof. As pointed out after Theorem 3.2.2 we have a convergent spectral sequence lncoh(QCohκ(X, Λ)). Ei,j 2 = Rif∗(H jV •) =⇒ Ri+jf∗V •. By Proposition 2.1.5 the subcategory LNilκ (resp. LNilCohκ) is a Serre subcategories of QCohκ. Hence, in order to conclude with the preceding Lemma 3.2.10 we only have to check that Rif∗ 3At this point, we have not yet established that the categories QCohκ(. . .) have enough injectives. So we cannot say anything general about existence or convergence. 24 3 Cohomological operations for Cartier sheaves preserves LNilκ (resp. LNilCohκ). For LNilκ, since Rif∗ commutes with direct limits, it is enough to show that Rif∗ preserves nilpotence. But this follows since Rif∗ is a functor: If (V , κ) κn is nilpotent, then (σ × id)n −−−→ V is zero for some n. This implies that Rif∗(κn) is also zero, and hence Rif∗V is nilpotent. ∗ V For LNilCohκ, we note that for proper f all Rif∗ also preserve coherence. Since, by definition a nil-coherent κ-sheaf is an extension of coherent and locally nilpotent, it follows that Rif∗ preserves nil-coherence. Combining Proposition 3.2.9 and Corollary 3.2.11 we deduce: Corollary 3.2.12. The functor Rf∗ on Cartier sheaves induces a functor Rf∗ : D∗(QCrysκ(Y, Λ)) −→ D∗(QCrysκ(X, Λ)). If f is proper, it restricts to a functor Rf∗ : D∗ crys(QCrysκ(Y, Λ)) −→ D∗ crys(QCrysκ(X, Λ)). Recall, the subscript crys indicates that the cohomology lies in LNilCrysκ. Remark 3.2.13. The functor Rf∗ on quasi-crystals is in fact a derived functor: the Cech resolution also exists for Cartier sheaves, and hence for Cartier quasi-crystals. It is clearly functorial and thus extends to complexes. Therefore the analog of Theorem 3.2.2 holds for Cartier quasi-crystals as well. The explicit construction shows that this derived functor agrees with the functor constructed in Corollary 3.2.12 via Theorem 3.2.6 and Corollary 3.2.11. For a general morphism f : Y −→ X push-forward f∗ does not preserve coherence. The following result shows that for Cartier sheaves and morphisms of finite type, morally the situation is better. Theorem 3.2.14. Let f : Y −→ X be a morphism of finite type and ∗ ∈ {b, −, +, ∅}. Then Rf∗(cid:0)D∗(LNilCohκ(Y, Λ))(cid:1) ⊂ D∗(LNilCohκ(X, Λ)). Proof. The proof is an adaption of an argument of Anderson from [And00]. The assertion is local, since Rf∗ may be defined via f∗ applied to Cech resolutions. Thus it suffices to prove the theorem for affine morphisms f : Spec S −→ Spec R of finite type given by ϕ : R −→ S in the following two cases: (a) f is a closed immersion, (b) ϕ is the canonical homomorphism R −→ R[X]. Since now f is affine, the functor Rf∗ has vanishing higher cohomology and thus in both cases we need to show that f∗(cid:0) LNilCohκ(Spec S, Λ)(cid:1) ⊂ LNilCohκ(Spec R, Λ). Case (a) is trivial since any closed immersion of finite type is proper. To show (b) we need to verify the following Claim. Let (M, κ) be a finitely generated R[X] ⊗ Λ-module and κ : M −→ M be (σR[X] ⊗ idΛ)−1- linear map giving M the structure of a Cartier module. Then M contains a finitely generated R⊗Λ-submodule N preserved under κ such that (M/N, κ) is a locally nilpotent R[σ]⊗Λ-module. 25 3 Cohomological operations for Cartier sheaves On R[X] we consider the filtration by the degree in X, i.e., we set R[X]≤d := {f ∈ R[X] degX f ≤ d}. Next we choose generators m1, . . . , mt of M as an R[X] ⊗ Λ-module. They yield a non-canonical exhaustive filtration M≤d := R[X]≤d ⊗ Λ · mi tXi=1 of M by finitely generated R ⊗ Λ-submodules. The key observation in [And00] is that the Cartier semi-linear operator κ on M has a strong contractivity property which can be described as follows: Choose, typically non-unique, elements ai,j,d ∈ R[X] such that κ(X dmi) =Xj ai,j,dmj for each d ∈ {0, . . . , q − 1} and i ∈ {1, . . . , t} and set C := max{degX ai,j,d i, j ∈ {1, . . . , t}, d ∈ {0, . . . , q − 1}}. Claim. For ℓ > Cq/(q − 1) one has κ(M≤ℓ) ⊂ M≤ℓ−1. If this claim is shown then the first claim above is immediate for N = M≤ℓ0 with ℓ0 = ⌊Cq/(q − 1)⌋, the largest integer less or equal to Cq/(q − 1). To prove the claim, fix ℓ ≥ Cq/(q − 1). Then for any d ≤ ℓ we have d q + C ≤ ℓ q + C < ℓ. Let d′ ∈ {0, . . . , q − 1} be the remainder of d by division by q. For i ∈ {1, . . . , t} it follows that κ(X dmi) = κ(X d−d′ X d′ mi) = X d−d′ q Xj ai,j,d′mj ∈ M≤ d−d′ q +C ⊂ M≤ℓ−1 finishing the proof of the claim and thereby the proof of the theorem. Corollary 3.2.15. Let f : Y −→ X be essentially of finite type and ∗ ∈ {b, −, +, ∅}. Then Rf∗(cid:0)D∗(IndCohκ(Y, Λ))(cid:1) ⊂ D∗(IndCohκ(X, Λ)). Proof. As in the proof of Theorem 3.2.14 the assertion is local, and so it suffices to prove it when X = Spec R and Y = Spec S are affine schemes. In that case we need to prove that f∗(cid:0) IndCohκ(Y, Λ)(cid:1) ⊂ IndCohκ(X, Λ). Since all ind-Cartier sheaves are direct limits of coherent Cartier sheaves, it suffices to show that f∗(cid:0) Cohκ(Y, Λ)(cid:1) ⊂ IndCohκ(X, Λ). (3.5) For f of finite type this follows from Theorem 3.2.14 due to the inclusions in (2.1). By Theorem 1.2.3 any essentially finite type morphism f factors as the composite of a finite type morphism and a localized immersion, and so it suffices to show (3.5) for Spec S −→ Spec R induced from a localization homomorphism α : R −→ S = S −1R, S ⊂ R a multiplicative system. 26 3 Cohomological operations for Cartier sheaves Let I be the collection of all multiplicatively closed subszstems S0 of S which are finitely generated. Let (M, κ) be a finitely generated S ⊗ Λ-module and κ : M −→ M be (σS ⊗ idΛ)−1- linear. We first construct a finitely generated S −1 0 R ⊗ Λ-module N0 ⊂ M such that κ(N0) ⊂ N0 for some S0 ∈ I: Consider a presentation S −1Rm ⊗ Λ A−−→ S −1Rn ⊗ Λ −→ M −→ 0. The coefficients of the presentation matrix A involve only finitely many denominators. Thus it is defined already over some S1 ∈ I. Let N1 denote the S −1 1 R ⊗ Λ-module which is the cokernel of the map A over S −1 1 R ⊗ Λ. By enlarging S1 if necessary, we may assume N1 ⊂ M . Since σR is finite by hypothesis, (σ × id)∗N1 is finitely generated over S −1 1 R ⊗ Λ. Any finite set of S −1 1 R ⊗ Λ-generators of (σ × id)∗N1 is also a set of S ⊗ Λ-generators of (σ × id)∗M . Hence there exists S0 ∈ I containing S1 such that κ(N1) ⊂ M is contained in N0 := S −1 0 N1. From κ(s−1m) = κ(s−qsq−1m) = s−1κ(sq−1m) for s ∈ S, one deduces κ(N0) ⊂ N0, as desired. Now observe that (M, κ) = S −1(N0, κ) = lim−→ S0⊆T ∈I T −1(N0, κ). By Theorem 3.2.14, the Cartier sheaves attached to T −1(N0, κ) are locally nil-coherent. It follows that their direct limit (M, κ) lies in IndCohκ(Spec R, Λ), and this completes the proof of the corollary. Remark 3.2.16. The corollary is not true if one replaces IndCohκ by LNilCohκ while keeping f not of finite type, as the following simple example shows: let k be an algebraically closed field, let α : k −→ k(x) be the natural homomorphism and let M = k(x)dx be the module of differentials on k(x) with the usual Cartier operator C given by sending xndx to x(n+1)/p+1 and taking pth roots on k. Then all elements of the form mc := dx x−c , c ∈ k are fixed points of C: C( dx x − c ) = (x − c)−1C((x − c)p−1dx) = (x − c)−1C(xp−1dx) = dx x − c using that C(xidx) = 0 for 0 ≤ i ≤ p − 2. In particular C acts not nilpotently on the mc's. As (mc)c∈k is linearly independent, α∗M is not an element of LNilCohκ(Spec k, k). Corollary 3.2.17. Suppose ∗ ∈ {b, +, −, ∅} and ? ∈ {lnil, lncoh, indcoh}. If f is of finite type, ?(QCohκ(X, Λ)). If f is essentially of finite type, the conclu- sion still holds for ? = ind. then Rf∗(cid:0)D∗ ?(QCohκ(Y, Λ))(cid:1) ⊂ D∗ Proof. Since f is of finite type, the derived functor Rf∗ has finite cohomological dimension. This ensures the convergence of the spectral sequence in Lemma 3.2.10 for any choice of ∗. Condition (a) of Lemma 3.2.10 was established in Corollary 3.2.11, Theorem 3.2.14 and Corollary 3.2.15, respectively. Therefore the proof is an application of Lemma 3.2.10. Corollary 3.2.17 together with Theorem 3.2.6 and Proposition 3.2.9 directly yield: 27 3 Cohomological operations for Cartier sheaves Theorem 3.2.18. Suppose ∗ ∈ {b, +, −, ∅}. If f is of finite type, then D∗(LNilCrysκ(Y, Λ)) D∗ ?(IndCrysκ(Y, Λ)) D∗ ?(QCrysκ(Y, Λ)) Rf∗ Rf∗ Rf∗ / D∗(LNilCrysκ(X, Λ)) / D∗ ?(IndCrysκ(X, Λ)) / D∗ ?(QCrysκ(X, Λ)) commutes for ? ∈ {crys, ∅}. If f is essentially of finite type, the lower square commutes for ? ∈ {ind, ∅}. Moreover via the equivalence Crysκ(Y, Λ) ≃−−→ LNilCrysκ(Y, Λ) from Proposition 2.3.5, for f of finite type the first row induces a functor D∗(Crysκ(Y, Λ)) Rf∗ / D∗(Crysκ(X, Λ)) Remark 3.2.19. Regarding the vertical arrows in Theorem 3.2.18, it is shown in [BB11b, Reference !] that for a regular scheme with finite σX the natural inclusion D∗ ?(IndCrysκ(X, Λ)) ֒−→ D∗ ?(QCrysκ(X, Λ)) is an equivalences of categories for ∗ ∈ {∅, +, −, b}, ? ∈ {indcrys, crys}. For ∗ ∈ {−, b} the same holds for D∗(LNilCrysκ(X, Λ)) ֒−→ D∗ crys(IndCrysκ(X, Λ)). This can be combined with the equivalence Crysκ(X, Λ) ≃−−→ LNilCrysκ(X, Λ) to obtain an −/b equivalence of categories D−/b(Crysκ(X, Λ)) crys(QCrysκ(X, Λ)). This observation can be used to define for proper f : Y −→ X and X regular the pushforward Rf∗ : D−/b(Crysκ(Y, Λ)) −→ D−/b(Crysκ(X, Λ)) without appealing to Corollary 3.2.15. This yields an alternative construction of Rf∗ of Theorem 3.2.18 in this case. ≃−−→ D It is unclear to us if the regularity assumption in the equivalence of categories above is really necessary; in the context of τ -crystals it is not, as shown in [BP09, Chapter 5]. 3.3 Twisted inverse image We first recall the key properties of the twisted inverse image functor for quasi-coherent sheaves f ! for essentially finite type morphisms from [Har66, Cor. VII.3.4], [Lip09, Thm. 4.8.1] and [Nay09, 5.2]. Let C be the category of Noetherian schemes with essentially finite type morphisms (see Definition 1.2.2). Theorem 3.3.1. On the category C, there exists a pseudofunctor ( in C assigns a functor f ! : D+(QCoh(X)) −→ D+(QCoh(Y )) )! which to each f : Y −→ X that is uniquely determined up to isomorphism by the following three properties : (a) On the subcategory of Noetherian schemes with essentially ´etale morphisms, f 7→ f ! is naturally isomorphic to the inverse image pseudofunctor f 7→ f ∗. 28 _ / _ _ / _ / / 3 Cohomological operations for Cartier sheaves (b) On the subcategory of Noetherian schemes with proper morphisms, f 7→ f ! is naturally isomorphic to the right adjoint of the direct image pseudofunctor f 7→ Rf∗. (c) Functorially over C, for any base change diagram of Noetherian schemes j i U g V X f / Y with f proper and i flat there is a flat base change isomorphism j∗f ! ∼−→ g!i∗. If in addition i and j are essentially ´etale, then the flat base change isomorphism agrees with the natural isomorphism j∗f ! ∼−→ j!f ! ∼−→ (f j)! = (ig)! ∼−→ g!i! ∼−→ g!i∗. On the subcategory Cp that possesses a dualizing complex ω• additional results: (d) Let DX denote the dualizing functor on X ∈ Cp S of Noetherian schemes essentially of finite type over a base scheme S S and where all morphisms are proper, one has the following f 7→ DY ◦ Rf∗ ◦ DX are naturally isomorphic pseudo-functors. S induced from ω• S, then f 7→ Rf∗ and (e) The adjoint pair (Lf ∗, Rf∗) yields an adjoint pair (DX◦Rf∗ ◦ DY , DY ◦Lf ∗ ◦ DX) and in particular f 7→ f ! and f 7→ DY ◦ Lf ∗ ◦ DX are naturally isomorphic pseudofunctors. coh(QCoh(Y )). (f ) For each f : Y −→ X in CS one has f !(cid:0)D+ coh(QCoh(X))(cid:1) ⊂ D+ Remark 3.3.2. The abstract existence result of f ! in (b) is attributed by Deligne to Verdier, see [Har66, App. IV]. The construction of f ! for general f proceeds as follows: By Nagata's compactification theorem, any finite type morphism f can be factored as the composite f = ¯f ◦ j of an open immersion j and a proper morphism ¯f . The extension by Nayak recalled in Theorem 1.2.3, yields an analogous factorization for essentially finite type morphisms. One then verifies that f ! = j! ◦ ¯f ! using (a) and (b). In [Har66] f ! is actually defined via such a factorization and a great deal of work goes into showing that this is independent of the factorization. In the presence of a compatible system of dualizing complexes, as in Cp S, the approach indicated in (d) and (e) gives an alternative construction to f ! for proper morphisms. This was carried out in [Har66] based on notes by Grothendieck. The compatibility of the two approaches is explained in [Lip09, App. 4.10]. Since an adjoint can only be defined up to natural isomorphisms of pseudofunctors, in a setting as in (d)-(f), one may simply take f ! = DY ◦ Lf ∗ ◦ DX as the definition. In general there is no explicit description of the functor f !, however in many important cases the situation is better: Example 3.3.3. If f is a finite morphism then f !( is a right derived functor in this case. If in addition f is flat, then f !( and one can verify the identity f !( ) ∼= f !OX ⊗ f ∗( ) ∼= f −1 RH om (f∗OX, ) by hand, cf. [BB11c, Lemma 5.7]. ); in particular f ! ) ) ∼= f −1 H om (f∗OX , More generally, in the case of a smooth morphism f : Y −→ X one has the explicit relationship between f ! and Lf ∗ given by f !( ), and f !OX = ωf [−d] is the relative dualizing sheaf for f placed in degree d = relative dimension of f . This can be generalized as follows: ⊗OY Lf ∗( ) = f !OX L 29 / / / 3 Cohomological operations for Cartier sheaves Remark 3.3.4. A morphism f : Y −→ X is called essentially perfect if it is essentially of finite type, if Rf∗OY ∈ Db coh(QCoh(X)) and if there exist integers m ≤ n such that for all y ∈ Y the module OY,y is quasi-isomorphic to a complex of flat OX,f (y)-modules concentrated in degrees [Har66, between m and n. This notion is a relative version of being of finite Tor-dimension, cf. II.4]. As shown in [Nay09] the key points are: (a) A morphism f is essentially perfect if and only if f !OX is bounded and there is a natural isomorphism of functors f !( ) ∼= f !OX L ⊗OY Lf ∗( ). (3.6) (b) If f is essentially perfect, then f ! is bounded. The isomorphism (3.6) is obtained as follows: First assume that f is proper, then Rf∗(cid:0)Lf ∗( f !OX(cid:1) Proj. formula ⊗OY ) is just the adjoint of (3.6) via the adjunction (Rf∗, f !). In general one uses the factorization of f into h ◦ i where i is essentially ´etale and h is proper, see [Nay09, Section 5.7] for details. ⊗OX Rf∗f !OX −−−−−−−−−→ ( Trf−−−→ ( ) ) L L Since f factors as f = h ◦ i where i is localizing and h is perfect, and i! = i∗ is exact, we may assume that f is perfect. This means OX is f -perfect, i.e. there is m ∈ N such that for all x ∈ X one has that OX,x has a flat OY,f (x)-resolution of length less than m. This immediately implies that Lf ∗ is bounded. Hence by the formula (3.6) and the fact that for essentially perfect f one has f !OY is f -prefect [Nay09, Theorem 5.9], [Lip09, Theorem 4.9.4] it follows that f is bounded. Example 3.3.5. Examples for essentially perfect morphisms are [Nay09]: (a) Any locally complete intersection morphism, as can be seen from the Koszul complex. (b) Any Cohen-Macaulay morphism f : Such f are flat by definition, so that Lf ∗ = f ∗. Moreover they satisfy f !OY ∼= ωf [−d], where ωf is the dualizing sheaf relative to f and d is the relative dimension (locally constant on X). Special cases are Gorenstein morphisms, where in addition ωf locally free of rank 1, and essentially smooth morphism in which case ωf = Vd ΩY /X for ΩY /X is the relative sheaf of differentials. (c) Any finite type morphism f with X, Y regular; this will be discussed in detail in Lemma 4.2.2 below. In order to show that the functor f ! descends to κ-sheaves we use the duality for a finite morphism applied to the Frobenius (σ × id). This states that the functors (σ × id)∗ and (σ × id)! form an adjoint pair of functors. Observing Theorem 3.3.1, it says that in the derived category D+ +/b (σ × id)!V •. To state this observation, let us define (ad hoc) the category D ? ? (QCoh(X × C)) giving a map κ : (σ × id)∗V • −→ V • is equivalent to giving a mapeκ : V • −→ (QCoh(X × C)) andeκ : V • −→ as the category whose objects are pairs (V •,eκ) with V • ∈ D (σ × id)!V •. The homomorphisms are the ones which commute witheκ. ? (QCoh(X × C))κ to (V •,eκ) whereeκ and +/b Lemma 3.3.6. The derived category D ? C))eκ. The equivalence is given by sending (V •, κ) ∈ D+ κ correspond to one another via the adjunction (duality of finite maps) +/b (QCoh(X × C))κ is equivalent to D ? HomD+ (V •, (σ × id)!V •) ∼= HomD+((σ × id)∗V •, V •). (QCoh(X × C))eκ (QCoh(X × +/b ? recalled in Theorem 3.3.1. 30 3 Cohomological operations for Cartier sheaves ? (QCoh(X × C))eκ, just by the compatibility of ( It is clear how for f : Y −→ X essentially of finite type the functor (f × id)! from Theorem 3.3.1 )! with composition. ? (QCoh(X × C))κ for ? ∈ {∅, coh}. induces a functor on D+ By the preceding Lemma 3.3.6 this induces an (f × id)! on D+ Concretely, given an element V • ∈ D+ we apply (f × id)! and obtain a functorial morphism ? (QCoh(X × C))κ with its adjointeκ : V • −→ (σ × id)!V • (f × id)!V • (f ×id)!(eκ) −−−−−−−→ (f × id)!(σ × id)!V • ∼= (σ × id)!((f × id)!V •). (3.7) This naturally gives (f × id)!V • the structure of an element of D+ defines a functor ? (QCoh(X × C))κ and hence f ! : D+ ? (QCoh(X × C))κ −→ D+ ? (QCoh(Y × C))κ. (3.8) Definition 3.3.7. From (3.8), for ? ∈ {coh, ∅}4, we define the functor f ! : D+ ? (QCohκ(X, Λ)) −→ D+ ? (QCohκ(Y, Λ)) via the equivalence D+ ? (QCohκ( , Λ)) −→ D+ ? (QCoh( × C))κ of Theorem 3.1.1. If f ! has bounded cohomological dimension, then by viewing the datum of (3.7) as an ob- ?(. . .)κ. In particular, by ?(QCoh(Y × C))κ one easily verifies that f !V • lies in Db ject V • ∈ Db Remark 3.3.4, we obtain: Proposition 3.3.8. Suppose f : Y −→ X is essentially perfect. Then for ? ∈ {coh, ∅} one has This holds in particular if X and Y are regular. f !(cid:0)Db ?(QCohκ(X, Λ))(cid:1) ⊂ Db ?(QCohκ(Y, Λ)). Remark 3.3.9. For general essentially finite f one cannot expect that f ! is bounded, since this already fails for a closed immersion Spec R/I ֒−→ Spec R if R/I is not of finite projective dimension over R. However, in Section 4.2 we shall prove that the induced functor f ! on complexes of crystals is bounded. In light of a duality theory that is the content of [BB11b] such a boundedness result is expected. The duality relates κ-crystals to τ -crystals of [BP09] and the functor corresponding to f ! is Lf ∗ which is shown to be exact in [BP09, Prop. 6.1.5, Thm. 5.3.1(c)]. Remark 3.3.10. In (3.7) we defined the structural map for f ! by a twofold application of the adjunction of pairs (V •, κ) and (V •,eκ). If one goes through the definitions, then the sheaf un- derlying f !V • is (f × id)!V • ∈ D+ the following commutative diagram: coh(QCoh(Y, Λ)) and the homomorphism κf !V • is defined by (σ × id)∗(f × id)!V • id−→σ!σ∗ / (σ × id)∗(f × id)!(σ × id)!(σ × id)∗V • (3.9) ∼= κ f !V • (σ × id)∗(σ × id)!(f × id)!(σ × id)∗V • (f × id)!V • (f ×id)!(κ) σ∗σ!−→id (f × id)!(σ × id)∗V •. For later use, we call the map from top left to bottom right a base change map: (σ × id)∗(f × id)! base change −−−−−−−−→ (f × id)!(σ × id)∗. (3.10) 4In Proposition 3.3.21 we shall prove the same for ? ∈ {lnil, lncoh} if f is essentially of finite type. 31 / o o 3 Cohomological operations for Cartier sheaves Remark 3.3.11. For finite maps f this base change morphism (3.10) can be made more explicit as follows. Let us first assume that X = Spec R and Y = Spec S are affine, and f is given by a finite map of rings f : R −→ S. To R f S σR σ∗R f σS / σ∗S and a single Cartier module V over R the base change morphism (3.10) on zeroth cohomology is given by the composition Homσ∗R(f∗σ∗S, V ) −→ Homσ∗R(f∗σ∗S, HomR(σ∗R, σ∗V )) −→ HomR(f∗σ∗S, σ∗V ) −→ HomS(σ∗S, HomR(f∗S, σ∗V )) −→ HomR(f∗S, σ∗V ) , Disregarding the first map for a moment, the middle two steps are given by Hom-⊗ ad- junction and express the commutation of f !σ! ∼= σ!f ! and the last is just evaluation at 1. But tracing through these last three steps one observes that their composition is just given by precomposition with the Frobenius σS and post-composition with the evaluation at 1 on HomR(σ∗R, σ∗V ) −→ σ∗V .5 But the first map is just given by post-composition with the natural map V −→ HomR(σ∗R, σ∗V ) which composed with the evaluation at 1 is the identity. It follows that the whole composition of all four maps is just given by precomposition with σS. This explicit computation in the affine case shows that more generally, for any finite map f : Y −→ X and a complex of Cartier sheaves V • the base change σ∗f !V • = σ∗f −1 RH om (f∗OX , V •) −→ f −1 RH om (f∗OX , σ∗V •) = f !σ∗V • is given on underlying OX -modules and after choosing some injective resolution I • of V • by the composition σ∗f −1 H om (f∗OX, I •) −→ f −1 H om (f∗σ∗OX , σ∗I •) −→ f −1 H om (f∗OX, σ∗I •) where the first map sends ϕ to σ∗ϕ and the second is composition with the Frobenius action σX on OX . Therefore, the extraordinary pullback f !V • is given by the pair consisting of the complex of sheaves (f × id)−1 RH om ((f × id)∗OX×C , V •) and the morphism of complexes (σ × id)∗(f × id)−1 RH om ((f × id)∗OX×C, V •) −→ (f × id)−1 RH om ((f × id)∗(σ × id)∗OX×C, (σ × id)∗V •) −→ (f × id)−1 RH om ((f × id)∗OX×C, V •), where the second morphism uses on the right argument the structural homomorphism κV : (σ × id)∗V • −→ V • and on the left simply the Frobenius morphism (σ × id) : OX×C −→ (σ × id)∗OX×C. 5The second map is given by ψ 7→ [α 7→ (ψ(α))(1)]. The third map by ϕ 7→ [γ 7→ [β 7→ ϕ(γσS(β))]] and the third is evaluation at 1. Combining these we get that ψ under this composition is a map that sends s to ψ(σS(s))(1) = ev1 ◦ ψ ◦ σS(s) as claimed. 32 / / / 3 Cohomological operations for Cartier sheaves Example 3.3.12. If i : Y ֒−→ X is a hypersurface in affine n-space given by the vanishing of a regular element f ∈ R = k[x1, . . . , xn]. The Koszul complex K•(f, R) = 0 −→ R −−→ R −→ 0 is a R-free resolution of S = R/f . The Frobenius on R/f lifts uniquely to the Koszul complex (homological degree on the right is zero) f 0 0 f / R f p−1σ R σ / σ∗R σ∗f / σ∗R yielding a map τ : K•(f, R) −→ σ∗K•(f, R) which is compatible with the Frobenius action σ : R/f −→ σ∗R/f . Using this free resolution of R/f we compute that for any Cartier mod- ule (M, κ) the restriction i!M is represented by the complex HomR(K•(f, R), M ). As by the above recipe we may use the lift τ of the Frobenius to the Koszul complex to compute that the structural map on i!M is given by σ∗ HomR(K•(f, R), M ) −→ HomR(σ∗K•(f, R), σ∗M ) κ◦ ◦τ −−−−−→ HomR(K•(f, R), M ). Concretely, identifying i!M = HomR(K•(f, R), M ) with the complex M at 1, the structural map i!κ is given by f −−→ M via evaluation σ∗M σ∗f σ∗M κ M κf p−1 f / M More generally, a similar description can be obtained for any complete intesection Y ֒−→ X by noting that the Koszul complex K(f1, . . . , ft, R) on a regular sequence is just the tensor product of the Koszul complexes K(fi, R) for i = 1, . . . , t. In this case, if the Cartier module (M, κ) is locally free, then i!M is equal to (M/(f1, . . . , ft)M )[−t] and the structural map is given by κ((f1 · . . . · ft)p−1 · ). Next we establish that the previously defined functor j∗ for essentially ´etale morphisms j : U −→ X agrees with j! here. Proposition 3.3.13. Suppose j : Y −→ X is essentially ´etale, then j∗ as defined in Definition 2.2.2 is isomorphic to j! as defined above. Proof. This comes down to verifying that the flat base change isomorphism (j × id)∗(σ × id)∗ −→ (σ × id)∗(j × id)∗ is inverse to the base change isomorphism (σ × id)∗(j × id)! ≃−−→ (j × id)!(σ × id)∗ from (3.10) under the functorial isomorphism (j × id)∗ ≃−−→ (j × id)! of Theorem 3.3.1. To ease notation we omit × id until the end of this proof. The two morphisms are the top and bottom row of the diagram j∗σ∗ σ∗σ∗j∗σ∗ e❏❏❏❏❏❏❏❏❏ ∼= σ∗σ!j∗σ∗ ∼= / σ∗j∗σ∗σ∗ σ∗j∗ zttttttttt σ∗j∗σ!σ∗ ∼= ∼= ∼= j!σ∗ σ∗σ!j!σ∗ σ∗j!σ!σ∗ σ∗j! ∼= 33 / / / / / / / / / / / / / z e o o o o o o o o 3 Cohomological operations for Cartier sheaves By Theorem 3.3.1, the vertical isomorphisms are just the functorial identifications of j∗ −→ j! and hence the bottom parts of the diagram commute. Therefore one has to check that the composition around the top-left pentagon starting at the top-left vertex is the identity. This will be done locally where one can write down all maps explicitly: let j : R −→ S be an essentially ´etale map of rings. Denote σ : R −→ R′ the Frobenius and similarly for S. We have by Lemma 2.2.1 that S ⊗R R′ ∼= S′ which is needed to identify the bottom isomorphism below as given by sending 1 ⊗ ϕ to idS ⊗ϕ. With M = σ∗N an R-module we can now make the top left diagram more explicit: S ⊗R M / σ∗(S′ ⊗S S ⊗R M ) i❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ev1 σ∗ HomS(S ⊗R R′, S ⊗R M ) ∼= ∼= / σ∗(S′ ⊗R′ R′ ⊗R M ) σ∗(S′ ⊗R′ HomR(R′, M )) Since the maps are S-linear, it is enough to check on generators, i.e. on elements of the form 1 ⊗ m, that the total composite is the identity: 1 ⊗ m 7→ 1 ⊗ 1 ⊗ m 7→ 1 ⊗ 1 ⊗ m 7→ 1 ⊗ [ϕm : r 7→ rm] 7→ idS ⊗ϕm 7→ 1 ⊗ m This shows our claim. Remark 3.3.14. The compatibility of the two base changes for the Frobenius under an essentially ´etale morphism is a consequence of a more general result about cartesian squares: Suresh Nayak explained to us [Nay12] that for any cartesian square Y ′ s′ X ′ j′ j Y s / X with j essentially ´etale and s proper, the composition j∗s∗ −→ s′ ∗s′∗j∗s∗ −→ s′ ∗j′∗s∗s∗ −→ s′ ∗j′∗ −→ s′ ∗j′∗s!s∗ −→ s′ ∗s′!j∗s∗ −→ j∗s∗ is the identity. Here composition of the first three maps is the base change under ( composition of the last three maps the base change under ( j! by Theorem 3.3.1. )∗ and the )! after the identification of j∗ with Our next goal is to show that f ! preserves cohomological nilpotence. Since f ! was not con- structed as a derived functor, for arbitrary essentially finite type f this does not follow from a spectral sequence as was the case with Rf∗ before. One way to bypass this problem, is to factor f in a convenient way and reduce thereby to cases which can either be handled directly or by a spectral sequence argument. To prepare for this, we give some basic results on injective κ-sheaves and on adjunction. Let us point out, that in Section 4.3 we will develop a theory of stalks. This will an give alternative proof of cohomological nilpotence by reduction to points. Then morphisms correspond to inclusion of fields, and cohomological nilpotence will be obvious. 34 / / i o o / / / 3 Cohomological operations for Cartier sheaves Adjunction for (j∗, Rj∗) and injective κ-sheaves Proposition 3.3.15. Let j : U ֒−→ X be an essentially ´etale morphism. Then the functor j∗ = j! : D+(QCohκ(X, Λ)) −→ D+(QCohκ(U, Λ)) of Proposition 3.3.13 is naturally left adjoint to the functor Rj∗ : D+(QCohκ(U, Λ)) −→ D+(QCohκ(X, Λ)) of Theorem 3.2.2. Proof. There is a well-known adjunction between Rj∗ and j∗ on the derived category of quasi- coherent sheaves. The adjunction is given by the natural transformations ad : j∗Rj∗ −→ id and cad : id −→ Rj∗j∗ satisfying the usual compatibilities, see [Mac98, Thm. VI.1.2]. We need to show that both of these transformations are compatible with the κ-structure. As we shall see, this will ∼= σ∗ ◦ cad. come down to the verification of natural isomorphism ad ◦σ∗ We now give details. To simplify notation, we omit × id from it until the end of the proof. ∼= σ∗ ◦ ad and cad ◦σ∗ For the natural transformation ad we need to verify that the trapezoid in the following diagram commutes: σ∗ ad σ∗j∗Rj∗ σ∗ κ id b.c. ∼ j∗σ∗Rj∗ can ∼ j∗Rj∗σ∗ ad σ∗ ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ κj∗Rj∗ j∗Rj∗κ ad / σ∗ κ j∗Rj∗ ad / id . Here by b.c. : j∗σ∗ −→ σ∗j∗ we denote the base change isomorphism from Lemma 2.2.1 and ∼−−→ Rj∗σ∗ the canonical isomorphism. By functoriality of the adjunction, the by can : σ∗Rj∗ square on the right is commutative. The triangle in the middle is the definition of κj∗Rj∗ as (j∗Rj∗κ) ◦ can−1 ◦ (b.c.)−1. We claim that the two maps from j∗Rj∗σ∗ to σ∗ in the top row are equal; this will complete the proof of commutativity of the above diagram. For the claim, consider σ∗j∗Rj∗ Rj∗σ∗j∗Rj∗ (3.11) σ∗ σ∗ ad v♠♠♠♠♠♠♠♠♠♠♠♠ h◗◗◗◗◗◗◗◗◗◗◗◗ ad σ∗ b.c. j∗σ∗Rj∗ j∗can j∗Rj∗σ∗ Rj∗σ∗ Rj∗σ∗ ad u❥❥❥❥❥❥❥❥❥❥❥❥ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ad(b.c.) σ∗Rj∗ can Rj∗σ∗, where the right diagram is the adjoint of the left one. It suffices to verify the commutativity on the right. By the definition of b.c, its adjoint ad(b.c) is the composite σ∗Rj∗ σ∗ cad Rj∗ −−−−−−−→ σ∗Rj∗j∗Rj∗ can−−−→ Rj∗σ∗j∗Rj∗. The commutativity of (3.11) now follows from the fact that the composite Rj∗ cad Rj∗ −−−−−−→ Rj∗j∗Rj∗ Rj∗ ad −−−−−→ Rj∗ is the identity. (Switch all σ∗ in the right diagram of (3.11) to the left using can.) 35 o o o o o o / o o / v u O O O O h O O O O 3 Cohomological operations for Cartier sheaves We leave it to the reader to draw the diagram that needs to commute for the compatibility of κ with cad. Here the proof comes down to the commutativity of σ∗ cad σ∗ 6❧❧❧❧❧❧❧❧❧❧❧❧ (❘❘❘❘❘❘❘❘❘❘❘❘ /❴❴❴❴❴❴ σ∗ cad δ Rj ∗ j∗σ∗ b.c. Rj∗σ∗j∗ can j∗Rj∗σ∗, (3.12) where δ is defined as δ = can−1 ◦ σ∗ cad. This time the base change map is the left adjoint of δ, hence δ is the right adjoint of b.c. But the right adjoint of b.c is precisely the composite around the top triangle, and hence that triangle commutes. The adjunction in Proposition 3.3.15 yields some basic results on injectives: Lemma 3.3.16. Let j : U ֒−→ X be an open immersion. If I ∈ QCohκ(U, Λ) is injective, then so is j∗I ∈ QCohκ(X, Λ). The same assertion holds for IndCohκ in place of QCohκ. Finally if I is any κ-sheaf in QCohκ(U, Λ) whose underlying sheaf is injective in QCoh(U × C), then Rj∗I = j∗I and the sheaf underlying j∗I is injective in QCoh(X × C). Proof. By [Wei94, 2.3.10] a functor that is right adjoint to an exact functor, preserves injectives. In the case at hand the functor j∗ is right adjoint to the exact functor j∗ since j is an open immersion. The first assertion is then immediate; for the second assertion, one also needs to invoke Corollary 3.2.15. The last assertion follows in the same way. Proposition 3.3.17. Suppose that {Ui}i∈I is a finite affine cover of a scheme X with corre- sponding open immersions ji : Ui ֒−→ X. Then QCohκ(X, Λ) contains enough injectives of the In particular QCohκ(X, Λ) contains enough injectives. form Li∈I ji∗I i where the I i ∈ QCohκ(Ui, Λ) are injective. Proof. By the previous lemma, the objectLi∈I ji∗I i is injective in QCohκ(X, Λ). Next observe that the defining property of sheaves implies that the natural transformation id ֒−→Li∈I ji∗j∗ i is a monomorphism. Hence it suffices to prove that each category QCohκ(Ui, Λ) for Ui = Spec Ri has enough injectives. This is clear, however, since QCohκ(Ui, Λ) is equivalent to the category of right Ri[σ] ⊗ Λ-modules, and as a module category, the latter has enough injectives, e.g. [Lam99, § 3D]. Lemma 3.3.18. Let I be an injective object of QCohκ(X, Λ). Then its underlying sheaf is injective as a sheaf of OX×C -modules. Proof. Let ϕ : I ֒−→ J be an injective homomorphism of OX×C -modules with J injective, and consider the diagram (σ × id)∗I (σ×id)∗ϕ (σ × id)∗J κI I κJ / J ϕ 36 6 / ( / / ✤ ✤ ✤ / 3 Cohomological operations for Cartier sheaves with the dashed arrow missing at first. Since σ is finite and affine, the morphism (σ × id)∗ϕ is injective. The injectivity of J as an OX×C-module now yields the dashed morphism, making J := (J , κJ ) into a κ-sheaf and ϕ : I −→ J into a monomorphism of κ-sheaves. Because I is an injective κ-sheaf, ϕ is split and thus I is a direct summand of J as a OX×C-module. As direct summands of injectives are injective, the proof is now complete. Cohomological nilpotence of f ! We begin with a consequence of the above results on injective κ-sheaves. Proposition 3.3.19. Suppose f : Y −→ X is a finite morphism. Then the pullback functor f ♭ = f −1 H om Y (f∗OX , ) on QCohκ(X, Λ) is left exact. It induces a right derived functor on D+(QCohκ(X, Λ)) that agrees with f !. In particular there exists a convergent spectral sequence Ei,j 2 = Rif !(H jV •) =⇒ Ri+jf !V •. Proof. Clearly f ♭ is left exact. Since by Proposition 3.3.17 the category QCohκ(X, Λ) possesses enough injectives, the functor f ♭ induces a right derived functor Rf ♭ on D+(QCohκ(X, Λ)). By standard arguments, it follows that Rf ♭ has a convergent spectral sequence as claimed for f !. It remains to show that f ! ∼= Rf ♭. The functor f ! is defined as the functor (f × id)! on D+(QCoh(X × C)) with an induced endomorphism (of complexes), see Remark 3.3.10. It therefore suffices to show that f ! can be computed by resolutions by injective κ-sheaves. But this follows from Lemma 3.3.18 which asserts that the sheaves underlying an injective κ-sheaf are injective in QCoh(X × C). The next two results clarify in particular cohomological nilpotence for f !. Lemma 3.3.20. Let ? be in {lnil, lncoh, indcoh}. If f : Y −→ X is essentially ´etale or finite, then f !(cid:0)D+ ? (QCohκ(X, Λ))(cid:1) ⊂ D+ ? (QCohκ(Y, Λ)). Proof. If f is essentially ´etale it is in particular flat and therefore f ! ∼= f ∗ is exact so that it commutes with taking cohomology. The result now follows from Lemma 2.2.3. If f is finite, we apply Lemma 3.2.10 with A = A ′ =?, and so we need to verify conditions (a) and (b) of that lemma. The required spectral sequence (b) was established in Proposition 3.3.19 above. For (a), we need to show that the functors Rif ! = Rif ♭ = f −1Ri H om Y (f∗OX , ) preserve local nilpotence and local nil-coherence. By functoriality, Rif ♭ preserves nilpotence: if the n-th iterate of κV is zero, the same holds for κRif ♭V . The preservation of coherence under the Rif ♭ is clear, since it holds already in the absence of a κ-structure. Moreover the functors Rif ♭ commute with direct limits, since they are constructed via injective resolutions. From the three properties just explained, condition (a) of Lemma 3.2.10 is immediate. Proposition 3.3.21. Let ? be in {lnil, lncoh, indcoh}. If f : Y −→ X is essentially of finite type, then f !(cid:0)D+ ? (QCohκ(X, Λ))(cid:1) ⊂ D+ ? (QCohκ(Y, Λ)). Proof. By the factorization result Theorem 1.2.3 of Nayak, and the previous lemma, it suffices to prove the result when f is a morphism of finite type. Let U be an affine cover of X and V an affine cover of Y refining the open cover f −1(U). Since by Proposition 2.2.5 containment in ? is a local property, it suffices to prove the proposition for the affine finite type morphisms V −→ U for 37 3 Cohomological operations for Cartier sheaves V ∈ V and U ∈ U which arise by restriction from f . Such a morphism factors as V ֒−→ An U −→ U for a closed immersion V −→ An U −→ U . By the previous lemma, it remains to prove the assertion for the smooth morphism pr. By (3.6) we find that f ! = f ∗ ⊗ ωf with ωf is an exact functor. It clearly preserves nilpotence and coherence and it commutes with filtered direct limits. As in the previous lemma, these three properties suffice to complete the proof. U and the canonical projection pr : An ∼= OAn X Together with Theorem 3.2.6 and Proposition 3.2.9, the previous proposition yields: Theorem 3.3.22. If Y induces an exact Λ-linear functor f −−→ X is essentially of finite type, then the functor f ! from Definition 3.3.7 f ! : D+ crys(QCrysκ(X, Λ)) −→ D+ crys(QCrysκ(Y, Λ)) . In Theorem 4.2.1 we shall prove the stronger result that f ! maps Db crys(QCrysκ(Y, Λ)), which means that f ! is a bounded functor on crystals. Db crys(QCrysκ(X, Λ)) into Adjunctions for crystals. Proposition 3.3.23. Let f : Y −→ X be a proper morphism. As functors on categories of the form D+ ? (QCohκ(. . .)) for ? ∈ {lncoh, ind, ∅}, the functor f ! of Proposition 3.3.21 is naturally right adjoint to the functor Rf∗ of Corollary 3.2.17. Proof. The proof is parallel to that of Proposition 3.3.15. The adjunction between Rf∗ and f ! on the derived category of quasi-coherent sheaves can be formulated in terms of the trace homomor- phism tr : Rf∗f ! −→ id and its bi-adjoint ctr : id −→ f !Rf∗; see [Har66, p. 383 or p. 417] or [Lip09, 4.1].6 To extend this adjunction to functors on D+ coh(QCohκ(. . .)), one has to show that tr and ctr are compatible with the structural map κ. This can be expressed as the commutativity of di- agrams like (3.11) and (3.12) with ad and cad replaced by tr and ctr. As with Proposition 3.3.15, the argument only uses formal properties of adjunction. Details are left to the reader. Corollary 3.3.24. Let f : Y −→ X be proper and j : U −→ X be essentially ´etale. Then the adjoint pairs (Rf∗, f !) from Proposition 3.3.23 and (j∗, Rj∗) from Proposition 3.3.15 induce adjuctions on the corresponding categories D+ := D+ crys(QCrysκ(. . .))7 of κ-crystals. Concretely, there are natural functorial isomorphisms and HomD+(Rf∗ , ) ∼= HomD+( , f ! HomD+(j∗ , ) ∼= HomD+( , Rj∗ ) ) Proof. Let (F, G) be either (j∗, Rj∗) for an essentially ´etale morphism j : U ֒−→ X, (Rf∗, f !) for a proper morphism f : Y −→ X. By Proposition 3.3.15 and Proposition 3.3.23, we have natural transformations a : F G −→ id and c : lncoh(QCohκ(. . .)) such that F F c−−−→ F GF aF−−−→ F and G cG−−−→ GF G Ga−−−→ G are the identity. Under localization at nil- isomorphisms we obtain induced natural transformations on categories D+ crys(QCrysκ(. . .)) and the formulae yielding idF and idG, respectively, are preserved; see [BP09, Prop. 2.2.5]. id −→ GF on D+ 6Note that for proper f neither Deligne's appendix to [Har66] or [Lip09] require complexes with coherent cohomol- ogy. Note also that since our schemes are separated, we may stay within the derived categories of quasi-coherent sheaves. 7The adjunction (j∗, Rj∗) also holds in the unbounded derived category by the same argument. 38 4 Support, Kashiwara equivalence, and applications 4 Support, Kashiwara equivalence, and applications In this section we shall establish a basic exact triangle which allows us to decompose complexes of Cartier crystals on a scheme X into their restriction to an open subscheme U and a closed com- plement Z. Applications of this triangle are given in the subsections. First we show in Section 4.2 that f ! on Cartier crystals has bounded cohomological dimension. In Section 4.3 we obtain some basic results on the crystalline support of a complex of Cartier crystals. 4.1 An exact triangle and Kashiwara equivalence Throughout this subsection, we fix an open immersion j : U ֒−→ X and a complement i : Z ֒−→ X, i.e. a closed immersion i : Z ֒−→ X such that U = X \ i(Z) on points. Due to Corollary 4.1.3 there is no harm in assuming that Z carries its reduced scheme structure. The following is the first main result of this subsection, see also [BB11c, Section 3.3]: Theorem 4.1.1. In D+ ? (QCrysκ(X, Λ)), for ? ∈ {∅, crys}, there is a natural exact triangle i∗i! −→ id −→ Rj∗j∗ −→ i∗i![1] Proof. Let a ⊂ OX denote the ideal sheaf corresponding to i : Z ֒−→ X. Since i is finite, the functor i! is the right derived functor i−1RH om OX×C (OZ×C, ). Moreover the homomorphism i∗i! −→ id is the evaluation at 1 map RH om OX×C (OZ×C, ) −→ . Explicitly, if one is given a complex I • of injective κ-sheaves, this is the map of complexes which in each degree i is the evaluation at the 1 section H om OX×C (OZ×C, I i) −→ I i. Next note that if I i is an injective κ-sheaf, then by Lemma 3.3.18 its underlying sheaf I i is injective as an OX×C -module. Hence by [Har77, Exer. III.3.6] the sheaf I i is flasque and hence so is j∗I i. Thus Rj∗j∗I • is given by j∗j∗I • since Rj∗ can be computed via flasque resolutions. Since we may compute all terms in the asserted triangle of the theorem via injective resolutions of κ-sheaves, to prove the theorem it suffices to show that the natural short sequence of complexes H om OX×C (OZ×C, I •) −→ I • −→ j∗j∗I • is degree wise exact as a complex of κ quasi-crystals. To verify this we may assume that X = Spec R is affine and hence a is an ideal of R. We shall simply write I for the injective κ-module on R × Λ and eI for the associated sheaf on Spec R. Denoting by I[a] = Hom(R/a, I) the a-torsion submodule of I, we need to show that (4.1) 0 −→ I[a] −→ I −→ Γ(U, j∗j∗eI) −→ 0 is exact in the category of κ quasi-crystals. By [Har77, Exer. III.3.7] the right hand term is given HomR×Λ(an × Λ, I). The injectivity of I and the by a formula of Deligne as Γ(U, j∗j∗eI) = lim−→n exactness of 0 −→ an −→ R −→ R/an −→ 0 imply that for all n 0 −→ I[an] −→ I −→ HomR×Λ(an × Λ, I) −→ 0 39 4 Support, Kashiwara equivalence, and applications is a short exact sequence, and hence so is the direct limit over this. The key observation now is that for all n the inclusion I[a] ֒−→ I[an] is a local nil-isomorphism: this is so, because κe(I[a[pe]]) ⊂ I[a] for all e ∈ N (cf. the proof of Lemma 4.3.1) and because the powers an and a[pe] generate the same topology on R, since in the ordinary- and bracket-powers are co-final within one another. It follows that (4.1) is exact in D+(QCrysκ(X, Λ) and hence the proof of the theorem is complete for ? = ∅. For ? = crys we only need to observe that the triangle preserves the subcategory D+ crys(QCrysκ(X, Λ). As an immediate consequence we record here an analog of the Kashiwara equivalence. As above let i : Y ֒−→ X be a closed subset with open complement i ֒−→ X Y j ←−֓ U . We define, that a κ-crystal M on X is supported in Y if and only if its restriction to U is zero (as a crystal), i.e. j!M = j∗M = 0. Likewise, a complex M • ∈ D+(QCrysκ(X, Λ)) is supported in Y if j∗M • is a acyclic as a complex of κ-crystals.8 Since j∗ is exact for open immersions, this is equivalent to the cohomology H i(M •) of M • being supported in Y for all i. We obtain the following equivalence of derived categories of crystals on Y with crystals on X which are supported on Y . Theorem 4.1.2 (Kashiwara equivalence). Let i : Y ֒−→ X be a closed immersion. For ? ∈ {∅, crys} and ∗ ∈ {b, +}, the essential image of the functor i∗ : D∗ ?(QCrysκ(Y, Λ)) −→ D∗ ?(QCrysκ(X, Λ)) is the full triangulated subcategory D∗ ?(QCrysκ(X, Λ) consisting of com- plexes which are supported in Y . The inverse to i∗ is given by i! which hence yield a pair of inverse equivalences of categories ?,Y (QCrysκ(X, Λ)) of D∗ D∗ ?(QCrysκ(Y, Λ)) i∗ i! / D∗ ?,Y (QCrysκ(X, Λ)) . Proof. Let j : U ֒−→ X be the open complement of Y in X. For N • ∈ D+ ? (QCrysκ(Y, Λ)) clearly j∗i∗(N •) = 0 since this is already the case on underlying quasi-coherent sheaves. Hence i∗N • is supported in Y . Likewise, the natural transformation id −→ i!i∗ is an isomorphism since this is already the case on the underlying quasi-coherent sheaves.9 Now, for M ∈ D+ ? (QCrysκ(X, Λ)) consider the triangle from Theorem 4.1.1 i∗i!M • −→ M • −→ Rj∗j∗M • −→ i∗i![1] 8Below in Section 4.3 we will introduce stalks for κ-crystal and a notion of crystalline support of a κ-sheaf. The condition that M • is supported in Y means then precisely that its crystalline support is contained in Y , justifying the ad hoc definition above. 9We have i!i∗( ) = i−1 RH om (i∗OY , i∗ ) ∼= RH om (Li∗i∗OY , ∼=←−− RH om (OY , ) ) = id( ) where the last isomorphism is induced from Li∗i∗OY −→ OY which is an isomorphism for any closed immersion. 40 / o o 4 Support, Kashiwara equivalence, and applications and note that by definition, if M • is supported in Y , then the term Rj∗j∗M • is zero, and hence we have a quasi-isomorphism i∗i!M • −→ M •, showing essential surjectivity of i∗. This shows the claimed equivalence of categories for ∗ = +. The case ∗ = b follows from the case ∗ = + since i∗ and i! both have bounded cohomological dimension on crystals by Theorem 3.2.18 and Theorem 4.2.1.10 As an immediate corollary we obtain: Corollary 4.1.3. Let i : Xred −→ X be the closed immersion of the reduced subscheme of a scheme. Let ? be in {∅, crys} and ∗ ∈ {b, +}. Then i! and i∗ define natural isomorphisms D∗ ?(QCrysκ(Xred, Λ)) i∗ i! / D∗ ?(QCrysκ(X, Λ)) Proof. We apply the above with Y = Xred and hence U = ∅. From this equivalence on the level of the derived category we immediately obtain an equivalence on the underlying abelian category which was already shown in the case of Crysκ in [BB11c, Prop. 3.19] Corollary 4.1.4. Let i : Y −→ X be a closed immersion. Denote by QCrysκ,Y (X, Λ) the category of Cartier crystals supported on Y . Then i! and i∗ define natural isomorphisms QCrysκ(Y, Λ) i∗ i! / QCrysκ,Y (X, Λ) and in particular Rai! is zero on QCrysκ,Y (X, Λ) for a 6= 0. The same statement hold for Crysκ. Proof. For a single (coherent) κ-crystal M Theorem 4.1.2 shows that i!i∗M ∼= M . In particular, the higher cohomology Rai! vanishes for a 6= 0. Now all claims follow from Theorem 4.1.2. 4.2 Finite cohomological dimension of the twisted inverse image As noted before in Remark 3.3.9, given a morphism f : Y −→ X of essentially finite type, the func- tor f ! on quasi-coherent sheaves may have infinite cohomological dimension. However, in [BB11b] we show that Grothendieck-Serre duality induces a duality between Db crys(QCrysκ(X, Λ)) and the corresponding category of τ -crystals of [BP09]; the functor f ! then corresponds to f ∗ which is shown to be exact for κ-crystals in op. cit. Hence f ! has finite cohomological dimension of κ-crystals. The bounds on the cohomological dimension of f ! one obtains via this reasoning are however not optimal. In fact, we show in this section directly that for κ-crystals the cohomlogical width of f ! on D+ crys(QCrysκ((. . .)) satisfies the same bounds as one would expect if X and Y were both smooth. In the smooth case one has -- already on the underlying sheaves -- that f !(M ) for M a sheaf has cohomology only in the range [− dim Y, dim X]; this is well known and explained below in some detail. 10The proof of Theorem 4.2.1 below only uses Theorem 4.1.1 and not Theorem 4.1.2, so there is no circular reasoning. 41 / o o / o o 4 Support, Kashiwara equivalence, and applications Theorem 4.2.1. Let f : Y −→ X be a morphism essentially of finite type and suppose that X has finite Krull dimension dim X. Then for any V in Crysκ(X, Λ), the complex f !V has cohomology supported in degrees [− dim Y, dim X]. The proof is a typical application of our triangle Theorem 4.1.1: One chooses U to be a dense open subset with desirable properties (in our case a dense affine regular subset). On U the statement is known. Then the triangle Theorem 4.1.1 allows to reduce the general case to the corresponding statement on the complement of U , which is dealt with by induction on dimension. In the present situation this type of argument is applied twice, on the base X and then on Y . But before we carry out the proof in detail, let us recall the situation when X and Y are regular: Lemma 4.2.2. Let f : Y −→ X be a finite type morphism of equidimensional regular schemes with X (and hence Y ) of finite Krull dimension. Then f !OX is quasi-isomorphic to a complex ωf [dim Y − dim X] with ωf locally free of rank 1. Furthermore, one has a functorial isomorphism f !( ) = f !OX L ⊗ Lf ∗( ), so in particular f is perfect. Proof. To prove the assertion on the shape of f !OX it suffices to show that its representing complex is acyclic outside degree dim X − dim Y and has a locally free sheaf of rank 1 as the cohomology in that degree. Since this is a local assertion, we may factor f (locally but without changing the notation) as i−→ An × X Y pr2−−−→ X with i a closed immersion. Because Y and An × X are regular, i is a locally complete intersection morphism. Via the Koszul complex and the explicit formula for i! for the finite morphism i, the latter implies that i!OAn×X is represented by a complex of the form ωi[dim Y − dim X − n] as stated. The assertion on f !OX follows since pr! 2 OX = OAn×X[n]. The remaining assertions follow from [Nay09], see also (3.6) on page 30. This description of f !( ) in the case of a finite type morphism between regular schemes im- mediately implies the following result which shows Theorem 4.2.1 in this case. Corollary 4.2.3. Let f : Y −→ X be a finite type morphism of equi-dimensional regular schemes with X (and hence Y ) of finite Krull dimension. Then for any quasi-coherent sheaf V on X × C, the complex (f × id)!V has cohomology concentrated in degrees [− dim Y, dim X − dim Y ]. Proof. Observe that the argument in the proof of the previous lemma applies to f × idC where C is any essentially finite type Fq-scheme C. As a finite type morphism between regular schemes, f and hence f × idC are essentially perfect, due to the fact that essentially perfect is preserved by base change. It yields that (f × id)!OX×C is quasi-isomorphic to ωf ×id[dim Y − dim X] for some locally free sheaf ωf ×id of rank one on Y × C and that (f × id)!( ) = ωf ×id ⊗OX×C L(f × id)∗( )[dim Y − dim X]. Thus it remains to prove that L(f × id)∗ has cohomology concentrated in degrees [− dim X, 0] which is a local assertion. So suppose X = Spec R, Y = Spec S for regular rings R, S, f is a ring 42 4 Support, Kashiwara equivalence, and applications homomorphism f : R −→ S and the module underlying V is M on R ⊗ Λ. Then Li(f × id)∗V is the sheaf associated to the module TorR⊗Λ −i (S ⊗ Λ, M ) ∼= TorR −i(S, M ). Using that R is regular of dimension dim X, the right hand side shows that Li vanishes if i is not in [− dim X, 0]. If one drops the assumption of equi-dimensionality in the preceding Corollary we can only conclude that the cohomology of (f × idC)!(V ) is concentrated in degrees [− dim Y, dim X]. The following lemma prepares the induction step of our main theorem: Lemma 4.2.4. Let X be a scheme of finite Krull dimension. Let j : U ֒−→ X be an affine open immersion with i : Z −→ X the closed complement such that dim Z < dim X. (a) The right derived functor of R0i! : Crysκ(X, Λ) −→ Crysκ(Z, Λ) has cohomological dimension at most one. (b) For a morphism f : Y −→ X essentially of finite type consider the fiber squares U ×X Y fU U ej j Y f / X ei i Y ×X Z fZ Z. U and f ! If f ! Z take Cartier-crystals to complexes whose cohomology is supported in degrees [a, b] and [a, b−1], respectively, then f ! takes Cartier-crystals to complexes whose cohomology is supported in degrees [a, b]. Proof. Since the map j is affine, the functor j∗ and hence j∗j∗ is exact, as a composite of exact functors on Crysκ(X, Λ). In the exact triangle i∗i! −→ id −→ j∗j∗ of functors on the category D+ crys(QCrysκ(X, Λ)), and for V in Crysκ(X, Λ) it follows that both, id V = V and j∗j∗V are concentrated in degree zero. Hence i∗i!V is concentrated in degrees 0 and 1 only. Since i∗ is exact and faithful part (a) follows. For part (b) we use the same triangle applied to f !(V ) (ei∗ei!f !(V ) −→ f !(V ) −→ej∗ej∗f !(V )) ∼= (ei∗f ! Zi!(V ) −→ f !(V ) −→ej∗f ! and henceej∗ is affine. Thusej∗ is exact. Hence by our assumption on f ! By the definition of fiber squares, for any affine V in Y the subscheme U ×X V ⊂ V is affine U the last entry of the the triangle has cohomology only in degrees [a, b]. By part (a) and the assumption on f ! Z the first entry of the triangle has cohomology only in degrees [a, b − 1 + 1]. The long exact cohomology sequence shows now that f !(V ) has cohomology only in degrees [a, b]. U j∗(V )). 43 / / o o / o o 4 Support, Kashiwara equivalence, and applications Proof of Theorem 4.2.1. Since we may factor f as the composite of a finite type morphism and a localizing immersion and since for a localizing immersion the functor extra-ordinary inverse image is exact, we may assume that f is of finite type. Note also that by Corollary 4.1.3 we may assume that X, and if necessary Y , is reduced. We induct on the dimension of X and for a fixed such dimension, on the dimension of Y : So let f : Y −→ X be a morphism of finite type with X and Y reduced. Let j : U ֒−→ X be the embedding of an affine open regular subscheme U of X such that U is equi-dimensional and dense in all components of X of maximal dimension -- the existence of U is provided by generic regularity, cf. [Mat92, Thm. 24.4]. Let i : Z −→ X be the closed reduced complement of U . By Lemma 4.2.4 and the induction hypothesis applied to Z, it suffices to prove the result if X = U is regular and equi-dimensional. This is treated by induction on dim Y : We choose an affine open embeddingej : V ֒−→ Y such that V is regular, equi-dimensional and dense in all components of Y of maximal dimension and we denote byei : W −→ Y the inclusion of the reduced closed complement. As above, we consider the triangle ei∗(fei)! −→ f ! −→ej∗(fej)! obtained from Theorem 4.1.1. Since the underlying morphisms are affineei∗ and ej∗ are exact. Moreover the cohomological dimension of fei is controlled by induction hypothesis. Thus it remains to show that (fej)! maps Cartier-crystals to complexes with cohomology concentrated in degrees [− dim Y, dim X]. But with notation as in the diagram in Lemma 4.2.4 we have that (fej)! = f ! U ◦ej∗, and since fU is a finite type morphism between equi-dimensional regular schemes, this follows from Corollary 4.2.3. := π!k is a complex concentrated in degree − dim An Remark 4.2.5. By considering the projection π : An k −→ Spec k to the prefect ground field, we see that ω• k = −n. This shows that the An k left bound in our theorem is sharp. Similarly, considering the inclusion i : Spec k ֒−→ An k and the = k[−n], showing that the right bound is Cartier module ωAn k also sharp. These two cases can be pieced together to get an example where there is cohomology precisely in the range allowed by Theorem 4.2.1. := π!k[−n], we get that i!ωAn k 4.3 Stalks of Cartier crystals and the crystalline support In this section, we discuss a notion of stalks for κ-crystals. Their main use will be a criterion for nilpotence of κ-sheaves. Related to this, we also define a notion of crystalline support. In parts we are able to clarify the relation between the crystalline support and the non-vanishing of stalks. For a (not necessarily closed) point x ∈ X denote by ix : Spec kx = x ֒−→ X the natural inclusion. For any κ-crystal V on X over Λ, we define i! xV as the stalk of V at x. In general this is a complex not necessarily concentrated in one degree. The reason for this name is that we shall show that a κ-sheaf V is locally nilpotent if and only if for all x ∈ X the complex i! xV has locally nilpotent cohomology. For x ∈ X we observe that the inclusion ix : Spec kx = x ֒−→ X may be factored as i′ x֒−−→ Spec OX,x x jx−−→ X where i′ essentially ´etale). In Lemma 3.3.20 we observed that if V is (locally) nilpotent, then so is i! x is a closed immersion (and hence finite) and jx is a localization (hence in particular xV 44 4 Support, Kashiwara equivalence, and applications due to its factorization into an essentially ´etale map followed by a finite map. In Lemma 2.2.4 we show that V is locally nilpotent if and only if this is the case for all j! xV = V x. To handle the case of the closed immersion i′ x : x ֒−→ Spec OX,x we first show the following simple, but crucial, lemma. xV = j∗ Lemma 4.3.1. Let (R, m) be a local Noetherian ring, and denote by ix : {m} = x −→ X = Spec R the inclusion of the closed point. Assume that V = (V , κ) is a κ-sheaf that is supported at x. (a) If V is coherent then it is nilpotent if and only if R0i! xV is nilpotent. (b) If V is arbitrary and if σR is finite, then V is locally nilpotent if and only if R0i! xV is locally nilpotent. Proof. If V is (locally) nilpotent, then, by functoriality, and since i! limits, so is R0i! xV . This shows the direction from left to right for both claims. x commutes with filtered direct Assume now that V is finitely generated. Since x is the closed point and ix is a finite map, then R0i! x(V ) = V [m] := HomR⊗Λ(R/m ⊗ Λ, V ) is just the m-torsion part of V viewed as an R/m ⊗ Λ- submodule of V . Note that the Cartier structure on V [m] is just the one given on V = (V , κ) restricted to this submodule. Since V is finitely generated and supported in x we may find e sufficiently big such that m[pe] annihilates V . Then we have κe(V ) = κe(V [m[pe]]) ⊆ V [m] since 0 = κe(rpe xV is nilpotent, then κs(V [m]) = 0 for some s ≥ 0. But this implies that κe+s(V ) ⊆ κs(V [m]) = 0, and hence V itself is nilpotent. This shows part (a). v) = rκe(v) for all r ∈ m ⊗ Λ and v ∈ V . If now R0i! For part (b) we use the local nilpotence criterion of Lemma 2.1.3. Let v ∈ V be arbitrary and set M = (R ⊗ Λ)v ⊆ V . Since M is supported at x and finitely generated we argue exactly as above that we may find e ≥ 0 such that κe((σ × id)e xV . The assumption that σ is finite on R implies that (σ × id)e ∗M is finitely generated such that this is also true for κe((σ × id)e xV is locally nilpotent, it is in particular ind-coherent by Proposition 2.1.5. Hence by Lemma 2.0.6 κe((σ × id)e ∗M ) is contained in a coherent, and hence nilpotent, κ-subsheaf of R0i! ∗ M = 0 for some e′. By the local nilpotence criterion of Lemma 2.1.3 this shows that V is locally nilpotent. xV . As above we conclude that κe+e′(σ × id)e+e′ ∗M ). Assuming now that R0i! ∗M ) ⊆ V [m] = i! An alternative approach to prove the above lemma would be to use the Kashiwara equivalence of the preceding subsection. However, since one point of our notion of stalks is a point-wise nilpotence criterion, which will lead to a new proof that f ! preserves local nilpotence, we have given a direct proof above. We now turn to our point-wise criterion for nilpotence of a κ-sheaf. Theorem 4.3.2. Let V be a κ-sheaf on X. Assuming that either V is ind-coherent or σ is finite on X, then V is locally nilpotent if and only if R0i! xV is locally nilpotent for all x ∈ X. Proof. We show the result under either hypothesis at once. The 'only if' direction has been shown before, so we concentrate on the converse and assume that R0i! xV is locally nilpotent for all x ∈ X. Let M = M0 be a coherent OX×C-subsheaf of V and consider Mi := κi((σ × id)i ∗M ) = κ((σ × id)∗Mi−1) for i ≥ 0. If V is ind-coherent we may and do assume that M is a coherent κ-subsheaf. By Corollary 2.1.4, we have to show that Mi = 0 for i ≫ 0. We first show that all Mi are coherent OX×C -subsheaves of V : In the ind-coherent case this is clear since M is a coherent κ-sheaf and Mi ⊆ M for all i. If σ is finite on X then Mi is coherent because it is the image under κ of the coherent subsheaf (σ × id)i ∗M ⊂ V . 45 4 Support, Kashiwara equivalence, and applications Since, if rMi = 0 then rMi+1 = rκ((σ × id)∗Mi) = κ(r(σ × id)∗Mi) = κ((σ × id)∗rqMi) = 0 def= Supp Mi ⊇ Supp Mi+1. Thus the Zi form we have Ann Mi ⊆ Ann Mi+1. This implies that Zi a descending sequence of closed subsets of X × C which stabilizes to Z. Replacing M by Mi for i ≫ 0 we may assume Supp Mi = Z for all i. Replacing V by the κ-sheaf generated by M , i.e. byPi Mi we may also assume that Supp V = Z as well. Let η be the generic point of a maximal dimensional component of Z and let x = pr1 η be the projection of η to X. Localizing at x we are now in the situation of Lemma 4.3.1, so that V x is locally nilpotent. Using the local nilpotence criterion of Corollary 2.1.4 it follows that (Mi)x = 0 for i ≫ 0. But by the choice of x this implies that Mi itself is zero for i ≫ 0, which finishes our proof. The above yields the following criterion for local nilpotence: Corollary 4.3.3. For a κ-sheaf V ∈ IndCohκ(X, Λ) (or V ∈ QCohκ(X, Λ) if σ is finite on X) the following conditions are equivalent. (a) V is locally nilpotent. (b) R0i! xV is locally nilpotent for all x ∈ X. (c) i! xV is cohomologically nilpotent for all x ∈ X (i.e. Rai! x ∈ X). xV is locally nilpotent for all a and Remark 4.3.4. If X is of finite type over a field, we shall see in Corollary 4.3.21 that the above conditions are equivalent to (d) i! xV is cohomologically nilpotent for all closed points x ∈ X. Example 4.3.5. For arbitrary X, condition (d) is not equivalent to the others: Let Λ = Fq and let X = Spec V be the spectrum of a discrete valuation ring V with uniformizer π and (for simplicity) perfect residue field k. Let i : s −→ X be its closed point and j : η −→ X be its generic ∼= V dπ be the dualizing sheaf of X with κ-structure obtained from an isomorphism point. Let ωX ωX ∼=−−→ σ!ωX. Setting V = j∗j∗ωX we get from the triangle i∗i!ωX −→ ωX −→ j∗j∗ωX = V by applying i! (and using that i!i∗ = id) that i!M = 0. However, V is cleraly nonzero. Corollary 4.3.6. For V • ∈ D+ X) the following conditions are equivalent: ind(QCohκ(X, Λ)) (or V • ∈ D+(QCohκ(X, Λ)) if σ is finite on (a) V • is cohomologically nilpotent. (b) For all x ∈ X the complexes i! xV • are cohomologically nilpotent. Proof. The implication (a) ⇒ (b) is a particular case of Lemma 3.3.20. The proof (b) ⇒ (a) is a standard inductive devissage argument based on Corollary 4.3.3: Assume that H l(V •) is locally nilpotent for all l < l0. Let the truncation τ≤l0V • be the sub- complex of V • defined by (τ≤l0V •)l = V l im(V l−1 −→ V l) 0 for l < l0 for l = l0, and for l > l0. 46 4 Support, Kashiwara equivalence, and applications Note that by our choice of l0 the complex τ≤l0V • is cohomologically nilpotent. If H • denotes the degree-wise quotient of V by τ≤l0V • one has the exact triangle τ≤l0V • −→ V • −→ H • +1−−−→ . x to this triangle and the direction (a) ⇒ (b) for τ≤l0V • we see that i! Applying i! xH • is cohomologically nilpotent for all x ∈ X. We will be done by induction once we show that H l0(H •) ∼= H l0(V •) is locally nilpotent. By construction, H l = 0 for all l < l0. The lowest term of the spectral sequence Rii! ) derived from Proposition 3.3.19 applied to H • asserts that R0i! xH •). Since for all x ∈ X the right term is locally nilpotent by assumption, Corollary 4.3.3 implies that H l0(H •) is locally nilpotent as desired. xH l0(H •) ∼= H l0(i! ∗H j( ) ⇒ Ri+ji! x( Corollary 4.3.6, which was shown without using any of the results in Section 3.3, yields a second proof of Proposition 3.3.21 that f ! preserves cohomological nilpotence because it allows one to reduce the statement to the case where f is an essentially finite type map between the spectra of fields. Namely if f : Y −→ X is essentially of finite type and V • ∈ D+ lncoh(QCohκ(X, Λ) is cohomologically nilpotent, let y ∈ Y be arbitrary and x = f (y) such that one has the commutative diagram y fy iy Y f By Corollary 4.3.6 it suffices to show that i! xV •) is cohomologically nilpotent. Since i! xV • is cohomologically nilpotent we are reduced to the case of a map of a field K into a finitely generated extension field L. In this case f ! can be described explicitly, is exact and hence clearly preserves cohomological nilpotence: y(i! ix x / X yf !V • = f ! Example 4.3.7. If f : Spec L −→ Spec K is induced by the inclusion K ⊆ L of finitely generated field extensions, then f ! can be described explicitly as follows: (a) If L/K is purely transcendental of transcendence degree n, then f ! = L[n] ⊗K . (b) If L/K is finite algebraic, then f ! = HomK (L, ). (c) If L/K is finite separable, the trace map trL/K : L −→ K induces an isomorphism L ⊗K l⊗v7→trL/K (l)v −−−−−−−−−−−→ HomK(L, ) which identifies f ! ∼= f ∗ = L ⊗K In each of these cases f ! is an exact functor and hence preserves nilpotence. Definition 4.3.8. Let V be a coherent κ-sheaf on X over Λ. The crystalline support of V is defined as Suppcrys V := pr1(cid:16)\e≥0 Supp κe(V )(cid:17). Remark 4.3.9. For V ∈ Cohκ(X, Λ), the sets Supp κe(V ) form a descending sequence of closed subsets of the noetherian scheme X × C. Therefore the sequence becomes stationary, and in particular Suppcrys V := pr1(cid:0) Supp κe(V )(cid:1) for all e ≫ 0. 47 / / / 4 Support, Kashiwara equivalence, and applications Lemma 4.3.10. For an affine morphism f : Y −→ X which is a bijection on points and a coherent sheaf F on Y one has Supp(f∗F ) = f (Supp F ). Proof. Since f is affine and the assertion is local, we may assume that Y = Spec S and X = Spec R are affine and that f is given by a ring homomorphism f # : R −→ S. Because f : Spec S −→ Spec R is a bijection on points, the kernel of f # is nilpotent. Therefore it suffices to consider the case where S is an extension ring of R. Again by the bijectivity of f , for any prime ideal p of R the ring S ⊗R Rp is local and thus equal to SP for P = f −1(p). It follows that for any finite type S-module M we have Mp = MP, which proves the lemma. Lemma 4.3.11. For V ∈ Cohκ(X, Λ) and T := Supp(κe(V )) with e ≫ 0, one has T = (σ×id)T . Proof. Suppose that Supp(κe(V )) becomes stationary for e ≥ e0. Then for any e ≥ e0 we have Supp κeV = Supp κe+1V = Supp(cid:0)κ((σ × id)∗κeV )(cid:1) ⊂ Supp(σ × id)∗(κeV ) = (σ × id)(Supp κeV ), where in the equality is by Lemma 4.3.10 using that (σ × id) is a topological isomorphism. This shows that T ⊂ (σ × id)T . But then (σ × id)−n(T ) is a descending sequence of closed subsets of X × C, and so it becomes stationary. Using again that (σ × id) is a topological isomorphism, we must have (σ × id)T = T . By [BP09, Lem. 4.1.6], any closed subset T of X × C such that T = (σ × id)T is a finite union T =Si Yi × Di for closed subsets Yi ⊂ X and Di ⊂ C. We deduce: Corollary 4.3.12. The crystalline support of any V ∈ Cohκ(X, Λ) is closed in X. Proposition 4.3.13. If V , W ∈ Cohκ(X, Λ) are nil-isomorphic, then Suppcrys V = Suppcrys W . Proof. Let ϕ : V −→ W be a nil-isomorphism. After replacing both V and W by their respective image under κe for e ≫ 0, see Remark 4.3.9, we may assume that Supp(κeV ) = Supp V for all e ≥ 0 and the same for W . Now choose e ≥ 0 such that there is a diagram (σ × id)e ∗V (σ×id)e ∗ϕ (σ × id)e ∗W κe α 9ttttttttttt κe V ϕ / W . as in Proposition 2.3.2, where α is a morphism of κ-sheaves. Note that (σ × id)e is regarded as a morphism X × C −→ X × C. Next note that if ψ : F −→ G is a homomorphism of sheaves then Supp(im(ψ)) ⊂ Supp F . Hence κe W = ϕ ◦ α implies that Supp W = Supp κe W ((σe × id)∗W ) ⊂ Supp(im(α)) ⊂ Supp V . The same argument for κe V = α ◦ (σ × id)e ∗ϕ yields Supp V = Supp κe V ((σe × id)∗V ) ⊂ Supp(im((σ × id)e ∗ϕ)) ⊂ Supp((σe × id)∗W ) = (σe × id) Supp W 4.3.11 = Supp W . Combining both inclusions, the proposition follows. 48 / / 9 / 4 Support, Kashiwara equivalence, and applications The following is an immediate consequence: Corollary 4.3.14. Let V be in LNilCohκ(X, Λ). If ϕi : V −→ W i, i = 1, 2, are nil-isomorphisms with W i ∈ Cohκ(X, Λ) (and hence ker ϕi ∈ LNilκ(X, Λ) for i = 1, 2), then Suppcrys W 1 = Suppcrys W 2. This corollary allows us to extend the crystalline support to LNilCohκ(X, Λ) and to complexes in the following way. Definition 4.3.15. Suppose V ∈ LNilCohκ(X, Λ) and ϕ : V −→ W is a nil-isomorphism with W ∈ Cohκ(X, Λ). Then one defines the crystalline support of V as For V • in Db lnilcoh(QCohκ(X, Λ)) we define the crystalline support as Suppcrys V := Suppcrys W . Suppcrys V • :=[i∈Z Suppcrys(H i(V •)). The following result is now obvious: Corollary 4.3.16. (a) The crystalline support of V ∈ LNilCohκ(X, Λ) only depends on the crystal in LNilCrysκ(X, Λ) that it represents. (b) The crystalline support of V • ∈ Db lnilcoh(QCohκ(X, Λ)) only depends on the complex in Db crys(QCrysκ(X, Λ)) that it represents. In both cases, the crystalline support is a closed subset of X. We summarize some properties of the crystalline supports with regards to its behavior under open and closed immersions. Lemma 4.3.17. Denote by j : U −→ X an open immersion with i : Z −→ X a closed complement. For W ∈ {X, U, Z}, let V • crys(QCrysκ(W, Λ)). Then the following hold: W be in Db (a) Suppcrys j∗V • X = U ∩ Suppcrys V • X. (b) Suppcrys j∗V • U ⊂ Suppcrys V • U . (c) If U ∩ Suppcrys V • = ∅, then Suppcrys i!V • X = Suppcrys V • X. (d) Suppcrys i∗V • Z = Suppcrys V • Z . Proof. Part (a) is obvious, since the formation of im κe commutes with restriction to U × C. Part (d) holds because the exact functor i∗ also commutes with the formation of im κe. Next we prove (c): Define W := Suppcrys i!V •. By (d), the complex i∗i!V • has crystalline support W . By our hypothesis j∗V • = 0 and thus j∗j∗V • = 0 holds as well. It follows from the exact triangle in Theorem 4.1.1 that V • ∼= i∗i!V • has crystalline support equal to W , and this proves (c). To prove (b), write ZU for Suppcrys V • U and ZX for the closure of ZU in X, and consider the diagram U iU ZU j ej / X ei / ZX 49 / ? O O / ? O O 4 Support, Kashiwara equivalence, and applications where the horizontal arrows are closed immersions and the vertical ones are open embeddings. From the exact triangle for ZU ֒−→ U ←֓ U \ ZU we obtain V • U U . Therefore ∼= iU ∗i! U V • j∗V • U = j∗iU ∗(i! U V •). U V •) =ei∗(ej∗i! By part (d), the crystalline support of the right hand side lies in ZX, which is (b). Corollary 4.3.18. For V • ∈ Db crys(QCrysκ(X, Λ)) one has {x ∈ X i! xV • 6= 0 ∈ Db crys(QCrysκ(x, Λ))} ⊂ Suppcrys V •. Proof. Denote by U the complement of Suppcrys V • and by j : U ֒−→ X the corresponding open immersion. By Lemma 4.3.17(a) we have j∗V • = 0. Since for any x ∈ U we have ix factors through j, we deduce i! xV • = 0 for all such x. Lemma 4.3.19. Let X be a regular connected scheme and V ∈ Crysκ(X, Λ) with Suppcrys V = X. Then there is a dense open subset U ⊂ X such that for all x ∈ U one has Rdxi! xV 6= 0 in Crysκ(x, Λ) where dx is the dimension of x. Proof. We follow the argument in [BP09, Thm. 4.6.2]. We may replace V by V /V nil and assume that the κ-sheaf V has no nilpotent Cartier subsheaves. This means, that the adjoint to the isomorphism, since (σ × id)!V and V have the same length, cf. [BB11c, proof of Proposition 4.1]. Yi ⊂ X and Di ⊂ C. By passing to a dense open subset of X, we may assume that Yi = X for all i, so that T = X × D for some closed D ⊂ C. structural map eκ : V −→ (σ × id)!V is injective. Let T := Supp κe(V ) for e ≫ 0. As recalled above Corollary 4.3.12, the set T is a finite union T = Si Yi × Di for suitable closed subsets The injectivity of eκ : V −→ (σ × id)!V shows that for any generic point of X × D, eκ is an Let Z ⊂ X × D to be the support of cokereκ. By construction it is a nowhere dense closed subset. Choose D′ a scheme of finite type over Fq such that D is a localization of D′ and let Z ′ be the closure of Z in X × D′. Then W ′ := (X × D′) \ Z ′ is an open dense subset of X × D′. Let V be the set of points x ∈ X for which W ′ ∩ x × D′ is dense in x × D′. As the projection pr1 : X ×D′ −→ X is a morphism of finite type, [Gro67, Prop. 9.5.3] shows that V is constructible. By assumption it contains the generic point of X; hence V contains an open dense subset U ⊂ X. By construction, for every x ∈ U the induced morphismeκ : (ix × id)!V −→ (σ × id)!(ix × id)!V is also an isomorphism at the generic points of x × D. Hence in particular, Rdx i!V is nonzero as a Cartier crystal, provided it is non-zero as a sheaf. Using the Koszul complex for a system of parameters at x one computes that Rdx(i × id)!V ∼= V /mxV which is nonzero since x × D ⊆ Supp V . Remark 4.3.20. Note that in the situation of Lemma 4.3.19, we have Rdi! xV = 0 for all x ∈ X and d > dx. This is so because the Koszul resolution has length exactly dx and thus already Rdi! xV vanishes for all d > dx. Corollary 4.3.21. For V • ∈ Db crys(QCrysκ(X, Λ)) one has {x ∈ X i! xV • 6= 0 ∈ Db crys(QCrysκ(x, Λ))} = Suppcrys V •. If moreover X is of finite type over a field k, then {x ∈ X closed i! xV • 6= 0 ∈ Db crys(QCrysκ(x, Λ))} = Suppcrys V •. 50 References Proof. For any closed immersion i : Z −→ X and x ∈ Z we have i! xi!. Taking Z = Suppcrys V • and applying Lemma 4.3.17(c), we may thus assume that X = Suppcrys V •. To prove the assertion we may now replace X by any dense open subset. By generic regularity, we can therefore assume that X is regular. By passing to a component, we may in addition assume that X is connected. To prove the first assertion, we choose d minimal such that Suppcrys H d(V •) contains the generic point ξ of X. Since dξ = 0, by Remark 4.3.20 the spectral sequence of Proposition 3.3.19 degenerates to x = i! ξ(H jV •) ∼= Rji! i! ξV •. Now by Lemma 4.3.19, we have i! establishes the first assertion. ξ(H dV •) 6= 0. This shows that Rdi! ξV • is non-zero and thus For the second assertion, let d be maximal such that Suppcrys H d(V •) contains ξ. By passing from X to a dense open subscheme we may assume that Suppcrys H d′(V •) = 0 for all d′ > d. Then for any closed point in x, the spectral sequence of Proposition 3.3.19 gives Rji! x(H kV •) =⇒ Rj+ki! xV •. Since X is of finite type over k, we have dx = dim X for any closed point. Using Remark 4.3.20, we deduce for j + k = dx + d the isomorphism Rdxi! x(H dV •) ∼= Rd+dxi! xV •. (4.2) We choose a dense open subset U for H d(V •) as in Lemma 4.3.19. Then for any closed point x ∈ U the left hand side of (4.2) is non-zero and thus for any such x we deduce that i! xV • is non-zero. But since X is of finite type over k, the closed points in U are dense in X and this concludes the proof. Remark 4.3.22. By Example 4.3.5 the set {x ∈ X i! crys(QCrysκ(x, Λ))} is not closed in general. However, we presently do not know if it is closed if on assumes that X is of finite type. xV • 6= 0 ∈ Db References [And00] G. W. Anderson, An elementary approach to L-functions mod p, J. Number Theory 80(2), 291 -- 303 (2000). [BB11a] M. Blickle and G. Bockle, Cartier Crystals and unit OF,X-modules, (in preparation), 2011. [BB11b] M. Blickle and G. Bockle, Cartier Crystals: Duality, (in preparation), 2011. [BB11c] M. Blickle and G. Bockle, Cartier modules: finiteness results, J. Reine Angew. Math. 661, 85 -- 123 (2011). [Bli08] M. Blickle, Minimal γ-sheaves, Algebra & Number Theory 2(3), 347 -- 368 (2008). [Bli13] M. Blickle, Test ideals via algebras of p−e-linear maps, J. Algebraic Geom. 22(1), 49 -- 83 (2013). 51 References [BMS08] M. Blickle, M. Mustat¸a and K. E. Smith, Discreteness and rationality of F -thresholds, Michigan Mathematical Journal 57, 43 -- 61 (2008), Special volume in honor of Melvin Hochster. [BP09] G. Bockle and R. Pink, Cohomological Theory of Crystals over Function Fields, European Mathematical Society, Zurich, 2009, Tracts in Mathematics. [BS05] N. Budur and M. Saito, Multiplier ideals, V -filtration, and spectrum, J. Algebraic Geom. 14(2), 269 -- 282 (2005). [BST11] M. Blickle, K. Schwede and K. Tucker, F-singularities via alterations, ArXiv e-prints (July 2011), 1107.3807. [Del77] P. Deligne, Cohomologie ´etale, Lecture Notes in Mathematics, Vol. 569, Springer- Verlag, Berlin, 1977, S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie SGA, Avec la collaboration de J. F. Boutot, A. Grothendieck, L. Illusie et J. L. Verdier. [EK04] M. Emerton and M. Kisin, Riemann -- Hilbert Correspondence for unit F -crystals, Ast´erisque 293, vi+257 pp. (2004). [EV92] H. Esnault and E. Viehweg, Lectures on vanishing theorems, volume 20 of DMV Seminar, Birkhauser Verlag, Basel, 1992. [Gab04] O. Gabber, Notes on some t-structures, in Geometric aspects of Dwork theory. Vol. I, II, page 711 -- 734, Walter de Gruyter GmbH & Co. KG, Berlin, 2004. [Gro67] A. Grothendieck, ´El´ements de g´eom´etrie alg´ebrique. IV. ´Etude locale des sch´emas et des morphismes de sch´emas IV, Inst. Hautes ´Etudes Sci. Publ. Math. (32), 361 (1967). [Har66] R. Hartshorne, Residues and duality, Lecture notes of a seminar on the work of A. Grothendieck, given at Harvard 1963/64. With an appendix by P. Deligne. Lecture Notes in Mathematics, No. 20, Springer-Verlag, Berlin, 1966. [Har77] R. Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977, Graduate Texts in Mathematics, No. 52. [HH90] M. Hochster and C. Huneke, Tight Closure, Invariant Theory and the Briancon-Skoda Theorem, Journal of the American Mathematical Society 3, 31 -- 116 (1990). [HT04] N. Hara and S. Takagi, On a generalization of test ideals, Nagoya Math. J. 175, 59 -- 74 (2004). [Lam99] T. Y. Lam, Lectures on modules and rings, volume 189 of Graduate Texts in Mathe- matics, Springer-Verlag, New York, 1999. [Lip09] J. Lipman, Notes on derived functors and Grothendieck duality, in Foundations of Grothendieck duality for diagrams of schemes, volume 1960 of Lecture Notes in Math., pages 1 -- 259, Springer, Berlin, 2009. [Lyu97] G. Lyubeznik, F -modules: an application to local cohomology and D-modules in characteristic p > 0, Journal fur reine und angewandte Mathematik 491, 65 -- 130 (1997). 52 References [Mac98] S. MacLane, Categories for the Working Mathematician, Number 5 in Graduate Texts in Mathematics, Springer-Verlag, New York, 1998. [Man63] J. I. Manin, Theory of commutative formal groups over fields of finite characteristic, Uspehi Mat. Nauk 18(6 (114)), 3 -- 90 (1963). [Mat92] H. Matsumura, Commutative ring theory, Cambridge University Press, Cambridge, UK, 1992. [Miy91] J.-I. Miyachi, Localization of triangulated categories and derived categories, J. Algebra 141(2), 463 -- 483 (1991). [Nay09] S. Nayak, Compactification for essentially finite-type maps, Adv. Math. 222(2), 527 -- 546 (2009). [Nay12] S. Nayak, Letter to the authors, 2012. [Sch11] K. Schwede, Test ideals in non-Q-Gorenstein rings, Trans. Amer. Math. Soc. 363(11), 5925 -- 5941 (2011). [ST13] K. Schwede and K. Tucker, On the behavior of test ideals under finite morphisms, J. Algebraic Geometry (2013), arXiv:1003.4333, to appear. [Ver96] J.-L. Verdier, Des cat´egories d´eriv´ees des cat´egories ab´eliennes, Ast´erisque (239), xii+253 pp. (1996), With a preface by Luc Illusie, Edited and with a note by Georges Maltsiniotis. [Wei94] C. A. Weibel, An introduction to homological algebra, Cambridge University Press, Cambridge, 1994. [Zin84] T. Zink, Cartiertheorie kommutativer formaler Gruppen, volume 68 of Teubner- Texte zur Mathematik [Teubner Texts in Mathematics], BSB B. G. Teubner Verlags- gesellschaft, Leipzig, 1984, With English, French and Russian summaries. 53
1603.00566	1	1603	2016-03-02T03:38:39	Point Counting on Non-Hyperelliptic Genus 3 Curves with Automorphism Group $\mathbb{Z} / 2 \mathbb{Z}$ using Monsky-Washnitzer Cohomology	[ "math.AG" ]	We describe an algorithm to compute the zeta function of any non-hyperelliptic genus 3 plane curve $C$ over a finite field with automorphism group $G = \mathbb{Z} / 2 \mathbb{Z}$. This algorithm computes in the Monsky-Washnitzer cohomology of~the curve. Using the relation between the Monsky-Washnitzer cohomology of $C$ and its quotient $E := C/G$, the computation splits into 2 parts: one in a subspace of the Monsky-Washnitzer cohomology and a second which reduces to the point counting on an elliptic curve $E$. The former corresponds to the dimension $2$ abelian surface $\mathrm{ker}(\mathrm{Jac}(C) \rightarrow E)$, on which we can compute with lower precision and with matrices of smaller dimension. Hence we obtain a faster algorithm than working directly on the curve $C$.	math.AG	math	Point Counting on Non-Hyperelliptic Genus 3 Curves with Automorphism Group Z/2Z using Monsky-Washnitzer Cohomology Yih-Dar SHIEH Abstract. We describe an algorithm to compute the zeta function of any non- hyperelliptic genus 3 plane curve C over a finite field with automorphism group G = Z/2Z. This algorithm computes in the Monsky-Washnitzer cohomology of the curve. Using the relation between the Monsky-Washnitzer cohomology of C and its quotient E := C/G, the computation splits into 2 parts: one in a subspace of the Monsky-Washnitzer cohomology and a second which reduces to the point counting on an elliptic curve E. The former corresponds to the dimension 2 abelian surface ker(Jac(C) → E), on which we can compute with lower precision and with matrices of smaller dimension. Hence we obtain a faster algorithm than working directly on the curve C. 1. Introduction Henn gave the table of the possible non-trivial groups which appear as auto- morphism groups of a non-hyperelliptic genus 3 curves, which can be found in Ver- meulen's thesis [11]. The dimension of the set of moduli points of non-hyperelliptic genus 3 curves whose automorphism group contain Z/2Z is 4 inside the moduli of genus 3 curves M3 of dimension 6. We thus obtain an algorithm to compute the zeta function of a large family of genus 3 curves. In [5], Kedlaya used Monsky-Washnitzer cohomology to compute the zeta func- tions of hyperelliptic curves over finite fields. This method could be applied to general varieties, and there are already generalizations to superelliptic curves, Ca,b curves and non-degenerate curves, see [1], [2] and [4]. This work also uses Monsky- Washnitzer cohomology but focuses on a smaller dimensional space associated to an abelian surface in the jacobian of C. The remainder of this paper is organized as follows: Section 2 recalls the defi- nition of Monsky-Washnitzer cohomology and results in this theory. In Section 3, we compute a basis of the cohomology and describe a way to do the reduction of a differential form. Reduction means to write a differential form as a linear com- bination of the basis. In particular, we give an upper bound of the denominator after a differential form is reduced. This bound makes the algorithm practical since it establishes a finite precision bound for the computation. Section 4 describes a This work was supported by a project of the Agence Nationale de la Recherche, reference ANR-12-BS01-0010-01. 1 2 YIH-DAR SHIEH way to compute a lift of Frobenius and Section 5 explains why the computation splits into 2 eigenspaces. Finally, Section 6 gives the algorithm and an analysis of its complexity. 2. Monsky-Washnitzer Cohomology In this section, we recall the definition of Monsky-Washnitzer cohomology which is introduced by Monsky and Washnitzer in [7][8][9]. Monsky-Washnitzer cohomology is a p-adic cohomology theory defined for smooth affine varieties over finite fields. Let X be a smooth affine variety defined over a finite field k := Fq of characteristic p with coordinate ring A which is a finitely gen- erated k-algebra. In [3], Elkik showed that there exists a finitely generated smooth Zq-algebra A such that A/pA ∼= A, here Zq is the valuation ring of Qq, the degree n := logpq unramified extension of Qp. In general, A does not admit a lift of the Frobenius endomorphism F on A , but its p-adic completion A∞ does. However, the dimension of the de Rham cohomology n=0 pnxpn−1dx is not an is not in A∞, but each term of this sum is exact. The problem is that this differential form does not converge fast enough for its integral to converge as well. of A∞ may be too big. For example, if A = Zp[x], then P∞ exact differential form since P∞ n=0 xpn Monsky and Washnitzer work with a subalgebra A† of A∞ consisting of series which converge fast enough to solve the above problem. For A = Zq[x1, x2,··· , xd]/(f1, f2,··· , fr), the weak completion or dagger ring of A is A† := Zqhx1, x2,··· , xdi/(f1, f2,··· , fr) where Zqhx1, x2,··· , xdi is the subring of A∞ which consists of overconvergent power series (Xα with α := (α1,··· , αd), xα := xα1 valuation on Zq. aαxα ∈ Zq[[x1, x2,··· , xd]] (cid:12)(cid:12)(cid:12) d , α =Pd 1 ··· xαd lim inf α→∞ vp(aα) α > 0) i=1 αi and vp is the usual p-adic Definition 2.1. The Monsky-Washnitizer cohomology of X/Fq is the de Rham cohomology of A†⊗Zq Qq. More precisely, let D0(A†) := A†, D1(A†) be its universal module of differentials dxd(cid:19)  D1(A†) :=(cid:0)A† dx1 + ··· + A† dxd(cid:1) / Xj=1  and Di(A†) be the i-th exterior product of D1(A†). Let H i(A, Zq) be the i-th cohomology group of the complex A†(cid:18) ∂fi dx1 + ··· + ∂fi ∂xd ∂x1 r 0 −→ D0(A†) d3−→ ··· where di is the usual differentiation. Then the i-th Monsky-Washnitzer cohomology group of X (or of A) is H i(A, Zq) ⊗Zq Qq, which is denoted by H i MW (X/Fq) (or H i(A, Qq)). d0−→ D1(A†) d1−→ D2(A†) d2−→ D3(A†) The Monsky-Washnitzer cohomology has the following properties, see [10]. POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 3 Theorem 2.2. For a smooth finitely generated Fq-algebra A, we have (a) The map A 7→ H i(A, Qq) is well defined and functorial. (b) There exists a Zq-algebra homomorphism Fq : A† → A† which lifts the Frobenius endomorphism of A. Furthermore, any two lifts induce homo- topic maps on the complex Di(A†). Hence they induce the same map Fq,∗ : H i(A, Qq) → H i(A, Qq) on the Monsky-Washnitzer cohomology. The following Lefschetz fixed point formula allows us to compute the zeta function of X = Spec(A) using Monsky-Washnitzer cohomology. Theorem 2.3 (Lefschetz fixed point formula). Let X/Fq be a smooth affine variety of dimension d. Then we have X(Fq) = d Xi=0 q,∗(cid:12)(cid:12)(cid:12) (−1)iTr(cid:16)qdF −1 H i MW (X/Fq)(cid:17) . 3. Cohomology of Non-Hyperelliptic Genus 3 Plane Curves with Automorphism Group Z/2Z In this article, we consider non-hyperelliptic smooth projective plane curves C of genus 3 whose automorphism group contains Z/2Z over a finite field Fq of characteristic p 6= 2. Such curves can be written (up to isomorphism) as C : F := Y 4 + G(X, Z)X 2 + H(X, Z) = 0, with G(X, Z) and H(X, Z) ∈ Fq[X, Z] which are homogeneous of degree 2 and 4 respectly. We assume that C is smooth. Since the Monsky-Washnitzer cohomology is defined for smooth affine varieties, we also consider the affine part of C Caff : f := y4 + g(x)y2 + h(x) = 0, where g(x) = G(x, 1) and h(x) = H(x, 1) are the dehomogenizations of G(X, Z) and H(X, Z) with respect to Z. In this section, we compute the Monsky-Washnitzer cohomology H i MW (Caff /Fq) of Caff and relate the zeta function of C/Fq to the characteristic polynomial of the Frobenius action Fq,∗ on H i MW (Caff /Fq). Choose arbitrary lifts G(X, Z) and H(X, Z) ∈ Zq[X, Z] of G(X, Z) and H(X, Z) such that degX G = degX G and degX H = degX H. Let g(x) := G(x, 1) and h(x) := H(X, 1) be the dehomogenizations. Consider the following two curves C : F := Y 4 + G(X, Z)Y 2 + H(X, Z) = 0, and (3.1) Caff : f := y4 + g(x)y2 + h(x) = 0 Since the reduction of F modulo the maximal ideal (p) of Zq is F which defines a smooth projective curve C, the generic fiber Cξ := C ×Zq Qq of C is also smooth. Using the three facts that the reduction of f modulo the maximal ideal (p) of Zq equals to f which is not zero in Fq[x, y], that A := Fq[x, y]/(f (x, y)) is an integral domain and that p is a prime element in the unique factorization domain Zq[x, y], one sees that A is an integral domain and hence it is flat over Zq. This shows that A is a finitely generated smooth Zq-algebra, so we can work with A to apply the theory of Monsky-Washnitzer cohomology. The above arguments also show that the generic fiber Cξ of C is a geometrically integral smooth projective curve over Qq. 4 YIH-DAR SHIEH Although we can compute the Monsky-Washnitzer cohomology of the affine curve Caff /Fq by explicit reduction algorithms and the control of denominators, we use the following theorem instead, see [6], and compute the algebraic de Rham dR( Cξ,aff /Qq) of the curve Cξ,aff /Qq, the affine part of the generical cohomology H i fiber Cξ. Note that we are concerned with curves, hence the divisors are always normal crossings. Theorem 3.1. Let Y /Zq be a smooth proper scheme, Z be a relative normal crossings divisor and X := Y \Z is affine. Then there is a canonical isomorphism H i dR(Xξ/Qq) → H i MW (Xp/Fq), where Xξ is the generic fiber and Xp is the special fiber of X/Zq, namely, the fibers of X at the closed point (p) of Spec(Zq). 3.1. Geometry. Before we compute the algebraic de Rham cohomology of the affine curve Cξ,aff , we need to study its geometry. The coordinate ring of Cξ,aff is A := Qq[x, y]/(f (x, y)), here f (x, y) = y4 + g(x)y2 + h(x) ∈ Zq[x, y] with deg(g) ≤ 2 and deg(h) ≤ 4. We write g(x) and h(x) as following (3.2) g(x) = a2x2 + a1x + a0, ai ∈ Zq h(x) = b4x4 + b3x3 + b2x2 + b1x + b0, bi ∈ Zq There are four cases to consider: Case 1. b4 = 0 and a2 = 0 There is only one point at infinity which is P∞ := (1 : 0 : 0). Using the fact that Cξ is smooth at P∞, one shows that b3 6= 0. The dimension of the first algebraic de Rham cohomology of Cξ,aff /Qq is 2g + N∞ − 1 = 2 · 3 + 1 − 1 = 6, here g is the genus of Cξ, which equals to the genus of C, and N∞ is the number of points at infinity. We have υP∞ (x) = −4 and υP∞ (y) = −3. The local parameter at P∞ is t := b3x2/y3. The expansions of x and y as Laurent series of the local parameter t are x = −b3/t4 + ··· and y = b3/t3 + ··· . Case 2. b4 = 0 and a2 6= 0 There are 3 points at infinity: P∞ := (1 : 0 : 0) and P∞,± := (1 : ±α : 0) with α2 = −a2. Using the fact that Cξ is smooth at P∞, one shows that b3 6= 0. (The condition a2 6= 0 implies the smoothness at P∞,+ and P∞,−.) We have dR( Cξ,aff /Qq) = 2 · 3 + 3 − 1 = 8, υP∞,±(x) = υP∞,±(y) = −1, υP∞ (x) = −2 dimQq H 1 and υP∞ (y) = −1. The local parameters at P∞ and P∞,± are t := 1/y and t± := 1/x. The expansions of x and y at P∞ and P∞,± as Laurent series of the local parameters are x = β/t2 + γ + δt2 + ··· with β = −a2/b3, y = 1/t, and x = 1/t±, y = ±α/t + ··· . Case 3. b4 6= 0 and a2 There are 2 points at infinity: P∞,± := (1 : ±α : 0) with α = (−a2/2)1/2. Using the fact that Cξ is smooth at P∞,+, one shows that a1a2 − 2b3 6= 0. We have dR( Cξ,aff /Qq) = 2· 3 + 2− 1 = 7 and υP∞,±(x) = υP∞,±(y) = −2. The local dimQq H 1 parameters are t± := y/x ∓ α. The expansions of x and y as Laurent series of the local parameters are x = β/t2 ± + (β ± αγ)/t± + ··· with β = −(a1a2 − 2b3)/4a2 and γ = α(a1a2 + 2b3)/2a2 2. Case 4. b4 6= 0 and a2 There are 4 points at infinity which are P∞,i,± := (1 : ±αi : 0) with ±α1 and ± + γ/t± + ··· and y = ±αβ/t2 2 − 4b4 6= 0 2 − 4b4 = 0 POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 5 dR( Cξ,aff /Qq) = ±α2 are the four roots of y4 + a2y2 + b4 = 0. We have dimQq H 1 2·3+4−1 = 9 and υP∞,i,±(x) = υP∞,i,±(y) = −1. The local parameters are t := 1/x. The expansions of x and y as Laurent series of t are x = 1/t and y = ±αi/t + ··· . In order to analyze the control of denominators later, we need to impose further assumptions on the choice of the lift F (X, Y, Z). Assumption 3.2. The coefficients ai and bj of g(x) and h(x) in (3.2) are either 0 or units in Zq. Furthermore, a2 2 − 4b4 is either 0 or a unit in Zq. Remark 3.3. A lift which satisfies the above assumptions could be constructed by using Teichmuller lift. The choice of such a lift is to preserve the geometric structure. Under these assumptions, we introduce some facts which will be used later. The expansions of x and y as Laurent series of the local parameters have integral coefficients.1 This means that x and y are in O((t)), here O is the integral closure of Zq in a finite extension Qq(α) (Qq(α1, α2) in Case 4) of Qq and α is the Y -coordinate of the points at infinity discussed above. Furthermore, the coefficients of the lowest terms 2 in these Laurent series are units in O. (In Case 3, one shows that a1a2 − 2b3 is a unit in Zq by using C is smooth.) Cξ and C have the same geometry. In Case 4, pa2 3.2. The Reduction Algorithm and Algebraic de Rham Cohomol- ogy. In this subsection, we present the reduction algorithm and use it to compute dR( Cξ,aff /Qq). First of all, since (i + 1)xjyidy = d(xj yi+1) − jxj−1yi+1dx for H 1 all i ≥ 0, the universal module of differential Ω1 of Cξ,aff /Qq is generated by {xjyidx i ≥ 0, j ≥ 0}. From the defining equation (3.1), Ω1 is generated by {xjyidx 1 ≤ i ≤ 3, j ≥ 0}. Since 0 = df = d(cid:0)y4 + g(x)y2 + h(x)(cid:1) =(cid:0)g′(x)y2 + h′(x)(cid:1) dx +(cid:0)4y3 + 2g(x)y(cid:1) dy, 2 − 4b4 is a unit in O. we have Combine with the following equation 2 2 l + 4 l + 2 yl+4 + kxk−1yl+4 + g(x)yl+2(cid:19)(cid:19) − xk(cid:0)4y3 + 2g(x)y(cid:1) yldy xk(cid:0)g′(x)y2 + h′(x)(cid:1) yldx + xk(cid:0)4y3 + 2g(x)y(cid:1) yldy = 0. d(cid:18)xk(cid:18) 4 =(cid:18) 4 xk(cid:18) l = d(Sl,k), where Sl,k := −xk(cid:18) 4 l + 2(cid:0)kxk−1g(x) + xkg′(x)(cid:1) yl+2(cid:19) dx, g′(x)y2 + h′(x)(cid:19) yldx − kxk−1(cid:18) 4 g(x)yl+2(cid:19) l + 4 2 y4 + yl+4 + l + 4 l + 2 l + 4 l + 2 2 l + 2 g(x)y2(cid:19) yldx one gets Using y4 = −(g(x)y2 + h(x)) and the explicit description of g(x) and h(x) in (3.2), the above equation becomes (3.3) Γk,l,0,j xk+j yl + 3 Xj=−1 1 Xj=−1   Γk,l,2,j xk+j yl+2  dx = d(Sl,k), 1Use Hensel's lemma. For Case 1, one needs a 2-variable version of Hensel's lemma. 2 In particular, α, β ∈ O∗ if we use the notations in the above classification. 0 · · 0 0 · · 0 0 · · 0 0 Γl,k,0,−1 Γl,k,0,0 Γl,k,0,1 Γl,k,0,2 Γl,k,0,3 0 0 Γl,k,2,−1 Γl,k,2,0 Γl,k,2,1 0 · · 0 0 · · 0 0 · · 0 0 0 0 0 0 0 0     6 YIH-DAR SHIEH here the coefficients Γk,l,0,j and Γk,l,2,j are defined as following Γk,l,0,j := (j + 1 + 4k l + 4 ) bj+1, Γk,l,2,j := l l + 2 (j + 1 + 2k l + 4 )aj+1. In order to make things more clear, we use the following notation: Definition 3.4. A family of matrices Mk of size m× n with entries (Mk)i,j = mk,i,j ∈ Qq is called a family of reduction matrices if mk,i,j xk+j−3yi−1 m n Xi=1 Xj=1    dx ≡ 0 in Ω1, for all k. For reduction matrices Mk, we define Mk dx to be m n Xi=1 Xj=1   mk,i,j xk+j−3yi−1  dx. For example, from (3.3), we have reduction matrices of size (l + 3) × 6 which has non-zero entries only at the (l + 1)-th and (l + 3)-th rows M 0 l,k := l,kdx = d(Sl,k). The superscript 0 that appears in M 0 and M 0 obtained from (3.3), without further reduction. l,k means that it is We have to consider l = 1, 2 and 3, which give the reductions of xk+3y2dx or xk+2y2dx: l = 1. As mentioned above, we have reduction matrices of size 4 × 6 0 0 0 0 0 0 0 0 1 2ka0 M1,k = (2k + 5)a1 (12k + 60)b4 (12k + 45)b3 (12k + 30)b2 (12k + 15)b1 0 0 12kb0 0 0 15 ·    From 3.1, we know that one of b4 and b3 is non-zero. Hence xk+3ydx or xk+2ydx can be reduced to a linear combination of {xjyidx} with i = 1 or 3 and j ≤ k + 2 (if b4 6= 0) or j ≤ k + 1 (if b4 = 0), which have smaller degree in x. l = 2. From (3.2), we have reduction matrices of size 5 × 6, so y4 is involved. Using the defining equation (3.1) to reduce the degree in y, one gets a reduction matrix M2,k whose transpose is (2k + 10)a2 0 0 0 0 M t 2,k = − 1 6 · 0 0 0 0 0 0 0 0 0 0 0 0   (k + 3)(a2 k(a2 0 0 0 − 4b0) 0 0 1 + 2a0a2 − 4b2) 0 0 0 (2k + 3)(a0a1 − 2b1) (2k + 9)(a1a2 − 2b3) (k + 6)(a2 2 − 4b4)   POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 7 This gives us the reductions of xk+3y2dx or xk+2y2dx depending on the nullity of a2 2 − 4b4. l = 3. As at the start of l = 2 (dealing with y5), one gets M 1 3,k whose transpose is 1 35 · −   0 6ka0b0 6k(a1b0 + a0b1) + 21a1b0 0 0 0 0 6k(a2b0 + a1b1 + a0b2) + 21a1b1 + 42a2b0 0 6k(a2b1 + a1b2 + a0b3) + 21a1b2 + 42a2b1 0 6k(a2b2 + a1b3 + a0b4) + 21a1b3 + 42a2b2 0 0 (6k + 21)(a2b3 + a1b4) + 21a2b3 (6k + 42)a2b4 0 0 0 0 0 0 0 0 (2k + 7)(3a2 k(6a2 0 0 − 20b0) 1 + 6a0a2 − 10b2) (4k + 7)(3a0a1 − 5b1) (4k + 21)(3a1a2 − 5b3) (2k + 14)(3a2 2 − 10b4) 0 0   Since we want to reduce xj y3dx to those with smaller degree in x, M 1 3,k is not suitable since it has (possible) non-zero entries which correspond to xk+5y1, xk+4y1 and xk+3y1. We use l = 1 to reduce xk+3ydx to {xj1 ydx, xj2 y3dx j1 ≤ k + 2, j2 ≤ k + 1} if b4 6= 0, or reduce xk+2ydx to {xj1 ydx, xj2 y3dx j1 ≤ k + 1, j2 ≤ k + 1} ) if b4 = 0. Then use this result to reduce xk+3ydx (if b4 = 0), xk+4ydx and xk+5ydx (if b4 6= 0) iteratively to {xj1 ydx, xj2 y3dx j1 ≤ k + 2, j2 ≤ k + 3}, depending the nullity of b4. Finally, use these reductions in M 1 Case 3 and Case 4: b4 6= 0 3,k, we get reduction matrices in ′ M 0 0 0 0 0 0 ∗ ∗ 0 0 ∗ ∗ 0 0 ∗ ∗2,5 0 0 ∗ ∗4,5 3,k = c ·  4(k + 4)(k + 5)(k + 6)(cid:0)(8k + 44)a1a2b4 − a2 with c = −1/(cid:0)2688(k + 4)(k + 5)(k + 6)b3 ∗4,5 = 96b2 which equals to 384b3 4(cid:1) and 2 − 4b4)   2b3 − (16k + 84)b3b4(cid:1) 4(k + 4)(k + 5)(k + 6)(2k + 11)(a1a2 − 2b3) if a2 384(k + 4)(k + 5)(k + 6)(k + 7)b3 4(a2 0 0 0 2 − 4b4 = 0, and ∗2,5 = 36(k + 4)b4 (cid:18)(6k2 + 136k + 285)a2b3 3 − 8(6k2 + 56k + 125)a2b2b3 + 48(k2 + 11k + 30)a2b1b2 4 + 16(k2 + 6k)a0b3b2 4 − 2(8k2 + 58k + 105)a1b2 4(cid:19) Notice that ∗2,5 may be non-zero, but it corresponds to xk+2y and since b4 6= 0, one can use M1,k−1 in the case l = 1 to reduce xk+2ydx and get a new reduction matrices 3b4 + 16(2k2 + 17k + 35)a1b2b2 M3,k = c ·  0 ∗ 0 ∗ 0 0 ∗ ∗ 0 0 ∗ ∗ 0 0 ∗ 0 0 0 ∗ ∗4,5 0 0 0 384(k + 4)(k + 5)(k + 6)(k + 7)b3 4(a2 2 − 4b4)   Notice that the reduction of xk+2ydx using M1,k−1 only involve xjydx with k− 2 ≤ j ≤ k+1 and xj y3 with k−2 ≤ j ≤ k, the last two columns of M3,k and M 2 3,k are the same except the (2, 5)-entry, and M3,k satisfies the condition in Definition 3.4, so they are indeed reduction matrices. The reduction involves division by 12(k + 4)b4, hence c−1M3,k has integral coefficients. 8 YIH-DAR SHIEH Case 1 and Case 2: b4 = 0 M3,k = c ·  0 0 0 0 0 ∗ 0 ∗ 0 0 ∗ ∗ 0 0 ∗ ∗ 0 0 0 ∗4,5 0 0 0 2(k + 7)(2k + 11)(4k + 15)(4k + 19)b2 3a2 2   with c = −1/(cid:0)7(4k + 15)(4k + 19)(4k + 23)b2 ∗4,5 = (4k + 15)b3 (cid:18)(32k3 + 504k2 + 2648k + 4641)a1a2b3 − (4k2 + 52k + 168)a2 3(cid:1) and 2b2 3(cid:19) − (64k3 + 1008k2 + 5276k + 9177)b2 which equals to −(4k + 15)(4k + 19)(4k + 21)(4k + 23)b3 Cξ,aff /Qq. 3 if a2 = 0. Now we can compute the algebraic de Rham cohomology H 1 dR( Cξ,aff /Qq) of Proposition 3.5. The algebraic de Rham cohomology H 1 dR( Cξ,aff /Qq) has a basis (1) {ydx, y2dx, y3dx, xydx, xy2dx, xy3dx}, if b4 = 0 and a2 = 0. (2) {ydx, y2dx, y3dx, xydx, xy2dx, xy3dx, x2y2dx, x2y3dx}, if b4 = 0 and (3) {ydx, y2dx, y3dx, xydx, xy2dx, xy3dx, x2ydx}, if b4 6= 0 and a2 2 − 4b4 = 0. (4) {ydx, y2dx, y3dx, xydx, xy2dx, xy3dx, x2ydx, x2y2dx, x2y3dx}, if b4 6= 0 and a2 6= 0. a2 2 − 4b4 6= 0. Proof. We give the proof for (c). For other parts, the proofs are all similar. Suppose b4 6= 0 and a2 2− 4b4 = 0. The reduction matrices M1,k shows that xk+3ydx is a linear combination of {xj1 ydx, xj2 y3dx k − 1 ≤ j1 ≤ k + 2, k − 1 ≤ j2 ≤ k + 1}, since b4 6= 0. So each xj ydx with j ≥ 3 can be reduced. The (3, 6)- entry of the reduction matrix M2,k is (k + 6)(a2 2 − 4b4) = 0, but its (3, 5)-entry is (2k + 9)(a1a2 − 2b3) which is non-zero by Case 3 in Section 3.1. So xk+2y2dx is a linear combination of {xjy2 k − 1 ≤ j ≤ k + 1} and hence each xjy2dx with j ≥ 2 can be reduced. The reduction matrix M3,k in Case 3 has (k + 7)(a2 2 − 4b4) = 0 at the (4, 6)-entry, but its (4, 5)-entry is ∗4,5 = (2k + 11)(a1a2− 2b3) 6= 0. So xk+2y3dx is a linear combination of {xj1 ydx, xj2 y3dx k − 2 ≤ j1 ≤ k + 1, k− 2 ≤ j2 ≤ k + 1} and hence each xjy3 with j ≥ 2 can be reduced. This completes the proof for (c). (cid:3) The following table give a more clear description of these basis. y y2 dx 1 x x2 • × • × • × (1.a) Case 1 • • • y3 y y2 dx 1 x x2 • × • • • • (1.b) Case 2 • • • y3 y y2 dx 1 x x2 • • • × • × (1.c) Case 3 • • • y3 y y2 dx 1 x x2 • • • (1.d) Case 4 • • • • • • y3 Table 1. basis of H 1 dR( Cξ,aff /Qq) POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 9 3.3. Control of the denominators in the reduction algorithm and Monsky-Washnitzer cohomology. The reduction algorithm in subsection 3.2 dR( Cξ,aff /Qq). By Theorem 3.1, this basis also allows us to obtain a basis of H 1 forms a basis of the Monsky-Washnitzer cohomology H 1 MW (Caff /Fq). One can also prove this by the following upper bound on the denominators that appear during the reduction process. This bound provides the precision necessary for our algorithm. Before stating the main result of this subsection, we fix some notations. For a local parameter t at a point at infinity P∞, we write the Laurent series expansion of x, y and xj yi with respect to t as following: (3.4) x = ∞ Xs=υp(x) δ0,1 s ts, y = ∞ Xs=υp(y) δ1,0 s ts and xjyi = ∞ Xs=υp(xj yi) δi,j s ts. If a subscript is used to denote a local parameter at some point, we use this subscript in the coefficients of the above expansion. For example, in Case 3, we write xjyi = − at P∞,−. Recall that all the coefficents s,− ts s,+ ts P∞ s δi,j δi,j s are in O in all cases that we are concerned, see Remark 3.3. + at P∞,+ and xjyi =P∞ s δi,j Proposition 3.6. Write (3.5) xkyldx = 3 2 Xi=1 Xj=0 ai,jxjyidx + dS, where S = 3 Xi=0Xj≥0 bi,jxjyi with ai,j and bi,j ∈ Qq, ai,j = 0 if xjyi is not in the basis in Proposition 3.5, 1 ≤ l ≤ 3 and k ∈ N. Then (a) One can choose S with the property that bi,j = 0 if i − l 6= 0 (mod 2). (b) For any S in (3.5) which satisfies the property in (a), we have bi,j−i = 0 for all 0 ≤ i ≤ 3 and j ≥ k +5. Furthermore pmbi,j−i ∈ O for all 0 ≤ i ≤ 3 and j ≥ 7, where m = ⌊logp(4k + 8)⌋. (c) pm+∆+1ai,j ∈ Zq, where ∆ := 11(cid:0)⌊logp(63)⌋ + τ (p)(cid:1) with τ (3) = 5, τ (5) = 3, τ (p) = 1 for p = 7, 11, 13 and τ (p) = 0 if p > 13. Proof. (a) Using the automorphism y → −y, or by a direct analysis on the reduction process discussed in the previous subsection. (b) We prove this for Case 4, proofs for other cases are all similar. From the expansions in Case 4, one obtains (3.6) xj yi = ∞ Xs=−(i+j) δi,j s,µ,±ts µ,± with δi,j bi,j = 0 for all j > M . Hence −(i+j),µ,± = (±αµ)i and µ = 1, 2. There is an integer M > 0 such that bi,jxjyi = 3 M Xi=0 Xj≥0 bi,j−iδi,j−i −j,µ,± + S = = 3 M M+3 Xj≥0 Xi=0  Xi=0 Xj=0  3 ∞ bi,j Xs=−(i+j) Xj′>j 3 Xi=0 M+3 δi,j s,µ,±ts µ,± bi,j′−iδi,j′−i −j,µ,±!  t−j µ,± 10 YIH-DAR SHIEH Since υp∞,µ,±(ai,j xjyidx) ≥ 7 and υp∞,µ,±(xkyldx) ≥ −(k + 5) and the expansions of xkyldx have integral coefficients, we have (3.7) 3 j · Xi=0  bi,j−iδi,j−i −j,µ,± + M+3 Xj′>j 3 Xi=0 bi,j′−iδi,j′−i −j,µ,±!  ∈ O for all j ≥ 7 and it is zero if j ≥ k + 5. Combine (3.7) with the property in (a) and the fact that α1 and α2 are units in O (Fact 3.3), we get Xi=0 ∗  ∈ O for i = 0, 1, j ≥ 7 and µ = 1, 2, and it is zero if j ≥ k + 5. Since α2 here D = a2 j · 1 · bi,j−i + α2 µ · bi+2,j−i−2 + 2 = √D, 1 − α2 Xj′>j M+3 3 2 − 4b4 6= 0, one obtains (3.8) j · 1 · bi,j−i + √D · bi+2,j−i−2 + M+3 3 Xj′>j Xi=0 ∗  ∈ O for i = 0, 1, j ≥ 7, and it is zero if j ≥ k + 5, here ∗ involves only bi,j′−i with j′ > j and elements in O. Remember that √D ∈ O∗. Apply j = M + 3 to (3.8), we know that bi,M+3−i = 0 for all 0 ≤ i ≤ 3. Repeat the same argument, one shows that bi,j−i = 0 for all 0 ≤ i ≤ 3 and j ≥ k + 5. Now apply j = k + 4 to (3.8), we get pmbi,k+4−i ∈ O for all 0 ≤ i ≤ 3. Repeat the same argument and notice that the terms ∗ in (3.8) are in O in each step (since all the δ•,• and bi,j′−i ∈ O if j′ > j in each step),one proves that pmbi,j−i ∈ O for all 0 ≤ i ≤ 3 and j ≥ 7. So pmbi,j−i ∈ O ∩ Qq = Zq for all 0 ≤ i ≤ 3 and j ≥ 7. (c) Consider s (3.9a) (3.9b) 3 k+5 ω := pm xkyldx − d Xi=0  = pm Xi=1  Xj≥7−i ai,jxjyidx + d Xi=0  bi,jxjyi    bi,jxjyi Xj=0  Xj=0 6−i 3 2 3   From (b) and (3.9a), one knows that ω has integral coefficients, so we can choose ϕ1(x, y) and ψ1(x, y) in Zq[x, y] such that ω = ψ1 dx + ϕ1 dy. On the other hand, from (3.9b), one know that ω = ψ2 dx + ϕ2 dy for some ϕ2 and ψ2 in Qq[x, y] with deg(ϕ2) ≤ 5, deg(ψ2) ≤ 5. Consider (3.10) fy ω = fy (ψi dx + ϕi dy) = (ψify − ϕifx) dx fx ω = fx (ψi dx + ϕi dy) = (ϕifx − ψify) dy (using fx dx + fy dy = df = 0). Let λi(x, y) := ψify − ϕifx. It is clear that λ1 ∈ Zq[x, y] and λ2 ∈ Qq[x, y] with deg(λ2) ≤ 8. Using the defining equation f to reduce the degree of y in λi, we get fy ω = λi dx and fx ω = −λi dy with λ1 ∈ Zq[x, y], degy( λ1) ≤ 3, λ2 ∈ Qq[x, y], deg( λ2) ≤ 8 and degy( λ2) ≤ 3. Since ( λ1 − λ2)dx = fy ω − fy ω = 0 and degy( λ1 − λ2) ≤ 3, we have λ1 = λ2. This means that fy ω = λ dx and fx ω = −λ dy with λ := λ1 = λ2 which is in Zq[x, y] of deg(λ) ≤ 8. By Corollary 3.9, there exist α and β in Zq[x, y] with deg(α) ≤ 5 and POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 11 i=1P2 deg(β) ≤ 5 such that αfy + βfx = 1 in A. So ω = (αfy + βfx) ω = (αλ)dx− (βλ)dy. Notice that ω ≡ P3 j=0 pmai,j xjyidx, we can use the reduction of (αλ)dx − (βλ)dy to get the denominators of pmai,j. Since deg(αλ) ≤ 13 and deg(βλ) ≤ 13, we need only to know the denominators of the final reductions of xjyidx and xjyidy with 0 ≤ i + j ≤ 13. Using xjyidy ≡ −j/(i + 1)xj−1yi+1dx, the defining equation f , and xjdx ≡ 0, we only need to consider the reductions of xj yidx with 1 ≤ i ≤ 3 and i + j ≤ 13, but if p = 3, the extra denominator 3 should be counted. The reduction of xkyldx (1 ≤ l ≤ 3) using the reduction matrices Mi,j in subsection 3.2 involve divisions by some of the following: 12(k + 2)b4, (12k + 21)b3, (k + 3)(a2 2 − 4b4), 384(k + 2)(k + 3)(k + 4)(2k + 7)b3 3a2 2, −(4k + 7)(4k + 11)(4k + 13)(4k + 15)b3 3, depending on each case, and the numbers 2, a2, b4, b3, a2 2 − 4b4 and a1a2 − 2b3 that we need to consider (depending on each case) are units of Zq. So in each step, we get extra denominators which are at most 2 − 4b4), (2k + 5)(a1a2 − 2b3), 384(k + 1)(k + 2)(k + 3)(k + 4)b3 4(a1a2 − 2b3), 2(k + 4)(2k + 5)(4k + 3)(4k + 7)b2 4(a2 p⌊logp(4j+15)⌋+τ (p), here τ (3) = 5, τ (5) = 3, τ (p) = 1 for p = 7, 1, 13 and τ (p) = 0 if p > 13. Since we are concerned with 2 ≤ j ≤ 12, we need at most 11 reduction steps, so the denominators of the reductions of xjyidx with 1 ≤ i ≤ 3 and i + j ≤ 13 are at most p11·(⌊logp(63)⌋+τ (p)). Hence pm+∆+1ai,j ∈ Zq. (cid:3) Remark 3.7. Proposition 3.6 gives an upper bound for the denominators after a differential form (with integral coefficients) is reduced to the linear combination of the basis we found in Propsition 3.5. Along with the rate of convergence of the Frobenius Fp (see Corollary 4.3), one can determine how much p-adic precision we need to work with (and determine an integer N3 such that one can work with modulo xN3), see Section 6. But one needs an upper bound for all the denominators that will appear during the computation (in the reduction step) in order to know how much precision of the reduction matrices Mi,j are required and to have an analysis of the bit complexity. It turns out that one has a similar bound as in Proposition 3.6. The proof is completely similar. Theorem 3.8. Let R be a field or a discrete valuation ring and m be the maximal ideal of R. Let f0, . . . , fn ∈ R[x1, . . . , xn] with degfi = di and define ρ = d0 + . . . + dn − n − 1. Denote the homogenization of fi by f h i point in Pn(R/m) satisfies f h g0, . . . , gn ∈ R[x1, . . . , xn] with deg gi ≤ ρ + 1 − di for i = 0, . . . , n such that for i = 0, . . . , n. Assume that there is no n = 0. Then there exist polynomials 1 = . . . = f h 0 = f h n gifi = 1. Xi=0 Proof. This appears as Theorem 2 in Denef-Vercauteren [2]. (cid:3) Corollary 3.9. There exist α and β in Zq[x, y] with deg(α) ≤ 5 and deg(β) ≤ 5 such that αfy + βfx = 1 in A. Furthermore, one can find such α and β such that α has only odd degrees in y and β has only even degrees in y. 12 YIH-DAR SHIEH Proof. Apply 3.8 to R = Zq, f0 = f , f1 = fy and f2 = fx. Apply 3.8 to R = Zq, f0 = f , f1 = fy and f2 = fx. If α and β don't satisfy the last property, consider the equality α(x,−y)fy(x,−y) + β(x,−y)fx(x,−y) = 1 in A. From (3.1), it is clear that fy(x,−y) = −fy(x, y) and fx(x,−y) = fx(x, y). We thus have (cid:18) α(x, y) − α(x,−y) 2 (cid:19) fy +(cid:18) β(x, y) + β(x,−y) 2 (cid:19) fx = 1, which completes the proof. (cid:3) 4. Lift of Frobenius In this section, we describe a lift Fp of the absolute Frobeninus endomorphism F p : a → ap on the coordinate ring A of C aff to A†. This means that Fp is a Zp- algebra endomorphism on A† such that π ◦ Fp = F p ◦ π, where π is the reduction modulo p. The lift Fq of the q-th Frobenius endomorphism of A is F n p , hence one can work with Fp for the purpose of computation. Denote by σ the p-th power Frobenius endomorphism on Fp and also its lift on Zp. Any lift Fp satisfies Fp(x) ≡ xp mod p, Fp(y) ≡ yp mod p, Fp(f (x, y)) = 0. From Corollary 3.9, we know that there exist α and β in Zq[x, y] such that αfy+ βfx = 1. Define δy := αp, δx := βp and consider the equation G(Z) := Fp(f (x, y)) = f σ(xp + δxZ, yp + δyZ) = 0 (4.1) in A†[Z]. Then G(0) = f σ(xp, yp) ≡ f σ(xσ, yσ) = 0 mod p. Also G′(0) = f σ y (xp, yp)δy + f σ y αp + x βp = (fyα + fxβ)p = 1 mod p. Hence by Hensel's lemma, there is a unique so- f p lution of (4.1) in A∞. Use the following proposition and its corollary, this solution is in fact in A†. In fact, Corollary 4.3 below gives an explicit lower bound on the rate of convergence, which allows us (together with Proposition 3.5) to work with a finite and explicit p-adic precision. x (xσ, yσ)δx ≡ f p y δy + f p x δx = f p x (xp, yp)δx ≡ f σ y (xσ, yσ)δy + f σ Lemma 4.1. Let H(Z) = P hk(x)Z k ∈ Zq[x][Z] and ∆1,k = dk := deg(hk). Assume h0(x) ≡ 0 mod p and h1(x) ≡ 1 mod p. Let 0 ≤ ∆n,0 ≤ ∆n,1 ≤ ··· ≤ ∆n,j ≤ ··· (n ≥ 1, j ≥ 0) with δj = ∆1,j be integers stastifies the following conditions: 1. ∆n+1,j ≥ max{∆n,j−l + δl 0 ≤ l ≤ j} for all n ≥ 0, j ≥ 0. 2. δ0 ≥ d0. 3. δj − δj−1 ≥ d1 for all j ≥ 1. 4. δk−1+j ≥ ∆k,j + dk for all k ≥ 2, j ≥ 0. Then the unique solution α = P∞ 2 if i ≥ δj + 1. One can always find such ∆n,j. Proof. Let T := {P∞ i=0 aixi ∈ Zqhxi vp(ai) ≥ j + 2 if i ≥ δj + 1} be a subset of Zqhxi, here −1 = δ−1 < 0 ≤ δ0 ≤ δ1 ≤ ··· ≤ δj ≤ ··· are integers which we will determine for which T satisfies some properties that are used in the proof. For each n, one can write T n = {P∞ i=0 aixi ∈ Zqhxi vp(ai) ≥ n + j + 1 if i ≥ ∆n,j + 1} for some −1 = ∆n,−1 < 0 ≤ ∆n,0 ≤ ∆n,1 ≤ ··· ≤ ∆n,j ≤ ··· . A sufficient condition i=0 aixi ∈ Zqhxi has the property: vp(ai) ≥ j + POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 13 for T n ⊂ T is ∆n+1,j ≥ max{∆n,j−l + δl 0 ≤ l ≤ j} for all n. Notice that T is closed under addition. In Zqhxi, we can use Newton method: αi+1 = αi − H(αi) H ′(αi) = αi − H(αi) 1 + (1 − H ′(αi))k! . ∞ Xk=1 k=2 hk(x)αk Our goal is to determine a condition on δj for which the result in each iteration above is in T . We use induction: assume αi ∈ T and to prove αi+1 ∈ T . It is sufficient to show: H(αi) ∈ T and (1 − H ′(αi)) T ⊂ T . H(αi) = h0(x) + h1(x)αi + P∞ i . Since vp(h0(x)) ≥ 1, h0(x) ∈ T if d0 ≤ δ0. Since h1(x)αi = αi− (1 − h1(x)) αi, αi ∈ T and vp (1 − h1(x)) ≥ 1, h1(x)αi ∈ T if δj − δj−1 ≥ d1 for all j ≥ 1. Similarly for k ≥ 2, since αk i ∈ T k, hk(x)αk i ∈ T if ∆k,j + dk ≤ δk−1+j. In fact, the above conditions imply (1 − h1(x)) T ⊂ T and hk(x)T k ⊂ T . Using this fact with (1 − H ′(αi)) T = (cid:0)(1 − h1(x)) +P∞ (cid:1) T and αk−1 ∈ T k−1, we know (1 − H ′(αi)) T ⊂ T . Hence αi+1 ∈ T and this implies the solution α ∈ T . For the existence ∆n,j, notice that the conditions 2, 3 and 4 are equivalent to: ∆1,n+1 ≥ max{∆1,n + d1, ∆k,n+2−k + dk 2 ≤ k ≤ n + 2} for all n ≥ 0. Suppose one has determined ∆n′,j′ for all n′ + j′ ≤ n + 1. Use condition 1, one can determine ∆k,n+2−k for all 2 ≤ k ≤ n + 2 (i.e for ∆n′,j′ with n′ + j′ ≤ n + 2 and n′ ≥ 2). Finally, one determines ∆1,n+1. Therefore, one determines all the ∆n′,j′ with n′ + j′ ≤ n + 2. This shows that one can find ∆n,j recursively. k=2 −khk(x)αk−1 (cid:3) i i Lemma 4.2. Suppose deg(hk(x)) ≤ (k + 1) d in Lemma 4.1. Then ∆i,j := (i + 4j) d satisfy the conditions in Lemma 4.1 and all the inequalities are equalities. In particular, δj = (4j + 1) d. Proof. This follows by induction in i and j. (cid:3) Corollary 4.3. There exists a lift Fp of the absolute Frobeninus endomor- phism a → ap on the coordinate ring A of Caff to A† such that Fp(x) = xp + δxZ0 and Fp(y) = yp + δyZ0 with Z0 = Pi,j ai,jxjyi, ai,j ∈ Zq and ordp(ai,j) > i+j 16p . Also the coefficient of xjyi in Fp(y) and Fp(x) has p-adic order > i+j 16p if i + j 6= p. Finally, Fp(xkyldx) =P3 16p − 4. i=1Pj bi,jxj yidx with ordp(bi,j) > i+j Proof. Using Corollary 3.9 and equation (4.1), one can apply d = 4p to (cid:3) Lemma 4.2. Theorem 4.4. There exists a lift of Frobenius Fp on A† which commutes with the involution τ : y → −y and has the rate of convergence in Corollary 4.3. Proof. We choose α and β such that α has only odd degrees in y and β has only even degrees in y as in Corollary 3.9. Since δy = αp and δx = βp, they have the same property as α and β. For solving G(Z) = 0 by Newton's method, we use Lemma 4.1 with H = G = f σ(xp + δxZ, yp + δyZ). It is clear that G has only even degrees in y, hence so does the solution Z0. From this, it is clear that the lift of Frobenius Fp : A† → A† commutes with the involution τ . (cid:3) 5. Quotient by Automorphism We have study H 1 MW (Caff /Fq). In this section, we consider the quotient of C by the automorphism τ : Y → −Y . We denote the quotient map by π : C → E := 14 YIH-DAR SHIEH π C/hτi. One can show that C/hτi has genus 1 either by Riemann-Hurwitz genus formula or from the affine equation directly, using the fact that Caff is stable under τ and Caff /hτi is smooth, hence the notation E is justified. The affine part Eaff of E is Caff /hτi, which has the defining equation: v2 + g(u)v + h(u) = 0. We have π−→ Eaff , (x, y) → (x, y2), and the corresponding map on the coordinate ring is Caff π∗ : u → x, v → y2. MW (Eaff /Fq), the induced map π∗ : Our goal is to study the followings: H 1 H i MW (Eaff /Fq) → H i MW (Caff /Fq) and its interplay with Frobenius endomorphism. −→ Eaff , (x, y) → (x, y2) lifts π, here Eaff is the lift of Eaff , whose defining Since Caff equation is v2 + g(u)v + h(u) = 0, we can study H i MW (Caff /Fq) dR( Caff,ξ/Qq). For i 6= 1, these are isomorphisms. For by H i i = 1, since {π∗(ujvdu) = xj y2dx} 0 ≤ j ≤ 1 (resp. 0 ≤ j ≤ 2)} are linear in Case 2 and Case 4), one sees that independent in Case 1 and Case 3 (resp. {ujvdu 0 ≤ j ≤ 1 (resp. 0 ≤ j ≤ 2)} are linear independent. Let δE be the number of points at infinity of E. We have δE = 1, 2, 1, 2 in each case, hence dR( Eaff,ξ/Qq) = 2·gE − 1 + δE = 2, 3, 2, 3. This shows that {ujvdu 0 ≤ j ≤ dimZq H 1 dR( Eaff,ξ/Qq) which is isomorphic via π∗ to the 1 (resp. 0 ≤ j ≤ 2)} is the basis of H 1 dR( Caff,ξ/Qq) generated by {xjy2dx 0 ≤ j ≤ 1 (resp. 0 ≤ j ≤ 2)}. subsapce V of H 1 E of the Frobenius endomorphism F q,E on the coordinate ring AE of E. The left diagram below is not necessary commutative, but its reduction mod p is commutative As in Section 4, there is a lift Fq,E : A† dR( Eaff,ξ/Qq) π∗ MW (Eaff /Fq) π∗ E → A† −→ H i −→ H i π∗ A† A† E A π∗ AE Fq Fq,E F q F q,E A† A† E π∗ A AE π∗ π∗ E is the natural lift of homomorphism A π∗ Here A† ←−− AE on the coordinate rings which corresponds to the morphism π : Caff,ξ/Qq → Eaff,ξ/Qq, so the reduc- tion of A† E modulo p is just the natural homomorphism on the coordinate rings of Caff ←−− A† ←−− A† π−→ Eaff . Since F q ◦ π∗ = π∗ ◦ F q,E, we know that MW (Eaff /Fq) MW (Caff /Fq) H i H i π∗ π∗ Fq,∗ Fq,E,∗ H i MW (Caff /Fq) π∗ H i MW (Eaff /Fq) So the point counting on Eaff is the same as computing on the subsapce of H 1 generated by {xjy2dx 0 ≤ j ≤ 1 (resp. 0 ≤ j ≤ 2)}. From Lefschetz fixed point formula (Theorem 2.3), we have q,∗ )rH 1 q,∗ )rH 1 #Caff (Fqr ) = Tr(cid:0)(qF −1 #Eaff (Fqr ) = Tr(cid:0)(qF −1 MW (Caff )(cid:1) − Tr(cid:0)(qF −1 MW (Eaff )(cid:1) − Tr(cid:0)(qF −1 q,∗ )rH 0 q,∗ )rH 0 MW (Caff )(cid:1) MW (Eaff )(cid:1) MW (Caff /Fq) o o o o o o o o o o o o POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 15 1 + βr Let PE(X) = (X − β1)(X − β2) be the Weil polynomial of E and Sr(E) := βr 2. Then #Eaff (Fqr ) = qr + 1− Sr(E)− δE, here δE is the number of points at infinity of E. Use #Caff (Fqr ) = #C(Fqr ) − δC , we get #C(Fqr ) = qr + 1 − Sr(E) − Tr(cid:0)(qF −1 q,∗ )rV(cid:1) + (δC − δE), MW (Caff /Fq) generated by {xjy2dx 0 ≤ j ≤ 1 (resp. 0 ≤ here V is the subsapce of H 1 j ≤ 2)}, whose dimension is 4+δC−δE. This implies that the Weil polynomial P (X) of C/Fq equals to PE(X)QV (X)(X − 1)−(δC −δE ), here QV (X) is the characteristic polynomial of qF −1 q,∗ acts on V . The characteristic polynomial PV (X) of Fq,∗ = q · (qF −1 q,∗ )−1 is QV (X)(X − q)δC −δE (X − 1)−(δC −δE ), so P (X) = PE(X)PV (X)(X − q)−(δC −δE ). Summary 5.1. The Weil polynomial P (X) of C/Fq is equal to PE(X)PV (X)(X − q)−(δC −δE ), where PE(X) is the Weil polynomial of E/Fq and PV (X) is the characteristic polynomial of Fq,∗ on V . 6. The Algorithm In order to compute PV (X), one needs to compute PV (X) with a precision N1 = ⌊logp30+2n⌋+1 with n = logpq, which is determined by the Weil bound. Due to the fact that the matrix Mp of the Frobenius action Fp,∗ may have denominators, we need Mp with a precision N2 := N1 + (6n − 1)c with c = ⌊c1 + logp(c1 + logp(2c1))⌋ + 1 and c1 = 6 + logp80 + ∆. From this, we only need to compute (for 1 ≤ l ≤ 3 and 0 ≤ k ≤ 2) Z0, Fp(x), Fp(y) and Fp(ylxkdx) modulo (xN3 , pN4) with N3 = ⌊16p(c2 + logp(2c2))⌋ + 1, N4 = ⌊N2 + c1 + logp(c2 + logp(2c2))⌋ + 1 and c2 = 6 + logp80 + ∆ + N2. Finally, the above discussion is based on the reduction matrices Mi,k (1 ≤ i ≤ 3) introduced in 3.2. But since one can only work with an approximation of Mi,k, one need Mi,k modulo pN5 with a slightly higher precision N5 = N4 + 8⌊logpN3⌋ + 14 . We have N3 = O(pn), N4 = O(n) and N5 = O(n). We work in Zq/pN5 3, whose elements can be stored in O(n2logp) space and the arithmetic on it could be done in O(n2logp) bit operations. This gives the algorithm: Algorithm Step 1. Compute α and β in Corollary 3.9 modulo p.4 Step 2. Compute Z0 in Corollary 4.3 using Newton's method, then Fp(xkyldx) for l = 1, 3 and 0 ≤ k ≤ 2, all of them are modulo (xN3 , pN4).5 Step 3. Use reduction matrices Mi,j (1 ≤ i ≤ 3, 2 ≤ j ≤ N3) to reduce Fp(xkyldx) and get Mp. Step 4. Compute Mq = MpM σ Step 5. Finally, compute the characteristic polynomial PV (X) modulo P N1. p ··· M σn−1 by repeated squaring. p 3More precisely, with p-adic precsion N4 but with denominators at most p(N5−N4). 4In Corollary 3.9, we only need αfy + βfx ≡ 1 modulo p in order to compute the lift of Frobenius. 5In the proof of Lemma 4.1, we showed that the results during the Newton's iteration all have the same rate of convergence as in Corollary 4.3, so we can work modulo xN3 during the Newton's iteration. 16 YIH-DAR SHIEH Theorem 6.1. The above algorithm requires O(n3p) bit operations. Proof. Step 1 consists of solving a system of linear equations over Fq of size at most 16. Hence it requires O(n2 log p) bit operations. Step 2 requires O(log N4) Newton's iterations, and each iteration requires O(N3n2log p) bit operations. Hence this step requires O(pn3) bit operations. Step 3 requires O(N3) operations in Zq/pN5, hence O(pn3) bit operations. Step 4 requires O(log n) squarings and the application of the lift of the p-th power Frobenius σ : Zq → Zq modulo pN4 on matrices of size 6 × 6. Squaring requires O(n2log p) bit operations. For σ, we use Newton's method which needs to evaluate a polynomial of degree n with coeffi- cients in Zq/pN4, which requires O(n) O(n2 log p) bit operations. Hence we need O(n3 log p) bit operations in this step. Step 5 requires O(n2 log p) bit operations. Hence the algorithm requires O(n3p) bit operations. (cid:3) If one works directly on H 1 MW (Caff ) and denote the precision needed by N ′ 3 N ′ 5 . (It is 4 5 instead of 2 i . Also the matrix Mp is of size 6×6 and M ′ i , then Ni ≈ 2 p is of size 9×9. From these, we give a comparison of speed. Step 2 is reduced by a factor of ( 2 3 )2 ≈ 0.45. In Step 3, we have 6 differential forms Fp(xkyldx), l = 1, 3 and 0 ≤ k ≤ 2, to reduce. This contributes a factor of 4 3 because the reductions of Fp(xky2dx) involve fewer operations than the reductions of Fp(xky1dx) and Fp(xky3dx). See the reduction matrices in 3.2.) Since each of these Fp(xkyldx) is computed modulo (xN3, pN4) in Step 2, we work with smaller powers on x and fewer p-adic precision in Step 3. This means that we have fewer reduction steps and the basic arithmetic operations are faster, which contribute a factor of ( 2 3 )2. So Step 3 is reduced by a factor of 4 3 )3 ≈ 0.3, due to the smaller size of Mp and fewer precision. 3 )2 ≈ 0.36. Step 4 is reduced by a factor at least of ( 2 5 · ( 2 References [1] Castryck, W., Point counting on nondegenerate curves. Katholieke Universiteit Leuven, The- sis, 135 pp. (2006). Avaiable at: http://wis.kuleuven.be/algebra/castryck/thesis.pdf [2] Denef, J.; Vercauteren, F., Counting Points on Cab Curves using Monsky-Washnitzer Coho- mology. Finite Fields and Their Applications, Volume 12, Issue 1, pp. 78-102 (2006) [3] Elkik, R., Solutions d'´equations `a coefficients dans un anneau hens´elien. Ann. Sci. ´Ecole Norm. Sup. (4), 6:553-603 (1973) [4] Gaudry, P.; Gurel, N., An extension of Kedlaya's algorithm to superelliptic curves. Proc. Asiacrypt 2001, Springer, Berlin, Lecture Notes in Comput. Sic., 2248: 480-494 (2001) [5] Kedlaya, K., Counting Points on Hyperelliptic Curves using Monsky-Washnitzer Cohomol- ogy. J. Ramanujan Math. Soc. (4), 16: 323-338 (2001) [6] Kedlaya, K., Computing zeta functions via p-adic cohomology. Proc. ANTS-VI, Springer LNCS, 3076: 1-17 (2004) [7] Monsky, P.; Washnitzer, G., Formal cohomology I. Ann. Math. (2), 88: 181-217 (1968) [8] Monsky, P., Formal cohomology II: The cohomology sequence of a pair. Ann. Math. (2), 88: 218-238 (1968) [9] Monsky, P., Formal cohomology III: Fixed point theorems. Ann. Math. (2), 93: 315-343 (1971) [10] van Der Put, M., The cohomology of Monsky and Washnitzer. M´em. Soc. Math. France (N.S.) (4): 23:33-59 (1986) [11] Vermeulen, A.M., Weierstrass pointss of weight two on curves of genus three. Amsterdam University Thesis, 1983. POINT COUNTING ON GENUS 3 CURVES WITH AUTOMORPHISM GROUP Z/2Z 17 Aix Marseille Universit´e, CNRS, Centrale Marseille, I2M, UMR 7373, 13453 Mar- seille, France Current address: Institut de Math´ematiques de Marseille (I2M) 163 avenue de Luminy, Case 907 13288 Marseille, cedex 9 France E-mail address: [email protected]
1502.07060	3	1502	2015-11-30T12:57:41	Compact K\"ahler manifolds admitting large solvable groups of automorphisms	[ "math.AG", "math.DS" ]	Let G be a group of automorphisms of a compact K\"ahler manifold X of dimension n and N(G) the subset of null-entropy elements. Suppose G admits no non-abelian free subgroup. Improving the known Tits alternative, we obtain that, up to replace G by a finite-index subgroup, either G/N(G) is a free abelian group of rank < n-1, or G/N(G) is a free abelian group of rank n-1 and X is a complex torus, or G is a free abelian group of rank n-1. If the last case occurs, X is G-equivariant birational to the quotient of an abelian variety provided that X is a projective manifold of dimension n > 2 and is not rationally connected. We also prove and use a generalization of a theorem by Fujiki and Lieberman on the structure of Aut(X).	math.AG	math	COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS OF AUTOMORPHISMS TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG Abstract. Let G be a group of automorphisms of a compact Kahler manifold X of dimension n and N (G) the subset of null-entropy elements. Suppose G admits no non- abelian free subgroup. Improving the known Tits alternative, we obtain that, up to replace G by a finite-index subgroup, either G/N (G) is a free abelian group of rank ≤ n − 2, or G/N (G) is a free abelian group of rank n − 1 and X is a complex torus, or G is a free abelian group of rank n − 1. If the last case occurs, X is G-equivariant birational to the quotient of an abelian variety provided that X is a projective manifold of dimension n ≥ 3 and is not rationally connected. We also prove and use a generalization of a theorem by Fujiki and Lieberman on the structure of Aut(X). 5 1 0 2 v o N 0 3 ] . G A h t a m [ 3 v 0 6 0 7 0 . 2 0 5 1 : v i X r a 1. Introduction We work over the field C of complex numbers. Let X be a compact Kahler manifold of dimension n and Aut(X) the group of all (holomorphic) automorphisms of X. Fujiki and Lieberman proved that Aut(X) is a complex Lie group of finite dimension but it may have an infinite number of connected components. The identity connected component Aut0(X) of Aut(X) is the set of automorphisms obtained by integrating holomorphic vector fields on X. It is a normal subgroup of Aut(X) and its action on the Hodge cohomology groups H p,q(X, C) is trivial. See [10, 14] for details. Elements in Aut0(X) have zero (topological) entropy. Indeed, for any automorphism g ∈ Aut(X), the entropy of g, defined in the theory of dynamical systems, turns out to be equal to the logarithm of the spectral radius of the pull-back operator g∗ acting on ⊕0≤p≤nH p,p(X, R). This is a consequence of theorems due to Gromov and Yomdin. Topological entropy is always a non-negative number and also equals the logarithm of the spectral radius of g∗ acting on the whole cohomology group ⊕0≤p,q≤nH p,q(X, C). This number is strictly positive if and only if the spectral radius of g∗ acting on H p,p(X, R) is strictly larger than 1 for all or for some p with 1 ≤ p ≤ n − 1. See [5, 12, 16] for details. The group Aut(X) satisfies the following Tits alternative type result which was proved in [3, Theorem 1.5] (generalizing [17, Theorem 1.1]). See [5] for a survey. Recall that a group H is virtually solvable (resp. free abelian, unipotent...), if a finite-index subgroup of H is solvable (resp. free abelian, unipotent...). 2010 Mathematics Subject Classification. 14J50, 32M05, 32H50, 37B40. Key words and phrases. automorphism, complex dynamics, iteration, topological entropy. 1 2 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG Theorem 1.1. Let X be a compact Kahler manifold of dimension n ≥ 2 and G ≤ Aut(X) a group of automorphisms. Then one of the following two alternative assertions holds: (1) G contains a subgroup isomorphic to the non-abelian free group Z ∗ Z, and hence G contains subgroups isomorphic to non-abelian free groups of all countable ranks. (2) G is virtually solvable. In the second case, or more generally, when the representation GH 2(X,C) is virtually solvable, G contains a finite-index subgroup G1 such that the null-entropy subset N(G1) := {g ∈ G1 g is of null entropy} is a normal subgroup of G1 and the quotient G1/N(G1) is a free abelian group of rank r ≤ n − 1. When the action of G on H 2(X, C) is virtually solvable, the integer r in the theorem is called the dynamical rank of G and we denote it as r = r(G). It does not depend on the choice of G1. When GH 2(X,C) is abelian, Theorem 1.1 can be deduced from [6, Theorem I] and the classical Tits alternative for linear algebraic groups. In [6, Remarque 4.9], the authors mentioned the interest of studying these X admitting a commutative G of positive entropy and maximal dynamical rank n − 1. In this paper, we study virtually solvable but not necessarily commutative groups G of maximal dynamical rank r(G) = n − 1. From Theorem 1.2 below, it turns out that N(G) is then virtually contained in Aut0(X), hence G/(G ∩ Aut0(X)) is virtually a free abelian group of rank n − 1. In particular, [17, Question 2.17] is affirmatively answered by Theorem 1.2 (or Theorem 4.1) below. Theorem 1.2. Let X be a compact Kahler manifold of dimension n ≥ 2 and G ≤ Aut(X) a group of automorphisms such that the null-entropy subset N(G) is a subgroup of (and hence normal in) G and G/N(G) ∼= Z⊕n−1. Then either X is a complex torus, or N(G) is a finite group and hence G is virtually a free abelian group of rank n − 1. Note that N(G) is normal in G because if g ∈ G has zero entropy, by Gromov and Yomdin, hgh−1 also has zero entropy for every h ∈ Aut(X). Remark 1.3. We can deduce from Theorem 1.1 that the representation GH 2(X,C) is virtually solvable if and only if, for some finite-index subgroup G1 of G, N(G1) is a normal subgroup of G1 and G1/N(G1) ∼= Z⊕r for some r ≤ n − 1 (cf. [3, Theorem 2.3]). So the condition of Theorem 1.2 is, up to replace G by a finite-index subgroup, equivalent to that GH 2(X,C) is virtually solvable and G has maximal dynamical rank n−1. Proposition 1.4 below gives a more precise description of the present situation. Recall that by Theorem 1.1, the first hypothesis in the following result is satisfied when G admits no non-abelian free subgroup. Note also that Theorem 1.2 can be applied to the following group G0. COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 3 Proposition 1.4. Let X be a compact Kahler manifold of dimension n ≥ 2 and G ≤ Aut(X) a group of automorphisms such that GH 2(X,C) is virtually solvable and of maximal dynamical rank n − 1. Then there is a normal subgroup G0 of G such that (1) G/G0 is isomorphic to a subgroup of the symmetric group Sn. In particular, it contains at most n! elements. (2) The null-entropy subset N(G0) is a normal subgroup of G0 and G0/N(G0) ∼= Z⊕n−1. Remark 1.5. Let X, G be as in Proposition 1.4, and let H be the normalizer of G in Aut(X). Then either H : G < ∞, or X is a complex torus. And in the latter case H/(H ∩ Aut0(X)) : G/(G ∩ Aut0(X)) < ∞. Theorem 1.2 (or Theorem 4.1 (2)) enables us to apply [19, Theorem 1.1] and immedi- ately get the following result. Corollary 1.6. Assume the condition for X and G in Theorem 1.2 (or 4.1). Assume further that n ≥ 3, X is a projective manifold and X is not rationally connected. Then, replacing G by a finite-index subgroup, we have the following: (1) There is a birational map X 99K Y such that the induced action of G on Y is biregular and Y = T /F , where T is an abelian variety and F is a finite group whose action on T is free outside a finite subset of T . (2) There is a faithful action of G on T such that the quotient map T → T /F = Y is G-equivariant. Every G-periodic irreducible proper subvariety of Y or T is a point. When X is rationally connected, we can also state some necessary and sufficient hy- pothesis for X to be G-equivariant birational to a quotient of an abelian variety, see [19] for more details. Generalizing Theorem 4.1 and Corollary 1.6, we propose the following. See [5, Problem 1.5] and the remarks afterwards for a related problem. See Oguiso and Truong [15, Theorem 1.4] for the first example of rational threefold or Calabi -- Yau threefold admitting a primitive automorphism of positive entropy. Problem 1.7. (1) Classify compact Kahler manifolds of dimension n ≥ 3 admitting a solvable group G of automorphisms of dynamical rank r for some 2 ≤ r ≤ n− 1, such that the pair (X, G) is primitive, i.e., there is no non-trivial G-equivariant meromorphic fibration even after replacing G by a finite-index subgroup. (2) One may strengthen the condition on G to that every non-trivial element of G is of positive entropy. So G is isomorphic to Z⊕r. (3) One may also weaken the condition on G to that the representation GH 1,1(X,C) is an abelian or solvable group. 4 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG One may ask similar problems for the case of manifolds defined over fields of positive characteristic but new tools need to be developed. We refer the reader to a pioneering work in this direction by Esnault and Srinivas in [8]. The present paper is organized as follows. In Section 2, we give a generalization of a theorem by Fujiki and Lieberman that we will need in the proof of the main theorem. Theorem 2.1 and Proposition 2.6 should be of independent interest. In Section 3, we study a notion of privileged vector space which allows us to construct common eigenvectors with maximal eigenvalues. The main theorem and its proof, as well as the proofs of Proposition 1.4 and Remark 1.5 are given in the last section. Idea of the proof of the main theorems 1.2 and 4.1. Assume G/N(G) is free abelian of maximal rank n − 1. The key step is to show that the null-entropy subset N(G) is virtually contained in the identity connected component Aut0(X) of Aut(X), i.e., N(G) is a finite extension of N(G) ∩ Aut0(X). Indeed, once this assertion is proved, either Aut0(X) is trivial and hence G is virtually free abelian of rank n − 1, or Aut0(X) is non-trivial. In the latter case, since the maximality of the dynamical rank of G implies that every G-equivariant fibration is trivial, we can show that -- either X is almost homogeneous under a linear algebraic group and hence G is of null entropy, which contradicts the fact that the dynamical rank of G is n − 1 ≥ 1 (cf. [9]). -- or the albanese map albX : X → Alb(X) is bimeromorphic (and indeed is biholomor- phic), so X is a complex torus. The desired assertion is equivalent to that the representation of N(G) on H 2(X, C) is a finite group, in view of the results of Fujiki and Lieberman [10, 14]. However, their theorems cannot be directly applied in our setting. We will give in Theorem 2.1 or Proposition 2.6 some generalization of their results. Hence, for our purpose, we only need to prove that after replacing G by a finite-index subgroup, the null-entropy subset N(G) fixes a nef and big class A in H 1,1(X, R). Such a class A will be constructed as follows (see Proof of Theorem 4.1 (1)). Note the representation N(G)H 2(X,C) is virtually a unipotent group (say a unipotent group for simplicity), and hence has a non-trivial centre ZH 2(X,C) for some Z E G. We then introduce a notion of privileged subspaces of H 2(X, R), defined over Q, in which we can control classes in H 2,0(X, C)⊕ H 0,2(X, C) by classes in H 1,1(X, R), see Definition 3.1. The key Proposition 3.2 shows the existence of a G-invariant privileged subspace F on which the action of Z is trivial. By considering the upper central series of N(G)H 2(X,C) and doing induction on its length, we can shrink F such that the action of N(G) on such F is also trivial. It follows from the above discussion that GF is abelian. Following the spirit of [6], we can find s+1 nonzero common nef classes L1, . . . , Ls+1 ∈ F of GF , such that L1 · · · Ls+1 6= COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 5 0 in H s+1,s+1(X, R) and the group homomorphism below ψ : GF → (R⊕s, +), g∗F 7→ (χ1(g), . . . , χs(g)) satisfies that Ker ψ = N(GF ) and Im ψ is a discrete subgroup of R⊕s generating R⊕s, where each χi : G → (R, +) is the corresponding group character of Li. In other words, we have GF /N(GF ) ≃ Z⊕s. We next claim that the above s = n − 1. Otherwise, we can extend those Li to a full quasi-nef sequence as in [17, §2.7]. Let eN denote the inverse image of N(GF ) under the natural restriction π : G → GF . It is easy to see that eN /N(G) is a free abelian group of rank n− 1− s. While by analysing the image of N(GF ) under the group homomorphism induced by the above quasi-nef sequence, we will see that eN/N(G) can be embedded into Now the nef and big class A := L1 + · · · + Ln satisfies the desired property. R⊕n−s−2 as a discrete subgroup. This is a contradiction. Acknowledgement. The authors would like to thank the referee for the suggestions which allow us to improve the presentation of the paper. The first and last authors are supported by ARFs of NUS. 2. Fujiki and Lieberman type theorem Let X be a compact Kahler manifold of dimension n. Let ω be a Kahler form on X. Its class in H 1,1(X, R) := H 1,1(X, C) ∩ H 2(X, R) is denoted by {ω}. Let Aut{ω}(X) denote the group of all automorphisms g of X preserving {ω}, i.e., g∗{ω} = {ω}. Fujiki and Lieberman proved in [10, Theorem 4.8] and [14, Proposition 2.2] that Aut0(X) is a finite-index subgroup of Aut{ω}(X). The purpose of this section is to prove a more general version of this theorem that we will use later. Let φ be a differential (p, p)-form on X. We say that φ is positive if in any local holomorphic coordinates z we can write φ(z) as a finite linear combination, with non- negative functions as coefficients, of (p, p)-forms of type (il1(z) ∧ l1(z)) ∧ · · · ∧ (ilp(z) ∧ lp(z)), where lj(z) are C-linear functions in z. This notion does not depend on the choice of local holomorphic coordinates z. A (p, p)-current T on X is said to be weakly positive if T ∧ φ defines a positive measure for any positive (n − p, n − p)-form φ. A (p, p)-form is weakly positive if it is weakly positive in the sense of currents. A (p, p)-current T on X is said to be positive if T ∧ φ defines a positive measure for any weakly positive (n − p, n − p)-form φ. It turns out that a form is positive if it is positive in the sense of currents and positivity implies weak positivity. These two notions coincide for p = 0, 1, n− 1, n and are different otherwise. If 6 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG α is a holomorphic p-form on X, then ip2 It vanishes only when α = 0. This can be easily checked using local coordinates on X. α∧ ¯α is an example of weakly positive (p, p)-form. Weakly positive and positive forms and currents are real, that is, they are invariant under the complex conjugation. If ω is a Kahler form on X as above, then ωp is a positive (p, p)-form for every p. A positive (resp. weakly positive) (p, p)-form or current T is said to be strictly positive (resp. strictly weakly positive) if there is an ǫ > 0 such that T − ǫωp is positive (resp. weakly positive). We refer to Demailly [4], or [7] for more details. Recall from Hodge theory that the group H p,q(X, C) can be obtained as the quotient of the space of closed differential (p, q)-forms by the subspace of exact (p, q)-forms. It is not difficult to see that every closed (p, q)-current T , defines via the pairing hT, φi with φ a closed (n− p, n− q)-form, an element in the dual of H n−p,n−q(X, C), because T vanishes on exact (n − p, n − q)-forms by Stokes' theorem. So by Poincar´e duality, T defines a class in H p,q(X, C) that will be denoted by {T}. Denote by Kp(X) (resp. Kw p (X)) the set of classes of strictly positive (resp. strictly weakly positive) closed (p, p)-forms. They are strictly (i.e., salient) convex open cones in H p,p(X, R) := H p,p(X, C)∩ H 2p(X, R). Denote by Ep(X) (resp. E w p (X)) the set of classes of positive (resp. weakly positive) closed (p, p)-currents. They are strictly (i.e., salient) convex closed cones in H p,p(X, R). Indeed, if T is a (weakly) positive closed (p, p)-current, the quantity hT, ωn−pi is equivalent to the mass-norm of T and it only depends on the cohomology class of T . One often calls this quantity the mass of T . So T vanishes if and only if its class {T} vanishes. Moreover, if Tn are such that {Tn} converge to some class c, then since {Tn} are bounded, the masses of Tn are bounded and hence, up to extract a subsequence, we can assume Tn converge to some (weakly) positive closed current T . It follows that c is represented by T and hence Ep(X) (resp. E w The Kahler cone of X, denoted by K(X), consists of all Kahler classes. So it is equal to K1(X) and also to Kw 1 (X) because positivity and weak positivity coincide for bidegree (1, 1). The closure K(X) ⊂ H 1,1(X, R) of K(X) is called the nef cone. The cones E1(X) and E w 1 (X) are also equal and are denoted simply by E(X). This is called the pseudo- effective cone. Classes in the interior of E(X) are called big and are the classes of Kahler currents, i.e., strictly positive closed (1, 1)-currents. In general, we have p (X)) are closed. Kp(X) ⊂ Kw p (X) ⊂ K p (X) ⊂ E w p (X) and Kp(X) ⊂ Kp(X) ⊂ Ep(X) ⊂ E w w p (X). The classes in the interior of Ep(X) (resp. E w p (X)) are the classes of strictly positive (resp. strictly weakly positive) closed (p, p)-currents. In what follows, we will write c ≤ c′ or c′ ≥ c for classes c, c′ ∈ H p,p(X, R) such that c′ − c belongs to E w The automorphism group Aut(X) acts naturally on cohomology groups and preserves all the above cones, their closures, interiors and boundaries. We will fix a norm k · k for each cohomology group and a Kahler form ω of X. Our results do not depend on the choice of these norms or ω. The main result in this section is the following theorem. p (X). COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 7 Theorem 2.1. Let X be a compact Kahler manifold of dimension n and G ≤ Aut(X) a group of automorphisms. Assume that for every element g ∈ G there is a class c of a strictly weakly positive closed (p, p)-current with 1 ≤ p ≤ n−1, such that k(gm)∗ck = o(m) as m → +∞. Here, c and p may depend on g. Then G is virtually contained in Aut0(X), i.e., G : G ∩ Aut0(X) < ∞. The following immediate corollary can be applied when c is a big class in H 1,1(X, R). If c is the class of a Kahler form, the result is exactly the theorem by Fujiki and Lieberman quoted above. Corollary 2.2. Let c be the class of a strictly weakly positive closed (p, p)-current, 1 ≤ p ≤ n− 1. Let Autc(X) denote the group of all automorphisms g such that g∗c = c. Then Autc(X) is virtually contained in Aut0(X). We will see in Lemma 2.3 below that the condition that g∗c = c for c as in the above corollary is equivalent to the condition that g∗c is parallel to c. We prove now Theorem 2.1. We will identify H n,n(X, R) with R by taking the integrals of (n, n)-forms on X. The cup product of two cohomology classes c, c′ is simply denoted by cc′ or c · c′. We need the following lemmas. Lemma 2.3. Let g ∈ Aut(X). Assume there is a class c of a strictly weakly positive closed (p, p)-current for some 1 ≤ p ≤ n − 1, such that k(gm)∗ck = o(m) as m → +∞. Then, for every 0 ≤ q ≤ n, the norm of (gm)∗ acting on H q,q(X, R) is bounded by a constant independent of m ∈ N. In particular, the conclusion holds if g∗c is parallel to c and in this case we have g∗c = c. Proof. Using the Jordan canonical form of g∗ acting on H p,p(X, C), we see that the hypothesis k(gm)∗ck = o(m) implies that k(gm)∗ck is bounded by a constant independent of m ∈ N. Let cp be any class in E w p (X), there is a constant λ > 0 such that cp ≤ λc. Since g∗ preserves E w p (X) and this cone is salient and closed, we deduce that the sequence k(gm)∗cpk is also bounded. We will use this property for the class {ωp} of ωp, where ω is the fixed Kahler form of X. p (X). Since c is in the interior of E w For 0 ≤ q ≤ n, we define Iq,m :=ZX (gm)∗(ωq) ∧ ωn−q = (gm)∗{ωq} · {ωn−q} and Jq := lim m→+∞ log Iq,m log m · By the Jordan canonical form of g∗ acting on H q,q(X, C), we see that the limit in the definition of Jq exists in Z≥0 ∪{±∞}. The above discussion implies that Jp ≤ 0. We also have J0 = Jn = 0 since g∗ is the identity on H 0,0(X, R) ∼= R and H n,n(X, R) ∼= R. We apply Theorem 2.4 below to ω1 := ω, ω2 := (gm)∗ω, ωi := (gm)∗ω for arbitrary q−1 indices 3 ≤ i ≤ n and ωi := ω for the other n − q − 1 indices 3 ≤ i ≤ n. We obtain that I 2 q,m ≥ Iq−1,mIq+1,m. It follows that the function q 7→ Jq is concave, i.e., 2Jq ≥ Jq−1 + Jq+1. We then deduce that Jq = 0 for every 0 ≤ q ≤ n. In particular, using again the Jordan canonical form of g∗ acting on H q,q(X, C), we see that the sequence Iq,m is bounded. 8 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG Since the integral Iq,m is the mass of the positive closed (q, q)-current (gm)∗(ωq), we deduce that the sequence k(gm)∗{ωq}k is also bounded. In the same way as it was just p (X) at the beginning of the proof, we obtain that the sequence k(gm)∗cqk is done for E w bounded for any cq ∈ E w q (X). Thus, the property holds for any cq ∈ H q,q(X, R) because E w q (X) spans H q,q(X, R). So the first assertion follows. Suppose now that g∗c = λc for some λ ∈ R. Since g∗ preserves the weak positivity, we have λ > 0. Replacing g by g−1 if necessary, we may assume that λ ≤ 1. Then the condition of the first assertion is satisfied, so the norm of (gm)∗ acting on all H q,q(X, R) is bounded. Thus the spectral radius of g∗ on all H q,q(X, R) are less than or equal to 1. The same is true for (g−1)∗, because the action of (g−m)∗ on H q,q(X, R) is the adjoint action of (gm)∗ on H n−q,n−q(X, R) by Poincar´e duality. Thus, all eigenvalues of g∗ on all H q,q(X, R) are of modulus 1. In particular, if g∗c = λc, then λ = 1. (cid:3) Recall the following result of Gromov, that we used above, see [5, Corollary 2.2] or [11]. It is a consequence of the mixed version of the classical Hodge -- Riemann theorem. Theorem 2.4 (Gromov). Let ωi be Kahler forms on X, 1 ≤ i ≤ n. Define for 1 ≤ i, j ≤ 2 Iij :=ZX 12 ≥ I11I22. Then we have I 2 ωi ∧ ωj ∧ ω3 ∧ · · · ∧ ωn = {ωi}{ωj}{ω3} · · ·{ωn}. Lemma 2.5. Let g ∈ Aut(X). Assume that the norm of (gm)∗ acting on H 1,1(X, R) is bounded by a constant independent of m ∈ N. Then the action g∗H 2(X,R) has finite order. Proof. We first show that the action of (gm)∗ on H 2(X, R) is bounded independently of m ∈ N. Recall that (gm)∗ preserves the Hodge decomposition H 2(X, C) = H 2,0(X, C) ⊕ H 1,1(X, C) ⊕ H 0,2(X, C). So it is enough to prove that the action of (gm)∗ on H 2,0(X, C) is bounded independently of m ∈ N. The similar property for H 0,2(X, C) is then obtained by complex conjugation. Observe that if v is a class in H 2,0(X, C), it can be represented by a unique closed holomorphic 2-form α. The form α ∧ ¯α is weakly positive and closed, which vanishes if and only if α = 0. We deduce that v¯v = 0 if and only if v = 0. Hence, kvk is bounded if and only if kv¯vk is bounded. For any (2, 0)-class v in H 2,0(X, C), there is a Kahler class u satisfying v¯v ≤ u2. Indeed, 2 (X), it is enough to choose u as a since the class {ω2} is in the interior of the cone E w large multiple of the fixed Kahler class {ω}. We deduce that (gm)∗v · (gm)∗¯v = (gm)∗(v¯v) ≤ (gm)∗(u2) = (gm)∗(u)2. Therefore, by hypothesis, k(gm)∗v · (gm)∗¯vk is bounded independently of m ∈ N. Thus k(gm)∗vk satisfies the same property. COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 9 We conclude that the action of (gm)∗ on H 2(X, R) is bounded independently of m ∈ In particular, all eigenvalues of g∗H 2(X,R) are of modulus ≤ 1, hence they are of N. modulus 1, and indeed are roots of unity by Kronecker's theorem, since g∗ preserves (and is an isomorphism of) H 2(X, Z) and H 2(X, R) = H 2(X, Z) ⊗Z R. This and the above boundedness result imply that the action g∗H 2(X,R) is periodic, by looking at the Jordan canonical form of the action. (cid:3) End of the proof of Theorem 2.1. By Lemmas 2.3 and 2.5, for any g ∈ G, the auto- morphism g∗H 2(X,R) is periodic. The characteristic polynomials of g∗H 2(X,R) are defined over Z and we have seen that their zeros are roots of unity for all g ∈ G. Since these poly- nomials are of degree dim H 2(X, R), the orders of these roots of unity are bounded by the same constant, i.e., the group GH 2(X,R) has bounded exponent. Thus, it is a finite group by the classical Burnside's theorem, see [2]. Set v :=P h∗{ω}, where h runs through the finite group GH 1,1(X,R) and ω is the fixed Kahler form of X. Then G is contained in Autv(X) := {g ∈ Aut(X) g∗v = v}. Since v is a Kahler class, the theorem by Fujiki and Lieberman quoted above says that Autv(X) : Aut0(X) < ∞. Now G/(G ∩ Aut0(X)) ∼= (G · Aut0(X))/ Aut0(X) ≤ Autv(X)/ Aut0(X) and the latter is a finite group. The theorem follows. (cid:3) Since Aut0(X) acts trivially on H p,q(X, C), the above discussion immediately implies the following proposition. Proposition 2.6. Let X be a compact Kahler manifold of dimension n and G ≤ Aut(X) a group of automorphisms. Then G is virtually contained in Aut0(X) if and only if the representation of G on H p,p(X, R) for some 1 ≤ p ≤ n − 1 (or on ⊕0≤p,q≤nH p,q(X, C)) is a finite group. 3. Invariant privileged vector spaces Let X be a compact Kahler manifold of dimension n. By Hodge theory, every class v in H 2(X, R) admits a unique decomposition v = v20 + v11 + v02, where v11 ∈ H 1,1(X, R), v20 ∈ H 2,0(X, C), v02 ∈ H 0,2(X, C) with v02 = ¯v20. For any vector subspace V ⊆ H 2(X, R) define V 0 := V ∩ (H 2,0(X, C) ⊕ H 0,2(X, C)), V 11 := V ∩ H 1,1(X, R). In general, we have V 0 ⊕ V 11 ⊆ V , but the inclusion may be strict. Definition 3.1. We say that V is privileged if it satisfies the following properties: 10 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG (a) V is defined over Q, V 6= 0 and V = V 0 ⊕ V 11. (b) The intersection of V 11 with the nef cone K(X) has non-empty interior in V 11, or equivalently this intersection spans V 11 (because K(X) is convex). (c) For every v ∈ V 0 there is a nef class u in V 11 such that v20¯v20 ≤ u2. The main purpose of this section is to construct some privileged subspace of H 2(X, R) as stated in the introduction. More precisely, we have the following result. Proposition 3.2. Let G and Z be subgroups of Aut(X). Assume that Z is normalized by G and that ZH 2(X,C) (or equivalently ZH 2(X,R)) is a unipotent commutative matrix group. Then, there is a privileged subspace F ⊆ H 2(X, R) which is G-stable such that the action of Z on F is the identity. Recall here that the representation ZH 2(X,R) of Z is defined over Q because Aut(X) acts on H 2(X, Z) and H 2(X, R) = H 2(X, Z) ⊗Z R. Notation 3.3. From here till Proposition 3.5, for simplicity, the pull-back action of g ∈ Aut(X) on H ∗(X, C), will be denoted by the same g, instead of g∗. So for g, h ∈ Aut(X) and v ∈ H ∗(X, C) the identity (gh)∗v = h∗(g∗v) will be written as (gh)v = h(g(v)). Lemma 3.4. There is a privileged Z-stable subspace V ⊆ H 2(X, R) such that the action of Z on V is the identity. Proof. Set V0 := H 2(X, R), and K0 := K(X) which has non-empty interior in V 11 0 = H 1,1(X, R). Take any z ∈ Z. Recall that the action of z on H 2(X, R) is unipotent, i.e., all eigenvalues are equal to 1. Therefore, in what follows, for vector spaces, we can use bases defined over Q. Let l ≥ 0 be the minimal integer such that the norm of zm acting on V0 satisfies kzmk = O(ml) as m → +∞. If l = 0 then z is the identity on V0. This can be seen by using the Jordan canonical form of z. So we can take V = V0 when the property l = 0 holds for all z. Otherwise, fix a z ∈ Z such that l ≥ 1. Define π : V0 → V0 to be the limit of m−lzm and V1 the image of V0 by π. Using the Jordan canonical form of z with a basis of V0 defined over Q, we see that V1 6= 0, V1 6= V0 and V1 is defined over Q. Since zm preserves the decomposition V 0 0 ⊕ V 11 Moreover, π(K0) has non-empty interior in V 11 0 , our V1 satisfies Property 3.1 (a). 1 by open mapping theorem applied to 1 . By the definition of π, we have π(K0) ⊆ K0. Denote by K1 the intersection 1 . So V1 satisfies Property 3.1 (b). 0 such that v = lim m−lzm(v⋆). Let u⋆ ∈ K0 such 0 → V 11 π : V 11 1 ∩ K0. It has non-empty interior in V 11 V 11 For any v ∈ V 0 1 , there is a v⋆ ∈ V 0 that v⋆ 20¯v⋆ 20 ≤ u⋆2. Define u := lim m−lzm(u⋆). COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 11 We have v20¯v20 ≤ u2 because u2 − v20¯v20 = lim m−2lzm(u⋆2 − v⋆ 20¯v⋆ 20). Indeed, the class m−2lzm(u⋆2 − v⋆ 20¯v⋆ 20) is represented by some weakly positive closed (2, 2)-current and the cone E w 2 (X) of weakly positive classes in H 2,2(X, R) is closed. So Property 3.1 (c) holds for V1. And eventually our V1 is privileged. For any arbitrary w ∈ Z, as actions on H 2(X, R), we have wzm = zmw. It follows that wπ = πw. So V1 is w-stable for all w ∈ Z, and hence Z-stable. We can now apply the same argument in order to construct inductively Vi+1 ⊂ Vi and Ki+1 ⊂ Ki. For dimension reason, there is an i such that Z is the identity on Vi. Take V = Vi. We are done. (cid:3) Proof of Proposition 3.2. Let V be as in Lemma 3.4. We show that for any g ∈ G, g(V ) is also privileged. Property 3.1 (a) is clear because g is defined over Q and preserves the Hodge decomposition H 2,0(X, C) ⊕ H 1,1(X, C) ⊕ H 0,2(X, C). Property 3.1 (b) is a consequence of the fact that g is an isomorphism of H 1,1(X, R) and preserves the nef cone K(X). Property 3.1 (c) is also clear because g preserves the weak positivity. For any z ∈ Z and g ∈ G, we have z′ := gzg−1 ∈ Z since Z is normalized by G, and hence for any v ∈ V (see Notation 3.3) z(g(v)) = g(z′(v)) = g(v). Thus, the action of Z on g(V ) is the identity. Consider F = P V , where V runs over all subspaces satisfying Lemma 3.4. We have g(F ) ⊆ F and hence g(F ) = F for dimension reason. Also by dimension reason, the last sum is in fact a finite sum. Therefore, F is defined over Q because it has a system of generators defined over Q. Like on V , the group Z acts as the identity on F , and F 11 is spanned by nef classes. Hence F satisfies Properties 3.1 (a) and 3.1 (b). To finish the proof of Proposition 3.2, we still have to show that F satisfies Property 20 ∈ H 2,0(X, C) and u, u′ ∈ 20) ≤ 2(u + u′)2. We 20 by a closed holomorphic 2-form α. Since the (2, 2)-form α ∧ ¯α is 3.1 (c). For this purpose, it is enough to show that if v20, v′ K(X) satisfy v20¯v20 ≤ u2 and v′ can represent v20 − v′ weakly positive, we deduce the following Cauchy -- Schwarz inequality 20 ≤ u′2, then (v20 + v′ 20)(¯v20 + ¯v′ 20¯v′ It follows that (v20 − v′ 20)(¯v20 − ¯v′ 20) ≥ 0. (v20 + v′ 20)(¯v20 + ¯v′ 20 ≤ 2u2 + 2u′2. 20) ≤ 2v20¯v20 + 2v′ 20¯v′ On the other hand, we can approximate u and u′ by classes of Kahler forms and hence 2 (X) is closed in H 2,2(X, R), we uu′ by classes of positive (2, 2)-forms. Since the cone E w 12 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG obtain uu′ ≥ 0. Therefore, the right-hand side class in the latter chain of inequalities is bounded by 2(u + u′)2. This ends the proof of Proposition 3.2. (cid:3) Proposition 3.5. Let G, Z and F be as in Proposition 3.2. Then, for every g ∈ G there exists a nef class v 6= 0 in F 11 ⊆ F such that gv = λv, where λ is the spectral radius of g acting on F . Proof. If γ is a complex eigenvalue of g acting on F 0, then up to replace γ with ¯γ, there are vectors v, v′ ∈ F 0 such that v′′ 20. Therefore, we have 20 6= 0 and g(v′′ 20 := v20 + iv′ 20) = γv′′ gm(v′′ 20¯v′′ 20) = γ2mv′′ 20¯v′′ 20. Let u, u′ ∈ F 11 be as in Property 3.1 (c) for v, v′ respectively. As in the proof of Propo- sition 3.2, for u′′ = √2(u + u′), we have v′′ 20¯v′′ 20 ≤ u′′2 and hence It follows that γ2mv′′ 20¯v′′ 20 ≤ gm(u′′2). γm . kgm(u′′)k. Therefore, γ ≤ ρ(gF 11), where ρ(gF 11) is the spectral radius of g acting on F 11. So g acts on F 11 with spectral radius λ and preserves the cone C := F 11 ∩ K(X). Since this cone has non-empty interior in F 11, the proposition follows from Theorem 3.6 below. (cid:3) We quote Birkhoff's generalization of the Perron -- Frobenius theorem. Theorem 3.6 (cf. [1]). Let C be a strictly (i.e., salient) convex closed cone of a finite- dimensional R-vector space V such that C spans V as a vector space. Let g : V → V be an R-linear endomorphism such that g(C) ⊆ C. Then the spectral radius ρ(g) is an eigenvalue of g and there is an eigenvector Lg ∈ C corresponding to the eigenvalue ρ(g). The following lemma slightly generalizes [6, Proposition 4.1] which only deals with commutative groups. We will use this lemma later to obtain the existence of common eigenvectors in the proof of Proposition 1.4. Lemma 3.7. Let C be a strictly (i.e., salient) convex closed cone of a finite-dimensional R-vector space V such that C spans V as a vector space. Let G ≤ GL(V ) and let N EG be a finite normal subgroup such that G(C) ⊆ C and the quotient group G/N is commutative. Then for every g ∈ G there is a non-zero element Lg in C such that g(Lg) = ρ(g)Lg with ρ(g) the spectral radius of g, G(Lg) ⊆ R+Lg, and n(Lg) = Lg for any n ∈ N. Proof. Consider the linear map π : V → V defined by n(L), π(L) := 1 N Xn∈N and denote by V ′ its image. This is the subspace of all vectors in V which are fixed by N. Hence, since N is normal in G, it is not difficult to see that V ′ is invariant by G. Define COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 13 C ′ := C ∩ V ′ = π(C). Since C is closed, strictly convex and spans V , C ′ is also closed, strictly convex, and spans V ′. By taking L in the interior of C, we see that C ′ contains vectors in the interior of C. Observe that since C is strictly convex and invariant by G, for every vector L in the interior of C and for any g ∈ G, when n → ∞, the growth of kgn(L)k is comparable with the growth of the norm kgnk of gn acting on V . We then conclude that the norm kgnk of gn acting on V ′ is comparable with the one on V as n → ∞. In particular, the spectral radius of g on V ′ is equal to the one on V . Now replacing G, V, C by G/N, V ′ and C ′ reduces the problem to the case of the action (cid:3) of a commutative group. This case was proved in [6, Proposition 4.1]. Before the end of this section, we recall the cone theorem of Lie -- Kolchin type in [13], which will be used in the proof of our main Theorem 4.1. Theorem 3.8 (cf. [13, Theorem 1.1]). Let V be a finite-dimensional real vector space and {0} 6= C ⊂ V a strictly (i.e., salient) convex closed cone. Suppose that G ≤ GL(V ) is a solvable group, and it has connected Zariski closure in GL(VC) (always possible by replac- ing G with a finite-index subgroup) and G(C) ⊆ C. Then G has a common eigenvector in the cone C. Remark 3.9. Note that in this theorem the connectedness of the Zariski closure of G in GL(VC) is necessary. Indeed, we have the following example. Let G be the subgroup of GL(2, R) consisting of all invertible 2 × 2 matrices (aij)1≤i,j≤2 with non-negative entries which are diagonal (i.e., a12 = a21 = 0) or anti-diagonal (i.e., a11 = a22 = 0). Then G is a solvable Lie group and its Zariski closure in GL(2, C) has two connected components. It preserves the cone of vectors with non-negative coordinates. However, there is no common eigenvector of G in this cone. Remark 3.10. Let V be a finite-dimensional real vector space and {0} 6= C ⊂ V a strictly (i.e., salient) convex closed cone. Suppose that G ≤ GL(V ) ≤ GL(VC) is a virtually solvable group and G(C) ⊆ C. Let G1 E G be the preimage, via the natural homomorphism G → GL(VC), of the identity connected component of the Zariski closure of G in GL(VC). This connected component is exactly the Zariski closure of G1 in GL(VC) and is denoted by G1. So G1 is solvable because it is virtually solvable and connected. We then deduce that G1 is solvable and we can apply Theorem 3.8 to G1. 4. Generalization of Theorem 1.2 and proof of Proposition 1.4 Theorem 1.2 will follow from the more general form below. Theorem 4.1. Let X be a compact Kahler manifold of dimension n ≥ 2 and G ≤ Aut(X) a group of automorphisms such that the null-entropy subset N(G) is a subgroup of (and hence normal in) G and G/N(G) ∼= Z⊕n−1. Then we have: 14 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG (1) The representation of N(G) on ⊕0≤p,q≤nH p,q(X, C) is a finite group. Hence N(G) is virtually contained in Aut0(X). (2) Either N(G) is a finite group and hence G is virtually a free abelian group of rank n − 1, or X is a complex torus and G ∩ Aut0(X) = N(G) ∩ Aut0(X) is Zariski-dense in Aut0(X) ∼= X. (3) G/(G ∩ Aut0(X)) and GE are virtually free abelian groups of rank n − 1 for any G-stable (real or complex) vector subspace E of ⊕0≤p,q≤nH p,q(X, C) containing H p,p(X, R) for some 1 ≤ p ≤ n − 1, e.g., H 2(X, R) or H 1,1(X, R). Remark 4.2. The condition of Theorem 4.1 (except the requirement for the dynamical rank of G) is, up to replace G by a finite-index subgroup, equivalent to that the repre- sentation GH 1,1(X,C) (or GNSR(X) when X is projective) is virtually solvable (see Remark 1.3, [17, Theorem 1.2], or [3, Theorem 2.3]). Proof of Theorem 4.1 (1). Since N(G) is of null-entropy and preserves H 2(X, Z), the restriction N(G)H 2(X,C) is virtually unipotent (cf. [3, Theorem 2.2]). Indeed, U := {g ∈ N(G) : g∗H 2(X,C) is unipotent} is a normal subgroup of N(G) with finite-index. Note that UH 2(X,C) is automatically nilpotent (with nilpotency class c say). If it is trivial, the proof is easy (see the argument below). Thus we assume it is non-trivial, and hence has a non-trivial centre ZH 2(X,C) for some normal subgroup Z of U. Note that Z EG (since Z is a characteristic subgroup of G) and ZH 2(X,C) is a unipotent commutative matrix group. Then by Proposition 3.2, there is a non-trivial G-stable privileged vector subspace F1 ⊆ H 2(X, R) (defined over Q) such that ZF1 is trivial. Hence as the image of the nilpotent group UH 2(X,C) by the natural restriction, UF1 is also a nilpotent group of nilpotency class at most c − 1. Repeating the above process several times, we may find a non-trivial G-stable privileged vector subspace F ⊆ H 2(X, R) (defined over Q) such that UF is trivial, and hence N(G)F is finite. Set FZ := F ∩ H 2(X, Z). Then F = FZ ⊗Z R. Replacing G by a finite-index subgroup, we may assume that N(G)F is trivial, see [3, In other words, if π : G → GF denotes the natural restriction, then Lemma 2.4]. N(G) E Ker π. Thus GF ≃ G/ Ker π is the quotient group of G/N(G) ≃ Z⊕n−1 by Ker π/N(G), and hence is abelian. As in [6, Proof of Theorem I], let s be the maximal number of non-trivial linearly independent characters χi : GF → (R, +) corresponding to nonzero common nef classes Li ∈ F of GF , such that g∗Li = exp χi(g∗F )Li for all g ∈ G. By taking the composition with π : G → GF , these χi can be thought of as characters of G itself. Define a group homomorphism as follows ψ : GF → (R⊕s, +), g∗F 7→ (χ1(g), . . . , χs(g)). COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 15 One can show that Ker ψ = N(GF ) (cf. [6, Proposition 4.2]), and Im ψ is a discrete subgroup of R⊕s (cf. [6, Corollary 2.2]) because our FZ is defined over Q. Further, the lattice Im ψ generates R⊕s. In other words, we have GF /N(GF ) ≃ Z⊕s. If s = 0, let Ls+1 be any nonzero common nef eigenvector of the commutative group GF . If s > 0, since ψ(G) is a spanning lattice, one may choose an automorphism f ∗F ∈ GF such that every coordinate of ψ(f ∗F ) is negative; let Ls+1 ∈ F be a nonzero common nef eigenvector of GF such that f ∗Ls+1 = ρ(f ∗F )Ls+1, where ρ(f ∗F ) denotes the spectral radius of f ∗F (cf. [6, Proposition 4.1], Proposition 3.5). Note that ρ((f −1)∗F ) > 1 by the choice of f , and hence ρ(f ∗F ) > 1, since FZ is defined over Z and G-stable entailing det(f ∗F ) = ±1. As in [6], we claim that L1 · · · Ls+1 6= 0. Indeed, we first prove that L1 · · · Ls 6= 0. Suppose to the contrary that L1 · · · Ls = 0. Let 2 ≤ k ≤ s be the least number such that Li1 · · · Lik = 0 for some {i1, . . . , ik} ⊆ {1, . . . , s}. Then Li1 · · · Lik−1 6= 0, Li1 · · · Lik−2 · Lik 6= 0, and Li1 · · · Lik = 0, imply that χik−1(g∗F ) = χik (g∗F ) for any g∗F ∈ GF by [6, Lemma 4.4]. This is a contradiction since Im ψ generates R⊕s. We return back to the proof of the claim. Suppose to the contrary that L1 · · · Ls+1 = 0. Since L1 · · · Ls 6= 0 and exp χs(f ∗F ) < 1 < ρ(f ∗F ), we have L1 · · · Ls−1 · Ls+1 = 0 by [6, Lemma 4.4]. Repeating this process, we would reach that L1 · Ls+1 = 0, hence L1 and Ls+1 are parallel to each other (cf. [6, Corollary 3.2]). But this is impossible by looking at the action of f ∗F on them. We next claim that the above s = n − 1. Suppose to the contrary that s < n − 1. Extend L1 · · · Ls+1 to a full quasi-nef sequence L1 · · · Lk ∈ L1 · · · Lk−1 · Nef(X) ⊆ H k,k(X, R) (1 ≤ k ≤ n) as in [17, §2.7], such that for any g ∈ G, g∗(L1 · · · Lk) = exp χ1(g) · · · exp χk(g)L1 · · · Lk, where χi : G → (R, +) are group characters. Then the group homomorphism ϕ : G → (R⊕n−1, +), g 7→ (χ1(g), . . . , χn−1(g)) has Ker ϕ = N(G) and image discrete in R⊕n−1. Denote by eN the inverse image π−1(cid:0)N(GF )(cid:1) of N(GF ) under the natural restriction π : G → GF . Then eN/N(G) is a free abelian subgroup of G/N(G) ≃ Z⊕n−1 of rank n − 1 − s, because G/eN ≃ π(G)/π(eN) = GF /N(GF ) ≃ Z⊕s. On the other hand, since the null-entropy group N(GF ) of course fixes those s+1 common nef classes L1, . . . , Ls+1 ∈ F of GF , all the first s + 1 coordinates of ϕ(eN ) are zero. Hence one can embed ϕ(eN) ≃ eN/N(G) into R⊕n−1−(s+1). Recall that Im ϕ is discrete in R⊕n−1. 16 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG Therefore, ϕ(eN) is discrete in R⊕n−s−2, and hence of rank ≤ n− s − 2, which contradicts the rank calculation above. So the claim holds. Therefore, we obtain n common nef classes Li of G, which are of course fixed by the null-entropy group N(G). Thus N(G) fixes the nef class A := L1 +· · · + Ln, which is also big because L1 · · · Ln is nonzero and hence positive. Hence N(G) is virtually contained in Aut0(X) by Corollary 2.2. Then Theorem 4.1 (1) follows from Proposition 2.6. (cid:3) Proof of Theorem 4.1 (2). By the assertion (1), N(G) is virtually contained in Aut0(X), i.e., N(G) : N(G) ∩ Aut0(X) < ∞. Since G/N(G) ∼= Z⊕n−1 and by [3, Lemma 2.4], we only have to consider the case that the group N(G) ∩ Aut0(X) is infinite. We will prove that X is a complex torus in this case. To do so, let H be the identity connected component of the Zariski closure of the latter group in Aut0(X). Then H is normalized by G because N(G) ∩ Aut0(X) is normal in G. Consider the quotient map X 99K X/H (and its graph) as in [10, Lemma 4.2], where G acts on X/H bi-regularly. The maximality of the dynamical rank r(G) of G implies that H has a Zariski-dense open orbit in X, see [17, Lemma 2.14]. Consider the case that the irregularity q(X) = 0. Then Aut0(X) is a linear algebraic group, see [10, Theorem 5.5] or [14, Theorem 3.12]. Hence X is an almost homogeneous variety under the action of a linear algebraic group H (≤ Aut0(X)). Then Aut(X) : Aut0(X) < ∞ by [9, Theorem 1.2]. Hence G ≤ Aut(X) is of null entropy, contradicting the assumption r(G) = n − 1 ≥ 1 on the dynamical rank of G. So we may assume that q(X) > 0. Then the Albanese map albX : X → A := Alb(X) is bimeromorphic by the maximality of r(G), see [17, Lemma 2.13]. Since the action HA of H on A also has a Zariski-dense open orbit in A, we have HA = Aut0(X) ∼= A. Let B ⊂ A be the locus over which albX is not an isomorphism. Note that B and its inverse in X are both H-stable. Since HA = Aut0(X) ∼= A, we have B = ∅. Hence albX is an isomorphism. This proves Theorem 4.1 (2). (cid:3) Proof of Theorem 4.1 (3). Let φ be one of the following natural homomorphisms: G → G/(G ∩ Aut0(X)) or G → GE. Then K := Ker(φ) ≤ Autc(X) for any class c in H p,p(X, R). Choose a class c in the interior of E w p (X). Since Autc(X): Aut0(X) < ∞, by Corollary 2.2, this K is of null entropy, see Proposition 2.6. Thus K ≤ N(G). Now Z⊕n−1 ∼= G/N(G) ∼= (G/K)/(N(G)/K) ∼= φ(G)/φ(N(G)). By Theorem 4.1 (1), φ(N(G)) is a finite group. So there is a finite-index subgroup G1 of G such that φ(G1) ∼= Z⊕n−1, see [3, Lemma 2.4]. This proves Theorem 4.1 (3), hence the whole of Theorem 4.1. (cid:3) Now the finiteness of N(G) in [6, Theorem I] when r(G) = n − 1, has a shorter proof: COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 17 Corollary 4.3. Let X be a compact Kahler manifold of dimension n ≥ 2 and G a group of automorphisms of X. Assume that G is commutative of maximal dynamical rank n−1. Then N(G) is a finite group and hence G is virtually isomorphic to Z⊕n−1. Proof. Suppose the contrary that N(G) is infinite. By Theorem 4.1 (2), X is a complex torus and N(G) ∩ Aut0(X) is Zariski-dense in Aut0(X). Take g0 ∈ G of positive entropy. Since g0 commutes with all elements in N(G), it commutes with any translation Tb ∈ Aut0(X). Write g0 = h ◦ Ta for some translation Ta and group automorphism h for X (here consider X as an additive group). Then for every x ∈ X, we have h(a + b + x) = (g0 ◦ Tb)(x) = (Tb ◦ g0)(x) = b + h(a + x). Hence h(b) = b for all b ∈ X. Thus h = idX. So g0 = Ta, and g0 is of null entropy, a contradiction. This proves Corollary 4.3. (cid:3) Proof of Proposition 1.4. In Theorem 1.1, we can take G1 as the preimage in G of the identity connected component of the Zariski closure of GH 1,1(X,C) in GL(H 1,1(X, C)), via the natural homomorphism G → GH 1,1(X,C), see [17, Theorem 1.2] and Remark 3.10. Theorem 4.1 can be applied to G1 instead of G. So N(G1)H 1,1(X,R) is a finite group and G1/N(G1) ∼= (G1H 1,1(X,R))/(N(G1)H 1,1(X,R)) ∼= Z⊕n−1. Further, G1 E G and G : G1 < ∞. For simplicity, we still use the same notation introduced in the proof of Theorem 4.1. Applying Lemma 3.7 to G1H 1,1(X,R) and the nef cone of X, we may assume that nef classes Li with 1 ≤ i ≤ n are common eigenvectors for the action of G1 on H 1,1(X, R). These n classes Li form a basis of an n-dimensional real vector space that we denote by W ⊆ H 1,1(X, R). For any fixed index 1 ≤ i ≤ n, we can find gi ∈ G1 such that g∗ i Lj = λi,jLj with λi,j < 1, for any j 6= i. This is true for i = n because ϕ(G1) is a spanning lattice in (R⊕n−1, +). The same is true for other i since Pn j=1 exp χj = 1. Since W contains a nef and big class A :=Pn i W ), and hence d1(gi) = maxj{λi,j}. Since λi,j < 1 for j 6= i and λi,i = 1/Qj6=i λi,j > 1, we have λi,i = d1(gi). Further, we have the following uniqueness property by [6, Lemma 4.4 and Corollary 3.2]. j=1 χj = 0 or Qn j=1 Lj, we have d1(gi) = ρ(g∗ Uniqueness Property. Let N be a nef class such that g∗ is parallel to Li. i N = λN with λ > 1. Then N Claim 4.4. For every g ∈ G, g∗ permutes the half-lines R+Li for i = 1, . . . , n. particular, W is invariant by G. In Proof. Since G1 is normal in G, we have hi := g−1gig ∈ G1. It follows that all Lj are eigenvectors of h∗ iW , then h∗ i Lj = λLj for some j. Note that the eigenvalue associated with Lj is always a positive number because Lj is nef. Since W contains big and nef classes, we deduce that i -stable, and if λ is the spectral radius of h∗ i . In particular, W is h∗ 18 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG λ is the first dynamical degree of hi. Moreover, since hi and gi are conjugated, they have the same first dynamical degree, i.e., λ = d1(gi). We deduce from the definition of hi that (g−1)∗Lj is an eigenvector of g∗ i associated with the maximal eigenvalue λ = d1(gi). By the above uniqueness property, (g−1)∗Lj is parallel to Li. It follows that g∗Li is parallel to Lj. This proves the claim. We go back to the proof of Proposition 1.4. By Claim 4.4, we can associate g with an element σg in the symmetric group Sn that we identify with the group of all permutations of the half-lines R+Li. So we have a natural group homomorphism from G to Sn. Let G0 denote its kernel. Recall that we use the basis L1, . . . , Ln for the vector space W . We see that G0 is the set of g ∈ G such that the action of g∗ on W is given by an n × n diagonal matrix with non-negative entries and determinant 1. Indeed, the entries of this matrix are non-negative because the classes Lj are nef; the determinant is 1 because g∗ is identity on H n,n(X, R) and hence preserves the class L1 · · · Ln. We also see that N(G0) is the set of all elements g ∈ G0 which act as the identity on W , see Theorem 2.1. Hence N(G0) is a normal subgroup of G0. Now, G0/N(G0) can be identified with a subgroup of the group of all the diagonal n × n matrices with non-negative entries and determinant 1. The later one is isomorphic to the torsion-free additive group (R⊕n−1, +). Moreover, since G0 contains the finite- index subgroup G1, the group G0/N(G0) contains a finite-index subgroup isomorphic to G1/N(G1) ∼= Z⊕n−1. We necessarily have G0/N(G0) ∼= Z⊕n−1. This completes the proof of Proposition 1.4. (cid:3) Proof of Remark 1.5. As in the proof of Proposition 1.4, let G1 be the preimage in G of the identity connected component of the Zariski closure of GH 1,1(X,C) in GL(H 1,1(X, C)), via the natural homomorphism G → GH 1,1(X,C). Then we have G1/N(G1) ∼= Z⊕n−1 and G : G1 < ∞. Further, by definition, G1 is normalized by H. Now we follow the proof of Proposition 1.4, with G there, replaced by H here. In particular, Claim 4.4 holds for all g ∈ H. In the same way as for G0, we construct a group H0 such that G1 E H0 E H, H/H0 is identified with a subgroup of Sn and H0/N(H0) is isomorphic to a subgroup of the torsion-free additive group (R⊕n−1, +). Hence by [5, Lemma 5.5], H0H 1,1(X,R) is virtually solvable since N(H0)H 1,1(X,R) is virtually unipotent (cf. [3, Theorem 2.2]). So, according to [17, Theorem 1.2], the torsion-free abelian group H0/N(H0) is isomorphic to Z⊕n−1 since it contains G1/N(G1) ∼= Z⊕n−1. By Theorem 4.1, N(H0) and N(G1) are virtually contained in Aut0(X). Thus we have a commutative diagram: H/(H ∩ Aut0(X)) finite-index H0/(H0 ∩ Aut0(X)) finite-kernel / H0/N(H0) ∼= Z⊕n−1 finite-index G/(G ∩ Aut0(X)) finite-index G1/(G1 ∩ Aut0(X)) finite-kernel / G1/N(G1) ∼= Z⊕n−1. ? _ o o / / / ? O O ? O O ? _ o o / / / ? O O COMPACT K AHLER MANIFOLDS ADMITTING LARGE SOLVABLE GROUPS 19 It follows that H/(H ∩ Aut0(X)) is a finite extension of G/(G ∩ Aut0(X)) and both are virtually free abelian groups of rank n − 1, see [3, Lemma 2.4]. If G1 ∩ Aut0(X) (or equivalently N(G1)) is infinite, then it is Zariski-dense in Aut0(X) and X is a complex torus, see Theorem 4.1. The same property holds if we replace G1 by H0. Otherwise, H0 : G1 < ∞. The remark follows. (cid:3) References [1] G. Birkhoff, Linear transformations with invariant cones, Amer. Math. Monthly 74 (1967), 274 -- 276. [2] W. Burnside, On criteria for the finiteness of the order of a group of linear substitutions, Proc. London Math. Soc. 2 (1905), no. 1, 435 -- 440. [3] F. Campana, F. Wang, and D.-Q. Zhang, Automorphism groups of positive entropy on projective threefolds, Trans. Amer. Math. Soc. 366 (2014), no. 3, 1621 -- 1638. [4] J.-P. Demailly, Complex analytic and differential geometry, available online at http://www-fourier.ujf-grenoble.fr/~demailly. [5] T.-C. Dinh, Tits alternative for automorphism groups of compact Kahler manifolds, Acta Math. Vietnam. 37 (2012), no. 4, 513 -- 529. [6] T.-C. Dinh and N. Sibony, Groupes commutatifs d'automorphismes d'une vari´et´e kahl´erienne com- pacte, Duke Math. J. 123 (2004), no. 2, 311 -- 328. [7] T.-C. Dinh and N. Sibony, Dynamics in several complex variables: endomorphisms of projective spaces and polynomial-like mappings, Holomorphic dynamical systems, Lecture Notes in Math., vol. 1998, Springer, Berlin, 2010, pp. 165 -- 294. [8] H. Esnault and V. Srinivas, Algebraic versus topological entropy for surfaces over finite fields, Osaka J. Math. 50 (2013), no. 3, 827 -- 846. [9] B. Fu and D.-Q. Zhang, A characterization of compact complex tori via automorphism groups, Math. Ann. 357 (2013), no. 3, 961 -- 968. [10] A. Fujiki, On automorphism groups of compact Kahler manifolds, Invent. Math. 44 (1978), no. 3, 225 -- 258. [11] M. Gromov, Convex sets and Kahler manifolds, Advances in differential geometry and topology, World Sci. Publ., Teaneck, NJ, 1990, pp. 1 -- 38. [12] M. Gromov, On the entropy of holomorphic maps, Enseign. Math. (2) 49 (2003), no. 3-4, 217 -- 235. Manuscript (1977). [13] J. Keum, K. Oguiso, and D.-Q. Zhang, Conjecture of Tits type for complex varieties and theorem of Lie-Kolchin type for a cone, Math. Res. Lett. 16 (2009), no. 1, 133 -- 148. [14] D. I. Lieberman, Compactness of the Chow scheme: applications to automorphisms and deformations of Kahler manifolds, Fonctions de plusieurs variables complexes, III (S´em. Fran¸cois Norguet, 1975 -- 1977), Lecture Notes in Math., vol. 670, Springer, Berlin, 1978, pp. 140 -- 186. [15] K. Oguiso and T.-T. Truong, Explicit Examples of rational and Calabi -- Yau threefolds with primitive automorphisms of positive entropy, preprint (2013), arXiv:1306.1590. [16] Y. Yomdin, Volume growth and entropy, Israel J. Math. 57 (1987), no. 3, 285 -- 300. [17] D.-Q. Zhang, A theorem of Tits type for compact Kahler manifolds, Invent. Math. 176 (2009), no. 3, 449 -- 459. [18] D.-Q. Zhang, Compact Kahler manifolds with automorphism groups of maximal rank, Trans. Amer. Math. Soc. 366 (2014), no. 7, 3675 -- 3692. [19] D.-Q. Zhang, n-dimensional projective varieties with the action of an abelian group of rank n − 1, Trans. Amer. Math. Soc. (to appear), 2015, arXiv:1412.5779. 20 TIEN-CUONG DINH, FEI HU, AND DE-QI ZHANG Department of Mathematics National University of Singapore, 10 Lower Kent Ridge Road, Singapore 119076 E-mail address: [email protected] E-mail address: [email protected] E-mail address: [email protected]
1003.1968	2	1003	2010-11-13T14:53:49	On tensors of border rank $l$ in $\C^{m\times n\times l}$	[ "math.AG" ]	We study tensors in $\C^{m\times n\times l}$ whose border rank is $l$. We characterize the tensors in $\C^{3\times 3\times 4}$ and in $\C^{4\times 4\times 4}$ of border rank 4 at most.	math.AG	math	On tensors of border rank l in Cm×n×l Department of Mathematics, Statistics and Computer Science Shmuel Friedland University of Illinois at Chicago Chicago, Illinois 60607-7045, USA E-mail: [email protected] November 12, 2010 Abstract We study tensors in Cm×n×l whose border rank is l. We give a set-theoretic char- acterization of tensors in C3×3×4 and in C4×4×4 of border rank 4 at most. Key words: rank of tensors, border rank of tensors, the salmon conjecture. 2000 Mathematics Subject Classification. 14A25, 14P10, 15A69. 1 Introduction T is the minimal number r := rank T , such that T = Pr Denote by Cm×n, S(m, C), Cm×n×l the linear spaces of m × n matrices, m × m symmetric matrices and 3-tensors T = [ti,j,k]m,n,l i=j=k=1 of dimension m × n × l over the field of complex numbers C respectively. We identify Cm×n×l with Cm ⊗ Cn ⊗ Cl. A rank one tensor is u ⊗ v ⊗ w = [uivjwk] ∈ Cm×n×l, where u = (u1, . . . , um)⊤ ∈ Cm, v = (v1, . . . , vn)⊤ ∈ Cn, w = (w1, . . . , wl)⊤ ∈ Cl, and u, v, w are nonzero vectors. The rank of a nonzero tensor i=1 ui ⊗ vi ⊗ wi. The border rank of T 6= 0, denoted by brank T , is a positive integer q, such that the following conditions hold. First, T is a limit of a sequence of tensors Tν ∈ Cm×n×l, rank Tν = q, ν ∈ N. Second, T is not a limit of any sequence of tensors, such that each tensor in the sequence has rank q − 1 at most. For m, n, l ≥ 2 there exists tensors with rank T > brank T . (This inequality does not hold for matrices, i.e. Cm×n.) The maximal border rank of T ∈ Cm×n×l is called the generic rank of Cm×n×l and is denoted by grank(m, n, l). The value of grank(m, n, l) is known for many triples of (m, n, l). The conjectural value of grank(m, n, l) is given in [5]. For r ≤ grank(m, n, l) denote by Vr(m, n, l) ⊂ Cm×n×l the set of all 3-tensors of border rank r at most. It is easy to see that Vr(m, n, l) is an irreducible variety in Cm×n×l, which is a zero set of a number of homogeneous polynomials. In fact, its projectivization is the r-secant variety of Pm−1 × Pn−1 × Pl−1. A general problem is to characterize Vr(m, n, l) in one of the following ways: set-theoretic, scheme theoretic and ideal theoretic. An elegant result of Strassen characterizes V4(3, 3, 3) [16]. It is a hypersurface given by a homogeneous polynomial of degree 9. This paper solves the set-theoretic aspect of the Challenge Problem posed by Elizabeth S. Allman in March 2007 (http://www.dms.uaf.edu/∼eallman/): Determine the ideal defining the fourth secant variety of P3 × P3 × P3. See [1] for more details how this particular problem related to phylogenic ideals and varieties for general Markov models. The Salmon conjecture [15, Conjecture 3.24] stated that V4(4, 4, 4) is defined by polynomials of degree 5 and 9. A first nontrivial step in characterizing V4(4, 4, 4) is to characterize V4(3, 3, 4). It is shown in [10] that V4(3, 3, 4) satisfies certain polynomial equations of degree 6. (See also [11, Remark 5.7] and [14].) Hence the corrected version of the Salmon conjecture states that V4(4, 4, 4) is defined by polynomials of degree 5, 6 and 9 [17, §2]. 1 The first main result of this paper shows that V4(3, 3, 4) is cut out by a set of polynomials of degree 9 and 16. Our second main result shows that V4(4, 4, 4) is cut out by a set of polynomials of degree 5, 9 and 16. Most of the results in this paper are derived from results in matrix theory and relatively basic results in algebraic geometry. Whenever we could, we stated our results in a general setting. We first explain briefly the main steps in the set-theoretic characterization of V4(4, 4, 4). First observe that T ∈ C4×4×4 is given by any four p-slices of T , for p = 1, 2, 3. For example, the (i, 1)-slice of T is Ti,1 = [ti,j,k]4 j,k=1 ∈ C4×4 for i = 1, 2, 3, 4. Let T1(T ) = span(T1,1, . . . , T4,1). Assume first the generic case that dim T1(T ) = 4 and rank T = 4. [5, Theorem 2.4] yields that T1(T ) = span(u1v⊤ 4 ). Assume the generic case that u1, . . . , u4 ∈ C4 and v1, . . . , v4 ∈ C4 are linearly independent. Let Q ∈ GL(4, C) satisfy Qui = wi, v⊤ i wj = δij, i, j = 1, . . . , 4. Then any two matrices in QT1(T ) commute. (This result is well known to the experts, e.g. [4] and references therein.) This result is equivalent to the statement that for any X, Y, Z ∈ T1(T ) the following condition holds. 1 , . . . , u4v⊤ X(adj Y )Z − Z(adj Y )X = 0. (1.1) (adj Y is the adjoint matrix of Y .) These identities give rise to a system of homogeneous equations of degrees 5 in the entries of X, Y, Z, which always hold if brank T ≤ 4. Vice versa, if the above equalities hold and T1(T ) contains an invertible matrix then the results in [7] yields that brank T ≤ 4. We next consider the case where T1(T ), T2(T ), T3(T ) does not contain an invertible matrix, and each three matrices in Ti(T ) satisfies (1.1) for i = 1, 2, 3. In §5 we show that either brank T ≤ 4 or by permuting factors in C4 ⊗ C4 ⊗ C4, if necessary, and changing bases in the first two components of C4 × C4 × C4 T can be viewed as a tensor C3×3×4. This is Corollary 5.6 of [11]. In [11] this corollary is deduced from [11, Prop. 5.4]. Unfortunately, Proposition 5.4 is wrong, see §5. Assume that T ∈ V4(3, 3, 4) and rank T = 4. Then T3(T ) ⊂ C3×3 is spanned by uiv⊤ i for i = 1, . . . , 4. Assume the generic case where any three vectors out of u1, . . . , u4 and out of v1, . . . , v4 are linearly independent. Then there exists L, R ∈ GL(3, C) such that LT3(T ) and T3(T )R are 4 dimensional subspaces of the subspace of 3×3 symmetric matrices S(3, C), which are spanned by 4 rank one symmetric matrices. Furthermore L, R are unique up to a multiplication by a nonzero scalar, and LR⊤ = R⊤L = tr(LR⊤) 3 I3. (1.2) The existence of nonzero L, R such that LT3(T ), T3(T )R ⊂ S(3, C) is equivalent to the condition that the corresponding system of homogeneous linear equations in the entries of L and R respectively, given respectively by the coefficient matrices CL(T ), CR(T ) ∈ C12×9, have nontrivial solution. (Note that L and R have 9 entries.) The entries of CL(T ) and CR(T ) are linear combinations of the entries of T with coefficients 0, 1, −1. A necessary and sufficient condition for a nontrivial solution R and L is that all 9 × 9 minors of CL(T ) and CR(T ) are zero. These gives rise to a number of polynomial equations of degree 9 that the entries of T ∈ C3×3×4 must satisfy. We show that the Strassen condition corresponds to some of the above polynomial equations. If the ranks of CL(T ) and CR(T ) is 8 then L and R are determined uniquely up to a multiplication of a nonzero scalar. The entries of L and R are polynomials of degree 8 in the entries of CL(T ) and CR(T ). The condition (1.2) translates to a system of polynomial equations of degree 16 in the entries of T . Assume first that L is invertible. Then LT3(T ) is a 4-dimensional subspace of S(3, C). A 4-dimensional generic subspace of S(3, C) spanned by 4 rank one matrices. Hence the assumption that L ∈ GL(3, C) yields that T ∈ V4(3, 3, 4). (This case does not require (1.2).) Assume that neither L or R are invertible. Then LR = 0. In that case we also show that the above conditions imply that T ∈ V4(3, 3, 4), by considering a few cases. (We need (1.2) to rule out certain cases.) 2 We survey briefly the contents of the paper. In §2 we discuss some known results which are needed in the next sections. We recall a simple known condition that rank T is the dimension of the minimal subspace spanned by rank one matrices that contains the subspace spanned by p-slices of T , denoted by Tp(T ), for each p = 1, 2, 3. Next we discuss a simple dimension condition on a generic subspace in U ⊂ Cm×n which implies that U is spanned by rank one matrices. We translate this result to the border rank of T . We recall the Strassen characterization of V4(3, 3, 3). Next we show that for a generic tensor T ∈ Cm×m×l of rank m one can change a basis in the first factor of Cm such that the l 3-slices of T are commuting matrices. These conditions give rise to the equations of type (1.1). In §3 we characterize subspaces U ⊂ Cm×m such that any 3 matrices satisfy the condition (1.1) and most of the matrices in U have rank m − 1. In §4 we characterize V4(3, 3, 4). §5 we discuss necessary and sufficient conditions for a T ∈ C4×4×4 to have border rank 4 at most. The nontrivial case is when T1(T ), T2(T ), T3(T ) do not contain an invertible matrix, and T can not be viewed as a tensor in C3×3×4. We analyze the case where most of the matrices in T3(T ) have rank 2. In this case the condition (1.1) holds trivially for T3(T ). We show that in some cases brank T = 5 if we do not assume the conditions (1.1) for T1(T ). This gives a counter-example to [11, Proposition 5.4], and invalidates the proof of [11, Corollary 5.6]. On the other hand we show that if (1.1) holds for T1(T ) and of the matrices in T3(T ) have rank 2, then brank T ≤ 4. Most of §5 is devoted to the proof of [11, Corollary 5.6]. We need to consider the case where most of the matrices in Tp(T ) have rank 3 for p = 1, 2, 3. Our analysis depends on the results in §3. In §6 we outline how to translate the problem of classifying tensors in Cm×n×l of rank l if either 2 ≤ l ≤ m, n or m = n = l − 1 and l ≥ 4. It turns out that in the generic case this problem boils down to the condition that a corresponding l subspace denote by S(T3(T )) ⊂ S(l, C) is congruent to a subspace spanned by l diagonal matrices. Note that the a simultaneous matrix diagonalization by congruence arises naturally in finding the rank decomposition of tensors [4]. We point out how some of these results can be generalized to tensors of border rank l at most. 2 Preliminary results We first recall a basic result on the rank of 3 tensor T = [ti,j,k]m,n,l is well known to the experts. denote the matrix Tk,3(T ) = Tk,3 := [ti,j,k]m,n span(T1,3, . . . , Tl,3) ⊂ Cm×n. We call T3(T ) the 3 − rd subspace induced by T . i=j=k ∈ Cn×m×l which (See for example [5, Theorem 2.4].) By a (k, 3)-slice, we i=j=1 ∈ Cm×n for k = 1, . . . , l. Let T3(T ) := Theorem 2.1 Let T ∈ Cm×n×l. Then rank T is the minimal dimension of a subspace U ⊂ Cm×n that contains T3(T ) and is spanned by rank one matrices. We can define similarly (p, 1) and (q, 2) slices of T and the corresponding subspaces T1(T ), T2(T ). Hence Theorem 2.1 can be stated for T1(T ) and T2(T ) respectively. Also note that the space Cm×n×l can be identified with Cm ⊗ Cn ⊗ Cl, and hence if we permute the three tensor factors Cm, Cn, Cl we obtain 6 isomorphic spaces of tensors. Let P = [pi′i] ∈ GL(m, C), Q = [qj′j] ∈ GL(n, C), R = [rk′ k] ∈ GL(l, C). Then we can change the standard bases in Cm, Cn, Cl to the bases obtained from the columns of P −1, Q−1, R−1 respectively. In the new bases T is represented by T ′ = [ti′,j′,k′ ] = T (P, Q, R). So T (P, Q, R) = P ⊗ Q ⊗ R(T ) := T ′ = [t′ i′,j′,k′ ], ti′,j′,k′ = m,n,l Xi=j=k=1 pi′iqj′jrk′ kti,j,k. (2.1) Clearly rank T (P, Q, R) = rank T . The following lemma is derived straightforward. 3 Lemma 2.2 Let T ∈ Cm×n×l, P ∈ GL(m, C), Q ∈ GL(n, C), R ∈ GL(l, C). Let T (P, Q, R) be defined as in (2.1). Then T1(T (P, Q, R)) = QT1(T )R⊤, T2(T (P, Q, R)) = P T2(T )R⊤, T3(T (P, Q, R)) = P T3(T )Q⊤. For a finite dimensional space W of dimension N denote by Gr(p, W) the Grassmannian variety of all subspaces of dimension 1 ≤ p ≤ N . Recall that dim Gr(p, W) = p(N − p). Let Γ(p, Cm×n) ⊆ Gr(p, Cm×n) and Γ(p, S(m, C)) ⊆ Gr(p, S(m, C)) be the varieties of all p- dimensional subspaces in Cm×n and S(m, C) that can be obtained as limit of p-dimensional subspaces in Cm×n and S(m, C) respectively, which are spanned by rank one matrices. For integers i ≤ j denote by [i, j] the set of all integers k, i ≤ k ≤ j. The following result is known, e.g. [3, Prop, 3.1, (iv)]. Lemma 2.3 For 1 < m, n ∈ N Γ(p, Cm×n) = Gr(p, Cm×n) for p ∈ [(m − 1)(n − 1) + 1, mn], Γ(p, S(m, C)) = Gr(p, S(m, C)) for p ∈ [(cid:18)m 2(cid:19) + 1,(cid:18)m + 1 2 (cid:19)] (2.2) (2.3) Proof. To prove (2.2) it is enough to show the case p = (m − 1)(n − 1) + 1. Clearly Γ(p, Cm×n) is an irreducible variety of Gr(p, Cm×n). It is left to show that dim Γ(p, Cm×n) = dim Gr(p, Cm×n). Let PV (r, m, n) ⊂ Pmn−1 be the projectivized variety of all matrices in Cm×n \ {0} of rank r at most. It is well known that dim PV (r, m, n) = r(m + n − r) − 1, e.g. [6]. Hence any generic projective linear subspace of dimension (m − 1)(n − 1) in Pmn−1 will intersect the Segre variety PV (1, m, n) = Pm−1 × Pn−1 at a fixed number of points [6, §6] deg Pm−1 × Pn−1 = (cid:18)m + n − 2 m − 1 (cid:19). (2.4) Thus, a generic projective subspace spanned by (m − 1)(n − 1) + 1 points on Segre variety, i.e. (m − 1)(n − 1) + 1 rank one matrices. Therefore dim Γ((m − 1)(n − 1) + 1, Cm×n) = ((m − 1)(n − 1) + 1)(m + n − 2) = dim Gr((m − 1)(n − 1) + 1, Cm×n) ⇒ Γ(p, Cm×n) = Gr((m − 1)(n − 1) + 1, Cm×n). variety of all A ∈ S(m, C) \ {0} of rank r at most. It is well known that codim PS(r, m) = To prove (2.3) is enough to show the case p = (cid:0)m 2(cid:1) + 1. Let PS(r, m) be the projectivized (cid:1) in PS(m, C), e.g. [6]. Hence a generic projective linear subspace of dimension (cid:0)m 2(cid:1) in PS(m, C) intersects PS(1, m) at a fixed number of points [6]: (cid:0)m−r+1 2 (cid:0) m+j m−1−j(cid:1) (cid:0)2j+1 j (cid:1) Similar arguments for the previous case show that Γ((cid:0)m deg PS(1, m) = Yj=0 m−2 . 1, S(m, C)) have the same dimensions. Hence the two varieties are equal. (2.5) 2(cid:1) + 1, S(m, C)) and Gr((cid:0)m 2(cid:1) + ✷ Lemma 2.4 Let 0 6= T ∈ Cm×n×l. Then r ≥ brank T , if there exists U ∈ Γ(r, Cm×n) such that U ⊇ T3(T ). Furthermore, r = brank T if there is no V ∈ Γ(r − 1, Cm×n) such that V ⊇ T3(T ). In particular, brank T ≥ dim T3(T ). Proof. Suppose that rank T = r. Then Theorem 2.1 yields the existence U ∈ Γ(r, Cm×n) such that U ⊇ T3(T ). Suppose now that T is a limit of T ′ q , q ∈ N of rank r′ ≤ r. It is trivial to see that T is a limit of Tq, q ∈ N of rank r. Then T3(Tq) ⊆ Uq ∈ Γ(r, Cm×n) for each q ∈ N. Take a convergent subsequence Uqs → U ∈ Γ(r, Cm×n). So Tk,3(T ) ∈ U, k = 1, . . . , l. Hence 4 T3(T ) ⊆ U. If brank T = r there is no V ∈ Γ(r − 1, Cm×n) such that V ⊇ T3(T ). Clearly, if U ⊇ T3(T ) then dim U ≥ dim T3(T ). Hence brank T ≥ dim T3(T ). ✷ k k l [12, 2]. For m ∈ N let hmi := {1, . . . , m}. For k ∈ hmi denote by 2hmi of hmi of cardinality k. Then α ∈ 2hmi We now recall some basic results for matrices we need here. Consult for example with the set of subsets is viewed as α = {α1, . . . , αk}, where 1 ≤ α1 < . . . < αk ≤ m. Denote kαk := Pk i=j=1 ∈ Cm×n, α ∈ 2hmi i=j=1 ∈ Ck×l. Recall that for p ∈ hmin(m, n)i the k p − th compound of A, denoted as Cp(A) ∈ C(m p), is a matrix whose rows and columns are indexed by α ∈ 2hmi and its (α, β) entry is given by det A[α, β]. If we view A as a linear transformation A : Cn → Cm given by x 7→ Ax, then Cp(A) represents j=1 αj , αc := hmi\α. For A = [ai,j]m,n denote A[α, β] := [aαi,βj ]k,l , β ∈ 2hni , β ∈ 2hni p )×(n p p Vp A : Vp Cn → Vp Cm in the corresponding bases. Clearly Cp(A⊤) = Cp(A)⊤, Cp(Im) = I(m p ). (Here Im is m × m identity matrix.) The Cauchy-Binet formula yields Cp(AB) = Cp(A)Cp(B) for any A ∈ Cm×n, B ∈ Cn×l. (2.6) Let m = n. For p ∈ hm − 1i denote by C−p(A) a matrix whose rows and columns are indexed by α ∈ 2hmi and its (α, β) entry is given by (−1)kαk+kβk det A[αc, βc]. So C−1(A)⊤ is the adjoint of A, denoted as adj A. The Laplace expansion yields that , β ∈ 2hmi p p Cp(A)C−p(A)⊤ = C−p(A)⊤Cp(A) = (det A) I(m p ), p = 1, . . . , m − 1. (2.7) Hence Cp(A−1) = 1 det A C−p(A)⊤ for A ∈ GL(m, C). (2.8) We now recall Strassen's result [16]. Theorem 2.5 Let T = [ti,j,k]3 i=j=1 ∈ C3×3, k = 1, 2, 3. Let f (X1, X2, X3) := det(X1(adj X2)X3 − X3(adj X2)X1) be a polynomial of degree 12 in the entries of the matrices X1, X2, X3. Then f (X1, X2, X3) = s(X1, X2, X3) det X2. The variety of all T of border rank 4 at most is a hypersurface in C3×3×3 of degree 9 given by equation s(X1, X2, X3) = 0. i=j=k ∈ C3×3×3. Denote Xk := [ti,j,k]3 The following result is straightforward. Lemma 2.6 . Let A ∈ Cm×n and assume that rank A ≤ k < n. Fix α ∈ 2hmi , β = {β1, . . . , βk+1} ∈ 2hni k+1. Let x(α, β) = (x1, . . . , xn)⊤ ∈ Cn be defined as follows. xj = 0 if j 6∈ β. If j = βi then xj = (−1)i−1 det A[α, β\{βi}]. Then Ax(α, β) = 0. Furthermore, x(α, β) 6= 0 for some α, β if and only if rank A = k. k The following result is known [11]. Theorem 2.7 Let T ∈ Cm×m×l. Assume that brank T ≤ m. Then for any A, B, C ∈ T3(T ) the following equalities hold. Cp(A)C−p(B)⊤Cp(C) = Cp(C)C−p(B)⊤Cp(A) for p = 1, . . . , m − 1. (2.9) Proof. Assume that rank T = m and T3(T ) contains an invertible matrix B. So U = span(u1v⊤ i . Since rank B = m we have that bi 6= 0, i = 1, . . . , m, and u1, . . . , um and v1, . . . , vm linearly independent. There exist P, Q ∈ GL(m, C) such that P ui = Qvi = ei := (δi1, . . . , δim)⊤, i = 1, . . . , m. So any m) ⊃ T3(T ). Hence B = Pm 1 , . . . , umv⊤ i=1 biuiv⊤ 5 matrix in P UQ⊤ is a diagonal matrix. Hence P AQ(P BQ)−1P CQ = P CQ(P BQ)−1P AQ, i.e. AB−1C = CB−1A, (2.10) for any A, B, C ∈ U. Take the p − th compound of the above equality, use (2.6) and (2.8) to deduce (2.9) for any A, B, C ∈ U. Since almost any B ∈ U is invertible (2.9) holds for any A, B, C ∈ U. In particular it holds for any A, B, C ∈ T3(T ). Assume now that brank T ≤ m. Then T is a limit of Tq, q ∈ N of rank m. Furthermore, it is easy to see that we can assume that each T3(Tq) contains an invertible matrix. Let T3(Tq) ⊂ Uq, where Uq is a span of m matrices of rank one. Hence any three matrices A, B, C ∈ Uq satisfy (2.9). Assume that Uq, q ∈ N converges to U ∈ Γ(m, Cm×n). Then any 3 matrices in U satisfy (2.9). The proof of Lemma 2.4 yields that T3(T ) ⊂ U. Hence any 3 matrices in T3(T ) satisfy (2.9). ✷ The following result is well known, e.g. [5]. brank T ≤ min(n, 2m) for any T ∈ C2×m×n where 2 ≤ m ≤ n. (2.11) 3 Some subspaces of singular matrices satisfying (1.1) For a subspace U ⊂ Cm×n define mrank U = {max rank A, A ∈ U}. The following theorem analyzes the condition (1.1) for a subspace U ⊂ Cm×m satisfying mrank U = m − 1. Theorem 3.1 Let U ⊂ Cm×m and assume that mrank U = m − 1. Then any three matrices in U satisfy (1.1) if and only if one of the following mutually exclusive conditions hold. 1. There exists a nonzero u ∈ Cm such that either Uu = 0 or u⊤U = 0⊤. 2. m ≥ 3, dim U = k + 1 ≥ 2. There exists P, Q ∈ GL(m, C) such that P UQ has a following basis F0, . . . , Fk. The last row and column of F0, . . . , Fk−1 is zero, i.e. Fi = Gi ⊕ 0, Fi ∈ C(m−1)×(m−1), i = 0, . . . , k − 1, G0 = Im−1, and Fk = (cid:20) Gk g⊤ 0 (cid:21) , Gk ∈ C(m−1)×(m−1), 0 6= f , g ∈ Cm−1, g⊤f = 0. f (3.1) Furthermore there exists two subspace X, Y ⊂ C(m−1) with the following properties f ∈ X, g ∈ Y, g⊤X = f ⊤Y = 0⊤, GkX ⊆ X, G⊤ GiX = 0, G⊤ k Y ⊆ Y, i Y = 0, i = 1, . . . , k − 1. (3.2) (3.3) Proof. Let A ∈ U, rank A = m − 1. Then adj A = u(A)v(A)⊤ for some nonzero u(A), v(A) ∈ Cm. Since A(adj A) = (adj A)A = 0 we deduce that Au(A) = A⊤v(A) = 0. Suppose first that Uu = 0 for some nonzero u ∈ Cm. So for each A ∈ U, rank A = m − 1 we must have that span(u(A)) = span(u). So we may assume that u(A) = u. Hence for any B ∈ U Badj (A) = 0. Since adj (Y ) = 0 if rank Y < m − 1, we deduce that any three matrices in U satisfy (1.1). Similarly, (1.1) holds if there exists nonzero u such that u⊤U = 0. Assume now that condition 1 does not hold. Then for most of B ∈ U Bu(A) 6= 0, B⊤v(A) 6= 0. (3.4) Assume now that (1.1) holds. Let X = B, Y = A, Z = C, and B, C satisfy (3.4). Then span(Bu(A)) = span(Cu(A)) = span(x(A)), span(B⊤v(A)) = span(C⊤v(A)) = span(y(A)), 6 for some nonzero x(A), y(A) ∈ Cm. Hence, there exists two nontrivial linear functionals φ, ψ : U → C, depending on A, such that Bu(A) = φ(B)x(A), B⊤v(A) = ψ(B)y(A) for all B ∈ U. Using (1.1) for X = B, Y = A, Z = C and the above assumptions we obtain the equality φ(B)ψ(C) = φ(C)ψ(B). Choosing B, C satisfying (3.4) we get that φ(B) ψ(C) . Hence ψ(B) = aφ(B) for all B ∈ U, and a 6= 0. By replacing y(A) by ay(A) we may assume that ψ = φ. Hence for each A ∈ U, rank A = m − 1 we have the equality ψ(B) = φ(C) Bu(A) = φA(B)x(A), B⊤v(A) = φA(B)y(A) for all B ∈ U, (3.5) for a corresponding nontrivial linear functional φA : U → C. Choose P, Q ∈ GL(m, C) such that F0 = Im−1 ⊕ 0 ∈ U′ = P UQ. Note that u(F0) = v(F0) = em = (δ1m, . . . , δmm)⊤. Then there exists a nonzero linear functional φ0 : U′ → C such that Bem = φ0(B)x0, B⊤em = φ0(B)y0 for some nonzero x0, y0 ∈ Cm and B ∈ U′. Observe that for any B ∈ Cm×m we have det(F0 + tB) = tb + O(t2), where b is the (m, m) entry of B. Hence b = 0 for B ∈ U′. Since (m, m) entry of B is zero it follows that x0, y0 0 = (g⊤, 0). Consider next the strict have the last coordinate zero, i.e. x⊤ subspace of U0 ⊂ U′ satisfying φ0(B) = 0. For B ∈ U0 we have that Bem = B⊤em = 0, i.e. the last row and column matrices in U0 are zero. Clearly F0 ∈ U0. Let F0, . . . , Fk−1 be a basis in U0. Let φ0(Fk) = 1. The assumption that φ0(Fk) = 1 yields that Fk is of the form given in (3.1). (We will show the condition g⊤f = 0 later.) So F0, . . . , Fk is a basis of U′. Let 0 = (f ⊤, 0), y⊤ F (z) = F0 + k Xi=1 ziFi, G(z) = Im−1 + k Xi=1 ziGi, z = (z1, . . . , zk)⊤ ∈ Ck. Assume that G(z) is invertible. A straightforward calculation shows u(F (z))⊤ = (zkf (z)⊤, −1), v(F (z))⊤ = (det G(z))(zkg(z)⊤, −1), where f (z) = G(z)−1f , g(z)⊤ = g⊤G(z)−1, and g⊤G(z)−1f = 0. (3.6) Indeed, the equalities F (z)u(z) = F (z)⊤v(z) = 0 are verified straightforward. The equality g⊤G(z)−1f = 0 must hold if zk 6= 0. The continuity argument yields this condition for zk = 0 if G(z) is invertible. Note that the condition g⊤G(z)−1f = 0 for z = 0 yields that g⊤f = 0. To see that adj F (z) = u(z)v(z)⊤ just observe that the (m, m) entry of adj F (z) = det G(z). Observe next that (3.5) yields G(w)u(z) = φz(w)x(z), (3.7) for some nonzero affine functional φz(w). (φz(w) is affine since in the definition of F (z) the coefficient of A0 is 1.) If zk 6= 0 we claim that we can choose x(z)⊤ = (f (z)⊤, 0) (3.8) Indeed, chose w = 0 so G(0) = Im−1. Clearly F (0)u(z) = (zkf (z)⊤, 0)⊤ = zk(f (z)⊤, 0)⊤. Let w = (w1, . . . , wk−1, 0)⊤. Then we get the equality F ( w)u(z) = (cid:20) zkG( w)f (z) 0 (cid:21) = φz( w)(cid:20) f (z) 0 (cid:21) . (3.9) If zk 6= 0 then f (z) is an eigenvector of G( w) for each w. Assume that G(z) is invertible for any z satisfying satisfying kzkmax ≤ r for some r > 0. Use the continuity argument to deduce that f (z) is an eigenvector of G( w) for any z satisfying satisfying kzkmax ≤ r. Letting z = 0 we get that f = f (0) is an eigenvector for each G( w). Hence Gif = λif for 7 i = 0, . . . , k − 1, where λ0 = 1. By replacing Gi with Gi − λiIm−1 we may assume without loss of generality that Gif = 0 for i = 1, . . . , k − 1. Let kzkmax ≤ r. Since f (z) is an eigenvector of Gi and f (0) corresponds to the zero eigenvalue of Gi it follows Gif (z) = 0, were i = 1, . . . , k − 1. Let z = (0, . . . , 0, zk)⊤ and zk < r. So f (z) = (Im−1 + zkGk)−1f = ∞ Xj=0 (−zk)jGj kf . Let X, Y be the cyclic subspaces spanned by Gj k )j g, j = 0, . . . , re- spectively. Clearly, GkX ⊆ X, G⊤ k Y ⊆ Y. The condition that Gif (z) = 0 yields that GiX = 0 for i = 1, . . . , k − 1. So G( w)f (z) = f (z). The condition g⊤f (z) = 0 yields that g⊤X = f ⊤Y = 0⊤. Observe next that (3.9) yields that φz( w) = zk. In view of (3.5) it follows that kf , j = 0, . . . , and (G⊤ F ( w)⊤v(z) = zky(z), y(z)⊤ = a(z)(g(z)⊤, 0) for some 0 6= a(z) ∈ C. Hence G⊤ the theorem. i g(z) = 0 and G⊤ i Y = 0 for i = 1, . . . , k − 1. This establishes the conditions 2 of Vice versa, suppose that the conditions 2 of the theorem hold. Let U′ = P UQ. Define F (z), G(z), u(z), v(z), f (z), g(z) as above. It is enough to show the condition (3.5), where A = F (z), B = F (w), C = F (w′)) and det G(z) 6= 0. Observe next that (3.2-3.3) yield that f (z) = (Im−1 + zkGk)−1f , g(z)⊤ = g⊤(Im−1 + zkGk)−1, g⊤f (z) = f ⊤g(z) = 0. Then Similarly, F (w)u(z) = ((zk(Im−1 + wkGk)(Im−1 + zkGk)−1f − wkf )⊤, 0)⊤ = (zk − wk)(f (z)⊤, 0)⊤. v(z)⊤F (w′) = (zk − w′ k) det G(z)(g(z)⊤, 0). Hence the condition (1.1) holds for Y = F (z), X = F (w), Z = F (w′) when det G(z) 6= 0. Since most of F (z) ∈ U′ satisfy the condition that det G(z) 6= 0 we deduce that (1.1) for each X, Y, Z ∈ U′. Theorem 3.2 A tensor T ∈ C3×3×3 has border rank 3 at most if and only if Tp(T ) satisfies the condition (1.1) for some p ∈ {1, 2, 3}. Proof. Theorem 2.7 implies that each Tp(T ) satisfies the condition (1.1). It is enough to consider the case where T3(T ) 6= {0} satisfies the condition (1.1). Suppose first that dim Tp(T ) ≤ 2 for some p ∈ {1, 2, 3}. Then by changing basis in the p-th component of C3 ⊗ C3 ⊗ C3 and interchanging the first and the p-th component, we can assume that T as 2 × 3 × 3 tensor. (2.11) yields that brank T ≤ 3. Assume now that dim Tp(T ) = 3 for p = 1, 2, 3. Suppose first that max rank T3(T ) = 1. So T3(T ) has a basis consisting of rank one matrices. Theorem 2.1 implies that rank T = 3, hence 3 ≥ brank T . Assume now that mrank T3(T ) = 3, i.e. there exists an invertible Y ∈ T3(T ). By considering P = Y −1 and changing a basis in the first factor of C3 ⊗ C3 ⊗ C3 we may that Y = I ∈ T3(T ). Let T3 = span(I, A1, A2). So A1A2 = A2A1. Recall that the variety of all commuting pairs (X1, X2) ∈ (C3×3)2 is irreducible [13]. Hence a pair (A1, A2) is a limit of generic commuting pairs (X1, X2). For a generic pair, X1 has 3 distinct eigenvalues. So X2 is a polynomial in X1. Thus, there exists Q ∈ GL(3, C) such that Q−1span(X1, X2)Q is a two dimensional subspace of 3 × 3 diagonal matrices D ⊂ C3×3. Clearly I ∈ D and D is spanned by 3 rank one diagonal matrices. Hence span(I, X1, X2) ⊆ Q−1DQ ∈ Γ(3, C3×3). Thus T3(T ) ⊆ U ∈ Γ(3, C3×3), and brank T ≤ 3. 8 Assume now that mrank T3(T ) = 2. We claim that there is no nonzero u ∈ C3 such that either T3(T )u = 0 or u⊤T3(T ) = 0⊤. Assume to the contrary that u⊤T3(T ) = 0⊤ for some nonzero u. By change of basis in the first component of C3 we may assume that u = e3. Hence the third row of each matrix in T3(T ) is zero. Hence T3,1 = 0, i.e. dim T1(T ) ≤ 2 contradicting our assumptions. Similarly there is no nonzero u such that T3(T )u = 0. Hence U = T3(T ) satisfies the condition of part 2 of Theorem 3.1. Since m = 3 it follows that the subspaces X, Y ⊂ C3 are one dimensional. By changing bases in C2 we can assume that f = e1, g = e2. Hence the three 3-slices of T are F0 =   1 0 0 1 0 0 0 0 0   , F1 =   0 1 0 0 0 0 0 0 0   , F3 =   ∗ 0 0 ∗ 1 ∗ 0 1 0 .   Do the following elementary row and column operations on F2 to bring it to the form F ′ 2 =   0 0 0 0 0 1 1 0 0  . First subtract a mulitple of the third row from the second and first row. Then subtract a multiple of the third column from the first column. Apply the same row and column operations on F0 and F1 to obtains the three 3-slices F0, F1, F ′ 2 of the tensor T ′. Consider the three 2-slices T ′: T1,2 =   1 0 0 0 0 0 0 0 0   , T2,2 =   0 1 1 0 0 0 0 0 1   , T3,2 =   0 0 0 0 1 0 0 0 0 .   Interchange the first two columns in each of the above matrices to obtain the matrices A1, I, A2. Note that A1A2 = A2A1 = 0. The previous arguments show that brank T = brank T ′ ≤ 3. ✷ We conclude this section with the following proposition. Proposition 3.3 Let T ∈ Cm×m×m. Assume that dim T3(T ) ≤ m − 1. Then any three matrices A, B, C ∈ Tk(T ) satisfy the conditions (2.9) for p = 1, m − 1 and k = 1, 2. For m = 4 the condition (2.9) holds also for p = 2 and k = 1, 2. Proof. By changing a basis in the last component of Cm ⊗ Cm ⊗ Cm we may assume that Tm,3 = 0. Hence the last row of each Ti,1 and the last column of each Ti,2 is zero. Theorem 3.1 yields that any three matrices in T1(T ), T2(T ) satisfy the conditions (1.1). Note that for any two matrices A, B ∈ Tk(T ) either A(adj B) = 0 or (adj B)A = 0. Hence for any three matrices in Tk(T ) (2.9) holds for p = m − 1 and k = 1, 2. Assume now that m = 4. We now show that for any three matrices in Tk(T ) and k = 1, 2 we have the equality C2(A)C−2(B)⊤C2(C) = 0. It is enough to show this identity for k = 1. Since the last row of A ∈ T1(T ) is zero, it follows that C2(A) has three zero rows labeled (1, 4), (2, 4), (3, 4). Hence the zero rows of C−2(B) are the rows (1, 2), (1, 3), (2, 3). So C−2(B)⊤ has three zero columns (1, 2), (1, 3), (2, 3). A straightforward calculation shows that C2(A)C−2(B)⊤C2(C) = 0. Hence C2(A)C−2(B)⊤C2(C) = C2(C)C−2(B)⊤C2(A) = 0. ✷ For m = 4 it seems to us that the condition (1.1) always implies the conditions (2.9) for p = 2, 3. 4 Tensors in C(m−1)×(m−1)×m of border rank m Let T ∈ C(m−1)×(m−1)×m be of rank m. So T = m Xi=1 ui ⊗ vi ⊗ wi, ui, vi ∈ Cm−1, wi ∈ Cm, i = 1, . . . , m. 9 We call T of rank m generic if any m − 1 vectors out u1, . . . , um and v1, . . . , vm are linearly independent. Lemma 4.1 Let T = [ti,j,k] ∈ C(m−1)×(m−1)×m be a generic rank m tensor. Then there exists unique L, R 6= 0 (up to a nonzero scalars), such that LT3(T ) ⊂ S(m − 1, C), T3(T )R ⊂ S(m − 1, C), LR⊤ = R⊤L = ( 1 m − 1 trace (LR⊤))Im−1. (4.1) (4.2) Furthermore L, R ∈ GL(m − 1, C). Proof. Let U, V ∈ GL(m−1, C) such that U ui = V vi = ei, i = 1, . . . , m−1. Let U um = x = (x1, . . . , xm−1)⊤, V vm = y = (y1, . . . , ym−1)⊤. The assumption that any m − 1 vectors from u1, . . . , um and v1, . . . , vm are linearly independent imply that all the coordinates of x and y are nonzero. Hence there exists a diagonal D ∈ GL(m − 1, C) such that Dx = D−1y. So (Dei)(D−1ei)⊤, i = 1, . . . , m − 1 are m − 1 commuting diagonal matrices. Furthermore the matrix (Dx)(D−1y)⊤ is symmetric. Hence DU T3(T )V ⊤D−1 ⊂ S(m − 1, C). Thus L = V −1D2U and R = (L−1) will satisfy the conditions of the lemma. It is left to show that L and R are unique up to a multiple of a nonzero constant. For that we may assume i , i = 1, . . . , m − 1 and zz⊤ for some z with nonzero already that T3(T ) is spanned by eie⊤ coordinates. The assumptions that Leie⊤ is symmetric for i = 1, . . . , m − 1 yields that L is i a diagonal matrix. The assumption that Lzz⊤ is symmetric implies that L = dIm−1. So if L 6= 0 then it is a nonzero multiple of Im−1. Similar results hold for R. In particular, R⊤ is an inverse of L times a nonzero constant. ✷ ⊤ Lemma 4.2 Let T = [ti,j,k] ∈ C(m−1)×(m−1)×m be a border rank m at most. Then there exist L, R ∈ C(m−1)×(m−1)\{0} such that (4.2) holds. Proof. There exist a sequence of Tk ∈ C(m−1)×(m−1)×m of rank m at most that con- verge to T . By perturbing each Tk we can assume that each Tk ia generic tensor of rank m. So there exists Lk, Rk ∈ GL(m − 1, C) satisfying (4.2). Normalize Lk, Rk to have k) = 1. Since the set {A ∈ C(m−1)×(m−1), trace (AA∗) = 1} is trace (LkL∗ compact, there exists a subsequence kp, p ∈ N, such that Lkp → L, Rkp → R and T3(Tkp ) converges to U ∈ Γ(m, Cm×m). Clearly LU, UR ⊂ S(m, C), and L, R satisfy the equality in (4.1)-(4.2). As U ⊇ T3(T ) we deduce the lemma. ✷ k) = trace (RkR∗ Lemma 4.3 Let T = [ti,j,k] ∈ C(m−1)×(m−1)×r, where 3 ≤ r. Denote by Tk := [ti,j,k]m−1 i=j=1 the (k, 3)-slice of T for k = 1, . . . , r. Then the two systems TkR − R⊤T ⊤ LTk − T ⊤ k = 0, k = 1, . . . , r, R ∈ C(m−1)×(m−1), k L⊤ = 0, k = 1, . . . , r, L ∈ C(m−1)×(m−1) (4.3) (4.4) have nontrivial solutions R, L if and only if the following conditions hold. Let CR(T1, . . . , Tr), CL(T1, . . . , Tr) ∈ Cr(m−1)2×(m−1)2 be the coefficient matrices of the systems (4.3) and (4.4) in (m − 1)2 variables, (the entries of R and L respectively), and r(cid:0)m−1 2 (cid:1) equations. Then rank CR(T1, . . . , Tr) < (m − 1)2 and rank CL(T1, . . . , Tr) < (m − 1)2. Equivalently any (m − 1)2 × (m − 1)2 minors of CR(T1, . . . , Tr) and CL(T1, . . . , Tr) vanishes. This assumption is equivalent to the assumption that the entries of T1, . . . , Tr satisfy corresponding system of r(m−1)(m−2) 2 2(cid:0) (cid:1) homogeneous polynomial equations of degree (m − 1)2. (m−1)2 Assume furthermore that rank CR(T1, . . . , Tr) = rank CL(T1, . . . , Tr) = (m − 1)2 − 1. Then nonzero solutions R, L of (4.3) and (4.4) are unique up to multiples by nonzero con- stants. The equalities (4.2) are equivalent to 2(cid:0) mial equations of degree 2((m − 1)2 − 1). 10 2 r(m−1)(m−2) 2 (m−1)2−1 (cid:1) (m − 1)2 homogeneous polyno- r(m−1)(m−2) 2 (m−1)2 2 )×(m−1)2 rows. Hence the total number of polynomial conditions for the existence of nonzero solution 2 (cid:1) linearly independent conditions. So equivalent to the fact that the entries of X satisfy (cid:0)m−1 CL(T1, . . . , Tr), CR(T1, . . . , Tr) ∈ Cr(m−1 . Note that any element of CR(T1, . . . , Tr) and CL(T1, . . . , Tr) is a linear function in the entries of some matrix Tk. Hence any (m − 1)2 × (m − 1)2 minor of CR(T1, . . . , Tl) and CL(T1, . . . , Tl) is a polynomial of degree (m − 1)2 (cid:1) distinct minors of order (m − 1)2 of CR(T1, . . . , Tr) in entries of T . There are (cid:0) and CL(T1, . . . , Tr) respectively, which corresponds to a choice of (m− 1)2 rows from r(cid:0)m−1 2 (cid:1) of (4.3) and (4.4) is equivalent to the vanishing of all 2(cid:0) (cid:1) minors of CR(T1, . . . , Tr) and CL(T1, . . . , Tr) of order (m − 1)2. Suppose now that rank CR(T1, . . . , Tr) = rank CL(T1, . . . , Tr) = (m − 1)2 − 1. Choose If either L or R are zero matrices then (4.2) a solution for L and R as in Lemma 2.6. holds trivially. If R, L 6= 0 then the conditions (4.2) are equivalent to 2(m − 1)2 polynomial identities of degree 2((m − 1)2 − 1) in the entries of T1, . . . , Tr. The number of choices of L and R as described in Lemma 2.6 is (cid:0) respectively. r(m−1)(m−2) 2 (m−1)2−1 (cid:1) r(m−1)(m−2) 2 (m−1)2 Proof. As X − X ⊤ is a skew symmetric matrix, the condition that X ∈ S(m − 1, C) is 2 ✷ We now discuss in detail the cases m = 4 and r = 3, 4. The case r = 3 is the Strassen condition. Theorem 4.4 Let T = [ti,j,k]3 i=j=k ∈ C3×3×3. Denote by T1, T2, T3 ∈ C3×3 the three 3-slices of T . Let CR(T1, T2, T3), CL(T1, T2, T3) ∈ C9×9 be the matrix coefficients of the systems (4.3) and (4.4) in the 9 entries of R and L respectively. Then the border rank of T is 4 at most if and only if one of the following condition hold. 1. det CR(T1, T2, T3) = 0. 2. det CL(T1, T2, T3) = 0. Equivalently, for any T ∈ C3×3×3 det CR(T1, T2, T3) = as(T1, T2, T3), det CL(T1, T2, T3) = bs(T1, T2, T3) (4.5) for some nonzero a, b ∈ C, where s(T1, T2, T3) is the Strassen polynomial described in The- orem 2.5. Proof. Suppose first that T1, T2, T3 ∈ S(3, C). Assume that T1, T2, T3 are three generic matrices. Add a generic matrix T4 ∈ S(3, C). The proof of Lemma 2.3 yields that span(T1, T2, T3, T4) is spanned by 4 rank one symmetric matrices. Theorem 2.1 yields that rank T ≤ 4. Assume that T1(adj T2)T3 6= T3(adj T2)T1. Theorem 2.7 implies that brank T ≥ 4. Hence rank T = 4. Since any T1, T2, T3 ∈ S(3, C) can be approximated by three symmetric matrices in general position we deduce that brank T ≤ 4 if the three 3-slices of T are symmetric matrices. Thus if (4.3) has a solution R ∈ GL(3, C) then brank T ≤ 4. We now show that there exists T1, T2, T3 ∈ C3×3 such that (4.3) has only the trivial solution R = 0. Let T1 = I, T2 a diagonal matrix with 3 distinct eigenvalues and T3 = [sij]3 i=j=1, were all sij 6= 0. The first condition of (4.3) yields that R ∈ S(3, C). The second condition of (4.3) imply that R commutes with T2. Hence R is a diagonal matrix diag(r1, r2, r3). The third condition of (4.3) is the condition risij = sjirj, i, j = 1, . . . , 3. So if s12 = s21, s13 = s32 and s23 6= s32 it follows that r1 = r2 = r3 = 0. On the other hand if T3 is also a symmetric matrix with nonzero entries, then (4.3) implies that R = rI3. Hence the condition det CR(T1, T2, T3) = 0 yield in the generic case, i.e. det R 6= 0, that brank T ≤ 4. By Strassen's theorem the set of T ∈ C3×3×3 of border rank 4 is a hypersurface given by the equation s(T1, T2, T3) = 0. Hence det CR(T1, T2, T3) = aS(T1, T2, T3) for some a 6= 0. Similar results apply to CL(T1, T2, T3). ✷ 11 Assume T ∈ C3×3×3 has border rank 4 at most. Then det CR(T1, T2, T3) = 0, and det CL(T1, T2, T3) = 0. Then the choice of R and L given by Lemma 2.6 is a column of adj CR and adj CL respectively. So the entries of R and L are homogeneous polynomials of degree 8 in the entries of T . Assume the generic case det R 6= 0. Then the arguments in the proof of Theorem 4.4 show that (4.2) hold. Note that since each entry of R and L are polynomials of degree 8 in the entries of T . So (4.2) are 18 polynomial equations of degree 16. Since the only condition for T ∈ C3×3×3 is the vanishing of the Strassen polynomial, we deduce that each polynomial equation of (4.2) is given by the Strassen polynomial times a homogeneous polynomial of degree 7. In conclusion, in this case, (4.2) do not give any additional restriction on T . i=j=1 ∈ C3×3, k = 1, . . . , 4. Let R = [xij ]3 We now discuss the case m = r = 4. So T = [ti,j,k] ∈ C3×3×4. We have four 3- slices Tk = [ti,j,k]3 i=j=1 be 3 × 3 matrices with unknown entries. Then (4.3) and (4.4) are 12 equations homoge- neous equations in 9 variables x11, . . . , x33 and y11, . . . , y33, which are given by the co- efficient matrices CR(T1, T2, T3, T4), CL(T1, T2, T3, T4) ∈ C12×9 respectively. The condi- tion that there exists nonzero R and L satisfying (4.3) and (4.4) respectively, are equiv- alent to the conditions rank CR(T1, T2, T3, T4) ≤ 8, rank CL(T1, T2, T3, T4) ≤ 8. So each 9 × 9 minor of CR(T1, T2, T3, T4), CL(T1, T2, T3, T4) is zero. The number of these condi- i=j=1, L = [yij ]3 tions is 2(cid:0)12 9(cid:1) = 440 polynomial equations of degree 9. Fix submatrices A, B ∈ C9×9 of CR(T1, T2, T3, T4), CL(T1, T2, T3, T4) respectively. Then each column of adj A, adj B respec- tively, represents a solution R, L of (4.3) and (4.4) respectively. If rank A < 8 then R = 0. If rank A = 8, then one of the 9 columns of adj A is nonzero. Similar conditions hold for B. So the number of the above choices of R and L is 9 × 220 = 1980 for each of them. Hence the total number of the above choices of pairs R, L is 19802. For each choice of R, L we assume that 18 conditions given by (4.2) hold. (To be precise, since tr LR⊤ = tr R⊤L we need at most 17 equations of (4.2).) It is not known the the author if the conditions that rank CR(T1, T2, T3, T4) ≤ 8, rank CL(T1, T2, T3, T4) ≤ 8 imply (4.2), as in the case of T ∈ C3×3×3. Theorem 4.5 T = [ti,j,k]3,3,4 i=j=k=1 ∈ C3×3×4 has a border rank 4 at most if and only the following conditions hold. 1. Let Tk := [ti,j,k]3 i=j=1 ∈ C3×3, k = 1, . . . , 4 be the four 3-slices of T . Then the ranks of CL(T1, . . . , T4), CR(T1, . . . , T4) are less than 9. (Those are 9 − th degree equations.) 2. Let R, L be solutions of (4.3) and (4.4) respectively as given in Lemma 2.6, (as de- scribed above). Then (4.2) holds. (Those are 16 − th degree equations.) Proof. Lemma 4.2 implies that if brank T ≤ 4 then the conditions 1-2 hold. We now assume that the conditions 1-2 hold. Let U := T3(T ). Suppose first that dim U ≤ 3. Pick A1, A2, A3 ∈ U such that span(A1, A2, A3) = U. Since each Ai is a linear combination of T1, . . . , T4, our assumption implies that there exists nonzero R such that AiR − R⊤A⊤ i = 0 for i = 1, 2, 3. Hence det CR(A1, A2, A3) = 0 which is equivalent to the Strassen condition s(A1, A2, A3) = 0. Strassen's theorem implies that rank T3 ≤ 4. Assume now that dim U = 4. Lemma 2.4 implies that brank T ≥ 4. Let R ∈ C3×3\{0} be a solution of (4.3) for m = 4. If R ∈ C3×3 has rank 3 then T ′ 4 := T4R are 4 linearly independent symmetric matrices. Use Lemma 2.3 to deduce that T ′ ∈ C3×3×4 has border rank 4. Similar results hold if rank L = 3. It is left to consider the case where max(rank R, rank L) ≤ 2. We now consider a number of cases. 1 := T1R, . . . , T ′ A: rank CR(T1, . . . , T4) = rank CL(T1, . . . , T4) = 8 I: rank L = rank R = 1. So after change of basis we can assume that L = e3e⊤ 3 . Then the condition that LT − T ⊤L⊤ = 0 is equivalent to T ⊤e3 = te3 for any T ∈ U. We now consider the following mutually exclusive subcases. 1 : T ⊤ i e3 = 0 for i = 1, . . . , 4. Hence T can be viewed as a tensor in C2×3×4. (2.11) implies that brank T ≤ 4. 12 2 : U contains F4 := e3e⊤ i e3 = 0, i = 1, 2, 3. Hence the tensor T ′ ∈ C3×3×3, whose three 3-slices are F1, F3, F3, can be viewed as a tensor in C2×3×3. (2.11) implies that brank T ′ ≤ 3 and and the border rank of T is 4 at most. 3 . So we can choose a basis F1, F2, F3, F4 such that F ⊤ 3 : Let T ′ k be obtained from Tk by deleting the last row for k = 1, 2, 3, 4. We claim that T ′ 4 are linearly independent. Otherwise, there is a nontrivial combination F ∈ C3×3 1, . . . , T ′ of T1, . . . , T4 such that the first two rows of F are zero rows. Since T1, . . . , T4 are linearly independent F 6= 0. As F ⊤e3 = te3 it follows that F = te3e⊤ e , t 6= 0. This contradicts our assumption that the case 2 does not hold. We now use the assumption that T R−R⊤T ⊤ = 0 and R = xy⊤, x = (x1, x2, x3)⊤, y = (y1, y2, y3)⊤ for each T ∈ U. So Tkx = sky for k = 1, . . . , 4. Suppose first that that all sk = 0. Then we are done as in the case 1. So we assume that si 6= 0 for some i. Since R and L have rank one, it follows that the condition 3 y = 0. Let Tk ∈ C2×2 (4.2) implies that R⊤L = LR⊤ = 0. Hence x3 = x⊤e3 = 0, y3 = e⊤ obtained from Tk by erasing the last row and column for k = 1, . . . , 4. Let x ∈ C2 be obtained from x ∈ C3 by deleting the the last coordinate. Then Tk x = sk x. So by changing the coordinates in C2 we may assume that x = (0, 1)⊤. Combine the above conditions with the conditions that T ⊤ i e3 = tie3, i = 1, . . . , 4 to deduce that there exists P, Q ∈ GL(3, C) with the following properties. Let Tk = P TkQ = [ti,j,k]3 i=j=1 ∈ C3×3, k = 1, . . . , 4. Then ti,j,k = 0 for (i, j) = (1, 2), (i, j) = (3, 1), (i, j) = (3, 2) and k = 1, . . . , 4. (4.6) Take a generic subspace V ⊂ C3×3 of dimension 4 whose entries are zero at the places (i, j) given by (4.6). We claim that V ∈ Γ(4, C3×3). First take a matrix D = [dij ]3 i=j=1 ∈ V such that dij = 0 for (1, 1), (2, 1), (2, 2). Generically there would one matrix, up to multiplication by a scalar, such that d33 6= 0. D has rank one. Now consider the 3-dimensional subspace of V where the (3, 3) entry of each matrix is zero. Then V can be viewed as a 3-dimensional subspace in V ⊂ C2×3. By Lemma 2.3 V ∈ Γ(3, C2×3). Hence V ∈ Γ(4, C3×3) and brank T ≤ 4. II: max(rank L, rank R) = 2. By considering U⊤ if necessary we may assume that rank L = 2. So there exist P, Q ∈ GL(3, C) such that P LQ = diag(1, 1, 0). Without loss of generality we may assume that P = Q = I. Then each LTk is symmetric. In particular Tke3 = tke3 and the 2 × 2 submatrix [ti,j,k]2 i,j=1 is symmetric. We now claim that any four dimensional subspace V ⊂ C3×3, such that each T = [tij ]3 i,j=1 ∈ V satisfies t12 = t21, t13 = t23 = 0, is in Γ(4, C3×3). As dim V = 4 there exists 0 6= S = [sij ]3 i=j=1 ∈ V such that 0 = s11 = s22 = s12(= s21). Hence rank S = 1. For a generic V satisfying the above conditions s33 6= 0. Consider now the 3 -dimensional subspace W of V with t33 = 0. Since W can be viewed as a 3-dimensional subspace of C3×2, Lemma 2.3 yields that W ∈ Γ(3, C3×3). Hence V ∈ Γ(4, C3×3) and brank T ≤ 4. B: min(rank CR(T1, . . . , T4), rank CL(T1, . . . , T4) < 8. By considering U⊤ if necessary we can assume that rank CL(T1, . . . , T4) < 8. So there exist at least two linearly independent matrices L1, L2 ∈ C3×3 such that (4.4) holds. If max(rank L1, rank L2) = 3 we deduce that brank T ≤ 4 as in the beginning of our proof. If max(rank L1, rank L2) = 2 we deduce that brank T ≤ 4 as in the case A.II. So it is left to consider the case where L1 and L2 are rank one matrices such any their linear combination is also rank a one matrix. It is easy to show that we can choose P, Q ∈ GL(3, C) such that P L1Q = e3e⊤ 3 or e3e⊤ 3 and P L2Q is either e2e⊤ 2 . So we have two cases. 3 , L2 = e2e⊤ I: L1 = e3e⊤ k e3 = tke3 for k = 1, 2, 3, 4,. I.e. any T ∈ U has zero entries at the places (3, 1), (3, 2). The condition (4.4) for L1 yields T ⊤ k = 0. Thus the third row of each Tk is zero. So T ∈ C2×3×4 and (2.11) yields that brank T ≤ 4. 3 . The condition (4.4) for L1 yields that T ⊤ ke2. Hence tk = t′ k e3 = t′ II: L1 = e3e⊤ 3 , L2 = e3e⊤ entries at the places (3, 1), (3, 2). The condition (4.4) for L2 yields that T ⊤ k e2 = t′ k = 1, 2, 3, 4. So the entries (2, 1), (2, 2) are zero for each T ∈ U. Take a nonzero T ′ 2 . The condition (4.4) for L1 yields that any T ∈ U has zero ke3 for 4 ∈ U 13 whose first row is zero. It is a rank one matrix. Then either (3, 3) entry or (2, 2) entry of T ′ 4 is not equal to zero. Assume for simplicity of the argument that (3, 3) entry of T ′ 4 is nonzero. Hence U contains a three dimensional subspace U′ whose last row is zero. Since a generic 3 dimensional subspace of 2 × 3 matrices is spanned by rank one matrices it follows that T has border rank 4 at most. ✷ Note that in the proof of Theorem 4.5 we used the condition 2, which are degree 16 polynomial equations, only in the proof of the case A.I.3. Thus one can eliminate the use of degree 16 polynomial equations, if one can show directly that a generic 4-dimensional subspace of matrices satisfying (4.3) and (4.4) for R and L of rank one, such that RL⊤ 6= 0, is in Γ(4, C3×3). As an example, consider the case where R = L = e3e⊤ 3 , which is ruled out by (4.2). Then the conditions (4.3) and (4.4) are equivalent to the assumption that T3(T ) is a four dimensional subspace of block diagonal matrices of the form   a b c 0 0 d 0 0 e .   (4.7) Hence Theorem 4.5 yields that T3(T ) 6∈ Γ(4, C3×3). It was shown in [14] that the corre- sponding T does not satisfy the degree 6 polynomial equations found in [10]. 5 Tensors in C4×4×4 of border rank 4 at most Theorem 5.1 T = [ti,j,k]4 i=j=k=1 ∈ C4×4×4 has a border rank 4 at most if and only the following conditions hold. 1. Any three matrices in T1(T ), T2(T ), T3(T ) satisfy the conditions (1.1). (These are 5 − th order degree equations on entries of X, Y, Z.) 2. For each P1, P2, P3 ∈ C4×4 let T (P1, P2, P3) = [ti,j,k(P1, P2, P3)]4 i=j=k=3 ∈ C4×4×4 be the tensor given by (2.1). Let Sip,p(P1, P2, P3) ∈ C4×4, ip = 1, . . . , 4 be the four p-slices of T (P1, P2, P3) for p = 1, 2, 3. (The entries of Sip,p(P1, P2, P3) ∈ C4×4 are given by ti1,i2,i3 (P1, P2, P3), where ip is fixed for a given p ∈ {1, 2, 2} and ip ∈ {1, 2, 3, 4}.) Denote by Tip,p(P1, P2, P3) ∈ C3×3 the submatrix obtained from Sip,p(P1, P2, P3) by deleting the last row and column, for ip = 1, 2, 3, 4. Then rank CL(T1,p(P1, P2, P3), . . . , T4,p(P1, P2, P3)) ≤ 8, rank CR(T1,p(P1, P2, P3), . . . , T4,p(P1, P2, P3)) ≤ 8, (5.1) (5.2) for p = 1, 2, 3. (Those are degree 9 − th degree equations.) Moreover the following conditions are satisfied for each p ∈ {1, 2, 3}. Let Rp(P1, P2, P3), Lp(P1, P2, P3) be solutions of (4.3) and (4.4) respectively as given in Lemma 2.6. Then (4.2) holds. (Those are degree 16 − th degree equations.) To prove Theorem 5.1 we need to prove Corollary 5.6 of [11]. Theorem 5.2 Let T ∈ C4×4×4 and assume that any three matrices X, Y, Z in Tp(T ) satisfy (1.1) for p = 1, 2, 3. Then either brank T ≤ 4 or or dim Tp(T ) ≤ 3, dim Tq(T ) ≤ 3 for two integers 1 ≤ p < q ≤ 3. Equivalently, by permuting factors in C4 ⊗ C4 ⊗ C4, if necessary, and changing bases in the first two components of C4 × C4 × C4 the tensor T can be viewed as a tensor C3×3×4. The proof of this theorem is completed by considering a number of lemmas. Lemma 5.3 Let T = [ti,j,k] ∈ C4×4×4 and p ∈ {1, 2, 3}. Assume that Tp(T ) contains an invertible matrix. Then the condition (1.1) for any three matrices in Tp(T ) implies that brank T ≤ 4. 14 Proof. It is enough to consider the case p = 3. Assume that Y ∈ T3(T ) is invertible. By considering P = Y −1 and changing a basis in the first factor of C4 ⊗ C4 ⊗ C4 we may that Y = I ∈ T3(T ). Let T3 = span(I, A1, A2, A3). So AiAj = AjAi for i, j = 1, 2, 3. Recall that the variety of all (X1, X2, X3) ∈ (C4×4) such that XiXj = XjXi, i, j = 1, 2, 3 is irre- ducible [7]. Hence a triple (A1, A2, A3) is a limit of generic commuting triples (X1, X2, X3). For a generic triple, X1 has 4 distinct eigenvalues. So X2, X3 are polynomial in X1. Thus, there exists Q ∈ GL(4, C) such that Q−1span(X1, X2, X3)Q is a three dimensional sub- space of 4 × 4 diagonal matrices D ⊂ C4×4. Clearly I ∈ D and D is spanned by 4 rank one diagonal matrices. Hence span(X1, X2, X3, I) ⊆ Q−1DQ ∈ Γ(4, C4×4). Thus T3(T ) ⊆ U ∈ Γ(4, C4×4), and brank T ≤ 4. ✷ In view of Lemma 5.3 we need to show Theorem 5.2 only in the case mrank Tp(T ) ≤ 3 for p = 1, 2, 3. Clearly, it is enough to assume that T 6= 0. If mrank Tp(T ) = 1 for some p ∈ {1, 2, 3}, then Tp(T ) spanned by rank one matrices. Theorem 2.1 implies that rank T ≤ 4. Thus we need to consider the case 2 ≤ mrank Tp(T ) ≤ 3 for p = 1, 2, 3. (5.3) We now consider the case mrank T3(T ) = 2. Lemma 5.4 Let T ∈ C4×4×4. Suppose that the (i, j) entry of each 3-slice Ti,3 is zero if min(i, j) ≥ 2. Then mrank T3(T ) ≤ 2 and any three matrices A, B, C ∈ T3(T ) satisfy (2.9) for p = 1, 2, 3. (In particular (1.1) holds for T3(T ).) For generic choices of the four 3-slices T1,3, . . . , T4,3 of the above form brank T = 5. Furthermore, if (1.1) holds for T1(T ) and T2(T ), then brank T ≤ 4. Proof. Let A = [aij]4 i=j=1 ∈ C4×4. Assume that aij = 0 if max(i, j) ≥ 2. So the nonzero entries of A are on the first row and column. Clearly rank A ≤ 2. Hence adj A = 0. This implies that any three matrices A, B, C ∈ T3(T ) satisfy (2.9) for p = 1, 3. Next observe that C2(A) has zero (2, 3), (2, 4), (3, 4) rows and columns. So C−2(B) has zero (1, 2), (1, 3), (1, 4) rows and columns. Hence C2(A)C−2(B)⊤C2(C) = 0, and (2.9) holds for p = 2. Assume now Ti,3 = ai ei fi gi   bi 0 0 0 ci 0 0 0 di 0 0 0   , i = 1, 2, 3, 4. Consider now T1,1, . . . , T4,1. Note that T1,1 is a full matrix, while Ti,1 has a full first column, while the other 3 columns are equal to zero. Suppose first that det Ti,1 6= 0. (This is true of T1,3, . . . , T4,3 are generic.) Consider the Ai = T −1 1,1 Ti,1 for i = 1, 2, 3, 4. So A1 = I4 and Ai has the first nonzero column ai, i.e. Ai = aie⊤ 1 for i = 2, 3, 4. The commutation condition (1.1) with X = Ai, Y = I4, Z = Aj for 2 ≤ i < j ≤ 4 is equivalent to (e⊤ 1 = (e⊤ 1 ai) 6= 0 we deduce that the commutation condition holds of and only if a2, a3, a4 are colinear. The assumption that T1,3, . . . , T4,3 are generic matrices with nonzero entries in the first row and column yield that a2, a3, a4 are 3 generic vectors. Hence a2, a3, a4 are not colinear, and the commutation condition (1.1) does not hold for T1(T ). Therefore brank T ≥ 5. 1 . Assuming that (e⊤ 1 aj )aie⊤ 1 ai)aj e⊤ 1 aj)(e⊤ To show that brank T = 5 we add to the space spanned by I4, a1e⊤ 1 the rank one matrix e1e⊤ i = a′ 1 = Ai − (e⊤ 3 be the four 3-slices of T ′ ∈ C4×4×4. 1 ai)e1e⊤ The proof of Lemma 5.3 yields that brank T ≤ 4. Hence brank T ≤ 5 and we conclude that brank T = 5. 1 , a3e⊤ 1 ai)e1, i = 1, 2, 3. Then the three matrices A′ 1 . Let a′ 1 , i = 1, 2, 3 commute. Let I4, A′ i = ai − (e⊤ 1 , a2e⊤ ie⊤ 1, A′ 2, A′ We now consider non-generic T ∈ C4×4×4 satisfying the conditions of our lemma. Sup- pose first that dim T3(T ) ≤ 2. By changing a basis in the last component of C4 ⊗ C4 ⊗ C4 we may assume that that T3,3 = T4,3 = 0. Then T can be viewed as a tensor in C4×4×2. (2.11) yields that brank T ≤ 4. Assume that dim T3(T ) = 3. By changing a basis in 15 1 + e1e⊤ i+1 + ei+1e⊤ the last component of C4 ⊗ C4 ⊗ C4 we may assume that T4,3 = 0. Assume now that the entries on the first row and column of T1,3, T2,3, T3,3 are in general position. Then there exists P = [1] ⊕ P1, Q = [1] ⊕ Q1, where P1, Q1 ∈ GL(3, C) we may assume that T ′ i,3 = P Ti,3Q = xie1e⊤ 1 , and de- note by T ′ ∈ C4×4×4 the tensor whose four 3-slices are T ′ i,3, i = 1, 2, 3, 4. We claim that brank T ′ ≤ 4. Consider the following basis in T3(T ′): e1e⊤ 1 and e1e⊤ 1 for i = 2, 3, 4. The above arguments show that T1(T ′) is a 4 dimensional subspace of commuting matri- ces, which contain I4. Hence by Lemma 5.3 brank T ′ ≤ 4. Hence brank T ≤ 4. Since any three matrices T1,3, T2,3, T3,3 with zero entries in the position (i, j) for min(i, j) ≥ 2 can be approximated by generic matrices of this kind, we deduce that brank T ≤ 4. 1 for i = 1, 2, 3. Let T ′ 4,3 = e1e⊤ i + eie⊤ We now assume that dim T3(T ) = 4. Assume now that T1(T ) and T2(T ) satisfy the condition (1.1). If either T1(T ) or T2(T ) contain an invertible matrix then by Lemma 5.3 brank T ≤ 4. So assume that mrank T1(T ), mrank T2(T ) ≤ 3. Hence, the four first rows and columns of T1,3, . . . , T4,1 are linearly dependent. By changing a basis in the last component of C4 ⊗ C4 ⊗ C4 we may assume that the first row of T4,3 is zero. As dim T3(T ) = 4 it follows that that T4,3 has zero first row and nonzero first column. Apply elementary row operations to the last three rows of T4,1 to assume that T4,1 = e4e⊤ 1 . Apply the same elementary row operations to T1,3, T2,3, T3,3. Hence, we can assume without loss of generality that T4,1 = e4e⊤ 1 . By considering Ti,3 − tiT4,3 for i = 1, 2, 3 we may assume the (4, 1) entry of Ti,3 is zero for i = 1, 2, 3. Consider again the column space spanned by the first three columns of T1,3, T2,3, T3,3. It must be two dimensional, otherwise the column space T is four dimensional and T2(T ) will contain an invertible matrix. So by changing a basis in span(T1,3, T2,3, T3,3) we can assume that the first column of T3,3 is zero. As dim T3(T ) = 4 the first row of T3,3 is nonzero. Apply elementary column operations to the last three columns of T3,3 we may assume that T3,3 = e1e⊤ 4 . Apply the same elementary column operations to T1,3, T2,3, T4,3. We still have that T4,3 = e4e⊤ 1 . Apply the above arguments to deduce that without loss of generality we may assume that the last row and column of T1,3, T2,3 are zero. Consider first Ti,2, i = 1, 2, 3, 4. Observe that T1,2 = T3,2 =     ∗ ∗ ∗ 0 ∗ 0 0 0 ∗ 0 ∗ 0 ∗ 0 0 0 ∗ 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0         ∗ 0 0 0 ∗ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 , .     , T2,2 = , T4,2 = The assumption that T1,2 + T1,4 is singular yields that the second and the third row of T1,2 are linearly dependent. Do elementary row operations on the second and the third row of T1,2 to obtain that a zero third row. Do the same elementary row operations on Ti,2, i = 2, 3, 4 to deduce that we may assume that each Ti,2 has zero third row. Translating back to T1(T ) we deduce that we may assume in addition to all our above assumptions on T1,3, . . . , T4,3 the third row of T1,3, T2,3 are zero. So all matrices in T3(T ) have zero third row. Consider now Ti,1 for i = 1, 2, 3, 4. Observe that T3,1 = 0. Apply the same arguments as for Ti,2 for i = 1, 2, 3, 4 to deduce that we can assume that the third column of T1,1 is zero. This implies that in this case we can assume that after suitable change of basis in the first two components of C4 ⊗ C4 ⊗ C4, in addition to the assumption that all (i, j) entries of matrices in T3(T ) are zero if min(i, j) ≥ 3, the third row and column of each matrix in T3(T ) is zero. It is left to show that brank T ≤ 4. Observe that the last two rows and columns of T1,3, T2,3 are zero. Let T ′ ∈ C4×4×2 be the tensor whose two 3-slices are T1,3, T2,3. So T ′ can be viewed as a tensor in C2×2×2. (2.11) yields that brank T ′ ≤ 2. As T3,4 = e1e⊤ 4 , T4,4 = e4e⊤ ✷ 1 we deduce that brank T ≤ 4. 16 We remark that Lemma 5.4 refutes Proposition 5.4 of [11], which claims that for any tensor T ∈ C4×4×4 for which T3(T ) satisfies (1.1) either brank T ≤ 4 or there exists nonzero u ∈ C4 such that either T3(T )u = 0 or u⊤T3(T ) = 0⊤. Indeed, if we assume as in the first part of Lemma 5.4 that T = [ti,j,k] ∈ C4×4×4 is a generic tensor such that ti,j,k = 0 if min(i, j) ≥ 2, then there is no nonzero u ∈ C4 such that either T3(T )u = 0 or u⊤T3(T ) = 0⊤, and by this lemma brank T = 5 . Lemma 5.5 Let T ′ ∈ C4×4×4 and assume that mrank T3(T ′) = 2. Then either brank T ′ ≤ 4 or it is possible to change bases in the first two components of C4 ⊗ C4 ⊗ C4 to obtain the following two possibilities. Either T ∈ C4×4×4 satisfies the conditions of Lemma 5.4, or the last row and column of each matrix in T3(T ) is zero. (In the last case, in addition every submatrix of A ∈ T3(T ) based on the first three rows and columns is singular.) A21 Proof. Change a basis in the first component of C4 ⊗ C4 ⊗ C4 such that T1,3 = I2 ⊕ 0. Since mrank T3(T ′) = 2 it follows that the four entries (3, 3), (3, 4), (4, 3), (4, 4) of each matrix in T3(T ′) are zero. So any matrix in A ∈ T3(T ′) has the block form (cid:20) A11 A12 0 (cid:21), where each Aij ∈ C2×2. Consider A + tT1,3. Assume that det(A11 + tI2) 6= 0. Then rank (A + tT1,3) = 2 if and only if A21(A11 + tI2)−1A12 = 0. Assume that rank A12 = 2 for some A ∈ T3(T ′). Then for most of A ∈ T3(T ′) rank A12 = 2. Hence for most A's A12 = 0. So the last two rows of each A ∈ T3(T ′) are zero. Hence T ′ can be viewed as a tensor in C2×4×4 and brank T ′ ≤ 4. Similarly if for rank A21 = 2 for some A ∈ T3(T ′) we deduce that brank T ′ ≤ 4. It is left to assume that rank A12, rank A21 ≤ 1 for each A ∈ T3(T ′), and rank A12 = rank A21 = 1 for some matrix A. It is easy to see for 2 × 2 matrices that A12 is either x(A12)u⊤ or ux(A12)⊤, and A21 is either y(A21)v⊤ or vy(A21)⊤, where u, v ∈ C2 \ {0} and x(A12), y(A21) ∈ C2 are nonidentically zero vectors depending linearly on the entries of A12 and A21 respectively. By using elementary row or column operations on the first two rows and columns, and then on the last two rows and columns we may assume that u = v = e1. We now obtain the four 3-slices of T . Note that for some T1,3 = B ⊕ 0, where B ∈ C2×2 is invertible. Suppose first that A12 = e1x(A12)⊤, A21 = y(A21)e⊤ 1 . Since x(A12) and y(A21) are not zero identically, and rank A ≤ 2 we deduce that the (2, 2) entry of each A is zero. (For example if (3, 1) and (1, 3) entries of A are nonzero then consider the minor with the first three rows and columns.) Hence T3(T ) is of the form given by Lemma 5.4. Suppose next that A12 = x(A12)e⊤ 1 , A21 = e1y(A21)⊤. Then the last row and column of each matrix in T3(T ) is zero. Moreover, each 3 × 3 submatrix of T3(T ) is singular. 1 , A21 = y(A21)e⊤ The next case is A12 = x(A12)e⊤ 1 . Thus the last column of each A ∈ T3(T ) is zero. Since x(A21) is not identically zero, and mrank T2(T ) = 2 we deduce the minor based on rows 1, 2 and columns 2, 3 must be identically zero. So we have two possibilities. First possibility: by elementary row operations on the first two rows of ma- trices on T3(T ) we can bring T3(T ) to matrices of the form given by Lemma 5.4 with an addition condition, that the last column of all these matrices is zero. Second possibility: by elementary column operations we can achieve that also the third column of all matrices in T3(T ) are zero. So T can be viewed as a tensor in C4×2×4. Hence brank T ′ ≤ 4. Similar results hold for the last case A12 = e1x(A12)⊤, A21 = e1y(A21)⊤. ✷ We state the precise version of Theorem 3.1 for m = 4. Lemma 5.6 Let U ⊂ C4×4 and assume that mrank U = 3. Then any three matrices in U satisfy (1.1) if and only if one of the following mutually exclusive conditions hold. 1. There exists a nonzero u ∈ C4 such that either Uu = 0 or u⊤U = 0⊤. 2. dim U = k +1 ≥ 2. There exists P, Q ∈ GL(4, C) such that P UQ has a following basis F0, . . . , Fk. The last row and column of F0, . . . , Fk−1 is zero, i.e. Fi = Gi ⊕ 0, Fi ∈ 17 Gk =   0 0 0 g h 0 0 f 0   . (5.8) C3×3, i = 0, . . . , k − 1, G0 = I3, and Fk = (cid:20) Gk e⊤ 2 e1 0 (cid:21) , Gk ∈ C3×3. (5.4) Furthermore Gk, G1, . . . , Gk−1 have one the following possible three forms. (a) (b) (c) Gk =   0 0 0 0 0 0 0 0 0   , Gi =   0 ai 0 0 0 ci Gk =   0 0 0 0 h 0 Gk =   0 0 0 0 0 0 0 0 0 0 f 0  , Gi =      , Gi =   0 ai 0 0 0 ci 0 ai 0 0 0 0 bi 0 di 0 0 0 bi 0 0 , i = 1, . . . , k − 1. (5.5) , i = 1, . . . , k − 1. (5.6)       , i = 1, . . . , k − 1. (5.7) Proof. By changing a basis in C3 we can assume that f , g ∈ C3 appearing in Theorem 3.1 for m = 4 are of the form f = e1, g = e2. Next we observe that Gk can be always assumed to be of the form Indeed, first replace Fk by F ′ k = Fk − tF0 such that the (3, 3) entry of Fk is zero. Next, use the following elementary row and column operations without changing the form of F0, . . . , Fk−1. Subtract a multiple of a row four from row i for i = 1, 2, 3. Similarly, subtract a multiple of a column four from column i for i = 1, 2, 3. The exact forms of Gk, G1, . . . , Gk−1 are obtained by choosing subspaces X, Y appearing in Theorem 3.1 to be of the following forms: X = span(e1), Y = span(e2); X = span(e1, e3), Y = span(e2); X = span(e1), Y = span(e2, e3). Consider now the choice X = span(e1, e3), Y = span(e2, e3). Then Gi is of the form given by (5.5) where bi = ci = di = 0 for i = 1, . . . , k − 1. Furthermore Gk = 0. This can be considered as a special case of 2a. ✷ Proof of Theorem 5.2. In view of Theorem 2.1 and Lemmas 5.3, 5.5, 5.4 we need only to consider the case mrank Tp(T ) = 3 for p = 1, 2, 3. Assume first that dim Tp(T ) ≤ 2 for some p ∈ {1, 2, 3}. Then by changing a basis in the p-th factor of C4 ⊗ C4 ⊗ C4 and permuting the factors, we obtain that T can be viewed as a tensor in C2×4×4. (2.11) yields that brank T ≤ 4. Hence we assume that dim Tp(T ) ≥ 3 for p = 1, 2, 3. Assume next that dim Tp(T ) = dim Tq(T ) = 3 for some 1 ≤ p < q ≤ 3. Then Theorem 5.2 holds. Assume next that dim Tp(T ) = 3 for some p ∈ {1, 2, 3} and dim Tq(T ) = 4 for q ∈ {1, 2, 3} \ {p}. By permuting the factors of C4 ⊗ C4 ⊗ C4 we may assume that dim T1(T ) = dim T2(T ) = 4, dim T3(T ) = 3. Observe first that there is no nonzero u ∈ C4 such that either T3(T )u = 0 or u⊤T3(T ) = 0⊤. Indeed, assume first that T3(T )u = 0 for some nonzero u. By change of coordinates in the second component of C4 ⊗ C4 ⊗ C4 we can assume that u = e4. So the fourth column of each matrix in T3(T ) is zero. Hence T4,2 = 0 which contradicts that dim T2(T ) = 4. Similar arguments apply if u⊤T3(T ) = 0⊤. given in 2.a. Consider the matrix G1 =    . Assume the generic case d 6= 0 and We now apply part 2 of Lemma 5.6. Here k = 2. Assume first that F1, F2 have the form 0 a b 0 0 0 d 0 c 18 ad − bc 6= 0. Then the three eigenvalues of G1 are 0, 0, d. The Jordan canonical form of G1 is J =   form 0 0 e 0 0 0 0 0 d  , where e 6= 0. Furthermore, P G1P −1 = J where and P, P −1 have the P =   1 −α + βγ −β 0 0 0 1   , P −1 =   1 α β 0 0 0 1   . 1 −γ (5.9) 1 γ (Note that e1, e⊤ 2 are right and left eigenvectors of G1, J corresponding to 0 eigenvalue.) Let Q = P ⊕ [1] ∈ GL(4, C). Then QF0Q−1 = F0, QF1Q−1 = J ⊕ [0], QF2D−1 = F2. 3 ∈ C4×4 and consider Equivalently, we may assume that G1 is equal to J. Let F3 = e3e⊤ the tensor T ′ ∈ C4×4×4 whose four 3-slices are T ′ 1,3 = F0 − e3e⊤ 3 , T ′ 2,3 = 1 e (F1 − de3e⊤ 3 ), T ′ 3,3 = F2, T4,3 = e3e⊤ 3 . Let T ′′ ∈ C4×4×3 is obtained from T ′ by deleting the last 3-slice of T ′. We claim that brank T ′′ ≤ 3. Observe that the three 3-slices of T ′′ have zero third row and column. So we can view T ′′ as S ∈ C3×3×3 whose three slices are given as in the last part of the proof of Theorem 3.2. Hence brank T ′′ ≤ 3 and brank T ≤ 4. Assume now that F1, F2 have the form given in 2.c. So G1 =   0 a b 0 0 0 0 0 0   0 0 0 0 0 0 f 0 0 , G2 =   i = (P ⊕ [1])Fi(P ⊕ [1])−1 for i = 0, 1, 2.   . Assume the generic case that b, f 6= 0. Consider F ′ Assume that P is of the form (5.9), where α = β = 0, γ = a b . Then   b 0 0 , G′ . e1 2 e⊤ 2 0 f f a b 1 = G′ 2 − f a 1 ⊕ [0], F ′ 0 0 0 0 0 0 F ′ 0 = F0, F ′ 0 (cid:21) , G′ 2 = (cid:20) G′ 2 =   1 =   0 0 0 − f a b 0 − f a2 b2 2 to obtain Fi, i = 0, 1, 2. Add to column one f a   Let F ′′ 2 = F ′ b F0. Then do the following elementary column and row operations on 1, F ′′ F0, F ′ times column four, add to row three f a2 times row four. Observe that F0, F1, F2 are b2 times row four and add to row two 2f a of the form F0, F1, F2 we started with, and with the addional fact that a = 0 in G1. Since b 6= 0, by replacing F1 with 1 b F1 we may assume that b = 1. It is left to show that our tensor T ∈ C4×4×3 has border rank 4 at most. Let T (t) ∈ C4×4×3 be the tensors with the following three 3-slices F0(t) = I3 ⊕ [t], F1, F2. It suffices to show that for any t 6= 0, brank T (t) ≤ 4. Mulitply the last row of F0(t), F1, F2 by 1 t to obtain I4, F1, F2(t). Let T (t)′ ∈ C4×4×3 be the tensor with the above three 3-slices. Observe that F1F2(t) = F2(t)F1 = 0. Apply the arguments in the proof of Lemma 5.3 to deduce that brank T (t)′ ≤ 4. Assume now that F1, F2 have the form given in 2.b. Consider F ⊤ 0 = F0, F ⊤ 2 . Let 1 , F ⊤ b b . Then F0 = (P ⊕ [1])F ⊤ 0 (P ⊕ [1])−1 and (P ⊕ [1])F ⊤ i (P ⊕ [1])−1, i = 1, 2 P =   0 1 0 1 0 0 0 0 1   are of the form 2.c. Hence brank T ≤ 4. It is left to discuss the case where dim Tp(T ) = 4, mrank Tp = 3, p = 1, 2, 3. We show that this case does not exists. As above we observe that there is no nonzero u such that either Tp(T )u = 0 or u⊤Tp(T ) = 0⊤. We now apply Lemma 5.6 to T3(T ). We change bases in the first two components of C4 ⊗ C4 ⊗ C4 to obtain one in the three possibilities discussed in Lemma 5.6. We start with the case 2.a, where k = 3. Our assumption that dim T3(T ) = 4 implies that the matrices G1 and G2 in (5.5) are linearly independent. 19 Observe next that T1,2 =     T3,2 = 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 b1 0 0 1 d1 0 0 b2 0 d2 0   0 0 0 0 , T2,2 =   , T4,2 =   0 a1 a2 0 1 0 c2 0 0 0 c1 0 0 0 0 1 0 0 0 0 0 0 0 0   0 1 0 0 0 0 0 0 , .     The assumption that mrank T2(T ) = 3 yields that det(T2,2 + x1T1,2 + x3T3,2 + x4T4,2) = 0. Hence (a1 + x3b1)(c2 + x3d2) − (a2 + x3b2)(c1 + x3d1) is identically zero. Let A1 = (cid:20) a1 a2 c2 (cid:21) , A2 = (cid:20) b1 c1 d1 b2 d2 (cid:21) . Then there exists a nonzero v ∈ C2 such that either A1v = A2v = 0 or v⊤A1 = v⊤A2 = 0⊤. The first possibility yields that there exists nonzero u ∈ C4 such that T2(T )u = 0 which contradicts our assumptions. Hence, there exists a nonzero v ∈ C2 such that v⊤A1 = v⊤A2 = 0⊤. The assumption that mrank T1(T ) = 3 yields that (a1 + x3c1)(b2 + x3d2) − (a2 + x3c2)(b1 + x3d1) is identically zero. Let B1 = (cid:20) a1 a2 b1 b2 (cid:21) , A2 = (cid:20) c1 c2 d1 d2 (cid:21) . As above we deduce that there exists nonzero w ∈ C2 such that B1w = B2w = 0. In particular we deduce that the rows (a1, b1), (a2, b2) and (c1, d1), (c2, d2) are linearly depen- dent. Hence, by choosing a new basis in span(G1, G2) we can assume that c1 = d1 = 0 and a2 = b2 = 0. So A1, A2 are both diagonal, singular, and have a common left zero eigenvector. So either a1 = b1 = 0 or c2 = d2 = 0. That is either G1 = 0 or G2 = 0 which contradicts our assumption that dim T3(T ) = 4. Assume the condition 2.b. Consider the matrices T1,2, T2,2, T3,2, T4,2. Note that T2,2 = 0 a1 0 1 0 c1 0 0 . The assumption that mrank T2(T ) = 3 yields that a1c2 − a2c1 = 0. So a2 0 c2 0 0 0 0 1 the vectors (a1, c1)⊤, (a2, c2)⊤ are linearly dependent. Hence F1, F2 are linearly dependent. This contradicts the assumption that dim T3(T ) = 4. Assume the condition 2.c. Consider the matrices T1,1, T2,1, T3,1, T4,1. The assumption that det T1,1 = 0 yields that the vectors (a1, b1), (a2, b2) are linearly dependent. Hence F1, F2 are linearly dependent. This contradicts the assumption that dim T3(T ) = 4. ✷ Proof of Theorem 5.1. We first show that if brank T ≤ 4 then the conditions 1-2 hold. Theorem 2.7 yields the condition 1. Let P1, P2, P3 ∈ GL(4, C). Then brank T (P1, P2, P3) ≤ 4. Let T ′(P1, P2, P3) = [ti,j,k(P1, P2, P3)]3,3,4 i=j=k ∈ C3×4. Clearly brank T ′(P1, P2, P3) ≤ 4. Theorem 4.5 yields the conditions (5.1)-(5.2) for p = 3 and (4.2) for R3(P1, P2, P3), L3(P1, P2, P3). Similar arguments imply the conditions 2 for p = 1, 2. Since GL(4, C) is dense in C4×4 we deduce the condition 2 for any P1, P2, P3 ∈ C4×4. Assume to that T satisfies the conditions 1-2. Suppose to the contrary that brank T > 4. Theorem 5.2 yields that there exists nonzero u ∈ C4 such that either T3(T )u = 0 or u⊤T3(T ) = 0. By changing the first two factors if needed we can assume that u⊤T3(T ) = 0. By changing a basis in the first factor we may assume that u = e4 = (0, 0, 0, 1)⊤. Con- sider now T1(T ). Note that the fourth 1-slice T4,1 is zero matrix. Since brank T > 4 20     Theorem 5.2 yields that there exists nonzero v ∈ C4 such that either T1(T )v = 0 or v⊤T1(T ) = 0. By permuting the two last factors C4, if needed, we may assume that vT1(T ) = 0. By changing a basis in the second factor we can assume that v = e4. This finally means that after permuting the factors of C4 × C4 × C4, and changing bases in the first two factors, we obtain a new tensor T ′ whose four 3-slices are matrices with the zero last row and column. Permuting back to the original factors and changing bases correspond- ingly using P1, P2, P3 ∈ GL(4, C), (one of these matrices is an identity matrix), we deduce that for some p ∈ {1, 2, 3}, the four p-slices S1,p(P1, P2, P3), . . . , S4,p(P1, P2, P3) have zero last row and column. Combine conditions 2 with Theorem 4.5 to deduce that the tensor T ′ ∈ C3×3×4, whose four 3-slices are T1,p(P1, P2, P3), . . . , T4,p(P1, P2, P3), has border rank 4 at most. Hence 4 ≥ brank T (P1, P2, P3) = brank T which contradicts our assumption. ✷ We conclude this section by showing that the conditions in of Theorem 5.1 can be stated as a finite number of polynomial equations in degrees 5, 9 and 16 in the entries of T . First we consider the conditions 1. In the notation of 1 Tp(T ) is spanned by S1,p(I, I, I), . . . , S4,p(I, I, I). Fix p ∈ {1, 2, 3}. Let X = 4 Xi=1 xiSi,p(I, I, I), Y = 4 Xj=1 yjSj,p(I, I, I), Z = 4 Xk=1 zkSk,p(I, I, I). Then adj Y is 4 × 4 matrix whose entries are homogeneous polynomials of degree 3 in y = (y1, . . . , y4)⊤. Note that the coefficients of the monomials of these polynomials are polynomials of degree 3 in the entries of T . Substitute the expressions of X, adj Y, Z into the conditions (1.1) to deduce that a finite number of polynomials in x, y, z of degree 5 must vanish identically. That is, the corresponding coefficient of each monomial must be zero. This procedure gives rise to a finite number of polynomial equations of degree 5 that the entries of T satisfy. Clearly, if these conditions hold then the condition 1 hold. We now discuss the conditions 2. Write Pq = [xij,q]4 i=j=1 for q = 1, 2, 3. We view each xij,q as a variable. So the entries of P1, P2, P3 give rise to 48 variables. The entries S1,p(P1, P2, P3), . . . , S4,p(P1, P2, P3), p = 1, 2, 3, are multilinear polynomials of degree 3 in 48 variables. Hence the entries of T1,p(P1, P2, P3), . . . , T4,p(P1, P2, P3), are multilinear polyno- mials of degree 3, whose coefficients are homogeneous polynomials of degree 1 in the entries of T . The conditions (5.1-5.2) yield that each 9 × 9 minor of the two matrices in (5.1-5.2) must be identically zero. Such a minor is a homogeneous polynomial of degree27 = 3 × 9 in the entries of P1, P2, P3. The coefficient of each monomial is a polynomial of degree 9 in the entries of T . Hence the coefficient of each monomial appearing in the expansion of each minor must equal to zero. (Recall that for each p ∈ {1, 2, 3} we have 440 such 9 × 9 minors.) These conditions give rise to a finite number polynomial conditions of degree 9 on the entries of T . Repeat the same procedure for the conditions on Rp(P1, P2, P3), Lp(P1, P2, P3) given by (4.3-4.4) to deduce a finite number of polynomial conditions of degree 16 on the entries of T . Clearly, these polynomial equations of degree 9 and 16 imply that the condition 2 hold for each P1, P2, P3 ∈ C4×4×4. 6 Tensors in Cm×n×l of rank l Let m, n, l ≥ 2 and and assume that T = [ti,j,k] ∈ Cm×n×l. In this section we study mainly the conditions when rank T = dim T3(T ). We point out briefly how to state some of these conditions for tensors of border rank l at most. We consider the generic case dim T3(T ) = l. Equivalently we study the conditions on a subspace T ⊂ Cm×n, of dimension l, to be spanned by l linearly independent rank one matrices. First choose a basis T1, . . . , Tl ∈ Cm×n of T. k=1 zkTk. Recall that T (z) has rank at Let z = (z1, . . . , zl)⊤ ∈ Cl and denote T (z) = Pl 21 most rank 1 if all 2 × 2 minors of T (z) are zero. Let α = (i1, i2), 1 ≤ i1 < i2 ≤ m, β = (j1, j2), 1 ≤ j1 < j2 ≤ n. As in §2 denote by 2hmi the set of all allowable α, β respectively. Then T (z)[α, β] is the 2 × 2 minor of T (z) based on the rows i1, i2 and the columns j1, j2. Clearly T (z)[α, β] is a quadratic form in z. For A = [aij], B = [bij] ∈ Cm×n denote , 2hni 2 2 b(A, B)[α, β] = det(cid:20) ai1j1 ai2j1 bi1j2 bi2j2 (cid:21) . Then the condition that T (z) has rank one at most is given by the system of quadratic equations T (z)[α, β] = l Xp,q=1 b(Ap, Aq)[α, β]zpzq = 0, for all α ∈ 2hmi 2 , β ∈ 2hni 2 . (6.1) Note the number of these equations is (cid:0)m associate a symmetric matrix S(α, β)(T1, . . . , Tl) ∈ S(l, C), Denote by S(T1, . . . , Tl) ⊂ S(lC) the subspace spanned by S(α, β)(T1, . . . , Tl), α ∈ 2hmi 2 . Suppose that we change a basis of T from T1, . . . , Tl to T ′ l . This is equivalent to the change of variables z = Ry. Hence 2(cid:1). With each of the above quadratic form we 2(cid:1)(cid:0)n 1, . . . , T ′ , 2hni 2 S(α, β)(T ′ 1, . . . , T ′ l ) = R⊤S(α, β)(T1, . . . , Tl)R, for some R ∈ GL(l, C), (6.2) and α ∈ 2hmi we let S(T) := S(T1, . . . , Tl). So the subspace S(T) is defined up to congruence. l ) = R⊤S(T1, . . . , Tl)R. For simplicity of notation 2 . Thus S(T ′ , β ∈ 2hni 1, . . . , T ′ 2 Proposition 6.1 Let T1, . . . , Tl be a basis in T ⊂ Cm×n. Then for any P ∈ GL(m, C), Q ∈ GL(n, C) S(α, β)(P T1Q, . . . , P TlQ) = Xγ∈2hmi ,δ∈2hni 2 2 P [α, γ]Q[δ, β]S(γ, δ)(T1, . . . , Tl). (6.3) Hence S(P TQ) = S(T). Proof. Recall that for any A ∈ Cm×n the 2 × 2 minors of P AQ are given by the Cauchy-Binet formula (P AQ)[α, β] = Xγ∈2hmi ,δ∈2hni 2 2 P [α, γ]A[γ, δ]Q[δ, β]. Clearly P T (z)Q = Pl S(P TQ) ⊆ S(T). Since T = P −1(P TQ)Q−1 we obtain that S(P TQ) ⊇ S(T). k=1 zkP TkQ. Apply the Cauchy-Binet formula to deduce (6.3). Hence ✷ Let S(T1, . . . , Tl)⊥ ⊂ S(l, C) be the orthogonal complement of with respect the symmetric product on Cl×l: hA, Bi := tr AB⊤. Lemma 6.2 Let T ⊂ Cm×n be an l-dimensional subspace. Then T contains r-linearly independent rank one matrices if and only the subspace S(T)⊥ contains r linearly indepen- dent rank one symmetric matrices which are simultaneously diagonable by congruency. That is, if T1, . . . , Tl is a basis of T, then there exist R ∈ GL(l, C) such that R⊤ diag(δ1k, . . . , δlk)R ∈ S(T)⊥ for k = 1, . . . , r. 22 Proof. Choose a basis in T: T1, . . . , Tl such that T1, . . . , Tr are r-rank one linearly in- dependent matrices. Hence the quadratic form T (z)[α, β] does not contain terms z2 1 , . . . , z2 r . Thus the diagonal entries (1, 1), . . . , (r, r) are zero for each S(α, β)(T1, . . . , Tl). Therefore Dk := diag(δ1k, . . . , δlk) ∈ S(T1, . . . , Tl)⊥ for k = 1, . . . , r. Assume now that S(T1, . . . , Tk)⊥ contains R−1Dk(R−1) ⊤ alent to the fact that Dk ∈ S(T ′ T ′ 1, . . . , T ′ r are r linearly independent rank one matrices in T. 1, . . . , T ′ l )⊥ for a corresponding basis T ′ for k = 1, . . . , r. This is equiv- l of T. Hence ✷ 1, . . . , T ′ Corollary 6.3 Let T ⊂ Cm×n be an l-dimensional subspace. Assume that T contains r-linearly independent rank one matrices. Then dimension S(T) is at most (cid:0)l+1 The data induced by S(α, β)(T1, . . . , Tl) for α ∈ 2hmi following (cid:0)m 2 (cid:1) matrix C(T) = C(T1, . . . , Tl) = [c(α,β)(p,q)]. Here c(α,β)(p,q) = b(Tp, Tq)[α, β] + b(Tq, Tp)[α, β], 1 ≤ p ≤ q ≤ l, α ∈ 2hmi 2(cid:1) ×(cid:0)l+1 , β ∈ 2hni 2(cid:1)(cid:0)n 2 2 , β ∈ 2hni 2 . 2 2 (cid:1) − r. can be arranged in the (6.4) Corollary 6.3 equivalent to the statement that rank C(T) ≤ (cid:0)l+1 of order (cid:0)l+1 2 (cid:1) − r. That is, all minors 2 (cid:1) − r + 1 of C(T) are zero. We now give two examples of generic subspaces T ⊂ Cm×n of dimension l spanned by rank one matrices which satisfy dim S(T)⊥ = l. In these case we obtain necessary conditions for T ∈ Cm×n×l to have a border rank l at most. Lemma 6.4 Assume that one of the following conditions hold. 1. 2 ≤ l ≤ m, n and T ⊂ Cm×n is an l-dimensional subspace spanned by l rank one l , where u1, . . . , ul ∈ Cm and v1, . . . , vl ∈ Cn are linearly 1 , . . . , ulv⊤ matrices u1v⊤ independent. 2. m = n = l − 1 ≥ 3 and T ⊂ C(l−1)×(l−1) is an l-dimensional subspace spanned by l l , where any l − 1 vectors out of u1, . . . , ul ∈ Cl−1 rank one matrices u1v⊤ and v1, . . . , vl ∈ Cl−1 are linearly independent. 1 , . . . , ulv⊤ trices which are simultaneously diagonable by congruency. Hence for any T ∈ Γ(l, Cm×n) Then dim S(T) = (cid:0) l the matrix C(T) has rank at most (cid:0) l rank not more than l, then the l 3-slices Tk := [ti,j,k]m,n identities given by the vanishing of all (cid:0) l 2(cid:1), dim S(T)⊥ = l, and S(T)⊥ spanned by l ranks one symmetric ma- 2(cid:1). In particular if T = [ti,j,k] ∈ Cm×n×l has border i=j=1 ∈ Cm×n, k = 1, . . . , l satisfy the 2(cid:1) + 1 minors of C(T1, . . . , Tl). Proof. Let ei := (δi1, . . . , δim)⊤, fj := (δj1, . . . , δjn)⊤ for i = 1, . . . , m and j = 1, . . . , n. Consider first that the case 1. Since u1, . . . , ul and v1, . . . , vl are linearly independent, it follows that there exist P ∈ GL(m, C), Q ∈ GL(n, C) such that P ui = ei, Qvi = fi, i = 1, . . . , l. Let T ′ for i = 1, . . . , l. Thus T ′(z) = diag(z1, . . . , zl, 0, . . . , ). The only nonzero 2 × 2 minors of T ′(z) are T ′(z)[α, α] where α = (p, q) and 1 ≤ p < q ≤ l. So S(P T1Q⊤, . . . , P TlQ⊤) consists of all symmetric matrices A ∈ S(l, C) with zero diago- i )Q⊤ = eif ⊤ i i := P (uiv⊤ nal. Hence dim S(T) = (cid:0) l matrices which are simultaneously diagonable by congruency. 2(cid:1), dim S(T)⊥ = l and S(T)⊥ spanned by l ranks one symmetric Consider now the case 2. The arguments of the proof of Lemma 4.1 yield that we may assume that ui = vi = ei for i = 1, . . . , l − 1 and ul = vl = w = (w1, . . . , wl−1)⊤, where wi 6= 0 for i = 1, . . . , l − 1. Let Ti = uiv⊤ for i = 1, . . . , l. A straightforward calculation i shows that det T (z)[(1, 3), (1, 2)] = w2w3z1zl. Similarly, we have all the quadratic forms of the form wj wkzizl for i = 2, . . . , l − 1, were i, j, k are three distinct elements in hl − 1i. By considering det T (z)[α, α], α ∈ 2hl−1i we deduce that the space of quadratic polynomials spanned by det T (z)[α, β], α, β ∈ 2hl−1i contains all the monomials zizj for i 6= j. Hence S(T) is all A ∈ S(l, C) with zero diagonal entries. I.e. dim S(T) = (cid:0) l S(T)⊥ spanned by l ranks one symmetric matrices which are simultaneously diagonable by 2(cid:1), dim S(T)⊥ = l and 2 2 23 congruency. Other claims of the lemma follow straightforward from the continuity argu- ment. ✷ Theorem 6.5 Assume that T ∈ Cm×n×l, dim T3(T ) = l and dim S(T3(T ))⊥ = l. Then 1. rank T = l if and only if the following conditions hold (a) S(T3(T ))⊥ contains an invertible matrix. (b) The condition (1.1) holds for any three matrices A, B, C ∈ S(T3(T ))⊥. (c) A adj B have l distinct eigenvalues for some A, B ∈ S(T3(T ))⊥. 2. Assume that either 2 ≤ l ≤ m, n or m = n = l − 1 ≥ 3. dim S(T3(T ))⊥ = l then S(T3(T ))⊥ ∈ Γ(l, S(l, C)). (2.9) holds for p = 1, . . . , l − 1 and any three matrices A, B, C ∈ S(T3(T ))⊥. If brank T = l and In particular, the conditions Proof. Let U := S(T3(T ))⊥ ⊂ S(l, C). Suppose that rank T = l. Lemma 6.2 yields that U contains l linearly independent rank one symmetric matrices which are simultaneously diagonable by a congruency. Since dim U = l we deduce that U has a basis of the form R⊤ diag(δ1k, . . . , δlk)R ∈ S(T)⊥ for k = 1, . . . , l. Hence the conditions 1a-1c hold. Suppose that the condition 1a holds. By considering T (Im, In, R) for some R ∈ GL(l, C) we may assume that I ∈ U. The condition 1b for B = I yields that U is a subspace of commuting matrices. Hence, there exists a unitary matrix V such that V UV ∗ is an upper triangular matrix, e.g. [9, §24.2, Fact 3]. The assumption 1c yields that most of the matrices of the the form Aadj B have simple eigenvalues. Choose B ∈ GL(l, C) ∩ U. So adj B = 1 det B B−1. Thus, most of the matrices of the form AB−1 have simple eigenvalues. Hence most of the matrices in U have simple eigenvalues. Choose A ∈ U such that A has simple eigenvalues. Thus any C ∈ U is a polynomial in A. Since dim U = l it follows that U has a basis of l-rank one commuting matrices which are simultaneously diagonable by an orthogonal matrix. Lemma 6.2 yields that rank T = l. Assume that brank T = l. Hence T is a limit of rank l tensors Tq, q ∈ N satisfying the conditions of Lemma 6.4. Since dim S(T3(T ))⊥ = l we deduce that S(T3(T ))⊥ is the limit of S(T3(Tq))⊥, q ∈ N. (Without this assumption we can only deduce that any convergent subsequence of subspaces in the sequence S(T3(Tq))⊥, q ∈ N converges to a subspace of S(T3(T ))⊥.) Apply for each Tq part 1 to deduce part 2. ✷ Note that the a simultaneous matrix diagonalization by congruence arises naturally in finding the rank decomposition of tensors [4]. As in our characterization of V4(3, 3, 4) we can restate the conditions (1b) of Theorem 6.5 in terms of some polynomial equations. These equations will also hold for any T satisfying brank T ≤ l. References [1] E.S. Allman and J.A. Rhodes, Phylogenic ideals and varieties for general Markov model, Advances in Appl. Math., 40 (2008) 127-148. [2] R.A. Brualdi, S. Friedland and A. Pothen, The sparse basis problem and multilinear algebra, SIAM J. Matrix Anal. Appl. 16 (1995), 1-20. [3] J. Buczynski and J. M. Landsberg, Ranks of tensors and a generalization of secant varieties, arXiv:0909.4262. [4] L. DeLathauwer, A link between the canonical decomposition in multilinear algebra and simultaneous matrix diagonalization , SIAM Journal of Matrix Analysis and Applications 28 (2006), 642-666. 24 [5] S. Friedland, On the generic rank of 3-tensors, arXiv:0805.3777v4. [6] S. Friedland and C. Krattenthaler, 2-adic valuations of certain ratios of products of factorials and applications, Linear Algebra and its Applications, 426 (2007), 159- 189. [7] R.M. Guralnick and B.A. Sethuraman, Commuting pairs and triples of matrices and related varieties Linear Algebra and its Applications, 310 (2000), 139-148. [8] J. Harris and L.W. Tu, On symmetric and skew-symmetric determinantal varieties, Topology 23 (1984), 71 -- 84. [9] L. Hogben, Handbook of Linear Algebra CRC Press, 2007. [10] J.M. Landsberg and L. Manivel, On the ideals of secant varieties of Segre varieties, Found. Comput. Math. 4 (2004), 397422. [11] J.M. Landsberg and L. Manivel, Generalizations of Strassen's equations for secant varieties of Segre varieties, Comm. Algebra 36 (2008), 405 -- 422. [12] M. Marcus and H. Minc, A Survey of Matrix Theory and Matrix Inequalities, Prindle, Weber & Schmidt, Boston, 1964. [13] T. Motzkin and O. Taussky-Todd, pairs of matrices with property L. II, Trans. AMS 80 (1955), 387-401. [14] L. Oeding and D.J. Bates, Toward a salmon conjecture, arXiv:1009.6181v1. [15] L. Pachter and B. Sturmfels, Algebraic Statistics for Computational Biology, Cam- bridge University Press, 2005. [16] V. Strassen, Rank and optimal computations of generic tensors, Linear Algebra Appl. 52/53: 645-685, 1983. [17] B. Sturmfels, Open problems in algebraic statistics, in Emerging Applications of Algebraic Geometry, (editors M. Putinar and S. Sullivant), I.M.A. Volumes in Math- ematics and its Applications, 149, Springer, New York, 2008, pp. 351-364. Acknowledgement : I thank J. M. Landsberg and the referees for useful remarks. 25
1008.0621	1	1008	2010-08-03T19:11:21	Moduli of varieties of general type	[ "math.AG" ]	This is a survey paper discussing the moduli problem for varieties of general type.	math.AG	math	May 29, 2018 MODULI OF VARIETIES OF GENERAL TYPE J ´ANOS KOLL ´AR Contents 1. Introduction Basic principles 2. Canonical models 3. Semi log canonical models Singularities of semi log canonical surfaces Du Bois singularities 4. Moduli of semi log canonical models 5. Coarse moduli spaces 6. Moduli of slc pairs References 1 6 7 8 9 11 14 18 19 21 1. Introduction 1 (Moduli theories). The development of a good moduli theory consists of four basic steps. (1.1) Identify a class of objects whose moduli theory is nice. In some cases the answer is obvious, for instance, we should study the moduli theory of smooth projective curves. In other cases, it took some time to understand what the correct objects should be: Abelian varieties should be replaced by polarized Abelian varieties, K3 surfaces by marked K3 surfaces and vector bundles by stable vector bundles. We see later that smooth, projective varieties of general type should be replaced by their canonical models. We discuss this in Section 2. In all these cases, we get non-proper moduli spaces. This is inconvenient for many reasons, for instance it is hard to count objects with various properties. To remedy this, one should look for a compactification whose points have clear geometric meaning. (1.2) Choose a larger class of objects to form a proper moduli space. The choice of these objects is usually neither obvious nor unique. It was not until the 1960's that the importance of this step was understood and stable curves and semi-stable sheaves were identified and studied in detail. For surfaces of general type the right class was described in [KSB88] and for polarized Abelian varieties in [Ale02]. The solution for varieties of general type is treated in Section 3. (1.3) Establish the correct moduli functor. Once a class of objects V is established, the corresponding moduli functor is ususally declared to consist of all flat families whose fibers are in V. However, for varieties of general type, allowing all flat families gives the wrong moduli functor. The problem and the solution are analyzed in Section 4. 1 2 J ´ANOS KOLL ´AR (1.4) Study the resulting moduli spaces and their applications. The moduli of curves and the moduli of semi-stable vector bundles on curves appear in many contexts and by now established themselves as one of the richest applications of algebraic geometry. We are only at the beginning of the development of the moduli of higher dimensional varieties; the basic results are outlined in Section 5. I hope to see many more applications in the future. These notes are intended to give a survey of the subject, stressing key examples and results. The forthcoming book [Kol10b] aims to give a complete treatment. Definition 2 (Moduli functors). Let V be a "reasonable" class of projective varieties (or schemes, or sheaves, or ...). For the next definition we only need to assume that if K ⊃ k is a field extension then a k-variety Xk is in V iff XK := Xk ×Spec k Spec K is in V. Define the corresponding moduli functor as Flat families X → T such that every fiber is in V, modulo isomorphisms over T . (2.1)   V arietiesV(T ) :=   (As noted in (1.3), we will need to impose additional restrictions eventually, but for now let us ignore these.) 3 (Moduli spaces). We say that a scheme ModuliV, or, more precisely, a flat mor- phism u : UnivV → ModuliV is a fine moduli space for the functor V arietiesV if, for every scheme T , pulling back gives an equality V arietiesV(T ) = Mor(cid:0)T, ModuliV(cid:1). Our aim is to understand all families whose fibers are in V and a fine moduli space presents the answer in the most succinct way. Applying the definition to T = Spec K, where K is a field, we see that every fiber of u : UnivV → ModuliV is in V and the K-points of ModuliV are in one-to-one correspondence with the K-isomorphism classes of objects in V. We consider the existence of a fine moduli space as the ideal possibility. Unfor- tunately, it is rarely achieved. When there is no fine moduli space, we still can ask for a scheme that best approximates its properties. Therefore, we look for schemes M for which there is a natural transformation of functors TM : V arietiesg(∗) −→ Mor(∗, M ). Such schemes certainly exist, for instance, if we work over a field k then M = Spec k. All schemes M for which TM exists form an inverse system which is closed under fiber products. Thus, as long as we are not unlucky, there is a universal (or largest) scheme with this property. Though it is not usually done, it should be called the categorical moduli space. This object can be rather useless in general. For instance, fix n, d and let Hn,d up to isomorphisms. One can be the class of all hypersurfaces of degree d in Pn+1 see that a categorical moduli space exists and it is Spec k. k In order to get something reasonable, we impose extra conditions. A scheme ModuliV is a coarse moduli space for V if the following hold. MODULI OF VARIETIES OF GENERAL TYPE 3 (1) There is a natural transformation of functors ModMap : V arietiesV(∗) −→ Mor(∗, ModuliV), (2) ModuliV is universal satisfying (1), and (3) for any algebraically closed field K ⊃ k, ModMap : V arietiesV(Spec K) ∼=−→ Mor(Spec K, ModuliV) = ModuliV(K) is an isomorphism (of sets). In many cases, the naturally occurring moduli spaces have a further very useful property. (4) There is a VU → U in V such that the corresponding moduli map U → ModuliV is open and quasi finite. Following woodworking terminology, I propose to call a moduli space satisfying conditions (1–4) a bastard moduli space. 4 (Problems with the moduli of smooth varieties). In contrast with curves, the moduli theory for higher dimensional smooth vari- eties can be very badly behaved, as shown by the following examples. (4.1) (Ruled surfaces) Let C be a smooth curve and L a line bundle on C that t , with first projecion π1, consider is generated by 2 sections f, g. On S := C × A1 the exact sequence 0 → π∗ 1 L−1 (t,f,g) −→ π∗ 1L−1 + OS + OS → Q → 0. t . We can view PC×A1 t Q is a P1-bundle over Q is a rank 2 vector bundle on C × A1 S, or as a flat family of ruled surfaces over A1. If t 6= 0 then t : π∗ 1L−1 → π∗ 1 L−1 is an isomorphism, thus Qt ∼= OC + OC . If t = 0 then we get Q0 ∼= L−1 + coker(cid:2)L−1 (f,g) −→ OC + OC(cid:3) ∼= L−1 + L. Thus we get a flat family of smooth ruled surfaces whose general member is P1 × C and whose special member is PC (L−1 + L). In a coarse moduli space over C both of these should correspond to C-points, but the above family shows that the impossible (at least for schemes but not for stacks). One can be even more specific for C = P1. Minimal ruled surfaces over P1 moduli point (cid:2)PC(L−1 + L)(cid:3) is in the closure of the moduli point (cid:2)P1 × C(cid:3). This is are Fm := PP1(cid:0)OP1 + OP1(m)(cid:1) for m ≥ 0. The "moduli space" has 2 connected There are no closed points in this "moduli space;" the closure of (cid:2)Fm(cid:3) consists components, corresponding to even and odd values of m. of all the points (cid:8)(cid:2)Fm(cid:3),(cid:2)Fm+2(cid:3),(cid:2)Fm+4(cid:3), . . .(cid:9). (4.2) (Abelian, elliptic and K3 surfaces) A general problem in all these cases is that a typical deformation of such an algebraic surface over C is a non-algebraic complex analytic surface. Thus any algebraic theory captures only a small part of the full moduli theory. (4.3) For Abelian varieties and for K3 surfaces, the moduli spaces look very strange topologically. For instance, the 3-dimensional space of Kummer surfaces is dense in the 20-dimensional space of all K3 surfaces [PSS71]. 4 J ´ANOS KOLL ´AR This can be corrected by fixing a basis in H 2(∗, Z), but it is not clear how similar tricks work in general. Also, as it happens already for stable curves, we would like to consider families where not all fibers are homeomorphic to each other. Then it is no clear what one means by " fixing a basis in H 2(∗, Z)." (4.4) (Repeated blow-ups lead to non-separatedness) Let f : X → B be a smooth family of projective surfaces over a smooth (affine) pointed curve b ∈ B. Let C1, C2, C3 ⊂ X be three sections of f , all passing through a point xb ∈ Xb that intersect pairwise transversally at xb and are disjoint elsewhere. Set X 1 := BC1 BC2BC3X, where we first blow-up C3 ⊂ X, then the birational transform of C2 in BC3X and finally the birational transform of C1 in BC2BC3X. Similarly, set X 2 := BC1BC3BC2 X. Since the Ci are sections, all these blow-ups are smooth families of projective surfaces over B. It is easy to check that (1) all the fibers are smooth, projective surfaces of general type, (2) X 1 → B and X 2 → B are isomorphic over B \ {b}, (3) the fibers X 1 This type of behavior happens every time we look at deformations of a surface b are not isomorphic. b and X 2 with at least 3 points blown-up. (4.5) (Non-separatedness for minimal resolutions.) Let X0 := (cid:0)f (x1, . . . , x4) = 0(cid:1) ⊂ P3 be a surface of degree n that has an ordinary double point at p = (0:0:0:1) as its sole singularity and contains the pair of lines (x1x2 = x3 = 0). Let g be homogeneous of degree n − 1 such that xn−1 appears in it with nonzero coefficient. Consider the family of surfaces 4 X := (cid:0)f (x1, . . . , x4) + tx3g(x1, . . . , x4) = 0(cid:1) ⊂ P3 x × A1 t . Note that Xt is smooth for general t 6= 0 and X contains the pair of smooth surfaces (x1x2 = x3 = 0). For i = 1, 2, let X i := B(xi,x3)X denote the blow-up of (xi = x3 = 0) with induced morphisms πi : X i → X and fi : X i → A1. There is a natural birational map φ := π−1 2 ◦ π1 : X 1 99K X 2. Let BpX denote the blow-up of p = (cid:0)(0:0:0:1), 0(cid:1) with exceptional divisor E ⊂ BpX. One checks that (1) all the fibers are smooth, projective minimal models, (2) X 1 → B and X 2 → B are isomorphic over B \ {b}, (3) the fibers X 1 b are isomorphic, but (4) X 1 → B and X 2 → B are not isomorphic. While it is not clear from our construction, similar problems happen for any smooth family of surfaces where the general fiber has ample canonical class and a special fiber has nef (but not ample) canonical class, see [Art74, Bri68, Rei80]. b and X 2 5 (Answers to these problems). The problems (4.1–3) come from the global geom- etry of the varieties that we work with. The current assumption is that the moduli problem of uniruled varieties is usually pathological. Furthermore, any general attempt to create a good moduli functor ends up with a theory that is not compatible with birational equivalence. (Although there are examples, like the moduli of smooth hypersurfaces of degree n in Pn for n ≥ 4, where the biregular moduli theory ends up being birationally invariant. However, even in these cases, it is not clear that a sensible compactification exists.) For varieties with trivial canonical class one should get a nice moduli theory only after some suitable "rigidification." This can consist of choosing a basis in MODULI OF VARIETIES OF GENERAL TYPE 5 H 2(∗, Z) or fixing an ample divisor. The compactification question is mostly still unsolved. For instance, a geometrically meaningful compactification of the moduli of K3 surfaces is yet to be found. The problems (4.4–5) are more local. The aim of these notes is to explain how to deal with them for varieties of general type. The solution is to work with canonical models instead of smooth varieties of general type. Following [Iit71] (see [Laz04, Sec.2.1.C] for details) a smooth projective variety X of dimension n is of general type if the following equivalent conditions hold: (1) h0(cid:0)X, OX (mKX )(cid:1) ≥ ǫ · mn for some ǫ > 0 and m ≫ 1. (2) Proj R(X, KX) has dimension n. (3) The natural map X 99K Proj R(X, KX) is birational. The main reason, however, why we do not study the moduli functor of smooth varieties up to isomorphism is that, in dimensions two and up, smooth projective varieties do not form the smallest basic class. Given any smooth projective variety X, one can blow up any set of points or subvarieties Z ⊂ X to get another smooth projective variety BZX which is very similar to X. Therefore, the basic object should be not a single smooth projective variety but a whole birational equivalence class of smooth projective varieties. Thus it would be better to work with smooth, proper families X → S modulo birational equivalence over S. That is, with the moduli functor GenT ypebir(S) :=   Smooth, proper families X → S, every fiber is of general type, modulo birational equivalence over S. (5.4)   In essence this is what we end up doing, but it is very cumbersome do deal with birational equivalence over a base scheme. The following result, or rather, its proof, shows the way to a good moduli theory for varieties of general type. Proposition 6. Let fi : X i → B be two smooth families of projective varieties over a smooth curve B such that the canonical classes KX i are fi-ample. Then every isomorphism between the generic fibers φ : X 1 k(B) extends to an isomorphism Φ : X 1 ∼= X 2. ∼= X 2 k(B) Proof. Let Γ ⊂ X 1 ×B X 2 be the closure of the graph of φ. Let Y → Γ be the normalization, with projections pi : Y → X i and f : Y → B. We use the canonical class to compare the X i. Since the X i are smooth, KY ∼ p∗ i KX i + Ei where Ei is effective and pi-exceptional. (6.1) Since (pi)∗OY (mEi) = OX i for every m ≥ 0, we get that (fi)∗OX i(cid:0)mKX i(cid:1) = (fi)∗(pi)∗OX i(cid:0)mp∗ = (fi)∗(pi)∗OX i(cid:0)mp∗ = (fi)∗(pi)∗OY(cid:0)mKY(cid:1) = i KX i(cid:1) i KX i + mEi(cid:1) = = f∗OY(cid:0)mKY(cid:1). Since the KX i are fi-ample, X i = ProjB Pm≥0(fi)∗OX i(cid:0)mKX i(cid:1). Putting these together, we get the isomorphism Φ : X 1 ∼= ProjB Pm≥0(f1)∗OX 1(cid:0)mKX 1(cid:1) ∼= ∼= ∼= ProjB Pm≥0 f∗OY(cid:0)mKY(cid:1) ∼= ProjB Pm≥0(f2)∗OX 2(cid:0)mKX 2(cid:1) ∼= X 2. (cid:3) 6 J ´ANOS KOLL ´AR Note that the smoothness of the X i is used only through the pull-back formula (6.1). This leads to the first major definition: Preliminary Definition 7. Let B be a smooth curve, X a normal variety and f : X → B a non-constant projective morphism. Assume that (1) m0KX is Cartier for some m0 > 0. (This is needed to make sense of (2) and also to define the pull-back in (3).) (2) m0KX is f -ample. (3) If p : Y → X is a resolution of singularities then where E(m0) is effective and p-exceptional. m0KY ∼ p∗(cid:0)m0KX(cid:1) + E(m0) We do not want to keep carrying the m0 along, thus we switch to Q-divisors and Q-linear equivalence and write (3) as (3') If p : Y → X is a resolution of singularities then KY ∼Q p∗KX + E where E is an effective, p-exceptional Q-divisor. With these assumptions we can make the following informal definitions: (4) A "general" fiber of f is a canonical model. (5) A "special" fiber of f is a semi log canonical model if (1–3') continue to hold for every base change X ′ := X ×B B′ → B′, for every smooth curve B′. In this area, "semi" refers to allowing non-normal schemes and "log" refers to allowing exceptional divisors with coefficients ≥ −1, see (15). (Note that (1–2) are inherited by every base change, thus the only question is (3'). Already for curves, this condition is necessary. Indeed, let (f (x, y, z) = 0) be any plane curve with isolated singularities and (g(x, y, z) = 0) a general smooth plane curve. Then is smooth near the t = 0 fiber. After the base change t = sm we get t (cid:0)f (x, y, z) + tg(x, y, z) = 0(cid:1) ⊂ P2 × A1 (cid:0)f (x, y, z) + smg(x, y, z) = 0(cid:1) ⊂ P2 × A1 s and now (3') is satisfied for m ≥ 3 iff (f (x, y, z) = 0) has only ordinary nodes.) Basic principles. The moduli theory of higher dimensional varieties of general type is governed by the following four basic principles. Principle 8. Canonical models are the correct higher dimensional analogs of smooth projective curves of genus ≥ 2. Principle 9. Semi log canonical models are the correct higher dimensional analogs of stable projective curves of genus ≥ 2. Principle 10. Flat families of canonical models form the correct higher dimen- sional open moduli problem. Principle 11. Flat families of semi log canonical models do not form the correct higher dimensional compactified moduli problem. MODULI OF VARIETIES OF GENERAL TYPE 7 2. Canonical models As noted in (8), canonical models are the basic objects of our moduli theory. Definition 12. A normal projective variety X is a canonical model iff the following hold. (1) m0KX is Cartier for some m0 > 0. (2) for every (equivalently, for one) resolution of singularities f : X ′ → X there is an effective, f -exceptional Q-divisor E such that KX ′ ∼Q f ∗KX + E. (3) KX is ample. Singularities satisfying (1–2) are called canonical. This assumption implies that the canonical ring of X R(X, KX) := Xm≥0 H 0(cid:0)X, OX (mKX )(cid:1) is isomorphic to the canonical ring of X ′ In particular, R(X ′, KX ′) := Xm≥0 H 0(cid:0)X ′, OX ′(mKX ′ )(cid:1). X = Projk R(X, KX) = Projk R(X ′, KX ′) is the unique canonical model in the birational equivalence class of X. Now we know [BCHM06, Siu08] that the canonical ring R(X, KX) of a smooth projective variety of general type is always finitely generated, thus X can is a pro- jective variety. It is not obvious, but true, that X can is a canonical model [Rei80]. Definition 13 (Moduli functor of canonical models). The moduli functor of canon- ical models is Flat, proper families X → S, every fiber is a canonical model, modulo isomorphisms over S. (13.1) CanM od(S) :=     This is an improved version of the birational moduli functor GenT ypebir(∗) (5.1). (Traditionally this was considered to be the obviously correct definition, but, in view of Principle 11, it needs an explanation. For details, see (30).) By a theorem of [Siu98], in a smooth, proper family of varieties of general type the canonical rings form a flat family and so do the canonical models. Thus there is a natural transformation TCanMod : GenT ypebir(∗) → CanM od(∗). By definition and by (6), if Xi → S are two smooth, proper families of varieties of general type then TCanMod(X1/S) = TCanMod(X2/S) iff X1 and X2 are birational, thus TCanMod is injective. It is, however, not surjective, but we have the following partial surjectivity statement. 8 J ´ANOS KOLL ´AR Let Y → S be a flat family of canonical models. Then there is a dense open subset S0 ⊂ S and a smooth, proper family of varieties of general type Y 0 → S0 such that TCanMod(Y 0/S0) = [X 0/S0]. Some of the obstruction to surjectivity are obvious but some, as in (4.3), are quite subtle. Canonical curves are exactly the smooth, projective curves of genus ≥ 2. Canon- ical surfaces have at most Du Val singularities (also called rational double points). Starting with dimension 3, we get more complicated singularities. For instance, (cid:0)xa0 0 + · · · + xan n = 0(cid:1) ⊂ An+1 > 1 and a quotient of An be a finite subgroup G without quasi-reflections is canonical iff for every g ∈ G (g 6= 1) with eigenvalues e2πicj ai is canonical iff P 1 (where 0 ≤ cj < 1) we have Pj cj ≥ 1. The most important general property of canonical singularities is the following. For short proofs see [KM98, Sec.5.1] or [Kol10a]. Theorem 14. [Elk81] Let X be a canonical model over a field of characteristic 0. Then X has rational singularities. That is, Rif∗OY = 0 for i > 0 for every resolution of singularities f : Y → X. Using the covering trick of [Rei80] this implies that the reflexive hulls ω[m] X are CM (=Cohen-Macaulay) for every m. 3. Semi log canonical models First we translate (7.5) into a proper definition. Let B be a smooth curve, X a normal variety and f : X → B a non-constant projective morphism. When is the fiber Xb a semi log canonical model? By [KKMSD73] there is a smooth pointed curve b′ ∈ B′ and a finite morphism (b′ ∈ B′) → (b ∈ B) such that the base change f ′ : X ′ := X ×B B′ → B′ has a resolution π : Y ′ → X ′ such that (f ′ ◦ π)−1(b′) ⊂ Y ′ is a reduced simple normal crossing divisor. We can also assume that the birational transform Y ′ b′ is smooth. Thus (f ′ ◦ π)−1(b′) = Y ′ b′ + F where F is a reduced simple normal crossing divisor. By (7.3), b′ := π−1 ∗ X ′ where E′ is effective. Since Y ′ KY ′ + Y ′ Restricting to Y ′ b′ we get KY ′ ∼Q π∗KX ′ + E′ b′ + F ′ = π∗(cid:0)X ′ b′ ∼Q π∗(cid:0)KX ′ + X ′ b′ ∼Q π∗KX ′ KY ′ b′ +(cid:0)E′ − F(cid:1)Y ′ b′ . b′(cid:1), we can rewrite the above as b′(cid:1) + E′ − F. Since E′ is effective, its restriction is again effective, but the restriction of −F brings in negative coefficients. However, none of these is smaller than −1 since F and Y ′ b′ intersect transversally. Definition 15. A reduced, projective variety X is a semi log canonical model or slc model iff the following hold. (1) m0KX is Cartier for some m0 > 0. (2) X has only ordinary nodes in codimension 1. MODULI OF VARIETIES OF GENERAL TYPE 9 (3) For every resolution of singularities f : X ′ → X there is an f -exceptional Q-divisor E = Pi aiEi such that KX ′ ∼Q f ∗KX +PiaiEi (4) KX is ample. and ai ≥ −1 for every i. One should think of this as combining a global condition (4) with purely local conditions (1–3). Singularities satisfying (1–3) are called semi log canonical or slc. For slc models it is usually better to use semi resolutions, that is, a proper birational morphism g : X s → X such that X s has only double normal crossing points (xy = 0) ⊂ Cn+1 and pinch points (x2 = y2z) ⊂ Cn+1 and g maps the double locus of X s birationally on the double locus of X. Let E denote the (reduced) exceptional divisor of g. Then the canonical ring of X R(X, KX) := Xm≥0 H 0(cid:0)X, OX (mKX )(cid:1) is isomorphic to the semi log canonical ring of X s R(X s, KX s + E) := Xm≥0 H 0(cid:0)X s, OX s(mKX s + mE)(cid:1). This actually creates a lot of problems since semi log canonical rings are not always finitely generated [Kol07]. It is a quite subtle theorem that semi log canonical models actually satisfy the preliminary definion (7.5). This is proved in [K+92, 17.4] and [Kaw07]. To get a feeling for semi log canonical, let us review the classification of slc surface singularities. Singularities of semi log canonical surfaces. It is convenient to describe the singularities of log canonical surfaces by the dual graph of their minimal resolution. That is, given a singularity (s ∈ S) with minimal resolution g : X → S we draw a graph Γ whose vertices are the g-exceptional curves and two vertices are connected by an edge iff the corresponding curves intersect. We use the number −(Ei · Ei) to represent a vertex. In our examples, save in (16.4.a), all the exceptional curves are isomorphic to P1. Let det(Γ) denote the determinant of the negative of the intersection matrix of the dual graph. This matrix is positive definite for exceptional curves. For instance, if Γ = {2 − 2 − 2} then det(Γ) = det  2 −1 −1 0 −1 0 2 −1 2   = 4. 16 (List of log canonical surface singularities). Each case includes all previous ones. (16.1) Terminal = smooth. (16.2) Canonical = Du Val (= rational double point). (16.3) Log terminal = quotient of C2 by a finite subgroup of GL(2, C) that acts freely outside the origin. The order of the group is det(Γ). A more detailed list is the following: 10 J ´ANOS KOLL ´AR (a) (Cyclic quotient) (b) (Dihedral quotient) Here n ≥ 2 with dual graph c1 − · · · − cn c1 − · · · − cn (cid:30) (cid:31) 2 2 (c) (Other quotients) The dual graph has 1 fork Γ1 − c0 − Γ2 Γ3 whith 3 cases for (cid:0)det(Γ1), det(Γ2), det(Γ3)(cid:1): (Tetrahedral) (2,3,3) (Octahedral) (2,3,4) (Icosahedral) (2,3,5). (16.4) Log canonical (a) (Simple elliptic) Γ = {E} has a single vertex which is a smooth elliptic curve with self intersection ≤ −1. (b) (Cusp) Γ is a circle of smooth rational curves, at least one of them with with ci ≥ 3. (The cases n = 1, 2 are somewhat special.) cn − · · · − cr+1 c1 (cid:30) (cid:31) (cid:31) (cid:30) cr (c) (Z/2-quotient of a cusp or simple elliptic) Γ has 2 forks. c2 − · · · − cr−1 2 2 (cid:31) (cid:30) c1 − · · · − cn (cid:30) (cid:31) 2 2 (d) (Other quotients of a simple elliptic) The dual graph is as in (16.3.c) with 3 possibilities for (cid:0)det(Γ1), det(Γ2), det(Γ3)(cid:1): (Z/3-quotient) (3,3,3) (Z/4-quotient) (2,4,4) (Z/6-quotient) (2,3,6). If X is a non-normal semi log canonical surface singularity, then we describe its normalization ¯X together with the preimage of the double curve ¯B ⊂ ¯X. The extended dual graph (Γ, ¯B) has an additional vertex (repesented by •) for each local branch of ¯B connected to Ci if ( ¯B · Ci) 6= 0. 17 (List of semi log canonical surface singularities). There are 3 irreducible cases. (The number on some edges is the different, which we do not define here [K+92, Sec.16]. Their role is explained in (17.4). MODULI OF VARIETIES OF GENERAL TYPE 11 (17.1) (Cyclic quotient, one branch of ¯B) det Γ 1− 1 −−−− c1 − · · · − cn • (17.2) (Cyclic quotient, two branches of ¯B) 1 − c1 − · · · − cn 1 − • • (17.3) (Dihedral quotient) Here n ≥ 2 with dual graph 1 − c1 − · · · − cn • (cid:30) (cid:31) 2 2 (17.4) (Reducible cases) We can take several components as above and glue them together along two local branches of ¯B. The gluing is allowed only if we see the same numbers on the edges. Thus we can glue 2 copies as in (17.1) as long as both have the same det(Γ). Or we can take any number of those in (17.2), make a chain out of them and then either turn the chain into a circle or end it with a copy of (17.3). We are also allowed to glue a local branch of ¯B to itself by an involution. For instance, • − 1 glued to itself gives the pinch point (x2 = y2z) ⊂ A3. Note: The above dual graphs are correct in any characteristic, the descriptions as quotients are correct as long as the the characteristic does not divide the order of the group mentioned. Du Bois singularities. Semi log canonical singularities need not be rational, not even CM (=Cohen- Macaulay) and their most important property is that they are Du Bois. After some examples, we discuss Du Bois singularities and their useful properties. Example 18. It is easy to see that a cone over a smooth variety X ⊂ PN is log canonical iff KX ∼Q r · H for some r ≤ 0 where H is the hyperplane class. For us the interesting case is when KX ∼Q 0 (hence r = 0). For these, the cone is CM (resp. rational) iff H i(X, OX ) = 0 for 0 < i < dim X (resp. for 0 < i ≤ dim X). Thus we see the following. (1) A cone over an Abelian variety A is CM iff dim A = 1. (2) A cone over a K3 surface is CM but not rational. (3) A cone over an Enriques surface is CM and rational. (4) A cone over a smooth Calabi-Yau complete intersection is CM but not rational. The concept of Du Bois singularities was introduced by Steenbrink in [Ste83] as a weakening of rationality. The precise definition is rather involved, but our main applications rely only on the following consequence. Theorem 19. [KK09], [Kol10b, Chap3] Let X be a proper slc scheme over C. Then the natural map H i(X an, C) → H i(X an, OX an ) ∼= H i(X, OX ) 12 J ´ANOS KOLL ´AR is surjective for all i. (In fact, with a functorial splitting.) In studying moduli questions, it is very useful to know that certain numerical invariants are locally constant. All of these follow from the Du Bois property, via the following base-change theorem [DJ74, DB81]. Proposition 20. Let f : X → S be a flat, proper morphism. Assume that the fiber Xs is Du Bois for some s ∈ S. Then there is an open neighborhood s ∈ S0 ⊂ S such that, for all i, (1) Rif∗OX is locally free and compatible with base change over S0 and (2) s 7→ H i(Xs, OXs(cid:1) is a locally constant function on S0. Proof. By Cohomology and Base Change [Har77, III.12.11], the theorem is equivalent to proving that the restriction maps φi s : Rif∗OX → H i(Xs, OXs(cid:1) (20.3) are surjective for every i. By the Theorem on Formal Functions [Har77, III.11.1], it is enough to prove this when S is replaced by any 0-dimensional scheme Sn whose closed point is s. Thus assume from now on that we have a flat, proper morphism fn : Xn → Sn, s ∈ Sn is the only closed point and Xs is Du Bois. Then H 0(cid:0)Sn, Rif∗OX(cid:1) = H i(cid:0)Xn, OXn(cid:1), hence we can identify the φi s with the maps (20.4) ψi : H i(cid:0)Xn, OXn(cid:1) → H i(Xs, OXs(cid:1) By the Lefschetz principle we may assume that everything is defined over C. By GAGA (cf. [Har77, App.B]), both sides of (20.4) are unchanged if we replace Xn by the corresponding analytic space X an n . Let CXn (resp. CXs) denote the sheaf of locally constant fucntions on Xn (resp. Xs) and jn : CXn → OXn (resp. js : CXs → OXs ) the natural inclusions. We have a commutative diagram ji n ↓ H i(cid:0)Xn, CXn(cid:1) H i(cid:0)Xn, OXn(cid:1) αi → H i(Xs, CXs(cid:1) → H i(Xs, OXs(cid:1) ↓ ji s ψi Note that αi is an isomorphism since the inclusion Xs ֒→ Xn is a homeomorphism and ji (cid:3) s is surjective since Xs is Du Bois. Thus ψi is also surjective. A line bundle L on X is called f -semi ample if there is an m > 0 such that Lm is f -generated by global sections. Using cyclic coverings [KM98, Sec.2.4], there is r=0 L−r and f ◦ π : Y → S also a finite morphism π : Y → X such that π∗OY = Pm−1 has slc fiber over s. Thus (20) implies the following. Corollary 21. Let f : X → S be a proper and flat morphism with slc fibers over closed points; S connected. Let L be an f -semi ample line bundle on X. Then, for all i, (1) Rif∗(cid:0)L−1(cid:1) is locally free and compatible with base change and (2) H i(Xs, L−1 s (cid:1) is independent of s ∈ S. Choose L to be f -ample above. By [KM98, 5.72], Xs is CM iff H i(Xs, L−m for all m ≫ 1 and i < dim X. The latter properties are deformation invariant for slc fibers by (21). Thus we conclude: (cid:3) s (cid:1) = 0 MODULI OF VARIETIES OF GENERAL TYPE 13 Corollary 22. Let f : X → S be a projective and flat morphism with slc fibers over closed points; S connected. Then, if one fiber of f is CM then all fibers of f are CM. (cid:3) (Nore that for arbitrary flat, projective morphisms f : X → S, the set of points s ∈ S such that the fiber Xs is CM is open, but usually not closed.) The next example shows that non-CM varieties occur among the irreducible components of smoothable, CM and slc varieties. Example 23. Here is an example of a stable family of projective varieties {Yt : t ∈ T } such that (1) Yt is smooth, projective for t 6= 0, (2) KYt is ample and Cartier for every t, (3) Y0 is slc and CM, (4) the irreducible components of Y0 are normal, but (5) one of the irreducible components of Y0 is not CM. Let Z be a smooth Fano variety of dimension n ≥ 2 such that −KZ is very ample, for instance Z = P2. Set X := P1 × Z and view it as embedded by − KX into PN for suitable N . Let C(X) ⊂ PN +1 be the cone over X. Let M ∈ − KZ be a smooth member and consider the following divisors in X: D0 := {(0 : 1)} × Z, D1 := {(1 : 0)} × Z and D2 := P1 × M. Note that D0 + D1 + D2 ∼ −KX. Let Ei ⊂ C(X) denote the cone over Di. Then E0 + E1 + E2 is a hyperplane section of C(X) and (cid:0)C(X), E0 + E1 + E2(cid:1) is lc. For some m > 0, let Hm ⊂ C(X) be a general intersection with a degree m hypersurface. Then (cid:0)C(X), E0 + E1 + E2 + Hm(cid:1) is snc outside the vertex and is lc at the vertex. Set Y0 := E0 + E1 + E2 + Hm Since Y0 ∼ OC(X)(m + 1), we can view it as a slc limit of a family of smooth hypersurface sections Yt ⊂ C(X). The cone over X is CM, hence its hyperplane section E0 + E1 + E2 + Hm is also CM. However, E2 is not CM. To see this, note that E2 is the cone over P1 × M and, by the Kuneth formula, H i(P1 × M, OP1×M ) = H i(M, OM ) = (cid:26) k if i = 0, n − 1, otherwise. 0 Thus E2 is not CM. As in the proof of [Har77, III.9.9], we get from (21) the following. Proposition 24. Let f : X → S be a projective, flat morphism with slc fibers over closed points. Then ωX/S exists and is compatible with base change. That is, for any g : T → S the natural map g∗ XωX/S → ωXT /T is an isomorphism where gX : XT := X ×S T → X is the first projection. (cid:3) (This seems like a very complicated way to prove that ωX/S behaves as expected, but, as far as I can tell, this was not known before. A proof for non-projective algebraic maps is given in [Kol10a]. I do not know how to prove (24) for analytic morphisms f : X → S.) 14 J ´ANOS KOLL ´AR If the fibers Xs are CM, then H i(Xs, ωXs ⊗ Ls(cid:1) is dual to H n−i(Xs, L−1 In general, a more detailed inductive argument is needed s (cid:1), and the following is clear. [Kol10b, Chap.4]. Corollary 25. Let f : X → S be a projective, flat morphism with slc fibers over closed points; S connected. Then, for all i, (1) Rif∗ωX/S is locally free and compatible with base change and (2) H i(Xs, ωXs(cid:1) is independent of s ∈ S. (cid:3) 4. Moduli of semi log canonical models Let us illustrate (11) with the an example of a flat, projective family of surfaces with log canonical singularties over the pair of lines (xy = 0) ⊂ C2 such that over one line we have smooth surfaces with ample canonical class and over the other line we have smooth elliptic surfaces. Example 26 (Jump of Kodaira dimension). There are 2 families of nondegenerate degree 4 smooth surfaces in P5. One family consists of Veronese surfaces P2 ⊂ P5 embedded by O(2). The general member of the other family is P1 × P1 ⊂ P5 embedded by O(2, 1), special members are embeddings of the ruled surface F2. The two families are distinct since K 2 P2 = 9 and K 2 P1×P1 = 8. For both of these surface, a smooth hyperplane section gives a degree 4 rational normal curve in P4. Let T0 ⊂ P5 be the cone over the degree 4 rational normal curve in P4. T0 has a log canonical (even log terminal) singularity and K 2 T0 = 9. For us the intersting feature is that one can write T0 as a limit of smooth surfaces in two distinct ways, corresponding to the two possibilities of writing the degree 4 rational normal curve in P4 as a hyperplane section of a surface. From the first family, we get T0 as the special fiber of a flat family whose general fiber is P2. This family is denoted by {Tt : t ∈ C}. From the second family, we get T0 as the special fiber of a flat family whose general fiber is P1 × P1. This family is denoted by {T ′ t : t ∈ C}. (In general, one needs to worry about the possibility of getting embedded points at the vertex, but in both cases the special fiber is indeed T0.) Note that K 2 is constant in the family {Tt : t ∈ C} but jumps at t = 0 in the family {T ′ t : t ∈ C}. Next we take a suitable cyclic cover of the two families to get similar examples with ample canonical class. Let π0 : S0 → T0 be a double cover, ramified along a smooth quartic hypersurface 2 H where H is the hyperplane class. Thus, by the section. Note that KT0 ∼Q − 3 Hurwitz formula, KS0 ∼Q π∗ 0(cid:0)KT0 + 2H(cid:1) ∼Q 1 2 π∗ 0H. So S0 has ample canonical class and K 2 S0 = 2. Since π0 is ´etale over the vertex of T0, S0 has 2 singular points, locally (in the analytic or ´etale topology) isomorphic to the singularity on T0. Thus S0 is a log canonical surface. Both of the smoothings lift to smoothings of S0. MODULI OF VARIETIES OF GENERAL TYPE 15 From the family {Tt : t ∈ C} we get a smoothing {St : t ∈ C} where πt : St → P2 is a double cover, ramified along a smooth octic. Thus St is smooth, KSt ∼Q π∗ t OP2(1) is ample and K 2 From the family {T ′ t → P1 × P1 is a double cover, ramified along a smooth curve of bidegree (8, 4). One of the families of lines on P1 × P1 pulls back to an elliptic pencil on S′ = 0. t : t ∈ C} we get a smoothing {S′ t : t ∈ C} where π′ St = 2. t : S′ t and K 2 S ′ t In order to exclude such examples, we concentrate on the Hilbert function of a slc model. Definition 27 (Hilbert function of slc models). Let X be an slc model. Note that ωX is locally free outside a subscheme Z ⊂ X such that Z has codimension ≥ 2. Hence the reflexive hull ω[m] X over X \ Z. The Hilbert function of X is X (cid:1)∗∗ X := (cid:0)ω⊗m is isomorphic to ω⊗m If ω[N ] X is locally free, then HX (m) := χ(cid:0)X, ω[m] X (cid:1). ω[m0+mN ] X ∼= ω[m0] X ⊗ ω[mN ] X , thus HX (m0+mN ) is a polynomial in m. Thus we can view HX (m) as a polynomial in m whose coefficients are periodic functions (with period N ). We view χ(cid:0)X, ω[m] to insist that they stay constant in "good" families of slc models. Over a reduced base, this is enough to get the correct definition. X (cid:1) as the basic numerical invariants of X. It is then natural Definition 28 (Moduli of slc models over reduced bases). Let H(m) be an integer valued function. On reduced schemes, the moduli functor of semi log canonical models with Hilbert function H is SlcM odH(S) :=   Flat, proper families X → S, fibers are slc models with ample canonical class and Hilbert function H(m), modulo isomorphisms over S.   Over an arbitrary base, let f : X → S be a flat, proper family of slc models. Note that ωX/S is locally free outside a subscheme Z ⊂ X such that Z ∩ Xs has codimension ≥ 2 in each fiber. Each ω⊗m X/S is also locally free on X \ Z, hence it has a reflexive hull ω[m] X/S. If s ∈ S is a general point, then ω[m] X/SXs ∼= ω[m] Xs but for an abitrary s ∈ S we only have a restriction map rs : ω[m] X/SXs → ω[m] Xs which is, in general, neither injective nor surjective. The best way to ensure that every fiber of Xs has the same Hilbert function is to require these restriction maps to be isomorphisms for every s ∈ S. (It turns out that this is the only way, that is, the kernel and the cokernel of rs can nor cancel each other for every s, unless they are both zero.) This leads to our final definition.     16 J ´ANOS KOLL ´AR Definition 29 (Moduli of slc models). Let H(m) be an integer valued function. The moduli functor of semi log canonical models with Hilbert function H is SlcM odH(S) := Flat, proper families X → S, fibers are slc models with ample canonical class and Hilbert function H(m), ω[m] X/S is flat over S and commutes with base change, modulo isomorphisms over S. Aside 30. We can now explain Principle 10. The reason is that for flat families of canonical models, ω[m] X/S is automatically flat over S and commutes with base change. This follows from two special properties of canonical singularities. For simplicity, consider a flat family X → Spec k[ǫ] whose special fiber X0 is affine. First we use that, as a result of the classification of canonical surface singularities (16.2), there is an open subset j : U ֒→ X whose complement Z has codimension ≥ 3 such that ωU0 is locally free. Thus we have an exact sequence 0 → ǫ · ωm U0 → ωm U → ωm U0 → 0. By pushing it forward, we get 0 → ǫ · j∗ωm U0 → j∗ωm U → j∗ωm U0 → ǫ · R1j∗ωm U0 As noted after (14), ω[m] Therefore, This implies that ω[m] X is a CM sheaf, hence has depth ≥ 3 at every point of Z. U0 = H 1(cid:0)U0, ωm Z0(cid:0)X0, ω[m] U0(cid:1) = H 2 X0 (cid:1) = 0. R1j∗ωm X equals j∗ωm U and it is flat over k[ǫ]. Now that we have the correct definition, we need to prove that the corresponding deformation theory is reasonable. The key result is the following. Theorem 31. Let f : X → S be flat, projective morphism whose fibers are slc models. Let H be an integer valued function. Then there is a locally closed embedding SH ֒→ S such that a morphism g : T → S factors through SH iff X ×S T → T is in SlcM odH(T ). For surfaces, a proof of this is outlined in [Hac04], a general solution is in [AH09]. The following general theory of hulls [Kol08] applies in many similar contexts as well. Definition 32. Let X be a scheme over a field k and F a coherent sheaf on X. Set n := dim Supp F . The hull of F is the unique q : F → F [∗∗] such that (1) q is an isomorphism at all generic points of Supp F , (2) q is surjective at all codimension 1 points of Supp F , (3) F [∗∗] is S2. If X itself is normal, F is coherent and Supp F = X, then F [∗∗] is the usual reflexive hull F ∗∗ of F . The hull of a nonzero sheaf is also nonzero, in contrast with the reflexive hull which kills all torsion sheaves. One can construct F [∗∗] as follows. First replace F by F/ torsn−1(F ) where torsn−1(F ) is the largest subsheaf whose support has dimension ≤ n − 1. Then there is a closed subscheme Z ⊂ Supp F of codimension ≥ 2 such that F/ tors(F ) is S2 on X \ Z. Let j : X \ Z → X be the open embedding and take F [∗∗] = j∗(cid:16)(cid:0)F/ tors(F )(cid:1)X\Z(cid:17). MODULI OF VARIETIES OF GENERAL TYPE 17 Definition 33. Let f : X → S be a morphism and F a coherent sheaf. A hull of F is a coherent sheaf G together with a map q : F → G such that, (1) G is flat over S and (2) for every s ∈ S, the induced map qs : Fs → Gs is a hull (32). It is easy to see that a hull is unique if it exists. It is clear from the definition that hulls are preserved by base change. That is, if g : T → S is a morphism, XT := X ×S T and gX : XT → X the first projection then g∗ X G is also a hull. XF → g∗ Xq : g∗ Definition 34. Let f : X → S be a projective morphism and F a coherent sheaf on X. For a scheme g : T → S set Hull(F )(T ) = 1 if g∗ X F has a hull and Hull(F )(T ) = ∅ if g∗ X F does not have a hull, where gX : T ×S X → X is the projection. The main existence theorem is the following. Theorem 35 (Flattening decomposition for hulls). Let f : X → S be a projective morphism and F a coherent sheaf on X. Then (1) Hull (F ) has a fine moduli space Hull(F ). (2) The structure map η : Hull(F ) → S is a locally closed decomposition, that is, η is one-to-one and onto on geometric points and a locally closed embedding on every connected component. Applying (35) to the relative dualizing sheaf gives the following result. Corollary 36. Let f : X → S be projective and equidimensional. Assume that there is a closed subscheme Z ⊂ X such that codim(Xs, Z ∩ Xs) ≥ 2 for every s ∈ S, (X \ Z) → S is flat and ωX/S is locally free on X \ Z. Then, for any m there is a locally closed decomposition Sm → S such that for any g : T → S the following are equivalent X×S T /T is flat over T and commutes with base change. (1) ω[m] (2) g factors through Sm → S. Proof. We claim that Sm = Hull(cid:0)ω⊗m Z ×S T → X ×S T be the inclusion. Then X/S(cid:1). Given g : T → S, let jT : X ×S T \ ω[m] X×ST /T = (cid:0)jT(cid:1)∗g∗ X ω⊗m X\Z/S. X×ST /T commutes with restrictions to the fibers of X ×S T → T , then X×S T /T is the hull of ω⊗m X×ST /T . If T 7→ ω[m] X×ST /T has S2 fibers, hence ω[m] ω[m] Conversely, if ω⊗m further base changes by (33). X×S T /T has a hull then it is ω[m] X×ST /T and it commutes with (cid:3) In order to prove (31), choose N such that ω[N ] Xs is locally free for every s ∈ S. For 1 ≤ i ≤ N , let Si → S be as in (36). Take T to be the fiber product S1 ×S · · · ×S SN → S. Then ω[m] X×ST /T is flat over T and commutes with base change for every m. Thus SH is the disjoint union of those connected components of T where the Hilbert function is H. (cid:3) 18 J ´ANOS KOLL ´AR 5. Coarse moduli spaces Having defined the correct moduli functor for slc models, we can now get down to studying its properties and the corresponding moduli spaces. 37 (Valuative criterion of separatedness). The functor SlcM od satisfies the valu- ative criterion of separatedness, which is essentially (6). This was built into our construction. 38 (Valuative criterion of properness). The short answer is that SlcM od satisfies the valuative criterion of properness, but some warnings are in order. We proceed very much as for curves. We start with a family of canonical models over an open curve X 0 → B0. By the semi-stable reduction theorem of [KKMSD73], after a base change C0 → B0 and extending the family over a proper curve C ⊃ C0, there is a resolution g : Y → C all of whose fibers are reduced simple normal crossing divisors. Finally we replace Y by its relative canonical model Y c := ProjC Xm≥0 g∗(cid:0)ωm Y /C(cid:1). It is not hard to see that Y c → C is in SlcM od(C) extending X ×B0 C0 → C0. This establishes the valuative criterion of properness if canonical models are dense in the moduli of slc models. We probably mostly care about the irreducible components where canonical models are dense, so we could take this as the final answer. However, not all irreducible components are such, and it would be better to understand all of them. So let us start with a family of slc models X 0 → B0. We can proceed as above, but instead of the relative canonical model of Y we need to take the relative semi log canonical model. As we noted, semi log canonical rings are not always finitely generated [Kol07]. Here the solution is to normalize the family, construct the models of the nor- malization over C and then try to reconstruct the desired extension of the original family. This is actually quite subtle, see [Kol10b, Chap.3]. 39 (Existence of coarse moduli spaces). Fix a function H and an integer m. Let SlcM odH,m be the moduli functor of slc models with Hilbert funcion H for which ω[m] is locally free, very ample and has no higher cohomologies. All of these thus embedd into PN for N = H(m) − 1. We use a variant of (31) to show that there is a locally closed subscheme SH,m of the Hilbert scheme Hilb(PN ) that parametrizes families of m-canonically embedded slc models with Hilbert function H. The general quotient theorems of [Kol97], [KM97] apply and we obtain the coarse moduli space SlcModH,m of SlcM odH,m as the geometric quotient SH,m/ Aut(PN ). Finally we let m run through the sequence 2!, 3!, 4!, . . . to get an increasing sequence of coarse moduli spaces whose union is the coarse moduli space SlcModH . For now we know only that it is a separated algebraic space which is locally of finite type. 40 (Properness). We saw that SlcModH satisfies the valuative criterion of proper- ness, hence it is proper iff it is of finite type. The components where the canonical models are dense were studied by [Kar00]. He proves that one can control the procedure outlined in (38) uniformly. Thus every such component is of finite type, hence proper. MODULI OF VARIETIES OF GENERAL TYPE 19 With some modifications, this implies that every irreducible component of SlcModH is proper. Thus the only remaining question is: can there by infinitely many irreducibe components? To illustrate some of the difficulties, let us consider a much simpler question: can we bound the number of irreducible components of a slc surface S with Hilbert function H? For curves the answer is easy. If C = ∪iCi then 2g(C) − 2 = deg ωC = Pi deg(cid:0)ωCCi(cid:1). Each Ci on the right hand side contributes at least 1, hence there are at most 2g − 2 irreducible components. In the surface case, we have something very similar. If S = ∪iSi then we can compute the self intersection of the canonical class as The unexpected propblem is that KS is only a Q-Cartier divisor, hence each sum- mand on the right hand side is a positive rational number, not an integer. (cid:0)KS · KS(cid:1) = Pi(cid:0)KSSi · KSSi(cid:1). We are, however, saved if the contributions on the right are bounded away from 0. This, and much more that is needed for boundedness was proved in [Ale94] and improved in [AM04]. The lower bound 1 1764 was established in [Kol94]. (I do not know the optimal bound, but(cid:0)P(3, 4, 5), (x3y+y2z+z2x = 0)(cid:1) has (KS +D)2 = 1 60 .) In higher dimensions, recent work of Hacon-McKernan-Xu establishes a lower bound for (cid:0)KXXi(cid:1)n ; the methods are likely to give boundedness as well. 41 (Projectivity). The method of [Kol90] shows that every proper subscheme of SlcMod is projective. For m sufficiently divisible, the 1-dimensional vector spaces det H 0(cid:0)X, ω[m] The proof uses the Nakai-Moishezon ampleness criterion, thus it works only for X (cid:1) naturally glue together to an ample line bundle. proper schemes. It seems very hard to give quasi-projectivity criteria. For instance, [Kol07] gives an example of a normal crossing surface S with a line bundle L and normalization π : ¯S → S such that π∗L is ample yet L is not ample, in fact no power of L is generated by global sections. 6. Moduli of slc pairs In dimension 1, it is useful to consider not just the moduli of curves but also the moduli of pointed curves. Similarly, in higher dimensions, one should consider the moduli of pairs (X, ∆) where ∆ = Pi aiDi is a linear combination of divisors with coefficients 0 ≤ ai ≤ 1. These were first considered in [Ale96]. The first task is to define slc singularities of pairs. This is actually quite natural, see [K+92]. By contrast, finding the correct analog of (29) turns out to be a quite thorny problem. Instead of going into details, let me just present a key example, due to Hassett, which shows that in general we can not view a deformation of a pair as first a deformation of X and then a deformation of the Di. One must view (X, ∆) as an inseparable unit. Example 42. Let S ⊂ P5 be the cone over the degree 4 rational normal curve. Fix r ≥ 1 and let DS be the sum of 2r lines. Then (S, 1 = 4. There are two different deformations of the pair (S, DS). r DS) is lc and (cid:0)KS + 1 r DS(cid:1)2 20 J ´ANOS KOLL ´AR (42.1) First, set P := P2 and let DP be the sum of r general lines. Then (P, 1 is lc (even canonical if r ≥ 2) and(cid:0)KP + 1 r DP ) = 4. The usual smoothing of S ⊂ P2 to the Veronese surface gives a family f : (X, DX ) → P1 with general fiber (P, DP ) and special fiber (S, DS). We can concretely realize this as deforming (P, DP ) ⊂ P5 to the cone over a general hyperplane section. Note that for any general DS there is a choice of lines DP such that the above limit is exactly DS. r DP(cid:1)2 The total space (X, DX ) is the cone over (P, DP ) (blown up along curve) and X is Q-factorial. The structure sheaf of an effective divisor on X is CM. In particular, DS is a flat limit of DP . Since the DP is a plane curve of degree r, we conclude that χ(ODS ) = χ(ODP ) = − r(r − 3) 2 . Let DQ be the sum of r lines from the A-family. Then (Q, 1 (cid:0)KQ + 1 (42.2) Second, set Q := P1 × P1 and let A, B denote the classes of the 2 rulings. r DQ) is canonical and = 4. The usual smoothing of S ⊂ P2 to P1 × P1 embedded by H := A + 2B gives a family g : (Y, DY ) → P1 with general fiber (Q, DQ) and special fiber (S, DS). We can concretely realize this as deforming (Q, DQ) ⊂ P5 to the cone over a general hyperplane section. r DQ(cid:1)2 is not Q-factorial. However, KQ + 1 (Y, 1 The total space (Y, DY ) is the cone over (Q, DQ) (blown up along curve) and Y r DY is Q-Cartier and r DQ ∼Q −H, thus KY + 1 r DY ) is lc. In this case, however, DQ is not a flat limit of DP for r > 1. Thus follows, for instance, from comparing their Euler characteristic: χ(ODS ) = − r(r − 3) 2 and χ(ODQ ) = r. (42.3) Because of their role in the canonical algebra, we are also interested in the sheaves O(mK + ⌊ m r D⌋). Let HP be the hyperplane class of P ⊂ P5 (that is, 2 times a line L ⊂ P ) and write m = br + a where 0 ≤ a < r. One computes that 2 2 χ(cid:0)P, OP (mKP + ⌊ m χ(cid:0)S, OS(mKS + ⌊ m χ(cid:0)Q, OQ(mKQ + ⌊ m r DP ⌋ + nHP )(cid:1) = (cid:0)2n−2m+2 r DS⌋ + nHS)(cid:1) = (cid:0)2n−2m+2 r DQ⌋ + nHQ)(cid:1) = (cid:0)2n−2m+2 2 (cid:1) − a(2n − 2m + 1) +(cid:0)a 2(cid:1), (cid:1) − a(2n − 2m + 1) +(cid:0)a 2(cid:1), (cid:1) − a(2n − 2m + 1). From this we conclude that the restriction of OY (mKY +⌊mDY ⌋) to the central fiber S agrees with OS(mKS + ⌊mDS⌋) only if a ∈ {0, 1}, that is when m ≡ 0, 1 mod r. The if part was clear from the beginning. Indeed, if a = 0 then OY (mKY +⌊mDY ⌋) is locally free and if a = 1 then OY (mKY + ⌊mDY ⌋) is OY (KY ) tensored with a locally free sheaf. Both of these commute with restrictions. In the other cases we only get an injection OY (mKY + ⌊mDY ⌋)S ֒→ OS(mKS + ⌊mDS⌋) whose quotient is a torsion sheaf of length (cid:0)a 2(cid:1) supported at the vertex. There are several ways to overcome these problems; all them will be discussed in [Kol10b]. (1) Embedded points do not appear if all the coeffcients ai are > 1 2 . (2) By wiggling the ai suitably, one again avoids embedded points. MODULI OF VARIETIES OF GENERAL TYPE 21 (X, ∆) with the corresponding map ω⊗m (3) Fix m such that OX (mKX + m∆) is locally free. One can identify a pair X → OX (mKX + m∆). It turns out X → L(cid:1) for some line to be easier to deal with the moduli of triples (cid:0)X, ω⊗m bundle L. (4) The branch varieties of [AK06] give another approach. Acknowledgments. Parts of this paper were written in connection with a lecture series on moduli at IHP, Paris. I thank my audience and especially C. Voisin for the invitation, useful comments and corrections. Partial financial support was provided by the NSF under grant number DMS-0758275. References [AH09] [AK06] [Ale94] [Ale96] [Ale02] [AM04] [Art74] Dan Abramovich and Brendan Hassett, http://www.citebase.org/abstract?id=oai:arXiv.org:0904.2797, 2009. Valery Alexeev and Allen Knutson, Complete moduli spaces of branchvarieties, 2006. Valery Alexeev, Boundedness and K 2 for log surfaces, Internat. J. Math. 5 (1994), no. 6, 779–810. MR MR1298994 (95k:14048) varieties with a twist, Stable , Moduli spaces M g,n(W ) for surfaces, Higher-dimensional complex varieties (Trento, 1994), de Gruyter, Berlin, 1996, pp. 1–22. MR MR1463171 (99b:14010) , Complete moduli in the presence of semiabelian group action, Ann. of Math. (2) 155 (2002), no. 3, 611–708. MR MR1923963 (2003g:14059) Valery Alexeev and Shigefumi Mori, Bounding singular surfaces of general type, Algebra, arithmetic and geometry with applications (West Lafayette, IN, 2000), Springer, Berlin, 2004, pp. 143–174. MR MR2037085 (2005f:14077) M. Artin, Algebraic construction of Brieskorn's resolutions, J. Algebra 29 (1974), 330–348. MR MR0354665 (50 #7143) [BCHM06] Caucher Birkar, Paolo Cascini, Christopher D. Hacon, and James McK- type, varieties of for [Bri68] [DB81] [DJ74] [Elk81] [Hac04] [Har77] [Iit71] [K+92] [Kar00] [Kaw07] [KK09] log general ernan, Existence of minimal models http://www.citebase.org/abstract?id=oai:arXiv.org:math/0610203, 2006. Egbert Brieskorn, Die Auflosung der rationalen Singularitaten holomorpher Abbil- dungen, Math. Ann. 178 (1968), 255–270. MR MR0233819 (38 #2140) Philippe Du Bois, Complexe de de Rham filtr´e d'une vari´et´e singuli`ere, Bull. Soc. Math. France 109 (1981), no. 1, 41–81. MR MR613848 (82j:14006) Philippe Dubois and Pierre Jarraud, Une propri´et´e de commutation au changement de base des images directes sup´erieures du faisceau structural, C. R. Acad. Sci. Paris S´er. A 279 (1974), 745–747. MR MR0376678 (51 #12853) R. Elkik, Rationalit´e des singularit´es canoniques, Inv. Math. 64 (1981), 1–6. Paul Hacking, Compact moduli of plane curves, Duke Math. J. 124 (2004), no. 2, 213–257. MR MR2078368 (2005f:14056) Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977, Graduate Texts in Mathematics, No. 52. MR MR0463157 (57 #3116) Shigeru Iitaka, On D-dimensions of algebraic varieties, J. Math. Soc. Japan 23 (1971), 356–373. MR MR0285531 (44 #2749) J. Koll´ar et al., Flips and abundance for algebraic threefolds, Soc. Math. France, Ast´erisque vol. 211, 1992. Kalle Karu, Minimal models and boundedness of stable varieties, J. Algebraic Geom. 9 (2000), no. 1, 93–109. MR MR1713521 (2001g:14059) Masayuki Kawakita, Inversion of adjunction on log canonicity, Invent. Math. 167 (2007), no. 1, 129–133. MR MR2264806 (2008a:14025) J´anos Koll´ar and S´andor J. Kov´acs, Log canonical singularities are Du Bois, http://www.citebase.org/abstract?id=oai:arXiv.org:0902.0648, 2009. [KKMSD73] G. Kempf, F. F. Knudsen, D. Mumford, and B. Saint-Donat, Toroidal embed- dings. I, Lecture Notes in Mathematics, Vol. 339, Springer-Verlag, Berlin, 1973. MR MR0335518 (49 #299) Se´an Keel and Shigefumi Mori, Quotients by groupoids, Ann. of Math. (2) 145 (1997), no. 1, 193–213. MR MR1432041 (97m:14014) [KM97] 22 [KM98] [Kol90] [Kol94] [Kol97] [Kol07] [Kol08] J ´ANOS KOLL ´AR J´anos Koll´ar and Shigefumi Mori, Birational geometry of algebraic varieties, Cam- bridge Tracts in Mathematics, vol. 134, Cambridge University Press, Cambridge, 1998, With the collaboration of C. H. Clemens and A. Corti, Translated from the 1998 Japanese original. MR MR1658959 (2000b:14018) J´anos Koll´ar, Projectivity of complete moduli, J. Differential Geom. 32 (1990), no. 1, 235–268. MR MR1064874 (92e:14008) , Log surfaces of general type; some conjectures, Classification of algebraic varieties (L'Aquila, 1992), Contemp. Math., vol. 162, Amer. Math. Soc., Providence, RI, 1994, pp. 261–275. MR MR1272703 (95c:14042) , Quotient spaces modulo algebraic groups, Ann. of Math. (2) 145 (1997), no. 1, 33–79. MR MR1432036 (97m:14013) , Two crossing http://www.citebase.org/abstract?id=oai:arXiv.org:0705.0926, 2007. surfaces with normal examples of singularities, , Hulls and husks, http://www.citebase.org/abstract?id=oai:arXiv.org:0805.0576, 2008. [Kol10a] , A local version of the Kawamata-Viehweg vanishing theorem, [Kol10b] [KSB88] [Laz04] [PSS71] [Rei80] [Siu98] [Siu08] [Ste83] http://www.citebase.org/abstract?id=oai:arXiv.org:1005.4843, 2010. , Moduli of varieties of general type, (book in prepapration), 2010. J. Koll´ar and N. I. Shepherd-Barron, Threefolds and deformations of surface singu- larities, Invent. Math. 91 (1988), no. 2, 299–338. MR MR922803 (88m:14022) Robert Lazarsfeld, Positivity in algebraic geometry. I-II, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Math- ematics], vol. 48–49, Springer-Verlag, Berlin, 2004. MR MR2095471 (2005k:14001a) I. I. Pjateckiı-Sapiro and I. R. Safarevic, Torelli's theorem for algebraic surfaces of type K3, Izv. Akad. Nauk SSSR Ser. Mat. 35 (1971), 530–572. MR MR0284440 (44 #1666) Miles Reid, Canonical 3-folds, Journ´ees de G´eometrie Alg´ebrique d'Angers, Juillet 1979/Algebraic Geometry, Angers, 1979, Sijthoff & Noordhoff, Alphen aan den Rijn, 1980, pp. 273–310. MR MR605348 (82i:14025) Yum-Tong Siu, Invariance of plurigenera, Invent. Math. 134 (1998), no. 3, 661–673. MR MR1660941 (99i:32035) , Finite generation of canonical ring by analytic method, Sci. China Ser. A 51 (2008), no. 4, 481–502. MR MR2395400 J. H. M. Steenbrink, Mixed Hodge structures associated with isolated singularities, Singularities, Part 2 (Arcata, Calif., 1981), Proc. Sympos. Pure Math., vol. 40, Amer. Math. Soc., Providence, RI, 1983, pp. 513–536. MR MR713277 (85d:32044) Princeton University, Princeton NJ 08544-1000 [email protected]
1705.00655	1	1705	2017-05-01T18:31:02	Generators in formal deformations of categories	[ "math.AG", "math.CT" ]	In this paper we use the theory of formal moduli problems developed by Lurie in order to study the space of formal deformations of a $k$-linear $\infty$-category for a field $k$. Our main result states that if $\mathcal{C}$ is a $k$-linear $\infty$-category which has a compact generator whose groups of self extensions vanish for sufficiently high positive degrees, then every formal deformation of $\mathcal{C}$ has zero curvature and moreover admits a compact generator.	math.AG	math	Generators in formal deformations of categories Anthony Blanc, Ludmil Katzarkov, Pranav Pandit May 3, 2017 Abstract In this paper we use the theory of formal moduli problems developed by Lurie in order to study the space of formal deformations of a k-linear ∞-category for a field k. Our main result states that if C is a k-linear ∞-category which has a compact generator whose groups of self extensions vanish for sufficiently high positive degrees, then every formal deformation of C has zero curvature and moreover admits a compact generator. Contents 1 Introduction 1.1 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 General conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Generalities on formal moduli problems 2.1 Reminders on formal moduli problems . . . . . . . . . . . . . . . . . 2.2 Approximation of formal moduli and loop spaces . . . . . . . . . . . 3 Reminders on deformations of categories 3.1 Linear ∞-categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Deformations of objects 3.3 Deformations of categories . . . . . . . . . . . . . . . . . . . . . . . . 4 Generators and curvature in deformations of categories . . . . . . . . . . . . . . . . . . . . . . . 4.1 Simultaneous deformations 4.2 Deformations of associative algebras . . . . . . . . . . . . . . . . . . 4.3 Formal deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . A Deformations as !-group actions 2 3 5 6 6 11 15 16 18 19 21 21 27 30 35 1 1 Introduction Let k be a field of characteristic zero and let B be an associative k-algebra. It is known since the work of Gerstenhaber that the formal deformations of B up to isomorphism are in bijection with the set of solutions of the Maurer–Cartan equation in Hochschild cohomology up to gauge equivalence. In their seminal work [KS05] Kontsevich and Soibelman have pionereed the approach of studying deformations of dg-algebras and dg-categories using differential graded Lie algebras. They show that if B is a connective differential graded algebra over k, its deformations as a dg-algebra up to quasi-isomorphism are controlled by the truncated Hochschild cohomology complex viewed as a differential graded Lie algebra. Given a k-linear dg-category C, it has been a long standing problem to give the precise relation between the dg-category deformations of C up to Morita equivalence and its Hochschild cohomology complex. In the case of first order infinitesimal deformations, each Hochschild 2-cocycle corresponds to a curved A∞-deformation of C (see [Low08]). In [KL09], B. Keller and W. Lowen gave examples of curved deformations which are not Morita equivalent to uncurved deformations. In [LB15], W. Lowen and M. Van den Bergh study deformations of a dg-algebra B up to torsion Morita equivalence and show that when B is cohomologically bounded above, the set of formal deformations of B up to torsion Morita equivalence is in bijection with the set of Maurer–Cartan solutions up to gauge equivalence. In order to study the whole space of deformations of a category up to Morita equivalence and not just merely its set of connected components, and also to exhibit the precise relation between this space and Hochschild cohomology, it is useful to place ourselves in the setting of derived algebraic geometry and more pre- cisely in the language of formal moduli problems developed by Jacob Lurie in [Lurb]. Formal moduli problems. Deformation theory was first formalized through the notion of functors on artinian rings satisfying the Schlessinger conditions. Formal moduli problems form a derived and higher analog of these functors. We refer the reader to [Lurb] for an introduction to formal moduli problems and their classification. Roughly they are functors from certain artinian E∞-algebrasa to spaces satisfying the derived analogs of the Schlessinger conditions. The main result states that formal moduli problems are classified by differential graded Lie algebras (over a field of characteristic zero). In loc. cit. Lurie studies noncommutative analogs of formal moduli problems known as formal En-moduli problems for n ≥ 1, which are functors from artinian En-algebras to spaces satisfying the derived Schlessinger conditions, where En is the ∞-operad of little n-cubes. In this context, the formal En-moduli problems are classified by augmented En-algebras. It is often useful to find the minimum n for which a moduli problem is defined, thereby obtaining the true algebraic structure on the tangent complex (see [Fra13] for a discussion of these ideas). aIn characteristic zero, we can also think of these as functors on artinian commutative differential graded algebras. 2 Formal moduli of a category after Lurie. Let k be a field of arbitrary In loc. cit. Lurie studies deformations of a presentable stable k- characteristic. linear ∞-categoryb C up to equivalencec by looking at the functor CatDefC : Alg(2),art k → S from the ∞-category of artinian E2-algebras to the ∞-category of spaces, classifying deformations of C. That is, for an artinian E2-algebra A, the space CatDefC(A) classifies pairs (CA, u) with C a right A-linear ∞-category and u : CA ⊗A k (cid:39) C an equivalence. In general, the functor CatDefC does not satisfy the derived Schlessinger conditions, the moral reason for this being the existence of curved deformations (see Example 4.11). When C is compactly generatedd, the subfunctor CatDef cC ⊆ CatDefC spanned by compactly generated deformations is better behaved and closer to satisfying the Schlessinger conditions. We denote by HH∗(C) the E2-algebra given by the center of C whose homotopy groups are the Hochschild cohomology groups of C. Lurie constructs a natural transformation θC : CatDef cC → ΨHH∗(C) = CatDef∧ C from CatDef cC to the formal moduli problem controlled by the augmented E2-algebra k ⊕ HH∗(C). He shows that under certain reasonable boundedness assumptions on C the map θC is close to being a homotopy equivalence: it induces an isomorphism on πi for i ≥ 1 (at any basepoint) and an injection on π0. The formal moduli CatDef∧ C can be regarded as a substitute for the space of curved A∞-deformations and the map θC as the embedding from uncurved deformations to curved deformations. By the universal property of the center, for each artinian E2-algebra A, the space CatDef∧ C (A) is given by the space of linear actions of the E2-algebra D(2)(A) on the ∞-category C, where D(2)(A) is the E2-Koszul dual of A. 1.1 Main results Our main results provide a new understanding of the behavior of the map θC in the case of formal deformations, that is, over the adic ring k[[t]] of formal power series. Namely, we give conditions on C under which the map θC induces a homo- topy equivalence on formal deformations, therefore providing the missing step to relate deformations of categories to Hochschild cohomology. It can also be viewed as a solution to the curvature problem of [LB15], where similar computations are performed. bAs a matter of facility for citations, from now on we study deformations of linear ∞-categories instead of differential graded categories. See the work of L. Cohn [Coh13] for a comparison between the two theories, recalled in §3.1. cRecall that the notion of equivalence between presentable ∞-categories corresponds to the notion of Morita equivalence between the corresponding small subcategories of compact objects. See §3.1. dAn ∞-category is compactly generated if it is generated under filtered colimits by its subcat- egory of compact objects (as for example the derived ∞-category of quasi-coherent sheaves on a finite type scheme). 3 Let k be a field and let C be compactly generated k-linear ∞-category. The space of compactly generated deformations of C is the space CatDef cC(k[[t]]) (cid:39) lim i CatDef cC(k[t]/ti). That is, a point in this space is essentially given by a formal family {Ci}i≥1 where each Ci is a compactly generated deformation of C over k[t]/ti and Ci+1⊗k[t]/ti+1 k[t]/ti (cid:39) Ci is an equivalence, plus higher coherences. In this case, the space CatDef∧ C (k[[t]]) is given by the space of linear actions of the E2-algebra D(2)(k[[t]]) (cid:39) k[u] on the ∞- category C, where k[u] is the free graded commutative algebra on a generator u in cohomological degree 2 viewed as an E2-algebra. Theorem 1.1. (See Theorem 4.27). - Let k be a field and let C be a compactly generated k-linear ∞-category. Suppose that C admits a single compact generator E such that ExtmC (E, E) = 0 for m (cid:29) 0. Then the natural transformation θC induces a homotopy equivalence CatDef cC(k[[t]]) (cid:39) CatDef∧ C (k[[t]]) = {k[u] − linear structures on C}. The idea of the proof is to exhibit another compact generator of C on which k[u] acts trivially, which implies that it is unobstructed. For this it suffices to kill the multiplication by u on the generator E by forming its cocone. As a by product of the proof of Theorem 4.27 we obtain the following existence result for compact generators. Theorem 1.2. (See Theorem 4.29). - In the situation of Theorem 1.1, let Ct = {Ci}i be any compactly generated formal deformation of C. Then there exists a formal family of compact generators {Ei}i of the family {Ci}i which satisfies ExtmCi (Ei, Ei) = 0 for m (cid:29) 0. We emphasize that our method in the proof of Theorems 1.1 and 1.2 relies on the fact that the E2-Koszul dual of the algebra k[[t]] of formal power series (viewed as an E2-algebra) is given by the free graded associative (also commutative) algebra k[u] with u in degree 2 (viewed as an E2-algebra). The argument cannot be adapted verbatim to first order infinitesimal deformations where the E2-algebra D(2)(k[t]/t2) is given by the more complicated free E2-algebra FreeE2(k[−2]) on a generator in degree 2. F. Petit has proven in [Pet12] the existence of compact generators in derived categories of DQ-modules. We expect to give an alternative proof of this fact using Theorem 1.2. On a more fundamental level, we prove a new result concerning the loop functor on the ∞-category of formal moduli problems over a deformation context (Propo- sition 2.15). Moreover it implies a new description of the formal moduli associated to any m-proximate formal moduli (Corollary 2.19). As a consequence, we obtain the following description of the formal moduli CatDef∧ C in terms of !-group actions (which also appeared in [Pre12, Prop 5.3.3.4]). 4 Proposition 1.3. (See Corollary A.11). - Let k be a field and let C be a k-linear ∞-category. Then the formal E2-moduli problem CatDef∧ C is given by the formula CatDef∧ C (A) (cid:39) {Left actions of QCoh!(ΩSpf(A))⊗ on C} where QCoh!(ΩSpf(A)) is the ∞-category of Ind-coherent sheaves on ΩSpf(A) en- dowed with the convolution monoidal structure. Acknowledgments A special acknowledgment and many thanks are due to Bertrand Toen for generously sharing his ideas on the subject matter of this paper and for many enlightening discussions. We would like to thank Mohammed Abouzaid, Alexander Efimov, David Gepner, Benjamin Hennion, Julian Holstein, Maxim Kontsevich, Wendy Lowen, Tony Pantev, Fran¸cois Petit, Marco Robalo, Pavel Safronov, Carlos Simpson, Michel Van den Bergh and Gabriele Vezzosi for useful discussions related to the subject matter of this paper. The second and third named authors were partially supported by ERC Gemis grant. The second named author was supported by Simons research grant, NSF DMS 150908, ERC Gemis, DMS-1265230, DMS-1201475 OISE-1242272 PASI. Si- mons collaborative Grant - HMS, and partially supported by the Laboratory of Mirror Symmetry NRU HSE, RF government grant, ag. 14.641.31.000. 1.2 General conventions • Let U be a Grothendieck universe, with U satisfying the axiom of infinity. The U-small mathematical objects will be called only small. We assume the axiom of Universes. Some arguments in this article will require to enlarge the universe U, which is always possible by assuming the axiom of Universes. If V is such an enlargement in which U is small, the V-small mathematical objects will be called not necessarily small. • We work within the theory of ∞-categories in the sense of Lurie [Lur09], a.k.a quasicategories. We follow the terminology and the conventions of loc.cit. regarding the theory of ∞-categories. We denote by Cat∞ the ∞-category of small ∞-categories. • For two ∞-categories C and C(cid:48), we denote by Fun(C,C(cid:48)) the ∞-category of functors from C to C(cid:48). • If C is an ∞-category admitting a final object ∗. The ∞-category C∗ of pointed objects in C is by definition the full subcategory of Fun(∆1,C) consisting of morphisms C → C(cid:48) such that C is a final object of C. egory of small Kan complexes. We denote by (cid:98)S the ∞-category of not nec- • We denote by S the ∞-category of small spaces in the sense of [Lur09, Def 1.2.16.1]. It is therefore defined as the simplicial nerve of the simplicial cat- essarily small spaces. The word space means an object of the ∞-category 5 S. The expression not necessarily small space means an object of (cid:98)S for some appropriate universe V. We denote by ∗ the final object ∆0 of S. • We follow the terminology of [Lurd] concerning n-connective spaces. Let n ≥ 0 be an integer. A space X is n-connected if πi(X, x) = 0 for every i < n and every vertex x ∈ X. Every space is declared to be (−1)-connective. • We follow the terminology of [Lurd] concerning n-truncated spaces. Let n ≥ −1 be an integer. A space X is n-truncated if πi(X, x) = 0 for every i > n and every vertex x ∈ X. By convention a space is (−1)-truncated if it is either empty or contractible, and is (−2)-truncated if it is contractible. A map of pointed spaces X → Y is called n-truncated if its fiber is n-truncated. • We denote by Sp the stable ∞-category of spectra (see [Lurd, Def 1.4.3.1]). • Let n ∈ Z be an integer. We consider the usual t-structure on the stable ∞- category Sp. A spectrum X is n-connective (resp. n-truncated) if πiX = 0 for every i < n (resp. i > n). 2 Generalities on formal moduli problems We start by recalling from [Lurb, §4] the notion of formal moduli problems defined over En-algebras and their classification. Then we recall the setting of axiomatic deformation theory from [Lurb, §1], in which our results from the next subsection §2.2 concerning the approximation of formal moduli via loop spaces are proven. 2.1 Reminders on formal moduli problems Reminders on formal En-moduli problems Let k be a field of arbitrary characteristic. Let Modk denote the ∞-category of k- module spectra. Let n ≥ 0 be an integer and let Alg(n) := AlgEn(Modk) denote the ∞-category of En-algebras over k in the sense of [Lurd, Def 7.1.3.5]. Let Alg(n),aug = (Alg(n) A over k artinian if the following conditions are satisfied k )/k denote the ∞-category of augmented En-algebras. We call an En-algebra k k • πiA = 0 for i < 0 and for i (cid:29) 0. • For every i, the k-vector space πiA is finite dimensional. • Let r denote the radical of the algebra π0A, then the natural map (π0A)/r → k is an isomorphism. the full subcategory spanned by artinian En- We denote by Alg(n),art algebras. For every integer m ≥ 0, we denote by k ⊕ k[m] the shifted square zero extension viewed as En-algebras over k, which are examples of artinian En-algebras. k k ⊆ Alg(n) 6 A functor X : Alg(n),art k → S is called a formal En-moduli problem if the space X(k) is contractible and if for every pullback diagram A A0 A1 / A01 in Alg(n),art k square such that the maps π0A0 → π0A01 ← π0A1 are surjective, the induced X(A) X(A0) X(A1) / X(A01) ,S) denote the is a pullback diagram in S. Let Fun∗(Alg(n),art → S such that X(k) is a full subcategory spanned by those functors X : Alg(n),art ,S) the full subcate- contractible space. We denote by Moduli gory spanned by formal En-moduli problems. The class of pullback diagrams of the form k ⊆ Fun∗(Alg(n),art En ,S) ⊆ Fun(Alg(n),art k k k k k ⊕ k[m] k k / k ⊕ k[m + 1] for m ≥ 0 are used to defined the tangent complex of a formal En-moduli problem. Indeed if X is a formal En-moduli problem over k, the induced map X(k ⊕ k[m]) → ΩX(k ⊕ k[m + 1]) is a homotopy equivalence, and defines a spectrum TX called the tangent complex of X (see [Lurb, Def 1.2.5]). By [Lurb, Prop 1.2.10], a map of formal En-moduli problems X → Y is an equivalence if and only if the induced map TX → TY is an equivalence of spectra. By [Lurb, Thm 4.0.8] there exists a functor Ψ : Alg(n),aug k −→ Moduli En k which is an equivalence of ∞-categories. For an augmented En-algebra B and for an artinian En-algebra A over k, there is an equivalence ΨB(A) (cid:39) )op → Alg(n),aug is the Koszul du- Map (D(n)(A), B) where D(n) : (Alg(n),aug k k Alg(n),aug k ality functor defined in [Lurb, Prop 4.4.1]. Moreover there exists a commutative diagram of ∞-categories Alg(n),aug k Ψ / En Moduli k m Modk T [−n] / Sp where m is given by taking the augmentation ideal. 7 / / / / / / / / / / / Reminders on axiomatic deformation theory We recall some definitions and constructions from [Lurb, §1]. For A a presentable ∞-category, we denote by Sp(A) the ∞-category of spectra objects of A in the sense of [Lurd, Def 1.4.2.8]. By [Lurd, Rem 1.4.2.25] there exists an equivalence of ∞-categories Sp(A) (cid:39) lim(. . . → A∗ Ω→ A∗ Ω→ A∗). We denote by Ω∞ : Sp(A) → A the evaluation on the 0-sphere S0 and for every integer n ∈ Z, we denote by Ω∞−n : Sp(A) → A the composition of Ω∞ with the nth shift functor on Sp(A). A deformation context in the sense of [Lurb, Def 1.1.3] is a pair (A, (Eα)α∈T ) where A is a presentable ∞-category and (Eα)α∈T is a sequence of objects of the ∞- category Sp(A) of spectra objects in A. Given a deformation context (A, (Eα)α∈T ), a map f : A → A(cid:48) in A is called elementary if there exists an index α ∈ T , an integer n ≥ 0 and a pullback diagram A f A(cid:48) ∗A / Eα[n] in A, where ∗A is a final object of A. A map f : A → A(cid:48) in A is called small if it can be written as a finite composition of elementary maps A0 = A → A1 → . . . → Am = A(cid:48) in A. An object A ∈ A is called artinian if the map A → ∗A is small. We denote by Aart the full subcategory of A spanned by artinian objects. Given a deformation context (A, (Eα)α∈T ), recall that a formal moduli problem in the sense of [Lurb, Def 1.1.14] is a functor X : Aart → S which satisfies the two conditions 1. The space X(∗A) is contractible. 2. For every pullback diagram A A0 A1 / A01 in Aart such that the map A0 → A01 is small, the induced square is a pullback diagram in S. X(A) X(A0) X(A1) / X(A01) 8 / / / / / / / / / By [Lurb, Prop 1.1.15] we are allowed to test condition (2) on the smaller class of pullback diagrams of the form A A(cid:48) ∗A / Eα[n] k in Aart where α ∈ T and n ≥ 0. We denote by Fun∗(Aart,S) ⊆ Fun(Aart,S) the subcategory spanned functors satisfying condition (1) above, or in other words sending ∗A to a contractible space. We denote by ModuliA ⊆ Fun∗(Aart,S) the subcategory spanned by the formal moduli problems. Example 2.1. Let k be a field. Let n ≥ 1 and A = Alg(n),aug be the presentable ∞-category of augmented En-algebras over k. Consider the deformation context (A,{k}) where k is seen as a module over itself. The deloopings of k are given by the shifted square zero extensions k ⊕ k[m] for m ≥ 0. By [Lurb, Prop 4.5.1], an augmented En-algebra A is artinian with respect to the deformation context (A,{k}) if and only if it is artinian in the sense of §2.1. By [Lurb, Prop 4.5.1], a morphism A → A(cid:48) in Alg(n),aug is small with respect to the deformation context (A,{k}) if and only if the induced map π0A → π0A(cid:48) is surjective. Then [Lurb, Cor 4.5.4] implies that X is a formal moduli problem with respect to the deformation context (A,{k}) if and only if it is a formal En-moduli problem in the sense of §2.1. Notation 2.2. Let (A, (Eα)α∈T ) be a deformation context. Consider the embedding iA : ModuliA (cid:44)→ Fun∗(Aart,S) into the ∞-category of functors X : Aart → S such that X(∗A) is a contractible space. The ∞-categories ModuliA and Fun∗(Aart,S) are presentable and by definition of formal moduli problems the functor iA preserves all small limits. By the adjoint functor theorem [Lur09, Cor 5.5.2.9], this implies that iA admits a left adjoint denoted by k ModuliA LA iA / Fun∗(Aart,S) . For every functor X : Aart → S sending ∗A to a contractible space we have the unit natural transformation X → iA(LA(X)) of this adjunction. The formal moduli problem LA(X) is therefore initial among formal moduli problems Y having a natural transformation X → iAY . Let (A, (Eα)α∈T ) be a deformation context. Recall from [Lurb, Def 1.2.5] that every formal moduli problem X : Aart → S has a tangent spectrum X(Eα) ∈ Sp of X at α. Moreover a map X → Y of formal moduli problems is an equivalence if and only if for every α the map X(Eα) → Y (Eα) is an equivalence of spectra. In [Lurb, Def 1.3.1 and 1.3.9], Lurie defines the notion of weak deformation theory and of deformation theory. In short a weak deformation theory for (A, (Eα)α∈T ) 9 / / / / t t consists of a functor D : Aop → B such that B is a presentable ∞-category, D has a left adjoint D(cid:48) which behaves like an inverse equivalence to D on a certain subcategory of B. In this situation, [Lurb, Cor 1.3.6] asserts the existence of an ∞-functor Ψ : B → ModuliA which is given on objects by ΨB(A) = MapB(D(A), B). By applying the tangent spectrum we obtain for every α a functor eα : B → Sp given on objects by eα(B) = ΨB(Eα). A deformation theory for (A, (Eα)α∈T ) is a weak deformation theory such that each eα preserves small sifted colimits and such that a morphism f in B is an equivalence if and only if each eα(f ) is an equivalence of spectra. The main theorem [Lurb, Thm 1.3.12] then asserts that under the existence of a deformation theory for (A, (Eα)α∈T ), the functor Ψ is an equivalence of ∞-categories. Formal spectrum and Koszul duality We now set up some notations and intermediate results concerning the formal spec- trum, pro-objects and formal Koszul duality. Notation 2.3. Let (A, (Eα)α∈T ) be a deformation context and let A ∈ A be any object. Consider the functor Spf(A) : Aart → S which is co-represented by A given by the formula Spf(A)(R) = MapA(A, R). Then Spf(A) is a formal moduli problem which is called the formal spectrum of A. Moreover the construction A (cid:55)→ Spf(A) determines a functor Spf : Aop −→ ModuliA which is fully faithful when restricted to the subcategory (Aart)op spanned by ar- tinian objects. Example 2.4. Let n ≥ 1 and consider the deformation context (Alg(n),aug , k) formed by augmented En-algebras over k. Let : A → k be an augmented En- algebra over k. If we think about the En-algebra A as determining a noncommu- tative scheme Spec(A), then the formal spectrum Spf(A) can be thought of as the functor parametrizing deformations of the point of Spec(A) determined by over artinian En-algebras. k The following Lemma will be useful in the sequel and follows from the fact that the Koszul duality functor D(n) is a deformation theory. Lemma 2.5. Let n ≥ 1 be an integer and denote by t(n) : Moduli k → Alg(n),aug En the equivalence given by [Lurb, Thm 4.0.8]. Let B be an augmented En-algebra over k. Spf(B) (cid:39) D(n)(B) Then there exists a natural equivalence of augmented En-algebras t(n) or equivalently an equivalence of formal moduli problems Spf(B) (cid:39) ΨD(n)(B). Proof. Consider the natural transformation α : Spf(B) → ΨD(n)(B) induced by the natural map k Map Alg(n),aug k (B, A) αA−→ Map Alg(n),aug k (D(n)(A), D(n)(B)) 10 k Alg(n),aug for each artinian En-algebra A over k. By adjunction the target space of αA (B, D(n)D(n)(A)) and the map is naturally equivalent to the space Map is induced by the unit map A → D(n)D(n)(A). Let m ≥ 0 be an integer and A = k ⊕ k[m] be the trivial square zero extension. We claim that the biduality map A → D(n)D(n)(A) is an equivalence. Indeed by [Lurb, Prop 4.5.6] there ex- (k[−m − n]) (cid:39) k ⊕ k[m] of augmented En-algebras ists an equivalence D(n)(Freeaug En (k[−m− n]) is n-coconnective and and by [Lurb, Prop 4.1.13] the En-algebra Freeaug En locally finite whenever m ≥ 0, which by [Lurb, Thm 4.4.5] implies that the bidu- (k[−m − n]), and thus also ality map is an equivalence for the En-algebra Freeaug En for A = k ⊕ k[m]. We deduce that the map αA is a homotopy equivalence and therefore that α induces an equivalence on the tangent spectrum, and is therefore an equivalence, which proves the lemma. 2.2 Approximation of formal moduli and loop spaces As explained above, the deformation functor of a linear ∞-category over a field does not in general satisfy the derived Schlessinger conditions, but turns out to be rather close to satisfy them. Many examples of deformation functors of some algebro- In [Lurb, §5.1], Lurie developed a formalism geometric objects behave similarly. to study these functors under the name of m-proximate formal moduli problems for m ≥ 0 an integer. We will prove that the functor L = LA : Fun∗(Aart,S) → ModuliA from Notation 2.2 admits a special description when restricted to the sub- category spanned by m-proximate formal moduli problems in terms of Em-maps between certain loop Em-spaces (see Corollary 2.19). For this we will prove a result concerning the loop space functor on the ∞-category of formal moduli problem (see Proposition 2.15). Let (A, (Eα)α∈T ) be a deformation context and m ≥ 0 an integer. Recall from [Lurb, Def 5.1.5] that a functor X : Aart → S is called an m-proximate formal moduli problem if the space X(∗A) is contractible and if for every pullback square A A0 A1 / A01 in Aart such that the map A0 → A01 is small, the induced map X(A) → X(A0) ×X(A01) X(A1) is (m− 2)-truncated (that is, all homotopy fibers of this map are (m− 2)-truncated). We observe that 0-proximate formal moduli problems are exactly the formal moduli problems. Notation 2.6. We denote by Prox(m)A ⊆ Fun∗(Aart,S) the full subcategory spanned by m-proximate formal moduli problems. There is a tower of ∞-categories ModuliA = Prox(0)A ⊆ Prox(1)A ⊆ Prox(2)A ⊆ . . . ⊆ Fun∗(Aart,S) 11 / / / , k) of Example 2.1. We will denote by Prox(m)En ⊆ Fun∗(Alg(n),art For example, we will see below that the deformation functor of an object in a linear ∞-category is in general 1-proximate and that the deformation functor of a fixed linear ∞-category itself is in general 2-proximate. Notation 2.7. Let n ≥ 1 and m ≥ 0, and consider the deformation context ,S) (Alg(n),aug the full subcategory of m-proximate formal En-moduli problem. Notation 2.8. Let (A, (Eα)α∈T ) be a deformation context and suppose that A is a pointed ∞-category in the sense that any final object is also initial (e.g. the ∞-category Alg(n),aug of augmented En-algebras over k). Then the ∞-category Fun∗(Aart,S) is also pointed. We denote by k k k Ω : Fun∗(Aart,S) −→ Fun∗(Aart,S) the loop functor in the sense of [Lurd, Rem 1.1.2.9]. The embedding ModuliA (cid:44)→ Fun∗(Aart,S) commutes with small limits and the loop functor restricts to a functor Ω : ModuliA → ModuliA. For m ≥ 0 an integer, we denote by Ωm = Ω ◦ . . . ◦ Ω (m times) the iterated loop functor. Remark 2.9. In the situation of Notation 2.8, let X : Aart → S be any functor. Then the functor ΩX is given on objects by (ΩX)(A) = ΩX(A) where the loop space is taken at the base point ∗ → X(A) corresponding to the map ∗ (cid:39) X(∗A) → X(A) induced by the essentially unique map ∗A → A. If we imagine that the functor X parametrizes deformations of an object of an ∞-category corresponding to the connected component π0X(∗A) = ∗, then the functor ΩX parametrizes deformations of the identity of this object, or in other words infinitesimal autoequivalences of this object. Remark 2.10. Let f : X → Y be a Kan fibration between Kan complexes. Let m ≥ 0 be an integer and suppose that the map f is (m − 2)-truncated, that is, all its homotopy fibers are (m − 2)-truncated. Then we deduce from the long exact sequence of homotopy groups for any choice of basepoint that the induced map ΩmX → ΩmY is a homotopy equivalence, and whenever m ≥ 1 that the morphism πm−1(X, x0) → πm−1(Y, f (x0)) is injective for any basepoint x0 ∈ X. Besides, we deduce from the same long exact sequence that the induced map ΩX → ΩY is (m − 3)-truncated. Lemma 2.11. Let (A, (Eα)α∈T ) be a deformation context such that A is a pointed ∞-category. Let m ≥ 0 and let X : Aart → S be an m-proximate formal moduli problem. Then the natural map ΩmX → ΩmLX induced by the unit map X → LX is an equivalence. In particular the functor ΩmX is a formal moduli problem. Proof. By [Lurb, Thm 5.1.9] the functor X is m-proximate if and only if the unit map X → LX is (m − 2)-truncated. If it is the case, we deduce from Remark 2.10 that the induced map ΩmX → ΩmY is an equivalence. 12 Notation 2.12. Let X be an ∞-category which admits finite products. Consider the cartesian symmetric monoidal structure X × with monoidal product given by the cartesian product (see [Lurd, §2.4.1]). Let 0 ≤ m ≤ ∞ and consider the ∞-category MonEm(X ) = AlgEm(X ×) of Em-monoid objects in X . Recall that an E1-monoid object M in X with multiplication map m : M × M → M is called grouplike if the two maps (m, p1) : M × M → M × M, (m, p2) : M × M → M × M 1 (cid:44)→ E⊗ m) is grouplike. Let Mongp Em are equivalences in X , where pi is the ith projection map. Recall that an Em-monoid object M in X is called grouplike if the underlying E1-monoid (for any embedding of ∞-operads E⊗ (X ) ⊆ MonEm(X ) denote the full subcategory spanned by grouplike Em-monoid objects (see [Lurd, Def 5.2.6.6]). If X is a presentable ∞-category, then an Em-monoid object M in X is grouplike if and only if the monoid π0M is a group. If X is a pointed presentable ∞-category, by [Lurd, Not 5.2.6.11] the iterated loop functor Ωm : X → X factorizes as a functor Ωm : X → Mongp Em (X ), E∞(X ). 1 (cid:44)→ E⊗ (X ) (cid:39) Mongp if m = ∞ we set Ω∞ = limm Ωm via the equivalence where by convention, limm MonEm(X ) (cid:39) MonE∞(X ). Remark 2.13. Let X be an ∞-category which admits finite products. The prop- erty of being grouplike for an Em-monoid object is independent of the choice of the embedding of ∞-operads E⊗ m (see [Lurd, Rem 5.2.6.8]). This implies that the equivalence limm MonEm(X ) (cid:39) MonE∞(X ) restricts to an equivalence limm Mongp Em Remark 2.14. Let X be an ∞-category which admits finite products. Consider (X ) ⊆ MonEm(X ) of grouplike Em-monoid objects into all the full embedding Mongp Em Em-monoid objects. This inclusion preserves finite product and therefore extends to (X )) → a symmetric monoidal functor. We have an induced functor MonE1(Mongp Em MonE1(MonEm(X )) on the ∞-categories of E1-monoid objects which is fully faithful. By the additivity theorem [Lurd, Thm 5.1.2.2] the target of this functor is equiv- alent to the ∞-category MonEm+1(X ) of Em+1-monoid objects and by definition of grouplike objects the equivalence MonE1(MonEm(X )) (cid:39) MonEm+1(X ) restricts to (X ). Moreover the above embedding an equivalence Mongp E1 restricts to an embedding (MonEm(X )) (cid:39) Mongp Em+1 Mongp E1 (Mongp Em (X )) → Mongp E1 (MonEm(X )) (cid:39) Mongp Em+1 (X ) on the subcategories of grouplike E1-monoid objects. This embedding is an equiva- lence of ∞-categories. Indeed, it is also essentially surjective, for if M is a grouplike E1-object in MonEm(X ), it follows from Remark 2.13 that the underlying Em-monoid object of M is grouplike and lies in the essential image. The following result provides conditions on the underlying deformation context under which the iterated loop functor is an equivalence for the ∞-category ModuliA of formal moduli problems. 13 Proposition 2.15. Let (A, (Eα)α∈T ) be a deformation context admitting a deforma- tion theory and such that A is a pointed ∞-category. Then for every 0 ≤ m ≤ ∞, the iterated loop functor Ωm : ModuliA → Mongp (ModuliA) is an equivalence of Em ∞-categories. Proposition 2.15 is a consequence of the following Lemma. Lemma 2.16. Let X be a pointed presentable ∞-category. Suppose that there exists a presentable stable ∞-category A and a functor f : X → A which is conservative, preserves geometric realizations of simplicial objects and preserves finite limits. Then (X ) is an equivalence of ∞- for any 0 ≤ m ≤ ∞ the functor Ωm : X → Mongp Em categories. Proof. If m = 0 the assertion follows from the assumption that X is pointed, since (X ) (cid:39) MonE0(X ) (cid:39) X∗, where X∗ is the ∞-category of there are equivalences Mongp E0 pointed objects. Otherwise for finite m, by Remark 2.14 and ascending induction on m it suffices to prove the result for m = 1. For this, consider the diagram X ΩX / Mongp E1 (X ) f f gp A ΩA / / Mongp E1 (A). Because f preserves finite limits, there exists a natural equivalence f gpΩX (cid:39) ΩAf making this diagram commutative up to equivalence. The ∞-category A being stable, the functor ΩA is an equivalence . The functors ΩX and ΩA admit as left adjoint the functors BX and BA respectively given by the bar construction. By virtue of the assumption that f preserves geometric realizations of simplicial objects, and that the bar construction is given by the geometric realization of a simplicial object (see [Lurd, Rem 5.2.2.8]), there exists a natural equivalence f BX (cid:39) BAf gp. Let : BX ΩX → idX be the counit map. The composition of with the composite functor ΩAf is equivalent to the natural transformation f gpΩX (cid:39) ΩABAf gpΩX (cid:39) ΩAf BX ΩX → ΩAf which is an equivalence because f preserves finite limits. Because ΩA is an equiv- alence and f is conservative, this implies that is an equivalence. We can prove similarly that the unit map η : id → ΩX BX is an equivalence by composing with the functor f gp. The assertion for m = ∞ follows by passing to the limit over m and form Remark 2.13. Proof of Proposition 2.15. Let D : Aop → B be a deformation theory for the de- formation context (A, (Eα)α∈T ). By definition, we have at our disposal for every α ∈ T , a functor eα : B → Sp which preserves small sifted colimits and such that the family {eα}α is jointly conservative (see [Lurb, Cor 1.3.8, Def 1.3.9]). This implies that the functor Dα = A (cid:89) α∈T e = eα : B → (cid:89) α∈T 14 / where Aα = Sp for every α, is conservative and preserves small sifted colimits. By construction, the functor eα preserves finite limits, so that e has the same property. Denote by t : ModuliA → B the inverse of the equivalence Ψ : B → ModuliA (see [Lurb, Thm 1.3.12]). Then the composite functor e ◦ t : ModuliA → A satisfies all the assumptions of Lemma 2.16, by which we deduce the claim. Notation 2.17. In the situation of Proposition 2.15, we denote by Bm : (ModuliA) → ModuliA the inverse of the equivalence Ωm : ModuliA → Mongp Em Mongp (ModuliA). Em Notation 2.18. Let (A, (Eα)α∈T ) be a deformation context such that A is a pointed (Fun∗(Aart,S)) ∞-category. Consider the restriction Ωm (Fun∗(Aart,S)). By of the iterated loop functor Ωm : Fun∗(Aart,S) → Mongp Em prox : Prox(m)A → Lemma 2.11, the functor Ωm prox factorizes as a functor Ωm Mongp Em prox : Prox(m)A → Mongp Em (ModuliA). Corollary 2.19. Let (A, (Eα)α∈T ) be a deformation context admitting a defor- mation theory and such that A is a pointed ∞-category. Let m ≥ 0 and let Lm : Prox(m)A → ModuliA denote the restriction of the functor L to the sub- category Prox(m)A of m-proximate formal moduli problems. Then there exists an prox : Prox(m)A → ModuliA. In particular, the equivalence of functors Lm (cid:39) BmΩm functor Lm preserves all small limits. As a consequence, for every m-proximate formal moduli problem X, the functor LmX is given by the formula LmX(A) (cid:39) MapMongpEm (ModuliA)(ΩmSpf(A), ΩmX). prox (cid:39) ΩmLm, which Proof. By Lemma 2.11, there is an equivalence of functors Ωm after composition with Bm provides by Proposition 2.15 the expected natural equiv- prox is given by a limit and that Bm is an equiva- alence BmΩm lence, we have that Lm preserves all small limits. Let X be an m-proximate formal moduli problem. Then for A ∈ Aart we have natural homotopy equivalences prox (cid:39) Lm. Because Ωm LmX(A) (cid:39) BmΩm proxX(A) (cid:39) MapModuliA (cid:39) MapMongpEm (Spf(A), BmΩmX) (ModuliA)(ΩmSpf(A), ΩmX). 3 Reminders on deformations of categories We start with notations and terminology related to presentable, compactly generated and linear ∞-categories, as well as the relation with differential graded categories. We follow the terminology of [Lur09] and [Lurd]. We also point out the reference [Rob14] where we can find an overview on this subject. 15 3.1 Linear ∞-categories We denote by PrL the ∞-category of not necessarily small presentable ∞-categories together with functors which are left adjoints (see [Lur09, Def 5.5.3.1] for a con- struction of PrL). An ∞-category C is said to be compactly generated if it is presentable and if in addition there exists a small ∞-category C0 which admits finite colimits and an equivalence Ind(C0) (cid:39) C. In this situation, by [Lur09, Thm 5.4.2.2] the small ∞- category C0 which generates C can be choosen to be the small idempotent complete full subcategory Cc ⊆ C consisting of compact objects of C. By [Lurd, Prop 4.8.1.14], the ∞-category PrL admits a symmetric monoidal structure in the sense of [Lurd, Def 2.1.2.13] and we denote the induced symmetric monoidal product by ⊗. If C,C(cid:48) are presentable ∞-categories of the form C = P(C0) and C(cid:48) = P(C(cid:48) 0) with C0, C0 small ∞-categories, then we have C ⊗ C = P(C0 × C0) where × is the cartesian symmetric monoidal structure on Cat∞. The unit of PrL,⊗ is the ∞-category S of small spaces. Moreover this symmetric monoidal structure is closed and for two presentable ∞-categories C, C, the ∞-category FunL(C,C) of colimit preserving ∞-functors is presentable. By [Lurd, Lem 5.3.2.11], if C and C are compactly generated k-linear ∞-categories, the tensor product C ⊗ C is again compactly generated and the subcategory (C ⊗C)c is generated under finite colimits by Cc × Cc. Let k be an E∞-ring. The ∞-category Modk of k-module spectra admits a sym- metric monoidal structure Mod⊗ k given by the smash product over k. Considering Mod⊗ k as an E∞-algebra object in the symmetric monoidal ∞-category PrL,⊗, we define the ∞-category of k-linear ∞-categories to be (PrL,⊗). PrL k := ModMod⊗ k k are called k-linear functors. Because Modk is a stable ∞-category Morphisms in PrL in the sense of [Lurd, Def 1.1.1.9], the underlying ∞-category of any k-linear ∞- category is stable. For example if k = S the sphere spectrum, then PrLS is the ∞- category of presentable stable ∞-categories with exact functors between them. The symmetric monoidal structure on PrL induces a symmetric monoidal structure on k . We denote the corresponding tensor product by ⊗k. Moreover the symmetric PrL k is closed. For k-linear ∞-categories C, C(cid:48) and C(cid:48)(cid:48) there monoidal structure on PrL k (C,C) of k-linear colimit preserving ∞-functors, exists a k-linear ∞-category FunL such that there exists an equivalence MapPrL k (C ⊗k C(cid:48),C(cid:48)(cid:48)) (cid:39) MapPrL k (C, FunL k (C(cid:48),C(cid:48)(cid:48))). In the case C = C(cid:48), we denote by Endk(C) = FunL k (C,C) the k-linear ∞-category of k-linear endofunctors of C. Any k-linear ∞-category C ∈ PrL k is enriched over Modk. Indeed if E, F ∈ C are objects, the ∞-functor Modk → S given on objects by V (cid:55)→ MapC(V ⊗ E, F ) commutes with small colimits and by presentability of C is therefore representable by an object MapC(E, F ) ∈ Modk satisfying the universal property MapC(V ⊗ E, F ) (cid:39) MapModk (M, MapC(E, F )). 16 We set the notation ExtmC (E, F ) = π−mMapC(E, F ) for the Ext groups in C. We now recall the relation of the above with the homotopy theory of k-linear differential graded categories. Let k be a discrete commutative ring. We consider two ∞-categories of k-linear dg-categories (we use the terminology of [Rob14, §6.1.1]): • Dg(k) : the ∞-category encoding the homotopy theory of presentable k-linear dg-categories up to quasi-equivalence. Morphisms are dg-functors which com- mute with small sums. The dg-nerve construction provides an ∞-functor dg : Dg(k) −→ PrL. NL • Dgc(k) : the ∞-category encoding the homotopy theory of presentable com- pactly generated k-linear dg-categories up to quasi-equivalence. Morphisms are dg-functors which commute with small sums and preserve compact ob- jects. The dg-nerve restricts to an ∞-functor dg : Dgc(k) −→ PrL Nc ω. By the work of Cohn [Coh13] there exists a commutative diagram of ∞-categories Dg(k) NL dg / PrL k Dgc(k) Nc dg / PrL ω,k k (PrL ω,k = ModMod⊗ ω) is the ∞-category of k-linear compactly generated ∞- where PrL categories together with k-linear functors which preserve small colimits. Moreover the bottom horizontal arrow is an equivalence of ∞-categories. Notation 3.1. Let k be an E∞-ring and C a k-linear ∞-category. If E is a collection of objects of C, we denote by E⊥ the full subcategory of C consisting of objects C such that MapC(E, C) (cid:39) 0 for every E ∈ E. Definition 3.2. Let k be an E∞-ring. Let C be a k-linear ∞-category. 1. A collection E of objects of C generates C if E⊥ consists of zero objects of C. 2. A collection E of objects of C is a family of compact generators of C if each object E ∈ E is compact and if the collection E generates C. 3. A compact generator of C is a compact object E of C such that the family E = {E} generates C. Definition 3.3. Let k be an E∞-ring. A k-linear ∞-category C is said to be tamely compactly generated if C is compactly generated and if for any compact objects E, E(cid:48) of C we have ExtmC (E, E(cid:48)) = 0 for m (cid:29) 0. Remark 3.4. Let k be an E∞-ring and C a k-linear ∞-category. Suppose that C admits a single compact generator E. Then C is tamely compactly generated if and only if ExtmC (E, E) = 0 for m (cid:29) 0. 17 / / ? O O ? O O 3.2 Deformations of objects We recall from [Lurb, §5.2] the fundamental results about the deformation functor of an object in a linear ∞-category over a field. Let k be a field, let C be a k-linear ∞-category and let E ∈ C be an object. Recall from [Lurb, Const 5.2.1] that we can define a functor ObjDef E : Algart k → S from artinian E1-algebras over k to spaces which satisfies the following properties. • For every artinian E1-algebra A over k, the space ObjDef E(A) classifies de- formations of E over A in C, or in other words it is the classifying space of pairs (EA, u) with EA ∈ RModA(C) a right A-module object in C and u : EA ⊗A k (cid:39) E an equivalence in C. • In general, the functor ObjDef E does not satisfy the derived Schlessinger con- E = L(ObjDef E) ditions and is not a formal E1-moduli problem. Let ObjDef∧ denote the associated formal E1-moduli problem. problem, or equivalently the unit map ObjDef E → ObjDef∧ • By [Lurb, Cor 5.2.5], the functor ObjDef E is a 1-proximate formal E1-moduli E is (−1)-truncated. • Let EndC(E) be the E1-algebra of endomorphisms of E in C. By [Lurb, Thm 5.2.8] the augmented E1-algebra which controls the formal E1-moduli problem ObjDef∧ In other words there exists for every A ∈ Algart E is given by k ⊕ EndC(E). k a natural equivalence ObjDef∧ E(A) (cid:39) MapAlgk (D(1)(A), EndC(E)). In particular the tangent spectrum of ObjDef∧ EndC(E)[1]. E is given by TObjDef∧ E (cid:39) • Koszul duality for modules implies that for every artinian E1-algebra A we have a natural homotopy equivalence ObjDef∧ E(A) (cid:39) LMod! A(C)(cid:39) ×C(cid:39) {E} with the space of deformations of E as an Ind-coherent right A-modules in C, where LMod! A ⊗k C is the ∞-category of Ind-coherent right A-modules in C (see [Lurb, Rem 5.2.16]). A(C) = LMod! • By [Lurb, Prop 5.2.14], under the assumption that C admits a left complete t-structure and that E is connective, then ObjDef E is a formal E1-moduli problem. Notation 3.5. Let k be a field, let C be a k-linear ∞-category and let E ∈ C be E2 an object. We will denote by ObjDef E the restriction of the functor ObjDef E to artinian E2-algebras along the forgetful functor Alg(2),art → Algart k . k 18 3.3 Deformations of categories We first recall from [Lurb, §5.3] the fundamental results about the deformation functor of a linear ∞-category over a field. Then we recall the construction of this functor. In [Lurb, §5.3] is constructed a Let k be field and C a k-linear ∞-category. CatDefC : Alg(2),art −→ (cid:98)S functor k from artinian E2-algebras over k to not necessarily small spaces which satisfies the following properties. • For any artinian E2-algebra A over k, the space CatDefC(A) classifies defor- mations of C over A, or in other words pairs (CA, u) with CA a right A-linear ∞-category and u : CA ⊗A k (cid:39) C an equivalence in PrL k (see below for the def- inition of linear ∞-category over an E2-algebra). We observe that the space CatDefC(k) is contractible. • In general, the functor CatDefC does not satisfy the derived Schlessinger con- C := L(CatDefC) ditions, and is not an E2-formal moduli problem. Let CatDef∧ denote the associated E2-formal moduli problem and σC : CatDefC → CatDef∧ C denote the unit map. In other words, the map σC is in general not an equiva- lence. We will see a simple example below. • However, the map σC is rather close to be an equivalence as shown by Lurie. Namely, by [Lurb, Cor 5.3.8] the functor CatDefC is a 2-proximate E2-formal moduli problem in the sense of [Lurb, Def 5.1.5] (with values in essentially small spaces). By [Lurb, Thm 5.1.9], it means that the map σC has 0-truncated fibers, or in other words, that the map σC induces an isomorphism on πi for i ≥ 2 and is injective on π1 (all homotopy groups are based at the trivial deformation of C). • Let ξ(C) denote the E2-algebra over k given by the center of the ∞-category C in the sense of [Lurb, Def 5.3.10]. The underlying k-module spectrum of ξ(C) is given by EndEndk(C)(idC) so that the homotopy groups of ξ(C) are isomorphic to the Hochschild cohomology groups of C. By [Lurb, Thm 3.5.1], the augmented E2-algebra which controls the E2-formal moduli problem CatDef∧ C is given by k ⊕ ξ(C). In other words, for any artinian E2-algebra A over k there exists a natural equivalence CatDef∧ C (A) (cid:39) Map Alg(2) k (D(2)(A), ξ(C)) where D(2) is the E2-Kozsul duality functor. In particular the tangent spec- trum of CatDef∧ C is given by TCatDef∧ C (cid:39) ξ(C)[2]. 19 We have stronger properties if we restrict to compactly generated deformations as observed by Lurie (see [Lurb, Variant 5.3.4]), a condition which is reasonable in most examples of interest. Let CatDef cC denote the subfunctor of CatDefC spanned by pairs (CA, u) where CA is compactly generated. By [Lurb, Variant 5.3.6], the functor CatDef cC has values in essentially small spaces, and can therefore be viewed as a functor CatDef cC : Alg(2),art k −→ S which satisfies the following properties. • The natural embedding CatDef cC → CatDefC has 0-truncated fibers, which implies that CatDef cC is a 2-proximate E2-formal moduli problem and that the induced map L(CatDef cC) → L(CatDefC) = CatDef∧ C is an equivalence. We denote by θC : CatDef cC → CatDef∧ C the unit map. • Suppose that C is tamely compactly generated (see Definition 3.3). Then the functor CatDef cC is a 1-proximate E2-formal moduli problem by [Lurb, Prop 5.3.21]. In other words, the map θC induces an isomorphism on πi for i ≥ 1 and an injection on π0. Our main result gives additional conditions under which the surjectivity on π0 holds for formal deformations. We recall now the construction of the functor CatDefC from [Lurb, Const 5.3.2]. By [Lurd, Thm 4.8.5.16] there exists a functor Algk −→ PrL k A (cid:55)−→ LModA which extends to a symmetric monoidal functor. Passing to algebra objects we obtain a functor Alg(2) k (cid:39) Alg(Algk) −→ Alg(PrL k ) A (cid:55)−→ LMod⊗ A A-module in PrL which to an E2-algebra A associates the monoidal ∞-category of left A-modules. Let A be an E2-algebra over k. A right A-linear ∞-category (resp. left A-linear ∞-category) is by definition a right LMod⊗ k (resp. a left LMod⊗ A- linear ∞-category, we denote by CA⊗AA(cid:48) the extension of scalars CA(cid:98)⊗LModALModA(cid:48). k ). If A → A(cid:48) is a morphism of E2-algebras over k, and CA a right A- module in PrL k ) with fiber over a k ) of ∞- k-linear monoidal ∞-category D⊗ given by the ∞-category RModD⊗(PrL categories right tensored over D⊗. Set RCatk = Alg(2) k ) and consider the induced cocartesian fibration p : RCatk → Alg(2) k with fiber over an E2- algebra A given by the ∞-category RModLMod⊗ k ) of right A-linear ∞-categories. Consider the subcategory RCatcocart consisting of p-cocartesian morphisms so that k ) → Alg(PrL k ×Alg(PrL Consider the cocartesian fibration RMod(PrL k ) RMod(PrL (PrL k A 20 → Alg(2) k with fiber over A the space of right Let C be a k-linear ∞-category and consider the pair (k,C) as an object of . The ∞-category of deformations of C is the ∞-category Deform[C] = )/(k,C). The induced left fibration Deform[C] → (Alg(2) is we obtain a left fibration RCatcocart A-linear ∞-categories. RCatcocart (RCatcocart classified by a functor Alg(2),aug denoted by CatDefC : Alg(2),art → (cid:98)S whose restriction to artinian E2-algebras is k )/k (cid:39) Alg(2),aug −→ (cid:98)S. k k k k k k 4 Generators and curvature in deformations of cate- gories In this section we first set up some preliminary results concerning simultaneous deformations and deformations of E1-algebras. Finally, in the last subsection, we use them to prove our main results concerning formal deformations. 4.1 Simultaneous deformations Let k be a field. We now study the deformation functor of a pair (C, E) with C a k-linear ∞-category and E ∈ C an object. construction of CatDefC in [Lurb, Const 5.3.2] but replacing the ∞-category PrL the slice ∞-category (PrL an object of C. We first give a construction of this functor which consists in mimicking the k by k )Modk/ of pairs (C, E) with C a k-linear ∞-category and E Construction 4.1. [Lurd, Thm 4.8.5.16] provides a symmetric monoidal functor Algk −→ (PrL k )Modk/ A (cid:55)−→ (LModA, A) where the ∞-category (PrL k )Modk/ is endowed with the natural symmetric monoidal structure from [Lurd, Rem 2.2.2.25] coming from the fact that Modk is the unit in PrL,⊗ . Applying algebra objects we obtain a functor k Alg(2) k )Modk/) (cid:39) Alg(PrL k (cid:39) Alg(Algk) → Alg((PrL k ) A (cid:55)−→ LMod⊗ A. Consider the cocartesian fibration RMod((PrL k )Modk/) → Alg(PrL k ) with fiber over a monoidal ∞-category D⊗ the ∞-category RModD⊗((PrL k )Modk/) of pairs (M, M ) where M is a k-linear ∞-category right tensored over D⊗ and M ∈ M an object. Set RCat∗ k = Alg(2) k ) RMod((PrL k )Modk/) and consider the induced cocartesian fibration q : RCat∗ k with fiber over an E2-algebra A the ∞-category of right A-linear ∞-categories together with a distinguished object. k ×Alg(PrL k → Alg(2) 21 Let (RCat∗ induced left fibration (RCat∗ an object of C. The ∞-category of deformations of (C, E) is the ∞-category k)cocart be the subcategory of q-cocartesian morphisms so that we have an k . Let C be a k-linear ∞-category and E k)cocart → Alg(2) Deform[C, E] := (RCat∗ k)cocart /(k,C,E). k )/k (cid:39) Alg(2),aug −→ (cid:98)S. k The induced left fibration Deform[C, E] → (Alg(2) functor Alg(2),aug → (cid:98)S whose restriction to artinian E2-algebras is denoted by is classified by a k SimDef (C,E) : Alg(2),art k For any artinian E2-algebra A over k, the space SimDef (C,E)(A) classifies 4- uples (CA, EA, u, v) where CA is a right A-linear ∞-category, u : CA ⊗A k (cid:39) C is an equivalence, EA ∈ CA is an object and v : EA ⊗A k (cid:39) A is an equivalence in C. We have that SimDef (C,E)(k) is a contractible space. The deformation functor SimDef (C,E) is closely related to the deformation func- tors CatDefC and ObjDef E. We construct the evident maps between them. Construction 4.2. Let k be a field, C a k-linear ∞-category and E an object of C. k → RCatk which in The forgetful functor (PrL turn induces a functor Deform[C, E] → Deform[C] commuting with the projection to Alg(2),aug k induces a functor RCat∗ . We obtain natural transformations k )Modk/ → PrL k τ(C,E) : SimDef (C,E) −→ CatDefC ρ(C,E) : SimDef c (C,E) −→ CatDef cC given on objects by forgetting the distinguished object deformation. The homotopy fiber of the forgetful functor (PrL k at C is equivalent to the Kan complex MapPrL k )Modk/ is given on object by sending an object E of C(cid:39) to the pair (C, E). This functor induces a functor Deform[E] → Deform[C, E] commuting with the projection to Alg(2),aug (Modk,C) (cid:39) C(cid:39). The induced functor C(cid:39) → (PrL . We therefore obtain a natural transformation k )Modk/ → PrL k k ϕ(C,E) : ObjDef E −→ SimDef (C,E) E2 E2 is the restriction of the functor ObjDef E to artinian E2-algebras where ObjDef over k along the forgetful functor Alg(2),art k . The functor ϕ(C,E) is given on object by sending a deformation EA of E over an E2-algebra A to the pair (C ⊗k A, EA). Moreover, whenever C is compactly generated, we have that the map ϕ(C,E) factors as a map ObjDef Proposition 4.3. Let k be a field, C a k-linear ∞-category and E ∈ C an object. The sequence of natural transformations E → SimDef c E2 → Algart (C,E). k ObjDef E −→ SimDef (C,E) −→ CatDefC E2 22 ,(cid:98)S). Consequently, the from Construction 4.2 is a fiber sequence in Fun∗(Alg(2),art functor SimDef (C,E) is a 2-proximate formal E2-moduli problem (in a larger uni- verse). Moreover it induces a homotopy commutative diagram / CatDefC SimDef (C,E) ObjDef k E2 E ObjDef E2,∧ E / SimDef∧ (C,E) / CatDef∧ C where the bottom line is a fiber sequence in Moduli E2 k . k )Modk/ → PrL Proof. The first assertion follows from the fact that the homotopy fiber of the functor k at C is given by the Kan complex C(cid:39). The second assertion follows (PrL E2 directly from the existence of the fiber sequence and the fact that ObjDef E and CatDefC are 1-proximate and 2-proximate respectively. The last assertion follows from the fact that the three functors are m-proximate formal moduli problems for some m and from the fact (after suitable enlargement of the universe) that the functor Lm : Prox(m)E2 → Moduli E2 k preserves small limits by Corollary 2.19. Remark 4.4. We can already deduce from Proposition 4.3 the tangent spectrum of the formal E2-moduli problem SimDef∧ C . However, to avoid dealing with the E2- E2,∧ algebra associated to the formal E2-moduli problem ObjDef E , we give a direct construction of the E2-algebra associated to SimDef∧ C , which will be useful in the sequel. Notation 4.5. Let k be a field, C a k-linear ∞-category and E ∈ C an object. Consider the object (C, E) of the slice ∞-category (PrL k )Modk/ and its endomorphism monoidal k-linear ∞-category Endk(C, E)⊗ := End(PrL ((C, E))⊗. Denote by ξ(C, E) = EndEndk(C,E)(id(C,E)) the E2-algebra over k given by endomorphisms of the unit. By adjunction, this E2-algebra satisfies the following universal property: for any E2-algebra A over k, there exits a homotopy equivalence k )Modk / Map Alg(2) k (A, ξ(C, E)) (cid:39) LModLMod⊗ A ((PrL k )Modk/)(cid:39) ×((PrL k )Modk /)(cid:39) {(C, E)} where the right hand side is the space of left A-linear structures on the pair (C, E), or in other words, the space of left A-linear structures on C acting trivially on E up to equivalence. Remark 4.6. Let k be a field, C a k-linear ∞-category and E ∈ C an object. By definition of the slice ∞-category (PrL k )Modk/, the k-linear ∞-category Endk(C, E) is the homotopy fiber of the k-linear functor evE : End(C) → C given by evaluation at E. We deduce that the induced sequence ξ(C, E) → ξ(C) → EndC(E) is a fiber sequence of k-module spectra. We observe that the first map in the sequence is a map of E2-algebras over k. 23 / / / / / Proposition 4.7. Let k be a field, C a k-linear ∞-category and E ∈ C an ob- ject. Consider the augmented E2-algebra given by k ⊕ ξ(C, E) and its associated E2-formal moduli problem Ψk⊕ξ(C,E). Then there exists a natural transformation SimDef (C,E) → Ψk⊕ξ(C,E) which is 0-truncated. Therefore the formal E2-moduli problem SimDef∧ (C,E) is given by the formula SimDef∧ (C,E)(A) (cid:39) Map Alg(2) k (D(2)(A), ξ(C, E)). Proof. We mimic the proof of [Lurb, Thm 5.3.16], replacing the center ξ(C) by ξ(C, E). Following the same lines as in [Lurb, Const 5.3.18], we see that there exits a homotopy commutative diagram Deform[C, E] (Alg(2),aug k )/ξ(C,E) (Alg(2),aug k )op D(2) / Alg(2),aug k where the upper functor sends a deformation (CA, EA) over an augmented E2- algebra A to C seen as left tensored over D(2)(A) via the equivalence (D(2)(A) ⊗k CA)⊗D(2)(A)⊗kAk (cid:39) C together with the distinguished object E. This functor induces a natural transformation SimDef (C,E) → Ψk⊕ξ(C,E). The same proof as [Lurb, Prop 5.3.17] but replacing the center ξ(C) by the E2-algebra ξ(C, E) shows that this map is 0-truncated. Finally [Lurb, Thm 5.1.9] implies that the induced map SimDef∧ Remark 4.8. In the situation of Proposition 4.7, the universal property of the E2- algebra ξ(C, E) (see Notation 4.5) implies that we have for each artinian E2-algebra A over k a natural homotopy equivalence (C,E) → Ψk⊕ξ(C,E) is an equivalence. SimDef∧ (C,E)(A) (cid:39) LModLMod⊗ D(2)(A) ((PrL k )Modk/)(cid:39) ×((PrL k )Modk /)(cid:39) {(C, E)} with the space of left D(2)(A)-linear structures on the ∞-category C such that D(2)(A) acts trivially on the object E. Remark 4.9. Let k be a field, C a k-linear ∞-category and E ∈ C an object. Recall from Proposition 4.3 the existence of a fiber sequence of formal E2-moduli problems ObjDef E2,∧ E −→ SimDef∧ (C,E) −→ CatDef∧ C . It induces a fiber sequence ΩSimDef∧ (C,E) −→ ΩCatDef∧ C −→ ObjDef E2,∧ E which after taking the tangent spectrum gives a fiber sequence of k-module spectra ξ(C, E) −→ ξ(C) −→ EndC(E). 24 / / / C (A) → It can be seen that for each artinian E2-algebra A over k, the map ΩCatDef∧ E2,∧ E (A) sends an A-linear autoequivalence fA : C ⊗k A (cid:39) C ⊗k A of the trivial ObjDef deformation to the image fA(E⊗k A) of the trivial deformation of E in C⊗k A. From this we can deduce that the fiber sequence of tangent spectra above is equivalent to the fiber sequence given in Remark 4.6, although we will not use this fact. We denote by χE : ξ(C) → EndC(E) the map of k-modules induced by evaluation at E and call it the obstruction map of E. Remark 4.10. Let k be a field, C a k-linear ∞-category and E ∈ C an object. Consider the homotopy commutative diagram CatDefC(k ⊕ k[m]) ΩCatDefC(k ⊕ k[m + 1]) / ObjDef E (k ⊕ k[m + 1]) E2 σC CatDef∧ C (k ⊕ k[m]) ∼ / / ΩCatDef∧ C (k ⊕ k[m + 1]) / ObjDef E2,∧ E (k ⊕ k[m + 1]) where the right square is induced by Proposition 4.3. The bottom left map is C is a formal moduli problem. By [Lurb, Thm 3.5.1], C (k⊕ k[m]) (cid:39) HHm+2(C). Moreover by [Lurb, (E, E). an equivalence because CatDef∧ there exists an isomorphism π0CatDef∧ Thm 5.2.8] there is an isomorphism π0ObjDef∧ From Proposition 4.3 and taking π0, we have an induced commutative diagram π0SimDef (C,E)(k ⊕ k[m]) E(k ⊕ k[m + 1]) (cid:39) Extm+2 π0CatDefC(k ⊕ k[m]) π0ObjDef E (k ⊕ k[m + 1]) E2 α / C σC C χE (E, E) / Extm+2 / HHm+2(C) (C,E)(k ⊕ k[m]) π0SimDef∧ where the bottom line is exact in the middle. Let C1 be a deformation of C over k ⊕ k[m] and let φ = σC(C1). Suppose E1 ∈ C1 deforms E or in other words that E1 ⊗k⊕k[m] k (cid:39) E in C1 ⊗k⊕k[m] k (cid:39) C. In this case we have α(C1, E1) = C1 and thus χE(φ) = 0. Example 4.11. We can now explain a simple example in which CatDefC fails to be a formal moduli problem. It is taken from [KL09, Ex 3.14]. Let k be a field of characteristic zero and A = k[u, u−1] the free graded commutative k-algebra gener- ated by a degree 2 variable u and its inverse over k. We regard A as an E∞-algebra and let C = ModA be the k-linear ∞-category of A-module spectra. A computation gives HH2(C) (cid:39) k generated by the multiplication by u whose corresponding cocycle is denoted by φu. Then for every non zero A-module M , we have that χM (φu) (cid:54)= 0 in Ext2 A(M, M ) because multiplication by u on M is an equivalence. Suppose now that φu is the image by the map σC : π0CatDefC(k[t]/t2) → HH2(C) of a deformation C1 of C over k[t]/t2. Then any M1 (cid:54)= 0 in C1 satisfies M = M1 ⊗k[t]/t2 k (cid:54)= 0 in C, and by Remark 4.10 we have χM (φu) = 0, which is a contradiction. We deduce that φu is not in the image of the map σC : π0CatDefC(k[t]/t2) → HH2(C) and that CatDefC is not a formal E2-moduli problem. 25 / / / / / / / / / Notation 4.12. Let k be a field, C a compactly generated k-linear ∞-category and E ∈ C an object. For each artinian E2-algebra A over k we denote by SimDef c (C,E)(A) the summand of SimDef (C,E)(A) consisting of 4-uplets (CA, u, EA, v) with CA being compactly generated. We obtain a functor SimDef c (C,E) : Alg(2),art k → (cid:98)S. Proposition 4.13. Let k be a field, C a compactly generated k-linear ∞-category and E ∈ C an object. The functor SimDef c (C,E) is a 2-proximate E2-moduli problem (C,E) → SimDef (C,E) induces an equivalence of formal E2- and the inclusion SimDef c (C,E). If moreover C is supposed to be (C,E)) (cid:39) SimDef∧ moduli problems L(SimDef c tamely compactly generated, then SimDef c (C,E) is a 1-proximate formal E2-moduli problem. Proof. A variant of Proposition 4.3 implies the existence of a homotopy commutative diagram ObjDef E2 E SimDef c (C,E) / CatDef cC ObjDef E2,∧ E / SimDef c,∧ (C,E) / CatDef∧ C where the two horizontal top maps are constructed in Construction 4.2. This dia- gram implies the first two assertions. The last one follows from this diagram and the fact that if C is tamely compactly generated then by [Lurb, Prop 5.3.21] the functor CatDef cC is 1-proximate. Remark 4.14. Let k be a field, C a tamely compactly generated k-linear ∞-category and E ∈ C an object. Let m ≥ 0 be an integer and consider the following commuta- tive diagram induced by the commutative diagram of the proof of Proposition 4.13, similarly to Remark 4.10. (C,E)(k ⊕ k[m]) π0SimDef c π0CatDef cC(k ⊕ k[m]) π0ObjDef E (k ⊕ k[m + 1]) E2 π0SimDef∧ (C,E)(k ⊕ k[m]) θC / HHm+2(C) χE / Ext2C(E, E) The bottom line is exact in the middle. All vertical maps are injective because when C is tamely compactly generated, the functors SimDef c (C,E) and CatDef cC are E2 1-proximate and moreover ObjDef E is 1-proximate. This implies that the upper line is also exact in the middle, as a sequence of pointed sets. We deduce from this that given a compactly generated deformation C1 of C over k ⊕ k[m], there exists an object E1 ∈ C1 deforming E if and only if χE(θC(C1)) = 0. 26 / / / / / / / / / / / 4.2 Deformations of associative algebras k → Alg(PrL k ) with the functor Alg : Alg(PrL We recall some known facts about the deformation functor of an associative algebra and construct maps to the formal E2-moduli problem of its ∞-category of left mod- ules and from the formal E2-moduli problem of simultaneous deformations of pairs (C, E) given by taking endomorphisms. Construction 4.15. Consider the functor Alg(2) k ) given on objects by A (cid:55)→ LMod⊗ A (see [Lurb, Not 5.3.1] or §3.3). For an E2-algebra A over k, we denote by AlgA = Alg(LMod⊗ A) the presentable ∞-category of A-algebras. Composing the k ) → PrL given by taking k → Alg(PrL functor Alg(2) k → PrL sending an E2-algebra A to the algebra objects, we obtain a functor Alg(2) ∞-category AlgA of A-algebras. This functor is classified by a cocartesian fibration s : Alg → Alg(2) k with fiber over an E2-algebra A given by the ∞-category AlgA. Consider the subcategory of s-cocartesian morphisms Algcocart and the induced left fibration Algcocart → Alg(2) k . Let B be an E1-algebra over k. The ∞-category of deformation of B is the → (cid:98)S whose restriction to artinian E2- ∞-category Deform[B] = (Algcocart)/(k,B). The induced left fibration Deform[B] → Alg(2),aug −→ (cid:98)S algebras is denoted by is classified by a functor Alg(2),aug For any artinian E2-algebra A, the space AlgDef B(A) classifies pairs (BA, u) with BA an A-algebra and u : BA ⊗A k (cid:39) B an equivalence of E1-algebras over k. Proposition 4.16. Let k be a field and B an E1-algebra over k. For every pullback diagram AlgDef B : Alg(2),art k k k A A1 A0 / A01 , the induced map in Alg(2),art k AlgDef B(A) → AlgDef B(A0) ×AlgDef B(A01) AlgDef B(A1) has (−1)-truncated homotopy fibers (or equivalently induces a homotopy equivalence onto its essential image). Proof. By [Lura, Thm 7.4], for every such pullback square in Alg(2),art functor , the induced k LModA → LModA0 ×LModA01 LModA1 is fully faithful. Because the functor Alg : Alg(PrL) → PrL given by applying alge- bra objects preserves small limits and preserves fully faithful functors, the induced functor AlgA → Alg(LModA0 ×LModA01 LModA1) (cid:39) AlgA0 ×AlgA01 AlgA1 is fully faithful, which implies our assertion. 27 / / / Corollary 4.17. Let k be a field and B an E1-algebra over k. For every artinian E2-algebra A over k, the space AlgDef B(A) is essentially small. Therefore AlgDef B can regarded as a functor with values in the ∞-category S of small spaces. Proof. Let m ≥ 0 be an integer. By Proposition 4.16 the map AlgDef B(k⊕ k[m]) → ΩAlgDef B(k ⊕ k[m + 1]) induces a homotopy equivalence onto its essential image. Moreover the target of this map is homotopy equivalent to the homotopy fiber of the restriction map MapAlgA (B, B) at idB and is therefore essentially small. This implies that AlgDef B(k ⊕ k[m]) is essentially small for any m ≥ 0. Let now A be an artinian E2-algebra and let A = A0 → A1 → . . . → An = k be a finite sequence of maps such that for every 0 ≤ i ≤ n− 1 there exists an integer mi ≥ 0 and a pullback diagram (B ⊗k A, B ⊗k A) → MapAlgk Ai k Ai+1 / k ⊕ k[mi]. By Proposition 4.16 the induced map AlgDef B(A) → AlgDef B(k) ×AlgDef B(k⊕k[mi]) AlgDef B(Ai+1) is a homotopy equivalence onto its essential image. We then deduce by descending induction on i, using the first part of the proof that AlgDef B(A) is essentially small. Corollary 4.18. Let k be a field and B an E1-algebra over k. The the functor AlgDef B : Alg(2),art → S is a 1-proximate formal E2-moduli problem. k Proof. It follows by combining Proposition 4.16 and Corollary 4.17. Proposition 4.19. Let k be a field and let B be an n-connective E1-algebra over k for some integer n (i.e. the underlying k-module of B is n-connective). Then the functor AlgDef B is a formal E2-moduli problem. ≥n Proof. For an E2-algebra A, let LMod A denote the full subcategory of LModA ≥n A denote the full sub- spanned by n-connective left A-modules. Moreover let Alg category of AlgA spanned by the n-connective A-algebras. By a variant of [Lura, Prop 7.6], for every pullback diagram in Alg(2),art k functor A A0 A1 / A01 such that the maps π0A0 → π0A01 ← π0A1 are surjective, the induced LMod ≥n A −→ LMod ≥n A0 ×LMod ≥n A01 LMod ≥n A1 28 / / / / / / ≥n A is the ∞-category of n-connective left A-modules. is an equivalence, where LMod Because for every E2-algebra A(cid:48) the forgetful functor AlgA(cid:48) → LModA(cid:48) is conserva- tive, the proof of [Lura, Prop 7.6] also shows that the induced functor Alg ≥n A −→ Alg ≥n A0 ×Alg Alg ≥n A1 ≥n A01 is an equivalence. Remark that if M is an n-connective k-module and if MA(cid:48) is a deformation of M over an artinian E2-algebra A(cid:48), then the A(cid:48)-module MA(cid:48) is n- connective (see the proof of [Lurb, Prop 5.2.14]). By the above, this implies that the induced map AlgDef B(A) −→ AlgDef B(A0) ×AlgDef B(A01) AlgDef B(A1) is a homotopy equivalence and therefore that AlgDef B is a formal E2-moduli prob- lem. Notation 4.20. Let k be a field and B an E1-algebra over k. Consider the co- cartesian fibration s : Alg → Alg(2) from Construction 4.15 and the cocartesian fibration p : RCatk → Alg(2) from §3.3. By [Lurd, Not 4.8.5.10] there exists a func- tor Alg → RCatk which to a pair (A, BA) with A an E2-algebra over k and BA an A-algebra, assigns the pair (A, LModBA(LModA)) where LModBA(LModA) is seen as right tensored over LMod⊗ A. By [Lurd, Prop 4.8.5.1] the functor Alg → RCatk sends s-cocartesian morphisms to p-cocartesian morphisms and therefore induces a natural transformation AlgDef B → CatDef LModB which factors as a natural trans- formation k k MB : AlgDef B −→ CatDef c LModB given on objects by applying the ∞-category of left modules. Notation 4.21. Let k be a field, let C be a compactly generated k-linear ∞-category and let E ∈ C be an object. Consider the cocartesian fibration s : Alg → Alg(2) from Construction 4.15 and the cocartesian fibration q : RCat∗ from Con- struction 4.1. By [Lurd, Thm 4.8.5.11], there exists a functor RCat∗ k → Alg which to a triple (A,CA, EA) with A an E2-algebra over k, CA a right A-linear ∞-category and EA ∈ CA an object, assigns the pair (A, EndCA (EA) is seen as an A-algebra. It is easy to see that the functor RCat∗ k → Alg sends q-cocartesian morphisms to s-cocartesian morphisms and therefore induces a natural transforma- tion (EA)) where EndCA k → Alg(2) k k E(C,E) : SimDef c (C,E) −→ AlgDef EndC(E) given on objects by taking endomorphisms. Remark 4.22. Let k be a field, let C be a tamely compactly generated k-linear ∞- category and let E ∈ C be a compact generator of C. Let A be an artinian E2-algebra over k, and suppose given a deformation CA of C over A and a deformation EA ∈ CA of E. Then by [Lurb, Lem 5.3.31] we have that CA is tamely compactly generated, and [Lurb, Lem 5.3.37] implies that the object EA is a compact generator of CA. 29 Therefore by [Lurd, Thm 7.1.2.1] (Schwede–Shipley result), there exists a natural A-linear equivalence LModEndCA (EA)(LModA) (cid:39) CA. This fact implies that the diagram SimDef c (C,E) E(C,E) / ρ(C,E) AlgDef EndC(E) MEndC (E) CatDef cC commutes up to natural homotopy, where ρ(C,E) is the natural transformation from Construction 4.2 given by forgetting the distinguished object. Indeed, unwinding the construction of each of these functors, we see that it suffices to produce for each artinian E2-algebra A over k an equivalence as above. 4.3 Formal deformations Let k be a field and let C be a k-linear ∞-category. Recall that under the assumption that C is tamely compactly generated (see Definition 3.3), we have by [Lurb, Prop 5.3.21] that the natural transformation θC : CatDef cC −→ CatDef∧ C is (−1)-truncated. Under some additional assumptions on C, we now prove that the map θC induces a surjection on π0 for formal deformations, and therefore is an equivalence on formal deformations (that is, over the adic ring k[[t]] of formal power series). To prove this, we prove a stronger statement which implies the existence of a compact generator for formal deformations. We first set up some conventions about formal deformations. Let n ≥ 1 be an integer, let k be a field and let X : Alg(n),art → S be a functor. Consider the discrete commutative algebra k[[t]] of formal power series in one variable k → Alg(n),aug as an augmented En-algebra over k via the forgetful functor CAlgaug . The space of formal deformations in X is by definition the space k k X(k[[t]]) = Map Fun∗(Alg(n),art k ,S) (Spf(k[[t]]), X) where Spf : Alg(n),aug k → Moduli En k is the formal spectrum (see Notation 2.3). Lemma 4.23. Let k be a field and let . . . → B2 → B1 → B0 be an inverse system of discrete local artinian commutative k-algebras having same residue field k. Let B = limi Bi be the limit. Let n ≥ 2 and consider for each i the algebra Bi as an artinian En-algebra and B as an augmented En-algebra via . Then the natural map colimiSpf(Bi) → the forgetful functor CAlgaug Spf(B) is an equivalence of formal En-moduli problems. Equivalently, the natural map colimiD(n)(Bi) → D(n)(B) is an equivalence of augmented En-algebras over k. k → Alg(n),aug k 30 / ( ( Proof. The statement for n = ∞ is given by [Lurc, Lem 6.3.3]. More precisely, let R = k ⊕ k[m] as an artinian E∞-algebra for some integer m ≥ 0 and consider the homotopy commutative diagram colimiMapCAlgaug k (Bi, R) MapCAlgaug k (B, R) colimiMap Alg(n),aug k (Bi, R) / Map Alg(n),aug k (B, R) k k k and Alg(n),aug,♥ k → Alg(n),aug where the vertical maps are given by the forgetful functor CAlgaug . By [Lurc, Lem 6.3.3] the horizontal top map is a homotopy equivalence. Denote by CAlgaug,♥ the subcategories of discrete augmented E∞-algebras over k and of discrete augmented En-algebras over k respectively. As n ≥ 2 the forgetful functor induces an equivalence of categories CAlgaug,♥ . The E∞- algebras Bi and B being discrete, this implies that the vertical maps in the diagram are homotopy equivalences. Therefore the horizontal bottom map is a homotopy equivalence as well. We have proven that the map colimiSpf(Bi) → Spf(B) induces an equivalence on tangent spectra, which implies that it is an equivalence of formal En-moduli problems. The last statement follows from Lemma 2.5 by applying the equivalence t(n) : Moduli Remark 4.24. Let n ≥ 2 be an integer, let k be a field and let X : Alg(n),art → S be a functor. Applying Lemma 4.23 to the ring k[[t]] = limi k[t]/ti of formal power series in one variable over k, we obtain a natural homotopy equivalence: k → Alg(n),aug En (cid:39) Alg(n),aug,♥ k k k k . X(k[[t]]) (cid:39) lim X(k[t]/ti). i Therefore a point in X(k[[t]]) is essentially given by a formal family {ηi}i of points ηi ∈ X(k[t]/ti) together with for every i ≥ 1, a path pi(ηi+1) (cid:39) ηi in X(k[t]/ti), where pi : X(k[t]/ti+1) → X(k[t]/ti) is the map induced by the pro- jection k[t]/ti+1 → k[t]/ti, plus higher coherences. Remark 4.25. Let k be a field and let C be a compactly generated k-linear ∞- category. Remark 4.24 applied to the functor CatDef cC : Alg(2),art → S implies that a point in CatDef cC(k[[t]]) is essentially given by a formal family {Ci}i with Ci a compactly generated k[t]/ti-linear ∞-category together with for each i ≥ 1 a k[t]/ti-linear equivalence Ci+1 ⊗k[t]/ti+1 k[t]/ti (cid:39) Ci. Remark 4.26. Let k be a field and let C be a k-linear ∞-category. Consider the space of formal deformations CatDef∧(k[[t]]) = Map (Spf(k[[t]]),CatDef∧). k → Alg(2),aug E2 Applying the equivalence t(2) : Moduli Thm 3.5.1] there exists natural homotopy equivalences Moduli k , by Lemma 2.5 and [Lurb, E2 k k CatDef∧(k[[t]]) = Map (cid:39) Map (cid:39) Map E2 Moduli k Alg(2),aug k Alg(2) k 31 (Spf(k[[t]]),CatDef∧ C ) (D(2)(k[[t]]), k ⊕ ξ(C)) (D(2)(k[[t]]), ξ(C)), / / / where ξ(C) is the E2-algebra given by the center of C. The usual computation using the Koszul–Tate resolution gives D(2)(k[[t]]) (cid:39) k[u] where k[u] is the augmented E2-algebra over k associated to the free augmented graded commutative k-algebra on one generator in cohomological degree 2 via the forgetful functor. We deduce from this and from the universal property of the center ([Lurb, Rem 5.3.11]) that there exists a natural homotopy equivalence C (k[[t]]) (cid:39) Map CatDef∧ with the space of left k[u]-linear structures on the ∞-category C. (k[u], ξ(C)) (cid:39) LModLMod⊗ (PrL Alg(2) k )(cid:39) ×(PrL k[u] k k )(cid:39) {C} Here is our first main result. Theorem 4.27. Let k be a field and let C be a tamely compactly generated k-linear ∞-category which has a compact generator. Then the map θtC : CatDef cC(k[[t]]) → CatDef∧ C (k[[t]]) induced by θC is a homotopy equivalence. We will deduce Theorem 4.27 from the following intermediate statement. Proposition 4.28. Let k be a field and let C be tamely compactly generated k- linear ∞-category which has a compact generator E ∈ C. Let α ∈ CatDef∧ C (k[[t]]) be any point. Then there exists another compact generator Eu of C and a formal deformation Bt of its endomorphism algebra B = EndC(Eu) such that the connected component of Bt is sent to the connected component of α by the composite map π0AlgDef B(k[[t]]) Mt B−→ π0CatDef cC(k[[t]]) θtC−→ π0CatDef∧ C (k[[t]]). In particular, the map θtC : CatDef cC(k[[t]]) → CatDef∧ C (k[[t]]) is surjective on π0. Proof of Theorem 4.27. By [Lurb, Prop 5.3.21] the map θC : CatDef cC → CatDef∧ (−1)-truncated. This implies that the induced map (Spf(k[[t]]),CatDef cC) → Map (Spf(k[[t]]),CatDef∧ C ) θtC : Map E2 E2 C is Moduli k Moduli k is (−1)-truncated as well. By Proposition 4.28 we have that the map θtC is surjective on π0, which proves that it is a homotopy equivalence. Here is another consequence of Proposition 4.28. Theorem 4.29. Let k be a field and let C be tamely compactly generated k-linear ∞-category which has a compact generator E ∈ C. Let Ct = {Ci}i be a compactly generated formal deformation of C. Then there exists another compact generator Eu of C and a formal deformation Bt = {Bi}i of its endomorphism algebra B = EndC(Eu) such that for every i ≥ 1 there exists a k-linear equivalence Ci (cid:39) LModBi. In particular, each Ci is tamely compactly generated and admits a compact generator. 32 Proof. Let Ct = {Ci}i be a compactly generated formal deformation of C. Consider the vertex α = θtC(Ct) ∈ CatDef∧ C (k[[t]]). Let Eu ∈ C be the compact generator provided by Proposition 4.28 so that there exists a formal deformation Bt = {Bi}i of B = EndC(Eu) such that the connected component of Bt is sent to the connected component of α by the map θtC ◦ mt B. But θtC is injective on π0 by Theorem 4.27, so that for every i ≥ 1 there exists a k-linear equivalence LModBi (cid:39) Ci. As C is tamely compactly generated, the E1-algebra B is n0-connective for some integer n0. By Proposition 4.19 this implies that for every i ≥ 1 the E1-algebra Bi is n0-connective, and therefore that Ci is tamely compactly generated. Remark 4.30. In the situation of Theorem 4.29, the fact that each Ci is tamely compactly generated for every compactly generated formal deformation Ct of C is already a consequence of [Lurb, Lem 5.3.31] and does not require the assumption that C has a compact generator. Proof of Proposition 4.28. Let α ∈ CatDef∧ C (k[[t]]). Recall from Remark 4.26 the existence of a homotopy equivalence CatDef∧ (k[u], ξ(C)) with C (k[[t]]) (cid:39) Map the space of left k[u]-linear structures on C. We denote by Cu the left k[u]-linear structure corresponding to α. Let E be a compact generator of C. We exhibit another compact generator of C on which k[u] acts trivially. The left k[u]-action gives a map of k-modules k[u] → EndC(E) and consider the map φE : E → E[2] given by the image of u. Define Eu to be the homotopy fiber of φE in C, it is well defined up to a contractible space of choice. We claim that the object Eu is still a compact generator of C. That it is compact follows from the fact that the subcategory Cc ⊆ C spanned by compact objects is stable under finite colimits in C and that C is stable. To see that it is a generator, let F be an object of C such that MapC(Eu, F ) (cid:39) 0 and we must prove that F (cid:39) 0. For this we can suppose that F is compact. The fiber sequence Eu → E → E[2] induces a fiber sequence of k-module spectra Alg(2) k MapC(E, F )[−2] → MapC(E, F ) → MapC(Eu, F ) (cid:39) 0 which implies that MapC(E, F ) (cid:39) MapC(E, F )[−2]. Because C is tamely compactly generated, we deduce that MapC(E, F ) (cid:39) 0. But E is a generator of C and thus F (cid:39) 0. We adopt the notation LModLMod⊗ k[u] (PrL k )(cid:39) C = LModLMod⊗ k[u] (PrL k )(cid:39) ×(PrL k )(cid:39) {C} for the space of left k[u]-linear structures on C and similarly for left actions of k[u] on the pair (C, E). Consider the diagram 33 SimDef c (C,Eu)(k[[t]]) SimDef∧ (C,Eu)(k[[t]]) ∼ / / LModLMod⊗ k[u] E ((PrL k )Modk/)(cid:39) (C,Eu) AlgDef EndC(Eu)(k[[t]]) υ AlgDef∧ EndC(Eu)(k[[t]]) π M θ ∼ k[u] k[u] / LModLMod⊗ k[u] (PrL C (k[[t]]) / CatDef∧ We claim that the object (Cu, Eu) of (PrL (Modk,Cu) → FunLMod⊗ CatDef cC(k[[t]]) k )(cid:39) C where the notations E, M, and θ have been removed indices for simplicity. By construction, the two smaller squares in this diagram commutes up to natural ho- motopy. Moreover the outer square commutes up to natural homotopy by virtue of Remark 4.22. Therefore the right square is homotopy commutative as well, up to natural homotopy. The two horizontal maps in the right square are equivalences by [Lurb, Thm 5.3.16] and Proposition 4.7. Because C is supposed to be tamely com- pactly generated and E compact, the E1-algebra EndC(E) is n0-connective for some integer n0 which implies that the map υ is a homotopy equivalence by Proposition 4.19. k )Modk/ provides a lift of Cu through the map π. Unwinding the definitions, we must show that Eu admits a lift through (LModk[u],Cu) (cid:39) Cu induced by the functor FunLMod⊗ the monoidal functor LMod⊗ k itself induced by the augmentation map k[u] → k. By the right handed variant of [Lurd, Thm 4.8.4.1], there exists an (Modk,Cu) (cid:39) RModk(Cu). Therefore we equivalence of ∞-categories FunLMod⊗ must show that there exists a lift of Eu through the forgetful functor RModk(Cu) → Cu. By a variant of [Lurd, Cor 4.2.4.4] this forgetful functor admits a right adjoint Cu → RModk(Cu) given by the cofree right k-module functor which to an object F ∈ Cu assigns the right k-module given by the exponential F k of F by the left k[u]-module k. There is a fiber sequence k[u] → k[u][−2] → k in LModk[u], where the first map is multiplication by u. This implies the existence of a fiber sequence Ek → E[2] → E in Cu and thus of an equivalence Ek (cid:39) Eu in Cu, which shows that Ek is sent to Eu via the forgetful functor RModk(Cu) → Cu. We deduce that the pair (Cu, Eu) provides a well defined object in the ∞-category k )Modk/)(C,Eu) and therefore, by virtue of the commutative dia- LModLMod⊗ gram above, of a well defined object Bt in AlgDef EndC(Eu)(k[[t]]) corresponding to a formal deformation of the E1-algebra B = EndC(Eu). By commutativity of the diagram, we have the equality of connected components [θ ◦ M(Bt)] = [α] in π0CatDef∧ k[u] → Mod⊗ ((PrL k[u] C (k[[t]]). k[u] 34 / / / / / / C induces a natural transformation AutC → Aut!C. A Deformations as !-group actions As a consequence of Corollary 2.19, we give a description of the formal E2-moduli problem CatDef∧ C for C a k-linear ∞-category over a field k in terms of !-actions of the formal group ΩSpf(A) on C. Notation A.1. Let k be a field and let C be a k-linear ∞-category, we set the notations AutC = ΩCatDefC and Aut!C = ΩCatDef∧ C . The natural transformation θC : CatDefC → CatDef∧ Remark A.2. Let k be a field and let C be a compactly generated k-linear ∞- category. Because the embedding CatDef cC → CatDefC of compactly generated de- formations into all deformations is fully faithful, the induced map ΩCatDef cC → ΩCatDefC = AutC is a homotopy equivalence. Remark A.3. Let k be a field and let C be a k-linear ∞-category. By [Lurb, Cor 5.3.8] the functor CatDefC is a 2-proximate formal E2-moduli problem, which implies by Remark 2.10 that the functor AutC is a 1-proximate formal E2-moduli problem and that the induced map ΩAutC → ΩAut!C is an equivalence. By Corollary 2.19, we deduce that the induced map L2(AutC) → Aut!C is an equivalence. If in addition C is tamely compactly generated, then by [Lurb, Prop 5.3.21] we have that CatDef cC is 1-proximate and we deduce from Remark A.2 that AutC is a formal E2-moduli problem and that the natural map AutC → Aut!C is an equivalence. Remark A.4. Let k be a field and let C be a k-linear ∞-category. It can be shown that there exists a natural transformation AutC → ObjDef E2 idC which is an equivalence, where idC is seen as an object of the k-linear ∞-category Endk(C) of k-linear endofunctors of C. Having this, we see that for every artinian E2-algebra A the space AutC(A) parametrizes pairs (FA, u) with FA an object of the ∞-category End(C)⊗k ModA (cid:39) FunL A(C⊗k A,C⊗k A) (which is necessarily an A-linear autoequiv- alence of C⊗k A) and u : FA⊗Ak (cid:39) idC an equivalence in End(C). Moreover the space Aut!C(A) is the classifying space of right Ind-coherent A-modules in End(C) which deforms idC, or in other words of pairs (FA, u) with FA an object in End(C)⊗k Mod! and u : FA ⊗A k (cid:39) idC an equivalence. Corollary A.5. Let k be a field and C a k-linear ∞-category. Then the formal E2-moduli problem CatDef∧ C is given by the formula A CatDef∧ C (A) (cid:39) Map MongpE1 (Moduli E2 k ) (ΩSpf(A), Aut!C). If moreover C is tamely compactly generated, then the formal E2-moduli problem CatDef∧ C is given by the formula CatDef∧ C (A) (cid:39) Map MongpE1 (Moduli E2 k ) (ΩSpf(A), AutC). Proof. The first assertion follows from the definition of Aut!C and from Proposition 2.15. The second follows from the first combined with Remark A.3. 35 k k is given by the !-pullback f ! : Mod! k → (cid:100)Cat∞ given by ν(A) = RMod! Construction A.6. (Ind-coherent sheaves on formal E1-moduli problems as a sym- metric monoidal functor). Let k be a field. Consider the cocartesian fibration RMod!(Modk) → Algart of [Lurb, Const 3.4.11] which is classified by the functor ν : Algart A and functoriality for a morphism f : A → B in Algart PrL ⊆ (cid:100)Cat∞ of presentable ∞-categories with functors that are left adjoints. The B. The func- tor f ! preserves small colimits. It follows that ν factors through the subcategory k ⊆ Algk is stable under tensor product and therefore in- full subcategory Algart herits a symmetric monoidal structure Algart,⊗ k . A variant of [Lurd, Thm 4.8.5.16] allows us to promote the functor ν to a symmetric monoidal functor ν⊗ : Algart,⊗ k ,S)op can be pro- moted to a symmetric monoidal functor j⊗ : Algart,⊗ k ,S)op,× where k ,S) with its cartesian monoidal structure. Ap- we endow the ∞-category Fun∗(Algart plying the monoidal variant of [Lur09, Thm 5.1.5.6], we deduce that ν⊗ admits an essentially unique factorization as a composition Recall that the Yoneda embedding j : Algart k → Fun∗(Algart k → Fun∗(Algart k → PrL,⊗. A → Mod! from Alg⊗ k Algart,⊗ k j⊗−→ Fun∗(Algart k ,S)op,× (QCoh!,⊗ R ,(−)!) −→ PrL,⊗ such that the functor (QCoh! k → S, the ∞-category QCoh! PrL can be computed as the limit of the corresponding diagram in(cid:100)Cat∞). For any R, (−)!) preserves small limits (recall that a limit in R(X) is the k-linear ∞-category of functor X : CAlgart right Ind-coherent sheaves on X and the symmetric monoidal structure expresses the k → S, the natural functor QCoh!(X) ⊗k fact that for any two functors X, Y : Algart R(X × Y ) is an equivalence in PrL. Similarly we can define left R(Y ) → QCoh! QCoh! Ind-coherent sheaves. (PrL)op,⊗ (cid:39) PrR,⊗), we obtain a symmetric monoidal functor (through the symmetric monoidal equivalence Passing to right adjoints Fun∗(Algart k ,S)× (QCoh!,⊗ R ,(−)∗) −→ PrR,⊗ where the functoriality is now given by the pushforward. Consider the lax symmetric monoidal inclusion PrR,⊗ ⊆(cid:100)Cat PrR,⊗ ⊆(cid:100)Cat k ,S)× (QCoh!,⊗ R ,(−)∗) −→ Lemma A.7. Let k be a field. The restriction of the composition Fun∗(Algart × ∞ × ∞. E1,× to the symmetric monoidal full subcategory Moduli k problems factors through the lax symmetric monoidal inclusion PrL,⊗ ⊆ (cid:100)Cat spanned by formal E1-moduli × ∞ and is symmetric monoidal, providing a symmetric monoidal functor (QCoh!,⊗ R , (−)∗) : Moduli E1,× k → PrL,⊗. 36 R(X) → QCoh! Proof. Let f : X → Y be a map of E1-formal moduli problems over k and let f∗ : QCoh! R(Y ) be the induced functor. We wish to show that f∗ admits a right adjoint, or equivalently by the adjoint functor theorem [Lur09, Cor 5.5.2.9] that it preserves small colimits. Let A and B be the augmented E1-algebras k → E1 over k corresponding to X and Y respectively through the equivalence Moduli . Let ϕ : A → B be the map corresponding to f . By the Koszul duality Algaug R(X) (cid:39) LModA and for modules [Lurb, Prop 3.5.1] there exists equivalences QCoh! R(Y ) (cid:39) LModB and the induced diagram QCoh! k QCoh! R(X) (cid:111) LModA f∗ ϕ∗ QCoh! R(Y ) (cid:111) / LModB . k commutes up to homotopy, where ϕ∗ is the left adjoint to the forgetful functor ϕ∗ : LModB → LModA along ϕ. The functor α∗ preserves small colimits which implies that f∗ has the same property. The second assertion follows directly from the existence of the symmetric monoidal functor (QCoh!,⊗ )op → PrL,⊗ Notation A.8. The symmetric monoidal functor (QCoh!,⊗ Lemma A.7 will be simply denoted by QCoh!,⊗ E1,× Moduli k tive algebra objects QCoh! a symmetric monoidal functor R , (−)∗) provided by R . Consider the unit Spec(k) in as a commutative algebra object. We have an equivalence of commuta- R(Spec(k)) (cid:39) Modk in PrL,⊗. Therefore QCoh!,⊗ induces R , (−)!) : (Moduli E1,× k E1,× Moduli k (PrL)⊗ = PrL,⊗ into the symmetric monoidal ∞-category of presentable k-linear ∞-categories. k )× QCoh!,⊗−→ ModMod⊗ (cid:39) ModSpec(k)(Moduli E1 k k E1-formal moduli k, For Moduli Map problem X over any space (Spec(k), X) (cid:39) X(k) is contractible, which implies that the pro- E1 k )Spec(k)/ → Modulik is an equivalence of ∞-categories. Because E1 jection (Moduli Spec(k) and Modk are the units of the symmetric monoidal ∞-categories E1,× respectively, we have natural symmetric monoidal structures Moduli k k )Modk/)⊗ on the slice ∞-categories (see [Lurd, Rem k )Spec(k)/)× and ((PrL E1 ((Moduli 2.2.2.5]). Moreover the symmetric monoidal functor QCoh!,⊗ R induces a symmetric monoidal functor and PrL,⊗ the k k E1,× Moduli k (cid:39) ((Moduli E1 k )Spec(k)/)× QCoh!,⊗ R,∗−→ ((PrL k )Modk/)⊗. The data of an object of the ∞-category (PrL k )Modk/ is essentially given by a pair (C, E) with C a k-linear ∞-category and E an object of C. The functor QCoh! R,∗ sends a formal moduli problem X to the pair (QCoh! R(X), k) where QCoh!(X) is the k-linear ∞-category of Ind-coherent sheaves on X and k is the skyscraper sheaf with value k at the k-point of X. If A is the augmented E1-algebra of X, the sheaf k R(X) (cid:39) LModA. corresponds to the left A-module A through the equivalence QCoh! 37 / / / We have the following consequence of [Lurd, Thm 4.8.5.11]. Proposition A.9. Let k be a field. The symmetric monoidal functor QCoh!,⊗ R,∗ : Moduli E1,× k → ((PrL k )Modk/)⊗ admits a lax symmetric monoidal right adjoint which sends a pair (C, E) to the formal E1-moduli problem ObjDef∧ (C,E) of deformations of E in C. Proof. By [Lurd, Cor 7.3.2.7], it suffices to show that the underlying functor QCoh! of QCoh!,⊗ there exists a homotopy commutative diagram R,∗ R,∗ admits a right adjoint. By Koszul duality for modules [Lurb, Prop 3.5.1], E1 Moduli k QCoh! R,∗ (PrL k )Modk/ t(1) Algaug k Θ∗ where t(1) is the equivalence of [Lurb, Thm 3.0.4] and Θ∗ is the composition of the k → functor constructed in [Lurd, Const 4.8.3.24] with the forgetful functor Algaug Algk. Therefore Θ∗ is given on objects by Θ∗(A) = (LModA, A). By [Lurd, Thm 4.8.5.11], Θ∗ has a right adjoint which sends a pair (C, E) to the augmented E1- algebra k ⊕ EndC(E), with EndC(E) the E1-algebra of endomorphisms of E in C. This implies that QCoh! R,∗ has a right adjoint, which by [Lurb, Thm 5.2.8] recalled in §3.2, is given on objects by (C, E) (cid:55)→ ObjDef∧ (C,E). k → E1 Remark A.10. Let k be a field. Consider the restriction functor Rest : Moduli k → Moduli E2 E2 E1 Moduli k which admits a left adjoint Ext : Moduli k given by the left → Algart Kan extension along the forgetful functor Alg(2),art k . By composition we obtain from Lemma A.9 that the symmetric monoidal functor k )Modk/)⊗ E2,× k → ((PrL R,∗ : Moduli QCoh!,⊗ k admits as right adjoint the lax symmetric monoidal functor k )Modk/)⊗ → Moduli E2,∧ : ((PrL ObjDef E2,× k E2,∧ (C,E) is the restriction of the formal E1-moduli problem ObjDef∧ where ObjDef on E2-algebras. Passing to algebra objects, we obtain a functor k ) → Alg(PrL E2 k ) E2,∧ sending a pair (D⊗, 1) where with right adjoint given by the functor ObjDef D⊗ is a monoidal k-linear ∞-category and 1 its unit object to the formal moduli problem ObjDef E2,∧ (D,1) of deformations of 1 in D. R,∗) : MonE1(Moduli Alg(QCoh!,⊗ (C,E) 38 / / 5 5 Corollary A.11. Let k be a field and let C be a k-linear ∞-category. Then the formal E2-moduli problem CatDef∧ C (A) (cid:39) Fun⊗ C (A) is naturally homotopy equivalent to the space CatDef∧ In other words CatDef∧ C is given by the formula k (QCoh!,⊗ R (ΩSpf(A)), End(C)⊗)(cid:39). LModQCoh!,⊗ R (ΩSpf(A))(PrL k )(cid:39) ×(PrL k )(cid:39) {C} of left actions of the monoidal ∞-category QCoh!,⊗ Proof. By Corollary A.5, for A an artinian E2-algebra, there is a natural homotopy equivalence R (ΩSpf(A)) on C. Besides, there exists an equivalence of functors AutC (cid:39) ObjDef natural equivalence Aut!C (cid:39) ObjDef we obtain a natural homotopy equivalence (ΩSpf(A), Aut!C). E2 idC which induces a E2,∧ idC . Applying the adjunction of Remark A.10, MongpE1 (Moduli E2 k ) CatDef∧ C (A) (cid:39) Map Map MongpE1 (Moduli E2 k ) (ΩSpf(A), Aut!C) (cid:39) Fun⊗ k (QCoh!,⊗ R (ΩSpf(A)), End(C)⊗)(cid:39) which proves our claim. References [Coh13] L. Cohn, Differential graded categories are k-linear stable infinity categories, arXiv preprint arXiv:1308.2587 (2013). [Fra13] J. Francis, The tangent complex and Hochschild cohomology of En-rings, Composi- tio Math 149 (2013), 430–480. [KL09] B. Keller and W. Lowen, On hochschild cohomology and morita deformations, In- ternational Mathematics Research Notices 2009 (2009), no. 17, 3221–3235. [KS05] M. Kontsevich and Y. Soibelman, Deformation theory, vol. 1, Unpublished book draft. Available at www. math. ksu. edu/soibel/Book-vol1. ps (2005). [LB15] W. Lowen and M. Van den Bergh, The curvature problem for formal and infinites- imal deformations, arXiv preprint arXiv:1505.03698 (2015). [Low08] W. Lowen, Hochschild cohomology, the characteristic morphism and derived defor- mations, Compos. Math 144 (2008), no. 6, 1557–1580. [Lura] J. Lurie, Derived Algebraic Geometry IX: Closed Immersions http://www. math. harvard. edu/ lurie/papers, DAG-IX. pdf. [Lurb] , Derived Algebraic Geometry X: Formal Moduli Problems http://www. math. harvard. edu/ lurie/papers, DAG-X. pdf. [Lurc] , Derived Algebraic Geometry XII: Proper Morphisms, Completions and the Grothendieck Existence Theorem http://www. math. harvard. edu/ lurie/papers, DAG-XII. pdf. 39 [Lurd] , Higher algebra. 2014, Book, available at http://www. math. harvard. edu/ lurie. [Lur09] , Higher topos theory, no. 170, Princeton University Press, 2009. [Pet12] F. Petit, DG affinity of DQ-modules, International Mathematics Research Notices 2012 (2012), no. 6, 1414–1438. [Pre12] A. Preygel, Thom-Sebastiani and duality for matrix factorizations, and results on the higher structures of the Hochschild invariants, Ph.D. thesis, Massachusetts In- stitute of Technology, 2012. [Rob14] M. Robalo, Motivic Homotopy Theory of noncommutative Spaces, Ph.D. thesis, Universit´e Montpellier 2, 2014. Anthony Blanc Max Planck Institute for Mathematics, 53111 Bonn, Deutschland Email: [email protected] Ludmil Katzarkov Universitat Wien, Fakultat fur Mathematik, 1090 Wien, Osterreich Laboratory of Mirror Symmetry NRU HSE, Moscow, Russia Email: [email protected] Pranav Pandit Universitat Wien, Fakultat fur Mathematik, 1090 Wien, Osterreich Email: [email protected] 40
1612.09439	2	1612	2017-06-18T06:05:08	Hodge Numbers from Picard-Fuchs Equations	[ "math.AG" ]	Given a variation of Hodge structure over $\mathbb{P}^1$ with Hodge numbers $(1,1,\dots,1)$, we show how to compute the degrees of the Deligne extension of its Hodge bundles, following Eskin-Kontsevich-M\"oller-Zorich, by using the local exponents of the corresponding Picard-Fuchs equation. This allows us to compute the Hodge numbers of Zucker's Hodge structure on the corresponding parabolic cohomology groups. We also apply this to families of elliptic curves, K3 surfaces and Calabi-Yau threefolds.	math.AG	math	Symmetry, Integrability and Geometry: Methods and Applications SIGMA 13 (2017), 045, 23 pages Hodge Numbers from Picard–Fuchs Equations Charles F. DORAN †1 †1 , Andrew HARDER †2 and Alan THOMPSON †3†4 Department of Mathematical and Statistical Sciences, 632 CAB, University of Alberta, Edmonton, AB, T6G 2G1, Canada E-mail: [email protected] 7 1 0 2 n u J 8 1 ] . G A h t a m [ 2 v 9 3 4 9 0 . 2 1 6 1 : v i X r a †2 †3 †4 Department of Mathematics, University of Miami, 1365 Memorial Drive, Ungar 515, Coral Gables, FL, 33146, USA E-mail: [email protected] Mathematics Institute, Zeeman Building, University of Warwick, Coventry, CV4 7AL, UK E-mail: [email protected] DPMMS, Centre for Mathematical Sciences, University of Cambridge, Wilberforce Road, Cambridge, CB3 0WB, UK Received January 20, 2017, in final form June 12, 2017; Published online June 18, 2017 https://doi.org/10.3842/SIGMA.2017.045 Abstract. Given a variation of Hodge structure over P1 with Hodge numbers (1, 1, . . . , 1), we show how to compute the degrees of the Deligne extension of its Hodge bundles, following Eskin–Kontsevich–Moller–Zorich, by using the local exponents of the corresponding Picard– Fuchs equation. This allows us to compute the Hodge numbers of Zucker's Hodge structure on the corresponding parabolic cohomology groups. We also apply this to families of elliptic curves, K3 surfaces and Calabi–Yau threefolds. Key words: variation of Hodge structures; Calabi–Yau manifolds 2010 Mathematics Subject Classification: 14D07; 14D05; 14J32 1 Introduction The goal of this paper is to compute the Hodge numbers of the parabolic cohomology groups of a variation of Hodge structure (VHS) in the case where we know the Picard–Fuchs equation. In more detail, let us assume that C is a smooth quasiprojective curve which bears an R-VHS of weight k. If V is the underlying local system and j : C (cid:44)→ C is the embedding of C into its smooth completion, then Zucker [31] showed that H 1(C, j∗V) (which we will call the parabolic cohomology of V) bears a pure Hodge structure of weight k + 1. Moreover, Zucker also showed that this Hodge structure is closely related to the filtrands of the Leray spectral sequence when our VHS comes from a family of manifolds, so one can use the Hodge structure on the parabolic cohomology groups to understand the Hodge structure of fibrations and vice versa. More recently, following work of Morrison and Walcher [26, 30], interest has arisen in deter- mining when the Hodge structure on the parabolic cohomology of specific families of Calabi–Yau threefolds admits rational (2, 2) classes, since such classes correspond to possible normal func- tions. Morrison and Walcher [26] gave an exciting interpretation of such normal functions of families of Calabi–Yau threefolds in terms of D-branes. Following this, del Angel, Muller-Stach, van Straten, and Zuo [3] sought further examples of such normal functions, by studying Hodge structures on the parabolic cohomology of families of This paper is a contribution to the Special Issue on Modular Forms and String Theory in honor of Noriko Yui. The full collection is available at http://www.emis.de/journals/SIGMA/modular-forms.html 2 C.F. Doran, A. Harder and A. Thompson Calabi–Yau threefolds whose underlying VHS's are pull-backs of VHS's on the thrice-punctured sphere with b3 = 4; such VHS's with b3 = 4 were classified by Doran and Morgan [12], who found 14 cases. They reduced these computations to the computation of the degree of the quasi- canonical extension of the Hodge bundles of these VHS's. However, it appears as if the authors of [3] were not able to compute these degrees in great generality, thus many entries in their tables are left blank. Further progress was made by Holborn and Muller-Stach [23], however it appears that they were still not able to complete the table from [3]. Some related computations of degrees of quasi-canonical extensions of Hodge bundles have also been carried out by Green, Griffiths, and Kerr [17]. Recently, a preprint of Eskin, Kontsevich, Moller, and Zorich [13, Section 6] suggested a way to compute the degrees of Hodge bundles of R-VHS's of (1, 1, . . . , 1)-type, when the Picard– Fuchs operators controlling these R-VHS's are known. The goal of this paper is to present the technique of [13] in greater generality1 and apply it to an array of situations. In particular, we offer a completion of the tables of [3]. The application most germane to the previous work of the authors is to the study of threefolds fibred by lattice polarized K3 surfaces. In [7, 8, 9, 10], we constructed Calabi–Yau threefolds by first choosing convenient families of lattice polarized K3 surfaces X → P1, then performing base change along suitable maps g : P1 → P1. One of the main results of this work [10] is a classification of all Calabi–Yau threefolds fibred by Mn-polarized K3 surfaces, where Mn denotes the rank 19 lattice Mn := E8 ⊕ E8 ⊕ H ⊕ (cid:104)−2n(cid:105). In order to restrict the number of cases that we needed to check geometrically, the results and techniques of this paper helped immensely. In more generality, our results allow us to compute the Hodge numbers of the parabolic cohomology H 1(P1, j∗V), where X → P1 is any smooth projective threefold fibred by K3 surfaces over P1 with generically Picard rank 19 fibres, and V is the corresponding local system of cohomology over the complement of the critical values of the fibration. By the Leray spectral sequence [31, Section 15], this is a direct summand of the Hodge structure on H 3(X , Q), and its complement has no (3, 0) or (0, 3) part, so in fact this computes h3,0(X ) and places a lower bound on h2,1(X ). Finally, we note that in the case of an elliptic fibration, the degree of the Hodge bundle can be computed geometrically by putting the fibration into Weierstrass normal form Y 2 = X 3 + g2X + g3 and computing the degrees of g2 and g3. In the case of K3 surface fibrations (and to an even greater extent for Calabi–Yau threefold fibrations) we do not have such a normal form available, because there are many families of K3 surfaces polarized by rank 19 lattices over P1. One can construct normal forms corresponding to each polarizing lattice, as we have done in [9, 10] for some specifically chosen lattices, but this approach quickly becomes intractable: given an arbitrary rank 19 lattice L, it is very difficult to find explicit representatives for all L-polarized K3 surfaces (see [5] for some discussion of this problem). The work in this paper allows us to bypass such considerations in the presence of a known Picard–Fuchs equation. In a sense this is antipodal to the work of Fujino [16], who computes the Hodge bundles of K3 surface fibrations with only unipotent monodromy. 1.1 Structure of this paper Section 2 begins with a brief discussion of the necessary background in Hodge theory and the theory of ordinary differential equations, following [13], which we require in order to state the 1As stated, the results of [13] apply only to VHS's of (1, 1, 1, 1)-type, though it is clear, even in their exposition, that their approach is more general. Hodge Numbers from Picard–Fuchs Equations 3 main results in Section 2.4. Section 2 concludes with a discussion of how these results are affected by base change. The remainder of the paper presents some applications of this theory. Firstly, in Section 3, we apply the results of Section 2 to the case of elliptic fibrations. Here we show how to compute the Hodge numbers of an elliptic surface over P1 directly from its Picard–Fuchs equation. In Section 4, we move on to analyze (1, 1, 1)-type R-VHS's over P1, which arise in the context of K3 fibrations. We again compute their Hodge numbers using the corresponding Picard–Fuchs equations, proving a Hodge bundle formula for such fibrations. In particular, we can use this to place constraints on the possible K3 fibrations that can arise on a Calabi–Yau threefold, which allows us to recover an approximate form of a result from [7]. Finally, in Section 5, we will consider the case of Calabi–Yau threefold fibrations over P1. Here we achieve our main goal and complete the computations of [3] and [23]. Finally, we conclude with a result that allows us to constrain the possible Calabi–Yau fourfolds which may admit fibrations by the quintic mirror Calabi–Yau threefold. The results we use to do this apply in greater generality; we expect similar results to hold for any Calabi–Yau threefolds X with (1, 1, 1, 1)-type Hodge structure on H 3(X, Q). 2 Background and important results In this section we will develop the necessary background material and state a number of results that are necessary for the subsequent computations in Sections 3, 4 and 5. 2.1 Variations of Hodge structure Let us begin with a real variation of polarizable Hodge structure over a quasi-projective curve C. We denote this R-VHS by (V, F•,∇), where V is the underlying real local system, F• is the Hodge filtration on V := V ⊗ OC and ∇ is the Gauss–Manin connection. We require that ∇ satisfy Griffiths transversality [18, 19, 22]; in other words, ∇(Fi) should be contained in Fi+1 ⊗ ΩC. Following Deligne [4], we can canonically extend the bundle V to a bundle on the smooth completion C of C, as follows. Near a point p ∈ ∆ = C \ C, we may choose a chart so that p is the center of a disc D. Let V0 be a fibre of V near p and suppose that monodromy around p acts on V0 by a quasi-unipotent transformation T . Define subspaces Wα of V0 by Wα :=(cid:8)v ∈ V0 : (T − ζα)kv = 0(cid:9), but finitely many values of α ∈ [0, 1), and we have a direct sum decomposition V0 =(cid:76) where ζα = e2πiα and α is chosen to be in the interval [0, 1). These vector spaces are zero for all α Wα. The vector spaces Wα over the different points of the punctured disc D\{p} define a sub-bundle Wα of V. Let Tα denote the action of monodromy on Wα; note that ζ−1 α Tα is unipotent by construction. Define Nα := −1 2πi log(Tα), where the branch of the logarithm is chosen so that the unique eigenvalue of Nα is in the interval [0, 1). Then for any section v of e∗Wα, where e(z) := exp(2πiz) denotes the complex exponential, we may define v(z) = exp(2πiz(Nα − α))v(z), where z ∈ H is an element of the upper half-plane. Since v(z + 1) = T (v(z)), we see that v(z) is invariant under the translation z (cid:55)→ z + 1, so v defines a holomorphic section of V on the 4 C.F. Doran, A. Harder and A. Thompson punctured disc D∗. Deligne's canonical extension of V is the bundle on D which is the OD-module spanned by all such vectors over all α ∈ [0, 1). This is denoted V. The Hodge filtration F• extends to a Hodge filtration on V, which we denote by F • • will be denoted Ep,(cid:96)−p = Fp /Fp−1 . The , where (cid:96) is the length of the Hodge Following [13, Section 2], there is a parabolic filtration of the fibre Vp of V over each point α≥β Vα, where Vα is the subspace of Vp spanned by local sections of V coming from sections of e∗Wα. The parabolic degree of V is given by degpar V := deg V + α dim Vα. (2.1) graded pieces of F filtration. p ∈ ∆, given by V ≥β :=(cid:76) (cid:88) p∈∆ α∈[0,1) This parabolic filtration on the fibres of V over points in ∆ extends to a parabolic filtration . The connection ∇ extends to on the fibres of Ep,(cid:96)−p; we denote its graded pieces by Ep,(cid:96)−p a connection on V with logarithmic poles along ∆ or, in other words, α ∇ : V −→ V ⊗ ΩC(∆). This connection is horizontal with respect to the Hodge filtration F of bundles θi−1 : E(cid:96)−i,i −→ E(cid:96)−i−1,i+1 ⊗ ΩC(∆). This map also respects the local filtration at each point p ∈ ∆, i.e., • , so we obtain morphisms ∇ : V ≥β p −→ V ≥β p ⊗ ΩC(∆). The following useful result appears as part of the proof of [13, Theorem 6.1]. Lemma 2.1. Assume that C = P1. Then degpar Ep,q = − degpar Eq,p. The following two examples will be useful in later portions of this paper. Example 2.2 (elliptic curves). Let us take a degeneration of elliptic curves over the unit disc in C, with central fibre of Kodaira type IV∗. In this case, one can check (see [25, Table VI.2.1]) that the local monodromy action on the first integral cohomology of a general fibre has matrix (cid:18)−1 −1 (cid:19) 1 0 . This matrix has eigenvalues equal to e2πi/3 and e4πi/3. Therefore, the spaces V1/3 and V2/3 are both 1-dimensional. We can ask how these spaces interact with the canonical extension of the Hodge filtration. We know that the map ∇ : F0 → F1 ⊗ ΩD(0) respects the filtration C2 = V ≥1/3 ⊃ V ≥2/3 = C and thus the image of ∇ in F1⊗ ΩD(0) must be V ≥2/3 = V2/3, since ∇ is not zero. Therefore E0,1 1/3 = E1,0 and E1,0 2/3 = 0. 1/3 = 0. Similar arguments show that E1,0 2/3 = E0,1 and E0,1 Using Kodaira's classification of germs of one parameter degenerations of elliptic curves, one can compute parabolic filtrations on the Hodge bundles corresponding to all non-unipotent degenerations of elliptic curves in the same way. The result is given in Table 1, which lists the values of α such that E1,0 α = E0,1 for each type of singular fibre. α = E1,0 and E0,1 Using this, the parabolic degrees of E1,0 and E0,1 can be determined easily once one knows their degrees. Furthermore, for an elliptic fibration over P1, we may combine this with equation (2.1) and Lemma 2.1 to show that − deg E1,0 − deg E0,1 is equal to the number of fibres around which monodromy is not unipotent. This is precisely [23, Remark 2.4]. Hodge Numbers from Picard–Fuchs Equations 5 0 In I∗ Fibre type E1,0 E0,1 0 1/2 5/6 3/4 2/3 II or II∗ III or III∗ IV or IV∗ 1/2 1/6 1/4 1/3 n Table 1. Values of α such that Ep,q α = Ep,q for degenerations of elliptic curves. Monodromy matrix E2,0 E1,1 E0,2 1/2 0 3/4 3/4 1/2 5/6 2/3 0 Tun T 2 un Ti −Ti T 2 i Tω T 2 ω Tnod 1/2 0 1/4 1/4 1/2 1/6 1/3 0 1/2 0 1/2 0 0 1/2 0 1/2 Table 2. Values of α such that Ep,q α = Ep,q for (1, 1, 1)-VHS's. Example 2.3 ((1, 1, 1) variations of Hodge structure). Let us take a degeneration X of K3 surfaces over the complex unit disc D, so that the restriction of X to D∗ := D \ {0} is an M - polarized family of K3 surfaces, in the sense of [7, Definition 2.1], for some rank 19 lattice M . The transcendental lattices of the general fibres give rise to a VHS of type (1, 1, 1) over D∗. One can generally describe all possible monodromy matrices that can underlie such a (1, 1, 1)- VHS. Up to sign, they are just symmetric squares of parabolic and elliptic elements of SL2(R); this comes from the fact that O(2, 1) is, up to sign, a symmetric square of the group SL2(R). By [23, Lemma 3.1], the monodromy of X around 0 ∈ D can be written in Jordan canonical form as either 0 λ 1 0 0 λ  λ 1 0 −1 −e2πi/3 Tun = Tω = 0 0 or 0 0 λ1  ,  0 λ2 0 0 0 λ3 3(cid:80) i=1 Ti = 1 0 0 −1 1 0 −1 0 0  , −e4πi/3 0 0 0 −1 for some λ, λi ∈ C roots of unity with λi ∈ Z. Cases that are relevant to our work [10] are  , 0 0 −1 0 −√−1 √−1 −1 0 0 0 0 0  , 0 0 1 0 0 1 Tnod = and their powers. As in the previous example, we can compute the filtration on the degenerate Hodge structure from these matrices. The result is given in Table 2, which lists the values of α such that α = E2,0, E1,1 E2,0 For a fibration of K3 surfaces over P1, we may combine this with equation (2.1) and Lemma 2.1 to show that − deg E2,0 − deg E0,2 computes the number of fibres around which we have mono- dromy matrix with Jordan normal form Tun, Ti, −Ti, T 2 ω. If we adjust the definition α = E0,2 for each type of VHS. α = E1,1, and E0,2 i , Tω or T 2 6 C.F. Doran, A. Harder and A. Thompson of II in [23, Section 3] to be the set of points in ∆ whose monodromy matrices have at least two eigenvalues not equal to 1, then [23, Remark 3.4] holds. 2.2 Picard–Fuchs equation of a variation of Hodge structure Now let us take a look at how the Picard–Fuchs equation of a variation of Hodge structure arises; more details for everything in this section and the next may be found in [13, Section 6.4]. We start with the same basic setup as before: a variation of Hodge structure (V, F•,∇) over a quasi-projective curve C with its canonical extension to C. For the sake of simplicity, we will assume that C = P1. If we now choose a global section2 σ of the Serre twist F(cid:96) (i), where (cid:96) is the length of the Hodge filtration and i is some integer, then we can build a differential equation associated to σ. Taking a polarization (cid:104)−,−(cid:105) on V, we can assign to σ a set of multivalued functions sv(t) = (cid:104)σ, v(cid:105), where v is some flat local section on V (or any flat local section v of V := V ⊗ OC). The functions sv(t) give multivalued meromorphic functions on P1, with monodromy occurring only around points in ∆ = C \ C. One can then produce a global differential equation Lσ, of rank at most the rank of V on C, whose solution set is the set of functions sv(t). The differential operator Lσ is a Fuchsian ODE (which will be discussed in the following section), called the Picard–Fuchs equation. The equation Lσ is defined as follows. For an appropriate choice of affine coordinate t on C = P1, we may assume that ∞ ∈ ∆, and therefore that the vector field d/dt is a global vector field on C = P1 \ ∆. If α and β are sections of V, then we have (cid:104)α, β(cid:105) = (cid:104)∇d/dtα, β(cid:105) ± (cid:104)α,∇d/dtβ(cid:105), d dt where ∇d/dtα is defined to be ∇(α) ∈ V ⊗ ΩC paired with d/dt. Therefore, dsv(t) dt = (cid:104)∇d/dtσ, v(cid:105), since v is chosen to be a flat local section of V. Since ∇i d/dt(σ) all live in V, if r = rank V, then σ,∇d/dtσ, . . . ,∇r d/dtσ must be linearly depen- dent over the field of meromorphic functions on C. Therefore, we can write an equation d/dtσ + f1(t)∇n−1 ∇n d/dtσ + ··· + fn(t)σ = 0 for some minimal n ≤ r. Therefore, for any flat local section v of V, we obtain the equation dnsv(t) dtn + f1(t) dn−1sv(t) dtn−1 + ··· + fn(t)sv(t) = 0, by linearity of the operator (cid:104)−, v(cid:105). In other words, sv(t) is annihilated by the differential operator Lσ = dn dtn + f1(t) dn−1 dtn−1 + ··· + fn(t). On the other hand, to a rank r ordinary differential equation L with regular singularities, we can assign to L its solution sheaf Sol(L), which is a subsheaf of OC spanned over C by the local solutions of L. The solution sheaf of Lσ is precisely the subsheaf of OC spanned locally by the functions sv(t). 2More generally, if such a section does not exist, one may instead construct a Picard–Fuchs D-module; however this will not be relevant to our work. Hodge Numbers from Picard–Fuchs Equations 7 2.3 Fuchsian ODE's Let L be a linear ordinary differential equation in one variable, written as where the fi are meromorphic functions in t. This equation can be rewritten in terms of the operator δt = t d dt as δn t + gi(t)δn−i t (2.2) (cid:18) d (cid:19)n−i , dn dtn + fi(t) dt n(cid:88) i=1 n(cid:88) (cid:19)n (cid:18) d i dt for some meromorphic functions g1, . . . , gn. Indeed, one can check that tn = δt(δt − 1)··· (δt − n − 1), so it is easy to translate between the two expressions for L. A point a ∈ P1 is called nonsingular if fi(t) is holomorphic at a, and is a regular singular point if (t − a)ifi(t) is holomorphic at a, for all i ∈ {1, . . . , n}. Note that 0 is a regular singular point if and only if the coefficients gi(t) in equation (2.2) are all holomorphic at 0. A linear differential equation of one variable with only regular singularities at all points in P1 is called a Fuchsian ODE. Regular singular points can be divided into apparent singularities, around which monodromy acts as the identity on the solution sheaf, and actual singularities, where it does not. Let L be a Fuchsian ODE with a regular singular point at 0 ∈ P1. The polynomial n(cid:88) i=1 T n + gi(0)T n−i 1, . . . , µa 1 ≤ µa 2 ≤ ··· ≤ µa n; throughout the paper we will assume that the µa is called the indicial equation of L at 0, and its roots are called the characteristic exponents of L at 0. For an arbitrary point a ∈ P1, one may compute the indicial equation and characteristic exponents of L at a by making the variable change t (cid:55)→ t − a. We will denote the characteristic exponents of L at a by µa i are ordered by magnitude, so that µa n. Often, the data of the regular singular points ai and their characteristic exponents µaj i are arranged into a matrix known as the Riemann scheme of L, which looks like ··· ak ··· µak ... . . . ··· µak  .  1 a2 µa2 1 ... µa2 n a1 µa1 1 ... µa1 n n Information about the monodromy of the solution sheaf of L can be read off from the Riemann scheme. For instance, if a is a regular singular point whose characteristic exponents are all integers, then monodromy around a is unipotent, i.e., all of its eigenvalues are 1. As a partial converse, if a is an apparent singularity, then the characteristic exponents at a are all integers. Finally, note that it also makes sense to compute the characteristic exponents at a nonsingular n) = (j, j + 1, . . . , j + n − 1), for some point a of L. If one does so, one obtains (µa integer j. 2, . . . , µa 1, µa 8 C.F. Doran, A. Harder and A. Thompson Remark 2.4. It is often natural to scale the solutions to a differential equation by a meromor- phic function h(t) on P1 to obtain a twisted differential equation, denoted hL. This operation acts on the characteristic exponents of L by an overall scaling hµa i +orda(h), where orda(h) is the order of h at a. Importantly, the differences between the characteristic exponents are not affected by the process of twisting. In the main results of this paper, we will only need to know the differences between the characteristic exponents, so knowing Lσ up to twist is enough. i = µa This is important in applications, since one often only knows the Picard–Fuchs equation of a variation of Hodge structure up to twist: for instance, this is all that the Griffiths–Dwork method [20, 21] for computing Picard–Fuchs equations of hypersurfaces guarantees. In Theo- rem 2.7, only the difference in characteristic exponents is relevant; therefore, our Hodge bundle computations can be done with hL for any meromorphic function h. Remark 2.5. One can also twist by multivalued functions h on P1. A particular choice of (possibly) multivalued function h can be made so that hL has fn−1(t) = 0. This is called the projective normal form of L. The process of twisting by a multivalued function does affect the differences of characteristic exponents in a significant way. For this reason, we will avoid using the projective normal form in this paper. Remark 2.6. The characteristic exponents of L at a point a allow us to compute local solutions for L near a. We remark that in general, the solutions of Lf = 0 around the point a have the form tµa i Pi(t) for some holomorphic functions Pi(t), modulo terms with logarithmic singularities i 's are the same, then there is an ordered basis of local solutions {si(t)} at a. For instance, if all µa i−1(cid:88) around a given inductively by sj(t) log t si(t) = tµa i Pi(t) + j! j=1 , where the Pi are holomorphic functions which do not vanish at a. If none of the differences between characteristic exponents are integers, then there are no terms with logarithmic singularities and the local solutions around a all have the form tµa i Pi(t), for some holomorphic functions Pi(t). In this case, if L is the Picard–Fuchs equation of a variation of Hodge structure (V, F•,∇), then Deligne's canonical extension V of V is spanned by the set of i (cid:99)Pi(t)} and the subspaces spanned by t(cid:98)µa local sections {t(cid:98)µa i (cid:99)Pi(t) are the parabolic filtrands Vα, i (cid:99). A more in-depth analysis of the Frobenius method shows that this is true i −(cid:98)µa where α := µa in general. 2.4 Important results Here we state some important results which, in combination, will allow us to compute Hodge numbers of a variation of Hodge structure of (1, 1, . . . , 1)-type if we know the Picard–Fuchs equation Lσ. The first is a result of Eskin, Kontsevich, Moller, and Zorich [13], which says that the characteristic exponents of the Picard–Fuchs equation Lσ corresponding to a variation of Hodge structure of (1, 1, . . . , 1)-type can be used to compute the Kodaira–Spencer maps on the bundles Ep,q. Theorem 2.7 ([13, Lemma 6.3]). Let (V, F•,∇) be a polarized R-VHS of (1, 1, . . . , 1)-type over a Zariski open subset P1 \ ∆, let F be its quasi-canonical extension to P1, let Ep,q be the graded pieces of F , and let Lσ be the Picard–Fuchs differential equation associated to a meromorphic section of σ of E(cid:96),0. Then for i < (cid:96)/2, the Kodaira–Spencer map θi : E(cid:96)−i,i → E(cid:96)−1−i,1+i⊗ΩP1(∆) is described as follows. • • 1. If p is a nonsingular point of Lσ, then θi is a local isomorphism at p. Hodge Numbers from Picard–Fuchs Equations 9 2. If p is an apparent singularity of Lσ, then θi has cokernel of length µp 3. If p is an actual singularity of Lσ, then θi has cokernel of length (cid:98)µp i+2 − µp i+2(cid:99) − (cid:98)µp i+1 − 1. i+1(cid:99). Remark 2.8. We record the observation that Eskin, Kontsevich, Moller, and Zorich use Theo- rem 2.7 to compute the parabolic degrees of the Hodge bundles of hypergeometric variations of Hodge structure. Specifically, they are interested in the 14 VHS's of (1, 1, 1, 1)-type on P1 found by Doran and Morgan [12]. However, it seems as if the authors of [13] were unaware of the fact that this result follows easily from the data in [23], in which the degrees of E3,0 and E2,1 are computed for all fourteen (1, 1, 1, 1)-type variations of Hodge structure of hypergeometric type. We will see later that Theorem 2.7 can be used to perform computations that neither the authors of [3] nor [23] were able to perform. This result is useful because, if we know the parabolic filtration of each bundle Ep,q and the Picard–Fuchs equation Lσ, then knowing the degrees of all maps θi allows us to easily compute the degrees of Ep,q when each Ep,q is a line bundle (i.e. when Lσ underlies a variation of Hodge structure of weight d and type (1, 1, . . . , 1)). Next we will state several results due to del Angel, Muller-Stach, van Straten and Zuo, and Hollborn and Muller-Stach, which appear in [3, 23]. These results say that, in the situations that interest us, we can compute Hodge numbers of H 1(P1, j∗V) (where j : C (cid:44)→ C = P1 denotes the inclusion) using the degrees of Ep,q and the local monodromy data of V. According to Theorem 2.7, all of this information can be obtained from Lσ. Theorem 2.9 ([3, 23]). Let V be a local system on P1 \ ∆ which supports a variation of Hodge structure of weight n and (1, 1, . . . , 1)-type. Then the Hodge number h0,n+1 of H 1(P1, j∗V) is given by h0,n+1 = h1(cid:0)P1, E0,n(cid:1) = h0(cid:0)P1, OP1 (cid:0)−2 − deg E0,n(cid:1)(cid:1). In the case where n = 1 or 2 in the above theorem, then we can deduce the rest of the Hodge numbers of the parabolic cohomology H 1(P1, j∗V) by Hodge symmetry and the following result, which generalizes the classical Riemann–Hurwitz formula (presumably this result is classically known, but the only reference that we know is [3]). Proposition 2.10 ([3, Proposition 3.6]). Let V be a local system on a quasi-projective curve C and let j : C (cid:44)→ C be the smooth completion. Let γq be the local monodromy matrix around the point q in ∆ := C \ C, acting on Vp for some basepoint p in C, and define R(q) := dim Vp − dim Vγq p denotes the part of Vp fixed under γq. Then p , where Vγq h1(cid:0)C, j∗V(cid:1) = R(q) +(cid:0)2g(cid:0)C(cid:1) − 2(cid:1) rank V. (cid:88) q∈∆ Finally, if V is a VHS of (1, 1, 1, 1)-type, then Hollborn and Muller-Stach show how to compute the Hodge number h1,3 of the parabolic cohomology H 1(P1, j∗V) from the degrees of its Hodge bundles. Theorem 2.11 ([23, Theorem 4.3]). If V supports a variation of Hodge structure of type (1, 1, 1, 1) over a Zariski open subset j : C (cid:44)→ P1, then the Hodge number h1,3 of H 1(P1, j∗V) is given by h1,3 = −2 + b − a + II + III + IV, 10 C.F. Doran, A. Harder and A. Thompson where a = deg(E3,0) and b = deg(E2,1). The sets II and III are the subsets of ∆ = P1 \ C whose local monodromy matrices have Jordan normal forms 1 1 0 0  , 0 1 1 0 0 0 1 1 0 0 0 1 1 1 0 0  0 1 0 0 0 0 1 1 0 0 0 1 and respectively. The set IV is the subset of ∆ for which the monodromy matrix is strictly quasi- unipotent. 2.5 Base change and characteristic exponents The following instructive computation will be useful to us later. We want to understand how base change of a Fuchsian ODE affects its Riemann scheme. More precisely, let us take a Fuchsian ODE L and a map g : P1 → P1. Pulling-back the solution sheaf of L by g yields a local sys- tem g∗Sol(L) and, in particular, another ODE g∗L, of the same rank as L, whose solution sheaf is g∗Sol(L). Our goal in this section is to analyze the local exponents of g∗L. Away from the ramification points of g, the differential equation g∗L will have the same behaviour as L at the corresponding point. So we can restrict ourselves to studying the local behaviour of L at ramification points, and this can be done in the power series ring C[[t]]. Therefore, our question reduces to a question of base change of a Fuchsian ODE along the map sk = t, for a positive integer k. The following computation appears for rank 2 ODEs in [6]. Proposition 2.12 ([6, Lemma 3.21]). Base change of order k at a point p ∈ P1 affects the characteristic exponents of L by multiplying everything by k. Proof . Assume that w(t) is a local solution to Lw = 0. By the chain rule, we have that (cid:0)sk(cid:1) = w(cid:0)sk(cid:1). δs k dw ds s k Thus if w is a solution to the differential equation dw ds therefore (cid:0)sk(cid:1) = ksk−1 dw (cid:0)sk(cid:1), (cid:0)sk(cid:1) = δtw(cid:0)sk(cid:1) = sk dw dt dt n(cid:88) n(cid:88) i=1 i=1 δn t + fi(t)δn−i t , then w(sk) solves δn s + kifi (cid:0)sk(cid:1)δn−i s . Suppose that the indicial equation of L at 0 is given by n(cid:89) i=1 n(cid:89) (T − µi), (T − kµi). for µi the local exponents of L at 0. Then the indicial equation of the differential equation in terms of s is written as i=1 This proves the proposition. (cid:4) Hodge Numbers from Picard–Fuchs Equations 11 Corollary 2.13. After base change, the ramification points of g : P1 → P1 become apparent singularities of g∗L. If p is a point with characteristic exponents all 0, then for any point q ∈ g−1(p), the characteristic exponents of g∗L at q are all 0. Combining this with Theorem 2.7, we can say quite a bit about how the Hodge numbers of a variation of Hodge structure of type (1, 1, . . . , 1) are altered by base change. Remark 2.14. We will take this opportunity to emphasize that performing base change on a variation of Hodge structure and base change on a linear differential equation may not in In other words, there is a natural variation of Hodge structure on P1 fact be compatible. underlying g∗V with appropriate canonical extensions at the points in g−1(∆). The Hodge g of this variation of Hodge structure may not agree with the pull-backs g∗Ep,q of the bundles Ep,q Hodge bundles associated to V. However, the monodromy representation and period maps are correct, so if Lg is the Picard–Fuchs system of the pulled back variation of Hodge structure, then g∗L = hLg for some meromorphic function h. Thus the differences between the characteristic exponents of g∗L are the same as for Lg, so we can apply Theorem 2.7 to compute the degrees of Ep,q g . The underlying reason behind this incompatibility is that Deligne's canonical extension does not commute with base change. Indeed, Deligne's canonical extension of g∗V chooses a branch of the logarithm so that the eigenvalues of the maps Nα at each point of g−1(∆) lie in the interval [0, 1) (see Section 2.1). However base change acts additively on these eigenvalues, so may take them out of this interval; the pulled-back variation of Hodge structure may thus correspond to a different choice of extension of g∗V. 3 Families of elliptic curves In the remainder of this paper we will study some applications of these results in various settings. We begin with the case of a family of elliptic curves. Let L be a rank 2 ODE which underlies a real variation of Hodge structure of weight 1. We will show that if L is the Picard–Fuchs equation of a family of elliptic curves S, then we can determine the geometry of S from L. First, we will restate Theorem 2.7 in the case of a (1, 1)-VHS. Corollary 3.1. Let L be a rank 2 Fuchsian ODE which underlies an R-VHS of weight 1, with regular singular points at ∆ in P1. Then the map θ : E1,0 → E0,1 ⊗ ΩP1(∆) has the following structure. 1) if p is a nonsingular point of L, then θ is an isomorphism at p, 2) if L has an apparent singularity at p, then θ has cokernel of length µp 3) if p ∈ ∆, then θ has cokernel of length (cid:98)µp 2(cid:99) − (cid:98)µp 1(cid:99). 2 − µp 1 − 1 at p, The following theorem follows easily. Theorem 3.2. Suppose that L underlies a family of elliptic curves over P1. Let aL denote the number of points in ∆ at which the local monodromy of L is strictly quasi-unipotent, and let ∆a denote the set of points at which L has an apparent singularity. Then 1(cid:99) − 1(cid:1) + (cid:0)µp (cid:88) p∈∆a 1 − 1(cid:1)(cid:19) . 2 − µp (cid:18) (cid:0)(cid:98)µp (cid:88) p∈∆ deg E0,1 = 1 2 2 − aL + 2(cid:99) − (cid:98)µp 12 C.F. Doran, A. Harder and A. Thompson Remark 3.3. If we note that the characteristic exponents µp or an apparent singularity, and that µp simplify the formula in Theorem 3.2 to 2 = µp i are integers when p is nonsingular 1 + 1 when p is a non-singular point, then we can (cid:18) 1(cid:99) − 1(cid:1)(cid:19) . 2(cid:99) − (cid:98)µp (cid:0)(cid:98)µp (cid:88) p∈P1 deg E0,1 = 1 2 2 − aL + However, the form given in Theorem 3.2 is simpler to work with from a computational stand- point, as both sums are clearly finite. Proof of Theorem 3.2. By Example 2.2, we see that deg E1,0 + deg E0,1 = −aL, and Corollary 3.1 gives that the map θ has cokernel of length (cid:0)µp (cid:88) p∈∆a 1 − 1(cid:1) + 2 − µp (cid:0)(cid:98)µp (cid:88) p∈∆ 2(cid:99) − (cid:98)µp Since deg(E0,1 ⊗ ΩP1(∆)) = ∆ − 2 + deg E0,1, it follows that deg E0,1 + ∆ − 2 = − deg E0,1 − aL + 1(cid:99)(cid:1). (cid:88) (cid:0)µp 2 − µp (cid:88) 1 − 1(cid:1) + p∈∆a p∈∆ 1(cid:99)(cid:1). 2(cid:99) − (cid:98)µp 1 − 1(cid:1) + (cid:0)(cid:98)µp 2(cid:99) − (cid:98)µp (cid:88) (cid:0)(cid:98)µp 1(cid:99) − 1(cid:1)(cid:19) p∈∆ (cid:18) 1 2 2 − aL + (cid:88) p∈∆a (cid:0)µp 2 − µp Thus deg E0,1 = as required. The term (µp 2 − µp (cid:4) 1 − 1) should be thought of as an invariant which counts the ramification of the j-function at p ∈ ∆a; Doran makes this precise in [6]. Remark 3.4. From this, we may use Theorem 2.9 to compute h0,2 for the parabolic cohomology of our elliptic fibration. In [2], Cox and Zucker show that this h0,2 computes the dimension of a particular space of modular forms. The computations in this section can be thought of as giving a way to compute the dimension of this space of modular forms using the Picard–Fuchs equation of the elliptic surface. 4 Families of K3 surfaces Next we turn our attention to families of K3 surfaces, as studied in Example 2.3. Theorem 2.7 can be rephrased in the context of the corresponding (1, 1, 1)-variations of Hodge structure, as follows. Recall that in this case we have the Kodaira–Spencer maps θ0 : E2,0 → E1,1 ⊗ ΩP1(∆) and θ1 : E1,1 → E0,2 ⊗ ΩP1(∆). Corollary 4.1. If L is a Picard Fuchs equation corresponding to a (1, 1, 1)-type variation of Hodge structure, then 1) if p is a nonsingular point of L, then θ0 is an isomorphism, 2) if p is an apparent singularity of L, then θ0 has cokernel of length µp 3) if p is a regular singular point of L, then θ0 has cokernel of length (cid:98)µp 1 − 1 at p, 2 − µp 2(cid:99) − (cid:98)µp 1(cid:99). Hodge Numbers from Picard–Fuchs Equations 13 Let U = P1 \ ∆ denote a Zariski open set and let j : U (cid:44)→ P1 denote the embedding. Suppose that X → U is an M -polarized family of K3 surfaces, in the sense of [7, Definition 2.1], for some rank 19 lattice M . The transcendental lattices of the fibres of X define a VHS V of type (1, 1, 1) over U ; we assume that all of the monodromy matrices of this VHS take the forms Tx considered in Example 2.3. Finally, let L denote the Picard–Fuchs equation underlying this (1, 1, 1)-VHS. E1,1 = 0, so the degree of E1,1 can be computed indeed, if a1/2 is the number of points p at which monodromy is 2 a1/2. We let af denote the ω. Finally, easily using equation (2.1): of type Tun, Ti, Tω or Tnod (see Example 2.3), then deg E1,1 = − 1 number of points whose monodromy matrices look like Tun, Ti, −Ti, T 2 let ∆a denote the set of points at which L has an apparent singularity. Proposition 4.2. If L underlies a family of K3 surfaces X → U as above, then we have By Lemma 2.1, it is easy to see that degpar i , Tω and T 2 deg(cid:0)E0,2(cid:1) = 2 + (cid:0)µp (cid:88) p∈∆a 1 − 1(cid:1) + 2 − µp (cid:0)(cid:98)µp (cid:88) p∈∆ 1(cid:99) − 1(cid:1). 2(cid:99) − (cid:98)µp a1/2 − af + 1 2 Proof . This is proved by a similar calculation to Theorem 3.2. (cid:4) Remark 4.3. As in the case of elliptic curves, we may simplify this formula to the shorter, but less transparent, form deg(cid:0)E0,2(cid:1) = 2 + (cid:0)(cid:98)µp (cid:88) p∈P1 1(cid:99) − 1(cid:1). 2(cid:99) − (cid:98)µp a1/2 − af + 1 2 Using this, one can easily use Theorem 2.9 to compute h0,3 of the Hodge structure on H 1(P1, j∗V). Corollary 4.4. The Hodge number h0,3 of the Hodge structure on H 1(P1, j∗V) is (cid:18) (cid:18) h0,3 = h0 P1, O1P af − 4 − 1 2 a1/2 − (cid:88) (cid:0)µp p∈∆a 1 − 1(cid:1) −(cid:88) (cid:0)(cid:98)µp p∈∆ 2 − µp 1(cid:99) − 1(cid:1)(cid:19)(cid:19) . 2(cid:99) − (cid:98)µp Now we may ask ourselves: when can such a fibration by K3 surfaces have a Calabi–Yau threefold total space? Definition 4.5. A weight n Hodge structure with Hodge numbers hp,q is called Calabi–Yau if hn,0 = 1. If a variety X is Calabi–Yau of dimension n, then the Hodge structure on H n(X, Q) is certainly Calabi–Yau, but the converse is not true. For example, look at any blow-up of an honest Calabi–Yau variety, or a Kulikov surface. According to Corollary 4.4, a family of K3 surfaces X → U has Calabi–Yau Hodge structure on H 1(P1, j∗V) if and only if af = 4 + 1 2 a1/2 + (cid:0)µp (cid:88) p∈∆a 1 − 1(cid:1) + 2 − µp (cid:0)(cid:98)µp (cid:88) p∈∆ 1(cid:99) − 1(cid:1). 2(cid:99) − (cid:98)µp (4.1) By the degeneration of the Leray spectral sequence [31, Section 15], for any compactification X of X , this parabolic cohomology group is a direct summand of the Hodge structure on H 3(X , Q), and its complement has no (3, 0) or (0, 3) part (see [10]). Thus, if equation (4.1) does not hold, then X → P1 has no compactification which is a Calabi–Yau threefold. In [10] we will study the converse problem in specific examples. 14 C.F. Doran, A. Harder and A. Thompson 4.1 Base change of a hypergeometric family We illustrate these results with an example regarding families of K3 surfaces and their pullbacks. Let us take the family of mirror quartic K3 surfaces written as the compactifications in P3 of the fibres of the Laurent polynomial f (x, y, z) = (x + y + z + 1)4 xyz . Narumiya and Shiga [27, Section 5] show that this family has Picard–Fuchs equation which is hypergeometric and may be written as The local monodromies around its singular points are of type T 2 respectively, so the Riemann scheme of this differential operator is un, Ti, and Tnod, over 0, ∞, and 1 (cid:1)(cid:0)δ + 3 (cid:1) . 4 2 4 δ3 + t(cid:0)δ + 1  0 ∞ 1 0 0 1/4 0 1/2 1/2 0 3/4 (cid:1)(cid:0)δ + 1  . 1 (cid:88) r = p∈P1\{0,1,∞} (ep − 1) Dg := af − a1/2 − 4 − (cid:88) Now consider the base change of this local system along a map g : P1 → P1. Let Ei,j g denote the appropriate Hodge bundles of the pulled-back local system g∗V and let hi,j g be the Hodge numbers of its parabolic cohomology. Let d be the degree of g, and set k, (cid:96), m to be the numbers of points over 0, ∞ and 1 respectively. Let denote the degree of ramification of g away from {0, 1,∞}. We will write [y1, . . . , y(cid:96)] for the partition of d encoding the ramification profile over ∞. Finally, define 1 − 1) −(cid:88) (cid:0)(cid:98)µp p∈∆ 1(cid:99) − 1(cid:1), 2(cid:99) − (cid:98)µp (µp 2 − µp p∈∆a where all terms in this expression are computed for the pulled-back local system g∗V. Then we have the following proposition. Proposition 4.6. With assumptions and notation as above, (cid:96)(cid:88) (cid:106) yi (cid:107) Dg = af − 2 + . 4 g = h0(cid:0)P1, OP1(Dg)(cid:1). i=1 In particular, h0,3 Proof . It follows from the definition that af is equal to the number of points over 0 with ramification of order not divisible by 4. Let (cid:96)odd and modd denote the number of points over 0 and 1, respectively, where the order of ramification is odd. Then a1/2 = modd + (cid:96)odd. Apparent singularities appear only at ramification points of g away from points over {0, 1,∞}, and each ramification point has µp (cid:88) (cid:0)µp ∆a 1 − 1(cid:1) = 2 − µp p∈P1\{0,1,∞} (cid:88) 2 − µp 1 − 1 equal to ep − 1. Therefore (ep − 1) = r. Hodge Numbers from Picard–Fuchs Equations 15 We split the set ∆ into three components, corresponding to points over 0, 1, and ∞ respec- tively, and denoted ∆0, ∆1, and ∆∞. The corresponding sums are (d − (cid:96)odd) − (cid:96)(cid:88) (cid:0)(cid:98) yi (cid:99)(cid:1) − (cid:96), 4 i=1 (d − modd) − m, p∈∆0 (cid:88) (cid:88) (cid:88) p∈∆1 (cid:0)(cid:98)µp (cid:0)(cid:98)µp (cid:0)(cid:98)µp p∈∆∞ 1 2 1(cid:99) − 1(cid:1) = 1(cid:99) − 1(cid:1) = 1(cid:99) − 1(cid:1) = k. 1 2 2(cid:99) − (cid:98)µp 2(cid:99) − (cid:98)µp 2(cid:99) − (cid:98)µp Substituting everything into Dg, we obtain Dg = af − 4 − d − r + k + (cid:96) + m + (cid:106) yi (cid:107) 4 . (cid:96)(cid:88) i=1 Now, since g is a map from P1 to P1, the Riemann–Hurwitz formula gives 2d − 2 = (ep − 1) = r + (d − (cid:96)) + (d − m) + (d − k), (cid:88) p which implies that k + (cid:96) + m − d − r = 2. Making this substitution, we find that Dg = af − 2 + (cid:96)(cid:88) (cid:106) yi (cid:107) 4 i=1 , which proves the proposition. (cid:4) As a corollary, we can approximately recover [9, Proposition 2.4] (see also [10] for some further applications of this result). Corollary 4.7. The Hodge structure of g∗V is Calabi–Yau if and only if one of the following two conditions holds: 1) the number (cid:96) is equal to 2 and 1 ≤ y1, y2 ≤ 4; or 2) the number (cid:96) is equal to 1 and 5 ≤ y1 ≤ 8. Proof . By Corollary 4.4 and Proposition 4.6, we may assume that (cid:96)(cid:88) i=1 af + (cid:99) = 2. (cid:98) yi 4 Therefore, we have at most two points over 0. If we have two points over 0, then they must both have 1 ≤ yi ≤ 4. If there is a single point over 0, then 5 ≤ y1 ≤ 8. (cid:4) 5 Families of Calabi–Yau threefolds In this section, we will first apply the results of Section 2.4 to complete the computations of [3] and [23]. Then we will determine the Hodge numbers of some specific fourfolds fibred by quintic mirror Calabi–Yau threefolds. 16 C.F. Doran, A. Harder and A. Thompson 5.1 Inhomogeneous Picard–Fuchs equations and Calabi–Yau threefolds In this section, we will perform some computations related to work of del Angel, Muller-Stach, van Straten, and Zuo [3]. These authors were interested in determining situations in which families of Calabi–Yau threefolds admit normal functions; here a normal function is an algebraic cycle on each fibre of a family of varieties, which is homologous to zero on all fibres. In particular, if we have some algebraic (2, 2)-class Z on a family of Calabi–Yau threefolds, whose restriction to each fibre is homologous to zero, then Z determines a multivalued func- tion ΦZ on the base of the fibration. From such a normal function, one obtains an inhomogeneous extension of the Picard–Fuchs equation Lσ which is satisfied by ΦZ. In other words, LσΦZ = gZ, for some holomorphic function gZ. Morrison and Walcher [26, 30] studied a specific such normal function on the Fermat family of quintics, and showed that ΦZ can be interpreted as the domain wall tension of certain D-branes. The computations in [3] aimed to determine whether such normal functions exist on other 1-parameter families of Calabi–Yau threefolds. Evidence for the existence of normal functions can be found by determining whether the parabolic cohomology of Lσ has classes of type (2, 2) or not. They looked at the 14 hypergeometric examples of Doran and Morgan [12] and their pullbacks along maps [t : s] (cid:55)→ [td : sd], for certain values of d, in an effort to determine whether, after such base change, the 14 families could admit interesting normal functions. In both [3] and the subsequent work of Hollborn and Muller-Stach [23], the authors were not able to compute the degrees of the bundles E2,1 in many situations, and therefore were not able to determine the Hodge numbers of the parabolic cohomology groups that they were interested in. By applying the results of Section 2.4, we are now able to finish their computations. Example 5.1. Let's compute the Hodge numbers of the VHS obtained from the second hy- pergeometric example in the list of Doran and Morgan [12]. This example is obtained in the following way. Begin with the hypergeometric operator This operator underlies a (1, 1, 1, 1)-type rational VHS, and has local monodromy matrices T0 = T1 = T∞ = where ζ10 is a primitive tenth root of unity. Its Riemann scheme is ζ10 0 0 0  , 0 ζ 3 10 0 0 0 0 ζ 7 10 0 0 0 0 ζ 9 10 (cid:19)(cid:18) 3 10 δ + . δ + 9 10 7 10 (cid:19)(cid:18) (cid:19) 1 0 0 0  , 0 1 0 0 0 0 1 1 0 0 0 1 δ4 − t δ + δ + 1 10 (cid:18) (cid:19)(cid:18) 1 1 0 0  ,  . 0 1 1 0 0 0 1 1 0 0 0 1 0 1 ∞ 0 0 1/10 0 1 3/10 0 1 7/10 0 2 9/10  Let us take the base change of this differential equation under the map of degree 5 which is totally ramified over 0 and ∞. The new differential equation has seven singular points and Hodge Numbers from Picard–Fuchs Equations 17 Riemann scheme  0 t5 − 1 = 0 ∞ 1/2 0 3/2 0 0 7/2 9/2 0 0 1 1 2  . The points over 1 all have monodromy of type T1, the point over 0 has monodromy of type T0, and the point over ∞ has monodromy matrix equal to −Id4. Therefore, the only point with nontrivial parabolic filtration is ∞ and the only filtrands are of weight 1/2. Thus equation (2.1) and Lemma 2.1 give deg E1,2 + deg E2,1 = −1. By Theorem 2.7, the map θ1 : E2,1 → E1,2 ⊗ ΩP1(∆) has cokernel of length 2. Therefore, deg E2,1 − deg E1,2 = 3. Hence deg E 2,1 = 1. This gives the correct b value in Theorem 2.11, which in turn allows us to complete the corresponding row of the table in [23, Section 4]; we obtain that the Hodge numbers of g∗L are (0, 1, 2, 1, 0). Similarly, if we take g to be the degree 10 map ramified completely at 0 and ∞, then deg E1,2 + deg E2,1 = 0. In this case θ1 has cokernel of length 4, so deg E2,1 − deg E1,2 = 6, and therefore deg E2,1 = 3. Once again, this allows us to complete the corresponding row of the table in [23, Section 4]; we obtain that the Hodge numbers of g∗L are (0, 1, 3, 1, 0). Using this, we may complete the calculations left unfinished by [3] and [23]. Table 3 is a reproduction of the table from [23, Section 4], with all missing pieces completed; our new results are highlighted in gray. Our notation is as follows. # is the case number from the table in [23, Section 4], and the numbers (α1, α2, α3, α4) are the numbers determining the corresponding hypergeometric differential equations L = δ4 − t(δ + α1)(δ + α2)(δ + α3)(δ + α4). The number d indicates that we have pulled back by the covering [t : s] (cid:55)→ [td : sd], to obtain a VHS over P1 with Hodge bundles E3,0 of degree a and E2,1 of degree b. We then apply the results of Section 2.4 to compute the Hodge numbers of the corresponding parabolic cohomology groups in the final column. Remark 5.2. Recently there has been a great deal of interest in studying the monodromy subgroups of Sp(4, R) in the 14 hypergeometric examples of Doran and Morgan [12]. These subgroups may be either arithmetic or non-arithmetic (more commonly called thin). Singh and Venkataramana [28, 29] have proved that the monodromy subgroup is arithmetic in seven of the 14 cases, and Brav and Thomas [1] have proved that it is thin in the remaining seven. The arithmetic/thin status of each of the examples is given in the "Monodromy" column of Table 3. Computations carried out in [13] suggest that the arithmetic/thin dichotomy for monodromy groups of hypergeometric local systems is closely related to the value of the Lyapunov exponents 18 C.F. Doran, A. Harder and A. Thompson of the corresponding variations of Hodge structure. Lyapunov exponents are dynamical invari- ants associated to a flat vector bundle over a hyperbolic Riemann surface, that are computed using the monodromy of the flat bundle as well as the metric structure on the hyperbolic curve. The main goal of [13] is to study these Lyapunov exponents; [13, Lemma 6.3] (restated above as Theorem 2.7) is a step towards this. In [13] it is conjectured that for the 14 hypergeometric variations of Hodge structure of Doran and Morgan [12], the associated Lyapunov exponents are closely related to the parabolic degrees of the bundles E3,0 and E2,1, provided that these degrees satisfy a certain relation (see [13, Conjecture 6.4] and [15] for the higher rank case). It is not clear whether there is any relation between the computations presented here and Lyapunov exponents; however, it would be interesting to see whether there is any relation E2,1 as the between Lyapunov exponents and the asymptotic growth of degpar degree d of the covering [t : s] (cid:55)→ [td : sd] grows. Such growth has been investigated in the rank 2 case by Kappes [24]. E3,0 and degpar Finally we note that, according to work of Doran and Malmendier [11], most of Doran's and Morgan's [12] 14 hypergeometric local systems can be obtained from hypergeometric local systems of rank 3. These rank 3 hypergeometric local systems underlie variations of Hodge structure corresponding to families of K3 surfaces, for which the Lyapunov exponents have been computed by Filip [14]. It seems plausible that the iterated construction of [11] along with the computations of [14] could allow one to compute Lyapunov exponents for the fourteen rank 4 hypergeometric local systems of [12]. Remark 5.3. We pose the question as to whether the above Hodge structures are reducible or not. It is possible that the h2,2 value in each of the sets of Hodge numbers in the above table all correspond to rational Hodge classes, thus induce normal functions on the corresponding families of Calabi–Yau threefolds. The maps along which we have pulled back are rigid in the sense that any deformation of this map will change the Hodge numbers, so this indeed seems possible. 5.2 Calabi–Yau fourfolds f ibred by mirror quintics Finally, we will apply the same approach to classifying Calabi–Yau fourfolds fibred by Calabi– Yau threefolds as we did with Calabi–Yau threefolds fibred by K3 surfaces. We do not obtain concrete classification results, since modularity results for families of Calabi–Yau varieties of higher dimension are scarce, but we are at least able to place bounds on the space of possibilities. We start with the mirror quintic family of K3 surfaces. This family can be represented as a compactification of the fibres of the Laurent polynomial (x + y + z + w + 1)5 xyzw . The periods of this family satsify the hypergeometric differential equation L δ4 − t 1 5 δ + δ + (cid:18) (cid:19)(cid:18) 1 1 0 0  , 0 1 1 0 0 0 1 1 0 0 0 1 T0 = T1 = (cid:19)(cid:18) 2 5 δ + 3 5 , 4 5 δ + (cid:19)(cid:18) (cid:19) 1 0 0 0  , 0 1 0 0 0 0 1 1 0 0 0 1 which has local monodromy matrices ζ5 0 0 0 0 ζ 2 5 0 0 0 0 ζ 3 5 0  , 0 0 0 ζ 4 5 T∞ = Hodge Numbers from Picard–Fuchs Equations 19 # 1 (α1, α2, α3, α4) (1/5, 2/5, 3/5, 4/5) 2 (1/10, 3/10, 7/10, 9/10) 3 4 5 6 7 8 9 (1/2, 1/2, 1/2, 1/2) (1/3, 1/3, 2/3, 2/3) (1/3, 1/2, 1/2, 2/3) (1/4, 1/2, 1/2, 3/4) (1/8, 3/8, 5/8, 7/8) (1/6, 1/3, 2/3, 5/6) (1/12, 5/12, 7/12, 11/12) 10 (1/4, 1/4, 3/4, 3/4) 11 (1/6, 1/4, 3/4, 5/6) 12 (1/4, 1/3, 2/3, 3/4) 13 (1/6, 1/6, 5/6, 5/6) 14 (1/6, 1/2, 1/2, 5/6) d 1 2 5 10 1 2 5 10 1 2 2k 1 2 3 6 1 6 1 4 8 1 2 4 8 1 2 6 1 2 3 4 6 12 1 2 4 8 1 2 12 1 2 3 12 1 2 3 6 1 3 6 a 0 0 1 2 0 0 0 1 0 1 k 0 0 1 2 0 1 0 1 2 0 0 0 1 0 0 1 0 0 0 0 0 1 0 0 1 2 0 0 1 0 0 0 1 0 0 0 1 0 0 1 b 0 0 2 4 0 0 1 3 0 1 k 0 0 1 2 0 2 0 2 4 0 0 1 3 0 0 2 0 0 1 1 2 5 0 0 1 2 0 0 3 0 0 1 4 0 0 0 1 0 1 3 Hodge numbers Monodromy (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 0, 0, 0, 0) (1, 1, 1, 1, 1) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 1, 2, 1, 0) (0, 1, 3, 1, 0) (0, 0, 0, 0, 0) (0, 0, 0, 0, 0) (k − 1, 0, 0, 0, k − 1) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 0, 0, 0, 0) (1, 0, 1, 0, 1) (0, 0, 0, 0, 0) (0, 0, 2, 0, 0) (0, 0, 0, 0, 0) (0, 0, 0, 0, 0) (1, 1, 0, 1, 1) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 1, 1, 1, 0) (0, 1, 1, 1, 0) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 0, 1, 0, 0) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 1, 0, 1, 0) (0, 1, 1, 1, 0) (0, 2, 1, 2, 0) (0, 3, 1, 3, 0) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 0, 1, 0, 0) (1, 0, 3, 0, 1) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 1, 5, 1, 0) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 1, 0, 1, 0) (0, 2, 3, 2, 0) (0, 0, 0, 0, 0) (0, 0, 1, 0, 0) (0, 0, 2, 0, 0) (0, 0, 3, 0, 0) (0, 0, 0, 0, 0) (0, 1, 0, 1, 0) (0, 1, 0, 1, 0) Thin Arithmetic Thin Arithmetic Thin Thin Thin Arithmetic Thin Arithmetic Arithmetic Arithmetic Arithmetic Thin Table 3. Invariants for families of Calabi–Yau threefolds. 20 C.F. Doran, A. Harder and A. Thompson and Riemann scheme  0 1 ∞ 0 0 1/5 0 1 2/5 0 1 3/5 0 2 4/5  . We can ask: along which maps g : P1 → P1 does the pull-back g∗L have Hodge number h0,4 = 1? In other words, when can the fourfold total space obtained from the mirror quintic family be a Calabi–Yau fourfold? We have all the tools to perform this calculation. Let us take a map g : P1 → P1 of degree d. Let k, (cid:96), and m denote the numbers of points over 0, ∞, and 1 respectively and let [y1, . . . , y(cid:96)] denote the partition of d encoding the ramification profile over ∞. Finally, let (cid:96)0 be the number of values of i such that 5 (cid:45) yi. We compute the Hodge numbers of the parabolic cohomology associated to g∗L. Proposition 5.4. The degrees of the Hodge bundles of the variation of Hodge structure on g∗L are given by (cid:18) (cid:18) d − (cid:96)0 − (cid:96)(cid:88) d − (cid:96)0 − (cid:96)(cid:88) i=1 (cid:0)(cid:98) 3yi (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1)(cid:19) 5 (cid:99)(cid:1)(cid:19) 5 (cid:99) − (cid:98) 2yi 5 (cid:99) − (cid:98) 2yi i=1 i=1 , deg E3,0 = deg E2,1 = 1 2 1 2 − (cid:96)(cid:88) (cid:0)(cid:98) 2yi 5 (cid:99)(cid:1) 5 (cid:99) − (cid:98) yi and the equations deg E3,0 + deg E0,3 = −(cid:96)0 and deg E2,1 + deg E1,2 = −(cid:96)0. Remark 5.5. From this data, one can compute all Hodge numbers of the parabolic cohomology groups, by applying the results of Section 2.4. Proof . By Theorem 2.7, the map θ1 : E2,1 → E1,2 ⊗ ΩP1(∆) has cokernel of length (cid:0)(cid:98) 3yi (cid:96)(cid:88) i=1 5 (cid:99)(cid:1) − ((cid:96) − (cid:96)0) 5 (cid:99) − (cid:98) 2yi r + (note that ((cid:96) − (cid:96)0) counts the number of values of i so that 5 yi). Thus deg E2,1 − deg E1,2 = k + (cid:96) + m − 2 − r − (cid:96)(cid:88) (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1). 5 (cid:99) − (cid:98) 2yi i=1 Applying the Riemann–Hurwitz formula, we see that 2d − 2 = (ep − 1) = r + (d − k) + (d − (cid:96)) + (d − m), (5.1) (cid:88) p which implies that d = k + (cid:96) + m − r − 2. Therefore, deg E2,1 − deg E1,2 = d − (cid:96)(cid:88) (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1). 5 (cid:99) − (cid:98) 2yi i=1 Hodge Numbers from Picard–Fuchs Equations 21 Equation (2.1) and Lemma 2.1 also give that deg E3−i,i + deg Ei,3−i = −(cid:96)0. So we conclude that deg E2,1 = 5 (cid:99)(cid:1)(cid:19) . 5 (cid:99) − (cid:98) 2yi (cid:18) 1 2 d − (cid:96)0 − (cid:96)(cid:88) (cid:0)(cid:98) 2yi (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1) − ((cid:96) − (cid:96)0), 5 (cid:99) − (cid:98) yi i=1 (cid:96)(cid:88) r + d + We also have that the cokernel of the map θ0 : E3,0 → E2,1 ⊗ ΩP1(∆) has length i=1 thus deg E3,0 − deg E2,1 = k + (cid:96) + m − 2 − r − d − (cid:96)(cid:88) (cid:0)(cid:98) 2yi 5 (cid:99)(cid:1) = − (cid:96)(cid:88) (cid:0)(cid:98) 2yi 5 (cid:99) − (cid:98) yi 5 (cid:99)(cid:1), 5 (cid:99) − (cid:98) yi i=1 i=1 where the second line follows by applying equation (5.1). Adding the expression for deg E2,1 obtained above, we get (cid:18) d − (cid:96)0 − (cid:96)(cid:88) (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1)(cid:19) − (cid:96)(cid:88) (cid:0)(cid:98) 2yi 5 (cid:99)(cid:1), 5 (cid:99) − (cid:98) yi 5 (cid:99) − (cid:98) 2yi i=1 i=1 deg E3,0 = 1 2 which completes the proof of the theorem. (cid:4) Since we have that deg E3,0 + deg E0,3 = −(cid:96)0, it follows that (cid:18) d + (cid:96)0 − (cid:96)(cid:88) (cid:0)(cid:98) 3yi 5 (cid:99)(cid:1)(cid:19) (cid:96)(cid:88) (cid:0)(cid:98) 2yi + 5 (cid:99)(cid:1). 5 (cid:99) − (cid:98) yi 5 (cid:99) − (cid:98) 2yi deg E0,3 = − 1 2 i=1 i=1 Therefore, Corollary 5.6. The Hodge structure on g∗L is Calabi–Yau if and only if (cid:18) d + (cid:96)0 − (cid:96)(cid:88) (cid:0)(cid:98) 3yi 1 2 5 (cid:99)(cid:1)(cid:19) − (cid:96)(cid:88) (cid:0)(cid:98) 2yi 5 (cid:99)(cid:1) = 2. 5 (cid:99) − (cid:98) yi 5 (cid:99) − (cid:98) 2yi i=1 i=1 Remark 5.7. Note that the value of h0,3 of the parabolic cohomology only depends on the structure of ramification over the point ∞, just as in the case of quartic threefolds. A consequence of this corollary is that there is a finite and enumerable set of values for [y1, . . . , y(cid:96)] so that g∗L is Calabi–Yau. This set strictly contains the possible ramification data over ∞ under which the total space of the pullback of the family of mirror quintics is Calabi–Yau. The list of all values of [y1, . . . , y(cid:96)] are as follows: [1, 1], [1, 2], [1, 3], [1, 4], [1, 5], [2, 2], [2, 3], [2, 4], [2, 5], [3, 3], [3, 4], [3, 5], [4, 4], [4, 5], [5, 5], [6], [7], [8], [9], [10]. We expect that a proper subset of these ramification indices should correspond to maps along which we can pull back the family of mirror quintics and get a variety that admits a smooth (or terminal) Calabi–Yau resolution. cohomology is 0 are the cases where (cid:96) = 1 and y1 ∈ {1, 2, 3, 4, 5}. One can also check that the only cases in which the Hodge number h3,0 of the parabolic 22 C.F. Doran, A. Harder and A. Thompson Remark 5.8. Similar calculations can presumably be completed for any family of Calabi– Yau threefolds realising one of the fourteen hypergeometric variations of Hodge structure of type (1, 1, 1, 1) classified by Doran and Morgan [12]. This includes the so-called "mirror twin" families [12, Section 3], which have the same Picard–Fuchs equations and R-VHS's as some well-known toric examples, but have different underlying Z-VHS's and Hodge number h1,1. In particular, the calculations above apply equally to the "quintic mirror twin" family, which shares its Picard–Fuchs equation and R-VHS with the quintic mirror example above. Acknowledgements C.F. Doran (University of Alberta) was supported by the Natural Sciences and Engineering Research Council of Canada, the Pacific Institute for the Mathematical Sciences, and the Visiting Campobassi Professorship at the University of Maryland. A. Harder (University of Miami) was partially supported by the Simons Collaboration Grant in Homological Mirror Symmetry. A. Thompson (University of Warwick/University of Cambridge) was supported by the Engi- neering and Physical Sciences Research Council programme grant Classification, Computation, and Construction: New Methods in Geometry. References [1] Brav C., Thomas H., Thin monodromy in Sp(4), Compos. Math. 150 (2014), 333–343, arXiv:1210.0523. [2] Cox D.A., Zucker S., Intersection numbers of sections of elliptic surfaces, Invent. Math. 53 (1979), 1–44. [3] Del Angel P.L., Muller-Stach S., Van Straten D., Zuo K., Hodge classes associated to 1-parameter families of Calabi–Yau 3-folds, Acta Math. Vietnam. 35 (2010), 7–22, arXiv:0911.0277. [4] Deligne P., ´Equations diff´erentielles `a points singuliers r´eguliers, Lecture Notes in Math., Vol. 163, Springer- Verlag, Berlin – New York, 1970. [5] Doran B., Doran C.F., Harder A., Picard–Fuchs uniformization of modular subvarieties, in Uniformization, Riemann–Hilbert Correspondence, Calabi–Yau Manifolds, and Picard–Fuchs Equations (Institut Mittag- Leffler, 2017), Editors L. Ji, S.-T. Yau, to appear. [6] Doran C.F., Picard–Fuchs uniformization: modularity of the mirror map and mirror-moonshine, in The Arithmetic and Geometry of Algebraic Cycles (Banff, AB, 1998), CRM Proc. Lecture Notes, Vol. 24, Amer. Math. Soc., Providence, RI, 2000, 257–281, math.AG/9812162. [7] Doran C.F., Harder A., Novoseltsev A.Y., Thompson A., Families of lattice polarized K3 surfaces with monodromy, Int. Math. Res. Not. 2015 (2015), 12265–12318, arXiv:1312.6434. [8] Doran C.F., Harder A., Novoseltsev A.Y., Thompson A., Calabi–Yau threefolds fibred by Kummer surfaces associated to products of elliptic curves, in String-Math 2014, Proc. Sympos. Pure Math., Vol. 93, Amer. Math. Soc., Providence, RI, 2016, 263–287, arXiv:1501.04024. [9] Doran C.F., Harder A., Novoseltsev A.Y., Thompson A., Calabi–Yau threefolds fibred by mirror quartic K3 surfaces, Adv. Math. 298 (2016), 369–392, arXiv:1501.04019. [10] Doran C.F., Harder A., Novoseltsev A.Y., Thompson A., Calabi–Yau threefolds fibred by high rank lattice Polarized K3 surfaces, arXiv:1701.03279. [11] Doran C.F., Malmendier A., Calabi–Yau manifolds realizing symplectically rigid monodromy tuples, arXiv:1503.07500. [12] Doran C.F., Morgan J.W., Mirror symmetry and integral variations of Hodge structure underlying one- parameter families of Calabi–Yau threefolds, in Mirror Symmetry. V, AMS/IP Stud. Adv. Math., Vol. 38, Editors J.D. Lewis, S.-T. Yau, N. Yui, Amer. Math. Soc., Providence, RI, 2006, 517–537, math.AG/0505272. [13] Eskin A., Kontsevich M., Moeller M., Zorich A., Lower bounds for Lyapunov exponents of flat bundles on curves, arXiv:1609.01170. [14] Filip S., Families of K3 surfaces and Lyapunov exponents, arXiv:1412.1779. [15] Fougeron C., Parabolic degrees and Lyapunov exponents for hypergeometric local systems, arXiv:1701.08387. Hodge Numbers from Picard–Fuchs Equations 23 [16] Fujino O., A canonical bundle formula for certain algebraic fiber spaces and its applications, Nagoya Math. J. 172 (2003), 129–171. [17] Green M., Griffiths P., Kerr M., Some enumerative global properties of variations of Hodge structures, Mosc. Math. J. 9 (2009), 469–530. [18] Griffiths P.A., Periods of integrals on algebraic manifolds. I. Construction and properties of the modular varieties, Amer. J. Math. 90 (1968), 568–626. [19] Griffiths P.A., Periods of integrals on algebraic manifolds. II. Local study of the period mapping, Amer. J. Math. 90 (1968), 805–865. [20] Griffiths P.A., On the periods of certain rational integrals. I, Ann. of Math. 90 (1969), 460–495. [21] Griffiths P.A., On the periods of certain rational integrals. II, Ann. of Math. 90 (1969), 496–541. [22] Griffiths P.A., Periods of integrals on algebraic manifolds. III. Some global differential-geometric properties of the period mapping, Inst. Hautes ´Etudes Sci. Publ. Math. 38 (1970), 125–180. [23] Hollborn H., Muller-Stach S., Hodge numbers for the cohomology of Calabi–Yau type local systems, in Algebraic and Complex Geometry, Springer Proc. Math. Stat., Vol. 71, Springer, Cham, 2014, 225–240, arXiv:1302.3047. [24] Kappes A., Lyapunov exponents of rank 2-variations of Hodge structures and modular embeddings, Ann. Inst. Fourier (Grenoble) 64 (2014), 2037–2066, arXiv:1303.1088. [25] Miranda R., The basic theory of elliptic surfaces, Dottorato di Ricerca in Matematica, ETS Editrice, Pisa, 1989. [26] Morrison D.R., Walcher J., D-branes and normal functions, Adv. Theor. Math. Phys. 13 (2009), 553–598, arXiv:0709.4028. [27] Narumiya N., Shiga H., The mirror map for a family of K3 surfaces induced from the simplest 3-dimensional reflexive polytope, in Proceedings on Moonshine and Related Topics (Montr´eal, QC, 1999), CRM Proc. Lecture Notes, Vol. 30, Amer. Math. Soc., Providence, RI, 2001, 139–161. [28] Singh S., Arithmeticity of four hypergeometric monodromy groups associated to Calabi–Yau threefolds, Int. Math. Res. Not. 2015 (2015), 8874–8889, arXiv:1308.4039. [29] Singh S., Venkataramana T.N., Arithmeticity of certain symplectic hypergeometric groups, Duke Math. J. 163 (2014), 591–617, arXiv:1208.6460. [30] Walcher J., Extended holomorphic anomaly and loop amplitudes in open topological string, Nuclear Phys. B 817 (2009), 167–207, arXiv:0705.4098. [31] Zucker S., Hodge theory with degenerating coefficients. L2 cohomology in the Poincar´e metric, Ann. of Math. 109 (1979), 415–476.
1104.5581	2	1104	2012-01-02T11:50:05	Torsors over Luna strata	[ "math.AG" ]	Let G be a reductive group and X be a Luna stratum on the quotient space V//G of a rational G-module V. We consider torsors over X with both non-commutative and commutative structure groups. It allows us to compute the divisor class group and the Cox ring of a Luna stratum under mild assumptions. This techniques gives a simple cause why many Luna strata are singular along their boundary.	math.AG	math	TORSORS OVER LUNA STRATA IVAN V. ARZHANTSEV Abstract. Let G be a reductive group and XH be a Luna stratum on the quotient space V //G of a rational G-module V . We consider torsors over XH with both non-commutative and commutative structure groups. It allows us to describe the divisor class group and the Cox ring of a Luna stratum under mild assumptions. This approach gives a simple cause why many Luna strata are singular along their boundary. Introduction Consider a reductive group G and a rational finite-dimensional G-module V over an al- gebraically closed field K of characteristic zero. The inclusion of the algebra of invariants K[V ]G in the polynomial algebra K[V ] gives rise to the quotient morphism π : V → V //G := Spec K[V ]G. In [8], D. Luna introduced a stratification of the quotient space X := V //G by smooth locally closed subvarieties. Every stratum is defined by the conjugacy class of the isotropy group H of a closed G-orbit in the fibre of π. With any such subgroup H one associates the group W = NG(H)/H. Then the stratum XH comes with a W -torsor π : V hHi → XH , where V hHi is an open subset in the subspace V H of H-fixed vectors. The aim of this work is to use this observation to describe the Cox rings of Luna strata and to apply them to the study of geometric properties of the strata and their closures. It is well known that a smooth variety Y with a finitely generated divisor class group Cl(Y ) admits a canonical presentation as a geometric quotient of a quasiaffine variety by an action of a quasitorus Q. This presentation also is known as a universal torsor over Y . The quasiaffine variety mentioned above is the relative spectrum of the so-called Cox sheaf on Y and Q is a direct product of a torus and a finite abelian group such that the group of characters X(Q) is identified with Cl(Y ). The W -torsor over a stratum XH is not a good candidate for a universal torsor, because the group W in general is not commutative. A naive idea is to consider the commutant S = [W, W ], the factor group Q = W/S, and to decompose the W -torsor into two steps, namely 2 1 0 2 n a J 2 ] . G A h t a m [ 2 v 1 8 5 5 . 4 0 1 1 : v i X r a V hHi /W XH /Q ;✇✇✇✇✇✇✇✇✇ /S $❍❍❍❍❍❍❍❍❍❍ V hHi/S Date: June 4, 2018. 2010 Mathematics Subject Classification. Primary 14R20; Secondary 14L30. Key words and phrases. Reductive group, rational module, quotient, Luna stratification, torsor, Cox ring. Supported by RFBR grants 09-01-00648-a. 1 / / $ ; 2 IVAN ARZHANTSEV Note that the groups H, W , S, and Q are reductive. In Theorem 2 we show that this indeed gives a universal Q-torsor V hHi/S → XH under some "codimension two" assumption on the stratum. In particular, the divisor class group of XH is isomorphic to X(Q). It turns out that in many cases a Luna stratum is singular along its boundary, i.e. the stra- tum coincides with the smooth locus of its closure in X, see [15, Theorem 8.1], [7, Theorem 7] and [6, Theorem 1.2]. In [6], J. Kuttler and Z. Reichstein use this property to prove that quite often the Luna stratification is intrinsic. The latter means that every automorphism of X as of an abstract affine variety preserves the stratification possibly permuting the strata. In our terms the reason for XH to be singular along its boundary may be formulated as follows. Since the canonical quotient presentation of the normalization Norm(XH) of the closure of the stratum is given in Theorem 2 by the quotient morphism p : V H//S → V H//W with the acting group Q, it is natural to expect that V hHi/S is the preimage of the principal Luna stratum in V H//W for the Q-variety V H//S. Thus the fibre of p over a point x ∈ V H//W is a free Q-orbit if and only if x is in XH , and this looks like a characterization of smooth points on V H //W . We realize this approach in Theorem 3 under the assumption that the group W is commutative. The text is organized as follows. In Sections 1 and 2 we recall basic facts on the Luna stratification and on Cox rings respectively. The class of admissible Luna strata which we are going to deal with in the theorems is defined and discussed in Section 3. In particular, we prove in Proposition 1 that if H is the isotropy group of a point with a closed G-orbit in a module V , then the Luna stratum corresponding to H and to the G-module V ⊕k with k ≥ 2 is admissible. The idea to obtain strata with good properties by replacing a module V by V ⊕k is taken from [6]. In Section 4 we formulate the main result (Theorem 2) and obtain some corollaries. The proof of Theorem 2 is given in Section 5. Finally, Section 6 is devoted to examples. 1. The Luna stratification In this section we recall basic facts on the Luna stratification obtained in [8], see also [14, Section 6] and [16]. Let G be a reductive affine algebraic group over an algebraically closed field K of characteristic zero and V be a rational finite-dimensional G-module. Denote by K[V ] the algebra of polynomial functions on V and by K[V ]G the subalgebra of G-invariants. Let V //G be the spectrum of the algebra K[V ]G. The inclusion K[V ]G ⊆ K[V ] gives rise to a morphism π : V → V //G called the quotient morphism for the G-module V . It is well known that the morphism π is a categorical quotient for the action of the group G on V in the category of algebraic varieties, see [14, 4.6]. In particular, π is surjective. The affine variety X := V //G is irreducible and normal. It is smooth if and only if the point π(0) is smooth on X. In the latter case X is an affine space. Every fibre of the morphism π contains a unique closed G-orbit. For any closed G-invariant subset A ⊆ V its image π(A) is closed in X. These and other properties of the quotient morphism may be found in [14, 4.6]. By Matsushima's criterion, if an orbit G · v is closed in V , then the isotropy group Stab(v) is reductive, see [11], [12], or [14, 4.7]. Moreover, there exists a finite collection {H1, . . . , Hr} of reductive subgroups in G such that if an orbit G · v is closed in V , then Stab(v) is conjugate to one of these subgroups. This implies that every fibre of the morphism π contains a point whose isotropy group coincides with some Hi. TORSORS OVER LUNA STRATA 3 For every isotropy group H of a closed G-orbit in V the subset VH := {w ∈ V ; there exists v ∈ V such that G · w ⊃ G · v = G · v and Stab(v) = H} is G-invariant and locally closed in V . The image XH := π(VH) turns out to be a smooth locally closed subset of X. In particular, XH is a smooth quasiaffine variety. By stratification of a variety X we mean a decomposition of X into disjoint union of smooth locally closed subsets. Definition 1. The stratification X = rGi=1 XHi is called the Luna stratification of the quotient space X. There is a unique open dense stratum called the principal stratum of X. A stratum XHi is contained in the closure of a stratum XHj if and only if the subgroup Hi contains a subgroup conjugate to Hj. This induces a partial ordering of the set of strata compatible with the (reverse) ordering on the set of conjugacy classes. Consider the subsets V H = {v ∈ V ; H · v = v} and V hHi = {v ∈ V ; Stab(v) = H and G · v = G · v}. Then V hHi is an open subset of V H. Moreover, the restriction of the morphism π to V H maps V hHi to XH surjectively. Let NG(H) be the normalizer of the subgroup H in G. Since H is reductive, the subgroup NG(H) is reductive and the connected component NG(H)0 is the product of H 0 and CG(H)0, where CG(H)0 is the connected component of the centralizer CG(H) of H in G, see [10, Lemma 1.1]. The group NG(H) preserves the subspace V H. Moreover, the kernel of the NG(H)-action on V H is H, and we get an effective action of the (reductive) group W := NG(H)/H on V H . By [9], for any point v ∈ V H the orbit G · v is closed in V if and only if the orbit W · v is closed in V H. In particular, V hHi is the union of closed free W -orbits in V H. The following definition first appeared in [13] plays an important role in Invariant Theory. Definition 2. An action of a reductive group F on an affine variety Z is stable if there exists an open dense subset U ⊆ Z such that the orbit F · z is closed in Z for any z ∈ U. Equivalently, an F -action on Z is stable if the general fibre of the quotient morphism π : Z → Z//F := Spec K[Z]F is an F -orbit, i.e. there is an open dense subset U ′ ⊆ Z//F such that π−1(u) is an F -orbit for any u ∈ U ′. Since V hHi is open in V H, the action of W on V H is stable. This action enjoys an additional nice property, namely, the isotropy group of the general point in V H is trivial. Summarizing, we observe that the restriction of the quotient morphism π to V hHi defines a W -torsor π : V hHi → XH. Restricting the algebra of invariants K[V ]G to the subspace V H , we get a subalgebra ResH(K[V ]G) in K[V H ]W . The closure XH may be identified with Spec ResH(K[V ]G). By [9], the morphism πH : V H//W → XH given by the inclusion ResH(K[V ]G) ⊆ K[V H]W is the normalization morphism. For the principal stratum, the restriction ResH is injective, the closure XH = X is normal, and we 4 IVAN ARZHANTSEV get an isomorphism K[V ]G → K[V H ]W . Several results on normality of the closures of Luna strata and on strata of codimension one may be found in [16] and [17]. 2. Cox rings and characteristic spaces Let X be a normal algebraic variety with finitely generated divisor class group Cl(X). Assume that any regular invertible function f ∈ K[X]× is constant. Roughly speaking, the Cox ring of X may be defined as R(X) := MD∈Cl(X) Γ(X, OX (D)). In order to obtain a multiplicative structure on R(X) some technical work is needed, especially when the group Cl(X) has torsion. We refer for details to [1, Section 4] and [5]. In a similar way one defines the Cox sheaf R, which is a sheaf of Cl(X)-graded algebras on X. The Cox ring R(X) is the ring of global sections of the sheaf R. Assume that this sheaf is locally of By definition, a quasitorus 1 is a commutative reductive algebraic group. It is not difficult to check that any quasitorus is a direct product of a torus and a finite abelian group. Since finite type. Then the relative spectrum bX := SpecX(R) is a quasiaffine variety. R is a sheaf of Cl(X)-graded algebras, the variety bX comes with an action of a quasitorus QX whose group of characters X(QX) is identified with Cl(X). Now we need a notion from Geometric Invariant Theory. Let G be a reductive group and Z be a normal G-variety. An invariant morphism π : Z → X is a good quotient, if π is affine and the pullback map π∗ : OX → OG Z is an isomorphism. An example of a good quotient is the quotient morphism π : V → V //G discussed in Section 1. Moreover, for any open subset U ⊆ V //G the restriction π : π−1(U) → U is a good quotient as well. is called the characteristic space over X. If the variety X is smooth, then the QX -action on It turns out that the QX -action on bX admits a good quotient pX : bX → X. This morphism bX is free, and the map pX : bX → X is a QX-torsor. Moreover, this is a universal torsor over X in the sense of [18]. The following definition appeared in [1, Definition 6.4.1], see also [5, Definition 1.37]. Definition 3. An action of a reductive group F on an affine variety Z is said to be strongly stable if there exists an open dense invariant subset U ⊆ Z such that 1. the complement Z \ U is of codimension at least two in Z; 2. the group F acts freely on U; 3. for every z ∈ U the orbit F · z is closed in Z. One may check that the QX -action on bX defined above is strongly stable with U being the X (Xreg) of the smooth locus of X. preimage p−1 Definition 4. A normal variety Y equipped with an action of a quasitorus Q is called Q- factorial if any Q-invariant Weil divisor on Y is principal. Again the characteristic space bX is a QX -factorial variety, see [1, Lemma 5.3.6] and [5, Theorem 1.14]. 1Also called a diagonalizable group. TORSORS OVER LUNA STRATA 5 The following theorem given in [1, Theorem 6.4.3] and [5, Theorem 1.40] shows that strong stability and Q-factoriality define exactly the class of characteristic spaces. This result is our main tool in the proof of Theorem 2. Theorem 1. Let a quasitorus Q act on a normal quasiaffine variety X with a good quotient p : X → X. Assume that K[X ]× = K× holds, X is Q-factorial and the Q-action is strongly stable. Then there is a commutative diagram X ❄❄❄❄❄❄❄❄ p µ ∼= X bX ⑦⑦⑦⑦⑦⑦⑦⑦ pX where the quotient space X is a normal variety with K[X]× = K×, we have Cl(X) = X(Q), the morphism pX : bX → X is a characteristic space over X, and the isomorphism µ : X → bX is equivariant with respect to actions of Q = QX . 3. Admissible Luna strata Let us fix settings and notation for the remaining part of the text. Let G be a reductive group, V be a rational finite-dimensional G-module, π : V → V //G := Spec K[V ]G be the quotient morphism with the quotient space X := V //G, and H be an isotropy group of a closed G-orbit in V . Define the group W = NG(H)/H, its commutant S := [W, W ] and the factor group Q := W/S. Recall that the groups W and S are reductive. Moreover, S is semisimple provided W is connected. Being reductive and commutative, the group Q is a quasitorus. We adapt Definition 3 to the situation we are dealing with. Definition 5. A triple (G, V, H) is admissible, if the action of the group W on V H is strongly stable. An admissible stratum is the Luna stratum XH corresponding to an admissible triple (G, V, H). Remark 1. A triple (G, V, H) is admissible if and only if the complement of V hHi in V H has codimension at least two. Indeed, we know from Section 1 that the action of W on V hHi is free, all W -orbits on V hHi are closed in V H and the subset V hHi is characterized by these properties. In other words, the subset V hHi is the preimage of the principal Luna stratum in the W -module V H. This shows that in order to check admissibility it suffices to work with principal strata. Since XH = π(V H), the complement of an admissible stratum in its closure has codimension at least two. In this case the algebra of regular functions K[XH ] coincides with K[Norm(XH)], and the latter algebra is isomorphic to K[V H ]W , see Section 1. In particular, K[XH]× = K× holds for any admissible stratum. Following [6] we say that a Luna stratum XH is singular along its boundary if the singular locus of the closure XH is precisely the complement XH \ XH. This property is known for the principal stratum on the quotient variety of the space of representations of a quiver except for some low dimensional anomalies [7, Theorem 7]. Moreover, there are many modules where every Luna stratum is singular along its boundary, see [6, Theorem 1.2]. In particular, this is the case for the module V ⊕r, where V is any rational G-module and r ≥ 2 dim(V ), or where / / 6 IVAN ARZHANTSEV V is the adjoint module and r ≥ 3. If the closure of a Luna stratum is normal, then "singular along its boundary" implies that the boundary has codimension at least two. This implication always holds for the principal stratum. If we know additionally that the morphism π : V → X does not contract invariant divisors 2, then the principal stratum singular along its boundary is admissible. The following proposition provides us with a wide class of admissible strata. Proposition 1. Let V be a rational G-module and H be an isotropy group of a closed G-orbit in V . For any positive integer k consider the module V ⊕k = V ⊕ · · · ⊕ V (k times) with the diagonal G-action. Then the subgroup H is an isotropy group of a closed G-orbit in V ⊕k and for k ≥ 2 the triple (G, V ⊕k, H) is admissible. Proof. Since H is an isotropy group of a closed G-orbit in V , there is a point v ∈ V such that the orbit G · v is closed and the isotropy group Stab(v) coincides with H. The G-orbit of the point (v, . . . , v) ∈ V ⊕k is contained in the diagonal ∆V ⊆ V ⊕k. Since ∆V ∼= V , this orbit is closed in ∆V and thus in V ⊕k. This proves that H is an isotropy group of a closed G-orbit in V ⊕k. Consider the diagonal action of W on (V ⊕k)H = (V H )⊕k with k ≥ 2. We claim that if at least one component vi of a point v = (v1, . . . , vk) is contained in V hHi, then v is contained in (V ⊕k)hHi. Indeed, assume that v1 ∈ V hHi. Then the W -isotropy group of v1, and thus of v, is trivial. The orbit W · v is contained in a closed subset (W · v1) × (V H )⊕(k−1). If this orbit is not closed, its closure contains a W -orbit of smaller dimension. On the other hand, any point in (W · v1) × (V H)⊕(k−1) has the trivial isotropy group. We conclude that the complement of (V ⊕k)hHi in (V ⊕k)H is contained in the set of points (v1, . . . , vk), where every vi is not in V hHi. Clearly, the latter set has codimension at least two in (V H)⊕k. (cid:3) Finally let us consider the case when the group G is finite. By Remark 1 it suffices to know when the principal stratum is admissible. Recall that a linear operator is called pseudoreflection if it has a finite order and its subspace of fixed vectors is a hyperplane. Proposition 2. Let G be a finite group and V be a G-module. Then the principal Luna stratum in X is admissible if and only if the image of G in GL(V ) does not contain pseu- doreflections. Proof. Any G-orbit in V is closed, so the principal stratum is admissible if and only if the set of points in V with non-trivial isotropy groups has codimension at least two. This means that for every g ∈ G, g 6= e, its subspace of fixed vectors has codimension at least two, or, equivalently, g is not a pseudoreflection. (cid:3) 4. The Cox ring of an admissible stratum We preserve the settings of the previous section. The quasitorus Q acts on the quotient space V H//S and this action defines an X(Q)-grading on the algebra K[V H ]S. 2This is the case if G is semisimple. TORSORS OVER LUNA STRATA 7 Theorem 2. Let G be a reductive group, V be a rational finite-dimensional G-module, and H be an isotropy group of a closed G-orbit in V . Assume that the stratum XH is admissible. Then the Q-quotient morphism p : V H//S −→ V H//W ∼= Norm(XH ) is a characteristic space over the normalization Norm(XH). Moreover, if q : V H → V H//S is the S-quotient morphism and Y := q(V hHi), then the restriction of p to Y defines a Q-torsor p : Y −→ XH, which is a universal torsor over XH . In particular, the divisor class group Cl(XH) may be identified with X(Q), and the Cox ring R(XH) is isomorphic to K[V H]S as an X(Q)-graded ring. Corollary 1. The Cox ring of an admissible Luna stratum is finitely generated. Proof. Since the group S is reductive, the algebra of invariants K[V H ]S is finitely generated. (cid:3) Corollary 2. If a Luna stratum XH is admissible and the group W is connected, then the group Cl(XH) is free and the Cox ring R(XH) is factorial. Proof. Since the group W is connected and reductive, the factor group Q is a torus, and the group X(Q) ∼= Cl(XH) is a lattice. Moreover, the commutant S is semisimple and the algebra of invariants K[V H ]S ∼= R(XH ) is factorial, see [14, Theorem 3.17]. (cid:3) Remark 2. In more general settings, the Cox ring of a variety with a free divisor class group is a unique factorization domain, see. e.g. [5, Theorem 1.14]. Corollary 3. Let (G, V, H) be an admissible triple. Then the divisor class group of the variety Norm(XH) is isomorphic to X(Q), while the Picard group of Norm(XH) is trivial. Proof. The first statement follows directly from Theorem 2. For the second one we note that the origin in V H is a W -fixed point which projects to a Q-fixed point on V H//S. Since for the characteristic space p : V H//S → Norm(XH) we have a Q-fixed point above, it follows from [5, Corollary 1.36] that the Picard group of Norm(XH ) is trivial. (cid:3) The following theorem shows that many admissible strata are singular along their boundary. Theorem 3. Let G be a reductive group, V be a rational finite-dimensional G-module, and H be an isotropy group of a closed G-orbit in V . Assume that the stratum XH is admissible and the group W is commutative. Then the stratum XH is singular along its boundary. Proof. Under our assumptions the group S is trivial and so Q coincides with W . By Theorem 2, the quotient morphism p : V H → V H //W is the characteristic space over V H//W ∼= Norm(XH ). Since the variety V H is smooth, a point x ∈ Norm(XH) is smooth if and only if the fibre p−1(x) consists of a unique free W -orbit, see [4, Proposition 4.12]. This shows that the preimage of the smooth locus in Norm(XH) coincides with V hHi and so the smooth locus of Norm(XH) is XH. This implies that the smooth locus of XH is XH too. (cid:3) 8 IVAN ARZHANTSEV Finally let us specialize Theorem 2 to the case when the group G is a torus T . Any rational T -module V admits the weight decomposition V = Mλ∈X(T ) V (λ), where V (λ) = {v ∈ V ; t · v = λ(t)v}. The multiplicity of a weight λ in V is defined as dim V (λ). Proposition 3. Let V be a rational T -module such that every weight in V has multiplicity at least two. Then every Luna stratum XH is admissible, the divisor class group of XH is a subgroup of the lattice X(T ) and the Cox ring R(XH) is polynomial. Proof. For any subgroup H in T which is an isotropy group of a closed G-orbit in V the group W is just T /H. Thus it is commutative, S = {e}, and Q = W . Since V H = L V (λ) with λH = 1, the multiplicity of any weight of the torus W in V H is at least two. To get admissibility, it suffices to show that an effective stable linear W -action satisfying this condition is strongly stable. For any vector v ∈ V H define its weight system as the set of λ's such that the component of v in V (λ) is non-zero. Then the isotropy group of v in W is trivial if and only if the weight system generates the lattice X(W ). Moreover, the orbit W · v is closed in V H if and only if the convex hull of the weight system contains zero in its relative interior. By assumptions these properties are fulfilled if all weight components of v are non-zero. As all weight multiplicities are at least two, the codimension of the complement of V hHi in V H is at least two as well, and the stratum XH is admissible. Further, Cl(XH) ∼= X(Q) = X(W ) = X(T /H), and X(T /H) is a subgroup of X(T ). Finally, R(XH) ∼= K[V H]S = K[V H] is a polynomial algebra. (cid:3) Remark 3. More generally, every Luna stratum as in Theorem 3 is a toric variety, so its Cox ring is polynomial [3]. The rays of the polyhedral cone defining the affine toric variety V H//W are obtained via Gale duality from the weights of the quasitorus W on V H , see [1, Chapter II, 2.1], while the stratum XH may be recovered as the smooth locus of V H//W . Corollary 4. Under the assumptions of Proposition 3, every Luna stratum in V //T is sin- gular along its boundary. In particular, the Luna stratification in V //T is intrinsic. Proof. The first statement follows from Theorem 3 and Proposition 3. For the second one, see [6, Lemma 3.1]. (cid:3) We get the result by checking all the conditions of Theorem 1. 5. Proof of Theorem 2 Proposition 4. Let (G, V, H) be an admissible triple. Then the action of the quasitorus Q on V H//S is strongly stable. Proof. We begin with the following general observation. TORSORS OVER LUNA STRATA 9 Lemma 1. Let F be a reductive group, S = [F, F ] be its commutant and Q be the factor group F/S. If an action of the group F on an irreducible affine variety Z is strongly stable, then the induced action of Q on Z//S is strongly stable as well. Proof. Let U be an open subset of Z as in Definition 3. Then the group S acts on U freely and all S-orbits on U are closed in Z. Let q : Z → Z//S be the quotient morphism and Y = q(U). Then Y is open in Z//S, U = q−1(Y ) and since q is surjective, the complement of Y in Z//S has codimension at least two. Moreover, for any z ∈ U the image q(F · z) is a closed Q-orbit of the point q(z) with trivial isotropy group. This proves that the group Q acts on Y freely and all Q-orbits on Y are closed in Z//S. (cid:3) Let us return to the proof of the proposition. By definition of an addmisible triple the action of the group W on V H is strongly stable, and Lemma 1 implies that the induced action of Q on V H//S is strongly stable as well. (cid:3) Proposition 5. Let (G, V, H) be an admissible triple. Then the Q-variety V H//S is Q- factorial. Proof. Take any Q-invariant Weil divisor D on V H//S. Without loss of generality we may assume that D is effective. The preimage q∗(D) of D under the quotient morphism q : V H → V H//S is a W -invariant divisor on V H. Since V H is an affine space, any such divisor is a principal divisor of some regular W -semiinvariant function. Such a function is S-invariant, thus it may be considered as a regular function on V H//S having D as its principal divisor. (cid:3) Since K[Norm(XH)]× ∼= K[V H//W ]× = K×, we get the first statement of Theorem 2 from Theorem 1. The second one follows from the fact that the complement of XH in Norm(XH) has codimension at least two and [1, Lemma 5.1.2]. The proof of Theorem 2 is completed. 6. Examples In this section we consider several examples. They illustrate Theorem 2 and show that the condition for a stratum to be admissible is essential. Example 1. Let G = SL(2) and V be the tangent algebra of G considered as adjoint G- module. This module has conjugacy classes of subgroups which are isotropy groups of closed G-orbits, namely the maximal torus T and the group SL(2) itself. Since V T is an affine line and the complement of V hT i in V T is a point, the triple (SL(2), V, T ) is not admissible. In this case the quotient space X := V //G is an affine line A1, and the stratum XT coincides with A1 \ {0}. The group W = NG(T )/T has order two, so S = {e} and Q = W . The divisor class group Cl(A1 \ {0}) is trivial, thus does not coincide with X(Q), while K[A1 \ {0}]× 6= K× and the Cox ring of XT is not defined. Further, the principal stratum for V ⊕ V corresponds to H = {±E}. One easily checks that it is also not admissible. In this case W = PSL(2) = S, Q is trivial, and so all conclusions of Theorem 2 hold. (V ⊕ V )H//S = (V ⊕ V )//G ∼= A3, Example 2. Let us take an arbitrary connected semisimple group G and the adjoint G- module V . The principal stratum for V corresponds to H = T , where T is a maximal torus in G, and V H is a Cartan subalgebra. In this case W is the (finite) Weyl group associated 10 IVAN ARZHANTSEV with the group G and Q is its maximal commutative factor. The principal stratum is not admissible, but starting from k = 2 the stratum for V ⊕k associated with H = T is admissible. Its divisor class group does not depend on k and is isomorphic to X(Q) ∼= Q. In order to describe the Cox ring one should calculate the algebra of invariants of the diagonal action of the group S = [W, W ] on the direct sum of Cartan subalgebras. It is important to note that by [15, Theorem 8.1] the principal stratum for V ⊕k is singular along its boundary for any k ≥ 2 provided G has no simple factors of rank 1. Example 3. Take G = SL(n) and V = Kn ⊕ (Kn)∗, where Kn is the tautological SL(n)- module and (Kn)∗ is its dual. The principal stratum for V corresponds to H ∼= SL(n−1), and W = Q is a one-dimensional torus. Since V //G ∼= A1, the principal stratum is not admissible. But starting from k = 2 the subgroup H defines an admissible statum for V ⊕k whose divisor class group is isomorphic to Z. By Theorem 3, this stratum is singular along its boundary. Let us show that the Cox ring of an admissible Luna stratum is not always factorial, cf. Corollary 2. The example below is taken from [2, Section 3], where the Cox ring of the quotient space of an arbitrary finite linear group is described. Example 4. Let V = K2 and G be the quaternion group Q8 =(cid:26)±E, ±(cid:18)i 0 −i(cid:19) , ±(cid:18) 0 0 −1 0(cid:19) , ±(cid:18)0 i i 0(cid:19)(cid:27) , 1 where i2 = −1. By Proposition 2, the principal Luna stratum in X corresponding to H = {e} is admissible. As S = [Q8, Q8] = {±E}, the Cox ring of the principal stratum K[V ]S is the ring of functions on the two-dimensional quadratic cone. This ring is not factorial. Note that Example 2 also leads to non-factorial Cox rings. Say, this is already the case for G = SL(3) and k = 2. But the description of the Cox ring here is more complicated. Acknowledgement The preparation of this text was inspired by the workshop "Torsors: theory and appli- cations", January 10 -- 14, 2011, International Centre for Mathematical Sciences, Edinburgh, organized by Victor Batyrev and Alexei Skorobogatov. The final version was written during a stay of the author at the Institut Fourier, Grenoble. He wishes to thank these institutions for generous support and hospitality. Special thanks are due to the referee for the careful reading, useful remarks and suggestions. References [1] I.V. Arzhantsev, U. Derenthal, J. Hausen, and A. Laface. Cox rings. arXiv:1003.4229. [2] I.V. Arzhantsev and S.A. Gaifullin. Cox rings, semigroups and automorphisms of affine algebraic varieties. Mat. Sbornik 201 (2010), no. 1, 3 -- 24 (Russian); English transl.: Sbornik: Math. 201 (2010), no. 1, 1 -- 21. [3] D.A. Cox. The homogeneous coordinate ring of a toric variety. J. Alg. Geom. 4 (1995), no. 1, 17 -- 50. [4] J. Hausen. Cox rings and combinatorics II. Moscow Math. J. 8 (2008), no. 4, 711 -- 757. [5] J. Hausen. Three letures on Cox rings. This volume. [6] J. Kuttler and Z. Reichstein. Is the Luna stratification intrinsic? Ann. Inst. Fourier (Grenoble) 58 (2008), no. 2, 689 -- 721. [7] L. Le Bruyn and C. Procesi. Semisimple representations of quivers. Trans. Amer. Math. Soc. 317 (1990). no. 2, 585 -- 598. [8] D. Luna. Slices ´etales. Bull. Soc. Math. France, Memoire 33 (1973), 81 -- 105. TORSORS OVER LUNA STRATA 11 [9] D. Luna. Adh´erences d'orbite et invariants. Invent. Math. 29 (1975), 231 -- 238. [10] D. Luna and R.W. Richardson. A generalization of the Chevalley restriction theorem. Duke Math. J. 46 (1979), no. 3, 487 -- 496. [11] Y. Matsushima. Espaces homog´enes de Stein des groupes de Lie complexes. Nagoya Math. J. 16 (1960), 205 -- 218. [12] A.L. Onishchik. Complex hulls of compact homogeneous spaces. Dokl. Akad. Nauk SSSR 130 (1960), no. 4, 726 -- 729 (Russian); English Transl.: Soviet Math. Dokl. 1 (1960), 88 -- 91. [13] V.L. Popov. Criteria for the stability of the action of a semisimple group on a factorial manifold. Izv. Akad. Nauk SSSR 34 (1970), 523 -- 531 (Russian); English transl.: Math. USSR, Izv. 4 (1970), 527 -- 535. [14] V.L. Popov and E.B. Vinberg. Invariant Theory. Algebraic Geometry IV, Encyclopaedia Math. Sciences, vol. 55, Springer-Verlag Berlin, 1994, pp. 123 -- 278. [15] R.W. Richardson. Conjugacy classes of n-tuples in Lie algebras and algebraic groups. Duke Math. J. 57 (1988), no. 1, 1 -- 35. [16] G.W. Schwarz. Lifting smooth homotopies of orbit spaces. Inst. Hautes ´Etudes Sci. Publ. Math. 51 (1980), 37 -- 135. [17] D.A. Shmel'kin. On algebras of invariants and codimension 1 Luna strata for nonconnected groups. Geom. Dedicata 72 (1998), no. 2, 189 -- 215. [18] A.N. Skorobogatov. Torsors and rational points. Cambridge Tracts in Mathematics 144, Cambridge University Press, Cambridge, 2001, 187 pp. Department of Higher Algebra, Faculty of Mechanics and Mathematics, Moscow State University, Leninskie Gory 1, GSP-1, Moscow, 119991, Russia E-mail address: [email protected]
1803.01068	3	1803	2019-09-26T17:16:10	Complexity of deciding whether a tropical linear prevariety is a tropical variety	[ "math.AG" ]	We give an algorithm, with a singly exponential complexity, deciding whether a tropical linear prevariety is a tropical linear variety. The algorithm relies on a criterion to be a tropical linear variety in terms of a duality between the tropical orthogonalization $A^\perp$ and the double tropical orthogonalization $A^{\perp \perp}$ of a subset $A$ of the vector space $({\mathbb R} \cup \{ \infty \})^n$. We also give an example of a countable family of tropical hyperplanes such that their intersection is not a tropical prevariety.	math.AG	math	COMPLEXITY OF DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY DIMA GRIGORIEV AND NICOLAI VOROBJOV Abstract. We give an algorithm, with a singly exponential complexity, de- ciding whether a tropical linear prevariety is a tropical linear variety. The algorithm relies on a criterion to be a tropical linear variety in terms of a duality between the tropical orthogonalization A⊥ and the double tropical or- thogonalization A⊥⊥ of a subset A of the vector space (R ∪ {∞})n. We also give an example of a countable family of tropical hyperplanes such that their intersection is not a tropical prevariety. Introduction In this paper we use the operation A⊥ of tropical orthogonalization, applied to a subset A of a vector space (R ∪ {∞})n, and its iteration, A⊥⊥, to formulate a criterion and an algorithm, deciding whether a tropical linear prevariety is a tropical linear variety. General concepts of tropical algebra can be found in [20, 25, 29]. Specific ques- tions of tropical linear algebra were considered in [6, 7, 30, 31]. In Section 1 we list basic definitions, including the concept of a tropical linear hull of a subset in (R ∪ {∞})n. We recall a theorem, proved in [3, 9], stating that any tropical linear prevariety is the tropical linear hull of a finite set of vectors. We list some properties of double orthogonalization, in particular, we prove that A⊥⊥ is the minimal tropical linear prevariety containing the finite set A. Further in Section 1, we recall a theorem, implicitly proved in [30] (see also [9] and [27]), stating that for two mutually complementary and orthogonal linear subspaces P and Q of the vector space (C((t1/∞)))n over Puiseux series, there exists a finite set A ⊂ (R ∪ {∞})n such that tropicalizations of P and Q coincide with A⊥ and A⊥⊥ respectively. This theorem is essential for the main result in Section 3, and for reader's convenience we present its new short proof. Sections 2 and 3 contain descriptions of some algorithms and estimates of their complexities. Each of these algorithms can, in principle, be modelled by a Turing machine, and complexity is understood in terms of its steps, i.e., as bit complexity. In Section 2 we describe an algorithm, with singly exponential complexity in the number of variables, which for a given (classical) algebraic set V ⊂ (C((t1/∞)))n and a point x ∈ Qn decides whether or not x belongs to the tropicalization of V , and, if it does, produces a lifting of x in V . This algorithm serves as a key subroutine for algorithms in Section 3, but also is of independent interest. An algorithm for the same problem follows from [18] and apparently has doubly exponential complexity bound in the number of variables. Section 3 contains the main result of the paper: a criterion and deciding algo- rithms for a tropical linear prevariety to be a tropical linear variety. We propose 2010 Mathematics Subject Classification 14T05 1 2 DIMA GRIGORIEV AND NICOLAI VOROBJOV two deciding algorithms. One has the complexity exponential in bit-sizes of ratio- nal generators of the tropical linear hull representing tropical linear prevariety and singly exponential in the number of variables. Another algorithm is polynomial in bit-size of generators. Its complexity also depends on the complexity of computing of a tropical basis (see [2, 17, 22, 23]) of a system of (classical) multivariate poly- nomial equations. Apparently, the latter complexity is doubly exponential in the number of variables, in which case the complexity of our second algorithm is also doubly exponential in the number of variables. We also describe an algorithm which for a given tropical linear variety A⊥ pro- duces a linear subspace P whose tropicalization coincides with A⊥. This algorithm has a singly exponential complexity in the number of variables. In Section 4 we give an example of a countable family of tropical hyperplanes in (R∪{∞})6 such that their intersection is not a tropical prevariety. This strengthens examples in [8] (example of T. Theobald) and [16] about countable intersections of non-linear tropical hypersurfaces. The extended abstract of this paper appeared in [15]. 1. Preliminaries 1.1. Tropical linear prevarieties and tropical linear hulls. We use the nota- tion R∞ for R ∪ {∞}. We assume that for all a ∈ R the rules a < ∞, a + ∞ = ∞, ∞ + ∞ = ∞, and, for positive a, a · ∞ = ∞ hold. The element ∞ is a "tropical zero", being the neutral element with respect to taking minimum. ∞ is the set of all points (x1, . . . , xn) ∈ Rn Definition 1.1 ([7], [25]). For a given (a1, . . . , an) ∈ Rn Rn an} has at least two minimal elements. A tropical linear prevariety in Rn intersection of a finite number of tropical hyperplanes. ∞, a tropical hyperplane in ∞ at which the set {x1 + a1, . . . , xn + ∞ is the Remark 1.2. The point (∞, . . . , ∞) belongs to every tropical linear prevariety. A tropical hyperplane according to Definition 1.1 corresponds to the notion of a codimension one linear subspace in classical linear algebra. It can be identi- fied with a special case, when an+1 = ∞, of a more general notion of a tropi- cal hyperplane, defined as a set of all points (x1, . . . , xn) ∈ Rn ∞ at which a set {x1 + a1, . . . , xn + an, an+1}, where ai ∈ R∞, 1 ≤ i ≤ n + 1, has at least two minimal elements. Definition 1.3 ([7], [25]). Vectors v = (v1, . . . , vn), a = (a1, . . . , an) ∈ Rn ∞ are called tropically orthogonal if among numbers vi + ai, 1 ≤ i ≤ n there are at least two minimal. Note that (∞, . . . , ∞) is tropically orthogonal to every vector ∞ denote by A⊥ the set of all a ∈ Rn vectors in Rn ∞. For a set of vectors A = {a1, . . . , ak} ⊂ Rn ∞ tropically orthogonal to each ai, 1 ≤ i ≤ k. It is clear that {a}⊥ is a tropical hyperplane for a vector a ∈ Rn ∞, while A⊥ is a ∞ is finite. Conversely, every tropical linear tropical linear prevariety when A ⊂ Rn prevariety in Rn ∞ coincides with A⊥ for a suitable finite set of vectors A ⊂ Rn ∞. Definition 1.4 ([7], [25]). For a finite set of vectors A = {a1, . . . , ak} ⊂ Rn its tropical linear hull Trophull(A) as the set of all vectors in Rn ∞ of the kind ∞ define min 1≤i≤k {ti1n + ai}, DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 3 where t1, . . . , tk are arbitrary elements in R∞, min1≤i≤k denotes the component- wise minimum of a set of vectors, and 1n = (1, . . . , 1) is the unit vector in Rn. For an arbitrary subset X ⊂ Rn ∞ define Trophull(X) as the union of sets Trophull(A) over all finite subsets A ⊂ X. Note that Trophull(A) is the direct tropical analogy of the concept of linear hull of a finite set of vectors in classical linear algebra. It always contains the point (∞, . . . , ∞) ∈ Rn ∞, because all ti, 1 ≤ i ≤ k can be chosen to be ∞. Definition 1.5. For every partition {i1, . . . , ip} ∪ {ip+1, . . . , in} of {1, . . . , n} a chart in Rn ∞ is an open convex polyhedron ∞ is the union of all 2n pair-wise disjoint charts. Ci1,...,ip := {xi1 = · · · = xip = ∞} ∩ {xip+1 < ∞} ∩ · · · ∩ {xin < ∞} ⊂ Rn ∞. Clearly, Rn One can extend the standard concepts of a convex polyhedron and a finite poly- hedral complex to the case of the subsets of the space Rn ∞ (see [9]). Restriction of a convex polyhedron P ⊂ Rn ∞ to a chart Ci1,...,ip coincides with a usual convex poly- hedron in Rn−p translated by a vector in {0, ∞}n−p with ∞ in positions i1, . . . , ip. Hence, P is a finite union of translated usual convex polyhedra, and we define the dimension dim(P ) as the maximum of the dimensions of restrictions of P to all charts. The dimension of a finite polyhedral complex is defined as the maximum of dimensions of its convex polyhedra. The following theorem directly follows from [9, Theorem 1] (part (2) of the theorem, except the complexity bound, was proved earlier in [3, Proposition 2]). Let A = {a1, . . . , ak} ⊂ Rn ∞ be a set of vectors. Theorem 1.6. (1) The set Trophull(A) is a union of all convex polyhedra of a polyhedral complex in Rn ∞. (2) For the tropical linear prevariety A⊥ ⊂ Rn ∞ there exists a finite set of vec- ∞ such that A⊥ = Trophull({b1, . . . , bN }). More- tors {b1, . . . , bN } ⊂ Rn over, there is an algorithm which for a given set {a1, . . . , ak} of vectors in Qn ∞, with bit-sizes of coordinates not exceeding L, computes the set {b1, . . . , bN } ⊂ Qn ∞. The complexity of this algorithm is polynomial in L and nk. Bit-sizes of coordinates of the computed vectors b1, . . . , bN do not exceed L + log k, while N = O(nk). Corollary 1.7. Every tropical linear prevariety A⊥ ⊂ Rn polyhedra of a polyhedral complex in Rn ∞. ∞ is the union of all convex Theorem 1.6, (2) states that any tropical linear prevariety A⊥ = {a1, . . . , an}⊥ ⊂ ∞ coincides with the tropical linear hull of a finite subset of its vectors. This is Rn not necessarily true for the restriction A⊥ ∩ Rn (see Example 1.11 below). 1.2. Dual tropical linear prevarieties. We extend the operation X ⊥, intro- duced in Definition 1.3, so that it can be applied to arbitrary (not necessarily finite) subsets X ⊂ Rn ∞ orthogonal to each a ∈ X. We will use notations X ⊥⊥ := (X ⊥)⊥ and X ⊥⊥⊥ := (X ⊥⊥)⊥. ∞. Namely, denote by X ⊥ the set of all vectors in Rn Remark 1.8. Observe that by the definition, for a finite subset A ⊂ Rn ∞, the set A⊥⊥ is an intersection of an infinite number of tropical hyperplanes in Rn ∞. As we will show in Section 4 below, not every intersection of even countable number of tropical hyperplanes is a union of cells of a finite polyhedral complex, let alone 4 DIMA GRIGORIEV AND NICOLAI VOROBJOV tropical linear prevariety. However, in the special case of a finite A, the set A⊥⊥ is a tropical linear prevariety (Proposition 1.10). Lemma 1.9. For any subset X ⊂ Rn ∞ we have: (1) Trophull(X) ⊂ X ⊥⊥; (2) X ⊥ = X ⊥⊥⊥. Proof. (1) Directly follows from definitions. (2) Inclusions X ⊂ X ⊥⊥ and X ⊥ ⊂ X ⊥⊥⊥ are trivial. The first of these inclusions implies that every x ∈ X ⊥⊥⊥ is orthogonal to every y ∈ X. Hence x ∈ X ⊥. (cid:3) Proposition 1.10. Let A be a finite set of vectors in Rn ∞. Then A⊥⊥ is the minimal (with respect to the subset relation) tropical linear prevariety containing A. Proof. By Theorem 1.6, A⊥ = Trophull({b1, . . . bN }) for some b1, . . . bN ∈ Rn hence A⊥⊥ = (Trophull({b1, . . . bN }))⊥. According to [6, 7], ∞, (Trophull({b1, . . . bN }))⊥ = {b1, . . . bN }⊥, which implies that A⊥⊥ is a tropical linear prevariety. Let C be any tropical linear prevariety containing A. Then C = {c1, . . . , cM }⊥ ∞. Since c1, . . . , cM ∈ A⊥, we get the inclusion for some vectors c1, . . . , cM ∈ Rn A⊥⊥ ⊂ {c1, . . . , cM }⊥ = C. Because A ⊂ A⊥⊥ (cf. proof of Lemma 1.9 (2)), we conclude that A⊥⊥ is the minimal tropical linear prevariety containing A. (cid:3) Lemma 12 in [14] (also Theorem 4.2 in [6]) implies that dim(A⊥)+dim(A⊥⊥) ≥ n ∞. The following example shows that the equality dim(A⊥) = for finite A ⊂ Rn dim(A⊥⊥) = n − 1 is possible for some A ⊂ Rn, which does not happen in classical linear algebra. Example 1.11 (cf. [14]). Let A0 = {a1, . . . , an−1} ⊂ Rn, where ai = (1, . . . , 1, 0 , 1, . . . , 1, 0, 0) for 1 ≤ i ≤ n − 2 and an−1 = (1, . . . , 1, 0, 0). {z i } It is easy to see that every vector x = (x1, . . . , xn) ∈ A⊥ ∞ should have minimal elements xn−1, xn, and, conversely, every vector x with minimal elements xn−1, xn is in A⊥ 0 ⊂ Rn 0 . Therefore, A⊥ 0 = {t1n + (c1, . . . , cn−2, 0, 0) for all 0 ≤ ci ∈ R∞ and t ∈ R∞}. From the example, we see that every vector y = (y1, . . . , yn) ∈ A⊥⊥ 0 ⊂ Rn ∞ should have minimal elements yn−1, yn, and, conversely, every vector y with mini- mal elements yn−1, yn is in A⊥⊥ 0 ) = dim(A⊥⊥ 0 . Note that dim(A⊥ . It follows that A⊥⊥ 0 = A⊥ ) = n − 1. 0 0 For another property of this example, observe that A⊥ 0 = Trophull({b1, . . . , bn−1}), where bi = (∞, . . . , ∞, 0 , ∞, . . . , ∞, 0, 0) for 1 ≤ i ≤ n − 2 and bn−1 = (∞, . . . , ∞, 0, 0). {z i } DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 5 On the other hand, take n = 3 in the example. For any finite set {v1, . . . , vN } ⊂ 0 ∩ R3 we have A⊥ A⊥ 0 ∩ R3 6= Trophull({v1, . . . , vN }) ∩ R3. Indeed, assume, for contradiction, that there is a set of vectors V = {v1, . . . , vN } such that A⊥ 0 ∩ R3 = Trophull(V ) ∩ R3. Without loss of generality, assume that vj = (cj, 0, 0) for each 1 ≤ j ≤ N , where cj ≥ 0. Denoting M := max1≤j≤N {cj}, we consider the vector (M + 1, 0, 0) ∈ A⊥ 0 ∩ R3 and prove that it is not in Trophull(V ). Indeed, if (1.1) (M + 1, 0, 0) = min 1≤j≤N (tj 13 + vj) for some t1, . . . , tN , then, considering either second or third coordinate, we conclude that every tj ≥ 0 and at least one of them, tj0 = 0. It follows that the first coordinate of the vector min1≤j≤N (tj 13 + vj) in (1.1) is at most cj0 , which is less than M + 1. 1.3. Tropicalization of linear subspaces. Let K be an algebraically closed field of characteristic 0 and F denote the field K((t1/∞)) of Puiseux series over K. For an element y ∈ F different from 0, let val(y) ∈ Q denote the valuation of the element y in F, i.e., the power in the lowest term of the Puiseux series y. Separately define val(0) = ∞. Definition 1.12 (cf. [25]). Let f := a1x1 + · · · + anxn, where 0 6= ai ∈ F for all 1 ≤ i ≤ n. The formal tropicalization of the hyperplane {f = 0} ⊂ Fn is the tropical hyperplane, Tropf({f = 0}) ⊂ Rn, defined by the set {y1+val(a1), . . . , yn+val(an)} (see Definition 1.1). By Kapranov's Theorem [25, Theorem 3.1.3], Tropf({f = 0}) coincides with the (Euclidean) closure in Rn of the countable set {(val(x1), . . . , val(xn)) (x1, . . . , xn) ∈ {f = 0} ∩ (F \ {0})n}. The following definition is "dual" to Definition 1.5. Definition 1.13. For every partition {i1, . . . , ir} ∪ {ir+1, . . . , in} of {1, . . . , n} a chart in Fn is a set Di1,...,ir := {xi1 = · · · = xir = 0} ∩ {xir+1 6= 0} ∩ · · · ∩ {xin 6= 0} ⊂ Fn. For any X ⊂ Fn we obviously have X = [ {i1,...,ir } (X ∩ Di1,...,ir ), where the union is taken over all subsets {i1, . . . , ir} of {1, . . . , n}. Definition 1.14 (cf. [25]). The tropicalization Trop(X ∩ Di1,...,ir ) of X ∩ Di1,...,ir is the set of all points (y1, . . . , yn) ∈ Rn ∞ such that yi1 = · · · = yir = ∞ and (yir+1 , . . . , yin ) belongs to the Euclidean closure in Rn−r of the set {(val(xir+1 ), . . . , val(xin )) (x1, . . . , xn) ∈ X ∩ Di1,...,ir }. The tropicalization Trop(X) of X is defined as S{i1,...,ir} Trop(X ∩ Di1,...,ir ). Remark 1.15. Definition 1.14 immediately implies that for any two sets X, Y ⊂ Fn there is the inclusion Trop(X ∩ Y ) ⊂ Trop(X) ∩ Trop(Y ). The inverse inclusion ⊃ is not generally true even for linear subspaces. 6 DIMA GRIGORIEV AND NICOLAI VOROBJOV Let P ⊂ Fn be a linear subspace, with dim(P ) = d, and z1, . . . , zd ∈ P be a basis of P . Recall that Plucker coordinates of P in the Grassmanian Gr(d, Fn) are all (d × d)-minors pj1,...,jd of the matrix with rows z1, . . . , zd, corresponding to the columns 1 ≤ j1 < · · · < jd ≤ n. Any z = (z1, . . . , zn) ∈ P satisfies the relation (1.2) (−1)ipj1,...,ji−1,ji+1,...,jd+1zji = 0, X 1≤i≤d+1 for every subset of columns 1 ≤ j1 < · · · < jd+1 ≤ n. Note that the relations in (1.2) are independent of the choice of a basis in P . Denote the set of points z satisfying (1.2) by Pj1,...,jd+1 . The following statement is a strengthening for R∞ of [30, Proposition 4.2] (also [25, Theorem 4.3.17]), and can be proved analogously. Lemma 1.16. (1.3) Trop(P ) = \ j1,...,jd+1 Trop(Pj1,...,jd+1). Let Q ⊂ Fn be a linear subspace orthogonal to P with dim(Q) = n−d. According to [2, 30], the tropicalizations Trop(P ) and Trop(Q) are tropically orthogonal, with dim(Trop(P )) = d and dim(Trop(Q)) = n − d. The following theorem implicitly appears in [30]. Formulated in the language of valuated matroids, an analogous statement can be found in [27]. To make our ex- position closed, for reader's convenience, we give another proof, using the technique of Plucker relations. Theorem 1.17. There is a finite subset A ⊂ Rn Trop(Q) = A⊥⊥. ∞ such that Trop(P ) = A⊥ and Proof. Consider the relations (1.2) for the subspace P . For every subset 1 ≤ j1 < · · · < jd+1 ≤ n introduce a vector pj1,...,jd+1 ∈ Fn such that its ji-coordinate is (−1)ipj1,...,ji−1,ji+1,...,jd+1 for 1 ≤ i ≤ d + 1, and all the rest of coordinates are equal to 0. Then every vector in Pj1,...,jd+1 is orthogonal to the vector pj1,...,jd+1, hence every vector in T{j1,...,jd+1} Pj1,...,jd+1 is orthogonal to vectors pj1,...,jd+1 for all subsets 1 ≤ j1 < · · · < jd+1 ≤ n. By val(pj1,...,jd+1) denote the vector in Rn ∞ obtained from pj1,...,jd+1 by taking val(·) of all non-zero coordinates and replacing all zero coordinates by ∞. As A take the set of all vectors val(pj1,...,jd+1) ∈ Rn ∞ for all 1 ≤ j1 < · · · < jd+1 ≤ n. By Definition 1.12, A⊥ = Tropf(Pj1,...,jd+1 ∩ DI ), \ [ I 1≤j1<···<jd≤n where the union is taken over all I ⊂ {1, . . . , n} and DI are the corresponding charts in Fn. By Kapranov's Theorem, Tropf(Pj1,...,jd+1 ∩ DI ) = Trop(Pj1,...,jd+1 ∩ DI ), while, by Definition 1.14, SI Trop(Pj1,...,jd+1 ∩ DI ) = Trop(Pj1,...,jd+1). Then, by Lemma 1.16, A⊥ = Trop(P ). Since the linear subspaces P and Q are orthogonal, for every y ∈ Q we have P ⊂ {y·z = 0}. Tropicalizing both sides of this inclusion, we conclude that Trop(Q) and Trop(P ) are tropically orthogonal. Hence, Trop(Q) ⊂ (Trop(P ))⊥ = A⊥⊥. Vectors pj1,...,jd+1 lie in Q, since they are orthogonal to P . It follows that A ⊂ Trop(Q). By Proposition 1.10, A⊥⊥ ⊂ Trop(Q), since Trop(Q) is a tropical linear prevariety. We conclude that A⊥⊥ = Trop(Q). (cid:3) DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 7 Corollary 1.18. Let P, Q ⊂ Fn be orthogonal complements of one another and Trop(P ) = A⊥. Then Trop(Q) = A⊥⊥. Proof. By Theorem 1.17, there is a finite subset B ⊂ Rn ∞ such that Trop(P ) = B⊥ and Trop(Q) = B⊥⊥. Since Trop(P ) = A⊥, we get A⊥ = B⊥, hence Trop(Q) = A⊥⊥. (cid:3) Remark 1.19. Corollary 1.18 implies, in particular, that in Example 1.11 the trop- ical linear prevariety A⊥ 0 is not a tropical linear variety. 2. Testing membership in a tropical variety and computing a lifting For computational purposes, from now on we will assume that the field K = Q, thus F := Q((t1/∞)) is the field of formal Puiseux series in a variable t with complex algebraic coefficients. Definition 2.1. Let x ∈ (Q ∪ {∞})n and v ∈ Fn such that Trop(v) = x. Then v is called a lifting of x. In this section we describe an algorithm for testing membership of a given point x ∈ Qn in a tropical variety. If x does belong to the variety, the algorithm computes its lifting. The complexity of the algorithm is singly exponential in n. For elements α1, . . . , αm ∈ F that are algebraic over Q(t) let η be the primi- tive element of the algebraic extension Q(t)(α1, . . . , αm) of the field Q(t), hence Q(t)(α1, . . . , αm) = Q(t)[η] (see [24, Theorem 4.6]). The primitive element η is de- fined (up to a conjugacy class) by its minimal (irreducible) polynomial ψ ∈ Z[t][Z], i.e., ψ(η) = 0. Then αi = ζi(η) for a polynomial ζi ∈ Q(t)[Z], for every 1 ≤ i ≤ m. Consider (classical) polynomials f1, . . . , fk ∈ F[X1, . . . , Xn], where the set of all coefficients α1, . . . , αm ∈ F of f1, . . . , fk is represented, as above, via a primitive element. Let V (f1, . . . , fk) := {f1 = · · · = fk = 0} ⊂ Fn. We describe an algorithm, with singly exponential complexity, which for given f1, . . . , fk and a point x = (x1, . . . , xn) ∈ Qn decides whether or not x belongs to Trop(V (f1, . . . , fk)) ⊂ Rn. Moreover, if x ∈ Trop(V (f1, . . . , fk)) the algorithm produces a lifting of x (see Definition 2.1). According to [13], this computational problem is NP-hard. Without loss of generality, we assume that x = 0, performing if necessary the coordinate transformation Xi → Xit−xi in polynomials fj, 1 ≤ j ≤ k. We keep the same notation, fj, 1 ≤ j ≤ k, for polynomials appearing after the change of coordinates. The rest of the algorithm consists of three parts. Firstly, it reduces testing whether 0 belongs to Trop(V (f1, . . . , fk)) to testing whether 0 lies in the tropical- ization of one of 0-dimensional algebraic sets. Next, the algorithm does the latter testing. Finally, in the case of positive result, it constructs the lifting of 0. 2.1. Reduction to 0-dimensional algebraic sets. Applying algorithms from [4, 11], represent V (f1, . . . , fk) as the finite union of its irreducible over Q(t) com- ponents, V (f1, . . . , fk) = Sν Vν. Each irreducible component is represented in two forms: by a system of equations with coefficients in Q(t), and by a generic point (a point such that the field generated by its coordinates has the transcendence degree, over the field generated by the coefficients of the equations, equal to the dimension of the component). More precisely, a generic point of an irreducible component Vν, with dim(Vν ) = s, is represented as follows. Among coordinates X1, . . . , Xn a tran- scendence basis is chosen, let it be X1, . . . , Xs. Then coordinates Xs+1, . . . , Xn are 8 DIMA GRIGORIEV AND NICOLAI VOROBJOV algebraic over Q(X1, . . . , Xs)(t), and algorithms from [4, 11] describe the primitive element µ of the extension Q(X1, . . . , Xn)(t) as µ = λs+1Xs+1 +· · ·+λnXn for posi- tive integers λs+1, . . . , λn, and by its minimal polynomial χ ∈ Q(X1, . . . , Xs)(t)[Z]. Each coordinate Xi, s + 1 ≤ i ≤ n is then represented by a polynomial expression in µ with coefficients in Q(X1, . . . , Xs)(t). Let Y0, . . . , Yn be new variables. Fix a component Vν with dim(Vν ) > 0, and consider the algebraic set eVν defined by the same equations as Vν over the field Q(Y0, Y1, . . . , Yn)((t1/∞)). Let fWν := eVν ∩  Y0 + X 1≤i≤n YiXi = 0  . Then [18, Proposition 4.6] (see also [28, Theorem 2.7.8], [19, Theorem 3.5]) implies Let dim(Vν ) = s. Introducing new sets of variables, Yp,q, 0 ≤ q ≤ n, where that if 0 ∈ Trop(Vν ), then 0 ∈ Trop(fWν ). The converse statement is obvious. 1 ≤ p ≤ s, and adding equations Yp,0 +P1≤i≤n Yp,iXi = 0 to the system of equa- tions defining eVν , we will obtain a system of equations defining a zero-dimensional algebraic set eUν over the field Q({Yp,q}1≤p≤s,0≤q≤n)((t1/∞)). Find all points in eUν and only if for some point v ∈ eUν its tropicalization is 0. We now describe how the using algorithms from [4, 11]. The algorithm will conclude that 0 ∈ Trop(Vν ) if algorithm checks the latter condition. 2.2. Testing existence of the lifting of a point in 0-dimensional algebraic set. Algorithms from [4, 11] represent each point in v = (v1, . . . , vn) ∈ eUν as follows. For every 1 ≤ i ≤ n the coordinate vi is represented as vi = ξi(θ), where θ is a root of an irreducible polynomial ϕ ∈ Z({Yp,q}p,q)(t)[Z], with degZ (ϕ) = r, while the expression ξi(θ) is of the form 1 ξi(θ) = ci,ℓ θℓ, bi X 0≤ℓ<r where bi, ci,ℓ ∈ Z[{Yp,q}p,q](t). We can represent ξi(θ) in a form ξi(θ) = ai + R(t), where R(t) = o(1) is the series of terms containing t in positive degree, while ai is the sum of terms of degrees in t at most 0. More precisely, let α be the smallest integral (possibly negative) degree in t of the rational in t coefficients in the expression ξi. Let β be the degree of the first term in the Puiseux series θ = θ(t). For a positive integer γ ≥ β, let Θ be the sum of all consecutive terms in the Puiseux series θ starting with the term of degree β and ending with the term of the smallest degree not less than γ. If β > 0, then choose γ to be the smallest integer such that α + γ ≥ 0. Then ξi(Θ) contains all terms in ξi(θ) having non-positive degrees in t. If β ≤ 0, then Θr−1 contains the term with the degree not less than α + (r − 2)β + γ, hence choosing γ to be the smallest integer such that α + (r − 2)β + γ ≥ 0, we guarantee that ξi(Θ) contains all terms in ξi(θ) having non-positive degrees in t. Using the algorithm from [5], find sum Θ of all first consecutive terms in the Puiseux expansion of θ in t up to the term of the smallest degree not less than γ. Substituting Θ into ξi, we compute the sum ai of terms in ξi(θ) of degrees in t at most 0. Let for each i, among the terms in ai, the lowest degree in t be 0, i.e., ai is DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 9 independent of t. Then we conclude that 0 belongs to Trop(V (f1, . . . , fk)) ⊂ Rn. If among the terms in ai the lowest degree in t is different from 0 for some i, then Trop(v) 6= 0. 0 6∈ Trop(Vν ). If the latter condition holds for every point v ∈ eUν, then 2.3. Constructing the lifting of a point. To find a lifting of 0 (in case of ai independent of t for all i), observe that for each i the element ai belongs to a finite extension of the field Q({Yp,q}p,q), and is defined by a polynomial expression via the primitive element of this extension, while the primitive element is defined by its minimal polynomial (cf. representation of algebraic elements α1, . . . αm at the beginning of this section). All coefficients in the polynomial expressions and the minimal polynomial belong to Q({Yp,q}p,q). Choose a point y ∈ Zs(n+1) such that (1) after the substitution y in variables {Yp,q}p,q neither of the denominators of these coefficients vanish; (2) after the substitution y in variables {Yp,q}p,q neither of the numerators at leading terms of minimal polynomials vanish. Coordinates of y can be taken to be non-negative integers not exceeding the degrees of the polynomials mentioned in (1), (2). Thus, one can estimate the complexity of choosing point y by the bound on the degrees of these polynomials which is provided below. The condition (2) implies that after the substitution of y in (Y1, . . . , Ys(n+1)) each minimal polynomial may become reducible but its degree does not change. Recall that each coordinate vi in v = (v1, . . . , vn) ∈ eUν is represented by polyno- mials ϕ, ξi ∈ Q({Yp,q}p,q)(t)[Z]. Substituting in coefficients of these polynomials the vector of variables (Y1, . . . , Ys(n+1)) by y, we obtain a representation of a point in (Q(t))n which is a lifting of 0. This ends the description of the algorithm. 2.4. Complexity of the algorithm. Let d = max1≤j≤k degX1,...,Xn (fj) and δ be the maximum among the degrees in t of polynomials ψ, ζi, which define coefficients αi of polynomials fj. Let M be the maximum among the common denominator of rational numbers x1, . . . , xn and the absolute values of their numerators. By the Bezout Theorem, the degree of the algebraic set V := V (f1, . . . , fk) does not exceed D := dn. After the change of coordinates Xi → Xit−xi in polynomials fj, 1 ≤ j ≤ k, the bound δ will be multiplied a positive number not exceeding M 2. Algorithms in [4, 11], applied to intersections of the irreducible components of V (considered over the field Q({Yp,q}p,q)((t1/∞))) with s hyperplanes  Yp,0 + X  1≤i≤n Yp,iXi = 0  , produce a union of 0-dimensional irreducible (over Q({Yp,q}p,q)(t)) algebraic sets (conjugacy classes) each defined by some polynomials g1, . . . , gr ∈ Q({Yp,q}p,q)(t)[X1, . . . , Xn], where r ≤ Dn, of degrees at most Dn2 with respect to O(n2) variables {Yp,q}p,q, and with coefficients represented by univariate polynomials ϕ, ξi of degrees at most D2 with respect to Z and at most δM 2Dn2 with respect to t. Complexity of the algorithm from [5], applied to polynomials ϕ, ξi, is polynomial in δM Dn2 and bit-sizes of their rational coefficients. 10 DIMA GRIGORIEV AND NICOLAI VOROBJOV As a result, the complexity of the algorithm is polynomial in bit-sizes of rational coefficients of polynomials ψ, ζi, which define coefficients αi of the input polynomials fj, in the number k of these polynomials, in δM , and in Dn4 = dn5 . We now summarise the results of this section. Theorem 2.2. There is an algorithm which for given polynomials f1, . . . , fk ∈ F[X1, . . . , Xn] and a point x = (x1, . . . , xn) ∈ Qn decides whether or not x belongs to Trop(V (f1, . . . , fk)) ⊂ Rn. Moreover, if x ∈ Trop(V (f1, . . . , fk)) the algorithm produces a lifting of x. To describe the complexity, let d = max1≤j≤k degX1,...,Xn (fj), let δ be the max- imum among the degrees in t of polynomials ψ, ζi, which define coefficients αi of polynomials fj. Let M be the maximum among the common denominator of rational numbers x1, . . . , xn and absolute values of their numerators. Then the complexity of the algorithm is polynomial in bit-sizes of rational coefficients of ψ, ζi, in the number k of polynomials fj, in δM , and in dn5 . Remark 2.3. The part of algorithm in the proof of Theorem 2.2, which checks whether 0 is a tropicalization of a point in eUν, actually computes tropicalization of the (0-dimensional) eUν. Since the dimension of a (classical) algebraic set in Fn coincides with the dimension of its tropicalization, the same procedure can be used to compute the tropicalization of any 0-dimensional algebraic set. Remark 2.4. In case of tropical linear varieties (d = 1), a deciding algorithm, with complexity similar to the one in Theorem 2.2, is given in [13]. 3. Criterion and deciding algorithm for being a tropical linear variety Let A ⊂ (Q∪{∞})n be a set of k vectors with bit-sizes of coordinates not exceed- ing L. Since A⊥ and A⊥⊥ are tropical linear prevarieties (see Proposition 1.10), Theorem 1.6 implies that A⊥ = Trophull({x1, . . . , xp}) and A⊥⊥ = Trophull({y1, . . . , yq}) for some vectors x1, . . . , xp, y1, . . . , yq ∈ (Q ∪ {∞})n. Moreover, the algorithm, stated in Theorem 1.6, computes vectors x1, . . . , xp, with complexity polynomial in L, nk, and vectors y1, . . . , yq with complexity polynomial in L, nnk . Theorem 3.1. The following three statements are equivalent. (1) There exist mutually complementary and orthogonal linear subspaces P, Q in Fn such that A⊥ = Trop(P ), and A⊥⊥ = Trop(Q) (in particular, A⊥ , A⊥⊥ are tropical linear varieties). (2) There exist liftings v1, . . . vp, w1, . . . , wq ∈ Fn of vectors x1, . . . xp, y1, . . . , yq ∈ (Q ∪ {∞})n respectively, such that (vi, wj) = 0 for all 1 ≤ i ≤ p, 1 ≤ j ≤ q. (3) A⊥ is a tropical linear variety. For the proof of this theorem we recall the following definition. Definition 3.2 ([6]). A square (r×r)-matrix with elements mij ∈ R∞, 1 ≤ i, j ≤ r, is called tropically singular if the minimum among {m1σ1 + · · · + mrσr for all permutations (σ1, . . . , σr) of (1, . . . , r)} DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 11 is attained on at least two elements, and tropically non-singular otherwise. The tropical rank, trk(M ), of a matrix M is the largest integer r such that this matrix has a non-singular (r × r)-submatrix. Proof of Theorem 3.1. Implication (1) ⇒ (2). Suppose there exist subspaces P, Q ⊂ Fn as in (1). Then arbitrary liftings v1, . . . , vp ∈ P and w1, . . . , wq ∈ Q, of x1, . . . , xp and y1, . . . , yq respectively, satisfy (2). We now prove implication (2) ⇒ (1). Let P (respectively, Q) be the linear hull of v1, . . . , vp (respectively, of w1, . . . , wq). Since (vi, wj) = 0 for all 1 ≤ i ≤ p, 1 ≤ j ≤ q, each vector v ∈ P is orthogonal to each vector wj, in particular, subspaces P and Q are orthogonal. Hence, Trop(v) ⊥ yj for each 1 ≤ j ≤ q. Then Trop(v) ⊥ y for each y ∈ Trophull({y1, . . . , yq}) = A⊥⊥, thus Trop(v) ∈ A⊥⊥⊥. By Lemma 1.9, (2), this means Trop(v) ∈ A⊥, thus Trop(P ) ⊂ A⊥. Similarly, we can prove that Trop(Q) ⊂ A⊥⊥. We prove the inclusion A⊥ ⊂ Trop(P ) by restricting it to charts in Rn ∞ (see ∞, where I ⊂ {1, . . . , n}, in Rn, and the Definition 1.5). Choose a chart CI in Rn corresponding chart DI in Fn (see Definition 1.13). Take x ∈ A⊥ ∩ CI . Consider first the case when x = (x1, . . . , xn) is representable, via generators of A⊥, as x = min1≤i≤p{ri1n + xi} with ri ∈ Q, where the minimum is taken component-wise. Recall that P is the linear hull of v1, . . . , vp ∈ Fn, let vi = (vi1, . . . , vin). For each 1 ≤ j ≤ n let α1j, . . . , αpj be coefficients of lowest degrees in Puiseux series v1jtr1, . . . , vpjtrp respectively. Choose s1, . . . , sp ∈ C so that s1α1j + · · · + spαpj 6= 0 for each 1 ≤ j ≤ n. Then Trop(s1v1j tr1 + · · · + spvpj trp) = min{x1j + r1, . . . , xpj + rp} = xj, i.e., x = (x1, . . . , xn) ∈ Trop(P ∩ DI ). In case when in the representation x = min1≤i≤p{ri1n + xi} elements ri belong in A⊥ ∩ CI to R, consider a sequence of points nmin1≤i≤p{r(ℓ) i 1n + xi}o1≤ℓ<∞ converging to x, such that each r(ℓ) ∈ Q. Then x ∈ Trop(P ∩ DI ), because Trop(P ∩ DI ) is closed. Taking the union over all subsets I ⊂ {1, . . . , n}, in the inclusions A⊥ ∩ CI ⊂ Trop(P ∩ DI ) we get A⊥ ⊂ Trop(P ). Similarly, we can prove that A⊥⊥ ⊂ Trop(Q). We conclude that A⊥ = Trop(P ) and A⊥⊥ = Trop(Q). We now prove that subspaces P and Q are mutually complementary. In [14] it is shown that dim(A⊥) + dim(A⊥⊥) ≥ n. Hence, by [6, Theorem 4.2], trk({x1, . . . , xp}) + trk({y1, . . . , yq}) = dim(A⊥) + dim(A⊥⊥) ≥ n. On the other hand, [6] implies that rk({v1, . . . , vp}) ≥ trk({x1, . . . , xp}) and rk({w1, . . . , wp}) ≥ trk({y1, . . . , yq}). It follows that rk({v1, . . . , vp})+rk({w1, . . . , wp}) ≥ n, hence P and Q are mutually complementary, given that they are orthogonal. Finally, implication (1) ⇒ (3) is trivial. To prove (3) ⇒ (1) notice that since there is a subspace P ⊂ Fn such that Trop(P ) = A⊥, by Corollary 1.18, Trop(Q) = A⊥⊥ for the subspace Q complement orthogonal to P . (cid:3) 12 DIMA GRIGORIEV AND NICOLAI VOROBJOV Corollary 3.3. There is an algorithm which for given tropical linear prevarieties A⊥ = Trophull({x1, . . . , xp}) and A⊥⊥ = Trophull({y1, . . . , yq}), where x1, . . . , xp, y1, . . . , yq ∈ (Q ∪ {∞})n, decides whether A⊥ is a tropical lin- ear variety. The complexity of the algorithm is exponential in bit-sizes of rational coordinates of vectors xi, yj, 1 ≤ i ≤ p, 1 ≤ j ≤ q, and in n, p, q. Proof. The input of the algorithm under construction is the set {x1, . . . , xp, y1, . . . , yq} ⊂ (Q ∪ {∞})n. Consider, over F∗ ∼= F \ {0}, the system of equations (3.1) Viν Wjν = 0 X 1≤ν≤n for all 1 ≤ i ≤ p, 1 ≤ j ≤ q such that xiν 6= ∞ and yjν 6= ∞ for all 1 ≤ ν ≤ n, where (Vi1, . . . , Vin), (Wj1, . . . , Wjn) are vectors of variables. Consider vectors xi, yj, 1 ≤ i ≤ p, 1 ≤ j ≤ q in the input, from which co- ordinates xiν and yjν with either xiν = ∞ or yjν = ∞ are removed. Applying Theorem 2.2, the algorithm checks whether there exist liftings of these vectors, If no, then A⊥ is not a tropical linear satisfying the system of equations (3.1). variety, by Theorem 3.1. Otherwise, let P be the linear hull of vectors v1, . . . vp such that in every vi each coordinate corresponding to xiν 6= ∞ is the lifting viν , while each coordinate corresponding to xiν = ∞ is 0. Similarly, let Q be the linear hull of vectors w1, . . . wq such that in every wj each coordinate corresponding to yjν 6= ∞ is the lifting wjν , while each coordinate corresponding to yjν = ∞ is 0. Then, by Theorem 3.1, P and Q are mutually complementary and orthogonal linear subspaces of Fn, while A⊥ = Trop(P ) and A⊥⊥ = Trop(Q). In particular, A⊥ and A⊥⊥ are tropical linear varieties. Theorem 2.2 implies that the complexity of the algorithm is exponential in bit- sizes of rational coordinates of vectors xi, 1 ≤ i ≤ p, yj, 1 ≤ j ≤ q, and in n, p, q. (cid:3) Remark 3.4. There is an alternative algorithm for the problem described in Corol- lary 3.3. Using [2, 17] or [23], it constructs a tropical basis of the system of equations (3.1), which is a finite set of polynomials Hℓ, with integer coefficients and (p + q)n variables Viν , Wjν such that xiν 6= ∞ and yjν 6= ∞. (See a detailed definition and properties of a tropical basis in [25, Section 2.6].) The algorithm checks whether, for all Hℓ, vectors x1, . . . xp, y1, . . . , yq, from which coordinates xiν and yjν with either xiν = ∞ or yjν = ∞ are removed, satisfy tropicalizations Trop(Hℓ). If yes, then, by the definition of tropical basis, there exist liftings viν , wjν ∈ F∗ of all xiν 6= ∞, yjν 6= ∞ respectively, which satisfy the system (3.1). Then the algorithm continues as the algorithm from Corollary 3.3. If vectors x1, . . . xp, y1, . . . , yq from which coordinates xiν and yjν with either xiν = ∞ or yjν = ∞ are removed, do not satisfy Trop(Hℓ) for all Hℓ, then A⊥ is not a tropical linear variety, by Theorem 3.1. The complexity of this algorithm is polynomial in the bit-size of rational coor- dinates of vectors xi, 1 ≤ i ≤ p, yj, 1 ≤ j ≤ q. The complexity also depends on the complexity of computing of a tropical basis for (3.1). Apparently, the latter complexity is doubly exponential in n, p, q, though we are unaware of a proof of this bound in literature. Note that there is a doubly exponential upper bound on the degree of a tropical basis [22] (cf. also [26, 10]). DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 13 Corollary 3.5. There is an algorithm which for tropical linear prevarieties A⊥ = Trophull({x1, . . . , xp}) and A⊥⊥ = Trophull({y1, . . . , yq}), where x1, . . . , xp, y1, . . . , yq ∈ (Q∪{∞})n, produces bases of linear subspaces P and Q, such that A⊥ = Trop(P ) and A⊥⊥ = Trop(Q), in case these subspaces exist. The complexity of the algorithm is exponential in bit-sizes of rational coordinates of vectors xi, 1 ≤ i ≤ p, yj, 1 ≤ j ≤ q, and in n, p, q. Proof. Apply the algorithm from Theorem 2.2 to vectors xi, yj, 1 ≤ i ≤ p, 1 ≤ j ≤ q, from which coordinates xiν and yjν with either xiν = ∞ or yjν = ∞ are removed, and to the system (3.1). The algorithm from Theorem 2.2 will either produce liftings viν , wjν ∈ F∗ of all xiν 6= ∞, yjν 6= ∞ respectively, which satisfy (3.1), or will indicate that liftings do not exist, i.e., vectors xi, yj with removed coordinates do not belong to the tropicalization of (3.1). If liftings exist, then P is the linear hull of vectors v1, . . . vp such that in every vi each coordinate corresponding to xiν 6= ∞ is the lifting viν of xiν , while each coordinate corresponding to xiν = ∞ is 0. Similarly, Q is the linear hull of vectors w1, . . . wq such that in every wj each coordinate corresponding to yjν 6= ∞ is the lifting wjν of yjν , while each coordinate corresponding to yjν = ∞ is 0. Herewith, all coordinates of vectors vi, wj are Puiseux series in Q((t1/∞)) represented as algebraic elements over the field Q(t), as described in the proof of Theorem 2.2. Choosing maximal linearly independent vectors among v1, . . . vp (respectively, among w1, . . . wq), we get a basis of P (respectively, of Q). Theorem 2.2 implies that the complexity of this algorithm is exponential in the bit-size of rational coordinates of vectors xi, 1 ≤ i ≤ p, yj, 1 ≤ j ≤ q, and exponential in n, p, q. (cid:3) 4. Infinite intersections of tropical linear prevarieties By the definition, the intersection of a finite number of tropical hyperplanes is a tropical linear prevariety. In this section we give an example of a countable family of tropical hyperplanes in R6 ∞ such that their intersection is not a finite union of convex polyhedra, in particular, not a tropical prevariety. This strengthens examples in [8] (example of T. Theobald) and [16], in which intersections of a countable families of tropical (non-linear) prevarieties were shown not to be finite unions of convex polyhedra. Choose a sequence {εi}∞ i=1 of pair-wise distinct real numbers εi such that 0 < εi < 1/4, and consider the tropical hyperplane Li ⊂ R6 ∞ defined by the set Li(x1, x2, y1, y2, z1, z2) := {−i + x1, −i + x2, −i/2 − εi + y1, −i/2 + y2, z1, z2} (see Definition 1.1). Let M := \ 1≤i<∞ Li ⊂ R6 ∞. Proposition 4.1. The set M ∩ R6 is not a finite union of convex polyhedra. In particular, M is not a tropical prevariety. Proof. Choose an integer 2 ≤ j < ∞ and consider the point pj := (0, 0, −j/2 − 1/4 + εj, −j/2 − 1/4, −j, −j) ∈ R6. 14 DIMA GRIGORIEV AND NICOLAI VOROBJOV A direct calculation shows that pj ∈ M because minima in the set Li(pj ) are attained at • coordinates x1, x2 if i > j, • coordinates y1, y2 if i = j, • coordinates z1, z2 if i < j. Moreover, the same calculation shows that any point pj +(δ1, δ1, δ2, δ2, δ3, δ3), for all sufficiently small positive δ1, δ2, δ3, also belongs to M . Hence, a neighbourhood of pj in M contains a 3-cube, denote it by Cj. Conversely, a neighbourhood of pj in M is contained in Cj . Indeed, consider a point qj ∈ M which is sufficiently close to pj in M . It can be represented as qj = pj + (α1, . . . , α6) = = (α1, α2, −j/2 − 1/4 + εj + α3, −j/2 − 1/4 + α4, −j + α5, −j + α6) for some small positive αℓ, 1 ≤ ℓ ≤ 6. Since qj ∈ Lj−1 ∩ Lj ∩ Lj+1, we conclude that each of the following three sets has at least two minimal elements: (1) {1 + α1, 1 + α2, 1/4 + α3, 1/4 + α4, α5, α6}, (2) {α1, α2, −1/4 + α3, −1/4 + α4, α5, α6}, (3) {−1 + α1, −1 + α2, −3/4 + α3, −3/4 + α4, α5, α6}. Since all αℓ are small, minimal (hence equal) elements in (1) are α5, α6, in (2) they are −1/4 + α3, −1/4 + α4, and in (3) they are −1 + α1, −1 + α2. It follows that α1 = α2, α3 = α4, and α5 = α6, hence qj ∈ Cj. We have proved that the 3-cube Cj is a neighbourhood of pj in M . By a direct calculation, any two points of the kind pi + (a1, a1, a2, a2, a3, a3) and pj + (b1, b1, b2, b2, b3, b3), where i 6= j and all aℓ, bℓ 6= ∞, 1 ≤ ℓ ≤ 3, are distinct due to εi 6= εj. This means that affine hulls of cubes Ci and Cj are disjoint in R6 for any two i 6= j, 2 ≤ i, j < ∞, and therefore, M ∩ R6 is not a finite union of convex polyhedra. Since every tropical prevariety is a finite polyhedral complex (see, e.g., [25]), M is not a tropical prevariety. (cid:3) Acknowledgements We thank M. Joswig, N. Kalinin, H. Markwig, and T. Theobald for useful dis- cussions, and anonymous referees for constructive remarks and suggestions. Part of this research was carried out during our joint visit in September 2017 to the Haus- dorff Research Institute for Mathematics at Bonn University, under the program Applied and Computational Algebraic Topology, to which we are very grateful. D. Grigoriev was partly supported by the RSF grant 16-11-10075. References [1] M. Akian, S. Gaubert, A. Guterman, Tropical polyhedra are equivalent to mean payoff games, Internat. J. Algebra Comput., 22, 1 (2012), 43 pp. [2] T. Bogart, A. N. Jensen, D. Speyer, B. Sturmfels, R.R. Thomas, Computing tropical varieties, J. Symbolic Comput. 42, 1 -- 2 (2007) 54 -- 73. [3] P. Butkovic, G.Hegedus, An elimination method for finding all solutions of the system of linear equations over an extremal algebra, Ekon.-Mat. Obzor 20 (1984) 203 -- 214. [4] A. Chistov, An algorithm of polynomial complexity for factoring polynomials, and determi- nation of the components of a variety in a subexponential time, J.Soviet Math., 34 (1986) 1838 -- 1882. DECIDING WHETHER A TROPICAL LINEAR PREVARIETY IS A TROPICAL VARIETY 15 [5] A. Chistov, Polynomial complexity of Newton-Puiseux algorithm, Lect. Notes Comput. Sci., 233 (1986) 247 -- 255. [6] M. Develin, F. Santos, B. Sturmfels, On the rank of a tropical matrix, In: Combinatorial and Computational Geometry, MSRI Publications, v. 52 (2005). [7] M. Develin, B. Sturmfels, Tropical convexity, Doc. Math., 9 (2004) 1 -- 27. [8] A. Dress, W. Wenzel, Algebraic, tropical, and fuzzy geometry, Beitr. Algebra Geom. 52, 2 (2011) 431-461. [9] S. Gaubert, R.D. Katz, Minimal half-spaces and external representation of tropical polyhedra, J. Algebraic Combin. 33, 3 (2011) 325-348. [10] P. Gorlach, Y. Ren, J. Sommars, Detecting tropical defects of polynomial equations, arXiv:1809.03350, 2018. [11] D. Grigoriev, Polynomial factoring over a finite field and solving systems of algebraic equa- tions, J. Soviet Math., 34 (1986) 1762 -- 1803. [12] D. Grigoriev, Complexity of solving tropical linear systems, Computational Complexity, 22 (2013) 71 -- 88. [13] D. Grigoriev, Polynomial complexity recognizing a tropical linear variety, Lect. Notes Com- put. Sci., v. 9301 (2015) 152 -- 157. [14] D. Grigoriev, V. Podolskii, Complexity of tropical and min-plus linear prevarieties, Compu- tational Complexity, 24, 1 (2015) 31 -- 64. [15] D. Grigoriev, N. Vorobjov, Orthogonal tropical linear prevarieties, in: CASC 2018, Lect. Notes Comput. Sci., v. 11077 (2018) 187 -- 196. [16] D. Grigoriev, N. Vorobjov, Upper bounds on Betti numbers of tropical prevarieties, Arnold Math. J., 4, 1 (2018) 127 -- 136. [17] K. Hept, T. Theobald, Tropical bases by regular projections, Proc. Amer. Math. Soc., 137, 7 (2009) 2233 -- 2241. [18] A. Jensen, H. Markwig, T. Markwig, An algorithm for lifting points in a tropical variety, Collect. Math., 59, 2 (2008) 129 -- 165. [19] A. Jensen, J. Yu, Stable intersections of tropical varieties, Journal of Algebraic Combinatorics, 43, 1 (2016) 101 -- 128. [20] M. Joswig, Essentials of tropical combinatorics, Springer, 2014. [21] M. Joswig, The Cayley trick for tropical hypersurfaces with a view toward Ricardian econom- ics, in: Homological and Computational Methods in Commutative Algebra, Springer INdAM Series, 20, Springer, 2017. [22] M. Joswig, B. Schroter, The degree of a tropical basis, Proc. Amer. Math. Soc. 146, 3 (2018) 961-970. [23] Ya. Kazarnovskii, A. G. Khovanskii, Tropical noetherity and Grobner bases, St. Petersburg Math. J., 26, 5 (2015) 797 -- 811. [24] S. Leng, Algebra, Springer, 2002. [25] D. Maclagan, B. Sturmfels, Introduction to Tropical Geometry, American Math. Society, 2015. [26] T. Markwig, Y. Ren, Computing Tropical Varieties Over Fields with Valuation, Found. Com- put. Math., (2019). [27] K. Murota, A. Tamura, On circuit valuation of matroids, Advances in Applied Mathematics, 26 (2001) 192-225. [28] B. Osserman, S. Payne, Lifting tropical intersections, Documenta Mathematica, 18 (2013) 121 -- 175. [29] J. Richter-Gebert, B. Sturmfels, T. Theobald, First steps in tropical geometry, In: G. Litvi- nov, V. Maslov (Eds.), Idempotent Mathematics and Mathematical Physics (Proceedings Vienna 2003), Contemporary Mathematics, 377, American Math. Society (2005) 289 -- 317. [30] D. Speyer, Tropical linear spaces, arXiv:0410455, 2008. [31] J. Yu, D.S. Yuster, Representing tropical linear spaces by circuits, The 19th International Conference on Formal Power Series and Algebraic Combinatorics, 2007, arXiv:0611579, 2006. CNRS, Math´ematiques, Universit´e de Lille, Villeneuve d'Ascq, 59655, France E-mail address: [email protected] Department of Computer Science, University of Bath, Bath BA2 7AY, England, UK E-mail address: [email protected]
1712.08901	3	1712	2018-09-22T19:01:54	Bott-Chern blow-up formula and bimeromorphic invariance of the $\partial\bar{\partial}$-Lemma for threefolds	[ "math.AG", "math.AT", "math.DG" ]	The purpose of this paper is to study the bimeromorphic invariants of compact complex manifolds in terms of Bott-Chern cohomology. We prove a blow-up formula for Bott-Chern cohomology. As an application, we show that for compact complex threefolds the non-K\"{a}hlerness degrees, introduced by Angella-Tomassini [Invent. Math. 192, (2013), 71-81], are bimeromorphic invariants. Consequently, the $\partial\bar{\partial}$-Lemma on threefolds admits the bimeromorphic invariance.	math.AG	math	BOTT-CHERN BLOW-UP FORMULA AND BIMEROMORPHIC INVARIANCE OF THE ∂ ¯∂-LEMMA FOR THREEFOLDS SONG YANG AND XIANGDONG YANG Abstract. The purpose of this paper is to study the bimeromorphic invariants of compact complex manifolds in terms of Bott-Chern cohomology. We prove a blow-up formula for Bott-Chern cohomology. As an application, we show that for compact complex threefolds the non-Kahlerness degrees, introduced by Angella-Tomassini [Invent. Math. 192, (2013), 71-81], are bimeromorphic invariants. Consequently, the ∂ ¯∂-Lemma on threefolds admits the bimeromorphic invariance. 1. Introduction Let X be a complex manifold then it admits a huge space of smooth Hermitian metrics. In general, to study the geometry and topology of X it is of usefulness to pick a special Hermitian metric on X, such as the Kahler metric and the balanced metric. Among these metrics, the most distinguished ones on complex manifolds are Kahler metrics. In the past decades, there have been extensive study of Kahler geometry from differential and algebraic geometry points of view. The existence of a Kahler metric on a compact complex manifold implies many special topological consequences, and one of them is the ∂ ¯∂-Lemma. According to the real homotopy theory, a direct result of the ∂ ¯∂-Lemma is that the real homotopy type of a compact Kahler manifold is a formal consequence of the de Rham cohomology ring (cf. [17]). Following the notion of Popovici [37], we say that a compact complex manifold X is a ∂ ¯∂-manifold if the ∂ ¯∂-Lemma holds on X. It is noteworthy that the class of ∂ ¯∂-manifolds consists of Moishezon manifolds and compact complex manifolds in the class C of Fujiki, which are not Kahler in general (cf. [33, 24]). On one hand, a great number of examples of ∂ ¯∂-manifolds that are known so far carry the balanced metrics. On the other hand, there exist many examples of balanced manifolds on which the ∂ ¯∂-Lemma does not hold, for instance, a classical example is the Iwasawa manifold (cf. [7]). In complex geometry, there exist two important geometric operations: modification and deformation. These operations provide us with useful methods to produce new complex manifolds from the original ones. In 1960, Hironaka [27] constructed a compact non-Kahler threefold X, with a modification f : X → CP3, on which the ∂ ¯∂-Lemma holds. This means that the Kahler metric on a compact complex manifold is not preserved by modifications. On balanced metric, Alessandrini-Bassanelli [5, 6] proved that the balanced structure is stable under modifications. Moreover, Popovici [36] showed that the strongly Gauduchon (sG) structure on a compact complex manifold admits the stability under modifications. On the Date: September 25, 2018. 2010 Mathematics Subject Classification. 32Q55; 32C35. Key words and phrases. Bott-Chern cohomology, blow-up, bimeromorphic invariant, ∂ ¯∂-Lemma. 1 2 S. YANG AND X. YANG small deformations of Kahler manifolds, a classical result of Kodaira-Spencer [29] shows that small deformations of a compact Kahler manifold remain Kahler; see also Rao-Wan- Zhao [38, 39] for a new proof. The example of Iwasawa manifold shows that the balanced structure is not preserved under small deformations (cf. [4]); whereas, it is stable under small deformations when the manifold satisfies some versions of the ∂ ¯∂-Lemma (cf. [23, 38, 50] etc). Particularly, the ∂ ¯∂-Lemma itself admits the stability under small deformations (cf. [48, 50, 12]). Assume that X and X are two compact complex manifolds with the same dimensions and let f : X → X be a modification. According to a result of Deligne-Griffiths-Morgan-Sullivan [17, Theorem 5.22], we know that if the ∂ ¯∂-Lemma holds for X then so does for X (see also [18, Theorem 12.9]). Conversely, it is natural to consider the following question (see also [3]): if the ∂ ¯∂-Lemma holds for X, then whether it also holds for X? In general, modifications are special cases of bimeromorphic maps, we have the following Problem 1.1. Is the ∂ ¯∂-Lemma a bimeromorphic invariant of compact complex manifolds? Here the bimeromorphic invariance means that if f : X 99K X is a bimeromorphic map of compact complex manifolds then the ∂ ¯∂-Lemma holds on X if and only if so does on X. In 2002, Abramovich-Karu-Matsuki-W lodarczyk [1, Theorem 0.3.1] proved the Weak Fac- torization Theorem which asserts that any bimeromorphic map of compact complex mani- folds f : X 99K X can be decomposed into a finite sequence of blow-ups and blow-downs with smooth centers. Therefore, to prove the bimeromorphic invariance of the ∂ ¯∂-Lemma, it is sufficient to show that the ∂ ¯∂-Lemma is stable under blow-ups. From cohomogical point of view, Angella-Tomassini [12] showed that the validity of the ∂ ¯∂-Lemma on a compact com- plex manifold can be characterized by Bott-Chern cohomology. This implies that to prove that the blow-up of a ∂ ¯∂-manifold satisfies the ∂ ¯∂-Lemma we need a blow-up formula for Bott-Chern cohomology. Using a sheaf-theoretic approach, we prove the following result. Theorem 1.2. Let X be a compact complex manifold with dimC X = n. Assume that Z ⊂ X is a closed complex submanifold such that codimC Z = r ≥ 2. Then we have the following isomorphisms: (i) H p,q (ii) H p,q BC ( X) ∼= H p,q BC ( X) ∼= H p,q BC(X)L(cid:18)H p,q BC(X), for p = 0 or q = 0. BC (E)/(πE )∗H p,q BC(Z)(cid:19), for 1 ≤ p, q ≤ n; Here π : X → X is the blow-up of X along Z and E = π−1(Z) is the exceptional divisor. Note that on compact complex surfaces the ∂ ¯∂-Lemma, the balanced condition, and the Kahler condition are equivalent; however, if the complex dimension is larger than 2 those conditions are not equivalent sharply. In particular, there exist examples of compact balanced manifolds on which the ∂ ¯∂-Lemma does not hold. Besides, all ∂ ¯∂-manifolds known so far are balanced manifolds, and hence this leads to a conjecture that every ∂ ¯∂-manifold carries a balanced metric (cf. [37, 40]). In 2012, Fu-Li-Yau [22, Theorem 1.2] proved that each Clemens manifold has a balanced metric; moreover, recently Friedman [20, Theorem 3.10] showed that for a general Clemens manifold the ∂ ¯∂-Lemma holds. In their paper [12], Angella-Tomassini introduced a series of numerical invariants called non-Kahlerness degrees for compact complex 3 manifolds. As an application of Theorem 1.2, we consider the bimeromorphic invariance of non-Kahlerness degrees for compact complex threefolds and we give a definite answer to Problem 1.1 in this case. Theorem 1.3. Let f : X 99K X be a bimeromorphic map of compact complex threefolds. Then for any k ∈ {0, 1, 2, 3, 4, 5, 6} the k-th non-Kahlerness degree of X is equal to the k-th non-Kahlerness degree of X. Moreover, X is a ∂ ¯∂-manifold if and only if X is a ∂ ¯∂-manifold. It is well-known that (p, 0)- and (0, q)-Hodge numbers are bimeromorphic invariants of compact complex manifolds (see [41, Corollary 1.4] for a uniform proof). Due to the sym- metric property of Bott-Chern cohomology the (0, q)-Bott-Chern number equals to the (q, 0)- Bott-Chern number. Analogously, on the (p, 0)-Bott-Chern number for a compact complex manifold X, a natural problem is: Problem 1.4. Does the (p, 0)-Bott-Chern number admit the bimeromorphic invariance? As an application of the second part of Theorem 1.2, we show that the (p, 0)-Bott-Chern numbers of a compact complex manifold are bimeromorphic invariants (Corollary 3.6). An outline of this paper is organized as follows. We devote Section 2 to the preliminaries of Bott-Chern cohomology and the cohomological characterization of the ∂ ¯∂-Lemma on a com- pact complex manifold. In Section 3, we prove a blow-up formula for Bott-Chern cohomology. In Section 4, we show that the non-Kahlerness degrees of threefolds are bimeromorphic in- variants. Finally, in Section 5 we give a brief remark on the bimeromorphic invariance of ∂ ¯∂-Lemma on compact complex manifolds in general. Recently, we were informed that, using different techniques, D. Angella, T. Suwa, N. Tar- dini and A. Tomassini also proved the stability of the ∂ ¯∂-Lemma under blow-ups with centers satisfying the ∂ ¯∂-Lemma and having holomorphically contractible neighbourhoods [10, Theo- rem 2.1]. The approach in [10] is a Cech cohomology theory for Dolbeault cohomology which gives rise to a more topological method from a complimentary point of view. Acknowledgement. We would like to express our gratitude to Prof. Guosong Zhao and Prof. Xiaojun Chen for their constant encouragements and supports, and sincerely thank the Departments of Mathematics, Universit`a degli Studi di Milano and Department of Math- ematics, Cornell University for the hospitalities during our respective visits. Specially, we would like to thank Sheng Rao for many useful discussions and comments which the present version is benefitted from. Last, but not least, we sincerely thank Prof. D. Angella, Dr. L. Meng and Dr. J. Stelzig for sending us their paper [10], [32] and [44], respectively. This work is partially supported by the NSFC (Grant No. 11571242, No. 11701414 and No. 11701051) and the China Scholarship Council. 2. Preliminaries 2.1. Cohomologies of complex manifolds. Assume that (X, J) is a compact complex manifold of complex dimension n. Let Ak(X; C) denote the space of C-valued differential forms of degree k on X. The complex structure J determines a decomposition of the exterior 4 S. YANG AND X. YANG differential d = ∂ + ¯∂; moreover, it induces a natural decomposition of the complexified cotangent bundle T ∗ CX = T ∗′X ⊕ T ∗′′X. The decomposition above gives a decomposition on the space of differential forms Ak(X; C) = Mp+q=k Ap,q(X; C), where Ap,q(X; C) is the space of smooth (p, q)-forms on X. Furthermore, we get a natural bi- differential Z2-graded algebra, namely, the double complex of X denoted by (A•,•(X; C); ∂, ¯∂). In particular, there are several important cohomologies X associated to (A•,•(X; C); ∂, ¯∂) as follows (cf. [19, 14, 2]). • The de Rham cohomology • The ∂-cohomology and Dolbeault cohomology H • dR(X; C) := ker d im d ; H •,• ∂ (X) := ker ∂ im ∂ , H •,• ¯∂ (X) := ker ¯∂ im ¯∂ ; • The Bott-Chern cohomology and Aeppli cohomology H •,• BC(X) := ker ∂ ∩ ker ¯∂ im ∂ ¯∂ , H •,• A (X) := ker ∂ ¯∂ im ∂ + im ¯∂ . From definition, the Bott-Chern and Aeppi cohomology admit the symmetric property by complex conjugation. It is noteworthy that all the above cohomologies can be thought of as bi-graded algebras In particular, the identity map induces the over C with respect to the wedge product. following natural maps of bi-graded C-vector spaces H •,• ∂ (X) H •,• ¯∂ (X) zttttttttt $❏❏❏❏❏❏❏❏❏ $❏❏❏❏❏❏❏❏❏ zttttttttt H •,• BC(X) H • dR(X; C) H •,• A (X) (2.1) In general, the maps in (2.1) are neither surjective nor injective. If the natural map H •,• BC(X) → H •,• dR (X; C) induced by the identity is injective, then we say that X satisfies the ∂ ¯∂-Lemma. In addition, we have the following proposition (cf. [17, Proposition 5.17]). Proposition 2.1. Let X be a compact complex manifold. Then the following conditions are equivalent. (i) The map H •,• (ii) All the maps in (2.1) are isomorphisms. BC(X) → H •,• dR (X; C) is injective. z $ $ z 5 (iii) The following equation holds on the space of C-valued differential forms ker ∂ ∩ ker ¯∂ ∩ im d = im ∂ ¯∂. If X is a Kahler manifold or more general a ∂ ¯∂-manifold then the Bott-Chern/Aeppli cohomology coincide with the Dolbeault cohomology via the maps in (2.1). In non-Kahler complex geometry, the Bott-Chern/Aeppli cohomology play important roles. According to the Hodge theory on compact Hermitian manifold we know that the de Rham and Dolbeault cohomology of a compact complex manifold are finite dimensional. Analogously, Schweitzer [43] established the Hodge theory for Bott-Chern and Aeppli cohomology. Given a Hermitian metric h on X then we can construct two 4-th order elliptic self-adjoint differential operators ∆BC := (∂ ¯∂)(∂ ¯∂)∗ + (∂ ¯∂)∗(∂ ¯∂) + ( ¯∂∗∂)( ¯∂∗∂)∗ + ( ¯∂∗∂)∗( ¯∂∗∂) + ¯∂∗ ¯∂ + ∂∗∂ and ∆A := ∂∂∗ + ¯∂ ¯∂∗ + (∂ ¯∂)∗(∂ ¯∂) + (∂ ¯∂)(∂ ¯∂)∗ + ( ¯∂∂∗)∗( ¯∂∂∗) + ( ¯∂∂∗)( ¯∂∂∗)∗. Moreover, we have H •,• BC(X) ∼= ker ∆BC and H •,• A ∼= ker ∆A. It follows that the Bott-Chern and Aeppli cohomology of a compact complex manifold are finite dimensional. In addition, the Hodge star operator with respect to the Hermitian metric h induces a duality ∗ : H p,q BC(X) ∼=−→ H n−p,n−q A (X) (2.2) for every p, q ∈ N. Consider the relationship between the de Rham and Dolbeault cohomology of X. There exists a spectral sequence {Er, dr} of the double complex (A•,•(X; C); ∂, ¯∂), called the Hodge- Frolicher spectral sequence starting from the Dolbeault cohomology and converging to the graded module associated to a filtration of the de Rham cohomology Ep,q 1 = H p,q ¯∂ (X) Ep,q r =⇒ Gr H p+q dR (X; C) ∼= H p+q dR (X; C). Moreover, this gives rise to the famous Frolicher-inequality [21]: dimC H p,q ¯∂ (X) ≥ dimC H k dR(X; C), Xp+q=k for any 0 ≤ p, q ≤ n. As an analogy, Angella-Tomassini [12] proved a Frolicher-type inequality which links Bott-Chern and Aeppli cohomology to de Rham cohomology and provides a new characteristic theorem of the ∂ ¯∂-Lemma. Theorem 2.2 ([12, Theorem A and Theorem B]). Let X be a compact complex manifold. Then, for every k ∈ N, the following inequality holds Xp+q=k(cid:0)dimC H p,q BC(X) + dimC H p,q Moreover, the equality in (2.3) holds for every k ∈ N if and only if X satisfies the ∂ ¯∂-Lemma. A (X)(cid:1) ≥ 2dimC H k dR(X; C). (2.3) 6 S. YANG AND X. YANG For any (p, q) ∈ N × N, k ∈ N and ♯ ∈ {∂, ¯∂, BC, A} we define h•,• ♯ (X) := dimC H •,• ♯ (X) and hk ♯ (X) := Xp+q=k dimC H p,q ♯ (X). Especially, we denote the k-th Betti number of X by bk(X) := dimC H k dR(X; C). Definition 2.3 (Non-Kahlerness degrees). For a compact complex manifold X of complex dimension n, and for any k ∈ {0, 1, · · · , 2n} the k-th non-Kahlerness degree is defined to be the non-negative integer ∆k(X) := hk BC (X) + h2n−k BC (X) − 2bk(X). In particular, we have ∆k(X) = ∆2n−k(X) and ∆0 = ∆2n = 0; therefore, the only non-trivial non-Kahlerness degrees are ∆k(X) for k ∈ {1, 2, · · · 2n − 1}. Equivalently, a quantitative characterization of the ∂ ¯∂-Lemma on compact complex man- ifolds is stated as follows. Theorem 2.4 ([11, Theorem 3.1]). A compact complex manifold X satisfies the ∂ ¯∂-Lemma if and only if, for any k ∈ N, there holds (hp,q BC (X) − hp,q A (X)) = 0. Xp+q=k The revised non-Kahlerness degrees of X is defined to be the following integers: N k(X) := hk BC (X) − hk A(X) = hk BC (X) − h2n−k BC (X) ∈ Z, where k ∈ {0, 1, · · · , 2n}. From definition, we get N k(X) = −N 2n−k(X); moreover, we have the following equivalent conditions (cf. [12, 11]): (i) X satisfies the ∂ ¯∂-Lemma; (ii) N k(X) = 0 for any k ∈ {0, 1, · · · , 2n}; (iii) ∆k(X) = 0 for any k ∈ {0, 1, · · · , 2n}. From the bimeromorphic geometry point of view, it would be interesting to confirm whether the non-Kahlerness degrees of a compact complex manifold are bimeromorphic invariants. It is noteworthy that for any compact complex surface the non-Kahlerness degree ∆1 is always zero and ∆2 = 0 if and only if the surface is Kahler (cf. [8]). In addition, if the surface is non-Kahler then ∆2 = 2 (cf. [13, 46]). 2.2. Bott-Chern hypercohomology. Assume that (X, J) is a compact complex manifold with dimC X = n. Let E s,t X be the sheaf of germs of differential (s, t)-forms on X. For a fixed bidegree (p, q) such that q ≥ p ≥ 1, let k = p + q. We can define the following sheaves and L l E s,t X for l ≤ k − 2 s<p,t<q X = Ms+t=l, X = Ms+t=l, s≥p,t≥q L l−1 E s,t X for l ≥ k. 7 Observe that L k−2 setting X = E p−1,q−1 X and L k−1 X = E p,q X . We define the operators acting on L • X by for l 6= k − 2 (in the case of l ≤ k − 3, we neglect the components which fall outside L l+1 and X ), δl := d : L l X → L l+1 X , X → L k−1 X . Since δi+1 ◦ δi = 0 for any i ≥ 0, we get a sheaf complex L • δk−2 := ∂ ¯∂ : L k−2 X as follows · · · d / L k−3 X d / L k−2 X ∂ ¯∂ / L k−1 X d / L k X d / · · · . From definition, we have H p,q BC (X) ∼= H k−1(L • X (X)). The (p, q)-Bott-Chern hypercohomol- ogy of X is defined to be the (k − 1)-th hypercohomology of L • X , i.e., Hk−1(X, L • X ) := Rk−1Γ(X, L • X ), where Rk−1Γ(X, −) is the right derived functor of the global section functor Γ(X, −). Because L • X is a complex of fine sheaves the following lemma holds (cf. [18, Chapter VI, §12]). Lemma 2.5. Let X be a compact complex manifold. Then for the fixed bi-degree (p, q) we have H p,q BC(X) ∼= Hp+q−1(X, L • X ). Let Ωl X (resp. ¯Ωl X ) be the sheaf of holomorphic (resp. anti-holomorphic) p-forms and CX the constant sheaf on X. There exist two sheaf complexes of holomorphic and anti- holomorphic forms 0 / OX ∂ / Ω1 X ∂ / · · · ∂ / Ωp−1 X / 0, 0 / ¯OX ¯∂ / ¯Ω1 X ¯∂ / · · · ¯∂ / ¯Ωq−1 X / 0; moreover, we can define a new sheaf complex S • X: 0 / OX + ¯OX ∂+ ¯∂ / Ω1 X ⊕ ¯Ω1 X / · · · / Ωp−1 X ⊕ ¯Ωp−1 X / ¯Ωp X / · · · ¯Ωq−1 X / 0. Similarly, we have a sheaf complex B• X: 0 / CX (+,−) / OX ⊕ ¯OX / Ω1 X ⊕ ¯Ω1 X / · · · / Ωp−1 X ⊕ ¯Ωp−1 X / ¯Ωp X / · · · ¯Ωq−1 X / 0. It is of importance that the inclusion S • X[−1] := S •−1 all quasi-isomorphic, where S • 4.3]). Consequently, by Lemma 2.5 we get the following isomorphisms. X ⊂ L • X (2.4) X[−1] are (cf. [18, Lemma 12.1] or [43, Proposition 4.2, X and the morphism B• X → S • Proposition 2.6. H p,q BC(X) ∼= Hp+q−1(X, L • X ) ∼= Hp+q−1(X, S • X ) ∼= Hp+q(X, B• X ). 3. Blow-up formula of Bott-Chern cohomology The purpose of this section is to give the proof of Theorem 1.2. Throughout of this section we assume that X is a compact complex manifold with dimCX = n and ι : Z ֒→ X be a closed complex submanifold with codimCZ = r ≥ 2. Set U = X − Z. / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / 8 S. YANG AND X. YANG 3.1. Case of 1 ≤ p, q ≤ n. Due to the symmetric property of Bott-Chern cohomology we only need to consider the case of q ≥ p ≥ 1. Recall that the space of sections of the constant sheaf CX over an open subset V is Γ(V, CX ) = {f : V → C locally constant function}. We define a surjective morphism of sheaves χ : CX → ι∗CZ by setting χ(V ) : Γ(V, CX ) → Γ(V, ι∗CZ ) = Γ(V ∩ Z, CZ ) f 7→ f V ∩Z , for any open subset V ⊂ X such that V ∩ Z 6= ∅, and zero map otherwise. Then there is a short exact sequence of sheaves 0 / KU / CX χ / ι∗CZ / 0 (3.1) where KU is the kernel sheaf of χ. For any integer 0 ≤ s ≤ n and any open subset V ⊂ X we can define a morphism ϕ(V ) : Γ(V, Ωs X ) → Γ(V, ι∗Ωs α 7→ (ιV ∩Z)∗α, Z) = Γ(V ∩ Z, Ωs Z ) where ιV ∩Z : V ∩ Z → V is the holomorphic inclusion and therefore we get a sheaf morphism ϕ : Ωs X → ι∗Ωs Z. X = ker (ϕ). Then we have the following sequence of sheaves; for s = 0 it is the structure Set Ks sheaf sequence. Lemma 3.1. The following short sequence of sheaves on X is exact 0 / Ks X / Ωs X ϕ / ι∗Ωs Z / 0. (3.2) Proof. We only need to show that the sheaf morphism ϕ is surjective. By definition, the stalk of the direct image ι∗Ωs Z)x if x ∈ Z and zero otherwise. It follows that the map of stalks ϕx : (Ωs Z)x is zero map if x ∈ U . The remaining thing to do in the proof is to verify that for any x ∈ Z the morphism of stalks Z at a point x is (Ωs X )x → (ι∗Ωs ϕx : (Ωs X)x → (ι∗Ωs Z)x is surjective. Let (W ; z1, z2, · · · , zn) be a local coordinate chart of x such that Then there exists a holomorphic projection given by W ∩ Z = {zn−r+1 = · · · = zn = 0}. τ : W → W ∩ Z, (z1, · · · , zn) 7→ (z1, · · · , zn−r). For any germ αx ∈ (ι∗Ωs Z)x we can choose a representative (V, α), here α is a holomorphic p-form on V ∩ Z. Particularly, we can choose V small enough such that V ⊂ W and then the restriction of τ on V gives rise to a holomorphic map τV : V → V ∩ Z satisfying / / / / / / / / 9 τV ◦ ιV ∩Z = idV ∩Z . Let β = (τV )∗(α), then (V, β) represents a germ, denoted by βx, in the stalk (Ωs X)x. From definition, we get It follows that ϕx(βx) = αx, namely, ϕx is surjective. (ιV ∩Z )∗β = (ιV ∩Z )∗(cid:0)(τV )∗(α)(cid:1) = (τV ◦ ιV ∩Z )∗(α) = α. (cid:3) Taking the complex conjugation of (3.2), we get the the anti-holomorphic sheaf sequence 0 / ¯Kt X / ¯Ωt X ¯ϕ / ι∗ ¯Ωt Z / 0, (3.3) for any integer 0 ≤ t ≤ n. Since the direct image functor ι∗ commutes with the direct sum we derive a short exact sequence by taking direct sum of (3.2) and (3.3) 0 / Ks X ⊕ ¯Kt X / Ωs X ⊕ ¯Ωt X ϕ⊕ ¯ϕ / ι∗(Ωs Z ⊕ ¯Ωt Z) / 0. (3.4) Observing that both operators ∂ and ¯∂ commute with the pullback of holomorphic maps and ι∗ is an exact functor, we get the following commutative diagram from (3.4) 0 0 / Ks X ⊕ ¯Kt X / Ωs X ⊕ ¯Ωt X / ι∗(Ωs Z ⊕ ¯Ωt Z) / 0 (3.5) ∂⊕ ¯∂ ∂⊕ ¯∂ ∂⊕ ¯∂ / Ks+1 X ⊕ ¯Kt+1 X / Ωs+1 X ⊕ ¯Ωt+1 X / ι∗(Ωs+1 Z ⊕ ¯Ωt+1 Z ) / 0. By the definition of the sheaf complex (2.4), the sequences (3.1) and (3.5) implies Proposition 3.2. For q ≥ p ≥ 1, there is a short exact sequence of sheaf complexes 0 / K • X / B• X / ι∗B• Z / 0, (3.6) where K • X is the sheaf complex (+,−) KU / K0 X ⊕ ¯K0 X / K1 X ⊕ ¯K1 X / · · · / Kp−1 X ⊕ ¯Kp−1 X / ¯Kp X / · · · ¯Kq−1 X / 0. (3.7) To prove the theorem we need the following basic result from homological algebra. Proposition 3.3. Consider a commutative diagram of two exact sequences of complex vector spaces · · · · · · / A1 f1 / A2 f2 / A3 f3 / A4 / · · · i1 i2 i3 i4 / B1 g1 / B2 g2 / B3 g3 / B4 / · · · . Assume that i1, i4 are isomorphic, and i2, i3 are injective. Then there is a natural isomor- phism Proof. By the hypothesis, i2 and i3 are injective and therefore we have B2 ∼= A2 ⊕(cid:16)B3/i3(A3)(cid:17). B2 ∼= A2 ⊕ coker i2, B3 ∼= A3 ⊕ coker i3. The commutativity of the second square implies g2(im(i2)) ⊂ im(i3) and therefore there exists a well-defined morphism of cokernels ¯g2 : coker i2 → coker i3. / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / 10 S. YANG AND X. YANG The injectivity and the surjectivity of ¯g2 can be obtained by the standard argument of diagram-chasing. (cid:3) We are now in a position to prove the first part of Theorem 1.2. Proof of Theorem 1.2 (i). Recall that X is a compact complex manifold with dimC X = n, ι : Z ֒→ X a closed complex submanifold such that codimCZ = r ≥ 2 and U = X − Z the complimentary open set. Assume that π : X → X is the blow-up of X with the center Z. Let E := π−1(Z) be the exceptional divisor and U := X − E. Then the restriction map is biholomorphic. Moreover, we have a commutative diagram π U : U → U πE E Z ι ι / X π / X, (3.8) where ι : E ֒→ X is the inclusion. Due to the symmetric property of Bott-Chern cohomology we only consider the case of q ≥ p ≥ 1. For any bi-degree (p, q) satisfying q ≥ p ≥ 1 we can construct the following short exact sequence of sheaves on X similar to (3.6) 0 / K • X / B• X / ι∗B• E / 0, (3.9) where B X and BE are defined as (2.4), and K • X is the counterpart of (3.7) on X. Note that π : X → X is a proper holomorphic map the pullback π∗ of holomorphic forms induces a morphism (over π) from Ωs ; moreover, the blow-up diagram (3.8) gives rise to a commutative diagram for the long exact sequences of hypercohomologies associated to (3.6) and (3.9), respectively. , and also a morphism from ¯Ωs X to ¯Ωs X X to Ωs X · · · / Hp+q(X, K • X ) / Hp+q(X, B• X ) / Hp+q(X, ι∗B• Z) / Hp+q+1(X, K • X ) / · · · π∗ π∗ (πE)∗ π∗ · · · / Hp+q( X, K • X ) / Hp+q( X, B• X ) / Hp+q( X, ι∗B• E) / Hp+q+1( X, K • X ) / · · · (3.10) Because the direct image functor ι∗ is exact and RΓ(X, ι∗(−)) = RΓ(Z, −), by Proposition 2.6 we obtain also by Proposition 2.6 we have Hp+q(X, B• BC(X). As a result, (3.10) becomes · · · / Hp+q(X, K • X ) / H p,q BC(X) / Hp+q+1(X, K • X ) / · · · (3.11) Hp+q(X, ι∗B• Z ) ∼= H p,q BC (Z), Z ) ∼= Hp+q(Z, B• X ) ∼= H p,q / H p,q BC(Z) π∗ π∗ (πE )∗ π∗ · · · / Hp+q( X, K • X ) / H p,q BC( X) / H p,q BC(E) / Hp+q+1( X, K • X ) / · · · We claim the following result and give its proof at the end of this subsection. / / / / / / / / / / / / / / / / / / / / / / / / / / 11 Lemma 3.4. For any k ∈ N, the morphism π∗ : Hk(X, K • X ) → Hk( X, K • X ) is isomorphic. From (3.11) and Lemma 3.4, we derive the commutative diagram · · · / Hp+q(X, K • X ) / H p,q BC (X) / H p,q BC(Z) / Hp+q+1(X, K • X ) / · · · (3.12) π∗ ∼= π∗ (πE)∗ π∗ ∼= · · · / Hp+q( X, K • X ) / H p,q BC ( X) / H p,q BC(E) / Hp+q+1( X, K • X ) / · · · . Since π : X → X is a proper and holomorphic map the induced morphism is injective (cf. [18, Theorem 12.9] or [9, Theorem 2.4.]) and so is the morphism π∗ : H p,q BC (X) → H p,q BC( X) (πE )∗ : H p,q BC(Z) → H p,q BC(E) (3.13) (3.14) by the Weak Five Lemma [30, Chapter 1, §3]. Applying Proposition 3.3 to (3.12), we obtain which means coker (3.13) ∼= coker (3.14), The proof of Theorem 1.2 (i) is now complete. (cid:3) H p,q BC( X) ∼= H p,q BC(E)/(πE )∗H p,q BC (X) ⊕(cid:16)H p,q BC(Z)(cid:17). Remark 3.5. Note that the hypercohomology groups of a sheaf complex are canonically iso- morphic to its Cech hypercohomology groups on manifolds (cf. [16, Theorem 1.3.13 (3)]). The diagram (3.10) is actually equal to a diagram of Cech hypercohomology groups whose com- mutativity can be also verified directly, because all morphisms are induced by the pullback of differential forms. Finally, we turn to the proof of Lemma 3.4. For the pair (X, Z), there exist two surjective sheaf morphisms χ : CX → ι∗CZ and ν : E l X → ι∗ E l Z defined by setting and Γ(V, CX) → Γ(V, ι∗CZ ), f 7→ f V ∩Z , Γ(V, E l X) → Γ(V, ι∗ E l Z ), α 7→ (ιV ∩Z )∗α, for any open subset V ⊂ X such that V ∩ Z 6= ∅, and zero map otherwise. Let KU and Kl U are the kernel sheaves of χ and ν, respectively. It can be directly verified that K• U is a fine resolution of KU , and hence the sheaf cohomology of KU is isomorphic to the cohomology of the complex (Γ(X, K• U ), d) called the relative de Rham cohomology. Moreover, by [31, Proposition 13.11], we have H l(X, KU ) ∼= H l dR(X, Z) ∼= H l dR,c(U ), where H l Chapitre XII] and H l dR(X, Z) is the l-th relative de Rham cohomology in the sense of Godbillon [25, dR,c(U ) is the l-th compactly supported de Rham cohomology of U . / / / / / / / / / / 12 S. YANG AND X. YANG Proof of Lemma 3.4. Recall that the sheaf complex K • X is defined to be (+,−) KU / K0 X ⊕ ¯K0 X / K1 X ⊕ ¯K1 X / · · · / Kp−1 X ⊕ ¯Kp−1 X / ¯Kp X / · · · ¯Kq−1 X / 0. The sheaf KU can be viewed as a complex concentrated in degree 0. Denote D • complex X be the sheaf 0 / K0 X ⊕ ¯K0 X / K1 X ⊕ ¯K1 X / · · · / Kp−1 X ⊕ ¯Kp−1 X / ¯Kp X / · · · ¯Kq−1 X / 0. Note that the following diagram is commutative ... ... ... 0 0 0 / K1 X ⊕ ¯K1 X ∂⊕ ¯∂ / K0 X ⊕ ¯K0 X id id / K1 X ⊕ ¯K1 X ∂⊕ ¯∂ / K0 X ⊕ ¯K0 X / 0 / 0 / 0 / 0 (+,1) id / KU / KU / 0. 0 0 / 0 0 We obtain a short exact sequence of sheaf complexes 0 / D • X [−1] / K • X / KU / 0, and hence a long exact sequence of hypercohomologies: / H l−1(X, KU ) / Hl(X, D • X [−1]) / Hl(X, K • X ) / H k(X, KU ) / Hl+1(X, D • X [−1]) which equals to · · · / H l−1(X, KU ) / Hl−1(X, D • X ) / Hl(X, K • X ) / H l(X, KU ) / Hl(X, D • X ) / · · · Similarly, for the sheaf complex K • X sequence with respect to the pair ( X, E), there is a long exact · · · / H l−1( X, K U ) / Hl−1( X, D • X ) / Hl( X, K • X ) / H l( X, K U ) / Hl( X, D • X ) / · · · Akin to (3.10), the blow-up diagram (3.8) induces a commutative diagram: · · · / H l−1(X, KU ) Hl−1(X, D • X ) Hl(X, K • X ) H l(X, KU ) Hl(X, D • X ) · · · π∗ U π∗ D π∗ π∗ U π∗ D · · · / H l−1( X, K U ) / Hl−1( X, D • X ) / Hl( X, K • X ) / H l( X, K U ) / Hl( X, D • X ) / · · · . On one hand, we have H l(X, KU ) ∼= H l dR,c(U ) and H l( X, K U ) ∼= H l dR,c( U ), (3.15) for any l ∈ N. / / / / / / / / / / / / / / / O O / O O / O O / / O O / O O / O O / / O O / O O / O O / O O O O O O / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / Now we claim that π∗ On the other hand, π U : U → U is biholomorphic. This follows that π∗ U in (3.15) is isomorphic. ) are iso- X ; D1, D2) and ; D1, D2), respectively (see Appendix A.1). According to [15, Theorem 14.14], there morphic to the total cohomologies of the bounded double complexes (K•,• (K•,• X exists a spectral sequence {Er, dr} converging to the total cohomology H(K• D is also isomorphic. Note that H(X, D • X ) and H( X, D • X X ) such that 13 Es,t 1 = H s,t D2 (K• X) = H t(Ks,• X ) ∼= H t(X, D s X ). Likewise, the double complex (K•,• X with the first term ; D1, D2) admits a spectral sequence, denoted by { Er, dr}, X ) = H t(Ks,• X which is converging to the total cohomology H(K• ). X Es,t 1 = H s,t D2 (K• ) ∼= H t( X, D s X ), Consider the Godement resolution D s X = H r(π∗Gs,•). By Lemma A.2, we have Rrπ∗D s X → Gs,•. Then the r-th higher direct image is Rrπ∗D s = 0 for r ≥ 1, i.e., H r(π∗Gs,•) = X 0. Equivalently, this means that π∗Gs,• is an exact sheaf complex and hence a flasque reso- lution of π∗D s X . As a result, by definition, we derive the canonical isomorphisms H l(X, π∗D s X) = H l(Γ(X, π∗Gs,•)) = H l(Γ( X, Gs,•)) = H l( X, D s X ). (3.16) From Lemma A.2, we have the isomorphism π∗ : D s X phism of sheaf cohomologies ≃−→ π∗D s X , which induces an isomor- π∗ : H l(X, D s X ) ≃−→ H l(X, π∗D s (3.17) Observe that the morphism of double complexes π∗ : K•,• the total cohomology groups π∗ : H(K• r : Er → Er for any r ≥ 1. From (3.16) and (3.17), we get an isomorphism π∗ X) → H(K• X X ). X → K•,• X induces a morphism of ) and a morphism of spectral sequences π∗ : H t(X, D s for any 0 ≤ t ≤ n and 0 ≤ s ≤ q − 1. This implies that π∗ to [26, Theorem 6.4.2], we get π∗ morphism of the total cohomologies for the simple complexes is isomorphic, i.e., π∗ isomorphism. ≃−→ H t( X, D s X ) 1 : E1 → E1 is isomorphic. Due r is isomorphic for any r > 1; furthermore, the induced D is an X ) As the morphisms π∗ U and π∗ is the middle one π∗. We have thus proved the lemma. D in (3.15) are isomorphic, from the standard Five Lemma, so (cid:3) 3.2. Case of p = 0 or q = 0. Due to the symmetric property we have H 0,q and therefore in the following we shall prove H p,0 by the same way as used in [47, Proposition 1.2] or [35, Proposition 4.1 of Chapter 1]. BC (−) BC (X). The result can be proved BC( X) ∼= H p,0 BC (−) ∼= H q,0 Proof of Theorem 1.2 (ii). From definition, we have H p,0 BC( X) = {α ∈ Γ( X, Ωp X ) ∂ α = 0} and H p,0 BC (X) = {α ∈ Γ(X, Ωp X ) ∂α = 0}. 14 S. YANG AND X. YANG Put U = X − Z and U = π−1(U ). Let j : U ֒→ X and j : U ֒→ X be the associated inclusions. Note that π is a proper holomorphic map. It follows that the morphism π∗ : H p,0 BC (X) → H p,0 BC( X) (3.18) is injective and hence hp,0 (3.18) is surjective, namely, hp,0 BC ( X) ≥ hp,0 BC ( X) ≤ hp,0 BC (X). BC (X). The next thing to do in the proof is to show that Since π U : U → U is biholomorphic we get an isomorphism H p,0 BC( U ) ∼= H p,0 BC (U ). The inclusion j induces a restriction morphism j∗ : H p,0 BC( X) → H p,0 BC( U ). (3.19) (3.20) We claim that (3.20) is injective. If the assertion was not true, then there exists a nonzero ∂- closed holomorphic p-form on X would vanish on the nonempty open subset U = X − E, and this leads to a contradiction. Using the same argument above, we can verify the injectivity of j∗ : H p,0 BC(X) → H p,0 BC(U ). Let α ∈ H p,0 BC(U ), which is a ∂-closed holomorphic p-form on U . Note that codimCZ ≥ 2, by the Hartogs Extension Theorem α extends over Z and we get a holomorphic p-form β on X such that βU = α. Let ζ = ∂β, then ζ ∈ Γ(X, Ωp+1 X ). Observe that the restriction map j∗ : H p+1,0 BC (X) → H p+1,0 BC (U ) is injective and ζU = ∂α = 0 we get ζ = 0. This means β ∈ H p,0 j∗ : H p,0 BC(U ) is surjective; moreover, we get BC (X) → H p,0 BC (X) and therefore BC(X) ∼= H p,0 Combining (3.19), (3.20) and (3.21) we obtain hp,0 H p,0 BC(U ). BC ( X) ≤ hp,0 BC (X). (3.21) (cid:3) Due to the Weak Factorization Theorem (see Section 4, Theorem 4.3), if an invariant of compact complex manifolds is stable under the blow-ups then it is a bimeromorphic invariant. As a direct result of Theorem 1.2 (ii), we get the following result. Corollary 3.6. If X and Y are two bimeromorphically equivalent compact complex manifolds with the complex dimension n, then for any 0 ≤ p ≤ n we have hp,0 BC (X) = hp,0 BC (Y ). 3.3. Pointed blow-up. If Z = {pt} is a single-pointed space, then we have Corollary 3.7. H p,q BC( X) =( H p,q BC(X) ⊕ C, 1 ≤ p = q ≤ n − 1; H p,q otherwise. BC(X), Proof. As Z = {pt} the exceptional divisor E is biholomorphic to CPn−1. Note that H p,q BC(pt) = 0 for other cases. Ob- serve that the de Rham cohomology ring of CPn−1 is ¯∂ (CPn−1), we have H 0,0 BC(pt) = C and H p,q BC(E) ∼= H p,q H ∗ dR(CPn−1; C) = C[x]/(xn). Because of the Hodge decomposition, we get that the cokernel of the morphism 15 (πE)∗ : H p,q ¯∂ (pt) → H p,q ¯∂ (E) is zero when (p, q) = (0, 0) and isomorphic to H p,q pletes the proof. ¯∂ (CPn−1) when (p, q) 6= (0, 0). This com- (cid:3) As an application of Corollary 3.7, we can prove the following result. Proposition 3.8. Assume that X is a ∂ ¯∂-manifold of dimCX = n. Let π : X → X be the blow-up of X at a point. Then X is a ∂ ¯∂-manifold. Proof. According to Corollary 3.7, we get hp,q BC ( X) =( hp,q BC (X) + 1, hp,q BC (X), if 1 ≤ p = q ≤ n − 1; otherwise. (3.22) For any 0 ≤ k ≤ 2n, using (3.22), the k-th revised non-Kahlerness degree of X is N k( X) = hk BC ( X) − h2n−k BC ( X) BC ( X) − hn−p,n−q BC BC (X) − hn−p,n−q BC ( X)(cid:1) (X)(cid:1) = Xp+q=k(cid:0)hp,q = Xp+q=k(cid:0)hp,q = hk = N k(X). BC (X) − h2n−k BC (X) Because X is a ∂ ¯∂-manifold we have N k(X) = 0, and hence N k( X) = 0. Thus we arrive at the conclusion. (cid:3) Remark 3.9. Compare also [10, Example 2.3] for another different proof of the stability of the ∂ ¯∂-Lemma under the pointed blow-ups. In particular, in [10] the authors consider the orbifold case [10, Theorem 3.1] and construct some new ∂ ¯∂-manifolds [10, Example 3.2]. Remark 3.10. By the duality (2.2) the blow-up formula in Theorem 1.2 is equivalent to H n−p,n−q A ( X) ∼= H n−p,n−q A (E)/(πE )∗H n−p,n−q A (X)M(cid:0)H n−p,n−q A (Z)(cid:1) and this gives rise to a blow-up formula of Aeppli cohomolgy groups. In particular, Alessandrini- Bassanelli [6] proved the following formula where H is the sheaf of germs of pluriharmonic functions. H 1,1 A ( X) = π∗H 1,1 A (X)M π∗H(Z)h[E]i, 4. Non-Kahlerness degrees of threefolds In this section we study the bimeromorphic invariance of non-Kahlerness degrees of com- pact complex threefolds. Definition 4.1. Let X be a compact complex space with dimCX = n. A modification of X is a proper holomorphic map f : X → X satisfying: (i) X is a compact complex space with complex dimension n; 16 S. YANG AND X. YANG (ii) there exists an analytic subset S ⊂ X of codimension ≥ 2 such that f : X − E ≃−→ X − S is a biholomorphism, where E := f −1(S) is called the exceptional set of the modifi- cation; and then clearly f ( X) = X. It is important to notice that the blow-up of a compact complex manifold with a smooth center is a special example of modifications; meanwhile, the modifications are bimeromorphic maps in the following sense. Definition 4.2. A meromorphic map f : X 99K Y of compact complex spaces is a map f of X to the set of subsets of Y satisfying the following conditions: (i) The graph Gf := {(x, y) ∈ X × Y y ∈ f (x)} is an analytic subset of X × Y ; (ii) The projection p1 : Gf → X is a modification. Moreover, if the projection p2 : Gf → Y is also a modification, then f : X 99K Y is called a bimeromorphic map. We say X and Y are bimeromorphically equivalent if there exists a bimeromorphic map f : X 99K Y between them. Suppose now that f : X 99K Y is a bimeromorphic map of compact complex manifolds. By Hironaka's singularity resolution theorem [28], there exists a compact complex manifold Z with a triangle Z π2 ❄❄❄❄❄❄❄❄ π1 ⑧⑧⑧⑧⑧⑧⑧⑧ Y, X := pi ◦ ρ is a modification. where ρ : Z → Gf is a resolution of the graph Gf , and πi There exist many examples of bimeromorphic maps which are not modifications and blow- ups; however, Abramovich-Karu-Matsuki-W lodarczyk [1] showed that any bimeromorphic map between compact complex manifolds is a composition of finite sequences of blow-ups and blow-downs of compact complex manifolds with smooth centers (see also [49]). Theorem 4.3 (Weak Factorization Theorem, [1, Theorem 0.3.1]). Let f : X 99K X be a bimeromorphic map of compact complex manifolds. Then there exists a diagram X = Y0 φ1 99K Y1 φ2 99K · · · φl−1 99K Yl−1 φl 99K Yl = X, where Yi is a compact complex manifold, f = φl ◦ φl−1 ◦ · · · ◦ φ1, and either φi : Yi−1 99K Yi or φ−1 : Yi 99K Yi−1 is a morphism obtained by blowing up a smooth center. i Due to the Theorem 2.2 and Theroem 4.3, to verify the bimeromorphic invariance of the ∂ ¯∂-Lemma for threefolds, it suffices to prove an explicit blow-up formula of Bott-Chern cohomology as Corollary 3.7. Lemma 4.4. Let X be a compact complex threefold and C ⊂ X a smooth curve. Then for any 0 ≤ p, q ≤ 3 we have H p,q BC( X) ∼= H p,q BC(X)M H p−1,q−1 ¯∂ (C), where π : X → X is the blow-up of X along C. Proof. According to Theorem 1.2, the assertion holds if and only if the co-kernel of the morphism 17 (πE)∗ : H p,q BC(C) → H p,q BC (E) ¯∂ equals to H p−1,q−1 (C), for any 0 ≤ p, q ≤ 3. By definition, we get that the exceptional divisor E = π−1(C) is biholomorphic to the projectivization of the normal bundle of C, i.e., E = P(NC/X). Since every compact complex curve is Kahler the ∂ ¯∂-Lemma holds on C. It follows that the Bott-Chern cohomology of C is isomorphic to its Dolbeault cohomology. In addition, note that the exceptional divisor E = P(NC/X ) is a Kahler surface in X. This implies H p,q ¯∂ (E). Let S be the universal subbundle over E and let t = c1(S∗) be the first Chern class of the dual bundle S∗, which is a real de Rham cohomology class of (1, 1)- type. According to the Leray-Hirsch theorem [15, Theorem 5.11], the de Rham cohomology H ∗ dR(E; C) is a free module over H ∗ dR(C; C) with the basis {1, t}, i.e., BC(E) ∼= H p,q H ∗ dR(E; C) ∼= H ∗ dR(C; C) ⊗ {1, t}. More precisely, we have H k dR(E; C) = π∗H k dR(C; C) ⊕ t ∧ π∗H k−2 dR (C; C), where k = p + q. It follows that the co-kernel of the morphism (πE )∗ : H k dR(C; C) → H k dR(E; C) is isomorphic to H k−2 checking we get dR (C; C). Due to the Hodge decomposition and via a careful degree H p,q ¯∂ (E)/(πE )∗H p,q ¯∂ (C) ∼= H p−1,q−1 ¯∂ (C). and hence H p,q BC(E)/(πE )∗H p,q BC(C) ∼= H p−1,q−1 ¯∂ (C). which completes the proof. (cid:3) Remark 4.5. In general, if the submanifold Z is in the class C of Fujiki, then the excep- tional divisor E is also in the class C of Fujiki. Consequently, the ∂ ¯∂-Lemma holds on E and hence the Bott-Chern cohomology H •,• BC(E) is isomorphic to the Dolbeaut cohomology H •,• ¯∂ (E). According to the Hodge decomposition and the projective bundle formula for de Rham cohomology, following the steps in Lemma 4.4, we can prove the following formula H p,q BC ( X) ∼= H p,q BC(X) ⊕(cid:16) r−1 Mi=1 H p−i,q−i ¯∂ (Z)(cid:17). We are ready to prove Theorem 1.3. Proof of Theorem 1.3. Let f : X 99K X be a bimeromorphic map of compact complex three- folds. From Theorem 4.3, there exists a diagram X = Y0 φ1 99K Y1 φ2 99K · · · φl−1 99K Yl−1 φl 99K Yl = X, (4.1) where either φi : Yi 99K Yi−1 is a morphism obtained by blowing up of a compact complex threefold along a point or a smooth curve. According to the : Yi−1 99K Yi or φ−1 i 18 S. YANG AND X. YANG diagram (4.1), to verify ∆k( X) = ∆k(X) for any k ∈ {0, 1, 2, 3, 4, 5, 6}, it suffices to prove ∆k(Yi) = ∆k(Yi+1). Note that X is a compact complex threefold. Without loss of the generality, we may / X is a blow-up of X at a point or along a smooth curve then we divide assume that g : Y the proof in two cases. Case 1. Assume that g : Y → X is a blow-up of X at a point. From Corollary 3.7 we have hp,q BC (Y ) =( hp,q BC (X) + 1, hp,q BC (X), if 1 ≤ p, q ≤ 2 and p = q; otherwise. (4.2) If k ∈ {0, 1, 3, 5, 6}, then from (4.2) and the de Rham blow-up formula [48, Theorem 7.31] we obtain ∆k(Y ) = hk = hk = ∆k(X). BC (Y ) − h6−k BC (X) − h6−k BC (Y ) − 2bk(Y ) BC (X) − 2bk(X) Moreover, in the case of k ∈ {2, 4}, via a direct check we have ∆k(Y ) = hk = (hk = ∆k(X). BC (Y ) + h6−k BC (X) + 1) + (h6−k BC (Y ) − 2bk(Y ) BC (X) + 1) − 2(bk(X) + 1) Case 2. Now we assume that g is the blow-up of X along a smooth curve C. On one hand, since C is a smooth curve, by Lemma 4.4 and for the dimension reason we get hp,q BC (Y ) = hp,q BC (X) + hp−1,q−1 ¯∂ (C) (4.3) (4.4) and h3−p,3−q BC where 2 ≤ p + q ≤ 4; otherwise, hp,q Rham blow-up formula we get (Y ) = h3−p,3−q BC (Y ) = hp,q BC (X) + h2−p,2−q ¯∂ (C) BC (X). On the other hand, according to the de bk(Y ) =( bk(X) + bk−2(C), bk(X), if 2 ≤ k ≤ 4; otherwise. (4.5) If k ∈ {0, 1, 3, 5, 6} then via a straightforward computation we can verify that the equality ∆k(Y ) = ∆k(X) holds. If k = 2 or k = 4, then from the Poincar´e duality on C we have b4−k(C) − bk−2(C) = 0; furthermore, by (4.3)-(4.5) we get ∆k(Y ) = hk BC (Y ) + h6−k BC (Y ) − 2bk(Y ) = (cid:0)hk BC (X) + bk−2(C)(cid:1) +(cid:0)h6−k = ∆k(X) +(cid:0)b4−k(C) − bk−2(C)(cid:1) = ∆k(X). BC (X) + b4−k(C)(cid:1) − 2(cid:0)bk(X) + bk−2(C)(cid:1) The results in Case 1 and Case 2 mean ∆k(Y ) = ∆k(X) for any k ∈ {0, 1, 2, 3, 4, 5, 6}. Likewise, we can show that ∆k(Yi) = ∆k(Yi+1). This implies ∆k( X) = ∆k(X), i.e., the non- Kahlerness degrees for compact complex threefolds are bimeromorphic invariants. According / 19 to Theorem 2.2, we get that X is a ∂ ¯∂-manifold if and only if X is a ∂ ¯∂-manifold and this completes the proof. (cid:3) For compact complex threefolds, Theorem 1.3 can be thought of as a generalization of [17, Theorem 5.22]. In particular, following the steps in the proof of Theorem 1.3 we can show that the revised non-Kahlerness degrees are also bimeromorphic invariants for threefolds. Example 4.6 (Iwasawa manifold). Recall that the 3-dimensional complex Heisenberg group H(3; C) is defined to be the Lie subgroup of GL(3; C) whose elements have the form 1 a b 0 1 c 0 0 1 g =    where a, b, c are complex numbers. It is noteworthy that H(3; C) is a connected and simply connected complex nilpotent Lie group. Let Γ ⊂ H(3; C) be a discrete subgroup such that all entries are Gaussian integers. The Iwasawa manifold is defined to be quotient space I3 := H(3; C)/Γ, which is a compact 6-dimensional smooth manifold. Besides, there exists a H(3; C)-invariant complex structure J0 on I3 determined by the standard complex structure on C3. This enables (I3, J0) to be a non-Kahler, no-formal, and holomorphically parallelizable manifold. In 1975, Nakamura [34] proved that the holomorphically parallelizable property of I3 is not stable under small deformations and the deformed objects of I3 can be divided into three classes. Via computing the Bott-Chern cohomology of Iwasawa manifold I3 and of its small deformations, Angella [7] refined Nakamura's classification. In particular, we have the following table (cf. [12, 11]). ∆1 = ∆5 ∆2 = ∆4 ∆3 I3 2 6 8 (i) 2 6 8 (ii.a) 2 3 8 (ii.b) 2 2 8 (iii.a) 2 1 8 (iii.b) 2 0 8 Table 1. Non-Kahlerness degrees of I3 and its small deformations 5. Concluding remark Based on the Weak Factorization Theorem (Theorem 4.3), to verify the bimeromorphic invariance of the ∂ ¯∂-Lemma in higher dimension case, we need to deal with two problems: to confirm whether the ∂ ¯∂-Lemma survive the closed complex submanifolds and prove a general projective bundle formula for Bott-Chern cohomology. More precisely, let X be a compact complex manifold with dimC X = n > 3. It is unknown whether the following explicit formula still holds If the formula (5.1) is true and any closed complex however, it seems to be reasonable. submanifold of a ∂ ¯∂-manifold is also a ∂ ¯∂-manifold, then by Theorem 2.4 we can verify that the ∂ ¯∂-Lemma is a bimeromorphic invariant of compact complex manifolds. H p,q BC ( X) ∼= H p,q BC(X) ⊕(cid:16) r−1 Mi=1 H p−i,q−i BC (Z)(cid:17); (5.1) 20 S. YANG AND X. YANG For a compact non-Kahler complex surface the non-Kahlerness degrees are trivial bimero- morphic invariants since ∆1 = 0 and ∆2 = 2. Let X be a compact complex threefold. By Theorem 1.3 we know that for each k ∈ {0, 1, 2, 3, 4, 5, 6}, the non-Kahlerness degree ∆k(X) is a bimeromorphic invariant. Example 4.6 means that for compact non-Kahler threefolds the non-Kahlerness degrees, as bimeromorphic invariants, are non-trivial. As a direct conse- quence, we get that the Iwasawa manifold I3 and its small deformations of class (ii) (resp. class (iii)) are not bimeromorphically equivalent. From the bimeromorphic geometry point of view, a natural problem is: Problem 5.1. Whether there exists a uniform upper bound for (revised) non-Kahlerness de- grees of compact non-Kahler threefolds; moreover, how to classify compact non-Kahler three- folds (up to the bimeromorphic equivalence) with respect to the (revised) non-Kahlerness degrees? Note that our proof of the bimeromorphic invariance of non-Kahlerness degrees for three- folds depends on the explicit Bott-Chern blow-up formula at a point or along a smooth curve. Unfortunately, our proof does not enable us to show the bimeromorphic invariance of non-Kahlerness degrees for compact complex manifolds with complex dimensions larger than 3. Update (August 2018). The Bott-Chern projective bundle formula has been claimed by Dr. J. Stelzig in [44] using a structure theory of double complexes [45]. Combining Theorem 1.2 with the result of Stelzig, one can show that the formula (5.1) holds. Consequently, Problem 1.1 descends to the following fundamental problem which is still open; see also [3, 10, 20, 41]. Problem 5.2. Let X be a ∂ ¯∂-manifold, is it true that the ∂ ¯∂-Lemma holds on any closed complex submanifold Z of X? Notice that if the above problem is confirmed positively then from the blow-up formula (5.1), Theorem 2.2 (or 2.4) and Weak Factorization Theorem we can verify that the ∂ ¯∂- Lemma is a bimeromorphic invariant of compact complex manifolds. Appendix A. Properties of D • X A.1. Double complex reinterpretation of D • X sider the sheaf complex . With the situation in Section 3, we con- D • X : 0 / K0 X ⊕ ¯K0 X / K1 X ⊕ ¯K1 X / · · · / Kp−1 X ⊕ ¯Kp−1 X / ¯Kp X / · · · ¯Kq−1 X / 0. ϕ Set Ks,t −→ ι∗ E s,t 3.29], we have a fine resolution X = ker(cid:0) E s,t X Z (cid:1). Due to the Dolbeault-Grothendieck lemma [18, Lemma K0,• X ⊕ ¯K0,• X / K1,• X ⊕ ¯K1,• X K0 X ⊕ ¯K0 X / K1 X ⊕ ¯K1 X / · · · / Kp−1,• X ⊕ ¯Kp−1,• X / ¯Kp,• X / · · · / ¯Kq−1,• X (A.1) / · · · / Kp−1 X ⊕ ¯Kp−1 X / ¯Kp X / · · · / ¯Kq−1 X / / / / / / / / / / / / / ? O O / ? O O / / ? O O / ? O O / / ? O O 21 For the simplicity, we set Ks,t = Γ(X, Ks,t X ), and ¯Ks,t = Γ(X, ¯Ks,t X ). Moreover, for any 0 ≤ t ≤ n, we define Ks,t X = Ks,t ⊕ ¯Ks,t, Ks,t X = ¯Ks,t, for any 0 ≤ s ≤ p − 1, for any p ≤ s ≤ q − 1. By taking the global sections of the resolutions (A.1), we obtain a double complex K•,• the horizontal and vertical differentials X with D1 = ∂ ⊕ ¯∂ : Ks,t X → Ks+1,t X , D2 = ¯∂ ⊕ ∂ : Ks,t X → Ks,t+1 X , satisfying D2 2 = 0 = D2 1 and D2D1 + D1D2 = 0. Denote the simple complex by Then we have an isomorphism Kl X = Ms+t=l Ks,t X . Analogously, we can construct a double complex K•,• X there exists an isomorphism H(X, D • X ) ∼= H(K• X). associated to the pair ( X, E). Moreover, H( X, D • X ) ∼= H(K• X ). A.2. Higher direct images. With the situation in Section 3, we have the blow-up diagram ρ:=πE E Z ι ι / X π / X. Recall that the pullback of differential forms defines a natural surjective morphism of sheaves ϕ : Ωs X E. Thus, there exists a short exact sequence → ι∗Ωs 0 / Ks X / Ωs X ϕ / ι∗Ωs E / 0, (A.2) and we call Ks X induces a long exact sequence of higher direct images along π: the relative Dolbeault sheaf with respect to ( X, E). The sequence (A.2) · · · / Rr−1π∗Ωs X / Rr−1π∗ι∗Ωs E / Rrπ∗ Ks X / Rrπ∗Ωs X / Rrπ∗ι∗Ωs E / · · · . In general, the higher direct image Rrπ∗Ωs X However, the higher direct images of the relative Dolbeault sheaves always vanish. is not vanishing, unless s = 0 or n = dimC X. The following proposition was pointed out to us by Dr. L. Meng [32, Lemma 2.4] and we figured out a detailed proof for the bundle-valued case in [42, Lemma 4.4]. Proposition A.1. Rrπ∗ Ks X = 0 for any r ≥ 1. In particular, for the sheaf D s X the following result holds. Lemma A.2. For any s ≥ 0, π∗ : D s X and Rrπ∗D s X = 0 for r ≥ 1. ⊕ ¯Ks X = Ks X ≃−→ π∗D s X / / / / / / / / / / / / 22 S. YANG AND X. YANG Proof. Note that the direct image functor commutes with the direct sum π∗(Ks X ⊕ ¯Ks X) = π∗ Ks X ⊕π∗ ¯Ks X. ∼= Ks ∼= ¯Ks To prove the assertion holds, it suffices to show that π∗ Ks X X by complex conjugation. More precisely, we only need to show that the induced morphism of stalks X and then we get π∗ ¯Ks X π∗ x : (D s X )x −→ (π∗D s X )x, for any x ∈ X is isomorphic. By definition, for any open subset V ⊂ X we have Γ(V, π∗ Ks X) = {α ∈ Γ( V , Ωs( V )) (ι V ∩E)∗ α = 0}, if V ∩ Z 6= ∅; and Γ(V, π∗ Ks X ) = Γ( V , Ks X) ∼= Γ(V, Ks X), if V ∩ Z = ∅; where V = π−1(V ). If β is a holomorphic s-form on V such that (ιV ∩Z )∗β = 0, the the pullback β := (π V )∗β is a holomorphic s-form on V that is satisfying (ι V ∩E)∗ β = 0. This implies that the pullback π∗ induces a sheaf morphism π∗ : Ks by assigning the map X → π∗ Ks X π∗(V ) : Γ(V, Ks X) → Γ(V, π∗ Ks X), α 7→ (π V )∗α to each open subset V ⊂ X. Moreover, from [42, Lemma 4.4] it can be shown that the sheaf morphism π∗ : Ks is actually isomorphic. X → π∗ Ks X It remains to prove Rrπ∗D s X = 0 for any r ≥ 1. We claim Rrπ∗ ¯Ks Rrπ∗ Ks X given sheaf F on X, its higher direct image Rrπ∗F is the sheafification of the presheaf = 0 for any r ≥ 1. According to Proposition A.1, we have X = 0 for any r ≥ 1. From definition, for a V 7→ H r( V , F) for any open set V ⊂ X and V := π−1(V ). On account of Rrπ∗ Ks X x ∈ X there is an open neighborhood V of x such that H r( V , Ks V complex (Γ( V , Ks,• V implies H r( V , ¯Ks V ) = 0 which concludes the claim Rrπ∗ ¯Ks direct images commutes with direct sums, we have = 0, for every point ) = 0. It means that the ), ∂). This V = 0. By the fact that the higher ), ¯∂) is exact, and hence its complex conjugation (Γ( V , ¯Ks,• V X = Rrπ∗(Ks This finishes the proof of Lemma A.2. Rrπ∗D s X ⊕ ¯Ks X) = Rrπ∗ Ks X ⊕Rrπ∗ ¯Ks X = 0 ⊕ 0. References (cid:3) [1] D. Abramovich, K. Karu, K. Matsuki and J. W lodarczyk, Torification and factorization of birational maps, J. Amer. Math. Soc. 15 (2002) 531-572. 2, 16 [2] A. Aeppli, On the cohomology structure of Stein manifolds, Proc. Conf. Complex Analysis (Minneapolis, Minn., 1964), Springer, Berlin, 1965, pp. 58-70. 4 [3] L. Alessandrini, Proper modifications of generalized p-Kahler manifolds, J. Geom. Anal. 27 (2017) 947- 967. 2, 20 [4] L. Alessandrini and G. Bassanelli, Small deformations of a class of non-Kahler manifolds, Proc. Amer. Math. Soc. 109 (1990) 1059-1062. 2 [5] L. Alessandrini and G. Bassanelli, Metric properties of manifolds bimeromorphic to compact Kahler spaces, J. Differential Geom. 37 (1993) 95-121. 1 23 [6] L. Alessandrini and G. Bassanelli, Modifications of compact balanced manifolds, C. R. Acad. Sci. Paris S´er. I Math. 320 (1995) 1517-1522. 1, 15 [7] D. Angella, The cohomologies of the Iwasawa manifold and of Its small deformations, J. Geom. Anal. 23 (2013) 1355-1378. 1, 19 [8] D. Angella, G. Dloussky and A. Tomassini, On Bott-Chern cohomology of compact complex surfaces, Ann. Mat. Pura Appl. 195 (2016) 199-217. 6 [9] D. Angella and H. Kasuya, Hodge theory for twisted differentials, Complex Manifolds 1 (2014) 64-85. 11 [10] D. Angella, T. Suwa, N. Tardini and A. Tomassini, Note on Dolbeault cohomology and Hodge structures up to bimeromorphisms, arXiv:1712.08889v1 (24 Dec 2017). 3, 15, 20 [11] D. Angella and N. Tardini, Quantitative and qualitative cohomological properties for non-Kahler mani- folds, Proc. Amer. Math. Soc. 145 (2017) 273-285. 6, 19 [12] D. Angella and A. Tomassini, On the ∂ ¯∂-Lemma and Bott-Chern cohomology, Invent. Math. 192 (2013) 71-81. 2, 5, 6, 19 [13] D. Angella, A. Tomassini and M. Verbitsky, On non-Kahler degrees of complex manifolds, to appear in Adv. Geom. arXiv:1605.03368v3. 6 [14] R. Bott and S.-S. Chern, Hermitian vector bundles and the equidistribution of the zeroes of their holo- morphic sections, Acta Math. 114 (1965) 71-112. 4 [15] R. Bott and L. Tu, Differential forms in algebraic topology, Graduate Texts in Mathematics 82, Springer- Verlag Berlin Heidelberg, (1982). 13, 17 [16] J.-L. Brylinski, Loop Spaces, characteristic classes and geometric quantization, Progr. Math. 107, Birkhauser, Boston, (1993). 11 [17] P. Deligne, P. Griffiths, J. Morgan and D. Sullivan, Real homotopy theory of Kahler manifolds, Invent. Math. 29 (1975) 245-274. 1, 2, 4, 19 [18] J.-P. Demailly, Complex analytic and differential geometry, J.-P. Demailly's CADG e-book. 2, 7, 11, 20 [19] P. Dolbeault, Sur la cohomologie des vari´et´es analytiques complexes, C. R. Acad. Sci. Paris 236 (1953) 175-177. 4 [20] R. Friedman, The ∂ ¯∂-lemma for general Clemens manifolds, arXiv:1708.00828v1. 2, 20 [21] A. Frolicher, Relations between the cohomology groups of Dolbeault and topological invariants, Proc. Nat. Acad. Sci. U.S.A. 41 (1955) 641-644. 5 [22] J. Fu, J. Li and S.-T. Yau, Balanced metrics on non-Kahler Calabi-Yau threefolds, J. Differential Geom. 90 (2012) 81-129. 2 [23] J. Fu and S.-T. Yau, A note on small deformations of balanced metrics, C. R. Math. Sci. Paris 349 (2011) 793-796. 2 [24] A. Fujiki, On automorphism groups of compact Kahler manifolds, Invent. Math. 44 (1978) 225-258. 1 [25] C. Godbillon, ´Element de topologie alg´ebrique, Hermann, Paris, 1971. 11 [26] V. Guillemin, S. Sternberg, Supersymmetry and equivariant de Rham theory, Springer-Verlag Berlin Heidelberg, 1999. 13 [27] H. Hironaka, On the theory of Birational blowing-up, Ph.D. thesis, Harvard University, 1960. 1 [28] H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic zero. I, II, Ann. of Math. 79 (1964) 109-203, 205-326. 16 [29] K. Kodaira and D. Spencer, On deformations of complex analytic structures. III: Stability theorems for complex structures, Ann. of Math. 71 (1960) 43-76. 2 [30] S. MacLane, Homology, Grund. Math. Wissen., 114, Academic Press, New York; Springer-Verlag, Berlin- Guottingen-Heidelberg (1963). 11 [31] I. Madsen, J. Tornehave, From calculus to cohomology: de Rham cohomology and characteristic classes, Cambridge University Press, Cambridge, (1997). 11 [32] L. Meng, An explicit formula of blow-ups for Dolbeault cohomology, arXiv:1806.11435v2 (Aug 2018). 3, 21 [33] B. Moishezon, On n-dimensional compact varieties with n algebraically independent meromorphic func- tions, I, Izv. Akad. Nauk SSSR Ser. Mat. 30 (1966) 133-174. 1 24 S. YANG AND X. YANG [34] I. Nakamura, Complex parallelisable manifolds and their small deformations, J. Differential Geom. 10 (1975) 85-112. 19 [35] M. Popa, Modern aspects of the cohomological study of varieties, Lecture notes (2011). 13 [36] D. Popovici, Stability of strongly Gauduchon manifolds under modifications, J. Geom. Anal. 23 (2013) 653-659. 1 [37] D. Popovici, Volume and self-intersection of differences of two nef classes, Ann. Sc. Norm. Super. Pisa Cl. Sci. 17 (2017) 1255-1299. 1, 2 [38] S. Rao, X. Wan and Q. Zhao, Power series proofs for local stabilities of Kahler and balanced structures with ∂ ¯∂-lemma, arXiv:1609.05637v1. 2 [39] S. Rao, X. Wan and Q. Zhao, On local stabilities of p-Kahler structures, arXiv:1801.01031v2. 2 [40] S. Rao and Q. Zhao, Several special complex structures and their deformation properties, J. Geom. Anal. (2017), https://doi.org/10.1007/s12220-017-9944-7. 2 [41] S. Rao, S. Yang and X. Yang, Dolbeault cohomologies of blowing up complex manifolds, arXiv:1712.06749v4 (Aug 2018) to appear in J. Math. Pures Appl. 3, 20 [42] S. Rao, S. Yang and X. Yang, Dolbeault cohomologies of blowing up complex manifolds II: bundle-valued case, arXiv:1809.07277v1. 21, 22 [43] M. Schweitzer, Autour de la cohomologie de Bott-Chern, arXiv:0709.3528v1. 5, 7 [44] J. Stelzig, The double complex of a blow-up, arXiv:1808.02882v1 (Aug 2018). 3, 20 [45] J. Stelzig, On the structure of double complexes, In preparation. 20 [46] A. Teleman, The pseudo-effective cone of a non-Kahlerian surface and applications, Math. Ann. 335 (2006) 965-989. 6 [47] K. Ueno, Classification of algebraic varieties. I, Compositio Math. 27 (1973) 277-342. 13 [48] C. Voisin, Hodge theory and complex algebraic geometry I, Cambridge Studies in Advanced Mathematics, 76. Cambridge University Press, Cambridge, (2007). 2, 18 [49] J. W lodarczyk, Toroidal varieties and the weak factorization theorem, Invent. Math. 154 (2003) 223-331. 16 [50] C.-C. Wu, On the geometry of superstrings with torsion, Ph.D. Thesis, Harvard University (2006). 2 Center for Applied Mathematics, Tianjin University, Tianjin 300072, P. R. China E-mail address: [email protected] Department of Mathematics, Chongqing University, Chongqing 401331, P. R. China E-mail address: [email protected]
0907.0298	3	0907	2010-07-09T10:53:33	Elliptic Surfaces	[ "math.AG", "math.NT" ]	This survey paper concerns elliptic surfaces with section. We give a detailed overview of the theory including many examples. Emphasis is placed on rational elliptic surfaces and elliptic K3 surfaces. To this end, we particularly review the theory of Mordell-Weil lattices and address arithmetic questions.	math.AG	math	ELLIPTIC SURFACES MATTHIAS SCH UTT AND TETSUJI SHIODA Abstract. This survey paper concerns elliptic surfaces with section. We give a detailed overview of the theory including many examples. Emphasis is placed on rational elliptic surfaces and elliptic K3 surfaces. To this end, we particularly review the theory of Mordell-Weil lattices and address arithmetic questions. Contents 1. Introduction 2. Elliptic curves 3. Elliptic surfaces 4. Singular fibres 5. Base change and quadratic twisting 6. Mordell-Weil group and N´eron-Severi lattice 7. Torsion sections 8. Rational elliptic surfaces 9. Extremal rational elliptic surfaces 10. Semi-stable elliptic surfaces 11. Mordell-Weil lattice 12. Elliptic K3 surfaces 13. Arithmetic of elliptic K3 surfaces 14. Ranks of elliptic curves References 1 3 7 10 17 22 32 38 42 45 48 58 72 84 88 1. Introduction Elliptic surfaces are ubiquitous in the theory of algebraic surfaces. They play a key role for many arithmetic and geometric considerations. While this feature Date : February 22, 2010. 2000 Mathematics Subject Classification. Primary 14J27; Secondary 06B05, 11G05, 11G07, 11G50, 14J20, 14J26, 14J28. Key words and phrases. Elliptic surface, elliptic curve, N´eron-Severi group, singular fibre, Mordell-Weil group, Mordell-Weil lattice, Tate algorithm, K3 surface. Partial funding from DFG under grant Schu 2266/2-2 and JSPS under Grant-in-Aid for Scientific Research (C) No. 20540051 is gratefully acknowledged. 1 2 MATTHIAS SCH UTT AND TETSUJI SHIODA has become ever more clear during the last two decades, extensive survey papers and monographs seem to date back exclusively to the 80’s and early 90’s. Hence we found it a good time to comprise a detailed overview of the theory of elliptic surfaces that includes the recent developments. Since we are mainly aiming at algebraic and in particular arithmetic aspects, it is natural to restrict to elliptic surfaces with sections. Albeit this deters us from several interesting phenomena (Enriques surfaces to name but one example), elliptic surfaces with section are useful in general context as well since one can always derive some basic information about a given elliptic surface without section from its Jacobian (cf. [31]; for the general theory see e.g. [50] and the references therein). While we will develop the theory of elliptic surfaces with section in full gener- ality, we will put particular emphasis on the following three related sub jects: • rational elliptic surfaces, • Mordell-Weil lattices, • elliptic K3 surfaces and their arithmetic. Throughout we will discuss examples whenever they become available. Elliptic surfaces play a prominent role in the classification of algebraic surfaces and fit well with the general theory. As we will see, they admit relatively sim- ple foundations which in particular do not require too many prerequisites. Yet elliptic surfaces are endowed with many different applications for several areas of research, such as lattice theory, singularities, group theory and modular curves. To get an idea, consider height theory. On elliptic curves over number fields, height theory requires quite a bit of fairly technical machinery. In contrast, elliptic surfaces naturally lead to a definition of height that very clear concepts and proofs underlie. In particular, it is a key feature of elliptic surfaces among all algebraic surfaces that the structure of their N´eron-Severi groups is well-understood; this property makes elliptic surfaces accessible to direct computations. Throughout the paper, we will rarely give complete proofs; most of the time, we will only briefly sketch the main ideas and concepts and include a reference for the reader interested in the details. There are a few key references that should be mentioned separately: textbooks on elliptic curves: Cassels [10], Silverman [106], [107] textbooks on algebraic surfaces: Mumford [58], Beauville [5], Shafarevich [86], Barth-Hulek-Peters-van de Ven [4] Conway-Sloane [12] textbook on lattices: references for elliptic surfaces: Kodaira [37], Tate [113], Shioda [93], Cossec-Dolgachev [15], Miranda [53] Elliptic surfaces form such a rich sub ject that we could not possibly have treated all aspects that we would have liked to include in this survey. Along the same lines, it is possible, if not likely that at one point or another we might have ELLIPTIC SURFACES 3 missed an original reference, but we have always tried to give credit to the best of our knowledge, and our apologies go to those colleagues who might have evaded out attention. 2. Elliptic curves We start by reviewing the basic theory of elliptic curves. Let K denote a field; that could be a number field, the real or complex numbers, a local field, a finite field or a function field over either of the former fields. Definition 2.1. An elliptic curve E is a smooth pro jective curve of genus one with a point O defined over K . One crucial property of elliptic curves is that for any field K ′ ⊇ K , the set of K ′ -rational points E (K ′ ) will form a group with origin O . Let us consider the case K = C where we have the following analytic derivation of the group law. 2.1. Analytic description. Every complex elliptic curve is a one-dimensional torus. There is a lattice Λ ⊂ C of rank two such that E ∼= C/Λ. Then the composition in the group is induced by addition in C, and the neutral element is the equivalence class of the origin. We can always normalise to the situation where Λ = Λτ = Z + τ Z, im(τ ) > 0. For later reference, we denote the corresponding elliptic curve by Eτ = C/Λτ . Two complex tori are isomorphic if and only if the corresponding lattices are homothetic. (1) 2.2. Algebraic description. Any elliptic curve E has a model as a smooth cubic in P2 . As such, it has a point of inflection. Usually a flex O is chosen as the neutral element of the group law. It was classically shown how the general case can be reduced to this situation (cf. for instance [10, §8]). By Bezout’s theorem, any line ℓ intersects a cubic in P2 in three points P , Q, R, possibly with multiplicity. This fact forms the basis for defining the group law with O as neutral element: Definition 2.2. Any three collinear points P , Q, R on an elliptic curve E ⊂ P2 add up to zero: P + Q + R = O . Hence P +Q is obtained as the third intersection point of the line through O and R with E . The hard part about the group structure is proving the associativity, but we will not go into the details here. MATTHIAS SCH UTT AND TETSUJI SHIODA 4 2.3. Weierstrass equation. Given an elliptic curve E ⊂ P2 , we can always find a linear transformation that takes the origin of the group law O to [0, 1, 0] and the flex tangent to E at O to the line ℓ = {z = 0}. In the affine chart z = 1, the equation of E then takes the generalised Weierstrass form (2) E : y 2 + a1 x y + a3 y = x3 + a2 x2 + a4 x + a6 . Outside this chart, E only has the point O = [0, 1, 0]. The lines through O are exactly ℓ and the vertical lines {x = const} in the usual coordinate system. The indices refer to some weighting (cf. 2.6). Except for small characteristics, there are further simplifications: If char(K ) 6= 2, then we can complete the square on the left-hand side of (2) to achieve a1 = a3 = 0. Similarly, if char(K ) 6= 3, we can assume a2 = 0 after completing the cube on the right-hand side of (2). Hence if char(K ) 6= 2, 3, then we can transform E to the Weierstrass form (3) E : y 2 = x3 + a4 x + a6 . The name stems from the fact that the Weierstrass ℘-function for a given lattice Λ satisfies a differential equation of the above shape. Hence (℘, ℘′ ) gives a map from the complex torus Eτ to its Weierstrass form with a4 , a6 expressed through Eisenstein series in τ . 2.4. Group operations. One can easily write down the group operations for an elliptic curve E given in (generalised) Weierstrass form. They are given as rational functions in the coefficients of K -rational points P = (x1 , y1), Q = (x2 , y2) on E : P + Q operation −P generalised Weierstrass form Weierstrass form (x1 , −y1 − a1 x1 − a3 ) (x1 , −y1) x3 = λ2 + a1λ − a2 − x1 − x2 x3 = λ2 − x1 − x2 y3 = −(λ + a1 )x3 − ν − a3 y3 = −λx3 − ν Here y = λ x + ν parametrises the line through P , Q if P 6= Q, resp. the tangent to E at P if P = Q (under the assumption that Q 6= −P ). 2.5. Discriminant. Recall that we require an elliptic curve to be smooth. If E is given in Weierstrass form (3), then smoothness can be phrased in terms of the discriminant of E : (4) 4 + 27 a2 ∆ = −16 (4 a3 6). For the generalised Weierstrass form (2), the discriminant can be obtained by running through the completions of square and cube in the reverse direction. It is a polynomial in the ai with integral coefficients (cf. [113, §1]). Lemma 2.3. A cubic curve E is smooth if and only if ∆ 6= 0. ELLIPTIC SURFACES 5 (5) 2.6. j-invariant. We have already seen when two complex tori are isomorphic. In this context, it is important to note that the Weierstrass form (3) only allows one kind of coordinate transformation: y 7→ u3 y . x 7→ u2 x, Such coordinate transformations are often called admissible. An admissible trans- formation affects the coefficients and discriminant as follows: a4 7→ a4/u4 , a6 7→ a6/u6 , ∆ 7→ ∆/u12 . It becomes clear that the indices of the coefficients refer to the respective expo- nents of u which we consider as weights. All such transformation yield the same so-called j-invariant: j = −1728 Theorem 2.4. Let E , E ′ be el liptic curves in Weierstrass form (3) over K . If E ∼= E ′ , then j = j ′ . The converse holds if additional ly a4/a′ 4 is a fourth power and a6/a′ 6 is a sixth power in K , in particular if K is algebraical ly closed. . (4 a4)3 ∆ The extra assumption of the Theorem is neccessary, since we have to consider quadratic twists. For each d ∈ K ∗ , y 2 = x3 + d2 a4 x + d3 a6 . (6) Ed : defines an elliptic curve with j ′ = j . However, the isomorphism E ∼= Ed involves √d, thus it is a priori only defined over K (√d). The two elliptic curves with extra automorphisms in 2.7 moreover admit cubic (and thus sextic) resp. quartic twists. 2.7. Normal form for given j-invariant. We want to show that for any field K and any fixed j ∈ K , there is an elliptic curve E over K with j-invariant j . For this purpose, we will cover j = 0 and 123 separately in the sequel. Definition 2.5 (Normal form for j). Let j ∈ K \ {0, 123}. Then the following elliptic curve in generalised Weierstrass form has j-invariant j : (7) E : 1 36 y 2 + x y = x3 − j − 123 x − j − 123 . The remaining two elliptic curve are special because they admit extra auto- morphisms. They can be given as follows: j = 0 : j = 123 : y 2 + y = x3 , y 2 = x3 + x, char(K ) 6= 3, char(K ) 6= 2. 6 MATTHIAS SCH UTT AND TETSUJI SHIODA 2.8. Group structure over number fields. The results for elliptic curves over number fields motivated many developments in the study of elliptic surfaces. We give a brief review of some of the main results. In view of Faltings’ theorem [24], elliptic curves can be considered the most intriguing curves if it comes to rational points over number fields. Namely, a curve of genus zero has either no points at all, or it has infinitely many rational points and is isomorphic to P1 . On the other hand, curves of genus g > 1 have only finitely many points over any number field. Elliptic curves meanwhile feature both cases: (8) Theorem 2.6 (Mordell-Weil). Let E be an el liptic curve over a number field K . Then E (K ) is a finitely generated abelian group: E (K ) ∼= E (K )tor ⊕ Zr . One of the main ingredients of the proof is a quadratic height function on the K -rational points. For elliptic surfaces, the definition of such a height function will feature prominently in the theory of Mordell-Weil lattices. In fact, the height arises naturally from geometric considerations (cf. 11). There are two basic quantities in the Mordell-Weil theorem: the torsion sub- group and the rank of E (K ). The possible torsion subgroups can be bounded in terms of the degree of K by work of Mazur and Merel (cf. [52]). Meanwhile it is an open question whether the ranks are bounded for given number fields K . In fact, elliptic surfaces prove a convenient tool to find elliptic curves of high rank by specialisation, as we will briefly discuss in section 14. For fixed classes of elliptic surfaces, one can establish analogous classifications, bounding torsion and rank. The case of rational elliptic surfaces will be treated in Cor. 11.8. 2.9. Integral points. If we want to talk about integral points on a variety, we always have to refer to some affine model of it. For elliptic curves, we can use the Weierstrass form (3) or the generalised Weierstrass form (2) for instance. Theorem 2.7 (Siegel). Let E be an el liptic curve with some affine model over Q. Then #E (Z) < ∞. We will see that the same result holds for elliptic surfaces under mild conditions. In fact, we can even allow bounded denominators without losing the finiteness property. 2.10. Bad places. On a similar note, one can consider the set of bad places v for an elliptic curve over a global field K . Here E attains a singularity or degenerates over the residue field Kv ; this behaviour is called bad reduction. In case K is a number fields, one usually refers to the bad primes. Since the bad places divide the discriminant by Lemma 2.3, any elliptic curve has good reduction outside a finite set of bad places. The following finiteness result for elliptic curves over Q addresses the inverse problem: ELLIPTIC SURFACES 7 Theorem 2.8 (Shafarevich). Let S be a finite set of primes in Z. Then there are only finitely many el liptic curves over Q up to Q-isomorphism with good reduction outside S . This result also generalises to elliptic surfaces under mild conditions which will appear quite naturally in the following (cf. 6.12). In fact, Shafarevich’s original idea of proof generalises to almost all cases. The generalisation is based on the theory of Mordell-Weil lattices as we will sketch in 11.18. 3. Elliptic surfaces In this section, we introduce elliptic surfaces. We explain the interplay with elliptic curves over one-dimensional function fields using the generic fibre and the Kodaira-N´eron model. The central ingredient is the identification of sections of the surface and points on the generic fibre. 3.1. We shall define elliptic surfaces in a geometric way. Therefore we let k = ¯k denote an algebraically closed field, and C a smooth pro jective curve over k . Later on, we will also consider elliptic surfaces over non-algebraically closed fields (such as number fields and finite fields); in that case we will require the following conditions to be valid for the base change to the algebraic closure. Definition 3.1. An elliptic surface S over C is a smooth pro jective surface S with an elliptic fibration over C , i.e. a surjective morphism f : S → C, such that (1) almost all fibres are smooth curves of genus 1; (2) no fibre contains an exceptional curve of the first kind. The second condition stems from the classification of algebraic surfaces. An exceptional curve of the first kind is a smooth rational curve of self-intersection −1 (also called (−1)-curve). Naturally, (−1)-curves occur as exceptional divisors of blow-ups of surfaces at smooth points. In fact, one can always successively blow-down (−1)-curves to reduce to a smooth minimal model. In the context of elliptic surface, we ask for a relatively minimal model with respect to the elliptic fibration. The next example will show that an elliptic surface need not be minimal as an algebraic surface when we forget about the structure of the elliptic fibration. 3.2. Example: Cubic pencil. Let g , h denote homogeneous cubic polynomials in three variables. Unless g and h have a common factor or involve only two of the coordinates of P2 in total, the cubic pencil S = {λ g + µ h = 0} ⊂ P2 × P1 . defines an elliptic surface S over P1 with pro jection onto the second factor. This surface is isomorphic to P2 blown up in the nine base points of the pencil. This 8 MATTHIAS SCH UTT AND TETSUJI SHIODA is visible from the pro jection onto the first factor. The exceptional divisors are not contained in the fibres: they will be sections. If the base points are not distinct, then S has ordinary double points as sin- gularities. In this case, the minimal desingularisation gives rise to the elliptic surface associated to the cubic pencil. For instance, the normal form for given j-invariant from 2.7 is seen as a cubic pencil with parameter j after multiplying its equation by (j − 123). Thus the associated elliptic surface S is rational with function field k(S ) = k(x, y ). The pencil has a multiple base point at [0, 1, 0]; the multiplicity is seven if char(k) 6= 2, 3, and nine otherwise. We will check in 4.7 how these multiplicities account for a singular fibre with many components. 3.3. Sections. A section of an elliptic surface f : S → C is a morphism f ◦ π = idC . π : C → S such that The existence of a section is very convenient since then we can work with a Weierstrass form (3) where we regard the generic fibre E as an elliptic curve over the function field k(C ). In particular, this implies that the sections form an abelian group: E (k(C )). Here we choose one section as the origin of the group law. We call it zero section and denote it by O . In order to exploit the analogies with the number field case, we shall employ the following conventions throughout this paper: Conventions: (1) Every elliptic surface has a section. (2) Every elliptic surface S has a singular fibre. In particular, S is not iso- morphic to a product E × C . O X ❄ C r Figure 1. Elliptic surface with section and singular fibres In terms of the Weierstrass form (3), the second convention guarantees that the coefficients are not contained in the ground field k , but that they really involve ELLIPTIC SURFACES 9 the function field k(C ). We will discuss elliptic surfaces without singular fibres briefly in relation with minimal Weierstrass forms (cf. 4.8, 4.9). The assumption of a section is fairly strong. In fact, it rules out a number of surfaces that we could have studied otherwise. For instance, every Enriques surface admits an elliptic fibration, albeit never with section. However, we can always pass to the Jacobian elliptic surface. This process guarantees the existence of a section and preserves many properties, for instance the Picard number. The Jacobian of an elliptic fibration on an Enriques surface is a rational elliptic surface. 3.4. Interplay between sections and points on the generic fibre. An el- liptic surface S over C (with section) gives rise to an elliptic curve E over the function field k(C ) by way of the generic fibre. Explicitly, a section π : C → S produces a k(C ) rational point P on E as follows: Let Γ denote the image of C under π in S . Then P = E ∩ Γ. If we return to the example of a cubic pencil, then the base points will in general give nine sections, i.e. points on the generic fibre E . Choosing one of them as the origin for the group law, we are led to ask whether the other sections give independent points. In general, this turns out to hold true (cf. 8.8). Conversely, let P be a k(C )-rational point on the generic fibre E . A priori, P is only defined on the smooth fibres, but we can consider the closure Γ of P in S (so that Γ ∩ E = P ). Restricting the fibration to Γ, we obtain a birational morphism of Γ onto the non-singular curve C . f Γ : Γ → C By Zariski’s main theorem, f Γ is an isomorphism. The inverse gives the unique section associated to the k(C )-rational point P . 3.5. Kodaira-N´eron model. Suppose we are given an elliptic curve E over the function field k(C ) of a curve C . Then the Kodaira-N´eron model describes how to associate an elliptic surface S → C over k to E whose generic fibre returns exactly E . At first, we can omit the singular fibres. Here we remove all those points from C where the discriminant ∆ vanishes (cf. (4)). Denote the resulting punctured curve by C ◦ . Above every point of C ◦ , we read off the fibre – a smooth elliptic curve – from E . This gives a quasi-pro jective surface S ◦ with a smooth elliptic fibration f ◦ : S ◦ → C ◦ . Here one can simply think of the Weierstrass equation restricted to C ◦ (after adding the point at ∞ to every smooth fibre). It remains to fill in suitable singular fibres at the points omitted from C . For instance, if the Weierstrass form (3) defines a smooth surface everywhere, then all fibres turn out to be irreducible. The singular fibres are either nodal or cuspidal rational curves: 10 MATTHIAS SCH UTT AND TETSUJI SHIODA Figure 2. Irreducible singular fibres If the surface is not smooth somewhere, then we resolve singularities minimally. We will give an explicit description of the desingularisation process in the next section where we study Tate’s algorithm in some detail. We will also recall Kodaira’s classification of the possible singular fibres. We conclude this section with a very rough discussion of the uniqueness of the Kodaira-N´eron model. Assume we have two desingularisations S, S which are relatively minimal with respect to the elliptic fibration over C : S S ∼bir S ◦ ∼bir ց ւ C By assumption there is a birational morphism S − − → S . The surface classification builds on the fact that every birational morphism is a succession of smooth blow-ups and blow-downs. By construction, the desingular- isations S and S are isomorphic outside the singular fibres. But then the fibres do not contain any (−1)-curves, since the surfaces are relatively minimal. Hence S ∼= S , and we have proven the uniqueness of the Kodaira-N´eron model. 4. Singular fibres We have seen that the main difference between an elliptic surface and its generic fibre consists in the singular fibres. In this section, we discuss the classification of singular fibres. This centers around Tate’s algorithm. We will mainly focus on the case outside characteristics 2, 3, but also mention pathologies in characteristic p such as wild ramification. 4.1. Classification. For complex elliptic surfaces, the singular fibres were first classified by Kodaira [37]. The fibre type is determined locally by the monodromy. In the general situation which includes positive characteristic, N´eron pioneered a canonical way to determine the singular fibres [59]; later Tate exhibited a simplified algorithm that is valid over perfect fields [113]. It turns out that there are no other types of singular fibres than already known to Kodaira. Except for ELLIPTIC SURFACES 11 characteristics 2 and 3, the fibre types are again determined by very simple local information. The case of non-perfect base fields has recently been analysed by M. Szydlo [110]. In characteristics 2 and 3, Tate’s algorithm can fail as there are new types of singular fibres appearing. Here we will only consider perfect base fields and follow Tate’s exposition [113] (see also [107]). It will turn out that all the irreducible components of a singular fibre are rational curves. If a singular fibre is irreducible, then it is a rational curve with a node or a cusp. In Kodaira’s classification of singular fibres, it is denoted by type I1 resp. II. If a singular fibre is reducible, i.e. with more than one irreducible components, then each irreducible component is a smooth rational curve with self-intersection number -2 (a (-2)-curve) on the elliptic surface. Reducible singular fibres are classified into the following types: • two infinite series In (n > 1) and I∗ n (n ≥ 0), and • five types III, IV, II∗ , III∗ , IV∗ . In general, an elliptic surface can also have multiple fibres. For instance, this happens for Enriques surfaces. In the present situation, however, multiple fibres are ruled out by the existence of a section. 4.2. Tate’s algorithm. In the sequel, we shall sketch the first few steps in Tate’s algorithm to give a flavour of the ideas involved. For simplicity, we shall only consider fields of characteristic different from two. Hence we can work with an extended Weierstrass form y 2 = x3 + a2 x2 + a4 x + a6 . (9) (10) Here the discriminant is given by 2 a6 − 4 a3 4 − 4 a3 ∆ = −27 a2 6 + 18 a2 a4 a6 + a2 2 a2 4 . Recall that singular fibres are characterised by the vanishing of ∆. Since we can work locally, we fix a local parameter t on C with normalised valuation v . (This set-up generalises to the number field case where one studies the reduction modulo some prime ideal and picks a uniformiser.) Assume that there is a singular fibre at t, that is v (∆) > 0. By a translation in x, we can move the singularity to (0, 0). Then the extended Weierstrass form transforms as (11) y 2 = x3 + a2 x2 + t a′ 4 x + t a′ 6 . If t ∤ a2 , then the above equation at t = 0 describes a nodal rational curve. We call the fibre multiplicative. If t a2 , then (11) defines a cuspidal rational curve at t = 0. We refer to additive reduction. The terminology is motivated by the group structure of the rational points on the special fibre. In either situation, the special point (0, 0) is a surface singularity if and only if ta′ 6 . 12 MATTHIAS SCH UTT AND TETSUJI SHIODA 4.3. Multiplicative reduction. Let t ∤ a2 in (11). From the summand a3 2 a6 of ∆ in (10), it is immediate that (0, 0) is a surface singularity if and only if v (∆) > 1. If v (∆) = 1, then (0, 0) is only a singularity of the fibre, but not of the surface. Hence the singular fibre at t = 0 is the irreducible nodal rational curve given by (11), cf. Fig. 2. Kodaira introduced the notation I1 for this fibre type. So let us now assume that v (∆) = n > 1. We have to resolve the singularity at (0, 0) of the surface given by (11). This can be achieved as follows: Let m = ⌊ n 2 ⌋ be the greatest integer not exceeding n/2. Then translate x such that tm+1 a4 : y 2 = x3 + a2 x2 + tm+1 a′ (12) 4 x + a6 . Then v (∆) = n is equivalent to v (a6) = n. Now we blow-up m times at the point (0, 0). The first (m − 1) blow-ups introduce two P1 ’s each. These exceptional divisors are locally given by y ′2 = a2 (0) x′2 , so they will be conjugate over k(pa2 (0))/k . This extension appears on the nodal curve in (11) as the splitting field of the tangents to the singular point. According to whether pa2 (0) ∈ k , one refers to split and non-split multiplicative reduction. After each blow-up, we continue blowing up at (0, 0), the intersection of the exceptional divisors. At the final step, the local equation of the special fibre is y 2 = a2 (0) x2 + (a6/t2m )(0). This describes a single rational component, if n = 2m is even, or again two components if n = 2m + 1 is odd. At any rate, the blown-up surface is smooth locally around t = 0. In total, the desingularisation has added (n − 1) rational curves of self-intersection −2. Hence the singular fibre consists of a cycle of n rational curves, meeting transversally. Following Kodaira, we refer to this fibre type as In . I2 ✁✁❆ ❆ ❆ ✁ ✁ I3 ❆ ❆ ❆❆ ✁ ✁ ✁ ❆ ✁ . . . ✁ ❆ ✁ ❆ ✁ n comp ❆ ❆ ❆ ✁ ✁ . . . In (n > 3) ❆ ✁ ❆ ✁ Figure 3. Singular fibre of type In (n > 1) ELLIPTIC SURFACES 13 4.4. Additive reduction. We now let t a2 in (11). We have to distinguish whether (0, 0) is a surface singularity, i.e. whether t a′ 6 . If the characteristic is different from 2 and 3, then this is equivalent to v (∆) > 2. In characteristics 2 and 3, however, the picture is not as uniform. It is visible from (10) that v (∆) ≥ 3 if char(k) = 3. In characteristic 2, one even has v (∆) ≥ 4 in case of additive reduction. This is due to the existence of wild ramification which we will briefly comment on in section 4.6. In any characteristic, if (0, 0) is a smooth surface point, then the singular fibre is a cuspidal rational curve – type II, cf. Fig. 2. If (0, 0) is a surface singularity, we sketch the possible singular fibres resulting from the desingularisation process. There are three possibilities for the excep- tional divisor of the first blow-up: (1) a rational curve of degree two, meeting the strict transform of the cuspidal curve tangentially in one point; (2) Two lines, possibly conjugate in a quadratic extension of k , meeting the strict transform of the cuspidal curve in one point; (3) a double line. In the first two cases, we have reached the desingularisation and refer to type III resp. IV. The third case requires further blow-ups, each introducing lines of multiplicity up to six and self-intersection −2. The resulting non-reduced fibres are visualised in the following figure. Multiple components appear in thick print. Unless they have a label, their multiplicity will be two. III III∗ 4 3 3 ❆ ✁ ❆ ❆ ✁ ✁ ✁ ✁ ❆ ❆ ✁✁ IV ❆❆ II∗ 3 6 4 5 4 ✘✘✘✘ ✘✘✘✘ I∗ 0 ❳❳❳❳ ❳❳❳❳ I∗ 1 ✁ ❆ ✁✁ ✁ ❆ ❆❆ IV∗ 3 ✘✘✘✘ ❳❳❳❳ ✘✘✘✘ ❳❳❳❳ ✘✘✘✘ ❳❳❳❳ I∗ n (n > 1) ✡✡ ✡ ... (n + 1) double comp ❏ ❏❏ Figure 4. Reducible additive singular fibres 14 MATTHIAS SCH UTT AND TETSUJI SHIODA 4.5. Additive fibres in characteristics different from 2,3. If the character- istic differs from 2 and 3, then we can determine the type of the singular fibre directly from the discriminant and very few local information – as opposed to having to go through Tate’s algorithm in full detail. In essence, this is due to the Weierstrass form (3) that we can reduce to. Here we only have to consider the vanishing orders of a4 and a6 next to ∆. Assume that v (∆) = n > 0. Then the type of the singular fibre can be read off from the following table. A single entry means equality, while otherwise we will indicate the possible range of valuations. fibre type v (a6 ) ≥ 0 0 I0 I∗ 0 In (n > 0) II III IV v (a4) ( 0 ≥ 0 0 ≥ 1 1 ≥ 2 ( 2 ≥ 2 I∗ 2 n−6 (n > 6) IV∗ ≥ 3 III∗ 3 II∗ ≥ 4 Table 1. Vanishing orders of Weierstrass form coefficients 0 1 ≥ 2 2 ≥ 3 3 3 4 ≥ 5 5 (13) in the multiplicative case; in the additive case; The following relation is worth noticing: singular fibre (cid:19) = (v (∆) # (cid:18)components of v (∆) − 1 4.6. Wild ramification. The notion of wild ramification was introduced to ac- count for a possible discrepancy in the additive case in (13). Following Ogg’s foundational paper [68] (see also [76]), we denote the index of wild ramification at an additive fibre by δv : δv = v (∆) − 1 − #(components of an additive fibre). In characteristics 2 and 3, some additive fibre types imply higher vanishing order of ∆ than predicted by the number of components. We have seen this happen for the cuspidal rational curve (type II) in 4.4. In order to determine the fibre type, one thus has to go through Tate’s algorithm – as opposed to the other ELLIPTIC SURFACES 15 characteristics where we read off the fibre type from discriminant and Weierstrass form. If a fibre type admits wild ramification, then the index of wild ramification is unbounded within all elliptic surfaces. By a simple case-by-case analysis (see [84, Prop. 16]), one finds the following information on the index of wild ramification where again a single entry indicates equality: fibre type p = 2 p = 3 In (n ≥ 0) 0 0 ≥ 1 ≥ 2 II ≥ 1 III 0 ≥ 1 IV 0 1 0 (n = 1) ≥ 2 (n 6= 1) 0 IV∗ ≥ 1 0 III∗ ≥ 1 0 II∗ ≥ 1 ≥ 1 Table 2. Wild ramification indices I∗ n S : (14) 4.7. Example: Normal form for j. Recall the normal form for an elliptic surface of given j-invariant from 2.7. Here we want to interpret the normal form itself as an elliptic surface over P1 with parameter j = t. Hence the generic fibre is defined over k(t). An integral model is obtain from (7) by a simple scaling of x and y as in (5): y 2 + (t − 123) x y = x3 − 36 (t − 123)3 x − (t − 123)5 . We want to determine the singular fibres of S . We first compute the discriminant ∆ = t2 (t − 123)9 . The fibres at 0 and 123 are easily seen to be additive. If char(k) 6= 2, 3, then translations in y and subsequently x yield the Weierstrass form 1 1 y 2 = x3 − t (t − 123)3 x + t (t − 123)5 . 48 864 By the above table, the singular fibres have type II at t = 0 and III∗ at t = 123 . If char(k) = 2 or 3, then the two additive fibres collapse so that vt=0 (∆) = 11. Tate’s algorithm only terminates at fibre type II∗ . Hence there is wild ramification of index 1. The extra fibre component is due to the collision of all base points mod 2 and 3 (cf. 3.2). In any characteristic, we obtain an integral model at ∞ by the variable change y 7→ y/s3 . t 7→ 1/s, x 7→ x/s2 , (15) 16 MATTHIAS SCH UTT AND TETSUJI SHIODA This results in the discriminant ∆ = s (1 − 123 s)9 . Since the vanishing order of ∆ at s = 0 is one, there has to be a singular fibre of type I1 , a nodal rational curve. 4.8. Minimal Weierstrass form. From the table in section 4.5 we also see why Tate’s algorithm terminates: If it returns none of the above fibre types, then v (a6) ≥ 6. v (a4 ) ≥ 4, But then we can apply the change of variables x 7→ t2 x, y 7→ t3 y to obtain an isomorphic equation that is integral at t. (For the generalised Weier- strass form (2), this requires v (ai ) ≥ i after suitable translations of x, y .) Such a transformation lets v (∆) drop by 12, hence it can only happen a finite num- ber of times. We call this process minimalising and the resulting equation the minimal Weierstrass form (locally at t = 0). The notion of minimality involves a little subtlety. Namely, we work in the coordinate ring k [C ] of a chosen affine open of the base curve C . For instance, if the base curve C is isomorphic to P1 , then k [C ] = k [t] and every ideal in k [C ] is principal. In consequence, an elliptic surface over P1 admits a globally minimal Weierstrass form. In general, however, k [C ] need not be a principal ideal domain, and thus there might not be a globally minimal Weierstrass form. As a strange twist, the Tate algorithm itself can also have the converse effect and make a minimal Weierstrass form non-minimal. For instance, in the multi- plicative case the translation in x yielding (12) may well be non-minimal at ∞ if m is big compared to the degrees of the coefficients ai (cf. 4.10). 4.9. Application: Elliptic surfaces without singular fibres. We give an example of an elliptic surface without singular fibres which is nonetheless not isomorphic to a product of curves. The example builds on the local nature of a minimal Weierstrass form. Let C be a hyperelliptic curve of genus g (C ) > 0. We can give C by a polyno- mial h ∈ k [u] of degree at least three without multiple factors: v 2 = h(u). C : Let u0 denote any ramification point of the double cover C → P1 , so h(u) = (u − u0) h′ . Consider the following elliptic surface S over C : y 2 = x3 + (u − u0)2 x + (u − u0)3 . S : Then S has discriminant ∆ = c (u − u0)6 , but S does not have a singular fibre at (u0 , 0). The above Weierstrass form for S is not minimal at u0 , since locally h′ is invertible. Hence we can write h′ (cid:19)3 h′ (cid:19)2 y 2 = x3 + (cid:18) v 2 x + (cid:18) v 2 S : . ELLIPTIC SURFACES 17 Then rescaling x, y eliminates the local parameter v . The resulting surface is smooth and thus locally minimal at (u0 , 0). 4.10. Elliptic surfaces over the pro jective line. In this survey, we will mostly consider elliptic surfaces with base curve P1 . Such surfaces have a globally mini- mal generalised Weierstrass form. In terms of the Weierstrass form with polyno- mial coefficients, minimality requires v (a4) < 4 or v (a6 ) < 6 at each place of P1 . After minimalising at all finite places, we fix the smallest integer n such that deg(ai ) ≤ ni. As in 4.7, we derive a local equation at ∞ with coefficients i = sni ai (1/s). a′ Alternatively, we can homogenise the coefficients ai (t) as polynomials in two variables s, t of degree n i. Then the discriminant is a homogeneous polynomial of degree 12n. This approach leads to the common interpretation of the minimal Weierstrass form of an elliptic surface over P1 with section as a hypersurface in weighted pro jective space P[1, 1, 2n, 3n]. Here the non-trivial weights refer to x and y respectively. This interpretation allows one to apply several results from the general theory of hypersurfaces in weighted pro jective spaces to our situation (cf. 14.1). The integer n has an important property: It determines how an elliptic surface over P1 with section fits into the classification of pro jective surfaces. We let κ denote the Kodaira dimension. rational elliptic surface, κ = −∞, n = 1 n = 2 K3 surface, κ = 0, honestly elliptic surface, κ = 1. n > 2 The invariant n is equal to the arithmetic genus χ which will be introduced in 6.7. In general, elliptic surfaces over curves of positive genus do also have Kodaira dimension one. We will cover rational elliptic surfaces and elliptic K3 surfaces with section in some detail in sections 8 and 12. 5. Base change and quadratic twisting This section is devoted to the study of base change as one of the most fun- damental operations in the context of elliptic surfaces. There are immediate applications to the problem of constructing specific elliptic surfaces that will re- cur throughout this survey. Base change also motivates other considerations such as quadratic twisting or inseparability. 18 MATTHIAS SCH UTT AND TETSUJI SHIODA 5.1. Base change provides a convenient method to produce new elliptic surfaces from old ones. The set-up is as follows: Suppose we have an elliptic surface f −→ C. In order to apply a base change, we only need another pro jective curve B mapping surjectively to C . Formally the base change of S from C to B then is defined as a fibre product: S S ×C B −→ B ↓ ↓ f−→ C S In practice, we simply pull-back the Weierstrass form (or in general the equation) of S over C via the morphism B → C . Clearly this replaces smooth fibres by smooth fibres. 5.2. Singular fibres under base change. The effect of a base change on the singular fibres depends on the local ramification of the morphism B → C . Singu- lar fibres at unramified points are replaced by a fixed number of copies in the base change (the number being the degree of the morphism). If there is ramification, we have to be more careful. Of course the vanishing orders of the polynomials of the Weierstrass form and of the discriminant multiply by the ramification index. This suffices to solve the base change problem for multiplicative fibres: a base change of local ramification index d replaces a fibre of type In (n ≥ 0) by a fibre of type Ind . For additive fibres, however, the pull-back of the Weierstrass form might become non-minimal. In the absence of wild ramification, the singular fibre resulting from the base change can be predicted exactly. For instance, if we have a Weierstrass form, then this can be read off from the vanishing orders of a4 , a6 and ∆. In particular it is also evident how often we have to minimalise. Depending on the index d of ramification modulo some small integer, Table 3 collects the behaviour of singular fibres without wild ramification: II, II∗ d mod 6 fibre 1 II IV 2 I∗ 3 0 IV∗ 4 II∗ 5 6 I0 III, III∗ d mod 4 fibre 1 III I∗ 2 0 III∗ 3 4 I0 IV, IV∗ d mod 3 fibre 1 IV IV∗ 2 3 I0 d mod 2 fibre I∗ 1 nd 2 Ind I∗ n Table 3. Singular fibres under ramified base change ELLIPTIC SURFACES 19 In the presence of wild ramification, the local analysis can be much more com- plicated. We will see an example in 5.5. Confer [56] for a general account of that situation. By inspection of the tables, additive fibres are potentially semi-stable: after a suitable base change, they are replaced by semi-stable fibres, i.e. multiplicative fibres of type In (n ≥ 0). 5.3. Base change by the j-map. Often it is convenient to consider the j- invariant of an elliptic surface S over C as a morphism from C to P1 . To em- phasise this situation, we will also refer to the j-map. Assume that j 6= const. If j : C → P1 is separable, then it is unramified outside a finite sets of points. Usu- ally the ramification locus includes 0 and 123 and often also ∞. Recall that these are exactly the places of the singular fibres of the normal form for the j-invariant. Thus the question arises to which extent elliptic surfaces admit an interpre- tation as base change from the normal form via the j-map. In this context, it is crucial that the j-invariant determines a unique elliptic curve up to isomor- phism only under the assumption that the ground field K is algebraically closed (Thm. 2.4). Since here we work over function fields, there might in general be some discrepancy between the given elliptic surface with j-map j and the base change of the normal form by j . By 2.6, this discrepancy is related to quadratic twisting. 5.4. Quadratic twisting. Two elliptic surfaces with the same j-map have the same singular fibres up to some quadratic twist. In particular, the smooth fibres are isomorphic over ¯k , the algebraic closure of the ground field. If char(k) 6= 2, then any elliptic surface with section admits an extended Weier- strass form (9). Hence quadratic twisting can be understood in analogy with (6): Sd : y 2 = x3 + da2x2 + d2a4x + d3a6 . (16) Here we point out the effect of a quadratic twist on the singular fibres. Again, this can be read off directly from Tate’s algorithm: IV ↔ II∗ . III ↔ III∗ , II ↔ IV∗ , In ↔ I∗ n , Note that any two elliptic surfaces that are quadratic twists of each other become isomorphic after a suitable finite base change. From 5.2 and (16), it is immediate that this only requires even ramification order at all twisted fibres. In characteristic 2, there is a different notion of quadratic twisting, related to the Artin-Schreier map t 7→ t2 + t (see for instance [84, p. 10]). 5.5. Example: Quadratic base change of normal form for j. As in 4.7, we consider the normal form for given j-invariant as an elliptic surface over P1 with coordinate t. To look into an example we want to apply the following quadratic 20 MATTHIAS SCH UTT AND TETSUJI SHIODA base change π : P1 → P1 7→ t2 + 123 . t The base change ramifies at 123 and ∞. By 5.2, the base changed surface S ′ has singular fibres II, II, I∗ 0 , I2 unless char(k) = 2, 3. To verify this, we pull back the integral model (14) via π ∗ : y 2 + t2 x y = x3 − 36 t6 x − t10 . S ′ : This elliptic curve is not minimal at t = 0. A change of variables x 7→ x t2 , y 7→ y t3 yields the minimal form y 2 + t x y = x3 − 36 t2 x − t4 S ′ : with I∗ 0 fibre at t = 0 and I2 at ∞. If char(k) 6= 2, then we can complete the square on the left-hand side: 1 t2 x2 − 36 t2 x − t4 . 4 Now we can apply quadratic twists by choosing an even number of places in P1 . For instance, twisting at 0 and α in effect moves the singular fibre of type I∗ 0 from t = 0 to α: y 2 = x3 + S ′ : 1 S : y 2 = x3 + t (t − α) x2 − 36 (t − α)2 x − t (t − α)3 . 4 By choosing α = ±24√−3, we can also merge fibres of type II and I∗ 0 to obtain type IV∗ . Similarly, twisting at 0 and ∞ results in a fibre of type I∗ 2 at ∞. We conclude by noting the singular fibres and ramification indices of S ′ in characteristics 2 and 3. In either characteristic, S ′ has two singular fibres, one of which at ∞ has type I2 . The other singular fibre at 0 has type III∗ with ramification index one if char(k) = 2, and IV∗ with ramification index two if char(k) = 3. Both claims are easily verified with Tate’s algorithm. 5.6. Construction of elliptic surfaces. The ideas from this section can also be used in the reversed direction to construct elliptic surfaces over a given curve C with prescribed singular fibres. Namely one has to find a polynomial map C → P1 whose ramification at 0, 123 , ∞ is compatible with the singular fibres. Here com- patibility might involve a quadratic twist. For instance, an elliptic surface over P1 with singular fibres of types II, II, I∗ 2 comes from the normal form after the base change in 5.5, but one has to apply the quadratic twist at 0 and ∞ after the base change. As a special case, we are led to consider polynomial maps to P1 which are unramified outside three points (usually normalised as 0, 1, ∞). These maps are often called Belyi maps. Belyi proved that a pro jective curve can be defined over ELLIPTIC SURFACES 21 a number field if and only if it admits a Belyi map. We will see in 10.2 how this problem can be translated into a purely combinatorial question. Thus we will be able to answer existence questions by relatively easy means. 5.7. Uniqueness. By 2.6, an elliptic curve is determined by its j-invariant if it is defined over an algebraically closed field. With elliptic surfaces resp. curves over function fields k(C ), we have seen that quadratic twisting causes some serious trouble. Obviously, another obstruction occurs when j is constant – despite the surface not being isomorphic to a product. Such elliptic surfaces are called isotrivial, see 9.8 for details. If on the other hand j is not constant, we deduce that the elliptic surface is uniquely determined (up to ¯k -isomorphism) by the pair (j, ∆). For semi-stable surfaces (i.e. without additive fibres), this criterion can be improved as follows: Theorem 5.1. Let S, S ′ denote semi-stable el liptic surfaces with section over the same projective curve C . Then S and S ′ are isomorphic as el liptic surfaces if and only if ∆ = u12∆′ for some u ∈ k∗ . Over ¯k , the theorem can be rephrased as an equality of divisors on the curve C (again only valid for semi-stable elliptic surfaces): (∆) = (∆′ ). Here we have to distinguish from isomorphisms of the underlying surfaces which forget about the structure of an elliptic fibration. For instance, K3 surfaces may very well admit several non-isomorphic elliptic fibrations (cf. 12.2). 5.8. Inseparable base change. Since the discriminant encodes the bad places of an elliptic curve over a global field, Thm. 5.1 brings us back to the theme of Shafarevich’s finiteness result for elliptic curves over Q (Thm. 2.8). A proof of the function field analogue would only require to show that there are only finitely many discriminant with given zeroes on the curve C . However, such a general statement is bound to fail in positive characteristic. This failure is due to the existence of inseparable base change. Namely, let the elliptic surface S → C be defined over the finite field Fq . Then we can base change S by the Frobenius morphism C → C which raises coordinates to their q -th powers. Clearly this does not change the set of bad places, but only the vanishing orders of ∆. For instance, at the In fibres the vanishing orders are multiplied by q . Hence the inseparable base change results in a fibre of type Iqn . 5.9. Conductor. We explain one possibility to take inseparability and wild ram- ification into account. We associate to the discriminant ∆ a divisor N on the base curve C which we call the conductor: uv =  if the fibre at v is smooth; 0, if the fibre at v is multiplicative; 1,  2 + δv , if the fibre at v is additive. N = Xv∈C uv v with 22 MATTHIAS SCH UTT AND TETSUJI SHIODA By definition (∆) − N is effective, in particular deg(∆) ≥ deg(N ). The two degree are related by the following result: Theorem 5.2 (Pesenti-Szpiro [73, Thm. 0.1]). Let S → C be a non-isotrivial el liptic surface over ¯Fp . Let the j-map have inseparability degree e. Then deg(∆) ≤ 6 pe (deg(N ) + 2g (C ) − 2). In characteristic zero the above result has been known to hold after removing the factor pe (see [91] and also [107, p. 287]). Once we understand the structure of the N´eron-Severi group of elliptic surfaces, we will apply the theorem to prove a function field anlogue of Shafarevich’s finiteness result over Q in 6.12. 6. Mordell-Weil group and N´eron-Severi lattice In 3.4 we have seen how a rational point on the generic fibre gives rise to a section on the corresponding elliptic surface and vice versa. In this section we in- vestigate this relation in precise detail. Along the way, we will also compute many fundamental invariants of elliptic surfaces such as canonical divisor, arithmetic genus, Euler number, Betti and Hodge numbers. 6.1. We fix some notation and terminology: As usual, we let S → C be an elliptic surface over k with generic fibre E over k(C ). The K -rational points E (K ) form a group which is traditionally called Mordell-Weil group. We will usually denote the points on E by P , Q etc. Each point P determines a section C → S which we interprete as a divisor on S . To avoid confusion, we shall denote this curve by ¯P . Without further distinguishing, we will also consider ¯P as an element in the N´eron-Severi group of S NS(S ) = {divisors}/ ≈ . Here ≈ denotes algebraic equivalence. For instance, all fibres of the elliptic fibration S → C are algebraically equivalent. The rank of NS(S ) is called the Picard number: ρ(S ) = rank(NS(S )). By the Hodge index theorem, NS(S ) has signature (1, ρ(S ) − 1). 6.2. Mordell-Weil group vs. N´eron-Severi group. We now state the three fundamental results that relate the Mordell-Weil group and the N´eron-Severi group of an elliptic surface with section. All theorems require our assumption that the elliptic surface has a singular fibre. Theorem 6.1. E (K ) is a finitely generated group. This result is a special case of the Mordell-Weil theorem, in generality for abelian varieties over suitable global fields (cf. [45, §6], [85, §4]). Here we will sketch the geometric argument from [93]. The first step is to prove the corre- sponding result for the N´eron-Severi group. ELLIPTIC SURFACES 23 Theorem 6.2. NS(S ) is finitely generated and torsion-free. The finiteness part is again valid in more generality for pro jective varieties as a special case of the theorem of the base [45, §5]. On an elliptic surface, one can use intersection theory to prove both claims (cf. Thm. 6.5). The connection between Thm. 6.1 and 6.2 is provided by a third result: Theorem 6.3. Let T denote the subgroup of NS(S ) generated by the zero section and fibre components. Then the map P 7→ ¯P mod T gives an isomorphism E (K ) ∼= NS(S )/T . The following terminology has established itself for elliptic surfaces: fibre com- ponents are called vertical divisors, sections horizontal. In this terminology, the above theorem states that NS(S ) is generated by vertical and horizontal di- visors (as opposed to involving multisections). ✡ ❏ ✡ ❏ ✡ ✡ ✡ ❏ ❏ ❏ O r Figure 5. Horizontal and vertical divisors X ❄ C The idea here is to prove Thm.s 6.2 and 6.3, since then Thm. 6.1 follows immediately. In the sequel, we sketch the main line of proof. Details can be found in [93]. We will restrict ourselves to those ideas which shed special light on elliptic surfaces or which are particularly relevant for our later issues. 6.3. Numerical Equivalence. On any pro jective surface, algebraic equivalence implies numerical equivalence ≡ where we identify divisors with the same intersection behaviour. Hence the intersection of divisors defines a symmetric bilinear pairing on NS(S ). It endows NS(S ) (up to torsion) with the structure of an integral lattice, the N´eron-Severi lattice. The reader is referred to [12] for generalities on lattices; note that in this survey we are only concerned with non-degenerate lattices (which will sometimes be stated explicitly in order to stress the condition, but often be omitted). In the following, we will use the phrases N´eron-Severi group and N´eron-Severi lattice interchangeably. Lemma 6.4. Modulo numerical equivalence, the N´eron-Severi group is finitely generated. 24 MATTHIAS SCH UTT AND TETSUJI SHIODA The lemma is a direct consequence of the existence of the cycle map γ : NS(S ) → H 2(S ). Here one can work with ℓ-adic ´etale cohomology in general (ℓ a prime differ- In particular H 2(S ) is a finite-dimensional Qℓ - ent from the characteristic). vectorspace, equipped with cup-product which is a non-degenerate pairing with values in H 4(S ) ∼= Qℓ . Notably, γ preserves the pairing. In consequence, the kernel of γ comprises exactly the elements which are numerically equivalent to zero. But then NS(S )/ ≡ embeds into a finite-dimensional vector space. Hence this quotient is finitely generated and torsion free. Thus, Thm. 6.2 will follow once we have proved the following result: Theorem 6.5. On an el liptic surface, algebraic and numerical equivalence coin- cide. This result is very convenient for practical reasons, since we can now solve problems concerning divisor classes in NS(S ) simply by calculating intersection numbers. 6.4. The trivial lattice. Thm. 6.3 introduced the trivial lattice T , generated by the zero section and fibre components. Since any two fibres are algebraically equivalent, we only have to consider a general fibre and fibre components not met by the zero section. We set up some notation for the remainder of this paper. As f−→ C with zero section O . usual it refers to an elliptic surface S F Fv mv Σ R Θv,0 Θv,i Tv a general fibre the fibre f −1(v ) above v ∈ C the number of components of the fibre Fv the points of C underneath singular fibres: Σ = {v ∈ C ; Fv is singular} the points of C underneath reducible fibres: R = {v ∈ C ; Fv is reducible} the component of Fv met by the zero section – the zero component the other components of Fv (i = 1, . . . , mv − 1) the lattice generated by fibre components in Fv not meeting the zero section: Tv = hΘv,i ; 1 ≤ i ≤ mv − 1i. In this notation, the trivial lattice T ⊆ NS(S ) is defined as the orthogonal sum T = h ¯O, F i ⊕ Mv∈R (17) Tv . Proposition 6.6. The divisor classes of { ¯O, F , Θv,i ; v ∈ R, 1 ≤ i ≤ mv − 1} form a Z-basis of T . In particular, rank(T ) = 2 + Xv∈R (mv − 1). ELLIPTIC SURFACES 25 To see that the given divisor classes form a basis, one can compute the inter- section matrix M of T and verify that det(M ) 6= 0. Here it suffices to consider the separate summands from (17). The first summand has intersection form (cid:18) ¯O2 1 0(cid:19) 1 This matrix has determinant −1 and signature (1, 1). We claim that all other summands Tv are negative-definite even lattices, so that T has signature (1, 1 + Pv∈R (mv − 1)). 6.5. Dynkin diagrams. Recall that every component of a reducible fibre is a rational curve of self-intersection −2. We associate a graph to each fibre by drawing a vertex for each component and connecting two vertices by an edge for each intersection point of the corresponding components. For instance, type In for n > 1 gives a circle of n vertices, and the same applies to III (n = 2) and IV (n = 3). It is an elementary observation going through all fibre types that these graphs correspond exactly to the extended Dynkin diagrams Am , Dm , Em . One can re- cover the Dynkin diagrams Am , Dm , Em by omitting the vertex corresponding to the zero component and the edges attached to it. The following figure sketches the Dynkin diagrams. The circles indicate components intersecting the zero com- ponent (as opposed to the balls). A1 ❜ r ❜ D4 r ❜ E6 r ❜ r r r r r r r r r r r A2 ❜ ❜ r r D5 r E7 r ❜ An (n > 2) . . . ❜ Dn (n > 5) . . . ❜ r ❜ r r r r r r r E8 r r r ❜ Figure 6. Dynkin diagrams Lemma 6.7. Each Dynkin diagram defines an even negative-definite lattice, de- noted by the same symbol. The determinants are: det(An ) = (−1)n (n + 1), det(Dm) = (−1)m 4 (m ≥ 4), det(E6 ) = 3, det(E7) = −2, det(E8) = 1. 26 MATTHIAS SCH UTT AND TETSUJI SHIODA = 0. Note that in each case, the absolute value of the determinant equals the number of simple components of the singular fibre (including the zero component). The multiplicities ni of the other components of the singular fibre (which one can compute with Tate’s algorithm) are uniquely determined by the condition niΘi!2 niΘi , F 2 = F .Θj = Θ0 + mv −1 mv −1 Xi=1 Xi=1 F ≈ Θ0 + The lattices An , Dn , En above are called the (negative) root lattices of type An , . . . (cf. 11.13). The positive-definite lattices which are obtained from them by changing the sign of the intersection form are also called the (positive) root lattices, and they are often denoted by the same symbol (see e.g. [12]). It will be clear from the context which (negative or positive) is meant in the statements in this survey. Most of the time (such as in this section), we will refer to the negative root lattices; but specifically in section 11, we define Mordell-Weil lattices in such a way that they are positive-definite. Definite lattices have been classified to some extent. For instance, the root lattice E8 with the two signs of the intersection form gives the unique positive- definite and negative-definite even unimodular lattice of rank 8. Such classifi- cations will play an important role in two areas: the study of rational elliptic surfaces (cf. section 11) and the classification of elliptic fibrations on a given K3 surface (cf. 12.19). 6.6. Canonical divisor. In order to prove that algebraic and numerical equiv- alence coincide on an elliptic surface S , it is useful to know the canonical divisor KS . The following formula goes back to Kodaira for complex elliptic surfaces. A characteristic-free proof was given in [9]. The statement involves the Euler char- acteristic χ(S ) = χ(S, OS ) which we will investigate further in the next section. In this survey, we shall refer to χ = χ(S ) as the arithmetic genus of S . N.B. In the theory of algebraic surfaces, the invariant pa = χ − 1 is usually called the arithmetic genus, but we find it more convenient to adopt our definition in this survey. surface S Theorem 6.8 (Canonical bundle formula). The canonical bund le of an el liptic f−→ C is given by ωS = f ∗(ωC ⊗ L−1) where L is a certain line bund le of degree −χ(S ) on C . In particular, we have KS ≈ (2g (C ) − 2 + χ(S )) F , K 2 S = 0. The main idea of proof is that KS is vertical. This comes from the fact that F .KS = 0 by adjunction. Then one applies Zariski’s Lemma to show that KS is a fibre multiple. Finally one computes the degree with the help of spectral sequences and the Riemann-Roch theorem. ELLIPTIC SURFACES 27 As an application, we compute the self-intersection of any section by the ad- junction formula: Corollary 6.9. For any P ∈ E (K ), we have ¯P 2 = −χ(S ). 6.7. Arithmetic genus and Euler number. We want to review a formula for the Euler number e(S ) of an elliptic surface S . Here one can think of the topological Euler number in case we are working over the complex numbers; in general we work with the alternating sum of the Betti numbers, the dimensions of the ℓ-adic ´etale cohomology groups. For the fibres of an elliptic surface, we thus obtain e(Fv ) =  0, mv ,  mv + 1, By (13), this local Euler number of the fibre agrees exactly with the vanishing order of the discriminant if there is no wild ramification. Otherwise it involves a contribution from the index of wild ramification δv at v . Recall from 4.6 that δv = 0 unless char(k) = 2, 3 and Fv is additive. if Fv is smooth; if Fv is multiplicative; if Fv is additive. Theorem 6.10 ([15, Prop. 5.16]). For an el liptic surface S over C , we have e(S ) = Xv∈C Here the sum is actually finite, running over the singular fibres of S → C , i.e. v ∈ Σ. By assumption, the elliptic surface has a singular fibre, hence e(S ) > 0. As a corollary, we obtain the arithmetic genus χ(S ) through Noether’s formula 12 χ(S ) = K 2 S + e(S ). (e(Fv ) + δv ). χ(S ) = e(S ) > 0. Corollary 6.11. For an el liptic surface S , we have 1 12 The proof of Thm. 6.5 now proceeds as follows. Assuming D ≡ 0, one shows that KS − D is vertical using Riemann-Roch with χ(S ) > 0 and Serre duality. Then Lemma 6.7 implies that KS − D is algebraically equivalent to some fibre multiple. By the canonical bundle formula, this also holds for D . But then the degree has to be zero since O .D = 0 by assumption. Hence D ≈ 0. As we have seen, Thm. 6.5 implies Thm. 6.2. It remains to prove Thm. 6.3. 6.8. Sections vs. horizontal components. In order to prove Thm. 6.3, we shall exhibit the inverse of the map E (K ) → NS(S )/T P 7→ ¯P mod T . 28 MATTHIAS SCH UTT AND TETSUJI SHIODA For this purpose, it will be convenient to view the generic curve E as a curve on S . We start by defining a homomorphism Div(S) → Div(E) as follows: Any divisor D on S decomposes into a horizontal part, consisting of sections and multisections, and a vertical divisor consisting of fibre components: D = D ′ + D ′′ , D ′ horizontal, D ′′ vertical. Then the horizontal part D ′ and E intersect properly, giving a divisor on E of degree D ′ .E . This (K -rational) divisor is called the restriction of D to E : D E := D ′ ∩ E ∈ Div(E ). In terms of linear equivalence ∼ (on E resp. S ), it is easy to see that D E ∼E 0 ⇔ D ∼S D ′′ for some vertical D ′′ . By Abel’s theorem for E over K , the divisor D thus determines a unique point P ∈ E (K ) by the following linear equivalence of degree zero divisors: D E − (D ′ .E ) O ∼E P − O . Writing ψ(D) = P , we obtain a homomorphism ψ : Div(S ) → E (K ). The kernel of ψ can be seen to be ker(ψ) = hD ∈ Div(S ); D ≈ 0i + Z ¯O + hD ∈ Div(S ); D verticali. Hence ψ induces the claimed isomorphism ψ : NS(S )/T ∼= E (K ). Recall that we introduced the terminology “horizontal divisor” for sections on an elliptic surface. The above construction identifies any divisor D (or its horizontal component D ′ ) with the horizontal divisor ¯P . All these considerations will be made more explicit in section 11 in order to endow the Mordell-Weil group (modulo torsion) with a lattice structure. 6.9. Picard variety. A crucial step in the proof of Thm. 6.3 concerns the Picard varieties of the elliptic surface S and the base curve C . Recall that generally on a pro jective variety X , the Picard variety Pic0 (X ) arises as the following quotient: Pic0 (X ) = hD ∈ Div(X ); D ≈ 0i/hD ∈ Div(X ); D ∼ 0i. More specifically, if X is a curve, then D ≈ 0 ⇔ deg(D) = 0, and Pic0 (X ) is the Jacobian of X . f → C , pull-back with f ∗ defines an injection In the case of an elliptic surface S f ∗ : Pic0(C ) ֒→ Pic0 (S ), since the section π : C → S provides a left inverse π ∗ . With the first bit of the above consideration, one can in fact show that f ∗ is surjective: ELLIPTIC SURFACES 29 Theorem 6.12. For an el liptic surface S over C with section, Pic0(S ) ∼= Pic0(C ). 6.10. Betti and Hodge numbers. From Thm. 6.12 (or [15, Cor. 5.2.2]), we deduce that an elliptic surface S and its base curve C have the same first Betti number: b1 (S ) = b1 (C ). Then Poincar´e duality allows us to express the second Betti number of S through the Euler number: b2 (S ) = e(S ) − 2 (1 − b1 (C )). For a complex elliptic surface S , we can describe the Hodge diamond explicitly. We shall give its entries, the dimensions hi,j = dim H j (S, Ωi S ), g = g (C ) = q(S ), χ(S ) = χ, in terms of the genus of the base curve C and the arithmetic genus of S (or equivalently the Euler number). For this we abbreviate pg = pg (S ) = h2,0 (S ) = χ − 1 + g . For the Euler number, we compare two expressions: e(S ) = 12χ and the definition of e(S ) as alternating sum of Betti numbers to derive e(S ) = 12χ = 2χ − 2g + h1,1(S ). Hence the Hodge diamond of a complex elliptic surface takes the shape pg g g 1 10χ + 2g 1 g g pg 6.11. Picard number. As a corollary of Thm. 6.3, we obtain a formula for the Picard number of an elliptic surface with section. Sometimes, this formula is referred to as Shioda-Tate formula. Corollary 6.13. Let S be an el liptic surface with section. Denote the generic fibre by E . Then ρ(S ) = rank T + rank E (K ) = 2 + Xv∈R Here we can also express the rank of the trivial lattice in terms of the Euler number. From Prop. 6.6 and Thm. 6.10, we deduce e(Fv ) = Xv∈Σ e(S ) = Xv∈Σ (mv + δv ) + #{v ∈ Σ; Fv is additive}. (mv − 1) + rank E (K ). 30 MATTHIAS SCH UTT AND TETSUJI SHIODA Hence we obtain rank T = e(S ) − #{v ∈ Σ; Fv is multiplicative} −2 #{v ∈ Σ; Fv is additive} − Xv∈Σ We conclude by recalling that over the complex numbers, Lefschetz’ bound ρ(S ) ≤ h1,1 (S ) holds true, while in general, we only have Igusa’s inequality δv . ρ(S ) ≤ b2(S ). With Cor. 6.13, these inequalities translate into estimates for the number of fibre components that an elliptic surface of given Euler number may admit. 6.12. Finiteness. Combined with Thms 5.1 and 5.2, the results from this section enable us to prove the following function field anlogue of Shafarevich’s finiteness result over Q (Thm. 2.8): Theorem 6.14. Assume that the field k is algebraical ly closed. Let C be a projective curve over k and S a finite set of places on C . Up to isomorphism, there are only finitely many el liptic surfaces S → C with section satisfying the fol lowing conditions: (1) S has good reduction, i.e. smooth fibres outside S ; (2) S → C is separable and not isotrivial: the j-map is not constant and separable; (3) there is no wild ramification. We shall only describe the proof in the semi-stable case. Let s = #S . Since ∆ has support ⊂ S , the conductor has degree deg(N ) ≤ s by semi-stability. Hence Thm. 5.2 yields the bound deg(∆) = e(S ) ≤ 6(s + 2g (C ) − 2). This tells us how to bound the possible configurations of singular fibres – either using the trivial bound from Thm. 6.10 (which is immanent in the above formula) or the improved bounds from the previous section depending on the characteristic. At any rate, there are only finitely many possible configurations and thus also only finitely many possibilities for ∆. (The last fact relies on the specification of the set S containing all bad places.) But by Thm. 5.1, each admissible choice of ∆ belongs to at most one semi-stable elliptic surface over C , if any. For a conceptual partial approach in the spirit of Shafarevich’s original proof that is based on the theory of Mordell-Weil lattices, see 11.18. ELLIPTIC SURFACES 31 1 36 6.13. Example: Mordell-Weil group of the normal form. Let us return to the normal form S for given j-invariant, considered as an elliptic surface over P1 with coordinate t. Since S can be written as a cubic pencil (cf. 3.2), it follows that S is rational. At any rate, the singular fibres give e(S ) = 12 and thus b2 (S ) = 10 by the previous section. This gives an upper bound for the Picard number ρ(S ) which we will see immediately to be attained. We start by investigating the trivial lattice. By Cor. 6.9, we have ¯O2 = −χ(S ) = −1. Hence the sublattice of T generated by ¯O and F is isomorphic to h1i ⊕ h−1i. In consequence, the trivial lattice depends on the characteristic as follows: T = h1i ⊕ h−1i ⊕ (E7 , if char(k) 6= 2, 3; E8 , if char(k) = 2, 3. If char(k) = 2, 3, then T thus has rank ten and discriminant −1. Since NS(S ) is an integral lattice, these conditions imply NS(S ) = T . In case char(k) 6= 2, 3, then T has only rank nine and discriminant 2. In addition, we find the following section on the integral model (14) of S : 3 + 2√2 P = − (t − 123)3! . (t − 123)2 , 216 This point is induced from one of the simple base point of the corresponding cubic pencil – which has x = −1/36. It is easily checked that the other simple base point induces −P . Hence we claim that E (K ) = hP i. To prove this, we consider the sublattice of NS(S ) generated by ¯P and the trivial lattice. By Thm. 6.5, the study of sublattices of NS(S ) basically amounts to the com- putation of intersection numbers. In the affine chart (14), ¯P and ¯O do not meet since P is given by polynomials. On the other hand, one easily checks that also in the chart at ∞, the section is polynomial. Indeed, on the integral model from (15), the section is given as 3 + 2√2 P = − (1 − 123s)3! . (1 − 123s)2 , 216 Hence ¯P . ¯O = 0. Using Tate’s algorithm, one finds that P meets the non- zero simple component of the III∗ fibre. The resulting intersection matrix of size 10 × 10 has determinant −1. As before, we deduce that NS(S ) = T + Z ¯P . Thm. 6.3 then gives the claim E (K ) = hP i. The above example indicates that the Tate algorithm might actually be needed in order to find the fibre component met by a section. Of course, this will always be a simple component, so for the present fibre type as well as I2 , II, III, II∗ , there is not much choice, but for the remaining fibre types the task becomes non-trivial. 1 36 32 MATTHIAS SCH UTT AND TETSUJI SHIODA 7. Torsion sections Obviously there is a substantial difference between torsion sections and so- called infinite sections. On an elliptic surface, this can be detected lattice theo- retically by the means of the previous chapter. In the following we will exploit torsion sections in detail. They admit a rich theory specifically due to the con- nection with universal elliptic curves, but they will also be interesting to us for other reasons, for instance in the context of singular fibres or automorphisms of surfaces. In the previous example, we have seen an elliptic surface with a non-torsion 7.1. section in characteristics other than 2, 3. We will also refer to non-torsion sections as infinite sections. Here’s how one can see that P is not torsion: Otherwise, we would have mP = O for some integer m. But then m · ¯P ∈ T by Thm. 6.3. In particular, adjoining P to T produces an overlattice of the same rank, i.e. rank nine. Since the intersection matrix has rank ten, we obtain a contradiction. Conversely, the existence of a torsion section is equivalent to the property that T does not embed primitively into NS(S ). That is, the quotient NS(S )/T is not torsion-free. Denote by T ′ the primitive closure of T inside NS(S ): T ′ = (T ⊗ Q) ∩ NS(S ). With this overlattice of T , we easily deduce the following statements: Lemma 7.1. Let P ∈ E (K ) be torsion. Then ¯P ∈ T ′ . Corollary 7.2. In the above notation, E (K )tors ∼= T ′ /T . 7.2. Group structure of singular fibres. The structure of the N´eron model induces a group structure on al l fibres, not only the smooth one. On a singular fibre, the smooth points form an algebraic group scheme over the base curve. Here the subgroup scheme of the identity component is Gm , if Fv is multiplicative; Ga , if Fv is additive. The quotient by this subgroup scheme is a finite abelian group which we denote by G(Fv ) depending on the fibre type. Lemma 7.3. The singular fibres of el liptic surfaces admit the fol lowing group structure: multiplicative Gm × G(Fv ) : G(In ) ∼= Z/nZ 2m ) ∼= (Z/2Z)2 additive Ga × G(Fv ) : G(I∗ 2m+1 ) ∼= Z/4Z, G(I∗ G(II) ∼= G(II∗) ∼= {0} G(III) ∼= G(III∗) ∼= Z/2Z, G(IV) ∼= G(IV∗ ) ∼= Z/3Z. ELLIPTIC SURFACES 33 7.3. Simple components vs. sections. By definition of the N´eron model, the group structures of the generic fibre and the special fibres are compatible. Here we only consider the simple fibre components met by a section. Since the simple components of a fibre exactly form the group G(Fv ), we derive the following property: Lemma 7.4. Consider the map ψ : E (K ) → Qv∈R G(Fv ), taking a section to the respective fibre components that it meets. Then ψ is a group homomorphism. 7.4. Simple components vs. torsion sections. Cor. 7.2 indicates that a tor- sion section P 6= O has to meet some fibre non-trivially. By this, we mean that P intersects a non-zero component (i.e. different from the zero component) of some fibre. Hence Lemma 7.4 specialises as follows: G(Fv ). E (K )tors Corollary 7.5. Restricted to the torsion subgroup of E (K ), the group homomor- phism ϕ is injective: ψ ֒→ Yv∈R The corollary is best-known in characteristic zero. A rigorous characteristic- free proof follows from the theory of Mordell-Weil lattices, see 11.9. Remark 7.6. If the characteristic p ≥ 0 does not divide the order of the torsion section P , then it follows from the theory of the N´eron model [59] that ¯P and ¯O are disjoint. Without the assumption, the statement is not valid as supersingular elliptic curves in characteristic p have no p-torsion. An example for an elliptic surface with peculiar p-torsion was given in [70, App. 2]. 7.5. Narrow Mordell-Weil group. The previous result yields a bound on the size of the torsion subgroup of E (K ). In particular, the order of a torsion section cannot exceed the least common multiple m of the annihilators of the G(Fv ). Generally, this tells us that upon multiplication by m on E , no section meets the trivial lattice anymore. Hence all these multiples lie in the following subgroup of E (K ): Definition 7.7. The narrow Mordell-Weil group E (K )0 consists of all those sections which meet the zero component of every fibre: E (K )0 = ker(ψ) = {P ∈ E (K ); P meets every fibre at Θ0}. In terms of the narrow Mordell-Weil group, our previous observation can be rephrased as follows: mE (K ) ⊆ E (K )0 . 7.6. Sections vs. automorphisms. We will gain a much better understanding of torsion sections by the following interpretation: Given a section P ∈ E (K ), we define an automorphism tP of the underlying surface S by translation by P : This is well-defined on the smooth fibres and can be extended to the singular 34 MATTHIAS SCH UTT AND TETSUJI SHIODA fibres. Note that translation by a section does not define an automorphism of the elliptic surface, since the zero-section is not preserved, but of the underlying surface where we forget the elliptic structure. However, tP respects the elliptic fibration, i.e. f = f ◦ tP . Often, an elliptic fibration with an infinite section allows one to conclude that some pro jective surface has infinite automorphism group. For instance, this holds true in characteristic 6= 2, 3 for the normal form for given j-invariant by 6.13. 7.7. Quotient by torsion section. If P is a torsion section, then tP defines an automorphism of finite order m on the underlying algebraic surface S . On the generic fibre, tP operates fixed point free. This translates to all smooth fibres unless p m, cf. Remark 7.6. In any case, the quotient is an isogenous elliptic curve E ′ over k(C ). If k contains the mth roots of unity, then E ′ is endowed with a k -rational m-torsion point which induces the dual isogeny E ′ → E . The composition of the two isogenies is just multiplication by m if p ∤ m, resp. by p2em′ if m = pem′ with p ∤ m′ . The Kodaira-N´eron model of E ′ is another elliptic surface S ′ over C with section. If p ∤ m, this surface is obtained from the quotient S/htP i by resolving the ordinary double point singularities resulting from the fixed points of tP : S ′ = ^S/htP i. Because of the isogeny E → E ′ , the elliptic surfaces S, S ′ are often also called isogenous. Thanks to the dual isogeny, the surfaces share the same Betti numbers. In fact, their Lefschetz numbers have to coincide as well; in consequence ρ(S ) = ρ(S ′ ) (cf. [27] for a similar result for complex K3 surfaces). 7.8. Relation between singular fibres. The quotient S/htP i is best studied when the characteristic p does not divide m, the order of the section P , since then the operation is separable. One can study the action of tP on the singular fibres componentwise. This is very convenient, since each component is a rational curve. Since the quotient has to give rise to one of Kodaira’s fibre types after the desingularisation, one can classify the possible actions on the singular fibres. 7.8.1. Multiplicative fibres. For simplicity, let us consider only the case where the order m of P is prime. The general case can easily be deduced from this simplification. If P meets the zero component of an In fibre, then each component is fixed by tP . Hence there are exactly n fixed points at the intersections of the com- ponents. Each of them attains an Am−1 singularity in the quotient. Hence the desingularisation results in a fibre of type Imn . If P does not meet the zero section, then m n by Cor. 7.5. In particular, tP rotates the singular fibre (i.e. the cycle of rational curves) by an angle of 2π/m. Thus tP acts fixed-point free and identifies m components. The resulting fibre in the quotient surface S ′ has type In/m . ELLIPTIC SURFACES 35 7.8.2. Additive fibres. If the torsion section P were to meet the zero component of an additive fibre, then this would impose a second fixed point apart from the node (or cusp). It is easily checked that the resolution cannot result in one of Kodaira’s types of singular fibres (under the assumption p ∤ m). Hence P has to meet the singular fibre non-trivially. Lemma 7.8. Let E (K )′ denote the prime to p-torsion in E (K ) where p =char(K ). Then any additive fibre Fv gives an injection E (K )′ ֒→ G(Fv ). The lemma is a direct consequence of Cor. 7.5, since ϕ is a group homomor- phism. In consequence, additive fibres can easily rule out torsion (prime to p), like fibres of type II and II∗ or combinations of other fibre types. The condition e(S ) = e(S ′ ) imposes further restrictions on the singular fibres. These restrictions apply particularly to the possible configurations of multiplicative fibres (cf. sec- tion 8). 7.9. Universal elliptic curves. In characteristic zero, torsion points give rise to universal elliptic curve. Here the universal property refers to some level N > 3 as follows: Y1(N ) is an elliptic curve over some function field K over Q such that • E (K ) ∼= Z/N Z, and • conversely, any elliptic curve with a rational N -torsion point is a member of the family. The assumption N > 3 guarantees that the universal elliptic curve Y1(N ) forms a unique one-dimensional family over the modular curve X1 (N ). By filling in suitable singular fibres through the Kodaira-N´eron model, we can thus interpret Y1(N ) as an elliptic surface with N -torsion in the Mordell-Weil group. The universal property then guarantees that any other elliptic surface with an N - torsion section arises from Y1(N ) through a base change. These universal elliptic curves have important applications to arithmetic and in particular to modular forms. Namely, one can relate all newforms of level N and weight > 2 to (desingularisations of ) the self fibre products of Y1(N ) by [16]. In this arithmetic context, it becomes important that Y1(N ) admits a model over Z[ 1 N ]. 7.10. Tate’s normal form. In order to explicitly parametrise universal elliptic curves, it is convenient to apply two normalisations. Starting from an elliptic curve in generalised Weierstrass form (2) with section P , we first translate the section to (0, 0). Then a6 ≡ 0, and P is two-torsion if and only if a3 ≡ 0. Otherwise a translation of y yields a4 ≡ 0. The resulting equation is Tate’s normal form for a cubic with a rational point P = (0, 0): E : y 2 + a1 x y + a3 y = x3 + a2 x2 . As a reconfirmation, we check −P = (0, −a3 ) 6= P , so P is not 2-torsion. 36 MATTHIAS SCH UTT AND TETSUJI SHIODA 7.10.1. 3-Torsion. Since −2P = (a2 , 0), the point P has order 3 if and only if a2 ≡ 0. But then there is still much freedom in choosing a1 , a3 – too much freedom actually in order to have a universal elliptic curve. Moreover, there are two principal possibilities for the j-invariant: If a1 ≡ 0, then j ≡ 0, otherwise j is not constant. The following two curve can be regarded as minimal in some sense. (Namely, they are extremal, cf. section 9.) IV∗ , I3 , I1 : S : y 2 + xy + ty = x3 IV∗ , IV : S ′ : y 2 + t y = x3 Both surfaces are apparently rational since their function fields equal k(x, y ). In either case, the trivial lattice T has rank ten. Since e(S ) = e(S ′ ) = 12 and ρ = 10, we can deduce from Cor. 6.13 that the Mordell-Weil rank is zero. Since T has discriminant −9, there can only be 3-torsion in either case: E (K ) = E ′ (K ) = {O , P , −P }. 7.10.2. Higher torsion. If a2 6≡ 0, then we can rescale x and y in such a way that a2 = a3 . Writing c = 1 − a1 , b = −a2 = −a3 , we can express the coefficients of nP through b, c. For the first few multiples, we obtain 3P = (c, b − c) 2P = (b, bc) P = (0, 0) −3P = (c, c2) −2P = (b, 0) −P = (0, b) From the coefficients, we obtain affine equations in b, c for the modular curves parametrising elliptic curves with a torsion point. For instance, for 4-torsion, we obtain c = 0. The resulting elliptic surface over the affine b-line has reducible singular fibres I4 at 0 and I∗ 1 at ∞. Similarly, 5-torsion implies b = c. In consequence, there are two fibres of type I5 at 0 and ∞. Subsequently, 6-torsion is parametrised by the conic b = c + c2 while 7- torsion yields a nodal cubic. The corresponding elliptic surface of level 7 is the first non-rational surface – it is a K3 surface. The above elliptic surfaces are well-defined in any characteristic, but the fibres degenerate in characteristics p N . In fact, there the j-map becomes inseparable (cf. [30, §12]). 7.11. 2-Torsion and quadratic twisting. The problems with 2-torsion are two-fold. On the one hand, we experience the same phenomenon with different j-invariants – constant vs. non-constant – as in the 3-torsion case. But there is also a similar structural difference related to quadratic twisting. We work with fields of characteristic other than 2. Hence we can reduce to an extended Weierstrass form (9). Here the 2-torsion points, if any, are given by the ELLIPTIC SURFACES 37 zeroes αi of the right-hand side polynomial in x – with y -coordinate zero, since the lines through O are exactly the vertical lines (and the line at ∞). Now consider the quadratic twist by some d ∈ K as in (6). It is immediate that the resulting elliptic curve has 2-torsion points (d αi , 0). In other words, quadratic twists preserve the 2-torsion. Thus the question for a universal elliptic curve with a 2-torsion section or full 2-torsion in the Mordell-Weil group is ill-posed, since we can always twist at arbitrary places. Note that in general, quadratic twists of elliptic surface introduce I∗ 0 fibres. Hence it makes sense to ask for elliptic surface with Z/2- or (Z/2)2 -torsion sections and the minimal number of I∗ 0 fibres. All these surfaces are rational by inspection of their function fields. For (Z/2)2 , this leads to the so-called Legendre family on the one hand y 2 = x (x − 1) (x − t) with singular fibres I2 at t = 0, 1 and I∗ 2 at ∞. On the other hand, we can fix any elliptic curve E over k . Since k is alge- braically closed, E has full 2-torsion over k . Then consider the quadratic twist by t. This yields an elliptic surface over P1 with coordinate t. It has two sin- gular fibres of type I∗ 0 at the two places of ramification, 0 and ∞, and constant j-invariant j = j (E ). In particular, these elliptic surface form a one-dimensional family parametrised by j . For a single 2-torsion section, we derive two surfaces without I∗ 0 fibres: S : y 2 = x3 + x2 + tx III∗ , I2 , I1 III∗ , III S ′ : y 2 = x3 + t x S : Both admit the section (0, 0). As before, one finds that E (K ) = E ′ (K ) = {O , (0, 0)}. Note that E ′ has j-invariant j = 123 . As for S , we can still twist the generic fibre quadratically at two of the singular places without introducing any I∗ 0 fibres. 7.12. We take this opportunity to note that with respect to non-2-torsion, qua- dratic twisting has exactly the opposite effect: no sections are preserved. This is related to quadratic base change as follows: Let S → C be an elliptic surface and C ′ → C a double cover with set of ramification points S . Then the Mordell-Weil group of the base change S ′ is generated (up to finite index, a 2-power) by the independent Mordell-Weil groups of S and its quadratic twist S over S : 2E (k(C ′ )) ֒→ E (k(C )) ⊕ E (k(C )) ֒→ E (k(C ′ )). This decomposition holds true up to two-torsion because the two given Mordell- Weil groups correspond exactly to the invariant and anti-invariant sections under the deck transformation. 38 MATTHIAS SCH UTT AND TETSUJI SHIODA 7.13. Elliptic modular surfaces. Over C there is direct construction of the elliptic surfaces associated to the universal elliptic curves. This was introduced by the second author in [91]. In brief, let Γ ⊂ SL2(Z) denote a subgroup of finite index. If −1 6∈ Γ, then one can associate a complex elliptic surface S (Γ) to Γ in a canonical way. For the congruence subgroups Γ1(N ), we recover the elliptic surface associated to Y1 (N ). The common theme is a generalisation of the relation with modular forms of weight 3. Namely, there is an intrinsic isomorphism H 2,0(S (Γ)) ∼= S3(Γ) between holomorphic 2-forms on S (Γ) and cusp forms of weight 3 for Γ. However, if Γ is not a congruence subgroup, then there is no general way to translate this isomorphism into arithmetic properties, say using the L-function of S (Γ) over some number field. Instead, there are conjectural Atkin–Swinnerton-Dyer congruence relations, but these have only been proven in very special cases so far. We remark that this construction was generalised by M. Nori in [65] to gen- eralised modular elliptic surfaces. Over C these have always finite Mordell-Weil group, thus yielding most extremal elliptic surfaces (see 9.2). 8. Rational elliptic surfaces This section initiates our study of rational elliptic surfaces. The previous sec- tions allow us to develop a fairly precise picture. Another aspect will be added by the introduction of Mordell-Weil lattices, see section 11. f −→ C be an elliptic surface over k . The fibration 8.1. Luroth’s theorem. Let S induces an inclusion of function fields k(C ) ⊆ k(S ). If S is rational, Luroth’s theorem implies that C ∼= P1 . Thus any rational elliptic surface is fibred over the pro jective line. Alternatively, we could have argued with Thm. 6.12. Before we continue the investigation of rational elliptic surfaces, we derive a general fact for elliptic surfaces over P1 . 8.2. Elliptic surfaces over the pro jective line cont’d. As a consequence of the Tate algorithm, we have seen in 4.10 that every elliptic surface with a section over P1 has a globally minimal generalised Weierstrass form (2). Equivalently, there is n ∈ N such that • deg(ai ) ≤ ni for all i, but • there is some i such that deg(ai ) ≥ (n − 1)i (minimality at ∞) and • for any finite place on P1 with valuation v , there is some i such that v (ai ) < i (minimality at finite places). ELLIPTIC SURFACES 39 Since the indices refer to some weight, the discriminant of a globally minimal elliptic surface S has degree On the other hand, we have at any place deg(∆) = 12n. v (∆) = e(Fv ) + δv by definition of the wild ramification in 4.6. Hence Thm. 6.10 gives deg(∆) = e(S ) = 12χ(S ). Thus we deduce the factor n of the weights of the coefficients ai of the generalised Weierstrass form: n = χ(S ). 8.3. Canonical bundle. For an elliptic surface S with section over P1 , the canonical bundle formula in 6.6 returns KS = (χ(S ) − 2) F in NS(S ). A rational surface has KS anti-ample. Recall that χ(S ) > 0 since we assumed the existence of a singular fibre. Hence any rational elliptic surface has χ(S ) = 1, e(S ) = 12. 8.4. Normal form of a rational elliptic surface. Combining 8.2 and 8.3, we deduce that a rational elliptic surface S with section is given by a generalised Weierstrass form (2) S : y 2 + a1 x y + a3 y = x3 + a2 x2 + a4 x + a6 . with deg(ai ) ≤ i. Conversely, such a generalised Weierstrass form defines a rational elliptic surface if and only if it has a singular fibre. That is, no admissible transformation makes each ai into an ith power. In particular, this holds if the discriminant ∆ is not a twelfth power. The last characterisation is an equivalence if char(k) 6= 2, 3 because then there is no wild ramification. Since there would be only one singular fibre F∞ , we would have deg(∆) = e(F∞ ). Hence F∞ would have at least 11 components. But then the trivial lattice would have rank(T ) ≥ 12 > ρ(S ) = 10, contradiction. In characteristics 2, 3, however, there are (rational) elliptic surfaces with ex- actly one singular fibre. Hence it is necessary to check the minimality at the twelve-fold zero of ∆. 40 MATTHIAS SCH UTT AND TETSUJI SHIODA 8.5. Cubic pencils. Our first examples for (rational) elliptic surfaces were cubic pencils. This characterisation is in fact complete: Proposition 8.1 ([15, Thm. 5.6.1]). Any rational el liptic surface S is isomorphic to the blow-up of P2 in the base points of a cubic pencil. The proof is based on the consequence of the adjuction formula that any smooth rational curve D on S has self-intersection D2 ≥ −2. Blowing down (−1)-curves successively to a minimal model X (in the sense of algebraic surfaces), one can show that S also admits a morphism π to P2 . But then π(KS ) = KP2 ∼ −3H for a hyperplane section H . Since F = −KS , the images of the fibres are plane cubic curves and generally smooth. (“The pencil descends from S to P2 .”) 8.6. Frame. Even lattices have been classified to great extent. Here we investi- gate whether NS(S ) is even for an elliptic surface – or whether it admits mean- ingful even sublattices. Recall the trivial lattice T as a sublattice of NS(S ). This lattice may vary greatly with the elliptic surface, but it always contains the rank two lattice gen- erated by zero section and general fibre: T ⊇ h ¯O, F i = (cid:18)−χ 1 0(cid:19) 1 Note that this sublattice L is even, i.e. D2 ∈ 2Z for all D ∈ L, if and only if χ(S ) is even: 0(cid:19) ∼=  h1i ⊕ h−1i, h ¯O, F i = (cid:18)−χ 1 U = 0 1 1 0! , 1  The even lattice U is often called the hyperbolic plane. Thus NS(S ) may not be even for an elliptic surface. However, NS(S ) always contains an even sublattice of corank two: the frame if χ(S ) is even. if χ(S ) is odd; W (S ) = h ¯O, F i⊥ ⊂ NS(S ). Lemma 8.2. For any el liptic surface S with section, the frame W (S ) is a negative-definite even lattice of rank ρ(S ) − 2. Proof. By the algebraic index theorem, NS(S ) has signature (1, ρ(S ) − 1). But h ¯O, F i is visibly indefinite, so its orthogonal complement has signature (0, ρ(S ) − 2). Letting D ∈ h ¯O, F i⊥ , we have D .F = D .KS = 0. The adjunction formula yields D2 = 2g (D) − 2. (cid:3) ELLIPTIC SURFACES 41 (18) 8.7. Let ϕ denote the orthogonal pro jection in NS(S ) with respect to U ϕ : NS(S ) → W (S ). Then the image of a section P under ϕ is ϕ( ¯P ) = ¯P − ¯O − (χ(S ) + ¯P . ¯O) F . In particular this element has self-intersection ϕ( ¯P )2 = −(2 χ(S ) + 2 ¯P . ¯O). This orthogonal pro jection should be compared to the definition of the Mordell- Weil lattices in 11.3; the latter will be based on the orthogonal pro jection with respect to T (S ) in NS(S ) ⊗ Q. As soon as χ(S ) > 1, the relation (18) allows us to recover (the root lattices of ) the singular fibres from the root lattice of the frame: W (S )root = {x ∈ W (S ); x2 = −2} Lemma 8.3. Assume that χ(S ) > 1. Then Pv∈R Tv = W (S )root . In particular this shows that if χ(S ) > 1, then the trivial lattice can be com- puted as T (S ) = U ⊕ W (S )root. On a rational elliptic surface (i.e. χ(S ) = 1), the frame behaves completely dif- ferently as we will see in the next section. 8.8. N´eron-Severi lattice of a rational elliptic surface. Since a rational elliptic surface S is isomorphic to P2 blown up in nine points, NS(S ) has rank ten and discriminant −1. The latter property is called unimodular. Inside NS(S ), we have the indefinite lattice generated by zero section and general fibre and its orthogonal complement, the frame W (S ) = h ¯O, F i⊥ . By definition, both sublattices are unimodular. Moreover, W (S ) is even and negative definite by Lemma 8.2. It follows from lattice theory that there is only one such lattice up to isometry. This lattice corresponds to the Dynkin diagram E8 . Hence there is an isomorphism W (S ) = h ¯O, F i⊥ ∼= E8 . We point out the special case where ¯O and F already generate the trivial lattice: Corollary 8.4. Let S be a rational el liptic surface without reducible fibres. Then the orthogonal projection in NS(S ) with respect to ¯O, F induces an isomorphism E (K ) ∼= E8 For a general cubic pencil, the corollary tells us that the nine (disctinct) base points induce eight independent sections after choosing one of them as the zero section. In the general situation where all fibres are irreducible, one can thus endow E (K ) with a lattice structure. Up to torsion, this approach can be pursued in 42 MATTHIAS SCH UTT AND TETSUJI SHIODA full generality, leading to the notion of Mordell-Weil lattices which we introduce in section 11. 9. Extremal rational elliptic surfaces Before continuing our investigation of rational elliptic surfaces, we turn to an important special class among these surfaces in detail – extremal elliptic surfaces. Their relevance lies in several different aspects such as classification questions and arithmetic properties. 9.1. Maximal Picard number. We first introduce a more general notion that governs extremality of elliptic surfaces. Recall the bounds for the Picard number by Lefschetz and Igusa from 6.11 ρ(S ) ≤ h1,1(S ) ρ(S ) ≤ b2(S ). If the respective bound is attained, we refer to the Picard number as maximal. For elliptic surfaces, one can see that both bounds are attained infinitely often by (purely inseprable in characteristic p) base change from rational elliptic surfaces. For rational elliptic surfaces, both bounds coincide and are always attained. resp. 9.2. Extremal elliptic surfaces. Elliptic surfaces with maximal Picard number can be distinguished according to the contribution from the Mordell-Weil group. A special case occurs when the trivial lattice generates the N´eron-Severi lattice up to finite index. Definition 9.1. An elliptic surface S with section is called extremal if ρ(S ) is maximal and E (K ) is finite. One should think of extremal elliptic surfaces as very isolated and rare. Over the complex numbers, this holds with the exception of a few isotrivial families (cf. 9.8), but generally speaking there are quite a few examples. In postive char- acteristic, however, it follows from Thm. 5.2 that essentially all extremal elliptic surfaces arise through inseparable base change. By Thm. 6.3, the index of the trivial lattice T (S ) inside NS(S ) is exactly the size of the Mordell-Weil group of the extremal elliptic surface S : disc(T (S )) = (#E (K ))2 disc(NS(S )). (19) Often extremal elliptic surfaces are related to subgroups of SL2 (Z). We have seen instances of this in Section 7, see in particular 7.13. 9.3. Extremal rational elliptic surfaces. A rational elliptic surface S has unimodular N´eron-Severi group. If S is extremal, (19) thus simplifies as disc(T (S )) = −(#E (K ))2 Since the discriminants of the full trivial lattice T (S ) and the sublattice of the frame W (S ) which is the orthogonal complement of the unimodular sublattice ELLIPTIC SURFACES 43 generated by ¯O , F agree up to sign, we obtain the following information about the singular fibres: (e(Fv ) + δv ) Subtracting yields: (21) (20) disc(Tv ) = (#E (K ))2 . Yv∈R Here we employ the convention that the irreducible fibres have discriminant 1. On the other hand, we can compare the Euler number from Thm. 6.10 12 = e(S ) = Xv against the rank of the trivial lattice from Prop. 6.6: 10 = rank(T (S )) = 2 + Xv (mv − 1).  4 = Xv  In particular, we deduce that there are at most four singular fibres. If there is no wild ramification (e.g. if char(k) 6= 2, 3), there are in fact exactly three cases: (1) 4 multiplicative singular fibres; (2) 2 multiplicative and 1 additive singular fibres; (3) 2 additive singular fibres. For either case, we have seen examples in 7.10.1 – 7.11. In the presence of wild ramification, all combinations of fewer singular fibres (involving at least one additive fibre) occur as well. if Fv is smooth; if Fv is multiplicative; if Fv is additive. 0, 1, 2 + δv , 9.4. Classification. One classifies extremal rational elliptic surfaces in terms of the configurations of singular fibres which we write as tuples [n1 , n2 , . . .] with entries 1, 2, . . . , 0∗ , 1∗ , . . . , II, III, . . . , II∗ representing the fibre types. 9.4.1. Existence. Relation (21) yields a finite list of possible configurations of singular fibres. One can rule out many of these configurations already by (20). Additionally we only need the injection from Lemma 7.8. To prove the existence of the remaining configurations, one has to find explicit equations. This can be accomplished with Weierstrass equations and base change [54]. It seems that the first semi-stable example was exhibited by the Hesse pencil in [91], realising the elliptic modular surface for Γ(3). The first complete list of semi-stable surfaces is due to Beauville using cubic pencils [6]. (For the close connection between extremal elliptic surfaces and the abc-conjecture, see 12.14 and [100].) 44 MATTHIAS SCH UTT AND TETSUJI SHIODA 9.4.2. Mordel l-Weil group. In each case, the configuration determines the Mordell- Weil group. One can use quotients by (translation by) the torsion sections to limit the possible orders. It turns out that there is always only one abelian group of order given by the singular fibres respecting these bounds. (This is not valid in general; for instance, it fails for extremal elliptic K3 surfaces – and so does the uniqueness.) 9.4.3. Uniqueness. To prove uniqueness, one can often argue with universal el- liptic curves. Here we obviously have to exclude the one-dimensional family with configuration [0∗ , 0∗ ]. A helpful fact is also that the quotient by a torsion section is an isogenous surface with a section of the same order. On the other hand, upon deriving explicit Weierstrass equations, uniqueness up to Mobius transformations and admissible transformation of x and y is always more or less immediate. Theorem 9.2 (Miranda–Persson [54]). For any possible configuration except for [0∗ , 0∗ ], there exists a unique extremal rational el liptic surface with pre-determined Mordel l-Weil group. In the sequel, we give the possible configurations without wild ramification and elaborate on a few of the details of the proof of the theorem. The classification in characteristic 2 or 3 involving wild ramification has been established by W. Lang in [46], [47], but unfortunately there are some typos in the equations, so the interested reader is advised to compare with [28] for instance. 9.5. Case (1). In the multiplicative case, the criterion in (20) suffices to rule out all non-existing configurations. The next table lists the existing configurations. We also give the Mordell-Weil group and describe the degeneration of the singular fibres in special characteristics (cf. 7.10.2). [1, 2, 3, 6] [1, 1, 5, 5] [2, 2, 4, 4] [3, 3, 3, 3] M W Z/3 Z/4 Z/6 configuration [1, 1, 1, 9] [1, 1, 2, 8] char(k) degeneration 6= 3 [II, 9] 6= 2 [III, 8] 6= (2 [2, IV, 6] [III, 3, 6] 3 6= 5 [II, 5, 5] −− 6= 2 −− 6= 3 9.6. Case (2). In the mixed case of additive and multiplicative fibres, Lemma 7.8 becomes important. For instance, that lemma directly shows that the exceptional entries in the previous table are impossible outside the given characteristic. Hence we shall not reproduce them in the following list. For shortness, it only includes those configuration that exist in characteristic zero. By reduction, the same holds outside chararacteristics 2, 3. In the remaining characteristics, the existence of wild ramification complicates the classification. Z/5 Z/4 × Z/2 (Z/3)2 ELLIPTIC SURFACES 45 configuration M W [1, 1, II∗ ] {0} Z/2 [1, 2, III∗ ] Z/3 [1, 3, IV∗ ] configuration M W Z/2 [1, 1, 4∗ ] Z/4 [1, 4, 1∗ ] (Z/2)2 [2, 2, 2∗ ] 9.7. Case (3). The additive configurations are treated exactly along the same lines. Outside characteristics 2, 3, we obtain configuration M W [II, II∗ ] {0} Z/2 [III, III∗ ] configuration M W Z/3 [IV, IV∗ ] (Z/2)2 [0∗ , 0∗ ] 9.8. Isotrivial elliptic surfaces. All the rational elliptic surfaces in case (3) have constant j-invariant. Such fibrations are called isotrivial. The name has its origin in the property that after a suitable finite base change, the surface becomes isomorphic to a product of the new base curve and the (constant) general fibre. Among elliptic fibrations, the isotrivial ones play a special role. We have seen in the discussion of universal elliptic curves that isotrivial fibrations can come in families even if it is unexpected. In characteristic zero, extremal elliptic surfaces over P1 are known to have discrete moduli unless e ≤ 60. The proof combines the infinitesimal Torelli theorem for elliptic surfaces with non-constant j-invariant by M.-H. Sait¯o [75] with the classification of all extremal elliptic surfaces with constant j-invariant by Kloosterman [32]. The exceptional Euler numbers are due to families of isotrivial elliptic surfaces with j-invariant 0 or 123 . 10. Semi-stable elliptic surfaces We have classified all extremal rational elliptic surfaces in terms of the possible configurations of singular fibres. Now we would like to know all possible config- urations in general. Letting go the condition extremal, there will clearly be no uniqueness anymore. We will develop part of the theory in greater generality. For simplicity, we shall only consider semi-stable elliptic surfaces (i.e. without additive fibres). In case of additive fibres, the arguments have to be modified slightly. Moreover, we will restrict ourselves to the ground field k = C. This will enable us to use monodromy arguments and deformations. In positive characteristic, there are further issues with inseparable base change and wild ramification that need to be addressed, but the results do not differ essentially. We note that the first investigation of this kind seems to be due to U. Schmickler- Hirezebruch who classified elliptic surfaces over P1 with section and at most three singular fibres [77]. Such a configuration forces some fibre to be additive, so these elliptic surfaces will not appear in this section as we restrict to the semi-stable case. 46 MATTHIAS SCH UTT AND TETSUJI SHIODA 10.1. J-map of a semi-stable elliptic surface. For a semi-stable elliptic sur- face S over C , the j-map j : C → P1 is a morphism of degree e(S ) = 12χ(S ). By definition, j is branched above 0 and 123 with ramification orders divisible by 3 and 2, respectively. For simplicity, we will often apply a normalisation so that the third ramification point is 1 instead of 123 . According to 5.2 this guarantees that in the pull-back of the normal form surface via j , the additive fibres II and III∗ are replaced by 4χ resp. 6χ fibres of type I∗ 0 . Then these can be eliminated by quadratic twisting to recover the semi-stable fibration S → C . Instead of the fibre I1 at ∞, S acquires semi-stable fibres of types [n1 , . . . , ns ] where the ni denote the ramification indices at ∞. If C = P1 , then we can phrase this through the following factorisation of the discriminant: ∆ = (t − αi )ni with pairwise distinct αi . s Yi=1 10.2. Monodromy representation. Let π : P1 → P1 be a map of degree n with ramification points 0, 1, ∞ (possibly among others). It is a general fact that any such π is determined by its monodromy representation (σ0 , σ1 , σ∞ , τ1 , . . . , τr ). Here σi , τj ∈ Sn encode the monodromy around the branch points such that • σ0 ◦ σ1 ◦ σ∞ ◦ τ1 ◦ · · · ◦ τr = id, and • the σi , τj generate a transitive subgroup of Sn . In the case of a semi-stable elliptic surface over P1 with arithmetic genus χ, we can specify the σi as follows: • σ0 is the product of 4χ 3-cycles, • σ1 is the product of 6χ 2-cycles, and • σ∞ is the product of s cycles γi of lengths ni . 10.3. Deformation of j-map. In the sequel, we shall say that a tuple [n1 , . . . , ns ] exists if there is an elliptic surface over P1 with exactly this configuration of sin- gular fibres. Over C, we have the following deformation argument at hand: Lemma 10.1. Let [n1 , . . . , ns ] exist. Let i ≤ s and a, b ≥ 1 such that a + b = ni . Then [n1 , . . . , ni−1 , a, b, ni+1 . . . , ns ] exists. Assume that [n1 , . . . , ns ] exists. The proof of the lemma builds on the fact that ((1, . . . , a) (a + 1, . . . , n)) (a, n) = (1, . . . , n). Hence we can write the ni -cycle γi as the product of two disjoint cycles γ ′ , γ ′′ of lengths a resp. b times a 2-cycle exchanging the “end points” of γ ′ and γ ′′ . This extra 2-cycle accounts for an additional branch point. ELLIPTIC SURFACES 47 It is easily checked that the resulting monodromy representation has product equalling the identity and generates a transitive group. By 10.2 there is a map j : P1 → P1 with the claimed ramification indices at ∞. 10.4. Semi-stable rational elliptic surfaces. Lemma 10.1 is very useful to prove the existence of elliptic surfaces with given configuration of singular fibres. Starting with the six semi-stable extremal rational elliptic surfaces, we obtain through the deformation lemma: Corollary 10.2. For rational el liptic surfaces, the fol lowing 5-tuples exist [1, 1, 1, 1, 8], [1, 1, 2, 3, 5], [1, 1, 1, 2, 7], [1, 1, 2, 4, 4], [1, 1, 1, 3, 6], [1, 2, 2, 3, 4], [1, 1, 2, 2, 6], [2, 2, 2, 2, 4], [1, 1, 1, 4, 5], [1, 2, 3, 3, 3]. Note that the tuple need not determine the torsion subgroup of the Mordell- Weil group. For instance, the first 5-tuple is realised by two different one- dimensional families: one has no torsion sections, while the other family arises through quadratic base change from the extremal rational elliptic surface with configuration [1, 1, 4∗ ]. In particular, it admits a 2-torsion section. Corollary 10.3. Every s-tuple [n1 , . . . , ns ] with s ≥ 6 and P ni = 12 exists on a rational el liptic surface. 10.5. It should be emphasised that the deformation construction is not algebraic in nature. Hence it does not provide any information about fields of definition or characteristic p. For the extremal elliptic surfaces, it is immediately clear that they are defined over some number fields. Compare Belyi’s theorem which is actually the converse of a characterisation of algebraic curves over ¯Q due to Weil. 10.6. Families. It follows from the deformation argument that semi-stable ra- tional elliptic surfaces with s singular fibres occur in (s − 4)-dimensional families. This statement does not generalise directly to elliptic surfaces with additive fi- bres, but it can be modified to work there as well as for arbitrary semi-stable elliptic surfaces. In consequence, rational elliptic surfaces with only I1 fibres have 8 moduli. This expected number agrees with the heuristic based on the Weierstrass from (3): the coefficients a4 , a6 have 5 + 7 = 12 coefficients. We have to subtract four degrees of freedom for applying Mobius transformations and rescaling x, y by an admissible transformation (5). 10.7. The remaining 5-tuples. There are three 5-tuples of natural numbers adding up to 12 (thus possibly realised by rational elliptic surfaces) that are not covered by Cor. 10.2. We claim that they are missing for a good reason: Proposition 10.4. The 5-tuples [1, 2, 2, 2, 5], [1, 1, 3, 3, 4], [2, 2, 2, 3, 3] do not ex- ist. 48 MATTHIAS SCH UTT AND TETSUJI SHIODA e(F ′ v ). The proof of the non-existence goes like this: First we prove that the above configurations do not allow torsion in the Mordell-Weil group. To see this, con- sider the quotient by translation by a hypothetical torsion section. By 7.7, this results in an isogenous rational elliptic surface S ′ . In particular, we obtain in the analogous notation (with R = R′ ) 12 = e(S ′ ) = Xv∈R But in the present situation, one easily checks that any torsion section would meet some fibres in the zero component. On the quotient surface, these fibres of type In result in “bigger” fibres Inp – causing e(S ′ ) to exceed 12. Since the Picard number is ten and the trivial lattice has rank nine, it follows from Thm. 6.3 that and NS(S ) = h ¯P , ¯O, F , Θv,i i. E (K ) = {P } Here we have not used yet that NS(S ) is unimodular. One could proceed with a case-by-case analysis, going though (enough of ) the possible intersection be- haviours of P with O and the singular fibres to conclude (by computing the intersection matrices) that it is an impossible task to set up P in such a way that NS(S ) has discriminant −1. However, this would be quite a lengthy and detailed undertaking. Instead, we turn to the theory of Mordell-Weil lattices which endows the Mordell-Weil group (up to torsion) with a lattice structure. This will enable us to read off the answer directly. In [70] Oguiso and Shioda proved more generally that no rational elliptic surface admits the corresponding trivial lattices. Yet another line of proof would involve the discriminant group, but we will only introduce this technique in the context of K3 surfaces where it becomes indispensable. 11. Mordell-Weil lattice It In this section we finally introduce the notion of Mordell-Weil lattices. goes back independently to Elkies [18] and one of us [93]. In practice, the idea of Mordell-Weil lattices makes the relations between the N´eron-Severi group and the Mordell-Weil group from Section 6 much more explicit. As a ma jor advantage, the extent of this theory is very feasible to actual computations. There are many important consequences to be noted. First we can complete the discussion of rational elliptic surfaces and classify their Mordell-Weil lattices completely (following [70]). We will also point out the functorial behaviour of Mordell-Weil lattices such as for base change and under Galois action. 11.1. Essential lattice. We aim at endowing the Mordell-Weil group (up to torsion) with the structure of a positive-definite lattice. Recall that the sections are in some sense complimentary to the trivial lattice T (S ) by Thm. 6.3. Thus ELLIPTIC SURFACES 49 the relevant part of NS(S ) sits inside the orthogonal complement of T (S ). This is how we define the essential lattice: L(S ) = T (S )⊥ ⊂ NS(S ). Since L(S ) is included in the frame of S by definition, we obtain Lemma 11.1. The essential lattice L(S ) is even and negative definite of rank ρ(S ) − 2 − Pv (mv − 1). By lattice theory, we can compute the discriminant of the essential lattice as disc(L(S )) = (cid:12)(cid:12)(cid:12)(cid:12) disc(T (S )) (cid:12)(cid:12)(cid:12)(cid:12) disc(NS(S )) Hence it is not necessarily true that L(S )⊕T (S ) = NS(S ) (although this will hold true if T (S ) is unimodular, e.g. if there are no reducible fibres). Thus it is unclear a priori whether there is a non-trivial group homomorphism E (K ) → L(S ). [NS(S ) : (L(S ) ⊕ T (S ))]2 . 11.2. To circumvent subtleties of finite index embeddings as sketched above, we shall now tensor all lattices in consideration with Q. Generally we shall abreviate this by a subscript Q: NS(S )Q = NS(S ) ⊗Z Q, T (S )Q , L(S )Q , . . . . Tensoring makes the lattices into Q vector spaces – which does not do any harm since NS(S ) is torsion-free by Thm. 6.2. Note that the subgroup of torsion sec- tions embeds into T (S )Q by Lemma 7.1. 11.3. Define the vector space homomorphism ϕ : NS(S )Q → L(S )Q as the orthogonal pro jection with respect to the sub-vector space T (S )Q . (Com- pare the analogous construction involving U and the frame W (S ) in 8.7.) For an element D ∈ NS(S ), the image ϕ(D) is uniquely determined by the universal properties ϕ(D) ⊥ T (S )Q, ϕ(D) ≡ D mod T (S )Q . Lemma 11.2. The orthogonal projection ϕ gives the unique map E (K ) → L(S )Q with the above properties. The uniqueness follows from the universal property since (upon tensoring E (K ) by Q) it identifies two vectorspaces of the same dimension. 50 MATTHIAS SCH UTT AND TETSUJI SHIODA 11.4. We can describe the map ϕ explicitly. For this purpose, we let Av denote the intersection matrix of the non-zero components of the (reducible) fibre Fv at v ∈ C : Av = (Θv,i , Θv,j )1≤i,j≤mv −1 . Then a section P ∈ E (K ) is mapped to ¯P .Θv,1 v   ... ϕ(P ) = ¯P − ¯O − ( ¯P . ¯O + χ(S )) F − Xv (Θv,1 , . . . , Θv,mv −1 )A−1   ¯P .Θv,mv −1 Obviously ϕ(P ) ≡ ¯P mod T (S )Q . Therefore the precise shape of ϕ(P ) is verified by checking that ϕ sits in the orthogonal complement of T (S )Q . The inverse matrix A−1 is in general not integral, but the coefficients are made v integral by multiplying with disc(Tv ). This shows that in general the definition of ϕ requires tensoring with Q unless there are no reducible fibres. Since the narrow Mordell-Weil group E (K )0 does not meet any non-zero fibre components by definition, we deduce that its image under ϕ is in fact integral: Corollary 11.3. The orthogonal projection ϕ defines an embedding E (K )0 ֒→ L(S ). 11.5. We now consider the full Mordell-Weil group. The following statement can easily be derived from the universal property of the orthogonal pro jection and from Abel’s theorem: Proposition 11.4. The map ϕ : E (K ) → L(S )Q is a group homomorphism with kernel ker(ϕ) = E (K )tor . We have seen in 7.5 that for m = Qv disc(Tv ) mE (K ) ⊆ E (K )0 . Hence the image of E (K ) under ϕ lies in fact in 1 m L(S ), and we obtain an embedding 1 m L(S ). E (K )/E (K )tor ֒→ 11.6. Height pairing. Using the orthogonal pro jection ϕ, we define a pairing on E (K ), the so-called height pairing: P , Q ∈ E (K ) ⇒ < P , Q >= −ϕ(P ).ϕ(Q). Theorem 11.5. The height pairing is a symmetric bilinear pairing on E (K ). It induces the structure of a positive-definite lattice on E (K )/E (K )tor . Symmetry and bilinearity are clear. The rest follows from the fact that L(S ) is negative-definite, since < P , P >= 0 ⇔ ϕ(P ) = 0 ⇔ P ∈ E (K )tor . ELLIPTIC SURFACES 51 11.7. Mordell-Weil lattice. We call the positive-definite lattice E (K )/E (K )tor with height pairing < ·, · > the Mordell-Weil lattice of the elliptic surface S . For ease of notation, we shall also denote the Mordell-Weil lattice by MWL(S ) (where we suppress the elliptic fibration S → C ). Note that the Mordell-Weil lattice will in general not be integral, but take values in Q (or sufficiently 1 m Z for the integer m from 7.5). In contrast, the narrow Mordell-Weil lattice MWL(S )0 is indeed always integral and even. contrv (P , Q) 11.8. Explicit formula. We shall now derive an explicit formula for the height pairing of two sections P , Q ∈ E (K ). Upon specialising to P = Q, we also obtain an expression for the height h(P ) of a section P : < P , Q >= χ(S ) + ¯P . ¯O + ¯Q. ¯O − ¯P . ¯Q − Xv h(P ) = < P , P >= 2χ(S ) + 2 ¯P . ¯O − Xv contrv (P ). Here the correction terms contrv (P , Q) only depend on the fibre components of Fv met. Specifically, let P meet Θv,i and Q meet Θv,j (with the convention that O intersects Θv,0 as usual). Then the local correction term at v is given by contrv (P , Q) = (0, if ij = 0 −(A−1 if ij 6= 0. v )i,j , The correction term for a single section P is defined by contrv (P ) =contrv (P , P ). The inverse matrices A−1 can be calculated in generality. We give their relevant v entries (i.e. those corresponding to the simple components) in the following table. This also verifies that the correction terms are always positive. Throughout we employ the following conventions: • 0 < i ≤ j after exchanging P and Q if necessary; • The components of an In fibre (n > 1) are numbered cyclically according to their intersection behaviour Θ0 , Θ1 , . . . , Θn−1 ; • At the additive fibres, we only number the simple components; • The simple components of an I∗ n fibre for n > 0 are distinguished into the near component Θ1 which intersects the same double component as the zero component, and the far components Θ2 , Θ3 . As claimed, we notice that the correction terms are always positive and become integral upon multiplication with the number of simple components of the fibre. 11.9. First application: proof of Corollary 7.5. With the above explicit formulas, it is easy to prove Corollary 7.5. Let P , Q be torsion sections that meet every (singular) fibre at the same component. In consequence contrv (P ) = contrv (Q) = contrv (P , Q) ∀ v . 52 MATTHIAS SCH UTT AND TETSUJI SHIODA fibre Dynkin diagram i = j i < j IV∗ E6 III∗ E7 In (n > 1), III, IV An−1 I∗ n Dn+4 4/3 3/2 i(n − i)/n (1, 1 + n/4, (1/2, i(n − j )/n − 1/2 + n/4, Table 4. Correction terms for the height pairing 2/3 i = 1 i = 2, 3 i = 1 i = 2 Thus the torsion conditions h(P ) = h(Q) = hP , Qi = 0 imply 2χ + 2 ¯P . ¯O = 2χ + 2 ¯Q. ¯O = χ + ¯P . ¯O + ¯Q. ¯O − ¯P . ¯Q. Whence ¯P . ¯O = ¯Q. ¯O and ¯P . ¯Q = −χ. As χ > 0 by Corollary 6.11 and ¯P , ¯Q are irreducible curves on S , we deduce ¯P = ¯Q, i.e. P = Q. 11.10. Since the Mordell-Weil lattice is defined via the orthogonal pro jection in NS(S )Q with respect to T (S )Q , we can compute the discriminant of NS(S ) as follows: (22) disc(NS(S )) = (−1)rankE (K )disc(T (S )) · disc(MWL(S ))/(#E (K )tor )2 . We can apply this relation to give a simple proof of Prop. 10.4. Recall that for those 5-tuples, the trivial lattice would have rank 9, so E (K ) would have rank 1. Moreover, the configuration of singular fibres ruled out torsion sections. Hence E (K ) would be generated by a single section P . The above equality simplifies as (23) disc(NS(S )) = −disc(T (S )) · h(P ). But then the denominator of the correction term contrv (P ) does not suffice to produce an equality in (22): NS(S ) is unimodular, but there are too many fibres with number of components divisible by 2 or 3. These fibres cause T (S ) to have highly divisible discriminant; hence, by inspection of the corresponding correction terms, the denominator of the height h(P ) cannot possibly be big enough to yield discriminant −1 in (23). 11.11. Conversely, the above argument shows how a fibre configuration can force the existence of a section. We illustrate this by considering the first few 5-tuples from Cor. 10.2. ELLIPTIC SURFACES 53 11.11.1. [1,1,1,1,8]. In this case, there are actually two possibilities: • E (K ) torsion-free with generator P disjoint from O , but meeting Θ3 of the I8 fibre: h(P ) = 2 − 3 · 5/8 = 1/8. • E (K ) with 2-torsion and a section P meeting Θ2 , but not O : h(P ) = 2 − 2 · 6/8 = 1/2. In this case, the 2-isogenous rational elliptic surface has configuration [2,2,2,2,4]. 11.11.2. [1,1,1,2,7]. The configuration rules out torsion sections, so E (K ) is gen- erated by a single section P , disjoint from O and intersecting the singular fibres I2 at Θ1 and I7 at Θ2 : h(P ) = 2 − 1/2 − 10/7 = 1/14. 6 Z, NS(S ) can only be 11.11.3. [1,1,1,3,6]. Since any section P has height P ∈ 1 unimodular if there is a 3-torsion section. 11.11.4. [1,1,2,2,6]. Again any section P has height h(P ) ∈ 1 6 Z. Hence there has to be a 2-torsion section. 4 Z. 11.11.5. [1,1,2,4,4]. With this configuration, any section has height h(P ) ∈ 1 Hence unimodularity requires 4 #E (K )tor by (22). However, there cannot be full 2-torsion in E (K ) since then there would be a 2-torsion section meeting an I4 fibre in the zero component. Hence the quotient by the translation would have singular fibres I2 , I2 , I8 (from I1 , I1 , I4) plus two more, resulting in an Euler number exceeding 12. We deduce Z/4Z ⊂ E (K ). With the given tools, we can classify the Mordell-Weil lattices of all rational elliptic surfaces of Mordell-Weil rank zero or one, since they have at most one non- torsion generator. The full classification was established by the second author jointly with Oguiso in [70] – we will briefly comment on the essential ingredients in the next few subsections. For general elliptic surfaces, we can also employ the above techniques. But then we have also to take the transcendental part of H 2 into account, since NS(S ) will in general not span H 2 . For the K3 case, see the discussion in 12.9, 12.10. 11.12. We can relate the Mordell-Weil lattice MWL(S ) and the essential lattice L(S ) more closely. For this purpose, we change the sign of the intersection form on L(S ) in order to make the lattice positive-definite; the lattice with new pairing is abbreviated by L(S )− . In the following, we shall frequently need the dual lattice M ∨ of a non-degenerate integral lattice M : M ∨ = {x ∈ MQ ; (x.y ) ∈ Z ∀ y ∈ M }. By integrality, one clearly has M ⊆ M ∨ as a sublattice of finite index det(M ), with equality if and only if M is unimodular. Using the setup of Mordell-Weil 54 MATTHIAS SCH UTT AND TETSUJI SHIODA lattices, one can show that the orthogonal pro jection ϕ induces a commutative diagram MWL(S )0 ∼= L(S )− ∩ ∩ ֒→ (L(S )− )∨ MWL(S ) The bottom row embedding need not be an isomorphism; however, it is an iso- morphism if NS(S ) is unimodular. This relation proves very useful in the case of rational elliptic surfaces. 11.13. Root lattice. For an even integral negative-definite lattice M , the roots are all those elements x ∈ M with maximal self-intersection x2 = −2. The lattice M is called a root lattice if it is generated by its roots. For instance, the Dynkin diagrams An , Dm , E6 , E7 , E8 encode root lattices. To each fibre Fv , we can thus associate a root lattice Tv spanned by the com- ponents not meeting the zero section. For an elliptic surface S , we define the root lattice R(S ) of S as the sublattice of NS(S ) generated by the root lattices Tv : R(S ) = ⊕v Tv . By definition, the root lattice R(S ) is exactly the intersection of the trivial lattice T (S ) and the frame W (S ). 11.14. Mordell-Weil lattices of rational elliptic surfaces. Recall that a rational elliptic surface has unimodular N´eron-Severi group NS(S ) ∼= h ¯O, F i ⊕ E8 . Since the root lattice R(S ) sits imside the frame W (S ), we obtain an embedding of R(S ) into E8 . From what we have seen before (in particular from MWL(S ) ∼= (L(S )− )∨ ), one can deduce the following relation between root lattice R(S ) and Mordell-Weil lattice MWL(S ): Theorem 11.6 (Shioda [93, Thm. 10.3]). Up to lattice automorphisms, the em- bedding R(S ) ֒→ E8 determines both E (K ) as an abstract group and MWL(S ) as a lattice. 11.15. The previous theorem creates the possibility to classify the Mordell-Weil lattices of rational elliptic surfaces by purely lattice-theoretic means. We sum up the relevant results. Theorem 11.7 (Dynkin). There are exactly 70 root lattices R 6= 0, E8 embedding into E8 . Here the embedding R ֒→ E8 is unique up to lattice automorphisms unless 1 , A4 3 , A5 ⊕ A1 , A3 ⊕ A2 R = A7 , A2 1 ELLIPTIC SURFACES 55 For each ambiguous root lattice, there are two different families of (generally semi-stable) rational elliptic surfaces associated: one with torsion sections and one without. We have already seen this for R = A7 in 11.11.1. On the other hand, there are three root lattices embedded into E8 as in Thm. 11.7 which are not compatible with the rational elliptic surfaces require- ment that e(R(S )) ≤ e(S ) = 12. These root lattices are A4 1 ⊕D4 , A8 1 , A7 1 . In consequence, there are (70+ 2+ 5−3) = 74 possible embeddings of root lattices R(S ) of rational elliptic surfaces S into E8 up to lattice automorphism. In each case, [70] determines the exact shape of MWL(S ), MWL(S )0 and E (K )tor . 11.16. Corollaries. Without details, we note the following corollaries of the clas- sification of Mordell-Weil lattices of rational elliptic surfaces in [70]. Corollary 11.8. For a rational el liptic surface, E (K ) takes the fol lowing shapes as an abstract group: E (K )tor {0} Z/2Z Z/3Z (Z/2Z)2 Z/4Z Z/5Z Z/6Z Z/4Z × Z/2Z (Z/3Z)2 r = rank E (K ) 0 ≤ r ≤ 8 0 ≤ r ≤ 4 0 ≤ r ≤ 2 0 ≤ r ≤ 2 0 ≤ r ≤ 1 0 0 0 0 Note that for all torsion groups (except for (Z/3Z)2 where one can use the Hesse pencil as in [91]) we have already seen examples in 7.10.1 – 7.11 – often even with the right number of parameters r =rank E (K ). Corollary 11.9. For a rational el liptic surface, E (K ) = hP ; h(P ) ≤ 2i. Even with the inequality h(P ) ≤ 2 replaced by h(P ) ≤ 2χ(S ), the above generation result does not hold true in general (cf. [70, Appendix 1]). Actually, there is a stronger result than Corollary 11.9 that follows from the classification in [70] (cf. [101, Thm 3.4]): Proposition 11.10. For a rational el liptic surface, E (K ) = hP ; ¯P . ¯O = 0i. 56 MATTHIAS SCH UTT AND TETSUJI SHIODA 11.17. Integral sections. The set in the preceeding two corollaries are always finite: For the points with bounded height, this follows from the general property of definite lattices of finite rank. For the second set, we only note that the sections disjoint from O have bounded height h ≤ 2χ(S ). It is this principle that allows us to generalise Siegel’s theorem for elliptic curves over Z. Here we only have to define the right notion of integrality. For an elliptic curve in Weierstrass form, the rational points other than O are given by two rational functions P = (X, Y ), X, Y ∈ k(C ). Here we can always make the Weierstrass form integral, i.e. after some transfor- mations the coefficients ai live in the coordinate ring k [C ]. Then any notion of an integral section should certainly require X, Y ∈ k [C ]. We call such a section polynomial. Note that X, Y 6∈ k [C ] ⇒ ¯P . ¯O > 0, since P and O meet at the poles of X, Y – which in fact have to be common poles since the Weierstrass form is integral. In other words, generally any section P on an integral elliptic curve over k(C ) admits polynomials X0 , Y0 , Z ∈ k [C ] such that P = (X0/Z 2 , Y0/Z 3). However, this only reflects integrality in the chosen affine model. So P could still have poles in the other affine charts – or attain poles by minimalising. Therefore it is convenient to write the coefficients X, Y as homogeneous functions of degree 2χ resp. 3χ over C . On a globally minimal integral model, it thus becomes immediate that an affine polynomial section has a pole at ∞ if and only if deg(X ) > 2χ, deg(Y ) > 3χ. Thus we call the section P integral if and only if it is polynomial and respects the degree bounds after minimalising. Equivalently, ¯P . ¯O = 0. Since integral sections have bounded height h ≤ 2χ as we have seen, we deduce the analogue of Siegel’s theorem: Theorem 11.11. Any integral el liptic curve admits only finitely many integral sections. Note that we can even allow the sections to have denominators of bounded degree. More precisely, if the bound for the x-coordinate is 2d and for the y - coordinate 3d, then there still is a height inequality h ≤ 2χ + 2d. Hence there are only finitely many such sections. Integral points are sometimes also called everywhere integral to stress inte- grality in all affine charts. They reveal interesting relations to Grobner bases and deformations of singularities. The situation on rational elliptic surfaces was recently studied in [101]. ELLIPTIC SURFACES 57 For polynomial sections, finiteness does only hold over fields of characteristic zero (cf. [106, §III.12]). In contrast, positive characteristic sees finiteness results for polynomial sections without further assumptions failing. For an example in every characteristic p > 3 (where polynomial sections are in fact called integral), see for instance [83, Prop. 9.1]. 11.18. Finiteness revisited. With the above finiteness results for the Mordell- Weil group, we can revisit Shafarevich’s finiteness problem for elliptic curves. Recall the set-up of Thm. 2.8 and 6.14 that predicted good reduction outside a finite set of places S . Over Q (or rather Z after switching to an integral model), Shafarevich’s line of proof was to consider all possible discriminants ∆ – there are only finitely many. Working with an integral Weierstrass forms y 2 = x3 − 3A x − 2B , A, B ∈ Z S : we have ∆ = 123 (A3 − B 2). Regrouping terms, this reads S ′ : B 2 = A3 − ∆/123 (24) The latter describes an elliptic curve in the affine A, B -plane. By Siegel’s theorem, this curve only has finitely many points with bounded denominator. For elliptic surfaces over a fixed curve C in characteristics 6= 2, 3, we can pursue exactly the same approach (with Z replaced by the coordinate ring k [C ]). Then the Weierstrass form (24) defines an (isotrivial) elliptic surface S ′ over k [C ]. As a little subtlety, this very Weierstrass form may not be minimal, since ∆ may have roots of high multiplicity; however, this issue does not cause any problems with integral points: they attain bounded denominators upon minimalising, but the height stays bounded. In particular, any elliptic surface over C with discriminant ∆ can be considered as an (almost) integral section on S ′ . By 11.17, these sections are finite in number. 11.19. Galois invariance. We conclude this investigation of Mordell-Weil lat- tices by noting two properties that are of relevance in arithmetic settings. These properties refer to the two ways of extending the function field k(C ): by extension of the base field k or base change C ′ → C . We retain the usual notation of an elliptic curve E over the functions field K = k(C ) of a pro jective curve C . Let ¯k denote the algebraic closure of k . Tacitly, we shall denote the base extension of E from k(C ) to ¯k(C ) also by E (distinguished by the respective sets of rational points). Proposition 11.12. For any σ ∈ Gal(¯k/k) and for any points P , Q ∈ E (¯k(C )), the height pairing is Galois-invariant: P σ , Qσ ∈ E (¯k(C )) with < P σ , Qσ >=< P , Q > . 58 MATTHIAS SCH UTT AND TETSUJI SHIODA The proof of the proposition builds on the associated elliptic surface S → C over k . The crucial property is that the orthogonal pro jection ϕ commutes with the Galois action: ϕ(P σ ) = ϕ(P )σ ∀ P ∈ E (¯k(C )). The reason for this commutativity lies in the fact that the trivial lattice T (S ) is Galois invariant: the Galois action may permute singular fibres and/or fibre components, but it fixes T (S ) as a whole. In consequence, the proposition follows from the Galois-invariance of intersec- tion numbers on pro jective varieties. Remark 11.13. The assumption in [93, Prop. 8.13] that the base field k be perfect is not necessary. 11.20. Base change. Assume we are given a morphism of pro jective curves C ′ → C of degree d over some field k (not necessarily algebraically closed). By base change, any elliptic curve E over k(C ) can also be interpreted as an elliptic curve over k(C ′ ). In particular, we can consider k(C )-rational points on E as elements in E (k(C ′ )). How does this base change effect the height pairing (specified by the function field over which we regard E )? Proposition 11.14. In the above notation, let P , Q ∈ E (k(C )). Then < P , Q >E (k(C ′ ))= d < P , Q >E (k(C )) . In essence, this property goes back to the corresponding elliptic surfaces S ′ , S with rational map g : S ′ 99K S of degree d. For any D , E ∈ NS(S ), we have g ∗(D).g ∗(E ) = d D .E Hence we only have to prove that pull-back from E (k(C )) resp. S commutes with the orthogonal pro jection ϕ. This can be seen by restricting to the narrow Mordell-Weil group E (K )0 and then applying the universal property of ϕ and bilinearity. The precise details are given in the proof of [93, Prop. 8.12]. 12. Elliptic K3 surfaces Elliptic fibrations are maybe most prominently featured in the theory of K3 surfaces. For example, K3 surfaces are singled out by the property that the same surface may admit more than one elliptic fibration (Lem. 12.18). On the other hand, there are also several conjectures concerning algebraic surfaces which are not known in general, but for elliptic K3 surfaces with section. Throughout this section, we will explore selected aspects of elliptic K3 surfaces, mainly concerning existence and classification questions. We start by reviewing some of the general theory before going into elliptic K3 surfaces (cf. [4] and the references therein). ELLIPTIC SURFACES 59 12.1. K3 surfaces. Definition 12.1. A smooth surface X is called K3 if • the canonical bundle is trivial, • h1(X, OX ) = 0. Mostly we will consider pro jective K3 surfaces. The name supposedly refers to Kahler, Kummer and Kodaira (the reasons for these references will become clear soon). The original definition albeit goes back to A. Weil who considered all smooth surfaces with the same differentiable structure as a quartic in P3 . Over C, all K3 surfaces are diffeomorphic and deformation equivalent. In particular, they are simply connected, as conjectured by Weil and Andreotti. The conjecture was proved by Kodaira in [38], using the theory of analytic elliptic surfaces. (i) Quartics in P3 – either smooth or with isolated rational dou- Example 12.2. ble point singularities after a minimal (crepant) desingularisation; (ii) smooth intersections of degree (2, 3) in P4 or (2, 2, 2) in P5 ; (iii) double sextics: double coverings of P2 branched along a sextic curve (smooth or after resolving isolated rational double point singularities); (iv) Kummer surfaces: starting from an abelian surface A, we consider the quo- tient by the involution −1 and resolve the 16 ordinary double points on A/h−1i (see Example 12.6). The Hodge diamond of a K3 surface is easily computed with Serre duality and Noether’s formula: 1 0 0 1 20 1 0 0 1 12.2. K3 lattice. We deduce that a K3 surface has second Betti number b2 = 22. Cup-product equips H 2(X, Z) with the structure of an integral lattice of rank 22. Often this lattice is called K3 lattice and denoted by Λ. The following properties of Λ are well-known: • Λ is unimodular by Poincar´e-duality; • Λ has signature (3, 19) by the topological index theorem; • Λ is even by Wu’s formula since the first Chern class is even. Hence the classification of even unimodular lattices implies that Λ ∼= U 3 ⊕ E 2 8 . A marking φ of a K3 surface X is given by an explicit isomorphism ∼=−→ Λ. φ : H 2(X, Z) 60 MATTHIAS SCH UTT AND TETSUJI SHIODA 12.3. Torelli theorem. Lattice theory is a crucial tool to study K3 surfaces. Its importance was pointed out in the pioneering work by Pjatetki-Shapiro and Shafarevich [74]. Recall that an isometry of lattices is an isomorphism of Z- modules preserving the intersection form. A Hodge isometry is moreover required to preserve the Hodge decomposition after C-linear extension. φ : H 2(X, Z) Theorem 12.3 (Weak Torelli Theorem). Two complex K3 surfaces X, X ′ are isomorphic if and only if there exists a Hodge isometry ∼=−→ H 2(X ′ , Z). A Hodge isometry is called effective if it preserves the Kahler cone (i.e. if X ia algebraic, then the isometry preserves ample divisors). The strong Torelli Theorem extends the weak Torelli Theorem to the point that the isomorphism ∼=−→ X is uniquely determined by the condition f ∗ = φ if φ is effective. f : X ′ The statements made in this paragraph are valid for all K3 surfaces, also non- algebraic ones (which are nonetheless Kahler by a theorem of Siu [108]). 12.4. Moduli. K3 surfaces admit a beautiful moduli theory. Over C, they can be studied through the period map. Here one fixes a nowhere vanishing holo- morphic 2-form ωX up to scalars and chooses a basis {γ1 , . . . , γ22} of H 2(X, Z). Then the period point pX associated to the K3 surface X is ωX (cid:21) ∈ P21 . pX = (cid:20)Zγ1 ωX , . . . , Zγ22 In fact, the image of the period map lies in a open subset of P20 as Z ωX ∧ ωX = 0, Z ωX ∧ ¯ωX > 0. This 20-dimensional space is called the period domain Ω of K3 surfaces. The nice moduli behaviour of K3 surfaces is illustrated by the following fact: If X varies in a local complete family of K3 surfaces, parametrised by a complex manifold M, then there is a local isomorphism M → Ω. For a global description, the degeneration at the boundary has to be taken into account, cf. [40]. Now we specialise to pro jective K3 surfaces. Let γ ∈ H 2(X, Z) be algebraic. Then Zγ since the integral over ωX vanishes along any irreducible curve in X . In the moduli space, this yields the intersection of Ω with a hyperplane. Thus algebraic K3 surfaces have 19-dimensional moduli, and we achieve a stratification of the moduli space in terms of the Picard number (note that ρ ≤ 20 in characteristic zero). ωX = 0, ELLIPTIC SURFACES 61 12.5. Polarisation. On an algebraic K3 surface, one fixes a polarisation, i.e. an ample divisor H with self-intersection H 2 = 2 d. For instance, the hyperplane section on a quartic in P3 is an ample divisor of self-intersection 4. Similarly, on a double sextic the hyperplane section on P2 pulls back to an ample divisor of self-intersection 2. Then one commonly considers moduli of polarised K3 surfaces. One can extend polarisations to the notion of lattice polarisations. Here we let M denote a non-degenerate even integral lattice of signature (1, r). A K3 surface X is M -polarised if there is a primitive embedding M ֒→ NS(X ). Proposition 12.4 ([57, Cor. 1.9]). Let M be an even non-degenerate integral lat- tice of signature (1, r). If M has a primitive embedding into Λ, then M -polarised complex K3 surfaces admit a local moduli space of dimension 19 − r . 12.6. Positive characteristic. In positive characteristic, the moduli question has a completely different flavour since there can be supersingular K3 surfaces with ρ = 22. Surprisingly, the supersingular K3 surfaces themselves admit a nice moduli theory which was discovered by Artin in [1] and greatly extended by Ogus [71]. Over ¯Fp , the Tate conjecture [111] implies that any K3 surface should have even Picard number. One reason (but certainly not the only one) to consider K3 surfaces admitting an elliptic fibration with section is that the Tate conjecture holds true for them by [2]. Remarkably there are also Torelli theorems for K3 surfaces in positive charac- teristic. This has been proved in the supersingular case by Ogus [72] and in the ordinary case by Nygaard [66]. Proposition 12.4 has an analogue for ordinary K3 surfaces in positive characteristic in the context of canonical liftings, see [17], [67]. 12.7. Elliptic K3 surfaces. An elliptic K3 surface X always has base curve P1 by Thm. 6.12. If X admits a section, then by 4.10 there is a globally minimal generalised Weierstrass form X : y 2 + a1 x y + a3 y = x3 + a2 x2 + a4 x + a6 ai ∈ k [t], deg(ai ) ≤ 2i. To give an impression of the possibilities, we discuss two instructive examples. Example 12.5. One can derive elliptic K3 surfaces X from rational elliptic surfaces S by quadratic base change over P1 . Here one only has to avoid ramification at non-reduced fibres. The quadratic twist S ′ of S at the two ramification points is also K3. Over C, the K3 surfaces X, S ′ are isogenous in a certain sense by [27]. In particular, they have the same Picard number. Note that ρ(X ) ≥ 10 by inspection of the trivial lattice and Prop. 11.14. Example 12.6. Let E , E ′ denote elliptic curves. Consider the Kummer surface X for E × E ′ (outside characteristic two). There are several simple models for X . 62 MATTHIAS SCH UTT AND TETSUJI SHIODA For instance, if E , E ′ are given in Weierstrass form y 2 = f (x), y 2 = g (x), and E ′ : E : then X can be defined as the desingularisation of the double sextic y 2 = f (x) g (x′). X : On the other hand, the pro jections onto the two factors induce elliptic fibrations of X . Pro jection onto E ′ yields the isotrivial elliptic fibration g (t) y 2 = f (x) (25) X : which is readily converted to an extended Weierstrass form. This elliptic fibration has four singular fibres of type I∗ 0 at the zeroes of g and ∞. There is full two- torsion in the Mordell-Weil group given by y = 0 and the zeroes of f (x). The Picard number of the Kummer surface X is determined by E , E ′ as follows: ρ(X ) = 16 + 2 + rank Hom(E , E ′ ). Over C, the Picard number thus only depends on the properties whether E , E ′ are isogenous and have complex multiplication (CM): ρ(X ) =   In each case, the trivial lattice stays the same: T (X ) = U ⊕ D4 4 . Comparing with Cor. 6.13, we deduce that the Mordell-Weil group of the given isotrivial fibrations has rank 0 resp. 1 resp. 2 in the above three cases. In Example 12.11, we will investigate further elliptic fibrations on such Kummer surfaces. if E , E ′ are not isogenous; if E , E ′ are isogenous, but do not have CM; if E , E ′ are isogenous and have CM. 18, 19, 20, (26) 12.8. Quasi-elliptic fibrations. In characteristic 2 and 3, smooth pro jective i.e. there is a morphism to a pro- surfaces may admit quasi-elliptic fibrations: jective curve such that the general fibre is a cuspidal rational curve. In other characteristics, the smoothness assumption for the surface prevents the general fibre from being singular. A detailed account of the theory of quasi-elliptic sur- faces can be found in the book by Cossec and Dolgachev [15]. Although many concepts from the theory of elliptic surfaces carry over to quasi-elliptic surfaces, there are a few exceptional properties: • Any quasi-elliptic surface X → C with section admits a purely inseparable cover of degree p by a ruled surface. In particular, if C ∼= P1 , then X is unirational and thus ρ(X ) = b2 (X ). • X admits only certain additive fibre types depending on the characteristic. • There is a different Euler number formula than Thm. 6.10, since the gen- eral fibre has e(F ) = 2. Accordingly, there is a different notion of dis- criminant to detect singular fibres. ELLIPTIC SURFACES 63 • The sections admit the structure of a finite abelian group MW(X ). This group is in fact p-elementary, but its length may exceed two. • The structure theorems from section 6 apply. For instance, they give ρ(X ) = rank(T (X )), MW(X ) = T (X )′/T (X ). Thus any “genus one fibration” with a section of infinite order is necessarily elliptic. To verify this, it suffices to compute the height of the section as in 11.8. Example 12.7. Consider a Kummer surface of product type as in Ex. 12.6. If the cubics f , g are defined over some number field and reduce nicely modulo (a prime above) 2 without multiple zeroes, then (25) defines a quasi-elliptic K3 surface with section over ¯F2 . In contrast to the usual situation, there are five singular fibres of type I∗ 0 , namely at the zeroes of g (t) including ∞ and at the (double) zero of the formal derivative of g (t). 12.9. It will be very convenient to have a characterisation of elliptic fibrations on K3 surfaces in terms of lattice theory. Recall that any fibre F of an elliptic surface has self-intersection F 2 = 0. On a K3 surface X , the converse statement also holds true. For ease of notation, we shall employ the following convention throughout the next two paragraphs: Disclaimer for 12.9, 12.10: In characteristic 2 and 3, the notion of elliptic fibration includes quasi-elliptic fibrations. This disclaimer is necessary since from the abstract information given in the propositions and corollaries, it is in general impossible to detect whether the fibration is elliptic or quasi-elliptic. Proposition 12.8 ([74, §3, Thm. 1]). A K3 surface X admits an el liptic fibration (not necessarily with section) if and only if there is a divisor 0 6= D ∈ NS(X ) with D2 = 0. The main idea to prove the proposition is to apply reflections and inversions to D until it becomes effective. Then the linear system D induces the elliptic fibration. Corollary 12.9. Any K3 surface with Picard number ρ ≥ 5 admits an el liptic fibration. By assumption, NS(X ) is an indefinite lattice of rank at least five. It is well- known that any such quadratic form represents zero. Hence the corollary follows from Prop. 12.8. The general situation on algebraic K3 surfaces is as follows: ρ ≥ 5 ρ = 2, 3, 4 ρ = 1 elliptic fibration depends on X no elliptic fibration 64 MATTHIAS SCH UTT AND TETSUJI SHIODA 12.10. The previous considerations did not address the question of a section. However, one can easily obtain a more precise description from [74, §3, Thm. 1]: Proposition 12.10. Let X be a K3 surface. Assume that D is an effective divisor on X that has the same type as a singular fibre of an el liptic fibration. Then X admits a unique el liptic fibration with D as a singular fibre. Moreover, any irreducible curve C on X with D .C = 1 induces a section of the el liptic fibration. Yet another equivalent way to phrase this abstractly is to ask for an embedding of the hyperbolic plane U into NS(X ): U ֒→ NS(X ). If NS(X ) = U ⊕ M , then M is related to the induced elliptic fibration by [39, Lem. 2.1]: Let F denote any fibre. Then M ∼= F ⊥/hF i ⊂ NS(X ). More specifically, if M decomposes into an orthogonal sum of root lattices associ- ated to Dynkin diagrams, then they correspond to singular fibres of the fibration by a result of Kond¯o [39, Lem. 2.2]. This last statement can also be viewed as a special case of Lem. 8.3. Example 12.11. On the Kummer surface of E × E ′ in Ex. 12.6 (char(k) 6= 2), we have singled out 24 rational curves on the two isotrivial fibrations: the com- ponents of the four I∗ 0 fibres and the torsion sections. This configuration is also known as the double Kummer pencil. In Fig. 7, the double components and two-torsion sections are represented by the vertical and horizontal lines. One can easily find further divisors of Kodaira type. For instance in [103], a II∗ fibre is found together with a section and two orthogonal effective divisors which yield further non-reduced singular fibres (see 13.4). The next figure sketches an elliptic fibration with two IV∗ fibres which are marked in blue and red. For this fibration, all other curves from the double Kummer pencil serve as sections. 12.11. With the tools exploited in the preceeding sections, we shall aim at two different kinds of classifications of elliptic K3 surfaces: • Configurations of singular fibres on all elliptic K3 surfaces with section; • Elliptic fibrations with section on a given K3 surfaces. These classifications will be sketched throughout the next sections. They will require some additional techniques from lattice theory which will be introduced along the way. ELLIPTIC SURFACES 65 Figure 7. Double Kummer pencil 12.12. Configurations on elliptic K3 surfaces: Existence. In Example 12.5, we have already seen two methods to produce elliptic K3 surfaces: by quadratic base change or quadratic twisting from rational elliptic surfaces. How- ever, since these surfaces do always have Picard number ρ ≥ 10, these ideas do not suffice to cover all elliptic K3 surfaces. Abstractly, we could also argue with Prop. 12.10 and try to find lattice polarised K3 surfaces, i.e. with any given shape of the N´eron-Severi group as in Prop. 12.4. In practice, however, this may be a difficult undertaking – often it is achieved using the theory of elliptic fibrations as developped in this section and not vice versa! Thirdly, we have the deformation argument of 10.3 at our hands. Like for rational elliptic surfaces, this argument reduces the existence problem to check- ing which configurations of extremal elliptic K3 surfaces can exist. Afterwards, we will review lattice theoretic means to prove non-existence for the remaining configurations. 12.13. Extremal elliptic K3 surfaces. Recall that an extremal elliptic surface has maximal Picard number, yet finite Mordell-Weil group. Contrary to rational elliptic surfaces, we encounter on the level of K3 surfaces that the notions of extremal elliptic K3 surfaces differ in characteristic zero and p > 0: In charac- teristic zero, the maximal Picard number is ρ = 20. Such K3 surfaces are called singular in the sense of exceptional, as they are lying isolated, but dense in the moduli space of K3 surfaces. Singular K3 surfaces play an improtant role in the arithmetic of K3 surfaces. We will briefly explore their properties in 13.1. Note that we have already seen singular K3 surfaces in Example 12.6: Kummer surfaces for isogenous elliptic curves with CM. However, we have also realised that the obvious elliptic fibrations on these Kummer surfaces are not extremal: they have Mordell-Weil rank two. In contrast, extremal elliptic K3 surfaces in positive characteristic are very rare. Since they are supersingular by definition, they can be traced back to 66 MATTHIAS SCH UTT AND TETSUJI SHIODA rational elliptic surfaces by purely inseparable base change. Confer [28] for a full treatment. 12.14. Dessins d’enfant. In this section, we work over C (or a number field). For an extremal elliptic fibration over P1 , all the ramification of the j-map occurs at three point 0, ∞ and 1 (or 123 depending on the normalisation). Hence we have in fact three equivalent representations: (1) through the j-map, a Belyi map of degree n; (2) through the monodromy representation σ0 , σ1 , σ∞ ∈ Sn ; (3) through the dessin d’enfant j −1 ([0, 1]) associated to the Belyi map j . In essence, the first maps are geometric (or arithmetic) ob jects while the second representations are a priori purely combinatorial. In this respect, the dessins d’enfant could be regarded as intermediate between the other two (see for instance [44] for details). In order to prove the existence of an extremal configuration of elliptic fibres, it suffices to derive one of the three representations with the corresponding prop- erties. In [55], Miranda and Persson exhibited monodromy representations for all semi-stable extremal elliptic K3 surfaces. Recently, Beukers and Montanus derived explicit defining equations for all these surfaces in [7]. A crucial ingre- dient for this explicit work was the fact that the number of inequivalent dessins d’enfant can be determined by representation theoretic means. Here we shall only sketch one example, which is the first case in the list of Miranda and Persson [55]. Prior to [7] its defining equation was determined by one of us in [97]. In fact, extremal elliptic surfaces are closely related to the extreme case of the abc-theorem and in particular to Davenport’s inequality (cf. [100]). Over C, one can thus prove existence and finiteness theorems for semi- stable extremal elliptic surfaces over P1 for any fixed arithmetic genus χ ≥ 1. Example 12.12. An elliptic K3 surface can maximally have singular fibres of type I19 or I∗ 14 (if one does not restrict to semistable case). In characteristic zero, this bound follows from Cor. 6.13. For positive characteristic ( 6= 2), the proof in [79] combines the theory of supersingular K3 surfaces with Mordell-Weil lattices. As for existence, we shall only prove this over some number field by drawing an appropriate dessin d’enfant. Explicitly we only visualise the preimages of 0 under j which appear as vertices with three adjacent edges. Note that the dessin d’enfant is symmetric, as the (unique) elliptic surface can in fact be defined over Q [97]. Thus it also reduces nicely mod p for almost all primes p (cf. [84] where also uniqueness is proven for any characteristic 6= 2). 12.15. Configurations on elliptic K3 surfaces: Non-existence. In order to prove that elliptic K3 surfaces do not admit certain configurations of singular fibres, we will use similar techniques as in section 10; in particular, we will again 67 r ELLIPTIC SURFACES r✟✟✟✟✟✟✟✟✟✟✟✟✟✟ r r r ❍❍❍❍❍❍❍❍❍❍❍❍❍❍ r r r Figure 8. Dessin d’enfant for the I19 fibration build the quotient by torsion sections. However, there is one substantial differ- ence: NS(X ) need not be unimodular (and unless ρ(X ) = 2, so that NS(X ) = U , it will not be unimodular in general). Hence we have to develop different criteria. The configuration of singular fibres is encoded in the trivial lattice T (X ) of an elliptic surface X with section. If e(X ) = 12χ(X ) is a multiple of 24 (e.g. if X is K3), then the zero section has even self-intersection ¯O2 = −χ(X ). Hence zero section and general fibre generate an even unimodular lattice, the hyperbolic plane U . In particular, NS(X ) is an even lattice by Lem. 8.2. Hence we can apply techniques from the theory of even lattices as developped by Nikulin [62]. In the context of elliptic fibrations, these techniques have first been employed systematically by Miranda and Persson [55]. 12.16. Discriminant group and discriminant form. Let L denote an even integral non-degenerate lattice L with quadratic form (·, ·). We define its dual L∨ by L∨ = {x ∈ L ⊗ Q; (x, y ) ∈ Z ∀ y ∈ L}. Then the discriminant group GL of L is the finite abelian group GL = L∨ /L. Example 12.13. For a Dynkin diagram W , let V be the corresponding type of singular fibre from 6.5 (regardless of the ambiguity in case W = A1 or A2 ). Then the discriminant group GW of W equals G(V ) in the notation of Lemma 7.3. We now consider the following set-up: N is an even unimodular lattice (think Λ). We assume that the non-degenerate lattice L embeds primitively into N . Let M denote the orthogonal complement of L in N : M = L⊥ ⊂ N . Proposition 12.14 ([62, Prop. 1.6.1]). Under the above assumptions, GL ∼= GM . 68 MATTHIAS SCH UTT AND TETSUJI SHIODA Most of the time, the above result is applied to the full N´eron-Severi group of a given surface X (with H 2(X, Z) even). Here, however, we are mainly concerned with the configuration of singular fibres. Hence we have to take the subtlety into account that the trivial lattice T (X ) of an elliptic surface need not be primitive in NS(X ). By 7.1, we obtain the primitive closure T (X )′ after saturating T (X ) by adding the torsion sections. The basic idea now is to compare the length of GT (X ) against the maximum possible length of the discriminant group of (T (X )′ )⊥ . The latter equals b2 (X ) − rank(T (X )). We now specialise to the K3 case where Λ is in fact even of rank b2 (X ) = 22: Length criterion: Let X be an elliptic K3 surface with section. If the p-length of GT (X ) exceeds 22 − rank(T (X )), then X admits a p-torsion section. But then we can consider the quotient by (translation by) the p-torsion section. Whenever the Euler numbers of the resulting singular fibres do not sum up to 24, we establish a contradiction. Lemma 12.15. The configurations [1, 27 , 9], [27 , 3, 7], [27 , 52 ] do not exist. Proof. By the length criterion, each of these configurations implies the existence of a 2-torsion section. But then the odd fibres imply that the quotient has Euler number at least 7 · 1 + 2 · 10 = 27, contradiction. (cid:3) For all other n-tuples (n ≥ 9), Miranda and Persson [55] prove existence through the deformation argument from 10.3. On the other hand, all 6-, 7- and 8-tuples deforming to either of those from Lemma 12.15 cannot exist. Prov- ing non-existence for other n-tuples with n = 6, 7, 8 requires some further criteria (see [55]). Here we shall only introduce the discriminant form which will be of later use. The discriminant form qL of an even, integral and non-degenerate lattice L is defined as the induced quadratic form on the discriminant group: qL : GL → Q/2Z x 7→ x2 mod 2Z. Proposition 12.16 ([62, Prop. 1.6.1]). Under the assumptions of Prop. 12.14, qL = −qM . Nikulin moreover showed that the genus of an even lattice (i.e. its isogeny class) is determined by signature and discriminant form [62, Cor. 1.9.4]. There are specific cases where the genus of a lattice consists of a single class. For definite lattices, this typically requires that both rank and (the absolute value of the) discriminant are relatively small (cf. [12, §15, Cor. 22]). For indefinite lattices, the situation differs significantly: ELLIPTIC SURFACES 69 Proposition 12.17 (Kneser [35], Nikulin [62, Cor. 1.13.3]). Let L be an indefinite even lattice. If the length of GL does not exceed rank(L) − 2, then the genus of L consists of a single class. 12.17. There is also a purely lattice-theoretic approach towards configurations of singular fibres. Namely we consider lattice polarised K3 surfaces and apply Prop. 12.4. If the trivial lattice of a given configuration embeds primitively into Λ, then this suffices to prove existence by Kond¯o’s argument from 12.10. The argument becomes more involved as soon as the configuration forces sec- tions by the length criterion. But then Lemma 8.3 shows how to recover the singular fibres from the frame. This approach has been pursued with great suc- cess by Shimada and Zhang: • In [87], Shimada classifies all possible configurations on elliptic K3 sur- faces including the torsion subgroups of MW. For the extra criterion implementing torsion sections, confer [87, Thm. 7.1]. • In [89], Shimada and Zhang classify all extremal elliptic K3 surfaces. This classification extends [55] in two directions: the authors also treat configu- rations that are not semi-stable, and they determine Mordell-Weil groups and transcendental lattices (cf. 13.1). 12.18. Elliptic fibrations. By the Kodaira-Enriques classification of algebraic surfaces (cf. [4]), only K3 surfaces and abelian surfaces have trivial canonical bundle. In consequence, any other surface can admit at most one elliptic fibration with section: Lemma 12.18. Assume that the surface S admits two distinct el liptic fibrations with section which are not of product type. Then S is a K3 surface. Proof. If KS is not trivial, then the canonical bundle formula (Thm. 6.8) implies that any two fibres of the two fibrations are algebraically equivalent. Hence the fibrations are isomorphic. If KS is trivial, then S cannot be abelian, since the fibrations are not of product type; thus S is K3. (cid:3) If we let go the condition of a section, then there are also Enriques surfaces to be considered. Enriques surfaces (outside characteristic two) are quotients of K3 surfaces by fixed point-free involutions. They can be studied purely in terms of their universal covers – which return the original K3 surfaces. In particular an elliptic fibration on an Enriques surface (without section) induces an elliptic fibration on the covering K3 surface (which may well admit a section). For general details, see [15]. Remarkably, several classification problems for elliptic fibrations and beyond have first been solved for Enriques surfaces (see [3] for an instance). 12.19. Elliptic fibrations on a K3 surface. On a single K3 surface, the elliptic fibrations with section can be classified by several means. To our knowledge, Oguiso was the first to establish a complete classification of elliptic fibrations on complex Kummer surfaces for E × E ′ as in Example 12.6 in case E , E ′ are 70 MATTHIAS SCH UTT AND TETSUJI SHIODA not isogenous [69]. Oguiso pursued an essentially geometric technique based on finiteness arguments of Sterk [109] that also allowed him to detect (infinitely many) non-isomorphic fibrations with the same configuration of singular fibres. On the other hand, Sterk showed in [109] that any K3 surface X does only have finitely many elliptic fibrations up to Aut(X ). Another approach towards this problem is based on lattice theoretic ideas. An essential ingredient is a gluing technique which can be traced back to Witt and Kneser [36]. From this technique and the classifcation of definite unimodular lattices of rank 24 in [60], Nishiyama developed an elegant method to classify all elliptic fibrations on a given K3 surface in [63]. This technique moreover builds on a converse of Nikulin’s results in 12.16: Theorem 12.19 ([62, Cor. 1.6]). Let L, M be even non-degenerate integral lat- tices such that GL ∼= GM , qL = −qM . Then there exists a unimodular overlattice N of L ⊕ M such that both L and M embed primitively into N and are each other’s orthogonal complement: L = M ⊥ ⊂ N , M = L⊥ ⊂ N . Of course, we can apply the above theorem to the N´eron-Severi lattice of any K3 surface X . What makes elliptic fibrations special is the fact that the frame W (X ) is even and negative-definite of rank ρ(X ) − 2 by Lem. 8.2. Hence we can try to embed the frame into a negative-definite lattice. This brings the advantage that negative-definite lattices of small rank have been fully classified. Note that in the present case, W (X ) has rank at most 18, so it suffices to consider unimodular negative-definite lattices of rank 24. Luckily these so-called Niemeier lattices admit a fairly simple classification in terms of their root lattices: Theorem 12.20 ([60, Satz 8.3]). A negative-definite even unimodular lattice N is determined by its root lattice Nroot up to isometry. There are 24 possibilities for N . Given an elliptic K3 surface X , Nishiyama aims at embedding the frames of all elliptic fibrations into Niemeier lattices. For this purpose he first determines an even negative-definite lattice M such that qM = −qNS(X ) , rank(M ) + ρ(X ) = 26. By Thm. 12.19, M ⊕ W has a Niemeier lattice as an overlattice for each frame W of an elliptic fibration on X . Thus one is bound to determine the (inequivalent) primitive embeddings of M into Niemeier lattices. To achieve this, it is essential to consider the root lattices involved (as shown in [63]). In each case, the orthogonal complement of M gives the corresponding frame W . Then one can read off Mordell-Weil lattice (plus the torsion in MW) and singular fibres from W by T (X ) = U ⊕ Wroot , MWL(X ) = W/W ′ root ELLIPTIC SURFACES 71 12.20. Transcendental lattice. The existence of a negative-definite lattice M as above was subsequently proven by Nishiyama in [64]. The argument starts from the transcendental lattice of X , i.e. the orthogonal complement of NS(X ) in H 2(X, Z) with respect to cup-product: T(X ) = NS(X )⊥ ⊂ H 2(X, Z). In general, T(X ) is an even lattice of rank r = 22 − ρ(X ) and signature (2, 20 − ρ(X )). Let t = r − 2. By [62, Thm. 1.12.4], T(X )[−1] admits a primitive embedding into the following indefinite unimodular lattice: T(X )[−1] ֒→ U t ⊕ E8 . Then we define M as the orthogonal complement of T(X )[−1] in U t ⊕ E8 . By construction, this lattice is negative definite of rank t + 6 = r + 4 = 26 − ρ(X ) with discriminant form qM = −qT(X )[−1] = qT(X ) = −qNS(X ) . Hence M takes exactly the shape required for Nishiyama’s technique in 12.19. 12.21. It remains to discuss when the above technique guarantees the existence of an elliptic fibration with fixed frame W on the given K3 surface X . First it is instructive to note that in any case there is some K3 surface with the elliptic fibration specified by W . Apply Thm. 12.19 to T(X ) and U ⊕ W to derive a primitive embedding of U ⊕ W into the K3 lattice Λ. By Prop. 12.4, there is a family of K3 surfaces with N´eron-Severi group NS = U ⊕ W . In particular, each of these K3 surfaces admits an elliptic fibration with section and frame W by 12.10. Now let us turn to the specific K3 surface X . Assume that we have found a candidate frame W for an elliptic fibration on X by Nishiyama’s technique. By construction, NS(X ) and U ⊕ W have the same signature and discriminant form. In particular, they lie in the same genus (cf. 12.16). We need to decide whether this genus consists of a single class. In order to apply Prop. 12.17 to the complex K3 case, the decisive property is that qNS(X ) = −qT(X ) by Prop. 12.16. Thus NS(X ) and T(X ) have the same length, and the lattice of bigger rank (unless both have rank 11) satisfies the condition of Prop. 12.17. Lemma 12.21 ([57, Cor. 2.9, 2.10]). Let X be a complex K3 surface. (a) If ρ(X ) ≥ 12, then NS(X ) is uniquely determined by signature and discrimi- nant form. (b) If ρ(X ) ≤ 10, then T(X ) is uniquely determined by signature and discrimi- nant form. Applied to the above situation arising from Nishiyama’s technique, the lemma implies that NS(X ) and U ⊕ W are isomorphic if ρ(X ) ≥ 12. By 12.10, X thus admits an elliptic fibration with frame W . 72 MATTHIAS SCH UTT AND TETSUJI SHIODA 12.22. Nishiyama applied this technique to several K3 surfaces. For instance, he determined all elliptic fibrations on the complex Kummer surfaces from Example 12.6 for the following cases (thus also recovering Oguiso’s result [69]). E , E ′ ρ(X ) no. of fibrations E ∼= E ′ , j (E ) = 0 30 20 E ∼= E ′ , j (E ) = 123 20 25 E ∼= E ′ without CM 19 34 E 6∼ E ′ 18 11 Subsequently, the classification begun by Oguiso in [69] was taken up by Kuwata and Shioda. In [43] they derive defining equations and elliptic parameters for all fibrations in the last case. 12.23. Given a K3 surface X one can also ask the general question whether X admits any elliptic fibration with section at all. Over C there is a uniform answer if the Picard number is big enough: Lemma 12.22. Every complex K3 surface of Picard number at least 13 admits an el liptic fibration with section. The claim follows directly from a result of Nikulin [62, Cor. 1.13.5]. In detail, Nikulin derived the following implication that relates the rank of an indefinite even integral lattice L with its length, i.e. the minimum number of generators of the discriminant group GL = L∨ /L: rank(L) ≥ length(L) + 3 =⇒ U ֒→ L. In the present situation, the length of NS(X ) is trivially bounded by both the rank of NS(X ) and the rank of T(X ). The latter is 22 − ρ(X ) ≤ 9 by assumption. Hence the inequality of rank against length holds for NS(X ), and the embedding of U into NS(X ) gives an elliptic fibration with section. 13. Arithmetic of elliptic K3 surfaces Throughout this paper, we have tried to emphasise the importance of elliptic fibrations for arithmetic considerations. In this section, we point out several applications to K3 surfaces. We start by reviewing singular K3 surfaces. This naturally leads to questions of Mordell-Weil ranks (13.8) and modularity (13.13). We shall also briefly look into connections with Shimura curves, abelian surfaces with real multiplication and rational points. 13.1. Singular K3 surfaces. Over C, the maximal Picard number for a K3 sur- face X is ρ(X ) = 20. The K3 surfaces attaining this bound are called singular. We have seen specific examples in two instances: • extremal elliptic K3 surfaces (12.13, Example 12.12); • Kummer surfaces of isogenous CM-elliptic curves (Example 12.6). ELLIPTIC SURFACES 73 Note that the first examples constitute a finite set up to isomorphism. Hence we shall concentrate on the Kummer surfaces. For a K3 surface X over C, recall the lattice T(X ), i.e. the orthogonal comple- ment of NS(X ) in H 2(X, Z) with respect to cup-product: T(X ) = NS(X )⊥ ⊂ H 2(X, Z). In general, T(X ) is an even lattice of signature (2, 20 − ρ(X )). For a singular K3 surface, the transcendental lattice is positive definite of rank two. Thus a singular K3 surface X is uniquely determined by T(X ) up to isomorphism due to the Torelli Theorem. The intersection form on T(X ) provides a useful link to class group theory – and thereby to elliptic curves with complex multiplication (CM). (27) 13.2. Let us first consider singular abelian surfaces, i.e. complex abelian sur- faces A with maximal Picard number ρ(A) = 4. One defines the transcendental lattice T(A) as above. In [104], Shioda and Mitani showed that every singular abelian surfaces is isomorphic to the product of isogenous CM elliptic curves. The curves can be given explicitly in terms of T(A) as follows. The intersection form on T(A) is expressed uniquely up to conjugation in SL2(Z) by a quadratic form Q(A) = (cid:18)2a 2c(cid:19) , b a, b, c ∈ Z. b We denote its discriminant by d = b2 − 4ac < 0. By [104], A is isomorphic to the product of the following elliptic curves E , E ′ (written as complex tori): E = Eτ , τ = −b + √d b + √d E ′ = Eτ ′ , τ ′ = (28) . , 2a 2 By construction, E and E ′ are isogenous with CM in K = Q(√d). In partic- ular, this result implies that every singular abelian surface can be defined over the Hilbert class field H (d) (a certain abelian Galois extension of K with pre- scribed ramification and Galois group C l(d)), since the same holds for E and E ′ by Shimura’s work [90, §7]. 13.3. The first approach towards singular K3 surfaces was to consider Kum- mer surfaces of singular abelian surfaces as in Example 12.6. In general, the transcendental lattice of a Kummer surface Km(A) is obtained from the abelian surface A by multiplying the intersection form Q(A) on T(A) by a factor of 2 by [74]: T(Km(A)) = T(A)[2]. Hence any singular K3 surface X with T(X ) two-divisible (as an even lattice) can be realised as a Kummer surface for appropriate E , E ′ . However, the tran- scendental lattice of a singular K3 surface need not be two-divisible as we will see for the following extremal elliptic K3 surface. 74 MATTHIAS SCH UTT AND TETSUJI SHIODA Example 13.1. Consider the elliptic K3 surface X with an I19 fibre from Example 12.12. By Thm. 6.3, X has N´eron-Severi group NS(X ) = U ⊕ A18 . As its orthogonal complement in H 2(X, Z), the transcendental lattice T(X ) has discriminant d = −19. As Q(√−19) has class number one, Q(X ) is uniquely represented by the quadratic form Q(X ) = (cid:18)2 1 10(cid:19) . 1 In particular, Q(X ) is not two-divisible. Hence X cannot be isomorphic to a Kummer surface over C. 13.4. Shioda-Inose structure. This failure of the Kummer map to be surjec- tive was overcome by Shioda and Inose by means of elliptic fibrations on Kummer surfaces [103]. They started with a Kummer surface of product type as in Exam- ple 12.6. Explicitly they identified a divisor D of type II∗ in the double Kummer pencil as mentioned in Example 12.11. This divisor is marked blue in the follow- ing diagram of the double Kummer pencil: Figure 9. Singular fibres in the double Kummer pencil By Prop. 12.10, the Kummer surface admits an elliptic fibration with D as a singular fibre and section corresponding to the rational curve marked in yellow. As the red marked curves are perpendicular to D , they constitute further fibre components. Thus there are two more reducible fibres. Shioda-Inose prove that 1 or possibly once IV∗ . their types can only be I∗ 0 , I∗ Consider the quadratic base change X of this elliptic fibration on Km(E × E ′ ) ramified at the last two singular fibres. Since these fibres are non-reduced, the base change has again e(X ) = 24. Thus X is a K3 surface, and it is singular since the surfaces are isogenous. Shioda and Inose proceeded to prove that X has the same transcendental lattice as the original abelian surface A. This concludes the construction of a singular K3 surface with any given transcendental lattice. ELLIPTIC SURFACES 75 Note that the deck transformation corresponding to the quadratic base change defines a Nikulin involution on X : It has exactly eight isolated fixed points and leaves the holomorphic 2-form invariant. The fixed points sit in the two branch fibres; their resolution yields the corresponding fibres on the Kummer surface. This general construction – abelian surface and K3 surface with the same transcendental lattice such that Kummer quotient and Nikulin involution yield the same Kummer surface – is nowadays referred to as Shioda-Inose structure. A ց Km(A) X ւ The terminology was introduced by Morrison who worked out lattice theoretic criteria to decide which K3 surfaces of Picard number ρ ≥ 17 admit a Shioda- Inose structure [57]. Recently Kumar extended Shioda-Inose’s construction for Kummer surfaces of product type to Jacobians of genus two curves [41]. This time, elliptic fibrations with one fibre of type I I ∗ and III∗ each play the central role. 13.5. Inose fibration. The construction of Shioda-Inose can be defined over some explicit number field: (1) The elliptic curves E , E ′ are defined over the Hilbert class field H (d). (2) The Kummer quotient respects the base field of the abelian surface (cf. Ex- ample 12.6). (3) The field of definition of the divisor D of type II∗ only involves the 2-torsion of E and E ′ . Thus D and with it the elliptic fibration D are defined over H (d)(E [2], E ′ [2]). (4) The quadratic base change can be defined over the same field, since the ramification points are either both rational or conjugate quadratic as there cannot be any further singular fibres of the same type by the Euler number e(Km(A)) = 24. The Shioda-Inose structure has some interesting consequences for the arithmetic of a singular K3 surface X . For instance, over some extension of H (d) the zeta function of X can be expressed in terms of Hecke characters [103, Thm. 6]. Subsequently Inose exhibited an explicit model for the elliptic fibration on X induced by the Shioda-Inose structure. By construction, this fibration has two singular fibres of type II∗ . Hence it admits a Weierstrass equation of the following shape (with II∗ ’s at 0, ∞): y 2 = x3 − 3 A t4 x + t5 (t2 − 2 B t + 1). (29) X : This fibration is often called Inose fibration. Inose showed in [27] that A and B are related to the j-invariants of E , E ′ as follows: A3 = j (E ) j (E ′ )/126 , B 2 = (1 − j (E )/123) (1 − j (E ′ )/123). 76 MATTHIAS SCH UTT AND TETSUJI SHIODA This gives a model of X over a degree six extension of H (d). In fact, the Inose fibration can be twisted in such a way that the coefficients are already defined over H (d) (cf. [81, Prop. 4.1]). The remaining singular fibres of Inose’s fibration are easily computed from E and E ′ . We also list the Mordell-Weil ranks and the discriminants in the case where X is singular, i.e. E and E ′ are isogenous with CM. In the non-CM case, the MW-rank drops by one if E ∼ E ′ , resp. by two if E 6∼ E ′ (exactly as in Example 12.6). E , E ′ E ∼= E ′ E 6∼= E ′ d sing fibres j −3 IV j (E ) = 0 j (E ) = 123 −4 2 I2 j (E ) 6= 0, 123 6= −3, −4 I2 , 2 I1 −3 N 2 (N > 1) j (E ) j (E ′) = 0 2 II j (E ) j (E ′) 6= 0 −16, −28 or h(d) > 1 4 I1 Table 5. Mordell-Weil ranks of Inose’s fibration (CM case) rank(MW) 0 0 1 2 2 13.6. Automorphisms. Except for two cases, Inose’s fibration always has pos- itive Mordell-Weil rank. Hence these singular K3 surfaces have infinite auto- morphisms groups. Shioda–Inose made this observation and extended it to all singular K3 surfaces: Lemma 13.2. Any singular K3 surface has infinite automorphism group. To complete the proof of the lemma, Shioda and Inose pursued an explicit geometric approach for the two remaining singular K3 surfaces [103, §5]. Later Vinberg calculated the automorphism groups of these two “most algebraic” K3 surfaces completely [117]. Alternatively, we could use the classification of elliptic fibrations on these K3 surfaces along the lines of 12.19. In [63], Nishiyama showed that there are exactly six elliptic fibrations for d = −3 and 13 for d = −4. In either case, X admits an elliptic fibration with Mordell-Weil rank one, so the automorphism group is infinite. 13.7. We shall now start with the Inose fibration (29) and apply cyclic base changes ramified at the II∗ fibres. For instance, a quadratic base change yields another K3 surface X ′ with two singular fibres of type IV∗ instead of II∗ . Shioda showed in [98] that X ′ returns in fact the Kummer surface Km(E × E ′ ) from the Shioda-Inose structure – we have identified the singular fibres in the double Kummer pencil in Example 12.11. Thus X is sandwiched by this very Kummer surface. In general, all such cyclic base changes of degree n up to six return isogenous K3 surfaces X (n) . By [99], the base change results in a multiplication of the ELLIPTIC SURFACES 77 intersection form on the transcendental lattice by the degree: T(X (n) ) = T(X )[n] (n = 1, . . . , 6). 13.8. Mordell-Weil ranks of elliptic K3 surfaces. By Cor. 6.13, a complex elliptic K3 surface could have any Mordell-Weil rank from 0 to 18, since ρ ≤ 20. Cox proved in [13] that all these ranks actually occur. This result is a consequence of the surjectivity of the period map; however, Cox did not give any explicit examples. To overcome this lack of examples, Kuwata considered the cyclic base changes of Inose’s fibration from 13.7. He noticed that since ρ stays constant under each base change, these surfaces provide explicit examples of complex elliptic K3 surfaces with any Mordell-Weil rank from 0 to 18 except for 15 [42]. Here we produce a table of the fibre configurations and Mordell-Weil ranks (Table 6). degree 1 2 II∗ , 4 I1 2 IV∗ , 8 I1 2 II∗ , I2 , 2 I1 2 I∗ 0 , 3 I2 , 6 I1 2 IV∗ , 2 I2 , 4 I1 E 6∼= E ′ E ∼= E ′ E ∼ E ′ E 6∼ E ′ config MW-rank MW-rank MW-rank config (2 (1 1 0 (6 (4 5 3 (10 (7 2 I∗ 0 , 12 I1 9 6 2 IV, 16 I1 (14 (10 13 9 (18 (13 2 II, 20 I1 17 12 (18 (12 17 11 Table 6. Mordell-Weil ranks of cyclic base changes of Inose’s fibration 2 II, 5 I2 , 10 I1 2 IV, 4 I2 , 8 I1 2 3 4 5 12 16 16 6 6 I2 , 12 I1 0 4 8 24 I1 The configurations are valid under the assumptions that in the first case j (E ) 6= 0, 123 , and in the second case j (E ) j (E ′ ) 6= 0. The Mordell-Weil ranks are computed from the Picard number given in Example 12.6 by Cor. 6.13. In the first two cases, we list both the Mordell-Weil ranks in the CM case and in the non-CM case. By inspection, the ranks cover all numbers from 0 to 18 except for 15. 78 MATTHIAS SCH UTT AND TETSUJI SHIODA 13.9. Kloosterman filled the gap in Kuwata’s list by exhibiting an elliptic K3 surface with Mordell-Weil rank 15 [32]. Here we briefly sketch his construction. Let S be a rational elliptic surface with a fibre of type III∗ and three I1 ’s. These surfaces have Mordell-Weil rank one by Cor. 6.13 and come in a one-dimensional family (write down the general Weierstrass form of a rational elliptic surface with an III∗ fibre). We construct an auxiliary K3 surface X ′ by applying a general quadratic base change to S . Then the trivial lattice T (X ′ ) has rank 16, as it equals U ⊕ E 2 7 . For general X ′ , the N´eron-Severi group is thus generated by T (X ′ ) and the pull- back of the MW-generator of S . Since quadratic base changes over P1 have two moduli, X ′ lives in a three-dimensional family with ρ(X ′ ) ≥ 17. Hence the general member actually has ρ = 17. Finally, we exhibit the cyclic base change of degree 4 to X ′ that ramifies at the two III∗ fibres. In general, this yields another elliptic K3 surface X with only I1 fibres. Since ρ(X ) = ρ(X ′ ) by construction, X has Mordell-Weil rank r ≥ 15. Again we have r = 15 in general. Within this three-dimensional family, it remains to single out a member with ρ = 17. Kloosterman achieved this as follows: he first determined a surface (over Q) where by means of good reduction mod p, the Lefschetz fixed point formula implied ρ ≤ 18 after point counting over Fq for q = p, p2 , p3 ; then he compared the reduction modulo another good prime q to prove ρ = 17 by an extension of a technique pioneered by van Luijk in [116] (related to the Tate conjecture, see 14.6). 13.10. Explicit Mordell-Weil lattices. Sometimes it is even possible to deter- mine the Mordell-Weil lattices abstractly or in terms of generators. For the base changes X (n) of Inose’s pencil, this was achieved in [99]. Kloosterman’s example above initiates a little detour to yet another area: sym- plectic automorphisms of K3 surfaces, i.e. automophisms that leave the holomor- phic two-form invariant. In consequence, the desingularisation of the quotient is again a K3 surface. Recall that we have seen one example in form of Nikulin involutions. Symplectic automorphism groups G on K3 surfaces X have been classified completely starting with Nikulin for the abelian case [61]. One crucial property is that the transcendental lattice T(X ) is G-invariant. Lattice theoretically this has the following effect. Let ΩG be defined as orthogonal complement ΩG = (H 2(X, Z)G )⊥ ⊂ H 2(X, Z). Then ΩG is in fact a primitive negative-definite sublattice of NS(X ) that does not depend on X . To compute ΩG , it thus suffices to work with one particular K3 surface X . In practice, it is often convenient to work with elliptic K3 surfaces since here we have several means of prooducing symplectic automorphisms: notably as quotients by translation by torsion sections and by base change (possibly ELLIPTIC SURFACES 79 composed with the hyperelliptic involution). This approach has been pursued very successfully by Garbagnati and Sarti [26]. Returning to Kloosterman’s example above, it has the dihedral group on four elements, D4 , as group of symplectic automorphisms. In [25] Garbagnati deter- mined ΩD4 explicitly. For the general member of Kloosterman’s family, it follows that MWL ∼= ΩD4 (−1). 13.11. Mordell-Weil rank in characteristic p. The corresponding MW rank problem in positive characteristic can be approached by similar methods. For instance, the Kummer construction from Example 12.6 is well-defined in any odd characteristic p > 2. The ma jor difference is the occurence of supersingular K3 surfaces, but the Picard number is expressed by the same formula as in characteristic zero: ρ(Km(E × E ′ )) = 18 + rank(Hom(E , E ′ )) ∈ {18, 19, 20, 22}. (30) Note that over ¯Fp , the Tate conjecture predicts an even parity of the Picard number (cf. 14.6). Hence ρ = 19 cannot occur over ¯Fp . In fact, in characteristic p > 3 one can still consider Inose’s fibration (29) with the same Picard number as the above Kummer surface. With Tables 5, 6 adjusted according to (30), we obtain elliptic K3 surfaces covering many MW ranks in characteristic p > 3. In particular, X (5) and X (6) yield K3 surfaces with MW rank 20 over ¯Fp as soon as there are at least two distinct supersingular elliptic curves over Fp (since maximal MW rank requires all fibres to be irreducible, so E and E ′ cannot be isomorphic). By the well-known formulas, the latter holds true for p = 11 and p > 13. Example 13.3. Choose E , E ′ with j-invariants 0 resp. 123 . Then y 2 = x3 + t12 + 1. y 2 = x3 + t (t10 + 1) and X (6) : X (5) : These fibrations have MW rank 20 over ¯Fp if and only if both E and E ′ are supersingular mod p, i.e. if p ≡ −1 mod 12. We note that all these supersingular K3 surfaces have Artin invariant σ = 1 (i.e. disc NS(X ) = −p2 ). By a result of Ogus [71], such a K3 surface is unique up to isomorphism. Through X (5) and X (6) , this K3 surface is endowed with two distinct elliptic fibrations of maximal rank 20 for any pair of non-isomorphic supersingular elliptic curves over Fp . In particular, the number of distinct MW rank 20 fibrations grows with the characteristic (cf. [102]). 13.12. Mordell-Weil ranks of elliptic surfaces over Q. So far we have mainly been concerned with N´eron-Severi group and Mordell-Weil group of el- liptic surfaces over algebraically closed ground fields k , thus regarding them as purely geometric ob jects. As soon as we do not assume k to be algebraically 80 MATTHIAS SCH UTT AND TETSUJI SHIODA closed, arithmetic problems enter much more prominently. In fact, these prob- lems can be very interesting in their own right, and often elliptic fibrations provide a genuine tool to study these problems. We shall discuss a few of them throughout the remainder of this section. Consider Kuwata’s and Kloosterman’s explicit examples of elliptic K3 surfaces over C for any given Mordell-Weil rank. In any case, NS and MW are defined over some finite extension of the ground field the details of which we ignored so far. However, the analogous MW-rank question for fixed ground field is much more delicate. Here we outline part of the solution to the problem over Q. The question of Mordell-Weil rank 18 over Q was formulated by Shioda in [96]. Elkies gave a negative answer in [19], [20]. His line of argument combined results specific to singular K3 surfaces with lattice theory. By Cor. 6.13, Mordell-Weil rank 18 implies that an elliptic K3 surface is sin- gular. Moreover the trivial lattice is only U , i.e. all fibres are irreducible. In consequence, the Mordell-Weil lattice is even and integral, but has no roots. This is a very special property, shared for instance by the Leech lattice. Thanks to the Shioda-Inose structure, any singular K3 surface X of discrim- inant d < 0 can be defined over the ring class field H (d) by means of Inose’s fibration (29). Here X may admit a model over some smaller field, but H (d) is always preserved through its Galois action on NS(X ) (cf. [82]). If the Mordell- Weil rank over Q is 18, this implies that d has class number one, so d ≤ 163. Because of the absence of roots, MWL would thus break the density records for sphere packings in R18 . By gluing up to a Niemeier lattice as in 12.19, Elkies is able to establish a contradiction. As for lower rank, Elkies found an elliptic K3 surface with Mordell-Weil rank 17 over Q (and necessarily ρ = 19) [20]. One can show that all intermediate ranks are also attained over Q. 13.13. Modularity. Modularity usually refers to varieties over Q. Here one asks for a relation to modular forms through the zeta function or L-series. For a sin- gular K3 surfaces, Inose’s fibration (29) defines a model over the ring class field H (d). In 13.12, we have reviewed an obstruction to descending X to smaller number fields: the Galois action on the N´eron-Severi group. On singular abelian surfaces, there is a similar, but stronger obstruction to the corresponding problem stemming from the Galois representation of H 1(A) (cf. [23, Lemma 6]). For sin- gular K3 surfaces, lattice theory imposes another substantially weaker condition related to the genus of the transcendental lattice (cf. [88], [81]). Specifically, we can ask for singular K3 surfaces over Q. By [81], this requires that the genus of T(X ) consists of a single class; thus we are concerned with imaginary quadratic fields of class group exponent two. Over some extension, the zeta functions is expressed through particular Hecke characters by [103, Thm. 6]. Over Q, this relation specialises to classical modularity by a result of Livn´e: ELLIPTIC SURFACES 81 Theorem 13.4 (Livn´e [49]). Every singular K3 surface X over Q is modular: the Galois representation on T(X ) is associated to a Hecke eigenform of weight 3. This result can be viewed as one of the first higher-dimensional analogues of the modularity of elliptic curves after Wiles [119]. The converse was recently established by Elkies and Schutt: Theorem 13.5 (Elkies, Schutt [23]). Every known Hecke eigenform of weight 3 with eigenvalues ap ∈ Z is associated to a singular K3 surface over Q. The solution was first reduced to a finite problem by a combination of a finite- ness result of Weinberger for imaginary quadratic fields with class group exponent two [118], and the classification of the above Hecke eigenforms up to twisting in [80]. Then it remained to find singular K3 surfaces over Q for the 65 discrimi- nants from Weinberger’s list. In support of exhibiting these surfaces, elliptic K3 surfaces enter through two key features: First parametrise suitable one- and two-dimensional families of elliptic K3 sur- faces over Q with Picard number ρ ≥ 19 resp. 18. Essentially, this is achieved through explicit calculations with extended Weierstrass forms and their discrim- inant, as it suffices to consider families with NS spanned by the trivial lattice. Then determine a singular member of the family with additional section(s) such that the discriminant equals some given d < 0. By the height formula (22), the discriminant translates into information on the intersection behaviour of the section(s). In particular, one can directly read off the fibre components that have to be met by the section(s). Often these extra information reduce the complexity enough as to facilitate an explicit computation of the singular member of the family and the section(s). 13.14. Shimura curves. There is another reason to be interested in one-dimensional families of (elliptic) K3 surfaces with ρ ≥ 19 as in the proof of Thm. 13.5: the parametrising curve is always a modular curve or a Shimura curve associated to a quaternion algebra over Q. In fact, as the explicit knowledge of Shimura curves is still fairly limited, K3 families enable us to draw interesting consequences for Shimura curves. This approach was pursued by Elkies in [22]; he derived explicit defining equations and CM points of certain Shimura curves and also determined the corresponding QM-abelian surfaces. The reason why K3 surfaces are easier to parametrise lies in the N´eron-Severi group: it has rank 19, so the level of the quaternion algebra (which gives the discriminant of NS) is spread over substantially more divisor classes. In practice, Elkies in [22] pursues a similar approach as in 13.13: he deter- mines a one-dimensional family of elliptic K3 surface with ρ ≥ 19 and given discriminant; then he computes CM points along the lines of 13.13. The main step now consists in finding the corresponding abelian surfaces. This can be achieved through a Shioda-Inose structure. Namely we need an elliptic 82 MATTHIAS SCH UTT AND TETSUJI SHIODA fibration with section and singular fibres of type II∗ and III∗ . Then we can apply Kumar’s construction [41] to find the Igusa-Klebsch invariants of a family of genus 2-curves. The Jacobians of these curves yield the corresponding abelian surfaces. In practice, it might actually not be so easy to parametrise the required elliptic fibrations directly. However, often the K3 surfaces admit other fibrations which are more readily found. By Prop. 12.10, it then suffices to identify perpendicular divisors of type E7 and E8 in NS(X ) and compute their linear systems. In [22], Elkies achieves this in several steps. Here we treat a related example. Example 13.6. Consider a K3 surface X given by Inose’s fibration (29) with two II∗ fibres. In the intersection graph consisting of zero section and E8 ’s, we directly identify a divisor of type D12 by omitting the vertices coresponding to the far double components of the II∗ fibre. The graph is sketched in Figure 10. E8 r r E8 r r r r r r r r r r r r r r ❍❍❍❍ r ✟✟✟✟ ❢r r O Figure 10. An D16 divisor supported on the zero-section and components of the singular fibres By Prop. 12.10, X admits an elliptic fibration with I∗ 12 fibre and sections in- duced by the omitted far components. By itself, this fibration might be not so easy to parametrise. However, the fibration can be deduced directly from (29) by choosing u = x/t2 as an elliptic parameter. One obtains the Weierstrass form y 2 = t2 u3 − 3 A t2 u + t (t2 − 2 B t + 1) X : over k(u) in (t, y )-coordinates with I∗ 12 fibre at u = ∞ and two-torsion section (0, 0). 13.15. Real multiplication abelian surfaces. Similar techniques can be ap- plied to study abelian surfaces A with real multiplication (RM): here an order in a real quadratic field K = Q(√D) embeds into the endomorphism ring of A. A general RM abelian surface A has Picard number ρ(A) = 2, but only in the very first cases the general A is known. Elkies and Kumar again argue with K3 surfaces: as in 13.14 the main idea is to exhibit a two-dimensional family of K3 surfaces with ρ ≥ 18 and the right discriminant. Then compute an appropri- ate elliptic fibration to find the Shioda-Inose structure. Elkies reported on some results for D = 33, 53 in [21]. ELLIPTIC SURFACES 83 13.16. Rational points. Let X be a variety over a number field K (or over the function field of a curve). The rational points are said to be potential ly dense on X if there is some finite extension K ′/K such that X (K ′ ) is dense. Behind this notion stands the expectation that potential density should be a geometric property, depending only on the canonical class KX . Thanks to Faltings’ theorem [24], this concept applies to curves: potential density holds true exactly for curves of genus zero and one. The same is true for surfaces with KX negative: Over a finite extension, they are all rational. On the other hand, the Lang-Bombieri conjecture rules out potential density for pro jective varieties of general type. Among surfaces with KX ≡ 0, potential density has been proved for abelian and Enriques surfaces. In this context, elliptic fibrations enter naturally since positive Mordell-Weil rank over K implies the density of K -rational points. In this spirit, a result by Bogomolov and Tschinkel covers a great range of K3 surfaces [8] – among them all elliptic K3 surfaces: Theorem 13.7 (Bogomolov, Tschinkel). Let X be a K3 surface over a number field. Assume that X has an el liptic fibration or infinite automorphism group. Then the rational points are potential ly dense on X . Note that the elliptic fibration of the theorem is not required to have a section. In consequence, Cor. 12.9 implies that every K3 surface over some number field with ρ ≥ 5 shares the property of potential density. 13.17. Hasse principle. The question of rational points is closely related to the Hasse principle: If a variety X has a Q-rational point, then it has Qv -rational points at every place v . Hence the set of adelic points X (A) is nonempty. The converse implication is known as the Hasse principle: ?=⇒ X (Q) 6= ∅. X (A) 6= ∅ The Hasse principle holds for conics, but is not true in general. In 1970, Manin [51] discovered that the failure of the Hasse principle can often be explained through the Brauer group Br(X ). He defined a subset X (A)Br ⊂ X (A) that contains X (Q), but can be empty even if X (A) is not. This case is referred to as Brauer-Manin obstruction. For K3 surfaces, it is an open problem whether the Brauer-Manin obstruction is the only obstruction to the Hasse principle. In the context of elliptic surfaces (not necessarily with section), Colliot-Th´el`ene, Skorobogatov and Swinnerton- Dyer pioneered an elaborate technique encompassing these problems [11]. There are direct applications to K3 surfaces; sub ject to standard hypotheses, they imply the existence of rational points on certain K3 surfaces (cf. [11, pp. 585, 625/626]). The techniques of [11] apply not only to K3 surfaces, but to several types of surfaces endowed with an elliptic fibration. They have been further devolped by Wittenberg in [120]. 84 MATTHIAS SCH UTT AND TETSUJI SHIODA 14. Ranks of elliptic curves In this section, we shall give a brief discussion of possible Mordell-Weil ranks of elliptic surfaces. In order for this problem to make sense, the base curve C has to be fixed; otherwise one can always apply a base change to a curve of higher genus to split some multisection and thus increase the MW rank. Here we will almost exclusively be concerned with the case C ∼= P1 . 14.1. Noether-Lefschetz loci. Before discussing Mordell-Weil ranks, we have to look into Picard numbers in generality. Let S → P1 denote an elliptic surface with section over C. Upon fixing the arithmetic genus χ(S ) = n, the moduli space of these surfaces has dimension 10 n − 2 by inspection of a globally minimal Weierstrass form as in 4.8: the degrees 4 n + 1 resp. 6 n + 1 of a4 and a6 minus 4 normalisations by Mobius transformations and scaling. Here we are not concerned with rational elliptic surfaces, since then always ρ = 10. Hence we shall assume that χ(S ) > 1. Recall that NS(S ) is generated by horizontal and vertical divisors by Cor. 6.13. Hence an increase of the Picard number within a family of elliptic surfaces can be achieved by two means: • by degenerating the singular fibres in order to produce an additional fibre component; • by forcing an additional, i.e. independent section. From the discriminant ∆, it is visible that the first degeneration causes just one condition on the parameters of the family: the degenerate surfaces ought to live in one-dimensional subspaces of the moduli space. Kloosterman gave a rigorous proof of this fact in [34] based on the model as hypersurface in weighted pro jective space from 4.10. Theorem 14.1. Consider non-isotrivial complex el liptic surfaces over P1 with section and fixed arithmetic genus χ ≥ 2. The surfaces with Picard number ρ ≥ r (r ≥ 2) have (10 χ − r)-dimensional Noether-Lefschetz locus To some extent, isotrivial elliptic surfaces can be dealt with as well (cf. [34, Cor. 1.2]). The K3 case of χ = 2 follows from the period map (12.4). For r = 3, the result was previously proven by Cox [14] as we shall discuss in the next section. 14.2. Section heuristics. While the above fibre degeneration argument seems to work in any characteristic, the problem of additional sections is more subtle; in particular, it depends on the characteristic. First we discuss a heuristic argument: Let P = (X (t), Y (t)) be an integral section on an integral model of the elliptic surface S → P1 in the notion of 11.17. Then X (t) has degree 2χ and deg(Y (t)) = 3χ in general. In total, these polynomials have 5χ + 2 coefficients; on the other hand, the Weierstrass form has ELLIPTIC SURFACES 85 degree 6χ in t. Hence there are (χ − 1) more equations (at ti , i = 0, . . . , 6 χ) than coefficients available. These heuristics suggest that increasing the Mordell-Weil rank by one should cost χ − 1 moduli dimensions. So far these arguments have been made explicit only in the first case by Cox [14]: Theorem 14.2. Consider non-isotrivial complex el liptic surfaces over P1 with section and fixed arithmetic genus χ ≥ 2. The surfaces with Mordel l-Weil rank ≥ 1 have (9 χ − 1)-dimensional moduli. For higher MW rank. Thm. 14.1 gives an upper bound for the dimension of the Noether-Lefschetz locus; however this bound cannot be expected to be optimal, as suggested by Thm. 14.2. 14.3. Mordell-Weil rank over C. The above heuristics show why it is hard to produce elliptic surfaces of high MW rank. At this point, the main hope seemingly consists in finding a canonical (geometric) construction as in 13.8. In particular, it is unclear whether Mordell-Weil ranks of complex elliptic surfaces over P1 can be arbitrary large: Question 14.3. Is there an upper bound for the Mordell-Weil ranks of complex elliptic surfaces over P1 ? One should point out that Lapin once claimed that the MW rank is unbounded [48]. However, some gaps were discovered in his proof (cf. [78]), so the question remains open. 14.4. To our knowledge, the current record for Mordell-Weil ranks of complex elliptic surfaces over P1 is 68. it is attained by some of the isotrivial elliptic surfaces y 2 = x3 + tm + 1 (m ∈ N). (31) S : Theorem 14.4 (Shioda (1991)). The el liptic surface S over C has Mordel l-Weil rank r ≤ 68. Moreover r = 68 ⇐⇒ m ≡ 0 mod 360. The proof relies on the fact that S is a Delsarte surface, i.e. covered by a Fermat surface. Since the Lefschetz number λ(S ) is a birational invariant, it can be computed from the Galois cover. Then b2 (S ) = λ(S ) + ρ(S ) gives the Picard number. The Mordell-Weil rank is easily obtained from Cor. 6.13 since there are no reducible fibres except possibly at ∞. This argument was exhibited in detail by Shioda in [92] where the previous rank record of 56 was achieved by another class of Delsarte elliptic surfaces which is non-isotrivial (i.e. with non-constant j -invariant). The lattice structure of the Mordell-Weil lattice for the elliptic surface (31) for any m has been determined by Usui [115]. For example, in case r = 68, its determinant is equal to 21363102540 . 86 MATTHIAS SCH UTT AND TETSUJI SHIODA 14.5. Mordell-Weil ranks in positive characteristic. The approach through Delsarte surfaces is also very useful in positive characteristic. In fact, it enables us to show that MW-ranks are not bounded. The reason for this lies in unirationality. Namely, by [29] the Fermat surface of degree n is unirational in charateristic p if and only if pν ≡ −1 mod n. ∃ ν ∈ N : Since unirationality implies supersingularity (b2 = ρ), one can thus construct supersingular elliptic surfaces in abundance: by controlling the trivial lattice, one can achieve arbitrarily large MW ranks. For instance, the elliptic surface in (31) is unirational in characteristic p if p ≡ −1 mod 3m (or if this holds for some power of p). The first proof that MW-ranks are unbounded in any positive characteristic goes back to Tate and Shafarevich [114] who used supersingular Fermat curves. Their proof is based on the validity of the Tate conjecture of abelian varieties over finite fields [112] that we will briefly discuss below in 14.6. In particular, the construction of Tate and Shafarevich works over fixed finite fields. This contrasts with the geometric reasoning in [29] over an algebraically closed field. Mordell-Weil lattices in positive characteristic had a great impact on the sphere packing problem. Namely they enabled Elkies and one of us independently to derive lattices with greater density than previously known (cf. [12, pp. xviii-xxii], [18], [94]). 14.6. Tate conjecture. Another feature of positive characteristic is that the Noether-Lefschetz loci behave in a different way. This observation can be easily checked for explicit examples; in general it depends on the validity of the Tate conjecture. The Tate conjecture [111] states that over a finite field Fq , the Picard number of an algebraic surface X can be read off from the induced action of the Frobe- nius morphism Frobq on H 2(X ). Namely, consider the subgroup N of NS(X ) generated by divisor class defined over Fq . Conjecturally, the rank of N equals q on H 2(X ) (for a suitable ´etale co- the multiplicity of q as an eigenvalue of Frob∗ homology with ℓ-adic coefficients). The Tate conjecture is only known for specific classes of surfaces, such as abelian surfaces and products of curves [112], elliptic K3 surfaces with section [2] and Fermat surfaces [29]. One particular consequence of the Tate conjecture concerns the (geometric) Picard number of X (i.e. over ¯Fq ): it equals the number of eigenvalues of Frob∗ q which have the shape q times a root of unity. By the Weil conjectures, all other eigenvalues come in complex conjugate pairs. Hence the Tate conjecture implies that the difference b2 (X ) − ρ( ¯X ) has even parity. In other words, divisors come always in pairs on surfaces over finite fields! The Tate conjecture has the following impact on an elliptic surfaces over ¯Fq : Suppose we degenerate or specialise a given surface over ¯Fq so that there is an ELLIPTIC SURFACES 87 additional fibre component (or section). Then the Tate conjecture forces the occurrence of another independent divisor. Generally this divisor will appear as an independent section. 14.7. Specialisation. When working over number fields, there is one particular reason to be interested in elliptic surfaces with high Mordell-Weil rank: spe- cialisation yields elliptic curves with high rank. Namely, by a result of N´eron and Silverman [105], the Mordell-Weil group of the general member in a one- dimensional family of abelian varieties injects into the Mordell-Weil groups of the specialisations except at finitely many places over any given (finite) number field. Applied to the elliptic K3 surface of MW rank 17 mentioned in 13.12, special- isation yields infinitely many elliptic curves over Q with rank at least 17. This construction was accelerated by Elkies in [19] as follows. First he exhibited a base change from P1 to an elliptic curves E of positive rank such that the MW rank increased to 18. Then he studied the fibres of the base changed elliptic surface over Q – generally elliptic curves of rank at least 18 over Q. Among them, Elkies found one particular fibre that attains the current record of an elliptic curve over Q with rank at least 28. In a way, this new record curve puts an interesting twist to the previous achieve- ments: the first record curves were based on geometric considerations, in partic- ular cubic pencils and rational elliptic surfaces. This approach culminated in the famous work of N´eron (1954) to construct infinitely many elliptic curves over Q of rank at least 11 (cf. Serre [85]). N´eron’s technique was later clarified with the method of Mordell-Weil lattices in [95]. Then, starting from an elliptic curve over Q(t) of rank at least 12 by Mestre and others, extensive computer searches played the dominant role in subsequent years, leading to the previous record curves of rank (at least) 24. It is now with Elkies’ work that geometric arguments are reclaiming their prominent role. Similar techniques have been pursued for elliptic curves over Q with fixed torsion subgroup (cf. [20]). Note that here the Mordell-Weil ranks of the elliptic surfaces are much smaller compared to Picard number and Euler number of the elliptic surface, since the torsion sections force reducible singular fibres by 7.7. Acknowledgements. We thank the organisers of the conference “Algebraic Ge- ometry in East Asia” for inviting us to participate in this fruitful meeting and contribute this paper to the proceedings. The first author is indepted to I. Shi- mada for supporting and arranging his visit. Most of this paper was written while the first author held a position at Uni- versity of Copenhagen, and the first draft was almost finished during a visit of the second author. We are grateful to the Department of Mathematics for their support. We would like to express our gratitude to all our colleagues for discussions and collaborations throughout the years. Particular thanks go to the students 88 MATTHIAS SCH UTT AND TETSUJI SHIODA who attended our lectures on the sub ject of this survey. It was on one of these occasions that we first met in Milano in 2006, and we would like to take this opportunity to thank B. van Geemen for making this initial meeting possible. We are grateful to I. Dolgachev, R. Kloosterman, D. Lorenzini, and H. Partsch for their helpful comments. Our thanks go to the referee for many comments that helped us improve the paper. References [1] Artin, M.: Supersingular K 3 surfaces, Ann. scient. ´Ec. Norm. Sup. (4) 7 (1974), 543–568. [2] Artin, M., Swinnerton-Dyer, P.: The Shafarevich-Tate conjecture for pencils of el liptic curves on K 3 surfaces, Invent. Math. 20 (1973), 249–266. [3] Barth, W., Peters, C.: Automorphisms of Enriques surfaces, Invent. Math. 73 (1983), no. 3, 383–411. [4] Barth, W., Hulek, K., Peters, C., van de Ven, A.: Compact complex surfaces. Second edition, Erg. der Math. und ihrer Grenzgebiete, 3. Folge, Band 4. Springer (2004), Berlin. [5] Beauville, A.: Complex algebraic surfaces. Second edition. LMS Student Texts, 34. Cam- bridge University Press, Cambridge, 1996 [6] Beauville, A.: Les famil les stables de courbes el liptiques sur P1 admettant quatre fibres singuli´eres, C.R. Acad. Sci. Paris 294 (1982), 657–660. [7] Beukers, F., Montanus, H.: Explicit calculation of el liptic fibrations of K3-surfaces and their Belyi-maps, Number theory and polynomials, 33–51, London Math. Soc. Lect. Note Ser. 352, Cambridge Univ. Press, Cambridge, 2008. [8] Bogomolov, F. A., Tschinkel, Y: Density of rational points on el liptic K 3 surfaces, Asian J. Math. 4 (2000), no. 2, 351–368. [9] Bombieri, E., Mumford, D.: Enriques’ classification of surfaces in char. p. II-III, in: Baily, W. L. Jr., Shioda, T. (eds.), Complex analysis and algebraic geometry, Iwanami Shoten, Tokyo (1977) 23–42; Invent. Math. 35 (1976), 197–232. [10] Cassels, J. W. S.: Lectures on El liptic Curves, London Math. Soc. Student texts 24, Cambr. Univ. Press (1991). [11] Colliot-Th´el`ene, J.-L., Skorobogatov, A. N., Swinnerton-Dyer, P.: Hasse principle for pencils of curves of genus one whose Jacobians have rational 2-division points, In- vent. Math. 134 (1998), no. 3, 579–650. [12] Conway, J. H., Sloane, N. J. A.: Sphere packings, lattices and groups, Grundlehren der Mathematischen Wissenschaften 290, Second Edition, Springer-Verlag, 1993. [13] Cox, D. A.: Mordel l Weil groups of el liptic curves over C (t) with pg = 0 or 1, Duke Math. J. 49 (1982), no. 3, 677–689. [14] Cox, D. A.: The Noether-Lefschetz locus of regular el liptic surfaces with section and pg ≥ 2, Amer. J. Math. 112 (1990), no. 2, 289–329. [15] Cossec, F. R., Dolgachev, I. V.: Enriques surfaces. I. Progress in Math. 76. Birkhauser (1989). [16] Deligne, P.: Formes modulaires et repr´esentations ℓ-adiques, in: S´em. Bourbaki 1968/69, no. 355 (Lect. Notes in Math. 179), Springer-Verlag (1971), 139–172. [17] Deligne, P.: Rel`evement des surfaces K3 en caract´eristique nul le, in: Algebraic surfaces (Orsay, 1976–78), Lect. Notes in Math. 868 (1981), 58–79. [18] Elkies, N. D.: Mordel l-Weil lattices in characteristic 2, I. Construction and first proper- ties, Int. Math. Res. Not. 1994, no. 8, 343–361; II. The Leech lattice as a Mordell-Weil lattice, Inv. Math. 128 (1997), 1–8; III. A Mordell-Weil lattice of rank 128, Exp. Math. 10 (2001), no. 3, 467-473. ELLIPTIC SURFACES 89 [19] Elkies, N. D.: The maximal Mordel l-Weil rank of an el liptic K3 surface over Q(t), talk at conference on Birational Automorphisms of Compact Complex Manifolds and Dynamical Systems at Nagoya University, Aug 28, 2007. [20] Elkies, N. D.: Three lectures on el liptic surfaces and curves of high rank, preprint(2007), arXiv: 0709.2908. [21] Elkies, N. D.: Real Multiplication abelian varieties over Q, preprint (2007). [22] Elkies, N. D.: Shimura Curve Computations Via K3 Surfaces of NS-Rank at Least 19, Lect. Notes in Comp. Sci. 5011 (ANTS VIII, 2008), Springer (2008), 196–211. [23] Elkies, N. D., Schutt, M.: Modular forms and K3 surfaces, preprint (2008), arXiv: 0809.0830. [24] Faltings, G.: End lichkeitssatze fur abelsche Varietaten uber Zahlkorpern, Invent. Math. 73 (1983), no. 3, 349–366. [25] Garbagnati, A.: El liptic K3 surfaces with abelian and dihedral groups of symplectic auto- morphisms, preprint (2009), arXiv: 0904.1519. [26] Garbagnati, A., Sarti, A.: El liptic fibrations and symplectic automorphisms on K3 sur- faces, Comm. in Algebra 37 (2009), 1-31. [27] Inose, H.: Defining equations of singular K 3 surfaces and a notion of isogeny, in: Pro- ceedings of the International Symposium on Algebraic Geometry (Kyoto Univ., Kyoto, 1977), Kinokuniya Book Store, Tokyo (1978), 495–502. [28] Ito, H.: On extremal el liptic surfaces in characteristic 2 and 3, Hiroshima Math. J. 32 (2002), pp. 179–188. [29] Katsura, T., Shioda, T.: On Fermat varieties, Tohoku Math. J. (2) 31 (1979), no. 1, 97–115. [30] Katz, N. M., Mazur, Barry: Arithmetic moduli of el liptic curves. Annals of Mathematics Studies 108. Princeton University Press, Princeton, NJ, 1985. [31] Keum, J.-H.: A note on el liptic K3 surfaces, Trans. AMS 352 (1999), no. 5, 2077–2086. [32] Kloosterman, R.: Extremal el liptic surfaces and infinitesimal Torel li, Michigan Math. J. 52 (2004), 141–161. [33] Kloosterman, R.: El liptic K3 surfaces with geometric Mordel l-Weil Canad. Math. Bull. 50 (2007), no. 2, 215–226. [34] Kloosterman, R.: Higher Noether-Lefschetz loci of el liptic surfaces, J. Differential Geom. 76 (2007), 293–316. [35] Kneser, M.: Klassenzahlen indefiniter quadratischer Formen in drei oder mehr Vernder- lichen, Arch. Math. 7 (1956), 323–332. [36] Kneser, M.: Klassenzahlen definiter quadratischer Formen, Arch. Math. 8 (1957), 241– 250. [37] Kodaira, K.: On compact analytic surfaces I-III, Ann. of Math., 71 (1960), 111–152; 77 (1963), 563–626; 78 (1963), 1–40. [38] Kodaira, K.: On the structure of compact complex analytic surfaces I, Amer. J. Math. 86 (1964), 751–798. [39] S. Kond¯o, Automorphisms of algebraic K3 surfaces which act trivial ly on Picard groups, J. Math. Soc. Japan, 44, No. 1 (1992), 75–98. [40] Kulikov, V. S.: Degenerations of K3-surfaces and Enriques surfaces, Math. USSR Izv. 11 (1977) 957–989. [41] Kumar, A.: K3 Surfaces Associated with Curves of Genus Two, Int. Math. Res. Notices Vol. 2008. [42] Kuwata, M.: El liptic K 3 given Mordel l-Weil surfaces with ment. Math. Univ. St. Pauli 49 (2000), no. 1, 91–100. rank, Com- rank 15, 90 MATTHIAS SCH UTT AND TETSUJI SHIODA [43] Kuwata, M., Shioda, T.: El liptic parameters and defining equations for el liptic fibra- tions on a Kummer surface, in: Algebraic geometry in East Asia (Hanoi 2005), 177–215, Adv. Stud. Pure Math. 50, Math. Soc. Japan, Tokyo, 2008. [44] Lando, S. K., Zvonkin, A. K.: Graphs on surfaces and their applications, Encyclopaedia of Mathematical Sciences 141. Low-Dimensional Topology, II. Springer-Verlag, Berlin, 2004. [45] Lang, S.: Fundamentals of Diophantine Geometry. Springer (1983). [46] Lang, W.: Extremal rational el liptic surfaces in characteristic p. I: Beauvil le surfaces, Math. Z. 207 (1991), 429–437. [47] Lang, W.: Extremal rational el liptic surfaces in characteristic p. II: Surfaces with three or fewer singular fibers, Ark. Math. 32 (1994), 423–448. [48] Lapin, A. I.: Subfields of hyperel liptic fields, Izv. Akad. Nauk. SSSR Ser. Mat. 28 (1964), 953–988; Transl. AMS 69 (1968), 204-240. [49] Livn´e, R.: Motivic orthogonal two-dimensional representations of Gal(Q/Q), Israel J. of Math. 92 (1995), 149–156. [50] Liu, Q., Lorenzini, D., Raynaud, M.: N´eron models, Lie algebras, and reduction of curves of genus one, Invent. Math. 157 (2004), no. 3, 455–518. [51] Manin, Y. I.: Le groupe de Brauer-Grothendieck en g´eom´etrie diophantienne, Actes du Congr`es International des Math´ematiciens (Nice, 1970), Tome 1, Gauthier-Villars (1971), 401–411. [52] Mazur, B.: Modular curves and the Eisenstein ideal, Sci. Publ. Math. 47 (1977), 33–186. [53] Miranda, R.: The basic theory of el liptic surfaces, Dottorato di ricerca in matematica, ETS Editrice, Pisa (1989). [54] Miranda, R., Persson, U.: On Extremal Rational El liptic Surfaces, Math. Z. 193 (1986), 537–558. [55] Miranda, R., Persson, U.: Configurations of In Fibers on El liptic K3 surfaces, Math. Z. 201 (1989), 339–361. [56] Miyamoto, R., Top, J.: Reduction of El liptic Curves in Equal Characteristic, Canad. Math. Bull. 48 (2005), 428–444. [57] Morrison, D. R.: On K3 surfaces with large Picard number, Invent. Math. 75 (1984), no. 1, 105–121. [58] Mumford, D.: Lectures on Curves on an Algebraic Surface, Princeton Univ. Press (1966). [59] N´eron, A: Mod`eles minimaux des vari´et´es ab´eliennes sur les corps locaux et globaux, Inst. Hautes ´Etudes Sci. Publ. Math. No. 21 (1964). [60] Niemeier, H.-V.: Definite quadratische Formen der Dimension 24 und Diskriminante 1, J. Number Theory 5 (1973), 142–178. [61] Nikulin, V. V.: Finite groups of automorphisms of Khlerian K3 surfaces, Trans. Moscow Math. Soc. 38 (1980), 71–135. [62] Nikulin, V. V.: Integral symmetric bilinear forms and some of their applications, Math. USSR Izv. 14, No. 1 (1980), 103–167. [63] Nishiyama, K.-I.: The Jacobian fibrations on some K3 surfaces and their Mordel l-Weil groups, Japan. J. Math. 22 (1996), 293–347. [64] Nishiyama, K.-I.: A remark on Jacobian fibrations on K3 surfaces, Saitama Math. J. 15 (1997), 67–71. [65] Nori, M., On certain el liptic surfaces with maximal Picard number, Topology 24 (1985), no. 2, 175–186 Inst. Hautes ´Etudes ELLIPTIC SURFACES 91 [66] Nygaard, N. O.: The Torel li theorem for ordinary K3 surfaces over finite fields, in: Arith- metic and geometry, Vol. I, 267–276, Progr. Math. 35, Birkhauser Boston, Boston, MA, 1983. [67] Nygaard, N. O.: The Tate conjecture for ordinary K3 surfaces over finite fields, In- vent. Math. 74 (1983), no. 2, 213–237. [68] Ogg, A.: El liptic curves and wild ramification, Amer. J. Math. 89 (1967), 1–21. [69] Oguiso, K,: On Jacobian fibrations on the Kummer surfaces of the product of nonisoge- nous el liptic curves, J. Math. Soc. Japan 41 (1989), no. 4, 651–680. [70] Oguiso, K., Shioda, T.: The Mordel l–Weil lattice of a rational el liptic surface, Com- ment. Math. Univ. St. Pauli 40 (1991), 83–99. [71] Ogus, A.: Supersingular K3 crystals, Journ´ees de G´eom´etrie Alg´ebrique de Rennes (Rennes 1978), Vol. II, 3–86, Ast´erisque 64, Soc. Math. France, Paris, 1979. [72] Ogus, A.: A crystal line Torel li theorem for supersingular K3 surfaces, in: Arithmetic and geometry, Vol. II, 361–394, Progr. Math. 36, Birkhuser Boston, Boston, MA, 1983. [73] Pesenti, J., Szpiro, L.: In´egalit´e du discriminant pour les pinceaux el liptiques `a r´eductions quelconques, Compositio Math. 120 (2000), no. 1, 83–117. [74] Piatetski-Shapiro , I. I., Shafarevich, I. R.: Torel li’s theorem for algebraic surfaces of type K3, Izv. Akad. Nauk SSSR Ser. Mat. 35 (1971), 530–572. [75] Sait¯o, M.-H.: On the infinitesimal Torel li problem of el liptic surfaces, J. Math. Kyoto Univ. 23 (1983), 441–460. [76] Saito, T.: Conductor, discriminant, and the Noether formula of arithmetic surfaces, Duke Math. J. 57 (1988), no. 1, 151–173. [77] Schmickler-Hirzebruch, U.: El liptische Flachen uber P1C mit drei Ausnahmefasern und die hypergeometrische Differentialgleichung, Schriftreihe des Mathematischen Instituts der Universitat Munster (1985). [78] Schoen, C.: Bounds for rational points on twists of constant hyperel liptic curves, J. reine angew. Math. 411 (1990), 196–204. [79] Schutt, M.: The maximal singular fibres of el liptic K3 surfaces, Arch. Math. 87, Nr. 4 (2006), 309–319. [80] Schutt, M.: CM newforms with rational coefficients, Ramanujan J. 19 (2009), 187–205. [81] Schutt, M.: Fields of definition of singular K3 surfaces, Communications in Number Theory and Physics 1, 2 (2007), 307–321. [82] Schutt, M.: K3 surfaces with Picard rank 20 over Q, Algebra & Number Theory 4 (2010), no. 3, 335–356. [83] Schutt, M., Schweizer, A.: Davenport-Stothers inequalities and el liptic surfaces in positive characteristic, Quarterly J. Math. 59 (2008), 499–522. [84] Schutt, M., Schweizer, A.: On the uniqueness of el liptic K3 surfaces with maximal singular fibre, preprint (2007), arXiv: 0712.3873. [85] Serre, J.-P.: Lectures on the Mordel l-Weil Theorem. Viehweg (1989). [86] Shafarevich, I. R: Basic algebraic geometry, Second edition, Springer-Verlag, Berlin, 1994. [87] Shimada, I: On el liptic K3 surfaces, Michigan Math. J. 47 (2000), no. 3, 423–446. [88] Shimada, I.: Transcendental lattices and supersingular reduction lattices of a singular K 3 surface, Trans. Amer. Math. Soc. 361 (2009), 909–949. [89] Shimada, I., Zhang, D. Q.: Classification of extremal el liptic K 3 surfaces and fundamental groups of open K 3 surfaces, Nagoya Math. J. 161 (2001), 23–54. [90] Shimura, G.: Introduction to the arithmetic theory of automorphic functions, Iwanami Shoten & Princeton University Press (1971). [91] Shioda, T.: On el liptic modular surfaces, J. Math. Soc. Japan 24 (1972), 20–59. [92] Shioda, T.: An explicit algorithm for computing the Picard number of certain algebraic surfaces, Amer. J. Math. 108, No. 2 (1986), 415–432. 92 MATTHIAS SCH UTT AND TETSUJI SHIODA certain surfaces, el liptic K3 [93] Shioda, T.: On the Mordel l-Weil lattices, Comm. Math. Univ. St. Pauli 39 (1990), 211– 240. [94] Shioda, T.: Mordel l-Weil lattices and sphere packings, Amer. J. Math. 113 (1991), 931– 948. [95] Shioda, T.: An infinite family of el liptiic curves over Q with large rank via N´eron’s method, Invent. math. 106 (1991), 109–119. [96] Shioda, T.: On the ranks of el liptic curves over Q(t) arising from K3 surfaces, Comm. Math. Univ. St. Pauli 43 (1994), 117–120. [97] Shioda, T.: The el liptic K3 surfaces with a maximal singular fibre, C. R. Acad. Sci. Paris, Ser. I, 337 (2003), 461–466. [98] Shioda, T.: theorem of sandwich Kummer Proc. Japan Acad. 82, Ser. A (2006), 137–140. [99] Shioda, T: K3 surfaces and sphere packings, J. Math. Soc. Japan 60 (2008), 1083–1105. [100] Shioda, T.: surfaces, The abc-theorem, Davenport’s inequality and el liptic Proc. Jap. Acad. 84, Ser. A, (2008), 51–56. [101] Shioda, T.: Grobner Basis, Mordel l-Weil Lattices and Deformation of Singularities, preprint (2009). [102] Shioda, T.: El liptic fibrations of maximal rank on supersingular K3 surfaces, in prepara- tion. [103] Shioda, T., Inose, H.: On Singular K 3 Surfaces, in: Baily, W. L. Jr., Shioda, T. (eds.), Complex analysis and algebraic geometry, Iwanami Shoten, Tokyo (1977), 119–136. [104] Shioda, T., Mitani, N.: Singular abelian surfaces and binary quadratic forms, in: Classi- fication of algebraic varieties and compact complex manifolds, Lect. Notes in Math. 412 (1974), 259–287. [105] Silverman, Joseph H.: Heights and the specialization map for families of abelian varieties, J. Reine Angew. Math. 342 (1983), 197–211. [106] Silverman, J. H.: The arithmetic of el liptic curves. Graduate Texts in Math. 106, Springer (1986). [107] Silverman, J. H.: Advanced Topics in the Arithmetic of El liptic Curves. Graduate Texts in Math. 151, Springer (1994). [108] Siu, Y. T.: Every K3 surface is Kahler, Invent. Math. 73 (1983), no. 1, 139–150. [109] Sterk, H.: Finiteness results for algebraic K 3 surfaces, Math. Z. 189 (1985), no. 4, 507– 513. [110] Szydlo, M.: El liptic fibers over non-perfect residue fields, J. Number Theory 104 (2004), no. 1, 75–99. [111] Tate, J.: Algebraic cycles and poles of zeta functions, in: Arithmetical Algebraic Geometry (Proc. Conf. Purdue Univ., 1963), 93–110, Harper & Row (1965). [112] Tate, J.: Endomorphisms of abelian varieties over finite fields, Invent. Math. 2 (1966), 134–144. [113] Tate, J.: Algorithm for determining the type of a singular fibre in an el liptic pencil, in: Modular functions of one variable IV (Antwerpen 1972), Lect. Notes in Math. 476 (1975), 33–52. [114] Tate, J., Shafarevich, I. R.: The rank of el liptic curves, Dokl. Akad. Nauk SSSR 175 (1967), 770–773. [115] Usui, H.: On the Mordel l-Weil lattice of the el liptic curve y 2 = x3 + tm + 1, I, II, III, IV, Comm. Math. Univ. St. Pauli 49 (2000), 71–78, 50 (2001), 65–87, 55 (2006), 173–194, 57 (2008), 23–63. [116] van Luijk, R.: K3 surfaces with Picard number one and infinitely many rational points, Algebra Number Theory 1 (2007), no. 1, 1–15. ELLIPTIC SURFACES 93 [117] Vinberg, `E. B.: The two most algebraic K 3 surfaces, Math. Ann. 265 (1983), no. 1, 1–21. [118] Weinberger, P. J.: Exponents of the class groups of complex quadratic fields, Acta Arith. 22 (1973), 117–124. [119] Wiles, A.: Modular el liptic curves and Fermat’s Last Theorem, Ann. Math. (2) 141, No. 3 (1995), 443–551. [120] Wittenberg, O.: Intersections de deux quadriques et pinceaux de courbes de genre 1, Lect. Notes in Math 1901. Springer, Berlin, 2007. Institut fur Algebraische Geometrie, Leibniz Universitat Hannover, Welfen- garten 1, 30167 Hannover, Germany E-mail address : [email protected] URL: http://www.iag.uni-hannover.de/~schuett/ Department of Mathematics, Rikkyo University, Tokyo 171-8501, Japan E-mail address : [email protected] URL: http://www.rkmath.rikkyo.ac.jp/math/shioda/ Research Institute, for Mathematical Sciences, Kyoto University, Kyoto 606- 8502, Japan E-mail address : [email protected]
1208.1590	4	1208	2013-06-07T23:59:22	A Wonderful Embedding of the Loop Group	[ "math.AG" ]	Index of notation added. Shortening of some section, simplification of some of the arguments, some small added results and strengthening of thm 3.10. Also a significant re-writing of the last section.	math.AG	math	A WONDERFUL EMBEDDING OF THE LOOP GROUP PABLO SOLIS Abstract. I describe the wonderful compactification of loop groups. These compactifications are ob- tained by adding normal-crossing boundary divisors to the group LG of loops in a reductive group G (or more accurately, to the semi-direct product C× LG) in a manner equivariant for the left and right C×LG-actions. The analogue for a torus group T is the theory of toric varieties; for an adjoint group G, this is the wonderful compactications of De Concini and Procesi. The loop group analogue is suggested by work of Faltings in relation to the compactification of moduli of G-bundles over nodal curves. Contents Introduction 3 1 0 2 n u J 7 ] . G A h t a m [ 4 v 0 9 5 1 . 8 0 2 1 : v i X r a 1. 1.1. Compactifications of G 1.2. Loop groups 1.3. LG and G-Bundles on Curves 1.4. Summary 2. Compactifications of Reductive Groups 2.1. Wonderful Compactification of an Adjoint Group 2.2. Extension to Reductive Groups 2.3. Stacky Compactifications 3. Extension to Loop groups 3.1. Preliminaries on Loop Groups 3.2. The Wonderful Embedding of the Loop Group 3.3. Completeness and Stacky Extension 4. Some toric geometry of LT 4.1. The group T 4.2. Voronoi and Delaunay Subdivisions and the fan 5. Bundles on Curves 5.1. The double coset construction 5.2. The connection with X af f ×VT 5.3. G-torsors 5.4. G-torsors on nodal curves Nomenclature References 1 2 3 3 4 5 5 7 8 11 11 13 18 20 20 22 24 24 25 26 26 30 31 poly and bundles on nodal curves 1. Introduction Let G be a simple simply connected algebraic group over C. This paper studies smooth compactifica- tions of G and limits of families of principal G bundles on nodal curves. The connection between the two can be seen as follows. If we fix a smooth curve C over Spec C then the stack BunG(C) of all G-bundles C satisfies the valuative criterion for completeness, which means given a commutative square as below we can always fill in the dotted arrow: Key words and phrases. Loop groups; affine Lie algebras; moduli of G bundles on curves; embeddings of reductive groups; representation theory; spherical varieties. 1 2 PABLO SOLIS BunG(C) Spec K Spec R / Spec C ∗ ∗ Ey→ π where R is a complete DVR with fraction field K. When C is a nodal curve then BunG(C) no longer has this property. To see this let E → C be a principal G-bundle and π∶ C→ C the normalization of C. Then E can be identified π Ez where y, z are the pre images of the node x∈ C. We can consider φ∈ G and isomorphism φ∶ π in families φ can go to infinity. Thus compactifications of G are relevant to any completion of BunG(C). Over a fixed nodal curve Bhosle in [3] has completed BunG(C) simply by compactifying G. However by Kausz for G= GL(V). The key innovation we demonstrate here is that in order to provide a similar this does not address how to compactify BunG over a family of curves and does not give a modular interpretation of what the boundary of the completion means. Such a modular interpretation was given E together with an construction for G simple one must not just compactifify G but in fact 'compactify' or complete the loop group LG. ∗ Much work has been done in both the subject of compactifying reductive groups and the study of bundles on curves via loop groups. Let us give a brief account of some of the relevant results in these areas. 1.1. Compactifications of G. In 1983 De Concini and Procesi studied the symmetric space G~H where a "wonderful" compactification G~H of G~H; see definition 2.1. After De Concini and Procesi's result paticular case is G= G× G~∆(G); when G is of adjoint type this gives a wonderful compactification of the properties of their compactification were axiomatized and such varieties were called wonderful. A G is a complex Lie group and H is the fixed point set of an involution σ of G; see [8]. They constructed G. A construction of the wonderful compactification which we will exploit uses representation theory: G= G× G.[id]⊂ PEnd(V(λ)) (1) where λ is a regular dominant weight. In fact smooth compactifications G for all reductive groups G exist [6, 6.2.4] but in general lack certain combinatorial properties required to be wonderful. Additionally, there is a so called canonical embedding of a semi simple group but this compactification is generally not smooth unless Z(G)= 1. The only exceptions are when G= Sp2n(C). Several places in the literature mistakenly state that the only exception is G= Sp2(C)= SL2(C). I thank Johan Martens for preventing another mistake here. The canonical embedding for semi simple G has finite quotient singularities and the singularities can be resolved by working with Deligne-Mumford stacks. In [23], Martens and Thaddeus carry this out explicitly by constructing certain moduli problems about G-bundles on chains of P1s that represent the compactification. In this paper, we give a different proof using representation theory. Namely for a regular dominant weight λ there is a quasiprojective variety Y with an action of a torus T such that the global quotientX=[Y~T] contains G as a dense open subvariety. Additionally,X contains a dense open substackX0 which is the closure of the open cell U Theorem 2.10. Let G be a semi simple group then there is a stacky compactification X of G with an action of G× G with the following properties (a) X is smooth and proper. (b) X−X0 is of pure codimension 1 and we have an exact sequence 0→ Zr→ P ic(X)→ Z→ 0 where Z is a finite group and the subgroup Zr is generated by the irreducible components ofX−X0. (c) The boundaryX−G consists of r divisors D1, . . . , Dr with simple normal crossings and the closure of the G× G-orbits are in bijective correspondence with subsets I⊂[1, r] in such a way that to I we associate∩i∈I Di. T U of G and − / / 8 8 / A WONDERFUL EMBEDDING OF THE LOOP GROUP 3 (see remark 13). chamber whose lattice points lie in M . important when we turn to the study of loop groups. is an ind-stack constructed using a regular dominant weight λ of Gaf f . Though this result is not new the fact that is can be proved using only representation theory will be (d) Let u1, . . . , ur be generators of the rays of the Weyl chamber and M be the monoid they generate. Any G equivariantX′→X determines and is determined by a fan supported in the negative Weyl 1.2. Loop groups. The algebraic loop group is LG= G((z))= G(Spec C((z))) and the smooth loop group is LsmG= C ∞(S1, G). We state the main result for LG but we have an analogous result for LsmG In fact, the group of interest is a semi-direct product C× LG; this means for u ∈ C× −1= γ(uz). The group Gaf f appearing in 3.10 is a central extension of C× LG. The object X af f uγ(z)u Theorem 3.10. If G is a simple, connected, simply connected algebraic group over C with center Z(G) then C× LG~Z(G) has a wonderful embedding X af f= C× LG~Z(G). There is a dense open substack which is the closure of the open cell in C× LG~Z(G) and is of pure codimension 1. It is a union of r+ 1 divisors that are Cartier and freely (b) X af f− X af f (c) The boundary X af f− L G~Z(G) consists of r+ 1 normal crossing divisors D0, . . . , Dr and the closure of the Gaf f× Gaf f -orbits are in bijective correspondence with subsets I⊂[0, r] in such a way that to I we associate∩i∈I Di. ′→ X af f determine and is determined by a Weyl equivariant morphism of ′→ T We use the word embedding because C× LG does not satisfy the valuative criterion for completeness. However we do have a completeness result for the polynomial loop group G(C[z ±]); see theorem 3.11. weight representation V(λ) of Gaf f . V(λ) is an infinite dimensional vector space which is a direct sum The strategy for proving 3.10 is to use the representation theory of Gaf f . More precisely, take a highest (a) X af f is independent of λ. generate the Picard group. (d) Any G equivariant X toric varieties T ad,0. we have X af f 0 0 of weight spaces Vµ for a maximal torus T af f of Gaf f . Now we consider X af f= Gaf f× Gaf f .[id]⊂ PV(λ) ⊗V(λ)∗ Where V(λ)⊗V(λ)∗∶=∏µ,ν hom(Vµ, Vν) and the product is over the weight spaces of the representation. There is also a stacky extension analogous to theorem 2.10, see theorem 3.12. morphism Theorem 1.1. For any connected topological group G and fixed Riemann Surface C we have the moduli stack 1.3. LG and G-Bundles on Curves. In this subsection we will focus on the smooth loop group because it will allow us to draw some intuitive pictures. BunG,hol(C) of holomorphic principal G-bundles on C. Atiyah and Segal [24] prove there is an iso- G consists of boundary values of holomorphic function from a small disc{z∈ C∶z< 1}≅ Here Lsm,+ C G consists of boundary values of holomorphic function C− Dp→ G. Dp⊂ C and Lsm Theorem 1.1 has a useful reformulation. We can write C = C 0∪ A where A is an annulus and C 0 is the disjoint union of a disc D and a noncompact Riemann surface C− D; see figure 1. Then for YA= LsmG× LsmG~Hol(A, G) we have BunG,hol(C)≅ Lsm C G(cid:131)LsmG~Lsm,+ G BunG,hol(C)= Hol(C 0, G) YA (2) Theorem 1.1 corresponds to gluing trivial bundles on a small disc and on the complement of a small disc with a transition function in LsmG. Equation (2) corresponds to gluing the same trivial bundles by a 4 PABLO SOLIS pair of elements in LsmG over a cylinder or annulus; in either case you produce a G-bundle on the same curve. Figure 1. The double coset construction. The left picture is the analytic picture un- derlying theorem 1.1 and the right is the analytic picture underlying equation (2). With these preliminaries in place, let us describe an approach for studying G-bundles on a smooth curve degenerating to a nodal one. Holomorphically (or algebraically formally) any family of curves over a 1-dimensional base with smooth generic fiber and nodal special fiber has an ´etale neighborhood that looks like the genus 0 degeneration represented by the morphism This is illustrated in figure 2. For a small positive constant δ, we can describe figure 2 as Au∶={(x, y)∈ C2x,y< δ, xy= u}; a family of annuli parametrized by u. In other words, if Cu is a family of smooth Spec C[x, y, u] xy− u → Spec C[u]. Figure 2. The local picture of a nodal degeneration. BunG,hol(Cu)= YAu curves degenerating to a nodal curve C0 then we can write Cu= C 0 u∪ Au. In analogy with (2) we have Hol(C 0 u, G) where YAu= LsmG× LsmG~Hol(Au, G). This construction allows us to reduce to studying the homoge- Let us now ignore the norm restrictions on x, y and allow and u∈ C= A1. We'll write this as YA p-→ A1 and for u∈ A1 we have p −1(0) = −1(C×) ≅ C× LG and p BunG,triv(A0) where the subscript "triv" means we are considering G-bundles together with a trivializa- −1(0) and completes this fiber and in turn allows one to complete BunG,hol(Cu). The cutting and gluing mentioned in this con- −1(C×) contains p tion on the boundary of the disc. The embedding X af f of p The main idea can now be presented as follows. One can show p neous space YAu . −1(u)= YAu . (3) struction posses no problems in the holomorphic setting but the appropriate algebraic analogue requires more care and will be addressed in a follow up paper. 1.4. Summary. In section 2, we recall the construction of the wonderful compactification of an adjoint group and use the theory of toric stacks to generalize the construction to produce wonderful compactifi- cations for all semisimple groups 2.10; this gives a representation theoretic construction to results in [23]. In section 3, we give basic definitions regarding loop groups and their representations. Additionally, we generalize the results in section 2 to the loop group setting to produce an embedding X af f . In section 4, A WONDERFUL EMBEDDING OF THE LOOP GROUP 5 we discuss in detail the positive energy representations of the loop groups LT for a torus T and describe the embeddings of LT in terms of combinatorics of positive definite forms. In section 5, we discuss in more detail the connection between LG and bundles on curves or more generally, torsors on curves. For the convenience of the reader I have included an index of notation which contains the notation defined in one section but used in multiple other sections. Acknowledgements I would like to thank Constantin Teleman for advising this project. I also would like to thank Michel Brion, Mathieu Huruguen, Carlos Simpson, Michael Thaddeus for useful discussions. I would like to especially thank Johan Martens for helpful discussions and also for pointing out some inaccuracies in an earlier version of this paper. 2. Compactifications of Reductive Groups This section reviews the construction of the wonderful compactification of a semisimple group of adjoint type, we recall the basic results regarding its structure and in section 2.2 describe an extension to reductive groups. In section 2.3 we use the theory of toric stacks to provide wonderful stacky compactifications for a general semi simple group. This method extends to to give compactifications of reductive groups but a further choice of a subdivision of a Weyl chamber must be made. 2.1. Wonderful Compactification of an Adjoint Group. This section largely follows chapter 6 of [6]. − − ∗ χ . weights. , their unipotent radicals U, U 2.1.1. Definitions. Let G be a semisimple group. opposite Borels B, B It has associated subgroups: a maximal torus T , . The character lattice we denote as ΛT , the co- character lattice we denote as VT and if µ∈ ΛT , η∈ VT then the integer µ○ η we denote asµ, η,µ, η, or µ(η). Let r= rk(G) and let α1, . . . , αr be the positive simple roots. Let ω1, . . . , ωr be the fundamental For dominant weight λ let V(λ) denote the highest weight representation (HWR) of highest weight (HW) λ. By V(λ)µ we denote the weight space of V(λ) with weight µ. When no confusion is likely to Let PEnd(V(λ))0⊂ PEnd(V(λ)) be the open subset consisting of points whose projection to Vλ⊗ V ∗ End(V(λ))= ࣷ µ,χ∈ΛT arise write simply Vµ. We can decompose Vµ⊗ V λ is (4) not zero. Define for a regular dominant weight λ Lastly, let H be a reductive group and Y a normal H-variety X∶= G× G.[id]∈ PEnd(V(λ))= PV(λ)⊗ V(λ)∗(cid:6) X0∶= X∩ PEnd(V(λ))0 Let Z(G) denote the center of G. It is routine to see that Stab([id]) is Z(G)× Z(G)⋅ ∆(G) and consequently X contains Gad∶= G~Z(G)= G×G Z(G)×Z(G)∆(G) as a dense open subset. X is the wonderful compactification of Gad and X0 is the open cell of X. A maximal torus for Gad is Tad= T~Z(G). By Tad we mean the closure of Tad in X and Tad,0∶= X0∩ Tad. D1, . . . , Dr such that the H-orbit closures are give by intersections∩i∈I Di for any subset I⊂{1, . . . , r}. Theorem 2.2. Let X= Gad be as in (4). Then (c) X− X0 is of pure codimension 1; it consists of divisors that freely generate the Picard group. The notation of X, X0, Tad, Tad,0 does not reflect the dependence on λ; this is justified by theorem 2.2. Definition 2.1. Y is wonderful of rank r if Y is smooth, proper and has r normal corssing divisors 2.1.2. Properties of the Wonderful Compactification. The main result of this subsection is theorem (a) X is independent of λ. (b) X is smooth. 6 (d) The boundary X− Gad consists of r normal crossing divisors D1, . . . , Dr and the closure of the G× G-orbits are in bijective correspondence with subsets I ⊂[1, r] in such a way that to I we associate∩i∈I Di. ′→ X determines and is determined by a fan supported in the negative Weyl (e) Any G equivariant X PABLO SOLIS chamber. The proof will follow after a few lemmas which will also serve as our basic tools for studying stacky compactifications for reductive groups and loop groups. −α represent the regular function on T given by the character−α. Let t Lemma 2.3. T 0≅ Spec C[t −α1, . . . , t −αr]≅ Ar 2 ∈ X maps isomorphically −×U×T→ X sending(u1, u2, t) u1tu −1 (cid:3) Proof. See the proof of 2.7. Lemma 2.4. The action morphism U onto X0. Proof. See 3.6. (cid:3) Lemma 2.5. Let G be a reductive group and let V a HWR of G. Then the orbit of the HW vector is the only closed orbit in PV Proof. This follows from the Borel fixed point theorem [4, Pg.272] for reductive groups. (cid:3) we have (b). p λ and smooth. Observe that G× G.X0= X For (a), consider compactifications Xλ and Xµ associated to weights λ, µ. Let X∆ be the closure Remark 1. When we discuss loop groups we give a different proof (lemma 3.2) that circumvents the need for a Borel fixed point theorem in the Kac-Moody setting. ∗ (5) λ is the unique closed orbit in P and it clearly intersects P0. proof of theorem 2.2. To prove(a),(b), we note that lemmas 2.3,2.4 show T and X0 are independent of This follows from lemma 2.5: G× G.vλ⊗ v Now P−G×G.P0 is G×G-stable and closed therefore must be empty and hence we have (5). Consequently of ∆(G) in Xλ× Xµ. The projection pλ∶ X∆ → Xλ is equivariant and lemmas 2.3,2.4 imply X∆,0 ∶= λ (Xλ,0)≅ U −⋅ Tad,0⋅ U ; that is, the restriction of pλ to X∆,0 is an isomorphism and therefor induces a −1 Gad× Gad equivariant isomorphism on This proves(a). Part(c) uses the result that if U ⊂ Y is a dense open affine of any scheme Y then Y − U is of pure In the case at hand, X0 is affine and P ic(X0) = 0; this immediately shows P ic(X) is generated by the irreducible components of X− X0. A relation among these generator is a principal divisor(f) which is invertible on X0; such a function is a constant c and f− c is zero on a dense open set hence there are no relations. Part(d) is proved by applying lemma 2.4 to reduce it to the case T 0 for which it is obvious. Part(e) is addressed in the next section. algebra g⊕ g consider the point ∆(g)∈ Grr,2r and consider the closure of G× G.∆(g) in Grr,2r; this gives Alternatively in [31], Vinberg constructs an affine scheme S with the action of G× T and identifies an open set S0⊂ S such that the GIT quotient S0~~T is also isomorphic to the wonderful compactification. g.X∆,0= X∆→ Xλ= g∈G×G Remark 2. There are alternative constuctions of the wonderful compactification. Starting with the Lie another construction of the wonderful compactification of G; see [8, pg.19]. codimesion 1. This is proved in [25, 2.4]. g∈G×G Thaddeus and Martens use this construction and generalize it to provide stacky compactifications of reductive groups; [23, § 5,6]. g.X0. (cid:3) A WONDERFUL EMBEDDING OF THE LOOP GROUP 7 2.2. Extension to Reductive Groups. Let G be a connected reductive group. In [6, 6.2], Brion and Kumar define a G-embedding Y to be a normal G× G variety containing G=(G× G)~diag(G) as an open orbit. They call X toroidal if the quotient map G→ G~Z(G)= Gad extends to a map Y → Gad=∶ X= the wonderful compactification of Gad. In fact, toroidal has a more general definition in the theory of spherical varieties but we will not need this level of generality. They prove the following proposition; see [6, 6.2.4]. Proposition 2.6. Any toroical G embedding is determined by its associated toric variety. In particular, given reductive group G, a maximal torus T and a fan of the form Σ= W⋅ Σ0 where W is the Weyl group Proof. Let Y be a toroidal G-embedding. We have a map p∶ Y → Gad and the associated toric variety is −1(Tad). Equivariance implies that and Σ0 is a fan with support in the negative Weyl chamber, one can construct a G-embedding and an equivariant morphism to X. p −1(X0)≅ U −× p −1(Tad,0)× U. p We now describe how to go from a Weyl equivariant toric variety to a G-embedding. The above obser- vation will show these are inverse constructions. Let Σ0 be as in the proposition and for a cone σ ⊂ Σ0 let Tσ,0 be the corresponding toric variety. pσ-→ X such that σ (X0)= U Tσ,0U ; further if τ⊂ σ⊂ Σ0 then we have an open immersion Gτ → Gσ. Using these open −1 The idea is that given σ we can construct a toroidal G-variety Gσ with a map Gσ p immersions we can glue together the various Gσ to get the desired toroidal variety GΣ0 . − p i faces. with a map Gσ This we do in 2.7 and 2.8 −× Tσ,0× U pσ-→ X such that It remains to actually construct the spaces Gσ and show the gluing maps satisfy the co-cycle condition. (cid:3) G× G.[id]⊂ PEnd(V(λ))ࣷ σ (X0)≅ U −1 ∨]; choose generators µ1, . . . , µm of ΛT ∩ σ End(V(λ+ µi))=∶ P[λ, λ+ µi] Lemma 2.7. Let Σ0 be as in proposition 2.6 and let σ⊂ Σ0 be a cone. Then there is toroidal variety Gσ and the closure Tσ of the torus in T ⊂ Gσ is given by the fan whose cones are wσ for w∈ W and their Proof. We have Tσ,0 = Spec C[ΛT ∩ σ dominant weight λ such that λ+ µi are all regular dominant. Consider and the rational map pσ∶ P[λ, λ+ µi]⇢ PEnd(V(λ)). Set Gσ to be the subset of G× G.[id] such that pσ(Gσ)⊂ X; it is evident that this is a G× G equivariant map. Then restriction of pσ∶ Gσ→ X to G is just the projection G→ Gad. As U . From the equivariance of pσ and from X0≅ U of G we have pσ(U σ (Tad,0)× U −1 P[λ, λ+ µi] can be depicted as: λ(t) ⊕ σ (Tad,0)= Tσ,0. Let µ1, . . . , µm be generators for σ −1 ∨ λ(t)µm(t) λ(t)µ1(t) ± −× Tad,0× U we have λ(t)µm(t) α1(t) σ (X0)= U −1 −× p So we are reduced to showing p ±)= U ± λ(t)µ1(t) α1(t) do not intersect the center ⊕⊕ . Choose a regular . The image of T in ∨ p λ(t) α1(t) λ(t) α2(t) λ(t)µ1(t) α2(t) λ(t)µm(t) α2(t) of Tad. . It follows 8 PABLO SOLIS −1. ⊕ 1 α The variety p µ1(t) α1(t) µ1(t) µm(t) µ1(t) α2(t) 2 (t) −1 σ (Tad,0) consists of those points where λ(t)≠ 0. This is: −1 1 (t) −1 ⊕⊕ µm(t) α1(t) µm(t) α2(t) Let σW denote the negative Weyl chamber in VT . Recall we require σ⊂ σW so all−αi∈ σ W ⊂ σ ∨ ∨ that all the diagonal entires are polynomials in µ1, . . . , µm so the projection to Spec k[µ1(t), . . . , µm(t)]≅ Tσ,0 is an isomorphism. The last statement of the proposition follows because Tσ= p σ (Tad) and −1 This in turn follows from the G×G equivariance of the compactification X which gives W×W equivariance Lemma 2.8. If Σ0 is as in 2.6 and σ1, σ2, σ3⊂ Σ0 then the Gσi can be glued along Gσi∩σj . Proof. Using lemma 2.7 we reduce the gluing construction to the torus: for any cone σ∈ Σ0 let Fσ denote . If τ⊂ σ then we can choose Fσ⊂ Fτ . Let λτ be such that λτ+ Fτ are all regular dominant weights. There is a projection P[λτ , τ]→ P[λτ , σ] which maps Gτ → Gσ and using lemma 2.7, equation (5) and the fact that Tτ,0→ Tσ,0 is an open immersion we conclude the same for Gτ→ Gσ. which it is manifestly obvious; namely, by G× G equivariance we can move any point q∈ Gσ1∩σ2∩σ3 to lie in T σ1∩σ2∩σ3,0. The cocycle condition follows immediately because it can be reduced to checking it for the tours for Tad= w∈W a choice for a set of generators for σ (cid:3) (cid:3) α w Tad,0 w ∨ 2.3. Stacky Compactifications. In general the fans Σ that arise for reductive groups produce singular toric varieties and hence singular G-embeddings. However these toric varieties are always smooth as stacks. Here we describe a modification of the above construction that produces a smooth stack. We briefly recall the basic theory of toric stacks and then incorporate them into the above construction. 2.3.1. Preliminaries on Toric Stacks. Following [14], we define toric stacks as[Y(Σ)~Z] where Y(Σ) is a normal toric variety with associated fan Σ and Z is a subgroup of the torus TΣ⊂ Y(Σ). The stack [Y(Σ)~Z] contains the torus T= TΣ~Z as a dense open subscheme. Just as in the theory of toric varieties, toric stacks are encoded by combinatorial data called stacky fans. A stacky fan is a pair(Σ, β∶ L→ N) where L, N are lattices, Σ is a fan in L⊗ R, and β is a homorphism of finite index. The equivalence between toric stacks and stacky fans is given as follows. Given[Y(Σ)~Z] we get a surjection TΣ→ T which induces a map β∶ VTΣ→ VT . Thus we get the stacky fan (Σ, β∶ VTΣ→ VT). Starting from(Σ, β∶ N→ L) we note that the hypothesis on β implies that the dual morphism ∗∶ hom(L, Z)= L ∗= hom(N, Z) ∗→ N , C×) and T∶= hom(L is injective. Consider the tori TΣ∶= hom(N ∗ ∗ ---------→ T→ 0 0→ Z(β)→ TΣ hom(β∗,C×) and thus we get the toric stack[Y(Σ)~Z(β)]. C={v∈ VT⊗ R∶= VT,Rαi(v)≥ 0} is a fundamental domain for the action of the Weyl group on VT⊗ R and defines a rationally smooth Here is how toric stacks appear in our situation. Recall the notation G, T, α1, . . . , αr from section 2. , C×). Dualizing β The Weyl chamber: we get a surjection fan. Let u1, . . . , ur be generators of the rays of the fan C and M be the monoid they generate. We get a homomorphism of lattices ∗ β β∶ Zr eiui---→ VT (6) ′ and let C A WONDERFUL EMBEDDING OF THE LOOP GROUP be the standard cone generated by the coordinate rays in Zr⊗ R giving rise to the toric variety Ar. We get a smooth toric stack [Ar~Z(β)] associated to the stacky fan (C , β) whose coarse ′ moduli space is the toric variety associated to the Weyl chamber in VT⊗ R. For general toric stacks the subgroup Z(β) is arbitrary but in the present situation Z(β) is always finite. In what follows we use the isomorphism[Ar~Z(β)]≅[(Ar× T)~(C×)r] where(C×)r acts diagonally on the product. Remark 3. Notice β(C ′⊂ L is any monoid such ′)= M and if L is any lattice with a map L β′-→ VT and M 9 ′(M ′)⊂ M then we can lift β ′ : that β / VT β β′ Zr l L induces a map ′)⊂ C ′ . This is used in the proof of theorem 2.10(d). Definition 2.9. The stacky wonderful compactification of G is ∆-→(C×)r×(C×)r (id,β)∗----→ VT,R⊕ VT,R arbitrary semi simple group G analogous to the wonderful compactification of Gad. We use proposition 2.6 together with the theory of toric stacks to give stacky compactifications of an and l(M The stacky compactifications will be quotients of an embedding of the group H = G×(C×)r where r= rk(G). The embedding is determined, via proposition 2.6, by a fan C∆⊂ VT,R⊕ V(C×)r,R in negative Weyl chamber of H. Using the map β in (6) we get homormorphism T×(C×)r (id,β)---→ T× T . This map VT,R⊕ V(C×)r,R and the desired cone C∆ is the inverse image of−C⊕ C⊂ VT,R⊕ VT,R. By 2.6 we get an H-embedding Y(C∆) ; Y(C∆) is a strictly quasi-projective variety. Set Tβ ∶=(C×)r and consider a subgroup of H via Tβ (id,β)---→ T×(C×)r ⊂ H the embedding Y(C∆) carries an action of H×H so we get an action of Tβ×Tβ. Identify Tβ with Tβ×Tβ~∆(Tβ). The construction of the embedding Y(C∆) and thus X depend on a choice of a regular dominant Theorem 2.10. Let G be a semi simple group andX as in definition 2.9. Then: (a) X is smooth and proper. (b) X−X0 is of pure codimension 1 and we have an exact sequence 0→ Zr→ P ic(X)→ hom(Z(β), C×)→ 0 where the subgroup Zr is generated by the irreducible components ofX−X0. (c) The boundaryX−G consists of r divisors D1, . . . , Dr with simple normal crossings and the closure of the G× G-orbits are in bijective correspondence with subsets I⊂[1, r] in such a way that to I we associate∩i∈I Di. Any G equivariantX′→X determines and is determined by a fan supported in the negative Weyl ∆ is generated by(0, ai),(±ωi,±wi). It ∗(αi), wi= β ∨ Proof. Let ai= β follows that the toric variety associated to C∆ is isomorphic T× Ar and using lemma 2.7 we have weight λ but just as in the case of the wonderful compactification different choices of λ give isomorphic objects. (d) Let u1, . . . , ur be generators of the rays of the Weyl chamber and M be the monoid they generate. X= G∶=[Y(C∆)~Tβ]. chamber whose lattice points lie in M . ∗(ωi)∈ ΛVβ . One can verify that C X0∶=[Y(C∆)0~Tβ]=[(U −×[(T× Ar)~Tβ]× U ≅U ≅U −×[Ar~Z(β)]× U. −× T× Ar× U)~Tβ] / = = O O PABLO SOLIS Σ / C l f (7) 10 U a diagram Notice that p Gad hence (a). this reduces to showing f -----→ U (id,f,id) ′ is affine and finite; the first property is clear; finiteness follows because the is affine the complement is pure codimension 1 and we have an exact sequence Zr i-→ Cl(Xc)→ Cl(Xc,0)≅ Cl(Tc). −× Tad,0× U . The map f is the composition [Ar~Z(β)]→ Tc The open cell is smooth and its translates covers X . Further the restriction of X → Gad is given by −×[Ar~Z(β)]× U f′-→ Tad,0 where Tad,0 is the course moduli space of[Ar~Z(β)]. We will show f is affine and finite hence proper; roots of G are finite index in the weights of G. We conclude thatX is finite over the projective scheme Let Xc denote the coarse moduli space forX ; it comes with a surjection Xc pc-→ X. To prove (b) we use that P ic(X)= Cl(Xc); this is shown in remark 3.4 of [15]. Working in Xc we see that as Xc,0= U −×Tc×U c∶ Zr = P ic(X) → P ic(Xc) is injective; this follows for the same reason given in 2.2(c). ∗ Further after tensoring (7) with Q we get a surjection Qr iQ-→ P ic(Xc)Q and by the previous sentence iQ Now Cl(Tc)= P ic([Ar~Z(β)]) and the latter is equal to the group of Z(β)-equivariant line bundles on of hom(Z(β), C×). Altogether this gives (b). For (c) it follows similarly as for Gad by reducing to the torus T 0. We can present T 0 as[Ar~Z] where To prove (d) we first note that any equivariant map determines a map X′ 0 →X0 and consequently ′)~Z(β ′), β]. By [14, 3.4] this corresponds to determines a map of torics stacks[Y(Σ ′ VTΣ′ β′ VT′ Ar. Any line bundle on Ar is trivial and an equivariant structure is uniquely determined by an element Z is a finite group so the result follows by observing that it holds for the atlas Ar. ′)]→[Ar~Z]=[Y(C ′ VTβ= Zr must be an isomorphism and so i is injective. ′ denote by l ∆ which is the inverse image of Now apply proposition 2.6 for the group H H and we have an equivariant map This gives one direction. Conversely if we are given a fan whose lattice points lie in M then by remark (id,β′)----→ VT⊕ VT . This produces an embedding Y(C 3 we can lift this data to a diagram as above. In particular, we also have a map TΣ′→ Tβ which we also ′= G× TΣ′ and the cone C −C⊕ C under the map VT⊕ VTΣ′ ∆) and we claim ′ ∆)~TΣ′] is the desired G-equivariant embedding mapping toX . [Y(C ′ equivariant map to Y(C∆) (H ′× H To see this note that Y(C ∆) has an H ′ ′ ′ ∆)×TΣ′ Tβ is a Tβ-bundle ′ id,l--→ H). Further Y(C ∆)~TΣ′] and therefore Y(C ∆) is a TΣ′-bundle over[Y(C ′ ′ ′ ∆)×TΣ′ Tβ→ Y(C∆)×Tβ Tβ≅ Y(C∆). Y(C ′ ∆)~TΣ′]→X . By the definition ofX we get a map[Y(C ′ thus there are fans supported in −C which don't yield maps to X ; e.g. −C itself. canonical embedding Xcan of the group G does not admit a map toX unlessX= Xcan. monoid S it is an affine variety with an action of G× T . An open subset S0⊂ S is identified and their compactification is given by [S0~T]. Martens explained to me that these constructions are the same. Specifically results of [31] allow one to show S0 is a toroidal embedding of G× T and thus comes from a Remark 5. As noted in the introduction, most of this result has been proved by Martens and Thaddeus in [23]. For parts (a),(d) see specifically [23, 4.2,6.4]. Part (c) seems implicit in the treatment given in [23] but part (b) is new. The construction given in [23] uses a construction they call the Cox-Vinberg Remark 4. In general the generators of M don't generate all the lattice points of the Weyl chamber C In particular, the acts on the latter via fan supported in the negative Weyl chamber, this is the fan given by C∆ above. Alternatively one can (cid:3) β / VT / / / / A WONDERFUL EMBEDDING OF THE LOOP GROUP 11 also see the equivalence between the Vinberg construction and given construction using universal torsors; this is explained in [7, 3.2.4]. Remark 6. Of course this stacky construction extends to a general reductive group over C. But for a general reductive group the Weyl chamber is not a strongly convex rational polyhedral fan. That is, there is not a toric variety associated to the Weyl chamber; this happens for GLn. Thus to get an embedding of such groups we must further subdivide the Weyl chamber. Kausz does this for GLN as follows. Let φ∶ GLN → P GLN+1 be given by composing g g 0 (det g)−1 with the natural map SLN+1→ P GLN+1. This induces a map GLn⊃ T1 to T2⊂ P GLn+1 and an induced map on co-character φ∗∶ VT1 → VT2 . Let Σ⊂ VT1 be the inverse image of−C under φ∗. Then Kausz's compactification is the one associated to Σ via proposition 2.6. Certainly there are other possibilities. 0 3. Extension to Loop groups The aim of this section is to prove the analogue of theorems 2.2, 2.10 for loop groups. As mentioned in the introduction there is both the algebraic loop group and the smooth loop group. In this section we focus on the algebraic loop group. To a large extend the statements we prove also hold for the smooth loop group. We remark along the way what modifications, if any, are necessary to get a statement for the smooth loop group. When we apply these results to studying the moduli stack of bundles on curves we will be more explicit about the distinction between LG and LsmG. 3.1. Preliminaries on Loop Groups. Here we introduce the groups LG, Gaf f . Let G be a simple algebraic group over C with π1(G)= π0(G)= 1. The loop group LG is the functor from C-algebras to R LG(R)∶= G(R((z)))∶= G(Spec R((z))) G(R)∶= Gm(R) LG(R) which, for u∈ Gm(R), γ∈ LG(R) is defined by uγ(z)u −1= γ(uz). Finally the group As we are working over C we'll write C× 1→ Gm→ Gaf f→ L for Gm. Topologically, Gaf f= C×× LG×C× The functor LG is represented by an ind-scheme of infinite type; see [10]. Closely related is L Gaf f is a central extension of L c where the subscript G→ 1 groups given by G: c indicates the factor is central. G. − L sometimes we abbreviate T If α1, . . . , αr be the simple roots of G then the simple ∶= C×× T ⊂ L G(R)∶= G(R[[z]]) + B(R)∶={γ∈ L + U(R)∶={γ∈ L + To put this in perspective, semi simple groups are finite dimensional Kac-Moody groups. Gaf f is an infinite dimensional Kac-Moody group. Despite being infinite dimensional, Gaf f and LG behave much like a semi simple group. In particular, Gaf f has associated root data. c and characters are denoted as(n, µ, l)∈ Z⊕ ΛT⊕ Z; The maximal torus of Gaf f is T af f∶= C×× T× C× affine roots of Gaf f are(0, α1, 0), . . . ,(0, αr, 0), α0=(1,−θ, 0) where θ is the longest root of G. By abuse of notation we denote(0, αi, 0) simply by αi so the simple roots of Gaf f are α0, . . . , αr. In addition we have a standard Borel Baf f ∶= C××B× C× c whereB along with related subgroups is defined below. Further, the affine Weyl group W af f is by definition N(T af f)~T af f . It is isomorphic to −1]) G(R)∶= G(R[z −1G(R)γ∈ B G(R)γ∈ B(R) mod z} B−(R)∶={γ∈ L G(R)γ∈ U G(R)γ∈ U(R) mod z} U−(R)∶={γ∈ L − W VT and we have the Birkhoff factorization [20, pg.142] for Gaf f Gaf f= ࣶ w∈W af f poly where LG is replaced by LpolyG∶= G(C[z ±]). sm where LG is replaced by LsmG= C with{z= 1}⊂ C then, for example, L ∞(S1, G(C)). If we identify S1 +,smG= boundary values of holomorphic maps{z< 1}→ G and Of course this restrict to give a similar decomposition for LG and L Gaf f −(R) mod z −1} −1} −(R) mod z Similarly there is a group Gaf f G. Also we have an analogous group U− wBaf f (8) L 12 PABLO SOLIS L smooth version of (8) is [26, 8.7.3a]. G is the subgroup of boundary values of holomorphic maps {z> 1}→ G, etc. A reference for the − In addition to the simple affine roots, we also have roots(k, α, 0) where α is a root of G. For k≠ 0 α= 0 is allowed so in general the weight spaces are not 1 dimensional as dim g(k≠0,0,0)= dim t. W af f acts Weyl alcoves. Namely, the roots are linear forms on R⊕ tR and we can identify them with affine linear forms on tR by identifying the Lie algebra with 1⊕ tR. For α≠ 0 we can define affine hyperplane in tR via on the roots and one can talk about a Weyl chamber however it is standard practice to introduce instead Hk,α={ζ∈ tRα(ζ)=−k} The complement of all the Hk,α is known as the Weyl alcove decomposition of tR. W af f acts freely on tR and permutes the alcoves in the decomposition. A fundamental domain is given by the positive Weyl alcove Al0∶={ζ∈ tRαi(ζ)≥ 0, i= 0, . . . , r} − G,B− finite dimensional case. is an infinite dimensional scheme. of Gaf f with the following properties The former we call a thin flag variety and the later a thick flag variety. Remark 7. There are many parallels between G and Gaf f but there are also clear differences. For example, the negative versions L , etc. are neither conjugate nor abstractly isomorphic to the positive versions. This asymmetry is also reflected in the flag varieties of LG= G((z)): LG~B is a projective ind-variety (union of finite dimensional projective varieties) and LG~B− If ω1, . . . , ωr are the fundamental weights of G then the fundamental weights of Gaf f are ω0 = (0, 0, 1), (0, ω1, 1) . . . ,(0, ωr, 1). The definitions of dominant and regular weight carry over from the Proposition 3.1. Let(0, λ, h) be a regular dominant weight. There exists a representation V = V(0, λ, h) (a) If µ is any other wight of L then λ− µ is a sum of positive roots. (b) λ− αi is a weight of V for all i. (c) The stabilizer of the weight space Vλ in PV isBaf f . (d) The morphism Gaf fBaf f = LGB→ P(V) given by γB γVλ is injective and gives LGB the structure of a projective ind scheme; in particular LGB is closed in P(V). (e) The action of Gaf f on P(V) factors through a faithful action of Gaf fZ(Gaf f)= L G~Z(G). consider the reflection si(λ). It is of the form λ− nαi for n≥ 1 and all the weights λ− mαi for 0≤ m< n are weights of the representation. (a) [20, 1.3.22]. (b) This follows form [20, 1.3.22] and the representation theory of SL2. Namely, for each simple root, Proof. The existence claim is contained in [20, 13.2.8]. (c) [20, 7.1.2] (d) [20, Ch.7.1] (e) [20, 13.2.8]. More specifically we have the following commutative diagram GL(V) / P GL(V) Gaf f ψ G~Z(G) where ψ is a surjective group homomorphism and ker ψ= Z(Gaf f). L Remark 8. The statements (a) - (c) and (f) for LsmG are proved in [26, ch.8,9,11]. The first half of (d) also holds [26, 8.7.6] but there is no ind-statement to make about the flag varieties of the smooth loop group; they are infinite dimensional complex projective algebraic varieties. (cid:3) / / / A WONDERFUL EMBEDDING OF THE LOOP GROUP 13 For more details on Gaf f , its subgroup and its representation theory see [20, chap. 13] or [2] for and a third called a positive energy representation (PER) U respect to a norm for which elements in LsmK act as unitary operators; K is a compact form of G. is a thick flag variety. If we work with the polynomial loop group LpolyG(R)= G(R[z G= SLr. Remark 9. There are three duals associated to a HWR U=⊕µUµ. There is U ∗=∏µ Uµ ∗ ∗ . Again, statement(d) is different because Statements(a),(b),(c),(e) still hold for LG acting on U ∗ ±]) then LG~B− notion of closure in PV when dim V is uncountably infinite. We say Y ⊂ PU is closed if the inclusion Y → PU satisfies the valuative criterion for completeness. Colloquially Y is closed if any limit point of Y that exists in PU belongs to Y . For example the map C×→ U=∏n∈N C given by t(1, t, t −2, . . .) has no limit in PU as t→ 0 or t→∞ and consequently is a closed subset. We end this subsection with the following technical but important result. The result requires the proposition 3.1 holds equally well for both U, U pos. PER result from completing U res=⊕µU ∗ −1, t2, t res with µ and U res. ∗ ∗ ∗ ∗ set I res. . The same statement holds for Gaf f Lemma 3.2. Let V a HWR of Gaf f . The orbit of a HW vector is the only closed orbit in PV . The orbit of a lowest weight vector is the only closed orbit in PV poly acting on PV, PV Proof. Suppose the orbit of v is closed in PV . Then for any 1-parameter subgroup Ga⊂ Gaf f we have the closure Ga.v is also in the orbit of v. We use this observation to show that if G.[v] is closed in PV then Gaf f .[v]= Gaf f .[vλ] where λ is the HW of V . For a positive simple root αi set Ui(b)∶= exp(b⋅ Xi) with b∈ C. The representation V is in particular an integrable representation of gaf f thus Xi acts locally nilpotently on V . So for any v the vector Uα(b).v b 0 exists in PU as well as the limit b∞; we write the latter as the limit b 0 of Uα(1~b).v. is a vector whose finitely many nonzero entries are a polynomial function of b. It follows that the limit v= v(0)∶= lim b→0 where 1λ spans a one dimensional representation of T Thinking of V as a quotient of a Verma module we can write any vector as a sum over a finite index cn 1λ⊗ Y n0 × C× Qn=(n1,...,nr)∈I′ c ∗ 0 Y nr Qn=(n0,...,nr)∈I c with weight λ. The action of(1~b)X0 on v lowers the exponent of Y0, adds a factor of(1~b) and leaves the other Y ni ′n 1λ⊗ Y n1 U0(1~b).v= Define v(i) inductively by v(i)= limb→0 Ui(1~b).v(i− 1). For any vector v we have[v(r)]=[vλ]∈ PV . It follows that if G.[v] is closed then G.[v]= G.[vλ]. ∗ potentially infinite but we can correct this by replacing v with lims0 η(s).v for a generic η∈ VT. In this ′= lims0 η(s).v is supported in the weight spaces Vµ such that µ minimizes the function χχ, η. This function is a quadratic function on a lattice so there are only closed. Let Y = Gaf f .[vλ]⊂ PV . Suppose [w]∈ Y − Gaf f .[vλ]; we have Gaf f .[w]⊂ Y − Gaf f .[vλ] as Y − Gaf f .[v] is Gaf f -stable and closed. In particular, each[w(i)]∈ Y − Gaf f .[vλ] but this contradicts that[w(r)]=[vλ]. We have shown if there is a closed orbit it must be that of a HW vector. Let us show this is indeed finitely many µ that minimize it; see section 4 for more details. case η acts nontrivially on the vµ and v The same argument goes through for U res. All that changes for U 1 Y nr unchanged. It follows that is the sum (9) is now and U i r , r ∗ (9) (cid:3) Remark 10. This result only relies on basic properties of representations of LG also holds for LsmG acting on U pos; as U pos is a topological vector space the notion of closure is already well defined. 3.2. The Wonderful Embedding of the Loop Group. The main result of this subsection is theorem 3.10. We follow the same outline as in section 2.1. 14 PABLO SOLIS k Gaf f k X af f k k X af f k ad ∶= Gaf f× Gaf f .[id]⊂ P(V(λ)⊗ V(λ)∗) 3.2.1. Construction of X af f . Naively we would like to choose a regular dominant weight λ of Gaf f and define ; instead we work with V(λ)⊗V(λ)∗∶= however one immediately runs into the problem that id~∈ V(λ)⊗V(λ)∗ ∏µ,χ∈ΛT af f Vµ⊗V χ . We can write Gaf f=k(Gaf f)k where(Gaf f)k is an infinite dimensional scheme al- ∗ −1 is a closed embedding from (Gaf f)k into some ways containing Z(Gaf f); the inversion map g g (Gaf f)k′ and in this way (Gaf f)−1 fk-→ PV(λ)⊗V(λ)∗ given by(g, h) g.[id].h. Define (Gaf f)k×(Gaf f)−1 X af f∶= G~Z(G). ad = L Proposition 3.3. X af f is an ind scheme of infinite type. X af f contains an open sub ind-scheme isomorphic to Gaf f ∶= im fk=(Gaf f)k.[id].(Gaf f)−1 k ⊂ PV(λ)⊗V(λ)∗ inherits a scheme structure. We have a morphisms of schemes Remark 11. This construction immediately extends to Gaf f sm the construction is pos are Hilbert spaces and there is a good notion of tensor product for Proof. The first assertion is clear. For the second, it suffices to show im f =(Gaf f)k~Z(Gaf f). In fact k ([id])=(Gaf f)k× id Z(Gaf f)× Z(Gaf f)∆(Gaf f)= fk(Gaf f)k×id is surjective and (Gaf f)k× id∩ f −1 (Z(Gaf f), id); consequently fk((Gaf f)k× id) = (Gaf f)k~Z(Gaf f). The equality Gaf f G~Z(G) follows from proposition 3.1(e): Z(Gaf f)= C× c× Z(G). even simpler. Note V(λ)pos, V(λ)∗ Hilbert spaces. We still have id~∈ V(λ)pos⊗ V(λ)∗ pos but for a generic 1 parameter subgroup η∶ C×→ T and fort sufficiently small we have η(t).id∈ V(λ)pos⊗ V(λ)∗ pos. The stabilizer of[η(t)id] is conjugate to Z(Gaf f sm)∆(Gaf f sm)× Z(Gaf f sm). Set X af f,sm to be the closure of the orbit of[η(t)id] in P(V(λ)pos⊗ pos); it contains an orbit isomorphic to the adjoint of Gaf f V(λ)∗ projection to Vλ⊗ V ad we mean∪kT ~Z(G). By T ad,0≅ C[t =∪kX af f ∗ λ is not zero. X af f ad and X af f ad given by the character−αi. −αi be the regular function on T −αl]≅ Al+1. In particular T −α0 , . . . , t the negative Weyl alcove−Al0; the fan T 3.2.2. Properties of the Wonderful Embedding. In this subsection we prove theorem 3.10 which is the loop group analogue of theorem 2.2. Following section 2.1, we define X af f k,0 to be the open subset where is an open ind-subscheme of X af f . X af f is the adk where the subscript denotes closure in X af f wonderful embedding of Gaf f T in fact T Let t is the open cell of X af f . A maximal torus for Gaf f ad ad is given by the Weyl aclove decomposition of t ad,0 is smooth and its fan is given by ad,0∶= X af f R= Lie(T ad,0 is a constant ind-scheme. ad = L poly to give X af f poly. For Gaf f ad ∶= ∩ T ad; sm as an open orbit. Proposition 3.4. T ad)R. and T k is T 0 0 0 k,0 (10) (cid:3) T Proof. Using proposition 3.1(b) one can use the same proof as in lemma 2.7. The second statement follows because ad=∪w∈W af f wT −1 and−Al0 is a fundamental domain for the action of W af f on t the dual vector. Define Pv∶={v≠ 0}⊂ PV ∗∈ V Let V = V(λ) be a HWR and let v= vλ∈ V and v ∗ and Pv∗={v ∗ . In particular X0 isU×U− ∗∩ Pv∗=U.v Lemma 3.5. LG.v∩ Pv=U− ∗ . The proofs of lemma 3.5 and proposition 3.6 are adapted from [6, 6.1.7]; the extension for the loop group essentially requires replacing the Bruhat decomposition of a reductive group with the Birkhoff factorization of LG. ∗≠ 0}⊂ PV .v and LG.v ad,0w stable. (cid:3) R. . 0 . 15 (cid:3) ad,0 0 ∗ . acting on V equivariant isomorphism G or Gaf f in lemma 3.5. A WONDERFUL EMBEDDING OF THE LOOP GROUP then u.vµ still has nonzero projection to vµ Remark 12. Of course we can also replace LG with L ad is just the multiplication map and this is known to be open by the Birkhoff decomposition; consequently the morphism is birational. The next step is to Proof. Let π∶ V → C⋅v be the projection. Let πv(g)= π(g.v). We have πv(g)= 0 if and only if g.v∩Pv=࢝. By the Birkhoff decomposition any g= u−.w.b with u−∈U− , w∈ W af f and b∈B. It follows that πv(g)= 0 if and only if πv(u−.w)= 0. If v ′= vµ+∑ν<µ vν and u∈U− and all other weights are still less than µ. Consequently πv(u−.w)≠ 0 if and only if πv(w)≠ 0 if and only if w= 1. Therefore .v≅U− LG.v∩ Pv=U−B.v∩ Pv=U− The second equality follows because the stabilizer of[v] in PV isB. The same argument works with LG Proposition 3.6. There is anU×U− a-→X af f U−×U× T (l, u, t) l⋅ t⋅ u Proof. First note that the restriction toU−×U× T construct aU×U− 0 →U×U− then thinking of φ as an endomorphism defined away from ker φ we see that φ(v) is Let φ∈ X af f defined and is in Pv. On the other hand φ(v) is in the closure of the orbit of the HW vector, but this G.v. So by the previous lemma φ(v)= l.v for a unique l. We orbit is closed by lemma 3.2 hence φ(v)∈ L 0 →U− 0 →U. Altogether we have a map The compositionU−×U× T ad,0→ X af f 0 →U−×U is given by(l, t, u)(l, u). To finish we showU−×U−× b −1(1, 1) = T −1(1, 1) a-→ X af f −1(1, 1) is a subset of X af f it suffices to show that if a(l, t, u)= a(l , l= l ′ ′ ′ (l, u)= b○ a(l, t, u)= b(x)= b○ a(l ′). ′)=(l ′ ′ ′ Now surjectivity. Let φ∈ X af f and(l, u)= b(φ). Then t∶=(l −1(1, 1), hence(l, t, u) does the −1).φ∈ b −1, u ad,0. Clearly we have ⊃ as b −1(1, 1) = T −1 is closed and contains T a is birartional it follows that they have the same dimension. Now π0(LG) = π1(G) = 1. Further, the map G→ G~Z(G)=∶ Gad induces a map LG→ LGad; the image is the connected component of via φ l. Similarly we can get a map X af f ′)= x then u= u ad,0. For injectivity note , u . b-→U−×U -equivariant map X af f that as b This follows It remains to show b get a map X af f is bijective and b ad and as 0 0 ad,0 me by Sharwan Kumar ~U×U−≅ b are irreducible hence so is (cid:3) The following result is crucial for proving theorem 3.10(b); the argument given below was conveyed to the identity, in particular it is irreducible. It follows that X af f and X af f X af f −1(1, 1). Thus it must equal T Proposition 3.7. P ic(U−)= P ic(X af f )= 0 ad,0U is an infinite dimen- ad,0U). The second factor is 0 because T Proof. P ic(X af f )= P ic(U−)× P ic(T sional affine space. So it remains to prove P ic(U−)= 0. w=U−∩BwB~B. In factU−⊂ LG~B For any w∈ W af f we have a Schubert varietyBwB~B⊂ LG~B; setU− w. We show P ic(Uw)= 0 for all w. n whereU− and we get an ind-structure onU−=nU− n=l(w)≤nU− ff k---→O∗ Fix w and abbreviate Y =U− w. For any k∈ N we have a short exact sequence Z~k→O∗ Y)≅ P ic(Y) and looking at the long exact sequence in ´etale cohomology we get et(Y, Z~k)→ et(Y, Z~k)→ P ic(Y)→ P ic(Y)→ H 2 → H 1 et(Y,O∗ using that H 1 Y ; Y 0 0 X af f 0 , t , u , u , t ′ 0 0 job. 0 16 PABLO SOLIS By the proof of [20, 7.4.17], Y is contractible and because H singular(−, Z~k) it follows that ∗ the outer terms vanish and P ic(Y) LL⊗k----→ P ic(Y) is an isomorphism for any k. We now show P ic(Y) is finitely generated and together with the previous statement it will follow that P ic(Y)= 0. Y is a normal variety with dim Y = l(w) so by [13, 2.1.1] P ic(Y) embeds in the Chow group P ic(Y)⊂ Al(w)−1(Y). So reduce to showing Al(w)−1(Y) is finitely generated. By [13, 1.8] there is a surjection Al(w)−1(BwB~B)→ Al(w)−1(Y). By [13, 19.1.11b] Al(w)−1(Y)= H2(l(w)−1)(BwB~B, Z) and finally the et(−, Z~k)= H ∗ Bruhat decomposition implies the latter group is finitely generated. G× L (cid:3) J . ࣶ w3∈WJ G: − L in distinct L where WJ is the Weyl group of LJ . G stable and so breaks up into a disjoint union of Proposition 3.8. The boundary of X af f is L J where LJ,ad is the wonderful compactification of LJ,ad. We know present two technical results that are in the proof of the main theorem. Proposition 3.8 is well known in the finite dimensional case and the same is likely true for proposition 3.9 but I couldn't find a reference. G orbits. J be the associated opposite parabolics with standard Levi decomposition P ad,0≅ Ar+1 for I⊂{0, . . . , r} lie orbits. Proposition 3.8 implies the standard idempotents eI=∑i∈I ei∈ T G× L J = LJ⋅U For J⊂[0, r] let P ± ± ± Set S(J)={(g1, g2)∈ PJ× P J ∶ g1LJ= g2LJ}= ∆(LJ)(UJ× U J). Set T(J)∶= Z(LJ)× Z(LJ). Then − − ad,0 we have Stab(eJ)= T(J)⋅ S(J) and for eJ=∑j~∈J ej∈ T GT(J)⋅ S(J) G× L J⊂[0,r] Orbit(eJ)= ࣶ X af f= ࣶ J⊂[0,r] L J and Orbit(eJ) is an LJ,ad-bundle over Equivalently Orbit(eJ) is an LJ,ad-bundle over L G~P G~PJ× L − G~P G~PJ× L G× L Corollary 3.9. We can further decompose Orbit(eJ) underB−×B⊂ L Orbit(eJ)= w1.eJ .w3w2B B− ࣶ (w1,w2)∈W af f~WJ×WJ(cid:131)W af f Note when J =࢝ we get P G and WJ = W af f and the disjoint union becomes the usual J = LJ = L ± Birkhoff decomposition. When J={α1, . . . , αr} then the parabolics are the Borels and WJ = 1 and the Theorem 3.10. Let X af f= Gaf f~Z(Gaf f) be as in (10). Then (b) X af f− X af f is of pure codimension 1. It is a union of r+ 1 divisors that are Cartier and freely (c) The boundary X af f− L G~Z(G) consists of r+ 1 normal crossing divisors D0, . . . , Dr and the closure of the Gaf f× Gaf f -orbits are in bijective correspondence with subsets I⊂[0, r] in such a way that to I we associate∩i∈I Di. ′→ X af f determine and is determined by a weyl equivariant morphism of ′→ T n − X af f Proof. The proof of (a) is the same as 2.2(a). For (b) we use [25, 2.4] which states that the complement of a dense open subset in any scheme is of pure codimension one. For every n, X af f n,0 is a dense affine scheme in X af f hence X af f itself is pure codimension 1. n,0 gives a sequence of compatible divisors and shows that X af f− X af f Alternatively, by corollary 3.9, any p∈ X af f− L G~Z(G) can be expressed as u w1η(0)w2b where η∈ − −Al0 is a one parameter subgroup such that η(0)= eJ for some J. Clearly p∈U−w1η(s)w2B=U−w1w2B. vB ( closure in Gaf f) U− Ei∶=U−siB= ࣶ v≥si We give the proofs of these results at the end of this sub section. disjoint union becomes the stratification by Shubert cells of L G~B×B−(cid:131)L G. 0 generate the Picard group. (d) Any G equivariant X toric varieties T ad,0. (a) X af f is independent of λ. n 0 Using w1,w2 then there is a morphism of the Ei which is pure codimension 1. A WONDERFUL EMBEDDING OF THE LOOP GROUP 17 w1,w2 w1,w2 . we have lims→0 w1,w2 We extend σi on each X af f w1,w2 . Namely, for p in the boundary of X af f use 3.9 again to write Gaf f→ Gaf f~Pi→ PVi To show the Ei are Cartier we use that the Ei are Cartier. In fact there is a maximal parabolic subgroup ν1.vλ,p(s)ν2.vλ w1.vλ,p(s)w2.vλ . σi(p(s))=vi, uν1η(s)ν2b.vi =ν −1.vi, η(s)ν2.vi=ν −1 1 u that is, the projection of g.vi to the vi weight space. We show that this section extends to all of X af f . be the open subset where projection where≥ is the Bruhat order on W af f , see [20, 1.3.15]. Note that Ei are codimension 1 in L G~Z(G) and G~Z(G) is the union for some i we have p∈U−w1w2B⊂ Ei (closure in X af f ). It follows that X af f− L Pi= LiUi and a line bundleOi(1) on Gaf f~Pi such that Ei is the inverse image of the vanishing of a section σi on Gaf f~Pi. Moreover choosing a HWR Vi such thatPi stabilizes the class of a HW vector vi g gPi[g.vi] andOi(1) is the pull back ofO(1) on PVi and further the section σi can be described as σi(g)=vi, g.vi; Consider X af f⊂ PV(λ)⊗V(λ)∗ and for w1, w2∈ W af f let X af f to Vw1.λ⊗ Vw2.λ is nonzero. Then 3.9 shows X af f is covered by X af f p= lims→0 uν1η(s)ν2b= lims→0 p(s). By definition of X af f ν1.vλ,p(s)ν2.vλ w1.vλ,p(s)w2.vλ exists for any ν1, ν2∈ W af f ; indeed thinking of p= p(0) as an infinite matrix then the limit in question is the entry of p(0) in "row" ν1.vλ and "column" ν2.vλ. Similarly, replacing vλ with vi, we conclude the limit also exists. We set σi(p)= lims→0 It remains to show the extension doesn't depend on the choice p(s). The numerator reduces to and likewise for the denominator; altogether we see σi(p) is a matrix coefficient for η(0) where the limit is taken in PV(i)⊗V(i)∗. Varying η just scales σ so it just remains to show σi doesn't depend on the choice of ν1, ν2. Let Stabλ(η(0)) denote the stabilizer of η(0) for the action in PV(λ)⊗V(λ)∗ and Stabi(η(0)) the stabilizer for the action in PV(i)⊗V(i)∗. The required statement follows because Stabλ(η(0))⊂ Stabi(η(0)). For(c) we note that any Gaf f× Gaf f -orbit intersects T The same argument for 2.2(d) works for(d) with one minor adjustment. In the finite dimensional case of the same level. Representations(0, λ, l) of level l are characterized by λ(θ ∨)≤ l; here θ there is a fixed l such that(0, λi, l) are all regular dominant. for X af f,sm but the argument is different: it is just necessary to show P ic(X af f,sm )= 0. Given 3.6 and the fact that for LsmG we have an abstract isomorphismU≅U− , the previous assertion reduces to P ic(U)= 0. its group of units. Then P ic(U)= H 1(U,O∗) and LetO be the sheaf of holomorphic functions andO∗ vanishing for the latter group follows from a standard argument that asserts line bundles on U are topological together with the result [26, 8.7.4 (ii)] which asserts that U is contractible. To show line bundles are topological letOcts be the sheaf of continuous functions onU andO∗ Then we have a commutative diagram (suppressing the mention ofU) ad-orbit. This follows from the description of the stabilizers given in proposition 3.8 and the fact that the standard parabolic subgroups in a Tits system are not conjugate [17, 30.1]. Therefore we can reduce the statement to the torus for which it is obvious. we utilized regular dominant weights with no further qualification. For LG we need all the HWR to be is the co-root associated to the longest root θ of G. It follows for any finite set λ1, ..., λm of regular dominant weights (cid:3) Remark 13. The proofs for (a),(c),(d) hold for X af f,sm without further comment. Statement (b) is true Finally, thanks to proposition 3.7, the same argument in 2.2(b) shows these divisors freely generate 1 vi, η(s)ν2.vi −1 ad,0 along a unique T ad× T cts its group of units. the Picard group. ∨ 0 H 1(O) H 1(Octs) H 1(O∗) / H 1(O∗ cts) H 2(Z) / H 2(Z) / H 2(O) / H 2(Octs) / / / / / / / / G× L dominant weight λJ or LJ,ad. −1 1 , w ′ ′ . In this case from general j (cid:3) 18 PABLO SOLIS − ad,0 given in the proof J . Further the fibers are compactifications of Lad,J . To conclude orbit closures are given intersections of π wonderful compactification of LJ,ad. The group L G and in particular the standard parabolic subgroups are generated by the root a higher dimensional group that stabilizes eJ . As L G is connected, there can't be other components. The vanishing of H i(Octs) for i≥ 1 follows becauseU is paracompact and the vanishing of H i(O) for i≥ 1 is proved in [22, 1.1]. Altogether we have H 1(O∗)= H 1(O∗ cts) as required. proof of prop. 3.8. That T(J) is in the stabilizer follows from the description of T proposition 3.4 and the fact that Z(LJ)=∩k~∈J ker αk. So let us focus on the group S(J). subgroups U(α)≅ Ga which it contains. So to check S(J) is in the stabilzer it suffices to check it for the root subgroups it contains. These break up into two cases. Roots subgroups of the form(U(α), 1) or (1, U(α)) and those of the form ∆(U(α)). We treat the first case; the second case follows similarly for some i~∈ J. It suffices In the first case α is not a root of Lie(LJ) and we have that α= αi+ α to check that Xα.eJ = 0. Recall eJ is an idempotent of End(V(λ)) and we can express eJ =∑j ej⊗ e ∗ where j ranges over some subset of the weights of the representation. Therefore to show Xα.eJ = 0 it suffices to show Xα.ej= 0∀j. Assume that Xα.ej≠ 0 for some j. The weight j has the property that λ− j∈∑i∈J niαi with ni≥ 0. But if eµ∶= Xα.ej is not zero then it is a weight vector of weight µ= α+ j. But then λ− µ fails to be a sum of positive roots, contradiction. This shows that T(J)S(J) is contained in the stabilzer, and for codimensional reasons there can't be Finally, according to our conventions, Orbit(eJ′)⊂ Orbit(eJ) off J ⊂ J properties of flag varieties we have surjections LG~PJ′→ LG~PJ with fiber PJ~PJ′. So indeed we get a surjection Orbit(eJ) πJ-→ LG~PJ× LG~P that is it the wonderful compactification one can note eJ is the identity in End(V(λJ)) for a regular Alternatively, Let m= rk(LJ,ad) and β1, . . . , βm its simple roots. Then for p= πJ(eJ) we have π J (p) −1 is Lad,J× Lad,J -stable, contains 2m orbits, one open orbit isomorphic to Lad,J and the Lad,J× Lad,J J (p) isomorphic to the −1 proof of corollary 3.9. The given expression is stable under the action ofB−×B and disjointness of the expression is implied by the fact thatB−×B∩ W af f= 1. Any p∈ Orbit(eJ) can be expressed as g1.eJ .g2 for gi∈ L G. Recall the Levi factorization PJ= LJ UJ . A variant of equation (8) shows we can write g1 = v 1 w1l1u1 and g2 = v2l2w2u2 where (vi, wi, liui) ∈ − J× W af f× LJ UJ . For l3= l1l − Lw3BL where BL= T UL is a Using the Brikhoff decomposition for LJ we write l3= v3w3t − −1u another element of LJ we can assume(w1, w2)∈ W af f~WL× WL(cid:131)W af f . Altogether we are reduced to g1.eJ .g2= v1w1v3w3.eJ tu3.w2u2. The result will follow if we can show(w1v3w 2 u3w2)∈U−×U. For this it is enough to show that if −1 w∈ W af f and−αi is a negative root then w(−αi) is a positive root↔ there exists an reduced expression w= si1si. The⇐ direction follows from [6, pg.61]. ′(−αi) is a negative root. The Using induction on the length of w we can assume w= sjw ′ ′(−αi)=−αj and that si must appear somewhere in an expression for w. We ′′= wsi. If wsi is not reduced then by the exchange property for reflection groups ′′(−αi) is a positive root by the⇐ direction. G = . Let Vpr1 be the associated 1-dimensional representation. For any p∈ X af f we can find ′′(−αi)= w(αi)=−w(−αi) and w(−αi) was assumed to be positive, a contradiction. pr1--→ C× 3.3. Completeness and Stacky Extension. Let us now address the issue of completeness. First we show we have a non constant regular function pr1 on X af f . Namely consider the character L we can find a reduced expression for w ending in si. Therefore we can prove the result if we can rule out the possibility that w (cid:3) But w is reduced. If it is reduced then w 2 , proposition 3.8 implies g1.eJ .g2= v1w1l3.eJ .w2u2. −1 hypothesis on w imply w abuse notation below and use w to indicate a reduced expression for w. J (p)∩ Orbit(J∪ βi). It follows that π −1 Borel for LJ and w3 is the Weyl group WL of LJ . In fact, by proposition 3.8, after replacing l3 with U the expression 3 ∈ U −1 where w Consider the word w ′′ C× LG η∶ C×→ T and g1, g2∈ LG such that p= p(0)= lims→0 g1η(s)g2. For any v∈ Vpr1− 0 define A WONDERFUL EMBEDDING OF THE LOOP GROUP Then pr1 extends pr1 and we have a cartesian diagram pr1(p)= lim s→0 η(s).v v = lim s→0η(s), pr1 v g1η(s)g2.v = lim s→0 G~Z(G) C× pr1 L / X af f pr1 19 (11) The special fiber over 0∈ A1 is the boundary and we denote it ∂X af f ; there is a similar diagram for / A1 X af f poly, X af f,sm. This shows X af f is not compete but for ∂X af f poly we have Theorem 3.11. An arbitrary morphism extends to C[[s]] Spec C((s))→ ∂X af f poly poly is the union of r+ 1 components Orbit(J={j}) and any map from C((s)) lands in one of Proof. ∂X af f them. Proposition 3.8 for X af f poly gives a diagram Spec C((s)) Spec C[[s]] / Orbit(J) f f′ / LpolyG~PJ× LpolyG~P πJ − The composition πJ○ f maps Spec C((s)) into a finite dimensional projective variety and the valuative J criterion for properness induces the map f . In turn, properness of πJ implies the result. (cid:3) ′ Remark 14. We do not have an analogous statement to 3.11 for ∂X af f,sm because the smooth flag varieties are not complete. 0 and 3.3.1. Stacky Extension. As in the finite dimensional case we can construct a stacky extension. With theorem 3.10 in hand, one check easily that the constructions in sections 2.2,2.3 carry over to the loop group case. Theorem 3.12. is of pure codimension 1 and we have an exact sequence There is only one slightly subtle point already mentioned in the proof of theorem 3.10 is that when dealing with multiple dominant weights λ, µ they must be of the same level. In this way, to a regular G; there is analogously an open cell dominant weight λ we can associate an ind stackX af f containing L X af f (a) X af f is formally smooth and independent of λ. (b) X af f−X af f 0→ Zr→ P ic(X)→ hom(Z(β), C×)→ 0 where the subgroup Zr is generated by the irreducible components ofX−X0. (c) The boundaryX af f− L G consists of r+ 1 normal crossing divisors D0, . . . , Dr and the closure G-orbits are in bijective correspondence with subsets I⊂[0, r] in such a way that to I we associate∩i∈I Di. Any Gaf f× Gaf f equivariantX′→X af f determines and is determined by a fan supported in the (e) The stabilizer of every point p∈X af f is finite. (d) Let u0, . . . , ur be generators of the rays of the Weyl alcove and M be the monoid they generate. negative Weyl chamber whose lattice points lie in M . G× L of the L 0 / / / / 20 PABLO SOLIS For statement (e) it suffices to look at points inX af f Proof. Statements (a) - (d) follow readily from the corresponding statements for theorems 3.10 and 2.10. . By proposition 3.6 (e) holds because it does for (cid:3) every point in the toric stack T 0 0 ≅[Ar+1~Z(β)]. G with C× LpolyG. We can also replace L 4. Some toric geometry of LT [24, 3.6]. An interesting aspect of the representation of loop groups is the appearance of a new intermediate abelian group. More precisely, instead of asking for a maximal torus T ⊂ LG, we can ask for a maximal abelian subgroup H⊂ LG. The standard choice is H= LT but there are others choices, see for example Let C× LT denote the inverse image of L the group structure on C× LT depends only on the Killing form. More generally, given any positive definite form Q onLie(T), one can construct an analogous central extensions of L therefore talk about a wonderful embedding of C× LT . × N and the orbit of the identity under N is in the sense that there is a group N⊂ L T ≅ T T . For any of these centrally extended groups, one can construct representations analogous to HWR or PER of Gaf f and T = C× LT under the projection Gaf f → L in these embeddings. This is the only interesting part In this section, we will study the closure of T G. In fact, T and L closed. The main result of this section is theorem 4.1 which states that the closure of is an infinite type toric variety whose fan can be realized as the cone on a Voronoi diagram for Q; the latter is an object studied in combinatorics. 4.1. The group T of L ×VT . Let L T be a central extension of L T and let V be an irreducible representation T . T ; we get a "wonderful embedding" Let us begin by pinning down the group structure of T It turns out understand the closure of T subgroup T central extension by looking at the case when T sits inside a simple group G. In this setting we have c . This action is computed in the proof of [20, 13.1.7]. We identify characters with their derivatives so that if we have an element LT∶= LT× LT.[id]⊂ P(V⊗V ∗). in LT× LT.[id] it suffices to look at the orbit under the ×VT ⊂ L × VT or more precisely its central extension T ×VT . We can determine the structure on the a linear map VT → ΛT , η η, given by the Killing form. The affine Weyl group W af f = W VT c = C×× T× C× has a natural action on the characters of T af f ∶= T × C× taf f=(er, et, ec)∈ T af f then a character λ=(n, λ, h) evaluates to λ(taf f)= enr+λ(t)+hc. With this notation the action of η∈ VT ⊂ W af f is 2η, η , hη,, 0=n− λ(η)+ h η-→η.λ= λ+− λ(η)+ h er expïïn− λ(η)+ h 2η, η(cid:6)⋅ r+λ(t)+ hη, t(cid:6)ï expïï[n− λ(η)]⋅ r + λ(t) + h⋅[η, t+ 1 2η, ηr]ï 2η, η, λ+ hη,, h et 1 λ A WONDERFUL EMBEDDING OF THE LOOP GROUP 21 r 2 1 η1 c and u1 H1 1 1 u1 H1 1 1 η u2 H2 η2 1 u1 H1 η1 1 η u2 η1η2 ⋅ u2 H2 η2 1 = Thus we can get a group structure for every positive definite symmetric form Q via We are using the semi-direct product to compute the middle row above. Comparing we see We can use this to understand conjugation in the central extension, for the action of η can also be interpreted as -----→ T af f T af f η○(−)○η−1 er et−η,r c(η, t, r) et 1 er λ-→ C× expnr+ λ(t)− λ(η)r+ hc(η, t, r) exp[n− λ(η)]⋅ r + λ(t) + h⋅ c(η, t, r) c(η, t, r)=η, t+η, η ⋅ 1 (u2)H2 −1 c(η1, H2, ρ2) cQ(η, H, u)= Q(η, H)+ u To emphasize the dependence on Q, we let Q denote T One can define cQ even if Q is not positive definite. However when we look at representations of these ×VT = C×× T× VT× C× ×VT : topologically T The cocycle c(η, t, r) determines the group law on T denoting elements by(u, H, η, w)T , the group structure has the form: ⋅ η ⋅ η1⋅ = −1 1 (ρ2)H1H2 −1 = c(η1, H2, ρ2) ×VT with its group structure determined by Q. groups we want want the weight spaces of the non central C×⊂ Q to be bounded below and we also want ×VT . In [26], Segal describes a representation of C× LT on a vector space of the form V1⊗V2 where V1 is a countable dimensional representation of C×T×VT and V2 is an uncountable dimensional representation of(LT)0~T . This produces a positive energy representation of LT . In this Fix a positive definite symmetric form Q on VT ; we view this as a linear map VT → ΛT . Consider the group Q from the previous section. We describe a representation of Q on V1=⊕χ∈Q(VT)V1,χ. In fact following the wonderful compactification recipe we look at the Q× Q orbit of the identity in End(V1). ×VT~ZQ for a finite subgroup ZQ⊂ T defined below. The embedding has Qad∶= Q× Q(ZQC× ~ZQ. The group ZQ is ker hom(Q, C×). In more words, we have an c× ZQC× isomorphism T≅ hom(ΛT , C×) and therefore we get a map and ZQ∶= ker hom(Q, C×). Remark 15. If T is a maximal torus of a semi simple lie group G and Q is the Killing form then ZQ= Z(G). as a dense open subset set. The connected components are indexed by VT and the connected component of the identity is the closure of T T≅ hom(ΛT , C×) hom(Q,C×) ------→ hom(VT , C×)≅(C×)dim T subsection we focus on V1 because this is all that is necessary to understand the closure the weight spaces to be finite dimensional; these conditions fail if Q is not positive definite. c)∆( Q)= T T⊂ LT 4.1.1. Representations of T ×VT ZQ u1u2 η1η2 Q(η,η) 2 . The embedding is of T η.vµ= vµ+Q(η) t.vµ= µ(t)vµ u.vµ= uQ(µ,µ)~2.vµ (12) 22 Let us return to the representation V1. We can describe it as follows. Let vµ∈ V1 be a weight vector PABLO SOLIS with weight µ, then: is given by a positive integer then = C×× T Remark 16. One can check that this indeed defines a representation; a verification of this appears in [26, pg.306]. An important property we can see from (12) is that if η ∈ VT is such that the composition pr1--→ C× C× η-→ T as an element of PV1⊗ V 1. To see this it is enough to check the function µ→µ, η from the weights of ∗ µ→µ, η is quadratic with leading order coefficient positive. Further the minimum is achieved at a finite → V1 and the toric variety we V1=∏µ∈ΛT(V1)µ. With this translation we have a representation of T V to Z has a well defined minimum value. This follows because the last line of (12) shows the function 1 as diagonal matrices. Such matrices we can identify with lims→0η(s) exists We can identify the image of T number of weight spaces µ. ∗ in V1⊗V Q∶= T.∏µ 1(cid:6)∈ P V1. are interested in is T . is v, v well studied and have many nice properties; see [29]. Let W be a real vector space equipped with an inner product Q. For v, v We can replace S with a regular lattice and apply the same procedure to get tiling of W . For example if 4.2. Voronoi and Delaunay Subdivisions and the fan. The goal of this subsection is to describe the fan for the toric variety T Q defined in the previous subsection. We begin by introducing the relevant combinatorics. ′)1~2 , v− v ′Q∶=Q(v− v ′ ′∈ S− s} ′Q∀s w− sQ≤w− s ′∈ W the distance between ′ v− v Given a finite set of points S⊂ W one can form the Voronoi diagram or Voronoi partition associated to S. This is partition W=∪s∈SCs where Cs={w∈ W W = R2 and Q= id and we consider the lattice Z2 then we get a tiling by squares centered around the points of Z2; the vertices of the squares are at the points(1~2, 1~2)+ Z2. For general Q these tilings are Dual to Voronoi diagrams is the notion of Delaunay subdivision. Let L⊂ W be a lattice. For r> 0 and p∈ W set B(p, r)={w∈ W ∶ w− pQ≤ r}. For any p∈ W set r(p)= min{t∈ R≥0 B(p, t)∩ L≠࢝} P(p)= Convex Hull(L∩ B(p, r(p)). Then we get a subdivision of W via W=∪p∈W P(p). Alternatively, the convex hull of p1, ..., pn∈ L is in the subdivision if and only if Cp1∩∩ Cpn≠࢝ Theorem 4.1. Let F be the fan of T Q. Then F is contained in VT,R⊕ R>0. For t> 0 identify VT,R with VT,R⊕ t; there is a lattice Lt⊂ VT,R such thatF is the cone on the Voronoi diagram associated to (Lt, VT,R, Q). Proof. Let (a, β)∈ VT = Z⊕ VT be a co-character. Recall Q acts on a vector space V1, (12). Fix a coordinate s on C× s be defined by replacing≤ with< in the definition of Cs. Then C 00={p∈ WP(p)= 0}. Now let us explain the connection between this combinatorics and T Q⊂ Qad. then we have a map Let C 0 and C× (a,β)---→ T ⊂ P V1 s M Q(η)∈Q(VT) s 2 ηQη+βQη a t 0 otherwise (cid:3) the auxiliary lattice the first claim. t A WONDERFUL EMBEDDING OF THE LOOP GROUP 23 2 ηQη+βQη). Set The limit s→ 0 only exists if a> 0. Moreover, if a> 0, then the limit exists for any value of β. This gives Fix now t∈ Z>0 and for every β∈ VT consider the 1 parameter subgroup diag(s t pβ= lim (s 2 ηQη+βQη)∈ P V1. s→0 M η∈VT For fixed t and β, consider η t 2 ηQη+ βQη as a function on VT ; it is quadratic and has a unique global minimum on VT,R. Consequently there are only finitely many co-characters η1, . . . , ηn∈ VT where 2 ηQη+ βQη attains its minimum, therefore f(η)=1 if η∈{η1, . . . , ηn} (f(η))∈ PV1 pβ= M η∈VT So in fact pβ∈ PV1. Fix any β∈ VT (with t still fixed as above). Then β determines a set ′∈ VTpβ= pβ′}. C(β)={β such that the cone on C(β) is a cone inF. We now relate this to the Voronoi subdivision∪β∈VT Cβ associated to(VT ⊂ VT,R, Q). For this we need ηQη+ βQη}. Lt={α∈ VT,Rα minimizes η t Basic properties of quadratic functions imply that Lt={β~tβ∈ VT} so Lt is indeed a lattice. We claim the C(β) are exactly the non empty intersections∩α1,...,αm∈VT Cα1∩∩ Cαm∩ Lt. Indeed, assume β~t, β ′~t is in the intersection. Then by properties above we have P(β~t)= P(β but this means the lattice points α1, . . . , αm are equidistant from β~t and no other α∈ VT are closer. This in turn implies that the function α t 2 αQα+ βQα is minimized on VT at exactly α1, . . . , αm. The same ′). One can run the statement in reverse to get the desired . In other words C(β)= C(β Remark 17. If we had worked with a the full representation V1⊗ V2 this would had the effect of adding 2 αQα+ βQα which doesn't effect where minimums occur. a positive constant to the function α t Remark 18. The fan of T in a HWR for LG is not the cone on a Voronoi subdivision associated to the More generally, for a fixed Q, the HWR of LT are indexed by the finite group ΛT~Q(VT). We computed the toric variety of T in the representation corresponding to 0∈ ΛT~Q(VT). For nonzero elements in ΛT~Q(VT) the fan of T is also a cone on a Voronoi diagram but the lattice is different. 4.2.1. Example. Consider the case T = C× then the form Q above is a positive integer; we take Q= 2. This choice comes from the killing form on C×⊂ SL2. Let us consider the Voronoi subdivision associated to Z in the normed vector space (R, Q). Clearly the set of points closest to n∈ Z consists of the line segment[n− 1~2, n+ 1~2]; in the notation of 4.2, Cn=[n− 1~2, n+ 1~2]. The fan of T ⊂ LC×~± 1 is shown in figure 3 where we have identified R with R⊕ 1. To differentiate T= C× in the semidirect product we write T= GL1. The associated toric variety T has a map to A1 extending the projection C× GL1 / T Killing form. This is because a HWR of LG is not irreducible as an LT representation. In a regular representation of Gaf f all the irreducible representations of L and fits into the following diagram 2 ′~t) ′ applies to β equality. T appear as a direct summand. from the C× C×× GL1 C× pr1--→ C× j∈Z P1 j / A1 0? / ? _ o o / _ o o 24 PABLO SOLIS Figure 3. Cone on the Voronoi subdivision. The lattice Z is shown in square dots and the lattice points of the Voronoi subdivision is shown in circle dots. The rays of the fan are shown in dashed dots. Note only a finite number of rays are drawn but every point in the upper half plane lies on a ray or in between two rays. curves due to Faltings. 5. Bundles on Curves In this section discuss the connection between LG and the moduli stack BunG(C) of principal G bundles on a curve C. We explain the connection between the embeddingX af f curves as well as a connection betweenX af f poly and a construction of a completion of BunG(C) on nodal 5.1. The double coset construction. Fix a smooth curve C~C and a point p∈ C. Let BunG(C), BunG,hol(C) 5.1.1. Holomorphic Version. Fix a local isomorphism z from{z> 1~2} in∞∈ P1(C) to a neighborhood ∗= C− Dp so that Dp∩ C∗= S1 identified with{z= 1}. Consider around p. Set Dp={z> 1} and C ∞(S1, G). Define now the smooth loop group LsmG= C C G= Hol(C ∗ ∗→ G. We have the following theorem denote the moduli stacks of algebraic and holomorphic principal G-bundles respectively. , G)⊂ LsmG, as the subgroup of loops that extend to give a holomorphic map C proved by Atiyah: poly and bundles on nodal Lsm Theorem 5.1. Let G be a connected topological group. The set of isomorphism classes of holomorphic principal G-bundles on C is equal to C G(cid:131)LsmG~Lsm,+ Lsm G ∗ Proof. [24, 8.11.5] Remark 19. One proves the result by noting a G-bundle on C out by the two subgroups amount to accounting for changes of trivialization. the bundle is trivial on Dp and so the bundle is determined by a "transition function" γ∈ LsmG; modding 5.1.2. Algebraic Version. In the algebraic setup the point a local coordinate z at p∈ C is used to identify C((z)) with f rac(OC,p) and thus identify LG with G(f rac(OC,p)). In this way homalg(C−p, G) embeds of algebraic maps(C− p)×C Spec R→ G. as a subgroup LCG of LG. More precisely, LCG, is the ind algebraic group which associates to a C-algebra R the group LCG(R)∶= G((C− p)R) is holomorphically trivial. Additionally The algebraic double coset construction is (cid:3) . (cid:3) Theorem 5.2. Let G be a semi simple group. There is a canonical isomorphism of stacks in [21]. ⊕rk(E) E≅ A over a Dedekind domain gives that [9]. This is the main ingredient in the proof of theorem 5.2. ⊕rk(E)−1⊕ det E= A Proof. This was proved for G= SLn by Beauville and Lazlo in [2] and for general G by Lazlo and Sorger The restriction to semi simple G allows us to conclude G-bundles are trivial over C− p. To see this for G= SLr note that we are considering locally free sheaves E with det E=OC. Over C− p, E is just a projective module over A=O(C− p) which is a Dedekind domain. A structure theorem for modules For general semi simple G choose a faithful irreducible representation G ⊂ SL(V) and identify a principal G-bundle with an SL(V)-bundle together with a reduction of the structure group to G. Then over C− p the SL(V) bundle is trivial and the reduction to G is preserved so the G bundle is trivial on C− p. More generally, for a family of curves, we have the following result due to Drinfeld and Simpson D→ S is an isomorphism. Set U= C− D. Then for any G-bundle F on C its restriction to U becomes S. If S is a scheme over Z[n ′ G~Z(G) which is the connected component (LGad)0 of loops in Gad. Consequently the interpretation is in terms of Gad-bundles on neighborhood is isomorphic to D= Spec C[[x, y]]xy; it has a unique closed point p=(x, y) and D− p= ∗= Spec(C((x))× C((y))). For any C-algebra R set DR= Spec R[[x, y]]xy and define D −1] where n is the order of π1(G(C)) then S Theorem 5.3. Let S be a scheme and C a smooth proper scheme over S with connected geometric fibers of pure dimension 1 and let G be a semisimple group. Let D be a subscheme of C such that the projection poly and bundles on nodal curves. Here we focus on a modular inter- poly. This is an embedding of the group L S. 5.2. The connection with X af f pretation of certain orbits in X af f To begin this discussion consider the formal neighborhood of a node on a fixed nodal curve C. The being locally of finite presentation over can be chosen to be ´etale over ′→ S with S trivial after a suitable faithfully flat base change S R similarly. curves. ∗ D ′ A WONDERFUL EMBEDDING OF THE LOOP GROUP 25 BunG(C)≅ LCG(cid:131)LG~L + G Define a stack BunG,τ D on C-algebras given by DR P↓ Set G Bun0 [z R× G ∗ where angled brackets denote groupoid and the superscript 0 denotes connected component of the trivial bundle. R→D , τ∶ PD∗ Gad,τ D(R)=G− bundles poly are parameterized by subsets J ⊂ ±] orbits of X af f [z ±]× G ±]= C× LpolyG. The G [z {0, . . . , r} (see proposition 3.8). In what follows we focus on J withJ= 1 and abbreviate Orbit({j}) by [z [z Oj; further define the subgroup Hj⊂ G ±]× G ±]=∶ G ±]×2 by Oj= G[z±]×2 [z stack by sheafifying the presheaf R G[z±]×2(R) Hj(R) Gad,τ D is represented by O0⊂ X af f ular, LGad stands for(LGad)0; we don't introduce notation for this to cut down on the notation. Set N= ker(G[[z]]→ G). Notice Gad(DR) (x,y)→(z,z) ------→ Gad(R)( N(R)× N(R)) is an isomorphism. This is because γ∈ Gad(DR) is equivalent to(γ1, γ2)∈ Gad[[x]]× Gad[[y]] such that γ1(0)= γ2(0). Gad,τ D(R) to Gad(R((z)))×Gad(R((z))) Gad(R)( N(R)× N(R)) Gad,τ D(R) we note for a faithfully flat extension R→ R For example, starting from(P, τ)∈ Bun0 ′ PR′ ≅ DR′× Gad; comparing with τR′ gives a γR′ ∈ Gad(R ′⊗R R ′′ = R ′ Proof. Any subgroup of LGad is to be interpreted as the connected component of the identity; in partic- Using faithfully flat descent one can construct inverse maps from Bun0 ′((x)))× Gad(R ′((y))). Set R Proposition 5.4. Bun0 . Consider Oj as a we get ; poly. Hj . . 26 PABLO SOLIS The other direction is similar. the two different pullbacks of PR′ to DR′′ are isomorphic hence differ by an element in Gad(DR′′) ≅ ′′)× N(R ′′)( N(R Gad(R ′′)) hence γR′ descends to a well defined coset[γR]∈ Gad(R((z)))×Gad(R((z))) Gad(R)( N(R)× N(R)) −) where ≅ O0. By proposition 3.8, H0= Z(G)× Z(G)G(N× N It remains to show Gad((z))×Gad((z)) Gad( N× N) N = ker(G[z]→ G). Consequently O0= Gad[z±]×Gad[z±] . From [20, pg.231] we get Gad((z))~Gad[[z]]≅ Gad(N×N−) ±]~Gad[z ±]~Gad[z]≅ Gad[z Gad[z −1] and the result follows. Fix a nodal curve C with a unique node x such that C− x is affine. Define Bun0 Gad,C−pC to be the ∗= C− x. A simple variant of the above stack parametrizing Gad-bundles on C with a trivialization on C proof gives (cid:3) . Gad,C∗C is represented by O0⊂ X af f poly. Proposition 5.5. Bun0 We can extend the above results to Oj for j≠ 0 by introducingG-torsors. 5.3. G-torsors. In general, given a curve C→ B and a sheaf of groupG on C we define aG-torsor to be a sheaf of setsF on C→ B together with a right action ofG such that (1) there is a finite flat cover {Ci→ C} such thatF(Ci)≠࢝ and (2) the action mapG×F→F×F is an isomorphism. Given a principal bundle F on C we can consider the sheaf of groupsGstd defined by U homalg(U, G) and the sheaf of sets U Sect(U, FU). In this way we associate to each principal G bundle aGstd-torsor. In fact there is a perfect dictionary betweenGstd-torsor and G-bundles. Let P⊂ G be a parabolic subgroup. Let L P G={γ∈ G[[z]]γ(0)∈ P}. Construct a sheaf of groupsGP + on Spec C[[z]]={(z),(0)} byGP({(z),(0)})= L P G andGP({0})= G((z)). Given a smooth curve C and + a point p we notice thatGstdC−p andGP agree over Spec C((z)) and thus define a sheaf of group which we also denoteGP . Clearly we can iterate over(xi)= x1, . . . , xm∈ C with parabolics(Pi)= P1, . . . , Pm. . ThenG(xi),(Pi) Call the resulting sheaf of groupsG(xi),(Pi) In the examples mentioned thus far all theG have the property thatG( Ox)⊂Gstd( Ox)= G[[z]] where Ox is the completion of the local ring with respect to the maximal ideal. This inclusion is the crucial G-bundles on C with reduction of structure group to Pi at xi. -torsors are exactly quasi parabolic bundles: P G. A is given by a nonzero G to one of the groups parahoric subgroup of L G that is conjugate in L We can construct non bundle torsors by considering parahoric subgroups; these generalize L property that allows a dictionary between torsors and principal bundles. This inclusion fails in general and consequently torsors are more general objects that bundles. G is any subgroupP⊂ L η(s)γη(s)−1 exists} η ={γ∈ L P G lim s→0 where η is a co-character such that the composition C× η-→ T = C×× T p1-→ C× integer. We use the notationPη to denote the quotientP η~C×=P η∩ LG and define parahoric subgroups of LG to be any subgroup conjugate to aPη. The groupsP Lη={γ∈P η lims→0 η(s)γη(s)−1= id}. η lims→0 η(s)γη(s)−1= γ} and Uη={γ∈P Given a set(Pi) of parahoric subgroups we can analogously construct a sheaf of groupsG(xi),(Pi) 5.4. G-torsors on nodal curves. Associate to j∈{0, . . . , r} the vertex ηj ∈ Al0 satisfying αj(ηj)> 0 and all other αi(ηj)= 0. SetPj=Pηj with Levi decomposition LjUj∶= LηUη; we call these the standard ∗= D−{(x, y)}. Define a sheaf of Now consider again D= Spec C[[x, y]]~xy={(x, y),(y),(x)} and D groupsGj byGj(D)= ∆(Lj)( Uj× Uj) andGjD∗=Gstd; the restriction mapGj(D)→Gj(D ∗) is given by ∆(Lj)(Uj× Uj)⊂Pj×Pj⊂ G((z))× G((z))≅ G((x))× G((y)). Similarly as in the smooth curve case we can extendGj to a sheaf groups on a nodal curve C with a unique node x with the property that GjC−x=Gstd. We can repeat this construction with Gad in place of G and take connected components of the identity throughout; call the resulting sheaf of groupsGj and consider its torsors. Such torsors, also known as quasi parahoric bundles, seemed to be first discussed by Teleman, see [27, sect. 9]. They have recently received more attention by Heinloth [16] and by Balaji and Seshadri [30]. η come with a natural Levi decomposition: maximal parahoric subgroups. ad. + 27 (13) , γ automorphism. poly. Bun0Gj F↓ C j R A WONDERFUL EMBEDDING OF THE LOOP GROUP Recall from proposition 3.8 that ∆(Lj)⊂ Hj; here Lj is a Levi factor of a maximal parahoric subgroup of LG and consequently Lj is a semi simple group. Let Zj = Z(Lj)~Z(G); for SLn these groups are trivial but not in general. An example for G= SO5 is given at the end of this section. x is replaced by a stacky point[pt~Zj]. Let C be the corse moduli space of C j; i.e. where x is replaced an ordinary non stacky point. Let C be the normalization. Define C j to be C with stacky points[pt~Zj] introduced at the pre images y, z of the node. We fix a map C j→ C j by identifying y, z via the trivial Define a stack Bun0Gj For Zj as above let C j be a curve as in before proposition 5.5 with the addendum that the nodal point R→Gj adC∗ , τ∶FC∗ R stacky node x such that C− x is affine and x≅[pt~Z j]. The stack Bun0Gj Proposition 5.6. Let Zj be the finite group described above. Let C j be a nodal curve with a unique orbit Oj⊂ X af f ad,C∗C j is represented by the Proof. If x had no stackyness then we can use the local model Spec Ox= Spec C[[y, z]]~yz and isomor- phisms of torsors over Spec Ox is by definition an element ofGj When working on the stacky curve we can lift the automorphisms of x=[pt~Zj] and so isomorphisms of torsors over Spec Ox is identified with Zj× Zj⋅ ∆(Lj~Z(G))( Uj× Uj). Further by [16], anyG-torsor can be trivialized on C− x so we are formally in the same situation as proposition 5.5 and we can use ad( Ox)= ∆(Lj~Z(G))( Uj× Uj). ad,C∗ C j on C-algebras by ad,C∗ C j(R)=Gj Bun0Gj ad− torsors LG× LG LGad× LGad polyG× L So it suffices to show the right hand side is isomorphic to Oj. From 3.8 we have Z(Lj)× Z(Lj)⋅ ∆(Lj)( Uj× Uj) Zj× Zj⋅ ∆(Lj~Z(G))( Uj× Uj)= ad,C∗ C j≅ Oj= Z(Lj)× Z(Lj)∆(Lj)(Uj× U j) − (u, ζ(t)) α0-→ uM i≥1 Z(Lj)= i≠j Choose a co-character ζ such that ζ(t)∈∩i≠j ker αi∩ T =∩i≠j,0 ker αi and αj(ζ(t))= t i=1 niαi with ni> 0. Then always possible if j≠ 0. Let θ be the longest root of G and write θ=∑r tniζ,αi= utnjζ,αj= utm with m≠ 0 and thus the co-character u ′(u)⊂ T(J) and for dimensional ′(u)×ζ ζ′-→(u, ζ(u)−1~m) satisfies ζ reasons we have equality. This means the class of(u, γ),(u polyG× L ′)∈ L ′ ′) where the γ's and h's differ by multiplication by an element of representative of the form(1, h),(1, h Z(Lj). Consequently, LpolyG× LpolyG Oj= Z(Lj)× Z(Lj)⋅ ∆(Lj)(Uj× U j) − ζ,α≠ 1; this is polyG in Oj has a unique the same argument to show and as in proposition 5.5 we establish that Oj is isomorphic to the right hand side of equation (13) (cid:3) Let C j be the normalization of C j as described before proposition 5.6 and let y, z be the preimages of x. And letPj be the parahoric subgroup of LG described at the start of subsection 5.4. Using the construction outlined after proposition 5.5 we associate to the data (y, z),(Pj,Pj) a sheaf of groups G(y,z),(Pj ,Pj) ad,C∗ C j can be identified with a quasi parahoric bundle F on C j, with the same Corollary 5.7. Bun0Gj trivialization τ and additionally a reduction of thePj×Pj torsor F( Oy)× F( Oz) toGj( Ox)⊂Pj×Pj. on C j and torsors forG(y,z),(Pj ,Pj) are quasi parahoric bundles. L polyG ker αi 28 PABLO SOLIS over C Extending to C ∗∪ y∪ z is handled similarly. ad,C∗ C j(R) we associate to this data a sheaf of groups Gj on C, a torsor Proof. Given an(F, τ)∈ Bun0Gj F and a section as in the statement of the proposition. Fix a local model for x∈ C where the point x corresponds to(y, z)⊂ Spec C[[y, z]]~yz where by abuse of notation we consider y, z as local coordinates near the points y, z∈ C. Then Gj=Gj on C− y− z and we define Gj( Oy) to be the image of the compositionGj( Ox)⊂ G((y))× G((z))→ G((y)) and similarly for Gj( Oz). To define F note that a trivialization of F over Spec Ox induces an element in P j = Gj( Oy) and comparing it with the trivialization τ defines 'transition function' ∈ G((y)) which we use to define F ∗∪ y; this is possible again by the uniformization of torsors over a smooth curve proved in [16]. To define the reduction note that an element e ∈ F( Ox) defines a point (e, e) ∈ F( Oy)×F( Oz) differs from e by an element of Gj( Ox), so we have a well defined section of ′ F( Oy)× F( Oz)~Gj( Ox). We show now given data on C we can get and element of Oj and composing with Oj→Bun0Gj we assume F has a trivialization gives the inverse to the above construction. Again extending to R→ R ad,C∗ C j over Spec( Oy× Oz) comparing with the trivialization τ we get elements(γ1, γ2)∈ G((y))× G((z)). We 2) and(γ 2) ′′ ′ ′′ ′′((z))). These elements are related by an isomorphism of F over Spec( Oy× Oz) ′′((y)))× G(R in G(R that preserves the section; such isomorphism come from multiplication by ∆(Lj)×( Uj× Uj). But there is also an additional factor of Z(Lj)× Z(Lj) that comes from lifting the automorphisms of the points y, z. Arguing as in proposition 5.5 we get an element of Oj(R) as desired. giving us two sets of elements(γ can extend in two different ways to R ′⊗R R ′′= R and another choice e 1 , γ (cid:3) ′ ′ ′ 1, γ over C ∗) on Bun0Gj Remark 20. The question of the modular interpretation of the remaining, higher codimensional orbits requires more care. There are some results in this direction when one restricts to the divisor O0. In this case higher codimensional orbits have been interpreted as torsion free sheaves for GLn and Spn. The other approach is to consider bundles on modifications of nodal curves as in [28, 18, 19]. In the analytic setting one can nevertheless fit together all the orbits of X af f into a complex analytic space that serves as a completion of bundles over nodal curves in families. The appropriate algebraic analogue is work in progress. poly. There is an action of G(C Even without a modular description of the other orbits, the previous proposition already tells us something interesting. Given that bundles on a nodal curve are equivalent to bundles on the normalization Remark 21. The space Bun0Gj one gets a space much closer to BunG(C). Indeed, the 0 and ad decorations can be removed by working ad,C∗ C j is a 'rigidified' moduli space and when one forgets the rigidifications withX af f ad,C∗ C j which changes the trivialization ∗ poly instead X af f . Quotienting by this action gives a moduli space which we can denote BunGj C j and in turn we can pass fromGj to G by working with certain equivariant G-bundles. The details of this construction together with a 'transition function'∈ G at the node, it is a natural first guess to try to complete BunG(C) We now give a couple of example of parahoric subgroups. We treat first the case G = SL2. Set O= C[[z]]. The standard parahoric subgroups of LG are P0= G(O)= G[[z]] P1∶=γ(z)= O z −1O O det γ(z)= 1 zO simply by compactifying G. However the previous proposition shows that this is not sufficient; namely, in families a principal bundle may develop parahoric structure at the node. Figure 4 is an illustrates this; compactifying G only tells you about the divisor D0. will be carried out in a follow up paper. The first corresponds to the co-character u(u, diag(1, 1))∈ T and the second corresponds to(u, diag(√u, 1√ u)). A WONDERFUL EMBEDDING OF THE LOOP GROUP 29 Figure 4. Schematic of BunG(C~B) over the base B. The general fiber is just BunG(Cb) for the smooth curve Cb. Over the special fiber of B we get a divisor with normal crossings and r+ 1 components. The divisor Di has a dense open subset, the complement of the squares, which corresponds to Orb({i}). In the case of SLr, all the maximal parahoric bundles are conjugate to G[[z]] by outer automorphisms which can either be interpreted as 'fractional' loops in G or as honest loops in GLr . For example g1P1g 1 = g2P1g −1 2 =P0 with g1= 1 −1 −1 , 0 z g2= z1~2 −1~2 0 z 0 0 hand if we require s1 to have a pole at the closed point and s2 to be non vanishing at the closed point As a result allG-torsors withPi structure always have interpretations in terms of vector bundles. In the case at hand if E is the trivial rank 2 vector bundle over D= Spec C[[z]] and s1, s2 are two non vanishing sections then any other such sections can be obtained as γ.s1, γ.s2 with γ∈P0. On the other then any other such sections can be obtained as γ.s1, γ.s2 with γ ∈ P1. The latter case corresponds to the vector bundle OD⊕OD(1) over D. In other words, if we have a smooth curve C and a point p then G(C− p)(cid:131)LG~P0 is the moduli space of rank 2 vector bundles with trivial determinant while G(C− p)(cid:131)LG~P1 corresponds to those vector bundles with determinantOC(p). G[[z]]. It will be enough to work with the Lie algebra which we present as ai, b, c, gi, hi∈ C We finish with one final example of a parahoric subgroup of SO5 which shows, unlike the SLr case, the parahoric structures that appear cannot always be interpreted as twists of the standard structure so5= the Lie algebra of a maximal torus is given by diagonal matrices and more generally we can pick out the root spaces a1 a3 0 −c g1 0 0 h2 h1 b 0 a2 a4 c 0 g2 −h1 −h2 −b −a1 −a3 −g1 −a2 −a4 −g2 (X1, Y1)↔(a2, a3) 2h2,−√ (X2, Y2)↔(√ 2g2) (X3, Y3)↔(√ 2h1,−√ 2g1) (X4, Y4)↔(b, c) 30 For example X1= Y1= 1 −1 PABLO SOLIS 1 −1 hyperplanes Nomenclature Y4⊗ z −H4 y= x, y= x~2, y= 1~2 X4⊗ z H4 −1 sl2× sl2≅ because of the different length roots in so5. × H1 X1 Y1 −H1 α0 Affine root of Gaf f , page 11 α1, . . . , αr Roots of a semi simple group, page 5 β Dual of the map β in a stacky fan, page 8 In fact the outter automorphisms of LG come from automorphisms of the corresponding affine Dynkin diagram. For The maximal parahoric subgroups are given by the vertices of the Weyl alcove ⊂ t. Identify t = C2 using the basis H1 =[X1, Y1] and H2 =[X2, Y2]. Then the Weyl alcove is a triangle with supporting The vertices are v0=(0, 0), v1=(1, 1~2) and v2=(1, 1~2) and they give rise to three maximal parahorics P0,P1,P2 . We have P0 = G[[z]] and P0 ≅ P1 via an outer automorphism of LSO(5). LSO(5) the diagram looks like 0 2 1 and the automorphism in question switches 0, 1. The parahoricP2 has a Levi factor isomorphic to SL2× SL2 the Lie algebra of this levi is given by In this case the groups Z2 referred to in proposition 5.6 is Z~2. The non isomorphic Levi subgroups occur (Σ, β∶ L→ N) Stacky fan., page 8 [Y(Σ)~Z(β)] Toric stack associated to a stacky fan., page 8 ∗ Oj Orbit in X af f associated to the subset{j}, page 25 G sheaf of groups on a family of curves, page 26 G( Ox) completed stalk of a sheaf of groups, page 26 Gstd Sheaf of groups associated to the constant group scheme., page 26 ad = L G~Z(G) Adjoint group of Gaf f , page 14 µ, η Paring between characters and co-characters, page 5 LpolyG Polynomial loop group, page 12 LT Embedding of the L R R Lie algebra of T t Tad Closure of Tad in X, page 5 T ad Closure of T Q toric variety inside an embedding for the group Q, page 22 T Tad,0 Affine toric variety, open cell of Tad, page 5 ad, page 14 ad,0 Affine toric variety, open cell of T T T central extension of C× LT , page 20 T af f maximal torus of Gaf f , page 11 Tad Maximal torus for Gad, page 5 L ( LT~T)0 subgroup of L Q The group T Qad adjoint group of Q, page 21 ×VT with group structure determined by Q, page 21 Gaf f Affine Kac-Moody group., page 11 Gaf f Gaf f L poly Polynomial affine Kac-Moody group associated to G, page 12 G Semi direct loop group, page 11 , page 14 T , page 20 ad in X af f , page 14 T , page 20 A WONDERFUL EMBEDDING OF THE LOOP GROUP 31 Maximal torus of L ad , page 14 G, page 11 G~Z(G), page 14 ×VT subgroup of L ad Maximal torus for Gaf f T T T , page 20 T W af f Affine Weyl group, page 12 X af f Wonderful embedding of L X af f Open cell of X af f , page 14 0 X af f poly Polynomial embedding, page 14 X af f,sm Smooth embedding, page 14 Al0 Positive Weyl alcove, page 12 C∆ Cone depending on the fundamental weights of a semisimple group, page 9 Gad Adjoint group of G, page 5 Hj Subgroup associated to the orbit Oj, page 25 L L P G Parahoric subgroup of LG associated to a parabolic of G, page 26 G,L + + P(p) convex hull of a finite number of points in a vector space., page 22 Y(C∆) Quasi-projective G×(C×)r-embedding, page 9 Z(β) Subgroup of a torus associated to a stacky fan., page 8 Z(G) Center of a group G, page 5 T , B,B u1, . . . , ur Ray generators for the cone given by the Weyl chamber., page 9 X Wonderful compactification of Gad, page 5 X0 Open cell of X, page 5 Subgroups of a semisimple group, page 5 Subgroups of the loop group, page 11 ,U,U− G,B,B− − − − ,U ,U ZQ Subgroup of torus assoicated to a quadratic form Q., page 21 HW highest weight, page 5 HWR Highest weight representation, page 5 PER positive energy representation, page 13 References [1] Arnaud Beauville. Conformal blocks, fusion rules and the Verlinde formula. In Proceedings of the Hirzebruch 65 Conference on Algebraic Geometry (Ramat Gan, 1993), volume 9 of Israel Math. Conf. Proc., pages 75–96, Ramat Gan, 1996. Bar-Ilan Univ. [2] Arnaud Beauville and Yves Laszlo. Conformal blocks and generalized theta functions. Comm. Math. Phys., 164(2):385– 419, 1994. [3] Usha N. Bhosle. Principal G-bundles on nodal curves. Proc. Indian Acad. Sci. Math. Sci., 111(3):271–291, 2001. [4] Armand Borel. Linear algebraic groups. Second edition. Graduate Texts in Mathematics, 126. Springer-Verlag, New York, NY, 1991. [5] Alexander Braverman and Michael Finkelberg. Pursuing the double affine Grassmannian. I. Transversal slices via instantons on Ak-singularities. Duke Math. J., 152(2):175–206, 2010. [6] Brion and Kumar. The Frobenius Splitting Method in Algebraic Geometry, Progress in Mathematics, 231. Birkhaeuser, Boston, MA, 2005. [7] Michel Brion. The total coordinate ring of a wonderful variety. J. Algebra, 313(1):61–99, 2007. [8] C. De Concini and C. Procesi. Complete symmetric varieties. In Invariant theory (Montecatini, 1982), volume 996 of [9] V. G. Drinfel′d and Carlos Simpson. B-structures on G-bundles and local triviality. Math. Res. Lett., 2(6):823–829, Lecture Notes in Math., pages 1–44. Springer, Berlin, 1983. 1995. [10] Gerd Faltings. A proof for the Verlinde formula. J. Algebraic Geom., 3(2):347–374, 1994. [11] Gerd Faltings. Moduli-stacks for bundles on semistable curves. Math. Ann., 304(3):489–515, 1996. [12] Jurgen Fuchs. Affine Lie algebras and quantum groups. Cambridge Monographs on Mathematical Physics. Cambridge University Press, Cambridge, 1992. An introduction, with applications in conformal field theory. [13] William Fulton. Intersection theory, volume 2 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics]. Springer-Verlag, Berlin, second edition, 1998. [14] A. Geraschenko and M. Sattriano. Toric stacks i: The theory of stacky fans. arxiv 1107.1906. [15] A. Geraschenko and M. Sattriano. Torus quotients as global quotients by finite groups. arxiv 1201.4807. [16] Jochen Heinloth. Uniformization ofG-bundles. Math. Ann., 347(3):499–528, 2010. [17] J. Humphreys. Linear Algebraic Groups. Graduate Texts in Mathematics, No. 21. Springer-Verlag, New York- Heidelberg, 1975. 32 PABLO SOLIS [18] Ivan Kausz. A modular compactification of the general linear group. Doc. Math., 5:553–594 (electronic), 2000. [19] Ivan Kausz. A Gieseker type degeneration of moduli stacks of vector bundles on curves. Trans. Amer. Math. Soc., 357(12):4897–4955 (electronic), 2005. [20] Kumar. Kac-moody groups, their flag varieties & representation theory. 2002. [21] Yves Laszlo and Christoph Sorger. The line bundles on the moduli of parabolic G-bundles over curves and their sections. Ann. Sci. ´Ecole Norm. Sup. (4), 30(4):499–525, 1997. [22] L´aszl´o Lempert. Vanishing cohomology for holomorphic vector bundles in a Banach setting. Asian J. Math., 8(1):65–85, 2004. [23] Johan Martens and Michael Thaddeus. Compactifications of reductive groups as moduli stacks of bundles. arxiv 1105.4830. [24] Presseley and Segal. Loop Groups. Oxford Science Publications. The Clarendon Press, Oxford University Press, New York, NY, 1986. [25] Mike Roth and Ravi Vakil. The affine stratification number and the moduli space of curves. In Algebraic structures and moduli spaces, volume 38 of CRM Proc. Lecture Notes, pages 213–227. Amer. Math. Soc., Providence, RI, 2004. [26] Graeme Segal. Unitary representations of some infinite-dimensional groups. Comm. Math. Phys., 80(3):301–342, 1981. [27] Constantin Teleman. The quantization conjecture revisited. Ann. of Math. (2), 152(1):1–43, 2000. [28] AJ Tolland, E Frenkel, and C Teleman. Gromov-witten gauge theory i. arxiv 0904.4834. [29] Delaunay Triangulations. Chapter 8 dirichlet - voronoi diagrams and. http://www.cis.upenn.edu/~cis610/convex8. [30] C.S. Seshadri V. Balaji. Moduli of parahoricG–torsors on a compact riemann surface. arxiv 1009.3485. pdf. [31] E. B. Vinberg. On reductive algebraic semigroups. In Lie groups and Lie algebras: E. B. Dynkin's Seminar, volume 169 of Amer. Math. Soc. Transl. Ser. 2, pages 145–182. Amer. Math. Soc., Providence, RI, 1995. Department of Mathematics, University of California, Berkeley, CA E-mail address: [email protected]
1312.0239	1	1312	2013-12-01T15:51:16	The augmented base locus of real divisors over arbitrary fields	[ "math.AG" ]	We show that the augmented base locus coincides with the exceptional locus (i.e. null locus) for any nef $\mathbb{R}$-Cartier divisor on any scheme projective over a field (of any characteristic). Next we prove a semi-ampleness criterion in terms of the augmented base locus generalizing a result of Keel. We also study nef divisors with positive top intersection number, and discuss some problems related to augmented base loci of log divisors.	math.AG	math	The augmented base locus of real divisors over arbitrary fields Caucher Birkar Abstract. We show that the augmented base locus coincides with the ex- ceptional locus (i.e. null locus) for any nef R-Cartier divisor on any scheme pro jective over a field (of any characteristic). Next we prove a semi-ampleness criterion in terms of the augmented base locus generalizing a result of Keel. We also study nef divisors with positive top intersection number, and discuss some problems related to augmented base loci of log divisors. 1. Introduction The base locus of a linear system is a fundamental notion in algebraic and especially birational geometry. The restricted base locus (also called the non- nef locus) and the augmented base locus (also called the non-ample locus) are refinements of the base locus which capture more essential properties of divisors and linear systems. These are closely related to important concepts and problems in birational geometry, eg see [4][16][15][9][8][6][2]. We start with some definitions. The augmented base locus. Let X be a scheme. An R-Cartier divisor is an element of Div(X ) ⊗Z R where Div(X ) is the group of Cartier divisors. A Q-Cartier divisor is defined similarly by tensoring with Q. Definition 1.1 Let X be a pro jective scheme over a field k . The stable base locus of a Q-Cartier divisor L is defined as B(L) = \ m∈N, mL Cartier Bs mL that is, it is the set of those points x ∈ X such that every section of every mL vanishes where m is a positive integer and mL is Cartier. The base locus, stable base locus, and all the other base loci defined below are considered with the reduced induced structure. The augmented base locus of L is defined as B+ (L) = \ m∈N B(mL − A) where A is any ample Cartier divisor. Date : December 3, 2013. 2010 MSC: 14E30, 14A15. 1 2 Caucher Birkar The augmented base locus of R-Cartier divisors on smooth pro jective varieties was defined in [8]. For basic properties of the augmented base locus in this context see [9][8]. We give a different definition which is more convenient for our purposes (the two definitions agree, by Lemma 3.1 (3) below). Definition 1.2 Let X be a pro jective scheme over a field k . Let L be an R-Cartier divisor on X . We can write L ∼R P tiAi where A1 , . . . , Ar are very ample Cartier divisors and ti ∈ R. The Ai are not necessarily distinct and the expression is obviously not unique. Define hmLi = P ⌊mti ⌋ Ai which depends on the above expression. Next define the augmented base locus of L as B+ (L) = \ m∈N B(hmLi − A) where A is any ample Cartier divisor. It turns out that B+ (L) does not depend on the choice of A nor the Ai nor the expression L ∼R P tiAi (see Lemma 3.1). Relation with the exceptional locus. Let X be a pro jective scheme over a field k and L an R-Cartier divisor on X . The exceptional locus of L (also called the null locus when L is nef ) is defined as E(L) := [ LV not big V where the union runs over the integral subschemes V ⊆ X with positive dimen- sion. Now we come to the first result of this paper. Theorem 1.3. Let X be a projective scheme over a field k . Assume that L is a nef R-Cartier divisor with a given expression L ∼R P tiAi as in 1.2, and A is a very ample Cartier divisor on X . Then B+ (L) = B(hnLi − A) = BshnLi − A = E(L) for any sufficiently divisible n ∈ N. The theorem was first proved for X smooth, Q-Cartier L, and k algebraically closed of characteristic zero by Nakamaye [16] using Kodaira type vanishing theorems, and this was generalized to R-Cartier divisors by Ein-Lazarsfeld- Mustat¸a-Nakamaye-Popa [8]. Nakamaye’s result was extended to log canonical varieties by Cacciola-Felice Lopez [5] again by using Kodaira type vanishing theorems. They also give some applications to the moduli spaces of curves. Related results concerning the restricted volume are proved on normal varieties by Boucksom-Cacciola-Felice Lopez [3]. The theorem was proved by Cascini-McKernan-Mustat¸a [6] when k is alge- braically closed of positive characteristic using techniques related to Keel [12]: the main ingredients are Serre vanishing and the Frobenius. Fujino-Tanaka [10] employ similar arguments on surfaces using Fujita vanishing and the Frobenius. The augmented base locus of real divisors over arbitrary fields 3 We will also use Fujita vanishing (but not the Frobenius). A semi-ampleness criterion. The following semi-ampleness result was first proved by Keel [12] when k has positive characteristic. A simplified proof of Keel’s result was given by Cascini-McKernan-Mustat¸a [6]. Theorem 1.4. Let X be a projective scheme over a field k . Assume that L is a nef Q-Cartier divisor on X . Then there is a closed subscheme Z ⊆ X such that • the reduced induced scheme associated to Z is equal to E(L), and • L is semi-ample if and only if LZ is semi-ample. When k has positive characteristic we can use the Frobenius to show that in fact we can take Z = E(L). However, when k has characteristic zero in general we cannot take Z = E(L), by Keel [12, §3]. Although Z is not unique but some choice can be calculated for any given X, L. It is interesting to see whether the theorem holds if L is only R-Cartier. Divisors with positive top intersection number. Assume that X is a normal pro jective variety of dimension d over an algebraically closed field k , and L a nef R-Cartier divisor with Ld > 0. If L is Q-Cartier, then it is a well- known fact that L is big. When L is only R-Cartier, Shokurov [17, Lemma 6.17] shows that L is big if k has characteristic zero, using resolution of singularities and Kawamata-Viehweg vanishing (although he only proves that L ∼Q M ≥ 0 but his proof can be extended to show that L is big). Here big means that L ∼R A + D where A is an ample R-Cartier divisor and D is an effective R- Cartier divisor. In [12, Remark 5.5.3], Keel asks whether the same holds in positive characteristic. We show that indeed it holds in a quite general setting. Theorem 1.5. Let X be a projective variety (i.e. projective integral scheme) of dimension d over a field k . Assume that L is a nef R-Cartier divisor on X . Then L is big if and only if Ld > 0. As Keel points out this result has interesting applications to the structure of extremal rays on varieties. In fact the characteristic zero case already plays a crucial role in the study of extremal rays on the moduli space of pointed ratio- nal curves (see Keel-McKernan [13, Lemma 2.2] and its many uses therein). The augmented base locus of log divisors. Let (X, B ) be a pro jective pair over an algebraically closed field k and A a nef and big R-divisor such that L = KX +B +A is nef. The locus B+ (L) is closely related to the geometry of X . In Section 6 we recall some results and pose some questions concerning such loci. I would like to thank Mircea Mustat¸a and Karl Acknowledgements. Schwede for discussions related to Section 6. This work was supported by a Leverhulme grant. 4 Caucher Birkar 2. Preliminaries 2.1. The operator h−i. Let L be an R-Cartier divisor with an expression L ∼R P tiAi as in 1.2. Let π : X ′ → X be a morphism such that A′ i := π ∗Ai is very ample for each i, eg π is a closed embedding or it is obtained by base change as in 2.7 below. Then we get the expression L′ := π ∗L ∼R P tiA′ i which we can use to define hmL′ i. It is clear that hmL′ i = π ∗ hmLi. For a coherent sheaf F on X , we often use the notation F hmLi instead of F (hmLi). 2.2. Growth of functions. Let h : Z → Z be a function. We say that the upper growth of h is like md (resp. at most like md ) if 0 < lim sup m→+∞ h(m) md < +∞ (resp. lim supm→+∞ h(m) md < +∞). 2.3. Divisors. Let X be a pro jective scheme over a field k . The group of Cartier divisors on X is denoted by Div(X ) (this group modulo linear equiv- alence is denoted Cl(X )). Recall that an R-Cartier divisor (resp. Q-Cartier divisor) is an element of Div(X ) ⊗Z R (resp. Div(X ) ⊗Z Q). Such a divisor can be represented as L = P liLi where li ∈ R (resp. li ∈ Q) and Li are Cartier divisors but this representation is not unique. Two R-Cartier divisors L, L′ are R-linearly equivalent (resp. Q-linearly equivalent ) if L − L′ = P aiNi where ai ∈ R (resp. ai ∈ Q) and Ni are Cartier divisors linearly equivalent to zero. We denote the equivalence by L ∼R L′ (resp. L ∼Q L′ ). An R-Cartier divisor L is: • nef if L · C ≥ 0 for every curve C ⊆ X (a curve is an integral closed subscheme of dimension one); • ample if L ∼R P liLi with li > 0 and Li ample Cartier divisors; • effective if L = P liLi with li ≥ 0 and Li effective Cartier divisors; • big if L ∼R A + D where A is an ample R-Cartier divisor and D is an effective R-Cartier divisor; • semi-ample if there is a pro jective morphism f : X → Y over k and an ample R-Cartier divisor H on Y such that L ∼R f ∗H . 2.4. Pairs. A pair (X, B ) over a field k consists of a normal quasi-pro jective variety over k and a Weil R-divisor B with coefficients in [0, 1] such that KX +B is R-Cartier where KX is the canonical divisor. The pair is klt if for every pro jective birational morphism f : Y → X from a normal variety the coefficients of BY are all < 1 where KY + BY = f ∗(KX + B ). 2.5. Fujita vanishing theorem. This is a generalization of Serre vanishing theorem. Let X be a pro jective scheme over a field k , A an ample Cartier divisor, and F a coherent sheaf on X . Then there is a number m0 such that hi (F (mA + L)) = 0 for any i > 0, m > m0 , and nef Cartier divisor L [11][14, Theorem 1.4.35]. The augmented base locus of real divisors over arbitrary fields 5 2.6. Restriction to a hyperplane section. Let X be a pro jective scheme over a field k , A an effective Cartier divisor, and F a coherent sheaf on X . Tensoring F with the exact sequence 0 → OX (−A) → OX → OA → 0 gives a sequence 0 → F (−A) → F → F ⊗ OA → 0 which is often not exact on the left. However, if A (considered as a closed subscheme) does not contain any of the finitely many associated points of F , then the latter sequence is also exact on the left. Now if A is a very ample Cartier divisor and if k is infinite, then perhaps after changing A up to linear equivalence we can make sure that A does not contain any associated point of F . To prove this we first reduce it to the case X = Pn k and OX (A) = OX (1). Then since k is infinite, there are infinitely many hyperplanes defined over k (hence infinitely many choices of A up to linear equivalence) so we can avoid the associated points of F . If k is not infinite we will do a base change to an infinite field to be able to use the above remarks. 2.7. Base loci and base change. Let X be a pro jective scheme over a field k and let L be a Cartier divisor on X . Recall that the base locus of L is defined as Bs L = {x ∈ X α vanishes at x for every α ∈ H 0(OX (L))} As pointed out earlier we consider Bs L (and other loci) with the reduced structure. Recall that B(L) = Tm∈N Bs mL. If n, n′ ∈ N, then each section α ∈ H 0(OX (nL)) gives a section α⊗n′ ∈ H 0(OX (n′nL)) hence Bs n′nL ⊆ Bs nL. In particular, B(L) = Bs mL for every sufficiently divisible m > 0. Assume that k ⊆ k ′ is a field extension and that X ′ and L′ are the scheme and Cartier divisor obtained by base change to k ′ . Let π : X ′ → X be the corresponding morphism. Since H 0(OX (L′ )) = H 0(OX (L)) ⊗k k ′ we can see that π−1 Bs L = Bs L′ . This in turn implies that π−1B(L) = B(L′ ). Now assume that L is R-Cartier with a given expression L ∼R P tiAi as in 1.2. As pointed out in 2.1, hmL′ i = π ∗ hmLi hence π−1B+(L) = B+(L′ ). With a little more work we can also see that π−1E(L) ⊇ E(L′ ). Indeed, let V ′ be a component of E(L′ ), let W be the closure of π(V ′ ), and let W ′ be the scheme obtained from W by base change. If LW is big then LW ∼R AW + DW where AW is ample and DW is effective. But then L′ W ′ ∼R A′ W ′ + D ′ W ′ where A′ W ′ is ample and D ′ W ′ is effective. Now V ′ * D ′ W ′ otherwise W ⊆ DW which V ′ + D ′ is not possible. So by restricting to V ′ we get L′ V ′ ∼R A′ V ′ which means that L′ V ′ is big, a contradiction. Therefore π−1E(L) ⊇ E(L′ ). So if in some situation we want to show that B+ (L) ⊆ E(L), then it is enough to show that B+(L′ ) ⊆ E(L′ ) because π is surjective. 6 Caucher Birkar 3. The augmented base locus is well-defined In this section, we show that the augmented base locus as defined in Definition 1.2 is well-defined. We also show that the definition agrees with 1.1 and the one in [8]. Lemma 3.1. Let X be a projective scheme over a field k and L an R-Cartier divisor with a given expression L ∼R P tiAi as in 1.2. Then B+ (L) as defined in 1.2 satisfies the fol lowing assertions: (1) B+ (L) does not depend on the choice of A nor the expression L ∼R P tiAi ; (2) for any positive rational number s we have B+(sL) = B+ (L); (3) B+ (L) = T B(L − H ) where H runs over al l ample R-Cartier divisors so that L − H is Q-Cartier; (4) if L is Q-Cartier then B+ (L) coincides with the one defined in 1.1. Proof. (1) First we show that B+ (L) is independent of the choice of A. Indeed let G be any other ample Cartier divisor. Assume x /∈ Tm∈N B(hmLi − A). Then x /∈ B(hmLi − A) = B(X ⌊mti ⌋ Ai − A) for some m > 0. Thus x /∈ B(P l ⌊mti ⌋ Ai − lA) for any sufficiently large l > 0. Since X ⌊lmti ⌋ Ai − X l ⌊mti ⌋ Ai is zero or ample, x /∈ B(P ⌊lmti ⌋ Ai − lA), and since lA is sufficiently ample, x /∈ B(P ⌊lmti ⌋ Ai − G) = B(hlmLi − G). This shows that B(hmLi − A) ⊇ \ \ B(hmLi − G) m∈N m∈N The opposite inclusion ⊆ can be proved similarly hence B+ (L) is independent of A. Now we show that B+ (L) is independent of the expression L ∼R P tiAi . Indeed assume that L ∼R P t′ iA′ i is another expression. Redefining the indexes we can assume that A′ i = Ai . Let A = P Ai + G with G ample. Assume that x /∈ Tm∈N B(P ⌊mti⌋ Ai − A). Then x /∈ B(P ⌊mti ⌋ Ai − A) for some m hence x /∈ B(P l ⌊mti⌋ Ai − lA) for any sufficiently large l > 0. Arguing as above we can show that x /∈ B(P ⌊lmti ⌋ Ai − A). Writing lmti = ⌊lmti ⌋ + ui and lmt′ i = ⌊lmt′ i ⌋ + u′ i , we see that (X ⌊lmt′ i ⌋ Ai − G) − (X ⌊lmti⌋ Ai − A) ∼R X uiAi − X u′ iAi + A − G is ample hence x /∈ B(P ⌊lmt′ i ⌋ Ai − G) so x /∈ Tm∈N B(P ⌊mt′ i ⌋ Ai − G). In other words, \ m∈N B(X ⌊mti ⌋ Ai − A) ⊇ \ m∈N B(X ⌊mt′ i ⌋ Ai − G) The augmented base locus of real divisors over arbitrary fields 7 The opposite inclusion ⊆ can be proved similarly bearing in mind that we are free to change A and G. (2) It is enough to treat the case when s ∈ N is sufficiently large. It is obvious that B+ (sL) ⊇ B+ (L). Assume that x /∈ B+(L). Then x /∈ B(hmLi − A) for some m hence x /∈ B(shmLi − sA). Arguing as in (1) we see that x /∈ B(hmsLi − A) which implies that x /∈ B+ (sL). That is, B+ (sL) ⊆ B+(L). (3) For each m > 0, B(hmLi − A) = B(mL − mHm ) = B(L − Hm ) for some ample R-Cartier divisor Hm . Thus B+ (L) ⊇ T B(L − H ). Conversely assume x /∈ T B(L − H ). Then x /∈ B(L − H ) for some H . Since L − H is assumed to be Q-Cartier, mL − mH is Cartier for some sufficiently divisible m > 0. Since mH is sufficiently ample, B(mL − mH ) ⊇ B(hmLi − A) hence x /∈ B(hmLi − A) which implies that x /∈ B+(L). (4) We can write L ∼Q P tiAi with all the ti rational numbers. Pick s ∈ N so that sL is Cartier, sti are all integers, and sL ∼ P stiAi . Then by (2) and (1) we have B+(L) = B+(sL) = \ m∈N But this is the same as B+ (L) in Definition 1.1. B(X ⌊msti ⌋ Ai−sA) = \ m∈N B(msL−sA) = \ m∈N B(mL−A) (cid:3) 4. Growth of cohomology The next lemma is similar to [6, Lemma 2.2]. Lemma 4.1. Let X be a scheme of dimension d projective over a field k . Assume that L is an R-Cartier divisor with a given expression L ∼R P tiAi as in 1.2. Let F be a coherent sheaf on X . Then the upper growth of h0 (F hmLi) is at most like md . Proof. By 2.7, we can extend k hence assume it is infinite. Let t be a positive integer such that ti ≤ t for every i. By 2.6, for each m > 0, we can change the Ai up to linear equivalence so that F hmLi = F (X ⌊mti ⌋ Ai ) ⊆ F (mt X Ai ) So by replacing L with t P Ai it is enough to assume that L is an effective very ample Cartier divisor and enough to show that the upper growth of h0(F (mL)) is at most like md . Once again by 2.6, we can change L up to linear equivalence so that the sequence 0 → F ((m − 1)L) → F (mL) → F (mL) ⊗ OL → 0 is exact. Now if m ≫ 0, by induction on dimension, the upper growth of h0(F (mL) ⊗ OL ) = h0(F (mL)) − h0 (F ((m − 1)L)) is at most like md−1 hence the upper growth of h0 (F (mL)) is at most like md . (cid:3) 8 Caucher Birkar Lemma 4.2. Let X be an integral scheme of dimension d projective over a field k . Assume that L is an R-Cartier divisor with a given expression L ∼R P tiAi as in 1.2. Then the fol lowing are equivalent: (1) the upper growth of h0(OX hmLi) is like md ; (2) for some coherent sheaf F , the upper growth of h0 (F hmLi) is like md ; (3) for any coherent sheaf F whose support is equal to X , the upper growth of h0 (F hmLi) is like md ; (4) L is big. Proof. (1) =⇒ (3): We can extend k to an infinite field hence we can use 2.6. Pick an effective sufficiently ample divisor A so that 0 → F → F (A) → F (A) ⊗ OA → 0 is exact. Applying Lemma 4.1 it is enough to show that the upper growth of h0(F (A)hmLi) is like md . Thus by replacing F with F (A) we can assume that F is generated by global sections. Each global section corresponds to a morphism OX → F . Since X is integral, the morphism is injective if and only if its image is not torsion. Therefore if α1 , . . . , αr form a basis of H 0(F ) and if φi : OX → F corresponds to αi , then φi is injective for at least one i otherwise F would be torsion which is not possible as the support of F is equal to X . Therefore h0(OX hmLi) ≤ h0(F hmLi) which implies that the upper growth of h0(F hmLi) is like md . (3) =⇒ (4): Take F = OX (−A) for some sufficiently ample divisor A. Then h0(F hmLi) ≫ 0 if m ≫ 0 hence in particular h0 (OX (P ⌊mti ⌋ Ai − A)) 6= 0 for some m ≫ 0. Thus since X is integral, P ⌊mti⌋ Ai − A ∼ D for some effective Cartier divisor D . Therefore L ∼R X(mti − ⌊mti ⌋)Ai + A + D which means that L is big. (4) =⇒ (1): By definition, L ∼R A + D where A is an ample R-Cartier divisor and D is an effective R-Cartier divisor. Replacing L with a large positive multiple and then replacing L with hLi allows us to assume that L is Cartier and hmLi = mL for each m > 0. By replacing A we can assume that L ∼Q A + D and that A is Q-Cartier. But then D is also Q-Cartier and h0 (OX hmLi) = h0(OX (mL)) ≥ h0 (OX (mA)) for any sufficiently divisible m > 0. Arguing as in the proof of Lemma 4.1 we can prove that the upper growth of h0(OX (mA)) is like md . (3) =⇒ (2): Obvious. (2) =⇒ (1): There is a filtration 0 = F0 ⊂ F1 ⊂ · · · ⊂ Fn = F of coherent sheaves such that for each 0 < j ≤ n, there exist a closed em- bedding f : S → X of an integral scheme S and an ideal sheaf J ⊂ OS such that Fj /Fj−1 ≃ f∗J (cf. The stacks pro ject [19], section on d´evissage of co- herent sheaves). Let j be the smallest number such that the upper growth of The augmented base locus of real divisors over arbitrary fields 9 h0(Fj hmLi) is like md . Let f : S → X and J be the corresponding embedding and ideal sheaf so that Fj /Fj−1 ≃ f∗J . Then from the exact sequence 0 → H 0(Fj−1hmLi) → H 0(Fj hmLi) → H 0(J hmLi) we deduce that the upper growth of h0(J hmLi) is like md . By Lemma 4.1, dim S = d hence S = X . But then the upper growth of h0(OX hmLi) is like md . (cid:3) Proposition 4.3. Let X be a projective scheme over a field k . Assume that L is a nef R-Cartier divisor with a given expression L ∼R P tiAi as in 1.2. Let F be a coherent sheaf on X , and let Y be its support and e the dimension of Y . Then (1) the upper growth of h0(F hmLi) is at most like me ; (2) the upper growth of hi (F hmLi) is at most like me−1 for any i > 0; (3) the upper growth of h0(F hmLi) is like me if and only if LZ is big for some component Z of Y with dim Z = e. Proof. We can assume that the theorem holds for coherent sheaves with support of dimension < e. We also may assume that the theorem holds for any closed subscheme of X other than X itself. (1) This follows from Lemma 4.1. (2) By extending k we can assume that it is infinite. Choose an effective sufficiently ample divisor A. For each m > 0 we can write hmLi + A = A′ + A′′ where A′ is ample Cartier and A′′ is sufficiently ample Cartier. Then by Fujita vanishing (2.5) we get hi (F (hmLi + A)) = 0 for every m > 0 and i > 0. By 2.6, we can choose A so that the sequence 0 → F → F (A) → F ⊗ OA (A) → 0 is exact. The dimension of the support of F ⊗ OA (A) is e − 1. Now using the exact sequence H i−1(F ⊗ OA (hmLi + A)) → H i(F (hmLi)) → H i(F (hmLi + A)) = 0 and induction on e we get the result. (3) As in the proof of Lemma 4.2, there is a filtration 0 = F0 ⊂ F1 ⊂ · · · ⊂ Fn = F of coherent sheaves such that for each 0 < j ≤ n, there exist a closed embedding f : S → X of an integral scheme S and an ideal sheaf J ⊂ OS such that Fj /Fj−1 ≃ f∗J . Assume that the upper growth of h0 (F hmLi) is like me . Let j be minimal with the property that the upper growth of h0(Fj hmLi) is like me . Let f : S → X and J be the corresponding embedding and ideal sheaf so that Fj /Fj−1 ≃ f∗J . Then the upper growth of h0 (J hmLi) is like me . So in particular J 6= 0 and since S is integral the support of J is equal to S . Moreover, since the upper growth of h0(J hmLi) is like me , Lemma 4.1 shows that dim S ≥ e. On the other hand, S is a subset of Y because F X \Y = 0 and because of the surjection 10 Caucher Birkar Fj → f∗J . Thus dim S ≤ e hence dim S = e. Now, by Lemma 4.2, LS is big and so we can take Z = S . Conversely, assume that there is a component Z of Y of dimension e such that LZ is big. In the filtration above, let j be the smallest number such that Z is a component of the support of Fj . Then Z is a subset of the support of the corresponding f∗J hence Z ⊆ S which in turn implies that Z = S because e = dim Z ≤ dim S ≤ e. It is then enough to show that the upper growth of h0(J hmLi) is like me because of the exact sequence 0 → H 0(Fj−1hmLi) → H 0(Fj hmLi) → H 0(J hmLi) → H 1(Fj−1hmLi) and the fact that the upper growth of h1(Fj−1hmLi) is at most like me−1 by (2). Now apply Lemma 4.2. (cid:3) 5. Proof of the theorems Proof. (of Theorem 1.3) By Noetherian induction we can assume that the the- orem already holds for any closed subscheme of X not equal to X . Step 1. We deal with the first equality in the theorem. By definition, B+(L) ⊆ B(hnLi − A) for any n > 0. Moreover, there are positive integers m1 , . . . , mr such that B+(L) = B(hm1Li − A) ∩ · · · ∩ B(hmrLi − A) If n = lmi for some positive integer l, then hnLi − lhmiLi is zero or ample hence B(hnLi − A) ⊆ B(lhmiLi − A) ⊆ B(lhmiLi − lA) = B(hmiLi − A) Therefore B+ (L) = B(hnLi − A) if each mi n. For the second equality: for any fixed n′ > 0 divisible by all the mi and any sufficiently divisible l > 0 we have B+(L) = B(hn′Li−A) = Bslhn′Li−lA ⊇ Bshln′Li−A ⊇ B(hln′Li−A) = B+(L) Now take n = ln′ . Step 2. The rest of the proof will be devoted to showing B+ (L) = E(L). It is obvious that B+(L) ⊇ E(L) so we will focus on the reverse inclusion. If LZ is not big for every component Z of X (with the reduced induced structure), then B+(L) ⊆ E(L) = X . So we may assume that there is a component Z such that LZ is big. Pick such a Z with maximal dimension, say e. Let Y be the union of the other components, again with the induced reduced structure. There are ideal sheaves I , J ⊂ OX such that the support of I is inside Z but the support of OX /I is inside Y , and the support of J is inside Y but the support of OX /J is inside Z (cf. [19], section on d´evissage of coherent sheaves). Let Y ′ , Z ′ be the closed subschemes defined by I , J respectively. On Z \ Y we have J = 0 and OZ ′ = OX . Thus the reduced scheme associated to Z ′ is nothing but Z . Similarly, one shows that the reduced scheme associated to Y ′ is Y . By construction, on Z \ Y we have OZ ′ = OX and I = OX , and on The augmented base locus of real divisors over arbitrary fields 11 Y \ Z we have I = 0. Step 3. We would like to find sections of OX (hnLi − A) which vanish on Y ′ but not on Z ′ . Let I → OZ ′ be the composition I ֒→ OX → OZ ′ and let K, L be its kernel and image respectively. Similarly, let L → OZ be the composition L ֒→ OZ ′ → OZ and let N , M be its kernel and image respectively. Then by Step 2 on Z \ Y we have L = OZ ′ and M = OZ , and on Y \ Z we have L = M = 0. Therefore the support of L, M, I are all equal to Z , and the support of K, N are subsets of Z . Now we have the exact sequences 0 → K(hnLi − A) → I (hnLi − A) → L(hnLi − A) → 0 and 0 → N (hnLi − A) → L(hnLi − A) → M(hnLi − A) → 0 By Proposition 4.3, the upper growth of h0 (M(hnLi − A)) is like ne but the upper growth of h1 (N (hnLi−A)) is at most like ne−1 . On the other hand, again by Proposition 4.3, the upper growth of h0(L(hnLi−A)) is like ne but the upper growth of h1 (K(hnLi − A)) is at most like ne−1 . Therefore for infinitely many n > 0 we can lift a nonzero section of M(hnLi − A) to a section of L(hnLi − A) and in turn to a section of I (hnLi − A). In other words, there is a section α ∈ H 0(I (hnLi − A)) whose restriction to Z is nonzero. Since I is the ideal sheaf of Y ′ , α vanishes on Y ′ when considered as a section of OX (hnLi − A) via the injection I (hnLi − A) → OX (hnLi − A). Step 4. From now on we consider α as a section of OX (hnLi − A). We can think of α as a morphism OX → OX (hnLi − A) such that if we tensor this with OZ then we obtain a nonzero morphism. Let α1 := α and let T1 be the kernel of α1 . Let α2 be the composition OX → OX (hnLi − A) → OX (2hnLi − 2A) → OX (h2nLi − 2A) where the first morphism is α1 , the second one is obtained by tensoring α1 with OX (hnLi − A), and the third one comes from the choice of an injective morphism OX → OX (h2nLi − 2hnLi) (which exists because h2nLi − 2hnLi is zero or very ample) and tensoring it with OX (2hnLi − 2A). Let T2 be the kernel of α2 . Obviously, T1 ⊆ T2 . Inductively we can define αi to be the composition OX → OX (h(i − 1)nLi−(i − 1)A) → OX (h(i − 1)nLi+hnLi−iA) → OX (hinLi−iA) where the first map is αi−1 , the second map is obtained by tensoring α1 with OX (h(i − 1)nLi − (i − 1)A), and third one is obtained from the choice of an injective morphism OX → OX (hinLi − h(i − 1)nLi − hnLi). Again it is obvious that Ti−1 ⊆ Ti . Step 5. By the Noetherian property, there is r such that Tr = Tr+1 = · · · . Since α1 restricted to Z is nonzero and since Z is integral, we can make sure that the restriction of each αi to Z is also nonzero. Indeed if U ⊂ Z is a small 12 Caucher Birkar nonempty open set, then the restriction to U of each map in the definition of αi is an isomorphism. Therefore each αi is nonzero hence Tr ( OX . Now tensor αr with OX (−hrnLi + rA) and let E be its image in OX . Then we get the exact sequence 0 → Tr (−hrnLi + rA) → OX (−hrnLi + rA) → E → 0 Let E be the closed subscheme defined by E , that is, E is the zero subscheme of αr . Note that E 6= X otherwise E = 0 hence αr = 0 which is not possible. We will argue that B(hmLi − A) ⊆ E if m > 0 is sufficiently divisible. If E = X topologically, then the claim is trivial. So we may assume that E 6= X topologically. By construction, αr does not vanish outside E hence B(hrnLi − rA) ⊆ E . If m = lrn, then B(hmLi−A) ⊆ B(lhrnLi−A) ⊆ B(lhrnLi−lA) = B(hrnLi−A) ⊆ B(hrnLi−rA) ⊆ E Step 6. We will assume that r ≫ 0. Consider the exact sequence 0 → Tr (hmLi−hrnLi+rA−aA) → OX (hmLi−hrnLi+rA−aA) → E (hmLi−aA) → 0 Since Tr does not depend on r ≫ 0, by Fujita vanishing, we may assume that H i(Tr (hmLi − hrnLi + rA − aA)) = 0 and H i(OX (hmLi − hrnLi + rA − aA)) = 0 for any i > 0, m > rn, and 0 ≤ a ≤ 1. Therefore H i(E (hmLi − aA)) = 0 if i > 0, m ≫ 0, and 0 ≤ a ≤ 1 (in this proof we only need to consider a = 1 but in the proof of Theorem 1.4 we need to take a = 0). On the other hand, we have the exact sequence 0 → E (hmLi − aA) → OX (hmLi − aA) → OE (hmLi − aA) → 0 from which we obtain the exact sequence H 0(OX (hmLi − aA)) → H 0(OE (hmLi − aA)) → H 1(E (hmLi − aA)) = 0 if m ≫ 0 and 0 ≤ a ≤ 1. Step 7. From the expression L ∼R P tiAi we obtain the expression LE ∼R P tiAi E . For each m > 0 we get hmLiE = hmLE i. Taking a = 1 in Step 6, recalling that B(hmLi − A) ⊆ E if m > 0 is sufficiently divisible, and using Step 1, we deduce that B+(L) = B(hmLi − A) = Bs hmLi − A = Bs hmLE i − AE = B(hmLE i − AE ) = B+ (LE ) for any sufficiently divisible m > 0. On the other hand, it is easy to see that E(LE ) ⊆ E(L). Indeed, let V be a component of E(LE ). Then (LE )V is not big so LV is not big hence V ⊆ E(L). Finally using the Noetherian induction and the above results we get E(L) ⊆ B+ (L) = B+ (LE ) = E(LE ) ⊆ E(L) The augmented base locus of real divisors over arbitrary fields 13 which in particular implies that B+ (L) = E(L). (cid:3) Proof. (of Theorem 1.4) We may assume that the theorem holds for closed subscheme of X other than X itself. Moreover, by replacing L with a multiple we can assume that it is Cartier. If E(L) = X , the theorem is trivial. So we assume this is not the case. Let E be the subscheme constructed in Step 5 of the proof of Theorem 1.3. We showed that if m ≫ 0, the map H 0(OX (mL)) → H 0(OE (mL)) is surjective (by taking a = 0). Moreover, we showed that E(LE ) = E(L). Since B(L) ⊂ B+(L) = E(L) ⊆ E we have B(L) = B(LE ). Thus L is semi-ample if and only if LE is semi- ample. Since the theorem already holds for E by assumptions, there is a closed subscheme Z of E whose reduction is E(LE ) and such that LE is semi-ample if and only if LZ is semi-ample. Now L is semi-ample if and only if LZ is semi-ample. (cid:3) Proof. (of Theorem 1.5) Assume that Ld > 0. Let A be a sufficiently ample effective Cartier divisor. Let L ∼R P tiAi be an expression as in 1.2. First we show that the upper growth of h0 (OX (hmLi + A)) is like md . By Fujita vanishing (2.5), hi (OX (hmLi + A)) = 0 for any i > 0 and m > 0. Thus X (OX (hmLi + A)) = h0(OX (hmLi + A)) On the other hand, by the asymptotic Riemann-Roch theorem the upper growth of X (OX (hmLi + A)) is like the upper growth of (hmLi + A)d . For each m > 0, there is an ample R-Cartier divisor Gm such that hmLi + A = mL + Gm hence (hmLi + A)d > mdLd which shows that the upper growth of (hmLi + A)d is like md . Therefore the upper growth of h0(OX (hmLi + A)) is like md . Now consider the exact sequence 0 → OX (hmLi − A) → OX (hmLi + A) → O2A (hmLi + A) → 0 By Lemma 4.1, the upper growth of h0(O2A (hmM i + A)) is at most like md−1 hence the upper growth of h0 (OX (hmLi − A)) is like md . Since X is integral, hmLi − A ∼ D for some effective Cartier divisor D . Therefore mL ∼R A′ + D for some ample R-Cartier divisor A′ hence L is big. Conversely assume that L is big. So by definition L ∼R A′ +D for some ample R-Cartier divisor A′ and effective R-Cartier divisor D . By replacing L, A′ , D appropriately we can assume that A′ is a general very ample Cartier divisor. Let S be an irreducible component of A′ with the induced reduced structure. Since dim S = d − 1 and since LS is big, by induction on dimension, (LS )d−1 > 0 hence Ld−1 · S > 0 which implies that Ld = Ld−1 · A′ + Ld−1 · D > 0. (cid:3) 14 Caucher Birkar 6. The augmented base locus of log divisors Assume that X is a normal pro jective variety of dimension d over an alge- braically closed field k , and that B , A ≥ 0 are R-divisors. Moreover, suppose A is nef and big and L = KX + B + A is nef. Theorem 6.1. Assume Ld = 0. Then B+ (L) = X is covered by rational curves C with L · C = 0. The theorem was proved by Cascini-Tanaka-Xu [7] and independently by McKernan, when k has positive characteristic. A short proof of this in any characteristic was given in [1]. Now if Ld > 0, what can we say about B+ (L)? For example, is it again covered by rational curves intersecting L trivially? We give a couple of examples to shed some light on this question. Example 6.2 Let E be an elliptic curve over an algebraically closed field k and let X = P(OE ⊕OE (1)). The surjection OE ⊕OE (1) → OE defines a section of the pro jection X → E whose image will be denoted by E again. Moreover, there is a birational contraction X → Z which contracts only E . Let B = E and A be the pullback of a sufficiently ample divisor on Z . Let L = KX + B + A. By construction, B+ (L) = E(L) = E which is not covered by rational curves but at least it is covered by curves intersecting L trivially. Example 6.3 There is a pro jective surface S over an algebraically closed field k so that there is a nef prime divisor M such that κ(M ) = 0, KS + 2M ∼ 0, and if M · C = 0 for any curve C then C = M (see Shokurov [18, Example 1.1] for such an example). Let X = P(OS ⊕ OS (1)). The surjection OS ⊕ OS (1) → OS defines a section of the pro jection π : X → S whose image will be denoted by S again. Moreover, there is a birational contraction X → Z which contracts only S to a point. Let B = S + 3π ∗M and let A be the pullback of a sufficiently ample divisor on Z . Let L = KX + B + A. Then B+ (L) = E(L) ⊆ S . Moreover, since LS = (KX + S + 3π ∗M + A)S ∼ KS + 3M ∼ M is not big, B+ (L) = E(L) = S . But there is no family of curves C covering S with the property L · C = 0. These examples show that we need to put some reasonably strong condition on X, B , A to be able to say something interesting about B+ (L). Question 6.4. Assume that (X, B ) is a projective klt pair over an algebraical ly closed field k and A a nef and big R-divisor. Assume that L = KX + B + A is nef and that Ld > 0. Is it true that B+(L) is covered by rational curves C with L · C = 0? Assume that k = C. Then L in the question is semi-ample by the base point free theorem hence it defines a contraction X → Y . Moreover, it is well-known that the fibres of X → Y are covered by rational curves. Note that B+(L) is nothing but the union of the fibres of X → Y . Now assume that k has characteristic p > 5 and dim X ≤ 3. One can show that L is again semi-ample (if dim X = 2, this holds for any p [20]). We sketch The augmented base locus of real divisors over arbitrary fields 15 the proof. Since A is nef and big, we can change the situation so that it is ample [1, Lemma 8.2]. Using boundedness of the length of extremal rays [12][1, 3.3] one can show that L = P ri (KX + Bi + Ai ) where ri > 0, P ri = 1, Bi , Ai are effective Q-divisors, Ai is ample, (X, Bi ) is klt, and KX + Bi + Ai is nef and big. Now applying [1, Theorem 1.4][21] each KX + Bi + Ai is semi-ample hence L is also semi-ample. Thus L defines a contraction X → Y . In particular, B+ (L) is covered by a family of curves intersecting L trivially. Using the results of [1] it does not seem hard to prove that the fibres of X → Y are actually covered by rational curves. Assume that k has positive characteristic and dim X ≥ 4. It seems hard to answer the question in this case because of the lack of resolution of singularities. However, if replace the klt condition with strongly F -regular, then it is likely that one can actually answer the question. References canonical pairs. theorem on log [1] C. Birkar; Existence of flips and minimal models for 3-folds in char p. arXiv:1311.3098. [2] C. Birkar, Z. Hu; Log canonical pairs with good augmented base loci. To appear in Com- positio Math. arXiv:1305.3569v2. [3] S. Boucksom, S. Cacciola, A. Felice Lopez; Augmented base loci and restricted volumes on normal varieties. arXiv:1305.4284. [4] S. Boucksom, J-P. Demailly, M. Paun, Th. Peternell; The pseudo-effective cone of a com- pact Kahler manifold and varieties of negative Kodaira dimension. J. Algebraic Geom. 22 (2013), 201-248 [5] S. Cacciola, A. Felice Lopez; Nakamaye’s arXiv:1303.7156. [6] P. Cascini, J. McKernan, M. Mustat¸a; The augmented base locus in positive character- istic. To appear in Shokurov’s conference proceedings. [7] P. Cascini, H. Tanaka, C. Xu; On base point freeness in positive characteristic. arXiv:1305.3502v1. [8] L. Ein, R. Lazarsfeld, M. Mustat¸a, M. Nakamaye, M. Popa; Restricted volumes and base loci of linear series. Amer. J. Math., 131(3), (2009), 607-651. [9] L. Ein, R. Lazarsfeld, M. Mustat¸a, M. Nakamaye, M. Popa; Asymptotic invariants of base loci. Ann. Inst. Fourier (Grenoble) 56, (2006) 1701-1734. [10] O. Fujino, H. Tanaka; On log surfaces. Proc. Japan Acad. Ser. A Math. Sci. Volume 88, Number 8 (2012), 109-114. [11] T. Fujita; Vanishing theorems for semipositive line bund les. In Algebraic geometry (Tokyo/Kyoto, 1982), 519-528, Lecture Notes in Math. 1016, Springer, Berlin, 1983. [12] S. Keel; Basepoint freeness for nef and big line bund les in positive characteristic. Annals of Math, Second Series, Vol. 149, No. 1 (1999), 253–286. [13] S. Keel, J. McKernan; Contractible extremal rays on M0,n . arXiv:alg-geom/9607009v1. [14] R. Lazarsfeld; Positivity in algebraic geometry. I. Springer-Verlag (2004). [15] M. Nakamaye; Base loci of linear series are numerical ly determined. Trans. Amer. Math. Soc. 355 (2002), 551-566. [16] M. Nakamaye; Stable base loci of linear series. Math. Ann. 318 (2000), 837-847. [17] V.V. Shokurov; 3-fold log models. Algebraic geometry, 4. J. Math. Sci. 81 (1996), no. 3, 2667-2699. [18] V.V. Shokurov; Complements on surfaces. Algebraic geometry, 10. J. Math. Sci. (New York) 102 (2000), no. 2, 3876–3932. [19] The stacks project. Website: http://stacks.math.columbia.edu. 16 Caucher Birkar [20] H. Tanaka; Minimal models and abundance for positive characteristic log surfaces. arXiv:1201.5699v2. [21] C. Xu; On base point arXiv:1311.3819. in positive characteristic. free theorem of threefolds DPMMS, Centre for Mathematical Sciences, Cambridge University, Wilberforce Road, Cambridge, CB3 0WB, UK email: [email protected]
1601.02911	2	1601	2016-04-26T13:18:58	Rank two aCM bundles on general determinantal quartic surfaces in $\mathbb{P}^3$	[ "math.AG", "math.AC" ]	Let $F\subseteq\mathbb{P}^3$ be a smooth determinantal quartic surface which is general in the N\"other-Lefschetz sense. In the present paper we give a complete classification of locally free sheaves $\mathcal{E}$ of rank $2$ on $F$ such that $h^1\big(F,\mathcal{E}(th)\big)=0$ for $t\in\mathbb{Z}$.	math.AG	math	ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 RANK TWO ACM BUNDLES GIANFRANCO CASNATI This paper is dedicated to the memory of A.T. Lascu. Abstract. Let F ⊆ P3 be a smooth determinantal quartic surface which is general in the Nother -- Lefschetz sense. In the present paper we give a complete classification of locally free sheaves E of rank 2 on F such that h1(cid:0)F,E(th)(cid:1) = 0 for t ∈ Z. 1. Introduction and Notation 6 1 0 2 r p A 6 2 ] . G A h t a m [ 2 v 1 1 9 2 0 . 1 0 6 1 : v i X r a acteristic 0 and we will denote by P3 the projective space of dimension 3 over k. is called aCM (i.e. arithmetically Cohen -- Macaulay) if h1(cid:0)F,E(th)(cid:1) = 0 for t ∈ Z. Throughout the whole paper we will work over an algebraically closed field k of char- If F ⊆ P3 is a smooth surface we set OF (h) := OP3(1) ⊗ OF . A vector bundle E on F When deg(F ) ≤ 2, the well -- known theorems of Horrocks and Knorrer (see [21], [1] and the references therein) provide a complete classification of aCM bundles on F . As deg(F ) increases a complete picture is missing. It is thus interesting to better understand aCM bundles on surfaces with deg(F ) ≥ 3. In order to simplify our study we restrict our attention to bundles satisfying some further technical non -- restrictive conditions. First, notice that the property of being aCM is invariant up to tensorizing by OF (th), thus we can consider in our study only initialized bundles, i.e. bundles E such that h0(cid:0)F,E(−h)(cid:1) = 0 and h0(cid:0)F,E(cid:1) (cid:54)= 0. Moreover, we can obviously focus onto indecomposable bundles, i.e. bundles which do not split in a direct sum of bundles of lower rank. When deg(F ) = 3 some scattered results are known. E.g., M. Casanellas and R. Hartshorne proved that each smooth cubic surface supports families of arbitrary dimension of non -- isomorphic, indecomposable initialized aCM bundles (see [4] and [5]). Moreover, D. Faenzi classified completely aCM bundles of rank 1 and 2 on such an F in [12]. A key point of all the quoted papers is the complete knowledge of Pic(F ). The case deg(F ) ≥ 4 is wide open. In this case the structure of Pic(F ) is not uniquely determined, thus it could be very difficult to classify initialized indecomposable aCM bundles E also in low ranks, unless assuming true some extra hypothesis on F . E.g. if Pic(F ) is generated by OF (h) and E has rank 2 on F with c1(E) ∼= OF (ch), then 3 − deg(F ) ≤ c ≤ deg(F ) − 1, as essentially proved by C. Madonna (see [19]: see also [6] where some slight improvement is explained). We are mostly interested in the case deg(F ) = 4. In this case F is a K3 surface, thus a lage amount of informations is available on Pic(F ). E.g. K. Watanabe in [24] classified initialized aCM bundles of rank 1 on such an F . 2010 Mathematics Subject Classification. Primary 14J60; Secondary 14J45. Key words and phrases. Vector bundle, Cohomology. The author is member of the GNSAGA group of INdAM. 1 2 GIANFRANCO CASNATI In the beautiful paper [9] E. Coskun, R.S. Kulkarni, Y. Mustopa dealt with the case of rank 2 initialized indecomposable bundles E on F with c1 = 3h. In [8] the authors gave a complete description of bundles with c1 = 2h: it is perhaps interesting to notice that in this case there is a large family of quartic surfaces, the determinantal ones (i.e. surfaces whose equation can be expressed as the determinant of a 4× 4 matrix with linear entries), supporting bundles whose non -- zero sections do not vanish in the expected dimension. The present paper is motivated by our interest in the classification of rank 2 initialized indecomposable bundles E on a determinantal quartic surface F ⊆ P3. In this case F contains a projectively normal smooth sextic curve A of genus 3. It is thus obvious that Pic(F ) must contain the rank 2 lattice Λ generated by h and A. Moreover, we actually have Pic(F ) = Λ if F is general in the sense of Nother -- Lefschetz, i.e. it represents a point of OP3(4) outside a countable union of suitable proper subvari- eties, due to some classic results proved in [18]. Each F satisfying the above property will be called a general determinantal quartic surface. Theorems 6.2 and 6.5 give the complete classification of the bundles we are interested in on such an F . For reader's benefit we summarize such a classification in the following simplified statement. For the complete statement see Theorems 6.2 and 6.5. Main Theorem. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. Then one of the following asseritions holds. (i) c1 ∈ { −h, A−h, 2h−A }: then c2 = 1. The zero locus E of a general s ∈ H 0(cid:0)F,E) then c2 = 2. The zero locus E ⊆ P3 of a general s ∈ H 0(cid:0)F,E) is a then c2 is 3, 4, or 5. The zero locus E of a general s ∈ H 0(cid:0)F,E) is 0 -- dimensional subscheme contained in exactly one line. (ii) c1 = 0: is a point. linearly independent quadrics. (iii) c1 = h: a 0 -- dimensional subscheme of degree c2 of a linear subspace of P3 of dimension c2 − 2. (iv) c1 ∈ { A, 3h − A }: then c2 is 3, or 4. In this case the zero locus E of a general (v) c1 ∈ { A + h, 4A − h }: then c2 = 8. In this case the zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional subscheme. s ∈ H 0(cid:0)F,E) is a 0 -- dimensional subscheme which is contained in exactly two (a) the zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional subscheme (b) the zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional subscheme lying (c) the zero locus E of each non -- zero s ∈ H 0(cid:0)F,E) is a divisor in either A, or (vii) c1 = 3h: then c2 = 14. In this case the zero locus E of a general s ∈ H 0(cid:0)F,E) is on exactly a twisted cubic; 3h − A. (vi) c1 = 2h: then c2 = 8. In this case one of the two following cases occur: which is the base locus of a net of quadrics; a 0 -- dimensional subscheme. Moreover, all the above cases actually occur on F . We now explain briefly the structure of the paper. In Section 2 we recall all the facts on a smooth quartic surface that we will need in the paper. In Section 3 we specialize to the case of determinantal quartic surfaces. In Section 4 we list a series of examples. In Section 5 we deal with the zero -- locus of the sections of initialized indecomposable aCM bundles of rank 2 on F , proving that it is 0 -- dimensional, but the two cases which we mentioned RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 3 before. Finally, in Section 6 we prove the above Main Theorem, showing that the list of examples in Section 4. The author would like to thank F. Galluzzi for several helpful discussions and the referee, whose comments helped to considerably improve the exposition of the paper. For all the notations and unproven results we refer to [13]. 2. Preliminary results In this section we will recall some preliminary facts about vector bundles on a smooth quartic surface F ⊆ P3, most of them collected from several papers (see [10], [22], [24] and [14]). ∼= OF , i.e. each such an F is a K3 surface, thus we can By adjunction in P3 we have ωF make use of all the results proved in [22]. The first important fact is that Serre duality for each locally free sheaf F on F becomes hi(cid:0)F,F(cid:1) = h2−i(cid:0)F,F∨(cid:1), i = 0, 1, 2. Moreover Riemann -- Roch theorem on F is h0(cid:0)F,F(cid:1) + h2(cid:0)F,F(cid:1) = h1(cid:0)F,F(cid:1) + 2rk(F) + (1) In particular, if D is a divisor with D2 ≥ −2, then either D or −D is necessarily effective. If D is an effective divisor on F , then h2(cid:0)F,OF (D)(cid:1) = h0(cid:0)F,OF (−D)(cid:1) = 0. Moreover h1(cid:0)F,OF (D)(cid:1) = h1(cid:0)F,OF (−D)(cid:1) = h0(cid:0)D,OD (cid:1) − 1 c1(F)2 2 − c2(F). (see [22], Lemma 2.2). It follows that for each irreducible divisor D on F , the dimension of D and the arithmetic genus of D satisfy h0(cid:0)F,OF (D)(cid:1) = 2 + (2) D2 2 , deg(D) = Dh, pa(D) = 1 + D2 2 , (see [22], Paragraph 2.4). In particular, the unique irreducible and reduced fixed divisors D on F satisfy D2 = −2, thus they are smooth rational curves. We summarize the other helpful results we will need in the following statement. Proposition 2.1. Let F ⊆ P3 be a smooth quartic surface. For each effective divisor D on F such that D has no fixed components the following assertions hold. (i) D2 ≥ 0 and OF (D) is globally generated. (ii) If D2 > 0, then the general element of D is irreducible and smooth: in this case (iii) If D2 = 0, then there is an irreducible curve D with pa(D) = 1 such that OF (D) ∼= h1(cid:0)F,OF (D)(cid:1) = 0. OF (eD) where e − 1 := h1(cid:0)F,OF (D)(cid:1): in this case the general element of D is smooth. Proof. See [22], Proposition 2.6 and Corollary 3.2. (cid:3) Now we turn our attention to aCM line bundles on F recalling a very helpful result from [24]. A first obvious fact is that for each divisor D on F , then OF (D) is aCM if and only if the same is true for OF (−D). The main result from [24] is the following statement. 4 GIANFRANCO CASNATI Proposition 2.2. Let F ⊆ P3 be a smooth quartic surface. of the following cases occurs. For each effective divisor D on F with D (cid:54)= 0, then OF (D) is aCM if and only if one (i) D2 = −2 and 1 ≤ hD ≤ 3. (ii) D2 = 0 and 3 ≤ hD ≤ 4. (iii) D2 = 2 and hD = 5. (iv) D2 = 4, hD = 6 and h0(cid:0)F,OF (D − h)(cid:1) = h0(cid:0)F,OF (2h − D)(cid:1) = 0. Proof. See [24], Theorem 1.1. Recall that a curve D ⊆ P3 is called aCM if h1(cid:0)P3,IDP3(th)(cid:1) = 0 for each t ∈ Z (see (cid:3) Notice that the general smooth quartic surface F ⊆ P3 does not support initialized aCM line bundles besides OF . Indeed, in this case, Pic(F ) is generated by OF (h), due to Nother -- Lefschetz theorem. In particular, for each line bundle OF (D) on F one has that both Dh and D2 are positive multiples of 4. [20], Lemma 1.2.3). A curve D ⊆ P3 is projectively normal if and only if it is aCM and smooth. We recall the following helpful fact. Lemma 2.3. Let F ⊆ P3 be a smooth quartic surface. aCM in P3. For each effective divisor D on F , then OF (D) is aCM if and only if the curve D is Proof. We have the exact sequence 0 −→ IFP3 −→ IDP3 −→ IDF −→ 0. only if h1(cid:0)F,OF (D − th)(cid:1) = h1(cid:0)F,IDF (th)(cid:1) = 0. Since IFP3 ∼= OP3(−4) and IDF a sheaf over P3, thus H 0∗ (F) :=(cid:76) ∼= OF (−D), it follows that h1(cid:0)P3,IDP3(t)(cid:1) = 0 if and t∈Z H 0(cid:0)F,F(th)(cid:1) is also a module over the polynomial We conclude this section by recalling some further facts facts on vector bundles on a smooth quartic surface F ⊆ P3. Let F be a vector bundle on F : we can consider it as (cid:3) ring S := k[x0, x1, x2, x3]. For the following fact see [4], Theorem 3.1. Lemma 2.4. Let F be an initialized aCM bundle of rank r on a smooth quartic surface F ⊆ P3. The minimal number of generators of H 0∗ (F) as a module over S is at most 4r. Definition 2.5. Let F be an initialized aCM bundle of rank r on a smooth quartic surface F ⊆ P3. We say that F is Ulrich if it is initialized, aCM and h0(cid:0)F,F(cid:1) = 4r. Remark 2.6. If F is Ulrich, then it is globally generated by Lemma 2.4. More generally, Ulrich bundles on F of rank r are exactly the bundles F on F such that H 0∗ (F) has linear minimal free resolution over S (see [4], Proposition 3.7). Sheafifying such a resolution we thus obtain an exact sequence P3 −→ F −→ 0. 0 −→ OP3(−1)⊕4r −→ O⊕4r 3. The general smooth determinantal quartic surfaces In this section we will summarize some facts about the smooth determinantal quartic surfaces in P3. We will denote by f the polynomial defining F . RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 5 Proposition 3.1. Let F ⊆ P3 be a smooth quartic surface. The following assertions are equivalent. (i) The polynomial f defining F is the determinant of a 4 × 4 matrix Φ with linear (ii) There exists a line bundle OF (A) such that the sheafified minimal free resolution entries. of H 0∗ (OF (A)) has the form (3) 0 −→ OP3(−1)⊕4 ϕ−→ O⊕4 P3 −→ OF (A) −→ 0. elliptic. (iii) There exists an initialized aCM line bundle OF (A) with A2 = 4. (iv) There exists a smooth integral sextic curve A ⊆ F of genus 3 which is not hyper- (v) There exists a smooth integral sextic curve A ⊆ F of genus 3 which is projectively (vi) There exists a smooth integral sextic curve A ⊆ F of genus 3 such that h0(cid:0)F,OF (2h − A)(cid:1) = h0(cid:0)F,OF (A − h)(cid:1) = 0. normal in P3. Proof. Assertion (i) is equivalent to assertions (ii), (iv), (v) due to Corollaries 1.8, 6.6 and Proposition 6.2 of [2]. Assertions (iii) and (vi) are equivalent thanks to Proposition 2.2. Assertion (ii) implies (iii). Indeed, we have 2 + A2 2 thus A2 = 4. Conversely, h0(cid:0)F,OF (A)(cid:1) = 4 for each initialized aCM line bundle with = χ(OF (A)) = 4(χ(OP3) − χ(OP3(−1))) = 4, A2 = 4. Hence, OF (A) is Ulrich, thus it fits into Sequence (3) (see e.g. Proposition 3.7 of (cid:3) [4]). Remark 3.2. Let F be determinantal, so that we have Sequence (3). Its dual is (cid:0)OF (A),OP3 0 −→ OP3(−1)⊕4 (cid:1) ∼= OF (4h − A). tϕ−→ O⊕4 (4) because Ex t1P3 Notice that OF (3h − A) (cid:54)∼= OF (A), because (3h − A)A = 14 (cid:54)= 4 = A2. Thus also the general element in 3h − A is a curve enjoying the same properties as A. P3 −→ OF (3h − A) −→ 0, From the above results we deduce that if the polynomial f defining the smooth quartic surface F is the determinant of a 4 × 4 matrix with linear entries, then Pic(F ) contains the even sublattice Λ generated by h and A such that h2 = A2 = 4 and hA = 6. By definition the general element in h is an irreducible smooth plane curve of degree 4. Thanks to Proposition 3.1 the general element in A is an integral smooth projectively normal curve of degree 6 and genus 3(see Equalities (2)). Conversely, let A ⊆ P3 be an integral projectively normal sextic of genus 3. We know that each smooth quartic surface F ⊆ P3 containing A is necessarily determinantal. The Moreover OF (4h − A) is globally generated (see e.g. Sequence (4)), thus the general element in OF (4h − A) is integral. It follows that Pic(F ) = Λ due to [18], Theorem II.3.1, if F is general in the Nother -- Lefschetz sense. minimal degree of a surface containing such an A is 3, because h0(cid:0)F,OF (2h − A)(cid:1) = 0. call it a general determinantal quartic surface. From now on, we will assume that the surface F has the above property and we will When we will consider such a surface F , we will always denote by OF (A) a line bundle corresponding to a projectively normal sextic of genus 3 on F . Notice that there is an 6 GIANFRANCO CASNATI ambiguity because, as we pointed out above, the line bundle OF (3h − A) has the same properties as OF (A). F ⊆ P3 satisfy rk(Pic(F )) = 2. In order to complete the picture, we recall that not all determinantal quartic surfaces E.g. a smooth quartic surface F ⊆ P3 containing a twisted cubic curve T and a line E without common points is determinantal (see [11], Exercise 4.14). Indeed the union of T with any plane cubic linked to E define a linear system whose general element A is a projectively normal sextic of genus 3. It is easy to check that h, T , E are linearly independent in Pic(F ), thus rk(Pic(F )) ≥ 3 in this case. In [7] the authors construct many examples of determinantal quartic surfaces F ⊆ P3 with rk(Pic(F )) = 3 and show that even the classification of indecomposable Ulrich bundles of rank 2 on them seems to be highly non -- trivial. Remark 3.3. A first interesting fact is that if D ∈ xh + yA, then (5) D2 = 4x2 + 12xy + 4y2, DA = 6x + 4y. Dh = 4x + 6y, More precisely it is very easy to check the following assertions: (i) D2 ≡ 0 (mod 16) if and only if both x and y are even; (ii) D2 ≡ 4 (mod 8) if and only if x and y are not simultaneously even. The two above assertions imply that D2 is always a multiple of 4. Thus Equality (1) imply that χ(OF (D)) is even. Moreover, Equalities (5) imply that the equation D2 = 0 has no non -- trivial integral solutions, thus: (iii) D2 = 0 if and only if D = 0. Irreducible and reduced fixed divisors D on F satisfy D2 = −2 (see the last Equality (2)), which is not a multiple of 4. It follows that there are no fixed divisors on F . In particular DU ≥ 0 for each divisor U ⊆ F , hence (see Equalities (5)): (iv) DU ≥ 0 is even for each divisor U ⊆ F . Combining the above assertions with Proposition 2.1, we deduce the following consequence: (v) each element in h, A, 3h − A is certainly integral (but maybe singular). Looking at Remark 3.3 we know that the following result holds. Lemma 3.4. Let F ⊆ P3 be a general determinantal quartic surface. If the divisor D on F is effective, then D has no fixed components. In particular for each effective divisor D on F we have D2 ≥ 0 and h0(cid:0)F,OF (D)(cid:1) ≥ 2. h0(cid:0)F,OF (D)(cid:1) ≥ 4. Finally, the ampleness of h implies Dh ≥ 2. Moreover, if D (cid:54)= 0, then from Remark 3.3 we know that D2 (cid:54)= 0, whence D2 ≥ 4 and The above discussion partially proves the following result. Lemma 3.5. Let F ⊆ P3 be a general determinantal quartic surface. The divisor D on F is effective if and only if either D = 0, or Dh ≥ 2 and D2 ≥ 4. Proof. The 'only if' part was proved above. Assume that Dh ≥ 2 and D2 ≥ 4. Either D or −D is effective, because D2 ≥ −2. We have (−D)h = −2, thus −D is not effective. we (cid:3) conclude that D is effective. Remark 3.6. Thus D ∈ xh + yA is effective if and only if either (x, y) = (0, 0), or x2 + 3xy + y2 ≥ 1, 2x + 3y ≥ 1. RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 7 The above inequalities give the closed connected subset of the affine plane with coordinates x and y not containing the point (x, y) = (0, 0) and whose border is the graph of the function −3x + 5x2 + 4 y = √ 2 (the 'upper branch' of the hyperbola x2 + 3xy + y2 = 1). Proposition 2.1 and the above Lemmas 3.4 and 3.5 prove the following corollary. Corollary 3.7. Let F ⊆ P3 be a general determinantal quartic surface. The divisor D on F is effective if and only if OF (D) is globally generated. 4. Examples of rank 2 aCM bundles of rank 2 aCM bundles on general determinantal quartic surfaces. In this section we briefly recall Hartshorne -- Serre correspondence, listing some examples Let E be a rank 2 vector bundle on a smooth quartic surface F ⊆ P3 with Chern classes ci := ci(E), i = 1, 2. If s ∈ H 0(cid:0)F,E(cid:1) is non -- zero, then its zero -- locus E := (s)0 ⊆ F has a possibly zero divisor. We thus obtain a section σ ∈ H 0(cid:0)F,E(−D)(cid:1) vanishing exactly on We can write (s)0 = E ∪ D where E has pure codimension 2 (or it is empty) and D is codimension at most 2. E. The Koszul complex of σ gives the following exact sequence 0 −→ OF (D) −→ E −→ IEF (c1 − D) −→ 0. (6) The degree of E is c2(E(−D)). Bertini's theorem for the sections of a vector bundle guarantees that if E is globally generated (e.g. if E is Ulrich), then D = 0 for the general s ∈ H 0(cid:0)F,E(cid:1). Assume that the divisorial part D of (s)0 is not empty for general s ∈ H 0(cid:0)F,E(cid:1). We can define a rational map H 0(cid:0)F,E) (cid:57)(cid:57)(cid:75) Pic(F ) defined by s (cid:55)→ OF (D). Since Pic(F ) is of the sections in a suitable non -- empty subset of H 0(cid:0)F,E) are linearly equivalent. discrete, it follows that such a map is constant. Thus the divisorial part of the zero -- loci The above construction can be reversed. To this purpose we give the following well -- known definition. Definition 4.1. Let F ⊆ P3 be a smooth quartic surface and let G be a coherent sheaf on F . We say that a locally complete intersection subscheme E ⊆ F of dimension zero is Cayley -- Bacharach (CB for short) with respect to G if, for each E(cid:48) ⊆ E of degree deg(E)−1, the natural morphism H 0(cid:0)F,IEF ⊗ G(cid:1) → H 0(cid:0)F,IE(cid:48)F ⊗ G(cid:1) is an isomorphism. For the following result see Theorem 5.1.1 in [15]. Theorem 4.2. Let F ⊆ P3 be a smooth quartic surface, E ⊆ F a locally complete inter- section subscheme of dimension 0. Then there exists a vector bundle E of rank 2 on F with det(E) = L and having a section s such that E = (s)0 if and only if E is CB with respect to L. If such an E exists, then it is unique up to isomorphism if h1(cid:0)F,L∨(cid:1) = 0. Let F ⊆ P3 be a smooth quartic surface. We recall that in [9] the author proved the existence on F of aCM bundles E with c1(E) = 3h and c2(E) = 14. To this purpose the authors proved the existence of suitable sets of 14 points which are CB with respect to OF (3h). 8 GIANFRANCO CASNATI and c2(E) = 8. Essentially two cases are possible. Similarly, in [8] the author proved the existence on F of aCM bundles E with c1(E) = 2h In the first case, the zero -- locus of a general section of E has pure dimension 0: this case In the second case, the zero -- loci of non -- zero sections of E define a linear system D whose general element is a smooth integral sextic of genus 3 such that h0(cid:0)F,OF (D−h)(cid:1) = h0(cid:0)F,OF (2h − D)(cid:1) = 0. In particular F must be a determinantal quartic surface in this occurs without further hypothesis on F . second case. Due to the results listed in the previous section we know that this case can occur only if F is determinantal. In particular it occurs when F is a general determinantal quartic surface F . E ⊆ F . Recall that for each closed subscheme X ⊆ F we have an exact sequence We now construct examples of rank 2 aCM bundles starting from suitable sets of points (7) 0 −→ IXF −→ OF −→ OX −→ 0. In what follows we will present a series of examples of initialized indecomposable aCM rank 2 bundles on a quartic surface F ⊆ P3 with plane section h. We stress that the bundles defined in Examples 4.3 and 4.4 exist on each smooth quartic surface. Example 4.3. Let E ⊆ F be a 0 -- dimensional subscheme arising as follows: • m = 3 E is the complete intersection of a line and a cubic surface; • m = 4 E is the complete intersection of two conics inside a plane; • m = 5 E is the degeneracy locus of a 5 × 5 skew symmetric matrix ∆ with linear entries. It is clear that E is CB with respect to OF (h) in the first two cases. In the third one, a minimal free resolution of H 0∗ (OE) over P3 is given by the Buchsbaum -- Eisenbud complex (see [3]). Thus again E is CB with respect to OF (h) thanks to Lemma (4.2) of [23]. Theorem 4.2 thus yields the existence of a rank 2 bundle E fitting into Sequence (6) with c1 = h, c2 = m and D = 0, i.e. (8) The cohomology of the above sequence twisted by OF (−h) gives 0 −→ OF −→ E −→ IEF (h) −→ 0, h2(cid:0)F,E(cid:1) = h0(cid:0)F,E∨(cid:1) = h0(cid:0)F,E(−h)(cid:1) = 0, 0 −→ H 1(cid:0)F,E(cid:1) −→ H 1(cid:0)F,IEF (h)(cid:1) −→ H 2(cid:0)F,OF i.e. E is initialized and the cohomology of Sequence (8) yields the exact sequence. By definition h0(cid:0)F,IEF (h)(cid:1) = 5 − m, whence h1(cid:0)F,IEF (h)(cid:1) = 1, from the cohomology of Sequence (7). We conclude that h1(cid:0)F,E(cid:1) = 0. (cid:1) −→ 0. The cohomology of Sequence (8) twisted by OF (th) gives h1(cid:0)F,E(th)(cid:1) = h1(cid:0)F,IEF ((t + 1)h)(cid:1), t ≥ 1. In the cases m = 3, 4 the minimal free resolution of H 0∗ (OE) over P3 is given by a Koszul complex. In the case m = 5, it is given by the Buchsbaum -- Eisenbud complex as already pointed out above. In any case it is easy to check that the Hilbert function of E is (1, m − 1, m, . . . ), hence the natural map H 0(cid:0)F,OF (th)(cid:1) → H 0(cid:0)E,OE(th)(cid:1) is surjective 9 RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 for t ≥ 1, thus h1(cid:0)F,IEF (th)(cid:1) = 0, in the same range (see the cohomology of Sequence (7)). We conclude that h1(cid:0)F,E(th)(cid:1) = 0 for t ≥ 0. By duality h1(cid:0)F,E(th)(cid:1) = h1(cid:0)F,E∨(−th)(cid:1) = h1(cid:0)F,E((−t − 1)h)(cid:1) = 0, i.e. t ≤ −1. We conclude that E is aCM. If E ∼= OF (U ) ⊕ OF (V ), then U + V = h and c2 = U V which is even (see Remark 3.3). Moreover both OF (U ) and OF (V ) are aCM and at least one of them must be initialized, the other being either initialized or without sections. Let OF (U ) be initialized. If U = A, then V = h − A, thus U V = 2 < 3 ≤ c2. The case U = 3h − A can be excluded in the same way. If U = 0, then V = h, a contradiction. We conclude that E is indecomposable. Example 4.4. Let E ⊆ F be a set of m = 1, 2 points. Then E is trivially CB with respect to OF ((m − 2)h), thus there is a rank 2 bundle E fitting into Sequence (6) with c1 = (m − 2)h, c2 = m and D = 0. As in the Example 4.3 one easily checks that such an E is aCM, initialized and inde- −t − 1 ≥ 0, composable. For all the following examples we obviously need to restrict to a determinantal quartic surface F ⊆ P3. As usual h denotes the class of the plane section and A the class of a projectively normal integral sextic curve of genus 3 on F . Example 4.5. Let E ⊆ F be a single point. The line bundle OF (A − h) is not effective, thus E is again trivially CB with respect to OF (A − h). In particular there is a rank 2 bundle E fitting into an exact sequence of the form (9) 0 −→ OF −→ E −→ IEF (A − h) −→ 0, for each t. OF (A − 2h) is not effective. Thus E is initialized. The same sequence twisted by OF (th) The cohomology of Sequence (9) twisted by OF (−h) gives h0(cid:0)F,E(−h)(cid:1) = 0, because also yields the existence of an injective map H 1(cid:0)F,E(th)(cid:1) → H 1(cid:0)F,IEF (A + (t − 1)h)(cid:1) The cohomology of Sequence (7) twisted by OF (A + (t− 1)h) yields h1(cid:0)F,IEF (A + (t− 1)h)(cid:1) = 0 for t ≥ 1. Hence h1(cid:0)F,E(th)(cid:1) = 0, for t ≥ 1. The cohomology of Sequence (7) twisted by OF (−th) yields h1(cid:0)F,IEF (−th)(cid:1) = 0 for Since OF (A) is aCM, the same is true for OF (−A), thus h1(cid:0)F,E(th)(cid:1) = h1(cid:0)F,E∨(−th)(cid:1) = H 1(cid:0)F,OF ((1 − t)h − A)(cid:1) → H 1(cid:0)F,E∨(−th)(cid:1), t ≤ 0. Sequence (9) yields the existene of surjective maps 0 for t ≤ 0. We conclude that E is aCM. If E ∼= OF (U ) ⊕ OF (V ), then c2 = U V which is even (see Remark 3.3). Since c2 = 1, it follows that E is indecomposable. Finally c1 = A − h by construction. Substituting A with 3h − A in the above construction we are also able to construct an example of initialized indecomposable aCM bundle of rank 2 with c1 = 2h − A. We can construct two more bundles. Indeed, E∨(h) is indecomposable and aCM too. Moreover the cohomology of Sequence (9) suitably twisted also implies that E∨(h) is initialized. Computing the Chern classes of E∨(h) we obtain c1 = 3h − A and c2 = 3. Again the substitution of A with 3h − A gives an example of initialized indecomposable aCM bundle of rank 2 with c1 = A and c2 = 3. t ≤ 0. 10 GIANFRANCO CASNATI F → O⊕2 Example 4.6. The curve A ⊆ F has genus 3 and it is not hyperelliptic (see Proposition 3.1), thus it supports infinitely many complete g1 4. Fix a smooth divisor E in one of these 4. The natural composite map O⊕2 A → OA(E) is surjective on global sections and g1 its kernel is a vector bundle F on F . The bundle E := F∨ is called the Lazarsfeld -- Mukai bundle of OA(E). By definition we have the exact sequence (10) Clearly c1 = A and c2 = 4. In what follows we will quickly show that E is aCM. The cohomology Sequence (10) yields h1(cid:0)F,E(cid:1) = h1(cid:0)F,E∨(cid:1) = 0 by definition. Twisting Sequence (10) by OF (−th), taking its cohomology we also obtain h1(cid:0)F,E(−th)(cid:1) = 0 for t ≥ 1, because h0(cid:0)A,OA(E−th)(cid:1) = 0, due to the inequality deg(OA(E−th)) = 4−6t ≤ −2 (cid:1) ∼= ωA(−E). Adjunction formula on F thus implies ωA (cid:0)OA(E),OF F −→ OA(E) −→ 0. 0 −→ E∨ −→ O⊕2 ∼= F We have Ex t1 in the same range. OA(A), hence the dual of Sequence (10) is (11) Thus, also for t ≥ 1, we obtain 0 −→ O⊕2 F −→ E −→ OA(A − E) −→ 0. h1(cid:0)F,E(th)(cid:1) ≤ h1(cid:0)A,OA(A − E + th)(cid:1) = h0(cid:0)A,OA(E − th)(cid:1) = 0. the vanishing h0(cid:0)A,OA(E − h)(cid:1) = 0 (see above) also imply that E is initialized. We conclude that E is aCM. The cohomology of Sequence (11) twisted by OF (−h) and Assume that E ∼= OF (U ) ⊕ OF (V ). We thus have U + V = A and U V = 4. As in Example 4.3 we can assume that OF (U ) is aCM and initialized and that OF (V ) is either initialized or without sections. If U = A, then V = 0, whence U V = 0 a contradiction. Similarly U (cid:54)= 0. If U = 3h − A, then V = 2A − 3h and again U V = 0. The same argument with 3h − A instead of A yields the existence of initialized inde- composable aCM bundles of rank 2 with c1 = 3h − A and c2 = 4. Also in this case we can construct two more bundles. Indeed E∨(h) is indecomposable and aCM too. Moreover, Equality (1) and the vanishing h2(cid:0)F,E∨(th)(cid:1) = h0(cid:0)F,E(−th)(cid:1) = 0, t = 1, 2, return h0(cid:0)F,E∨(h)(cid:1) = 0, h0(cid:0)F,E∨(2h)(cid:1) = 6. It follows that E∨(2h) is initialized. For such a bundle c1 = 4h − A and c2 = 8. The substitution of A with 3h− A gives an example of initialized indecomposable aCM bundle of rank 2 with c1 = A + h and c2 = 8. It is interesting to notice that the bundle E (hence also E∨(2h)) described above has an interesting property. Indeed, one can prove using Lemma 2.4 of [16] that both OF (A), and OF (3h − A) are very ample on F , thus they define two other different embeddings F ⊆ P3 with respective hyperplane line bundles OF (A) and OF (3h − A). Proposition 2.3 of [25] guarantees that E is aCM also for these embeddings: indeed, it is the bundle with m = 4 of Example 4.3 with respect to these embeddings. 5. On the zero -- locus of general sections In this section we will assume that F ⊆ P3 is a general determinantal quartic surface. Lemma 5.1. Let E be an indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. If s ∈ H 0(cid:0)F,E(cid:1), then (s)0 (cid:54)= ∅. RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 11 Proof. Assume that (s)0 = ∅. Then Sequence (6) becomes 0 −→ OF −→ E −→ OF (c1) −→ 0, where h1(cid:0)F,OF (c1)(cid:1) = h1(cid:0)F,OF (−c1)(cid:1) (cid:54)= 0, because E is indecomposable. The coho- mology of the above sequence gives a monomorphism H 1(cid:0)F,OF (c1)(cid:1) → H 2(cid:0)F,OF (cid:1), thus h1(cid:0)F,OF (c1)(cid:1) = 1. 1 ≥ 4 (see Lemma 3.5), thus h1(cid:0)F,OF (c1)(cid:1) = 0 (see Lemma Trivially h1(cid:0)F,E∨(2h)(cid:1) = 0: moreover h2(cid:0)F,E∨(h)(cid:1) = h0(cid:0)F,E(−h)(cid:1) = 0. The above Remark 3.3 and Equality (1) imply that χ(OF (c1)) is even, thus either OF (c1) or (cid:3) OF (−c1) is effective. Hence c2 3.4 and Proposition 2.1), a contradiction. vanishings prove the following lemma. Lemma 5.2. Let E be an initialized, indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. Then E∨(3h) is regular in the sense of Castelnuovo -- Mumford. In particular, both E∨(3h) and OF (6h − c1) are globally generated. Let E be as above. If s ∈ H 0(cid:0)F,E(cid:1) is a non -- zero section, then (s)0 (cid:54)= ∅. From now on Notice that the cohomology of Sequence (6) gives an injective map H 0(cid:0)F,OF (D−h)(cid:1) → H 0(cid:0)F,E(−h)(cid:1), thus OF (D) is initialized too, or, in other words, OF (D−h) is not effective. we will assume that (s)0 = E ∪ D where E has pure dimension 0 (or it is empty) and D is an effective divisor. Now assume that c1 = xh + yA is effective. It follows from Lemma 5.2 and Remark 3.6 that c1 is either 0, or 6h, or the inequalities y2 + 3xy + x2 ≥ 1, y2 − 3(6 − x)y + (6 − x)2 ≥ 1, 2x + 3y ≥ 1, 2(6 − x) − 3y ≥ 1, must hold. Thus, in the latter case, the point (x, y) must lie in the region delimited by the graphs of the two functions −3x + y = 5x2 + 4 , y = √ 2 3(6 − x) −(cid:112)5(6 − x)2 + 4 . 2 By drawing the graphs of the above functions (we made such computations using the software Grapher), we obtain Picture 1 below. Figure 1 -2-1012345678-3-2-112312 GIANFRANCO CASNATI The values of c1 we are interested in are in the region bounded by the graphs of the two functions above, plus the two points (0, 0) and (6, 0) (see above). Thus c1 is in the following list: (12) 0, A, 2A, 4h − A, h, 4h, h + A, 5h − A, 5h, 6h − 2A, 3h, 6h − A, 3h + A, 6h. 2h, 2h + A, 3h − A, We are ready to prove the main result of the present section. Theorem 5.3. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. Let s ∈ H 0(cid:0)F,E(cid:1) and (s)0 = E ∪ D be as above. Then exactly one of the following cases occurs: (i) c1 − D is effective; (ii) E is a point, D = 0 and c1 is not effective; (iii) E = ∅, D ∈ A ∪ 3h − A and c1 = 2h. Proof. Let c1 − D be effective. It is obviously true that c1 is effective, thus case (ii) above cannot occur. If c1 = 2h and D ∈ A ∪ 3h − A, then c1 − D is not effective, because (2h − A)2 = (A − h)2 = −4, (see Lemma 3.5). It turns out that case (iii) above cannot occur as well. In order to complete the proof we can thus assume that c1 − D is not effective proving that case either (ii), or (iii) must hold. The cohomology of Sequence (7) for X := E twisted by OF (c1 − D) implies h1(cid:0)F,IEF (c1 − D)(cid:1)≥ deg(E). The cohomology of Sequence (6) implies because D is effective and non -- zero. Thus E = ∅ and Sequence (6) becomes 0 −→ OF (D) −→ E −→ OF (c1 − D) −→ 0. The cohomology of the above sequence gives surjective maps H 0(cid:0)F,OF (c1 − D + th)(cid:1) → H 1(cid:0)F,OF (D + th)(cid:1), for t ∈ Z. It follows that h1(cid:0)F,OF (D + th)(cid:1) = 0 for each t ≤ 0. Moreover D is effective, thus h1(cid:0)F,OF (D + th)(cid:1) = 0 also for t ≥ 1. We conclude that D is aCM. We already know that OF (D) is initialized, because the same is true for E. Due to Proposition 2.2 and Lemma 3.5 the only initialized aCM line bundles OF (xh + yA) (cid:54)∼= OF on F satisfy the equalities (xh + yA)2 = 4 and (xh + yA)h = 6, which are equivalent to the system The unique solutions of this system correspond exactly to A and 3h − A. x2 + 3xy + y2 − 1 = 2x + 3y − 3 = 0. If D = 0, then c1 is not effective. We have E (cid:54)= ∅ due to Lemma 5.1. Moreover h1(cid:0)F,IEF (c1 − D)(cid:1) ≤ h2(cid:0)F,OF (D)(cid:1) = h0(cid:0)F,OF (−D)(cid:1). 1 ≤ deg(E) ≤ h1(cid:0)F,IEF (c1)(cid:1) ≤ h0(cid:0)F,OF (cid:1) = 1, deg(E) ≤ h1(cid:0)F,IEF (c1 − D)(cid:1) ≤ h0(cid:0)F,OF (−D)(cid:1) = 0, thus equality must hold. It follows that E is a point. If D (cid:54)= 0, then RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 13 If c1 is effective, then it is in the List (12). Taking into account that c1 − D is not effective, it follows that D is non -- zero, whence E = ∅ from the above discussion. The pair (c1, D) must be either (2h, A), or (2h, 3h − A), or it is in the following list; (0, 3h − A), (A, 3h − A), (2A, 3h − A), (0, A), (h + A, 3h − A), (3h − A, A), (2h + A, 3h − A), (h, A), (4h − A, A), (5h − A, A), (h, 3h − A), (6h − 2A, A). In order to complete the proof of the theorem we have to show that all the cases in the above list cannot occur. We will examine the cases one by one. Let (c1, D) = (0, A). Sequence (6) becomes Let (c1, D) = (h, A). Sequence (6) becomes 0 −→ OF (A) −→ E −→ OF (−A) −→ 0, same argument can be used to exclude also the case (0, 3h − A). which splits, because h1(cid:0)F,OF (2A)(cid:1) = 0, contradicting the indecomposability of E. The Since (2A − h)2 = −4, we deduce that h0(cid:0)F,OF (2A − h)(cid:1) = h1(cid:0)F,OF (2A − h)(cid:1) = 0. Equality (1) thus yields h1(cid:0)F,OF (2A − h)(cid:1) = 0, hence the above sequence must split, h0(cid:0)F,OF (2A − 3h)(cid:1) = 0, and h2(cid:0)F,OF (2A − 3h)(cid:1) = h0(cid:0)F,OF (3h − 2A)(cid:1) = 0. It follows from Equality (1) that h1(cid:0)F,OF (2A−3h)(cid:1) = 8. Since h1(cid:0)F,OF (A)(cid:1) = h2(cid:0)F,OF (A)(cid:1) = 0, it follows h1(cid:0)F,E(cid:1) = h1(cid:0)F,OF (2A − 3h)(cid:1) (cid:54)= 0 from the cohomology of Sequence (6). Let (c1, D) = (A, 3h − A). Since (2A − 3h)2 = (3h − 2A)2 = −20, we deduce that both again a contradiction. The same argument also excludes the case (h, 3h − A). 0 −→ OF (A) −→ E −→ OF (h − A) −→ 0. The same argument can be also used to exclude the cases (2A, 3h− A), (A + h, 3h− A), (3h− A, A), (4h− A, A), (6h− 2A, A). In the cases (A + 2h, 3h− A), (5h− A, A) we must modify slightly the above argument by taking the cohomology of Sequence (6) twisted by OF (−h). (cid:3) Remark 5.4. We have h1(cid:0)F,OF (2A − 2h)(cid:1) = h1(cid:0)F,OF (4h − 2A)(cid:1) = 6, thus there exist indecomposable extensions 0 −→ OF (A) −→ E −→ OF (2h − A) −→ 0, 0 −→ OF (3h − A) −→ E −→ OF (A − h) −→ 0. Both OF (A) and OF (−A) are aCM, thus the same is true for the above bundles E and it is easy to check that such bundles E are initialized as well. As pointed out in [8], Theorem 4.1 and Remark 4.2, we know that such bundles are indecomposable and that each non -- zero section of the above bundles vanishes on a divisor in A and 3h − A respectively. 6. Proof of the main Theorem In this section we will prove the main theorem stated in the introduction. We will first give the proof when c1 is not effective, then we will examine the case of effective c1. 6.1. The case of non -- effective c1. If s ∈ H 0(cid:0)F,E(cid:1) and (s)0 = E ∪ D, thanks to the previous Theorem 5.3, we know that D = 0 and E is a point. Lemma 6.1. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. If c1 is not effective, then c1 + h is effective. 14 GIANFRANCO CASNATI (13) Proof. If c1 is not effective, then D = 0. Twisting Sequence (6) by OF (−c1 + th), we obtain The cohomology of Sequence (13) with t = −1 yields 0 −→ OF (D − c1 + th) −→ E∨(th) −→ IEF (th − D) −→ 0. h0(cid:0)F,OF (c1 + h)(cid:1) = h2(cid:0)F,OF (−c1 − h)(cid:1) ≥ h1(cid:0)F,IEF (−h)(cid:1). yields h1(cid:0)F,IEF (−h)(cid:1) = 1. We also know that deg(E) = 1, hence the cohomology of Sequence (7) twisted by OF (−h) (cid:3) Assume that c1 = xh + yA. It follows from Lemmas 5.2 and 6.1 that either c1 = −h, or the inequalities y2 + 3(x + 1)y + (x + 1)2 ≥ 1, y2 − 3(6 − x)y + (6 − x)2 ≥ 1, 2(x + 1) + 3y ≥ 1, 2(6 − x) − 3y ≥ 1, 2A − h, 2h − A, − h, 5h − 2A, 7h − 3A. 3A − 2h, A − h, Proof. We know that the above three cases occur thanks to Examples 4.4 and 4.5. Looking at List (14) it remains to exclude the cases 3A− 2h, 2A− h, 5h− 2A, 7h− 3A. Recall that D = 0 and deg(E) = 1. must hold. We obtain a picture similar to Figure 1, with the bottom branch of hyperbola, shifted by −1 in the vertical direction. The admissible values of c1 correspond either to (−1, 0), or to the integral points (x, y) in the bounded region such that either y2+3xy+x2 ≤ 0 or 2x + 3y ≤ 0, because c1 is not effective. Thus c1 is in the following list: (14) Theorem 6.2. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. If c1 is not effective, then c1 ∈ { −h, A − h, 2h − A }. The zero locus E of a general s ∈ H 0(cid:0)F,E) is a point. Moreover all these cases occur on F , and E is never Ulrich. We have (3h− 2A)2 = −20, thus h0(cid:0)F,OF (3h− 2A)(cid:1) = h1(cid:0)F,OF (3h− 2A)(cid:1) = 0, hence h1(cid:0)F,OF (3h−2A)(cid:1) = 8 (see Equality (1)). Thus we obtain h1(cid:0)F,IEF (3h−2A)(cid:1) = 9 from 0 −→ H 1(cid:0)F,E(cid:1) −→ H 1(cid:0)F,IEF (3h − 2A)(cid:1) −→ H 2(cid:0)F,OF (cid:1), hence h1(cid:0)F,E) ≥ 8, a contradiction. One can argue similarly when c1 = 7h − 3A. Let c1 = 2A − h. We trivially have h0(cid:0)F,IEF (h)(cid:1) = 3 and h1(cid:0)F,IEF (h)(cid:1) = 0, hence H 0(cid:0)F,IEF (h)(cid:1) −→ H 1(cid:0)F,IEF (2h − 2A)(cid:1) −→ H 1(cid:0)F,E(h)(cid:1) −→ 0. Since (2h − 2A)2 = −16, as usual we deduce that h1(cid:0)F,IEF (2h − 2A)(cid:1) ≥ 6, hence h1(cid:0)F,E(h)(cid:1) ≥ 3 necessarily, a contradiction. The same argument excludes the case c1 = the cohomology of Sequence (7) twisted by OF (3h − 2A). The cohomology of Sequence (6) gives the exact sequence the cohomology of Sequence (6) twisted by OF (2h − 2A) gives the exact sequence 5h − 2A. Finally, notice that if c1 is not effective, then E is not globally generated, thus it cannot (cid:3) be Ulrich (see Remark 2.6). Remark 6.3. Let E be as in the statement above. We know that it fits into Sequence (6) with D = 0. Since h1(cid:0)F,OF (h)(cid:1) = h1(cid:0)F,OF (h − A)(cid:1) = h1(cid:0)F,OF (A − 2h)(cid:1) = 0 RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 15 it follows that such a bundle is necessarily unique once the point is fixed. Thus E actually coincides with the bundles constructed in Examples 4.4 and 4.5. Theorem 5.3, we know that c1 − D is effective, unless c1 = 2h and D (cid:54)= 0. Moreover equality holds if and only if E is Ulrich. If this is the case D = 0 for a general choice of s (see again Remark 2.6). 6.2. The case of effective c1. If s ∈ H 0(cid:0)F,E(cid:1) and (s)0 = E ∪ D, thanks to the previous Lemma 2.4 and Remark 2.6 imply that 1 ≤ h0(cid:0)F,E(cid:1) ≤ 8, because E is initialized. Assume D (cid:54)= 0. We have 1 ≤ h0(cid:0)F,E(cid:1) ≤ 7, then the cohomology of Sequence (6) and the effectiveness of D give h0(cid:0)F,OF (D)(cid:1) ≤ 7 and h2(cid:0)F,OF (D)(cid:1) = 0. Proposition 2.1 and Lemma 3.5 also give h1(cid:0)F,OF (D)(cid:1) = 0. It follows from Equality (1) that D2 ≤ 10. We deduce that D2 = 4 by Lemma 3.5 and Remark 3.3. If D ∈ xh + yA, then x2 + 3xy + y2 = 1. Moreover 2x + 3y ≥ 1 (see Remark 3.6), hence −3x + y = 5x2 + 4 . √ 2 Taking into account List (12), an analysis case by case of the inequality (c1 − D)h ≥ 2 shows that the possible cases with effective c1 − D are exactly the ones listed in the following table. Table A D − c1 + h D c1 6h 5h 3h − 6A, A − 2h 0, A, 3h − A −5h, A − 5h, −2h − A 6h − A 0, A, 3h − A A − 5h, 2A − 5h, −2h 0, 3h − A 6h − 2A 0, A, 3h − A −4h, A − 4h, −h − A 0, 3h − A 5h − A 0, A, 3h − A 4h − A 0, 3h − A 3h + A 0, A, 3h − A −2h − A, −2h, h − 2A 0, A, 3h − A −2h, A − 2h, h − A 3h − A 0, 3h − A 2h + A A − 4h, −h A − 3h, 0 A − 2h, h −h − A, −h −3h, A − 3h, −A 0, A 4h 3h 2h h + A h 2A A 0 0 0, A 0 0, A 0, A 0 −h −A, 0 0 h − 2A, h − A h − A, h 0 Is D − c1 + h effective? y/n. n, n, n n, n, n n, n n, n, n n, n n, n, n n, y n, n, n n, n, n n, y n, n n n, y y n, n n, y y The importance of the two last columns will be clear in a while. We know from Equality (1) that h0(cid:0)F,E(cid:1) + h2(cid:0)F,E(cid:1) = 4 + (15) h2(cid:0)F,E(−h)(cid:1) = 8 + Looking at Table A one deduces from Sequence (13) with t = 0 that h2(cid:0)F,E(cid:1) = 0 if − c1h − c2. − c2, c2 1 2 c2 1 2 (c1, D) is neither (3h − A, 3h − A), nor (A, A), nor (0, 0). 16 h2(cid:0)F,E(−h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) = 0 in all the listed cases but (c1, D) in the following short Looking at Table A one also deduces from Sequence (13) with t = 1 that if D (cid:54)= 0, then GIANFRANCO CASNATI list: (4h − A, 3h − A), (3h − A, 3h − A), (h + A, A), (A, A), (0, 0). 3h − A. Then exactly one of the following cases occurs: (16) Proposition 6.4. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. c1h ≤ 11, which turns out to be impossible by computing c1h from Table A. (i) the zero -- locus of a general section s ∈ H 0(cid:0)F,E(cid:1) has pure dimension 0; (ii) c1 = 2h and the zero locus of each non -- zero s ∈ H 0(cid:0)F,E(cid:1) is in either A, or Proof. Assume D (cid:54)= 0, so that 1 ≤ h0(cid:0)F,E(cid:1) ≤ 7. Let (c1, D) be not in List (16). By combining the above informations we obtain the equality h0(cid:0)F,E(cid:1) = c1h − 4, hence 5 ≤ In the case (A, A) we deduce h0(cid:0)F,E(cid:1) = 4, h2(cid:0)F,E(cid:1) = h0(cid:0)F,E∨(cid:1) = 1, h2(cid:0)F,E(−h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) = 4 Sequences (6) and (13). The first Equality (15) yields In the case (A+h, A) we deduce h2(cid:0)F,E(−h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) = 1 (from the cohomology of Sequence (7) twisted by OF (h) gives h1(cid:0)F,IEF (h)(cid:1) ≥ 3. Thus the cohomology of Sequence (6) implies h1(cid:0)F,E(cid:1) ≥ 3 because OF (A) is aCM. The case (4h − A, 3h − A) can ing to prove. c2 = 1, the second c2 = 0, a contradiction. The case (3h − A, 3h − A) can be excluded similarly. of Sequence (13) with t = 1). The second Equality (15) gives c2 = 7. The cohomology Thus we have to restrict to (c1, D) in List (16). In the case (0, 0) there is noth- be excluded similarly. (cid:3) in when c1 is effective. ∼= OX ⊗ OP3(α) for a suitable α ∈ Z. We are now ready to give the complete classification of the bundles we are interested For reader's benefit we recall that a closed subscheme X ⊆ P3 is aG (i.e. arithmetically Gorenstein) if it is aCM and ωX Closed subschemes in P3 of degree at most 2 and dimension 0 are always aG. This is no longer true when the degree is d ≥ 3. Anyhow, if the Hilbert function of the homogeneous coordinate ring SE of a 0-dimensional, aG, closed subscheme E ⊆ P3 is (1, n1, . . . , nσ−1, d, . . . ) with nσ−1 < d, then E is CB with respect to OP3(σ) (see [17], Theorem 1.1). Theorem 6.5. Let E be an initialized indecomposable aCM bundle of rank 2 on a general determinantal quartic surface F ⊆ P3. (i) c1 = 0: then c2 = 2. The zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional (ii) c1 = h: then c2 is 3, 4, or 5. The zero locus E of a general s ∈ H 0(cid:0)F,E) is a s ∈ H 0(cid:0)F,E) is 0 -- dimensional subscheme which is not aG. s ∈ H 0(cid:0)F,E) is 0 -- dimensional subscheme which is not aG and which is contained 0 -- dimensional aG subscheme of degree c2 of a linear subspace of P3 of dimension c2 − 2. (iii) c1 ∈ { A, 3h − A }: then c2 is 3, or 4. In this case the zero locus E of a general (iv) c1 ∈ { A + h, 4A − h }: then c2 = 8. In this case the zero locus E of a general If c1 is effective, then one of the following holds. aG subscheme contained in exactly one line. in exactly one pencil of quadrics. 17 RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 (v) c1 = 2h: then c2 = 8. In this case one of the two following cases occur: (a) the zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional aG subscheme (b) the zero locus E of a general s ∈ H 0(cid:0)F,E) is a 0 -- dimensional aG subscheme (c) the zero locus E of each non -- zero s ∈ H 0(cid:0)F,E) is a divisor in either A, or (vi) c1 = 3h: then c2 = 14. In this case the zero locus E of a general s ∈ H 0(cid:0)F,E) is lying on exactly one twisted cubic; 3h − A. which is the base locus of a net of quadrics; a 0 -- dimensional aG subscheme which is CB with respect to OF (3h). Moreover, all the above cases actually occur on F , and E is Ulrich if and only if c1 = 3h. Proof. Recall that the zero -- locus E of a general section of E has pure codimension 2, but possibly in the case c1 = 2h (see Proposition 6.4). With this restriction in mind, we will prove the statement by an analysis case by case with the help of Table A. We first assume that h − c1 is effective. Looking at the last column of Table A this is equivalent to c1 ∈ { 0, h }, because D = 0. Let c1 = 0. In this case E ∼= E∨. Computing the cohomology of Sequence (6) we obtain h0(cid:0)F,E(cid:1) = h2(cid:0)F,E(cid:1) = 1. The first equality (15) thus gives c2 = 2. Let c1 = h. In this case the cohomology of Sequence (6) implies As already pointed out, E is necessarily aG with Hilbert function (1, 1, . . . , 1, 2, . . . ). The homogeneous coordinate ring SE is generated in degree 1, thus there is only a single 1 in the sequence: in particular E is contained in exactly two linearly independent planes, hence in a single line. This case actually occurs as proved in Example 4.4. h0(cid:0)F,E(cid:1) = h0(cid:0)F,IEF (h)(cid:1) + 1 ≤ h0(cid:0)F,OF (h)(cid:1) + 1 = 5. Moreover h2(cid:0)F,E(cid:1) = h0(cid:0)F,E∨(cid:1) = h0(cid:0)F,E(−h)(cid:1) = 0. The first equality (15) thus gives Thus c2 ∈ { 1, 2, 3, 4, 5 }. If c2 = 1 (resp. 2), we would have h0(cid:0)F,IEF (h)(cid:1) = 4 (resp. c2 = 6 − h0(cid:0)F,E(cid:1) = 5 − h0(cid:0)F,IEF (h)(cid:1). 3). Nevertheless, in this case E is a point which is necessarily contained in at most three planes (resp. a closed subscheme of degree 2, which is necessarily contained in at most two planes). Thus we obtain a contradiction. When c2 ≥ 3, then E is contained in exactly 5 − c2 linearly independent planes, i.e. in a linear subspace of dimension c2 − 2 of P3. Since E must be CB with respect to OF (h), it follows that its closed subschemes of degree c2 − 1 are not contained in smaller linear [23], Lemma (4.2)). These cases occur as proved in subspaces, thus E is aG (see e.g. Example 4.3. From now on we will assume that h − c1, hence −c1, is not effective. Again we know from the last column of Table A which are the possible values of c1. The cohomology of Sequence (13) with t = 0 implies h2(cid:0)F,E(cid:1) = h0(cid:0)F,E∨(cid:1) = 0. We know that h0(cid:0)F,E(cid:1) ≤ 8. By combining this restriction with the Equalities (15) we obtain the folowing chain of inequalities c2 1 2 − 4 ≤ c2 = 8 + c2 1 2 − c1h − h2(cid:0)F,E(−h)(cid:1) ≤ 8 + − c1h, c2 1 2 (17) whence c1h ≤ 12. Looking at List (12) one can immediately exclude all the cases but A, h + A, 2A, 2h, 3h − A, 3h, 4h − A, 6h − 2A. We will now examine these cases. H := h2(cid:0)F,E(−h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) ≤ h0(cid:0)F,IEF (h)(cid:1) ≤ 4. Sequence (13) with t = 1 gives 18 GIANFRANCO CASNATI The above discussion proves the existence of an exact sequence of the form Let c1 = 2A. In this case Inequalities (17) force c2 = 4, hence H = 0. We will check that A2 = 4 and the elements of A are integral (see Remark 3.3). Notice that the coho- E is the intersection of two divisors in A, by checking that h0(cid:0)F,IEF (A)(cid:1) = 2: indeed mology of Sequence (6) twisted by OF (−A) gives h0(cid:0)F,IEF (A)(cid:1) = h0(cid:0)F,E(−A)(cid:1). Since h2(cid:0)F,E(−A)(cid:1) = h0(cid:0)F,E(−A)(cid:1), it follows that 2h0(cid:0)F,E(−A)(cid:1) = 4, thanks to Equality (1). h1(cid:0)F,OF (−2A)(cid:1) = 0. The same argument allows us to exclude the case c1 = 6h − 2A as Let c1 = A (the case c1 = 3h−A can be handled similarly), hence h0(cid:0)F,E(cid:1) = 6−c2 ≤ 6. well. Inequalities (17) imply H ∈ { 0, 1, 2, 3, 4 } and c2 = 4 − H. The case H = 4 cannot occur, because E (cid:54)= ∅ (see Lemma 5.1). The case H = 3 cannot occur too: indeed, in this case, E should be a point, thus it cannot be CB with respect to OF (A), because this line bundle is globally generated. besides Sequence (6). Thanks to Theorem 4.2 we deduce that E ∼= OF (A)⊕2, because 0 −→ OF −→ OF (A)⊕2 −→ IEF (2A) −→ 0, already pointed out above we have h0(cid:0)F,IEF (h)(cid:1) = 2. Thus, computing the cohomology Assume H = 2. In this case E is a 0 -- dimensional closed subscheme of degree 2. As of Sequence (13) for t = 0 and 1, one checks that E∨(h) is initialized. Moreover, it is also aCM and indecomposable (by hypothesis) because the same is true for E. We have c1(E∨(h)) = 2h − A and c2(E∨(h)) = 0. We thus deduce from Theorem 6.2 that this case cannot occur. The same argument when H = 1 implies that E∨(h) is initialized indecomposable aCM with c1(E∨(h)) = 2h − A and c2(E∨(h)) = 1. As pointed out at the end of Example 4.5 we can construct a bundle with all the above properties, thus such an E exists. Moreover, one easily checks h0(cid:0)F,IEF (h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) = 1, thus E is not contained in a line, properties, as shown in Example 4.6. We have h0(cid:0)F,IEF (h)(cid:1) = h0(cid:0)F,E∨(h)(cid:1) = 0 as well. Let H = 0. In this case c1 = A and c2 = 4. We can construct a bundle with these hence E is not aG in this case. Thus E is not contained in any plane, hence its Hilbert function is (1, 4, . . . ). Since every subscheme of E of degree 3 trivially lies in a plane, it follows that E is not CB with respect to OP3(1), hence it cannot be aG. Let c1 = h + A (the case c1 = 4h− A can be handled similarly). Inequalities (17) imply H ∈ { 0, 1, 2 } and c2 = 8 − H. Consider the exact sequence 0 −→ IFP3(t) −→ IEP3(t) −→ IEF (th) −→ 0 We have that OF (−A) is aCM and IFP3(t) ∼= OF (t − 4). Thus the cohomologies of Sequence (13) and of the above one for t = 0, 1, 2, imply that the Hilbert function of E is (1, c2 − 4, c2, . . . ). In particular we obtain (1, 4, 8, . . . ), (1, 3, 7, . . . ), (1, 2, 6, . . . ) if H = 0, 1, 2 respectively. Since SE is generated in degree 1, looking at the growth from degree 1 to 2, we deduce that the only admissible Hilbert function is the first one, hence H = 0. It follows that c2 = 8 and that E is contained in exactly one pencil of quadrics. In particular, E is not CB with respect to OP3(2), thus E cannot be aG. Moreover, h0(cid:0)F,E(cid:1) = 6. A bundle with In this case h2(cid:0)F,E(−h)(cid:1) = h0(cid:0)F,E(−h)(cid:1) = 0, thus Equality (1) for χ(E(−h)) yields c2 = 8, hence h0(cid:0)F,E(cid:1) = 4. The complete description and the existence these properties can be constructed as shown at the end of Example 4.6. Let c1 = 2h. of initialized, indecomposable aCM bundles of rank 2 on F with c1 = 2h and c2 = 8 RANK TWO ACM BUNDLES ON GENERAL DETERMINANTAL QUARTIC SURFACES IN P3 19 is in [8], Proposition 2.1 and Theorems 6.2, 7.1, 4.1, where the three family of bundles corresponding to the cases (a), (b), (c) are described in details. Notice that in all the above cases we showed that h0(cid:0)F,E(cid:1) ≤ 7, hence the considered E Let c1 = 3h. In this case H = 0 and c2 = 14, thus h0(cid:0)F,E(cid:1) = 8, thanks to the first are not Ulrich. Equality (15): hence E is Ulrich. We know from Theorem 1.1 of [9] the existence of such (cid:3) kind of bundles on each smooth quartic surface. Remark 6.6. Let E be as in the statement above. As in Remark 6.3 one easily checks that the bundles with c1 ∈ { 0, h, 2h, 3h } are uniquely determined by E, thus they actually coincide with the bundles constructed in Examples 4.4 and 4.3, or with the bundles described in [8] and [9] respectively. Similarly, when either c1 ∈ { A, 3h − A } and c2 = 3, or c1 ∈ { A + h, 4h − A }, then E is as at the end of Example either 4.5, or 4.6. Let us consider the case c1 = A more in details (analogous arguments hold in the cases 5 − c2 ≥ 1, thus there is a curve in C ∈ A containing E. If C is smooth, the cohomology of a suitable twist of the exact sequence c1 = 3h − A). The first Equality (15) implies h0(cid:0)F,E(cid:1) = 6 − c2, hence h0(cid:0)F,IEF (A)(cid:1) = ∼= OC(−E) yield h0(cid:0)C,OC(C − E)(cid:1) = 1. Riemann -- Roch theorem on C implies h0(cid:0)C,OC(E)(cid:1) = 2. Thus OC(E) is a complete g1 c2 = 8. As in the previous case one h0(cid:0)F,IEF (A)(cid:1) ≥ 1. and the isomorphisms ICF on C. In particular the bundle E is constructed exactly as in either Example 4.6, if c2 = 4, or at the end of Example 4.5, if c2 = 3. An analogous construction holds when c1 = A + h (or 4h − A). Indeed, in this case If there is a smooth curve 0 −→ ICF −→ IEF −→ IEC −→ 0, ∼= OF (−C), IEC C ∈ A + h, then we can argue as above that OC(E) is a complete g1 c2 8 on C. References [1] V. Ancona, G. Ottaviani: Some applications of Beilinson's theorem to projective spaces and quadrics. Forum Math. 3 (1991), 157 -- 176. [2] A. Beauville: Determinantal hypersurfaces. Michigan Math. J. 48 (2000), 39 -- 64. [3] D.A. Buchsbaum, D. Eisenbud: Algebra structures for finite free resolutions, and some structure theorems for ideals of codimension 3. Amer. J. Math. 99 (1977), 447 -- 485. [4] M. Casanellas, R. Hartshorne: ACM bundles on cubic surfaces. J. Eur. Math. Soc. 13 (2011), 709 -- 731. [5] M. Casanellas, R. Hartshorne, F. Geiss, F.O. Schreyer: Stable Ulrich bundles. Int. J. of Math. 23 1250083 (2012). [6] G. Casnati: On rank two aCM bundles. Preprint. [7] G. Casnati, F. Galluzzi: Stability of rank 2 Ulrich bundles on projective K3 surfaces. Work in progress. [8] G. Casnati, R. Notari: Examples of rank two aCM bundles on smooth quartic surfaces in P3. Preprint arXiv:1601.02907 [math.AG]. To appear in Rend. Circ. Mat. Palermo. [9] E. Coskun, R.S. Kulkarni, Y. Mustopa: Pfaffian quartic surfaces and representations of Clifford algebras. Doc. Math. 17 (2012), 1003 -- 1028. [10] E. Coskun, R.S. Kulkarni, Y. Mustopa: The geometry of Ulrich bundles on del Pezzo surfaces. J. Algebra 375 (2013), 280 -- 301. [11] I.V. Dolgachev: Classical algebraic geometry. A modern view. Cambridge University Press (2012). [12] D. Faenzi: Rank 2 arithmetically Cohen -- Macaulay bundles on a nonsingular cubic surface. J. Algebra 319 (2008), 143 -- 186. [13] R. Hartshorne: Algebraic geometry. G.T.M. 52, Springer (1977). [14] D. Huybrechts: appear: http://www.math.uni-bonn.de/people/huybrech/K3Global.pdf . surfaces. To Lectures on K3 draft notes available at 20 GIANFRANCO CASNATI [15] D. Huybrechts, M. Lehn: The geometry of moduli spaces of sheaves. Second edition. Cambridge Math- ematical Library, Cambridge U.P. (2010). [16] A. Knutsen: Smooth curves on projective K3 surfaces. Math. Scand. 90 (2002), 215 -- 231. [17] M. Kreuzer: On 0 -- dimensional complete intersections. Math. Ann. 292 (1992), 43 -- 58. [18] A. Lopez: Noether -- Lefschetz theory and the Picard group of projective surfaces. Memoirs A.M.S. 89 (1991). [19] C. Madonna: A splitting criterion for rank 2 vector bundles on hypersurfaces in P4. Rend. Sem. Mat. Univ. Pol. Torino 56 (1998), 43 -- 54. [20] J.C. Migliore: Introduction to liaison theory and deficiency modules. Progress in Mathematics 165, Birkhauser (1998). [21] C. Okonek, M. Schneider, H. Spindler: Vector bundles on complex projective spaces. Progress in Mathematics 3, (1980). [22] B. Saint -- Donat: Projective models of K -- 3 surfaces. Amer. J. Math. 96 (1974), 602 -- 639. [23] F.O. Schreyer: Syzygies of canonical curves and special linear series. Math. Ann. 275 (1986), 105 -- 137. [24] K. Watanabe: The classification of ACM line bundles on a quartic hypersurfaces on P3. Geom. Dedi- cata 175 (2015) 347 -- 353. [25] K. Watanabe: On the splitting of Lazarsfeld -- Mukai bundles on K3 surfaces. J. Algebra 447 (2016), 445 -- 454. Gianfranco Casnati, Dipartimento di Scienze Matematiche, Politecnico di Torino, c.so Duca degli Abruzzi 24, 10129 Torino, Italy e-mail: [email protected]
1810.11839	1	1810	2018-10-28T17:22:36	On homogeneous locally nilpotent derivations of trinomial algebras	[ "math.AG", "math.AC" ]	We provide an explicit description of homogeneous locally nilpotent derivations of the algebra of regular functions on affine trinomial hypersurfaces. As an application, we describe the set of roots of trinomial algebras.	math.AG	math	ON HOMOGENEOUS LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS SERGEY GAIFULLIN AND YULIA ZAITSEVA Abstract. We provide an explicit description of homogeneous locally nilpotent deriva- tions of the algebra of regular functions on affine trinomial hypersurfaces. As an applica- tion, we describe the set of roots of trinomial algebras. 1. Introduction Let K be an algebraically closed field of characteristic zero and R be an algebra over K. A K-linear map δ : R → R is called a derivation of the algebra R if it satisfies the Leibniz rule, that is δ(f g) = δ(f )g + f δ(g) for any f, g ∈ R. A derivation δ is said to be locally nilpotent if for any f ∈ R there exists k ∈ Z>0 such that δk(f ) = 0. Let K be an abelian group. An algebra R is said to be K-graded if R =Mw∈K Rw and Rw1Rw2 ⊆ Rw1+w2 for any w1, w2 ∈ K. Elements of Rw are called homogeneous of degree w ∈ K and Rw is called the homogeneous component of R of degree w. The weight monoid is the set S = {w ∈ K Rw 6= 0}, the weight cone is the convex cone in the rational vector space KQ = K ⊗Z Q generated by S. Every element of R has a decomposition into the sum of homogeneous elements, which are called its homogeneous components. An ideal I in R is said to be homogeneous if the homogeneous components of any element of I belong to I as well. In particular, any principal ideal generated by a homogeneous element is homogeneous. If I is a homogeneous ideal in R, then R/I inherits the grading from R. A derivation δ of R is called homogeneous if it maps homogeneous elements of R to homogeneous ones. In this case, there exists an element w0 ∈ K such that δ(Rw) ⊆ Rw+w0 for any w ∈ K. The element w0 is called the degree of δ. These notions have a geometric interpretation. Denote by X an irreducible algebraic variety over K and by Ga = (K, +) the additive group of the ground field. It is known that locally nilpotent derivations of the algebra K[X] of regular functions on X are in one- to-one correspondence with regular Ga-actions on X, see e.g. [5, Section 1.5]. Let H be a quasitorus, that is a direct product of a torus and a finite abelian group. Suppose that H acts on the variety X. Then we have a corresponding grading on the algebra K[X] by the group of characters K of the quasitorus H: K[X] =Mw∈K K[X]w, where K[X]w = {f h · f = w(h)f ∀h ∈ H}. 2010 Mathematics Subject Classification. Primary 13N15, 14R20; Secondary 13A50, 14J50. Key words and phrases. Affine hypersurface, torus action, graded algebra, derivation. The work was supported by the Foundation for the Advancement of Theoretical Physics and Mathematics "BASIS". 1 2 SERGEY GAIFULLIN AND YULIA ZAITSEVA It can be proved that a locally nilpotent derivation of K[X] is homogeneous with respect to this grading if and only if the quasitorus H normalizes the corresponding Ga-action on X. A description of homogeneous locally nilpotent derivations enables us to describe the automorphism group of an algebraic variety in some cases, see e.g. [2, Theorem 5.5], [1, Theorem 5.5]. Let us define a trinomial algebra. Fix n0, n1, n2 ∈ Z>0 and denote n = n0 + n1 + n2. For each i = 0, 1, 2, fix a tuple li = (li1, . . . , lini) ∈ Zni >0 and define a monomial T li i = T li1 i1 . . . T lini ini ∈ K[Tij i = 0, 1, 2, j = 1, . . . , ni]. A polynomial of the form g = T l0 2 ∈ K[Tij] is called a trinomial, the hypersurface X(g) given by {g = 0} in the affine space An is a trinomial hypersurface, and the algebra R(g) := K[Tij] / (g) of regular functions on X(g) is a trinomial algebra. 0 + T l1 1 + T l2 Our motivation to study trinomials comes from toric geometry. Consider an effective action T × X → X of an algebraic torus T on an irreducible variety X. The complexity of such an action is the codimension of a general T -orbit in X and equals dim X − dim T . Actions of complexity zero are actions with an open T -orbit. A normal variety admitting a torus action with an open orbit is called a toric variety. Let X be a toric variety. Then Ga-actions on X normalized by T (or, equivalently, locally nilpotent derivations of K[X] that are homogeneous with respect to the corresponding grading) can be described in terms of Demazure roots of the fan corresponding to X; see [4], [16, Section 3.4] for the original approach and [13, 3, 2] for generalizations. Let X admit a torus action of complexity one. A description of Ga-actions on X nor- malized by T (or homogeneous locally nilpotent derivations of K[X]) in terms of proper polyhedral divisors may be found in [14] and [13]. It is an interesting problem to find their explicit form in concrete cases. The study of toric varieties is related to binomials, see e.g. [18, Chapter 4]. At the same time, Cox rings establish a close relation between torus actions of complexity one and trinomials, see [10, 9, 8, 2, 7]. In particular, any trinomial hypersurface admits a torus action of complexity one. We consider the fine grading on a trinomial algebra R(g), it is defined in Construction 1. This grading corresponds to an effective action of a quasitorus on the trinomial hypersur- face X(g). The neutral component of the quasitorus is a torus of dimension n − 2, which acts on X(g) with complexity one. In this paper, homogeneous locally nilpotent derivations of a trinomial algebra R(g) are It is proved that they are elementary (Theorem 1), see Section 3 for defini- described. tions. In [2, Theorem 4.3]1, this statement was proved for primitive derivations, i.e. for homogeneous derivations with the degrees not belonging to the weight cone of R(g). Every primitive derivation is locally nilpotent; the converse is false (see [19, Example 7]). In [19], Theorem 1 was proved for some classes of trinomial algebras R(g), including non-factorial algebras R(g). In this paper, we use this result and obtain a description in a general case. Specifically, we reduce the remaining case to a description of homogeneous locally nilpotent derivations for X = {x + y + zk = 0}, see Proposition 4 and Lemma 9. The degrees of homogeneous locally nilpotent derivations are called the roots of the variety X. This definition imitates in some sense the notion of a root from Lie Theory. 1In [2], a more general class of affine varieties defined by a system of trinomials is studied, see [2, Construction 3.1] for details. LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 3 In [15], A. Liendo gives a description of roots of the affine Cremona group. This answers a question due to Popov, see [17]. The proof is based on the classification of locally nilpotent derivations in the case of torus actions of complexity one given in [13]. In [11], this result is obtained more directly. Theorem 1 enables us to find roots of X(g), see Section 6. Besides that, Theorem 1 gives a description of locally nilpotent derivations and roots in some important particular cases, see Examples 2, 4, and 3. The authors are grateful to Ivan Arzhantsev for useful discussions and comments and to the referee for careful reading of the paper and valuable suggestions. 2. Trinomial algebras In this section, the fine grading deg on a trinomial algebra is defined and some properties of homogeneous elements are proved. Construction 1. Fix n0, n1, n2 ∈ Z>0 and let n = n0 + n1 + n2. Consider the polynomial algebra K[Tij, i = 0, 1, 2, j = 1, . . . , ni] =: K[Tij]. For each i = 0, 1, 2, fix a tuple li = lini = (li1, . . . , lini) ∈ Zni ini ∈ K[Tij]. By a trinomial we mean a polynomial of the form >0 and define a monomial T li i = T li1 i1 . . . T g = T l0 0 + T l1 1 + T l2 2 . A trinomial hypersurface is the zero set X(g) = {g = 0} in the affine space An. It can be checked that the polynomial g is irreducible, hence the algebra R(g) := K[Tij] / (g) of regular functions on the trinomial hypersurface has no zero divisors. We call such algebras R(g) trinomial. We use the same notation for elements of K[Tij] and their projections to R(g). Following [8], we consider a (2 × n)-matrix L corresponding to the trinomial g as follows: L =(cid:18)−l0 −l0 0 l2(cid:19) . l1 0 Let L∗ be the transpose of L. Denote by K the factor group K = Zn / Im L∗ and by Q : Zn → K the canonical projection. Let eij ∈ Zn be the canonical basis vectors. The equalities deg Tij = Q(eij) (1) define a K-grading on the algebra K[Tij]. Note that the sums li1Q(ei1) + . . . + liniQ(eini) ∈ K are the same for i = 0, 1, 2. Hence g is a homogeneous polynomial and equalities (1) define a K-grading on R(g) = K[Tij] / (g) as well. We call this grading the fine grading and denote the degree with respect to this grading by deg. The derivations that are homogeneous with respect to the fine grading are called finely homogeneous. Example 1. Let all lij = 1, that is g = T01 . . . T0n0 + T11 . . . T1n1 + T21 . . . T2n2. Then L = (cid:18)−1 . . . −1 1 . . . 1 0 . . . 0 row and column operations to the form(cid:18)1 0 0 . . . 0 −1 . . . −1 0 . . . 0 1 . . . 1(cid:19). The matrix L can be reduced by elementary 0 1 0 . . . 0(cid:19), hence the group K = Zn / Im L∗ is isomorphic to Zn−2. Let us choose some basis {vij, (i, j) /∈ {(0, 1), (1, 1)}} of Zn−2. We 4 SERGEY GAIFULLIN AND YULIA ZAITSEVA can define the grading explicitly via deg T01 = − v0k + n0Xk=2 n2Xl=1 v2l, deg T11 = − v1k + n1Xk=2 n2Xl=1 v2l, and deg Tij = vij otherwise. The following lemma states that the constructed grading deg is the finest grading on the algebra R(g) such that all generators Tij are homogeneous. It follows easily from the definition of the grading deg. Lemma 1. Let deg be the fine grading on the algebra R(g) defined in Construction 1 and deg′ be any grading on R(g) by some abelian group such that all generators Tij are homogeneous. Then deg′ = ψ ◦ deg for some homomorphism ψ : K → K ′. In particular, any derivation that is homogeneous with respect to the fine grading deg is homogeneous with respect to the grading deg′ as well. For any algebra R graded by an abelian group, one can consider the subalgebra of degree zero fractions inside the field of fractions: Q(R)0 =(cid:26) f g (cid:12)(cid:12)(cid:12)(cid:12) f, g ∈ R, homogeneous, deg f = deg g, g 6= 0(cid:27) ⊆ Q(R). The following lemma is proved in [2, Proposition 3.4]. Lemma 2. Take any i, j ∈ {0, 1, 2}, i 6= j. Then the field of degree zero fractions of R(g) is the rational function field j ! . Q(R(g))0 = K T li T lj i For any w ∈ K, denote by Rw the finely homogeneous component of R(g) of degree w. This follows from the proof of [2, Proposition 3.5]. ij geneous polynomial F in two variables and some uij ∈ Z>0. Lemma 3. Let h ∈ R(g) be homogeneous. Then h =ehF (T l0 polynomial F in two variables and someeh ∈ R ew with dim R ew = 1. Lemma 4. Let h ∈ R(g) be homogeneous. Then h =Q T uij Proof. Let eh from Lemma 3 be equal to eh = monomials in Tij and deg hk = degeh = ew. Then dim R ew = 1 implieseh = h1. hypersurfaces (in notation of [2] that is r = 2, A = (cid:0) 0 −1 1 3. Elementary derivations R(g) = R(A, P0)). sPk=1 1 −1 0(cid:1), g = gI for I = {0, 1, 2}, The following construction is given in [2] and is described below in the case of trinomial 0 , T l1 1 ) for some homogeneous · F (T l0 0 , T l1 1 ) for some homo- hk, where hk 6= 0 are non-proportional (cid:3) Construction 2. Let us define a derivation δC,β of R(g). The input data are • a sequence C = (c0, c1, c2), where ci ∈ Z, 1 6 ci 6 ni; • a vector β = (β0, β1, β2) such that βi ∈ K, β0 + β1 + β2 = 0. It is clear that for β 6= 0 as above either all entries βi differ from zero or there is a unique i0 with βi0 = 0. According to these cases, we put further conditions and define: LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 5 (i) if βi 6= 0 for all i = 0, 1, 2 and there is at most one i1 with li1ci1 > 1, then we set (ii) if βi0 = 0 for a unique i0 and there is at most one i1 with i1 6= i0 and li1ci1 > 1, then 0, βiQk6=i δC,β(Tij) = δC,β(Tij) = βi Qk6=i,i0 0, lk k ∂T ∂Tkck , j = ci, j 6= ci. lk k ∂T ∂Tkck , j = ci, j 6= ci. These assignments define a map δC,β on the generators Tij. It can be extended uniquely to a derivation on K[Tij] by Leibniz rule. One can check that δC,β(g) = 0, whence constructed map induces a well-defined derivation of the factor algebra R(g). Lemma 5. Every derivation of the form δC,β of the algebra R(g) is homogeneous and locally nilpotent. The proof is given in [2, Construction 4.2]. Let h ∈ R(g) be a homogeneous element in the kernel of a derivation δC,β. The deriva- tion hδC,β is also locally nilpotent. Definition 1. We say that a derivation of a trinomial algebra R(g) is elementary if it has the form hδC,β, where h is a homogeneous element in the kernel of δC,β. In addition, elementary derivations of Type I are elementary derivations with δC,β corresponding to case (i) in Construction 2; elementary derivations of Type II are elementary derivations with δC,β corresponding to case (ii). Proposition 1. Let δC,β be finely homogeneous locally nilpotent derivation of R(g) defined in Construction 2. Then h is a homogeneous element of Ker δC,β if and only if h = α YTij ∈Ker δC,β T uij ij ·(cid:0)β1T l0 0 − β0T l1 , 1 (cid:1)m for some uij, m ∈ Z>0, α ∈ K. Proof. Let h be homogeneous, h ∈ Ker δC,β. According to Lemma 4, for some homogeneous polynomial F in two variables and some uij ∈ Z>0. Every homoge- neous polynomial in two variables over an algebraically closed field can be decomposed in a product of linear polynomials. Hence h =Y T uij ij · F (T l0 0 , T l1 1 ) h =Y T uij ij ·Yk (cid:0)ζkT l0 0 + ξkT l1 1 (cid:1) for some ζk, ξk ∈ K. The kernel of a locally nilpotent derivation is factorially closed (see, for example, [5, Principle 1]). Thus h ∈ Ker δC,β implies that all Tij with uij > 0 in the first product and 0 + ξkT l1 all(cid:0)ζkT l0 1 (cid:1) in the second product belong to Ker δC,β as well. By definition of δC,β, 0 = δC,β(cid:0)ζkT l0 (ζkβ0 + ξkβ1)Qi (ζkβ0 + ξkβ1) Qi6=i0 li ∂T i ∂Tici li ∂T i ∂Tici in the case (ii), in the case (i), 0 + ξkT l1 1 (cid:1) = 6 SERGEY GAIFULLIN AND YULIA ZAITSEVA whence ζkβ0 + ξkβ1 = 0. Consequently any pair (ζk, ξk) is proportional to (β1, −β0). This yields that h has the required form. Conversely, any h of such a form is homogeneous and belongs to Ker δC,β, since all factors (cid:3) are of this kind. Example 2. Suppose that all lij = 1, that is g = T01 . . . T0n0 + T11 . . . T1n1 + T21 . . . T2n2 (see Example 1). Let us find all elementary derivations of R(g). For any C = (c0, c1, c2), 1 6 ci 6 ni, we have lici = 1, so both cases of Construction 2 are possible. (i) Let β = (β0, β1, β2), β0 + β1 + β2 = 0, βi 6= 0 for all i = 0, 1, 2. Then we have 2 /T1c1T2c2 and two analogous formulas for T1c1 and T2c2, hence 1 T l2 2 δC,β(T0c0) = β0T l1 1 T l2 T l0 0 T l1 ∂ ∂ δC,β = T0c0T1c1T2c2(cid:18)β0 T0c0 T l0 0 + β1 T1c1 T l1 1 ∂T0c0 + β2 ∂T1c1 T2c2 T l2 2 ∂ ∂T2c2(cid:19) . (ii) Consider for example the case i0 = 2, that is β = (β0, −β0, 0). Then we get δC,β(T0c0) = β0T l1 1 /T1c1 and the same formula for T1c1, hence δC,β = β0 0 T l1 T l0 1 T0c0T1c1(cid:18) T0c0 T l0 0 ∂ ∂T0c0 − T1c1 T l1 1 ∂ ∂T1c1(cid:19) . Thus we have n0n1n2 + n0n1 + n1n2 + n2n0 classes of δC,β. Applying Proposition 1 we obtain that every elementary derivation δ has the form δ = hδC,β, where a homogeneous element h belongs to the kernel of corresponding δC,β. In the above cases, an elementary derivation δ is of the following forms for some uij, m ∈ Z>0: δ = Y(i,j)6=(i,ci) T uij ij 0 − β0T l1 ·(cid:0)β1T l0 δ = α Y(i,j)6=(0,c0) (i,j)6=(1,c1) T uij ij 1 T l2 2 0 T l1 T0c0T1c1T2c2(cid:18)β0 1 (cid:1)m T l0 1 (cid:1)m T l0 ·(cid:0)T l0 0 + T l1 T0c0 T l0 0 ∂ ∂T0c0 + β1 T1c1 T l1 1 ∂ ∂T1c1 + β2 0 T l1 1 T0c0T1c1(cid:18)T0c0 T l0 0 ∂ ∂T0c0 − T1c1 T l1 1 ∂ ∂T1c1(cid:19) . T2c2 T l2 2 ∂ ∂T2c2(cid:19) ; The following theorem is the main result of the paper. 4. Main results Theorem 1. Every finely homogeneous locally nilpotent derivation of a trinomial algebra is elementary. For example, all finely homogeneous locally nilpotent derivations of the trinomial alge- bra R(g) for g = T01 . . . T0n0 + T11 . . . T1n1 + T21 . . . T2n2 are described in Example 2. Let us start to prove Theorem 1. The following lemma includes basic properties of locally nilpotent derivations. The proof can be found, for example, in [5, Principle 5 and Corollary 1.20]. Lemma 6. Suppose δ : R → R is a locally nilpotent derivation of a domain R and f, g ∈ R. (a) If f δ(f ), then δ(f ) = 0. (b) If f δ(g) and g δ(f ), then δ(f ) = 0 or δ(g) = 0. Definition 2. Let δ be a derivation of a trinomial algebra R(g). We say that a variable Tij is a non-kernel variable if δ(Tij) 6= 0. The exponent lij of such variable is called a non-kernel exponent. LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 7 The following lemma is proved in [6, Lemma 3.4]. For convenience of the reader we give a short proof below. Lemma 7. Let δ be a finely homogeneous locally nilpotent derivation of an algebra R(g). Then there exists at most one non-kernel variable in every monomial T li i . Proof. Assume the converse. Then there is a monomial with at least two non-kernel vari- ables. We can assume that these variables are T01 and T02 in T l0 0 . Consider the following Z-grading on the algebra K[Tij]: deg′ T01 = l02, deg′ T02 = −l01, deg′ Tij = 0 for all other (i, j). The trinomial g is homogeneous (of degree 0) with respect to this grading, therefore the deg′-grading is a well-defined grading on the factor algebra R(g). By Lemma 1, it follows that the derivation δ is deg′-homogeneous. We have the following two cases. 1) deg′ δ > 0. Then deg′ δ(T01) = deg′ δ + deg′ T01 > 0. Note that T01 is a unique variable with a positive degree. Hence every monomial in δ(T01) includes T01 and therefore T01 divides δ(T01). But δ(T01) 6= 0 by assumption. This contradicts Lemma 6(a). 2) deg′ δ 6 0. Then deg′ δ(T02) = deg′ δ + deg′ T02 < 0. In the same way, T02 divides (cid:3) δ(T02) and this contradicts Lemma 6(a). Proof of the following Proposition is given in [19, Proposition 1]. Proposition 2. Let δ : R(g) → R(g) be a finely homogeneous locally nilpotent derivation. Suppose that δ(T l0 0 ), δ(T l1 2 ) lie in a subspace of R(g) of dimension 1. Then the derivation δ is elementary. 1 ), and δ(T l2 The following proposition is in fact proved in [19, Theorem 1]. However, the required formulation is stronger than in [19, Theorem 1], and for convenience of the reader a proof is given below. Proposition 3. Let δ be a finely homogeneous locally nilpotent derivation of a trinomial algebra R(g). Let at most one non-kernel exponent be equal to 1. Then δ(T l0 1 ), and δ(T l2 2 ) lie in a subspace of R(g) of dimension 1. 0 ), δ(T l1 Proof. According to Lemma 7, there are at most three non-kernel variables. Suppose that 1 ), and δ(T l2 there exist at most two non-kernel variables. Then at least one of δ(T l0 2 ) is equal to 0, and the assertion follows from δ(g) = δ(0) = 0 in R(g). Hence we can assume without loss of generality that there are three non-kernel variables T01, T11, T21 with non-kernel exponents l01 > 1, l11 > 1, l21. 0 ), δ(T l1 Denote by Zm = {[0]m, [1]m, . . . , [m − 1]m} the cyclic group of order m. Consider the following Zm-grading on K[Tij]: deg′ T01 = [1]l01, deg′ Tij = [0]l01 for all other (i, j). Since g is a homogeneous polynomial with respect to this grading (of degree [0]l01), we get a well-defined grading deg′ on R(g). If deg′ δ(T01) 6= [0]l01 then T01 divides δ(T01). Indeed, this inequality implies that the degrees of all monomials in δ(T01) do not equal [0]l01, and it is possible only if every monomial includes T01 (since T01 is a unique variable with a nonzero degree). On the other hand, the fact that T01 divides δ(T01) together with δ(T01) 6= 0 contradicts Lemma 6(a). Thus deg′ δ(T01) = [0]l01. 8 SERGEY GAIFULLIN AND YULIA ZAITSEVA Then deg′ δ = deg′ δ(T01) − deg′ T01 = [l01 − 1]l01 It follows that deg′ δ(T11) = deg′ δ + deg′ T11 6= [0]l01. Arguing as above, we conclude that T01 divides δ(T11). 6= [0]l01 (recall that l01 > 1). Similarly, T11 divides δ(T01). This contradicts Lemma 6(b). (cid:3) The proof of the following proposition is given in the next section. Proposition 4. Let δ be a finely homogeneous locally nilpotent derivation of a trinomial algebra R(g). Let at least two non-kernel exponents be equal to 1. Then δ(T l0 1 ), and δ(T l2 2 ) lie in a subspace of R(g) of dimension 1. 0 ), δ(T l1 Proof of Theorem 1. The result follows from Propositions 3, 4, and 2. (cid:3) 5. Proof of Proposition 4 Recall that an element of polynomial algebra v ∈ K[y, z] is a coordinate if there is an element u ∈ K[y, z] such that K[u, v] = K[y, z]. The proof of the following lemma can be found, for example, in [5, Corollary 4.7]. Lemma 8. Let ∂ be a derivation of K[y, z]. Then ∂ is locally nilpotent if and only if ∂ is of the form ∂( · ) = J( · , g), where J denotes the Jacobian and g ∈ K[v] for some coordinate v of K[y, z]. First let us prove the simplest case of Proposition 4. Lemma 9. Let g = x + y + zk. Let ∂ be a locally nilpotent finely homogeneous derivation of the algebra R(g). Then ∂(x), ∂(y), and ∂(zk) lie in a subspace of R(g) of dimension 1. Proof. The fine grading in this case is given by deg x = k, deg y = k, deg z = 1. The variety {x + y + zk = 0} in K3 is isomorphic to the plane K2 with coordinates y and z, the isomorphism is given by x = −y − zk. Hence we have the corresponding locally deg y = k, deg z = 1. to the grading nilpotent derivationb∂ of the polynomial algebra K[y, z], which is homogeneous with respect By Lemma 8, b∂ has the form b∂( · ) = J( · , g(v)), where g is a polynomial in one variable over K and v is a variable of K[y, z]. One can prove that a derivation J( · , u) is homogeneous if and only if u is a sum of a homogeneous element of K[y, z] and a constant from K. Since the constant does not affect J( · , u), we can suppose that g(v) is a homogeneous element of K[y, z] with respect to the grading (2). It is easily shown that g(v) is homogeneous as an element of K[y, z] if and only if g is homogeneous as a polynomial and v is homogeneous as an element of K[y, z]. Hence g(v) = ηvm for some η ∈ K, m ∈ Z>0, and v ∈ K[y, z] is homogeneous with respect to (2). (2) On the other side, v is irreducible as a polynomial in y and z since it is a variable of K[y, z]. It can be proved that irreducible polynomials in K[y, z], which are homogeneous with respect to (2), have the form ζy + ξzk, ζ, ξ ∈ K. Thus we get: where a, b, c ∈ K, m ∈ Z>0. b∂( · ) = J( · , g(v)) = J( · , ηvm) = J( · , η(ζy + ξzk)m), LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 9 Let us calculate b∂(y) and b∂(zk). Note that for any f ∈ K[y, z] ξkzk−1(cid:19) , b∂(f ) = J(f, ηvm) = ηmvm−1J(f, v) = ηmvm−1J(f, ζy + ξzk) = ηmvm−1 det(cid:18)fy whence we get b∂(y) = ηξmkvm−1zk−1 and b∂(zk) = −ηζmkvm−1zk−1. Thus we prove that b∂(y) and b∂(zk) are proportional over K in K[y, z]. Since ∂(y) = b∂(y), ∂(z) = b∂(z), ∂(x) = = −b∂(y) −b∂(zk), we conclude that ∂(x), ∂(y), and ∂(zk) are proportional in R(g). Now let us reduce Proposition 4 to the previous lemma. Proof of Proposition 4. Assume that there exist at most two non-kernel variables. Then at least one of δ(T l0 2 ) is equal to 0, and δ(g) = δ(0) = 0 implies the assertion. Thus without loss of generality it can be assumed that there are three non-kernel variables T01, T11, and T21 with non-kernel exponents l01 = l11 = 1, l21 = k. 0 ), δ(T l1 1 ), δ(T l2 fz (cid:3) ζ Let L be the algebraic closure of the field obtained from K by adding all variables from the kernel of δ: 0 + T l1 L = K(T02, . . . , T0n0, T12, . . . , T1n1, T22, . . . , T2n2). 1 + T l2 Consider {T l0 2 = 0} as an algebraic variety over L in 3-dimensional space with coordinates T01, T11, and T21. Denote L[Ti1] := L[T01, T11, T21]. Since δ(Tij) = 0 for j 6= 1, we can consider δ as a derivation of L[Ti1] / (g). Indeed, a map δ defined by the same δ(Tij) ∈ K[Tij] / (g) ⊆ L[Ti1] / (g) and extended to a map L[Ti1] → L[Ti1] / (g) by Leibniz rule, induces a well-defined derivation L[Ti1] / (g) → L[Ti1] / (g) since δ(g) = 0. Clearly, this δ is also locally nilpotent. Lemma 10. δ(T l0 2 ) are proportional over L. 1 ), and δ(T l2 0 ), δ(T l1 Proof. After a linear coordinate transformation x = T01 · T l02 y = T11 · T l12 z = T21 · kqT l22 22 . . . T 02 . . . T 12 . . . T l0n0 0n0 = T l0 0 l1n1 1n1 = T l1 1 l2n2 2n2 = kqT l2 2 we obtain the hypersurface {x + y + zk = 0} ⊆ L3 and the corresponding locally nilpotent derivation ∂. Notice that deg′ T01 = k, deg′ T11 = k, deg′ T21 = 1, deg′ Tij = 0 for all other (i, j) is a well-defined Z-grading on K[Tij] / (g) since g is homogeneous (of degree k). By Lemma 1, δ is homogeneous with respect to this grading, whence ∂ is homogeneous with respect to deg′ x = k, deg′ y = k, deg′ z = 1. Applying Lemma 9 to the derivation ∂ and the ground field L, we obtain that ∂(x), ∂(y), and ∂(zk) lie in a subspace of L[x, y, z] / (x + y + zk) of dimension 1. For the derivation δ it means that δ(T l0 2 ) lie in a subspace of L[Ti1] / (g) of dimension 1. (cid:3) 1 ), and δ(T l2 0 ), δ(T l1 Fix any s, t ∈ {0, 1, 2}, s 6= t. According to Lemma 10, there exists ℓ ∈ L such that δ(T ls s ) = ℓ δ(T lt t ) in L[Ti1] / (g). (3) Then δ(T ls s ) δ(T lt t ) = ℓ in Q (L[Ti1] / (g)) . 10 SERGEY GAIFULLIN AND YULIA ZAITSEVA On the other hand, deg δ(T ls s ) = deg δ(T lt t ). By Lemma 2 it follows that 2 (cid:19) in Q (K[Tij] / (g))0 ⊆ Q (L[Ti1] / (g)) . δ(T ls s ) δ(T lt t ) 1 T l2 ∈ K(cid:18) T l1 v(cid:17) can be written in the form Every element of K(cid:16)u = ℓ f1(T l1 f2(T l1 2 ) − ℓf2(T l1 2 ) − ℓf2(T l1 1 , T l2 2 ) 1 , T l2 2 ) 1 , T l2 2 ) = 0 1 , T l2 2 ) = g · h f1(T l1 f1(T l1 1 , T l2 1 , T l2 Since the left side does not depend on T01, f1(T l1 1 , T l2 2 ) − ℓf2(T l1 1 , T l2 2 ) = 0 f1(u, v) f2(u, v) , where f1, f2 ∈ K[u, v]. Whence in Q (L[Ti1] / (g)) ; in L[Ti1] / (g); in L[Ti1] for some h ∈ L[Ti1]. in L[Ti1], which follows that f1 and f2 are proportional with coefficient ℓ as polynomials in two variables. At the same time f1, f2 are polynomials over K. Thus we get that ℓ ∈ K in (3) and (3) holds in K[Tij] / (g). This concludes the proof of Proposition 4. (cid:3) 6. Roots of trinomial algebras Let a quasitorus H act on an algebraic variety X. Consider the corresponding grading on K[X]. Recall that a root of X is the degree of a homogeneous locally nilpotent derivation of K[X]. Proposition 5. Let R(g) = K[X(g)] be a trinomial algebra. For any a, b ∈ {0, 1, 2}, a 6= b, 1 6 ca 6 na, and 1 6 cb 6 nb such that lbcb = 1, denote E(Taca, Tbcb) =(cid:26)deg g − deg Taca − deg Tbcb + X(i,j)6=(a,ca) Then e ∈ K is a root of X(g) if and only if e ∈S E(Taca, Tbcb). (i,j)6=(b,cb) uij deg Tij (cid:12)(cid:12)(cid:12)(cid:12) uij ∈ Z>0(cid:27) ⊆ K. Proof. According to Theorem 1, we have to find the degrees of elementary derivations. Note that deg Tici in case (i) of Construction 2 in case (ii) of Construction 2 Every elementary derivation δ has the form hδC,β, where h is a homogeneous element uij deg Tij + m deg g for some uij, m ∈ Z>0. deg Tici 2 deg g −Pi deg g − Pi6=i0 deg δC,β = deg δ = 2 deg g −Pi deg g − Pi6=i0 of Ker δC,β. By Proposition 1, deg h = PTij ∈Ker δC,β deg Tici + PTij ∈Ker δC,β deg Tici + PTij ∈Ker δC,β Hence uij deg Tij + m deg g in case (i) uij deg Tij + m deg g in case (ii) (4) We claim that for any elementary δ = hδC,β of Type I there is δ′ = h′δC ′,β ′ of Type II with deg δ = deg δ′. Indeed, for any C = (c0, c1, c2), β satisfying the conditions of case (i), take LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 11 any i0 ∈ {0, 1, 2} and β′ = (β′ C ′ = C, β′ satisfies the conditions of case (ii), and for δ1 = T 2) 6= 0 with β′ i0 = 0, β′ 0, β′ 1, β′ 1 + β′ 0 + β′ li0 i0 /Ti0ci0 · δC ′,β ′ we have 2 = 0. Then the data deg Tici = 2 deg g −Xi deg Tici = deg δC,β. li0 i0 , and Tij ∈ Ker δC,β belong to Ker δC ′,β ′. Replacing the factor of , we get the element h1 ∈ Ker δC ′,β ′ of degree deg h. Then li0 Note that T li0 i0 /Ti0ci0 , T deg δ1 = deg g − deg Ti0ci0 + deg g −Xi6=i0 degree m deg g in h by(cid:16)T It remains to note that deg g =Pj i0 (cid:17)m We call a set of the form E(Taca, Tbcb) a basic set. δ′ = h1δ1 is elementary of Type II and deg δ′ = deg δ, as required. li0j deg Ti0j and set {a, b} = {0, 1, 2} \ {i0}. (cid:3) Corollary 1. For any e ∈ E(Taca, Tbcb) there exists locally nilpotent derivation δ with non- kernel variables Taca, Tbcb and deg δ = e. Proof. The statement follows from (4) and Proposition 1. (cid:3) The following lemma states that the degree of a locally nilpotent derivation determines a unique variable in every monomial T li i that can be non-kernel. Lemma 11. Let deg δ = deg δ′, Tic /∈ Ker δ, Tic′ /∈ Ker δ′. Then c = c′. Proof. Suppose that c 6= c′. Without loss of generality we can assume that c = 1, c′ = 2, i = 0. Consider the following Z-grading on the algebra K[Tij]: deg′ T01 = l02, deg′ T02 = −l01, deg′ Tij = 0 for all other (i, j). The trinomial g is homogeneous (of degree 0) with respect to this grading, therefore the deg′-grading is a well-defined grading on the factor algebra R(g). By Lemma 1, it follows that δ and δ′ are deg′-homogeneous and deg′ δ = deg′ δ′. Let deg′ δ > 0, then deg′ δ(T01) > 0. Note that T01 is a unique variable with a positive degree. Hence every monomial in δ(T01) includes T01 and therefore T01 divides δ(T01). But δ(T01) 6= 0 since T01 is non-kernel. This contradicts Lemma 6(a). Thus, deg′ δ < 0. By the same reason, deg′ δ′ > 0, but they are equal, a contradiction. (cid:3) Proposition 6. Every e ∈ K belongs to at most three basic sets. Proof. Let e ∈ E(Taca, Tbcb). According to Corollary 1, there exists locally nilpotent deriva- tion δ with deg δ = e and Taca, Tbcb /∈ Ker δ. By Lemma 11 there is at most one variable Tici in every monomial T li i that can be non-kernel for some locally nilpotent derivation with the degree e. Then e can belong only to E(T0c0, T1c1), E(T0c0, T2c2), E(T1c1, T2c2). (cid:3) Remark 1. It follows from the proof of Proposition 5 that any degree of a derivation of Type I belongs to all three possible basic sets. The converse is false, see Example 4. In [12], homogeneous locally nilpotent derivations for quadrics with a complexity one torus action are described. This description is obtained by applying the technique of proper polyhedral divisors (see [14] and [13]). We obtain the following statements from Theorem 1 and Proposition 1 (Examples 3, 4). 12 SERGEY GAIFULLIN AND YULIA ZAITSEVA Example 3. Every finely homogeneous locally nilpotent derivation of the algebra R(g), where g = T01T02 + T11T12 + T21T22, has the form 0i0T k1 T k0 1i1T k2 2i2(α2T11T12 − α1T21T22)p(cid:18)α0T1i1T2i2 ∂ ∂T0¯i0 + α1T0i0T2i2 ∂ ∂T1¯i1 + α2T0i0T1i1 ∂ ∂T2¯i2(cid:19) , where k0, k1, k2, p ∈ Z>0, {i0,¯i0} = {i1,¯i1} = {i2,¯i2} = {1, 2}, α0, α1, α2 ∈ K, and α0 + α1 + α2 = 0. Example 4. Every finely homogeneous locally nilpotent derivation of the algebra R(g), where g = T01T02 + T11T12 + T 2 21, has the form ∂ λT k 21(cid:18)T1j 1j(α1T01T02 − α0T11T12)p(cid:18)α0T1jT21 1jT p 0iT l ∂T1¯i ∂ ∂T0¯i T k 0iT l − T0i ∂ ∂T1¯j(cid:19) or + α1T0iT21 − ∂ ∂T1¯j α0 + α1 2 T0iT1j ∂ ∂T21(cid:19) , where k, l, p ∈ Z>0, {i,¯i} = {j, ¯j} = {1, 2}, α0, α1, λ ∈ K. Remark 2. There is no factor (T01T02 + T11T12)k in the first case since it can be written as (−1)kT 2k 21 . z z y y x x Figure 1. A ¾cone¿ E(T01, T11) Figure 2. A ¾cone¿ E(T01, T21) z z z y x x y x y Figure 3. A ¾layer¿ Figure 4. All roots Figure 5. Type I LOCALLY NILPOTENT DERIVATIONS OF TRINOMIAL ALGEBRAS 13 Let us describe the roots in Example 4. The grading group K is isomorphic to Z3, and the grading can be given explicitly via deg T01 = 1 0 1 , deg T02 = −1 0 1  , deg T11 = 0 1 1 , deg T12 = 0 −1 1  , deg T21 = 1 . 0 0 By applying Proposition 5, one can find the set of roots of the quadric. Namely, it is the union of eight basic sets. Four ¾corner¿ basic sets of the form E(T0c0, T1c1) consist of all integer points in cones with vertices at (±1, ±1, 0), one of them is shown in Figure 1. Four ¾lateral¿ basic sets of the form E(T0c0, T21) and E(T1c1, T21) consist of integer points with odd sum of coordinates in cones with vertices at (±1, 0, 0) and (0, ±1, 0), one of them is shown in Figure 2. All roots in ¾lateral¿ basic sets except four ¾layers¿ belong to ¾corner¿ basic sets as well (see Figure 3). Thus, the set of all roots consists of four ¾corner¿ basic sets and four ¾layers¿ between them, it is shown in Figure 4. Note that every root belongs to at most three basic sets, this agrees with Proposition 6. The degrees of derivations of Type I are the integer points in four 2-dimensional cones with vertices at (±1, ±1, 1), one of them is shown in Figure 5. Note that every point in this set belongs to three basic sets, but there are other points that belong to three basic sets. This agrees with Remark 1. Let us give one more application of Theorem 1. It is known that for a commutative K-domain R and locally nilpotent derivations δ1, δ2 on R the condition Ker δ1 = Ker δ2 implies h1δ1 = h2δ2 for some h1, h2 ∈ R; see [5, Principle 12]. For trinomial algebras we can prove a more precise statement. Proposition 7. Let R(g) be a trinomial algebra and δ1, δ2 be finely homogeneous locally nilpotent derivations of R(g) with Ker δ1 = Ker δ2. Then δi = hiδ for some locally nilpotent derivation δ of R(g) and h1, h2 ∈ R(g). Proof. By Theorem 1, δ1 and δ2 are elementary, that is they are of the form hδC,β for some h ∈ R(g). Since h does not affect the kernel of hδC,β, it is sufficient to prove that δC,β is defined by Ker δC,β up to a constant from K. First note that the set of variables Tij with δC,β(Tij) = 0 uniquely determines the se- quence C and the type of the derivation, see Construction 2. 0 − β0T l1 1 ∈ Ker δC,β. Let any other α1T l0 Now let us prove that β = (β0, β1, β2) is defined by Ker δC,β up to a constant. According to Proposition 1, we have β1T l0 1 belong to Ker δC,β. Suppose that the pairs (α0, α1) and (β0, β1) are non-proportional. Then T l0 0 , T l1 1 are linear combinations of the above elements from Ker δC,β and hence belong to Ker δC,β. Since Ker δC,β is factorially closed, this implies that Tij ∈ Ker δC,β for any i, j, a contra- diction. Thus α1T l0 1 belongs to Ker δC,β if and only if (α0, α1) is proportional to (β0, β1). Applying β2 = −β0 − β1 yields that (β0, β1, β2) is determined by Ker δC,β up to a constant. (cid:3) 0 − α0T l1 0 − α0T l1 References [1] Ivan Arzhantsev and Sergey Gaifullin. The automorphism group of a rigid affine variety. Math. Nachrichten 290 (2017), no. 5-6, 662-671 [2] Ivan Arzhantsev, Jurgen Hausen, Elaine Herppich, and Alvaro Liendo. The automorphism group of a variety with torus action of complexity one. Moscow Math. J. 14 (2014), no. 3, 429-471 14 SERGEY GAIFULLIN AND YULIA ZAITSEVA [3] Ivan Arzhantsev and Alvaro Liendo. Polyhedral divisors and SL2-actions on affine T-varieties. Michigan Math. J. 61 (2012), no. 4, 731-762 [4] Michel Demazure. Sous-groupes alg´ebriques de rang maximum du groupe de Cremona. Ann. Sci. ´Ecole Norm. Sup. 3 (1970), 507-588 [5] Gene Freudenburg. Algebraic theory of locally nilpotent derivations. Encyclopaedia Math. Sci., vol. 136, Springer, Berlin, 2006 [6] Sergey Gaifullin. Automorphisms of Danielewski varieties. arXiv:1709.09237, 22 pages [7] Jurgen Hausen and Milena Wrobel. Non-complete rational T -varieties of complexity one. Math. Nachrichten 290 (2017), no. 5-6, 815-826 [8] Jurgen Hausen and Elaine Herppich. Factorially graded rings of complexity one. In: Torsors, ´etale homotopy and applications to rational points. London Math. Soc. Lecture Note Series 405, 414-428, 2013 [9] Jurgen Hausen, Elaine Herppich, and Hendrik Suss. Multigraded factorial rings and Fano varieties with torus action. Documenta Math. 16 (2011), 71-109 [10] Jurgen Hausen and Hendrik Suss. The Cox ring of an algebraic variety with torus action. Advances Math. 225 (2010), no. 2, 977-1012 [11] Polina Kotenkova. On restriction of roots on affine T -varieties. Beitrage zur Algebra und Geome- trie 55 (2014), no. 2, 621-634 [12] Polina Kotenkova. Torus actions and locally nilpotent derivations. PhD Thesis, Moscow State Univer- sity, Moscow, 2014 [13] Alvaro Liendo. Affine T-varieties of complexity one and locally nilpotent derivations. Transform. Groups 15 (2010), no. 2, 398-425 [14] Alvaro Liendo. Ga-actions of fiber type on affine T-varieties. J. Algebra 324 (2010), no. 12, 3653-3665 [15] Alvaro Liendo. Roots of the affine Cremona group. Transform. Groups 16 (2011), no. 4, 1137-1142 [16] Tadao Oda. Convex bodies and algebraic geometry: an introduction to the theory of toric varieties. A Series of Modern Surveys in Math. 15, Springer, Berlin, 1988 [17] Vladimir Popov. Problems for problem session. In: Affine algebraic geometry, vol. 369 of Contemp. Math., 12-16. Amer. Math. Soc., Providence, RI, 2005 [18] Bernd Sturmfels. Grobner bases and convex polytopes. University Lecture Series, vol. 8, AMS, Provi- dence, RI, 1996 [19] Yulia Zaitseva. Homogeneous locally nilpotent derivations of non-factorial trinomial algebras. arXiv:1710.10610, 12 pages Lomonosov Moscow State University, Faculty of Mechanics and Mathemat- ics, Department of Higher Algebra, Leninskie Gory 1, Moscow, 119991 Russia; and National Research University Higher School of Economics, Faculty of Computer Science, Kochnovskiy Proezd 3, Moscow, 125319 Russia E-mail address: [email protected] Lomonosov Moscow State University, Faculty of Mechanics and Mathematics, Depart- ment of Higher Algebra, Leninskie Gory 1, Moscow, 119991 Russia; E-mail address: [email protected]
1703.09855	2	1703	2019-08-13T16:01:22	Derived $\ell$-adic zeta functions	[ "math.AG", "math.AT", "math.KT", "math.NT" ]	We lift the classical Hasse--Weil zeta function of varieties over a finite field to a map of spectra with domain the Grothendieck spectrum of varieties constructed by Campbell and Zakharevich. We use this map to prove that the Grothendieck spectrum of varieties contains nontrivial geometric information in its higher homotopy groups by showing that the map $\mathbb{S} \to K(Var_k)$ induced by the inclusion of $0$-dimensional varieties is not surjective on $\pi_1$ for a wide range of fields $k$. The methods used in this paper should generalize to lifting other motivic measures to maps of $K$-theory spectra.	math.AG	math	DERIVED ℓ-ADIC ZETA FUNCTIONS JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Abstract. Let K0(Vk) be the Grothendieck group of k-varieties. Campbell and Zakharevich have constructed a higher algebraic K-theory spectrum K(Vk) such that π0K(Vk) = K0(Vk). In this paper we construct non-trivial classes in the higher homotopy groups of K(Vk) when k is finite or a subfield of C. To do this we give a recipe for lifting motivic measures K0(Vk) / K0(E ) to / K(E ). We consider two special cases: the classical local zeta function, maps of spectra K(Vk) thought of as a homomorphism K0(VF / K0(End(Qℓ)), and the compactly-supported Euler characteristic, thought of as a homomorphism K0(VC) / K0(Q). We use lifts of these motivic measures to prove that the Grothendieck spectrum of varieties contains nontrivial geometric / K(Vk) is nontrivial information in its higher homotopy groups by showing that the map S in higher dimensions when k is finite or a subfield of C, and, moreover, that when k is finite this map is not surjective on higher homotopy groups. q ) 1. Introduction Many constructions in arithmetic and geometry give rise to "motivic measures", i.e. assign- ments X 7→ µ(X) ∈ A which send an algebraic variety X to µ(X) ∈ A, for A an abelian group, such that for any closed subvariety Z ⊂ X, µ(X) = µ(X − Z) + µ(Z). This observation led Grothendieck to introduce a universal motivic measure, via the "Grothendieck ring of varieties" K0(Vk): K0(Vk): = Z[isoclass(Vk)]/{[X] = [X − Z] + [Z] Z ⊂ X closed}. where Vk denotes the category of k-varieties, i.e. reduced separated k-schemes of finite type, and where isoclass(Vk) denotes the set of isomorphism classes of varieties. The Grothendieck ring of varieties arises by forgetting large parts of the category Vk. Similarly, many motivic measures of interest arise from forgetting large parts of functorial constructions on varieties, including (1) the point counting measure #(−), which comes from applying isoclass to the functor from Fq-varieties to finite sets VFq / FinSet X 7→ X(Fq) (2) the compactly supported Euler characteristic χ(−), which comes from applying K0 to the C) to the homotopy category functor from the category C-varieties and isomorphisms (V × of homologically bounded rational cochain complexes C V × X 7→ C • / Ho(Chb(Q)) c (X(C); Q) 1 / / / / / / / 2 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH (3) the zeta function ζ(−)(s),1 which, by the Weil conjectures [Del74], comes from applying K0 ) to the homotopy to the functor from the category of Fq-varieties and isomorphisms (V × Fq category of automorphisms of homologically bounded ℓ-adic cochain complexes / Ho(Chb(Aut(Qℓ))) V × Fq X 7→ F robq (cid:8) RΓc(XFq ; Qℓ) which sends a variety to the compactly supported ℓ-adic cohomology of its restriction to Fq along with the action of Frobenius. (For a more in-depth discussion of the Grothendieck ring of varieties, together with several other important example of motivic measures, see [NS11, Hal05, Sah].) Recent work of the first and third authors ([Cam, Zak17b, CZ]) constructs a higher Grothendieck ring of varieties K(Vk), namely a commutative ring spectrum (in the sense of stable homotopy theory) with π0K(Vk) ∼= K0(Vk). As with any higher algebraic K-theory, the structure of this spectrum is much richer than just its π0. For example, the third author shows that K(Vk) is tightly linked with birational geometry [Zak17a] and the first author shows that when k = C it is related to Waldhausen's algebraic K-theory of spaces [Cam]. On the other hand, despite these links, there are essentially no tools in the literature to exhibit nontrivial classes in πi(K(Vk)) for i > 0. Just as K(Vk) represents a substantial refinement of the approximation to Vk provided by the classical Grothendieck ring, one can ask for K-theoretic refinements of the approximations of important functors provided by classical motivic measures. The purpose of the present article is to show that the classical functorial constructions in the last two examples above2 determine such refinements, and to use them to exhibit the first examples of nontrivial classes in the higher homotopy groups of K(Vk). More precisely, let k be a field, ks a separable closure of k, and ℓ 6= char(k) a prime. Let Chb(Repcts(Gal(ks/k); Zℓ)) denote the category of homologically finite cochain complexes of con- tinuous integral ℓ-adic Gal(ks/k) representations. For a variety X, let [Gal(ks/k) (cid:8) H • c (X; Zℓ)] denote the compactly supported ℓ-adic cohomology of X, viewed as an object of the homotopy category Ho(Chb(Repcts(Gal(ks/k); Zℓ))) (see Section 3 for precise definitions). Recall that a "W - exact" functor is the data required to induce a map on K-theory in this context (see Section 2 for the precise definition). We prove: Theorem 1.1. In the notation above, the functor / Ho(Chb(Repcts(Gal(ks/k); Zℓ)))op V × k X 7→ [Gal(ks/k) (cid:8) H • c (X; Zℓ)] admits a strict model as a span of W -exact functors ∼ Vk fVk / Chb(Repcts(Gal(ks/k); Zℓ))op where the left arrow induces an equivalence on K-theory. Now let R be a ring, and let W rat(R) denote the ring of (big) rational Witt vectors of R. Work of Almkvist [Alm78], explicated beautifully in Grayson [Gra79], establishes the isomorphism K0(Aut(R)) det(1−t·f ) ∼= / W rat(R) 1While essentially classical, following from the Weil conjectures and work of Almkvist [Alm78] (also explicated beautifully in Grayson [Gra79]), this example does not appear to be sufficiently widely known. 2The first example is much simpler, and is already treated in [Cam, Zak17b]. / / o o / / DERIVED ℓ-ADIC ZETA FUNCTIONS 3 where Aut(R) denotes the category of R-modules equipped with an automorphism. Now take k = Fq and R = Qℓ. Recall that Gal(Fq/Fq) is canonically freely topologically generated by Frobenius F robq. Then specializing Theorem 1.1 and applying K-theory, we obtain the following. Corollary 1.2. Let k = Fq be a finite field and ℓ ∤ q a prime. The strict model of Theorem 1.1 determines a contractible space of maps of K-theory spectra which fits into a commuting square ζ: K(VFq ) / K(Aut(Zℓ)), K0(VFq ) π0ζ K0(Aut(Zℓ)) Z(−,t) det(1−t·F rob∗ q ) W rat(Z) / W rat(Zℓ) after applying π0. The K-theory spectrum K(Vk) can be thought of as a topological object that remembers not only that certain varieties decompose into other varieties, which is what K0(Vk) does, but also In the same way, the map in Thm 1.2 tells us how the decomposition of varieties gets how. reflected in the decomposition of zeta functions. This is a great deal of information, and already at the level of the first homotopy group gives subtle information about automorphisms of varieties. We note below that this lift is canonical, and fully determined by the functor in Theorem 1.1. We refer to ζ as the derived ℓ-adic zeta function. More generally, for any field k and any g ∈ Gal(ks/k), we obtain (Corollary 4.3) a "derived ℓ-adic g-Zeta function" which induces the map K0(Vk) / W (Zℓ) ζg : K(Vk) / K(Aut(Zℓ)) [X] 7→ 2 dim XXi=0 (−1)i det(1 − t · g∗ (cid:9) (H i c(X; Zℓ))). The method developed to prove Theorem 1.1 can also be used to construct other motivic measures. For example, when k is a subfield of C we can prove the following: Theorem 1.3. Let k be a subfield of C. Let R be a ring. There exists a contractible space of maps of K-theory spectra which fits into a commuting square X: K(Vk) / K(R), K0(Vk) π0X K0(R) χ Z rk Z after applying π0. We refer to X as the derived compactly supported Euler characteristic. A special case of this when R = Zℓ or Z can be obtained from Theorem 1.1 by using the isomorphism between ℓ-adic and singular cohomology of complex varieties. However, as the proof of this theorem is significantly simpler than that of Theorem 1.1, we present it separately (and, indeed, use it as an illustrative example of the general method). For more details and the proof, see Sections 4 and 5, in particular Corollary 4.2. / / / / / / / / / 4 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Remark 1.4. While we state the theorem for homotopy categories for conciseness, our proof shows more. For readers who prefer the language of ∞-categories, we can summarize this as follows: the classical construction of compactly supported ℓ-adic cohomology determines a pair of ∞-functors (−)! : V open (−)! : V closed,op k k / Chb(Repcts(Gal(ks/k); Qℓ)) / Chb(Repcts(Gal(ks/k); Qℓ)) k k (resp. V closed where V open ) denotes the subcategories of open (resp. closed) embeddings, and where the target denotes the ∞-category underlying the dg-category of homologically finite com- plexes of continuous ℓ-adic Galois representations. Our proof shows that this pair of ∞-functors admits a strict representative as a span of W -exact functors of 1-categories as in Theorem 1.1. The span associates to each k-variety a filtering diagram in Chb(Repcts(Gal(ks/k); Qℓ))) in which all maps are weak equivalences. The pair of ∞-functors are obtained by taking a homotopy in- verse limit over this (homotopically constant) diagram. Specializing to k = Fq, we conclude that compactly supported ℓ-adic cohomology determines a lift of the classical zeta function to a map of spectra K(VFq ) / K(Aut(Qℓ)) as in Corollary 1.2. Because compactly supported ℓ-adic cohomology is classically defined as a homotopy inverse limit of right derived functors (i.e. a "derived of a derived"), we refer to this map of spectra as a "derived zeta function". A similar, though simpler, discussion applies to Theorem 1.3. While the language of ∞-categories allows for useful conceptual statements, it frequently re- quires significant baggage if one wants to explicitly translate classical references into this frame- work. To cut down on this baggage,3 we have chosen to systematically eschew overt mention of ∞-categories and ∞-functors. However, the interested reader can easily verify the above claims in her favorite model of compactly supported ℓ-adic cohomology as an ∞-functor. Remark 1.5. One use of ∞-categories would be to prove a universality statement for our spectral lift of the zeta function. The most general statement of the universality of algebraic K-theory appears to be Barwick's [Bar16] phrased in terms of Waldhausen ∞-categories. In order to make such a universality statement in our context, one would have to port this machinery to SW- categories (see 2 for definitions). While we are confident that this can be done, given the formal similarities between SW and Waldhausen categories, to do so would substantially lengthen the paper. There are, however, other ways to see an appropriate universality statement. Steimle [Ste17] has shown that the K-theory of Waldhausen categories is an initial functor in a certain category of Euler characteristics [Ste17, Thm. 0.2]. Again, given the formal similarities between Waldhausen categories and SW-categories, it is clear that this statement can be made for SW-categories, and with significantly less technical upfront cost. However, we content ourselves with noting that such a universality statement can be made. The spectral lift of the zeta function we obtain is the (homotopy) initial map lifting the zeta function. Just as motivic measures allow one to probe the Grothendieck ring of varieties (and thus to coarsely probe the category of varieties itself), the derived motivic measures above allow us to probe the higher Grothendieck ring of varieties. Most ambitiously, one would like to understand the following: Question 1.6. What arithmetic or geometric information do the higher homotopy groups Ki(Vk) encode? 3e.g. Because categories with weak equivalences provide the only model for ∞-categories in which the first and third authors' extension of algebraic K-theory has been defined. / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 5 Given the complexity of higher algebraic K-theory in general, and given that several decades of effort have only begun to answer this question for K0(Vk), a full answer to Question 1.6 is likely a long ways off. One might instead begin by asking a milder question: are there even nonzero elements in Ki(Vk) for i > 0? One simple example of an SW -category is FinSet, the category of finite sets; by the theorem of Barratt-Priddy-Quillen [BP72, Seg74] the K-theory of this is equivalent to S, the sphere spectrum. Note that for any variety X we can define a map by thinking of S as K(FinSet) and sending the finite set F to `F X. When k is a subfield of C this gives enough information to detect some of these nonzero elements. σX: S / K(Vk) Theorem 1.7. Let k be a subfield of C. Then there are arbitrarily high nonzero homotopy groups of K(Vk). The proof proceeds by tracing through the map π∗σX and showing that for a fixed s and ap- propriately chosen X, it is nonzero in degrees 4s − 1. Thus in particular K4s−1(Vk) is non-zero for all s. Using an elaboration of this proof could produce many more non-trivial homotopy groups. / K1(Vk), For example, from the multiplicative structure of K(Vk) and the inclusion i: Aut(X) we have, for any automorphism f of X, the composite π∗S / K∗(Vk) ·i(f ) −−−→ K∗+1(Vk) / K∗+1(Z). We are hopeful that this gives a method for detecting other non-trivial homotopy groups of K∗(Vk), but leave it to future work. We note that this is an approach very similar to the one Bokstedt and Waldhausen [BW87] use to detect non-trivial homotopy groups in the algebraic K-theory of spaces, A(∗). When k is finite we can obtain more refined information. For a finite field k = Fq, the map / S which is a cosection of σSpec k, yielding K(VFq ) ≃ X 7→ X(Fq) defines a map K(VFq ) S ∨ eK(VFq ). Thus the analog of Theorem 1.7 for k finite is essentially trivial as it exhibits π∗S is a summand of K∗(Vk). Defining eK(Vk) = hocofib σSpec k, a more subtle question asks: Question 1.8. Do there exist nontrivial elements in eKi(Vk) for i > 0? Note that this statement makes sense for any k. We use the derived zeta function to answer this question affirmatively. Theorem 1.9. The group eK1(Vk) is nontrivial whenever k is a subfield of R, a finite field with k ≡ 3 (mod 4), or a global or local field with a place of cardinality 3 (mod 4). To prove this theorem we use the derived 2-adic zeta function. For any category C, let Aut(C) be the category of pairs (P, f ), where P ∈ C and f ∈ Aut(P ). Morphisms (P, f ) / (Q, g) / Q such that hf = gh. When C, is exact it induces an exact structure are morphisms h: P on Aut(C). In particular, for a ring R, we obtain an exact category Aut(R): = Aut(Modfg(R)). Specializing from a representation of Gal(Fq/Fq) to its value on F robq and tensoring by Q gives / Aut(Qℓ). Composing the derived 2-adic zeta function with a functor Repcts(Gal(Fq/Fq); Zℓ) the K-theory of this functor and applying π1, gives a homomorphism K1(Vk) / K1(g Aut(Q2)). The group K1(Aut(Q2)) is relatively well-understood. In [Gra79], Grayson constructs a homo- / K2(Q2). By Moore's Theorem (see e.g. [Mil71, Appendix]), the morphism σ2: K1(Aut(Q2)) 2-adic Hilbert symbol induces a (split) surjection (−, −)2 : K2(Q2) ։ Z/2Z, and by composing these maps, we produce a map h2 : K1(Vk) / K1(Aut(Q2)) σ2 / K2(Q2) (−,−)2 / Z/2Z. We then show h2 ◦ π1σSpec k is trivial, but h2 ◦ π1σP1 is surjective. / / / / / / / / / / / / / 6 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Corollary 1.2 and Theorem 1.9 are both applications of Theorem 1.1, in which ζ is constructed. In order to construct ζ, we use a K-theory machinery first created by the first author in [Cam]. The usual categories one wants to work with as inputs for a K-theory machine are Waldhausen categories [Wal85]. Unfortunately, these do not work to produce K(Vk), which is why the first and third authors introduced their formalisms. In [Cam], the difficulty is circumvented by defining a modification of Waldhausen categories called SW -categories (the S is for "scissors") where one can define algebraic K-theory for Vk in much the same way one does for Waldhausen categories. However, in order to get maps K(C) /K(W) where C is an SW -category and W is a Waldhausen category, one needs the notion of a "W -exact functor" introduced in [Cam]. It needs to satisfy certain variance conditions reminiscent of push-pull formulae (see Section 2 for details). To con- struct the derived ℓ-adic zeta function, we take the SW -category Vk and the Waldhausen category Chb(Repcts(Gal(ks/k); Zℓ)) of homologically finite and bounded chain complexes of continuous ℓ-adic Galois representations. To go from one to the other we need to use compactly-supported / X, right ´etale cohomology. Classically, this is defined by choosing a compactification j : X deriving the functor Γ ◦ j! and then taking a homotopy inverse limit over all such choices to get a well-defined object. The key technical observation of this paper is that all of this can be made "strict". We show that once we restrict to constant sheaves, the Godement resolution combines both the necessary functoriality and exactness properties required to produce a W -exact functor. By further replacing the homotopy inverse limit by the left-facing map in the span, we obtain the span of W -exact functors in Theorem 1.1. Using invariance properties of K-theory, we can then invert the left-facing arrow and obtain the derived ℓ-adic zeta function. We view this work as part of a larger program to lift motivic measures to the spectral/homotopical level. For example, the outline we follow should adapt to give lifts for other cohomologically defined motivic measures, e.g. p-adic zeta functions, Serre polynomials, and the Gillet -- Soul´e measure [GS96]. One might similarly ask for lifts of Kapranov's motivic zeta function, or of the motivic measure used by Larsen and Lunts [LL03] to show that motivic zeta function is not rational as a map out of K0(VC). See Section 7 for a more detailed discussion. This paper is organized as follows. In Section 2, we quickly review Waldhausen K-theory and introduce SW-categories. In Section 3, we review the background and necessary results from ℓ-adic cohomology and Galois representations. In Section 4 we review our general approach to constructing derived motivic measures, and provide the the construction of the motivic measure arising from singular cohomology. Section 5 contains the full construction of the derived ℓ-adic zeta function. In Section 6 we use the results of the previous section to construct nontrivial elements in the higher K-theory of varieties over both C and finite fields. We close, in Section 7, by discussing questions for future work. Acknowledgments. We thank Bhargav Bhatt, Denis-Charles Cisinski, Sean Howe, Keerthi Madapusi Pera and Nick Rozenblyum for helpful correspondence. We thank Oliver Braunling, Kiran Kedlaya, Dan Petersen, Ravi Vakil, Chuck Weibel, Kirsten Wickelgren and Ilya Zakharevich for many helpful questions and comments on an earlier draft. We thank the anonymous referee for a careful reading and many helpful comments which greatly improved the paper. J.W. was supported in part by NSF Grant No. DMS-1400349. I.Z. was supported in part by an NSF MSPRF grant and NSF Grant No. DMS-1654522. Notation 1.10. Throughout, when dealing with schemes or varieties, we let Z ֒→ Y denote a closed inclusion and X ◦−→ Y denote an open inclusion. / / DERIVED ℓ-ADIC ZETA FUNCTIONS 7 2. SW -categories and K-Theory In [Zak17b], the third author defines a spectrum K(Vk) whose zeroth homotopy group is the Grothendieck ring of varieties over k. In [Cam], the first author gives an alternate construction of this spectrum. In this paper, we use the latter construction to produce maps out of K(Vk), so we review the structure necessary to produce this spectrum. Most definitions of K-theory work with categories where a suitable notion of quotient exists, for example Quillen's exact categories [Qui73] or Waldhausen's categories [Wal85]. These notions of quotient are then used to define the exact sequences that K-theory is defined to "split." When dealing with the category of varieties, we have no such quotients. Instead, our "exact sequences" are sequences of the form Z ֒→ X ◦←− (X − Z) where the first map is a closed inclusion and the second is an open inclusion. The notion of an SW -category is meant to modify Waldhausen's definition of categories with cofibrations and weak equivalences to allow the use of such "exact sequences." For ease of reading, we review Waldhausen's construction before recalling the first author's construction. Definition 2.1 (Waldhausen category, [Wal85, Section 1.2]). A Waldhausen category 4 is a cat- egory C equipped with two distinguished subcategories: cofibrations and weak equivalences, de- noted co(C) and w(C). The arrows in co(C) are denoted by hooked arrows ֒→. Arrows repre- senting weak equivalences are decorated with ∼. These categories satisfy the following axioms: (1) C has a zero object 0. (2) All isomorphisms are contained in co(C) and w(C). (3) For all objects A of C, the morphism 0 (4) (pushouts) For any diagram / A is a cofibration. C A / B where A ֒→ B is a cofibration, the pushout exists and the morphism C ֒→ B ∪A C is a cofibration. (5) (gluing) For any diagram C ∼ C ′ A ∼ A′ B ∼ B′ where the vertical morphisms are weak equivalences the induced morphism is also a weak equivalence. B ∪A C ∼ / B′ ∪A′ C ′ We will be using a slightly more general definition of the K-theory of a Waldhausen category, as we need this flexibility for one of our main results. We begin with some preliminary definitions. Definition 2.2. Let C be a category. The category Ar C has, as its objects, the morphisms of C. The morphisms in Ar C from a morphism f : A / D are commutative squares / B to a morphism g: C A f B C . g / D 4Referred to as a "category with cofibrations and weak equivalences" by Waldhausen. / o o / / / o o / / o o / / / / / / 8 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH If C is equipped with a subcategory of weak equivalences then a morphism in Ar C is considered a weak equivalence if both horizontal morphisms in the diagram above are weak equivalences. Let [n] be the ordered set {0 < . . . < n} considered as a category. Then Ar[n] can be considered / C; to be the set of pairs (i, j) ∈ [n] × [n] with i ≤ j. We are often considering functors X: Ar[n] in this case, we write Xi,j for X(i, j). Definition 2.3. Let C be a Waldhausen category. A morphism in Ar C is a weak cofibration if it is weakly equivalent (via a zigzag of weak equivalences) in Ar C to a cofibration. A square in C is homotopy cocartesian if it is weakly equivalent (by a zigzag) to a pushout square where either both horizontal or both vertical morphisms are cofibrations. Definition 2.4 (S′ We define S′ nC to be the category of functors •-construction, see [BM11, Definition 2.3]). Let C be a Waldausen category. with morphisms natural transformations, subject to the conditions X: Ar[n] / C • the initial map 0 • When i ≤ j ≤ k, Xi,j • For any i ≤ j ≤ k the square / Ai,i is a weak equivalence for all i. / Xi,k is a weak cofibration. Xi,j Xi,k 0 / Xj,k is a homotopy cocartesian square. / B in S′ nC is a weak equivalence when each component map Ai,j A map A equivalence; it is a cofibration when each component map Ai,j map Ai,k ∪Ai,j Bi,j Remark 2.5. The S′ category of small categories). For more detail on this, see [BM11, Section 2]. / Bi,k is a weak cofibration. nC assemble to form a simplicial category (i.e. a simplicial object in the / Bi,j is a weak / Bi,j is a cofibration and the We now define the algebraic K-theory spectrum of a Waldhausen category. Unfortunately, the S′ •-construction does not work correctly for all Waldhausen categories, but only those satisfying a condition Blumberg -- Mandell call functorial factorization of weak cofibrations (FFWC). All examples that we are concerned with satisfy this condition, but a discussion of the condition is not illuminating for the sake of the current discussion; we prove that this is the case in Appendix A and restrict our attention to categories satisfying FFWC. Definition 2.6. Let C be a Waldhausen category satisfying FFWC. Let wS′ category of weak equivalences of S′ topological space K t(C) is defined by nC denote the sub- nC denote the nerve of that category. The nC and let N•wS′ K t(C) = ΩN•wS′ •C where − denotes the geometric realization of a bisimplicial set. The spectrum K(C) is defined by taking a (functorial) fibrant-cofibrant replacement in the stable model category of symmetric spectra [MMSS01, Sec. 9] of the spectrum whose m-th space is The most important example of a Waldhausen category for the purposes of this paper is the N•w S′ C. • · · · S′ • m times {z } following: / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 9 Example 2.7. Let E be any exact category. If we define the admissible monomorphisms to be the cofibrations and the isomorphisms to be the weak equivalences then the Waldhausen K-theory of E and Quillen's K-theory of E are equivalent. Let Chstr(E) be the category of bounded chain / Chstr(E) given by mapping E to complexes in E; by [TT90, Theorem 1.11.7], the incluson E the chain complexes concentrated at 0 is an equivalence on K-theory. The inverse map on K0 is exactly the Euler characteristic. However, we need a stronger version of this. The following lemma is probably well-known to experts, but we could not find a statement in the literature. We have included it since it is needed below. Lemma 2.8. Let R be a ring, and suppose that E is Modfg R, the category of finitely generated R-modules. Let Chb(R) be the category of chain complexes of (possibly infinitely generated) R- modules whose cohomology is bounded and in Modfg R. The induced map is an equivalence. K(Modfg R) / K(Chb(R)) Proof. We prove this by breaking the map up into three compositions: K(Modfg R) / K(Chstr(R)) / K(Chfb(R)) / K(Chb(R)). Here, Chfb(R) is the category of chain complexes of finitely generated R-modules which are homologically bounded. The first of these is an equivalence by [TT90, Theorem 1.11.7]. The second is an equivalence by [Wei13, Section V.2.7.1]. Thus it remains to consider the third. We prove that this is an equivalence by using Waldhausen's Approximation Theorem, [Wal85, Theorem 1.6.7]. To apply the theorem we must show that for any map f : A / B, where A ∈ Chfb(R) and B ∈ Chb(R) there exists a cofibration g: A ֒→ A′ in Chfb(R) and a weak equivalence f ′: A′ / B such that f = f ′g. Note that (by possibly first factoring f as a cofibration followed by a weak equivalence) it suffices to check this when f is itself a cofibration; in particular f is levelwise injective. Suppose that the cohomology of B is only nonzero below dimension k. We define A′ m = 0 if m > k. For m ≤ k we define A′ m to be the submodule of Bm generated by • the image of Am, • a choice of generators for H m(B), and • a choice of generators for the relations between the generators for H m+1(B). (Since the cohomology is bounded we can just construct these starting at m = k and working downwards.) Since each of these only involves a finite number of generators, A′ m is finitely generated. Thus A′ is levelwise a subcomplex of B which has the same cohomology and satisfies the desired conditions. Thus the map K(Chfb(R)) / K(Chb(R)) is a weak equivalence. It is not necessarily the case that for all exact categories, Chb(E) satisfies FFWC. In Appendix A we show that the specific cases we are interested in in this paper do; the results of this appendix also suggest that for almost all cases of E that are of interest FFWC holds. (cid:3) We now turn to defining SW -categories. As much of the intuition necessary for working with these comes from Waldhausen's S•-construction we omit the full definition and instead refer to [Cam]. Note also, we omit the "weakness" hypotheses, since we need to work inside SW -categories more strictly than in Waldhausen categories. Definition 2.9 (SW -category [Cam, Definition 3.23]). An SW -category is a category C equipped with three distinguished subcategories: cofibrations, complement maps, and weak equivalences, denoted co(C), comp(C) and w(C). The arrows in co(C) are denoted by hooked arrows ֒→ and the arrows in comp(C) are denoted by ◦−→. Arrows representing weak equivalences are / / / / / / / / 10 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH decorated with ∼. The category C is further equipped with a collection of subtraction sequences ◦←− U }. The data is required to satisfy axioms spelled out in [Cam, Definition 3.7, {Z ֒→ X Definition 3.13, Definition 3.24] which mimic those of Waldhausen, replacing cofiber sequences with subtraction sequences. There are many examples of these kinds of categories. The following is the motivating example. Example 2.10. The category Vk of varieties over a field k, is an SW -category, where cofibra- tions are closed immersions, complements are open immersions, and the weak equivalences are isomorphisms. (This is proven in detail in the results leading up to [Cam, Prop. 3.28]) The / X, i determines a subtraction sequences are defined as follows. Given a closed inclusion i: Z homeomorphism of Z onto a closed set i(Z). We consider the open set X − i(Z) and give it a scheme structure by restricting the structure sheaf on X. Thus X − Z = (X − i(Z), OX X−Z ). The definition of an SW -category is designed to provide exactly the structure needed to carry out a Waldhausen-style S•-construction when we have subtraction instead of quotients. However, we need one auxiliary definition. (i, j) with i ≤ j. Definition 2.11. We define fAr[n] to be the full subcategory of [n]op × [n] consisting of pairs Definition 2.12 (eS•-construction). Let C be an SW -category. We define eSnC to be the category with objects functors with morphisms natural transformations, subject to the conditions X:fAr[n] / C • Xi,i = ∅, the initial object • When j < k, Xi,j • The subdiagram / Xi,k is a cofibration. is a subtraction sequence for all i < j < k. Xi,j −→ Xi,k ←− Xj,k Remark 2.13. The eSnC assemble to form a simplicial category (i.e. a simplicial object in the category of small categories). Each of these is itself an SW -category, so this construction can be iterated. For details, see [Cam, Lem. 3.34]. We may finally define the algebraic K-theory spectrum of an SW -category. Definition 2.14. Let C be an SW -category. Let weSnC denote the subcategory of weak equiva- lence of eSnC and let N•weSnC denote the nerve of that category. The topological space K t(C) is defined by where − denotes the geometric realization of a bisimplicial set. The spectrum K(C) is defined by taking a (functorial) fibrant-cofibrant replacement in the stable model category of symmetric spectra of the spectrum whose m-th space is K t(C) = ΩN•weS•C There is a notion of exact functors for SW -categories: C. N•w eS• · · · eS• {z } m times Definition 2.15. Let C, D be SW -categories. A functor F : C / D is called exact if (1) F preserves the initial object: F (∅) = ∅. / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 11 (2) F preserves subtraction sequences (3) F preserves pushout diagrams. Proposition 2.16. Let C and D be SW -categories and let F : C F descends to a map of spectra K(C) / K(D). / D be an exact functor. Then Proof. This follows directly from the definition of exact functor. (cid:3) Most of the maps in which we are interested do not have SW -categories as codomains; instead, we wish to be able to construct a functor from an SW -category to a Waldhausen category. This requires we use a different definition in order to define the map of K-theories, since we cannot In fact, because of the change in variance, the proper notion is not a functor at all -- instead it is a triple of functors, two covariant (for weak equivalences and cofibrations) and one contravariant (for the complement maps). One should keep in mind here the dual of compactly supported cohomology, which is covariant on closed inclusions and contravariant on open. just hit the source and target with the eS• construction or S′ •-construction. Definition 2.17 (Based on [Cam, Defn. 5.2]). Let C be an SW -category and D a Waldhausen category. A W -exact functor (resp. weakly W -exact functor) from C to D is a triple of functors (F!, F !, F w) such that / D from the category of cofibrations in C to D. For a map i (1) F! is a functor F!: co(C) we abbreviate F!(i) to i!. (2) F ! a contravariant functor F !: comp(C)op / D, from the category of complement maps in C to D. For a map j we abbreviate F !(j) to j!. / w(D). (3) F w is a functor w(C) (4) For objects X ∈ C, F !(X) = F!(X) = F w(X) and we denote all three by F (X). (5) For every cartesian diagram in C on the left below, where the horizontal maps are cofi- brations and the vertical maps are complements, we obtain a commuting diagram on the right X i Y j Z i′ / W j ′ F (X) i! F (Y ) j! j ′ ! / F (Z) (i′)! / F (W ) (6) For a subtraction sequence in C on the left, the square in D on the right is cocartesian (resp. weakly cocartesian). i Z / X ◦ j X − Z F (X) i! F (Y ) j! F (0) / F (Y − X) (7) For any commutative diagram on the left below, where the horizontal morphisms are cofibrations and the vertical morphisms are weak equivalences, the diagram on the right commutes f X iX Y iY X ′ / Y ′ f ′ F (X) f! F (Y ) F w(iX ) F w(iY ) F (X ′) / F (Y ′) (f ′)! A similar statement holds for complement maps. / / / / / / / / / O O / O O / O O / / / / / / / / / 12 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Remark 2.18. For notational ease, we denote W -exact functors by (F!, F !, F w): C F : C / D, when no confusion can arise. / D or even Having defined this, one can prove the following. Proposition 2.19 (Based on [Cam, Prop. 5.3]). Let C be an SW -category and D a Waldhausen / D be a weakly W -exact functor. Then (F!, F !, F w) category with FFWC and let (F!, F !, F w): C determines a spectrum map K(C) F / K(D). Proof. It suffices to prove that this map exists before taking fibrant-cofibrant replacement, since our replacement is functorial. But this follows exactly from the definitions of K(C) and K(D), since a weakly W -exact functor takes a simplex in the spaces defining K(C) to a simplex in the spaces defining K(D). (cid:3) As a consequence of the definition of K-theory, we obtain the following result, which can be used to pick out interesting elements of K1 of an SW -category. The proof is just as for Waldhausen K-theory as in [Wal85, §1.5]. Proposition 2.20. Let X be an object in an SW -category (resp. Waldhausen category) C. There / K1(C), which is natural in C in the sense that for any exact is a homomorphism ξX: Aut(X) (resp. weakly W -exact) functor F : C / D, ξF (X) = π1F ◦ ξX. Proof. We begin by recalling how this statement works for Waldhausen categories (see [Wal85, /X there is a corresponding 1-simplex in wC. Composing p.341]). Given an automorphism f : X /K1C. This construction is visibly with the map wC / D, and uses only that automorphisms include into π1wC and natural for exact functors F : C that we have a natural map wC •-construction, an analogous statement holds. •C. As the same is true for the S′ /ΩwS•C gives the desired map Aut(X) / wS′ reduced and isomorphisms are a subcategory of weak equivalences. For a weakly W-exact functor, F = (F!, F !, F w) from an SW-category C to a Waldhausen cate- For SW-categories C, the construction works the same way since the simplicial set weS•C is still gory D we obtain a simplicial map weS•C / wD induced by F w. It is then clear that the diagram /weS′ and wC •D. For X ∈ C, there are maps Aut(X) / Aut(F (X)), Aut(X) π1wC Aut F (X) / π1wD commutes, which is the statement of the proposition. π1(ΩweS•C) / π1(ΩweS′ •D) (cid:3) An important tool in the general method discussed in Section 4 is a lemma designed to identify when the K-theories of two SW -categories are equivalent. Although the conditions of this lemma look complicated, in the geometric cases we are interested in they are generally very natural. For a detailed example of an application, see Example 2.24. Lemma 2.21. Let U : A extra conditions hold: / C be an exact functor of SW -categories. Suppose that the following (1) U is surjective on objects. Moreover, one of the following holds: (a) For all cofibrations f : X ֒→ X ′, if U (A′) = X ′ then there exists a cofibration A ֒→ A′ whose image under U is f . / / / / / / / / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 13 (b) For all cofibrations f : X ֒→ X ′, if U (A) = X then there exists a cofibration A ֒→ A′ whose image under U is f . (2) If for f, g: A / B in eSnA, there exists a weak equivalence h: X that U (f )h = U (g)h, then there exists a weak equivalence eh : Z factors through U (eh) and such that feh = geh. (3) Given any diagram / U (A) in eSnC such / A in eSnA such that h U (A) U (f ) U (A′) ∼ ∼ A f A′ X X ′ C C ′ ∼ ∼ / U (B) U (g) / U (B′) B g / B′ there exists a diagram in A making the diagram X ∼ ∼ ∼ U (A) U (C) U (B) U (f ) X ′ U (g) ∼ ∼ ∼ U (A′) U (C ′) / U (B′) commute. Then K(U ) is an equivalence. on geometric realizations. By Quillen's Theorem A [Qui73, Theorem A], it suffices to show that Proof. It suffices to prove that, for n ≥ 0, the map weSn(A) /weSn(C) induces a weak equivalence for any α ∈ weSn(C), the undercategory α/weSn(A) is cofiltering. Recall that α is represented by a diagram X1 / · · · / Xn . For this, we first observe that by condition (1) the category α/weSn(A) is non-empty. We use the second version of (1); the proof for the first works analogously. Since U is surjective on / X2. Now, since cofibrations by (1), there exists a cofibration A1 U (A2) = X2, there exists a cofibration A2 / X3. Working / An of cofibrations whose image under inductively, we see that there is a sequence A1 / A3 whose image under U is X2 / A2 that maps to X1 U is α. To show that α/weSn(A) is nonempty we take this sequence and the identity map from α. / · · · / / / _ _ o o / _ o o / _ o o / / _ _ o o / u u { { # # _ _ o o / / _ _ u u " " o o / / / / / / / / / / 14 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH We now show that α/weSn(A) is co-filtering. We start by showing that any two parallel arrows are equalized by a third. An object A ∈ α/weSn(A) is a diagram X1 · · · / Xn . ∼ ∼ U (A1) / · · · / U (An) Given two such objects A and B, and two morphisms f, g: A exists a map h : Z (2). / A in α/weSn(A) such that f h = gh. This is exactly guaranteed by condition It remains to show that for every pair of objects in α/weSn(A), there exists a third object which maps into each of them. This is guaranteed by condition (3). / B, we must show that there (cid:3) Our main example is the SW -category of varieties together with a choice of compactification; we show that forgetting the choices induces an equivalence on K-theory. Definition 2.22. Let k be a field. We define the SW -category V cptd are open embeddings X (X as follows. Objects of V cptd ◦−→ X where X is a k-variety and X is a proper k-variety. Morphisms ◦−→ Y ) are commuting squares ◦−→ X) / (Y k k X ◦ X . f Y f ◦ / Y A morphism (X ◦−→ X) / (Y ◦−→ Y ) is a cofibration: if f and f are closed embeddings, complement: if f is an open embedding and f is a closed embedding, and weak equivalence: if f is an isomorphism. ◦−→ Z) −→ (X A sequence (Z cofibration, the right map is a complement, Z ֒→ X the closed embedding U ◦−→ X) ◦←− (U ◦−→ U ) is a subtraction sequence if the left map is a ◦←− U is a subtraction sequence in Vk, and / X has set-theoretic image equal to the closure of X − Z in X. Lemma 2.23. The category V cptd traction defined as above satisfies the axioms of an SW -category. k with cofibrations, complements, weak equivalences, and sub- k Proof. We verify the axioms from [Cam] in turn. Note that limits and colimits are computed pointwise in V cptd . As a result, [Cam, Defn. 3.7(1)] is automatically satisfied. The axiom [Cam, Defn. 3.7(2)] follows immediately because for any varieties X, Y , the embedding X is both open and closed. Axiom [Cam, Defn. 3.7(3)] is also immediate, because isomorphisms in V cptd are pointwise, and any isomorphism of varieties is simultaneously an open and closed embedding. Axiom [Cam, Defn. 3.7(4)] holds for the same reason as in Vk. We now verify the remaining axioms in turn. / X` Y k • [Cam, Defn. 3.7(5a)] Let (Z ◦−→ X) be a cofibration. Then Z ֒→ X determines the open embedding X − Z ֒→ X up to unique isomorphism, and similarly, Z ֒→ X determines the closed embedding X − Z / X up to unique isomorphism. ◦−→ Z) ֒→ (X / / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 15 • [Cam, Defn. 3.7(5b)] Given a diagram (W, W ) (Z, Z) / (X, X) ◦ (U, U ) where the bottom row is a subtraction sequence, we have a subtraction sequence W ×X Z ◦ / W W ×X Y in Vk. Moreover, because taking closure commutes with pullback, we have W ×X U ∼= W ×X X − Z = W ×X (X − Z) where the outer (−) denotes closure in W . Further, because the pullback of the comple- ment of a subvariety is the complement of its pullback, this is isomorphic to = W − W ×X Z. Therefore, (W ×X Z, W ×X Z) sequence. / (W, W ) ◦ (W ×X U, W ×X U ) is a subtraction • [Cam, Defn. 3.7(5c)] Given a Cartesian square (W, W ) (X, X) (Y, Y ) / (Z, Z) in which all maps are cofibrations, the unique map X −W ֒→ Z −Y is an open embedding. Further, there is a unique map X−W / /Z is a closed emebdding, this extends uniquely to a closed embedding /Z−Y and, because X / X − W / / Z − Y . Axioms [Cam, Defn. 3.13(1)], [Cam, Defn. 3.13(2)], and [Cam, Defn. 3.13(3)] hold for the same reason they hold in Vk, namely [Sch05, Corollary 3.9] and the remarks following. Axioms [Cam, Defn. 3.24(2)] and [Cam, Defn. 3.24(3)] follow immediately from the definition of weak equiva- lences in V cptd along with Axioms [Cam, Defn. 3.13(1)] and [Cam, Defn. 3.7(5a)] respectively. (cid:3) k Example 2.24. Let C = Vk and let A = V cptd the conditions of Lemma 2.21 hold in this example. We check the conditions in turn: , with U being the forgetful functor. We claim that k (1) By Nagata [Nag62, Theorem 4.3]5 every k-variety X admits an open embedding X ◦−→ X into a proper k-variety X. Thus U is surjective on objects. Moreover, for any cofibration X ֒→ X ′, if X is a compactification of X. Thus condition (a) holds. is a compactification of X ′ then the closure of X in X ′ ′ 5For a modern treatment of Nagata's Theorem, see [Con07], esp. Theorem 4.1. / o o / o o / o o / / _ _ / 16 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH (2) Weak equivalences in Vk are isomorphisms. In this case this condition says that given two parallel maps f and g in eSnV cptd k (X1, X 1) · · · / (Xn, X n) . (Y1, Y 1) / · · · / (Yn, Y n) such that f and g are equal when restricted to Xi, then there exists a weak equivalence h in eSnV cptd k (X1, X 1) · · · / (Xn, X n) . (X1, X 1) / · · · / (Xn, X n) such that f h = gh. But for this, we can take X i to be the closure of Xi in X i, and h to be given by the canonical inclusions. Then, for each i, the ith components of f h and gh agree on a dense set in X i, and thus are equal. Therefore f h = gh. ′ (3) This property states that given a closed embedding of varieties X ֒→ X ′ together with induced embeddings of two choices of compactification (X, X) ֒→ (X ′, X ′) and (X, X) ֒→ (X ′, X ), there exists a third choice of embeddings of compactifications that dominates both of these. Note that if we can find a compactification of X ′ that dominates both X ′ and X ′ then we can take the closure of X in it to produce the desired embedding. Thus all that we must show is that for any two choices of compactification, there is a third that dominates them. For example, we can take the closure of X inside X × X , as in [Del77, §IV-10.5]). ′ 3. ℓ-adic Cohomology In this section we being by recalling the standard facts we will need about ℓ-adic cohomology with its continuous Galois action. We then review the Godement resolution, and use this to establish the key technical Lemma 3.11. We take [Del77] and [FK80] as standard references. 3.1. Continuous Galois Representations. Let k be a field, let ks be a separable closure, and let ℓ 6= char(k) be a prime. The separable Galois group Gal(ks/k) is a profinite group, and canonically carries the profinite topology Gal(ks/k) = lim←− L/k fin., sep. Gal(L/k) where the finite groups Gal(L/k) are discrete, and the limit is in the category of topological groups. Let R be a ring. Recall that for a (discrete) R-module A, a continuous representation Gal(ks/k) / AutR(A) is one which factors through a finite subgroup Gal(ks/k) / Gal(L/k) / AutR(A) for some separable L/k. Denote by Repcts(Gal(ks/k); R) the category of finitely generated con- tinuous representations of Gal(ks/k) over R. Recall the following (cf. [Del77, Arcata II.4.4]). / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 17 Proposition 3.1. Let k be a field, and ks a separable closure. Let R be a ring. Denote by Sh(Spec(k); R) the category of ´etale sheaves of (discrete) finitely generated R-modules on Spec(k). The functor Sh(Spec(k); R) / Repcts(Gal(ks/k); R) F 7→ F Spec(ks) is an equivalence of categories. We are especially interested in continuous ℓ-adic representations. Recall the following refor- mulation of the category of finitely generated Zℓ-modules (we follow the presentation of [FK80, Ch I.12]). We consider diagrams F : Z≥ / Mod(Zℓ), /n where Z≥ is the category whose objects are integers and where there is a unique morphism m / whenever m ≥ n. The category of diagrams is an abelian category, and in particular has images, kernels, cokernels, etc. defined pointwise. For r ∈ Z, denote by F [r] the shifted diagram, i.e. with F [r]m := Fr+m. Definition 3.2. A diagram F : Z≥ / Mod(Zℓ) satisfies: (1) the Mittag -- Leffler (ML) condition if for every n, there exists t ≥ n such that for all m ≥ t, Image(Fm / Fn) = Image(Ft / Fn), (2) the Mittag -- Leffler -- Artin -- Rees (MLAR) condition if there exists some t ≥ 0 such that for all r ≥ t, Image(F [r] / F ) = Image(F [t] / F ). Definition 3.3. Define ProM LAR(Zℓ) to be the category in which objects are MLAR diagrams in which each Fn is torsion, and for for two such diagrams F and G, homProM LAR(Zℓ)(F, G) := lim−→ r≥0 hom(F [r], G). The key purpose of Mittag -- Leffler diagrams is that on such diagrams, the inverse limit is an exact functor. The ML property is frequently satisfied. For instance, if all the modules Fn are finite length, then F is an ML diagram. Definition 3.4. An ℓ-adic diagram is a diagram F such that Fn = 0 for n < 0 and for all n (1) Fn is a module of finite length, (2) ℓn+1Fn = 0, and (3) the map Fn+1 / Fn induces an isomorphism Fn+1/ℓn+1Fn+1 ∼= Fn. An A -- R ℓ-adic diagram is any object of ProM LAR(Zℓ) which is isomorphic to an ℓ-adic diagram. Denote by AR(ℓ) ⊂ ProM LAR(Zℓ) the full sub-category of A -- R-ℓ-adic diagrams. We can now give the promised reformulation of the category Modfg(Zℓ) of finitely generated Zℓ-modules (see e.g. [FK80, Proposition I.12.4]). Proposition 3.5. The inverse limit is an equivalence of exact categories. lim←− : AR(ℓ) / Modfg(Zℓ) / / / / / / / / / 18 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Recall that Zℓ is a profinite ring, with the profinite (equivalently "adic") topology. Similarly, for any finitely generated Zℓ-module A, the group AutZℓ(A) is canonically a topological group, with the profinite topology. We can use the proposition to give a similar reformulation of the category Repcts(Gal(ks/k); Zℓ) of continuous representations Gal(ks/k) / AutZℓ(A). Mutatis mutandis, we obtain from Definitions 3.3 and 3.4 a notion of ℓ-adic diagrams of continuous cts (Gal(ks/k); ℓ) of such. Concretely, objects are Gal(ks/k) representations, and a category RepAR given by diagrams · · · / Fn / Fn−1 / · · · where for each n, Fn is a continuous representation of Gal(ks/k) in finitely generated (discrete) Z/ℓnZ-modules and the analogous conditions to those of Definition 3.4 hold. Analogously to Proposition 3.5, we have the following. Proposition 3.6. The inverse limit lim←− : RepAR cts (Gal(ks/k); ℓ) / Repcts(Gal(ks/k); Zℓ) is an equivalence of exact categories. 3.2. ℓ-adic Sheaves. We now extend the above to sheaves. We consider schemes X for which ℓ ∈ O(X) is invertible. Mutatis mutandis, we obtain from Definition 3.3 a definition of MLAR diagrams of ℓ-torsion ´etale sheaves, and the category ProM LAR(Sh(X), Zℓ) of such. Definition 3.7. Let X be a scheme and ℓ a prime invertible on X. An ℓ-adic sheaf on X is diagram F : Z≥ / Sh(X; Z) of ´etale sheaves of abelian groups on X such that (1) the sheaves Fn are constructible for all n, (2) Fn = 0 for n < 0, and (3) the map Fn+1 / Fn induces isomorphisms Fn+1 ⊗Z Z/ℓnZ ∼= Fn. An A -- R-ℓ-adic sheaf is any object of ProM LAR(Sh(X), Zℓ) which is isomorphic to an ℓ-adic sheaf. We denote the category of A -- R-ℓ-adic sheaves by Sh(X; Zℓ). The following is the key theorem we use for ℓ-adic sheaves (cf. [FK80, Theorem I.12.15]). Recall / S is compactifiable if f can be factored as an open embedding j : X ֒→ X / S and ν ≥ 0, recall that a map f : X followed by a proper map f X / S. Given a compactifiable map f : X (cf. [FK80, Definition/Proposition I.8.6]) that Rνf! denotes the functor Rνf ∗ ◦ j! : Sh(X) / Sh(S) By loc. cit., this functor is independent of the choice of j and f , and is referred to as the νth higher direct image with compact supports. Theorem 3.8 (Finiteness Theorem for ℓ-adic Sheaves). Let f : X / S be a compactifiable map, let ℓ a prime invertible on X, and n 7→ Fn an A -- R-ℓ-adic sheaf on X. For ν ≥ 0, let Then for all ν ≥ 0, the system n 7→ Rνf!(Fn) is an A -- R-ℓ-adic sheaf on S. Using the smooth base change theorem (see e.g. [FK80, Theorem I.7.3]) and Propositions 3.1 above, we immediately deduce the following. Corollary 3.9. Let k be a field, ks a separable closure, and let X be a variety of finite type over k. Then for any A -- R-ℓ-adic sheaf F on X, a choice of compactification j : X ֒→ X and a choice of functorial flasque resolution QX on Sh(X; Zℓ) determine an extension of the assignment F 7→ H ∗ c (X/ks; F ) / / / / / / / / / / / / to a functor DERIVED ℓ-ADIC ZETA FUNCTIONS 19 Sh(X; Zℓ) / Chb(RepAR F 7→ ΓXQX(j!F )ks cts (Gal(ks/k); ℓ)) 3.3. The Godement Resolution. Recall that the Godement resolution G•(F) of an ´etale sheaf F on a k-variety X is defined inductively as follows (see e.g. [FK80, p. 129]). Fix algebraic closures Ω0: = k and Ωn: = k(t1, . . . , tn) for all n. Let MX be the set of geometric points MX : = {x : Spec(Ωn) / X n ∈ N} Note that every point in X is associated to at least one geometric point in MX (since algebraically closed fields over k are distinguished up to isomorphism only by their transcendance degree, e.g. [Hun74, Theorem VI.1.12]). In particular, this is what allows us to define the Godement resolution with respect to this set of points of the ´etale site of X. Now define G0 X (F): = Yx∈MX X F where δX : `MX x∗x∗F Equivalently, G0 The functor G0 X(F) = δX∗δ∗ X (−) is exact (it is exact on stalks), and comes with a diagonal inclusion / X is the inclusion of points. x given by mapping a section to its stalks. Define F / G0 X (F) G1 X(F): = G0 X(coker(F / G0 X (F))). In general, define X (coker(Gi−1 Gi+1 X (F): = G0 X(F) has the following properties: / Chb(Sh(X)), X (F) / Gi X(F))) Then the assignment F 7→ G• (1) it defines an exact functor G• (2) the sheaves Gi (3) the map F X : Sh(X) X(F) are flabby for all i, X (F) is a resolution. / G• In particular, for any ring R the functor SX: = ΓX ◦ G• X : Sh(X; R) / Chb(Sh(Spec(k); R)) is exact. Using the smooth base change theorem (see e.g. [FK80, Theorem I.7.3]) and Propositions 3.1 and 3.6 above, we immediately deduce the following strengthening of Corollary 3.9. Corollary 3.10. Let k be a field, ks a separable closure. Fix a Godement resolution G• (−) for k-varieties as above, and write S(−): = Γ(−) ◦ G• (−) as above. Let X be a variety of finite type over k. Then for any A -- R-ℓ-adic sheaf F on X, a choice of compactification j : X ֒→ X determines an extension of the assignment to an exact functor F 7→ H ∗ c (X/ks; F ) Sh(X; Zℓ) / Chb(RepAR cts (Gal(ks/k); ℓ)) F 7→ SX(j!F )ks / / / / / / / / / / 20 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH 3.4. A Technical Lemma. We now establish the key technical lemma of the paper. As above, we fix a separable closure ks of k and a Godement resolution G• (−) for k-varieties. As above, we adopt the notation: S(−): = Γ(−)G• (−) i.e. S(−) is the strict model of the sheaf cohomology functor associated to the Godement resolu- tion. Lemma 3.11. Let X be a k-variety, let ι : U ֒→ X be the inclusion of an open subvariety. Let / Y be a proper map for which j ◦ ι is an embedding. Then for all sheaves F on U , there j : X is an induced isomorphism (of chain complexes of sheaves on Spec(k)) jΓ : SX(ι!F) ∼= / SY (j!ι!F). This isomorphism is natural in the following two senses: (1) For any map of sheaves g: F / F ′ on U , the diagram SX(ι!F) SX ι!(g) SX(ι!F ′) jΓ jΓ SY (j!ι!F) SY j!ι!(g) / SY (j!ι!F ′) commutes. (2) Given a commuting diagram (j1)U U ι1 V ι2 X j1 / Y j2 / Z where ιi is an open embedding and ji is a proper map such that ji ◦ ιi is an embedding for i = 1, 2, and j1(U ) ⊂ V , then for any sheaf F on U , (j2)Γ ◦ (j1)Γ = (j2 ◦ j1)Γ. Proof. Since j is proper, j! = j∗, so it suffices to prove the statement of the lemma with j∗ substituted for j!. Now let ι : U map such that jι : U chain complexes in Sh(Y )) /X be an open embedding, let F be a sheaf on U , and let j : X / /Y be a proper / Y is an embedding. We claim that there is a canonical isomorphism (of Granting the claim, we obtain jΓ by applying ΓY to this isomorphism and pre-composing with the natural isomorphism ΓX (−) ∼= ΓY ◦ j∗(−). j∗G• X(ι!F) ∼= / G• Y (j∗ι!F). In the notation of Section 3.3, the key observation underpinning the claim is that any map of / MZ (just by post-composing f with / Z determines a map of sets f : MW k-varieties f : W / each map x : Spec(Ωn) / W , and thus a commuting diagram (of non-finite type varieties) `MW w δW W δf f z `MZ δZ / Z / / / / / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 21 If f is an open embedding, then by the definition of f!, we have a canonical "base change" isomorphism Z f!F ∼= δf ∗δ∗ δ∗ W F. Since f is an embedding, δf is an embedding as well, and is both open and closed because the varieties in question are discrete. Therefore δf ! = δf ∗. We are now ready to prove the claim by induction. For the base of the induction, j∗G0 X (ι!F): = j∗δX∗δ∗ X ι!F ∼= j∗δX∗δι∗δ∗ U F ∼= δY ∗δj∗δι∗δ∗ U F (because ι is an embedding) (by the functoriality of (−)∗) Finally, because j ◦ ι is an embedding, by inspection of the definitions, we see that there is a canonical isomorphism δj∗δι∗δ∗ U F ∼= δ∗ Y j∗ι!F (i.e. y∗y∗j∗ι!F = y∗y∗F for y ∈ MU ⊂ MY and y∗y∗j∗ι!F = 0 otherwise.) We conclude that j∗G0 X (ι!F) ∼= δY ∗δ∗ Y j∗ι!F =: G0 Y (j∗ι!F). This settles the base of the induction. For the induction step, note that in the argument above, we have j∗G0 X (ι!F) ∼= j∗δX∗δι∗δ∗ U F ∼= j∗ι!δU ∗δ∗ U F where the second isomorphism follows by inspection from the definition of ι!. This implies that G1 X(ι!F): = coker(ι!F / ι!δU ∗δ∗ U F) ∼= ι!G1 U (F). In particular, G1 colimits, so X (ι!F) is again an extension by 0 of a sheaf on U . Because j is proper, j∗ preserves j∗G1 X (ι!F) ∼= G1 Y (j∗ι!F) and we can now apply the same argument as above to conclude that there exists a canonical isomorphism j∗Gi X (ι!F) ∼= Gi Y (j∗ι!F) for all i, thus proving the claim. It remains to show the two naturality properties. The naturality with respect to morphisms of sheaves g : F / F ′ follows immediately from the functoriality of the constructions above. For the second naturality property, a direct inspection of the construction above shows that the naturality property follows from the properness of the maps ji and the universal properties of cokernels and products (using that δX∗δ∗ Fx). Concretely, the naturality follows by writing out the explicit definitions of the maps and sheaves on Spec(k) and then repeatedly X F =Qx∈MX / K are injective, using that if I then for any sheaf of abelian groups A on the discrete space I, there are canonical isomorphisms / K are maps of sets such that ι : I / J and jι : I / J ι j YI Ai ∼=YI ∼=YI Ai × YJ−I Ai × YK−I 0 ∼=YJ 0 ∼=YK (ι!A)j ((jι)!A)k. (cid:3) / / / / / / 22 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH 4. The General Approach, by example In this section we prove a warm-up to our main result in order to illustrate the general ap- of Definition 2.22 (consisting of k-varieties with a choice proach. Recall the SW-category V cptd of compactification), and the exact functor k V cptd k / Vk given by forgetting the compactification. We now prove the following "warm-up" to Theorem 1.1, in order to illustrate our general approach. Theorem 4.1. Let k be a subfield of C, and R a commutative ring. Let Chb(R) be the category of homologically finite and bounded chain complexes of R-modules. Let [C • c (−; R)] denote the class of compactly supported R-valued singular cochains in the homotopy category. Then the functor V × k X 7→ [C • / Ho(Chb(R))op c (X(C); R)] admits a strict model as a span of weakly W -exact functors where the left-facing map induces an equivalence on K-theory. Vk ∼ V cptd k G / Chb(R)op Proof. For conciseness we abuse notation and write X for X(C) throughout this proof. We also suppress the coefficients on cohomology, which are always be taken to be R. The functor U : V cptd We define the functor G as follows. Let k / Vk forgets the choice of compactification. G(X, X) = C ∗ sing(X, X − X), be the chain complex of singular cochains with coefficients in R on the topological space X which vanish on all chains contained in X − X. The functor G!: co(V cptd k ) / Chb(R)op sends a closed embedding (Z, Z) is well-defined, since (as Z is closed in X) Z − Z ⊆ X − X. The functor / (X, X) to the usual pullback on cohomology. Note that this G!: comp(V cptd k ) / Chb(R)op sends an open embedding (U, U ) / (X, X) to the map which extends a cochain on U to a cochain on X by defining it to be 0 on all cochains which are not contained in U . The func- tor Gw: w(V cptd / Chb(R)op is defined similarly, with the analogous observation that a weak equivalence (X, X) / (X, X) produces a map of pairs (X, X − X) / (X, X − X). ) k By construction, the span Vk V cptd k G / Chb(R)op provides a strict model for the functor X 7→ [C • c (X(C); R)] It remains to prove that G is actually weakly W -exact. Conditions (1)-(4) hold by definition. We check the others in turn. / / o o / / / / / / / / / o o / DERIVED ℓ-ADIC ZETA FUNCTIONS 23 (5) By the definitions of the maps, it suffices to check that the diagram commutes on the com- pactification components. Thus the statement we need to check is that for any cartesian diagram X i Y j j ′ Z i′ / W where all maps are closed embeddings, (i′)! ◦ j′ compositions around the diagram take a cochain α: C∗(Y , Y − Y ) α′: C∗(Z, Z − Z) X, and to α(i ◦ σ) if it does. Thus the diagram commutes. / R which takes a chain σ: ∆n ! = j! ◦ i!. In this case, both of the / R to the cochain / Z to 0 if σ doesn't factor through (6) Suppose that we are given a subtraction sequence (Z, Z) i / (X, X) j ◦ (U, U ) . Note that we have the following commutative diagram of pairs of spaces: (U , U − U ) (X, X − U ) (X, X − X) (X − U, X − X) h◗◗◗◗◗◗◗◗◗◗◗◗ Applying C ∗ gives an exact sequence of cochain complexes across the middle, as it is the sequence associated to the triple (X, X −U, X −X). By excision, all vertical isomorphisms become quasi-isomorphisms on cochains. Thus we have the diagram (Z, Z − Z) C ∗(U , U − U ) ∼ i! (◗◗◗◗◗◗◗◗◗◗◗◗◗ C ∗(X, X − U ) C ∗(X, X − X) C ∗(X − U, X − X) )❙❙❙❙❙❙❙❙❙❙❙❙❙❙ j! ∼ C ∗(Z, Z − Z) where the dotted arrow is the result of applying C ∗, and the curved arrow is the quasi- inverse that extends chains by 0. Both of the solid arrow triangles commute. Thus the sequence G(U, U ) i! / G(X, X) j! / G(Z, Z) is weakly equivalent to an exact sequence, and thus is weakly exact. (7) The condition for complement maps holds because Gw and G! are given by the same functor. The condition for cofibrations holds because in both cases we extend a cochain on X ′ to a cochain on Y which is zero outside of X ′; the only difference is that applying the maps in one direction extends the cochain to Y ′ first, and then to Y , and in the other direction it extends to X first and then to Y . (cid:3) / / / / / / / o o o o o o O O h ( / / O O / / ) / / 24 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Now let R be any ring, and let χR denote the compactly supported R-valued Euler characteristic of homologically finite topological spaces, i.e. χR(X): = ∞Xi=0 (−1)i rkR H i c(X; R). We now deduce Theorem 1.3 as a corollary of Theorem 4.1. Corollary 4.2. Let k be a subfield of C. Then the strict model of Theorem 4.1 determines a contractible space of maps of K-theory spectra which fit into a commuting square XR: K(Vk) / K(R), K0(VC) π0XR K0(R) χR Z rk Z after applying π0. Proof. As discussed in Lemma 2.8, the inclusion of an exact category as the chain complexes concentrated at 0 is an equivalence, with the "Euler characteristic" as the inverse. Thus, we obtain from Theorem 4.1 a diagram of W -exact functors Vk V cptd k G / Chb(R)op Modfg(R)op By Example 2.24 and Lemma 2.8, the left-facing arrows induce weak equivalences on K-theory. Since a weak equivalence between fibrant-cofibrant spectra has a contractible space of inverses, we obtain a contractible space of maps XR: : K(Vk) h / K(V cptd k ) K(G) / K(Chb(R)) χ / K(Modfg(R)) = K(R) where h is any choice of inverse equivalence to K(V cptd K(Modfg(R)) /K(Vk) and χ is any choice of inverse to /K(Chb(R)). The commuting square after applying π0 now follows by inspection. (cid:3) ) k Granting the statement of Theorem 1.1, we can similarly deduce the following generalization of Corollary 1.2. Corollary 4.3. Let k be a field, ks a separable closure, ℓ 6= char(k) a prime, and g ∈ Gal(ks/k) any element. Then the strict model of Theorem 1.1 determines a contractible space of maps of K-theory spectra g∗ ◦ ζ : K(Vk) / K(Aut(Zℓ)). Passing to K0 and taking the characteristic polynomial, we recover the "g-Zeta-function" Zg : K0(Vk) / W (Zℓ) [X] 7→ 2 dim(X)Yi=0 det(1 − g∗t; H i et,c(X ×k ks; Zℓ))(−1)i . Proof. The map g∗ is the map induced on K-theory by the functor g∗ : Repcts(Gal(ks/k); Zℓ) / Aut(Zℓ) Gal(ks/k) (cid:9) V 7→ (V, g) / / / o o / o o / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 25 For the map ζ, consider the diagram of W -exact functors Vk V cptd k / Chb(Repcts(Gal(ks/k); Zℓ))op Repcts f g(Gal(ks/k))op from Theorem 1.1 (where the right-most arrow is again just inclusion of complexes concentrated in degree 0). The construction of ζ : K(Vk) / K(Repcts(Gal(ks/k); Zℓ)) now follows along identical lines to the construction of X in the proof of Corollary 4.2. The commuting square on π0 now follows from results of Almkvist [Alm78], explicated beau- (cid:3) tifully in Grayson [Gra79]. Remark 4.4. Taking k = Fq and g to be Frobenius, the "g-Zeta-function" is precisely the classical zeta function (see e.g. the discussion on p. 171-174 of [FK80]); in this special case Corollary 4.3 is exactly Corollary 1.2. 5. Proof of Theorem 1.1 We now prove Theorem 1.1, along the lines indicated in the proof of Theorem 4.1; the main difference is in the intricacy of the construction of the W -exact functor. By Proposition 3.6, it suffices to construct a W -exact functor (5.1) F : V cptd k / Chb(RepAR cts (Gal(ks/k); ℓ))op such that the cohomology of the inverse limit of the A -- R ℓ-adic chain complex produces compactly supported ℓ-adic cohomology (with the Galois action induced by the action on ks). Remark 5.2. (1) We prove in this section that the functor which assigns compactly supported cochains, defined via the Godement resolution, to the constant ℓ-adic sheaf Zℓ takes subtraction sequences of k-varieties to exact sequences of homologically bounded chain complexes of sheaves on Spec(k). (2) Our proof applies to the constant sheaf Zℓ, considered as a uniform system of sheaves / X, we have an on all k-varieties. The key property we use is that for any map f : Y ∼= Zℓ,Y . Our proof does not apply to any collection of sheaves on isomorphism f ∗Zℓ,X k-varieties for which this identity ever fails. In particular, we do not prove that there is a chain level realization of any cohomology functor which is exact on the category of all sheaves. (3) Our proof exploits a key difference between sheaf cohomology and singular cohomology of topological spaces. Namely, given a disjoint decomposition of a space X = Z ∪ (X − Z), it is not the case that a singular chain is either contained in Z or in X −Z. By contrast, every point of X is contained in one piece of the decomposition or the other. The Godement resolution is defined purely in terms of the points of the space; this is what allows our construction of compactly supported cochains to give an exact functor, as opposed to merely a weakly exact functor. As in Section 3.3, we fix once and for all a Godement resolution as our functorial flasque resolution for ´etale sheaves over a variety X, Notation 5.3. In this section, we make the following shorthand definitions. F 7−→ G• X F. X : denotes, for a k-variety X, the sheaf (Z/ℓnZ)Xks . An SX : denotes the functor ΓXG• X . Note that this functor lands in the category of chain complexes, not in the derived category, which is why we avoid the standard notation RΓ. o o / o o / / / 26 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH Following the discussion in Section 2, we proceed by defining functors F w n taking values in Chb(Repcts(Gal(ks/k); Z/ℓnZ)). We then show these fit together to define an A -- R-ℓ- adic system of continuous Galois representations. As all of the Gal(ks/k)-actions follow from the action on ks, we omit these from the notation, but the implication should be that all functors record this data as well. ! and F ! n , F n Throughout the construction below, we make repeated use of the isomorphism (of chain com- plexes of sheaves on Spec(k)) of Lemma 3.11 jΓ : SX(ι!F) ∼= / SY (j!ι!F) associated to an open embedding ιU ֒→ X and a proper map j : X to an open embedding denotes the "extension by 0" functor. /Y . Recall that (−)! applied We begin by constructing the functor n : comp(V cptd F ! k )op / Chb(Repcts(Gal(ks/k); Z/ℓnZ))op. On objects, it is given by (5.4) F ! n(γX : X ◦−→ X) := SX(γX!An X). By definition, a complement map j in V cptd k consists of a commuting square γU U U j j X γX / X where j is an open embedding and j is a closed embedding. Given such, we obtain a map via the following composition: F ! nj: SU (γU !An U ) / SX(γX!An X ) SU (γU !An U ) jΓ / SX (j!γU !An U ) ∼= / SX(γX!j!An U ) ∼= / SX(γX!j!j∗An X ) SX γX!(ǫ) / SX(γX!An X ). The first map is the isomorphism of Lemma 3.11. The middle two isomorphisms come from the canonical identification j!γU ! = γX!j! and An X . The last morphism ǫ is the counit of the U adjunction j! ⊣ j∗ (which exists because j is an open embedding). ∼= j∗An Lemma 5.5. The assignment j 7→ F ! nj is functorial on comp(V cptd k )op. Proof. The definition immediately implies that identities are mapped to identities, so we only need to check that composition is respected. For this, note that a composable pair of cofibrations in V cptd consists of a commuting diagram k γU U U j1 j1 j2 j2 X γX / X Y / Y γY / / / / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 27 in which the maps ji are open embeddings and the maps ji are closed embeddings. The above diagram yields the following commutative diagram: SU (γU !An U ) j1Γ (j2j1)Γ )❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ j2Γ SX(j1!γU !An U ) SY ((j2j1)!γU !An U ) ∼= SY ((j2j1)!γU !(j2j1)∗An Y ) ∼= ∼= ∼= SX(γX!j1!j∗ 1 An X) SX γX!ǫ1 SX(γX!An X ) j2Γ j2Γ / SY (j2!γX!j1!j∗ 1 An X ) SY (γY !(j2j1)!(j2j1)∗An Y ) SY j2!γX!ǫ1 / SY (j2!γX!An X ) ∼= SY γY !(ǫ21) SY (γY !j2!j∗ 2 An Y ) SY γY !ǫ2 / SY (γY !An Y ) The left-hand side of the diagram commutes by the naturality properties of j2Γ (by Lemma 3.11); the right-hand side of the diagram commutes because it commutes in Sh(Y ×k ks) before applying SY . The composition around the bottom is the map F ! n(j1), and the composition around the top is the map F ! n is functorial. (cid:3) n(j2) ◦ F ! n(j2 ◦ j1). Since the diagram commutes, F ! We now define the functor F n ! : co(V cptd k ) / Chb(Repcts(Gal(ks/k); Z/ℓnZ))op. On objects, it is equal to F ! commuting square n (see (5.4)). By definition, a cofibration i in V cptd k consists of a γZ Z Z i i X γX / X where the horizontal maps are closed embeddings. Given such, we obtain a map using the composition of morphisms ! i : SX(γX!An F n X ) / SZ(γZ!An Z ) SX(γX!An X) SX γX!(η) / SX(γX!i∗i∗An X ) ∼= / SX(i!γZ!An Z ) ∼= −1 Γ i / SX(γX!i!An Z ) / SZ(γZ!An Z). Here, η is the unit of the adjunction i∗ ⊣ i∗, and the last two isomorphisms come from the canonical identifications i! = i∗ (because i is proper), γX!i! = i!γZ! (by functoriality of (−)!) and An Z X. The last map is the inverse of the isomorphism of Lemma 3.11. ∼= i∗An Lemma 5.6. The functor F n ! is well-defined. Proof. The definition immediately implies that identities are mapped to identities, so we only need to check that composition is respected. A composable pair of complement maps in V cptd k ) / / / / / / / / / / / / / / / / 28 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH consists of a commuting diagram W γW W i2 i2 Z γZ / Z i1 i1 X γX / X in which the horizontal maps are closed embeddings. Given this diagram, we have the following diagram: SX(γX!An X ) SX γX!η1 SX γX!η12 +❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ SX γX!i1∗η2 SX(γX!i1∗i∗ 1An X ) SX(γX!(i1i2)∗(i1i2)∗An X) ∼= SX(i1!γZ!An Z ) i1Γ SZ(γZ!An Z ) SZ γZ!η2 SZ (γZ!i2∗i∗ 2An Z ) ∼= SX i1!γZ!η2 / SX(i1!γZ!i2∗i∗ 2An Z) ∼= SX(i1!γZ!i2!An W ) −1 1Γ i ∼= / SZ(i2!γW !An W ) ∼= SX(γX!(i1i2)!An W ) ∼= ∼= −1 2Γ i / SX((i1i2)!γW !An (i1i2)−1 Γ W ) / SW (γW !An W ) Here, ηa is the unit of the adjunction i∗ (i1i2)∗ ⊣ (i1i2)∗. The composition around the top is F n is F n Thus F n ! a ⊣ (ia)∗ for a = 1, 2 and η12 is the unit for the adjuction ! (i1i2); the composition around the bottom ! (i1). The diagram commutes by the naturality of η1,η2,η12 and by Lemma 3.11. (cid:3) ! (i2) ◦ F n is a functor. It now remains to construct F w n. Note that as the proof of Lemma 5.6 only used the results of Lemma 3.11, the proof works analogously to show that F w n . We define it using the same formula as for F ! n is well-defined. Lemma 5.7. The collection of functors {(F ! n )} defines a W -exact functor n, F n ! , F w cts (Gal(ks/k); ℓ))op. / Chb(RepAR F : V cptd k n}, {F n n+1⊗ZZ/ℓnZ /F ! Proof. The maps F ! n endow the collections n+1 {F ! ! } and {F w n } with the structure of projective systems. From the construction, for each fixed X, {F ! n } are obtained by applying SXγX! to A -- R-ℓ-adic systems of sheaves on X ×k ks. Therefore, by [FK80, Theorem I.12.15], {F ! n (X)} are A -- R-ℓ-adic complexes of sheaves on ks, i.e. A -- R-ℓ-adic complexes of continuous Gal(ks/k)- modules. So, we indeed have functors n and similarly for F n /F ! ! (X)} and {F w ! (X)} and {F w n(X)}, {F n n(X)}, {F n ! and F w / (Chb(RepAR cts (Gal(ks/k); ℓ))op)op k F ! : comp(V cptd F! : co(V cptd F w : w(V cptd ) ) k k )op / Chb(RepAR / Chb(RepAR cts (Gal(ks/k); ℓ))op cts (Gal(ks/k); ℓ))op. It remains to verify that F is W -exact. Axioms (1)-(4) hold by definition, so we check the remainder in turn. / / / / / / + / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 29 First, consider Axiom (5). It suffices to prove it for Fn. This is a large but straightforward ! . For those who would like to see the details, n and F n diagram chase using the definitions of F ! we present them in Appendix B. Now we check Axiom (6): that F takes subtraction sequences to exact sequences. Again, it suffices to show it for Fn. Given a subtraction sequence in V cptd k (Z, Z) j / (X, X) i ◦ (U, U ) we obtain a sequence in Sh(X ×k ks) 0 / j!An U / An X / i∗An Z / 0 and we see that this is exact by direct inspection (i.e. by verifying exactness on stalks). Applying cts (Gal(ks/k); ℓ)) (using that the Godement SXγX!, we obtain an exact sequence in Chb(RepAR resolution is an exact functorial flabby resolution, see e.g. [FK80, p. 129]). Consider the following diagram: SU (γU !An U ) jΓ F ! n(j) 'PPPPPPPPPPPP SX(γX!j!An U ) / SX(γX!An X ) SX(γX!i∗An Z) 'PPPPPPPPPPPP ! (i) F n −1 Γ i SZ (γZ!An Z ) The exact sequence across the middle is isomorphic to the diagonal sequence, which is thus also exact. Therefore Fn takes subtraction sequences to exact sequences, as desired. It remains to verify Axiom (7); as before, it suffices to prove that it holds for Fn. The proof of (7) for complement maps is identical to the proof of Lemma 5.6, since it is simply checking that the transformation defined in Lemma 3.11 respects composition. The proof of (7) for cofibrations is identical to the proof of (5), since any such commutative diagram is automatically a pullback square, and F w (cid:3) n is defined identically to F n ! . 6. Nontrivial Classes in the Higher K-Theory of Varieties functor FinSet Let FinSet be the category of finite sets. Consider the map S / K(Vk) induced by the exact / Vk given by A 7→`A Spec(k). This induces a homomorphism S / K∗(Vk). The stable homotopy groups of spheres have a rich higher structure, but it is not clear that this structure is not annihilated when passing to varieties. In this section we explore this map in the cases where k is a subfield of C and when k is a finite field. In the case when k is a subfield of C we show that the image of this map is nontrivial even above the 0-th homotopy group; in the case when k is finite we show that K∗(S) is a direct summand of K∗(Vk) and prove that this map is not surjective when ∗ = 1. Moreover, we are able to extend this result to local and global fields which contain a place of cardinality congruent to 3 mod 4, as well as to subfields of R. Our analysis considers a particularly simple family of maps involving permutations of varieties. In order to construct these maps, it is useful to have a model of the sphere spectrum S within SW-categories. The following lemmas furnish this. Lemma 6.1. The category of finite sets FinSet is an SW-category where cofibrations are monomor- phisms and subtraction sequences are diagrams [m] ֒→ [n] ←֓ [n − m] where the images of the two maps are disjoint. / o o / / / / ' / / / ' / / / 30 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH The category FinSet+ of finite pointed sets can be considered to be a Waldhausen category by defining the cofibrations to be injections and the weak equivalences to be isomorphisms. Write S+ for a set S with a disjoint basepoint added. We can define a W -exact functor (I !, I!, I w): FinSet / FinSet+ by defining I(S) = S+ for any finite set S, I!(i) = i+ for any cofibration i and I w(i) = i+ for any isomorphism i. To define I ! we consider an injection j: S / T , considered as a complement / S+ by taking t to j−1(t) when t is in the image of map. We can define an "inverse" map T+ j, and to the basepoint otherwise. Proposition 6.2. The W -exact functor (I !, I!, I w): FinSet /FinSet+ induces an isomorphism K SW (FinSet) / K W (FinSet+); here, K SW takes the K-theory of FinSet as an SW -category, and K W takes the K-theory of FinSet+ as a Waldhausen category. • FinSet ∼= S(n) • FinSet+. To check this, it suffices to check that I induces an isomorphism of categories at each simplicial level; this follows because I bijectively renames the terms in each object of (cid:3) Proof. It suffices to check that I induces an isomorphism of n-simplicial categories eS(n) eS(n) We now define the family of maps we use. FinSet. i1···in Definition 6.3. For each X ∈ Vk, define to be the exact functor of SW -categories FinSet / Vk induced by σX : FinSet / Vk Thus we get a family of maps [n] 7−→ X ∐ · · · ∐ X . n times {z } π∗σX: π∗(S) / K∗(Vk). 6.1. Subfields of C. We appeal to some facts from Adams [Ada66] and Quillen's letter to Milnor [Qui76]. Recall that there is a homomorphism from the stable homotopy of the stable orthogonal group to the stable homotopy groups of spheres: J : πs / π∗S. This is constructed as follows. / Rn, by one point compactification, it defines Since every element of O(n) defines a map Rn / Map∗(Sn, Sn) ∼= ΩnSn. Taking a point map Sn homotopy groups and stabilizing yields the J-homomorphism. The following is due to Adams / Sn. Thus, there is a map of spaces O(n) ∗O / Theorem 6.4. [Ada66, Thm. 1.5] The map J: π4s−1O summand of π4s−1S and the image is / π4s−1S exhibits π4s−1O as a direct J(π4s−1(O)) ∼= Cds ds = denominator(cid:18) Bs 4s(cid:19) where Bs is the sth Bernoulli number, and Cds denotes the cyclic group of order ds. There is also a map pi : πiS / BGL(Z). For our purposes, it is more useful to regard it as the map induced by the exact functor of Waldhausen categories / Ki(Z) induced by the inclusion BΣ∞ Quillen's letter to Milnor gives the following. P : FinSet+ / Modfg(Z) [n]+ 7→ Z ⊕ · · · ⊕ Z / / / / / / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 31 Theorem 6.5. [Qui76] The composite map π4s−1(O) J / π4s−1(S) p4s−1 / K4s−1(Z) is injective. We can use these results to identify nontrivial elements in K2s−1(Vk). Theorem 6.6. Let k be a subfield of C. There are infinitely many non-trivial homotopy groups of K∗(Vk). In particular, for all s > 0, K4s−1(Vk) is non-trivial. Proof. Fix s > 0, and let X ∈ Vk be a projective variety such that its compactly-generated Euler characteristic χc(X) is relatively prime to ds. (Note that this is always possible, as for example χc(CP n) = n for all n.) Let C ∗ c (X) be the compactly-supported singular chains on the complex points of X. Consider the diagram Vk U f▲▲▲▲▲▲▲▲▲▲▲▲ σX V cptd k σ(X,X) /❴❴❴❴❴❴ Chb(Z) ·[0] Modfg(Z) hPPPPP ⊗C ∗ c (X) PPPP FinSet I /❴❴❴❴❴ FinSet+ P / Modfg(Z) where the left half of the diagram consists of SW -categories and the right half consists of Wald- hausen categories; both dashed arrows are W -exact functors. This diagram commutes, in the sense that the left-hand triangle commutes as a triangle of exact functors, and the middle penta- gon commutes as a diagram of W -exact functors (as W -exact functors can be pre/postcomposed with exact functors). The bottom dashed arrow is the W -exact functor I constructed above Proposition 6.2. Applying π0 and noting that the two solid arrows across the top become isomor- phisms we recover the motivic measure constructed in Corollary 4.2. Upon applying K-theory we get a diagram K4s−1(Vk) ∼= K4s−1(V cptd k ) / K4s−1(Chb(Z)) χ / K4s−1(Z) h❘❘❘❘❘❘ σX ❘❘❘❘❘❘ J σ(X,X) / π4s−1S π4s−1(O) ∼= π4s−1S i❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘ p4s−1 ·χc(X) / K4s−1(Z) where the isomorphism induced by I is the identity. By Theorems 6.4 and 6.5, the composition / K4s−1(Z) is injective; since χc(X) is relatively prime to ds, the across the bottom π4s−1(O) composition π4s−1(O) / K4s−1(Z) across the bottom and to the upper-right is also injective. Since this factors through K4s−1(Vk) the theorem follows. (cid:3) Remark 6.7. In the above proof we are not really using anything interesting about X; in particular, we could take X to be Spec k and the proof still goes through. This makes sense, as the image of J sees nontrivial elements of K-theory which can be described in terms of permutations, which are automorphisms of 0-dimensional varieties. However, we believe that the more complex proof is useful due to the possibility of generaliza- tion. It should be possible to use the same construction in the proof to show that there are more interesting nontrivial elements in higher homotopy groups of K(Vk) by exploiting the structure of the cohomology of C ∗ c (X). For example, if the measure can be enriched to land in the K-theory of mixed Hodge structures then by selecting an X with a nontrivial mixed Hodge structure the / K∗(V) induced above proof would detect an element which is not in the image of the map S by the inclusion of 0-dimensional varieties. / / o o / o o f O O / / h o o / / / h / / O O / i O O / / / 32 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH 6.2. Finite Fields. When k is finite the functor Vk X 7−→ X(k) / FinSet homotopy group). A priori it may be the case that all higher homotopy groups of K(Vk) are in the image of the homotopy groups of S. gives a splitting of the K-theory spectrum K(Vk) as K(Vk) ≃ S ∨ eK(Vk) (and thus of each We show that this is not the case by using ζ to identify a nonzero element in eK1(Vk) for k = Fq (with q ≡ 3 mod 4). For such k, we construct a surjective homomorphism h2: K1(Vk) / Z/2 / Z/2 is trivial. The map h2 is defined as follows. such that the composition π1(S) Let ⋆ be the operation defined in [Mil71, §8]. Grayson [Gra79] has shown that given a pair of commuting automorphisms f, g on a Qℓ-vector space V , the map / K1(Vk) h2 (f, g) 7→ f −1 ⋆ g, / K2(Qℓ). Moore's Theorem (see [Mil71, Appendix], induces a homomorphism sℓ: K1(Aut(Qℓ)) or [Wei05, Theorem 57] and its proof) shows that the mod-ℓ Hilbert symbol gives a split surjection (−, −)ℓ : K2(Qℓ) ։ µ(Qℓ) onto the roots of unity in Qℓ with kernel an uncountable uniquely divisible abelian group U2(Qℓ). For q odd, we then define h2 to be the composition K1(VFq ) π1(F rob∗◦ζ) / K1(Aut(Q2)) s2 / K2(Q2) (−,−)2 / Z/2, where F rob∗ is the map defined in Corollary 4.3 for g = F rob and (−, −)2 is the 2-adic Hilbert symbol. Fix a variety X, and consider the following diagram: (6.8) π1σ(X,X) K1(V cptd k ) h2 Z/2 K1(U ) π1S π1σX K1(Vk) If X is proper then σ(X,X) exists, and the diagram commutes with the dotted arrow added. Theorem 6.9. When k = Fq with q ≡ 3 (mod 4) the map h2 defined above detects a nontrivial class in eK1(Vk). in its kernel. This shows that eK1(Vk) 6= 0. Proof. Let η ∈ π1(S) be the nonzero element. We show that h2 is nonzero but contains (π∗σSpec k)(η) The difficult part of computing h2 is computing the image of s2 in K2(Q2). This is done using the ⋆ map, discussed in detail in [Mil71, Chapter 8]. As we are using many of the results in the section we will not cite them individually; we simply note that the important components are the definition of the general ⋆ on page 57, the proof that ⋆ is bilinear in Theorem 8.8 and Lemma 8.3. Let τ : {1, 2} The cohomology of a point is concentrated in degree 0, with Frobenius acting trivially; thus / {1, 2} be the transposition of two points. τ ∗ ⋆ F robq is trivial, and h2(π1σSpec k(t)) = 1 in Z/2. Now let X = P1, and consider the automorphism α: x 7→ 1/x acting on P1. For k = Fq, we can write α (cid:9) (H ∗ et,c(P1Fq ; Q2), F robP1 q ) as a direct sum (in Aut(Q2)) F robq acts on the first summand by 1, and on the second by q; α acts on the first summand by 1 and on the second by −1. Since both summands are in even degrees we can disregard degree (1 (cid:9) Q2(0)) ⊕ (−1 (cid:9) Q2(−1)). / / / / / / / / / / / / 2 2 / DERIVED ℓ-ADIC ZETA FUNCTIONS 33 in our computations; thus we have a module Q2 ⊕ Q2 with the automorphism (1 ⊕ q) as our object of Aut(Q2); we have the extra automorphism (1 ⊕ −1) acting on this. Since ⋆ is bilinear, to compute the image of this under s2 it suffices to compute q ⋆ −1. By definition q−1 1  ⋆   −1 1 −1   = {q, −1}{q−1, 1}{1, −1} = {q, −1} ∈ K2(Q2). The image of this under h2 is (q, −1)2, which, when q ≡ 3 mod 4, is −1. This gives the desired (cid:3) q q ⋆ −1 =  element in eK1. The last computation in the proof above shows that in general, for f and g automorphisms of a one-dimensional module, f ⋆ g = {f, g}. 6.3. Elements in K1 for other fields. Suppose that k is a global or local fields with a place of cardinality equivalent to 3 mod 4, or k ⊂ R. The results from the previous section hold with almost identical proofs, with slightly different definitions. For k a global or local field with a place of cardinality equivalent to 3 mod 4, we pick a Frobenius element φ for this place, and define h2 to be the composition K1(Vk) π1(φ∗◦ζ) / K1(Aut(Q2)) σ2 / K2(Q2) (−,−)2 / Z/2, Similarly for k ⊂ R, we define h2 to be the composition K1(Vk) π1((·)∗◦ζ) / K1(Aut(Q2)) σ2 / K2(Q2) (−,−)2 / Z/2, where · denotes complex conjugation and (·)∗ is the map defined in Corollary 4.3 for g = ·. Theorem 6.10. For k a global or local field with a place of cardinality equivalent to 3 mod 4, or for k ⊂ R, eK(Vk) has nontrivial higher homotopy groups. Remark 6.11. We expect that eK(Vk) has nontrivial higher homotopy groups in general, however the particular class that we use h2 to detect requires the assumptions on k. Using the methods of this paper, it should be possible to find a different example that gives a nontrivial class for all odd q, all global and local fields with a place of odd cardinality, and all non-algebraically closed subfields of C. For even q, one would need the p-adic (rather than the ℓ-adic) analogue of the derived zeta function to employ the present approach. Proof of Theorem 6.10. This proof works the same way as the proof of Theorem 6.9. If k is a global or local field with a place of cardinality 3 mod 4, then, for any Frobenius element φ for this place, the action of φ on the ´etale cohomology of P1 factors through the action on the special fiber. In particular, the above computation similarly shows that the map h2 takes the class α (cid:9) P1 to −1, and thus gives the desired element in eK1(Vk). For k ⊂ R, we can consider the same X. We write π1α (cid:9) (H ∗ et,c(XC; Q2), ·) as a direct sum (1 (cid:9) (Q2, 1)) ⊕ (−1 (cid:9) (Q2, −1)). The map s2 sends everything with the identity acting on it to the unit, so the image of this under (cid:3) h2 is (−1, −1)2 = −1. This gives the desired element in eK1. Theorem 1.9 is a direct consequence of Theorems 6.9 and 6.10. Remark 6.12. It should be possible to do a more powerful analysis on K1 by exploiting the rich structure of automorphism groups of varieties and applying Proposition 2.20. / / / / / / 34 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH 7. Questions for Future Work Indecomposable Elements in K(Vk). Theorems 6.9 and 6.10 establish that there are non- trivial classes in the higher K-theory of varieties that do not come from the sphere spectrum. However, one could ask a more refined question: since K(Vk) is an E∞-ring spectrum, its homo- topy groups K∗(Vk) form a ring. We therefore have a ready supply of elements of K∗(Vk): those in the image of the multiplication β: K0(Vk) ⊗ π∗(S) 1⊗σSpec k −−−−−−→ K0(Vk) ⊗ K∗(Vk) / K∗(Vk). We call such elements decomposable. A priori, it may be the case that this map is surjective, and that therefore all higher homotopy groups of K(Vk) are decomposable. The example constructed in Section 6 is decomposable, since this can just be written as η · [P1]. Question 7.1. Indecomposable elements. Do there exist indecomposable elements in K∗(Vk)? As we explain in Remark 7.2 below, we do not expect the map h2 to be able to distinguish decomposable from non-decomposable elements. Instead, we hope that by expanding the collec- tion of derived motivic measures and employing Proposition 2.20 judiciously, a suitable invariant could be found. Remark 7.2. We expect the derived ℓ-adic zeta function to factor through the K-theory of the Morel-Voevodsky category of motivic spectra, via the construction of compactly supported inte- gral motivic cohomology. This suggests that the invariant h2 : K1(Vk) / K2(Q2) in Section 6 factors through the composition We highlight three implications of this expected factoring: K1(Aut(Z)) / K2(Z) / K2(Q) / K2(Q2). (1) It underscores the importance of the 2-adic Hilbert symbol, as opposed to the ℓ-adic Hilbert symbol for ℓ 6= 2. [Mil71, Theorem 11.6]), the map K2(Z) / K2(Q) is split injective, with the splitting given by the 2-adic Hilbert symbol. In particular, the Hilbert symbols (−, −)ℓ for ℓ 6= 2 identically vanish on K2(Z). Further, no classes outside the summand µ(Q2) ⊂ K2(Q2) are in the image of K2(Z). Indeed, by Tate's computation of K2(Q) (see e.g. (2) It suggests that we should not expect the map h2 to be able to distinguish indecomposable elements in eK1(Vk). Indeed, Milnor's computation [Mil71, Corollary 10.2] shows that the / K2(Z) is surjective and the nontrivial class in K2(Z) = Z/2Z is map K1(Aut(Z)) mapped onto by decomposable classes. (3) It suggests that the higher invariants of the derived zeta functions should be in some sense independent of ℓ. It would be fruitful to understand this more precisely! Other Derived Motivic Measures. The recipe in this paper should work to construct derived motivic measures for other cohomological invariants. We took ℓ-adic cohomology as the basis for our derived zeta function. One would like analogous maps for the other Weil cohomology theories. Problem 7.3. Derived p-adic zeta functions. Let k be a perfect field of characteristic p with Witt vectors W (k). Construct a map of K-theory spectra K(Vk) / K(Aut(W (k))) which lifts the function sending a variety X to H ∗ cohomology (with constant coefficients) acted on by the Frobenius automorphism. rig,c(X/W (k)) to its compactly supported rigid We expect that the construction should parallel that in Section 4, with the category V cptd replaced by a category of varieties X equipped with a choice of compactification X ֒→ X, and a / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 35 /(X, Y , Y) extending X ֒→ X as in [Ber86, Section choice of map of admissible triples (X, Y, Y) 3], and with rigid cohomology replacing the ℓ-adic constructions. Note that, Tsuzuki's finiteness theorem [Tsu03, Theorem 5.1.1] plays an essential role in defining the W -exact functor. Problem 7.4. Derived Serre Polynomial. Let k be a field of characteristic 0. Construct a map of K-theory spectra taking values in the K-theory of integral mixed Hodge structures K(Vk) / K(M HSZ) which lifts the function sending a variety X to H ∗ structure. c (X(C); Z) with its canonical mixed Hodge We expect that the construction should parallel that in Section 4, with the category V cptd replaced by a category of varieties X equipped with a choice of compactification X ֒→ eX, and a choice of cubical hyperresolution eX• / eX of the pair (eX, eX − X) (see e.g. [PS08, Chapter 5]), and with logarithmic differential forms in lieu of the ℓ-adic constructions. The framework of motives suggests that the derived ℓ-adic zeta function, along with the two maps described above, should factor through a derived motivic measure built from motivic coho- mology. Problem 7.5. Derived Gillet -- Soul´e. Let k be a field admitting resolution of singularities. Con- struct a map of K-theory spectra taking values in the K-theory of integral Chow motives over k K(Vk) / K(Mk) which lifts the motivic measure of Gillet -- Soul´e [GS96] sending a k-variety X to its compactly supported integral Chow motive. Prove that the derived ℓ-adic zeta function and the derived Serre polynomial factor through this map. We expect that the replacement of V cptd should be the same as for the Serre polynomial. For general fields k, one might expect to have a motivic measure based on integral Voevodsky motives through which all of the above maps factor. Moving further away from cohomological invariants, one of the richest motivic measures is Kapranov's motivic zeta function K0(Vk) / W (K0(Vk)) [X] 7→ ∞Xi=0 [Symi(X)]ti. Question 7.6. Derived motivic zeta function. Does Kapranov's motivic zeta function lift to a map of K-theory spectra? For motivation, recall that Weil's realization that the classical zeta function of a variety over a finite field can be obtained cohomologically provided a robust strategy for proving that the zeta function of such varieties is rational. Similarly, a lift of Kapranov's motivic zeta function to a map of K-theory spectra might be expected to go a long way toward proving that the motivic zeta function is rational, in an appropriate sense. Recall, however, that purely as a map out of K0(VC), Kapranov's motivic zeta function is not rational. This was proven by Larsen and Lunts [LL03], and the key tool in their proof was a motivic measure µLL : K0(VC) / Z[SBC] taking values in the free abelian group on stable birational equivalences classes of complex vari- eties. Note that, since P1 ∼SB ∗, Larsen and Lunts' measure takes the class of the affine line to 0. In particular, it still may be the case that Kapranov's motivic zeta is rational after inverting / / / / / / 36 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH the affine line, or performing some other modification of K0(Vk) (cf. [LL04]). This underpins the "in the appropriate sense" above. Question 7.7. Derived Larsen -- Lunts. Does the Larsen -- Lunts measure lift to a map of K-theory spectra? Using the formalism of assemblers, the third-named author was able to accomplish this [Zak17a]. However, it would be desirable to have a direct construction of this motivic measure. For this, one would need an SW-category which naturally encodes stable birational equivalence of projective varieties. Appendix A. Functorial factorization of weak cofibrations In this appendix, we define and verify in the cases of interest the technical condition "functorial factorization of weak cofibrations" defined in [BM08]. The condition is meant to be a weakening of Waldhausen's cylinder functors [Wal85, Sect. 1.6]. Definition A.1. Let [1] denote the ordered set 0 < 1 considered as a category, and let [2] the ordered set 0 < 1 < 2, also considered as a category. A functorial factorization is a functor ϕ: Fun([1], C) / [2] takes 0 to 0 and 1 to 2. / Fun([2], C) such that (d1)∗ ◦ ϕ = 1Fun([1],C). Here, d1: [1] Definition A.2. Let C be a Waldhausen category. A weak equivalence between morphisms f : A / D is a diagram / B and g: C A ∼ C f g B ∼ / D A weak cofibration is a map A A′ ֒→ B′. / B that admits a zig-zig of weak equivalences to a cofibration Definition A.3. [BM08, Definition 2.2] Let C be a Waldhausen category. Write Funwc([1], C) for the full subcategory of Fun([1], C) consisting of those functors whose image is a weak cofibration. Let Func,w([2], C) be the full subcategory of Fun([2], C) consisting of those diagrams which are a cofibration followed by a weak equivalence. A functorial factorization of weak cofibrations is a functor ϕ: Funwc([1], C) / Func,w([2], C) such that (f 1)∗ ◦ ϕ = 1Funwc([1],C). Let C be a cofibrantly generated model category. We thus have a functorial factorization Fun([1], C) / Func,w([2], C), given by the functorial factorization of morphisms into a cofibration followed by an acyclic fibra- tion. If a Waldhausen category arises as a subcategory of a model category, we can often leverage this factorization to obtain functorial factorizations inside the Waldhausen category. The main problem is that objects in a Waldhausen category need to be small (in some sense), whereas functorial factorizations often produce very large objects. However, in Lemma 2.8 we showed that being homologically small is sufficient; when weak equivalences are quasi-isomorphisms this is therefore sufficient. To produce the functorial factorizations that we need we appeal to a theorem of Beke [Bek00] which produces model category structures on categories of chain complexes. A version of this theorem is also proved in [Hov01, Theorem 2.2]. Theorem A.4. [Bek00, Proposition 3.13] Let A be a Grothendieck abelian category, i.e. A admits a generator, has small colimits, and filtered colimits commute with finite limits. Then Ch(A) admits a cofibrantly generated model structure where / / / / / / / / / / DERIVED ℓ-ADIC ZETA FUNCTIONS 37 • weak equivalences are quasi-isomorphisms • cofibrations are injections • fibrations have the right lifting property with respect to trivial cofibrations This entitles us to the following two theorems. Theorem A.5. The category Chb(Repcts(Gk); ℓ) admits functorial factorization of weak equiva- lences. Proof. As noted in Proposition 3.1, the category Repctscts(Gk; ℓ) is equivalent to the category Shet(Spec(k); ℓ). This is the category of sheaves of an abelian group on a ringed site, as such, it is a Grothendieck category. By Beke's Theorem, Ch(Repctscts(Gk; ℓ)) admits a cofibrantly generated structure. Thus, all morphisms in Ch(Repctscts(Gk; ℓ) have functorial factorizations. Given a weak / B; it remains to show that C ∈ Chb(Repctscts(Gk; ℓ)). cofibration A / B is a weak equivalence and B is homologically bounded, C must be as However, since C well. (cid:3) / B factor it as A ֒→ C Theorem A.6. The category Chb(R) admits functorial factorization of weak equivalences. Proof. The category ModR is a Grothendieck abelian category. As such, Ch(ModR) has a cofi- brantly generated injective model structure. Thus, the category of all maps X /Y in Ch(ModR) has a functorial factorization of the required form. Restricting to Chb(ModR), as in the previous proof, we see that Chb(ModR) does as well. (cid:3) Appendix B. The proof of Axiom (5) Axiom (5) states the following. Suppose that we are given a cartesian diagram We must show that (X, X) ✤ f (Y, Y ) ✤ i j (Z, Z) g / (W, W ). Fn(X, X) F ! n(f ) Fn(Y, Y ) F ! n(i) F ! n(j) / Fn(Z, Z) F n ! (g) Fn(W, W ) commutes. Note that since f and g are both closed, we know that on sheaves f∗ = f! and g∗ = g!. / / / / ✤ / / _ _ ✤ / / / / O O O O 38 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH First, note that the following diagram (in Chb(RepAR cts (Gal(ks/k)); ℓ)) commutes. SY γY !An Y SY γY !η SY γY !f∗f ∗An Y ∼= SY γY !f!An X −1 Γ f SXγX!An X jΓ jΓ jΓ ∗ iΓ / SW γW !j!An Y SW γW !j!η SW γW !j!f∗f ∗An Y ∼= SW γW !j!f!An X ∼= SW γW !g!i!An X g−1 Γ / SZ γZ!i!An X ∼= ∼= ∼= ∼= ∼= / SW γW !j!j∗An W SW γW !j!ηj∗ / SW γW !j!f∗f ∗j∗An W ∼= SW γW !j!f∗i∗g∗An W ∼= / SW γW !j!f!i∗An Z ∼= / SW γW !g!i!i∗An Z SW γW !g!ǫ g−1 Γ / SZ γZ!i!i∗An Z SZ γZ!ǫ SW γW !g!An Z g−1 Γ / SZ γZ!An Z The pentagon marked ∗ commutes because of the naturality of (−)Γ. The composition around the bottom is F ! n(i) ◦ F n ! (f ). Now consider the following diagram in Sh(W ×k ks): j!j∗An W j!ηj∗ = ⋆ j!j∗An W j!j ∗η j!f∗f ∗j∗An W j!j∗g∗g∗An W ǫ ǫ An W η g∗g∗An W ∼= j!αg∗ ❧❧❧❧❧ j!f∗i∗g∗An W ❧❧❧❧ ∼= ∼= ∼= j!f!i∗An Z ∼= g!i!i∗An Z j!α / j!j∗g!An Z ǫ † g!ǫ / g!An Z ∼= g!An Z Here, it is important to keep in mind that since the original diagram of varieties is Cartesian (and since f∗ = f! and g∗ = g!), there is a natural isomorphism α: f!i∗ ∼= f∗i∗ =⇒ j∗g∗ ∼= j∗g!. The only two parts of this diagram which do not commute by definition are the two pentagons marked ⋆ and †. To see that these commute, it suffices to check that the two diagrams f∗f ∗j∗An W ∼= η(f )j∗ j∗An W j ∗η(g) f∗i∗g∗An W αg∗ / j∗g∗g∗An W / / / / / / / / / / / / / / / / / / / / 6 6 / / / / o o / and DERIVED ℓ-ADIC ZETA FUNCTIONS 39 j!f!i∗An Z ∼= g!i!i∗An Z j!α ǫ(i) j!j∗g!An Z ǫ(j) / g!An Z commute. That these commute follows directly from the definition of the base change homo- [FK80, p. 60]). More conceptually, base change (and proper push forward) morphism (see e.g. preserves the base change isomorphism. After applying SW γW ! to this diagram, it fits into the rectangle in the upper-right in the above diagram; then the composition around the top is F n ! (g) ◦ F ! n(j). This proves Axiom (5). References [Ada66] [Alm78] [Bar16] [Bek00] [Ber86] [BM08] [BM11] J. F. Adams, On the groups J(X). IV, Topology 5 (1966), 21 -- 71. MR 0198470 Gert Almkvist, K-theory of endomorphisms, J. Algebra 55 (1978), no. 2, 308 -- 340. MR 523461 (80i:18018) Clark Barwick, On the algebraic K-theory of higher categories, J. Topol. 9 (2016), no. 1, 245 -- 347. MR 3465850 Tibor Beke, Sheafifiable homotopy model categories, Math. Proc. Cambridge Philos. Soc. 129 (2000), no. 3, 447 -- 475. MR 1780498 P. Berthelot, G´eom´etrie rigide et cohomologie de vari´et´es alg´ebriques de caract´eristique p, M´emoires de la S.M.F. 23 (1986), 7 -- 32. Andrew J. Blumberg and Michael A. Mandell, The localization sequence for the algebraic K-theory of topological K-theory, Acta Math. 200 (2008), no. 2, 155 -- 179. MR 2413133 , Algebraic K-theory and abstract homotopy theory, Adv. Math. 226 (2011), no. 4, 3760 -- 3812. [BP72] MR 2764905 Michael Barratt and Stewart Priddy, On the homology of non-connected monoids and their associated groups, Comment. Math. Helv. 47 (1972), 1 -- 14. MR 0314940 [BW87] Marcel Bokstedt and Friedhelm Waldhausen, The map BSG / QS0, Algebraic topology and algebraic K-theory (Princeton, N.J., 1983), Ann. of Math. Stud., vol. 113, Princeton Univ. Press, Princeton, NJ, 1987, pp. 418 -- 431. MR 921487 J. Campbell, The K-theory spectrum of varieties, arXiv:1505.03136. B. Conrad, Deligne's notes on Nagata compactifications, J. Ramanujan Math. Soc. 22 (2007), no. 3, 205 -- 257. Jonathan Campbell and Inna Zakharevich, CGW-categories, arXiv: 1811.08014. P. Deligne, La conjecture de Weil: I, Inst. Hautes ´Etudes Sci. Publ. Math. 43 (1974), 273 -- 307. / A(∗) [Cam] [Con07] [CZ] [Del74] [Del77] [FK80] [Gra79] [GS96] [Hal05] , Cohomologie ´Etale (SGA 4 1/2), Springer, 1977. E. Freitag and R. Kiehl, ´Etale Cohomology and the Weil Conjectures, Springer-Verlag, 1980. D. R. Grayson, K2 and the K-theory of automorphisms, J. Algebra 58 (1979), no. 1, 12 -- 30. H. Gillet and C. Soul´e, Descent, motives and K-theory, J. reine angew. Math. 478 (1996), 127 -- 176. Thomas C. Hales, What is motivic measure?, Bull. Amer. Math. Soc. (N.S.) 42 (2005), no. 2, 119 -- 135. MR 2133307 [Hov01] Mark Hovey, Model category structures on chain complexes of sheaves, Trans. Amer. Math. Soc. 353 [Hun74] [LL03] [LL04] [Mil71] (2001), no. 6, 2441 -- 2457. MR 1814077 T. Hungerford, Algebra, Springer, 1974. M. Larsen and V. Lunts, Motivic measures and stable birational geometry, Mosc. Math. J. 3 (2003), no. 1, 85 -- 95. , Rationality criteria for motivic zeta functions, Compos. Math. 140 (2004), no. 6, 1537 -- 1560. J. Milnor, Introduction to algebraic K-theory, Princeton University Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1971, Annals of Mathematics Studies, No. 72. MR 0349811 [MMSS01] M. A. Mandell, J. P. May, S. Schwede, and B. Shipley, Model categories of diagram spectra, Proc. London Math. Soc. (3) 82 (2001), no. 2, 441 -- 512. MR 1806878 [Nag62] M. Nagata, Imbeddings of an abstract variety in a complete variety, J. Math. Kyoto Univ. 2 (1962), no. 1, 1 -- 10. / / / / / 40 JONATHAN CAMPBELL, JESSE WOLFSON, AND INNA ZAKHAREVICH [NS11] [PS08] [Qui73] [Qui76] [Sah] [Sch05] [Seg74] [Ste17] [Tsu03] [TT90] [Wal85] [Wei05] [Wei13] [Zak17a] Johannes Nicaise and Julien Sebag, The grothendieck ring of varieties, London Mathematical Society Lecture Note Series, vol. 1, p. 145188, Cambridge University Press, 2011. C.A.M. Peters and J.H.M. Steenbrink, Mixed Hodge structures, Springer, 2008. D. Quillen, Algebraic K-theory, I: Higher K-theories (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), ch. Higher algebraic K-theory: I, pp. 85 -- 147, Springer, 1973. , Letter from Quillen to Milnor on Im(πiO → πs i → KiZ), 182 -- 188. Lecture Notes in Math., thesis, Universit´e Bordeaux. Vol. 551. MR 0482758 Neeraja Sahasrabudhe, Grothendieck ring of varieties, Masters http://algant.eu/documents/theses/neeraja.pdf. K. Schwede, Gluing schemes and a scheme without closed points, Recent progress in arithmetic and algebraic geometry, Contemp. Math., vol. 386, Amer. Math. Soc., 2005, pp. 157 -- 172. Graeme Segal, Categories and cohomology theories, Topology 13 (1974), 293 -- 312. MR 0353298 Wolfgang Steimle, On the universal property of waldhausen's k-theory, 2017. N. Tsuzuki, Cohomological descent of rigid cohomology for proper coverings, Invent. Math. 151 (2003), 101 -- 133. R. Thomason and T. Trobaugh, Higher algebraic K-theory of schemes and of derived categories, The Grothendieck Festschrift, vol. III, Birkhauser Boston, 1990, pp. 247 -- 435. F. Waldhausen, Algebraic K-theory of spaces, Algebraic Geometry and Topology (New Brunswick, N.J, 1983), Lecture Notes in Mathematics, vol. 1126, Springer, 1985. C. A. Weibel, Algebraic K-theory of rings of integers in local and global fields, Handbook of K-Theory, vol. 1, Springer-Verlag, 2005, pp. 139 -- 190. Charles A. Weibel, The K-book, Graduate Studies in Mathematics, vol. 145, American Mathematical Society, Providence, RI, 2013, An introduction to algebraic K-theory. MR 3076731 Inna Zakharevich, The annihilator of the Lefschetz motive, Duke Math. J. 166 (2017), no. 11, 1989 -- 2022. MR 3694563 [Zak17b] , The K-theory of assemblers, Adv. Math. 304 (2017), 1176 -- 1218. MR 3558230 Department of Mathematics, Vanderbilt University E-mail address: [email protected] Department of Mathematics, University of California, Irvine E-mail address: [email protected] Department of Mathematics, Cornell University E-mail address: [email protected]
1707.00653	1	1707	2017-07-03T17:05:01	Hodge numbers and deformations of Fano 3-folds	[ "math.AG" ]	We calculate the Hodge numbers of quasismooth Fano 3-folds whose total anti-canonical embedding has small codimension, and relate these to the number of deformations.	math.AG	math	HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS GAVIN BROWN AND ENRICO FATIGHENTI Abstract. We calculate the Hodge numbers of quasismooth Fano 3-folds whose total anti- canonical embedding has small codimension, and relate these to the number of deformations. 1. Introduction In this paper we calculate the Hodge numbers and the number of moduli of all known (index 1) Fano 3-folds in codimensions 1, 2 and 3. These results are presented in Tables 1–3 respectively; the Picard rank is 1 in every case. We also calculate a few cases in codimensions 4 in §4.3, where the Picard rank is sometimes larger. A Fano 3-fold is a normal 3-dimensional complex projective variety X with ample anticanon- ical class −KX and Q-factorial terminal singularities, and we restrict consideration to those X whose singularities are terminal cyclic quotient singularities. Any such X is a projective orbifold, the quotient of a projective manifold by a finite cyclic group. It is known by Sano [San16] that any Fano 3-fold has a small deformation (a Q-smoothing) that has only quotient singularities. A K3 elephant of a Fano 3-fold X is an irreducible surface E ⊂ X with canonical singularities that is linearly equivalent to −KX. In particular, E has KE = 0, and so E is a K3 surface. There are two main ingredients. The first is an unprojection calculus (see §2.3 or [BKR12a]). The second is a relation betwen the Hodge numbers of a Fano 3-fold and the number of its moduli, together with an infinitesimal rigidity result, which we summarise as follows. Theorem 1. Let X be a Fano 3-fold with K3 elephant E ⊂ X and genus gX = h0(X, −KX ) − 2. (i) Setting αE = h1,1(E) − gX + 1, (1) h1(X, TX ) − h0(X, TX ) = αE + h2,1(X) − h2,2(X). (ii) If X is a complete intersection in weighted projective space or in a weighted Grassman- nian w Gr(2, 5), then h0(X, TX ) = 0. Part (i) is proved in §3.1 and part (ii) in §3.2. We work over C throughout. 2. Preliminaries 2.1. Fano 3-folds in their anticanonical embeddings. We study a Fano 3-fold X using its anticanonical graded ring R(X, −KX ) = Mm≥0 H 0(X, OX (−mKX)). A minimal set of generators x0, . . . , xn for R(X, −KX ), whose degrees are denoted a0, . . . , an, present X as a subvariety X ⊂ P(a0, . . . , an) defined by the relations holding in the ring. By definition, the codimension of a Fano 3-fold X is its codimension in this embedding: codim(X) = n − 3. Such embedded X is an orbifold if its equations satisfy the Jacobian condition. According to [KMMT00] (following [Kaw92] in the case of Mori–Fano 3-folds), the classifica- tion of Fano 3-folds consists of finitely many deformation familes. The Hilbert series of members of those families whose generic element lies in codimension at most 4 are known [Alt98, ABR02] and available on the Graded Ring Database [BK16]. They fall into 95+85+70+145 = 395 cases, according to codimension. There may be more than one irreducible family for any given Hilbert 1 2 GAVIN BROWN AND ENRICO FATIGHENTI series, and in codimension 4 there are usually two or more families in each case [BKR12a]; the different families are distinguished by the Euler characteristic of their general member. 2.2. The Hodge numbers of Fano 3-folds. On a quasi-smooth variety X is it possible to define the notion of a pure Hodge structure, see Steenbrink [Ste77, Theorem 1.12]. Consider in X0. Then we can define H p,q(X) as in the X ). The Hodge decomposition takes then the X := j∗Ωp fact the smooth locus j : X0 ֒→ X and bΩp smooth case and moreover H p,q(X) ∼= H q(X,bΩp H k(X, C) = Mp+q=k form X ). H q(X,bΩp Since there will be no danger of confusion, to avoide cumbersome notations when dealing with quasi-smooth varieties we will abuse the notation and write directly Ωp It follows at once from the Lefschetz hyperplane theorem and Kawamata–Viehweg vanishing that the Hodge diamond of a Fano 3-fold X has the form X instead of bΩp X. h3,0 h3,1 h2,0 h3,2 h2,1 h1,0 h3,3 h2,2 h1,1 h0,0 h2,3 h1,2 h0,1 h1,3 h0,2 h0,3 = 0 0 0 0 h2,1 0 1 h2,2 h1,1 1 0 h1,2 0 0 0 0 . The Euler characteristic e(X) of X can be expressed as e(X) = 2 + 2h1,1(X) − 2h2,1(X). We calculate these three integers for Fano 3-folds X lying in the known families of Fano 3-folds with small codimension. We explain the different strategies we employ in §2.4 below. The answer is well known in codimension 1: the Hodge numbers of weighted hypersurfaces are computed by results of Griffiths, Dolgachev and Dimca. (Recall that primitive cohomology is the kernel of the hyperplane operator: if X has dimension n and hyperplane class L, then H k(X, C)prim = kern∩Ln−k+1 : H k(X, C) → H 2n+2−k(X, C)o , prim(X) = H p,q(X) ∩ H p+q(X, C)prim. When X is a Fano 3-fold, then b5(X) = 0 and so and H p,q H 2,1 prim(X) = H 2,1(X).) Theorem ([Dol82, Dim85, IF00]). Let Xd : (f = 0) ⊂ P(a0, . . . , an) be a quasismooth hypersur- face, defined by a homogeneous polynomial f of degree d in weighted homogeneous coordinates x0, . . . , xn of degrees deg xi = ai. Then the Milnor algebra M of X is C[x0, . . . , xn]/Jf is finite dimensional, and there is an isomorphism H n−p,p−1 prim (X) ∼= Mpd−P ai. The Hilbert Series PM of the Milnor algebra M is given, in the notation of the theorem, by PM = (1 − tb0) · · · (1 − tbn) (1 − ta0) · · · (1 − tan) , where bi = d − ai. For example, X66 ⊂ P(1, 5, 6, 22, 33) has PM = Qb∈{65,61,60,44,33}(1 − tb) Qa∈{1,5,6,22,33}(1 − ta) = 1 + t + t2 + t3 + t4 + 2t5 + · · · + 118t64 + 120t65 + 122t66 + · · · + t196. Thus we read off h2,1(X) = dim M2·66−67 = dim M65 = 120. We list all 95 cases in Table 1. HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 3 In codimensions 2, 3 and 4, the Euler characteristic is known in most cases by [BKR12a], so knowing h1,1(X) completes the calculation. Blache [Bla96] describes a general theory of orbifold characteristic classes, and their relations with the usual topological notions, that we describe and compare in Appendix A.2. We calculate codimension 2 using our methods described below, and Theorem 1 is crucial in the higher codimension, non-complete intersection cases-and the cases with higher Picard rank in §4.3 use these in an essential way. Thus the first observation is that this is readily computed in low codimension, since every Fano 3-fold in codimension up to 3 appears in one of the two situations of the theorem. Theorem 2. If X is a quasismooth Fano 3-fold that is either (i) a complete intersection in weighted projective space, or (ii) a complete intersection in a weighted cone over a weighted Gr(2, 5), then h1,1(X) = 1. Proof. We prove that T 2 H 2(X, KX ) = 0 allows us to apply [DNFF15, Theorem 2.8], which says H 1,1 0, and so h1,1(X) = 1. AX (−1) = 0, where AX is the affine cone on X. This is enough since AX (−1) = prim(X) = T 2 In part (i), the vanishing is [Sch73, 1.3]. For part (ii), T 2 AX (−1) ∼= H 1(X, NX/ C P(−1)), where C P denotes the ambient projective space for the Grassmannian in its Plucker embedding with the addition of the cone variables. From [Ser07, §D.1, Lemma D.3] the flag of schemes X ⊂ C Gr ⊂ C P determines a sequence of sheaves on X: 0 → NX/ C Gr → NX/ C P → NC Gr / C P → 0, where the last map is exact since H 1(NX/ C Gr) = 0. Twisting by OX (−1) we get H 1(NX/ C P(−1)) ∼= H 1(NC Gr / C P(−1)) = 0. This proves part (ii). (cid:3) Part (i) of this result appeared in a recent preprint, [PST17], and we found (ii) stated several times in the literature, such as [KOW16], but we could not find a proof to cite. In this situation, one would like appeal to folklore and simply apply a weighted Lefschetz hyperplane theorem for ample systems. But unfortunately the linear systems we cut by to make X are rarely base-point free when there are nontrivial weights, so the strong results in the literature such as [RS06, Theorem 1] and [HL10, Corollary 2.8] do not apply directly. 2.3. Fano 3-folds and projection. Consider the following arrangement of projective 3-folds: (2) eY → X ↓ Y Y where X and Y are quasismooth, Y Y is a degeneration to a singular orbifold whose only number of vanishing cycles to nodes and then resolves the nodes by exceptional P1s, is known as a conifold transition. non-quasismooth points are ordinary nodes, Y ← eY is a projective small resolution of the nodes, and eY → X is the contraction of a divisor eD ⊂ eY . The passage from Y to eY , that shrinks a In our context, the exceptional divisor eD ∼= P(a, b, c) maps to a divisor P(a, b, c) → D ⊂ Y , and the nodes of Y lie on D. The small resolution is the relatively eD-ample resolution, so is projective, and eD → D is birational-often an isomorphism, in fact. With this setup, we recall from Clemens [Cle83] (see also [Rei87a, §5]): 4 GAVIN BROWN AND ENRICO FATIGHENTI Theorem 3 ([Cle83, Rei87a]). Let X and Y be Fano 3-folds related as in diagram (2). Then (3) e(X) = e(Y ) + 2n − 2, where n is the number of nodes of Y . In particular, if h1,1(X) = h1,1(Y ), then (4) h2,1(X) = h2,1(Y ) − n + 1. The relevance of this is as follows (see [CPR00, 2.6.3], [BKR12a, 3.2]). If X is a Fano 3-fold in codimension k, then it often happens that the Gorenstein projection from a quotient singularity sits in diagram (2) as X 99K Y , and that Y lies in codimension < k. If this nodal Fano Y deforms to a quasismooth Y whose Hodge numbers are known, then we may recover the invariants of X. 2.4. An overview of the calculations. We adopt different tactics to compute the Hodge numbers of a Fano 3-fold X according to its graded ring. 2.4.1. When X is a hypersurface, this calculation is well known (see §2.2). 2.4.2. If X is a complete intersection in weighted projective space or inside a weighted Grass- mannian, then h1,1(X) = 1 (Theorem 2). If X arises by (possibly multiple) unprojection from a hypersurface, then we can compute e(X) and hence the whole Hodge diamond. This applies to most X that lie in codimension 2 or 3; see §§4.1–4.2. Up to codimension 3, this calculation can be done by hand-the key point is to confirm the existence of a nodal degeneration. 2.4.3. Denoting the affine cone over X by AX , [DNFF15, Theorem 2.5] gives H 2,1(X) ∼= T 1 AX (−1). If X is given by explicit equations, we may use standard algorithms and implementations in computer algebra to calculate h2,1(X); see §2.5 and §A.1. In these cases we compute h2,1(X) for a single quasismooth member of each family, expressed in the format we expect. Since hp,q are deformation invariants for orbifolds (since Steenbrink [Ste77, Theorem 2] applies in the context of V-manifolds), the numbers we obtain are also the Hodge numbers of any orbifold Fano 3-fold in the family. 2.4.4. By [DNFF15, Theorem 2.8], H 1,1 prim(X)(X) ∼= T 2 AX (−1), and so if X is given by explicit equations we may compute h1,1(X); see Section 4.3 for an example. This algorithm seems to be more complicated, and in practice choosing good equations is delicate. 2.5. Calculating T 1 and h2,1(X) by computer algebra. We recall the context and results of [DNFF15]. A subcanonical pair (X, OX (1)) consists of a quasismooth projective variety X and an ample sheaf OX(1) which satisfies ωX ∼= OX (kX ) for some kX ∈ Z. Let (X, OX (1)) be a subcanonical pair. We denote by AX the affine cone over X and by UX = AX \ {v}, where v is the vertex of the cone. The results of [DNFF15] require that depthvAX ≥ 3, which holds in our context since H 1(X, OX (j)) = 0 for any j ∈ Z. Consider the space T 1 AX that parametrizes the set of isomorphism classes of first order in- finitesimal deformations of AX. This is defined (as in [Sch73], since X is projectively normal) by T 1 AX := Ext1(Ω1 AX , OAX ), and admits a natural Z-grading given by the natural C∗-action on AX . By [Sch73], the degree 0 component of the deformations of the affine cone parametrizes the embedded deformations of X; that is, the deformations of the pair (X, OX (1)). Furthermore, HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 5 the negative components are identified with the smoothings of the affine cone, while the pos- itive components parametrize equisingular deformations. In the case of a smooth projective hypersurface of degree d, the Jacobian ring of X, as in §2.2. AX (−d) ∼= C[x0, . . . , xn]/Jf , T 1 Theorem 4 ([DNFF15] Theorem 1.1). Let (X, OX (1)) be a subcanonical pair with ωX ∼= OX(kX ). Set n = dim X. Then there is an isomorphism T 1 AX (k) ∼= ker(cid:0)λ : H 1(X, Ωn−1(k − kX )) −→ H 2(X, ωX (k − kX)(cid:1) , where λ(η) = c1(OX (1)) ∧ η. When k = kX, the statement becomes T 1 AX (kX ) ∼= H n−1,1 prim (X), the primitive cohomology. 3. Moduli of Fano 3-folds We explain a relation between H 2,1(X) of a Fano threefold X and the tangent space to its versal deformation space H 1(X, TX ). Since deformations of quasismooth Fano 3-folds X are unobstructed (by [San16, Theorem 1.7]), this is the number of moduli of X. 3.1. Deforming a Fano with an elephant. The idea comes from Calabi–Yau 3-folds. Given such a V , it follows by standard Serre duality (non-canonically, involving a choice of determinant) that H 2,1(V ) ∼= H 1(V, TV ); or one may observe that both are isomorphic to the same graded piece T 1 AV (0) ⊂ T 1 AV . If a Fano 3-fold X has a K3 elephant E = (x = 0) ⊂ X, we may regard the pair (X, E) as a log Calabi–Yau and hope to mimic this relationship. In the index 1 case, one has H 2,1(X) ∼= AX (−1) and H 1(X, TX ) ∼= T 1 T 1 AX (0), and the analogue to the Calabi–Yau isomorphism is the multiplication map x : H 2,1(X) → H 1(X, TX ). This map is not an isomorphism, in general, but Theorem 6 below explains the difference in terms of the geometry of E. To make this intuition precise, we start with a more general lemma about Fano 3-folds of arbitrary index m > 0. Lemma 5. Let X a Fano threefold. If E ⊂ X a K3 elephant E ∈ −KX , then h1(X, TX ) − h0(X, TX ) = αE + h2,1(X) − h2,2(X), where αE = h1,1(E) − gX + 1. Proof. Suppose X is of index m with −KX lin ∼ mH, for an ample Q-Cartier divisor H. Consider the standard exact sequence of OX -modules twisted by Ω2(m), In cohomology this yields a long exact sequence 0 → Ω2 X → Ω2 X(m) → Ω2 X(m)E → 0. (5) 0 → H 0(Ω2 X (m)) → H 0(Ω2(m)X E) → H 1(Ω2 X) → H 1(Ω2 X(m)) → H 1(Ω2 X(m)E) → H 2(Ω2 X) → 0, where H 0(Ω2 X) = 0 and H 2(Ω2 X(m)) = 0 by Akizuki–Kodaira–Nakano vanishing. On the other hand the relative exact tangent sequence 0 → TE → TXE → OE(m) → 0 yields a long exact sequence (6) 0 → H 0(E, TX E) → H 0(E, OE (m)) → H 1(E, TE) → H 1(E, TX E) → 0, 6 GAVIN BROWN AND ENRICO FATIGHENTI where H 1(E, OE (m)) = 0 and H 0(E, TE ) = H 0(E, Ω1 (6) we get E) = 0, since E is K3 surface. By (5) and (7) and h0(X, Ω2 X E(m)) + h1(X, Ω2 X (m)) + h2,2(X) = h2,1(X) + h1(X, Ω2 X (m)E) + h0((X, Ω2 X (m)) h1(TX E) − h0(TX E) = h1(TE) − h0(OE (m)). We have Ω2 X(m) ∼= TX from the pairing X ⊗ Ω2 Ω1 X → ωX ∼= OX (−m). So with αE defined as in the statement, we get h1(X, TX ) − h0(X, TX ) = αE + h2,1(X) − h2,2(X) as required. (cid:3) Theorem 6. Let X be a Fano 3-fold with K3 elephant E ⊂ X and αE as defined in Lemma 5. If h0(X, TX ) = 0, then h1(X, TX ) − h2,1(X) = αE − h2,2(X). This gives an estimate of the difference between the moduli and Hodge theory of X: when b2 = h2,2(X) is small, we have a more moduli than h2,1, while if b2 >> 0 the opposite holds. Remark 1. The number αE is a function of the polarised K3 surface E (since h0(E, OE (E)) = gX − 1). When E is smooth h1,1(E) = 20, and so αE = 20 − h0(E, OE (1)). More generally, if E has canonical singularities with corresponding basket B =(cid:8) 1 (9.1)(III)]), then r (a, −a)(cid:9) (see [Rei87b, Theorem αE = 20 −XB (r − 1) − h0(E, OE (m)). In every case we know, when a general member X of a family of Fano 3-folds has a K3 elephant E ⊂ X, then both X and E are quasismooth; in particular, they both have only quotient singularities, and the basket of E is equal to the set of singularities of E. 3.2. Automorphisms of Fano 3-folds in Grassmannians. Lemma 7. Let X be a Fano 3-fold of index 1. If X is a weighted complete intersection (in its total anticanonical embedding), then H 0(X, TX ) = 0. Proof. Recall from Flenner [Fle81, Satz 8.11] that if X is an n-dimensional weighted complete intersection, then H p(X, Ωq X (t)) = 0 whenever p + q < dim X and t < q − p. The lemma follows by setting q = 2, p = 0, t = 1 together with Serre duality TX ∼= Ω2 X(1). (cid:3) We prove an analogous result for complete intersection in weighted Grassmannians. Our main interest is in Fano 3-folds of index 1 in codimension 3, X ⊂ P(a0, . . . , a6), most of which arise in this way. We show in Theorem 10 below that H 0(X, TX ) = 0 in this case. We first show the vanishing result in standard (non-weighted) Grassmannians. Lemma 8. Let X a Fano 3-fold of index 1 that is a complete intersection in a cone V = C Gr(2, n), on vertex a linear projective space that is disjoint from X, over a Grassmannian Gr(2, n) for some n ≥ 5. Then H 0(X, TX ) = 0. Proof. We show that H 0(X, Ω2 of index 1. X (1)) = 0, which suffices since TX ∼= Ω2 X(1) for X a Fano 3-fold HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 7 We consider the case V = Gr(2, n) first, with no cone structure. Suppose that X = (f1 = · · · = fc = 0) ⊂ G = Gr(2, n), and denote di = deg fi. The Koszul complex of OX -modules for OX twisted by Ω2(1)G is 0 → Ω2 G(1 − d1 − · · · − dc) → · · · →Mi,j,k Mi,j G(1 − di − dj) →Mi Ω2 Ω2 Ω2 G(1 − di − di − dk) → G(1 − di) → Ω2 G(1) → Ω2 G(1)X → 0. By [PW95, Lemma 0.1], H p(G, Ω2 H 0(G, Ω2 that G(t)) = 0 for each of p = 1, 2, 3 and any t ≤ 1, and also G(1)) = 0. It follows, by splitting the Koszul sequence above into short exact sequences, (8) H 0(X, Ω2 G(1)X ) = H 1(X, Ω2 G(1)X ) = H 1(X, Ω2 G(1 − di)X ) = 0. The conormal exact sequence of X ⊂ G is 0 → M1≤i≤c OX (−di) → Ω1 GX → Ω1 X → 0. Taking its second exterior power and twisting by OX (1) we get G(1 − di)X → Ω2 OX (1 − didj) → M1≤i≤c 0 → M1≤i,j≤c Ω2 G(1)X → Ω2 X(1) → 0. After splitting this into short exact sequences, the vanishing statements in (8) show at once that H 0(X, Ω2 X (1)) = 0, as required. The proof for a cone is the same, replacing Ω2 Gr to the complement of the vertex, in which X is a complete intersection; this restricts to X as above, and the proof follows. (cid:3) Gr by the extension of the pullback of Ω2 The proof of Lemma 8 suggests that we need a Bott vanishing type of result to extend the vanishing statements to complete intersections in w Gr(2, 5). The following lemma gives the precise statement we need. Lemma 9. Let wG = wGr(2, 5). Then H p(w Gr, Ω2 w Gr(−k)) = 0 for p = 1, 2, 3 and any k > 0. Proof. If A• have the following diagram G denotes the punctured affine cone over the (weighted or not) Grassmannian, we A• G π1 ւ ց π2 Gr(2, 5) wGr(2, 5) where π1 and π2 denote the quotients by the standard and the weighted C∗ actions respectively. We use the vanishing results from [PW95, Lemma 0.1] for the standard Gr(2, 5) repeatedly. The grading on the cohomology groups of A• is interpreted in terms of local cohomology at the maximal ideal m of the vertex of the affine cone A. Consider the short exact sequence (9) 0 → π∗ 1Ω1 G → Ω1 A• → OA• → 0. Since H i(G, OG(−k)) = 0 for any i < dim(G), we have H 1(A•, Ω1 A•)(−k) = H 1(G, Ω1 G(−k)) = 0. In the same way one also gets H 0(A•, Ω1 A•)(−k) = 0. Raising the short exact sequence (9) to the second exterior power we have 0 → π∗ 1Ω2 G → Ω2 A• → π∗ 1Ω1 G → 0; 8 GAVIN BROWN AND ENRICO FATIGHENTI by the vanishing statements above this reduces to H 1(A•, Ω2 A•)(−k) = H 1(G, Ω2 G(−k)) = 0. Comsidering analogous exact sequences for the second projection π2 gives 0 → π∗ 2Ω1 wG → Ω1 A• → OA• → 0, Putting all these vanishing statements together with H 0(OwG(−k)) = 0 we get 0 → π∗ 2Ω2 wG → Ω2 A• → π∗ 2Ω1 wG → 0. H 1(wG, Ω2 wG(−k)) = H 1(A•, Ω2 A•)(−k) = 0, as required. The results for i = 2, 3 follow similarly. (cid:3) Theorem 10. Let X a Fano 3-fold of index 1 that is a complete intersection in a weighted cone C Gr(2, 5), with vertex a linearly-embedded weighted projective space that is disjoint from X. Then H 0(X, TX ) = 0. Both the lemma and the theorem can be extended to weighted Grassmannians w Gr(2, n), for n ≥ 5, using Bott-type vanishing theorems, but we only need the Gr(2, 5) case here. 4. Explicit calculations It takes a few hundred calculations to complete Tables 1–3 below. In this section, we give illustrative examples of each type. 4.1. Codimension 2. There are 85 deformation families of Fano 3-folds in codimension 2 ([IF00, CCC11]), each one a complete intersection with h1,1(X) = 1. The case X2,3 ⊂ P5 is classical: e(X) = c3(TX ) can be calculated directly to give e(X2,3) = −36 and so h2,1(X2,3) = 20. Of the remaining 84 cases, 66 have a Type I projection (see §4.1.1), and a further 10 cases have a Type II1 projection (see §4.1.2); 8 cases have no projection of either type (see §4.1.3). 4.1.1. 66 cases with Type I projection. Consider one of the families of Fano 3-folds of the form X = Xa3+r,a4+r ⊂ P(1, a, r − a, a3, a4, r) with a < r. The general member has a quotient singularity 1 r (1, a, r − a), and admits a Type I projection, as in diagram (2), to a hypersurface: X ⊂ P(1, a, r − a, a3, a4, r) πr ↓ D ⊂ (x3A = x4B) = Y ⊂ P(1, a, r − a, a3, a4), where D = (x3 = x4 = 0) = P(1, a, r − a) and πr is the projection from the final coordinate point of index r. In each one of these 66 cases, the general Y is quasismooth away from n = deg(A) deg(B)/(a(r−a)) nodes that lie on D (by Bertini's theorem), and it admits a Q-smoothing to a general Y = Ya3+a4+r ⊂ P(1, a, r − a, a3, a4). Thus we calculate e(X) = e(Y )+ 2n − 2 by (3). Example 11. Working from the bottom up in diagram (2), let Y4 ⊂ P4 be a smooth quartic. We know e(Y4) = −56 and h2,1(Y4) = 30. Imposing a linear plane D = P2 on Y4 gives, in coordinates x, y, z, t, u of P4, P2 = D = (x = y = 0) ⊂ Y 4 = (Ax = By) ⊂ P 4, where A, B are general cubic forms. Such Y has 9 nodes at (A = B = 0) ⊂ D. The unprojection of D ⊂ Y is a quasismooth variety X3,3 ⊂ P(15, 2), which has Fano Hilbert series No. 20522. By (3) we have e(X3,3) = e(Y4) + 18 − 2 = −40, and so h2,1(X3,3) = 30. This calculation is recorded in Table 2, together with the numerical data described here. HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 9 4.1.2. 10 cases with Type II1 projection. Again we work from bottom up in diagram (2). Thus, for example, to study X whose Hilbert series PX is no. 6858 in the Grdb [BK], we observe from that database (or by hand from the methods of [ABR02]) that the numerics suggest a Type II1 projection to Y with Hilbert series PY no. 5837, whose general member we know to be of the form Y10 ⊂ P(1, 1, 2, 2, 2, 3). The task in this case is to impose a divisor D onto a special (nodal) member of this family, where the divisor D may be singular, but its normalisation is eD ∼= P2. Example 12. Consider X = X4,6 ⊂ P(1, 1, 2, 2, 2, 3), which has Fano Hilbert series no. 6858 in [BK]. As in Example 11 we work bottom up, first constructing D ⊂ Y 10 ⊂ P(1, 1, 2, 2, 5) and then unprojecting. We follow Reid [Rei00, §9] and Papadakis [Pap08a] for Type II1 unprojections. In coordinates x, y, z, t, u on P(1, 1, 2, 2, 5), the finite morphism P2 ∼= eD −→ D ⊂ P(1, 1, 2, 2, 5) 7→ (a, b, c2, (a − b)c, abc3 + c5) (a, b, c) has image D defined by the 2 × 2 minors of M =(cid:18) t u (x − y)z (xy + z)z2 x − y (xy + z)z t u (cid:19) . two points (1 : 1 : ±i). A general Y 10 containing this D has 34 nodes, all of which lie on D. (Two lie at the singu- The surface D has two singular points, each of which has a length 2 preimage in eD: the point (1 : 1 : 0 : 0 : 0) is the pinched image of (1 : 1 : 0) ∈ eD, and (1 : 1 : −1 : 0 : 0) is the image of larities of D, so the preimage in eD of the singular subscheme of Y has length 36 on eD.)     with entries of degrees   The unprojection of D ⊂ Y is given by the maximal Pfaffians of the skew 5 × 5 matrix 1 4 2 5 2 0 3 3 6 4 x − y (xy + z)z   s0 t 1 s1 s1 + A3 u B6 zs0 + C4 in P(1, 1, 2, 2, 5, 2, 3) with coordinates x, y, z, t, u, s0, s1, where A, B, C may be determined by the unprojection calculus if we wish to know them explicitly. Eliminating u using the linear equation gives X4,6 ⊂ P(1, 1, 2, 2, 2, 3), as required. We know e(Y ) = −124, so conclude that e(X) = −124 + 2 · 34 − 2 = −58 and h2,1(X) = 31. 4.1.3. 8 cases with no projection. Our projection techniques do not work in these cases. We use computer algebra instead. Example 13. Consider a quasismooth Fano 3-fold X6,6 : (f = g = 0) ⊂ P(1, 23, 32) with Fano Hilbert series number 3508, defined by f = x6 + y3 + z3 + t3 + u2 + v2 and g = y2z + z2t + t2y + uv. Iten's Macaulay2 package [Ilt12] works as follows (compressing blank lines in the output): Macaulay2, version 1.5 with packages: ConwayPolynomials, Elimination, IntegralClosure, LLLBases, PrimaryDecomposition, ReesAlgebra, TangentCone i1 : loadPackage "VersalDeformations" o1 = VersalDeformations o1 : Package i2 : R = QQ[x,y,z,t,u,v,Degrees=>{1,2,2,2,3,3}]; i3 : I = ideal ( x^6 + y^3 + z^3 + t^3 + u^2 + v^2, y^2z + z^2t + t^2y + uv ); 10 GAVIN BROWN AND ENRICO FATIGHENTI o3 : Ideal of R i4 : CT^1(-1,I) 2 24 o4 : Matrix R <--- R The answer is that h2,1(X) = dim T 1 AX (−1) = 24. Since X has a K3 elephant E = (x = 0) ⊂ X with basket 9× 1 2 (1, 1) quotient singularities, and h0(X, TX ) = 0 by Theorem 1(ii), we know at this stage from the moduli formula Theorem 1(i) that h1(X, TX ) = 34. This can also be calculated directly by Macaulay2 as follows: i5 : CT^1(0,I) 2 34 o5 : Matrix R <--- R Again, the answer is that h1(X, TX ) = dim T 1 AX (0) = 34. A similar calculation works with X12,14 : (f = g = 0) ⊂ P(2, 3, 4, 5, 6, 7), with Hilbert series number 37, with, for example, f = x6 + y4 + z3 − u2 + tv and g = x7 + z2u + xu2 + zt2 + v2. In this case there is no elephant E ⊂ X, so the moduli formula (1) does not apply as stated. However, the Macaulay2 results are that h2,1(X) = 18 and h1(X, TX ) = 23, and so in fact the formula holds with "αE = 6", which is the correct number calculated on X from its basket indices and h0(X, O(1)) = 0. 4.2. Codimension 3. There are 70 known deformation families of Fano 3-folds in codimen- sion 3. The complete intersection X = X2,2,2 ⊂ P5 is classical: the chern class calculation and Lefschetz gives e(X) = −24, ρX = 1 and h2,1(X) = 14. The remaining 69 cases are all complete intersections in weighted Grassmannians w Gr(2, 5), and so h1,1(X) = 1 in every case. 4.2.1. 64 cases Type I. We say that a Fano 3-fold X has a Type I staircase if it admits a sequence of alternate Type I projections and Q-smoothings to a hypersurface. Concretely, if X ⊂ wP6 lies in codimension 3, then the staircase is (10) eY → X ↓ eY → Y Y ↓ Z Z where X 99K Y ⊂ wP5 eliminates a single variable, Y ⊂ wP5 is a general Q-smoothing of Y , and Y 99K Z is a projection to a nodal hypersurface Z ⊂ wP4 as in §4.1. Counting nodes on Y and Z and using the formula of Theorem (3) completes the calculation of e(X) and h2,1(X). Of the 64 Fano 3-folds in codimension 3 with a Type I projection, 57 have a Type I staircase to a hypersurface. Example 14. Consider the family with Hilbert series no. 20523 in [BK]. A typical member X ⊂ P(1, 1, 1, 1, 1, 2, 3), in coordinates x1...5, y, z, is given by the five maximal Pfaffians of a skew 5 × 5 matrix of forms   x1 x2 A D x3 B E C F z   where the entries have degrees   1 1 2 2 1 2 2 2 2 3   . It has a quotient singularity 1 3 (1, 1, 2) at the z-coordinate point Pz ∈ X. HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 11 Projection from that point is calculated by eliminating z from these equations. Doing that leaves the two Pfaffians of degree 3, which define Y 3,3 :   D E F(cid:19) (cid:18)A B C  x3 −x2 x1  = 0   ⊂ P(1, 1, 1, 1, 1, 2). For general degree 2 forms A, . . . , F , the image Y has 6 nodes (by Hilbert–Burch) and a Q- smoothing Y3,3 which was computed in Example 11 above. Making the projection from Y3,3 as in Example 11 completes the staircase. In any case, using the result of Example 11 gives e(X) = e(Y ) + 2 · 6 − 2 = −40 + 12 − 2 = −30, and so h2,1(X) = 17. Of the remaining 7 cases, 4 have a Type I projection to a family that arises by Type II1 unprojection from a hypersurface, so again have a staircase, but with a more complicated second step. A fifth case has a Type I projection to the classical family Y2,3 ⊂ P5, so also works. But in two remaining cases, the image of the Type I projection lies in a family whose Hodge numbers were computed using the algorithms for dim T 1; in this paper, these cases remain dependent on computational algebra. 4.2.2. 2 cases Type II1. Of the cases without a Type I projection, two have a Type II1 pro- jection: X7,8,8,9,10 ⊂ P(1, 2, 3, 3, 4, 4, 5) has a Type II1 projection from 1 4 (1, 1, 3) and X10...14 ⊂ P(1, 3, 4, 5, 5, 6, 7) has a Type II1 projection from 1 5 (1, 2, 3). We consider the latter in detail, following Reid [Rei00, 9.5] and Papadakis [Pap08b, 4.4]. Consider D ⊂ P(1, 3, 4, 5, 6) defined by the maximal minors of MD =(cid:18)t v yz t y z z2 v(cid:19) . This D is the image of P(1, 2, 3) → P(1, 3, 4, 5, 6) given by (a, b, c) 7→ (a, c, b2, bc, b3); the normal- ising variable b is recovered as the ratio of the rows of MD. The general hypersurface Y 18 containing D has the form Y 18 = (A12m12 + B11m13 + 2B12m23 + B22m24 = 0) ⊂ P(1, 3, 4, 5, 6), where mij denotes the minor of MD involving columns i and j. The unprojection of D ⊂ Y 18 is a codimension 3 variety X ⊂ P(1, 3, 4, 5, 5, 6, 7), in coordinates x, y, z, t, u, v, w, defined by the maximal Pfaffians of the skew 5 × 5 matrix z −u t −B22 −w + B12 y   v w + B12 −B11 −uz − A12   . For example, setting A12 = yv + y3 + x9, B11 = yt + x8, B12 = 0 and B22 = v results in a quasismooth X, and Y 18 whose non-quasismooth locus is defined by the equations zt − yv, z3 − v2, yz2 − tv, y2z − t2, x9t + y3t + 2z2v + ytv, x9y + y4 + y2v + 2v2, x8y2 + y3t + z2v, x9z − 2x8t − yt2 + yzv, 2x8yz − x9v + y3v − yv2 and consists of 22 nodes, all of which lie on D ⊂ Y 18. The general Y18 ⊂ P(1, 3, 4, 5, 6) has e(Y18) = −80, so e(X) = −38 and h2,1(X) = 21. 12 GAVIN BROWN AND ENRICO FATIGHENTI 4.2.3. No Type I or II1 projection. The three remaining cases are X12...16 ⊂ P(1, 4, 5, 5, 6, 7, 8), X16...20 ⊂ P(1, 5, 6, 7, 8, 9, 10) and X14...18 ⊂ P(1, 5, 5, 6, 7, 8, 9). The first has only a type IV projection, while the other two do not have any Gorenstein projections at all. We compute T 1 in these cases: we work out the first in detail here; the other two are similar. Example 15. A particular X12...16 ⊂ P(1, 4, 5, 5, 6, 7, 8), in coordinates x, y, z, t, u, v, w, is given by the maximal Pfaffians of the skew 5 × 5 matrix z u y   v u v y2 + w −y2 + w x9 + yz zt + t2   in the usual antisymmetric notation. One checks that the scheme defined by those equations is quasismooth. We compute h2,1(X) = 20 and h1(X, TX ) = 23 by Macaulay2 as before. We verify the moduli formula (i) of Theorem 1. The basket of X is BX =(cid:26) 1 2 (1, 1, 1), 1 4 (1, 1, 3), 2 × 1 5 (1, 1, 4), 1 5 (1, 2, 3)(cid:27) . The K3 elephant E = (x = 0) ⊂ X is the unique member of −KX. It has h0(OE(1)) = 0 and h1,1(E) = 20 −P ri − 1, where the ri are the indices of singularities of BX. Thus h1(TX ) − h2,1(X) = αE − h2,2(X) = (20 − 1 − 3 − 3 · 4) − 1 = 3, which agrees with 23 − 20. The other two cases work similarly; in each case h2,1(X) = 20. 4.3. Codimension 4. All the calculations in codimensions 4 in this section depend on com- puter algebra: we use Magma [BCP97] to compute examples of the codimension 4 equations by unprojection, and Macaulay2 [GS, Ilt12] for the Hodge numbers. When a Hilbert series is realised by a Fano 3-fold in codimension 4, it frequently happens that there is more than one deformation family of such Fano 3-folds. For 116 of Hilbert series listed in [BK] in codimension 4, [BKR12a] computes the different families, and observes that they are distinguished by the Euler characteristic of a quasismooth member. However it does not compute the Picard rank of these Fano 3-folds, in part because there is no known format in which they lie as complete intersections, and so we have no Lefschetz theorem to apply directly. But the computational methods of this paper still apply, in conjunction with the unprojection construction of [BKR12a, Pap04]. We compute a few examples here as first calculations. Example 16. Fano Hilbert series 24097. By [BKR12a] there are 3 families of Fano 3-folds Y ⊂ P(16, 22) with (typically) two 1 2 (1, 1, 1) quotient singularities, each with the Hilbert series No.24097 in [BK]. They arise by unprojection of P2 = D ⊂ Y ⊂ P(16, 2), where D ⊂ P(16, 2) is a linearly embedded plane, and Y is defined by the vanishing of Pfaffians of a skew 5 × 5 matrix of forms of weights (11)   1 1 1 2 1 1 2 1 2 2   . The three famlies arise as so-called "Tom" and "Jerry" unprojections (see [BKR12a, §2.3] for details), and the three different results are listed in the Big Table [BKR12b]: Tom1, Jer12 and Jer15. Takagi's analysis [Tak02, Theorem 0.3] of prime Fano 3-folds with index 2 terminal HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 13 singularities shows that the first and third of these families have h1,1(X) = 1. Using the Macaulay2 computation, and Theorem 1(i) (which holds since each unprojection does indeed carry a quasismooth elephant E with αE = 19 − 1 − 5 = 13), we complete the table below. unproj type # nodes Tom1 Jer12 Jer15 6 8 9 eX −14 −10 −12 h1,1(X) h2,1(X) h1(X, TX ) h0(X, TX ) 1 3 1 9 9 8 21 19 20 0 0 0 For example, the Jer12 case above uses Y defined by Pfaffian matrix t u v v t + u x   w ux y2 − z2 yz + t2 + u2   in the coordinates x, y, z, t, u, v and w of P(16, 2). Such Y contains the plane D = (t = u = v = w = 0). Unprojecting D ⊂ Y gives X ⊂ P(16, 22), defined by xt − tu − u2 + v2, y2t − z2t − xu2 + vw, yzt + t3 + tu2 − xuv + tw + uw, yzu + t2u + u3 − y2v + z2v + xw, x2u − y2u + z2u − xu2 + yzv + t2v + u2v + vw, x2v − xw + ts, −xyz − xt2 − xu2 − xw − us, −x3 + xy2 − xz2 + x2u + vs, x2y2 − y4 − x2z2 + 3y2z2 − z4 + yzt2 − xy2u + xz2u + yzu2+ +y2uv − z2uv + xtuv + yzw − xuw − tuw + u2w − ws in coordinates x, y, z, t, u, v, w and unprojection variable s. Example 17. Fano Hilbert series 24078. By [BKR12a] there are 3 families of Fano 3-folds X ⊂ P(16, 2, 3) with (typically) two 1 3 (1, 1, 2) quotient singularities, each with the Hilbert series No.24078 in [BK]. They arise by unprojection of P2 = D ⊂ Y ⊂ P(16, 2), where D ⊂ P(16, 2) is a linearly embedded P(1, 1, 2), and Y is defined by the vanishing of Pfaffians of a skew 5 × 5 matrix of forms of the same weights as (11) above. The three different results [BKR12b] are: Tom1, Tom5 and Jer12. In this case the elephant E ⊂ X has αE = 13, and the table below summarises the results. unproj type # nodes Tom1 Tom5 Jer12 5 4 6 eX −16 −18 −14 h1,1(X) h2,1(X) h1(X, TX ) h0(X, TX ) 1 2 1 10 12 9 22 23 21 0 0 0 These calculations seem to be on the limit of what we can do, as they terminate only when the equations are relatively small. For example, the Tom5 case above uses Y defined by Pfaffian matrix   z t u v + u w t z xv + zu w − y2 x2 − v2   in the coordinates x, y, z, t, u, v and w of P(16, 2). Of the 145 Hilbert series of Fano 3-folds listed in [BK] as presented naturally in codimension 4, 116 have the numerical properties consistent with having a Type I unprojection. The unprojec- tion analysis of these is the subject of [BKR12a], with the results presented in [BKR12b], and 14 GAVIN BROWN AND ENRICO FATIGHENTI in principle they could all be computed as above. A further 16 Hilbert series have the numerical properties of a Type II1 projection, and a computational approach following Papadakis [Pap08a] is conceivable; the constructions are part of Taylor's thesis [Tay]. Some of the remaining 13 cases have more complicated projections that we do not know how to work with systematically yet, but four cases have no Gorenstein projections at all, and some other approach is required (even to write down examples by equations). These cases are: No. 25 X ⊂ P(2, 5, 6, 7, 8, 9, 10, 11) No. 166 X ⊂ P(2, 2, 3, 3, 4, 4, 5, 5) No. 282 X ⊂ P(1, 6, 6, 7, 8, 9, 10, 11) No. 308 X ⊂ P(1, 5, 6, 6, 7, 8, 9, 10). 4.4. A quasismooth unprojection from codimension 4. We construct a codimension 4, quasismooth Fano 3-fold X ⊂ P(16, 22) with Hilbert series number 24097 which contains a quasismooth divisor E ⊂ X that is itself a complete intersection. We adapt Example 16 so that the codimension 3 projection Y ⊂ P(16, 2) contains two divisors: the coordinate planes D = P2 and E = P(1, 1, 2) meeting along the coordinate line P1. Indeed define Y by the maximal Pfaffians of t u v v u z − t   w −zv − u2 yz − x2 y2 − t2   in the coordinates x, y, z, t, u, v and w of P(16, 2). Then D = (t = u = v = w = 0) = P2 lies inside Y in Jer12 format while E = (z = t = u = v = 0) = P(1, 1, 2) lies inside Y in Tom5 format. Altogether Y has 8 nodes; these all lie on D (in accordance with Jer12 unprojection of D to construct Hilbert series 24097), and 4 of them lie on the intersection D ∩ E (in accordance with the Tom5 unprojection or E to construct Hilbert series 24078). We may unproject either divisor, and we choose to unproject D ⊂ Y to give X ⊂ P(16, 22). All the 8 nodes are resolved by this, and X is quasismooth. The Fano 3-fold X has Picard rank ρX = 3 (as in Example 16 above). Furthermore, E ⊂ Y has birational image in X, which we also denote E ⊂ X defined by equations E = (z = t = u = v = 0) ∩ X ⊂ P(16, 22), in coordinates x, y, z, t, u, v, w, s. Computing the unprojection shows that E ∼= (x4 − y4 − w2 + ws = 0) ⊂ P(12, 22) in coordinates x, y, w, s, which is P(1, 1, 2) blown up in 4 points on the coordinate line L = P(1, 1) followed by the contraction of the resulting −2-curve eL, the birational transform of L. Thus it is a index 2 Fano surface with two 1 rank 4 and K 2 local coordinates, the cone on E) on an otherwise smooth complete intersection Z2,2,2 ⊂ P6. 2 (1, 1) quotient singularities, Picard E = 4. It can be unprojected to an ordinary, isolated cDV singular point (in new References [ABR02] [Alt98] Selma Altınok, Gavin Brown, and Miles Reid. Fano 3-folds, K3 surfaces and graded rings. In Topology and geometry: commemorating SISTAG, volume 314 of Contemp. Math., pages 25–53. Amer. Math. Soc., Providence, RI, 2002. Selma Altınok. Graded rings corresponding to polarised K3 surfaces and Fano 3 folds. PhD thesis, University of Warwick, 1998. [BCP97] Wieb Bosma, John Cannon, and Catherine Playoust. The Magma algebra system. I. The user lan- guage. J. Symbolic Comput., 24(3-4):235–265, 1997. Computational algebra and number theory (Lon- don, 1993). Janko Boehm and Anne Frubis-Kruger. A smoothness test for higher codimensions. preprint available online as arXiv:1602.04522, 2016. [BFK16] HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 15 [BG17] [BK] [BK16] Gavin Brown and Konstantinos Georgiadis. Polarized Calabi–Yau 3-folds in codimension 4. Mathe- matische Nachrichten, 290(5-6):710–725, 2017. Gavin Brown and Alexander M. Kasprzyk. The graded ring database. Online access via http://www.grdb.co.uk/. Gavin Brown and Alexander Kasprzyk. Four-dimensional projective orbifold hypersurfaces. Exp. Math., 25(2):176–193, 2016. [BKR12a] Gavin Brown, Michael Kerber, and Miles Reid. Fano 3-folds in codimension 4, Tom and Jerry. Part I. Compos. Math., 148(4):1171–1194, 2012. [BKR12b] Gavin Brown, Michael Kerber, and Miles Reid. Tom and Jerry: Big Table, 2012. Linked at [Bla96] [CCC11] [Cle83] [CPR00] http://grdb.co.uk/Downloads/. Raimund Blache. Chern classes and Hirzebruch-Riemann-Roch theorem for coherent sheaves on complex-projective orbifolds with isolated singularities. Math. Z., 222(1):7–57, 1996. Jheng-Jie Chen, Jungkai A. Chen, and Meng Chen. On quasismooth weighted complete intersections. J. Algebraic Geom., 20(2):239–262, 2011. C. Herbert Clemens. Double solids. Adv. in Math., 47(2):107–230, 1983. Alessio Corti, Aleksandr Pukhlikov, and Miles Reid. Fano 3-fold hypersurfaces. In Explicit birational geometry of 3-folds, volume 281 of London Math. Soc. Lecture Note Ser., pages 175–258. Cambridge Univ. Press, Cambridge, 2000. [DGPS16] Wolfram Decker, Gert-Martin Greuel, Gerhard Pfister, and Hans Schonemann. Singular 4-1-0 - A [Dim85] computer algebra system for polynomial computations. http://www.singular.uni-kl.de, 2016. Alexandru Dimca. Monodromy and Betti numbers of weighted complete intersections. Topology, 24(3):369–374, 1985. [DNFF15] C. Di Natale, E. Fatighenti, and D. Fiorenza. Hodge theory and deformations of affine cones of [Dol82] [Fle81] [GS] [HL10] [IF00] [Ilt12] [Kaw92] subcanonical projective varieties. preprint available online as arXiv:1512.00835, 2015. Igor Dolgachev. Weighted projective varieties. In Group actions and vector fields (Vancouver, B.C., 1981), volume 956 of Lecture Notes in Math., pages 34–71. Springer, Berlin, 1982. Hubert Flenner. Divisorenklassengruppen quasihomogener Singularitaten. J. Reine Angew. Math., 328:128–160, 1981. Daniel R. Grayson and Michael E. Stillman. Macaulay2, a software system for research in algebraic geometry. Available at http://www.math.uiuc.edu/Macaulay2/. Daniel Halpern-Leistner. Lefschetz Hyperplane Theorem for stacks. preprint available online as arXiv:1008.0891, 2010. A. R. Iano-Fletcher. Working with weighted complete intersections. In Explicit birational geometry of 3-folds, volume 281 of London Math. Soc. Lecture Note Ser., pages 101–173. Cambridge Univ. Press, Cambridge, 2000. Nathan Owen Ilten. Versal deformations and local Hilbert schemes. J Softw. Algebra Geom., (4):12– 16, 2012. Yujiro Kawamata. Boundedness of Q-Fano threefolds. In Proceedings of the International Conference on Algebra, Part 3 (Novosibirsk, 1989), volume 131 of Contemp. Math., pages 439–445. Amer. Math. Soc., Providence, RI, 1992. [KMMT00] J´anos Koll´ar, Yoichi Miyaoka, Shigefumi Mori, and Hiromichi Takagi. Boundedness of canonical Q- [KOW16] [Pap04] [Pap08a] [Pap08b] [PST17] [PW95] [Rei87a] Fano 3-folds. Proc. Japan Acad. Ser. A Math. Sci., 76(5):73–77, 2000. In-Kyun Kim, Takuzo Okada, and Joonyeong Won. Alpha invariants of birationally rigid fano three- folds. arXiv preprint arXiv:1604.00252, 2016. Stavros Argyrios Papadakis. Kustin-Miller unprojection with complexes. J. Algebraic Geom., 13(2):249–268, 2004. Stavros Argyrios Papadakis. The equations of type II1 unprojection. J. Pure Appl. Algebra, 212(10):2194–2208, 2008. Stavros Argyrios Papadakis. The equations of type II1 unprojection. J. Pure Appl. Algebra, 212(10):2194–2208, 2008. Marco Pizzato, Taro Sano, and Luca Tasin. Effective non-vanishing for fano weighted complete in- tersections. arXiv preprint arXiv:1703.07344, 2017. Thomas Peternell and Jaros law A. Wi´sniewski. On stability of tangent bundles of Fano manifolds with b2 = 1. J. Algebraic Geom., 4(2):363–384, 1995. Miles Reid. The moduli space of 3-folds with K = 0 may nevertheless be irreducible. Math. Ann., 278(1-4):329–334, 1987. 16 GAVIN BROWN AND ENRICO FATIGHENTI [Rei00] [RS06] [San16] [Sch73] [Rei87b] Miles Reid. Young person's guide to canonical singularities. In Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), volume 46 of Proc. Sympos. Pure Math., pages 345–414. Amer. Math. Soc., Providence, RI, 1987. Miles Reid. Graded rings and birational geometry. Proc. of algebraic geometry symposium (Kinosaki, Oct 2000), K. Ohno (Ed.), pages 1–72, 2000. G. V. Ravindra and V. Srinivas. The Grothendieck-Lefschetz theorem for normal projective varieties. J. Algebraic Geom., 15(3):563–590, 2006. Taro Sano. On deformations of Q-Fano 3-folds. J. Algebraic Geom., 25(1):141–176, 2016. Michael Schlessinger. On rigid singularities. Rice Univ. Studies, 59(1):147–162, 1973. Complex anal- ysis, 1972 (Proc. Conf., Rice Univ., Houston, Tex., 1972), Vol. I: Geometry of singularities. Edoardo Sernesi. Deformations of algebraic schemes, volume 334. Springer Science & Business Media, 2007. J. H. M. Steenbrink. Mixed Hodge structure on the vanishing cohomology. In Real and complex singularities (Proc. Ninth Nordic Summer School/NAVF Sympos. Math., Oslo, 1976), pages 525– 563. Sijthoff and Noordhoff, Alphen aan den Rijn, 1977. Hiromichi Takagi. On classification of Q-Fano 3-folds of Gorenstein index 2. I, II. Nagoya Math. J., 167:117–155, 157–216, 2002. Rosemary Taylor. Fano 3-folds and Type II Unprojection, PhD thesis, University of Warwick. In preparation. Fabio Tonoli. Construction of Calabi-Yau 3-folds in P6. J. Algebraic Geom., 13(2):209–232, 2004. [Ton04] [Ser07] [Ste77] [Tak02] [Tay] Appendix A. Hodge numbers of Fano 3-folds Tables 1–3 in A.3 below list the invariants for all known families of Fano 3-folds in codimension at most 3. The majority of the calculations can be carried out by hand. We use computer algebra where not, and also use it as a check on all results. In codimensions 1 and 2 respectively the Fano 3-folds come from Iano-Fletcher ([IF00] Tables 5 and 6 respectively; in codimensions 3 and 4 they are from Altınok ([Alt98]). The graded ring database identifier (denoted 'Grdb' in the tables) is that of [BK]. A.1. Our use of computer algebra. The explicit calculations we need are standard, although sometimes rather involved. There are three places computer algebra may assist. (i) Checking that a variety is quasismooth can usually be done with Bertini's theorem. In codimension 3 and 4, this can be carried out as in [BG17, §3–4], for example, when Type I projections (and staircases) are available. In other cases, we check the Jacobian condition by machine. This, or some equivalent (such as [Ton04, Theorem 5.5] or [BFK16]), can be checked by computer algebra given explicit equations. (ii) Checking that a variety has only ordinary nodes as singularities, and counting those nodes, can again usually be done by Bertini's theorem together with a chern class calculation when we have Type I projections; see for example [BG17, §4] for the nodes and [BKR12a, §7] for the count. In other cases, we use computer algebra following [BKR12a, §6]. (iii) Computing the dimensions of graded pieces of spaces T 1 AX seems too hard by hand in most cases, but there are algorithms to do this based on Grobner basis. We are indebted to the developers of the computer algebra systems Macaulay2 [GS], Magma [BCP97] and Singular [DGPS16] that we used for these calculations, and to Ilten [Ilt12] for the Versal Deformation package for Macaulay2. (The latter conveniently handles the gradings on variables automatically when computing graded pieces of T 1 AX ; on other systems we had to pick out the graded piece given generators for the whole module "by hand".) In practice, most computations here work when the equations of the Fano 3-fold are fairly sparse, and as the codimension increases it becomes harder to find such sparse representatives. HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 17 A.2. Blache's orbifold formula. Let V be a projective orbifold of dimension n, embedded as a quasismooth subvariety of weighted projective space V ⊂ P = P(a0, . . . , aN ). We suppose, in addition, that V is a manifold away from a finite set of strictly orbifold points Q1, . . . , Qs ∈ V . We define the orbifold total chern class corb(TP) = 1 + c1,orb(TP ) + · · · + cn,orb(TP) of P via 0 → OP → ⊕N i=0OP(ai) → TP → 0. Taking the restriction of this to V , we derive the top chern class corb(V ) of V from the tangent exact sequence exactly as in the smooth case: that is, we make the formal computation 0 → TV → TPV → NV P → 0 where H 2(P, Q) = hQ and corb,j ∈ H 2j(P, Q), and then 1 + c1,orb(TP) + · · · + corb,N (TP) = corb(TP) :=Y(1 + aih), (cid:0)1 + corb,1(TV ) + · · · + corb,n(TV )(cid:1) c(NV P) = corb(TP). Then we define the orbifold euler class eorb(V ) by eorb(V ) :=ZV cn,orb(V ) ∈ Q. This is a formal computation that ignores orbifold behaviour. However, it is related to the topological euler characteristic e(V ) by the following theorem of Blache [Bla96]. Theorem 18 ([Bla96] (2.11–14)). Let V be a projective orbifold with finite orbifold locus as above. Then eorb(X) ∈ Q satisfies e(X) = eorb(X) + XQ∈B r − 1 r , where r = r(Q) is the local index of the orbifold point Q. For a hypersurface Xd ⊂ P(a0, . . . , an+1) we have eorb(X) = the coefficient of hn in series expansion of (cid:18)Q(1 + aih) 1 + dh For example, Fano number 337 is X28 ⊂ P(1, 4, 6, 7, 11) and has basket B =(cid:26)2 × 1 2 (1, 1, 1), 1 6 (1, 1, 5), (1, 4, 7)(cid:27) . 1 11 Calculating as above gives e(X) = eorb(X) + 2 × + + 5 6 10 11 1 2 1 2 = coeffh3(cid:0)(1 + 29h + 309h2)(1 − 28h + 784h2 − 21952h3)(cid:1) +2 × + + 5 6 10 11 = coeffh3(1 + h + 281h2 − 6385h3) 1 66 + 2 × 1 2 + 5 6 + 10 11 −6385 = 66 = −94. + 1 + 5/6 + 10/11 This agrees with our calculation h2,1(X) = 49 and e(X) = 4 − 2 × 49. deg(X)(cid:19) . 28 4 · 6 · 7 · 11 18 GAVIN BROWN AND ENRICO FATIGHENTI A.3. Tables of results. Tables 1–3 list the Hodge number h2,1(X), the topological euler char- acteristic e(X) and the number of moduli h1(TX ) = dim H 1(X, TX ) for quasismooth members X of the families of Fano 3-folds in codimensions 1–3 respectively. In codimension 1, we apply the Griffith's Residue Theorem in §2.2 together with the formulas of Theorem 1. In codimension 2, Table 2 documents the method we use to compute the invari- ants. This could be the conventional chern class calculation, indicated by c3(TX), a computer AX , indicated by T 1, or a projection calculation, indicated by I or II1 depending calculation of T 1 on the type of the projection. Where we use a projection, we also list the centre 1 r of projection (leaving the polarising weights of 1 r (1, a, −a) implicit), the number of nodes on the image of projection, and the number of that image in the Grdb. Where there is more than one possible centre of projection, we list them all. Combining this data with the results of Table 1 and Theorems 1 and 3 calculates the invariants. For example, number 25022, X3,3 ⊂ P(15, 2) (the second line in Table 2) projects to number 20521 with 9 nodes; the Euler charactistic of the smoothed image is listed in Table 1 as −56, and so the for X3,3 it is −56 + 2 × 9 − 2 = −40, as displayed. In codimension 3, Table 3 documents the method we use in the 70 cases as follows: (i) 57 cases have at least one 'staircase' of two Type I projections to a hypersurface. This is indicated by I–I. (ii) 4 cases have a Type I projection to a codimension 2 family that has as a Type II1 projection to a hypersurface (indicated by I–II1). (iii) 2 cases have a Type II1 projection directly to a hypersurface (II1). (iv) 2 cases have a Type I projection to a codimension 2 family with no projection (I–T 1). (v) 1 case has a Type I projection to a known smooth Fano (I–smooth). (vi) 1 case is a known smooth Fano complete intersection (c3(TX )). (vii) 3 cases have no Type I or II1 projections at all (T 1). Again, where there is a projection from X we list the centre 1 r , the number of nodes and the Grdb identifier for each possibility, and applying Theorems 1 and 3 together with data from previous tables calculates the invariants. Table 1: Codimension 1: h1,1(X) = 1 and h0(X, TX ) = 0 in all cases. Grdb variety h2,1 e(X) h1(TX ) 20521 X4 ⊂ P4 16203 X5 ⊂ P(1, 1, 1, 1, 2) 16202 X6 ⊂ P(1, 1, 1, 1, 3) 11101 X6 ⊂ P(1, 1, 1, 2, 2) 10981 X7 ⊂ P(1, 1, 1, 2, 3) 10980 X8 ⊂ P(1, 1, 1, 2, 4) 10960 X9 ⊂ P(1, 1, 1, 3, 4) 10959 X10 ⊂ P(1, 1, 1, 3, 5) 10958 X12 ⊂ P(1, 1, 1, 4, 6) 5838 X8 ⊂ P(1, 1, 2, 2, 3) 5837 X10 ⊂ P(1, 1, 2, 2, 5) 5257 X9 ⊂ P(1, 1, 2, 3, 3) 5157 X10 ⊂ P(1, 1, 2, 3, 4) 5153 X11 ⊂ P(1, 1, 2, 3, 5) 5152 X12 ⊂ P(1, 1, 2, 3, 6) 30 −56 38 −72 52 −100 41 −78 51 −98 64 −124 71 −138 85 −166 111 −218 45 −86 64 −124 49 −94 56 −108 65 −126 75 −146 43 51 66 55 63 78 83 98 125 58 79 62 66 74 88 Continued on next page HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 19 Table 1 continued from previous page 5137 X12 ⊂ P(1, 1, 2, 4, 5) 5136 X14 ⊂ P(1, 1, 2, 4, 7) 5134 X15 ⊂ P(1, 1, 2, 5, 7) 5133 X16 ⊂ P(1, 1, 2, 5, 8) 5132 X18 ⊂ P(1, 1, 2, 6, 9) 4984 X12 ⊂ P(1, 1, 3, 4, 4) 4909 X13 ⊂ P(1, 1, 3, 4, 5) 4907 X15 ⊂ P(1, 1, 3, 4, 7) 4906 X16 ⊂ P(1, 1, 3, 4, 8) 4893 X15 ⊂ P(1, 1, 3, 5, 6) 4892 X18 ⊂ P(1, 1, 3, 5, 9) 4891 X21 ⊂ P(1, 1, 3, 7, 10) 4890 X22 ⊂ P(1, 1, 3, 7, 11) 4889 X24 ⊂ P(1, 1, 3, 8, 12) 4835 X16 ⊂ P(1, 1, 4, 5, 6) 4834 X20 ⊂ P(1, 1, 4, 5, 10) 4822 X18 ⊂ P(1, 1, 4, 6, 7) 4821 X22 ⊂ P(1, 1, 4, 6, 11) 4820 X28 ⊂ P(1, 1, 4, 9, 14) 4819 X30 ⊂ P(1, 1, 4, 10, 15) 4807 X21 ⊂ P(1, 1, 5, 7, 8) 4806 X26 ⊂ P(1, 1, 5, 7, 13) 4805 X36 ⊂ P(1, 1, 5, 12, 18) 4794 X24 ⊂ P(1, 1, 6, 8, 9) 4793 X30 ⊂ P(1, 1, 6, 8, 15) 4792 X42 ⊂ P(1, 1, 6, 14, 21) 2402 X12 ⊂ P(1, 2, 2, 3, 5) 2401 X14 ⊂ P(1, 2, 2, 3, 7) 1389 X12 ⊂ P(1, 2, 3, 3, 4) 1162 X14 ⊂ P(1, 2, 3, 4, 5) 1160 X16 ⊂ P(1, 2, 3, 4, 7) 1159 X18 ⊂ P(1, 2, 3, 4, 9) 1155 X15 ⊂ P(1, 2, 3, 5, 5) 1149 X17 ⊂ P(1, 2, 3, 5, 7) 1147 X18 ⊂ P(1, 2, 3, 5, 8) 1146 X20 ⊂ P(1, 2, 3, 5, 10) 1144 X21 ⊂ P(1, 2, 3, 7, 9) 1143 X24 ⊂ P(1, 2, 3, 7, 12) 1142 X24 ⊂ P(1, 2, 3, 8, 11) 1141 X26 ⊂ P(1, 2, 3, 8, 13) 1140 X30 ⊂ P(1, 2, 3, 10, 15) 1113 X20 ⊂ P(1, 2, 4, 5, 9) 1112 X22 ⊂ P(1, 2, 4, 5, 11) 1079 X20 ⊂ P(1, 2, 5, 6, 7) 1078 X26 ⊂ P(1, 2, 5, 6, 13) 1076 X27 ⊂ P(1, 2, 5, 9, 11) 1075 X32 ⊂ P(1, 2, 5, 9, 16) 70 −136 90 −176 97 −190 108 −212 128 −252 60 −116 66 −128 82 −160 91 −178 78 −152 104 −204 126 −248 136 −268 154 −304 77 −150 108 −212 88 −172 120 −236 165 −326 182 −360 99 −194 137 −270 211 −418 110 −216 154 −304 240 −476 47 −90 60 −116 40 −76 45 −86 54 −104 65 −126 48 −92 56 −108 61 −118 72 −140 72 −140 89 −174 87 −170 99 −194 121 −238 62 −120 72 −140 55 −106 80 −156 77 −150 100 −196 81 102 106 119 141 73 73 89 102 87 114 133 144 165 83 119 94 127 172 190 104 143 218 115 160 247 59 74 54 52 62 76 60 60 66 82 78 97 93 106 131 70 81 60 87 79 104 Continued on next page 20 GAVIN BROWN AND ENRICO FATIGHENTI Table 1 continued from previous page 1074 X42 ⊂ P(1, 2, 5, 14, 21) 1067 X30 ⊂ P(1, 2, 6, 7, 15) 866 X15 ⊂ P(1, 3, 3, 4, 5) 545 X18 ⊂ P(1, 3, 4, 5, 6) 539 X19 ⊂ P(1, 3, 4, 5, 7) 537 X20 ⊂ P(1, 3, 4, 5, 8) 536 X24 ⊂ P(1, 3, 4, 5, 12) 534 X24 ⊂ P(1, 3, 4, 7, 10) 533 X28 ⊂ P(1, 3, 4, 7, 14) 532 X30 ⊂ P(1, 3, 4, 10, 13) 531 X34 ⊂ P(1, 3, 4, 10, 17) 530 X36 ⊂ P(1, 3, 4, 11, 18) 529 X42 ⊂ P(1, 3, 4, 14, 21) 508 X21 ⊂ P(1, 3, 5, 6, 7) 507 X33 ⊂ P(1, 3, 5, 11, 14) 506 X38 ⊂ P(1, 3, 5, 11, 19) 505 X48 ⊂ P(1, 3, 5, 16, 24) 500 X24 ⊂ P(1, 3, 6, 7, 8) 356 X24 ⊂ P(1, 4, 5, 6, 9) 355 X30 ⊂ P(1, 4, 5, 6, 15) 353 X25 ⊂ P(1, 4, 5, 7, 9) 352 X32 ⊂ P(1, 4, 5, 7, 16) 351 X44 ⊂ P(1, 4, 5, 13, 22) 350 X54 ⊂ P(1, 4, 5, 18, 27) 337 X28 ⊂ P(1, 4, 6, 7, 11) 336 X34 ⊂ P(1, 4, 6, 7, 17) 296 X27 ⊂ P(1, 5, 6, 7, 9) 295 X30 ⊂ P(1, 5, 6, 8, 11) 294 X38 ⊂ P(1, 5, 6, 8, 19) 293 X66 ⊂ P(1, 5, 6, 22, 33) 289 X40 ⊂ P(1, 5, 7, 8, 20) 271 X36 ⊂ P(1, 7, 8, 9, 12) 270 X50 ⊂ P(1, 7, 8, 10, 25) 144 −284 88 −172 40 −76 42 −80 45 −86 48 −92 63 −122 57 −110 72 −140 74 −144 90 −176 97 −190 120 −236 45 −86 74 −144 92 −180 126 −248 48 −92 45 −86 62 −120 46 −88 65 −126 91 −178 120 −236 49 −94 65 −126 42 −80 46 −88 64 −124 120 −236 64 −124 42 −80 63 −122 150 96 52 49 47 53 71 58 80 75 92 101 125 51 74 93 130 56 47 69 46 69 91 121 50 67 42 45 64 120 68 41 62 Table 2: Codimension 2: h1,1(X) = 1 and h0(X, TX ) = 0 in all cases. grdb variety 24076 X2,3 ⊂ P5 20522 X3,3 ⊂ P(1, 1, 1, 1, 1, 2) 16225 X3,4 ⊂ P(1, 1, 1, 1, 2, 2) 16204 X4,4 ⊂ P(1, 1, 1, 1, 2, 3) 11435 X4,4 ⊂ P(1, 1, 1, 2, 2, 2) 11102 X4,5 ⊂ P(1, 1, 1, 2, 2, 3) 11002 X4,6 ⊂ P(1, 1, 1, 2, 3, 3) method c3(TX ) I I I I I I 1 r , #nodes, target id h2,1 e(X) h1(TX ) 1 2 , 9, 20521 1 2 , 12, 16203 1 3 , 8, 16203 1 2 , 16, 11101 1 3 , 12, 10981 1 2 , 20, 10981; 1 3 , 10, 11101 20 22 27 31 26 32 40 −36 −40 −50 −58 −48 −60 −76 34 36 41 45 39 45 53 Continued on next page Table 2 continued from previous page 10983 X5,6 ⊂ P(1, 1, 1, 2, 3, 4) 10982 X6,6 ⊂ P(1, 1, 1, 2, 3, 5) 10961 X6,8 ⊂ P(1, 1, 1, 3, 4, 5) 6858 X4,6 ⊂ P(1, 1, 2, 2, 2, 3) 5857 X5,6 ⊂ P(1, 1, 2, 2, 3, 3) 5843 X6,6 ⊂ P(1, 1, 2, 2, 3, 4) 5839 X6,7 ⊂ P(1, 1, 2, 2, 3, 5) 5514 X6,6 ⊂ P(1, 1, 2, 3, 3, 3) 5261 X6,7 ⊂ P(1, 1, 2, 3, 3, 4) 5258 X6,8 ⊂ P(1, 1, 2, 3, 3, 5) 5200 X6,8 ⊂ P(1, 1, 2, 3, 4, 4) 5161 X7,8 ⊂ P(1, 1, 2, 3, 4, 5) 5159 X6,9 ⊂ P(1, 1, 2, 3, 4, 5) 5158 X8,9 ⊂ P(1, 1, 2, 3, 4, 7) 5156 X6,10 ⊂ P(1, 1, 2, 3, 5, 5) 5155 X8,10 ⊂ P(1, 1, 2, 3, 5, 7) 5154 X9,10 ⊂ P(1, 1, 2, 3, 5, 8) 5138 X8,10 ⊂ P(1, 1, 2, 4, 5, 6) 5135 X10,14 ⊂ P(1, 1, 2, 5, 7, 9) 4985 X8,9 ⊂ P(1, 1, 3, 4, 4, 5) 4936 X8,10 ⊂ P(1, 1, 3, 4, 5, 5) 4912 X9,10 ⊂ P(1, 1, 3, 4, 5, 6) 4911 X8,12 ⊂ P(1, 1, 3, 4, 5, 7) 4910 X10,12 ⊂ P(1, 1, 3, 4, 5, 9) 4908 X12,14 ⊂ P(1, 1, 3, 4, 7, 11) 4894 X10,12 ⊂ P(1, 1, 3, 5, 6, 7) 4848 X10,12 ⊂ P(1, 1, 4, 5, 6, 6) 4837 X11,12 ⊂ P(1, 1, 4, 5, 6, 7) 4836 X12,15 ⊂ P(1, 1, 4, 5, 6, 11) 4823 X12,14 ⊂ P(1, 1, 4, 6, 7, 8) 4808 X14,16 ⊂ P(1, 1, 5, 7, 8, 9) 4795 X16,18 ⊂ P(1, 1, 6, 8, 9, 10) 3508 X6,6 ⊂ P(1, 2, 2, 2, 3, 3) 2419 X6,8 ⊂ P(1, 2, 2, 3, 3, 4) 2409 X6,10 ⊂ P(1, 2, 2, 3, 4, 5) 2403 X9,10 ⊂ P(1, 2, 2, 3, 5, 7) 1390 X8,9 ⊂ P(1, 2, 3, 3, 4, 5) 1249 X8,10 ⊂ P(1, 2, 3, 4, 4, 5) 1179 X9,10 ⊂ P(1, 2, 3, 4, 5, 5) 1171 X8,12 ⊂ P(1, 2, 3, 4, 5, 6) 1165 X10,11 ⊂ P(1, 2, 3, 4, 5, 7) 1164 X9,12 ⊂ P(1, 2, 3, 4, 5, 7) 1163 X10,12 ⊂ P(1, 2, 3, 4, 5, 8) 1161 X12,14 ⊂ P(1, 2, 3, 4, 7, 10) 1156 X10,12 ⊂ P(1, 2, 3, 5, 5, 7) 1154 X10,14 ⊂ P(1, 2, 3, 5, 7, 7) 1152 X10,15 ⊂ P(1, 2, 3, 5, 7, 8) I I I II1 I I I I I I I I I I I I I I I I I I I I I I I I I I I I T 1 II1 II1 I I II1 I II1 I I I I I I I 1 3 , 21, 5157; 1 3 , 24, 5153; 1 4 , 14, 5257 5 , 8, 5257 1 1 4 , 16, 5157 1 3 , 28, 5137; 1 4 , 18, 5153; 1 5 , 14, 5157 5 , 9, 5157 1 1 3 , 40, 5134; 1 7 , 8, 5153 1 4 , 24, 4909; 1 5 , 18, 4984 1 4 , 30, 4893; 1 5 , 24, 4907; 1 6 , 18, 4909 7 , 8, 4909 1 1 5 , 33, 4822; 1 7 , 22, 4835 1 5 , 6, 10981 1 5 , 12, 10960 1 2 , 34, 5837 1 3 , 15, 5838 1 4 , 12, 5838 1 5 , 7, 5838 1 3 , 18, 5257 1 7 , 6, 5157 1 5 , 10, 5153 1 8 , 6, 5153 1 6 , 16, 5137 1 9 , 10, 5134 1 5 , 20, 4909 1 9 , 6, 4909 1 11 , 6, 4907 1 7 , 20, 4893 1 6 , 24, 4835 1 11 , 6, 4835 1 8 , 24, 4822 1 9 , 28, 4807 1 10 , 32, 4794 1 3 , 33, 2401 1 4 , 25, 2401 1 7 , 9, 2402 1 5 , 12, 1389 1 4 , 36, 1159 1 5 , 15, 1162 1 5 , 30, 1159 1 7 , 11, 1162 1 8 , 8, 1162 1 10 , 8, 1160 1 7 , 14, 1149 HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 21 1 2 , 30, 10960; 1 4 , 10, 10981 −80 42 46 −88 60 −116 −58 31 −58 31 −64 34 39 −74 −60 32 −68 36 42 −80 −78 41 −82 43 −92 48 51 −98 56 −108 58 −112 60 −116 55 −106 88 −172 −82 43 47 −90 49 −94 59 −114 61 −118 77 −150 59 −114 54 −104 56 −108 72 −140 65 −126 72 −140 79 −154 −44 24 −52 28 36 −68 −74 39 −54 29 30 −56 −58 31 −68 36 −66 35 37 −70 −72 38 −90 47 −70 37 43 −82 −90 47 55 59 73 43 42 45 50 42 46 52 51 53 58 61 66 68 70 65 98 51 55 57 67 69 85 67 61 63 79 72 78 85 34 37 45 47 36 37 37 43 41 43 44 53 42 48 52 1 5 , 18, 1160; 1 7 , 9, 1162 1 5 , 20, 1149; 1 7 , 12, 1155 1 7 , 15, 1147; 1 8 , 10, 1149 Continued on next page 22 GAVIN BROWN AND ENRICO FATIGHENTI Table 2 continued from previous page 1151 X12,14 ⊂ P(1, 2, 3, 5, 7, 9) 1150 X14,15 ⊂ P(1, 2, 3, 5, 7, 12) 1148 X15,16 ⊂ P(1, 2, 3, 5, 8, 13) 1145 X14,18 ⊂ P(1, 2, 3, 7, 9, 11) 1121 X10,12 ⊂ P(1, 2, 4, 5, 5, 6) 1114 X10,14 ⊂ P(1, 2, 4, 5, 6, 7) 1083 X12,16 ⊂ P(1, 2, 5, 6, 7, 8) 1080 X14,15 ⊂ P(1, 2, 5, 6, 7, 9) 1077 X18,22 ⊂ P(1, 2, 5, 9, 11, 13) 1068 X14,18 ⊂ P(1, 2, 6, 7, 8, 9) 867 X10,12 ⊂ P(1, 3, 3, 4, 5, 7) 640 X10,12 ⊂ P(1, 3, 4, 4, 5, 6) 547 X12,13 ⊂ P(1, 3, 4, 5, 6, 7) 546 X12,15 ⊂ P(1, 3, 4, 5, 6, 9) 544 X12,14 ⊂ P(1, 3, 4, 5, 7, 7) 542 X12,15 ⊂ P(1, 3, 4, 5, 7, 8) 541 X14,15 ⊂ P(1, 3, 4, 5, 7, 10) 540 X14,16 ⊂ P(1, 3, 4, 5, 7, 11) 538 X15,16 ⊂ P(1, 3, 4, 5, 8, 11) 535 X20,21 ⊂ P(1, 3, 4, 7, 10, 17) 509 X14,15 ⊂ P(1, 3, 5, 6, 7, 8) 453 X12,14 ⊂ P(1, 4, 4, 5, 6, 7) 359 X14,16 ⊂ P(1, 4, 5, 6, 7, 8) 358 X12,20 ⊂ P(1, 4, 5, 6, 7, 10) 357 X18,20 ⊂ P(1, 4, 5, 6, 9, 14) 354 X18,20 ⊂ P(1, 4, 5, 7, 9, 13) 338 X16,18 ⊂ P(1, 4, 6, 7, 8, 9) 297 X18,20 ⊂ P(1, 5, 6, 7, 9, 11) 279 X18,30 ⊂ P(1, 6, 8, 9, 10, 15) 265 X24,30 ⊂ P(1, 8, 9, 10, 12, 15) 37 X12,14 ⊂ P(2, 3, 4, 5, 6, 7) I I I I II1 II1 II1 I I II1 I T 1 I I I I I I I I I T 1 T 1 II1 I I T 1 I T 1 T 1 T 1 1 5 , 28, 1144; 1 9 , 12, 1149 1 12 , 6, 1149 1 13 , 6, 1147 1 11 , 14, 1144 1 5 , 40, 1112 1 6 , 35, 1112 1 9 , 15, 1079 1 13 , 18, 1076 1 8 , 45, 1067 1 7 , 10, 866 1 5 , 48, 1067; 1 7 , 40, 1078 1 7 , 15, 537; 1 8 , 12, 539 1 7 , 13, 545 1 9 , 9, 545 1 7 , 14, 539 1 10 , 10, 539 1 11 , 8, 539 1 11 , 10, 537 1 17 , 6, 534 1 8 , 14, 508 1 7 , 27, 355 1 14 , 8, 356 1 13 , 10, 353 1 11 , 12, 296 −86 45 51 −98 56 −108 59 −114 −62 33 −72 38 41 −78 41 −78 60 −116 44 −84 −58 31 −52 28 −56 30 34 −64 −60 32 −64 34 36 −68 −72 38 39 −74 52 −100 32 −60 −52 28 −54 29 36 −68 −72 38 −70 37 −56 30 31 −58 −68 36 −56 30 18 −32 50 56 61 64 39 44 46 45 63 49 36 33 34 38 35 37 39 41 42 54 35 32 32 39 40 38 33 32 38 31 23 Table 3: Codimension 3: h1,1(X) = 1 and h0(X, TX ) = 0 in all cases. grdb variety method 1 r , #nodes, target id h2,1 e(X) h1(TX) 26988 X2,2... = X2,2,2 ⊂ P6 24077 X2,3... ⊂ P(1, 1, 1, 1, 1, 1, 2) 20543 X3,3... ⊂ P(1, 1, 1, 1, 1, 2, 2) 20523 X3,3... ⊂ P(1, 1, 1, 1, 1, 2, 3) 16338 X3,3... ⊂ P(1, 1, 1, 1, 2, 2, 2) 16226 X3,4... ⊂ P(1, 1, 1, 1, 2, 2, 3) 16205 X4,4... ⊂ P(1, 1, 1, 1, 2, 3, 4) 12062 X4,4... ⊂ P(1, 1, 1, 2, 2, 2, 2) 11436 X4,4... ⊂ P(1, 1, 1, 2, 2, 2, 3) c3(TX ) I – T 1 I – I I – I I – I I – I I – I I – I I – I 1 2 , 7, 24076 1 2 , 8, 20522 1 3 , 6, 20522 1 2 , 10, 16225 1 4 , 7, 16204 1 2 , 12, 11435 1 2 , 11, 16204; 1 3 , 7, 16225 1 2 , 14, 11102; 1 3 , 8, 11435 Continued on next page 14 14 15 17 18 21 25 15 19 −24 −24 −26 −30 −32 −38 −46 −26 −34 27 27 28 30 31 34 38 27 31 HODGE NUMBERS AND DEFORMATIONS OF FANO 3-FOLDS 23 Table 3 continued from previous page 11122 X4,4... ⊂ P(1, 1, 1, 2, 2, 3, 3) 11105 X4,5... ⊂ P(1, 1, 1, 2, 2, 3, 4) 11103 X4,5... ⊂ P(1, 1, 1, 2, 2, 3, 5) 11003 X4,5... ⊂ P(1, 1, 1, 2, 3, 3, 4) 10984 X5,6... ⊂ P(1, 1, 1, 2, 3, 4, 5) 10962 X6,7... ⊂ P(1, 1, 1, 3, 4, 5, 6) 6859 X4,5... ⊂ P(1, 1, 2, 2, 2, 3, 3) 5962 X5,5... ⊂ P(1, 1, 2, 2, 3, 3, 3) 5865 X5,6... ⊂ P(1, 1, 2, 2, 3, 3, 4) 5858 X5,6... ⊂ P(1, 1, 2, 2, 3, 3, 5) 5844 X6,6... ⊂ P(1, 1, 2, 2, 3, 4, 5) 5840 X6,7... ⊂ P(1, 1, 2, 2, 3, 5, 7) 5515 X6,6... ⊂ P(1, 1, 2, 3, 3, 3, 4) 5302 X6,6... ⊂ P(1, 1, 2, 3, 3, 4, 4) 5267 X6,7... ⊂ P(1, 1, 2, 3, 3, 4, 5) 5264 X6,6... ⊂ P(1, 1, 2, 3, 3, 4, 5) 5262 X6,7... ⊂ P(1, 1, 2, 3, 3, 4, 7) 5259 X6,8... ⊂ P(1, 1, 2, 3, 3, 5, 8) 5201 X6,7... ⊂ P(1, 1, 2, 3, 4, 4, 5) 5175 X6,7... ⊂ P(1, 1, 2, 3, 4, 5, 5) 5162 X7,8... ⊂ P(1, 1, 2, 3, 4, 5, 6) 5160 X6,8... ⊂ P(1, 1, 2, 3, 4, 5, 7) 5139 X8,9... ⊂ P(1, 1, 2, 4, 5, 6, 7) 4999 X8,8... ⊂ P(1, 1, 3, 4, 4, 5, 5) 4988 X8,9... ⊂ P(1, 1, 3, 4, 4, 5, 6) 4986 X8,9... ⊂ P(1, 1, 3, 4, 4, 5, 9) 4937 X8,9... ⊂ P(1, 1, 3, 4, 5, 5, 6) 4914 X9,10... ⊂ P(1, 1, 3, 4, 5, 6, 7) 4913 X8,9... ⊂ P(1, 1, 3, 4, 5, 6, 7) 4895 X10,11... ⊂ P(1, 1, 3, 5, 6, 7, 8) 4849 X10,11... ⊂ P(1, 1, 4, 5, 6, 6, 7) 4838 X11,12... ⊂ P(1, 1, 4, 5, 6, 7, 8) 4824 X12,13... ⊂ P(1, 1, 4, 6, 7, 8, 9) 4809 X14,15... ⊂ P(1, 1, 5, 7, 8, 9, 10) 4796 X16,17... ⊂ P(1, 1, 6, 8, 9, 10, 11) 2420 X6,7... ⊂ P(1, 2, 2, 3, 3, 4, 5) 2404 X9,10... ⊂ P(1, 2, 2, 3, 5, 7, 9) 1409 X7,8... ⊂ P(1, 2, 3, 3, 4, 4, 5) 1396 X8,8... ⊂ P(1, 2, 3, 3, 4, 5, 5) 1394 X8,9... ⊂ P(1, 2, 3, 3, 4, 5, 7) 1391 X8,9... ⊂ P(1, 2, 3, 3, 4, 5, 8) 1252 X8,9... ⊂ P(1, 2, 3, 4, 4, 5, 5) 1250 X8,9... ⊂ P(1, 2, 3, 4, 4, 5, 7) 1184 X8,9... ⊂ P(1, 2, 3, 4, 5, 5, 6) 1180 X9,10... ⊂ P(1, 2, 3, 4, 5, 5, 7) 1168 X9,10... ⊂ P(1, 2, 3, 4, 5, 7, 7) 1166 X10,11... ⊂ P(1, 2, 3, 4, 5, 7, 9) I – I I – I I – I I – I I – I I – I I – II1 I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – I I – II1 I – I II1 I – I I – I I – I I – II1 I – II1 I – II1 I – I I – I I – I 1 3 , 19, 5159; 1 4 , 13, 5258; 1 5 , 7, 5261 1 1 1 2 , 17, 11002; 1 2 , 18, 10983; 1 2 , 19, 10982; 1 3 , 11, 10983; 1 2 , 27, 10961; 1 3 , 9, 11102 4 , 8, 11102 5 , 5, 11102 4 , 9, 11002 5 , 9, 10983 1 1 1 6 , 11, 10961 1 3 , 11, 6858 1 3 , 12, 5857 1 3 , 13, 5843; 1 3 , 14, 5839; 1 4 , 10, 5857 5 , 6, 5857 1 1 5 , 10, 5843 1 7 , 6, 5839 1 1 3 , 15, 5261; 1 3 , 17, 5200; 1 3 , 18, 5161; 1 4 , 11, 5514 4 , 12, 5261 5 , 11, 5261 1 1 1 1 3 , 20, 5158; 1 3 , 23, 5154; 1 4 , 14, 5161; 1 5 , 13, 5159; 1 3 , 24, 5138; 1 4 , 17, 5155; 1 7 , 5, 5261 8 , 5, 5258 5 , 12, 5200 5 , 8, 5161 6 , 12, 5161 7 , 7, 5159 1 1 1 1 7 , 14, 5138 1 1 1 4 , 19, 4936; 1 4 , 20, 4912; 1 4 , 23, 4910; 1 5 , 17, 4912; 1 4 , 25, 4894; 1 6 , 17, 4911; 1 5 , 15, 4985 6 , 14, 4985 9 , 5, 4985 6 , 15, 4936 7 , 15, 4912 7 , 7, 4912 1 1 1 1 8 , 17, 4894 1 6 , 20, 4837; 1 5 , 27, 4823; 1 7 , 18, 4848 8 , 18, 4837 1 −44 24 −46 25 −52 28 −60 32 −64 34 −96 50 −38 21 −36 20 −40 22 −48 26 −46 25 −64 34 −40 22 −46 25 −48 26 −56 30 −60 32 −72 38 −56 30 −68 36 −60 32 −80 42 −80 42 −54 29 −56 30 −74 39 −62 33 −66 35 −82 43 −82 43 −70 37 −74 39 −88 46 50 −96 54 −104 −38 21 −60 32 20 −36 −36 20 −40 22 −44 24 20 −36 −44 24 −44 24 −42 23 28 −52 −50 27 36 37 40 44 46 62 32 30 32 36 35 44 31 34 35 39 41 47 39 45 41 51 51 36 37 46 40 42 50 50 43 45 52 55 59 29 39 27 26 28 30 26 30 30 28 33 32 1 9 , 20, 4823 1 10 , 23, 4808 1 11 , 26, 4795 1 5 , 8, 2419 1 9 , 8, 2403 1 4 , 21, 1389 1 5 , 10, 1390 1 7 , 8, 1390 1 8 , 6, 1390 1 5 , 11, 1249 1 7 , 7, 1249 1 5 , 13, 1171 1 5 , 13, 1165; 1 7 , 10, 1164; 1 7 , 9, 1179 7 , 8, 1165 1 1 9 , 9, 1165 Continued on next page 24 GAVIN BROWN AND ENRICO FATIGHENTI Table 3 continued from previous page 1157 X10,12... ⊂ P(1, 2, 3, 5, 5, 7, 12) 1153 X10,12... ⊂ P(1, 2, 3, 5, 7, 8, 9) 1090 X12,13... ⊂ P(1, 2, 5, 6, 7, 7, 8) 1081 X14,15... ⊂ P(1, 2, 5, 6, 7, 9, 11) 868 X10,12... ⊂ P(1, 3, 3, 4, 5, 7, 10) 641 X10,11... ⊂ P(1, 3, 4, 4, 5, 6, 7) 568 X10,11... ⊂ P(1, 3, 4, 5, 5, 6, 7) 548 X12,13... ⊂ P(1, 3, 4, 5, 6, 7, 10) 543 X12,14... ⊂ P(1, 3, 4, 5, 7, 8, 11) 510 X14,15... ⊂ P(1, 3, 5, 6, 7, 8, 11) 454 X12,13... ⊂ P(1, 4, 4, 5, 6, 7, 9) 392 X12,13... ⊂ P(1, 4, 5, 5, 6, 7, 8) 326 X14,15... ⊂ P(1, 5, 5, 6, 7, 8, 9) 298 X16,17... ⊂ P(1, 5, 6, 7, 8, 9, 10) I – I I – I I – II1 I – I I – I I – T 1 II1 I – I I – I I – I I – T 1 T 1 T 1 T 1 1 5 , 19, 1150; 1 8 , 9, 1151; 1 12 , 5, 1156 9 , 11, 1152 1 1 8 , 11, 540; 1 11 , 7, 542 1 7 , 15, 1083 1 11 , 12, 1080 1 10 , 7, 867 1 7 , 9, 640 1 5 , 22, 545 1 10 , 8, 547 1 11 , 9, 509 1 9 , 8, 453 33 37 27 30 25 20 21 23 28 24 21 20 20 20 −62 −70 −50 −56 −46 −36 −38 −42 −52 −44 −38 −36 −36 −36 37 41 31 33 29 24 25 26 30 26 24 23 22 22
1302.2316	3	1302	2013-02-17T13:13:06	Classification of involutions on Enriques surfaces	[ "math.AG" ]	We present the classification of involutions on Enriques surfaces. We classify those into 18 types with the help of the lattice theory due to Nikulin. We also give all examples of the classification.	math.AG	math	CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES HIROKI ITO AND HISANORI OHASHI Abstract. We present the classification of involutions on Enriques surfaces. We classify those into 18 types with the help of the lattice theory due to Nikulin. We also give all examples of the classification. 1. Introduction An Enriques surface Y is a compact complex surface satisfying the following conditions: (1) the geometric genus and the irregularity vanish, (2) the bi-canonical divisor on Y is linearly equivalent to 0. Every Enriques surface Y is a quotient of a K3 surface X by a fixed point free involution ε. In this work, we give the classification of involutions on Enriques surfaces. An involution ι on Y lifts to two involutions of X. One of them, which we denote by g, acts on H 0(X, Ω2) trivially. An involution with this property is called symplectic or Nikulin involution. To classify ι, we study the pair of involutions (g, ε). For our purpose, we use the theory of the classification of involutions of a lattice with condition on a sublattice, due to V. V. Nikulin [Nik4]. Let S be a lattice and θ an involution of S. In [Nik4], the determining condition of a triple (L, ϕ, i) with the condition (S, θ) satisfying the following commutative diagram is given: L i S ϕ / L i S θ Here L is a unimodular lattice, ϕ is an involution of L, and i : S → L is a primitive embedding. To investigate (L, ϕ, i), we use the following invariants: Let L± = {x ∈ L ϕ(x) = ±x} and S± = {x ∈ S θ(x) = ±x}. From the primitive embedding i : S → L, we get primitive embeddings i± : S± → L±. Hence we have Date: September 25, 2018. 2000 Mathematics Subject Classification. 14J28. Supported in part by the JSPS Grant-in-Aid for Scientific Research (S) 22224001 and for Young Scientists (B) 23740010. 1 / / / ? O O ? O O 2 H. ITO AND H. OHASHI the orthogonal complements K± = (S±)⊥ L± and images of projection H− = pS−((L ∩ (L+ ⊕ S−) ⊗ Q)/L+ ⊕ S−) ⊂ AS− , gH− = pS−((L ∩ (K+ ⊕ S−) ⊗ Q)/K+ ⊕ S−) ⊂ H−, where AS− is the discriminant group of S−. We apply this theory as L = H 2(X, Z), S = {x ∈ H 2(X, Z) g∗(x) = −x} and ϕ = ε∗. Next theorem is our main result. Theorem 1.1. Involutions of Enriques surfaces are classified as follows: Table 1: Invariants and the model No. S+( 1 2 ) S−( 1 2 ) qS−H− qS−gH− Horikawa model [1] [2] [3] {0} {0} {0} [4] A1 [5] A1 [6] A2 1 [7] A2 1 [8] A3 1 [9] A3 1 [10] D4 [11] D4 [12] D4 [13] D4 E8 E8 E8 E7 E7 D6 D6 u4 u3 ⊕ w u3 ⊕ z u3 ⊕ w u2 ⊕ w2 u2 ⊕ w2 u ⊕ w3 D4 ⊕ A1 u ⊕ w3 D4 ⊕ A1 w4 D4 D4 D4 D4 v ⊕ z2 v ⊕ z2 w ⊕ z2 w ⊕ z2 w ⊕ z2 z2 (See Subsection 6.2) CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 3 No. S+( 1 2 ) S−( 1 2 ) qS−H− qS−gH− Horikawa model [14] A4 1 A4 1 [15] D4 ⊕ A1 A3 1 [16] D6 [17] E7 [18] E8 A2 1 A1 {0} w4 w3 w2 w -- In Table 1, the blank in qS−gH− ants are collected in the next table. stands for the same as qS−H− . Further invari- Table 2: Further Invariants [1] [2] [3] [4] [5] [6] [7] No. k− K− U ⊕ U (2) U (2) ⊕ U (2) U (2) ⊕ U (2) U ⊕ U (2) ⊕ A1(2) U (2) ⊕ U (2) ⊕ A1(2) U ⊕ U (2) ⊕ A1(2)2 U (2) ⊕ U (2) ⊕ A1(2)2 U ⊕ U (2) ⊕ A1(2)3 U (2) ⊕ U (2) ⊕ A1(2)3 U ⊕ U (2) ⊕ D4(2) U ⊕ U (2) ⊕ D4(2) u u2 u2 u ⊕ h −1 4 i u2 ⊕ h −1 4 i u ⊕ h −1 4 i2 u2 ⊕ h −1 4 i2 u ⊕ h −1 4 i3 [8] u2 ⊕ h −1 4 i3 [9] [10] u ⊕ v ⊕ v(4) [11] u ⊕ v ⊕ v(4) [12] u2 ⊕ v ⊕ v(4) U (2) ⊕ U (2) ⊕ D4(2) [13] u2 ⊕ v ⊕ v(4) U (2) ⊕ U (2) ⊕ D4(2) [14] u ⊕ h 1 4i4 [15] u2 ⊕ h 1 4i3 [16] u3 ⊕ h 1 4i2 [17] u4 ⊕ h 1 4i [18] u5 U ⊕ U (2) ⊕ A1(2)4 U ⊕ U (2) ⊕ D4(2) ⊕ A1(2) U ⊕ U (2) ⊕ D6(2) U ⊕ U (2) ⊕ E7(2) U ⊕ U (2) ⊕ E8(2) Fixed curves (r, l, δ) (18, 2, 0) C(1) + 4P1 (18, 4, 0) (18, 4, 0) 4P1 4P1 (16, 4, 1) C(1) + 3P1 (16, 6, 1) 3P1 (14, 6, 1) C(1) + 2P1 (14, 8, 1) 2P1 (12, 8, 1) C(1) + P1 (12, 10, 1) P1 1 + C(1) (10, 6, 0) C(2) + P1 (10, 8, 0) C(1) (10, 8, 0) C(1) (10, 10, 0) ∅ (10, 10, 1) C(1) 2 (8, 8, 1) (6, 6, 1) (4, 4, 1) (2, 2, 0) C(2) C(3) C(4) C(5) 4 H. ITO AND H. OHASHI In Table 2, k− is the invariant defined in Section 4, (4.2) and (r, l, δ) is the main invariant of the non-symplectic involution θ = g ◦ ε, Section 6. "Fixed curves" stands for the 1-dimensional components of the fixed locus of ι on Y . We also note that K− corresponds generically to the transcendental lattice of the covering K3 surface X. The Enriques surface of type [1] was constructed by Horikawa [Hor], and studied by Dolgachev [Dol] and Barth-Peters [BP]. Type [2] was found by Kondo [Kon] and constructed generally by Mukai [Muk1]. Type [3] was constructed by Lieber- man (cf. [MN]). The Enriques surfaces of type [1] -- [3] were studied by Mukai- Namikawa [MN] and Mukai [Muk1] as numerically trivial involutions. Moreover type [5] was studied by Mukai [Muk2] as numerically reflective involutions. In Section 2 we collect some basic definitions and notation of lattice theory. In Section 3 we show that Nikulin's classification theory [Nik4] is useful for our purpose and we introduce this theory in Section 4. In Section 5 we classify the lattice structures of involutions into 18 types of the tables in Theorem 1.1. We determine the lattices S±, K− and forms qS−H−, qS−gH− , k− here. In Section 6 we determine the other invariants, give the examples and complete Theorem 1.1. The authors wish to express their gratitude to Professor Kondo for suggestions to this problem and many stimulating conversations. We are grateful to Professor Tokunaga for the construction of the curve in Example No. [14]. 2. Preliminaries Our main tool is the lattice theory. Here we recall some definitions and notations. A lattice is a pair (L, ( , )), where L is a free Z-module of finite rank and ( , ) is a non-degenerate integral symmetric bilinear form on L. We abbreviate (L, ( , )) to L. We will denote by L(m) the lattice (L, m( , )) for a given lattice (L, ( , )) and m ∈ Q. L is called even if (x, x) ∈ 2Z for all x ∈ L. For a lattice L, there exists an injective homomorphism α : L → L∗ = Hom(L, Z) defined by x 7→ (x,−). L is called unimodular if α is bijective. Let U (resp. hni) denote the rank 2 (resp. rank 1) lattice given by the matrix (cid:18)0 1 1 0(cid:19) (resp.(cid:0)n(cid:1)). The root lattices Al, Dm, En are considered to be negative definite. A finite quadratic form is a triple (A, b, q), where A is a finite abelian group, b : A × A → Q/Z is a symmetric bilinear form, and q is a map q : A → Q/2Z satisfying the following conditions: (1) q(na) = n2q(a) for all n ∈ Z, a ∈ A. (2) q(a + a′) ≡ q(a) + q(a′) + 2b(a, a′) (mod 2) for all a, a′ ∈ A. A finite quadratic form is called non-degenerate if b is non-degenerate. An element x ∈ A is called characteristic if b(x, a) ≡ q(a) (mod 1) for all a ∈ A. We abbriviate (A, b, q) (resp. b(a, a′), q(a)) to (A, qA) or just qA (resp. aa′, a2). We denote by w (resp. z) the finite quadratic form on Z/2Z whose value is 1 (resp. 0). Note that w and z are degenerate. CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 5 A discriminant (quadratic) form for an even lattice L is a non-degenerate finite quadratic form (AL, bL, qL), where AL := L∗/L, bL(¯x, ¯y) = (x, y) (mod Z), and qL(¯x) = (x, x) (mod 2Z). We denote by u (resp. v, h 1 ni) the associated discriminant form of the lattice U (2) (resp. D4, hni). We often use the following discriminant forms: (L, qL) = (A1(2),h −1 4 i), (D4(2), v ⊕ v(4)), (D6(2), u2 ⊕ h 1 4i), (E8(2), u4), where un denotes n copies of u and v(4) denotes 4i2), (E7(2), u3 ⊕ h 1 ((Z/4Z)2,(cid:18) 1 2(cid:19)). 2 1 4 1 4 1 An embedding i : S → L of lattices is called primitive if L/i(S) is free. Let S be a sublattice of L. We define the sublattices S⊥ := {x ∈ L (x, y) = 0 ∀y ∈ S}, S∧ := S ⊗ Q ∩ L of L called the orthogonal complement to S and the primitive closure of S respec- tively. Let T be an orthogonal sublattice to S. We write ΓST := (S ⊕ T )∧/(S ⊕ T ). Two primitive embeddings i : S → L and i′ : S → L′ are called isomorphic if there exists f ∈ Isom(L, L′) such that f ◦ i = i′. Let M and N be even lattices, and let M → N be an embedding. Then N is called an overlattice of M if N/M is a finite abelian group. Let l(A) denote the minimal number of generators of an abelian group A. Note that (2.1) rank M ≧ l(AM ), l(AN ) ≧ l(AM ) − 2l(N/M ) for a lattice M and an overlattice N of M . A lattice M is called 2-elementary if AM = M ∗/M is a 2-elementary group (Z/2Z)a. Proposition 2.1 ([Nik2, Theorem 3.6.2]). The isomorphism class of an even hy- perbolic 2-elementary lattice M is determined by the invariants (r, l, δ), where r is the rank of M , l is the minimal number of generators of AM , and δ is the parity of qM , that is, δ =(0 1 if qM (x) = 0 ∀x ∈ AM , otherwise. Let L be a lattice and σ an involution of L. Write Lhσi = {x ∈ L σ(x) = x}, Lhσi = (Lhσi)⊥ = {x ∈ L σ(x) = −x}. Note that if L is unimodular, then Lhσi and Lhσi are 2-elementary lattices. Next proposition is the analogue of Witt's theorem. 6 H. ITO AND H. OHASHI Proposition 2.2 ([Nik4, Prop 1.9.2]). Let q be a finite quadratic form on a finite 2-elementary group Q whose kernel is zero, that is, {x ∈ Q x ⊥ Q and q(x) = 0} = {0}. Let θ : H1 → H2 be an isomorphism of two subgroups of Q that preserves the restric- tions qH1 and qH2 and that maps the elements of the kernel and the characteristic elements of the bilinear form q into the same sort of elements if they belong to H1. Then θ extends to an automorphism of q. 3. Involutions on Enriques surfaces Let Y be an Enriques surface and X its covering K3 surface with the covering involution ε. Consider an involution ι of Y . Then ι lifts to two involutions of X. One of them acts on H 0(X, Ω2) trivially, which we denote by g. Then another involution is g ◦ ε = ε ◦ g. The second cohomology group H 2(X, Z) is an even unimodular lattice with the signature (3, 19). Let S = {x ∈ H 2(X, Z) g∗(x) = −x}, where g∗ is the involution of H 2(X, Z) induced by g. It is known that S is isomorphic to E8(2) and this does not depend on g ([Mor], [Nik1]). Lemma 3.1. Let L be a unimodular lattice and S a 2-elementary lattice. The followings are equivalent. (1) There exists an involution α of L such that Lhαi ∼= S. (2) There exists a primitive embedding S → L. Proof. Assume (1). Since S = (Lhαi)⊥, it follows that the sublattice S is primitive in L. Assume (2). Let K = S⊥. Since S and K are 2-elementary lattices, there exists an involution α ∈ O(L) such that αK = 1 and αS = −1, by [Nik2, Corollary 1.5.2]. (cid:3) Since S is primitive in L, it follows that S = Lhαi. To classify ι, it suffices to classify the pair of involutions (g, ε). From Torelli type theorem ([PS]), this is equivalent to classifying the pair (g∗, ε∗). By Lemma 3.1, this is equivalent to classifying a primitive embedding of S = E8(2) into H 2(X, Z) and an action of ε∗ on S. 4. Involutions of a lattice with condition on a sublattice In this section, we introduce the theory of involutions of a lattice with condition on a sublattice. Definition 4.1 ([Nik4, Definition 1.1.1]). By a condition on an involution we understand a pair (S, θ), where S is a non-degenerate lattice and θ is an involution of S. Remark 4.2. In [Nik4], a condition on an involution is defined as a triple (S, θ, G), where S is a (possibly degenerate) lattice, θ is an involution of S, and G ⊂ O(S, θ) is a distinguished subgroup of the normalizer of θ in O(S). In this paper, we assume that G = {idS}. CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 7 Definition 4.3 ([Nik4, Definition 1.1.2]). By a unimodular involution with the condition (S, θ) we understand a triple (L, ϕ, i), where L is a unimodular lattice, ϕ is an involution of L, and i : S → L is a primitive embedding satisfying ϕ◦ i = i◦ θ. Two unimodular involutions (L, ϕ, i) and (L′, ϕ′, i′) with the condition (S, θ) are called isomorphic if there exists an isomorphism f : L → L′ with ϕ′ ◦ f = f ◦ ϕ and f ◦ i = i′. Let S± = {x ∈ S θ(x) = ±x}. We write pS± : S/(S+ ⊕ S−) → AS± for the projections and Γ± = pS±(S/(S+ ⊕ S−)) ⊂ AS± for the images of S/(S+ ⊕ S−). Note that S/(S+ ⊕ S−) is the graph of γ := pS− ◦ p−1 : Γ+ → Γ−, so we write Γγ = S/(S+ ⊕ S−). Theorem 4.4 ([Nik4, Theorem 1.3.1]). Any unimodular involution with the con- dition (S, θ) is determined by the list S+ (H±, qr, q, γr, K±, γK±), ± ∩ 1 (4.1) where H± are subgroups with Γ± ⊂ H± ⊂ (S∗ 2 S±)/S±, qr is a finite qua- dratic form on the 2-elementary group (H+ ⊕ H−)/Γγ with qrH± = ±qS±H± , q is the isomorphism class of a non-degenerate 2-elementary finite quadratic form, γr : qr → q is an embedding of forms, K± are even lattices, and γK± : qK± → k± are isomorphisms of forms. Here k± are defined by (4.2) k± = ((−qS± ⊕ ±q)Γ⊥ γrH±)/ΓγrH± , ±, q′ r, K ′ r, q′, γ′ where ΓγrH± are the graphs of the embeddings H± → q induced by γr. Two lists (4.1) and (H ′ ) determine isomorphic unimodular r, q = q′, and ±, qr = q′ involutions with the condition (S, θ) if and only if H± = H ′ ±) such that ξ ◦ γr = γ′ there exist isomorphisms ξ ∈ O(q) and ψ± ∈ Isom(K±, K ′ r and (id, ξ)k± ◦ γK± = γ′ ± ◦ ψ±, where (id, ξ)k± are isomorphisms between k± and k′ ± induced by id ∈ O(qS±) and ξ, and ψ± are isomorphisms between qK± and qK′ induced by ψ±. ±, γ′ K′ ± K′ ± Proof. We prove only the assertion about the equivalence of the lists (4.1), which is omitted in [Nik4]. Let (L, ϕ, i) and (L′, ϕ′, i′) be the unimodular involutions with the condition (S, θ) determined by the lists (4.1) and (H ′ ) respectively. ±, γ′ K′ ± r, q′, γ′ r, K ′ ±, q′ Assume that two lists determine isomorphic unimodular involutions. There exists f ∈ Isom(L, L′) such that f ◦ i = i′ and ϕ′ ◦ f = f ◦ ϕ. It follows from ϕ′ ◦ f = f ◦ ϕ that f induces f± := fL± ∈ Isom(L±, L′ ±, where i± : S± → L± ± are primitive embeddings induced by i and i′ respectively. Since and i′ f induces fL+⊕S− = (f+, id) ∈ Isom(L+, L′ +) × O(S−), so does an isomorphism between (L+ ⊕ S−)∧ and (L′ + ⊕ S−)∧. Hence we have ±) with f± ◦ i± = i′ ± : S± → L′ H− = pS−(ΓL+S−) = pS−(ΓL′ +S−) = H ′ − and f+ ◦ γrH− = γ′ f+. rH′ − , where f+ is an isomorphism between q and q′ induced by 8 H. ITO AND H. OHASHI Similarly f induces an isomorphism between (L−⊕S+)∧ and (L′ −⊕S+)∧. Hence . From f−◦γL+L− = L and (L′ + ⊕ S−)∧ = − ◦ f+, we have f+ ◦ γr = γ′ L′ , there exists ψ− with the condition, by [Nik2, Corollary 1.5.2]. Similarly, we see that H+ = H ′ γL′ (K ′ we have ψ+ with the condition. It is clear that q = q′ and qr = q′ r. rH′ r. Since (L+ ⊕ S−)∧ = (K−)⊥ + and f−◦(γL+L−◦γrH+ ) = γL′ +L′ −)⊥ −◦γ′ +L′ + ±, qr = q′ We turn to the contrary. Assume that H± = H ′ r, q = q′ and there exist ξ = ξ+ ∈ O(q) and ψ± ∈ Isom(K±, K ′ ±) with the conditions. Note that invariants (H±, γrH±, K±, γK±) determine primitive embeddings i± : S± → L± with orthogonal complements K± by [Nik2, Proposition 1.15.1], where L± are the lattices with discriminant forms ±q respectively. Let T1 (resp. T2) be any lattice which is the unique in its genus and furthermore O(T1) → O(qT1 ) (resp. O(T2) → O(qT2)) is surjective and qT1 = q (resp. qT2 = −q). From q = q′ and K− ∼= K ′ − (resp. K+ ∼= K ′ +) are obtained as orthogonal complements of a primitive embedding T1 → L1 (resp. T2 → L2), where L1 (resp. L2) is a unimodular lattice with +), we see that L− and L′ − (resp. L+ and L′ Sign L1 = Sign L− + Sign T1 = Sign L′ − + Sign T1 (resp. Sign L2 = Sign L+ + Sign T2 = Sign L′ + + Sign T2). Moreover T1 is obtained as an orthogonal complement of a primitive embedding T2 → L3, where L3 is a unimodular lattice with Sign L3 = Sign T1 + Sign T2. Hence there exists ξ− ∈ O(−q) such that ξ− ◦ γT1T2 = γT1T2 ◦ ξ+. f1 = ξ+. By ξ+ ◦ γrH− = γ′ O(S−) extends to an isomorphism Since O(T1) → O(qT1 ) = O(q) is surjective, there exists f1 ∈ O(T1) such that −, it follows that (f1, id) ∈ O(T1) × and H− = H ′ rH′ − α1 : (T1 ⊕ S−)∧ → (T1 ⊕ S−)∧. L1 , and the latter is equal to L1 . From the condition of ψ−, it follows that (α1, ψ−) extends to an automor- Note that the former (T1 ⊕ S−)∧ is equal to (K−)⊥ −)⊥ (K ′ phism Similarly there exists an automorphism β2 : L2 → L2 such that β2T2 ∈ O(−q), β2S+ = id and β2K+ = ψ+. Therefore we have the following commutative diagram: β1 : L1 → L1. AL− AT1 AT2 AL+ β1L− ξ+ ξ− β2L+ AL′ − / AT1 / AT2 / AL′ + Hence (β2L+, β1L−) extends to an isomorphism β : L → L′ with β ◦ i = i′ and β ◦ ϕ = ϕ′ ◦ β, which is the desired isomorphism. (cid:3) Remark 4.5. In the proof of Theorem 4.4, we see that β2L+ = (ψ+, id)Γ⊥ K+S+ /ΓK+S+, β1L− = (ψ−, id)Γ⊥ K−S−/ΓK−S−. / / / / / / / / / CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 9 Moreover, if L± is indefinite, then we can take T1 and T2 as L+ and L− respectively. Hence we see that (4.3) ξ+ = (ψ+, id)Γ⊥ K+S+/ΓK+S+ , ξ− = (ψ−, id)Γ⊥ K−S−/ΓK−S−. 5. Classification The construction of the list (4.1) from the unimodular involution with condition is as follows (see [Nik4] for more details): Let (L, ϕ, i) be a unimodular involution with the condition (S, θ). We write L± = {x ∈ L ϕ(x) = ±x}. Define q := qL+. The primitive embedding i : S → L defines primitive embeddings i± : S± → L±. Hence we define 2-elementary groups ± ∩ 1 H± := pS±(ΓL∓S±) ⊂ (S∗ 2 S±)/S±. Note that pS± are injective, since i± are primitive. We can also say that ΓL−S+ (resp. ΓL+S−) is the graph of injective homomorphism γH+ : H+ → AL− (resp. γH− : H− → AL+ ). Note that the notation of γH± is slightly different from that of [Nik4]. We define the embedding of forms γr and the quadratic form qr on (H+ ⊕ H−)/Γγ as L+L− ◦ γH+ , γH− ) : H+ ⊕ H−/Γγ → q, γr := (γ−1 qr := q ◦ γr, where γL+L− is an isomorphism between AL+ and AL−. The even lattices K± are defined by K± := (S±)⊥ L±. The quadratic forms −k± in (4.2) are equal to discrimi- nant forms of (L∓⊕ S±)∧. Hence the sign reversing isometies give γK± : qK± → k±. From now on, we regard L = H 2(X, Z), ϕ = ε∗, and S = {x ∈ L g∗(x) = −x} ∼= E8(2). It is known that L+ ∼= U (2) ⊕ E8(2), L− ∼= U ⊕ U (2) ⊕ E8(2) and these do not depend on ε ([BP]). Lemma 5.1. Suppose that S = E8(2) and θ is an involution of S. Then the isomorphism class of (S+, S−) is one of the following: (S+( 1 2 ), S−( 1 2 )) = (E8,{0}), (E7, A1), (D6, A2 1, D4 ⊕ A1), (A2 1), (A3 1), (D4 ⊕ A1, A3 1), (D4, D4), 1, D6), (A1, E7), ({0}, E8). (A4 1, A4 Proof. It suffices to prove the lemma for S( 1 2 ) = E8. Since θ is an involution, it follows that S± are even 2-elementary lattices. We can assume that the rank of S+ is at most 4. By [Nik2, Theorem 3.6.2], invariants (r, l, δ) of S+ is one of the following: (0, 0, 0), (1, 1, 1), (2, 2, 1), (3, 3, 1), (4, 4, 1), (4, 2, 0). We see that {0}, A1, A2 1 and D4 have above invariants respectively, and these lattices have one class in their genus (cf. [Nik2, Remark 1.14.6]). Hence S+ (cid:3) is one of them. S− is obtained as orthogonal complement to S+. 1, A3 1, A4 10 H. ITO AND H. OHASHI From this lemma, we calculate the list (4.1) for each (S+, S−). (cid:3) 2 ) is an even negative definite lattice of rank less than 8 and 2 ) ∼= U ⊕ E8 is a unimodular lattice of signature (1, 9), the lemma follows from Lemma 5.2. Suppose that S+ is one of them in Lemma 5.1. Then there exists a unique primitive embedding S+ → L+. Proof. Since S+( 1 L+( 1 [Nik2, Theorem 1.14.4]. Corollary 5.3. We have K+ ∼= U (2) ⊕ S− in all cases. In particular, ΓS+S− ∼= ΓK+S+ . Proof. Recall that L+ ∼= U (2) ⊕ E8(2) and S ∼= E8(2). By Lemma 5.2, a primitive embedding S+ → L+ is unique. Hence K+ = (S+)⊥ L+ is uniquely determined as U (2) ⊕ S−. Therefore we see that ΓK+S+ = L+/(K+ ⊕ S+) ∼= (U (2) ⊕ S)/(U (2) ⊕ S− ⊕ S+) ∼= S/(S+ ⊕ S−) = ΓS+S−. Lemma 5.4. On H±, we have the following: (1) Γ+ ⊂ H+ = 1 2 S+/S+, Γ− ⊂ H− ⊂ 1 (2) qS±H± ≡ 0 (mod 1). (3) rank S− − 1 ≦ rank H− ≦ rank S−. (4) qS+Γ+ (resp. qS−Γ−) is a direct summand of qS+H+ (resp. qS−H− ). 2 S−/S−. (cid:3) 2 ) are even lattices, we have H± ⊂ 1 Proof. Since S±( 1 L+( 1 that x + y ∈ L, which implies x + y ∈ (S+ ⊕ L−)∧. Therefore H+ = 1 qr ≡ 0 (mod 1). Hence qS±H± = qrH± ≡ 0 (mod 1). 2 ) ∼= U ⊕ E8, we have x ∈ L∗ Since γr : qr → q is an embedding and q = u5 ≡ 0 (mod 1), qr also satisfies By K− = (S−)⊥ L−, we see that rank K− = rank L− − rank S− = 12 − rank S−. +. Since L is unimodular, there exists y ∈ L∗ 2 S+/S+. 2 S±/S±. Let x ∈ 1 2 S+. From − such From (2.1), we see that l(AK−) = l(A(L+⊕S−)∧) ≧ l(AL+⊕S−) − 2l(ΓL+S−) = 10 + l(AS−) − 2l(ΓL+S−). Obviously l(AS−) = rank S−. The primitivity of L+ in (L+⊕S−)∧ gives l(ΓL+S−) = l(H−) = rank H−. Therefore rank K− ≧ l(AK−) yields 12 − rank S− ≧ 10 + rank S− − 2 rank H−. Hence we have (3). In our S+, we can write AS+ = (Z/2Z)a ⊕ (Z/4Z)b and qS+ = q2 ⊕ q4 where q2 (resp. q4) is a finite quadratic form on (Z/2Z)a (resp. (Z/4Z)b). Since Γ+ = 2AS+ = {2x x ∈ AS+}, we have qS+Γ+ = 2q4 where 2q4 denotes a finite quadratic form whose generators are twice the size of those of q4. Since qS+ 1 2 S+/S+ = q2⊕2q4, 2 S+/S+ . Hence qS+Γ+ is also that we see that qS+Γ+ is a direct summand of qS+ 1 of qS+H+ . The same proof works for qS−Γ−. (cid:3) Lemma 5.5. (1) In cases S−( 1 2 ) = E8, E7, D6, D4 ⊕ A1, we have Γ+ = H+ = 1, A1,{0}, we have Γ− = H− = 1 2 S−/S−. 1 2 S+/S+. (2) In cases S−( 1 2 ) = A4 1, A3 1, A2 CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 11 Proof. We give the proof only for the case S−( 1 the reader. In case S−( 1 2 ) = E7, we have S+( 1 2 ) = E7; the other cases are left to 2 ) = A1. Hence we see that Γ+ = pS+ (S/(S+ ⊕ S−)) = pS+(E8(2)/(A1(2) ⊕ E7(2))) ∼= pS+ (E8/(A1 ⊕ E7)) = AA1 ∼= Z/2Z. At the same time, we see that 1 2 S+/S+ = 1 2 A1(2)/A1(2) ∼= Z/2Z. The lemma follows from Lemma 5.4 (1). (cid:3) We consider the behavior of γH± : H± → AL∓. Note that K±S± ∩ AK±)/(ΓK±S± ∩ AK± ) ⊂ Γ⊥ K±S± ∩ AK± = (Γ⊥ Γ⊥ K±S±/ΓK±S± = AL±. Definition 5.6. Let ]AK+ := Γ⊥ AL−. The subgroupgH− of H− andgH+ of H+ are defined by We see that (gH−, γH−gH− ) determine (K+ ⊕ S−)∧ and (S+ ⊕ K−)∧ respectively, since (H∓, γH∓ ) determine (L± ⊕ S∓)∧. It follows from Corol- lary 5.3 that Γ+ = pS+(ΓK+S+ ). Therefore we have K+S+ ∩ AK+ ⊂ AL+ and ]AK− := Γ⊥ (]AK+ ), gH+ := γ−1 gH− := γ−1 K−S− ∩ AK− ⊂ (]AK− ). H− H+ ) and (gH+, γH+gH+ Γ− ⊂gH− ⊂ H−. (5.1) From Theorem 4.4, if two unimodular involutions with the condition (S, θ) deter- ) respectively are isomorphic, ±) with the conditions. As mined by the lists (4.1) and (H ′ then there exist ξ± ∈ O(±q) and ψ± ∈ Isom(K±, K ′ stated in Remark 4.5, we have (4.3). It follows that ±, γ′ K′ ± r, q′, γ′ r, K ′ ±, q′ gH− =gH ′ − (resp.gH+ =gH ′ K+S+/ΓK+S+ (resp. (ψ−, id)Γ⊥ +), (resp. ]AK− and ]AK′ − since (ψ+, id)Γ⊥ phism between ]AK+ and ]AK′ equivalence relation: + K−S−/ΓK−S−) induces an isomor- ). Hence we define the following γH∓ ∼ γH′ ∓ def⇐⇒ there exists ξ± ∈ O(±q) such that ξ± ◦ γH∓ = γH′ ∓ The existence condition of ξ± follows from Proposition 2.2. Thus we have a one- ∓. andgH∓ =gH ′ Lemma 5.7. We have an equality to-one correspondence between {γH∓}/∼ and {gH∓}. H−/gH− = H+/gH+. Proof. It is easy to check that Hence the primitivity shows the lemma. (cid:3) ΓL+S−/ΓK+S− = ΓS+L−/ΓS+K−. 12 H. ITO AND H. OHASHI 2 S−. +, ν ∈ S∗ +, µ ∈ S∗ 2 S+, ν ∈ 1 −. If λ + µ + ν ∈ L, then λ ∈ 1 Lemma 5.8. Let λ ∈ K ∗ µ ∈ 1 Proof. Let T be a primitive sublattice of L spanned by K+ ⊕ S−, that is, T = (K+⊕ S−)∧. Since T is also the fixed part of the action of the involution (g◦ ε)∗ on L, it follows that L/(T ⊕T ⊥) is a 2-elementary group. Hence we have 2(λ+ν)+2µ ∈ T ⊕ T ⊥, in particular 2µ ∈ T ⊥ ⊂ L. Since S+ is a primitive sublattice of L, we see that 2µ ∈ S∗ 2 S+. The rest of the proof is (cid:3) left to the reader. L = S+. We thus get µ ∈ 1 + ∩ L ⊂ (S+)∧ 2 K+, From this lemma, we see that γH− (H−) ⊂ ( 1 (5.2) 2 K+/K+ ⊕ 1 2 S+/S+)/ΓK+S+ . 1 1, A3 2 ) = A4 Proof. In cases S−( 1 1, A2 1, A1, {0}, it follows from (5.1) and Lemma 5.5 2 ) = E8, E7, D6, D4 ⊕ A1, we have γH− (Γ−) ≡ It (cid:3) Lemma 5.9. We have gH± = H± unless S± = D4(2). that gH− = H−. In cases S−( 1 2 S+/S+ (mod ΓK+S+) by Lemma 5.5. From (5.2), we see that gH− = H−. follows from Lemma 5.7 thatgH+ = H+. the case S± = D4(2). In case S± = D4(2), we have to calculate (H−,gH−, K+, K−). Proof. By Lemmas 5.4 and 5.9, we calculate (H−, K+, K−) for each (S+, S−) except Theorem 5.10. The lists (4.1) are classified as Table 1 and Table 2 in Theorem 1.1. We first calculate H−. 2 S−/S− = u4. By Lemma 5.4 (3), rank H− = 2 S−/S−. For rank H− = 7, we have In case S− = E7(2), we see that qS− 1 8 or 7. For rank H− = 8, we have H− = 1 qS−H− = u3 ⊕ w or u3 ⊕ z by Lemma 5.4 (2). 2 S−/S− = u3 ⊕ w and qS±Γ± = w. By Lemma 5.4 (3), rank H− = 7 or 6. For rank H− = 7, we have H− = 1 2 S−/S−. For rank H− = 6, we have qS−H− = u2 ⊕ w2 by Lemma 5.4 (2) and (4) (note that we have w ⊕ z = w2). The same proof works for the cases S−( 1 2 ) = D6, D4 ⊕ A1. So we omit it. In case S− = E8(2), we see that qS− 1 In cases S−( 1 2 ) = A4 1, A3 1, A2 1, A1, {0}, we see that qS−H− = qS− 1 2 S−/S− by Lemma 5.5. We next deal with the case S± = D4(2). We see that qS− 1 2 S−/S− = v ⊕ z2 and qS±Γ± = z2. By Lemma 5.4 (3), rank H− = 4 or 3. For rank H− = 3, we have qS−H− = w ⊕ z2 by Lemma 5.4 (2) and (4). From (5.1), we have qS−gH− = w ⊕ z2 or z2. For rank H− = 4, we have H− = 1 2 S−/S−. From (5.1), a candidate = z2 is for qS−gH− impossible. is one of v ⊕ z2, w ⊕ z2 and z2. Here we claim that qS−gH− Suppose that qS−gH− GH− is a subgroup of H− whose quadratic form is v. Moreover let = z2. This yieldsgH− = Γ−. Let H− =gH− ⊕ GH− , where 1 2 K+/K+ = GK+ ⊕ pK+(ΓK+S+ ), 2 S+/S+ = GS+ ⊕ pS+(ΓK+S+ ), 1 CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 13 where GK+ (resp. GS+ ) is a subgroup of 1 2 K+/K+ (resp. 1 2 S+/S+) whose quadratic form is u ⊕ v (resp. v). SincegH− = Γ−, we have γH−(gH−) ≡ pS+ (ΓK+S+) ≡ pK+ (ΓK+S+ ) It follows from (5.2) that (mod ΓK+S+ ). γH−(GH− ) ⊂ GK+ ⊕ GS+ . Since GH− gives difference between ΓK+S− and ΓL+S−, a non-zero element of γH−(GH− ) is a sum of non-zero elements of GK+ and GS+ . This contradicts the fact that the quadratic form of GH− is v. Now we have Table 1. We proceed to calculate K±. By Lemma 5.2, K+( 1 2 ) is uniquely determined with the signature (1, 9 − rank S+) and the discriminant form −qS+( 1 2 ). By calculating (4.2) we have k−. From [Nik2, Theorem 1.14.2 and Corollary 1.9.4], K− is uniquely determined with the signature (2, 10 − rank S−) and the discriminant form k−. (cid:3) Therefore we have k− and K− in Table 2. 6. Examples In this section we construct examples of involutions on Enriques surfaces. In particular we show that all cases in Theorem 1.1 actually occur. We denote by ι an involution on an Enriques surface Y . The K3-cover is denoted by X with the covering transformation ε. The symplectic lift of ι to X is denoted by g and the other non-symplectic one is θ = g ◦ ε = ε ◦ g. We first note that the fixed locus of θ, can be computed from Theorem 5.10 via the following theorem. X θ = {x ∈ X θ(x) = x}, Theorem 6.1 ([Nik3, Theorem 4.2.2]). Let θ be a non-symplectic involution of X and let T = H 2(X, Z)hθ∗i. Since T is 2-elementary, the lattice T is determined by invariants (r, l, δ) by Proposition 2.1. Then, the fixed locus X θ has the following form. Here we denote by C(g) a non-singular curve of genus g and by Ei a non-singular rational curve. Proposition 6.2. The invariant (r, l, δ) for each case is as in Table 2. that Proof. We see that T = H 2(X, Z)hθ∗i is exactly the sublattice (K+ ⊕ S−)∧ = ((K+ ⊕ S−) ⊗ Q) ∩ L of L = H 2(X, Z). Therefore we get r = rank K+ + rank S−. Since T is 2-elementary, we have det T = 2l. By pS−(ΓK+S−) = gH−, it follows gH− = ΓK+S− =s det(K+ ⊕ S−) det(K+ ⊕ S−)∧ =r det(K+ ⊕ S−) 2l . C(g) +Pk C(1) 1 + C(1) ∅ 2 X θ = i=1 Ei where g = 22−r−l 2 and k = r−l 2 if r = 10, l = 8, δ = 0 . if r = 10, l = 10, δ = 0 14 H. ITO AND H. OHASHI From this equation we get l. Next we compute the invariant δ. In cases No. [4], [5], [8], [9], [15] -- [17], the invariants (r, l) already determine δ uniquely by the existence condition for the 2-elementary hyperbolic lattices, see [Nik3]. In cases No. [1] -- [3], [18], we have that the parity of K+ ⊕ S− is zero, hence the overlattice T has parity zero, too. In No. [6], we see from Table 1 that the length of gH− is 6, which equals the rank of S−. By straightforward computations, we see that the discriminant group of T has elements of non-integer square, that is, we have δ = 1 in this case. In No. [7], we see that T ⊥ has rank 8, signature (2, 6) and length 8. Therefore T ⊥( 1 2 ) is an integral unimodular lattice, which must be odd by the signature reason. We get T ⊥ ≃ A1(−1)2 ⊕ A6 The remaining five cases where rank S+ = rank S− = 4 are treated by the next 1 and so δ = 1. two lemmas. Lemma 6.3. Assume that S± = A1(2)4 and (r, l) = (10, 10). Then T = U (2)⊕ A8 and δ = 1. Proof. Let K+ = U (2) ⊕ A1(2)4 = U (2) ⊕ he1i ⊕ ··· ⊕ he4i where ei are generators of A1(2) respectively. Similarly let 1 1i ⊕ ··· ⊕ he′ 4i, 1i ⊕ ··· ⊕ he′′ 4i. S+ = A1(2)4 = he′ S− = A1(2)4 = he′′ 1 /2i⊕···⊕he′′ j + e′′ By pS−(ΓS+S−) = Γ− = he′′ 4 /2i, elements of norm 1 (mod 2) in Γ− is i /2 or (e′′ of the form either e′′ i/2 to either e′′ j /2 or (e′′ k + e′′ l )/2. In the former case, it contradicts the fact that S = E8(2) does not contain (−2)-vector. Similarly the patching pS+ (ΓK+S+ ) → pK+ (ΓK+S+ ) maps e′ i/2 to (ej + ek + el)/2. Hence ΓK+S− contains an element of the form of l )/2. Hence γ : Γ+ → Γ− maps e′ j + e′′ k + e′′ ei + ej + ek + e′′ l + e′′ m + e′′ n 2 . This element has norm (−6). Assumption (r, l) = (10, 10) yields that T ( 1 2 ) = U ⊕ E8 or U ⊕ h−1i8. Since U ⊕ E8 does not contain (−3)-vector, we conclude T = U (2) ⊕ A8 (cid:3) 1. Lemma 6.4. Assume that S± = D4(2). Then the parity δ of T = (K+ ⊕ S−)∧ is equal to 0. Proof. By Corollary 5.3, we see that K+ = U (2) ⊕ D4(2). Let qK+ = u ⊕ v ⊕ v(4) = u⊕he1, f1i⊕hg1, h1i where he1, f1i and hg1, h1i are generators of v and v(4) respectively. Similarly, let qS+ = v ⊕ v(4) = he2, f2i ⊕ hg2, h2i, qS− = v ⊕ v(4) = he3, f3i ⊕ hg3, h3i. Recall that L+ = U (2) ⊕ E8(2) and S = E8(2). We see that ΓK+S+ = h2g1 + 2g2, 2h1 + 2h2i and ΓS+S− = h2g2 + 2g3, 2h2 + 2h3i. Hence ΓK+S− contains h2g1 + 2g3, 2h1 + 2h3i. This shows that T is an overlattice of U (2)⊕ E8(2). Therefore the (cid:3) parity of T is equal to 0. CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES This completes the proofs for all cases. 15 (cid:3) 6.1. Horikawa constructions. The general construction is as follows. Proposition 6.5 ([BHPV, V. 23]). Let ψ be an involution on P1 × P1 given by ψ : (u, v) 7→ (−u,−v) where u and v are inhomogeneous coordinates of P1 respec- tively. Let B be a curve on P1 × P1 whose bidegree is (4, 4) with at worst simple singularities and preserved under ψ. Assume that B does not pass through any of fixed points of ψ. Then the minimal resolution X of the double cover of P1 × P1 whose branch locus is B is a K3 surface. Moreover, ψ lifts to two involutions of X. One of them is a fixed point free involution ε. In particular, Y = X/ε is an Enriques surface. In this construction, the other lift of ψ gives a symplectic involution g on X and induces an involution ι on Y (namely the construction always associates an involution on Y ). The covering involution θ of X/P1× P1 is the same as ε◦ g, which is a non-symplectic involution of X. In what follows, we exhibit many choices of branch B so that the resulting ι covers all involutions in Theorem 1.1 except for No. [13]. We remark that, the condition for B to have the expected number of components, types of singularities and not to pass through the fixed points of ψ is Zariski open, so that we will always assume that the coefficients (parameters) of the exhibited equation of B are general enough to satisfy these conditions. Example No. [1]. This example was constructed by Horikawa [Hor], and studied by Dolgachev [Dol] and Barth-Peters [BP]. Here we give another construction given by Mukai-Namikawa [MN]. Consider the following curves on P1 × P1 (Figure 1); X± : u = ±1, Y± : v = ±1, E : u2v2 − 1 + a1(u2 − 1) + a2(v2 − 1) = 0 (ai ∈ C). Figure 1. Figure 2. Blow up P1 × P1 at 4 intersection points of X±, Y± and E. Let F±,± be the exceptional curves over (±1,±1) respectively. Blow up again at 12 intersection points of F±,± and the strict transforms of X±, Y± and E. Let R be the blown ±,± and E′ the strict transforms of X±, Y±, up surface. We denote by X ′ ±, Y ′ ±, F ′ 16 H. ITO AND H. OHASHI ±, Y ′ ± and F ′ (−1)-curves. Let B′ =P(X ′ F±,± and E respectively. The configuration of curves in R is given in Figure 2. Note that X ′ ±,± are all (−4)-curves, and other rational curves are all ±,±) + E′. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of one elliptic curve and 8 rational curves, we see (r, l) = (18, 2), by Theorem 6.1. This is enough to conclude that this example belongs to No. [1] by Table 2. ± + Y ′ ± + F ′ Example No. [2]. This example was found by Kondo, and overlooked in [MN] (cf. [Muk1]). Consider the following curves on P1 × P1 (Figure 3); X± : u = ±1, Y± : v = ±1, C± : uv − 1 + a1(±u − 1) + a2(±v − 1) = 0 (ai ∈ C). Figure 3. Figure 4. Figure 5. Blow up P1 × P1 at 10 intersection points of X±, Y± and C±. Let F+ and F− be the exceptional curves over (1, 1) and (−1,−1) respectively. Blow up again at 6 intersection points of F± and the strict transforms of X±, Y± and C±. Let R be the blown up surface. We denote by X ′ ± the strict transforms of X±, Y±, C± and F± respectively. The configuration of curves in R is given in Figure 4. Note that X ′ ± are all (−4)-curves, and the others are all ± + F ′ ±). The K3 surface X is the double ± and F ′ ± + C′ ± and F ′ ± + Y ′ ±, Y ′ ±, C′ ±, Y ′ ±, C′ (−1)-curves. Let B′ =P(X ′ CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 17 cover of R whose branch locus is B′. Since X θ = B′ consists of 8 rational curves, we see (r, l) = (18, 4), by Theorem 6.1. Note that the configuration of curves in X is given in the same as Figure 4. We notice that there exists E7 ⊕ A1 diagram in Figure 4 (continuous lines in Figure 5). Let ei (i = 1, . . . , 8) denote the cohomology class of these curves respectively. The image of this diagram by ε is given by dashed lines in Figure 5. Let M be the lattice generated by ei − ε∗(ei) (i = 1, . . . , 8). We see that M ∼= E7(2) ⊕ A1(2) and M ⊂ S−. For (ei − ε∗(ei))/2 ∈ 1 2 M , there exists (ei + ε∗(ei))/2 ∈ L∗ + such that ei − ε∗(ei) + 2 ei + ε∗(ei) 2 = ei ∈ L. It follows that By calculation, we have qE8(2) 1 example of No. [2]. 2 (E7(2)⊕A1(2))/E8(2) = u3 ⊕ w. Therefore this is the 1 2 M/S− ⊂ H−. Example No. [3]. This example was constructed by Lieberman. Consider the following curves on P1 × P1 (Figure 6); X1± : u = ±1, Y1± : v = ±1, X2± : u = ±a1, Y2± : v = ±a2 (ai ∈ C). Figure 6. Figure 7. Figure 8. 18 H. ITO AND H. OHASHI 2±, Y ′ 2±, Y ′ 1±, X ′ 1±, X ′ 2± + Y ′ 1± + Y ′ 1± and Y ′ 1± and Y ′ 1± + X ′ are all (−1)-curves. Let B′ =P(X ′ Blow up P1 × P1 at 16 intersection points of X1±, X2±, Y1± and Y2±. Let R be the blown up surface. We denote by X ′ 2± the strict transforms of X1±, X2±, Y1± and Y2± respectively. The configuration of curves in R is given in Figure 7. Note that X ′ 2± are all (−4)-curves, and the others 2±). The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of 8 rational curves, we see (r, l) = (18, 4), by Theorem 6.1. Note that the configuration of curves in X is given in the same as Figure 7. We notice that there exists D8 diagram in Figure 7 (continuous lines in Figure 8). Let ei (i = 1, . . . , 8) denote the cohomology class of these curves respectively. The image of this diagram by ε is given by dashed lines in Figure 8. Let M be the lattice generated by ei − ε∗(ei) (i = 1, . . . , 8). We see that M ∼= D8(2) and M ⊂ S−. Similarly to the Example No. [2], we have 2 (D8(2))/E8(2) = u3 ⊕ z. Therefore 2 M/S− ⊂ H−. By calculation, we have qE8(2) 1 this is the example of No. [3]. Example No. [4]. Consider the following curves on P1 × P1 (Figure 9); 1 X± : u = ±1, Y± : v = ±1, E : u2v2 − 1 + a1(u2 − 1) + a2(v2 − 1) + a3(uv − 1) = 0 (ai ∈ C). Figure 9. Figure 10. Blow up P1 × P1 at 8 intersection points of X±, Y± and E. Let F+ and F− be the exceptional curves over (1, 1) and (−1,−1) respectively. Blow up again at 6 intersection points of F± and the strict transforms of X±, Y± and E. Let R be the blown up surface. We denote by X ′ ± and E′ the strict transforms of X±, Y±, F± and E respectively. The configuration of curves in R is given in Figure 10. Note that X ′ ± are all (−4)-curves, and other rational curves are all ±) + E′. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of one elliptic curve and 6 rational curves, we see (r, l) = (16, 4), by Theorem 6.1. Therefore this is the example of No. [4]. (−1)-curves. Let B′ =P(X ′ ± and F ′ ± + Y ′ ± + F ′ ±, Y ′ ±, Y ′ ±, F ′ Example No. [5]. This example was studied by Mukai [Muk2] as the example of numerically reflective involution. CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 19 Consider the following curves on P1 × P1 (Figure 11); X± : u = ±1, Y± : v = ±1, C± : uv ± a1u ± a2v + a3 = 0 (ai ∈ C). Figure 11. Figure 12. ±, Y ′ ± and C′ ±, Y ′ Blow up P1 × P1 at 14 intersection points of X±, Y± and C±. Let R be the blown up surface. We denote by X ′ ± the strict transforms of X±, Y± and C± respectively. The configuration of curves in R is given in Figure 12. Note that X ′ ± are all (−4)-curves and the others are all (−1)-curves. Let ±). The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of 6 rational curves, we see (r, l) = (16, 6), by Theorem 6.1. Therefore this is the example of No. [5]. Example No. [6]. Consider the following curves on P1 × P1 (Figure 13); B′ =P(X ′ ± and C′ ± + Y ′ ± + C′ X± : u = ±1, Y± : v = ±1, E : u2v2 + a1u2 + a2v2 + a3uv + a4 = 0 (ai ∈ C). Figure 13. Figure 14. Blow up P1 × P1 at 12 intersection points of X±, Y± and E. Let R be the blown up surface. We denote by X ′ ± and E′ the strict transforms of X±, Y± and E respectively. The configuration of curves in R is given in Figure 14. Note that X ′ ± are all (−4)-curves and other rational curves are all (−1)-curves. Let ±, Y ′ ±, Y ′ 20 H. ITO AND H. OHASHI B′ =P(X ′ ± + Y ′ ±) + E′. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of one elliptic curve and 4 rational curves, we see (r, l) = (14, 6), by Theorem 6.1. Therefore this is the example of No. [6]. Example No. [7]. Consider the following curves on P1 × P1 (Figure 15); Y± : v = ±1, C± : u2v ± uv ± a1u2 + a2u + a3v ± a4 = 0 (ai ∈ C). Figure 15. Figure 16. surface. We denote by Y ′ The configuration of curves in R is given in Figure 16. Note that Y ′ ± and C′ all (−4)-curves and the others are all (−1)-curves. Let B′ = P(Y ′ Blow up P1 × P1 at 12 intersection points of Y± and C±. Let R be the blown up ± the strict transforms of Y± and C± respectively. ± are ±). The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of 4 rational curves, we see (r, l) = (14, 8), by Theorem 6.1. Therefore this is the example of No. [7]. Example No. [8]. Consider the following curves on P1 × P1 (Figure 17); Y± : v = ±1, E : v2(u4 + a1u2 + a2) + 2a3uv(u2 − a4) + a5(u2 − a4)2 = 0 Note that E has 2 nodes at (u, v) = (±√a4, 0). ± and C′ ± + C′ (ai ∈ C). Figure 17. Figure 18. Blow up P1 × P1 at 8 intersection points of Y± and E, and at 2 nodes of E. ± and E′ the strict transforms Let R be the blown up surface. We denote by Y ′ CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 21 of Y± and E respectively. The configuration of curves in R is given in Figure 18. Note that Y ′ ± are (−4)-curves and other rational curves are all (−1)-curves. Let B′ = Y ′ + + Y ′ − + E′. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of one elliptic curve and 2 rational curves, we see (r, l) = (12, 8), by Theorem 6.1. Therefore this is the example of No. [8]. Example No. [9]. Consider the following curves on P1 × P1 (Figure 19); C± : v2(u2 ± a1u + a2) ± 2a3v(u ∓ a4)2 + a5(u ∓ a4)2 = 0 (ai ∈ C). Note that C+ and C− have a node at (u, v) = (a4, 0) and (−a4, 0) respectively. Figure 19. Figure 20. Blow up P1 × P1 at 8 intersection points of C±, and at 2 nodes of C±. Let R be the blown up surface. We denote by C′ ± the strict transforms of C± respectively. The configuration of curves in R is given in Figure 20. Note that C′ ± are (−4)- curves and the others are all (−1)-curves. Let B′ = C′ −. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of 2 rational curves, we see (r, l) = (12, 10), by Theorem 6.1. Therefore this is the example of No. [9]. Example No. [10]. Consider the following curves on P1 × P1 (Figure 21); Y± : v = ±1, C : v2(u4 + u2 + a1) + vu(a2u2 + a3) + a4u4 + a5u2 + a6 = 0 (ai ∈ C). + + C′ Figure 21. Figure 22. 22 H. ITO AND H. OHASHI Blow up P1 × P1 at 8 intersection points of Y± and C. Let R be the blown up surface. We denote by Y ′ ± and C′ the strict transforms of Y± and C respectively. The configuration of curves in R is given in Figure 22. Note that Y ′ ± are (−4)-curves + + Y ′ and other rational curves are all (−1)-curves. Let B′ = Y ′ − + C′. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of a curve of genus 3 and 2 rational curves, we see (r, l) = (10, 6), by Theorem 6.1. Therefore this is the example of No. [10]. Example No. [11]. Consider the following curves on P1 × P1; E1 : u2v2 + u2 + a1v2 + a2uv + a3 = 0, E2 : u2v2 + v2 + a4u2 + a5uv + a6 = 0 (ai ∈ C). Then Ei are smooth elliptic curves and preserved by ψ (Figure 23). Blow up P1 × P1 at 8 intersection points of E1 and E2. Let R be the blown up surface. We denote by E′ 2 the strict transforms of E1 and E2 respectively. Let B′ = E′ 2. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of two elliptic curves, we see (r, l, δ) = (10, 8, 0), by Theorem 6.1. To see to which No. this example belongs, we argue as follows. 1 and E′ 1 + E′ The involution ψ of P1 × P1 lifts to the rational elliptic surface R/P1, which acts on the base trivially. Hence, by choosing a zero-section, it corresponds to a translation by a 2-torsion section σ. In this case, the Horikawa construction corresponds exactly to the quadratic twist construction discussed in [Kon, HS]: the free involution ε is given by a lift of the translation automorphism by σ. We remark that generically the elliptic surface R has eight singular fibers 4I2 + 4I1 (Kodaira's notation). Here we consider a deformation of the K3 surface X: we move the branch locus B′ to B′ 1, the union of one I2 fiber plus one smooth fiber. We denote by X1 the smooth K3 surface obtained by the double cover branched along B′ 1 and the minimal desingularization. Since only rational double points appear in construction, X and X1 are connected by a smooth deformation. Now X1 has also an Enriques quotient Y1 via the quadratic twist construction. By definition of B′ 1, the main invariant of θ1 on X1 is (12, 8, 1) and the associated involution on Y1 has type No. [8]. We recall that a specialization of K3 surfaces X X1 exists if and only if TX1 ⊂ TX. Hence we see that our example belongs to No. [11]. Example No. [12]. Consider the following curves on P1 × P1; E± : v2(u2 ± a1u + a2) ± v(u2 ± a3u + a4) + (u2 ± a5u + a6) = 0 (ai ∈ C). Then E± are elliptic curves which are exchanged by ψ (Figure 24). ± the strict transforms of E± respectively. Let B′ = E′ Blow up P1 × P1 at 8 intersection points of E±. Let R be the blown up surface. We denote by E′ −. The K3 surface X is the double cover of R whose branch locus is B′. Since X θ = B′ consists of two elliptic curves, we see (r, l, δ) = (10, 8, 0), by Theorem 6.1. To check that they correspond to No. [12] in this case, we discuss as follows. + + E′ We remark that the case No. [9] is a specialization of our family: it is exactly the case where E± acquire nodes. By simultaneous resolution, we can regard the CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 23 Figure 23. Figure 24. K3 surface X0 in No. [9] as a special member of a smooth deformation with general fiber X1 from our family. Here, the two elliptic curves E± deform into the sums of two rational curves F± + F ′ ±) = 2 (double sign corresponds). ±)2) = −2 and (F±, F ′ ±, where (F 2 ±) = ((F ′ Moreover, since the formation of ε does not change under this specialization, our family is in fact a family of K3 surfaces with free involutions (X1, ε1) and (X0, ε0). (In other words, the free involutions are preserved under the specialization.) By the theory of period maps, we have an inclusion N S(X1) ⊂ N S(X0). The orthogonal complement is generated by the (−4)-vector F+ − F−, and the overlattice structure is given by F+ = F+ + F− 2 + F+ − F− 2 ∈ N S(X0). Hence, we can compute det N S(X0) = det N S(X1)· 4/22 = det N S(X1). Recalling that det N S is the same as det K− in each case, we can see that our example belongs to No. [12]. Example No. [14]. We need an irreducible curve on P1 × P1 which has 8 nodes and stable under ψ, but it seems not easy to construct them in a direct way. The following construction is due to H. Tokunaga. Let B0 be a smooth irreducible divisor of bidegree (2, 2) to which the four lines u = 0,∞; v = 0,∞ are tangent. We remark that in general, if a divisor is tangent to the branch curve (with local intersection number 2), then by pulling back to the double cover, the divisor acquires a node at the point of tangency. Thus in our case the following construction works: We consider the two self-morphisms ψ1 : (u, v) 7→ (u2, v) and ψ2 : (u, v) 7→ (u, v2) of P1 × P1. Then, the pullback C8 := (ψ1 ◦ ψ2)∗(B0) has bidegree (4, 4) with eight nodes and is stable under ψ (Figure 25). We can exhibit the equation for C8 as follows, for example. (c2u4 + 2cbu2 + b2)v4 + (2cau4 + du2 + 2b)v2 + (a2u4 + 2au2 + 1) = 0. Blow up P1 × P1 at 8 nodes of C8. Let R be the blown up surface. We denote by 8 the strict transforms of C8. The K3 surface X is the double cover of R whose C′ 24 H. ITO AND H. OHASHI Figure 25. 8. Since X θ = C′ 8 is a elliptic curve, we see (r, l, δ) = (10, 10, 1), branch locus is C′ by Theorem 6.1. Therefore this is the example of No. [14]. Example No.s [15] -- [18]. Let C2i (i = 0, 1, 2, 3) be irreducible curves on P1× P1 whose bidegree is (4, 4) with 2i nodes respectively. Blow up P1 × P1 at 2i nodes of C2i. Let R2i be the blown up surface. We denote by C′ 2i the strict transforms of C2i. The K3 surface X2i is the double cover of R2i whose branch locus is C′ 2i is a curve of genus 9 − 2i, we see (r, l) = (2i + 2, 2i + 2), by Theorem 6.1. Therefore the cases i = 3, 2, 1 and 0 are the examples of No. [15], [16], [17] and [18] respectively. 2i. Since X θ 2i = C′ 6.2. Enriques' sextics. The non-normal sextic surface in P3 which is singular along the six edges of a tetrahedron is a model of Enriques surface, the one first considered by Enriques himself. In fact its normalization gives a smooth Enriques surface, see [GH]. Setting the tetrahedron as xyzt = 0, the general equation of such surfaces is given by q(x, y, z, t)xyzt + (x2y2z2 + x2y2t2 + x2z2t2 + y2z2t2) = 0, where q is a quadratic equation. By considering various linear actions on P3, we can get many examples of involutions on Enriques surfaces. The most important for us among them is the following example exhibiting No. [13]. Example No. [13]. Let us consider the involution ι : (x : y : z : t) 7→ (y : x : t : z) on P3. The general equation of invariant Enriques' sextic Y looks as (cid:0)a1(x2 + y2) + a2(z2 + t2) + a3xy + a4zt + a5(xz + yt) + a6(xt + yz)(cid:1) xyzt +(x2y2z2 + x2y2t2 + x2z2t2 + y2z2t2) = 0, where ai ∈ C are general. Then the normalization Y is a smooth Enriques surface with the induced action by ι. Let us show that they belong to No. [13]. Since in this case θ is also fixed-point- free, this is equivalent to saying that the fixed locus Y ι is a finite set. Moreover ι since the normalization Y → Y is a finite morphism, it suffices to show that Y is a finite set. But this set is the intersection of Y with the fixed locus in P3, CLASSIFICATION OF INVOLUTIONS ON ENRIQUES SURFACES 25 {x = y, z = t} ∪ {x + y = 0, z + t = 0}. Since the general element does not contain these lines, the intersection is a finite set as desired. References [BHPV] W. Barth, K. Hulek, C. Peters and A. Van de Ven, Compact complex surfaces, Springer- Verlag, 2004. [BP] W. Barth and C. Peters, Automorphisms of Enriques surfaces, Invent. Math., 73 (1983), [Dol] [GH] [Hor] [HS] [Kon] [Mor] 383 -- 411. I. Dolgachev, On automorphisms of Enriques surfaces, Invent. Math., 76 (1984), 163 -- 177. P. Griffiths and J. Harris, Principles of algebraic geometry, Wiley Classics Library, John Wiley & Sons, Inc., New York, 1994. E. Horikawa, On the periods of Enriques surfaces. I -- II. Math. Ann., 234 (1978), 73 -- 88; 235 (1978), 217 -- 246. K. Hulek and M. Schutt, Enriques surfaces and Jacobian elliptic K3 surfaces, Math. Z., 268 (2011), 1025-1056. S. Kondo, Enriques surfaces with finite automorphism groups, Japan. J. Math., 12 (1986), 191 -- 282. D. R. Morrison, On K3 surfaces with large Picard number, Invent. Math., 75 (1984), 105 -- 121. [Muk1] S. Mukai, Numerically trivial involutions of Kummer type of an Enriques surface, Kyoto J. Math., 50 (2010), 889 -- 902. [Muk2] S. Mukai, Kummer's quartics and numerically reflective involutions of Enriques surfaces, [MN] J. Math. Soc. Japan, 64 (2012), 231 -- 246. S. Mukai and Y. Namikawa, Automorphisms of Enriques surfaces which act trivially on the cohomology groups, Invent. Math., 77 (1984), 383 -- 397. [Nik1] V. V. Nikulin, Finite automorphism groups of Kahler K3 surfaces, Trans. Moscow Math. Soc., 38 (1980), 71 -- 135. [Nik2] V. V. Nikulin, Integral symmetric bilinear forms and some of their applications (English translation), Math. USSR Izv., 14 (1980), 103 -- 167. [Nik3] V. V. Nikulin, Factor groups of groups of automorphisms of hyperbolic forms with respect to subgroups generated by 2-reflections, J. Soviet Math., 22 (1983), 1401 -- 1475. [Nik4] V. V. Nikulin, Involutions of integral quadratic forms and their applications to real [PS] algebraic geometry, Math. USSR Izv., 22 (1984), 99 -- 172. I. I. Piatetski-Shapiro and I. R. Shafarevich, A Torelli theorem for algebraic surfaces of type K3, Izv. Akad. Nauk SSSR Ser. Mat., 35 (1971), 530 -- 572. Graduate School of Mathematics, Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-8602 JAPAN E-mail address: [email protected] Department of Mathematics, Faculty of Science and Technology, Tokyo University of Science, 2641 Yamazaki, Noda, Chiba 278-8510, JAPAN E-mail address: [email protected]
1807.02082	2	1807	2018-11-19T01:06:57	Associated form morphism	[ "math.AG", "math.AC" ]	We study the geometry of the morphism between moduli spaces of hypersurfaces in $\mathbb P^{n-1}$ that sends a smooth hypersurface of degree $d+1$ to its associated hypersurface of degree $n(d-1)$. As a result, we obtain a compactification of the moduli space of smooth hypersurfaces such that the induced rational map from the standard GIT compactification often contracts the discriminant divisor.	math.AG	math	ASSOCIATED FORM MORPHISM MAKSYM FEDORCHUK AND ALEXANDER ISAEV Abstract. We study the geometry of the morphism that sends a smooth hyper- surface of degree d + 1 in Pn−1 to its associated hypersurface of degree n(d − 1) in the dual space (cid:0)Pn−1(cid:1)∨ . Contents Introduction 1. 2. Associated form of a balanced complete intersection 3. Preliminaries on dualities 4. The gradient morphism ∇ 5. The morphism AGr References 1 3 4 7 12 17 1. Introduction One of the first applications of Geometric Invariant Theory is a construction of the moduli space of smooth degree m hypersurfaces in a fixed projective space Pn−1 [15]. This moduli space is an affine GIT quotient given by a categorical quotient of the locus of GIT semistable hypersurfaces. We call Vm,n the GIT compactification of Um,n. The subject of this paper is a certain rational map Vm,n 99K Vn(m−2),n, where n ≥ 2, m ≥ 3 and where we exclude the (trivial) case (n, m) = (2, 3). While this map has a purely algebraic construction, which we shall recall soon, it has several surprising geometric properties that we establish in this paper. In particular, this rational map restricts to a locally closed immersion ¯A : Um,n → Vn(m−2),n, and often Mathematics Subject Classification: 14L24, 13A50, 13H10. Keywords: Geometric Invariant Theory, associated forms. The first author was partially supported by the NSA Young Investigator grant H98230-16-1-0061 and Alfred P. Sloan Research Fellowship. 1 where ∆ is the discriminant divisor parameterizing singular hypersurfaces. The GIT construction produces a natural compactification Um,n :=(cid:0)PH0(cid:0)Pn−1, O(m)(cid:1) \ ∆(cid:1) // PGL(n), Um,n ⊂ Vm,n :=(cid:0)PH0(cid:0)Pn−1, O(m)(cid:1)(cid:1)ss // PGL(n), 2 MAKSYM FEDORCHUK AND ALEXANDER ISAEV contracts the discriminant divisor in Vm,n. Consequently, the closure of the image of ¯A in Vn(m−2),n is a compactification of the GIT moduli space Um,n that is different from the GIT compactification Vm,n. To define ¯A, we consider the associated form morphism defined on the space of smooth homogeneous forms f ∈ C[x1, . . . , xn] of fixed degree m ≥ 3. Given such an f , its associated form A(f ) is a degree n(m − 2) homogeneous form in the graded dual polynomial ring C[z1, . . . , zn]. In our recent paper [10], we proved that the associated form A(f ) is always polystable in the sense of GIT. Consequently, we obtain a morphism ¯A from Um,n to Vn(m−2),n sending the image of f in Um,n to the image of A(f ) in Vn(m−2),n. Our first result is that the morphism ¯A is an isomorphism onto its image, a locally closed subvariety in the target. Theorem 1.1. The morphism is a locally closed immersion. ¯A : Um,n → Vn(m−2),n In the process of establishing Theorem 1.1, we generalize results of [2] to the case of an arbitrary number of variables, and, in particular, prove that the auxiliary gradient morphism sending a semistable form to the span of its partial derivatives gives rise to a closed immersion on the level of quotients (see Theorem 2.1). Our second main result is Theorem 2.2, which describes the rational map ¯A : Vm,n 99K Vn(m−2),n in codimension one. Namely, we study how ¯A extends to the generic point of the discriminant divisor in the GIT compactification (see Corol- lary 5.8), and prove that for n = 2, 3 and m ≥ 4, as well as for n ≥ 4, m ≫ 0, the morphism ¯A contracts the discriminant divisor to a lower-dimensional subvariety in the target (see Corollary 5.9). In the process, we prove that the image of ¯A contains the orbit of the Fermat hypersurface in its closure and as a result obtain a new proof of the generic smoothness of associated forms (see Corollary 5.10). 1.1. Notation and conventions. Let S := Sym V ≃ C[x1, . . . , xn] be a symmetric algebra of an n-dimensional vector space V , with its standard grading. Let D := Sym V ∨ ≃ C[z1, . . . , zn] be the graded dual of S, with the structure of the S-module given by the polar pairing S × D → D, which is defined by (1.1) g(x1, . . . , xn) ◦ F (z1, . . . , zn) := g(∂/∂z1, . . . , ∂/∂z1)F (z1, . . . , zn). A homogeneous polynomial f ∈ Sm is called a direct sum if, after a linear change of variables, it can be written as the sum of two non-zero polynomials in disjoint sets of variables: f = f1(x1, . . . , xa) + f2(xa+1, . . . , xn). We will use the recognition criteria for direct sums established in [8], and so we keep the pertinent terminology of that paper. We will say that f ∈ Sm is a k-partial ASSOCIATED FORM MORPHISM 3 Fermat form for some k ≤ n, if, after a linear change of variables, it can be written as follows: f = xm 1 + · · · + xm k + g(xk+1, . . . , xn). Clearly, any n-partial Fermat form is linearly equivalent to the standard Fermat form. Furthermore, all k-partial Fermat forms are direct sums. We denote by DSm the locus of direct sums in Sm. 2. Associated form of a balanced complete intersection Fix d ≥ 2. In what follows the trivial case (n, d) = (2, 2) will be excluded. A length n regular sequence g1, . . . , gn of elements of Sd will be called a balanced complete intersection of type (d)n. It defines a graded Gorenstein Artin C-algebra A(g1, . . . , gn) := S/(g1, . . . , gn), whose socle lies in degree n(d − 1). In [2] an element A(g1, . . . , gn) ∈ Dn(d−1), called the associated form of g1, . . . , gn, was introduced. The form A(g1, . . . , gn) is a homogeneous Macaulay inverse system, or a dual socle generator, of the algebra A(g1, . . . , gn). It follows that [A(g1, . . . , gn)] ∈ PDn(d−1) depends only on the linear span hg1, . . . , gni, which we regard as a point in Grass(n, Sd). Recall that g1, . . . , gn is a regular sequence in Sd if and only if hg1, . . . , gni does not in lie in the resultant divisor Res ⊂ Grass(n, Sd). Setting Grass(n, Sd)Res := Grass(n, Sd) \ Res, we obtain a morphism A : Grass(n, Sd)Res → PDn(d−1). Given f ∈ Sd+1, the partial derivatives ∂f /∂x1, . . . , ∂f /∂xn form a regular sequence if and only if f is non-degenerate. For a non-degenerate f ∈ Sd+1, in [1, 3] the associated form of f was defined to be A(f ) := A(∂f /∂x1, . . . , ∂f /∂xn) ∈ Dn(d−1). Summarizing, we obtain a commutative diagram P(Sd+1)∆ (PPPPPPPPPPPP ∇ A / P(Dn(d−1)) 6♠♠♠♠♠♠♠♠♠♠♠♠ A Grass(n, Sd)Res, where P(Sd+1)∆ denotes the complement to the discriminant divisor in P(Sd+1) and ∇ is the morphism sending a form into the linear span of its first partial derivatives. The above diagram is equivariant with respect to the standard SL(n)-actions on S and D. By [2], the morphism A is a locally closed immersion, and it was proved in [10] that A sends polystable orbits to polystable orbits. Passing to the GIT ( / 6 4 MAKSYM FEDORCHUK AND ALEXANDER ISAEV quotients, we thus obtain a commutative diagram (2.1) P(Sd+1)∆// SL(n) ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ e∇ ¯A / P(Dn(d−1))ss// SL(n) 4✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ¯A Grass(n, Sd)Res// SL(n), where e∇ := ∇// SL(n) is a finite injective morphism (see [9]) and ¯A := A// SL(n) is a locally closed immersion. The main focus of this paper is the geometry of diagram (2.1). Noting that by [9] the map ∇ extends to a morphism from P(Sd+1)ss to Grass(n, Sd)ss and thus induces a map ∇ of the corresponding GIT quotients, we will now state our two main results as follows: Theorem 2.1. The morphism ∇ : P(Sd+1)ss// SL(n) → Grass(n, Sd)ss// SL(n) is a closed immersion. Theorem 2.2. The rational map ¯A : P(Sd+1)ss// SL(n) 99K P(Dn(d−1))ss// SL(n) extends to the generic point of the discriminant divisor ∆// SL(n) in the GIT com- pactification and contracts the discriminant divisor to a lower-dimensional variety for all sufficiently large d as described in Corollaries 5.8 and 5.9. 3. Preliminaries on dualities In this section we collect results on Macaulay inverse systems of graded Gorenstein Artin C-algebras. We also recall the duality between the Hilbert points of such algebras and the gradient points of their inverse systems. Recall that we regard S = C[x1, . . . , xn] as a ring of polynomial differential oper- ators on the graded dual ring D := C[z1, . . . , zn] via polar pairing (1.1). For every positive m, the restricted pairing Sm × Dm → C is perfect and so defines an isomorphism (3.1) Dm ≃ S ∨ m, where, as usual, V ∨ stands for the dual of a vector space V . Given W ⊂ D, we define W ⊥ := {f ∈ S f ◦ g = 0, for all g ∈ W } ⊂ S. Similarly given U ⊂ S, we define U ⊥ := {g ∈ D f ◦ g = 0, for all f ∈ U } ⊂ D. / 4 ASSOCIATED FORM MORPHISM 5 Claim 3.1. Isomorphism (3.1) sends an element ω ∈ S ∨ m to the element Dω := Xi1+···+in=m ω(cid:0)xi1 1 · · · xin i1! · · · in! n(cid:1) zi1 1 · · · zin n ∈ Dm. Conversely, an element g ∈ Dm is mapped by isomorphism (3.1) to the projection Sm ։ Sm/(g⊥)m ≃ C, where the isomorphism with C is chosen so that 1 ∈ C pairs to 1 with g. Proof. One observes that f ◦ Dω = ω(f ) for every f ∈ Sm, and the first part of the claim follows. The second part is immediate from definitions. (cid:3) Corollary 3.2. Given ω ∈ S ∨ m, for every (a1, . . . , an) ∈ Cn we have (3.2) Proof. Dω(a1, . . . , an) = ω(cid:0)(a1x1 + · · · + anxn)m/m!(cid:1). ω(cid:0)(a1x1 + · · · + anxn)m/m!(cid:1) = = (a1x1 + · · · + anxn)m ◦ Dω (a1∂/∂z1 + · · · + an∂/∂zn)m m! m! Dω = Dω(a1, . . . , an), where the last equality is easily checked, say on monomials. (cid:3) Remark 3.3. It follows from Corollary 3.2 that all forms in a subset W ⊂ Dm vanish at a given point (a1, . . . , an) ∈ Cn if and only if (a1x1 + · · · + anxn)m ∈ W ⊥. Notice that the maps define isomorphisms (cid:2)hDωi ⊂ Dm(cid:3) 7→(cid:2)(D⊥ More generally, for any 1 ≤ m ≤(cid:0)m+n−1 ω )m ⊂ Sm(cid:3) =(cid:2)ker(ω) ⊂ Sm(cid:3) n−1 (cid:1) − 1 the correspondence Grass(1, Dm) ≃ Grass (dimC Sm − 1, Sm) . (cid:2)W ⊂ Dm(cid:3) 7→(cid:2)(W ⊥)m ⊂ Sm(cid:3) yields an isomorphism (3.3) Grass(k, Dm) ≃ Grass (dimC Sm − k, Sm) . Let I ⊂ S be a Gorenstein ideal and ν the socle degree of the algebra A = S/I. Recall that a (homogeneous) Macaulay inverse system of A is an element fA ∈ Dν such that f ⊥ A = I (see [11, Lemma 2.12] or [6, Exercise 21.7]). As (f ⊥ inverse systems are mutually proportional and hfAi = (cid:0)(Iν)⊥(cid:1)ν . Clearly, the line hfAi ∈ Grass(1, Dν ) maps to the νth Hilbert point Hν ∈ Grass(dimC Sν − 1, Sν ) of A under isomorphism (3.3) with k = 1. A )ν = Iν , we see that all Macaulay 6 MAKSYM FEDORCHUK AND ALEXANDER ISAEV Remark 3.4. Papers [3, 4], for any ω ∈ S ∨ ν with ker ω = Iν, introduced the associated form of A as the element of Dν given by the right-hand side of formula (3.2) with m = ν (up to the factor ν!). By Corollary 3.2, under isomorphism (3.3) with k = 1 the span of every associated form in Dν also maps to the νth Hilbert point Hν ∈ Grass(dimC Sν − 1, Sν ) of A. In particular, for the algebra A any associated form is simply one of its Macaulay inverse systems, and equation (3.2) with m = ν and ker ω = Iν is an explicit formula for a Macaulay inverse system of A (see [12] for more details). 3.1. Gradient points. Given a polynomial F ∈ Dm, we define the pth gradient point of F to be the linear span of all pth partial derivatives of F in Dm−p. We denote the pth gradient point by ∇p(F ). Note that ∇p(F ) = {g ◦ F g ∈ Sp} is simply the (m − p)th graded piece of the principal S-module SF . The 1st gradient point ∇F := h∂F/∂z1, . . . , ∂F/∂zni will be called simply the gradient point of F . Proposition 3.5 (Duality between gradient and Hilbert points). The pth gradient point of a Macaulay inverse system fA ∈ Dν maps to the (ν − p)th Hilbert point Hν−p of A under isomorphism (3.3). Proof. Let G be the pth gradient point of fA, that is G :=* ∂p ∂zi1 1 · · · ∂zin n fA i1 + · · · + in = p+ . We need to verify that Iν−p = (G⊥)ν−p. We have ∂zi1 1 · · · ∂zin n (G⊥)ν−p =(f ∈ Sν−p f ◦ =nf ∈ Sν−p f xi1 =nf ∈ Sν−p xi1 =nf ∈ Sν−p xi1 = Iν−p, ∂p fA = 0 for all i1 + · · · + in = p) n ◦ fA = 0 for all degree p monomialso A for all degree p monomialso n f ∈ Iν for all degree p monomialso n f ∈ f ⊥ 1 · · · xin 1 · · · xin 1 · · · xin where the last equality comes from the fact that I is Gorenstein. (cid:3) As a corollary of the above duality result, we recall in Proposition 3.6 below a gen- eralization of [1, Lemma 4.4]. Although this statement is well-known (it appears, for example, in [5, Proposition 4.1, p. 174]), we provide a short proof for completeness. We first recall that a non-zero homogeneous form f in n variables has multiplicity ℓ + 1 at a point p ∈ Pn−1 if and only if all partial derivatives of f of order ℓ (hence of all orders ≤ ℓ) vanish at p, and some partial derivative of f of order ℓ + 1 does ASSOCIATED FORM MORPHISM 7 not vanish at p. We define the Veronese cone Cm to be the variety of all degree m powers of linear forms in Sm: Cm :=(cid:8)Lm L ∈ S1(cid:9) ⊂ Sm. Proposition 3.6. Let I ⊂ S be a Gorenstein ideal and ν the socle degree of the algebra A = S/I. Then a Macaulay inverse system fA of A has a point of multi- plicity ℓ + 1 if and only if there exists a non-zero L ∈ S1 such that Lν−ℓ ∈ Iν−ℓ, and Lν−ℓ−1 6∈ Iν−ℓ−1. In particular, fA has no points of multiplicity ℓ + 1 or higher if and only if Iν−ℓ ∩ Cν−ℓ = (0). Proof. By Proposition 3.5, the ℓth gradient point of fA is dual to the (ν −ℓ)th Hilbert point of A Hν−ℓ : Sν−ℓ ։ Aν−ℓ. We conclude by Remark 3.3 that all partial derivatives of fA of order ℓ vanish at (a1, . . . , an) if and only if (a1x1 + · · · + anxn)ν−ℓ ∈ ker Hν−ℓ = Iν−ℓ. It follows that L = a1x1 + · · · + anxn satisfies Lν−ℓ ∈ Iν−ℓ and Lν−ℓ−1 6∈ Iν−ℓ−1 if and only if fA has multiplicity exactly ℓ + 1 at the point (a1, . . . , an). (cid:3) 4. The gradient morphism ∇ In this section, we prove Theorem 2.1. Recall that we have the commutative diagram P(cid:0)Sd+1)ss P(cid:0)Sd+1)ss// SL(n) π0 ∇ ∇ / Grass(cid:0)n, Sd(cid:1)ss π1 / Grass(cid:0)n, Sd(cid:1)ss// SL(n). Let DSss [8, Section 3], the set DSss positive fiber dimension. d+1 := P(DSd+1)ss be the locus of semistable direct sums in P(cid:0)Sd+1)ss. By d+1 is precisely the closed locus in P(cid:0)Sd+1)ss where ∇ has Suppose f ∈ Sd+1 is a semistable form. Then, after a linear change of variables, we have a maximally fine direct sum decomposition (4.1) f = fi(xi), kXi=1 where Vi = hxii are such that V = ⊕k i=1Vi, and where each fi is not a direct sum in Sym Vi. Set ni := dimC Vi. We define the canonical torus Θ(f ) ⊂ SL(n) associated to f as the connected component of the identity of the subgroup {g ∈ SL(n) Vi is an eigenspace of g, for every i = 1, . . . , k} ⊂ SL(n). / / 8 MAKSYM FEDORCHUK AND ALEXANDER ISAEV Clearly, Θ(f ) ≃ (C∗)k−1, and since ∇([f ]) = ∇([f1]) ⊕ · · · ⊕ ∇([fk]), where ∇([fi]) ∈ Grass(ni, Symd Vi), we also have Θ(f ) ⊂ Stab(∇([f ])), where Stab denotes the stabilizer under the SL(n)-action. From the definition of Θ(f ), it is clear that Θ(f ) · [f ] ⊂ ∇−1(∇([f ])), and in fact [8, Corollary 3.12] gives a set-theoretic equality ∇−1(∇([f ])) = Θ(f ) · [f ]. We will now obtain a stronger result: Lemma 4.1. One has ∇−1(∇([f ])) = Θ(f ) · [f ] scheme-theoretically, or, equiva- lently, where T[f ] denotes the tangent space at [f ]. ker(d∇[f ]) = T[f ](Θ(f ) · [f ]), P(Sd+1) with Sd+1/hf i, the sub- Proof. Under the standard identification of T[f ] space T[f ](Θ(f ) · [f ]) is identified with hf1, . . . , fki/hf i. It now suffices to show that every g ∈ Sd+1 that satisfies ∇[g] ⊂ ∇[f ] must lie in hf1, . . . , fki, where ∇[g] := h∂g/∂x1, . . . , ∂g/∂xni ⊂ Sd. This is precisely the statement of [8, Corollary 3.12]. (cid:3) We note an immediate consequence: Corollary 4.2. If f ∈ Sss d+1 is not a direct sum, then ∇ is unramified at [f ]. Further, since ∇ is equivariant with respect to the SL(n)-action, we have the inclusion Stab([f ]) ⊂ Stab(∇([f ])). As the following result shows, the difference between Stab([f ]) and Stab(∇([f ])) is controlled by the torus Θ(f ). Corollary 4.3. The subgroup Stab(∇([f ])) is generated by Θ(f ) and Stab([f ]). Proof. Suppose σ ∈ Stab(∇([f ])). Then ∇(σ · [f ]) = ∇([f ]) implies by Lemma 4.1 that σ · [f ] = τ · [f ] for some τ ∈ Θ(f ). Consequently, τ −1 ◦ σ ∈ Stab([f ]) as desired. (cid:3) Next, we obtain the following generalization of [2, Proposition 6.3], whose proof we follow almost verbatim. d+1 of all elements that are not direct sums. Proposition 4.4. The morphism ∇ is a closed immersion along the open locus U := P(Sd+1)ss \ DSss Proof. Since for every [f ] ∈ U we have that ∇ is unramified at [f ] and ∇−1(∇([f ])) = [f ], it suffices to show that ∇ is a finite morphism when restricted to U . Since, by [9], the induced morphism on the GIT quotients is finite, by [13, p. 89, Lemme] it suffices to verify that ∇ is quasi-finite and that ∇ sends closed orbits to closed orbits. The former has already been established, and the latter is proved below in Proposition 4.5. (cid:3) Proposition 4.5. Suppose f ∈ Sss image ∇([f ]) ∈ Grass(n, Sd)ss is polystable. d+1 is polystable and not a direct sum. Then the ASSOCIATED FORM MORPHISM 9 The above result is a generalization of [9, Theorem 1.1], whose method of proof we follow; we also keep the notation of loc.cit., especially as it relates to monomial orderings. We begin with a preliminary observation. Lemma 4.6. Suppose f ∈ Sd+1 is such that there exists a non-trivial one-parameter subgroup λ of SL(n) acting diagonally on x1, . . . , xn with weights λ1, . . . , λn and satisfying wλ(inλ(∂f /∂xi)) = dλi. Then f is a direct sum. Proof. We can assume that for some 1 ≤ a < n. Then the fact that λ1 ≤ · · · ≤ λa < λa+1 = · · · = λn wλ(inλ(∂f /∂xi)) = dλi = dλn, for all i = a + 1, . . . , n, implies ∂f /∂xa+1, . . . , ∂f /∂xn ∈ C[xa+1, . . . , xn]. Consequently, f = g1(x1, . . . , xa) + g2(xa+1, . . . , xn) is a direct sum. (cid:3) Proof of Proposition 4.5. Since f is polystable, by [9, Theorem 1.1] it follows that ∇([f ]) is semistable. Suppose ∇([f ]) is not polystable. Then there exists a one- parameter subgroup λ acting on the coordinates x1, . . . , xn with the weights λ1, . . . , λn such that the limit of ∇([f ]) under λ exists and does not lie in the orbit of ∇([f ]). In particular, the limit of [f ] under λ does not exist. Then by [9, Lemma 3.5], there is an upper triangular unipotent coordinate change x1 7→ x1 + c12x2 + · · · + c1nxn, x2 7→ x2 + · · · + c2nxn, ... xn 7→ xn such that for the transformed form h(x1, . . . , xn) := f (x1 + c12x2 + · · · + c1nxn, x2 + · · · + c2nxn, . . . , xn) the initial monomials are distinct. Now, setting inλ(∂h/∂x1), . . . , inλ(∂h/∂xn) by [9, Lemma 3.2] we have µi := wλ(inλ(∂h/∂xi)), µ1 + · · · + µn = 0. It follows that with the respect to the one-parameter subgroup λ′ acting on xi with the weight dλi − µi, all monomials of h have non-negative weights (cf. [9, the proof of Lemma 3.6]). Write h = h0 + h1, where all monomials of h0 have zero λ′-weights 10 MAKSYM FEDORCHUK AND ALEXANDER ISAEV and all monomials of h1 have positive λ′-weights. Then h0 ∈ SL(n) · h = SL(n) · h, by the polystability assumption on f . Furthermore, h0 is stabilized by λ′. If λ′ is a trivial one-parameter subgroup, then µi = dλi for all i = 1, . . . , n, and by Lemma 4.6 the form h is a direct sum, which is a contradiction. Suppose now that λ′ is a non-trivial one-parameter subgroup. Clearly, we have wλ(inλ(∂h0/∂xi) ≥ wλ(inλ(∂h/∂xi), since the state of h0 is a subset of the state of h. If one of the inequalities above is strict, then ∇([h0]) is destabilized by λ, contradicting the semistability of ∇([h0]) established in [9, Theorem 1.1]. Thus wλ(inλ(∂h0/∂xi)) = wλ(inλ(∂h/∂xi)) = µi. Moreover, since h0 is λ′-invariant, we have that ∂h0/∂xi is homogeneous of degree −wλ′(xi) = µi − dλi with respect to λ′. Let µ be the one-parameter subgroup acting on x1, . . . , xn with the weights µ1, . . . , µn. It follows that wµ(inµ(∂h0/∂xi)) = dwλ(inλ(∂h0/∂xi) + wλ′(inλ′(∂h0/∂xi) = dµi − µi + dλi. Then the one-parameter subgroup λ + µ acting on x1, . . . , xn with the weights λ1 + µ1, . . . , λn + µn satisfies wλ+µ(inλ+µ(∂h0/∂xi)) = wλ(inλ(∂h0/∂xi)) + wµ(inµ(∂h0/∂xi)) = dµi − µi + dλi + µi = d(µi + λi). Applying Lemma 4.6, we conclude that either h0 is a direct sum, or λi + µi = 0 for all i = 1, . . . , n. In the latter case, it follows that λ is proportional to λ′ = dλ − µ. Since the limit of h under λ′ exists and is equal to h0, the limit under λ of h must exist and be equal to h0 as well. Observing that the inverse of an upper-triangular matrix with 1's on the diagonal has the same form, we see that the limit of f (x1, . . . , xn) = h(x1 + c′ 2nxn, . . . , xn) 12x2 + · · · + c′ 1nxn, x2 + · · · + c′ under λ also exists. This contradiction concludes the proof. (cid:3) Corollary 4.7. The morphism ∇ : P(Sd+1)ss → Grass(n, Sd)ss preserves polystabil- ity. Proof. Suppose f = f1 + · · · + fk is the maximally fine direct sum decomposition of a polystable form f , where fi ∈ Symd+1 Vi, and where V = ⊕k i=1Vi. Then each fi is polystable and not a direct sum in Symd+1 Vi. Hence ∇([fi]) is polystable with respect to the SL(Vi)-action. Since Θ(f ) ⊂ Stab(∇([f ])) is a reductive subgroup, to prove that ∇([f ]) is polystable, it suffices to verify that ∇([f ]) is polystable with respect to the cen- tralizer CSL(n)(Θ(f )) of Stab(Θ(f )) in SL(n), see [14, Corollaire 1 and Remarque 1]. We have CSL(n)(Θ(f )) = (GL(V1) × · · · × GL(Vk)) ∩ SL(n). ASSOCIATED FORM MORPHISM 11 Arguing as on [9, p. 456], we see that every one-parameter subgroup λ of CSL(n)(Θ(f )) can be renormalized to a one-parameter subgroup of SL(V1) × · · · × SL(Vk) without changing its action on ∇([f ]). Since ∇([fi]) is polystable with respect to SL(Vi), it follows that ∇([f ]) = ∇([f1]) ⊕ · · · ⊕ ∇([fk]) is polystable with respect to the action of λ thus proving the claim. (cid:3) Proof of Theorem 2.1. Suppose that f is polystable, consider its maximally fine di- rect sum decomposition and the canonical torus Θ(f ) in Stab(∇([f ])) as constructed In what follows, we will write X to denote P(Sd+1)ss and Y to denote above. Grass(n, Sd)ss. Set p := π0([f ]) ∈ X// SL(n). We will prove that ∇ is unramified at p. Let N[f ] be the normal space to the SL(n)-orbit of [f ] in X at the point [f ], and N∇([f ]) the normal space to the SL(n)- orbit of ∇([f ]) in Y at the point ∇([f ]). We have a natural map ι : N[f ] → N∇([f ]) induced by the differential of ∇. The map ι is injective by Lemma 4.1. Since both [f ] and ∇([f ]) have closed orbits in X and Y , respectively (see Corol- lary 4.7), to verify that ∇ is unramified at p, it suffices, by Luna's ´etale slice theorem, to prove that the morphism (4.2) s(f ) : N[f ]// Stab([f ]) → N∇([f ])// Stab(∇([f ]) is unramified. As ∇ is not necessarily stabilizer-preserving at [f ] (i.e., Stab([f ]) may not be equal to Stab(∇([f ]))), we cannot directly appeal to the injectivity of ι. Instead, consider the Θ(f )-orbit, say F , of [f ] in X. Let NF/X be the Θ(f )-invariant normal bundle of F in X. Since by Lemma 4.1 we have ∇−1(∇([f ])) = F , there is a natural Θ(f )-equivariant map J : NF/X → N∇([f ]). We now make a key observation that for the induced map J : NF/X//Θ(f ) → N∇([f ]) one has J (NF/X//Θ(f )) = ι(cid:0)N[f ](cid:1) . Since ∇ is finite by [9, Proposition 2.1], the morphism s(f ) from Equation (4.2) is quasi-finite. Applying Lemma 4.8 (proved below), with Spec A = N[f ], Spec B = N∇([f ]), T = Θ(f ), H = Stab([f ]), G = Stab(∇([f ])), as well as Corollary 4.3, we obtain that s(f ) is in fact a closed immersion, and so is unramified. Note that here the group G is reductive by Matsushima's criterion. This proves that ∇ is unramified at p. We now note that ∇ is injective. Indeed, this follows as in the proof of [9, Part (2) of Proposition 2.1] from Corollary 4.7 and the finiteness of ∇. We then conclude that ∇ is a closed immersion. (cid:3) Lemma 4.8 (GIT lemma). Suppose G is a reductive group. Suppose T ⊂ G is a connected reductive subgroup, and H ⊂ G is a reductive subgroup such that G 12 MAKSYM FEDORCHUK AND ALEXANDER ISAEV is generated by T and H. Suppose we have a G-equivariant closed immersion of normal affine schemes admitting an action of G Spec A ֒→ Spec B. such that Spec AH → Spec BG is quasi-finite. Then Spec AG ≃ Spec AH and, con- sequently, Spec AH → Spec BG is a closed immersion. Proof. We have the following commutative diagram Spec AH / Spec BH +❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ (Spec AH)//T ≃ Spec AG / (Spec BH)//T ≃ Spec BG. Since the diagonal arrow is quasi-finite by assumption, and the bottom arrow is a closed immersion, we conclude that the GIT quotient Spec AH → (Spec AH )//T is quasi-finite as well. Since this is a good quotient by a connected group, the morphism Spec AH → (Spec AH )//T ≃ Spec AG must be an isomorphism. (cid:3) Corollary 4.9 (Theorem 1.1). The morphism ¯A : P(Sd+1)∆// SL(n) → P(Dn(d−1))ss// SL(n) is a locally closed immersion. 5. The morphism AGr In this section, we prove Theorem 2.2. In fact, we study in detail the rational map ¯A : (PSd+1)ss// SL(n) 99K P(Dn(d−1))ss// SL(n) in codimension one. As in Section 2, fix d ≥ 2. As always, we assume that n ≥ 2 and disregard the trivial case (n, d) = (2, 2). Given U ∈ Grass(n, Sd), we take IU to be the ideal in S generated by the elements in U . Consider the following locus in Grass(n, Sd): Wn,d = {U ∈ Grass(n, Sd) dimC(S/IU )n(d−1)−1 = n}. Since dimC(S/IU )n(d−1)−1 is an upper semi-continuous function on Grass(n, Sd) and for every U ∈ Grass(n, Sd) one has dimC(S/IU )n(d−1)−1 ≥ n, we conclude that Wn,d is an open subset of Grass(n, Sd). Moreover, since for U ∈ Grass(n, Sd)Res the ideal IU is Gorenstein of socle degree n(d − 1), we have Grass(n, Sd)Res ⊂ Wn,d. Applying polar pairing, we obtain a morphism AGr : Wn,d → Grass(n, Dn(d−1)−1), AGr(U ) =h(IU )⊥ n(d−1)−1 ⊂ Dn(d−1)−1i . From the duality between Hilbert and gradient points it follows that ∇(A(U )) = AGr(U ) for every U ∈ Grass(n, Sd)Res. + / / ASSOCIATED FORM MORPHISM 13 We conclude that we have the commutative diagram: P(Sd+1)ss// SL(n) π0 P(Sd+1)ss P(Dn(d−1))ss// SL(n) π2 P(Sd+1)∆ A P(Dn(d−1))ss ∇ ∇ Grass(n, Sd)ss π1 Grass(n, Sd)Res / Grass(n, Dn(d−1)−1)ss A 4✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ 4✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ AGr ∇ π3 Grass(n, Sd)ss// SL(n) Wn,d Grass(n, Dn(d−1)−1)ss// SL(n). Proposition 5.1. Suppose U ∈ Grass(n, Sd) is such that V(IU ) = {p1, . . . , pk} is scheme-theoretically a set of k distinct points in general linear position in Pn−1. Then U ∈ Wn,d. Remark 5.2. A set {p1, . . . , pk} points in Pn−1 is in general linear position if and only if k ≤ n, and, up to the PGL(n)-action, in the homogeneous coordinates [x1 : · · · : xn] on Pn−1. pi = {x1 = · · · = bxi = · · · = xn = 0}, i = 1, . . . , k, Proof of Proposition 5.1. Since depth(IU ) = n − 1, we can choose degree d gen- erators g1, . . . , gn of IU such that g1, . . . , gn−1 form a regular sequence. Then Γ := V(g1, . . . , gn−1) is a finite-dimensional subscheme of Pn−1. By B´ezout's theo- rem, Γ is a set of dn−1 points, counted with multiplicities. Set R := S/(g1, . . . , gn−1). Consider the Koszul complex K• := K•(g1, . . . , gn). We have Since g1, . . . , gn−1 is a regular sequence, we also have H0(K•) = S/(g1, . . . , gn) = S/IU . Hi(K•) = 0 for all i > 0 and H1(K•) =(cid:0)((g1, . . . , gn−1) : S (g1, . . . , gn))/(g1, . . . , gn−1)(cid:1)(−d) ≃ AnnR(gn)(−d). To establish the identity it suffices to prove codim(cid:0)(IU )n(d−1)−1, Sn(d−1)−1(cid:1) = n Indeed, in this case the graded degree n(d − 1) − 1 part of the Koszul complex will be an exact complex of vector spaces and so the dimension of (S/IU )n(d−1)−1 will H1(K•)n(d−1)−1 = 0. O O ? _ o o / / O O ? _ o o _ 4 / 4 14 MAKSYM FEDORCHUK AND ALEXANDER ISAEV coincide with that in the situation when g1, . . . , gn is a regular sequence, that is, with n. As we have already observed, we have H1(K•)n(d−1)−1 = AnnR(gn)n(d−1)−1(−d) = AnnR(gn)n(d−1)−1−d. Hence it suffices to prove that AnnR(gn)n(d−1)−1−d = 0. Write Γ = Γ′ ∪ Γ′′, where Γ′ 6= ∅ and Γ′′ := {p1, . . . , pk}. Since gn vanishes on all of Γ′′ but does not vanish at any point of Γ′, every element of AnnR(gn)n(d−1)−1−d comes from a degree n(d − 1) − 1 − d form that vanishes on all of Γ′. We apply the Cayley-Bacharach Theorem [7, Theorem CB6], which implies the following statement: Claim 5.3. Set s := d(n − 1) − (n − 1) − 1 = n(d − 1) − d. If r ≤ s is a non-negative integer, then the dimension of the family of projective hypersurfaces of degree r containing Γ′ modulo those containing all of Γ is equal to the failure of Γ′′ to impose independent conditions on projective hypersurfaces of complementary degree s − r. In our situation r = s − 1, and Γ′′ imposes independent conditions on hyperplanes by the general linear position assumption. Hence we conclude by Claim 5.3 that every form of degree n(d − 1) − 1 − d that vanishes on all of Γ′ also vanishes on all of Γ′′ and therefore, as the ideal (g1, . . . , gn−1) is saturated, maps to 0 in R. We thus see that AnnR(gn)n(d−1)−1−d = 0. This finishes the proof. (cid:3) Motivated by the result above, we consider the following partial stratification of the resultant divisor Res ⊂ Grass(n, Sd). For 1 ≤ k ≤ n, define Zk to be the locally closed subset of Grass(n, Sd) consisting of all subspaces U such that V(IU ) is scheme-theoretically a set of k distinct points in general linear position in Pn−1. Clearly, Z1 is dense in Res, and Z k ⊃ Zk+1 ∪ · · · ∪ Zn. We will also set Σk := ∇−1(Zk) ⊂ P(Sd+1). By the Jacobian criterion, Σk is the locus of hypersurfaces with only k ordinary double points in general linear position and no other singularities. Lemma 5.4. For every 1 ≤ k ≤ n, one has that Zk is a non-empty and irreducible subset of Grass(n, Sd), and Σk is a non-empty and irreducible subset of P(Sd+1)ss. Proof. It follows from the Hilbert-Mumford numerical criterion that any hypersur- face in Pn−1 of degree d + 1 with at worst ordinary double point singularities is semistable. Having k singularities at k fixed points p1, . . . , pk (resp., having k fixed base points p1, . . . , pk) in general linear position is a linear condition on the elements of P(Sd+1) (resp., the elements of the Stiefel variety over Grass(n, Sd)) and so defines an irreducible closed subvariety Σ(p1, . . . , pk) in P(Sd+1) (resp., Z(p1, . . . , pk) in Grass(n, Sd)). The property of having exactly ordinary double points at p1, . . . , pk (resp., having the base locus being equal to {p1, . . . , pk} scheme-theoretically) is an open condition in Σ(p1, . . . , pk) in P(Sd+1) (resp., Z(p1, . . . , pk) in Grass(n, Sd)) ASSOCIATED FORM MORPHISM 15 and so defines an irreducible subvariety Σ0(p1, . . . , pk) (resp., Z 0(p1, . . . , pk)). We conclude the proof of irreducibility by noting that Σk = PGL(n) · Σ0(p1, . . . , pk) (resp., Zk = PGL(n) · Z 0(p1, . . . , pk)). Since Σk = ∇−1(Zk), it suffices to check the non-emptiness of Σk. If F ∈ Σn has ordinary double points at p1, . . . , pn, then by the deformation theory of hypersur- faces, there exists a deformation of F with ordinary double points at p1, . . . , pk and no other singularities. Indeed, if G ∈ Sd+1 is a general form vanishing at p1, . . . , pk and non-vanishing at pk+1, . . . , pn, then F + tG ∈ Σ0(p1, . . . , pk) will have ordinary double points at p1, . . . , pk and no other singularities for 0 < t ≪ 1. It remains to prove that Σn is non-empty. Indeed, the following is an element of Σn: (d−1)(x1 +· · ·+xn)d+1 −(d+1)(x1 +· · ·+xn)d−1(x2 1 +· · ·+x2 n)+2(xd+1 1 +· · ·+xd+1 ). n In fact, a generic linear combination of all degree (d + 1) monomials with the ex- ception of xd+1 i xj, for i, j = 1, . . . , n, i < j, is a form with precisely n ordinary double point singularities in general linear position. (cid:3) , for i = 1, . . . , n, and xd i By Proposition 5.1, we know that AGr is defined at all points of Z1 ∪ · · · ∪ Zn. In fact, we can explicitly compute AGr(U ) for all U ∈ Zn, as well as the orbit closure of AGr(U ) for all U ∈ Zn−1. We need a preliminary fact. Proposition 5.5. Suppose U ∈ Grass(n, Sd) and p ∈ V(IU ) ⊂ PV ∨. Let L ∈ V ∨ be a non-zero linear form corresponding to p. Then Ln(d−1)−1 ∈ (IU )⊥ n(d−1)−1. Proof. Since p ∈ V(IU ), all elements of (IU )n(d−1)−1 vanish at p, and it follows that F ◦ Ln(d−1)−1 = 0 for all F ∈ (IU )n(d−1)−1 (cf. Remark 3.3). (cid:3) Corollary 5.6. Suppose U ∈ Zk is such that V(IU ) = {p1 := [1 : 0 : · · · : 0], p2 := [0 : 1 : · · · : 0], . . . , pk := [0 : · · · : 1 : · · · : 0]}. Then AGr(U ) = hzn(d−1)−1 1 , . . . , zn(d−1)−1 k , gk+1(z1, . . . , zn), . . . , gn(z1, . . . , zn)i, for some gk+1, . . . , gn ∈ Dn(d−1)−1. In particular, for U ∈ Zn one has AGr(U ) = hzn(d−1)−1 1 , . . . , zn(d−1) n + · · · + zn(d−1) n Moreover, for a generic U ∈ Zk, we have AGr(U ) ∈ Grass(n, Dn(d−1))Res. i = ∇(cid:0)(cid:2)zn(d−1) 1 (cid:3)(cid:1). form zi ∈ V ∨, Proposition 5.5 implies that zn(d−1)−1 Proof. Since the point pi = V(x1, . . . ,bxi, . . . , xn) ∈ PV ∨ corresponds to the linear ∈ AGr(U ) for every i = 1, . . . , k. As Zn ⊂ Z k and AGr(U ) ∈ Grass(n, Dn(d−1))Res for every U ∈ Zn, it follows that AGr(U ) is also spanned by a regular sequence for a generic U ∈ Zk. The claim follows. (cid:3) i 16 MAKSYM FEDORCHUK AND ALEXANDER ISAEV Consider the rational maps P(Sd+1)ss// SL(n) /❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴ ¯A P(Dn(d−1))ss// SL(n) ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ∇ 4✐ ✐ ✐ ✐ ✐ ¯A ✐ ✐ Grass(n, Sd)ss// SL(n) ✐ of projective GIT quotients. Theorem 5.7. There is a dense open subset Yk of Zk such that ¯A : Grass(n, Sd)ss// SL(n) 99K P(Dn(d−1))ss// SL(n) is defined on π1(Yk), k = 1, . . . , n. Moreover, for U ∈ Yk the value ¯A(π1(U )) is the image under π2 of a polystable k-partial Fermat form. In particular, for every U ∈ Zn and for a generic U ∈ Zn−1 ¯A(π1(U )) = π2(cid:16)zn(d−1) 1 + · · · + zn(d−1) n (cid:17) . Proof. Recall that Zk is non-empty by Lemma 5.4. Suppose U ∈ Zk is generic, then by Corollary 5.6 in suitable coordinates we have AGr(U ) = hzn(d−1)−1 1 , . . . , zn(d−1)−1 k , gk+1(z1, . . . , zn), . . . , gn(z1, . . . , zn)i, and AGr(U ) /∈ Res. It follows (as in the proof of [10, Proposition 2.7]) that the closure of the SL(n)-orbit of AGr(U ) contains (5.1) hzn(d−1)−1 1 , . . . , zn(d−1)−1 k , ¯gk+1(zk+1, . . . , zn), . . . , ¯gn(zk+1, . . . , zn)i, where ¯gi := gi(0, . . . , 0, zk+1, . . . , zn) for i = k + 1, . . . , n. Then the claim follows for for k = n − 1 and k = n as in these cases we necessarily have ¯gn = zn(d−1)−1 n . For k arbitrary, since ∇ is a closed immersion by Theorem 2.1, we conclude that ¯A is defined at π1(U ). Let F ∈ P(Dn(d−1))ss be a polystable element with π2(F ) = ¯A(π1(U )). Then we must have ∇(F ) ∈ SL(n) · AGr(U ), and so ∇(F ) is linearly equivalent to an element of the form (5.1). It follows at once that ¯A(π1(U )) = π2(cid:16)zn(d−1) 1 + · · · + zn(d−1)−1 k + G(zk+1, . . . , zn)(cid:17) is the image under π2 of a polystable k-partial Fermat form. (cid:3) We will now establish Theorem 2.2 as detailed in the next two corollaries. Corollary 5.8. The rational map ¯A : (PSd+1)ss// SL(n) 99K P(Dn(d−1))ss// SL(n) is defined at a generic point of π0(Σn−1) and at every point of π0(Σn). For a generic f ∈ Σn−1 and for every f ∈ Σn, we have ¯A(π0(f )) = π2(zn(d−1) 1 + · · · + zn(d−1) n ). / 4 ASSOCIATED FORM MORPHISM 17 Corollary 5.9. When n = 2, the rational map ¯A contracts the discriminant divisor to a point (corresponding to the orbit of the Fermat form in D2d−4) for all d ≥ 3. When n = 3, the rational map ¯A contracts the discriminant divisor to a lower- dimensional subvariety if d ≥ 3. More generally, for every n ≥ 4 there exists d0 such that for all d ≥ d0 the map ¯A contracts the discriminant divisor to a lower- dimensional subvariety. Proof. Notice that Σ1 is dense in the discriminant divisor ∆. Hence, for n = 2 the statement follows from Corollary 5.8. When n = 3, Theorem 5.7 implies that ¯A(π0(Σ1)) lies in the locus of a 1-partial Fermat form in D3(d−1). The linear equivalence classes of 1-partial ternary Fermat forms are in bijection with the linear equivalence classes of binary degree 3(d − 1) forms. The dimension of this locus is 3d − 6, which for d ≥ 3 is strictly less than 2 (cid:1) − 10 of the discriminant divisor. If n ≥ 4, by Theorem 5.7 the set ¯A(π0(Σ1)) lies in the locus of a 1-partial Fermat form in Dn(d−1). The linear equivalence classes of 1-partial Fermat forms in n variables are in bijection with the linear equivalence classes of degree n(d − 1) forms the dimension(cid:0)d+3 (cid:1), which for sufficiently in n−1 variables. The dimension of this locus is(cid:0)n(d−1)+(n−2) (cid:1) − large d is strictly less than the dimension of the discriminant divisor(cid:0)(d+1)+(n−1) (n2 + 1). n−1 (cid:3) n−2 We conclude the paper with an alternative proof of the main fact of [1] (see Proposition 4.3 therein). Corollary 5.10 (Generic smoothness of associated forms). The closure of Im A in P(Dn(d−1))ss contains the orbit SL(n) ·nzn(d−1) 1 + · · · + zn(d−1) n o of the Fermat hypersurface. Consequently, A(f ) is a smooth form for a generic smooth f ∈ Sd+1. Proof. By Corollary 5.8, we have π2(zn(d−1) 1 + · · · + zn(d−1) n ) ∈ Im( ¯A). Since the orbit of the Fermat hypersurface is closed in P(Dn(d−1))ss, it lies in the closure of Im A. (cid:3) References [1] Jarod Alper and Alexander Isaev. Associated forms in classical invariant theory and their applications to hypersurface singularities. Math. Ann., 360(3-4):799 -- 823, 2014. [2] Jarod Alper and Alexander Isaev. Associated forms and hypersurface singularities: The binary case. J. reine angew. Math., 2016. To appear, DOI: 10.1515/crelle-2016-0008. [3] Michael Eastwood and Alexander Isaev. Extracting invariants of isolated hypersurface singu- larities from their moduli algebras. Math. Ann., 356(1):73 -- 98, 2013. 18 MAKSYM FEDORCHUK AND ALEXANDER ISAEV [4] Michael Eastwood and Alexander Isaev. Invariants of Artinian Gorenstein algebras and isolated hypersurface singularities. In Developments and retrospectives in Lie theory, volume 38 of Dev. Math., pages 159 -- 173. Springer, Cham, 2014. [5] Richard Ehrenborg and Gian-Carlo Rota. Apolarity and canonical forms for homogeneous polynomials. European J. Combin., 14(3):157 -- 181, 1993. [6] David Eisenbud. Commutative algebra with a view toward algebraic geometry, volume 150 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995. [7] David Eisenbud, Mark Green, and Joe Harris. Cayley-Bacharach theorems and conjectures. Bull. Amer. Math. Soc. (N.S.), 33(3):295 -- 324, 1996. [8] Maksym Fedorchuk. Direct sum decomposability of polynomials and factorization of associated forms, 2017. arXiv:1705.03452. [9] Maksym Fedorchuk. GIT semistability of Hilbert points of Milnor algebras. Math. Ann., 367(1- 2):441 -- 460, 2017. [10] Maksym Fedorchuk and Alexander Isaev. Stability of associated forms. J. Algebraic Geom., to appear. arXiv:1703.00438. [11] Anthony Iarrobino and Vassil Kanev. Power sums, Gorenstein algebras, and determinantal loci, volume 1721 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1999. [12] Alexander Isaev. A criterion for isomorphism of Artinian Gorenstein algebras. J. Commut. Algebra, 8(1):89 -- 111, 2016. [13] Domingo Luna. Slices ´etal´es. M´emoires de la S. M. F., 33:81 -- 105, 1973. [14] Domingo Luna. Adh´erences d'orbite et invariants. Invent. Math., 29(3):231 -- 238, 1975. [15] David Mumford, John Fogarty, and Frances Kirwan. Geometric invariant theory, volume 34 of Ergebnisse der Mathematik und ihrer Grenzgebiete (2) [Results in Mathematics and Related Areas (2)]. Springer-Verlag, Berlin, third edition, 1994. (Fedorchuk) Department of Mathematics, Boston College, 140 Commonwealth Ave, Chestnut Hill, MA 02467, USA E-mail address: [email protected] (Isaev) Mathematical Sciences Institute, Australian National University, Acton, Canberra, ACT 2601, Australia E-mail address: [email protected]
1004.4379	1	1004	2010-04-25T21:10:23	A generalization of Fulton's conjecture for arbitrary groups	[ "math.AG", "math.RT" ]	We prove a generalization of Fulton's conjecture which relates intersection theory on an arbitrary flag variety to invariant theory.	math.AG	math	1 A GENERALIZATION OF FULTON'S CONJECTURE FOR ARBITRARY GROUPS PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE ABSTRACT. We prove a generalization of Fulton's conjecture which relates intersection theory on an arbitrary flag variety to invariant theory. 1. INTRODUCTION 1.1. The context of Fulton's original conjecture. Let L be a connected reductive complex al- gebraic group with a Borel subgroup BL and maximal torus H ⊂ BL. The set of isomorphism classes of finite dimensional irreducible representations of L are parametrized by the set X(H)+ of L-dominant characters of H via the highest weight. For λ ∈ X(H)+, let V (λ) = VL(λ) be the cor- responding irreducible representation of L with highest weight λ. Define the Littlewood-Richardson coefficients cν λ,µ by: V (λ) ⊗ V (µ) =Xν cν λ,µV (ν). The following result was conjectured by Fulton and proved by Knutson-Tao-Woodward [KTW]. (Subsequently, geometric proofs were given by Belkale [B2] and Ressayre [R2].) Theorem 1.1. Let L = GL(r) and let λ, µ, ν ∈ X(H)+. Then, if cν every positive integer n. λ,µ = 1, we have cnν nλ,nµ = 1 for (Conversely, if cnν nλ,nµ = 1 for some positive integer n, then cν λ,µ = 1. This follows from the saturation theorem of Knutson-Tao.) Replacing V (ν) by the dual V (ν)∗, the above theorem is equivalent to the following: Theorem 1.2. Let L = GL(r) and let λ, µ, ν ∈ X(H)+. Then, if [V (λ) ⊗ V (µ) ⊗ V (ν)]SL(r) = 1, we have [V (nλ) ⊗ V (nµ) ⊗ V (nν)]SL(r) = 1, for every positive integer n. The direct generalization of the above theorem for an arbitrary reductive L is false (see Exam- ple 8.3(3)). It is also known that the saturation theorem fails for arbitrary reductive groups. It is a challenge to find an appropriate version of the above result for GL(r) which holds in the setting of general reductive groups. The aim of this paper is to achieve one such generalization. This generalization is a relationship between the intersection theory of homogeneous spaces and the invariant theory. To obtain this generalization, we must first reinterpret the above result for GL(r) as follows. Without loss of generality, we only consider the irreducible polynomial representations of GL(r). These are parametrized by the sequences λ = (λ1 ≥ λ2 ≥ · · · ≥ λr ≥ 0), where we view any such λ as the dominant character diag(t1, . . . , tr) 7→ tλ1 r of the standard maximal torus consisting of the diagonal matrices in GL(r). Let P(r) be the set of such sequences (also called Young 1 . . . tλr 1P.B. and S.K. were supported by NSF grants. N.R. was supported by the French National Research Agency (ANR-09-JCJC-0102-01). Mathematics Subject Classification 2010: 14C17, 14L30, 20G05. 1 2 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE diagrams or partitions) λ = (λ1 ≥ λ2 ≥ · · · ≥ λr ≥ 0) and let Pk(r) be the subset of P(r) consisting of those partitions λ such that λ1 ≤ k. Then, the Schubert cells in the Grassmannian Gr(r, r + k) of r-planes in Cr+k are parametrized by Pk(r) (cf. [F2, §9.4]). For λ ∈ Pk(r), let σλ be the corresponding Schubert cell and ¯σλ its closure. By a classical theorem (cf. loc. cit.), the structure constants for the intersection product in H ∗(Gr(r, r + k), Z) in the basis [¯σλ] coincide with the corresponding Littlewood-Richardson coefficients for the representations of GL(r). Thus, the above theorem can be reinterpreted as follows: Theorem 1.3. Let L = GL(r) and let λ, µ, ν ∈ Pk(r) (for some k ≥ 1) be such that the intersection product [¯σλ] · [¯σµ] · [¯σν] = [¯σλo] in H ∗(Gr(r, r + k), Z), where λo := (k ≥ · · · ≥ k) (r copies of k). Then, [V (nλ) ⊗ V (nµ) ⊗ V (nν)]SL(r) = 1, for every positive integer n. 1.2. Generalization for arbitrary groups. Our generalization of Fulton's conjecture to an arbi- trary reductive group is by considering its equivalent formulation in Theorem 1.3. Moreover, the generalization replaces the intersection theory of the Grassmannians by the deformed product ⊙0 in the cohomology of G/P introduced in [BK]. The role of the representation theory of SL(r) is replaced by the representation theory of the semisimple part Lss of the Levi subgroup L of P . To be more precise, let G be a connected reductive complex algebraic group with a Borel sub- group B and a maximal torus H ⊂ B. Let P ⊇ B be a (standard) parabolic subgroup of G. Let L ⊃ H be the Levi subgroup of P , BL the Borel subgroup of L and Lss = [L, L] the semisimple part of L. Let W be the Weyl group of G, WP the Weyl group of P , and let W P be the set of mini- mal length coset representatives in W/WP . For any w ∈ W P , let Xw be the corresponding Schubert variety and [Xw] ∈ H 2(dim G/P −ℓ(w))(G/P, Z) the corresponding Poincar´e dual class (cf. Section 2). Also, recall the definition of the deformed product ⊙0 in the singular cohomology H ∗(G/P, Z) from [BK, Definition 18]. The following is our main theorem (cf. Theorem 8.2). Theorem 1.4. Let G be any connected reductive group and let P be any standard parabolic sub- group. Then, for any w1, . . . , ws ∈ W P such that (1) we have, for every positive integer n, [Xw1] ⊙0 · · · ⊙0 [Xws] = [Xe] ∈ H ∗(G/P ), (2) where VL(λ) is the irreducible representation of L with highest weight λ and χw is defined by the identity (16). dim(cid:0)(cid:2)VL(nχw1) ⊗ · · · ⊗ VL(nχws)(cid:3)Lss (cid:1) = 1, Remark 1.5. Let M be the GIT quotient of (L/BL)s by the diagonal action of Lss linearized by L(χw1) ⊠ · · · ⊠ L(χws). Then, the conclusion of Theorem 1.4 is equivalent to the rigidity statement that M = point. Theorem 1.4 can therefore be interpeted as the statement "multiplicity one in intersection theory leads to rigidity in representation theory". Our proof builds upon and further develops the connection between the deformed product ⊙0 and the representation theory of the Levi subgroup as established in [BK]. In loc. cit., for any w ∈ W P , the line bundle LP (χw) on P/BL was constructed (see Section 6 for the definitions). Further, the following result was proved in there (cf. [BK, Corollary 8 and Theorem 15]). FULTON'S CONJECTURE FOR ARBITRARY GROUPS 3 Proposition 1.6. Let w1, . . . , ws ∈ W P be such that [Xw1] ⊙0 · · · ⊙0 [Xws] = d[Xe] ∈ H ∗(G/P ), for some d 6= 0. Then, m := dim(cid:0)H 0(cid:0)(L/BL)s,(cid:0)LP (χw1) ⊠ · · · ⊠ LP (χws)(cid:1)(L/BL)s(cid:1)Lss (cid:1) 6= 0. Note that, by the Borel-Weil theorem, for any w ∈ W P , H 0(L/BL, LP (χw)(L/BL)) = VL(χw)∗. The condition (1) can be translated into the statement that a certain map of parameter spaces X → Y = (G/B)s appearing in Kleiman's theorem is birational. Here X is the "universal intersection" of Schubert varieties. It is well known that, for any birational map X → Y between smooth projective varieties, no multiple of the ramification divisor R in X can move even infinitesimally (i.e., the corresponding Hilbert scheme is reduced, and of dimension 0 at nR for every positive integer n). We may therefore conclude that h0(X, O(nR)) = 1 for every positive integer n. In our situation, X is not smooth, and moreover H 0(X, O(nR)) needs to be connected to the invariant theory. We overcome these difficulties by taking a closer look at the codimension one boundary of Schubert varieties. The proof also brings into focus the largest (standard) parabolic subgroup Qw acting on a Schu- bert variety Xw ⊆ G/P (where w ∈ W P ), the open Qw orbit Yw ⊆ Xw and the smooth locus Zw ⊆ Xw. The difference Xw \ Zw is of codimension at least two in Xw (since Xw is normal) and can effectively be ignored. The varieties Yw give us the link to invariant theory (see Proposition 6.2). The difference Zw \ Yw turns out to be quite subtle. A key result in the paper is that, in the setting of Proposition 6.2, the intersection ∩igiZwi of translates is non-transverse 'essentially' at any point which lies in (cid:0)∩i6=jgiZwi(cid:1) ∩ gj(Zwj \ Ywj ) for some j (cf. Proposition 8.1 for a precise statement). This re- veals the significance of Qw in the intersection theory of G/P and, in particular, to the deformed product ⊙0. The "complexity" of the varieties Zw \ Yw can therefore be expected to relate to the deformed product ⊙0. Note that by a result of Brion-Polo [BP], if P is a cominuscule maximal par- abolic subgroup, then Yw = Zw, and in this case the deformed cohomology product ⊙0 coincides with the standard intersection product as well (cf. [BK, Lemma 19]). As mentioned above, for any cominuscule flag variety G/P (in particular, for the Grassmanni- ans Gr(r, r + k)), the deformed product ⊙0 in H ∗(G/P ) coincides with the standard intersection product. In the case of G = GL(r + k) and G/P = Gr(r, r + k), the set W P can be identified with Pk(r). For any λ ∈ W P , the corresponding irreducible representation of the Levi subgroup L = GL(r)×GL(k) with the highest weight χλ coincides with V (λ)∗⊠V (λ) (cf. [B1]), where V (λ) is the irreducible representation of GL(r) as defined in Section 1.1 and λ is the conjugate partition giving rise to the irreducible representation V (λ) of GL(k). Thus, if we specialize Theorem 1.4 to G = GL(r + k), we get Theorem 1.3. Observe that in the case G = GL(r + k) and G/P = Gr(r, r + k), under the assumption of Proposition 1.6, from the above discussion and the discussion in Section 1.1, we get the stronger relation m = d2. In general, however, there are no known numerical relations between m and d (cf. Examples 8.3). We remark that if we replace the condition (1) in Theorem 1.4 by the standard cohomology product, then the conclusion of the theorem is false in general (see Example 8.3(4)). Also, the converse to Theorem 1.4 is not true in general (cf. Example 8.3(1)). 4 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE 2. NOTATION Let G be a connected reductive complex algebraic group. We choose a Borel subgroup B and a maximal torus H ⊂ B and let W = WG := NG(H)/H be the associated Weyl group, where NG(H) is the normalizer of H in G. Let P ⊇ B be a (standard) parabolic subgroup of G and let U = UP be its unipotent radical. Consider the Levi subgroup L = LP of P containing H, so that P is the semi-direct product of U and L. Then, BL := B ∩ L is a Borel subgroup of L. Let X(H) denote the character group of H, i.e., the group of all the algebraic group morphisms H → Gm. Then, BL being the semidirect product of its commutator [BL, BL] and H, any λ ∈ X(H) extends uniquely to a character of BL. We denote the Lie algebras of G, B, H, P, U, L, BL by the corresponding Gothic characters: g, b, h, p, u, l, bL respectively. Let R = Rg be the set of roots of g with respect to the Cartan subalgebra h and let R+ be the set of positive roots (i.e., the set of roots of b). Similarly, let Rl be the set of roots of l with respect to h and R+ l be the set of roots of bL. Let ∆ = {α1, . . . , αℓ} ⊂ R+ be the set of simple roots, where ℓ is the semisimple rank of G (i.e., the dimension of h′ := h ∩ [g, g]). We denote by ∆(P ) the set of simple roots contained in Rl. For any 1 ≤ j ≤ ℓ, define the element xj ∈ h′ by (3) αi(xj) = δi,j, ∀ 1 ≤ i ≤ ℓ. Recall that if WP is the Weyl group of P (which is, by definition, the Weyl Group WL of L; thus WP := WL), then in each coset of W/WP we have a unique member w of minimal length. This satisfies (cf. [K, Exercise 1.3.E]): (4) wBLw−1 ⊆ B. Let W P be the set of minimal length representatives in the cosets of W/WP . For any w ∈ W P , define the Schubert cell: Cw = C P w := BwP/P ⊂ G/P. Then, it is a locally-closed subvariety of G/P isomorphic to the affine space Aℓ(w), ℓ(w) being the length of w (cf. [J, Part II, Chapter 13]). Its closure is denoted by Xw = X P w , which is an irreducible (projective) subvariety of G/P of dimension ℓ(w). We denote the point wP ∈ Cw by w. We also need the shifted Schubert cell: Λw = ΛP w := w−1BwP/P ⊂ G/P. Let µ(Xw) denote the fundamental class of Xw considered as an element of the singular ho- mology with integral coefficients H2ℓ(w)(G/P, Z) of G/P . Then, from the Bruhat decomposition, the elements {µ(Xw)}w∈W P form a Z-basis of H∗(G/P, Z). Let {[Xw]}w∈W P be the Poincar´e dual basis of the singular cohomology with integral coefficients H ∗(G/P, Z). Thus, [Xw] ∈ H 2(dim G/P −ℓ(w))(G/P, Z). The tangent space T P = T e(G/P ) of G/P at e ∈ G/P carries a canonical action of P induced from the left multiplication of P on G/P . We recall the following definition from [BK, Definition 4]. Definition 2.1. Fix a positive integer s ≥ 1. Let w1, . . . , ws ∈ W P be such that (5) codim Λwj = dim G/P. s Xj=1 FULTON'S CONJECTURE FOR ARBITRARY GROUPS 5 This of course is equivalent to the condition: (6) ℓ(wj) = (s − 1) dim G/P. s Xj=1 We then call the s-tuple (w1, . . . , ws) Levi-movable (for short L-movable) if, for generic (l1, . . . , ls) ∈ Ls, the intersection l1Λw1 ∩ · · · ∩ lsΛws is transverse at e. All the schemes are considered over the base field of complex numbers C. The varieties are reduced (but not necessarily irreducible) schemes. 3. A CRUCIAL GEOMETRIC RESULT Let π : X → Y be a regular birational morphism of smooth irreducible varieties with Y projec- tive. Assume that we have a (not necessarily smooth) irreducible projective scheme ¯X containing X as an open subscheme such that (1) the codimension of each irreducible component of ¯X \ X in ¯X is at least two, (2) π extends to a regular map ¯π : ¯X → Y . Let R be the ramification divisor of π in X. It is, by definition, the effective Cartier divisor obtained as the zero scheme of the section of the line bundle L induced by the derivative map Dπx : Tx(X) → Tπ(x)(Y ), where the line bundle L has base X and fiber Lx at any x ∈ X is given by: In the above set up, one has the following crucial result. Lx = ∧top(Tx(X)∗) ⊗ ∧top(Tπ(x)(Y )). Proposition 3.1. For every n ≥ 1, h0(X, O(nR)) = 1, where h0 denotes the dimension of H 0. Proof. Clearly πX\R : X \ R → Y is an ´etale (and hence quasi-finite) birational morphism between smooth varieties. Hence, by the original form of Zariski's main theorem [M, Chap. III, §9], it is an open immersion, i.e., π(X \ R) is open in Y and π : X \ R → π(X \ R) is an isomorphism. We will show that V := Y \ π(X \ R) is of codimension at least two in Y . This will then imply that H 0(X, O(nR)) ⊆ H 0(X \ R, O) = H 0(Y \ V, O) = H 0(Y, O) = C. Since ¯π is surjective, a point v ∈ V is either in ¯π( ¯X \ X), or in π(R), i.e., V ⊆ ¯π( ¯X \ X) ∪ π(R). We show that π(R) is of codimension at least two in Y and thus conclude the proof (by assumption (1)). To do this let Z be the smallest closed subset of Y so that there exists a morphism σ : Y \ Z → ¯X representing the birational inverse to ¯π. It is known that the codimension of Z in Y is at least two (follow [H, Proof of Theorem 8.19 on page 181]). Clearly, ¯π ◦ σ = I on Y \ Z and similarly σ ◦ ¯π is identity on ¯π−1(Y \ Z) (for the last, note that σ ◦ ¯π is well defined as a morphism ¯π−1(Y \ Z) → ¯X which on an open subset is the identity). We therefore find that ¯π : ¯π−1(Y \ Z) → Y \ Z is an isomorphism. This tells us that ¯π−1(Y \ Z) is smooth and ¯π−1(Y \ Z) ∩ R = ∅. Hence, R is a subset of π−1(Z), (cid:3) or that π(R) ⊆ Z. 6 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE 4. SOME REMARKS ON RAMIFICATION DIVISORS Consider a linear map p : V → W between vector spaces of the same dimension. Let Det(p) := (∧topV )∗ ⊗ ∧topW = Hom(∧topV, ∧topW ). Denote by θ(p) the canonical element of Det(p) induced by p, i.e., θ(p) is the top exterior power of p. The following lemma is immediate. Lemma 4.1. Let p : V → W be as above and α : W ′ → W a surjective map. Let V ′ ⊂ V ⊕ W ′ consist of (v, w′) such that p(v) = α(w′) (i.e., V ′ is the fiber product of p and α). Let p′ : V ′ → W ′ be the projection. Then, the kernel of p′ is identified with the kernel of p via the surjective projection π : V ′ → V . Further, there is a canonical isomorphism of the vector spaces Det(p) and Det(p′) (defined below), which carries θ(p) to θ(p′). (Observe that V ′ and W ′ have the same dimension.) Hence, for any fiber diagram of irreducible smooth varieties: f f X ′ π′ Y ′ / X π / Y, where f is a smooth morphism and X, Y are of the same dimension with π a dominant morphism, we have the following identity between the ramification divisors: (7) f ∗(R(π)) = R(π′). The isomorphism ξ : Det(p) → Det(p′) is given by: ξ(θ)(e1 ∧ · · · ∧ ed ∧ ed+1 ∧ · · · ∧ en) = p′(e1) ∧ · · · ∧ p′(ed) ∧ ¯θ(π(ed+1) ∧ · · · ∧ π(en)), for any θ ∈ Det(p) = Hom(∧topV, ∧topW ), where {e1, . . . , en} is any basis of V ′ such that {e1, . . . , ed} is a basis of Ker(π) and ¯θ := σ ◦ θ (σ being any section of the map ∧n−d(W ′) → ∧n−d(W ) induced from α). It is easy to see that ξ does not depend upon the choice of the basis and the section σ. Gi be algebraic subgroups which keep Yi stable. Assume further thatPs Let X be an irreducible smooth variety and Y1, . . . , Ys irreducible smooth locally-closed subva- rieties of X. Assume that X has a transitive action by a connected linear algebraic group G and let i=1 codim(Yi) = dim X. Let Yi = G ×Gi Yi be the total space of the fiber bundle with fiber Yi associated to the principal Gi-bundle G → G/Gi. Then, we have the morphism mi : Yi → X, [g, yi] 7→ gyi, where [g, yi] denotes the equivalence class of (g, yi) ∈ G × Yi. Since Yi is smooth and G acts transitively on X, by the G-equivariance, mi is a smooth morphism (cf. [H, Corollary 10.7, Chap. III]). Taking their Cartesian product, we get the smooth morphism m : Y1 × · · · × Ys → X s. Let Y be the fiber product of m with the diagonal map δ : X → X s. We get a smooth morphism m : Y → X by restricting m to Y. Hence, Y is a smooth and irreducible variety (cf. the proof of Lemma 5.2). We also have the morphism π : Y → G/G1 × · · · × G/Gs obtained coordinatewise from the canonical projections πi : Yi → G/Gi. For any gi ∈ G and yi ∈ Yi, the map eyi : G → X, g 7→ gyi, induces the tangent map Ψ(gi,yi) : Tgi(G) → Tgiyi(X). Since Yi is Gi-stable, this map induces the map ¯Ψ(gi,yi) : T¯gi(G/Gi) → Tgiyi(X)/Tgiyi(giYi), where ¯gi = giGi. Moreover, for any hi ∈ Gi, (8) ¯Ψ(gi,yi) = ¯Ψ(gihi,h−1 i yi). / / FULTON'S CONJECTURE FOR ARBITRARY GROUPS 7 To see this, observe that the following diagram is commutative for any gi ∈ G and hi ∈ Gi. Tgi(G) &LLLLLLLLLL DRhi / Tgihi(G) xqqqqqqqqqqq T¯gi(G/Gi), where Rhi : G → G is the right multiplication by hi. Thus, ¯Ψ(gi,yi) depends only upon the equiv- alence class [gi, yi] ∈ G ×Gi Yi and we denote ¯Ψ(gi,yi) by Ψ[gi,yi]. Since G acts transitively on X, Ψ[gi,yi] is surjective. For any a = ([g1, y1], . . . , [gs, ys]) ∈ Y, we have the following diagram (for x = m(a)): (9) πa TaY T¯g1(G/G1) ⊕ · · · ⊕ T¯gs(G/Gs) D m TxX TxX Tx(giYi) , i=1 /Ls where ¯gi := giGi, the bottom horizontal map is the canonical projection in each factor, D m is surjective since m is a smooth morphism and the right vertical map is the coordinatewise surjective map Ψ[gi,yi]. Lemma 4.2. The above diagram is commutative. In fact, Ta(Y) is the fiber product of Tx(X) and T¯g1(G/G1) ⊕ · · · ⊕ T¯gs(G/Gs) via the above diagram. Proof. Let F be the fiber product of Tx(X) and T¯g1(G/G1) ⊕ · · · ⊕ T¯gs(G/Gs). It is easy to see that the above diagram is commutative. Moreover, since yi = g−1 i x for any a = ([g1, y1], . . . , [gs, ys]) ∈ Y with m(a) = x, Ta(Y) is a subspace of the fiber product F . Further, (10) (11) (12) dim Y = dim X + = dim X + = dim X + (dim Yi − dim X) (dim G/Gi + dim Yi − dim X) (dim G/Gi − codim Yi). s s Xi=1 Xi=1 Xi=1 s From this we see that dim F = dim Ta(Y). This proves the lemma. (cid:3) 5. INTERSECTION OF GENERAL TRANSLATES OF SCHUBERT VARIETIES We follow the notation from Section 2. For w ∈ W P , let Qw be the stabilizer of the Schubert variety Xw inside G/P under the left multiplication of G on G/P . Then, clearly, Qw is a standard parabolic subgroup of G. Let and let Zw denote the smooth locus of Xw. Clearly Yw := Qw w ⊂ Xw, Xw ⊃ Zw ⊃ Yw ⊃ Cw, & / x / / / 8 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE and each of Zw, Yw, Cw is an open subset of Xw. Remark 5.1. It is instructive to look at the example of G/P = Gr(r, n). Let F• : 0 ( F1 ( F2 ( · · · ( Fn = Cn be the standard flag in Cn, and let I = {i1 < · · · < ir} be a subset of {1, . . . , n} of cardinality r. Consider the (closed) Schubert variety ΩI(F•) = {V ∈ Gr(r, n) dim V ∩ Fik ≥ k, k = 1, . . . , r}. Let J = {i ∈ I : i + 1 6∈ I}. It is easy to see that ΩI (F•) = {V ∈ Gr(r, n) dim V ∩ Fik ≥ k, ∀ik ∈ J}. So I \ J is "redundant" for the definition of the closed Schubert variety ΩI(F•). It is easy to see that the stabilizer of the Schubert variety ΩI (F•) is QI := {g ∈ SL(n) : gFj ⊂ Fj, ∀j ∈ J}. We may think of QI as the set of elements of SL(n) that preserve the parts of F• "essential" for the definition of the closed Schubert variety ΩI (F•). It may be remarked that if P is a minuscule or cominuscule maximal parabloic, then Zw = Yw (cf. [BP]). Fix a positive integer s ≥ 1 and fix w1, . . . , ws ∈ W P , so that [Xw1] · . . . · [Xws] = d[Xe] ∈ H ∗(G/P ), for some d > 0. (13) There are four universal intersections that will be relevant here. Let δ : G/P → (G/P )s be the diagonal embedding. We denote its image by δ(G/P ) and identify it with G/P . For a locally- closed B-subvariety A ⊂ G/P , let A := G ×B A be the total space of the fiber bundle with fiber A associated to the principal B-bundle G → G/B. Then, there is a G-equivariant morphism mA : A → G/P defined by [g, x] 7→ gx, which is a smooth morphism if A is smooth. Now, consider the product X := Xw1 × · · · × Xws, where Xwi = G ×B Xwi, and similarly define Y, Z, C by replacing Xwi with Ywi, Zwi, Cwi re- spectively. Let mX : X → (G/P )s be the G-equivariant morphism mXw1 × · · · × mXws acting componentwise. We similarly define mY , mZ, mC. Finally, we define the (universal intersection) G-scheme X as the fiber product of δ with mX. We similarly define the G-schemes Y, Z, C by replacing mX with mY , mZ, mC respectively. Since δ is a closed embedding, X , Y, Z, C are the scheme theoretic inverse images of δ(G/P ) under mX , mY , mZ, mC respectively. Moreover, since mY , mZ, mC are smooth morphisms, Y, Z, C are reduced closed subschemes of Y, Z, C respectively. It is easy to see that (due to the assumption (13)) (14) dim X = s × dim(G/B). Observe that, set theoretically, X = {(g1B, . . . , gsB, x) ∈ (G/B)s × G/P : x ∈ ∩s i=1 giXwi}. There is a similar description for Y, Z, C. The open embeddings give rise to G-equivariant open embeddings: Cwi ⊂ Ywi ⊂ Zwi ⊂ Xwi C ⊂ Y ⊂ Z ⊂ X , and X is projective. FULTON'S CONJECTURE FOR ARBITRARY GROUPS 9 Lemma 5.2. (1) X is irreducible and so is Y, Z and C. (2) Z is a smooth variety (and hence so is Y and C). (3) The complement of Z in X is of codimension ≥ 2. Proof. (1) It is easy to see that each fiber of mXw : Xw → G/P is irreducible. Thus, each fiber of mX : X → (G/P )s is also irreducible. Now, take an irreducible component X1 of X such that X1 contains the full fiber of mX over the base point in δ(G/P ). Since X1 is G stable, X1 must contain the full fiber over any point in δ(G/P ). Thus, X1 = X , proving that X is irreducible. Since Y, Z and C are open subsets of X , they must be irreducible too. (2) For the second part, observe that the canonical map Z → δ(G/P ) is a smooth morphism. Since G/P is smooth, we get the smoothness of Z. (3) Since the Schubert varieties Xw are normal, the complement of Zw in Xw is of codimension ≥ 2 and is covered by Schubert cells. Thus, the complement of Z in X is of codimension ≥ 2. From this it is easy to see that the complement of Z in X is of codimension ≥ 2. (cid:3) We have a natural G-equivariant projection π : X → (G/B)s obtained coordinatewise from the projections Xwi → G/B. As observed in the identity (14), the domain and the range of π have the same dimension. The following lemma follows from Lemma 4.2. Lemma 5.3. For any point a = ([g1, x1], . . . , [gs, xs]) ∈ Z, the derivative (Dπ)a of π at a has Ker(Dπ)a ≃ ∩s i=1Tx(giZwi), where x = g1x1 = · · · = gsxs. In particular, π is regular at a if and only if the intersection ∩s i=1giZwi in G/P is transverse at x. Using Kleiman's transversality theorem [BK, Proposition 3] and our assumption (13), the map πZ : Z → (G/B)s is generically finite. Let R be the ramification divisor for the map πZ (equipped with the scheme structure described in Section 3). Under the assumption of the following corollary, the hypotheses of Proposition 3.1 are in place here and allow us to conclude the following: Corollary 5.4. Assume that d = 1 in equation (13). Then, for every n ≥ 1, (15) h0(Z, O(nR)) = 1. Proof. By Lemma 5.2, all the hypotheses of Proposition 3.1 are satisfied except the hypothesis that πZ is birational, which we now prove. i=1 giZwi is dense in ∩s By [BK, Proposition 3], there exists a nonempty open subset U ⊂ (G/B)s such that for each i=1 giZwi is transverse at each point of the intersection x = (g1B, . . . , gsB) ∈ U, the intersection ∩s i=1 giXwi. Moreover, since d = 1 (by assumption), the intersection and ∩s i=1 giZwi consists of a single point. From this we see that (πZ)−1(x) consists of exactly one ∩s point for each x ∈ U and, moreover, by Lemma 5.3, (πZ)−1(x) ⊂ Z \ R. Thus, π(πZ )−1(U ) : (πZ)−1(U) → U is an isomorphism, proving that πZ is birational. Now applying Proposition 3.1, we get the corollary. (cid:3) The aim now is to have equation (15) bear representation theoretic consequences. However, it is the space H 0(Y, O(nR)) which has clear relations to invariant theory. 10 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE 6. CONNECTING h0(Y, O(nR)) TO INVARIANT THEORY We first prove that Y and R ∩ Y are obtained from a base change with connected fibers. To do this, define Y′ := (G ×Qw1 Yw1) × · · · × (G ×Qws Yws). Ywi → G/P, [g, x] 7→ gx. Again, m′ Y : Y′ → (G/P )s obtained from the coordinatewise Similar to the map mY , we define the map m′ Y is a smooth morphism. Now, let Y ′ be the maps G ×Qwi fiber product of m′ Y with δ. Then, Y ′ is an irreducible smooth variety of the same dimension as that of (G/Qw1) × · · · × (G/Qws) (by virtue of the same proof given in the last section for the corresponding results for Y). Similar to the map πY : Y → (G/B)s, we have the map It is easy to see that the following diagram is Cartesian: π′ : Y ′ → (G/Qw1) × · · · × (G/Qws). Y π / Y ′ π′ (G/B)s / (G/Qw1) × · · · × (G/Qws), where the two horizontal maps are the canonical projections. (To prove this, observe that the above diagram is clearly Cartesian with Y, Y ′ in the above diagram replaced by Y, Y′ respectively.) Since π is a dominant morphism, so is π′. Thus, by Lemma 4.1, the ramification divisor S := R ∩ Y of πY is the pull-back of the ramification divisor R′ of π′. In particular, the line bundle O(nR)Y = O(nS). We therefore conclude that under the G-equivariant pull-back map, Lemma 6.1. For any n ∈ Z, H 0(Y, O(nR)Y) ≃ H 0(Y ′, O(nR′)), as G-modules. Define the P -variety (under the diagonal action of P ): P = (P/(w−1 1 Qw1w1 ∩ P )) × · · · × (P/(w−1 s Qwsws ∩ P )), and define the G-equivariant morphism of G-varieties: φ : G ×P P → Y ′, [g, (¯p1, . . . , ¯ps)] 7→ ([gp1w−1 1 , w1], . . . , [gpsw−1 s , ws]), where ¯pi = pi(w−1 i Qwiwi ∩ P ). It is easy to see that it is bijective. Since Y ′ is smooth and irreducible, φ is an isomorphism by [K, Theorem A.11]. For any w ∈ W P , it is easy to see that the Borel BL of the Levi subgroup L of P is contained in w−1Qww ∩ L (in fact, it is contained in w−1Bw by equation (4)). For any λ ∈ X(H), we have a P -equivariant line bundle LP (λ) on P/BL associated to the principal BL-bundle P → P/BL via the one dimensional BL-module λ−1. (As observed in Section 2, any λ ∈ X(H) extends uniquely to a character of BL.) The twist in the definition of L(λ) is introduced so that the dominant characters correspond to the dominant line bundles. For w ∈ W P , define the character χw ∈ h∗ by χw = Xβ∈(R+\R+ l )∩w−1R+ β . / / FULTON'S CONJECTURE FOR ARBITRARY GROUPS 11 Then, from [K, 1.3.22.3] and equation (4), χw = ρ − 2ρL + w−1ρ, (16) where ρ (resp. ρL) is half the sum of roots in R+ (resp. in R+ a character of w−1Qww ∩ P . Proposition 6.2. Assume that the s-tuple (w1, . . . , ws) satisfying the condition (13) is Levi-movable. Then, for any n ≥ 1, l ). It is easy to see that χw extends as H 0(Y, O(nR)Y)G ≃(cid:2)VL(n(χw1 − χ1))∗ ⊗ VL(nχw2)∗ ⊗ · · · ⊗ VL(nχws)∗(cid:3)L, where VL(χ) is the irreducible L-module with highest weight χ. (Observe that for w ∈ W P , χw is a L-dominant weight and so is χw − χ1.) Proof. Applying Lemma 4.2 to the case when X = G/P, Yi = Ywi, Gi = Qwi, and using the isomorphism φ : G ×P P → Y ′ as above, we get the following Cartesian diagram (for any g ∈ G and p = (¯p1, . . . , ¯ps) ∈ P): T[g,p](G ×P P) / ⊕s i=1Tgpiw−1 i Qwi (G/Qwi) TgP (G/P ) / ⊕s i=1 TgP (G/P ) TgP (gpiw−1 i Ywi ) , π′ where the top horizontal map is induced from the G-equivariant composite map π′ ◦ φ : G ×P P → i=1(G/Qwi) and the bottom horizontal map is the canonical projection in each factor. Thus, by Lemma 4.1, the ramification divisor φ−1(R′) is the same as the ramification divisor associated to the bundle map (between the vector bundles of the same rank over the base space G ×P P): Qs s G ×P (P × T P ) → Mi=1 G ×P (cid:0)P ×(w−1 i Qwi wi∩P ) (T P /T P wi)(cid:1), w is the tangent space T e(Λw), P acts diagonally on where T P is the tangent space T e(G/P ), T P P × T P and the map in the i-th factor is induced from the composite map P × T P → (P/(w−1 wi). Thus, by [BK, Lemma 6 and the discussion following it] and Lemma 4.1, the line bundle corre- sponding to the divisor φ−1(R′) is G-equivariantly isomorphic to the line bundle G ×P M over the base space G ×P P, where i Qwiwi ∩ P )) × T P ≃ P ×(w−1 i Qwi wi∩P ) T P → P ×(w−1 i Qwi wi∩P ) (T P /T P M = LP (χw1 − χ1) ⊠ LP (χw2) ⊠ · · · ⊠ LP (χws). Observe that, for any w ∈ W P , the line bundle LP (χw), though defined on P/BL, descends to a line bundle on P/(w−1Qww ∩ P ) since the character χw extends to a character of w−1Qww ∩ P . Thus, H 0(Y, O(nR)Y )G ≃ H 0(Y ′, O(nR′))G, by Lemma 6.1 ≃ H 0(G ×P P, G ×P M⊗n)G ≃ H 0(P, M⊗n)P L )L, ≃ H 0(L, M⊗n / / 12 where PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE L := (L/(w−1 1 Qw1w1 ∩ L)) × · · · × (L/(w−1 s Qwsws ∩ L)) and the last isomorphism follows from [BK, Theorem 15 and Remark 31(a)]. Thus, the proposition follows from the Borel-Weil theorem. (cid:3) 7. STUDY OF CODIMENSION ONE CELLS IN THE SCHUBERT VARIETIES We continue to follow the notation and assumptions from Section 2. The following lemma can be found in [BP, §2.6]. However, we include its proof for completeness. Lemma 7.1. For any w ∈ W P , the stabilizer Qw of Xw satisfies (17) where ∆w := ∆ ∩ w(R+ l ⊔ R−) and R− is the set of negative roots of g. ∆(Qw) = ∆w, Thus, (18) ∆(Qw) = ∆ ∩ (wwP o )R−, where wP element w in the coset wWL.) o is the longest element of the Weyl group WL of L. (Observe that wwP o is the longest Proof. We first prove equation (17). Observe that l ⊔ R− l ⊔ (R−\R− l )(cid:17) w(cid:16)R+ l ⊔ R−(cid:17) = w(cid:16)R+ = w(cid:16)Rl ⊔ (R−\R− l )(cid:17). ∆w = ∆ ∩ w(cid:16)Rl ⊔ (R−\R− l )(cid:17) = ∆ ∩ wR−, since w(R+ l ) ⊂ R−. Thus, (19) Take αi ∈ ∆w = ∆ ∩ wR−. Then, si BwP/P ⊂ (BwP/P ) ∪ (BsiwP/P ) = (BwP/P ) ∪ (Bsi wP/P ). But si w < w since ( w)−1 αi ∈ R−. Hence, siXw ⊂ Xw. This proves the inclusion ∆(Qw) ⊃ ∆w = ∆ ∩ wR−. Conversely, take αi ∈ ∆(Qw), i.e., siXw ⊂ Xw. Thus, si w < w and hence w−1αi ∈ R−. This proves the inclusion ∆(Qw) ⊂ ∆w and hence equation (17) is proved. The equation (18) follows by combining equations (17) and (19). (cid:3) β Proposition 7.2. Let v → w ∈ W P (i.e., v, w ∈ W P , β ∈ R+ such that w = sβv and ℓ(w) = ℓ(v) + 1). Then, the (codimension one) cell Cv of Xw is contained in QwwP/P if and only if β ∈ ∆w. In particular, β is a simple root in this case. FULTON'S CONJECTURE FOR ARBITRARY GROUPS 13 Proof. We first prove the implication '⇐': If β ∈ ∆w, then β ∈ ∆(Qw), by Lemma 7.1. Thus, v = sβwP ∈ QwwP/P . Conversely, we prove the implication '⇒': Assume, if possible, that v ∈ QwwP/P but β /∈ ∆w. We first show that Xv is stable under Qw (assuming β /∈ ∆w). By Lemma 7.1, it suffices to show that for any αj ∈ ∆w = ∆ ∩ wR−, we have αj ∈ ∆v. Since w−1αj ∈ R−, we get sj w < w. Take β→ w. Hence, there exists a a reduced decomposition w = sjsi1 · · · sid. Since v (unique) 1 ≤ p ≤ d such that v = sjsi1 · · · sip · · · sid and, of course, it is a reduced decomposition. (Here we have used the assumption that β /∈ ∆w.) β→ w, then so is v Thus, sj v < v, i.e., v−1αj ∈ R− and hence αj ∈ ∆v. This proves the assertion that Xv is stable under Qw. By assumption, v ∈ QwwP/P , i.e., v = q w for some q ∈ Qw. Thus, q−1 v = w and hence w ∈ QwXv = Xv, which is a contradiction. This contradiction shows that β ∈ ∆w and hence completes the proof of the proposition. (cid:3) For w ∈ W P , it is easy to see that the tangent space, as an H-module (induced from the left multiplication of H on Xw), is given by: (20) where gγ is the root space of g corresponding to the root γ. Hence, (21) gγ, T w(Xw) ≃ Mγ∈R+∩wR− T e(cid:0)w−1Xw(cid:1) ≃ Mγ∈R−∩w−1R+ gγ. The following lemma determines the tangent space along codimension one cells. Lemma 7.3. For v β→ w ∈ W P , the tangent space, as an H-module, is given by: Thus, as an H-module, T v(Xw) ≃ Mγ∈R+∩vR− T e(v−1Xw) ≃ Mγ∈R−∩v−1R+ gγ!M g−β. gγ!M g−v−1β. (Observe that v is a smooth point of Xw since Xw is normal; in particular, its singular locus is of codimension at least two.) Proof. Since v ∈ Xv ⊂ Xw, by (20), (22) gγ ⊂ T v(Xw). Mγ∈R+∩vR− For any root α ∈ R, let Uα := Exp(gα) ⊂ G be the corresponding 1-dimensional unipotent group. Then, UβU−β w = UβwU−w−1β e = Uβ w ⊂ Xw (since w−1β ∈ R−). Hence, UβHU−β w ⊂ Xw. But, from the SL(2)-theory, UβHU−β ⊃ U−βsβH. In particular, U−βsβH w ⊂ Xw, i.e., U−β v ⊂ Xw. 14 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE This proves that (23) Combining (22)–(23), we get g−β ⊂ T v(Xw). But, both the sides are of the same dimension ℓ(v) + 1, proving the lemma. (cid:0) Mγ∈R+∩vR− gγ(cid:1)M g−β ⊂ T v(Xw). (cid:3) As above, let P be any standard parabolic subgroup of G and let xP ∈ h′ = h ∩ [g, g] be the element defined by αi(xP ) = 0, = 1, for all the simple roots αi ∈ ∆(P ) for all the simple roots αi /∈ ∆(P ). Then, xP is in the center of the Lie algebra l. Set mo = θ(xP ), where θ is the highest root of g. (Observe that mo ≤ 2 for any maximal parabolic subgroup P of a classical group G.) Define a decomposition of T e(G/P ) as a direct sum of L-submodules as follows. First decompose T e(G/P ) as a direct sum of H-eigenspaces (induced from the canonical action of H on G/P ): For any 1 ≤ j ≤ mo, define T e(G/P ) = Mβ∈R+\R+ Vj = Mβ∈R+\R+ l : l β(xP )=j T e(G/P )−β. T e(G/P )−β. Clearly, each Vj is a L-submodule of T e(G/P ) and we have the decomposition (as L-modules) mo Define an increasing filtration of T e(G/P ) by P -submodules given by T e(G/P ) = Vj. Mj=1 where d F1 ⊂ F2 ⊂ F3 ⊂ · · · ⊂ Fmo = T e(G/P ), For any subvariety Z ⊂ G/P such that e is a smooth point of Z, define Fd = Vj. Mj=1 Also, we get the increasing filtration Fj(Z) of T e(Z) given by Fj(Z) := Fj ∩ T e(Z). Vj(Z) := Vj ∩ T e(Z). We set, for 1 ≤ j ≤ mo, dj(Z) = dimension of Vj(Z). (Observe that d0(Z) = 0 since T e(Z) ⊂ T e(G/P ) ≃Lα∈R+\R+ l g−α.) FULTON'S CONJECTURE FOR ARBITRARY GROUPS 15 Let z(L) be the center of L. If Z, as above, is z(L)-stable, we get the decomposition (as z(L)- modules) T e(Z) = Vj(Z). mo Mj=1 Theorem 7.4. For any v such that β → w in W P such that v is not in the Qw-orbit of w, there exists 1 ≤ j ≤ mo (24) dj(w−1Xw) 6= dj(v−1Xw). Proof. Let us set α := v−1β ∈ R+. For any 1 ≤ j ≤ mo, we get (by Lemma 7.3 and equation (21) applied to v) (25) dj(v−1Xw) = dj(v−1Xv) + δj,α(xP ). By the equation (21), the roots in T e(w−1Xw) are precisely R− ∩ w−1R+ (cid:16)i.e., T e(w−1Xw) ≃ Lγ∈R−∩w−1R+ gγ(cid:17). Set Φw−1 := R+ ∩ w−1R−. Then, as is well known, where ρ is half the sum of all the positive roots. δ = ρ − w−1ρ, Xδ∈Φw−1 Thus (abbreviating dj(w−1Xw) by dj and dj(v−1Xv) by d′ j), (26) Similarly, (27) Of course, (28) and (29) Now, (30) (ρ − w−1ρ)(xP ) = d1 + 2d2 + · · · + modmo. (ρ − v−1ρ)(xP ) = d′ 1 + 2d′ 2 + · · · + mod′ mo. d1 + d2 + · · · + dmo = ℓ(w), d′ 1 + d′ 2 + · · · + d′ mo = ℓ(v) = ℓ(w) − 1. (ρ − w−1ρ)(xP ) − (ρ − v−1ρ)(xP ) = (v−1ρ − w−1ρ)(xP ) = (v−1ρ − sαv−1ρ)(xP ), since w = vsα = hv−1ρ, α∨i α(xP ) = hρ, (vα)∨i α(xP ) = hρ, β∨i α(xP ). 16 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE On the other hand, by (26)–(29), (ρ − w−1ρ)(xP ) − (ρ − v−1ρ)(xP ) = (d1 − d′ = 1 + (d2 − d′ 1) + 2(d2 − d′ 2) + · · · + mo(dmo − d′ mo) 2) + 2(d3 − d′ 3) + · · · + (mo − 1)(dmo − d′ mo). (31) Combining (30)–(31), we get (32) 1 + (d2 − d′ 2) + 2(d3 − d′ 3) + · · · + (mo − 1)(dmo − d′ mo) = hρ, β∨i α(xP ). If (24) were false, we would get dj = dj(v−1Xw), for all 1 ≤ j ≤ mo, i.e., by the identity (25), we would get dj = d′ j for all j 6= α(xP ) and dα(xP ) = d′ α(xP ) + 1. Combining this with the identity (32), we would get (33) 1 + α(xP ) − 1 = hρ, β∨i α(xP ), i.e., α(xP ) = hρ, β∨i α(xP ). But, by the definition of β, it is easy to see that if β were a simple root, then β ∈ ∆w. Since, by assumption, v is not in the Qw-orbit of w, this contradicts Proposition 7.2. Hence, β is not a simple root and this contradicts the identity (33). (Observe that α(xP ) 6= 0, since v, w ∈ W P and w = vsα.) This contradiction arose because we assumed that (24) is false. This proves (24) and hence the theorem is proved. (cid:3) 8. MAIN THEOREM AND ITS PROOF We follow the notation and assumptions from Section 5. In particular, let w1, . . . , ws ∈ W P be such that identity (13) is satisfied for some d > 0. We assume further that the s-tuple (w1, . . . , ws) is Levi-movable. This will be our assumption through this section. Proposition 8.1. Under the above assumption, there exists a closed subset A of Z such that (34) Z \ Y ⊆ R ∪ A, codim(A, Z) ≥ 2. Proof. Let Z o := Z \ R. It suffices to show that for u1, . . . , us ∈ W P such that ui = wi for all i 6= io and uio → wio for some 1 ≤ io ≤ s and uio /∈ Ywio , Z o ∩ (Cu1 × · · · × Cus) = ∅. Since the s-tuple (w1, . . . , ws) is Levi-movable, there exist l1, . . . , ls ∈ L such that the standard quotient map s T e(G/P ) → T e(G/P )/T e(liΛwi) Mi=1 is an isomorphism. Hence, the eigenspaces corresponding to any eigenvalue 1 ≤ j ≤ mo under the action of xP also are isomorphic, i.e., Vj(G/P ) ≃ s Mi=1 Vj(G/P )/Vj(liΛwi), Xi=1 Xi=1 Xi=1 Xi=1 FULTON'S CONJECTURE FOR ARBITRARY GROUPS 17 where Vj is as in Section 7. (Here we have used the fact that liΛwi is z(L)-stable.) In particular, since the filtration Fj of T e(G/P ) is P -stable, for any p1, . . . , ps ∈ P , s s (35) dim Fj = (dim Fj − dim(Fj(liΛwi)) = (dim Fj − dim(Fj(piΛwi)). If nonempty, take a = ([g1, x1], . . . , [gs, xs]) ∈ Z o ∩(Cu1 ×· · ·×Cus), for gi ∈ G and xi ∈ Cui. In particular, g1x1 = · · · = gsxs. Let us denote this common element by gP . From the G-equivariance, we can assume that g = e. By Lemma 5.3, the quotient map T e(G/P ) → T e(G/P )/T e(piu−1 i Zwi) s Mi=1 is an isomorphism, where pi ∈ P is any element chosen such that gi ∈ piu−1 any j, the quotient map i B. In particular, for is injective. Thus, for any j, Fj → s Mi=1 Fj/(T e(piu−1 i Zwi) ∩ Fj) s s (36) dim Fj ≤ (dim Fj − dim(Fj(piu−1 i Zwi)) = (dim Fj − dim(Fj(u−1 i Zwi)). Considering the image of T e(g−1Zw) in T e(G/P )/Fj, for gP ∈ Zw, it is easy to see that, for any u, w ∈ W P such that u ∈ Zw and any j, we have (37) dim Fj(w−1Zw) ≤ dim Fj(u−1Zw). Now, let jo be an integer such that dim Fjo(w−1 (38) i Zwi) 6= dim Fjo(u−1 i Zwi) for i = io. This is possible by virtue of Theorem 7.4. This contradicts the inequality (36) for j = jo (by using (35), (37)–(38)). Hence the proposition is proved. (cid:3) Recall the definition of the deformed product ⊙0 in the singular cohomology H ∗(G/P, Z) from [BK, Definition 18]. We now come to our main theorem. Theorem 8.2. Let G be any connected reductive group and let P be any standard parabolic sub- group. Then, for any w1, . . . , ws ∈ W P such that [Xw1] ⊙0 · · · ⊙0 [Xws] = [Xe] ∈ H ∗(G/P ), we have (for any n ≥ 1) (39) dim HomL(VL(nχ1), VL(nχw1) ⊗ · · · ⊗ VL(nχws)) = 1, where χw is defined by identity (16). Equivalently, we have (for the commutator subgroup Lss := [L, L]): (40) dim(cid:0)(cid:2)VL(nχw1) ⊗ · · · ⊗ VL(nχws)(cid:3)Lss (cid:1) = 1, ∀ n ≥ 1. 18 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE Proof. By [BK, Theorem 15 and Proposition 17], the s-tuple (w1, . . . , ws) is L-movable. Hence, by Proposition 6.2, we have H 0(Y, O(nR)Y)G ≃(cid:2)VL(n(χw1 − χ1))∗ ⊗ VL(nχw2)∗ ⊗ · · · ⊗ VL(nχws)∗(cid:3)L Moreover, by Proposition 8.1, . H 0(Y, O(nR)Y) ֒→ H 0(Z, O(m(n)R)), for some m(n) > 0. Finally, by Corollary 5.4, for any m ≥ 1, h0(Z, O(mR)) = 1. But, since the constants belong to H 0(Y, O(nR)Y), we have dim(H 0(Y, O(nR)Y)G) ≥ 1. This proves the identity (39) since (χ1 being a trivial character on the maximal torus of Lss) (cid:2)VL(n(χw1−χ1))∗⊗VL(nχw2)∗⊗· · ·⊗VL(nχws)∗(cid:3)L ≃ HomL(VL(nχ1), VL(nχw1)⊗· · ·⊗VL(nχws)). The equivalence of (39) with (40) follows from [BK, Theorem 15]. (cid:3) Example 8.3. (1) The converse to the above theorem is false in general. For example, consider G = Sp(2ℓ), G/P the Lagrangian Grassmannian LG(ℓ, 2ℓ). It is cominuscule, so the structure constants for the singular cohomology and the deformed cohomology ⊙0 are the same. The cells in LG(ℓ, 2ℓ) are parametrized by the strict partitions a : (a1 > a2 > · · · > ar > 0) and a1 ≤ ℓ, r ≤ ℓ (cf. [FP, Page 29]). The corresponding Levi subgroup is GL(ℓ), so the Fulton conjecture (Theorem 1.1) holds. Now, take ℓ = 3 and consider the cells in LG(3, 6) corresponding to the strict partitions (1), (2 > 1), (2). The corresponding intersection number is 2. The corresponding representations of the Levi sub- group have Young diagrams (2 ≥ 0 ≥ 0), (3 ≥ 3 ≥ 0) and (3 ≥ 1 ≥ 0) respectively. Hence, the dimension of the invariant subspace for the corresponding tensor product of the Levi is 1. (2) In the above example, the intersection number is strictly larger than the dimension of the invariant subspace for the corresponding tensor product. We also have examples where the in- tersection number is strictly smaller than the dimension of the invariant subspace for the cor- responding tensor product. Take, for G/P the Lagrangian Grassmannian LG(5, 10) and con- sider the cells corresponding to the strict partitions (3 > 1), (3 > 2), (4 > 2). The intersec- tion number is 4. The corresponding representations of the Levi subgroup have Young diagrams (4 ≥ 3 ≥ 1 ≥ 0 ≥ 0), (4 ≥ 4 ≥ 2 ≥ 0 ≥ 0) and (5 ≥ 4 ≥ 2 ≥ 1 ≥ 0) respectively. Hence, the dimension of the invariant subspace for the corresponding tensor product of the Levi is 5. (3) Following the convention in [Bo], for L of type G2, [V (6ω1) ⊗ V (6ω2) ⊗ V (7ω2)]L = 1, and [V (12ω1) ⊗ V (12ω2) ⊗ V (14ω2)]L = 2. Similarly, [V (6ω1) ⊗ V (6ω2) ⊗ V (10ω1 + ω2)]L = 1 and [V (12ω1) ⊗ V (12ω2) ⊗ V (20ω1 + 2ω2)]L = 3, where {ω1, ω2} are the fundamental weights. Thus, the direct generalization of the Fulton's conjecture is false for general semisimple L. (4) There are examples of w1, w2, w3 ∈ W P such that (41) [Xw1] · [Xw2] · [Xw3] = [Xe] ∈ H ∗(G/P ), but (40) is false. Take, for example, G = Sp(6) and P to be the maximal parabolic with ∆ \ ∆(P ) = {α2} (following the convention in [Bo]). Now, take w1 = w2 = s1s3s2s1s3s2, w3 = s3s2. FULTON'S CONJECTURE FOR ARBITRARY GROUPS 19 Then, (41) is satisfied (cf. [KLM, Theorem 4.6]). In this case, restricted to the Cartan of Lss, we have χw1 = χw2 = ω1 + ω3, χw3 = 3ω1 + ω3. Thus, for any n ≥ 1, dim(cid:0)(cid:2)VL(nχw1) ⊗ VL(nχw2) ⊗ VL(nχw3)(cid:3)Lss (cid:1) = 0. Remark 8.4. (1) If we specialize Theorem 8.2 to G = GL(m) and P any maximal parabolic subgroup, then (as explained in the introduction) we readily obtain a proof of Fulton's conjecture proved by Knutson-Tao-Woodward [KTW] (Belkale [B2] and Ressayre [R2] gave other geometric proofs) asserting the following: Let VL(λ1), . . . , VL(λs) be finite dimensional irreducible representations of L = GL(r) with is one dimensional. highest weights λ1, . . . , λs respectively. Assume that(cid:2)VL(λ1)⊗· · ·⊗VL(λs)(cid:3)Lss Then, for any n ≥ 1,(cid:2)VL(nλ1) ⊗ · · · ⊗ VL(nλs)(cid:3)Lss In fact, since any maximal parabolic subgroup in GL(m) is cominuscule, by a result of Brion- Polo [BP], we have Z = Y. Hence, Proposition 8.1 and the results from Section 7 are not needed in this case. again is one dimensional. (2) We now specialize Theorem 8.2 to G = Sp(2ℓ) and G/P = LG(ℓ, 2ℓ) the Lagrangian Grassmannian. Under the assumption that some structure coefficient of (H ∗(LG(ℓ, 2ℓ)), ⊙0) in the Schubert basis is equal to one, the conclusion of the theorem is that some Littlewood-Richardson coefficient is equal to one. In [R3], it is shown that this assumption is fulfilled if and only if some Littlewood-Richardson coefficient is equal to one. Hence by combining Theorem 8.2 and [R3], we obtain the following result on Littlewwod-Richardson coefficients. Let λ, µ and ν be three partitions. We assume that the Young diagrams of λ, µ and ν are con- tained in the square of size ℓ and are symmetric relative to the diagonal. Then, for the Littlewood- Richardson coeffcients for GL(ℓ), where λ′ and µ′ are obtained from λ and µ by adding one to some initial parts (for the details, see [R3]), and ν′ is defined by dualizing the rule. λ, µ = 1 ⇒ cν′ cν λ′, µ′ = 1, REFERENCES [B1] P. Belkale, Invariant theory of GL(n) and intersection theory of Grassmannians, IMRN 2004, no. 69, p. 3709– 3721. [B2] P. Belkale, Geometric proof of a conjecture of Fulton, Advances Math., vol. 216 (2007), p. 346–357. [BK] P. Belkale, S. Kumar, Eigenvalue problem and a new product in cohomology of flag varieties, Inventiones Math., vol. 166 (2006), p. 185–228. [Bo] N. Bourbaki, "Groupes et Alg`ebres de Lie", Chap. 4–6, Masson, Paris, 1981. [BP] M. Brion, P. Polo, Generic singularities of certain Schubert varieties, Math. Z., vol. 231 (1999), p. 301–324. [F1] W. Fulton, "Intersection Theory", Second Edition, Springer-Verlag, 1998. [F2] W. Fulton, "Young Tableaux", London Mathematical Society, Cambridge University Press, 1997. [FP] W. Fulton, P. Pragacz, "Schubert Varieties and Degeneracy Loci", Lecture Notes in Mathematics, vol. 1689, Springer-Verlag, Berlin, 1998. J.C. Jantzen, "Representations of Algebraic Groups", 2nd edn., Am. Math. Soc., 2003. [H] R. Hartshorne, "Algebraic geometry", Springer-Verlag, 1977. [J] [Kl] S.L. Kleiman, The transversality of a general translate, Compositio Math., vol. 28 (1974), p. 287–297. [KTW] A. Knutson, T. Tao, C. Woodward, The honeycomb model of GLn(C) tensor products II: Puzzles determine facets of the Littlewood-Richardson cone, J. Amer. Math. Soc., vol. 17 (2004), p. 19–48. [K] S. Kumar, "Kac-Moody Groups, their Flag Varieties and Representation Theory", Progress in Mathematics, vol. 204, Birkhauser, 2002. 20 PRAKASH BELKALE, SHRAWAN KUMAR, AND NICOLAS RESSAYRE [KLM] S. Kumar, B. Leeb, J. Millson, The generalized triangle inequalities for rank 3 symmetric spaces of noncom- pact type, Contemp. Math., vol. 332 (2003), p. 171–195. [M] D. Mumford, "The Red Book of Varieties and Schemes", Springer Lecture Notes in Mathematics, vol. 1358, 1988. [R1] N. Ressayre, Geometric invariant theory and the generalized eigenvalue problem, Inventiones Math., vol. 180 (2010), p. 389–411. [R2] N. Ressayre, A short geometric proof of a conjecture of Fulton, Preprint (2009). [R3] N. Ressayre, A cohomology free description of eigencones in type A, B and C Preprint (2009). Addresses: P.B.: Department of Mathematics, University of North Carolina, Chapel Hill, NC 27599-3250, USA (email: [email protected]) S.K.: Department of Mathematics, University of North Carolina, Chapel Hill, NC 27599-3250, USA (email: [email protected]) N.R.: Department of Mathematics, Universit´e Montpellier II, CC 51-Place Eug`ene Bataillon, 34095 Montpellier, Cedex 5, France (email: [email protected])
1308.4888	1	1308	2013-08-22T14:55:04	Conformal blocks, Berenstein-Zelevinsky triangles and group-based models	[ "math.AG", "math.RT" ]	Work of Buczynska, Wisniewski, Sturmfels and Xu, and the second author has linked the group-based phylogenetic statistical model associated with the group Z/2Z with the Wess-Zumino-Witten (WZW) model of conformal field theory associated to SL(2,C). In this article we explain how this connection can be generalized to establish a relationship between the phylogenetic statistical model for the cyclic group Z/mZ and the WZW model for the special linear group SL(m,C). We use this relationship to also show how a combinatorial device from representation theory, the Berenstein-Zelevinsky triangles, correspond to elements in the affine semigroup algebra of the Z/3Z phylogenetic statistical model.	math.AG	math	CONFORMAL BLOCKS, BERENSTEIN-ZELEVINSKY TRIANGLES AND GROUP-BASED MODELS KAIE KUBJAS, CHRISTOPHER MANON Abstract. Work of Buczy´nska, Wi´sniewski, Sturmfels and Xu, and the sec- ond author has linked the group-based phylogenetic statistical model associ- ated with the group Z/2Z with the Wess-Zumino-Witten (WZW) model of conformal field theory associated to SL2(C). In this article we explain how this connection can be generalized to establish a relationship between the phy- logenetic statistical model for the cyclic group Z/mZ and the WZW model for the special linear group SLm(C). We use this relationship to also show how a combinatorial device from representation theory, the Berenstein-Zelevinsky (BZ) triangles, correspond to elements in the affine semigroup algebra of the Z/3Z phylogenetic statistical model. Contents Introduction 1. 2. Essentials of conformal blocks 3. Conformal blocks and invariants in tensor products 4. Phylogenetic algebraic geometry 5. Conformal block algebras and group-based models 6. Berenstein-Zelevinsky triangles and group-based models References 1 3 8 11 14 16 23 1. Introduction In this paper we study an algebraic relationship between two branches of applied mathematics, phylogenetic algebraic geometry and the Wess-Zumino-Witten model of conformal field theory. Phylogenetics utilizes algebraic varieties associated with a statistical model in order to construct phylogenies between empirically observed taxa. A special class of these models, the group-based models, derive some of their parameters from a finite abelian group A. There is one such model for each graph Γ and finite abelian group A. We consider the coordinate algebra of the variety MΓ(A) associated with this model. The Wess-Zumino-Witten model of conformal field theory constructs a vector space VC,~p(~λ, L) for each stable marked projective curve C, ~p, and tuple of repre- sentation theoretic data ~λ, L for a simple Lie algebra g. Summing over the possible values of the data ~λ, L produces an infinite dimensional vector space VC,~p(G) which carries the structure of a commutative algebra. This algebra is the total coordinate Date: June 8, 2018. 1 2 KAIE KUBJAS, CHRISTOPHER MANON ring of the moduli space MC,~p(G) of principal G bundles on the curve C with quasi-parabolic structure at marked points ~p ⊂ C. A connection between these objects was first made in the paper [SX10] of Sturm- fels and Xu, where they construct a flat degeneration of VC,~p(SL2(C)) to the coor- dinate ring of each MT (Z/2Z) for C, a generic n-pointed rational curve, and T , a trivalent tree with n leaves. This follows work of Buczy´nska and Wi´sniewski, who proved in [BW07] that the Hilbert functions of these spaces all agree. The result of Buczy´nska and Wi´sniewski left open the question whether or not the toric ideals of the varieties MT (Z/2Z) lie in the same irreducible component of their Hilbert scheme, this question is answered in the positive by Sturmfels and Xu's result. In [Buc12], Buczy´nska goes on to define a statistical model MΓ(Z/2Z) for every trivalent graph Γ, and shows that the Hilbert function of this space depends only on the first Betti number and number of leaves of Γ. Corresponding structures appear in [Man09], where the second author constructs a flat degeneration VΓ(G) of the algebra VC,~p(G) for each graph Γ with n leaves and first Betti number g equal to the genus of C. A higher genus version of the theorem of Sturmfels and Xu is a corollary of these results, namely the algebra VΓ(SL2(C)) is isomorphic to the coordinate ring of MΓ(Z/2Z). In this paper we generalize this relationship to Z/mZ and SLm(C). The algebra VΓ(G) carries an action of T E(Γ), where T ⊂ G is a maximal torus, and E(Γ) is the edge set of Γ, we let SΓ(G) denote the semigroup of the characters for this action, and πΓ : VΓ(G) → C[SΓ(G)] be the projection map onto the corresponding semigroup algebra. The following is our main theorem. Theorem 1.1. There is an inclusion of semigroup algebras. (1) C[MΓ(Z/mZ)] ⊂ C[SΓ(SLm(C))] The results of Buczy´nska and Wi´sniewski on the group Z/2Z lead one to wonder whether or not the Hilbert functions of other group-based phylogenetic statistical models are invariant of the graph Γ. This question was addressed by the first author in [Kub12], where it was shown that this property fails for the group Z/2Z × Z/2Z. Donten-Bury and Micha lek then showed in [DM12] that this property also fails for the group Z/2Z × Z/2Z × Z/2Z and the cyclic groups Z/mZ for m = 3, 4, 5, 7, 8, 9. We observe in Theorem 3.5 that the first value of the Hilbert function is always independent of the graph Γ. We prove Theorem 1.1 by carefully studying the conformal blocks of level 1 for the special linear groups. The WZW model is closely related to the classical rep- resentation theory of SLm(C) (see for example [Uen08, Corollary 6.2]), this allows us to transfer the combinatorial devices of SLm(C) into the theory of phylogenetic statistical models. We will show that there is a natural correspondence between the degree one elements of the phylogenetic semigroup associated with Z/mZ and a set of so-called SLm(C) BZ triangles. The latter objects are tools for counting SLm(C) tensor product multiplicities. As a consequence, we obtain a graphical in- terpretation of the degree one elements of these phylogenetic semigroups. Moreover, we show that a graded semigroup algebra C[BZgr Γ (SL3(C))] of these BZ triangles surjects onto the Z/3Z phylogenetic statistical models. Theorem 1.2. There is a surjective map of semigroup algebras. (2) φΓ : C[BZgr Γ (SL3(C))] → C[MΓ(Z/3Z)] CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 3 This allows a convenient graphical device for studying the phylogenetic statistical model. We restrict ourselves to the case m = 3 here because, the map φΓ is an isomorphism for m = 2 by the results in [SX10] and [Man09], and it is not yet known if VC,~p(SLm(C)) has the combinatorial features enjoyed by the m = 2, 3 cases in general. In particular a toric degeneration of this algebra has not been found. As a consequence of the techniques presented in this paper, such a degeneration would also provide convenient graphical devices for studying the coordinate algebras of MΓ(Z/mZ). Graphical interpretations of BZ triangles have been crucial for solving problems in representation theory [KT99, GP00, KTW04]. We hope that the relation be- tween BZ triangles and group-based models, and the graphical interpretation of group-based semigroups will provide a new approach to tackle open problems in phylogenetic algebraic geometry such as the Sturmfels-Sullivant conjecture about the maximal degrees of minimal generators of group-based ideals [SS05, Conjec- ture 29]. Sections 2 to 4 will be introductory chapters defining basic objects in this ar- ticle. In Section 2, we will define conformal block algebras, and in Section 3, we will study their representation theoretic properties. In Section 4, the introduction to phylogenetic algebraic geometry and to group-based models will be given. In Section 5, we will prove Theorem 1.1. In Section 6, we will establish a connection between BZ triangles and group-based models, and prove Theorem 1.2. 2. Essentials of conformal blocks 2.1. Basics of SLm(C) representation theory. We recall a few basics of the representation theory of the special linear group and its Lie algebra, for what follows see the book of Fulton and Harris, [FH91]. Recall that SLm(C) is the set of m × m invertible matrices with determinant 1, and its Lie algebra slm(C) is the set of m × m matrices with trace 0. A representation of these objects is a vector space V along with a map of groups SLm(C) → GL(V ), and respectively a map of Lie algebras slm(C) → Mm×m(C). Representations form a category, and it is a classical result that the categories of finite dimensional representations of SLm(C) and slm(C) are the same. Repre- sentations can be combined in a number of ways, most important to us are the direct sum V ⊕ W and the tensor product V ⊗ W, which are given by the expected operations on the underlying vector spaces. When a representation has no proper sub-representations, it is said to be irre- ducible. It can be shown that any representation in Rep(m) can be written as a direct sum of irreducible representations in a unique way. For this reason, a clas- sification of the irreducible representations of SLm(C) suffices to describe Rep(m). These irreducibles are in bijection with decreasing lists of non-negative integers of length m which end in 0. (3) (λ1, . . . , λm−1, 0) = λ → V (λ) Such lists are known as weights, and they are the integer points of simplicial cone ∆. This set is generated under addition by the weights ω1 = (1, . . . , 0), ω2 = (1, 1, . . . , 0), . . . , ωm−1 = (1, . . . , 1, 0). These are known as fundamental weights, and they correspond to the exterior powers of Cm. 4 KAIE KUBJAS, CHRISTOPHER MANON T Γ E(Γ) V (Γ) MΓ(A) PT (A) LT (A) RΓ(A) Lgr Γ (A) RΓ(A) Rpr Γ (A) SLm(C) slm(C) ωi ∆ λ ∈ ∆ V (λ) L ∈ Z≥0 ∆L αij Wm Rm Mg,n (C, ~p) ∈ Mg,n VC,~p(~λ, L) VC,~p(G) VΓ(G) V (G) SΓ(SLm(C)) BZ(SLm(C)) pr(BZ(SLm(C))) BZgr(SLm(C)) a tree a graph the set of edges of Γ the set of non-leaf vertices of Γ the algebraic variety associated with a group-based model with abelian group A and graph Γ the polytope associated with a group-based model with abelian group A and tree T the lattice generated by the vertices of PT (A) the graded affine semigroup associated with a group-based model with abelian group A and graph Γ the graded lattice associated with a group-based model with abelian group A and graph Γ the saturation of RΓ(A) the projection of RΓ(A) that forgets the grading the special linear group the Lie algebra of SLm(C) a fundamental weight of slm(C) a Weyl chamber of slm(C) a dominant weight of slm(C) the irreducible representation of SLm(C) with highest weight λ the level the level L alcove of ∆ a positive root of slm(C) the weight lattice of slm(C) the root lattice of slm(C) the moduli of stable genus g, n-marked curves a n-marked genus g stable curve the space of ~λ conformal blocks on (C, ~p) of level L the direct sum of the spaces of conformal blocks on (C, ~p) the degeneration of VC,~p(G) associated to Γ the algebra of conformal blocks the affine semigroup associated to VΓ(SLm(C)) the affine semigroup of BZ triangles for SLm(C) the projection of BZ(SLm(C)) to boundaries the graded affine semigroup of BZ triangles for SLm(C) Table 1. Table of notation (4) V (ωk) = k ^(Cm) Any representation V (λ) can be viewed as a (C∗)m−1 representation by restrict- ing the action of SLm(C) to its maximal torus of diagonal matrices. Since this is an abelian group, V (λ) decomposes into 1-dimensional (C∗)m−1 isotypical spaces, called weight spaces. Each such space is labelled by a corresponding (C∗)m−1 char- acter, which can be identified with an (m − 1)-tuple of integers. The set of all CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 5 such tuples forms a lattice Wm in the linear space Rm/(1, . . . , 1)R called the weight lattice. This lattice is generated by the fundamental weights ωi. The weights of the representation V (ωk) are easy to compute, they correspond to the exterior powers zi1 ∧ . . . ∧ zik , and are in bijection with m-length 0, 1 vectors with exactly k 1′s. In particular, the subspace (the "weight space") of V (ωk) with a particular weight is always 1-dimensional. The weight lattice contains the vectors αij = (0, . . . , 1, . . . , −1, . . . 0), 1 ≤ i < j ≤ m, which are known as the roots of the Lie algebra slm(C), i.e. these are the characters which appear in the decomposition of slm(C) as a representation of the group of diagonal matrices in SLm(C). The root lattice Rm ⊂ Wm is the lattice generated by the roots. It is a simple exercise to show the following. (5) Z/mZ = Wm/Rm As we will see, this is the principal reason that SLm(C) conformal blocks are related to phylogenetic statistical models for the group Z/mZ. 2.2. Graphs and curves. In this subsection we recall some of the combinatorial notions needed to discuss algebras of conformal blocks and phylogenetic statistical models. From now on Γ denotes a graph, we say Γ is of type g, n if its first Betti number is g and it has n leaves. We let V (Γ) be the set of non-leaf vertices of Γ. For two graphs Γ, Γ′ of the same type, we define a map of graphs φ : Γ′ → Γ to be a map on the underlying graphs which preserves leaves, obtained by collapsing a set of non-leaf edges of Γ′. Figure 1. The graph on the right is obtained by collapsing the highlighted edges on the left For a graph Γ we also define a forest Γ, obtained from Γ by splitting each non-leaf edge of Γ. The connected components Γv of Γ correspond to the non-leaf vertices v ∈ V (Γ), and each has one distinguished non-leaf vertex. We let πΓ : Γ → Γ be the quotient map which "glues" split edges back together. Graphs Γ act as combinatorial models for points in the moduli of stable curves Mg,n of genus g with n marked points. This stack has a stratification by substacks M(Γ), and a curve C, ~p ∈ M(Γ) ⊂ Mg,n is said to be of type Γ. Such a curve is generally reducible, with one irreducible component Cv, ~pv, ~qv for each non-leaf vertex v ∈ V (Γ). Here ~pv are the marked points of C which lie on the component Cv, 6 KAIE KUBJAS, CHRISTOPHER MANON and ~qv are the points on the component Cv which are shared by other components of C. The edges of the labelled tree Γv are in bijection with the points ~pv, ~qv. Figure 2. The forest Γ, along with the normalization of the as- sociated stable curve For a reducible curve C, ~p of type Γ, we can form the normalization C, ~p, ~q by splitting each nodal singularity into a pair of points, creating a disconnected, marked curve. Each pair of points introduced by the normalization become new marked points in C. The combinatorial type of C, ~p, ~q is given by the forest Γ, and the connected components Ci of C are the irreducible components of C, and correspond to the trees Γvi . The poset of the stratification of Mg,n given by the substacks M(Γ) is deter- mined by the graph maps π : Γ′ → Γ. The component M(Γ′) is in the closure of M(Γ) if and only if there is a graph map π : Γ′ → Γ. The lowest strata of Mg,n are given by trivalent graphs Γ, these are all isolated points in the moduli Mg,n. 2.3. Conformal Blocks. In this subsection we recall some properties of the alge- bra of conformal blocks, and define the semigroup SΓ(SLm(C)). Vector spaces of conformal blocks arise from the Wess-Zumino-Witten (WZW) model of conformal field theory. We let ∆L be the simplex {λλ(α1,m) ≤ L}, where α1,m is the longest root of slm(C), this polytope is called the level L alcove. There is a conformal blocks vector space VC,~p(~λ, L) for each n-marked genus g stable curve C, ~p ∈ Mg,n, set of n dominant weights ~λ = {λ1, . . . , λn}, pi → λi, and non-negative integer L, known as the level. Vectors in these spaces are possible partition functions for the WZW model on C. In [TUY89], Tsuchiya, Ueno, and Yamada show that for fixed ~λ, L, these spaces knit together into a vector bundle over Mg,n. This theorem is modified by the second author in [Man09] with the following theorem. Theorem 2.1. There is a flat sheaf V (SLm(C)) of algebras on the moduli space MC,~p of stable curves, such that the fiber over a point C, ~p is isomorphic to the direct sum VC,~p(SLm(C)) = L~λ,L VC,~p(~λ, L) of SLm(C) conformal blocks on C, ~p. CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 7 This theorem, along with work of Kumar, Narasimhan, Ramanathan [KNR94], Pauly [Pau96], Beauville, Laszlo, [BL94], and Faltings [Fal94], establishes that over a smooth curve C, ~p, the algebra VC,~p(SLm(C)) is isomorphic to the total coordinate ring of the moduli of rank m vector bundles on C with parabolic structure at the marked points ~p. Conformal blocks come with a number of structural features. For a graph Γ and a stable curve C, ~p of type Γ we can take the smooth normalization, C, ~p, ~q1, ~q2 by pulling apart the double point singularities, resulting in pairs of marked points ~q1, ~q2. The following theorem expresses how conformal blocks behave with respect to this operation, it is due to Tsuchiya, Ueno, and Yamada, [TUY89] and also Faltings [Fal94]. Theorem 2.2. For C, ~p a stable curve with normalization C, ~p, ~q1, ~q2, there is an isomorphism of vector spaces, canonical up to choice of scalar for each ~η = (η1, . . . , ηm) summand, where ~qi = m. (6) VC,~p(~λ, L) = M ηj ∈∆L V C,~p,~q1,~q2(~λ, ~η, ~η∗, L) Here the sum is over all assignments of weights ηj ∈ ∆L. This theorem allows us to express properties about conformal blocks over a In particular, for every graph Γ general curve C, ~p in terms of simpler curves. with first Betti number equal to g and n leaves there is a stable curve C, ~p with normalization C, ~p, ~q1, ~q2 equal to a disjoint union of marked copies of the rational curve P1. Spaces of conformal blocks over non-connected curves are tensor products of the spaces of conformal blocks over their connected components, so this implies that we can effectively study the conformal blocks for any curve by understanding these spaces over marked projective spaces. This approach extends to the algebraic structure introduced in [Man09]. We fix a marked curve C, ~p of type Γ with normalization C, ~p, ~q, and we consider the following tensor product. (7) O v∈V (Γ) VCv ,~pv ,~qv (SLm(C)) This algebra is a multigraded sum of tensor product spaces of conformal blocks. (8) VCv ,~pv,~qv (~λv, ~ηv, Lv) O v∈V (Γ) Here ~λv are the dominant weights assigned to the marked points of Cv which cor- respond to marked points on the original curve C, and ~ηv are the dominant weights assigned to marked points which are introduced by the normalization procedure. Recall that there is a way to pair these weights ηvi ↔ ηvj . Definition 2.3. The algebra VΓ(SLm(C)) is the sub-algebra of O v∈V (Γ) VCv ,~pv ,~qv (SLm(C)) 8 KAIE KUBJAS, CHRISTOPHER MANON defined by the components VCv ,~pv,~qv (~λv, ~ηv, Lv) with Lvi = Lvj for all v ∈ V (Γ), and ηvi = η∗ vj for all paired dominant weights. This algebra is related to the algebra of conformal blocks VC,~p(SLm(C)) for a curve of genus g with n markings by the following theorem from [Man09]. Theorem 2.4. There is a flat degeneration of commutative algebras VC,~p(SLm(C)) to VΓ(SLm(C)). Definition 2.5. We define the semigroup SΓ(SLm(C)) ⊂ ∆E(Γ) × Z≥0 to be the set of pairs (ω, L), where ω : E(Γ) → ∆E(Γ) and L ∈ Z≥0, such that the component of VΓ(SLm(C)) defined by (ω, L) is non-zero. There is a map of algebras, VΓ(SLm(C)) → C[SΓ(SLm(C))] given by sending conformal blocks to their weight and level information. Note that if π : Γ′ → Γ is a map of graphs, then there is a corresponding map of semigroups SΓ′(SLm(C)) → SΓ(SLm(C)) defined by forgetting the weights on the edges collapsed by π. 3. Conformal blocks and invariants in tensor products In this section we discuss the construction of the space of conformal blocks over a genus 0, triple marked curve. This involves spaces of invariant vectors in tensor products of SLm(C) representations in a non-trivial way. We use this construction to revisit a well-known combinatorial classification of level 1 SLm(C) conformal blocks. The combinatorics of these special conformal blocks allow us to make the connection with phylogenetic statistical models in Section 5. 3.1. Spaces of conformal blocks in spaces of invariants. In order to build the space V0,3(λ, η, µ, L) of conformal blocks we study the action of SLm(C) on the tensor product V (λ)⊗ V (η)⊗ V (µ). We let [V (λ)⊗ V (η)⊗ V (µ)]SLm(C) be the space of vectors fixed by this action. The space of conformal blocks can be constructed as a subspace of this space of invariants. (9) V0,3(λ, η, µ, L) ⊂ [V (λ) ⊗ V (η) ⊗ V (µ)]SLm(C) We introduce the subgroup SL2(1, m) ⊂ SLm(C), this is the copy of SL2(C) inside of SLm(C) which acts on the basis vectors with indices 1, m in the standard representation of SLm(C). The group SL2(1, m) is composed of the matrices of the following form.   a 0 · · · b 1 · · · 0 0 ... ... ... . . . 0 · · · d c   The subgroup SL2(1, m) corresponds to the longest root of SLm(C) in its stan- dard ordering of roots. By way of the inclusion SL2(1, m) ⊂ SLm(C), every SLm(C) representation can be restricted to a representation of SL2(1, m), and decomposed along the irreducible representations of SL2(C). (10) V (λ) = M i Wλ,i ⊗ V (i) CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 9 V (η) = M j V (µ) = M k Wη,j ⊗ V (j) Wµ,k ⊗ V (k) We let WL ⊂ V (λ) ⊗ V (η) ⊗ V (µ) be the following subspace, (11) M i+j+k≤2L [Wλ,i ⊗ V (i)] ⊗ [Wη,j ⊗ V (j)] ⊗ [Wµ,k ⊗ V (k)]. For the following proposition see [Uen08, Corollary 6.2]. Proposition 3.1. The space of conformal blocks V0,3(λ, η, µ, L) can be identified with the following subspace of [V (λ) ⊗ V (η) ⊗ V (µ)]SLm(C), (12) V0,3(λ, η, µ, L) = WL ∩ [V (λ) ⊗ V (η) ⊗ V (µ)]SLm(C). 3.2. Level 1, genus 0, 3-marked conformal blocks. A consequence of Proposi- tion 3.1 is that the dimension of V0,3(λ, η, µ, L) is bounded above by the dimension of the space [V (λ) ⊗ V (η) ⊗ V (µ)]SLm(C). This is particularly useful in the case we consider here, when the level L is restricted to be 1. The level 1 alcove ∆1 is precisely the 0 weight and the fundamental weights ωi, so all conformal blocks we consider will now have only these weights in their weight datum. We make use of the following fact from the representation theory of SLm(C). Lemma 3.2. The tensor product invariant space [V (λ) ⊗ V (ωi) ⊗ V (µ)]SLm(C) is multiplicity free, where ωi is the i-th fundamental weight of SLm(C). Proof. By a general theorem due to Zhelobenko [Zhe73] (see also [Man10]), the space [V (λ) ⊗ V (ωi) ⊗ V (µ)]SLm(C) can be identified with a subspace of a weight space of V (ωi) = Vi(Cm). As we remarked above, all weight spaces of these repre- sentations are multiplicity free. (cid:3) Proposition 3.3. The space V (ωi) ⊗ V (ωj) ⊗ V (ωk) has an invariant SLm(C) vector if and only if i + j + k = 0 modulo m. Proof. If the tensor product V (ωi) ⊗ V (ωj) ⊗ V (ωk) has an invariant, then the weight ωj + ωk − ωm−i is a member of the root lattice of SLm(C). This means that the class [ωj] + [ωk] − [ωm−i] = 0 in the group Wm/Rm. As Wm/Rm = Z/mZ with [ωi] = i, this implies that there is an invariant vector in V (ωi) ⊗ V (ωj) ⊗ V (ωk) only if i + j + k = 0 modulo m. This proves that the condition i + j + k ∈ mZ is necessary. For sufficiency, note that i + j + k above must be m or 2m. The space V (ωi) ⊗ V (ωj)⊗V (ωk) has an invariant vector if and only if V (ωm−i)⊗V (ωm−j)⊗V (ωm−k) does as well by duality, so we assume without loss of generality that i + j + k = m. The representation V (ωk) is the exterior product Vk(Cm) as a vector space, and has a basis of exterior forms zI = zi1 ∧ . . . ∧ zik , I = {i1, . . . , ik}. The determinant form z1 ∧ . . . ∧ zm can be expressed in these basis members. (13) z1 ∧ . . . ∧ zm = X I ` J ` K=[m] (−1)σ(I,J,K)zI ⊗ zJ ⊗ zK 10 KAIE KUBJAS, CHRISTOPHER MANON Here (−1)σ(I,J,K) is a sign that depends on the indices I, J, K. Recall that the determinant form is the unique basis member of Vm(Cm), which is the trivial representation of SLm(C). This implies that V (ωi)⊗V (ωj)⊗V (ωk) has an invariant vector when i + j + k = m, and when i + j + k = 2m by duality. (cid:3) Now we can determine the dimension of the spaces V0,3(ωi, ωj, ωk, 1). Note that this space has dimension 1 or 0, and the same duality considerations allow us to restrict to the case i + j + k = m, as above. Proposition 3.4. The space V0,3(ωi, ωj, ωk, 1) with i + j + k = 0 modulo m has dimension 1. Proof. We determine the space Wωi,s ⊗ V (s) ⊗ Wωj ,r ⊗ V (r) ⊗ Wωk ,t ⊗ V (t) ⊂ V (ωi) ⊗ V (ωj) ⊗ V (ωk) which contains the element zI ⊗ zJ ⊗ zK for each element in the expansion of the determinant in Equation (13). This is determined by which sets I, J, K contain the indices 1 and m. If both 1, m are contained in the same index set, say J, then this form restricts to the determinant form z1 ∧ zm of SL2(1, m). This implies that zI ⊗ zJ ⊗ zK lies in a component of the form Wωi,0 ⊗ V (0) ⊗ Wωj ,0 ⊗ V (0) ⊗ Wωk ,0 ⊗ V (0). If 1, m are in different index sets, say I, K, then these give basis members z1 and zm of the representation V (1) of SL2(1, m). This implies that zI ⊗ zJ ⊗ zK lies in a component of the form Wωi,1 ⊗ V (1) ⊗ Wωj ,0 ⊗ V (0) ⊗ Wωk,1 ⊗ V (1). It follows that the maximum value obtained by s + r + t is 2, and this value is obtained. (cid:3) 3.3. Level 1 conformal blocks. Now we extend the classification of level 1 con- formal blocks to curves C of any genus and number of markings. Our principal tool is the factorization property, Theorem 2.2. This allows us to express a space of level 1 conformal blocks VC,~p(ωi1 , . . . , ωin , 1) in terms of the spaces of genus 0, 3-marked conformal blocks studied above. There is one such expression for each trivalent graph Γ. (14) VC,~p(ωi1 , . . . , ωin , 1) = M[ O v∈V (Γ) V0,3(ωiv , ωjv , ωkv , 1)] Each of the tensor product spaces appearing on the right hand side of this equa- tion has dimension 1 or 0, so the space VC,~p(ωi1 , . . . , ωin , 1) has a basis in bijection with the set of labellings of the edges of Γ by fundamental weights such that the following conditions are satisfied. (1) For each v ∈ V (Γ) iv + jv + kv = 0 modulo m. (2) If ωjv and ωkv′ are assigned to edges which agree under π : Γ → Γ, then jv + kv′ = 0 modulo m. The second condition above follows from duality. This allows us to combinato- rially determine the dimension of any space of level 1 SLm(C) conformal blocks. Theorem 3.5. The space VC,~p(ωi1 , . . . , ωin , 1) has dimension 0 or mg. dimension mg precisely when P ij = 0 modulo m. Proof. It is a luxury of the factorization theorem that we may use any trivalent graph Γ we wish in the analysis of the dimension of VC,~p(ωi1 , . . . , ωin , 1). We employ the graph Γg,n, depicted in Figure 3. It has CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 11 ... ... Figure 3. The graph Γg,n The paired edges in Γg,n which correspond to the loops must each be assigned dual fundamental weights. This forces the third edge in each of these tripods to be assigned 0. This forces each edge to the left of e to be assigned a 0. Note that there are mg such assignments of fundamental weights and their duals on these loop edges. This reduces the problem to showing that when g = 0, there is a unique labelling if and only if P ij = 0 modulo m. We have already seen that this is the case if n = 3. The general result follows from the observation that in the case of a trivalent tree T , an assignment of weights to the leaves of T determines the assignment on the rest of the tree by the equation iv + jv + kv = 0 modulo m. This follows because every trivalent tree T must have a vertex v which is connected to two leaves. Using the equation iv + jv + kv = 0 modulo m to force the weighting on the third edge connected to this tripod reduces the case by 1 leaf, so we obtain the result by induction. (cid:3) There are mn−1 n-tuples of fundamental weights (ωi1 , . . . , ωin ) such that P ij = 0 modulo m. It follows that the space of all SLm(C) conformal blocks of level 1 has dimension mg+n−1. 4. Phylogenetic algebraic geometry Phylogenetic algebraic geometry studies algebraic varieties associated with sta- tistical models on phylogenetic trees, see [ERSS05] for an introductory article on the topic. These statistical models on phylogenetic trees can be parametrized by polynomial maps that give the explicit description of the associated algebraic vari- ety. Group-based models are statistical models invariant under the action of a finite abelian group. Algebraic varieties associated with group-based models are toric af- ter the change of coordinates given by the discrete Fourier transform [ES93, SSE93]. By the well-known correspondence between toric varieties and polyhedral fans, many properties of group-based models can be studied by considering correspond- ing discrete objects. In the following, we will define lattice polytopes and affine semigroups associated with group-based models. Definition 4.1. Let A be an additive finite abelian group and T a tree. Let T be the forest obtained from T by splitting every non-leaf edge of T . Label the edges of T with unit vectors indexed by the elements of A such that 12 KAIE KUBJAS, CHRISTOPHER MANON (1) at each inner vertex the indices of unit vectors sum up to zero, (2) the indices of unit vectors corresponding to an inner edge of T sum up to zero. Define the polytope PT (A) as the convex hull of vectors corresponding to all such labellings of the edges of T . Define the lattice LT (A) to be the lattice generated by the vertices of PT (A). We denote the polytope associated with a group A and the tripod by P0,3(A) and the corresponding lattice by L0,3(A). Let e be an edge of T and a an element of A. Denote the coordinate correspond- ing to the tuple (e, a) by xe a. For x ∈ PT (A), we have (15) xe a = 1. X a∈A Hence the projection of PT (A) that forgets the coordinates corresponding to 0 ∈ A is lattice equivalent to PT (A). From now on we will consider this projection of PT (A) and we will also denote it by PT (A). Example 4.2. The polytope associated with Z/2Z and a claw tree T is (16) PT (Z/2Z) = conv{x ∈ {0, 1}E(T ) X e∈E(T ) xe is even}, where E(T ) denotes the set of edges of T . In particular, the polytope associated with Z/2Z and tripod is (17) P0,3(Z/2Z) = conv{(0, 0, 0), (0, 1, 1), (1, 0, 1), (0, 1, 1)}. The group-based model associated with group Z/2Z is called the Jukes-Cantor binary model. Note the similarity of the conditions (1) and (2) in Definition 4.1 with the condi- tions at the beginning of Section 3.3. These conditions will be the main ingredient for establishing the relationship between group-based models, conformal block al- gebras and BZ triangles in Sections 5 and 6. Definition 4.1 differs from the standard definition of a lattice polytope associated with a group-based model and a tree. For the standard definition, one needs to direct the edges of T and the definition depends on the orientation we choose, see for example [SS05] and [Sul07, Section 3.4]. Different ways of directing the edges of T give lattice equivalent lattice polytopes. Also the polytope in Definition 4.1 is lattice equivalent to all of them. The advantage of Definition 4.1 is that it does not depend on the orientation of T we choose. This makes it simpler to establish a connection with conformal block algebras and BZ triangles. Moreover, Definition 4.1 makes it possible to define affine semigroups associated with a group A and a graph Γ generalizing the definition of the phylogenetic semigroup for the group Z/2Z and a graph Γ in [Buc12, BBKM13]. Definition 4.3. The affine semigroup associated with an additive finite abelian group A and a tree T is RT (A) = N((PT (A) ∩ LT (A)) × {1}). We denote the semigroup associated with a group A and the tripod by R0,3(A). The associated graded lattice is Lgr T (A) = LT (A) ⊕ Z. CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 13 Using the definition for trees, we can define the semigroup associated with a group-based model and a graph. Let Γ be a graph with first Betti number g. We associate a tree T with g pairs of distinguished leaves with Γ. Choose an edge e1 of Γ such that splitting e1 gives a graph with first Betti number g − 1. Denote the new leaf edges obtained from e1 by e1 and e1. Repeating this operation, we get an associated tree T with g pairs of distinguished leaves {e1, e1}, {e2, e2}, . . . , {eg, eg}. Definition 4.4. The affine semigroup associated with an additive finite abelian group A and a graph Γ is RΓ(A) = RT (A) ∩ \ i∈{1,...,g} and a∈A The associated graded lattice is Lgr Γ (A) = Lgr T (A) ∩ \ i∈{1,...,g} and a∈A ker(xei a − x ei −a). ker(xei a − x ei −a). Although the tree T associated with Γ is not unique, the definitions of RΓ(A) and Lgr Γ (A) do not depend on the choices we make. Lemma 4.5. Let A be an additive finite abelian group and Γ a graph. Denote the edges obtained by splitting an inner edge e of Γ by e and e. Then (18) RΓ(A) = Y v∈V (Γ) R0,3(A) ∩ \ e an inner edge and a∈A where V (Γ) is the set of non-leaf vertices of Γ. ker(xe a − xe −a), Proof. The statement follows from Definitions 4.1, 4.3 and 4.4. (cid:3) The algebraic variety MΓ(A) associated with an abelian group A and a graph Γ is defined as the projective spectrum Proj (C[cone(RΓ(A)) ∩ Lgr Γ (A)]). We will show that the affine semigroup cone(RΓ(A)) ∩ Lgr Γ (A) is the saturation of the affine semigroup RΓ(A). Recall that an affine semigroup S ⊂ L is said to be saturated if m · u ∈ S if and only if u ∈ S for all m ∈ Z≥0. An affine semigroup can always be completed to a saturated affine semigroup S by formally including all u ∈ L such that m · u ∈ S. Proposition 4.6. Let Γ be a graph and A an additive abelian group. Then RΓ(A) = cone(RΓ(A)) ∩ Lgr (19) Proof. We have RΓ(A) ⊆ cone(RΓ(A)) ∩ Lgr Γ (A) and cone(RΓ(A)) ∩ Lgr Lgr Γ (A) is saturated. Take a degree k element u ∈ cone(RΓ(A)) ∩ Lgr Γ (A). Γ (A), since RΓ(A) ⊆ cone(RΓ(A)) ∩ k u is a rational convex combination of degree 1 elements of RΓ(A). Let m be the lowest com- mon denominator of the coefficients in this convex combination. Then (km) · u ∈ RΓ(A) ⊆ RΓ(A). Since RΓ(A) is saturated, we have u ∈ RΓ(A). This proves RΓ(A) ⊇ cone(RΓ(A)) ∩ Lgr (cid:3) Γ (A). Then 1 Γ (A). Corollary 4.7. Let Γ be a graph and A an additive abelian group. Then (20) MΓ(A) = Proj (C[RΓ(A)]). Connections between group-based models, conformal block algebras and BZ triangles have developed from a result about the Jukes-Cantor binary model by Buczy´nska and Wi´sniewski, and its generalization by Buczy´nska. 14 KAIE KUBJAS, CHRISTOPHER MANON Theorem 4.8 ([BW07] Theorem 2.24, [Buc12] Theorem 3.5). The Hilbert polyno- mial of the affine semigroup associated with the Jukes-Cantor binary model on a trivalent graph depends only on the number of leaves and the first Betti number of the graph. It has been shown by the first author [Kub12], Donten-Bury and Micha lek [DM12] that this theorem cannot be generalized to group-based models with underlying groups Z2 × Z2, Z2 × Z2 × Z2, Z3, Z4, Z5, Z7, Z8, or Z9. This suggests that the invari- ance of the Hilbert polynomial of the Jukes-Cantor binary model is a consequence of the flat degenerations of SL2(C) conformal block algebras constructed by Sturm- fels and Xu [SX10], and the second author [Man09], and is not something typical to group-based models. This hypothesis was further confirmed by the second author in [Man12], where he constructs flat degenerations of SL3(C) conformal block alge- bras to semigroup algebras associated with SL3(C) BZ triangles on a graph. The Hilbert polynomials of these semigroups depend only on the combinatorial data of the graph, hence SL3(C) BZ triangles on a trivalent graph generalize Theorem 4.8. These results motivated our investigation of relations between group-based models, BZ triangles and conformal block algebras. In particular, we will show in the next sections that the affine semigroup associated with SL3(C) BZ triangles is closely related to the phylogenetic semigroup associated with Z/3Z. 5. Conformal block algebras and group-based models In this section we start by defining the map from the phylogenetic semigroup RΓ(Z/mZ) into the semigroup SΓ(SLm(C)). Theorem 5.1. Let Γ be a graph. Then (21) RΓ(Z/mZ) ⊂ SΓ(SLm(C)). Proof. For a tripod, R0,3(Z/mZ) ⊂ S0,3(SLm(C)) ⊂ ∆3 × Z≥0 is established by results in Section 3.3. Here ∆ is a Weyl chamber, considered as a simplicial cone, with extremal rays generated by the unit vectors e1, . . . , em−1. The semigroup R0,3(Z/mZ) is the sub-semigroup of this cone generated by elements in ∆3 × {1} where each edge is assigned the generator of such an extremal ray or e0 = (0, . . . , 0), such that the sum of indices i from the assignment of these generators is 0 modulo m. Each of these generators corresponds to a unique element of VΓ(SLm(C)). The unit vector ei is replaced by the fundamental weight ωi. This fundamental weight has image [i] in the quotient group Z/mZ = Wm/Rm. Similarly, the dual weight ω∗ i = ωm−i has image [m − i] in Z/mZ. For a general graph Γ, we form the covering forest Γ, and note that the following inclusions are immediate. (22) RΓ(Z/mZ) ⊂ SΓ(SLm(C)) ⊂ ∆3V (Γ) × Z≥0 For each pair of edges e, e′ ∈ Γ which are identified by its map to Γ, there is a subcone He,e′ ⊂ ∆3V (Γ) × Z≥0 of points with e weight dual to e′ weight. The theorem now follows from the fact that RΓ(Z/mZ) ⊂ SΓ(SLm(C)) ⊂ ∆E(Γ) × Z≥0 are obtained from RΓ(Z/mZ) ⊂ SΓ(SLm(C)) ⊂ ∆3V (Γ) × Z≥0 by intersecting everything in sight with the subcones He,e′ . (cid:3) CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 15 Corollary 5.2. Let Γ be a graph. Then (23) RΓ(Z/mZ) ⊂ SΓ(SLm(C)). Proof. A deep result of Belkale [Bel03, Section 3.1] implies that the semigroup SΓ(SLm(C)) is saturated, so we also get an inclusion of the saturation RΓ(Z/mZ) in SΓ(SLm(C)). (cid:3) Proof of Theorem 1.1. Corollary 5.2 gives a corresponding inclusion on semigroup algebras, (24) C[RΓ(Z/mZ)] ⊂ C[SΓ(SLm(C))]. Theorem 1.1 follows from Equation (24) and Corollary 4.7. (cid:3) 5.1. VΓ(SL2(C)). In the case m = 2, the spaces of conformal blocks V0,3(λ, µ, ν, L) for a genus 0, triple marked curve are known to be multiplicity free. This follows from Proposition 3.1 above, and the fact that invariant spaces for triple tensor products of SL2(C) representations are also multiplicity free. This implies that the algebra V0,3(SL2(C)) is an affine semigroup algebra on the semigroup given by the weight data λ, µ, ν, L for which the space V0,3(λ, µ, ν, L) is non-zero. This semi- group can be shown to be generated by the vectors (0, 0, 0, 1), (1, 1, 0, 1), (1, 0, 1, 1), and (0, 1, 1, 1) ∈ R4, where the last coordinate is the level coordinate L. The multiplicity-free property also implies that we have an isomorphism of commuta- tive algebras, (25) VΓ(SL2(C)) ∼= C[SΓ(SL2(C))]. Furthermore, by Example 4.2 we know that R0,3(Z/2Z) is generated by (0, 0, 0, 1), (1, 1, 0, 1), (1, 0, 1, 1), and (0, 1, 1, 1) ∈ R4, thus (26) VΓ(SL2(C)) ∼= C[RΓ(Z/2Z)]. The semigroup algebra C[RΓ(Z/2Z)] is analyzed by Buczy´nska, Buczy´nski, the first author, and Micha lek in [BBKM13], where they show it is always generated by elements with L ≤ g(Γ) + 1. In [Man09], a particular graph Γg,n is found for each g, n such that C[SΓg,n (SL2(C))] is generated by those elements with L = 1, 2. In [Man12] it is also shown that this degree of generation is strict when g, n > 1 or 4 ≤ g. 5.2. VΓ(SL3(C)). The case m = 3 is slightly more complicated than the m = 2 case in that Equation (25) no longer holds. Unlike in the m = 2 case, the algebra of conformal blocks V0,3(SL3(C)) is not a semigroup algebra. It has the following presentation by its L = 1 component, see [Man12], (27) where (28) V0,3(SL3(C)) = C[X, Y, Z, P12, P23, P31, P21, P32, P13]/I, I =< XY Z − P12P23P31 + P21P32P13 > . Note that three binomial relations can be obtained as degenerations of this alge- bra, XY Z − P12P23P31, XY Z + P21P32P13, and P12P23P31 − P21P32P13. The ideal generated by the union of these three relations cuts out the phylogenetic statistical model C[R0,3(Z/3Z)]. 16 KAIE KUBJAS, CHRISTOPHER MANON The Corollary 6.10 implies that there is a surjection (29) VT (SL3(C)) → C[ST (SL3(C))] ∼= C[RT (Z/3Z)]. Once again, it can be shown that when T is trivalent, the toric ideal which cuts out ST (SL3(C)) ∼= RT (Z/3Z) is generated by a union of the 3V (T ) binomial ideals obtained from the degenerations of VT (SL3(C)) constructed in [Man12]. We do not know at this time if a similar statement holds for T non-trivalent. Little is known about the semigroups RΓ(Z/3Z), SΓ(SL3(C)) or the algebra VΓ(SL3(C)) outside of the genus 0 case. For a particular graph Γ1,n it is shown that VΓ1,n (SL3(C)) is generated by conformal blocks of levels 1, 2, 3 in [Man12]. It would be interesting to obtain a general function which bounds the number of necessary generators for all Γ for both, the algebra VΓ(SL3(C)) and the semigroup SΓ(SL3(C)). 6. Berenstein-Zelevinsky triangles and group-based models BZ triangles provide a combinatorial tool for determining dimensions of triple tensor product invariants. There are several graphical interpretations of BZ tri- angles such as honeycomb diagrams of Knutson and Tao [KT99], web diagrams of Gleizer and Postnikov [GP00], and an alternate version of honeycombs by the second author and Zhou [MZ12]. In this section, we will show that there is a nat- ural injection from the degree one elements of phylogenetic semigroups associated with Z/mZ to SLm(C) BZ triangles. As a consequence, we obtain a graphical interpretation of the degree one elements of these phylogenetic semigroups. It would be interesting to find a graphical interpretation of R0,3(Z/mZ). If R0,3(Z/mZ) is not normal, then we need a graphical description of higher degree minimal generators of R0,3(Z/mZ). Even if R0,3(Z/mZ) is normal, then higher degree elements of R0,3(Z/mZ) might have several different honeycombs associated with it. This is caused by the fact that there are less algebraic relations between BZ triangles than between elements of phylogenetic semigroups. We hope that the connection between phylogenetic semigroups and BZ triangles will motivate further research on graphical interpretations of phylogenetic semigroups to approach open problems like the Sturmfels-Sullivant conjecture [SS05, Conjecture 29]. 6.1. BZ Triangles. The Littlewood-Richardson rule determines the dimension of the space of triple tensor product invariants cλµν = dim(Vλ ⊗ Vµ ⊗ Vν )SLm(C). In [BZ92], Berenstein and Zelevinsky showed that cλµν equals the number of integer points in the intersection of a polyhedral cone with an affine subspace. These integer points are called BZ triangles. The second author and Zhou extended this definition to BZ triangles on trivalent graphs [MZ12]. We will recall the definition of BZ triangles and projections to corresponding highest weights as in [BZ92], and then define BZ triangles on graphs similarly to [MZ12]. Let Tm be the set of integer points in the triangle with vertices (2m − 3, 0, 0), (0, 2m − 3, 0) and (0, 0, 2m − 3). Let Hm be the subset of Tm with all coordinates odd and Gm be the subset of Tm with exactly one coordinate odd. Geometrically, the elements of Gm correspond to the vertices of a graph consisting of hexagons and triangles, and the elements of Hm correspond to the centers of the hexagons, see Figure 4. CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 17 (0,3,0) (0,5,0) (1,4,0) (0,4,1) (2,3,0) (0,3,2) (1,2,0) (0,2,1) (3,2,0) (0,2,3) (2,2,1) (1,2,2) (0,1,0) (2,1,0) (0,1,2) (4,1,0) (2,1,2) (0,1,4) (1,0,0) (0,0,1) (3,0,0) (2,0,1) (1,0,2) (0,0,3) (5,0,0) (4,0,1) (3,0,2) (2,0,3) (1,0,4) (0,0,5) Figure 4. Graphs consisting of hexagons and triangles for m = 2, 3, 4 Let Lm be the subspace of RGm such that x(ξ1) + x(ξ2) = x(ξ′ 2) for all 2] of all hexagons. Define a polyhedral cone 1) + x(ξ′ opposite pairs of edges [ξ1, ξ2] and [ξ′ Km = Lm ∩ RGm 1, ξ′ ≥0 and an affine semigroup BZ(SLm(C)) = Km ∩ ZGm. Definition 6.1. Elements of the affine semigroup BZ(SLm(C)) are called BZ tri- angles. Example 6.2. Minimal generating sets for BZ(SL2(C)) and BZ(SL3(C)) are listed in Figures 5 and 6. 0 1 0 1 0 0 0 0 1 Figure 5. Minimal generating set for BZ(SL2(C)) Honeycombs provide a graphical interpretation of BZ triangles. We use the version of honeycombs defined by the second author and Zhou [MZ12, Section 2]. To construct the honeycomb associated with a BZ triangle, we replace each label x(ξ) at a vertex ξ of a hexagon by an edge weighted by x(ξ) from the vertex of the hexagon to the center of the hexagon. A label x(ξ) at the vertex ξ of the underlying triangle is replaced by a vertex weighted by x(ξ). An example of a BZ triangle and the corresponding honeycomb is depicted in Figure 7. We identify the lattice of SLm(C)-weights with the standard lattice Zm−1 us- ing as the standard basis the fundamental weights ω1 = (1, 0, 0, . . . , 0), ω2 = (1, 1, 0, . . . , 0), . . . , ωm−1 = (1, 1, 1, . . . , 1). Define a linear projection pr:Lm → R3(m−1) to the vector space of triples of SLm(C)-weights by (30) pr(x) = (λ1, . . . , λm−1; µ1, . . . , µm−1; ν1, . . . , νm−1), 18 KAIE KUBJAS, CHRISTOPHER MANON 0 0 0 1 1 0 1 0 0 0 0 1 0 0 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0 1 1 0 Figure 6. Minimal generating set for BZ(SL3(C)) 1 1 1 1 Figure 7. A BZ triangle and the corresponding honeycomb where λi = x(2(m − i) − 1, 2(i − 1), 0) + x(2(m − i) − 2, 2i − 1, 0), µi = x(0, 2(m − i) − 1, 2(i − 1)) + x(0, 2(m − i) − 2, 2i − 1), νi = x(2(i − 1), 0, 2(m − i) − 1) + x(2i − 1, 0, 2(m − i) − 2) for i ∈ 1, . . . , m − 1. The coordinates of pr(x) are pairwise sums of neighboring la- bels on the boundary of the underlying triangle starting from the lower left corner and going clockwise around the triangle. In particular, the coordinates λi corre- spond to the sums of labels on the northwest edge, µi on the northeast edge, and νi on the south edge of the underlying triangle. Example 6.3. The projections of minimal generators of BZ(SL2(C)) as in Figure 5 are (31) (1; 0; 1), (1; 1; 0), (0; 1; 1). CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 19 The projections of minimal generators of BZ(SL3(C)) as in Figure 6 are (1, 0; 1, 0; 1, 0), (0, 1; 0, 1; 0, 1), (1, 0; 0, 0; 0, 1), (0, 1; 1, 0; 0, 0), (0, 0; 0, 1; 1, 0), (1, 0; 0, 1; 0, 0), (0, 0; 1, 0; 0, 1), (0, 1; 0, 0; 1, 0). Theorem 6.4 ([BZ92], Theorem 1). Let λ = P λiωi, µ = P µiωi and ν = P νiωi be the three highest SLm(C)-weights. Then the triple multiplicity cλµν equals BZ(SLm(C)) ∩ pr−1(λ, µ, ν). Define a linear projection pre : Lm → Rm−1 to the vector space of SLm(C)- weights for each edge e of the underlying triangle by (32) pre(x) =   (λ1, . . . , λm−1) (µ1, . . . , µm−1) (ν1, . . . , νm−1) if e is the northwest edge of the triangle, if e is the northeast edge of the triangle, if e is the south edge of the triangle. The projection pre maps a BZ triangle to the corresponding λ, µ or ν. Let pr∗ e : Lm → Rm−1 be the projection that maps a BZ triangle to the dual λ∗, µ∗ or ν∗ of the corresponding weight vector λ, µ or ν, i.e. pr∗ e is obtained from pre by reversing the order of coordinates. Figure 8. The trivalent 4-leaf tree and the triangle complex dual to it Given any trivalent graph Γ consider the triangle complex C dual to Γ, as il- lustrated in Figure 8. Let v be an inner vertex of Γ and ∆v the triangle dual to v. Assign a copy of SLm(C) BZ triangles to v, denoted by BZ(SLm(C))(v), such that its underlying triangle is ∆v. For every edge e adjacent to v define pre : BZ(SLm(C))(v) → Zm−1 to be equal to pre∗ : BZ(SLm(C))(v) → Zm where e∗ ∈ C is the edge dual to e. All projections are taken in the clockwise direction of the corresponding triangle. Definition 6.5. Let Γ be a trivalent graph. The affine semigroup of SLm(C) BZ triangles on Γ is BZΓ(SLm(C)) = Y v∈V (Γ) BZ(SLm(C))(v) ∩ \ e=(v1,v2) ker(pre(x(v1)) − pr∗ e(x(v2))). If Γ is the tripod, then BZΓ(SLm(C)) = BZ(SLm(C)). Example 6.6. Let T be the trivalent 4-leaf tree and C the triangle complex dual to T as in Figure 8. A SL3(C) BZ triangle on T is shown in Figure 9. Similarly to the tripod case, one can construct a honeycomb to any BZ triangle on a trivalent graph, see [MZ12, Figure 4] for an example of this construction. 20 KAIE KUBJAS, CHRISTOPHER MANON 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 1 0 Figure 9. BZ triangle on the trivalent 4-leaf tree 6.2. Connection Between Group-Based Models and BZ Triangles. Lemma 6.7. Denote the edges of the tripod by e1, e2, e3. The polytope P0,3(Z/mZ) has m2 vertices. They are • the zero vertex, • xe1 • xe2 • xe1 • xe1 [i] = xe2 [i] = xe3 [i] = xe3 [i] = xe2 m and all other coordinates zero. [m−i] = 1 for i ∈ {1, . . . , m − 1} and all other coordinates zero, [m−i] = 1 for i ∈ {1, . . . , m − 1} and all other coordinates zero, [m−i] = 1 for i ∈ {1, . . . , m − 1} and all other coordinates zero, [j] = xe3 [k] = 1 for i, j, k ∈ {1, . . . , m − 1} with i + j + k = 0 modulo Proof. We can freely choose a label [i] on the edge e1 and a label [j] on the edge e2. The label [k] on the edge e3 is determined by the first two. The additive group Z/mZ has m elements, hence there are exactly m2 possibilities to label the edges of the tripod. All such possibilities are listed above. (cid:3) Example 6.8. The vertices of P0,3(Z/2Z) and P0,3(Z/3Z) are depicted in Fig- ures 10 and 11. Without the zero vertex, they are in one-to-one correspondence with the minimal generators of the semigroup of SL2(C) and SL3(C) BZ triangles as in Example 6.3. We will prove in Theorem 6.9 that there is a similar connection between SLm(C) BZ triangles and the group-based model with the underlying group Z/mZ. 0 0 1 0 1 1 0 1 0 1 0 1 Figure 10. Labellings of the tripod with elements of Z/2Z In the rest of this section we consider the semigroup associated with a group- based model without the grading induced by the last coordinate of RΓ(Z/mZ). We denote this semigroup by Rpr Γ (Z/mZ). Theorem 6.9. For m ∈ N≥2 Rpr (33) 0,3(Z/mZ) ⊂ pr(BZ(SLm(C))). For m ∈ {2, 3} the inclusion is equality. By Proposition 3.1, the set pr(BZ(SLm(C))) and the projection of S0,3(SLm(C)) that forgets the level are equal. Thus, the inclusion in Theorem 6.9 follows from CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 21 0 1 1 0 0 2 0 2 0 1 2 0 1 1 1 1 0 2 2 2 0 2 0 2 2 1 1 Figure 11. Labellings of the tripod with elements of Z/3Z Theorem 5.1. Here we will give a combinatorial proof with the aim to provide a graphical tool for studying group-based models associated with groups Z/mZ. Proof. We will prove that every vertex of P0,3(Z/mZ) ⊂ R3(m−1) corresponds to an element of pr(BZ(SLm(C))) ⊂ R3(m−1) (the unit vectors corresponding to the edges e1, e2, e3 are replaced by the fundamental weights λ, µ, ν, respectively). This implies Rpr 0,3(Z/mZ) is generated by the vertices of P0,3(Z/mZ). 0,3(Z/mZ) ⊆ pr(BZ(SLm(C))), since Rpr • The zero vertex of P0,3(Z/mZ) equals the projection of the zero BZ triangle. • The vertex of P0,3(Z/mZ) with xe2 [m−i] = 1 for fixed i ∈ {1, . . . , m−1} and all other coordinates 0 is the projection of a BZ triangle with 1's at the vertices that lie on a line segment parallel to the northwest edge, and 0's on all the other vertices. Figure 12 depicts honeycombs corresponding to all such BZ triangles for SL4(C). [i] = xe3 Figure 12. BZ triangles with 1's at the vertices that lie on a line segment parallel to the northwest edge [i] = xe3 [i] = xe2 • The vertex of P0,3(Z/mZ) with xe1 [m−i] = 1 and all other coordinates 0, and the vertex of P0,3(Z/mZ) with xe1 [m−i] = 1 and all other coordinates 0 are projections of BZ triangles with 1's at the vertices that lie on line segments parallel to the northeast and south edge, respectively. [k] = 1 for fixed i, j, k ∈ {1, . . . , m − 1} with i + j + k = m and all other coordinates 0 equals the projection of a BZ triangle with 1's at the vertices that lie on the three line segments that start at the middle point of a hexagon and go west, northeast • The vertex of P0,3(Z/mZ) with xe1 [i] = xe2 [j] = xe3 22 KAIE KUBJAS, CHRISTOPHER MANON and southeast. Figure 13 depicts honeycombs corresponding to all such BZ triangles for SL4(C). Figure 13. BZ triangles with 1's at the vertices that lie on the three line segments that start at the middle point of a hexagon and go west, northeast and southeast [i] = xe2 [j] = xe3 • The vertex of P0,3(Z/mZ) with xe1 [k] = 1 for fixed i, j, k ∈ {1, . . . , m − 1} with i + j + k = 2m and all other coordinates 0 equals the projection of a BZ triangle with 1's at the vertices that lie on the three line segments that start at the middle point of a hexagon and go east, southwest and northwest. Figure 14 depicts honeycombs corresponding to all such BZ triangles for SL4(C). Figure 14. BZ triangles with 1's on the vertices that lie on three line segments that start at the middle point of a hexagon and go east, southwest and northwest Example 6.8 proves the equality for m ∈ {2, 3}. The semigroups BZ(SL2(C)) and Rpr 0,3(Z/2Z) are isomorphic. The semigroups BZ(SL3(C)) and Rpr 0,3(Z/3Z) are not isomorphic (but we have pr(BZ(SL3(C))) = Rpr 0,3(Z/3Z)). The toric ideal corresponding to the semigroup BZ(SL3(C)) is gen- erated by one cubic polynomial and the toric ideal corresponding to Rpr 0,3(Z/3Z) is generated by two cubic polynomials. For m > 3, the inclusion in Theorem 6.9 is strict. The SLm(C) BZ triangle given by (cid:3) x(2m − 3, 0, 2) = x(2m − 2, 1, 0) = x(2m − 4, 2, 1) = x(2m − 6, 4, 1) = . . . = x(4, 2m − 6, 1) = x(2, 2m − 4, 1) = x(1, 2m − 2, 0) = x(0, 2m − 3, 2) = 1 and all other coordinates 0, see Figure 15 for the SL4(C) case, has projection (34) which does not belong to Rpr 0,3(Z/mZ). (1, 0, . . . , 0, 1; 0, 1, 0, . . . 0; 0, . . . 0, 1, 0), Let Γ be a trivalent graph. Define prΓ as the direct product of functions pr with one copy for each v ∈ V (Γ). CONFORMAL BLOCKS, BZ TRIANGLES AND GROUP-BASED MODELS 23 Figure 15. A SL4(C) BZ triangle whose projection does not be- long to Rpr 0,3(Z/4Z) Corollary 6.10. Let Γ be a trivalent graph. For m ∈ N≥2 (35) Rpr Γ (Z/mZ) ⊂ prΓ(BZΓ(SLm(C))). Furthermore, for m ∈ {2, 3} the equality holds. Proof. By definition, we have prΓ(BZΓ(SLmC)) = Y v∈V (Γ) pr(BZ(SLm(C))(v))∩ \ e=(v1 ,v2) ker(pre(x(v1))−pr∗ e(x(v2))). By Theorem 6.9 and Lemma 4.5, this set contains RΓ(Z/mZ). (cid:3) The inclusion in Corollary 6.10 is strict for m > 3, since the inclusion in Theo- rem 6.9 is strict for m > 3. We now let BZgr(SL3(C)) ⊂ BZ(SL3(C)) × Z≥0 be the sub-semigroup generated by the eight BZ triangles in Figure 6 and the zero BZ triangle, lifted to height 1. This semigroup is studied in [KMSW93, Man12], in the context of conformal field theory. We can now state the proof of Theorem 1.2. Proof of Theorem 1.2. By Theorem 6.9, the generators of BZgr(SL3(C)) are in one to one correspondence with the degree one elements of R0,3(Z/3Z) = R0,3(Z/3Z). This defines a linear map φ : BZgr(SL3(C)) → R0,3(Z/3Z) given by passing to the boundaries of these elements. The result for general Γ then follows by modification of the argument used in Corollary 6.10. (cid:3) The assignment of height 1 to each of the generators of BZ(SL3(C)) is a result of the conformal field theory discussed in Section 3. This is not carried out for general m, because a generating set for each BZ(SLm(C)), and a way to assign heights to elements of this semigroup in a way consistent with the conformal field theory is still unknown. This is related to the problem of constructing a toric degeneration of V0,3(SLm(C)). References [BBKM13] W. Buczy´nska, J. Buczy´nski, K. Kubjas, and M. Micha lek. Degrees of generators of [Bel03] [BL94] [Buc12] phylogenetic semigroups on graphs. Cent. Eur. J. Math., 11(9):1577–1592, 2013. P. Belkale. The quantum Horn conjecture. arXiv:math/0303013v2, 2003. A. Beauville and Y. Laszlo. Conformal blocks and generalized theta functions. Com- mun. Math. Phys., 164(2):385–419, 1994. W. Buczy´nska. Phylogenetic toric varieties on graphs. J. Algebraic Comb., 35(3):421– 460, 2012. [BW07] W. Buczy´nska and J. Wi´sniewski. On the geometry of binary symmetric models of phylogenetic trees. J. Eur. Math. Soc., 9(3):609–635, 2007. 24 [BZ92] [DM12] KAIE KUBJAS, CHRISTOPHER MANON A. Berenstein and A. Zelevinsky. Triple multiplicities for sl(r + 1) and the spectrum of the exterior algebra of the adjoint representation. J. Algebraic Comb., 1(1):7–22, 1992. M. Donten-Bury and M. Micha lek. Phylogenetic invariants for group-based models. J. Algebraic Stat., 3(1):44–63, 2012. [ES93] [ERSS05] N. Eriksson, K. Ranestad, B. Sturmfels, and S. Sullivant. Phylogenetic algebraic ge- ometry. In Projective Varieties with Unexpected Properties, pages 237–255. Walter de Gruyter GmbH & Co. KG, Berlin, 2005. S. N. Evans and T. P. Speed. Invariants of some probability models used in phyloge- netic inference. Ann. Stat., 21(1):355–377, 1993. G. Faltings. A proof of the Verlinde formula. J. Algebraic Geom., 3(2):347–374, 1994. R. Fulton and J. Harris. Representation Theory: A First Course, volume 129 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1991. O. Gleizer and A. Postnikov. Littlewood-Richardson coefficients via Yang-Baxter equation. Int. Math. Res. Notices, 2000(14):741–774, 2000. [Fal94] [FH91] [GP00] [KMSW93] A. N. Kirillov, P. Mathieu, D. S´en´echal, and M. A. Walton. Can fusion coefficients be [KNR94] [KT99] calculated from the depth rule? Nucl. Phys. B, 391(3):651–674, 1993. S. Kumar, M.S. Narasimhan, and A. Ramanathan. Infinite Grassmannians and moduli spaces of G-bundles. Math. Ann., 300(1):41–75, 1994. A. Knutson and T. Tao. The honeycomb model of GLn(C) tensor products I: Proof of the saturation conjecture. J. Amer. Math. Soc., 12(4):1055–1090, 1999. [Kub12] [Pau96] [SS05] [SSE93] invariants. tensor product [Man12] [MZ12] [Man09] [Man10] [KTW04] A. Knutson, T. Tao, and C. Woodward. The honeycomb model of GLn(C) tensor products II: Puzzles determine facets of the Littlewood-Richardson cone. J. Amer. Math. Soc., 17(1):19–48, 2004. K. Kubjas. Hilbert polynomial of the Kimura 3-parameter model. J. Algebraic Stat., 3(1):64–69, 2012. C. Manon. The algebra of conformal blocks. arXiv:0910.0577v6, 2009. C. Manon. Presentations of semigroup algebras of weighted trees. J. Algebraic Comb., 31(4):467–489, 2010. C. Manon. The algebra of SL3(C) conformal blocks. arXiv:1206.2535v1, 2012. C. Manon and Z. Zhou. Semigroups of SL3(C) arXiv:1206.2529v1, 2012. C. Pauly. Espaces de modules de fibr´es paraboliques et blocs conformes. Duke Math. J., 84(1):217–235, 1996. B. Sturmfels and S. Sullivant. Toric ideals of phylogenetic invariants. J. Comput. Biol., 12(2):204–228, 2005. L. A. Sz´ekely, M. A. Steel, and P. L. Erdos. Fourier calculus on evolutionary trees. Adv. Appl. Math., 14(2):200–216, 1993. S. Sullivant. Toric fiber products. J. Algebra, 316(2):560–577, 2007. B. Sturmfels and Z. Xu. Sagbi bases of Cox-Nagata rings. J. Eur. Math. Soc., 12(2):429–459, 2010. A. Tsuchiya, K. Ueno, and Y. Yamada. Conformal field theory on universal family of stable curves with gauge symmetries. In Integrable Systems in Quantum Field Theory and Statistical Mechanics, volume 19 of Advanced Studies in Pure Mathematics, pages 459–566. Academic Press Inc., Boston, 1989. K. Ueno. Conformal Field Theory With Gauge Symmetry, volume 24 of Fields In- stitute Monographs. Fields Institute for Research in Mathematical Sciences, Toronto, 2008. D. P. Zhelobenko. Compact Lie Groups and Their Representations, volume 40 of Translations of Mathematical Monographs. American Mathematical Society, Provi- dence, 1973. [Sul07] [SX10] [TUY89] [Uen08] [Zhe73]
1206.6939	2	1206	2016-07-14T07:43:05	A hyperplane section theorem for Galois points and its application	[ "math.AG" ]	A point $P$ in projective space is said to be Galois with respect to a hypersurface if the function field extension induced by the projection from $P$ is Galois. We present a hyperplane section theorem for Galois points. Precisely, if $P$ is a Galois point for a hypersurface, then $P$ is Galois for a general hyperplane section passing through $P$. As an application, we determine hypersurfaces of dimension $n$ with $n$-dimensional sets of Galois points.	math.AG	math	A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION SATORU FUKASAWA Abstract. A point P in projective space is said to be Galois with respect to a hypersurface if the function field extension induced by the projection from P is Galois. We present a hyperplane section theorem for Galois points. Precisely, if P is a Galois point for a hypesurface, then P is Galois for a general hyperplane section passing through P . As an application, we determine hypersurfaces of dimension n with n-dimensional sets of Galois points. 1. Introduction Let the base field K be an algebraically closed field of characteristic p ≥ 0 and let X ⊂ Pn+1 be an irreducible and reduced hypersurface of dimension n and of degree d. H. Yoshihara introduced the notion of the Galois point (see [1, 7, 12, 13, 14]). If the function field extension K(X)/K(Pn) induced by the projection πP : X 99K Pn from a point P ∈ Pn+1 is Galois, then the point P is said to be Galois. Galois point theory has given a new viewpoint of classification of algebraic vari- eties, by the distribution of Galois points (see [1, 2, 3, 4, 7, 10, 12, 13, 14]). If the dimension of the singular locus SX is at most n − 2, then Fukasawa and Takahashi [4] presented upper bounds for the number of Galois points, as a generalization of a result of Yoshihara for smooth hypersurfaces [14]. To do this, they showed a "hyperplane section theorem" (see [4, Theorem 1.3]). In this article, we prove this theorem for arbitrary Galois points (which may be singular) with respect to arbi- trary hypersurfaces. The intersection of (almost) all tangent spaces of X is denoted by TX . 2000 Mathematics Subject Classification. 14J70, 12F10. Key words and phrases. Galois point, hypersurface. The author was partially supported by JSPS KAKENHI Grant Numbers 22740001 and 16K05088. 1 A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 2 Theorem 1. Let X ⊂ Pn+1 be a hypersurface of dimension n ≥ 2 and degree d ≥ 3 in characteristic p ≥ 0, and let P be a Galois point for X with multiplicity m, where 0 ≤ m ≤ d − 2. Then: (i) A general hyperplane H passing through P satisfies the following condition: (⋆) the hyperplane section XH := X ∩ H is an irreducible hypersurface in H ∼= Pn of degree d such that SXH = SX ∩ H, P 6∈ TXH , and the multiplicity of XH at P is equal to m. (ii) Let H be a hyperplane passing through P and satisfying the condition (⋆). Then, the point P is Galois for XH. (iii) In this case, the Galois groups are isomorphic: GP (X) ∼= GP (XH). As an application, we generalize results of Fukasawa and Hasegawa [2, 3] for plane curves with infinitely many Galois points. Let ∆(X) (resp. ∆′(X)) be the set of all Galois points contained in X \ SX (resp. Pn+1 \ X). Theorem 2. Let X ⊂ Pn+1 be a hypersurface of dimension n ≥ 1 and degree d ≥ 4 in characteristic p ≥ 0. Then, the following conditions are equivalent. (i) There exists a non-empty Zariski open subset U of X such that U ⊂ ∆(X). (ii) p > 0, d = pe for some e > 0, and X is projectively equivalent to the hypersurface defined by X pe−1 0 X1 − X pe 2 = 0. In this case, ∆(X) = X \ {X0 = X2 = 0}, and the induced Galois group GP is a cyclic group of order pe − 1 for any point P ∈ ∆(X). Theorem 3. Let X ⊂ Pn+1 be a hypersurface of dimension n ≥ 1 and degree d ≥ 3 in characteristic p ≥ 0. Then, the following conditions are equivalent. (i) There exist an irreducible Zariski closed subset Y ⊂ Pn+1 of dimension n and a non-empty open subset UY of Y such that UY ⊂ ∆′(X). (ii) p > 0, d = pe for some e > 0, and X is projectively equivalent to an irreducible hypersurface whose equation is of the form e n+1 X X j=0 i=1 αijX pe−pj 0 X pj i = 0, where αij ∈ K. In this case, ∆′(X) is a Zariski open set of a hyperplane (see Proposition 1), and the induced Galois group GP is isomorphic to (Z/pZ)⊕e for any point P ∈ ∆′(X). A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 3 2. Preliminaries and Lemmas Let X ⊂ Pn+1 be an irreducible hypersurface, let SX be the singular locus of X, and let Xsm = X \SX be the smooth locus. The projective tangent space at a smooth point P ∈ X \ SX is denoted by TP X ⊂ Pn+1. Let Pn+1 be the dual projective space which parameterizes hyperplanes of Pn+1. The Gauss map γ of X is a rational map from X to Pn+1 which sends a smooth point P to the tangent space TP X. If F is the defining polynomial of X, then γ is given by (∂F/∂X0 : · · · : ∂F/∂Xn+1). Let TP X. If TX 6= ∅, then X is said to be strange and TX is called a TX := TP ∈Xsm strange center. A strange center is a linear space. It is well known that (the function field extension induced by) the projection from TX is separable if and only if X is a cone with center TX. Therefore, any strange variety is a cone with center TX if p = 0. For distinct points P, Q ∈ Pn+1, the line passing through P and Q is denoted by P Q. For a point P ∈ Pn+1, the projective space Pn parameterizes lines passing through P . Then, we can identify πP (Q) = P Q for any point Q ∈ X \ {P }. If P = (0 : · · · : 0 : 1), then πP (X0 : · · · : Xn+1) = (X0 : · · · : Xn), up to the projective equivalence of Pn. We note the following Bertini theorem (see [5, Theorem 1.1], or [8, II. 6.1, Theorem 1] and [15, Lemma 5]). Lemma 1. Let P 6∈ TX (i.e., the projection πP : X 99K Pn is generically finite and separable). Then, for a general hyperplane H ⊂ Pn+1 with H ∋ P , the hyperplane section XH := X ∩ H is an integral scheme. Let P be a Galois point for X. The Galois group induced by the Galois extension is denoted by GP (X), or simply GP . Then, any element of GP (X) corresponds to a birational map from X to itself. Let Bir(X) be the group consisting of all birational map from X to itself. We can consider GP (X) as a subgroup of Bir(X). For σ ∈ GP (X), the maximal open subset of X on which σ is defined is denoted by Uσ. We set U[P ] = Tσ∈GP Uσ. Lemma 2. Let P ∈ Pn+1 be a Galois point and let σ ∈ GP be an induced birational map from X to itself. Suppose that H is a hyperplane such that P ∈ H and XH is an integral scheme. Then, XH ∩ Uσ 6= ∅. Furthermore, if the multiplicity of XH at P is less than d, then the restriction map σXH is a birational map from XH to itself. A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 4 Proof. For the first assertion, see the proof of [6, V. Lemma 5.1]. Now, XH cor- responds to a regular point of codimension one in the scheme X. We can prove the first assertion by using a valuative criterion of properness for X. We prove the second assertion. Let τ be the inverse of σ. Then, σ is an isomorphism from U[P ] ∩ σ−1U[P ] to U[P ] ∩ τ −1U[P ]. By definitions of πP and σ, σ(XH) ⊂ XH. If U[P ]∩σ−1U[P ]∩H 6= ∅, we have the conclusion. Assume that U[P ]∩σ−1U[P ]∩H = ∅. Then, σ(U[P ] ∩ H) ⊂ (X ∩ H) \ (U[P ] ∩ H). Since U[P ] ∩ H is of dimension n − 1 and (X ∩ H) \ (U[P ] ∩ H) is of dimension ≤ n − 2, σ−1(σ(Q)) is one-dimensional for a general point Q of X ∩ H. By definitions of πP and σ, P Q ⊂ X ∩ H. Then, XH is a cone with center P . Therefore, the multiplicity of XH at P is equal to d. This is a contradiction. (cid:3) Lemma 3. Let P be a Galois point for X with multiplicity m and let Q, R ∈ X be points such that πP (Q) = πP (R) and the intersection multiplicity of X and QR at P is equal to m. Assume that Q ∈ U[P ]. Then, there exists σ ∈ GP such that σ(Q) = R. Proof. Let (X0 : · · · : Xn+1) be a system of homogeneous coordinates, let xi = Xi/X0 for i = 1, . . . , n + 1, and let F be the defining homogeneous polynomial of X. We can assume that P = (0 : · · · : 0 : 1) and R = (1 : 0 : · · · : 0) for a suitable system of coordinates. Then, the line passing through P, Q, R is given by X1 = · · · = Xn = 0. Then, we can take πP (1 : x1 : · · · : xn+1) = (1 : x1 : · · · : xn) and πP (Q) = πP (R) = (1 : 0 : · · · : 0). We have F (X0, . . . , Xn+1) = Ad−mX d−m n+1 + · · · + A1Xn+1 + A0, where Ai ∈ K[X0, . . . , Xn]. We define f (x1, . . . , xn+1) := F (1, x1, . . . , xn+1) and ai(x1, . . . , xn) := Ai(1, x1, . . . , xn). Since f (R) = 0, a0(0, . . . , 0) = 0. Since the intersection multiplicity of X and QR at P is m, ad−m(0, . . . , 0) = Ad−m(1, . . . , 0) 6= 0. We consider a function xn+1. Let g = Qσ σ∗xn+1. Then, we have g(Q) = ((−1)d−ma0/ad−m)(Q) = 0. Assume that σ(Q) 6= R for any σ ∈ GP . Then, xn+1(σ(Q)) 6= 0 for any σ ∈ GP . Therefore, g(Q) 6= 0. This is a contradiction. (cid:3) 3. Proof of Theorem 1 Proof of Theorem 1. Suppose that P is Galois for X. Then, P 6∈ TX. By Lemma 1, for a general hyperplane H ∋ P , XH is an integral scheme. Let WP ⊂ Pn+1 be the A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 5 set of such hyperplanes. Since P 6∈ TX, WP \ γ(Xsm) 6= ∅. We have SXH = SX ∩ H for any H ∈ WP \ γ(Xsm), because, for a point Q ∈ Xsm ∩ H, TQX = H if and only if X ∩ H is singular at Q. Let UP ⊂ Xsm be the set of all smooth points Q such that the differential map of the projection πP at Q is surjective, and let ΣP ⊂ Pn+1 be the set of all hyperplanes H such that H ⊂ (X \ UP ). Since πP is separable, WP \ ΣP 6= ∅. We have P 6∈ TXH for any H ∈ WP \ (γ(Xsm) ∪ ΣP ). Let ΓP be the (finite) set of all hyperplanes H such that the multiplicity of X ∩ H at P is less than m. We have WP \ (γ(Xsm) ∪ ΣP ∪ ΓP ) 6= ∅, and any hyperplane H in this set satisfies (⋆). We have assertion (i). Let H be a hyperplane passing through P and satisfying (⋆). We consider a homomorphism of groups φ : GP → G; σ 7→ σXH , where G = {σ ∈ Bir(XH)σ(XH ∩l\{P }) ⊂ XH ∩l for a general line l such that P ∈ l ⊂ H}. Since the multiplicity of XH at P is m < d, it follows from Lemma 2 and the definition of a Galois point that φ is well-defined. In addition, by the condition P 6∈ TXH , XH ∩ l \ {P } consists of d − m points for a general line l ⊂ H containing P . It follows from Lemma 3 that φ is injective. Since the order of G is at most d − m, φ is an isomorphism. Then, P is Galois for XH . (cid:3) Remark 1. A general hyperplane section for a Galois point which is singular was studied by T. Takahashi in his Ph.D. thesis [11]. He proved that a Galois point P ∈ SX is also Galois for a general hyperplane section XH ∋ P , under the assumption that p = 0 and X ⊂ P3 is a normal surface. 4. Case of inner Galois points As an application of Theorem 1, we have the following. Corollary 1. Assume that the condition (i) in Theorem 2 holds. Then, for a general hyperplane H, there exists a non-empty Zariski open set UXH ⊂ XH such that UXH ⊂ ∆(XH). Proof. Let U be an open set as in Theorem 2(i) and let H be a general hyperplane. Since H is general, we can assume that U ∩ H 6= ∅. By Lemma 1, XH is an integral scheme and H satisfies the condition (⋆) for a general point P ∈ XH. By Theorem 1(ii), we can take UXH = (U ∩ XH) \ (SXH ∪ TXH ). (cid:3) A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 6 Proof of Theorem 2. We prove the implication (i) ⇒ (ii). We use induction on dimension n. If n = 1, then the assertion is nothing but a result of Fukasawa and Hasegawa [3]. We consider the case where n ≥ 2. Let H ⊂ Pn+1 be a general hyperplane. It follows from Corollary 1 that there exists a Zariski open set UXH ⊂ XH such that UXH ⊂ ∆(XH). By the assumption of induction, p > 0, d is a power of p, and XH ⊂ H ∼= Pn is projectively equivalent to X d−1 2 = 0. The Gauss map γXH for XH is given by (−X d−2 : 0 : · · · : 0), if XH ⊂ Pn is given by X d−1 2 = 0. Therefore, we find that TXH is a linear space of dimension n − 2 with TXH 6⊂ XH , and a general fiber of γXH is a linear space of dimension n − 2. Since H is general and TQXH = TQX ∩ H for any smooth point Q ∈ XH , TX is a linear space of dimension n − 1 with TX 6⊂ X, γXH coincides with the restriction map γXH , and a general fiber of γ is a linear space of dimension n − 1. 0 X1 : X d−1 0 X1 − X d 0 X1 − X d 0 Let P ∈ X be a general point. Since the linear spaces γ −1(γ(P )) of dimension n − 1 and TX of dimension n − 1 are contained in the projective space TP X of dimension n, these intersect along a linear space LP of dimension n − 2. If LP ′ 6= LP for a general point P ′ ∈ X, then TX is contained in X, since LP ′ ⊂ TX. This is a contradiction. Therefore, LP ′ = LP . Then, X is a cone with a (n − 2)-dimensional center L. For a suitable system of coordinates, we can assume that L is defined by X0 = X1 = X2 = 0 and X is defined by F (X0, X1, X2) = 0. We can assume that H is defined by Xn+1 − (a0X0 + · · · + anXn) = 0. Then, XH is given by the same equation F = 0 and there exists a linear transformation φ : H → H such that φ(XH) is defined by F1 := X d−1 2 = 0. Then, φ(L∩H) = L∩H. Therefore, φ gives an automorphism of the sublinear system hX0, X1, X2i of H 0(Pn, O(1)) which is given by L ∩ H. This implies that X is projectively equivalent to the hypersurface defined by X d−1 0 X1 −X d 0 X1 − X d 2 = 0. 0 X1 − X pe We consider the implication (ii) ⇒ (i). Let F = X pe−1 2 be the defining polynomial. It is not difficult to check that the singular locus SX of X is given by X0 = X2 = 0. Let P ∈ X \SX. Then, P = (1 : b1 : · · · : bn+1) for some b1, . . . , bn+1 ∈ K with b1 = bpe 2 . The projection πP is given by (X1−b1X0 : · · · : Xn+1−bn+1X0). Let Xi = Xi − biX0. Then, F (X0, X1 + b1X0, . . . , Xn+1 + bn+1X0) = ( X1 + b1X0)X pe−1 0 − ( X2 + b2X0)pe = X1X pe−1 2 = F (X0, X1, . . . , Xn). Then, πP = (X0 : X1 : · · · : Xn+1). Therefore, we have a field extension K(x0, x2, . . . , xn+1)/K(x2, . . . , xn+1) with a relation F (x0, 1, x2, . . . , xn+1) = xpe−1 2 = 0. It is not difficult to check 0 − xpe − X pe 0 A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 7 that this is a Galois extension, which is cyclic of degree pe − 1. Therefore, we have ∆(X) = X \ {X0 = X2 = 0}. (cid:3) 5. Case of outer Galois points As an application of Theorem 1, we have the following. Corollary 2. Assume the condition (i) in Theorem 3 holds. Then, for a general hyperplane H, there exists a non-empty Zariski open set UYH ⊂ YH such that UYH ⊂ ∆′(XH). Proof. Let Y, UY be as in Theorem 3(i) and let H be a general hyperplane. Since H is general, we can assume that U ∩ H 6= ∅. By Lemma 1, XH and YH are integral and H satisfies the condition (⋆) for a general point P ∈ YH. By Theorem 1(ii), we can take UYH = (UY ∩ H) \ TXH . (cid:3) Lemma 4. Let n = 1. Assume that there exists an irreducible plane curve Y ⊂ P2 and a non-empty open set UY ⊂ Y such that UY ⊂ ∆′(X). Then, (1) Y is a line, and (2) if we take a linear transformation φ such that φ(Y ) is defined by X0 = 0, i + c = 0, then the defining polynomial φ(X) is of the form Pe where αij, c ∈ K. i=1 αijxpj j=0 P2 Proof. Let (X0 : X1 : X2) be a system of homogeneous coordinates and let xi = Xi/X0 for i = 1, 2. By a result of Fukasawa [2], Y is a line and there exists a linear transformation ψ such that ψ(X) is defined by f := Pe i = 0. Then ψ(Y ) is defined by X0 = 0. Let φ be a linear transformation as in assumption (2). Then, φ(X) is given by (ψ ◦ φ−1)∗f = 0. Let xi := (ψ ◦ φ−1)∗xi = βi0 + βi1x1 + βi2x2, where βij ∈ K, for i = 1, 2. Then, φ(X) is given by f (x1, x2) = 0, where f (x1, x2) = Pe Proof of Theorem 3. We consider the following condition (Pn): If X ⊂ Pn+1 is an irreducible hypersurface, and there exists an irreducible hypersurface Y ⊂ Pn+1 and a non-empty open set UY of Y such that UY ⊂ ∆′(X), then i + c for some αij, c ∈ K. i=1 αijxpj j=0 P2 j=0 P2 i=1 γijxpj (cid:3) (1) Y is a hyperplane, and (2) if we take a linear transformation φ such that φ(Y ) is defined by X0 = 0, then i + c = 0, the defining polynomial of φ(X) is of the form Pe where αij, c ∈ K. j=0 Pn+1 i=1 αijxpj A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 8 If n = 1, then the assertion We prove (Pn). We use induction on dimension n. is nothing but Lemma 4. We consider the case where n ≥ 2. Let H ⊂ Pn+1 be a general hyperplane. It follows from Corollary 2 that there exists a non-empty Zariski open set UYH ⊂ ∆′(XH). By the assumption of induction, p > 0, d is a power of p, and XH ⊂ H ∼= Pn−1 is projectively equivalent to Pe j=0 Pn i + c = 0. Since H is general, Y is a hyperplane. We have result (1) of (Pn). i=1 αijxpj We take a linear transformation φ such that φ(Y ) is defined by X0 = 0. Let P = (1 : a1 : . . . : an+1) ∈ X be a general point and let H be a general hyperplane passing through P . Then, there exists an open set UYH ⊂ YH such that UYH ⊂ ∆′(XH). If we take a linear transformation ψ = (X0 : X1 − a1X0 : · · · : Xn+1 − an+1X0), then ψ(P ) = (1 : 0 : · · · : 0), ψ(φ(Y )) is defined by X0 = 0 and ψ(H) is defined by Xn+1 − b1X1 + · · · + bnXn = 0 for some bi ∈ K. Let F (X0, . . . , Xn+1) be the defining polynomial of ψ(X), let f = F (1, x1, . . . , xn+1) and let x = b1x1 +· · ·+bnxn. Then, f (0, . . . , 0) = 0 since (1 : 0 : · · · : 0) ∈ ψ(X). Since ψ(X) ∩ ψ(H) satisfies the condition (Pn−1) by induction, g(x1, . . . , xn) := f (1, x1, . . . , xn, x) = Pi,j βijxpj i for some βij ∈ K, similar to the proof of Lemma 4. If f has a term of degree not equal to some power of p, g has such a term for a general hyperplane H with P ∈ H. Therefore, f has only terms of degree equal to some power of p. Let fpi be the component of f of degree pi for i = 0, . . . , e. Then, fpi(x1, . . . , xn, x) be the component of g of degree pi. By condition (2) of (Pn−1), fpi(x1, . . . , xn, x) must be of the form hpi for some linear polynomial h(x1, . . . , xn). Since H is general, fpi(x1, . . . , xn+1) must be of the form hpi for some h(x1, . . . , xn+1). Therefore, ψ(X) is given by the polynomial as in condition (2) of (Pn) and hence, φ(X) also. The implication (ii) ⇒ (i) is derived from Proposition 1 below. (cid:3) Proposition 1 (cf. [2], Propositions 2 and 3). Let X be an irreducible hypersurface defined by the equation in Theorem 3(ii) and let H0 be the hyperplane defined by X0 = 0. Then, we have the following. (i) SX and TX are linear spaces of dimension n − 1 which are contained in H0. (ii) ∆′(X) = H0 \ (SX ∪ TX), and all points in SX \ TX are Galois. (Here, we consider a point P with π∗ P (Pn) = K(X) as a Galois point.) (iii) For any Galois point P ∈ H0 \ TX, any birational map induced by GP is a restriction of a linear transformation of P2. (iv) For any Galois point P ∈ H0 \ TX, the Galois group GP is isomorphic to (Z/pZ)⊕m for some m ≤ e. A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 9 (v) ∆(X) = Xsm if X is projectively equivalent to the hypersurface defined by X pe−1 0 X1 − X pe 2 = 0, and ∆(X) = ∅ otherwise. j=0 Pn+1 Proof. Let F = Pe difficult to check that SX is given by X0 = Pn+1 X0 = Pn+1 i=1 αi0Xi = 0. Therefore, we have (i). X pj i=1 αijX pe−pj 0 i be the defining polynomial. It is not i = 0 and TX is given by i=1 αieX pe We prove that all points in H0 \ TX are Galois. There exists i such that αi0 6= 0. We can assume that i = 1. Let X1 = Pn+1 i=1 αi0Xi. Then, φ(X0, . . . , Xn+1) = (X0, X1, X2 . . . , Xn+1) is a linear transformation and φ(TX ) is given by X0 = X1 = 0. By considering φ(X) as X, we can assume that TX is given by X0 = X1 = 0. Let P ∈ H0 \ TX. Then, P = (0 : 1 : b2 : · · · : bn+1) for some b2, . . . , bn+1. The projection πP is given by (1 : x2 − b2x1 : · · · : xn+1 − bn+1x1). Let xi = xi − bix1. Then, we have a field extension K(x1, x2, . . . , xn+1)/K(x2, . . . , xn+1) with a relation g(x1, x2, . . . , xn+1) = F (1, x1, x2 + b2x1, . . . , xn+1 + bn+1x1) = 0. By the form of F , g is of the form Pi,j βijxpj for some βij ∈ K. Since P 6∈ TX , this extension is Galois of degree pm for some m ≤ e and the Galois group is isomorphic to (Z/pZ)⊕m (see [9, pp. 117 -- 118]). Therefore, we have H0 \ (SX ∪ TX ) ⊂ ∆′(X), and any point in SX \ TX is Galois. By considering the form g, we find that assertions (iii) and (iv) hold for all Galois points in H0 \ TX. i We prove that ∆′(X) ⊂ H0 \ (SX ∪ TX ). If this is proved, then we have (ii), (iii) and (iv). If n = 1, then this is a result of Fukasawa (see [2, Proposition 2]). Assume that n ≥ 2 and there exists an outer Galois point P 6∈ H0 for X. By Theorem 1, there exists a hyperplane H ∋ P such that {P } ∪ (∆′(X) ∩ H0 ∩ H) \ TXH ⊂ ∆′(XH). When n = 2, XH is a curve and this is a contradiction. By using induction, we have a contradiction for any n ≥ 2. We prove (v). Assume that P ∈ ∆(X). Since X ∩ H0 = SX, P ∈ X \ H0. Let P ′ ∈ X \ H0 such that the line P P ′ intersects the set ∆′(X). Let R ∈ P P ′ ∩ ∆′(X). Since any element of GR is a linear transformation, it follows from Lemma 3 that there exists σ ∈ GR such that σ(P ) = P ′. Since P is Galois, P ′ is also Galois. Therefore, if one inner Galois point exists, then almost all points of X are inner Galois points. It follows from Theorem 2 that X is projectively equivalent to the hypersurface defined by X pe−1 2 = 0 and ∆(X) = X \ {X0 = X1 = 0} = Xsm. (cid:3) 0 X1 − X pe A HYPERPLANE SECTION THEOREM FOR GALOIS POINTS AND ITS APPLICATION 10 References [1] S. Fukasawa, Galois points for a plane curve in arbitrary characteristic, Proceedings of the IV Iberoamerican conference on complex geometry, Geom. Dedicata 139 (2009), 211 -- 218. [2] S. Fukasawa, Classification of plane curves with infinitely many Galois points, J. Math. Soc. Japan 63 (2011), 195 -- 209. [3] S. Fukasawa and T. Hasegawa, Singular plane curves with infinitely many Galois points, J. Algebra 323 (2010), 10 -- 13. [4] S. Fukasawa and T. Takahashi, Galois points for a normal hypersurface, Trans. Amer. Math. Soc. 366 (2014), 1639 -- 1658. [5] W. Fulton and R. Lazarsfeld, Connectivity and its applications in algebraic geometry, in: "Algebraic Geometry", Lecture Notes in Math. 862 (1981), pp. 26 -- 92. [6] R. Hartshorne, Algebraic geometry, GTM 52, Springer, 1977. [7] K. Miura and H. Yoshihara, Field theory for function fields of plane quartic curves, J. Algebra 226 (2000), 283 -- 294. [8] I. R. Shafarevich, Basic algebraic geometry I, Springer-Verlag, 1994. [9] H. Stichtenoth, Algebraic function fields and codes, Universitext, Springer-Verlag, Berlin, 1993. [10] T. Takahashi, Galois points on normal quartic surfaces, Osaka J. Math. 39 (2002), 647 -- 663. [11] T. Takahashi, Galois points on curves and surfaces, Ph.D. thesis, Niigata University, 2003. [12] H. Yoshihara, Function field theory of plane curves by dual curves, J. Algebra 239 (2001), 340 -- 355. [13] H. Yoshihara, Galois points on quartic surfaces, J. Math. Soc. Japan 53 (2001), 731 -- 743. [14] H. Yoshihara, Galois points for smooth hypersurfaces, J. Algebra 264 (2003), 520 -- 534. [15] O. Zariski, Pencils on an algebraic variety and a new proof of a theorem of Bertini, Trans. Amer. Math. Soc. 50 (1941), 48 -- 70. Department of Mathematical Sciences, Faculty of Science, Yamagata University, Kojirakawa-machi 1-4-12, Yamagata 990-8560, Japan E-mail address: [email protected]
1108.1727	1	1108	2011-08-08T14:55:48	Hilbert-Mumford criterion for nodal curves	[ "math.AG", "math.DG" ]	We prove by Hilbert-Mumford criterion that a slope stable polarized weighted pointed nodal curve is Chow asymptotic stable. This generalizes the result of Caporaso on stability of polarized nodal curves, and of Hasset on weighted pointed stable curves polarized by the weighted dualizing sheaves. It also solved a question raised by Mumford and Gieseker to prove the Chow asymptotic stability of stable nodal curves by Hilbert-Mumford criterion.	math.AG	math	HILBERT-MUMFORD CRITERION FOR NODAL CURVES JUN LI AND XIAOWEI WANG Abstract. We prove by Hilbert-Mumford criterion that a slope stable polarized weighted pointed nodal curve is Chow asymptotic stable. This generalizes the re- sult of Caporaso on stability of polarized nodal curves, and of Hasset on weighted pointed stable curves polarized by the weighted dualizing sheaves. It also solved a question raised by Mumford and Gieseker to prove the Chow asymptotic stability of stable nodal curves by Hilbert-Mumford criterion. 1. Introduction and summary of main result In late seventy, Mumford [16] and Gieseker [7] constructed the coarse moduli space Mg of stable curves using Mumford's Geometric Invariant Theory (GIT). They proved the sta- bility of smooth curves by verifying Hilbert-Mumford stability criterion; for nodal curves, they proved the stability indirectly by using semi-stable replacement and the numerical criterion to rule out curves with worse than nodal singularities. This construction has been very successful and is widely adopted subsequently for studying related to stability of curves, for instance, Caporaso's proof of asymptotic stablility of nodal curves [2]. In this paper, we will prove the Chow asymptotic stability of weighted pointed nodal curves by verifying Hilbert-Mumford criterion directly. As an application, we provide a GIT construction of the moduli of weighted pointed stable curves. An interesting con- sequence of this construction is that the GIT closure of the moduli of weighted pointed smooth curves, using Chow asymptotic stability, is identical to Hassett's coarse mod- uli of weighted pointed stable curves; nevertheless, its universal family includes strictly semistable weighted pointed nodal curves. Another application of our stability study is showing that a polarized nodal curve is K-stable (c.f. Section 7) if and only if the polarization is numerically equivalent to a multiple of its dualizing sheaf. This generalizes a theorem of Odaka that a stable nodal curve polarized with dualizing sheaf is K-stable. The primary goal of this work is to understand the GIT compactification of moduli of canonically polarized varieties. The recent work on the relation between various notions of K-stabilities and the existence of constant scalar metrics suggests that some deep and interesting geometry are yet to be uncovered in this area. This work is a first step toward this direction. We hope this study will help us understand the stability of high dimensional singular varieties. We now outline the results proved in this paper. Definition 1.1 (Hassett [9]). A weighted pointed nodal curve (X, x, a) consists of a re- duced, connected curve X, n ordered (not necessarily distinct) smooth points x = (x1, · · · , xn) of X, and weights a = (a1, · · · , an), ai ∈ ≥0, of x, such that the total weight at any point Date: May 31, 2018. 1 2 JUN LI AND XIAOWEI WANG is no more than one (i.e. curve is a weighted pointed curve together with a polarization OX(1). for any p ∈ X, Pxi=p ai ≤ 1). A polarized weighted pointed In this paper, we will use (X, OX (1), x, a) to denote such a polarized weighted pointed curve. In case OX (1) is very ample, we let (1.1) ı : X ⊂−→ W, W = H 0(OX (1))∨, be the tautological embedding; let Chow (X) ∈ Divd,d[(W ∨)2] := H 0(cid:0)O(W ∨)2 (d, d)(cid:1) be the Chow point of X, which is the bi-degree (d, d) hypersurface in (W ∨)2 consisting of points (V1, V2) ∈ (W ∨)2 such that Vi ⊂ W are hyperplanes satisfying V1 ∩V2 ∩ı(X) 6= ∅. We abbreviate (1.2) Ξ := Divd,d[(W ∨)2] × (W )n, and let the Chow point of (X, OX (1), x) be Chow (X, x) = (Chow (X), x) ∈ Ξ. The stability of this Chow point is tested by the positivity of the a-weight of any one × → SL(W ). (A one parameter subgroup, abbreviated to 1-PS, parameter subgroup λ : C is always non-trivial.) Given a 1-PS λ, its action on W induces an action on Ξ. Since Divd,d[(W ∨)2] is a projective space, it has a canonical polarization O(1). We let OΞ(1, a) be the -ample line bundle on Ξ that has degree 1 on Divd,d[(W ∨)2] and has degree ai on the i-th copy of the W in (W )n. Integral multiple of this line bundle is canonically linearized by SL(W ). Definition 1.2. With (X, OX (1), x, a) understood, we define the a-λ-weight of Chow (X, x) ∈ Ξ be the weight of the λ-action on the fiber OΞ(1, a)ζ , where ζ = limt→0 λ(t)·Chow (X, x) ∈ Ξ; we denote this weight to be ωa(λ). We define ω(λ) be the λ-weight of Chow (X) ∈ Divd,d[(W ∨)2] defined with Chow (X, x) (resp. OΞ(1, a)) replaced by Chow (X) (resp. O(1)). Definition 1.3. Given (X, OX (1), x, a), we say that it is stable (resp. semistable) if for any 1-PS λ of SL(W ), the a-λ-weight of Chow (X, x) is positive (resp. non-negative). To make an analogy with the slope stability of vector bundle, we introduce the no- tion of slope stable by testing on proper closed subcurves Y ⊂ X. First, with OX (1) understood, we denote deg X = deg OX (1), and for any subcurve Y ⊂ X, we denote deg Y = deg OX (1)Y . For any proper subcurve Y ⊂ X, we define the number of linking nodes of Y to be Definition 1.4. Given (X, OX (1), x, a), We say that it is slope (semi-)stable if X is nodal and if for any proper subcurve Y ( X we have (1.3) (1.4) ℓY =(cid:12)(cid:12)Y ∩ Y ∁(cid:12)(cid:12), Y ∁ = X \ Y . 2 +Pxj∈Y deg X +Pn j=1 h0(OX (1)) aj 2 < h0(OX (1)Y ) deg Y + ℓY aj 2 , (resp. ≤) . In this paper, we will prove by verifying the Hilbert-Mumford criterion the following theorem. For the weight a and g(X) = g, we denote (1.5) χa(X) := g − 1 + (a1 + · · · + an). HILBERT-MUMFORD CRITERION FOR NODAL CURVES 3 Theorem 1.5. Given g and a such that χa(X) > 0, there is an N so that a genus g polarized weighed pointed curve (X, OX (1), x, a) such that deg X ≥ N is (semi-)stable if and only if it is slope (semi-)stable. For (X, x), we abbreviate the -line bundle ωX (P aixi) to ωX (a · x). For integer k so X (P kaixi) is a line bundle. In Section 5, that k · ai ∈ Z for all i, then ωX(a · x)⊗k = ω⊗k we will show that in case deg X is sufficiently large, the slope stability is equivalent to the criterion: Proposition 1.6. Given g and a such that χa(X) > 0, there is an N so that a genus g polarized weighed pointed curve (X, OX (1), x, a) such that deg X ≥ N is slope (semi- )stable if and only if for any proper subcurve Y ( X satisfying h0(OX (1)Y ) < h0(OX (1)), we have aj 2(cid:17) − degY ωX (a · x) deg ωX (a · x) (cid:16)deg X + nXj=1 aj 2(cid:17)(cid:12)(cid:12)(cid:12) < ℓY 2 , (resp. ≤) . (1.6) (cid:12)(cid:12)(cid:12)(cid:16)deg Y + Xxj∈Y The case x = ∅ is a theorem of Caporaso [2] on the stability of polarized nodal curves. The case of the asymptotic Hilbert stability of smooth1 weighted pointed curves is a theorem of David Swinarski [22] (see also [15]). We now sketch the main ingredients of our proof. Our starting point is a theorem of Mumford that expresses the a-λ-weight of Chow (X, x) in terms of the leading coefficient of the Hilbert-Samuel polynomial of an ideal I ⊂ OX×A 1(1) (cf. Prop. 2.1). Our first observation is that this leading coefficient can be evaluated by the leading coefficient of the Hilber-Samuel polynomial of the pull back I of I to the normalization X of X. This transforms the evaluation of the a-λ-weight to the calculation of the areas of a class of Newton polygons associated to the pull back sheaf I. By dividing the Newton polygons into two kinds and studying them seperately, we obtain an effective bound of the areas, thus a bound of the a-λ-weight of Chow (X, x). This bound is linear in the weights of λ. We then apply linear programing to complete our proof of Theorem 1.5. Our GIT construction of the moduli of weighted pointed stable curves goes as follows. m of fixed degree. We form the Hilbert scheme H of pointed 1-dimensional subscheme of Let ψ : H → C be the Hilbert-Chow morphism (map) to the Chow variety of pointed 1- m of the same degree, equivariant under SL(m + 1). Applying our dimensional cycles in main theorem, we conclude that in case the degree is sufficiently large, the preimage under ψ of the set Css ⊂ C of GIT-semistable points is the set of semistable polarized weighted pointed nodal curves. Let K ⊂ H be the subset of canoncially polarized weighted pointed smooth curves. We prove that the GIT-quotient of the closure K is isomorphic to the Hassett's moduli of weighted pointed stable curves. An interesting observation is that the complement K − K contains polarized semistable but not canonically polarized weighted pointed curves. Thus though GIT gives the same compactification as that of Hassett of the moduli of canonically polarized weighted pointed smooth curves, the geometric objects added to obtain the compactification in the mentioned two constructions are markedly different. It is worth pursuing to see how this extends in the high dimensional case. In the end, using that the Donaldson-Futaki invariants can be expressed as the limit of certain Chow weights under a 1-PS, we apply our main theorem to prove that a polarized nodal curve (X, OX (1)) is K-stable if and only if OX (1) is numerically equivalent to a multiple of ωX . This implies that GIT compactification is same as the compactification of 1Notice if X is smooth then (1.6) becomes vacuous. 4 JUN LI AND XIAOWEI WANG smooth curve using K-stability. This is analogous to that the Uhlenbeck compactification coincides with the GIT compactification of the moduli of vector bundles over curves. The paper is organized as follows. In Section two, we show that the weights can be evaluated via the leading coefficients of the Hilbert-Samuel polynomial of a sheaf on the normalization X. In Section three, we reduce our study to a particular class of 1-PS: the staircase 1-PS. We will derive a sharp bound for each irreducible component in Section four. We complete the proof of our main theorems in Section five. The last two sections include the application of our stability study to constructing moduli of weighted pointed curves and to study the K-stability of polarized curves. Acknowledgement. The first author is partially supported by an NSF grant NSF0601002. The second author is partially supported by RGC grants CUHK 403106 and 403709 from the Hong Kong government. Part of the writing was completed when the second author was visiting the Mathematics Department of Stanford University in summer 2009 and the Mathematics Department of Harvard University in fall 2009. He thanks the faculty members there for the hospitality and the fine research environment they provided. List of notations 1(1); similarly defined on X (2.3) 1(k)/I(λ)k); similarly defined on X (2.4); after (2.6) = 0, for j ≥ i + 1}. (tρ0 s0, · · · ) ⊂ OX×A n.l.c. χ(OX×A contribution of e(I) at q ∈ X; along Xα λ-Chow weight the vanishing order of si at q max{i v(si, q) 6= ∞} mini{i sj Xα Newton polygon supported at q ∈ X (si, si+1, . . . , sm) ⊂ OX (1) {q ∈ Xα sα(q) = 0}; Λ(λ) = ∪r length(Ei/Ei+1)p or = 0 I(λ); I(λ) e(I(λ)); e(I) e(I)q; e(Iα) ω(λ) v(si, q) (q) α ∆q Ei = E(λ)i Λα(λ); Λ(λ) δ(si, p) incα(si) δα(si); δ(si) w(Ei, p); wα(Ei) α = α(λ) LY ; Lα; LY ; Lα Y ∩ Y ∁; LXα; π−1(LY ) ∩ Y ; LXα NY ; Nα; Nα ℓα; ℓα,β; ℓα,α pri α Eα(ρ) Wi = Wi(λ) ωa(λ) Φ : H → C K, ¯K ⊂ H ~δ(L) Pp∈ Xα δ(si, p)p and inc(si) =Pα incα(si); δ(si, p); δ(si) =Pα δα(si) Pp∈ Xα length(O X (1)/Ei)p; wα(Ei) =Pp∈ Xα {i ∈ inc(si) ∩ Xα 6= ∅ or i = α}; π−1(NY ) ∩ Y ; NXα ; NXα Lα; Xα ∩ Xβ; −Lα {i ∈ α wα(Ei+1) ≤ deg Xα − 2g(Xα) − ℓα − 1} Def. 3.12 upper bound of e(I)α {v ∈ W si(v) = · · · = sm(v) = 0} ⊂ W ω(λ) + µa(λ) . Hilbert-Chow map slice polarized by ωX /H(a · x) degree class for the line bundle L (4.3) (5.1) (5.5) before Lem. 6.2 before (6.4) after (6.8) (2.12) Prop. 2.1 (2.8) (2.9) (2.15) Def. 2.6 (3.1) Def. 3.1 Def. 3.2 Def. 3.1 Def. 3.1 Def. 3.2 (3.3) (3.9) and (3.11) (3.10) (3.11); (6.9) α=1Λα(λ) w(Ei, p) 2. Chow stability, Chow weight and Newton polygon In this section, we first recall some basic facts from [16] on stability of a polarized curve; we then localize the calculation of the weight of Chow (X) to a divisor on the HILBERT-MUMFORD CRITERION FOR NODAL CURVES 5 normalization of X, and interpret the contribution from each point of the divisor as the area of a generalized Newton polytope. Throughout the paper, we fix a polarized (connected) curve (X, OX (1)), its associated embedding ı : X → W (cf. (1.1)), and denote by Chow (X) the Chow point of ı once and for all. We also assume that X is nodal unless otherwise is mentioned. We will reserve the symbol λ for a 1-PS of SL(W ); for such λ, we diagonalize its action by choosing (2.1) s = {s0, · · · , sm} a basis of W ∨ so that under its dual bases the action λ is given by λ(t) := diag[tρ0 , · · · , tρm ] · t−ρave, ρ0 ≥ ρ1 ≥ · · · ≥ ρm = 0, (2.2) and ρave = 1 m+1P ρi. We will call s a diagonalizing basis of λ. In [16], Mumford introduced a subsheaf (2.3) I(λ) = (tρ0 s0, · · · , tρm sm) ⊂ OX×A 1(1) := p∗ X OX (1) 1 → X is the projection. Let generated by sections in the paranthesis, where pX : X × A e(I(λ)) be the normalized leading coefficient (abbreviate to n.l.c.) of the Hilbert-Samuel polynomial: (2.4) e(I(λ)) = n.l.c. χ(OX×A 1 (k)/I(λ)k) Proposition 2.1 (Mumford). The λ-weight Chow (X) is ω(λ) = 2 deg X m + 1 mXi=0 ρi − e(I(λ)). In the following, when the 1-PS λ and its diagonalizing basis s are understood, we will drop λ from I(λ) and abbreviate I(λ) to I. Our first step is to lift the calculation of e(I) (= e(I(λ))) to the normalization of X: π : X −→ X. We let (2.5) and let I be the pull-back of I: si = π∗si ∈ O X (1) := OX (1) ⊗OX O X , (2.6) I = (tρ0 s0, · · · , tρm sm) ⊂ O X×A 1(1) = O X (1) ⊗O X O X×A 1 . Like e(I), we define e(I) = n.l.c. χ(O X×A 1(k)/Ik). We have the following proposition whose proof will be given at the end of this section. Proposition 2.2. We have e(I) = e(I). This Proposition enables us to lift the evaluation of e(I) to X. Our next step is to localize the evaluation of e(I) to individual q ∈ X. In order to do that, let z be a uniformizing parameter of X at q; let t be the standard coordinates of A 1. We denote by O X,q the formal completion of the local ring O X,q at its maximal ideal. We fix an isomorphism of O X,q-modules (the first isomorphism below): (2.7) ϕq : O X (1) ⊗O X O X,q ∼= O X,q ∼= C[[z]], where the second isomorphism is induced by the choice of z. 6 JUN LI AND XIAOWEI WANG Definition 2.3. Let si ∈ H 0(O X (1)) be as in (2.5). We define v(si, q) = the vanishing order of si at q; (2.8) in case si ≡ 0 near q, we define v(si, q) = ∞. We set (2.9) (q) = max{i v(si, q) 6= ∞} and w(I, q) = v(s(q), q). The quantity w(I, q) is the width of the polygon ∆q associated to I (at q) to be defined later. We now look at the image of I under O X×A 1(1) → O X×A 1,(q,0). We let (2.10) Iq = (zv(sm,q), zv(sm−1,q)tρm−1 , · · · , zv(s0,q)tρ0 ) ⊂ R = C[[z, t]]. By construction, ϕq induces an isomorphism Notice that the right hand side is not a finite module when (q) < m. Since tρi ·ϕq(si) ∈ (2.11) tρ(q) R for all i, the map (cid:0)O X×A 1 O X×A 1 (k)/Ik(cid:1) ⊗O X×A tk·ρ(q) R/(cid:0)Iq ∩ tρ(q) R(cid:1)k 1,(q,0) ∼= R/I k q . −→ R/I k q induced by the inclusion tk·ρ(q) R ⊂ R is injective. This time the R-module on the left hand side is a finite module. We define (2.12) e(I)q = n.l.c. dim tk·ρ(q) R/(cid:0)Iq ∩ tρ(q) R(cid:1)k Lemma 2.4. We have the summation formula 2 e(I) =Pq∈ X e(I)q. + 2ρ(q) · w(I, q). We need some preparation to prove this Lemma. We begin with a geometric interpre- tation of the quantity e(I)q. Let I ⊂ C[z1, z2] be a monomial ideal and let Γ be the set of exponents of monomials in I; namely, I is the linear span of the monomials {xγ γ ∈ Γ}, 2 -- the xy-plane.) where Γ is a subset of ( ∪ {0})2 ⊂ R ≥0 +Γ. We let ¯Γ = Conv(R 2 ≥0 + 2; the integral closure ¯I of I is the ideal generated by {xγ γ ∈ ¯Γ} [6, Ex4.23]. We then form the closed convex hull Conv(R 2 ≥0 is the first quadrant of R 2 ≥0 +Γ) of R 2 ≥0. (R Γ) ∩ 2 We let ∆(I) be the Newton polygon of I: ∆(I) = R 2 ≥0 − Conv(R 2 ≥0 + Γ) ⊂ R 2 ≥0. Lemma 2.5. Let ∆(I) be the area of the ∆(I). Then dim C[z1, z2]/I k = ∆(I) · k2 + O(k). Proof. Since ¯I is the integral closure of I, by Briancon-Skoda theorem [13, Thm 9.6.26], I k ⊂ ¯I k ⊂ I k−1 for k sufficiently large. Since dim I k−1/I k is bounded from above by a linear function in k, dim C[z1, z2]/I k = dim C[z1, z2]/ ¯I k + O(k). Further, dim C[z1, z2]/ ¯I k is precisely the number of lattice points in k∆( ¯I) = k∆(I). From the work of Kantor and Khovanski [11, 5], the number of lattice points inside the polygon is given by ∆(I) · k2 + O(1). This proves the Lemma. (cid:3) We now come back to the 1-PS λ and its diagonalizing basis s = {si}. Definition 2.6. For any q ∈ X, we define Γq = {(v(si, q), ρi)}0≤i≤m ⊂ ( ∪ {0})2 and define the Newton polygon (of I = I(λ)) at q to be ∆q(λ) := (R 2 ≥0 − Conv(R 2 ≥0 + Γq)) ∩ ([0, w(I, q)] × R≥0). 2 We were informed that similar formula was obtained by Swinarski in 2008. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 7 We will abbreviate ∆q(λ) to ∆q when the choice of the basis s is understood. Let ∆q be the area of ∆q. Corollary 2.7. We have e(I)q = 2∆q; henceforth, e(I) = 2Pq∈ X ∆q. Proof. Since ∆q is the union of ∆q ∩ [0, w(I, q)] × [ρ(q), ∞) with [0, w(I, q)] × [0, ρ(q)], by (2.4), (2.12) and Lemma 2.5, e(I)q = 2 · ∆q ∩ [0, w(I, q)] × [ρ(q), ∞) + 2 · ρ(q) · w(I, q) = 2 ∆q. The second identity follows from Lemma 2.4. (cid:3) This formula will be used to estimate the quantity e(I) = e(I) in the next section. For now, we prove Proposition 2.2. Proof of Proposition 2.2. Let p1, · · · , pl be the nodes of X; let ξ = π×1A be the projection. Tensoring the exact sequence 1 : X×A 1 1 → X×A 0 −→ OX×A 1 −→ ξ∗O X×A 1 −→ ⊕l j=1 with OX×A 1(k)/Ik, we obtain an exact sequence OX×A 1 (k)/Ik fk−→ (OX×A 1(k)/Ik) ⊗O 1 ξ∗O X×A 1 −→ X×A are linear in k. This will prove the Proposition. We begin with the first claim. We let q be one of the node of X; let q+ and q− be the preimages π−1(q) ⊂ X, and let x and y be uniformizing parameters of X at q+ and OX,q ∼= OX,q near q and q−, respectively. Then after fixing an isomorphism OX (1) ⊗OX denoting R = k[[x, y]]/(xy), we have isomorphism (2.14) (OX×A 1 (k)/Ik) ⊗O 1 X×A Oq×A 1 ∼= (R[t]/I k) ⊗R[t] R[t]/(x, y), where I ⊂ R[t] is the ideal generated by tρi si, i = 0, · · · , m, and si are formal germs of si at q as elements in R. Since for some i the value si(q) 6= 0, iq = max{i si(q) 6= 0} is finite. Thus the right hand side of (2.14) is isomorphic to R[t]/(I k, x, y) = k[t]/(tk·iq ) whose dimension is linear in k. This proves the first claim. For the second claim, since the kernel of fk consists of torsion elements supported on 1. Hence to prove the claim, we only need to study the the union of p1 × A kernel of an analogue homomorphism 1, · · · , pl × A ¯fk : R[t]/I k −→ (R[t]/I k) ⊗R[t] (k[[x]][t] ⊕ k[[y]][t]), Opj ×A 1 −→ 0 (OX×A 1 (k)/Ik)pj ×A 1 −→ 0. rMα=1 1(k)/Ik) ⊗O 1. 1 ξ∗O X×A X×A 1 (k)/Ik) ⊗O 1 ξ∗O X×A X×A 1(cid:1), 1 (k)/Ik)pj ×A 1 )(cid:1). dim ker fk By projection formula, we have Thus 1 (k)/Ik(cid:1) = ξ∗(cid:0)ξ∗(OX×A ξ∗(cid:0)O X×A e(I) = n.l.c.χ(cid:0)ξ∗(cid:0)O X×A which equals We claim that both (2.13) n.l.c.(cid:0)χ(cid:0)OX×A χ(cid:0)(OX×A 1(k)/Ik)(cid:1) = (OX×A 1(k)/Ik(cid:1)(cid:1) = n.l.c.χ(cid:0)(OX×A lXi=1 χ(cid:0)(OX×A 1(cid:1) and 1 (k)/Ik(cid:1) − dim ker fk + 1(k)/Ik) ⊗O Opj ×A 1 X×A 8 JUN LI AND XIAOWEI WANG where I is as in the previous paragraph, and R[t] → k[[x]][t] ⊕ k[[y]][t] is the normalization homomorphism g(x, y, t) 7→ (g(x, 0, t), g(0, y, t)). Since the domain and the target of ¯fk are t-graded rings and ¯fk is a homomorphism of graded rings, as vector spaces ker ¯fk =Mj≥0 ker(cid:8)( ¯fk)j : tjR/(I k ∩ tjR) → (tj R/(I k ∩ tjR)) ⊗R (k[[x]] ⊕ k[[y]])(cid:9). Because R = k[[x, y]]/(xy), as R-modules, tjR/(I k ∩ tjR) is isomorphic to R/J for J one of the ideals in the list: R, (0), (xe), (ye), (xe, ye′ ), (xe + ye′ ), where e, e′ ∈ . One checks that for J of the first five kinds, ker( ¯fk)j = 0; for J of the last kind, ker( ¯fk)j ∼= k. Thus we always have dim ker( ¯fk)j ≤ 1. On the other hand, since siq (q) 6= 0, tρiq ∈ I and tkρiq ∈ I k. Thus ker( ¯fk)j = 0 for j ≥ kiq. This proves that dim ker fk is at most linear in k. This proves the Proposition. (cid:3) Because of this Proposition, we will work over the normalization X of X subsequently. To avoid possible confusion, we will reserve " " to denote the associated objects lifted to X. For instance, we will denote by X1, · · · , Xr the irreducible components of X, and denote by X1, · · · , Xr their respective normalizations. For the sections tρi si in I, tρi si are their lifts in I = I ⊗O 1 . For consistence, we reserve subindex i for the sections si, and reserve the greek α for the index of the irreducible components {Xα}1≤α≤r. Proof of Lemma 2.4. For each irreducible component Xα ⊂ X, we let Iα = I Xα×A O Xα×A 1(1). Then O X×A 1 ⊂ 1 X×A Thus to prove the Lemma, we only need to show that for each Xα, 1(k)/Ik e(Iα). α(cid:1) = rXα=1 e(Iα)q, e(I) = rXα=1 n.l.c. χ(cid:0)O Xα×A e(Iα) = Xq∈ Xα where e(Iα)q = e(I)q when q ∈ Xα. To proceed, we notice that (q)(cf.(2.9)) is a locally constant function on Xα; we let α be the values of (q) for q ∈ Xα. Then we have (2.15) = 0, for j ≥ i + 1}. α = min {i sj Xα i Thus tρα divides tρi si for all i > α. Since ρi ≥ ρi+1, the same division holds for all i. We let ¯ρi = ρi − ρα, and introduce ideal Rα = (t ¯ρ0 s0, t ¯ρ1 s1, · · · , t ¯ρα sα) ⊂ O Xα×A 1 (1). This way, Iα = tρα Rα ⊂ tρα O Xα×A 1 (1). We let (tkρα ) = tkρα O Xα×A 1 (k); it belongs to the exact sequence 0 −→ (tk·ρα )/Ik α −→ O Xα×A 1(k)/Ik α −→ O Xα×A 1 (k)/(tk·ρα ) −→ 0. Since (tk·ρα )/Ik have α = tk·ρα ·(cid:0)O Xα×A 1(k)/Ik χ(O X×A 1(k)/ Rk α(cid:1) and O Xα×A α) = χ(O Xα×A 1(k)/ Rk α) + χ(O Xα×A 1(k)/(tk·ρα )). 1 (k)/ Rk α is a finite module, we Taking the n.l.c. of individual term, and using χ(O Xα×A 1 (k)/(tk·ρα )) = k ρα · χ(O Xα (k)) = k2 ρα · deg Xα + O(k), HILBERT-MUMFORD CRITERION FOR NODAL CURVES 9 we obtain (2.16) e(Iα) = n.l.c. χ(O X×A 1(k)/Ik α) = n.l.c. χ(O Xα×A 1(k)/ Rk α) + 2ρα · deg Xα. Next let {q1, · · · , ql} be the support of (sα = 0) ∩ Xα. Following the convention in (2.11), we have an isomorphism O Xα×A 1 (k)/ Rk α ∼=−→ ⊕l a=1 tk·ρα R/(Iqa ∩ tρα R)k, induced by restricting to germs at qa after multiplying tk·ρα . Adding that deg Xα = dim O Xα (1)/(sα) = w(I, qa), lXa=1 (2.16) gives us e(Iα) = lXa=1(cid:16)n.l.c. h0(tk·ρα R/(Iqa ∩ tρα R)k) + 2ρα · w(I, qa)(cid:17) = Xq∈ Xα e(I)q. This proves the Lemma. (cid:3) Finally, we give one example that will be used later. Example 2.8. Let s = {si} be a basis of H 0(OX (1)); using weights ρ0 = 1 > ρ1 = · · · = ρm = 0 we form a 1-PS with diagonalizing basis s = {si}: λ = diag[t, 1, · · · , 1] · t− 1 m+1 . Suppose p = {s1 = · · · = sm = 0} ∈ X is a single point. Then e(I(λ)) = 1 (resp. = 2) when p is a smooth point (resp. nodal point) of X. Hence ω(λ) =( 2 deg X m+1 − 2, 2 deg X m+1 − 1, q is a nodal point; q is a regular point. 3. Staircase One-parameter subgroups We begin with some conventions attached to a fixed 1-PS λ and its diagonalizing basis {s0, · · · , sm}. For simplicity, we denote = {0, 1, . . . , m}. For each i ∈ , we introduce subsheaves (3.1) Ei = E(λ)i := (si, si+1, . . . , sm) ⊂ OX (1); they form a decreasing sequence of subsheaves. Similarly, we introduce O X -submodules Ei = E(λ)i := (si, si+1, . . . , sm) ⊂ O X (1). Definition 3.1. We call i ∈ a base index if i = α (cf. (2.15)) for some irreducible component Xα. For each Xα, we define Λα(λ) = {q ∈ Xα sα(q) = 0}; define Λ(λ) = α=1Λα(λ); define Λα(λ) = {p ∈ Xα sα(p) = 0}, and define Λ = Λ(λ) = ∪m ∪r In the following, for any sheaf of O X -modules F and p ∈ X, we denote Fp := F⊗O X O X,p, the localization of F at p. We remark that for any p ∈ Xα, (p) = α is the largest index i so that (Ei)p 6= 0. Λα(λ). α=1 10 JUN LI AND XIAOWEI WANG Definition 3.2. For closed point p ∈ Xα ⊂ X, we define δ(si, p) = length(Ei/Ei+1)p when i + 1 ≤ α(p); δ(si, p) = 0 otherwise. We define the increments of si, along Xα and X, be (cycles) incα(si) = Xp∈ Xα δ(si, p)p we define their degrees be δα(si) =Pp∈ Xα the width of Ei at p ∈ Xα and at Xα for i ≤ α be (3.2) w(Ei, p) := length(O X (1)/Ei)p incα(si); and inc(si) =Xα δ(si, p) and δ(si) =Pα δα(si). We also define and wα(Ei) := Xp∈ Xα w(Ei, p) . We remark that for p ∈ Xα, i + 1 ≤ (p) is equivalent to (Ei+1)p 6= 0. Definition 3.3. For any irreducible component Xα ⊂ X we introduce α = α(λ) = {i ∈ inc(si) ∩ Xα 6= ∅ or i = α}; (3.3) for mα + 1 = α, the order of α, we introduce a re-indexing map (3.4) Similarly, for p ∈ X, we introduce indα : α −→ [0, mα] ∩ Z, order preserving and bijective. p = {i ∈ p ∈ inc(si)}. For mp + 1 = p; we define similarly indp : p −→ [0, mp] ∩ Z, order preserving and bijective. To define the staircase 1-PS, we need the following (3.5) Y ∁(1), Y = Supp(Ei), Definition 3.4. For each Ei, we define its codegree codeg(Ei) = length(cid:0)OY (1)/EiY(cid:1) + deg O where Supp(Ei) is the smallest closed subscheme Y ⊂ X so that the tautological Ei → OY is injective, and Y ∁ = X \ Y . Since Ei is decreasing, codeg(Ei) is EiY := Ei ⊗OX increasing. Definition 3.5. We say a 1-PS λ is a semi-staircase after index i if for any i < j ≤ m, either codeg(Ej−1) < codeg(Ej), or j = α + 1 (cf. (2.15)) for some irreducible component Xα ⊂ X. We say λ is a semi-staircase when λ is a semi-staircase after index 1. Proposition 3.6. Given a 1-PS λ, there is a semi-staircase 1-PS λ′ so that ω(λ) ≥ ω(λ′). Proof. Suppose λ is a semi-staircase at index i but not at i − 1 then (3.6) codeg(Ei−1) = codeg(Ei) < codeg(Ei+1). We claim that Ei−1 = Ei. Since Supp(Ei) is always a subcurve of X and i 6= α + 1 for all α by the assumption, Y is also the support of Ei−1. Consequently, EiY ⊂ Ei−1Y and Ei−1Y /EiY is a finite module. Then codeg(Ei−1) = codeg(Ei) implies that length(Ei−1Y /EiY ) = 0. This proves that Ei−1 = Ei. As a consequence, we have Ei−1 = Ei ) Ei+1. Therefore, there is a point p ∈ X such that if we denote by sj ∈ OX,p(1) the formal germ of sj at p, then as OX,p-modules (3.7) OX,p(1) ⊃ (si−1, · · · , sm) = (si, · · · , sm) ) (si+1, · · · , sm). HILBERT-MUMFORD CRITERION FOR NODAL CURVES 11 By the middle equality, we can find cj ∈ OX,p such that si−1 =Pm We now construct a new basis s′. Let c = ci(p) ∈ k. We define j=i cj sj. (3.8) for j 6= i, i − 1; s′ j = sj i} is a basis of H 0(OX (1)). For j 6= i, because the linear span of j is the m}, we have Ej = E′ s′ i = si−1 − csi; j, where E′ s′ i−1 = si. j, · · · , s′ Clearly, s′ = {s′ {sj, · · · , sm} equals the linear span of {s′ Ei in (3.1) with si replaced by s′ i. For i, we claim that E′ i ( Ei. The inclusion E′ i ⊂ Ei follows from E′ i ⊂ Ei−1 = Ei. For the inequality, we claim that (si−1 − csi, si+1, · · · , sm) 6= (si, si+1, · · · , sm). Suppose instead the identity holds, then there are constants aj ∈ k such that aj sj =(cid:0)ai(si−1 − cisi) + mXj=i+1 aj sj(cid:1) + ai(ci − c)si. si = ai(si−1 − csi) + mXj=i+1 Combined with si−1 = Pm j=i cj sj, we conclude that si ∈ (si+1, · · · , sm) + sim, where m ⊂ OX,p is the maximal ideal. By Nakayama Lemma, si ∈ (si+1, · · · , sm), contradicting to (3.7). This proves the claim. Finally, we claim that if we define λ′ be the 1-PS with diagonalizing basis s′ and associated weights {ρ}i∈, then ω(λ′) ≤ ω(λ). By Mumford's formula (cf. Prop. 2.1), this is equivalent to e(I(λ′)) ≥ e(I(λ)). By our construction, E′ i ⊂ Ei for all i ∈ ; hence since 1(k)/I(λ)k. This proves ρi−1 ≥ ρi, I(λ′) ⊂ I(λ). Thus OX×A e(I(λ′)) ≥ e(I(λ)). 1(k)/I(λ′)k surjects onto OX×A So far, for any λ that is not a semi-staircase, we have constructed a new λ′ so that I(λ′) ⊂ I(λ). We now claim that by continuing this process, we eventually arrive at a semi-staircase λ′. Suppose not, then we can constructed an infinite sequence of 1-PS λ = λ0, λ1, · · · , λl, · · · so that E(λl+1)i ⊂ E(λl)i for all i ∈ , and for some i, E(λl+1)i 6= E(λl)i. (Here E(λl)i is the sheaf Ei in (3.1) with λ replaced by λl.) Because codeg(E(λl)i) ≤ deg OX(1), for each i, the sequence E(λ0)i ⊂ E(λ1)i ⊂ · · · ⊂ E(λl)i ⊂ E(λl+1)i ⊂ · · · stabilize at finite places. In particular, after finite place, we will have E(λl)i = E(λl+1)i for all i; or equivalently, I(λl) = I(λl+1), a contradiction. This proves that this process eventually provides us a semi-staircase λ′ such that ω(λ) ≥ ω(λ′). (cid:3) Remark 3.7. We remark that for a semi-staircase λ, the inclusions O X (1) = E0 ) E1 ) · · · ) Em 6= 0 are proper. Definition 3.8. We say a semi-staircase 1-PS λ is a staircase if for any p ∈ Λ, v(si, p) ≤ v(si+1, p) for all i (cf. Definition 2.3). Corollary 3.9. Proposition 3.6 holds with semi-staircase replaced by staircase. Proof. By Proposition 2.1, the λ-weight ω(λ) (of Chow (X)) depends only the sheaf I(λ) and the weights {ρi}. Thus, for any 1-PS λ′ with I(λ) = I(λ′) and having identical weights {ρ′ i} as that of λ, we have ω(λ) = ω(λ′). Given any 1-PS, we let λ be the corresponding semi-staircase constructed in Proposition 3.6. Let Λ and {si} be the associated objects of λ. Since Λ is a finite set, if we replace si 12 JUN LI AND XIAOWEI WANG by s′ but new basis {s′ i = si +Pj>i cij sj for a general choice of cij ∈ C, the new 1-PS with the same {ρi} i} will be a desired staircase 1-PS. (cid:3) Lemma 3.10. Suppose λ is a staircase 1-PS, then for p ∈ Xα and i < α, w(Ei, p) = v(si, p), and δ(si−1, p) = v(si, p) − v(si−1, p). Proof. The proof is a direct consequence of the definition of staircase 1-PS. (cid:3) As we will see, if λ is a staircase 1-PS then for most of i, δ(si) = 1. For those i with δ(si) > 1, we will give a detailed characterization (cf. Prop. 3.11). To this purpose, for any subcurve Y ⊂ X, we denote by NY to be the set of nodes of X in Y ; namely, NY = Xnode ∩ Y . We denote (cf. (1.3)) (3.9) LY := Y ∩ Y ∁ , and call it the linking nodes of Y . Moreover, let (3.10) NY := π−1(NY ) ∩ Y and LY := π−1(LY ) ∩ Y ⊂ NY . Since we reserve α for the index of the components Xα, we abbreviate (3.11) Nα := NXα, Nα := NXα, Lα := LXα, Lα := LXα, ℓα := Lα . We now state a characterization of those indices with δ(si) > 1. Proposition 3.11. Suppose λ is a staircase 1-PS. Let i ∈ α be a non-base index (cf. Definition 3.1) and let p ∈ inc(si) ∩ Xα. Suppose δ(si) ≥ 2, and suppose further that either deg Xα = 1 or (3.12) wα(Ei) + 1 ≤ deg Xα − 2g(Xα) − ℓα, then q = π(p) ∈ X is a node of X, indp(i) = 0 and δ(si, p) = 1. When this happens, let {p, p′} = π−1(q) and let Xβ be the component satisfying p′ ∈ inc(si) ∩ Xβ (possibly Xα = Xβ), and suppose further deg Xβ > 1 and (3.13) then inc(si) = p + p′. wβ (Ei) + 1 ≤ deg Xβ − 2g(Xβ) − ℓβ , Before its proof, we introduce a few notations. Since Xα is smooth, we can view a zero-subscheme of Xα as a divisor as well. This way, the union of two effective divisors is pp) =P max{np, n′ the union as zero subschemes, and the sum is as sum of divisors. For example, (P npp) ∪ (P n′ p}p and (P npp) + (P n′ Proof of Proposition 3.11. We will prove each part of the statement by repeatedly ap- plying the following strategy. Suppose i satisfies (3.12) and δ(si) ≥ 2, we will construct a section ζ ∈ H 0(OX (1)) so that the OX -modules Fj = (ζ, sj , · · · , sm) fits into a strict filtration pp) =P(np + n′ p)p. (3.14) Since Ej and Fj are generated by global sections of H 0(OX (1)), this implies h0(OX (1)) > m + 2, a contradiction. F0 ) · · · ) Fi ) Fi+1 ) Ei+1 ) · · · ) Em 6= 0. Let us assume deg Xα > 1 first, since for the case deg Xα = 1 the proof is rather easy. So wi(Ei) satisfies (3.12). We recall an easy consequence of a vanishing result. Let B ⊂ Xα be a closed zero-subscheme such that (3.15) deg B ≤ deg Xα − 2g(Xα) − ℓα + 1. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 13 Let Nα be as defined in (3.11). We claim that the γ in the exact sequence (1)(− Nα ∪ B)) −→ H 0(O Nα∪B(1)) −→ H 1(O Xα H 0(O Xα (3.16) (1)) γ is surjective. Indeed, using deg Nα = 2g(Xα) − 2g( Xα) + ℓα and (3.15), we obtain deg O Xα (1)(− Nα ∪ B) ≥ deg Xα − deg Nα − deg B ≥ 2g( Xα) − 1. Therefore, the last term in (3.16) vanishes, which shows that the γ in (3.16) is surjective. The section ζ mentioned before (3.14) will be chosen by picking an appropriate B and v ∈ H 0(O Nα∪B(1)) so that any element ζα ∈ γ−1(v) descends to a section in H 0(OXα (1)) and the descent glue with si+1X ∁ to form a desired section ζ. α We let (3.17) Zα,j := (sj = · · · = sm = 0) ∩ Xα ⊂ Xα. Since p ∈ inc(si) ∩ Xα, δα(si) ≥ 1. In case δα(si) = 1, we choose B = Zα,i + p, which is a subscheme of Zα,i+1. In case δα(si) ≥ 2 and δ(si, p) = 1, then there exists a p′ 6= p ∈ Xα such that p + p′ ≤ inc(si) ∩ Xα, (which is equivalent to Zα,i + p + p′ ⊂ Zα,i+1). In case δ(si, p) ≥ 2, we choose p′ = p. Combined, we let B = Zα,i + p + p′. We then let v1 = si+1 Nα ∈ H 0(O Nα (1)) and v2 6= 0 ∈ H 0(OB(1)) s.t. v2B−p = 0. We claim that when p 6∈ Nα, or indp(i) ≥ 1, or δ(si, p) ≥ 2, then both v1 Nα∩B and v2 Nα∩B are zero. Indeed, since Nα ∩ B ⊂ Zα,i+1 and si+1 Zα,i+1 = 0, we have v1 Nα∩B = si+1 Nα∩B = 0. For v2, we prove case by case. Suppose p 6∈ Nα, then Nα ∩ B = Nα ∩ (B − p); therefore since v2B−p = 0, v2 Nα∩B = 0. Now suppose p ∈ Nα. Since v2B−p = 0, v2(¯p) = 0 for all k=0 nkpk, pk distinct, as an effective divisor. Since p ∈ B, we can arrange p0 = p. In case indp(i) ≥ 1, we have n0 ≥ 2; in case δ(si, p) ≥ 2, since p′ = p we still have n0 ≥ 2. Thus p ∈ B − p and v2(p) = 0. This proves that v1 and v2 have identical images in H 0(O Nα∩B(1)). Consequently, (v1, v2) lifts to a section v ∈ H 0(O Nα∪B(1)) using the exact sequence ¯p ∈ ( Nα ∩ B) − {p}. We remain to show that v2(p) = 0. We write B =Pl H 0(O Nα∪B(1)) −→ H 0(O Nα (1)) ⊕ H 0(OB(1)) −→ H 0(O Nα∩B(1)) . α Since deg B ≤ wα(Ei) + 2 and i satisfies (3.12) (, because we assume deg Xα > 1), deg B satisfies the inequality (3.15). Therefore, the γ in (3.16) is surjective. We let ζα ∈ γ−1(v) ⊂ H 0(O Xα . This implies that ζα descends to a section ζα ∈ H 0(OXα (1)), and the descent ζα glues with si+1X ∁ (1)) be any lift. Because it is a lift of v1, ζα Nα to form a new section ζ ∈ H 0(OX (1)). = si+1 Nα We now prove the first part of the Proposition. We let Zα,j ⊂ Xα be the subscheme Zα,j = (sj = · · · = sm = 0) ∩ Xα. We decompose Zα,j into disjoint union Zα,j = Rj ∪ R′ j j is disjoint from q. We let ¯Zα = (ζ = si+1 = so that Rj is supported at q = π(p) and R′ · · · = sm = 0) ∩ Xα and decompose ¯Zα = ¯R ∪ ¯R′ accordingly. Suppose q is a smooth point of X. Then Rj and ¯R are divisors, and can be written as Rj = njq and ¯R = ¯nq. In case δα(si) = 1, the choice of B ensures that ni = ¯n = ni+1 − 1 and R′ i ⊂ ¯R′ ( R′ (si, · · · , sm) ⊗OX i+1. Thus OXα ⊆ (ζ, si+1, · · · , sm) ⊗OX OXα ( (si+1, · · · , sm) ⊗OX OXα. 14 JUN LI AND XIAOWEI WANG Further, since δ(si) ≥ 2 and ζX ∁ α = si+1X ∁ α , we have (si, · · · , sm) ⊗OX O X ∁ α ( (ζ, si+1, · · · , sm) ⊗OX O X ∁ α ⊆ (si+1, · · · , sm) ⊗OX O . X ∁ α Thus we have (3.18) In case δα(si) ≥ 2, the the choice of B ensures that Ri ( ¯R ( Ri+1. Thus Ei ( Fi+1 ( Ei+1. (si, · · · , sm) ⊗OX OXα ( (ζ, si+1, · · · , sm) ⊗OX OXα ( (si+1, · · · , sm) ⊗OX OXα. This implies (3.18) as well. In summary, by the argument at the beginning of the proof, (3.18) leads to a contradiction which proves that q must be a node of X. It remains to study the case where q is a node of X. A careful case by case study shows that when either indp(i) ≥ 1 or δ(si, p) ≥ 2, then Zα,i ( ¯Zα ( Zα,i+1. Thus (3.18) holds, which leads to a contradiction. This proves that q is a node, indp(i) = 0 and δ(si, p) = 1. We complete the proof of the first part by looking at the case deg Xα = 1. In this case indp(i) = 0 and δ(si, p) = 1, since otherwise deg Xα = 1 implies that i = α, contradicting to the assumption that i is not a base index. We next show that p ∈ Lα. But this is parallel to the proof of the case deg Xα > 1 by letting B = p because δα(si) = 1. This completes the proof of the first part. We now prove the further part. Let π−1(q) = {p, p′} with p′ ∈ inc(si) ∩ Xβ so that (3.13) holds. Then by the first part of the Proposition, we have indp(i) = indp′ (i) = 0; hence si(q) 6= 0. Thus for Zj = (sj = · · · = sm = 0) ⊂ X, we have p 6∈ Zi and Zi+1 = p ∪ S, where S is a zero-subscheme disjoint from p. Since Zi ( Zi+1 and p 6∈ Zi, we have Zi ⊂ S. In case Zi = S, then the further part of the Proposition holds. Suppose Zi ( S, then repeating the proof of the first part of the Proposition, we can find a section ζ ∈ H 0(OX (1)) so that p 6∈ (ζ = 0) and S ⊂ (ζ = 0). This way, we will have (3.18) again, which leads to a contradiction. This proves the further part of the Proposition. (cid:3) The Proposition above motivates the following Definition 3.12. For deg Xα > 1, we define the primary indices of Xα be pri α = {i ∈ α wα(Ei+1) ≤ deg Xα − 2g(Xα) − ℓα − 1}, for deg Xα = 1, we define inc(si)∩ Xα if i ∈ pri α ; otherwise we say it is secondary. We define ¯α := max{i i ∈ α (0) ⊂ α. We say i ∈ α is primary at p ∈ pri α } . α = ind−1 pri Note that in the proof above, the assumption δ(si) ≥ 2 is used only to show that (3.14) is strict. If i = α for some α, then length(Ei/Ei+1) = ∞. This time we choose ζ so that Ei/Fi+1 is finite. Since Ei/Ei+1 is infinite, (3.14) remains strict. Hence we have Proposition 3.13. Let i = α be a base index for some Xα, and let p ∈ inc(si) ∩ Xα. Suppose δ(si) ≥ 1 and deg Xα = 1, or wα(Ei) satisfies the inequality (3.12). Then indp(i) = 0, δ(si, p) = 1, and q = π(p) ∈ Xα is a linking node of Xα. Further, let {p, p′} = π−1(q), then i must be secondary at p′ (cf. Definition 3.12), and there is a component Xβ so that p′ ∈ Xβ and i = β. Proof. The proof is parallel to the proof of the previous Proposition. We will omit it here. (cid:3) Corollary 3.14. Denoting wpri α := wα(E¯α+1), suppose Xα ( X, then (3.19) 0 ≤ deg Xα − wpri α ≤ 2(g(Xα) + ℓα + 1). HILBERT-MUMFORD CRITERION FOR NODAL CURVES 15 Proof. The first inequality is trivial. We now prove the second one. If deg Xα = 1 we obtain deg Xα − wpri α = 0, from which the second inequality trivially follows. So from now on we assume deg Xα > 1. We let ¯i ∈ α be the index succeeding ¯α; namely, ¯i is the smallest index > ¯α so that δα(s¯i) ≥ 1. In particular, this implies that (3.20) Since ¯i 6∈ pri α , α = wα(E¯α+1) = wα(E¯i+1) − δα(s¯i) > deg Xα − 2g(Xα) − ℓα − 1 − δα(s¯i). (3.21) wpri δα(s¯α ) = · · · = δα(s¯i−1) = 0. Thus when δα(s¯i) ≤ 2, the second inequality follows from ℓα ≥ 1 (, since Xα ( X ). Suppose δα(s¯i) > 2. By our assumption ¯i is the index in α immediately succeeding ¯α, we have wα(E¯i) = wα(E¯α+1) because of (3.20). By Definition 3.12, wα(E¯i) satisfies (3.12). So we can apply Proposition 3.11 to the index ¯i to conclude that every p ∈ inc(s¯i) ∩ Xα lies in Nα and has δ(s¯i, p) = 1. We claim that inc(si) ∩ Xα ⊂ Lα. Indeed, let p ∈ inc(s¯i) ∩ ( Nα \ Lα), then the second part of Proposition 3.11 implies that inc(s¯i) = p + p′ and δ(s¯i) = 2, contradicting to the assumption δα(s¯i) > 2. This proves that inc(s¯i) ∩ Xα ⊂ Lα. Adding that δ(s¯i, p) = 1 for p ∈ inc(s¯i) ∩ Xα, we conclude that δα(s¯i) ≤ ℓα. These and (3.21) proves the second inequality in (3.19). (cid:3) 4. Main estimate for irreducible curves Throughout this section, we fix a staircase 1-PS λ, and an irreducible Xα. We will derive a sharp estimate of e(Iα(λ)) for the Xα ⊂ X. We let gα be the genus of Xα; we define the set of special points (4.1) Sα = (π−1(x) ∩ Xα)) ∪ Nα ⊂ Xα , where x = (x1, · · · , xn) ⊂ X is the set of weighted points. We continue to denote by ¯ρi = ρi − ρα. For each p ∈ Λα, we define the initial index (4.2) i0(p) := min{i i ∈ p}. For deg Xα > 1 and a fixed ǫ > 0, we define (4.3) Eα(ρ) :=(cid:16)2+ 2ǫ deg Xα(cid:17) Xi∈ pri α δα(si)¯ρi −(cid:16)1+ 2ǫ deg Xα(cid:17) Xq∈ Sα∩ Λα ¯ρi0(q) +2 deg Xα ·ρα; for deg Xα = 1 satisfying x ∩ Xα = ∅, we define (4.4) Eα(ρ) := δα(si0 )¯ρi0 + 2 · ρα; It is clear that in both cases Eα(ρ) are linear in ρ ∈ R is the following α (0). i0 = ind−1 m+1 + . Our main result of this section Theorem 4.1. For any 1 ≥ ǫ > 0 there is a constant M depending only on gα, ℓα and ǫ such that whenever deg Xα ≥ M , then e(Iα(λ)) ≤ Eα(ρ). In case deg Xα = 1 and x ∩ Xα = ∅, the same inequality holds for Eα(ρ) defined in (4.4). Note that the theorem implies that we can bounded e(I(λ))) in terms of the primary ρi's only. And for primary indices, we have a complete understanding of the multiplicity δα(si) due to the detailed study in the previous section. We begin with the following bound on the area of ∆p in terms of {ρi}. 16 JUN LI AND XIAOWEI WANG Lemma 4.2. Let λ be a staircase. Then for each p ∈ Λα and any 0 ≤ l ≤ k ≤ α, we have (4.5) ∆p ∩ ([w(El, p), w(Ek, p)] × R) − ρα· (w(Ek, p) − w(El, p)) ≤ ≤ Xi∈p∩[l,k−1] δ(si, p)¯ρi − (¯ρimin(p) + ¯ρimax(p)) 2 , where imin := min(p ∩ [l, k − 1]) and imax := max(p ∩ [l, k − 1]). Note that by letting l = 0 and k = α, we obtain (4.6) ∆p − ρα· w(Eα , p) ≤Xi∈p δ(si, p)¯ρi − ¯ρi0(p) 2 . Proof. First, we notice that the above inequality is invariant when varying ρα, thus to prove the Lemma we can and do assume from now on that ρα = 0; hence ¯ρi = ρi. Let Γp := {(w(Ei, p), ρi)}0≤i≤m; it follows from Definition 2.6 and 3.8 that (4.7) 2 + − Conv(R ∆p = (R 2 + + Γp)) ∩ ([0, w(I, p)] × R) . Fixing an indexing p = {i0(p), · · · , id(p)} ⊂ , (4.8) we let T be the continuous piecewise linear function on [0, w(I, p)] defined by linear inter- polating the points ij(p) increasing and d + 1 = p, {(0, ρi0), · · · , (w(Eik , p), ρik ), · · · , (w(Eid , p), ρα)} ⊂ R 2 , and let ∆T be the polygon bounded on two sides by x = 0 and x = w(Ek, p), from below by y = 0 and from above by the graph of y = T. By the convexity of ∆p, we have ∆p ∩ ([w(El, p), w(Ek, p)] × R) ⊂ ∆T ∩ ([w(El, p), w(Ek, p)] × R) ⊂ R 2 . By Lemma 3.10, w(Ei, p) =Pi−1 j=0 δ(sj, p); hence ∆p ∩ ([w(El, p), w(Ek, p)] × R) ≤ ∆T ∩ ([w(El, p), w(Ek, p)] × R) ≤ Xi∈p∩[l,k−1] δ(si, p)ρi − 1 2 (ρimin(p) + ρimax(p)). This proves the Lemma. (cid:3) With this lemma in hand, we now explain the key ingredient in the proof of the theorem. We will divide our estimates into two cases according to the size of Λα (cf. Definition 3.1). When Λα is large, applying Lemma 4.2, we will gain a sizable multiple of 1 2 ρi0(p)'s (cf. (4.6)) in the estimate of ∆p; these extra gains will take care of the contributions from non-primary ρi's. When Λα is small, one large ∆p is sufficient to cancel the contribution from the non-primary ρi's. We need a few more notions. For any p ∈ Λα, we let pri p := pri α ∩ p, and define (4.9) ¯p := max{i ∈ pri p }, wpri(p) := w(E¯α+1, p), and w(p) := w(I, p) (cf. (2.9)). Note that w(p) is the base-width of the Newton polygon ∆p. Using ¯p, we truncate the Newton polygon ∆p by intersecting it with the strip [0, wpri(p)] × R: ∆pri p := ∆p ∩ [0, wpri(p)] × R. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 17 Our next Lemma says that if one ∆p is big enough, the contribution from non-primary ρi's will be absorbed by the difference between Eα(ρ) and e(Iα(ρ)). Lemma 4.3. For any 1 > ǫ > 0, there is an M depending only on ǫ, gα and ℓα such that whenever w(p) ≥ M (cf.(4.9)), ∆pri p + 2(deg Xα − wpri α )¯ρ¯α ≤(cid:16)1 + ǫ w(p)(cid:17)Xi∈ pri p δ(si, p)¯ρi + ρα · wpri(p) −(cid:16) 1 2 + ǫ w(p)(cid:17)¯ρi0(p), where wpri α is defined in Corollary 3.14. Proof. By the same reason as in the proof of Lemma 4.2, we only need to treat the case that ρα = 0; hence ¯ρi = ρi. lying in the (open) first quadrant) with the lattice points (w(Ei, p), ρi) (cf. (3.2)). case they differ slightly, then the term ǫ term 2(deg Xα − wpri Our proof relies on the proximity of ∂+∆p (∂+∆p is the boundary component of ∆p In δ(si, p)¯ρi is sufficient to absorb the α )ρ¯α in the inequality (note ¯ρi = ρi by assumption). Otherwise, the p ∩[c,¯p] δ(si, p)¯ρi (for some c that will be specified below ) and difference between Pi∈ Let us assume M > 4, then w(p) −pw(p) ≥ 2 whenever w(p) ≥ M . We introduce ∆p is sufficient to imply the desired estimate. w(p)Pi∈ 2} and wc(p) := w(Ec, p) , (w(Ei, p), pri c = max{i ∈ p ) ∈ ∆p ⊂ R pri p pri ρi 2 and let ∆≤c p = ∆p ∩ [0, wc(p)] × R . We divide our study into two cases. The first is when w(p) − wc(p) ≤ pw(p). We introduce trapezoid Θ to be the region between x-axis and the line passing through the points (w(p), 0) and (wc(p), ρc 2 ) intersecting with the strip [1, wc(p)] × R. Then by our assumption · ρc 4 ≥ w(p)3/2 · ρc 32 . Since the piecewise linear ∂+∆p is convex, Θ lies inside ∆p, hence ∆p − ρi0(p) 2 > Θ > w(p)3/2 32 ρc . By the definition of ∆pri bounded by w(p) − wpri(p); therefore by Lemma 4.2 we have ρi0(p) p , the difference between the base-width of ∆pri p ρi0(p) + 2(deg Xα − wpri α )ρ¯α ≥ ∆pri p − ∆pri p − + (w(p) − wpri(p))ρ¯α and of ∆p is 2 ≥ ∆p − 2 ρi0(p) 2 ≥ w(p)3/2 32 ρc . Since the length of the two vertical edges of Θ are of ρc w(p)−1 w(p)−wc(p) · ρc 2 , we deduce w(p) −pw(p) 2 2 and . wc(p) − 1 ≥ w(p) −pw(p) − 1 ≥ + 1(cid:17)(wc(p) − 1) w(p) −pw(p) Θ = (cid:16) w(p) − 1 ≥ (cid:16)pw(p) + 1(cid:17) · w(p) − wc(p) 2 2 ρc 4 18 JUN LI AND XIAOWEI WANG Since ρ¯α ≤ ρc, this implies (4.10) ∆pri p − ρi0(p) 2 >(cid:16) w(p)3/2 32 − 2(deg Xα − wpri α )(cid:17)ρc . We now choose M so that M 3/2 ≥ 28(gα + ℓα + 1). By Corollary 3.14, we have deg Xα − wpri α ≤ 2(gα + ℓα + 1). Therefore when w(p) ≥ M , we have 2(deg Xα − wpri α ) ≤ 4(gα + ℓα + 1) ≤ w(p)3/2 64 . Pluging this into (4.10), we obtain ∆pri p − ρi0(p) 2 > w(p)3/2 64 ρc , equivalently, ρc ≤ hence 2(deg Xα − wpri α )ρ¯α ≤ 2(deg Xα − wpri α )ρc ≤ p − 26 w(p)3/2(cid:0)∆pri (cid:16)∆pri ρi0(p) 2 (cid:1) , 2 (cid:17). 26(deg Xα − wpri α ) w(p)3/2 ρi0(p) p − So if we assume further M ≥ 214(gα + ℓα + 1)2/ǫ2, then whenever w(p) ≥ M we have 26(deg Xα − wpri α )w(p)−3/2 ≤ ǫ/w(p), thus ∆pri p − 1 2 ρi0(p) + 2(deg Xα − wpri α )ρ¯α ≤ (cid:16)1 + ≤ (cid:16)1 + ǫ w(p)(cid:17)(cid:16)∆pri w(p)(cid:17)(cid:0)Xi∈ pri p p − ρi0(p) 2 (cid:17) δ(si, p)ρi − ρi0(p)(cid:1), The second case is when w(p) − wc(p) >pw(p). By the definition of c, for j ∈ := p ∩ (c, ¯α], (w(Ej , p), ρj/2) /∈ ∆p. Since ∂+∆p is convex, by Lemma 4.2, we have where the last inequality follows from Lemma 4.2. This proves the Lemma in this case. ǫ Xi∈ δ(si, p)ρi − ∆pri p \ ∆≤c p ≥Xi∈ δ(si, p)ρi/2. δ(si, p) = wpri(p) − wc(p) = w(p) − wc(p) − (w(p) − wpri(p)) Notice that Xi∈ > pw(p) − (deg Xα − wpri α ) , since deg Xα − wpri we choose α ≥ w(p) − wpri(p) and w(p) − wc(p) >pw(p) by our assumption. If M ≥ 102(gα + ℓα + 1)2 ≥ 52(deg Xα − wpri α )2 and require w(p) ≥ M , thenPi∈ δ(si, p)ρi − ∆pri p \ ∆≤c Xi∈ p ≥Xi∈ δ(si, p) ≥ 4(deg Xα − wpri α ). This implies δ(si, p)ρi/2 ≥ 2(deg Xα − wpri α )ρ¯α, HILBERT-MUMFORD CRITERION FOR NODAL CURVES 19 and combined with Lemma 4.2, we obtain ∆pri p + 2(deg Xα − wpri α )ρ¯α p \ ∆≤c p \ ∆≤c p + ∆pri p + ∆pri ≤ ∆≤c ≤ ∆≤c p + 2(deg Xα − wpri α )ρ¯α p −Xi∈ δ(si, p)ρi ≤ Xi∈ pri p δ(si, p)ρi − δ(si, p)ρi +Xi∈ δ(si, p)ρi − ρi0(p) 2 ρi0(p) 2 (cid:1) + ρi0(p) 2 . ≤ ∆≤c < (cid:0)1 + p +Xi∈ w(p)(cid:1)(cid:0)Xi∈ ǫ pri p δ(si, p)ρi + 2(deg Xα − wpri α )ρ¯α In the end, since ǫ < 1 we choose M := 214(gα + ℓα + 1)2/ǫ2. Then for w(p) > M , (4.3) (cid:3) holds. This proves the Lemma. Proof of Theorem 4.1. First, by the same reason as in the proof of Lemma 4.3 we only need to deal with the case ρα = 0 and ¯ρi = ρi. Also, when deg Xα = 1, then the statement is a direct consequence of Lemma 4.2. So from now on we assume that deg Xα > 1. Let ≤ 1/2 whenever deg X ≥ M ≥ 2. 1 > ǫ > 0 be any constant. Since ǫ < 1, we have We define σ to be the number of Newton polytopes supported on Xα. We divide our study into two cases. deg Xα ǫ The first case is when σ > 10(gα + ℓα + 1) + Sα. Since Corollary 3.14 implies {p ∈ Λα ∩ Sα i0(p) > ¯α} ≤ Xi∈α\ pri α δα(si) ≤ (deg Xα − wpri α ) ≤ 2(gα + ℓα + 1) , the number of p ∈ Λα \ Sα satisfying ρi0(p) ≥ ρ¯α is at least 8(gα + ℓα + 1). By Lemma 4.2, for each p ∈ Λα, we gain an extra ρi0(p)/2 on the right hand side in the estimate ∆p in terms of {ρi}m i=0. This implies (4.11) Xi∈α\ pri α δα(si)ρi ≤ (deg Xα − wpri α )ρ¯α ≤ 2(gα + ℓα + 1)ρ¯α ≤ 1 4 Xp∈ Λα\ Sα ρi0(p) . So we obtain, via using Lemma 4.2 and summing over p ∈ Λα, δα(si)ρi + Xi∈α\ pri α δα(si)ρi − δα(si)ρi − ρi0(p) 1 2 Xp∈ Λα ρi0(p) − Xp∈ Λα (4.12) pri α ∆p ≤ Xi∈ = (cid:16)Xi∈ +(cid:16) Xi∈α\ pri α ǫ deg Xα Xp∈ Λα∩ Sα pri α i0(p)∈α\ ǫ deg Xα Xp∈ Λα∩ Sα i0(p)∈α\ pri α 1 2 Xp∈ Λα∩ Sα ρi0(p)(cid:17) + ρi0(p)(cid:17). 1 2 Xp∈ Λα\ Sα ρi0(p) − δα(si)ρi + pri α Using (4.11) and ǫ deg Xα Xp∈ Λα∩ Sα i0(p)∈α\ pri α ρi0(p) ≤ Xi∈α\ pri α δα(si)ρi ≤ 1 4 Xp∈ Λα\ Sα ρi0(p) , 20 JUN LI AND XIAOWEI WANG the sum in the line of (4.12) is non-positive. Therefore, for any 0 < ǫ < 1 we have pri α ∆p ≤ Xi∈ ≤(cid:16)1 + Xp∈ Λα since δα(si)ρi − ǫ pri α deg Xα(cid:17) Xi∈ deg Xα Xp∈ Λα∩ Sα ǫ i0(p)∈ pri α ρi0(p) ǫ i0(p)∈α\ deg Xα Xp∈ Λα∩ Sα δα(si)ρi −(cid:16) 1 2 pri α ρi0(p) − 1 2 Xp∈ Λα∩ Sα deg Xα(cid:17) Xp∈ Λα∩ Sα ǫ + ρi0(p) = Eα(ρ) 2 , ρi0(p) ≤ ǫ deg Xα Xi∈ pri α δα(si)ρi . This verifies the Theorem in this case. exists at least one p0 ∈ Λα such that The other case is when σ ≤ 10(gα + ℓα + 1) + Sα. By the pigeon hole principle, there (4.13) w(I, p0) ≥ deg Xα σ ≥ deg Xα 10(gα + ℓα + 1) + Sα . By Corollary 2.7, we have eXα(I(λ)) 2 = Xp∈ Λα ∆p. Our assumption ǫ ≤ 1, 1/ deg X ≤ 1/2 and Corollary 3.14 imply (4.14) + (cid:0) 1 2 So we obtain ǫ deg Xα(cid:1) Xp∈ Sα∩ Λα i0(p)∈α\ pri α ρi0(p) ≤ Xp∈ Sα∩ Λα i0(p)∈α\ pri α ρi0(p) ≤ (deg Xα − wpri α )ρ¯α . eXα (I(λ)) 2 − Xp∈ Sα∩ Λα ρi0(p) 2 = ∆pri p0 + ∆p0 \ ∆pri p0 + Xp06=p∈ Λα (∆pri p + ∆p \ ∆pri p ) − Xp∈ Sα∩ Λα ρi0(p) 2 . By Lemma 4.2 and the first inequality of (4.11), we have ∆p0 \ ∆pri p0 + Xp06=p∈ Λα ∆p \ ∆pri p = Xp∈ Λα ∆p \ ∆pri p ≤ (deg Xα − wpri α )ρ¯α . So eXα(I(λ)) 2 − Xp∈ Sα∩ Λα ρi0(p) 2 ≤ ∆pri p0 + (deg Xα − wpri α )ρ¯α + Xp06=p∈ Λα ∆pri p − Xp∈ Sα∩ Λα ρi0(p) 2 HILBERT-MUMFORD CRITERION FOR NODAL CURVES 21 (4.15) ≤ ∆pri p0 + 2(deg Xα − wpri α )ρ¯α − ρi0(p0) 2 {p0} ∩ Sα + Xp06=p∈ Λα ∆pri p − Xp06=p∈ Sα∩ Λα ρi0(p) 2 + −(cid:0) 1 2 ǫ deg Xα(cid:1) Xp∈ Sα∩ Λα i0(p)∈α\ pri α ρi0(p) where we have used (4.14) in (4.15). By definition, Sα ≤ n + ℓα + gα. Let ǫ0 = ǫ 11(gα + ℓα + 1) + n ǫ 10(gα + ℓα + 1) + Sα , by (4.13) we obtain (4.16) ǫ0 w(I, p0) ≤ w(I, p0)(10(gα + ℓα + 1) + Sα) ≤ ǫ deg Xα ≤ ǫ If we let M = M (ǫ0) be the constant fixed in Lemma 4.3 for ǫ0 and choose M ′ ≥ (11(gα + ℓα + 1) + n)M ≥ (10(gα + ℓα + 1) + Sα)M, then deg Xα ≥ M ′ implies w(I, p0) > M . In particular, we have i0(p0) ∈ term after (4.15) is equal to pri α . The whole = ∆pri p0 + 2(deg Xα − wpri α )ρ¯α − ρi0(p0) 2 {p0} ∩ Sα + Xp06=p∈ Λα ∆pri p − Xp06=p∈ Sα∩ Λα i0(p)∈ pri α ρi0(p) 2 −(cid:0)1 + ǫ deg Xα(cid:1) Xp∈ Sα∩ Λα i0(p)∈α\ pri α ρi0(p). Applying Lemma 4.2 to the term ∆pri p0 + 2(deg Xα − wpri α )ρ¯α, Lemma 4.3 to the term in the above identity and using (4.16) we ρi0 (p) 2 pri α Pp06=p∈ Λα obtain ∆pri ρi0(p) 2 eXα (I(λ)) p −Pp06=p∈ Sα∩ Λα,i0(p)∈ − Xp∈ Sα∩ Λα w(I, p0)(cid:17)(Xi∈ pri p0 2 ǫ0 ≤ (cid:16)1 + +(cid:0) Xp06=p∈ Λα Xi∈ ≤ (cid:16)1 + = (cid:16)1 + ǫ pri p ǫ deg Xα(cid:17)(cid:0)Xp∈ Λα Xi∈ deg Xα(cid:17)(cid:0)Xi∈ − Xp∈ Sα∩ Λα ρi0(p) pri α 2 . = Eα(ρ) 2 δ(si, p0)ρi − ρi0(p0)) + ρi0(p0) 2 (1 − {p0} ∩ Sα) + ǫ deg Xα(cid:1) Xp∈ Sα∩ Λα i0(p)∈α\ pri α ρi0(p) 2 δ(si, p)ρi − Xp06=p∈ Sα∩ Λα i0(p)∈ pri α ρi0(p)(cid:1) −(cid:0)1 + ρi0(p)(cid:1) pri p δ(si, p)ρi − Xp∈ Sα∩ Λα ρi0(p)(cid:1) δα(si)ρi − Xp∈ Sα∩ Λα This completes the proof the theorem. (cid:3) 22 JUN LI AND XIAOWEI WANG 5. Stability of weighted pointed nodal curve We prove Theorem 1.5 in this section. By Proposition 3.6, it suffices to verify the positivity of the a-λ-weight ωa(λ) of Chow(X, x) ∈ Ξ for any staircase λ. Let s be a diagonalizing basis of λ: λ(t) := diag[tρ0 , · · · , tρm] · t−ρave , with ρ0 ≥ ρ1 ≥ · · · ≥ ρm = 0. The a-λ-weight of Chow(X, x) is the sum of the contributions from Divd,d[(W ∨)2] and (W )n. By Proposition 2.1, the contribution from Divd,d[(W ∨)2] is ω(λ). For the contribution from (W )n, we introduce subspaces (5.1) Wi = Wi(λ) := {v ∈ W si(v) = · · · = sm(v) = 0} ⊂ W = H 0(OX (1))∨. They form a strictly increasing filtration of W . Also, for any closed subscheme Σ ⊂ X, we denote by (5.2) WΣ := {v ∈ W s(v) = 0 for all s ∈ H 0(OX (1) ⊗ IΣ)} ⊂ W the linear subspace spanned by Σ ⊂ X. For instance, for a marked point xi, Wxi is the line in W spanned by xi ∈ W ; for any i and (5.3) Zi = {si = · · · = sm = 0} ⊂ X, we let Wi = WZi . By [14, Prop 4.3], the a-λ-weight of x = (x1, · · · , xn) ∈ (W )n is (5.4) µa(λ) := aj(cid:16)Pm i=0 ρi m + 1 nXj=1 + (ρi+1 − ρi) dim(Wxj ∩ Wi+1(λ))(cid:17). m−1Xi=0 (µa(λ) implicitly depends on ρi, which we fix for the moment.) Therefore, the a-λ-weight ωa(λ) of Chow(X, x) ∈ Ξ is (5.5) ωa(λ) = ω(λ) + µa(λ) . We now argue that for the staircase λ′ constructed from λ by applying Proposition 3.6, we have (5.6) ωa(λ) ≥ ωa(λ′). Indeed, since ω(λ) ≥ ω(λ′), if suffices to show that µa(λ) ≥ µa(λ′). To see that, we first notice that (5.7) dim(Wxj ∩ Wi+1(λ)) = #(cid:0)xi ∩ Supp(OX (1)/E(λ)i+1)(cid:1). (Here E(λ)i = (si, si+1, . . . , sm) ⊂ OX (1).) On the other hand, by the proof of Proposition 3.6, we conclude Supp(OX (1)/E(λ)i) ⊂ Supp(OX (1)/E(λ′)i). This together with (5.7) proves dim(Wxj ∩ Wi+1(λ)) ≤ dim(Wxj ∩ Wi+1(λ′)). The inequality µa(λ) ≥ µa(λ′) then follows from ρi ≥ ρi+1. Therefore, to prove Theorem 1.5, we suffices to show ωa(λ) > 0 for all staircase 1-PS's λ. From now on we assume λ is a staircase. Before we proceed, we collect a few boundness results that are needed to pass from the estimates on single component in Section 4 to the entire curve. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 23 Lemma 5.1. Suppose (X, OX (1), x, a) is slope stable and χa(X) > 0 (cf. Theorem 1.5). n such that Then there are positive constants M and C depending only on g,n and a ∈ whenever deg X > M we have deg Y ≥ C deg X for any connected proper subcurve Y ⊂ X. In case (X, OX (1), x, a) is slope semi-stable, then the same conclusion deg Y ≥ C deg X holds except when Y is a line, Y ∩ x = ∅, Y ∩ Y ∁ = 2, and the inequality (1.6) is an equality. Proof. We continue to denote ℓY = Y ∩ Y ∁. Let gY be the arithmetic genus of Y . Suppose 2gY + ℓY ≥ 3, then since ai ≥ 0, the inequality (1.6) implies deg Y +Pxj∈Y aj/2 gY − 1 +Pxj ∈Y aj/2 + ℓY /2 j=1 aj/2 j=1 aj/2 − j=1 aj/2 j=1 aj/2 − 3. ℓY 2(gY − 1) +Pxj∈Y aj + ℓY ≥ ≥ deg X +Pn g − 1 +Pn deg X +Pn g − 1 +Pn deg Y ≥(cid:16) deg X 2χa − 6(cid:17) − n 2 . Denoting χa = χa(X), this inequality implies Therefore, if we choose M ′ = 4χa · (6 + n/2), choose C′ = 1/4χa and require deg X ≥ M ′, we obtain Now suppose 2gY + ℓY ≤ 2. Since Y ⊂ X is a proper connected subcurve, ℓY ≥ 1 and deg Y ≥ deg X/4χ = C′ · deg X. gY ≥ 0. Thus gY = 0 and ℓY ≤ 2. In this case, (1.6) becomes (cid:12)(cid:12)(cid:12) deg Y + Xxj ∈Y aj 2 − deg X +Pn g − 1 +Pn j=1 aj/2 j=1 aj/2 ·(cid:0)−1 + Xxj ∈Y aj/2 + ℓY /2(cid:1)(cid:12)(cid:12)(cid:12) ≤ 1. Let A := −1 +Pxj ∈Y aj/2 + ℓY /2. In case A ≤ 0, then we have deg Y = 1,Pxi∈Y ai = 0 and ℓY = 2. This is precisely the second case in the statement of the Lemma. In case A > 0. then (5.8) deg Y > j=1 aj/2 · 2A − 1 − Xxj∈Y aj 2 > deg X χa A, provided deg X > (2 + n)χa/2A. To obtain a universal constant, we intorduce Ck := min ai/2 + k/2 > 0}, ai/2 + k/2 Xi∈I and define Cmin := mink≥0{Ck}. Clearly, Cmin > 0. By our construction, A ≥ Cmin when A > 0. We choose C′′ = min{Cmin/χa, 1/2χa}, and choose M ′′ := max{6g + n/2 − 6, χa(2 + n)/2Cmin}. Our discussion shows that the statement (1) in the Lemma holds in the case under study with this choices of M ′′ and C′′. Finally, we let C = min{C′, C′′} and M = max{M ′, M ′′}. The Lemma holds with this choice of C and M in all cases. This completes the proof of the Lemma. (cid:3) Corollary 5.2. Let M be as in Lemma 5.1 and suppose deg X ≥ M . Then the number of irreducible components of X is at most 9g + 4n − 5; the number of nodes of X is at most 10g + 4n − 5. χa deg X +Pn {Xi∈I I⊂{1,··· ,n} 24 JUN LI AND XIAOWEI WANG Proof. We divide the irreducible components of X into three categories: the first (resp. second; resp. third) consists of irreducible components of X ∈ W that are not lines (resp. are lines that contains marked points; resp. are lines in W that contains no marked points). Applying the previous Lemma, for Xα in the first category, deg Xα > deg X/4χa; thus the first category contains no more than 4χa elements. Since every component in the second category contains at least one marked point, there are at most n elements in this category. We now bound the element in the third category. We let B1 be a maximal subcollection of the third category so that X −∪α∈B1Xα remains connected. We let Y = X − ∪α∈B1 Xα. Then g(Y ) = g − B1 ≥ 0 since Y is connected. We let B2 be the complement of B1 in the third category. By Lemma 5.1, Xα and Xα′ are disjoint for α 6= α′ ∈ B2. Thus if we let Y1, · · · , Yk be the connected components of Y − ∪α∈B2 Xα, then B2 ≤ k − 1. On the other hand, applying Lemma 5.1 we know each Yi has degree at least deg X/4χa. Thus k ≤ 4χa. Combined, the total number r of irreducible components of X is bounded by r ≤ 4χa + n + (g + 4χa − 1) ≤ 9g + 4n − 5. Next we bound the total nodes of X. We first pick a maximal set A1 ⊂ Xnode so that X − A1 is connected; then A1 ≤ g. We let A2 = Xnode − A1. Then A2 is less than the number of irreducible components of X. By the bound we just derived, we obtain Xnode ≤ g + (9g + 4n − 5). This proves the Corollary. (cid:3) We need the next consequence in our proof of Theorem 1.5 to replace inequality (1.6) by (1.4). Corollary 5.3. Let X ⊂ H 0(OX (1))∨ be a connected nodal curve of arithmetic genus g > 0. Suppose χa(X) > 0, then there is an M depending on g such that whenever deg X ≥ M , X satisfies (1.6) for any subcurve Y ⊂ X if and only if it satisfies (1.4) for any subcurve Y ⊂ X , Proof. First, we claim that there is an M depending only on g, n such that for deg X ≥ M , we have h1(OX (1)Y ) = 0 for any subcurve Y ⊂ X (not necessary proper) provided X satisfies either (1.6) or (1.4) for any proper subcurve Y ( X. To to that, let Y ⊂ X be any subcurve with arithmetic genus gY . Suppose h1(OX (1)Y ) ≥ 1, then by the vanishing theorem we must have gY > 1 and 2 < deg Y < 2gY − 1 ≤ 2g − 1. Suppose X satisfies (1.6), then the claim is an easy consequence of Lemma 5.1 by letting M = max{(2g + ν)/C, 2g + ν} with C being chosen in Lemma 5.1 and ν being the total number of nodes of X. Suppose X satisfies (1.4), then the claim was proved in [7, Proposition 1.0.7]. Finally, when h1(OX (1)Y ) = 0 for any subcurve Y ⊂ X the equivalence of (1.6) and (1.4) follows from an argument parallel to the one given in [2, Proposition 3.1] and [7, Proposition 1.0.7]. So we omit it. (cid:3) Before we state our key estimate of this section, we first notice that λ being a staircase α=1 α = {0, · · · , m}, where α is the index set of 1-PS (cf. Definition 3.8) implies thatSr the component Xα defined in (3.3). We define the shifted weights { ρi} by (5.9) ρi := min {ρi − ρα i ∈ α} ≥ 0. α (ρi may not be monotone. And ρi are only defined for staircase 1-PS.) We state out main estimate. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 25 Proposition 5.4. Suppose (X, OX(1), x, a) is slope stable (cf. (1.4)), and ωX(a · x) is ample. Then for any 0 < ǫ < 1 there exists an M depending only on χa(X) (cf. Theorem 1.5) and ǫ such that whenever deg X > M , then for any staircase 1-PS λ we have (5.10) EX (λ, ρ) 2 ≤ mXi=0 ρi −Xq∈ Sreg ρi0(q) 2 + rXα=1(cid:16)deg Xα + ℓα 2 − mα − 1(cid:17)·ρα + 2C−1ǫ m + 1 ρi . mXi=0 α=1(π−1(x) ∩ Xα ∩ Λ) is the support of weighted points and C > 0 is the constant fixed in Lemma 5.1. where Sreg :=Sr Proof. By the definition of Eα(ρ) (cf.(4.3)), EX (λ, ρ) =Pr By linear programming, (5.10) holds on m+1 + R := {(ρ0, · · · , ρm) ∈ R m+1 ρ0 ≥ ρ1 ≥ · · · ≥ ρm = 0 } α=1 Eα(ρ) is linear in ρ = (ρi). if and only if it holds on every edge of R (5.11) for every 0 < m0 < m. m+1 + ; that is on m0 ρ = ( 1, · · · , 1, 0, · · · , 0) z } { We now fix a 0 < m0 < m. By possibly reindexing the irreducible components of X, we can assume that for an ¯r ≤ r, 1 ≤ · · · ≤ ¯r < m0 ≤ ¯r+1 ≤ · · · ≤ r. In other words, (5.12) ρ1 = · · · = ρ¯r = 1, and ρ¯r+1 = · · · = ρr = 0. We let Y := [α>¯r Xα; let its complement Y ∁ = [α≤¯r Xα. We claim that Y ∁ is the maximal subcurve of X contained in the linear subspace Wm0 (cf. (5.1)). By definition, for any α, α is the largest index 0 < i ≤ m of which siXα 6= 0. On the other hand, because Wm0 = {sm0 = · · · = sm = 0}, Xα ⊂ Wm0 if and only if siXα = 0 for all i ≥ m0, which is equivalent to α < m0. This proves the claim. Let Xα be a component in Y ∁. Since ρα = 1, ρi = 1 for i ∈ α. Using the explicit expression of Eα(λ, ρ), we obtain Eα(λ, ρ) = 2 deg Xα. Thus Xα≤¯r 2 deg Xα = 2 deg Y ∁ . Eα(λ, ρ) =Xα≤¯r Y ∁) ∩ Y . We claim that LY ⊂ ΛY :=Sα>¯r We now look at Y . Following (3.9) and (3.10), ℓY := Y ∩ Y ∁, and LY := π−1(Y ∩ Λα. Indeed, for any α > ¯r, there is an i ≥ m0 so that siXα 6= 0. However for any β ≤ ¯r, i ≥ m0 implies siXβ = 0. Thus siXα∩Xβ = 0, and consequently, π−1(Xα ∩ Xβ) ∩ Xα ⊂ Λα. Summing over all α > ¯r and β ≤ ¯r, we obtain LY ⊂ ΛY . As a consequence, (5.13) ρi0(p) = ℓY . Xp∈ LY ρi0(p) := Xp∈A∩ Λ Xp∈A ρi0(p), To simplify the notation, in the remaining part of this section, we will abbreviate with the understanding that for any A ⊂ X, the summation Pp∈A only sums over p ∈ A ∩ Λ. 26 JUN LI AND XIAOWEI WANG Sublemma 5.5. Let the notation be as before. Then Eα(λ, ρ) 2 − ℓY 2 Xα>¯r ≤(cid:16)1+ ǫ deg X(cid:17)(cid:16)Xα>¯r i∈ pri α δα(si)ρi−Xp∈ LY ρi0(p)− Xp∈ NY \ LY ρi0(p) 2 (cid:17)− Xπ(p)∈x∩Y ρi0(p) 2 Proof. We let Xα ⊂ Y be an irreducible component, then α > ¯r and ρα = 0. Since (X, OX(1), x, a) is slope semi-stable, by Corollary 5.1, there are positive constants C and N such that whenever deg X ≥ N , either deg Xα > C deg X or deg Xα = 1. If deg Xα > C deg X, from the definition of Eα(ρ) (cf. (4.3) ) and ρα = 0, we have (5.14) Eα(λ, ρ)) 2 ≤ (cid:16)1 + C−1 · ǫ deg X (cid:17) Xi∈ pri α δα(si)ρi −(cid:16) 1 2 + C−1 · ǫ deg X (cid:17) Xp∈ Sα ρi0(p). If deg Xα = 1, (5.14) remain holds since by (4.4) and Definition 3.12, we have Eα(λ, ρ)) 2 = δα(sind−1 α (0)) · δα(si) · ρi 2 pri α ¯ρind−1 α (0) 2 + ρα = Xi∈ δα(si)ρi −(cid:16) 1 ρi0(p) − Xp∈ Nα\ LY pri α + 2 C−1 · ǫ deg X (cid:17) Xp∈ Sα ρi0(p) − Xp∈ LY ∩ Xα deg X (cid:17) Xi∈ C−1 · ǫ ≤ (cid:16)1 + ρi0(p) = − Xp∈π−1(x)∩ Xα ρi0(p). ρi0(p). Then using ρi ≥ 0, we get Next we split − Xp∈ Sα deg X (cid:17) Xp∈ Sα C−1 · ǫ −(cid:16) 1 2 + ρi0(p) ≤ − Xp∈π−1(x)∩ Xα C−1 · ǫ 2 2 + ρi0(p) −(cid:16) 1 deg X (cid:17) Xp∈ LY ∩ Xα −(cid:16)1 + deg X (cid:17) Xp∈ LY ∩ Xα ρi0(p) 2 C−1 · ǫ deg X (cid:17) Xp∈ Nα\ LY ρi0(p) + Xp∈ LY ∩ Xα deg X (cid:17) Xp∈ Nα\ LY ρi0(p) + Xp∈ LY ∩ Xα C−1 · ǫ C−1 · ǫ −(cid:16)1 + ≤ − Xp∈π−1(x)∩ Xα −(cid:16)1 + ρi0(p) − ρi0(p) 2 ρi0(p) 2 − ρi0(p) 2 . Putting together, we obtain Eα(λ, ρ)) 2 ≤ (cid:16)1 + − Xp∈ Nα\ LY C−1 · ǫ deg X (cid:17)(cid:16)Xi∈ pri α δα(si)ρi − Xp∈ LY ∩ Xα ρi0(p) − ρi0(p) 2 (cid:17) + Xp∈ LY ∩ Xα ρi0(p) 2 − Xp∈π−1(x)∩ Xα ρi0(p) 2 . Summing over α and applying (5.13) prove the Lemma. (cid:3) The following inequality is crucial for the proof of the Proposition. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 27 Lemma 5.6. For 1 ≤ k ≤ m0, we have δα(si)ρi − Xp∈ LY i0(p)<k ρi0(p) − Xp∈ NY \ LY i0(p)<k Xα>¯r i∈ pri α ∩[0,k) ρi0(p) 2 ≤ dim WY ∩ Wk − dim WY ∩Y ∁ ∩ Wk, where WY ∩Y ∁ is the linear subspace in W spanned by Y ∩ Y ∁. Proof. We prove the Lemma by induction on k. When k = 0, then both sides of the inequality are zero, and the inequality follows. Suppose the Lemma holds for a 0 ≤ k < m0. Then the Lemma holds for k + 1 if for the expressions Ak,1 := Xα>¯r, k∈ pri α δα(sk)ρk, Ak,2 := Xp∈ LY , i0(p)=k ρi0(p), Ak,3 := Xp∈ NY \ LY , i0(p)=k ρi0(p) 2 and Bk,1 := dim WY ∩Wk+1 − dim WY ∩Wk, Bk,2 := dim WY ∩Y ∁ ∩Wk+1 − dim WY ∩Y ∁ ∩Wk, the following inequality holds (5.15) Ak,1 − Ak,2 − Ak,3 ≤ Bk,1 − Bk,2. To study the left hand side of (5.15), we introduce the set (5.16) Rk = {p ∈ Y k ∈ pri p }. By Proposition 3.11 and 3.13, Rk can take three possibilities according to (5.17) Xα>¯r, k∈ δα(sk) pri α taking values 0, 1 or ≥ 2. Notice that if Ak,1 = 0, then Ak,1 − Ak,2 − Ak,3 ≤ 0. The Lemma holds trivially in this case since the right hand side of (5.15) is non-negative. So from now on, we will assume that Ak,1 ≥ 1, in particular, (5.17) is positive. We first observe that since dim Wk+1 − dim Wk = 1, both Bk,1 and Bk,2 can only take values 0 or 1. We now investigate the case when Bk,2 = 1. Claim 5.7. Suppose (5.17) is positive and Bk,2 = 1. Then there is a p ∈ Rk (cf.(5.16))such that i0(p) = k and (5.18) q = π(p) ∈ Y ∩ Y ∁ ∩ (Wk+1 − Wk). Proof. Suppose (5.17) is positive then there is a p ∈ inc(sk) ∩ Xα with α > ¯r and k ∈ pri α . Let Zk be the subscheme defined in (5.3) and WZk+q ) Wk be defined in (5.2). Then Wk+1 = WZk+q, since dim Wk+1 = dim Wk + 1. Suppose q = π(p) 6∈ Y ∩ Y ∁ and k ∈ pri α then by applying the argument parallel to Proposition 3.11 and 3.13, we deduce (5.19) WZk+q + WY ∩Y ∁ ) WZk + WY ∩Y ∁ . On the other hand, Bk,2 = 1 implies that dim(Wk + WY ∩Y ∁ ) = dim Wk + dim WY ∩Y ∁ − dim Wk ∩ WY ∩Y ∁ = dim Wk+1 + dim WY ∩Y ∁ − dim Wk+1 ∩ WY ∩Y ∁ = dim(Wk+1 + WY ∩Y ∁ ) , which means Wk + WY ∩Y ∁ = Wk+1 + WY ∩Y ∁ contradicting to (5.19). So we must have q ∈ Y ∩ Y ∁. 28 JUN LI AND XIAOWEI WANG By definition, q ∈ Wk+1 (cf. (5.1)) implies that si(q) = 0 for i ≥ k + 1; q 6∈ Wk implies that not all si(q), k ≤ i ≤ m, are zero. Combined, we have sk(q) 6= 0. This implies i0(q) = k. As an easy consequence, this shows that Bk,2 = 1 forces WY ∩Wk+1 6= WY ∩Wk, and hence Bk,1 = 1. In particular, the right hand side of (5.15) is non-negative. This proves the Claim. (cid:3) We complete our proof of Lemma (5.6). When (5.17) takes value 1, then Rk consists of a single point, say p ∈ Y . In case π(p) ∈ Y is a smooth point of X, Ak,1 = 1 and Ak,2 = Ak,3 = 0. We claim that Bk,1 = 1 and Bk,2 = 0. Indeed, if Bk,1 = 0, then WY ∩ Wk+1 = WY ∩ Wk, which is the same as Y ∩(sk = · · · = sm = 0) = Y ∩(sk+1 = δα(sk) = 1. Thus Bk,1 = 1. On the other hand, if Bk,2 = 1, then Claim 5.7 shows that Rk ∩ Y contains an element in LY , contracting to our assumption that Rk = {p} lies over a smooth point of X. · · · = sm = 0) as subschemes of Y . But this contradicts toPα>¯r, k∈ In case p ∈ LY , then the previous paragraph shows that Ak,1 = Bk,1 = 1, Ak,3 = 0. For the values of Ak,2 and Bk,2, when i0(p) = k, then both Ak,2 = Bk,2 = 1; when i0(p) 6= k, then both Ak,2 = Bk,2 = 0. Therefore, (5.15) holds. pri α The last case is when p ∈ NY − LY . In this case, since the point p′ in Y ∩ π−1(π(p)) pri p′ . In other than p is not contained in Rk, either i0(p) 6= k or i0(p) = i0(p′) = k and k 6∈ both cases, Ak,1 = Bk,1 = 1, and Bk,2 = 0; the inequality (5.15) holds. Lastly, when (5.17) is bigger than 1, by Proposition 3.11 and 3.13, either Rk = {p−, p+} such that π(p−) = π(p+) is a node of Y , i.e. p± ∈ NY , and i0(p−) = i0(p+) = k, or Rk = {p1, · · · , pl} so that i0(pi) = k and {π(pi)}1≤i≤s are distinct nodes of X. In case Rk = {p−, p+}; since p± ∈ NY \ LY , Ak,1 = 2, Ak,2 = Bk,2 = 0, and Ak,3 = Bk,1 = 1. The inequality (5.15) holds in this case. i) = k as well, but k 6∈ The other case is when Rk = {p1, · · · , pl}. By reindexing, we may assume p1, · · · , pl1 are in NY \ LY and pl1+1, · · · , pl are in LY . We let p′ i ∈ Y be such that π−1(π(pi)) = {pi, p′ i} pri for i ≤ l1. Then i0(p′ because of Proposition 3.11 and 3.13. p′ i This in particular implies that the interior linking nodes NY \ LY contributes once in Ak,1 i) appear but twice in Ak,3 (, e.g. only ρi0(pi) appears in Ak,1, but both ρi0(pi) and ρi0(p′ in Ak,3). Therefore, Ak,1 = l; Ak,2 = l − l1, and Ak,3 = 2l1/2 = l1. Hence the left hand side of (5.15) is 0. This proves (5.15) in this case; hence for all cases. This proves the Lemma. (cid:3) We continue our proof of Proposition 5.4. We apply Lemma 5.5 and Lemma 5.6 with k = m0. Noticing ρi0(p) = 0 for i0(p) > m0, we obtain (5.20) EY (λ, ρ) 2 − ℓY 2 = Eα(λ, ρ) 2 − ℓY 2 ≤ rXα=¯r+1 ≤ (cid:16)1 + ≤ (cid:16)1 + C−1ǫ deg X(cid:17)(cid:16)dim WY ∩ Wm0 − ℓY(cid:17) − deg X(cid:17)(cid:16)dim WY ∩ Wm0 − ℓY(cid:17) − C−1ǫ ρi0(p). 1 2 Xπ(p)∈x∩π( Λ)∩Y 2 Xp∈ Sreg ρi0(p). 1 HILBERT-MUMFORD CRITERION FOR NODAL CURVES 29 Here we used that for all p′ ∈ π( Sreg) − X ∩ π(Λ) ∩ Y , ρi0 (p′) = 0. And the last in- ρi0(q) ≤ equality holds since by the Definition of Sreg and ρi (cf. (5.9)), we havePq∈ Sreg Pπ(p)∈x∩π( Λ)∩Y ρi0(p). Using EY ∁ (λ, ρ) = 2 deg Y ∁ and deg X − g = m, we obtain EX (λ, ρ) 2 ≤ (cid:16)deg Y ∁ + =(cid:16)deg Y ∁ + 2 (cid:17) +(cid:16)1 + ℓY − 2 ℓY ℓY 2 (cid:17) +(cid:16) EY (λ, ρ) m + 1(cid:17)(cid:16)m0 + 1 − dim WY(cid:17) − 2 (cid:17) 2C−1ǫ 1 2 Xp∈ Sreg ρi0(p) , Here the last inequality follows from dim Wm0 ≥ dim(Wm0 ∩ WY + Wm0 ∩ WY ∁ ) = dim Wm0 ∩ WY + dim Wm0 ∩ WY ∁ − dim Wm0 ∩ WY ∩ WY ∁ = dim Wm0 ∩ WY + dim WY ∁ − ℓY . Now we consider the right hand side of (5.10) for ρ chosen as in (5.11), which gives i=0 ρi = m0 + 1. Since by our assumption, the embedding X ⊂ W is given by a complete linear system of a very ample line bundle OX(1), using our choice of weights ρi (cf. (5.12)), Pm rXα=1(cid:16)deg Xα + ℓα 2 − mα − 1(cid:17) · ρα = deg Y ∁ + ℓY 2 − dim WY ∁ . i=0 ρi = m0 + 1 − dim WY ∁ . Indeed, from our choice of ρ and the definition of ρ (cf. (5.9)), for any 0 ≤ i ≤ m, ρi = 1 or 0, and it is 0 if and only if either i > m0 of there is an Xα with i ∈ α (cf. (3.3)) such that ρα = 1, that is, We claim that Pm i ∈ Y ∁ =SXα⊂Y ∁ α. This proves mXi=0 ρi = m0 + 1 − Y ∁. Our claim will follow if once we prove Y ∁ = dim WY ∁ ; but this follows from the criteria (5.21) i ∈ Y ∁ if and only if dim Wi+1 ∩WXα −dim Wi ∩WXα = 1 for some Xα ⊂ Y ∁. To justify this criteria, we notice that dim Wi+1 ∩ WXα = dim Wi ∩ WXα for all Xα ⊂ Y ∁ is equivalent to Y ∁ ∩ {si = · · · = sm = 0} = Y ∁ ∩ {si+1 = · · · = sm = 0} as subschemes of Y ∁; that is, inc(si) ∩ Y ∁ = ∅. Since λ is a staircase i 6∈ α for all Xα ⊂ Y ∁ if and only if inc(si) ∩ Y ∁ = ∅ (cf. (3.3)) . This proves (5.21). 30 JUN LI AND XIAOWEI WANG With those in hand, we obtain EX (λ, ρ) 2 =(cid:16)deg Y ∁ + 2 (cid:17) +(cid:16)1 + ℓY ≤ (cid:16)deg Y ∁ + ≤ m0 + 1 − Xq∈ Sreg mXi=0 ρi −Xq∈ Sreg = ρi0(q) 2 ρi0(q) 2 + 2 ℓY − ℓY 2C−1ǫ 2 (cid:17) 2 (cid:17) +(cid:16) EY (λ, ρ) m + 1(cid:17)(cid:16)m0 + 1 − dim WY ∁(cid:17) − Xq∈ Sreg +(cid:16)deg Y ∁ + rXα=1(cid:16)deg Xα + − dim WY ∁(cid:17) + − mα − 1(cid:17) · ρα + 2C−1ǫ ℓα 2 ℓY 2 ρi0(q) 2 m + 1(cid:16)m0 + 1 − dim WY ∁(cid:17) 2C−1ǫ m + 1 ρi . mXi=0 So the proof of Proposition is completed. (cid:3) We state and prove the main result of this section. We introduce ρi − EX (λ, ρ) and ωa(λ) = ω(λ) + µa(λ). (5.22) ω(λ, ρ) := 2 deg X m + 1 where EX (λ, ρ) :=Pr mXi=0 α=1 Eα(ρ). By Theorem 4.1, we have ω(λ) ≥ ω(λ). Theorem 5.8. Let (X, OX (1), x, a) be a connected weighted pointed nodal curve that is slope stable. Suppose ωX(a · x) is ample. We let 1 > ǫ > 0 be such that 2(2C−1 + 1)ǫ < deg ωX (a · x) with C given in Lemma 5.1. Then there exists an M depending only on χa(X) and ǫ such that whenever deg X > M , then for any staircase 1-PS λ we have (5.23) ωa(λ) = ω(λ) + µa(λ) ≥ ω(λ) + µa(λ) ≥ ρi . By linear programming, we only need to prove the above estimate for ρ of the form (5.11). We will break the verification into several inequalities. First, we have (5.25) µa(λ, ρ) = aj m0 + 1 m + 1 nXj=1 − Xxj ∈Y ∩Wm0 Here xj runs through all marked points of the curve. We claim that (5.26) Xq∈ Sreg ρi0(q) 2 = x ∩ π(Λ) ∩ Y ∩ Wm0 2 To this purpose, we first show that (5.27) x ∩ π(Λ) ∩ Y ∩ Wm0 = x ∩ Y ∩ Wm0 . aj. aj − Xxj∈Y ∁∩Wm0 ≥ Xxj ∈Y ∩Wm0 aj 2 . 2 · ǫ m + 1 mXi=0 − mα − 1(cid:17) · ρα + (2C−1 + 1)ǫ m + 1 ρi ≤ mXi=0 Proof. By Proposition 5.4, it suffices to prove (5.24) mXi=0 ρi −Xq∈ Sreg mXi=0 deg X m + 1 ≤ ρi0(q) 2 + rXα=1(cid:16)deg Xα + ℓα 2 ρi + µa(λ, ρ) 2 . HILBERT-MUMFORD CRITERION FOR NODAL CURVES 31 Indeed, for any xi in x that lies in Y ∩ Wm0 , sk(xj ) = 0 for k ≥ m0. On the other hand, let xj ∈ Xα ⊂ Y ; since Y ∁ is the largest subcurve of X contained in Wm0 , for some k ≥ m0, skXα 6= 0. Combined with sk(xj) = 0, we conclude xj ∈ π(Λ) (cf. Definition 3.1). In particular x ∩ Y ∩ Wm0 ⊂ π(Λ). This proves (5.27). Applying (5.27), and using that for any colliding subset {xi1 , · · · , xis } (i.e. xi1 = · · · = xis ) necessarily ai1 + · · · + ais ≤ 1, we obtain (5.28) Xxj ∈Y ∩Wm0 aj 2 − x ∩ π(Λ) ∩ Y ∩ Wm0 2 = Xxj ∈Y ∩Wm0 aj 2 − x ∩ Y ∩ Wm0 2 ≤ 0, hence (5.26). By putting (5.25) and (5.26) together, we obtain (5.29) −Xq∈ Sreg ρi0(q) 2 − µa(λ, ρ) 2 ≤ − m0 + 1 m + 1 aj 2 nXj=1 + Xxj∈Y ∁∩Wm0 aj 2 . On the other hand, for ρ of the form in (5.11), we have (5.30) ρi + rXα=1(cid:0)deg Xα + mXi=0 = m0 + 1 +(cid:0)deg Y ∁ + ℓY 2 ℓα 2 − mα − 1(cid:1) · ρα + − dim WY ∁(cid:1) + (2C−1 + 1)ǫ m + 1 ρi mXi=0 (2C−1 + 1)ǫ m + 1 (cid:0)m0 + 1 − dim WY ∁(cid:1) . Pluging (5.30) and (5.29) into (5.24), and using the slope condition deg Y ∁ + ℓY 2 + Xxj∈Y ∁∩Wm0 aj 2 ≤ deg X +Pn m + 1 j=1 aj 2 dim WY ∁ , JUN LI AND XIAOWEI WANG 32 we obtain − µa(λ, ρ) 2 + EX (λ, ρ) 2 + ((2C−1 + 1)ǫ m + 1 ρi mXi=0 − dim WY ∁(cid:17) − m0 + 1 m + 1 ai 2 + nXi=1 dim WY ∁ − m0 + 1 m + 1 ai 2 + nXi=1 ≤ m0 + 1 +(cid:16)deg Y ∁ + (2C−1 + 1)ǫ + m + 1 deg Y ∁ + ℓY aj 2 ℓY 2 + Xxj∈Y ∁∩Wm0 (cid:16)m0 + 1 − dim WY ∁(cid:17) 2 +Pxj ∈Y ∁∩Wm0 aj 2 = ≤ ≤ = dim WY ∁ (2C−1 + 1)ǫ +(cid:16)1 + deg X +Pn deg X +Pn m + 1 m + 1 m + 1 aj 2 j=1 aj 2 j=1 (2C−1 + 1)ǫ (cid:17)(cid:16)m0 + 1 − dim WY ∁(cid:17) dim WY ∁ +(cid:16)1 + (cid:16)dim WY ∁ + m0 + 1 − dim WY ∁(cid:17) − m + 1 deg X m + 1 · (m0 + 1) = deg X m + 1 ρi , mXi=0 (cid:17)(cid:16)m0 + 1 − dim WY ∁(cid:17)− m0 + 1 m + 1 ai 2 nXi=1 m0 + 1 m + 1 ai 2 nXi=1 where we have use the assumption 2(2C−1 + 1)ǫ < deg ωX (a · x) to conclude deg X +Pn m + 1 j=1 aj 2 > 1 + (2C−1 + 1)ǫ m + 1 in the fourth inequality. This completes the proof. (cid:3) Proof of Theorem 1.5. Since ρi ≥ 0, the sufficiency follows from Theorem 5.8. We now prove the other direction. Let Y ⊂ X be any proper subcurve; let WY ⊂ W be the linear subspace spanned by Y , and let m0 + 1 = dim WY . We choose a 1-PS λ = diag[tρ0 , · · · , tρm] · t−ρave such that the corresponding filtration {Wi}m i=0 satisfies Wm0+1 = WY ; we choose the weights {ρi} be as in (5.11). Then µa = nXj=1 aj(cid:16) m0 + 1 m + 1(cid:17) − nXxj∈WY aj. Thus by Corollary 2.7 (cf. [16, Prop 5.5]), e(I)/2 = deg Y + ℓY /2; hence 0 ≤ ω + µa 2 = m0 + 1 m + 1 · deg X − (deg Y + ℓY 2 ) + m0 + 1 m + 1 = (m0 + 1)(cid:16) deg X +Pn m + 1 j=1 aj 2 − deg Y + ℓY which is (1.4). This completes the proof of the Theorem. − aj 2 nXj=1 2 +Pxj ∈Y m0 + 1 nXxj ∈WY (cid:17), aj 2 aj 2 (cid:3) HILBERT-MUMFORD CRITERION FOR NODAL CURVES 33 6. Re-construction of the moduli of weighted pointed curves In this section, we use GIT quotient of Hilbert scheme to construct the moduli of weighted pointed stable curves, first introduced and constructed by Hassett [9] using different method. Definition 6.1. A weighted pointed semi-stable curve is a weighted pointed curve (X, x, a) such that (1) ωX (a · x) is numerically non-negative; (2) the total degree 2χa(X) = deg ωX(a · x) is positive; (3) for any smooth subcurve E ⊂ X such that deg ωX (a·x)E = 0, necessarily E∩x = ∅ and E ∼= 1. We call E ⊂ X satisfying (3) exceptional components. We say (X, x, a) is weighted pointed stable if it does not contain exceptional components. n We fix integers n, g and weights a ∈ + satisfying χa(X) > 0; for a large integer k such that k · ai ∈ Z for all i, we let d = (a + 2g − 2) · k, and form (6.1) We denote by HilbP we define H be the fine moduli scheme of families of data m the Hilbert scheme of subschemes of P (t) = d · t + 1 − g ∈ Z[t], and set m = P (1). m of Hilbert polynomial P ; H =set {(X, ι, x) [ι : X → m] ∈ HilbP m, x = (x1, · · · , xn) ∈ X n}. Using that Hilbert schemes are projective, we see that H exists and is projective. We denote by (6.2) (πH, ϕ) : X −→ H × m, xi : H → X the universal family of H. We introduce a parallel space for the Chow variety. We let Chow d m be the Chow m. For any such cycle Z, we denote m)∨] its associated Chow point (cf. Section 1). We variety of degree d dimension one effective cycles in by Chow (Z) ∈ Divd,d[( define m)∨ × ( C := {(Z, x) ∈ Chow d m × ( m)n x = (x1, · · · , xn) ∈ supp(Z)n}. By Chow Theorem, C is projective. Using the Chow coordinate, we obtain an injective morphism (6.3) C ⊂−→ Divd,d[( m)∨ × ( m)∨] × ( m)n . Like before (cf. Section 1), we endow it with an ample -line bundle OC(1, a) (depending on the weights a); the line bundle is canonically linearized by the diagonal action of G := SL(m + 1) on (6.3). We let Css ⊂ C be the (open) set of semi-stable points with respect to the G linearization on OC(1, a). For any one-dimensional subscheme X ⊂ m, we denote by [X] its associated one- dimensional cycle. By sending (X, ι, x) ∈ H to ([X], x) ∈ C, we obtain the G-linear Hilbert-Chow morphism (cf. [14, Section 5.4]) Φ : H −→ C. Lemma 6.2. For fixed g, n and a satisfying χa(X) > 0 , there is an integer M depending on g, n and a so that for d ≥ M , Φ−1(Css) consists exactly of those (X, ι, x) ∈ H so that the associated data (X, ι∗O m(1), x, a) is a slope semi-stable weighted nodal curve. 34 JUN LI AND XIAOWEI WANG Proof. By an argument parallel to [16, Prop. 5.5], one proves that there is an M depending only on χa(X) such that for d ≥ M , Chow (X, x) ∈ Css implies that X is a nodal curve and the inclusion ι : X → m is given by a complete linear system. We now show that any (X, x, a) ∈ Φ−1(Css) is a weighted pointed nodal curve as defined in the beginning of the paper. We first check that the weighted points are away from the nodes, and the total weight at any point is no more than one. Let p ∈ X; we choose the 1-PS λ as in the Example 2.8; the λ-weight for Chow (X, x) is (cf.(5.5)) ω(λ) + µa(λ) = 2 deg X m + 1 − ǫp + 1 m + 1 nXj=1 aj − Xxj=p aj = 2 − ǫp + 2χa(X) m + 1 − Xxj =p aj, where ǫp = 2 if p is a node and 1 otherwise. Since Chow (X, x) is semistable, we must have 0 ≤ ω(λ) + µa(λ). Now we choose M ′ so that m + 1 = M ′ + 1 − g(X) > 2χa(X) minaj >0{ai} ; then 0 ≤ ω(λ) + µa(λ) implies that the weighted points are away from the nodes, and the total weight of marked points at p does not exceed one. Finally, the slope semi-stability of (X, x, a) follows from the necessity part of Theorem 1.5; the condition (3) of Definition 6.1 follows from the inequality (1.6) and Lemma 5.1. This proves that all (X, x, a) ∈ Φ−1(Cxx) are slope-semi-stable weighted nodal curves. The other direction is straightforward, and will be omitted. (cid:3) We define Hss = Φ−1(Css) ⊂ H. Corollary 6.3. For d ≥ M specified in Lemma 6.2, the restriction Φss := ΦHss : Hss → Css is an isomorphism Proof. First, since H is proper, Φss is surjective. For the injectivity of Φss, suppose there are (X, ι, x) and (X ′, ι′, x′) ∈ Hss such that Φ(X, ι, x) = Φ(X ′, ι′, x′) ∈ C ss, then by m. Since Φ(X, ι, x) = Φ(X ′, ι′, x′) ∈ Lemma 6.2, both X and X ′ are nodal subcurves of m; since both X and X ′ are nodal, we have X = X ′. This proves (X, ι, x) = (X ′, ι′, x′); thus Φss is bijective. Finally, Φss is an isomorphism since both Hss and Css are smooth. (cid:3) ss, the cycles [X] = [X ′] and x = x′ ⊂ C To construct the moduli of weighted pointed curves, taking the k specified before (6.1), we form K = {(X, ι, x) ∈ H X smooth weighted pointed curves, ι∗O m (1) ∼= ωX (a · x)⊗k}. It is locally closed, and is smooth. Since X in (X, ι, x) ∈ K are smooth, applying Theorem 1.5, we conclude that Φ(K) ⊂ Css, thus K ⊂ Hss. Let K ⊂ Hss be the closure of K in Hss. Because Φss is an isomorphism, and C is projective, the GIT quotient Hss//G ∼= Css//G exists and is projective. Because K is closed in Hss, the GIT quotient (6.4) q : K −→ K(cid:12)G exits and is projective. Theorem 6.4. The coarse moduli space Mg,a of stable genus g, a-weighted nodal curves constructed by Hassett is canonically isomorphic to the GIT quotient K (cid:12) G. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 35 The main technical part of the proof is to analyze the closed points of K(cid:12)G. We have the following preliminary results. For any (X, ι, x) ∈ K, since the associated weighted pointed curve (X, x, a) is semistable, we can form a new weighted pointed curve by contracting all of its exceptional components (cf. Definition 6.1). We denote the resulting curve by (6.5) (X st, xst, a), and call it the stablization of (X, x, a). Since the marked points never lie on the contracted components, the stabilization produces a weighted pointed nodal curve of the same genus. Further, the stabilization also applies to families of semistable weighted curves. Thus applying this to the restriction to K of the universal family of H, we obtain a family of weighted pointed stable curves on K. Since Mg,a is the coarse moduli space of stable weighted pointed nodal curve, we obtain a morphism (6.6) Ψ : K −→ Mg,a. As this morphism is G-equivariant with G acting trivially on Mg,a, it descends to a morphism (6.7) ψ : K(cid:12)G −→ Mg,a. We will prove Theorem 6.4 by proving that ψ is an isomorphism. 6.1. Surjectivity. Let (X, x, a) be a weighted pointed stable curve. We endow it the polarization OX (1) = ωX (a · x)⊗k together with the embedding ι : X → H 0(OX (1))∨. When X is smooth, (X, ι, x, a) lies in K; when X is singular, this may not necessarily hold. Our solution is to replace ωX (a · x)⊗k by its twist, to be defined momentarily. Given (X, x), we choose a smoothing π : X → T over a pointed curve 0 ∈ T such that X is smooth and X0 = X ×T 0 ∼= X. By shrinking T if necessary, we can extend the n-marked points of X to sections xi : T → X so that, denoting x = (x1, . . . , xn), (X , x, a) form a flat family of weighted pointed stable curves. Let X1, . . . , Xr be the irreducible components of X. The following Proposition gives the surjectivity of ψ. Proposition 6.5. Let (X, x, a) be a weighted pointed stable curve, and let (X , x, a) be the T -family constructed. Then there exist non-negative integers {bα}r α=1 so that after letting (6.8) OX (1) = ωX /T (a · s)⊗k ⊗OX OX (X bαXα), (X , OX (1), s) forms a family of slope semistable weighted pointed nodal curves. The Proposition was essentially proved by Caporaso in [2]. Since we need to use the same technique to prove the injectivity, we recall the notation used to prove this Proposition. The remainder part of this subsection essentially follows [2]. For any line bundle L on X, we denote δα(L) = deg LXα . We define the associated lattice point of L be ~δ(L) := (δ1(L), . . . , δr(L)) ∈ Z ⊕r. We call ~δ(L) the numerical class of L. We next introduce a subgroup ΓX ⊂ Z ⊕r. We let ℓα,β = ℓα,β(X) = Xα ∩ Xβ (6.9) We define ~ℓα = ~ℓα(X) = (ℓα,1(X), ℓα,2(X), · · · , ℓα,r(X)). We define ΓX ⊂ Z subgroup generated by ~ℓ1, . . . , ~ℓr. ⊕r be the if α 6= β; and ℓα,α = ℓα,α(X) = −Xα ∩ X ∁ α. 36 JUN LI AND XIAOWEI WANG Remark 6.6. Let L = ωX (a · x)⊗k. Since X is smooth, for the invertible sheaf OX (1) stated in (6.8) depending on the integers b1, . . . , br, we have ~δ(OX (1)X ) = ~δ(L) + bα~ℓα. rXα=1 This says that any two choices of OX (1) restricted to the central fiber have equivalent numerical classes modulo ΓX . This motivates the definition Definition 6.7. We define the degree class group of X be the quotient Z ⊕r/ΓX . We introduce one more notation. For any vector ~v = (v1, . . . , vr) ∈ Z ⊕r and any subcurve Y ⊂ X, mimicimg the notion of degree, we define degY ~v = XXα⊂Y vα. In particular, degX ~v =P vα. Let (X, x, a) be a weighted pointed nodal curve and L a line bundle on X of total degree d. For any subcurve Y ⊂ X, we introduce the extremes of Y (depending on d) be (6.10) M± Y := degY ωX (a · x) deg ωX (a · x) (cid:16)d + nXj=1 aj 2(cid:17) − Xxj ∈Y aj 2 ± ℓY 2 . Proposition 1.6 is reformulated as Lemma 6.8. Suppose χa(X) > 0, and let d > M , where M is defined in Corollary 5.3. Then a weighted pointed nodal curve (X, x, a) with a numerical effective line bundle L on X of deg L = d is slope semi-stable (cf. Proposition 1.6) if and only if (6.11) degY L ∈ [M− Y , M+ Y ] for any subcurve Y ⊂ X. We quote the basic properties of extremes. Y − M− Lemma 6.9. Let Y , Y1 and Y2 be surcurves of X. We have (1) M+ (2) suppose E ⊂ X is an exceptional component such that E∩Y = 1, then M± (3) suppose Y1 and Y2 have no common component, then M± Y = ℓY , and M− Y = vY ∁ + vY = d; Y ∁ + M+ Y1∪Y2 ±Y1 ∩Y2 = M± Y1 E∪Y = M± Y ; +M± . Y2 Proof. The proof is a direct check, and will be omitted. (cid:3) Let Z ⊕r ≥0 be those ~v = (vi) ∈ Z ⊕r so that vi ≥ 0. We define L replaced by degY ~v}. Bd X,x,a = {~v ∈ Z ⊕r ≥0 (cid:12)(cid:12) degX ~v = d, ~v satisfies (6.11) with degY Proposition 6.10. Let (X, x, a) be a weighted pointed semi-stable (cf. Definition 6.1) curve and d sufficiently large. Then for any ~v ∈ Z ⊕r satisfying degX ~v = d, we have (~v + ΓX ) ∩ Bd X,x,a 6= ∅. Proof. The proof is parallel to that of [2, Prop. 4.1], and will be omitted. (cid:3) Lemma 6.11. Let (X, x, a) be a weighted pointed nodal curve satisfying χa(X) > 0, then there is a constant K depends only on the genus g, χa(X) and a such that if d ≥ K, then M− Y > 0 for any connected subcurve Y ⊂ X satisfying degY ωX(a · x) > 0. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 37 Proof. Let Y ⊂ X be a connected subcurve such that degY ωX(a · x) > 0. Since the Y only involves the nodes LY = Y ∩ Y ∁, without lose of generality we can expression M− assume that X consists of two smooth irreducible components Y and Y ∁. Then ℓY ≤ g +1. Thus to prove that for large d we have M− Y > 0 whenever degY ωX (a · x) > 0, it suffices to show that (6.12) inf{degY ωX(a · x) degY ωX (a · x) > 0} is bounded below by a positive constant depending only on a and g. But this is true because (6.12) is bounded below by κ = infnXi∈I ai − l(cid:12)(cid:12)(cid:12) l ∈ Z, I ⊂ {1, · · · , n}, Xi∈I ai − l > 0o. Since a is fixed, κ is positive. Since d = k · deg ωX (a · x), for large k, which is the same as for large d, we have M− Y > 0 whenever degY ωX (a · x) > 0. This proves the Lemma. (cid:3) Proof of Proposition 6.5. By Proposition 6.10, there are {bα}'s such that for the OX (1) given in (6.8) and L = OX (1))X0 , ~δ(L) satisfies (6.11). To show that (X, L, x, a) is a polarized slope semi-stable curve, we need to show that L is ample. Since ~δ(L) satisfies (6.11), L is ample if M− > 0 for any component Xα ⊂ X; but this is precisely Lemma Xα 6.11 because (X, x, a) is weighted pointed implies that ωX (a · x) is positive. (cid:3) 6.2. Injectivity. In this subsection, we use the separatedness of K(cid:12)G to prove that ψ in (6.7) is injective. Definition 6.12. Let (X, x, a) be a weighted pointed nodal curve (cf. Definition 1.1). We say a weighted pointed nodal curve ( ¯X, ¯x, a) is a blow-up of (X, x, a) if there is a morphism π : ¯X → X that is derived by contracting some of the exceptional components of ( ¯X, ¯x, a). (Recall that exceptional component is defined in Definition (6.1).) Suppose ( ¯X, ¯x, a) is a blow-up of (X, x, a), then ( ¯X st, ¯xst, a) = (X st, xst, a) (cf. (6.5)). Since the restriction of ψ to K(cid:12)G is an isomorphism, ψ is a birational morphism. By Zariski's Main theorem and the properness of K(cid:12)G, the injectivity follows from Lemma 6.13. ψ−1(ψ(ξ)) is zero dimensional for each ξ ∈ K(cid:12)G. in (6.4). Proof. Let ξ ∈ K(cid:12)G \ (K(cid:12)G), and let ψ(ξ) = (X, x, a) ∈ Mg,a be the associated weighted pointed stable curve. We describe the set Θξ = q−1(cid:0)ψ−1(ψ(ξ))(cid:1) ⊂ K, where q is defined For any η = ( ¯X, ι, ¯x) ∈ Θξ ⊂ K, there is a smooth affine curve φ : 0 ∈ T → K so that m(1), ¯x) the pull back of the universal family of K, say π : (X , L, s) → T , contains ( ¯X, ι∗O as its central fiber and φ(T \ {0}) ⊂ K, and that the total space X is smooth. By Lemma 6.2, the central fiber ( ¯X, ¯x, a) is weighted pointed semi-stable (cf. Definition 6.1) and is a blow-up of (X, x, a) (cf. Definition 6.12). Since X is smooth, there are integers {bα} indexed by the irreducible components ¯Xα of ¯X, such that if we view ¯Xα as divisor in X , then ι∗O m(1) = ω ¯X/T (a · x)⊗k(Σ¯r α=1bα ¯Xα). Since the collection of blow-ups of X coupled with integers {bα}¯r the choices of ( ¯X, L, ¯x) are discrete. Thus {( ¯X, ι∗O Finally, any two ( ¯X, ι, ¯x) with isomorphic ( ¯X, ι∗O Θξ consists of a discrete collection of G-orbits. Hence ψ−1(ψ(ξ)) is discrete. α=1 is a discrete set, m(1), ¯x) ( ¯X, ι, ¯x) ∈ Θξ} is discrete. m(1), ¯x) lie in the same G-orbit. Thus (cid:3) 38 JUN LI AND XIAOWEI WANG We remark that this proof uses the existence of the coarse moduli space Mg,a con- structed by Hassett. 6.3. The coarse moduli space. We prove that K(cid:12)G is a coarse moduli space of weighted pointed stable curves, thus proving that ψ is an isomorphism. Proposition 6.14. Let T be any scheme and (X , x, a) be a T -family of weighted pointed stable curves. Then there is a unique morphism f : T → K (cid:12) G, canonical under base changes, such that for any closed point c ∈ T , the image ψ(f (c)) ∈ Mg,a is the closed point associated to the weighted pointed stable curve (X , x, a)c. have P = K. Γ ⊂ P × K(cid:12)G be the graph of the morphism F stated in the Lemma; we let Lemma 6.15. The composition F : P → K → K (cid:12) G extends to a unique morphism We define a subscheme eP ⊂ H: eP =(cid:8)(X, ι, x) ∈ H (X, x, a) weighted pointed stable curves, ωX (a · x)⊗k ∼= ι∗O m(1)(cid:9) A direct check shows that eP is a smooth, locally closed, and G-invariant subscheme of H. We let P ⊂ eP be the open subset of (X, ι, x) such that X are smooth. By definition, we eF : eP → K(cid:12)G. Proof. Applying deformation theory of nodal curves, we know that P is dense in eP. Let be the closure of Γ. Let p : Γ → eP be the projection. We claim that p is bijective. Indeed, Next, we claim that p is an isomorphism. Since eP is smooth, P ⊂ eP is dense, and Γ is isomorphic to P, we conclude that Γ is reduced. Then since p : Γ → eP is birational, a diffeomorphism and eP is smooth, p must be ´etale. Thus p is an isomorphim. Finally, by composing the isomorphism p−1 with the projection to the second factor of eP → K(cid:12)G, we obtain the desired extension eF of F . given ξ = (X, ι, x) ∈ ¯P, we let (X , OX (1), x) be the family given by Proposition 6.5, which shows that ξ ∈ p(Γ). This proves that p is surjective. On the other hand, repeating the proof of Lemma 6.13, we see that p is one-to-one. This proves that p is bijective. Γ ⊂ eP × K(cid:12)G (cid:3) Ta → K(cid:12)G. Proof of Proposition 6.14. We cover T by a collection of affine open {Ta}a∈A. Let πa : Xa → Ta with sections xa,i : Ta → Xa be the restriction of xi to Ta of the family on T . By fixing a trivialization πa∗ωXa/Ta(a · xa)⊗k ∼= O⊕m+1 , we obtain morphisms fa : Ta → eP. Composed with the morphism eF constructed in the previous Lemma, we obtain eF ◦ fa : Since the choice of picking the trivializations does not alter the morphism eF ◦ fa, this collection {eF ◦ fa}a∈A patches to a morphism T → K (cid:12)G. This proves the first part of Finally, that ψ(f (c)) is the point associated to the weighted pointed curve (X , x, a)c (cid:3) Proposition 6.14. follows from the construction. Ta Proof of Theorem 6.4. It follows from Proposition 6.5, 6.14, and Lemma 6.13. (cid:3) For completeness, we describe without proof the geometry of poly-stable3 points in Css. 3Recall a point ξ ∈ C ss is poly-stable if the G-orbit G · ξ is closed in C ss. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 39 Definition 6.16 ([2] when x = ∅). We call (X, OX (1), x, a) extremal if for any proper subcurve Y ⊂ X satisfying ~δY (OX (1)) = M− Y (cf. (6.11)), we have LY = Y ∩ Y ∁ ⊂ EX , where EX is the union of degree one rational curves in X ⊂ W . Let (X, OX (1), x, a) be a slope semi-stable weighted pointed nodal curve such that deg X > M with M given in Theorem 1.5. Then the Chow point of (X, OX (1), x) is poly-stable with respect to the polarization OΞ(1, a) if and only if it is extremal. (This in case x = ∅ was proves by Caporaso in [2].) 7. K-stability of nodal curve In this section, we give another application of Theorem 1.5, which was motivated by a question of Yuji Odaka on studying the K-stability of a polarized nodal curve. Theorem 7.1. A polarized connected nodal curve (X, OX(1)) is K-stable if and only if OX(1) is numerically equivalent to a multiple of ωX . We comment that Odaka has proved the K-stability for nodal curve X polarized by OX(1) = ω⊗k X for some k ∈ [18]. His proof uses birational geometry and a weight formula proved by himself and by the second named author independently. He also informed us that he was able to generalize his method to prove the above theorem. We recall the notion of K-stability of polarized varieties. ∗-equivariant flat projective morphism π : X → A 1, where C Definition 7.2 ([19, Sect. 3]). A test configuration for a polarized scheme (X, OX (1)) 1 via consists of a C ∗-linearized relative very ample line bundle L of π, such that for the usual action, and a C 1, (Xt, Lt) ∼= (X, OX(1)). (Here Lt = LXt .) We call such test configuration any t 6= 0 ∈ A (X , L) a product test configuration if X ∼= X × A 1; we call it a trivial test configuration if in addition to that it is a product test configuration, the line bundle L is a pull back from X and the C ∗-action is the product action that acts trivially on X. ∗ acts on A For notational simplicity, from now on we restrict ourselves to when (X, OX(1)) is a polarized nodal curve. Given a text configuration (X , L), we w(l) be the weight of the ∗-action on π∗L⊗l0; w(l) = a2l2+a1l+a0 is a degree 2 (= dim X +1) polynomial induced C in l. We then form the quotient w(l) l · χ(OX (l)) = e0 + e−1l−1 + . . . . Definition 7.3. We define the Donaldson-Futaki invariant of a test configuration (X , L) of (X, OX (1)) be DF(X , L) = e−1 = − an+1bn−1 − an · bn b2 n ; the polarized nodal curve (X, OX (1)) is K-stable if DF(X , L) < 0 for any nontrivial test configuration (X , L) of (X, OX(1)). For (X, OX (1)), and letting W ∨ = H 0(OX (l)) with X ⊂ W the tautological embed- 1 via ∗ action produces a test configuration of (X, OX(1)); we denote such test ding, then given any 1-PS subgroup λ of Aut W , the C the diagonal C configuration by (Xλ, Lλ). ∗-orbit of X in W × A Conversely, given (X, OX (1)), any test configuration of (X, OX (1)) can be derived from [19, Prop. 3.7]). Thus to prove the K-stability of (X, OX (1)), it a 1-PS of Aut W (cf. 40 JUN LI AND XIAOWEI WANG suffices to show that the Donaldson-Futaki invariant DF(Xλ, Lλ) < 0 for all 1-PS λ of Aut W . Our starting point is to relate DF(Xλ, Lλ) with the Chow weights of (X, OX(l)). We pick a λ-diagonalizing basis s = {s0, · · · , sm} of W ∨; namely, under its dual bases the action λ is given by the following (1-PS of GL(W ∨)): (7.1) λ(t) := diag[tρ0 , · · · , tρm ], ρ0 ≥ ρ1 ≥ · · · ≥ ρm = 0, ρi ∈ Z. Here by shifting all ρi by a common integer a, we can choose ρm = 0. Our next step is to construct a 1-PS of W ∨ l = H 0(OX (l)) that is most close to λ. First, since (X, OX (1)) is K-stable is equivalent to that of (X, OX (k)), k > 0. THus without lose of generality, we can assume OX(1) is sufficiently ample so that 1 · · · sim m ; then sI has (7.2) For convenience, for multi-indices I = (i0, · · · , im), we denote sI = si0 φl : SlW ∨ −→ W ∨ l = H 0(OX (l)). total degree I = P ij. For the weights ρ = (ρ0, · · · , ρm), we denote ρ(I) = P ρj · ij, which is the weight of sI under the induced λ action on SlW ∨. We let Sl be the set of monomials in SlW ∨. We order Sl as follows: sI ≻ sI ′ either ρ(I) < ρ(I ′), or when ρ(I) = ρ(I ′) and there is a 0 ≤ j0 ≤ m such that ij = i′ all j > j0 and ij0 > i′ j0 Further, sI ≻ sI ′ versa. when j for least) element in Sl. for any non-trivial monomial sJ , and vice if and only if sJ · sI ≻ sJ · sI ′ 0) is the largest (resp. m (resp. sl . Thus, sl We pick a basis of W ∨ l , which will be a diagonalizing basis of the λl we will construct momentarily. Let ml + 1 = dim W ∨ m. Suppose for an integer 0 ≤ k < ml, we have picked sl,k+1, · · · , sl,ml and their weights l,j, we let sl,k = sIk be the largest element in m, with weight l,ml = ρl l . We set sl,ml = sl (cid:8)sI ∈ Sl φl(sI ) 6∈ φl(Θl,k+1)(cid:9), where Θl,k+1 = φl(cid:0)C{sl,k+1, · · · , sl,ml}(cid:1), and C{·} is the C-linear span of elements in {·}. We let ρ(Ik), which is the weight of sIk in SlW ∨ under λ⊗l. We let sl,k = φl(sIk ). Then sl,0, · · · , sl,ml form a basis W ∨ l . By setting (7.3) λl(σ) · sl,k = σl,k sl,k we obtain a 1-PS of Aut Wl of diagonalizing basis {sl,0, · · · , sl,ml}. Following the discussion in Section 2, we define, (for q ∈ X and sl,k the lift of sl,k to the normalization X of X, ) (7.4) and (7.5) α(λl) = min{i sl,i+1 Xα = 0}, (λl, q) = max{i v(sl,i, q) 6= ∞}, Λα(λl) = {p ∈ Xα sl,α(λl)(p) = 0}, Λ(λl) = ∪r α=1 Λα(λl). (Here r is the number of irreducible components of X.) Lemma 7.4. For the 1-PS λl constructed, we have Λ(λl) = Λ(λ). Further, for each q ∈ Λ(λ), and for w(I, q) defined in (2.9), we have v(sl,(λ,q), q) = w(Il, q) = l · w(I, q) = l · v(s(q), q), and hence e(I(λl)) = n.l.c. χ(OX×A 1(k)/I(λ)kl). l be the staircase 1-PS obtained from λl by applying Proposition 3.6, then Let λ′ (1) the support Λ(λl) is the same as Λ(λ′ (2) for each q ∈ Λ(λl), ∆q(λl) = l · ∆q(λ) ⊂ ∆q(λ′ l). l); for each q ∈ Λ(λl), w(I(λl), q) = w(I(λ′ l), q); HILBERT-MUMFORD CRITERION FOR NODAL CURVES 41 Proof. Because of our choice of λl, we have the middle identity (the first the the third is by the definition) (s0, · · · , sm) = I(λl) = I(λ)l = (tρ0 s0, · · · , tρm sm)l ⊂ OX×A 1(l). This prove the first part of the Lemma. The second part, follows from the construction of staircase in Proposition 3.6, and from Lemma 2.5. (cid:3) We have the following useful Lemma, relating DF(Xλ, Lλ) to the weights ω(l) of λ. Lemma 7.5 ([5, Sect. 2.3], [19, Thm. 3.9]). Let X ⊂ W ∨ and λ a 1-PS be as before. Then lim l→∞ l−1 · ω(λl) = −b−1 1 · DF(Xλ, Lλ) < ∞ . Thus to prove DF(Xλ, Lλ) < 0, it suffices to show that (7.6) l−1 · ω(λl) > 0. lim l→∞ Let λ′ l be the staircase constructed from λl using Proposition 3.6, of the same weights l; namely, l,i. We let l,i be the shifted weights according to the rule (5.9) applied to λ′ l,i = minα{i − β (λ′ l) i ∈ β(λ′ l)}. Proof of Theorem 7.1. Suppose X is a stable (nodal) curve and OX (1) is numerically proportional to ωX , then (X, OX(1)) is slope stable. We will show in this case that for any 1-PS λ ⊂ SL(W ) we have DF(Xλ, Lλ) < 0. 0 < i0 < m such that i0 = 0 and Tk≥i0 The first case to study is when e(I(λ)) = 0. By Section 2 we know that there is a {sk = 0} = ∅. Stoppa proved that in this case either the test configuration (Xλ, L) induced by λ is trivial (cf. Definition 7.2) or DF(λ) < 0 [21, page 1405-1406]. This settles this case. The other case is when e(I(λ)) > 0. Applying Theorem 5.8 (since (X, OX (1)) is slope stable), and applying Proposition 3.6, we can find an ǫ > 0 so that (7.7) l−1 · ω(λl) ≥ l−1 · ω(λ′ l) ≥ 1 deg X + l−1(1 − gX ) · ǫ l2 · mlXi=0 l,i. Thus DF(Xλ, Lλ) < 0 follows from Lemma 7.6, which we will prove shortly. This proves the if part, once we prove Lemma 7.6. For the other direction, suppose (X, OX (1)) is K-stable, we show that (X, OX (1)) is slope stable. Suppose (X, OX(1)) is not slope stable, then there is a subcurve Y ⊂ X destabilizing the polarized curve (X, OX (1)), namely, (7.8) degY ωX deg ωX · deg X − deg Y − ℓY 2 ≥ 0. Let H 0(OX (1)Y )∨ ⊂ W = H 0(OX (1))∨, which is the linear subspace spanned by Y ; let m0 + 1 = dim H 0(OX (1)Y ). We choose a two-weight 1-PS λ as in the proof of Theorem 1.5 (at the end of Section 5) so that λ acts of weight 1 on H 0(OX (1)Y )∨ ⊂ H 0(OX (1))∨ and acts of weight 0 on its linear complement; we form its associated test configuration (Xλ, Lλ). We now evaluate DF(Xλ, Lλ) = lim l→∞ − ω(λl) l = lim l→∞ 1 l ·(cid:16) 2l deg XPml i=0 l,i l deg X + 1 − g − e(I(λl))(cid:17). 42 JUN LI AND XIAOWEI WANG First, the central fiber X0 is the union Y ∪ E ∪ Y ∁, where E consists of ℓY number λ X0) has a of lines that inserted in the linking nodes LY of X; the total space H 0(L⊗l decomposition H 0(L⊗l λ X0) ∼= H 0(OX (l)Y ) ⊕ H 0(OE(l)(−Y ∩ E)) ⊕ H 0(OX (l)Y ∁ (−E ∩ Y ∁)); elements in H 0(OX (l)Y ) have weights l; elements in H 0(OE(l)(−Y ∩ E)) have weights l − 1, · · · , 0 for each copy in E; elements in H 0(OX (l)Y ∁ (−E ∩ Y ∁)) have weights 0. Thus X l,i = h0(OX (l)Y ) · l + ℓY · i=0 l,i = e(I(λ)) · l2 Using thatPml (7.9) DF(Xλ, Lλ) = l(l − 1) 2 = (deg Y + ℓY 2 ) · l2 + (1 − gY − ℓY 2 ) · l. 2 + lower order term, and simplifying, we obtain g − 1 · deg X − deg Y − deg X(cid:16) degY ωX deg ωX ℓY 2 (cid:17). Since Y ⊂ X is destabilizing, by (7.8) we have DF(Xλ, Lλ) > 0, violating that (X, OX (1)) is K-stable. This proves that (X, OX (1)) is slope stable. On the other hand, since K-stability of (X, OX (l)) is independent of l > 0, (X, OX (1)) is K-stable implies that (X, OX (l)) is K-stable for l > 0, thus (X, OX (l)) is slope stable. Once we know that (X, OX (l)) is slope stable for large l, an easy argument shows that OX (l) satisfies (1.6) for all large l, which is possible only when OX (1) is numerically proportional to ωX . This proves the Theorem. (cid:3) Lemma 7.6. Let the notation be as stated. Suppose further that e(I(λ)) > 0. Then 4 lim l=lk→∞ 1 l2 · mlXi=0 l,i > 0. Proof. We comment that in case X is irreducible, the positivity is immediate. Indeed, applying [16, Prop. 2.11 ], we have (7.10) l,i = e(I(λ)) · l2 2 + a1 · l + a2, ai depending only on λ. mlXi=0 Suppose X is irreducible, we have l,i = l,i. Therefore, lim l=lk→∞ 1 l2 · l,i = lim l=lk→∞ 1 l2 · l,i = e(I(λ)) 2 > 0. mlXi=0 mlXi=0 We now prove the general case. We claim that there is a 1 ≤ β ≤ r and q ∈ Xβ so that (7.11) ∆q(λ) − ρβ(λ) · w(I(λ), q) > 0. Suppose for any q ∈ Xβ the inequality (7.11) does not hold. Since the ≥ always hold, we will have that i = α for every i ∈ α. Since e(I(λ)) > 0, we must have an α > 1 such that α > 0. Since X is connected, we can find a pair α 6= β so that Xα ∩ Xβ 6= ∅, and ρα(λ) > ρβ (λ) = 0. We next let q ∈ Xβ be a lift of a node in Xα ∩ Xβ. We show that the pair (β, q) satisfies the inequality (7.11). Let π : X → X be the projection. Since π(q) ∈ Xα, we have sj(q) = 0 for all j > α(λ); since ρα(λ) > ρβ (λ) = 0, we have i0(q) ≤ α(λ). Thus 4Here by l = lk → ∞ we mean that by passing to a subsequence we assume that the limit limlk→∞ does exist, and is finite. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 43 ρi0(q) ≥ ρα(λ) > 0, and then ∆q(λ) is two dimensional. Since ρβ (λ) = 0, this contradicts to initial assumption that (7.11) never holds. This proves the claim. Let (β, q) be a pair satisfying (7.11). We next establish the following two inequalities (7.12) and (7.13) lim l=lk→∞ l,i ≥ Xi l−2 ·(cid:0)∆q(λ′ mlXi=0 1 2(cid:0)∆q(λ′ l) − l,β (λ′ l) · w(I(λ′ l), q)(cid:1) l) − l,β(λ′ l) · w(I(λ′ l,i ≥ Xi∈α(λ′ l) l), q)(cid:1) ≥ ∆q(λ) − l,β(λ) · w(I(λ), q). l,i ≥ Xi∈ pri q (λ′ l) l,i, q (λ′ pri l) is the set of primary indices for We prove the inequality (7.12). Following the notation introduced in Section 4, we have where β(λ′ q ∈ Xβ, both with respect to the staircase λ′ l. l) is the set of indices for Xβ, and By Proposition 3.11 and 3.13, we know that for i0(q) 6= i ∈ q (λ′ pri l), we have l,i = l,i − l,β (λ′ l). (Note l,i0(q) = l,i0(q) − l, By the proof of Lemma 4.2, we have α′ (λ′ l) possibly for some α′ 6= β.) l,i − l,i0(p) ≥ ∆pri p (λ′ l) ∩ ([1, wpri(p, λ′ l)] × R) − l,β (λ′ l) · (wpri(p, λ′ l) − 1). Xi∈ pri q (λ′ l) Following (4.9), we continue to denote ¯p(λ′ w(E¯p(λ′ l, the effects to the shape of ∆q(λ′ thus for large l we have l) = l), p). By the boundness result from Corollary 3.14, for sufficiently large l) is marginal, l) from the secondary indices q(λ′ l)} and wpri(p, λ′ l) = max{i ∈ l)+1(λ′ pri q (λ′ p (λ′ pri l) \ q (λ′ l)∩([1, wpri(q, λ′ (cid:12)(cid:12)∆pri Combined, and adding i0(p) > 0, we obtain l)]×R)(cid:12)(cid:12)−l,β(λ′ Xi∈ l)·(cid:0)wpri(q, λ′ 2(cid:16)∆q(λ′ l,i ≥ pri q (λ′ l) 1 l)−1(cid:1) ≥ l) − l,β (λ′ l) · w(I(λ′ 1 2(cid:16)∆q(λ′ l), q)(cid:17). l)−l,β(λ′ l)·w(I(λ′ l), q)(cid:17). This proves (7.12). Before we move to (7.13), we claim that (7.14) Aβ := lim l=lk→∞ l,β(λ′ l) − l,β (λl) l = 0. Suppose not, say Aβ > 0, (it is non-negative,) then for l = lk large, l) − l,β (λ′ by examining the geometry of ∆q(λl) ⊂ ∆q(λ′ l,β(λ′ l) ≥ 1 2 · l · Aβ; l), we obtain ∆q(λ′ l) − ∆q(λl) ≥ Aβ 2 · l · w(I(λ), q) h∆q(λ) − l,β (λ) · l := C · l2 > 0, where h∆q(λ) is the height of ∆q(λ). This implies l−1 · ω(λl) = l−1 · ω(λ′ where we have used Theorem 1.5 to deduce ω(λ′ l) + l−1 ·(cid:0)e(I(λ′ l) − e(I(λl)(cid:1) ≥ l−1 ·(cid:0)e(I(λ′ l) ≥ 0. l) − e(I(λl)(cid:1), 44 JUN LI AND XIAOWEI WANG By Corollary 2.7 and our construction of staircase using Proposition 3.6, we deduce l−1 · (e(I(λ′ l) − e(I(λl)) ≥ l−1 · (∆q(λ′ l) − ∆q(λl)) > C · l . This is impossible since Lemma 7.5 implies that the left-hand-side remains bounded as l = lk → ∞. So we must have Aβ = 0. This proves the claim. We prove inequality (7.13). Because Aβ = 0, ∆q(λ′ l) ≥ ∆q(λl), and by Lemma 7.4, we obtain l) · w(I(λ′ l) − l,β (λ′ l) − l,β (λl) · w(I(λl), q) + l,β (λl) · w(I(λl), q) − l,β (λ′ ∆q(λ′ = ∆q(λ′ ≥ ∆q(λl) − l,β (λl) · w(I(λl), q) + (l,β (λl) − l,β (λ′ l), q) l)) · w(I(λl), q) l) · w(I(λ′ l), q) = l2· (∆q(λ) − l,β (λ) ·w(I(λ), q)) + l2 · l,β (λl) − l,β (λ′ l) l · w(I(λ), q). Taking limit as l = lk → ∞, and using Aβ = 0, we obtain (7.13). Finally, by (7.10), and that 0 ≤ l,i ≤ l,i, we conclude that the limit in the Lemma is (cid:3) finite; thus the limit is finite and positive; this proves the Lemma. Remark 7.7. Following [16, Sect. 3], (or the recent work of Odaka [18],) we know that a K-stable polarized curve has at worst nodal singularity. Thus results like Theorem 7.1 show that K-stability compactifies the moduli of smooth curves (of g ≥ 2). As K-stability is an analytic version of GIT stable via a CM-line bundle defined by Paul and Tian [20], which is a multiple of λ⊗12 ⊗ δ−1 for moduli of curves (cf. [16, Thm. 5.10]), Theorem 7.1 can be viewed as comparing compactifications via two versions of GIT stability, one via finite dimensional embedding and one via analysis. Generalizing this to high dimensional canonically polarized varieties remains a challenge. Lately, Yuji Odaka (cf. [18]) has made important progress along this direction. References [1] Arbarello, E.; Cornalba, M.; Griffiths, P. and Harris, J. Geometry of Algebraic Curves. Volume I, Springer-Verlag, New York, 1984. [2] Caporaso, L. A compactification of the universal Picard variety over the moduli space of stable curves. J. Amer. Math. Soc. 7 (1994), no. 3, 589 -- 660. [3] Caporaso, L.; Casagrande, C. and Cornalba, M. M oduli of roots of line bundles on curves. Trans. Amer. Math. Soc. 359 (2007), no. 8, 3733 -- 3768 [4] Cornalba, M. M oduli of curves and theta-characteristics. Lectures on Riemann surfaces (Trieste, 1987), 560 -- 589, World Sci. Publ., Teaneck, NJ, 1989. [5] Donaldson, S. K. Scalar curvature and stability of toric varieties. J. Differential Geom. 62 (2002), no. 2, 289 -- 349. [6] Eisenbud, D. Commutative algebra. With a view toward algebraic geometry. Graduate Texts in Math- ematics, 150. Springer-Verlag, New York, 1995. [7] Gieseker, D. Lectures on moduli of curves. Tata Institute of Fundamental Research Lectures on Mathematics and Physics, 69. Springer-Verlag, Berlin-New York, 1982. [8] Gieseker, D. and Morrison, I. Hilbert stability of rank-two bundles on curves. J. Differential Geom. 19 (1984), no. 1, 1 -- 29. [9] Hassett, B. Moduli spaces of weighted pointed stable curves. Adv. Math. 173 (2003), no. 2, 316 -- 352. [10] Harris, J. and Morrison, I. Moduli of Curves [11] Kantor, J. and Khovanskii, A. Une application du th´eor`eme de Riemann-Roch combinatoire au d. C. R. Acad. Sci. Paris S´er. I Math. 317 (1993), no. polynome d'Ehrhart des polytopes entiers de R 5, 501 -- 507. [12] Kempf, G. and Ness, L. The length of vectors in representation spaces. Algebraic geometry (Proc. Summer Meeting, Univ. Copenhagen, Copenhagen, 1978), pp. 233243, Lecture Notes in Math., 732, Springer, Berlin, 1979. HILBERT-MUMFORD CRITERION FOR NODAL CURVES 45 [13] Lazarsfeld, R. Positivity in algebraic geometry. II. Positivity for vector bundles, and multiplier ideals. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Springer-Verlag, Berlin, 2004. [14] Mumford, D., Fogarty, J. and Kirwan, F. Geometric invariant theory. Third edition. Springer-Verlag, Berlin, 1994. [15] Morrison, I., GIT Constructions of Moduli Spaces of Stable Curves and Maps arXiv:0810.2340 [16] Mumford, D. Stability of projective varieties. Enseignement Math. (2) 23 (1977), no. 1-2, 39 -- 110. [17] Ness,L. Stratification of null cone via the moment map Amer. J. Math. Vol. 106, No. 6 (1984),1281- 1329. [18] Odaka, Y. The Calabi conjecture and K-stability. arXiv:1010.3597 [19] Ross, J. and Thomas, R. A study of the Hilbert-Mumford criterion for the stability of projective varieties. J. Algebraic Geom. 16 (2007), no. 2, 201255. [20] Paul, S. and Tian, G. CM stability and the generalized Futaki invariant, arXiv: math.AG/0605278. [21] Stoppa, J. K -stability of constant scalar curvature Khler manifolds. Adv. Math.221 (2009), no. 4, 1397-1408. [22] Swinarski, D. GIT stability of weighted pointed curves. arXiv:0801.1288. Stanford University, Stanford, CA94305, USA E-mail address: [email protected] Rutgers University, Newark, NJ07102, USA E-mail address: [email protected]
1908.05605	1	1908	2019-08-13T19:14:25	Rigidity of Bott-Samelson-Demazure-Hansen variety for $PSO(2n+1, \mathbb{C})$	[ "math.AG", "math.CO", "math.RT" ]	Let $G=PSO(2n+1, \mathbb{C}) (n \ge 3)$ and $B$ be the Borel subgroup of $G$ containing maximal torus $T$ of $G.$ Let $w$ be an element of Weyl group $W$ and $X(w)$ be the Schubert variety in the flag variety $G/B$ corresponding to $w.$ Let $Z(w, \underline{i})$ be the Bott-Samelson-Demazure-Hansen variety (the desingularization of $X(w)$) corresponding to a reduced expression $\underline{i}$ of $w.$ In this article, we study the cohomology modules of the tangent bundle on $Z(w_{0}, \underline{i}),$ where $w_{0}$ is the longest element of the Weyl group $W.$ We describe all the reduced expressions of $w_{0}$ in terms of a Coxeter element such that all the higher cohomology modules of the tangent bundle on $Z(w_{0}, \underline{i})$ vanish (see Theorem \ref{theorem 8.1}).	math.AG	math	RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Abstract. Let G = P SO(2n+1, C)(n ≥ 3) and B be the Borel subgroup of G containing maximal torus T of G. Let w be an element of Weyl group W and X(w) be the Schubert variety in the flag variety G/B corresponding to w. Let Z(w, i) be the Bott-Samelson- Demazure-Hansen variety (the desingularization of X(w)) corresponding to a reduced expression i of w. In this article, we study the cohomology modules of the tangent bundle on Z(w0, i), where w0 is the longest element of the Weyl group W. We describe all the reduced expres- sions of w0 in terms of a Coxeter element such that all the higher cohomology modules of the tangent bundle on Z(w0, i) vanish (see Theorem 7.1). 1. introduction Let G be a simple algebraic group of adjoint type over the field C of complex numbers. We fix a maximal torus T of G and let W = NG(T )/T denote the Weyl group of G with respect to T . We denote by R the set of roots of G with respect to T and by R+ a set of positive roots. Let B+ be the Borel subgroup of G containing T with respect to R+. Let w0 denote the longest element of the Weyl group W . Let B be the Borel subgroup of G opposite to B+ determined by T , i.e. B = nw0B+n−1 w0 , where nw0 is a representative of w0 in NG(T ). Note that the roots of B is the set R− := −R+ of negative roots. We use the notation β < 0 for β ∈ R−. Let S = {α1, . . . , αn} denote the set of all simple roots in R+, where n is the rank of G. The simple reflection in the Weyl group corresponding to a simple root α is denoted by sα. For simplicity of notation, the simple reflection corresponding to a simple root αi is denoted by si. For w ∈ W , let X(w) := BwB/B denote the Schubert variety in G/B corresponding to w. Given a reduced expression w = si1si2 · · · sir of w, with the corresponding tuple i := (i1, . . . , ir), we denote by Z(w, i) the desingularization of the Schubert variety X(w), which is now known as Bott-Samelson-Demazure-Hansen variety. This was first intro- duced by Bott and Samelson in a differential geometric and topological context (see [2]). Demazure in [6] and Hansen in [8] independently adapted the construction in algebro- geometric situation, which explains the reason for the name. For the sake of simplicity, we will denote any Bott-Samelson-Demazure-Hansen variety by a BSDH-variety. The construction of the BSDH-variety Z(w, i) depends on the choice of the reduced expression i of w. In [5], the automorphism groups of these varieties were studied. There, the following vanishing results of the tangent bundle TZ(w,i) on Z(w, i) were proved (see [5, Section 3]): 1 2 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA (1) H j(Z(w, i), TZ(w,i)) = 0 for all j ≥ 2. (2) If G is simply laced, then H j(Z(w, i), TZ(w,i)) = 0 for all j ≥ 1. As a consequence, it follows that the BSDH-varieties are rigid for simply laced groups and their deformations are unobstructed in general (see [5, Section 3] ). The above vanishing result is independent of the choice of the reduced expression i of w. While computing the first cohomology module H 1(Z(w, i), TZ(w,i)) for non simply laced group, we observed that this cohomology module very much depend on the choice of a reduced expression i of w. It is a natural question to ask that for which reduced expressions i of w, the cohomology module H 1(Z(w, i), TZ(w,i)) does vanish ? In [4], a partial answer is given to this question for w = w0 when G = P Sp(2n, C). In this article, we give a partial answer to this question for w = w0 when G = P SO(2n + 1, C). Recall that a Coxeter element is an element of the Weyl group having a reduced ex- pression of the form si1si2 · · · sin such that ij 6= il whenever j 6= l (see [11, p.56, Section 4.4]). Note that for any Coxeter element c, there is a decreasing sequence of integers n ≥ a1 > a2 > . . . > ak = 1 such that c = [i, j] := sisi+1 · · · sj for i ≤ j. In this paper we prove the following theorem. kQ j=1 [aj, aj−1 − 1], where a0 := n + 1, Theorem 1.1. Let G = P SO(2n + 1, C) (n ≥ 3) and let c ∈ W be a Coxeter ele- ment. Let i = (i1, i2, . . . , in) be a sequence corresponding to a reduced expression of w0, where ir (1 ≤ r ≤ n) is a sequence of reduced expressions of c (see Lemma 2.8). Then, H j(Z(w0, i), T(w0,i)) = 0 for all j ≥ 1 if and only if c = and a2 6= n − 1. kQ j=1 [aj, aj−1 − 1], where a0 := n + 1 By the above vanishing results, we conclude that if G = P SO(2n + 1, C) (n ≥ 3) and i = (i1, i2, . . . , in) is a reduced expression of w0 as above, then the BSDH-variety Z(w0, i) is rigid. The main differences in the proof of main theorem between the case of type Cn and the case of type Bn are as follows: When G is of type Cn, there is only one long simple root, namely αn. Therefore, by [14, Corollary 5.6, p.778], we have H 1(w, αj) = 0 for any w ∈ W and for any j 6= n. But, if G is of type Bn, α1, α2, . . . , αn−1 are all long simple roots. So, we can not apply [14, Corollary 5.6, p.778]. Hence, we need to study the cohomology modules H 1(w, αj) ( w ∈ W , j 6= n − 1 ). While studying these modules, we prove that H 1(w, αj) = 0 for any w ∈ W and for any j 6= n − 1 (see Lemma 3.3). Also, in this article, we need to prove an additional statement namely, (s1s2 · · · sn)r−1s1s2 · · · sn−1(αj) < 0 for 2 ≤ r ≤ n and n + 1 − r ≤ j ≤ n − 1 (see Lemma 5.9). The organization of the paper is as follows: In Section 2, we recall some preliminaries on BSDH-varieties. We deal with the special case G = P SO(2n + 1, C) (n ≥ 3) in the later sections 3, 4, 5, 6 and 7. In Section 3, we prove H 1(w, αj) = 0 for j 6= n − 1 and w ∈ W. In Section 4 (respectively, Section 5) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 3 we compute the weight spaces of H 0 (respectively, H 1) of the relative tangent bundle of BSDH-varieties associated to some elements of the Weyl group. In Section 6, we prove some results on cohomology modules of the tangent bundle of BSDH varieties. In Section 7, we prove the main result using the results from the previous sections. 2. preliminaries In this section, we set up some notation and preliminaries. We refer to [3], [9], [10], [13] for preliminaries in algebraic groups and Lie algebras. Let G be a simple algebraic group of adjoint type over C and T be a maximal torus of G. Let W = NG(T )/T denote the Weyl group of G with respect to T and we denote the set of roots of G with respect to T by R. Let B+ be a Borel subgroup of G containing T . Let B be the Borel subgroup of G opposite to B+ determined by T . That is, B = n0B+n−1 0 , where n0 is a representative in NG(T ) of the longest element w0 of W . Let R+ ⊂ R be the set of positive roots of G with respect to the Borel subgroup B+. Note that the set of roots of B is equal to the set R− := −R+ of negative roots. Let S = {α1, . . . , αn} denote the set of simple roots in R+. For β ∈ R+, we also use the notation β > 0. The simple reflection in W corresponding to αi is denoted by sαi. Let g be the Lie algebra of G. Let h ⊂ g be the Lie algebra of T and b ⊂ g be the Lie algebra of B. Let X(T ) denote the group of all characters of T . We have X(T ) ⊗ R = HomR(hR, R), the dual of the real form of h. The positive definite W -invariant form on HomR(hR, R) induced by the Killing form of g is denoted by ( , ). We use the notation h , i to denote hµ, αi = 2(µ,α) (α,α) , for every µ ∈ X(T ) ⊗ R and α ∈ R. We denote by X(T )+ the set of dominant characters of T with respect to B+. Let ρ denote the half sum of all positive roots of G with respect to T and B+. For any simple root α, we denote the fundamental weight corresponding to α by ωα. For 1 ≤ i ≤ n, let h(αi) ∈ h be the fundamental coweight corresponding to αi. That is ; αi(h(αj)) = δij, where δij is Kronecker delta. For a simple root α ∈ S, let nα ∈ NG(T ) be a representative of sα. We denote the minimal parabolic subgroup of G containing B and nα by Pα. We recall that the BSDH- variety corresponds to a reduced expression i of w = si1si2 · · · sir defined by Z(w, i) = Pα1 × Pα2 × · · · × Pαir B × B × · · · × B , where the action of B × B × · · · × B on Pαi1 (p1, p2, . . . , pr)(b1, b2 . . . , br) = (p1 · b1, b−1 1 × Pαi2 · p2 · b2, . . . , b−1 × · · · × Pαir is given by r−1 · pr · br), pj ∈ Pαij , bj ∈ B and i = (i1, i2, . . . , ir) (see [6, Definition 1, p.73], [3, Definition 2.2.1, p.64]). We note that for each reduced expression i of w, Z(w, i) is a smooth projective variety. We denote by φw, the natural birational surjective morphism from Z(w, i) to X(w). Let fr : Z(w, i) −→ Z(wsir, i′) denote the map induced by the projection Pαi1 × · · · × Pαir −→ Pαi1 × Pαi2 × Pαi2 × · · · × Pαir−1 , where i′ = (i1, i2, . . . , ir−1). Then we observe that fr is a Pαir /B ≃ P1-fibration. For a B-module V, let L(w, V ) denote the restriction of the associated homogeneous vector bundle on G/B to X(w). By abuse of notation, we denote the pull back of L(w, V ) 4 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA via φw to Z(w, i) also by L(w, V ), when there is no confusion. Since for any B-module V the vector bundle L(w, V ) on Z(w, i) is the pull back of the homogeneous vector bundle from X(w), we conclude that the cohomology modules H j(Z(w, i), L(w, V )) ≃ H j(X(w), L(w, V )) for all j ≥ 0 (see [3, Theorem 3.3.4(b)]), are independent of choice of reduced expression i. Hence we denote H j(Z(w, i), L(w, V )) by H j(w, V ). In particular, if λ is character of B, then we denote the cohomology modules H j(Z(w, i), Lλ) by H j(w, λ). We recall the following short exact sequence of B-modules from [5], we call it SES : (1) H 0(w, V ) ≃ H 0(sγ, H 0(sγw, V )). (2) 0 → H 1(sγ, H 0(sγw, V )) → H 1(w, V ) → H 0(sγ, H 1(sγw, V )) → 0. Let α be a simple root and λ ∈ X(T ) be such that hλ, αi ≥ 0. Let Cλ denote one dimensional B-module associated to λ. Here, we recall the following result due to Demazure [7, p.271] on short exact sequence of B-modules: Lemma 2.1. Let α be a simple root and λ ∈ X(T ) be such that hλ, αi ≥ 0. Let ev : H 0(sα, λ) −→ Cλ be the evaluation map. Then we have (1) If hλ, αi = 0, then H 0(sα, λ) ≃ Cλ. (2) If hλ, αi ≥ 1, then Csα(λ) ֒→ H 0(sα, λ), and there is a short exact sequence of B-modules: 0 → H 0(sα, λ − α) −→ H 0(sα, λ)/Csα(λ) −→ Cλ → 0. Further more, H 0(sα, λ − α) = 0 when hλ, αi = 1. (3) Let n = hλ, αi. As a B-module, H 0(sα, λ) has a composition series 0 ⊆ Vn ⊆ Vn−1 ⊆ · · · ⊆ V0 = H 0(sα, λ) such that Vi/Vi+1 ≃ Cλ−iα for i = 0, 1, . . . , n − 1 and Vn = Csα(λ). We define the dot action by w · λ = w(λ + ρ) − ρ, where ρ is the half sum of positive roots. As a consequence of exact sequences of Lemma 2.1, we can prove the following. Let w ∈ W , α be a simple root, and set v = wsα. Lemma 2.2. If l(w) = l(v) + 1, then we have (1) If hλ, αi ≥ 0, then H j(w, λ) = H j(v, H 0(sα, λ)) for all j ≥ 0. (2) If hλ, αi ≥ 0, then H j(w, λ) = H j+1(w, sα · λ) for all j ≥ 0. (3) If hλ, αi ≤ −2, then H j+1(w, λ) = H j(w, sα · λ) for all j ≥ 0. (4) If hλ, αi = −1, then H j(w, λ) vanishes for every j ≥ 0. The following consequence of Lemma 2.2 will be used to compute the cohomology mod- ules in this paper. Now onwards we will denote the Levi subgroup of Pα(α ∈ S) containing T by Lα and the subgroup Lα ∩ B by Bα. Let π : G −→ G be the universal cover. Let Lα (respectively, Bα) be the inverse image of Lα (respectively, Bα). Lemma 2.3. Let V be an irreducible Lα-module. Let λ be a character of Bα. Then we have RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 5 (1) As Lα-modules, H j(Lα/Bα, V ⊗ Cλ) ≃ V ⊗ H j(Lα/Bα, Cλ). (2) If hλ, αi ≥ 0, then H 0(Lα/Bα, V ⊗ Cλ) is isomorphic as an Lα-module to the tensor product of V and H 0(Lα/Bα, Cλ). Further, we have H j(Lα/Bα, V ⊗ Cλ) = 0 for every j ≥ 1. (3) If hλ, αi ≤ −2, then H 0(Lα/Bα, V ⊗Cλ) = 0, and H 1(Lα/Bα, V ⊗Cλ) is isomorphic to the tensor product of V and H 0(Lα/Bα, Csα ·λ). (4) If hλ, αi = −1, then H j(Lα/Bα, V ⊗ Cλ) = 0 for every j ≥ 0. Proof. Proof (1). By [13, Proposition 4.8, p.53, I] and [13, Proposition 5.12, p.77, I], for all j ≥ 0, we have the following isomorphism of Lα-modules: H j(Lα/Bα, V ⊗ Cλ) ≃ V ⊗ H j(Lα/Bα, Cλ). Proof of (2), (3) and (4) follows from Lemma 2.2 by taking w = sα and the fact that Lα/Bα ≃ Pα/B. (cid:3) p.130]). Lemma 2.4. Recall the structure of indecomposable modules over Bα and eBα (see [1, Corollary 9.1, (1) Any finite dimensional indecomposable eBα-module V is isomorphic to V ′ ⊗ Cλ for some irreducible representation V ′ of eLα and for some character λ of eBα. (2) Any finite dimensional indecomposable Bα-module V is isomorphic to V ′ ⊗ Cλ for some irreducible representation V ′ of eLα and for some character λ of eBα. Now onwards we will assume that G = P SO(2n + 1, C)(n ≥ 3). Note that longest element w0 of the Weyl group W of G is equal to −id. We recall the following Proposition from [15, Proposition 1.3, p.858]. Proposition 2.5. Let c ∈ W be a Coxeter element, let ωi be the fundamental weight corresponding to the simple root αi. Then there exists a least positive integer h(i, c) such that ch(i,c)(ωi) = w0(ωi). Lemma 2.6. Let c ∈ W be a Coxeter element. Then we have (1) w0 = cn. (2) For any sequence ir(1 ≤ r ≤ n) of reduced expressions of c; the sequence i = (i1, i2, . . . , ir) is a reduced expression of w0. Proof. Note that for n ≥ 3, there is an isomorphism of Weyl group of Bn and Weyl group of Cn sending si 7→ si for (1 ≤ i ≤ n). Proof of the lemma holds in the case of type Cn for (n ≥ 3) (see [4, Lemma 4.2, p.441]). Therefore lemma holds for type Bn (n ≥ 3). (cid:3) Lemma 2.7. Let n ≥ a1 > a2 > · · · > ar−1 > ar ≥ 1 be a decreasing sequence of integers. Then, w = ( nQ j=a1 sj)( nQ j=a2 sj) · · · ( nQ j=ar−1 sj)( n−1Q j=ar sj) is a reduced expression of w. 6 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Proof. Note that for n ≥ 3, there is an isomorphism of Weyl group of Bn and Weyl group of Cn sending si 7→ si for (1 ≤ i ≤ n). Proof of the lemma holds in the case of type Cn for (n ≥ 3) (see [4, Lemma 4.3,p.441]). Therefore lemma holds for type Bn (n ≥ 3). (cid:3) Let c be a Coxeter element in W. We take a reduced expression c = [a1, n][a2, a1 − 1] · · · [ak, ak−1 − 1], where [i, j] = sisi+1 · · · sj for i ≤ j and n ≥ a1 > a2 > · · · > ak = 1. Then we have following. Lemma 2.8. (1) For all 1 ≤ i ≤ k − 1, ci = ( iQ l1=1 [al1, n])( kQ l2=i+1 [al2, al2−i − 1])( i−1Q l3=1 [alk , ak−i+l3 − 1]). (2) For all k ≤ j ≤ n, cj = ( k−1Q l1=1 [al1, n])([ak, n]j+1−k)( k−1Q l2=1 [ak, al2 − 1]). (3) The expressions of ci for 1 ≤ i ≤ n as in (1) and (2) are reduced. Proof. Note that for n ≥ 3, there is an isomorphism of Weyl group of Bn and Weyl group of Cn sending si 7→ si for (1 ≤ i ≤ n). Proof of the lemma holds in the case of type Cn for (n ≥ 3) (see [4, Lemma 4.4, p.442]). Therefore lemma holds for type Bn (n ≥ 3). (cid:3) 3. cohomology modules H 1(w, αj) where j 6= n − 1 and w ∈ W In this section, we prove that H 1(w, αj) = 0 for every w ∈ W and j 6= n − 1. Lemma 3.1. Let v ∈ W and α ∈ S. Then H 1(sj, H 0(v, α)) = 0 for j 6= n. Proof. By [14, Corollary 5.6, p.778] we have H 1(w, αn) = 0. Therefore, we may assume that α is a long simple root. If H 1(sj, H 0(v, α))µ 6= 0, then there exists an indecomposable Lαj - summand V of H 0(v, α) such that H 1(sj, V )µ 6= 0. By Lemma 2.4, we have V ≃ V ′ ⊗ Cλ for some character λ of Bαj and for some irreducible Lαj -module V ′. Since H 1(sj, V )µ 6= 0 from Lemma 2.3(3) we have hλ, αji ≤ −2. Since α is a long root, there exists w ∈ W such that w(α) = α0. Thus H 0(v, α) ⊆ H 0(vw, α0). Again, since α0 is highest long root, H 0(w0, α0) = g −→ H 0(vw, α0) is surjective. Let µ′ be the lowest weight of V. Then by the above argument µ′ is a root. Therefore we have µ′ = µ1 + λ, where µ1 is the lowest weight of V ′. Hence, we have hµ′, αji ≤ −2. Since αj is a long root and µ′ is a root, we have hµ′, αji = −1, 0, 1. This is a contradiction. Thus we have H 1(sj, H 0(v, α))µ = 0. (cid:3) Lemma 3.2. Let v ∈ W and αj ∈ S be such that j 6= n − 1. Then we have H 1(sk, H 0(v, αj)) = 0, for every k = 1, 2, . . . , n. Proof. Step 1: H 0(v, αj)−(αn−1+2αn) = 0. Case 1: Assume that j = n, choose an element u ∈ W of minimal length such that u−1(αn) = β0, the highest short root. Then we have H 0(v, αj) ⊆ H 0(vu, β0). RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 7 Since β0 is dominant weight the natural restriction map H 0(w0, β0) −→ H 0(vu, β0) is surjective. Hence H 0(v, αj)µ 6= 0 implies either µ = 0 or µ is a short root. Therefore, we have H 0(v, αj)−(αn−1+2αn) = 0. Case 2: Assume that 1 ≤ j ≤ n − 2. Note that if H 0(v, αj)µ 6= 0 then either µ = αj, 0 or µ ≤ −αj (see [4, Corollary 4.5, p.678]). Hence H 0(v, αj)−(αn−1+2αn) = 0. Step 2: If H 0(v, αj)−(βi+2αn) 6= 0 for some 1 ≤ i ≤ n − 2, then H 0(v, αj)−(βi+αn) 6= 0. Proof of Step 2: If v = id, we are done. So choose 1 ≤ t ≤ n such that l(stv) = l(v) − 1. Let v ′ = stv. Then H 0(v, αj) = H 0(st, H 0(v ′, αj)). Case 1: Assume that t = n. In this case h−(βi + 2αn), αti = −2. If H 0(v, αj)−(βi+2αn) 6= 0, then there is an indecomposable Bαn-summand V of H 0(v ′, αj) with highest weight −βi. Since h−βi, αni = 2, we have H 0(st, V )−(βi+αn) 6= 0. Therefore we have H 0(v, αj)−(βi+αn) 6= 0. Case 2: Assume that t = n − 1. In this case h−(βi + 2αn), αti = 1. If H 0(v, αj)−(βi+2αn) 6= 0, then there is an indecomposable Bαn-summand V of H 0(v ′, αj) with highest weight −(βi + 2αn). Thus by induction hypothesis we have H 0(v ′, αj)−(βi+αn) 6= 0. Since h−(βi + αn), αti = 0, we have H 0(v, αj)−(βi+αn) 6= 0. Case 3: Assume that 1 ≤ t ≤ n − 2. In this case h−(βi + 2αn), αti = −1, 0 or 1. Assume that i = t. Then we have h−(βi + 2αn), αti = −1. If further H 0(v, αj)−(βi+2αn) 6= 0, then there is an indecomposable Bαt-summand V of H 0(v ′, αj) with highest weight −(βi+1 + 2αn). Therefore we have H 0(v ′, αj)−(βi+1+2αn) 6= 0. It is clear from Step: 1 that t+1 ≤ n−2. Therefore by induction H 0(v ′, αj)−(βt+1+αn) 6= 0. Since h−(βt+1 +αn), αti = 1, we have H 0(v, αj)−(βt+αn) 6= 0. Assume that 1 ≤ t ≤ i−2 or i+ 1 ≤ t ≤ n−2. Then we have h−(βi + 2αn), αti = 0. Thus H 0(v, αj)−(βt+2αn) = H 0(v ′, αj)−(βt+2αn) 6= 0. Therefore by induction H 0(v, αj)−(βt+αn) 6= 0. Since h−(βi + αn), αti = 0, we have H 0(v, αj)−(βi+2αn) 6= 0. Assume that i = t + 1. Since h−(βi + αn), αti = 1, then there is an indecomposable Bαt-summand V of H 0(v ′, αj) such that: V = C−(βi+2αn) ⊕ C−(βi−1+2αn) or V = C−(βi+2αn). Then we have H 0(v ′, αj)−(βt+2αn) 6= 0. Therefore by induction H 0(v ′, αj)−(βi+αn) 6= 0. Since h−(βt + αn), αti = 1, we have H 0(v, αj)−(βi+αn) 6= 0 Hence the proof of Step 2. Proof of Lemma : Case 1: Assume that k 6= n. Then by Lemma 3.1 we have H 1(sk, H 0(v, αj)) = 0. Case 2: Assume that k = n. By Step 1 we see that H 0(v, αj)−(αn−1+2αn) = 0. Note that if β is a root such that H 0(v, αj)β 6= 0 and hβ, αni = −2, then we have β = −(βi + 2αn) for some 1 ≤ i ≤ n − 2. 8 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA By Step 2 if H 0(v, αj)−(βi+2αn) 6= 0 for some 1 ≤ i ≤ n − 2, then H 0(v, αj)−(βi+αn) 6= 0. Therefore C−(βi+αn) ⊕ C−(βi+2αn) is an indecomposable Bαn-summand of H 0(v, αj). By Lemma 2.4, C−(βi+αn) ⊕ C−(βi+2αn) is isomorphic to V ⊗ C−ωn, where V is an irreducible Lαn-module. Therefore by Lemma 2.3(4) we have H 1(sk, C−(βi+αn) ⊕ C−(βi+2αn)) = 0. Thus our result follows. (cid:3) Lemma 3.3. Let w be an element of W and αj be an element of S such that j 6= n − 1. Then H 1(w, αj) = 0. Proof. We will prove by induction on length of w. If length of w is 0, then w = id. Thus it follows trivially. Now suppose w ∈ W such that l(w) ≥ 1. Then there exists a simple root α ∈ S such that l(sαw) = l(w) − 1. Then using SES: 0 −→ H 1(sα, H 0(sαw, αj)) −→ H 1(w, αj) −→ H 0(sα, H 1(sαw, αj)) −→ 0. From the above SES using induction hypothesis and Lemma 3.2, we get H 1(w, αj)) = 0 for j 6= n − 1. (cid:3) 4. cohomology module H 0 of the relative tangent bundle In this section we describe the weights of H 0 of the relative tangent bundle. Notation: Let c be a Coxeter element of W. We take a reduced expression c = [a1, n][a2, a1 − 1] · · · [ak, ak−1 − 1], where [i, j] = sisi+1 · · · sj for i ≤ j and n ≥ a1 > a2 > · · · > ak = 1. Let βi = αi + αi+1 + · · · + αn−1 for all 1 ≤ i ≤ n − 1. For 1 ≤ r ≤ k, let n ≥ a1 > a2 > a3 > · · · > ar ≥ 1 be a decreasing sequence of integers. Let nQ j=a1 nQ j=a2 nQ j=a3 nQ j=ar−1 n−1Q j=ar wr = ( sj)( sj)( sj) · · · ( sj)( sj) and let τr = ( nQ j=a1 sj)( nQ j=a2 sj)( nQ j=a3 sj) · · · ( nQ j=ar−1 sj)( n−2Q j=ar sj). Note that l(wr) = l(τr) + 1. Lemma 4.1. Assume that r ≥ 3. (1) Let v = sar−1sar−1+1 · · · snsar sar+1 · · · sn−1. Then we have ar−1−1L i=ar (C−(βi+αn)⊕C−(βi+2αn)⊕C−(βi+2αn+βn−1)⊕· · ·⊕C−(βi+2αn+βar−1 )) (C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)) ⊕ C−(βn−1+2αn). H 0(v, αn−1) = n−2L i=ar−1 (2) Let v ′ = s1 · · · sns1 · · · sn−1. Then we have H 0(v ′, αn−1) = n−2L i=1 (C−(βi+2αn)⊕C−(βi+2αn+βn−1)⊕· · ·⊕C−(βi+2αn+βi+1))⊕C−(βn−1+2αn). RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 9 Proof. Proof of (1): Let u = sar sar+1 · · · sn−1. By using SES we have H 0(u, αn−1) = Ch(αn−1) ⊕ ( n−1L j=ar C−βj ). Since h−βj, αni = 2 for all ar ≤ j ≤ n − 1, by using SES we have H 0(snu, αn−1) = Ch(αn−1)⊕( n−1L j=ar C−βj ⊕C−(βj +αn)⊕C−(βj +2αn)). (4.1.1) Since Ch(αn−1)⊕C−αn−1 is an indecomposable Bαn−1-module (see [5, p.11] and [14, p.8]), by Lemma 2.4, we have Ch(αn−1) ⊕ C−αn−1 = V ⊗ C−ωn−1, where V is the two dimensional irreducible repres- entation of Lαn−1. Therefore by Lemma 2.3(4) we have H 0( Lαn−1/ Bαn−1, Ch(αn−1) ⊕ C−αn−1) = 0. Moreover, we have h−(βn−1 + αn), βn−1i = −1, h−(βn−1 + 2αn), βn−1i = 0, h−βj, βn−1i = −1, h−(βj + αn), βn−1i = 0 and h−(βj + 2αn), βn−1i = 1 for all ar ≤ j ≤ n − 2. Therefore we have H 0(sn−1snu, αn−1) = ( n−2L j=ar C−(βj+αn)⊕C−(βj +2αn)⊕C−(βj +2αn+βn−1))⊕C−(βn−1+2αn). (4.1.2) (C−(βj +αn)⊕C−(βj +2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βk)) Claim: For ar−1 ≤ k ≤ n − 2 k−1L H 0(sksk+1 · · · snu, αn−1) = n−2L j=ar j=k H 0(sk+1 · · · snu, αn−1) = n−2L j=k+1 (C−(βj+2αn) ⊕ C−(βj+2αn+βn−1) ⊕ · · · ⊕ C−(βj+2αn+βj+1)) ⊕ C−(βn−1+2αn). Proof of the claim: We will prove by descending induction on k. By hypothesis we have kL j=ar (C−(βj+αn)⊕C−(βj +2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βk+1)) (C−(βj+2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βj+1))⊕C−(βn−1+2αn). (4.1.3) Let V = k−1L j=ar (C−(βj +αn) ⊕ C−(βj +2αn) ⊕ C−(βj +2αn+βn−1) ⊕ · · · ⊕ C−(βj+2αn+βk+1)) and V ′ = j=k+2 (C−(βj+2αn) ⊕ C−(βj+2αn+βn−1) ⊕ · · · ⊕ C−(βj+2αn+βj+1)) ⊕ C−(βn−1+2αn). n−2L Then roots {(βj +αn), (βj +2αn), (βj +2αn+βn−1), . . . , (βj +2αn+βk+2) : ar ≤ j ≤ k−1}, {(βj + 2αn), (βj + 2αn + βn−1), . . . , (βj + 2αn + βj+1) : k + 2 ≤ j ≤ n − 2} and −(βn−1 + 2αn) are orthogonal to αk. Therefore V , V ′ are direct sums of irreducible Lαk−modules. By Lemma 2.3(2), we have H 0( Lαk/ Bαk , V ) = V, H 0( Lαk / Bαk , V ′) = V ′ and H 0( Lαk/ Bαk , C−(βn−1+2αn)) = C−(βn−1+2αn). 10 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Further the remaining roots of (4.1.3) are {−(βk + αn), −(βk + 2αn), −(βk + 2αn + βn−1), . . . , −(βk + 2αn + βk+2), −(βk + 2αn + βk+1)}, {−(βk+1 + 2αn), −(βk+1 + 2αn + βn−1), . . . , −(βk + 2αn + βk+2)} and {−(βj + 2αn + βk+1) : ar ≤ j ≤ k}. Since h−(βk + αn), αki = −1, by Lemma 2.3(4) we have H 0( Lαk/ Bαk , C−(βk+αn)) = 0. Since C−(βk+2αn)⊕C−(βk+1+2αn) is the irreducible two dimensional Lαk -module, by Lemma 2.3(2) we have H 0( Lαk/ Bαk , C−(βk+2αn) ⊕ C−(βk+1+2αn)) = C−(βk+2αn) ⊕ C−(βk+1+2αn). Similarly for each k + 2 ≤ j ≤ n − 1, C−(βk+2αn+βj) ⊕ C−(βk+1+2αn+βj) is the irreducible two dimensional Lαk−module. Therefore by Lemma 2.3(2) we have H 0( Lαk/ Bαk , C−(βk+2αn+βj)⊕C−(βk+1+2αn+βj)) = C−(βk+2αn+βj)⊕C−(βk+1+2αn+βj)) for each k + 2 ≤ j ≤ n − 1. Moreover, h−(βj + 2αn + βk+1), αki = 1 for all ar ≤ j ≤ k − 1. Therefore by Lemma 2.3(2) we have H 0( Lαk/ Bαk , C−(βj+2αn+βk+1)) = C−(βj +2αn+βk+1) ⊕ C−(βj+2αn+βk) for all ar ≤ j ≤ k − 1. Since h−(βk + 2αn + βk+1), αki = 0, by Lemma 2.3(2) we have H 0( Lαk/ Bαk , C−(βk+2αn+βk+1)) = C−(βk+2αn+βk+1). From the above discussion, we have k−1L j=ar (C−(βj +αn)⊕C−(βj +2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βk)) H 0(sksk+1 · · · snu, αn−1) = n−2L j=k (C−(βj+2αn) ⊕ C−(βj+2αn+βn−1) ⊕ · · · ⊕ C−(βj+2αn+βj+1)) ⊕ C−(βn−1+2αn). Therefore the claim follows. The claim implies H 0(v, αn−1) = H 0(sar−1sar+1 · · · snv ′ r, αn−1) = ar−1−1L j=ar (C−(βj +αn)⊕C−(βj +2αn)⊕C−(βj+2αn+βn−1)⊕ · · ·⊕C−(βj +2αn+βar−1)) n−2L j=ar−1 (C−(βj +2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βj+1))⊕C−(βn−1+2αn). Proof of (2): By (1) we have H 0(s2s3 · · · sns1s2 · · · sn−1, αn−1) = (C−(β1+αn) ⊕ C−(β1+2αn) ⊕ C−(β1+2αn+βn−1) ⊕ · · · ⊕ C−(β1+2αn+β2)) n−2L j=2 (C−(βj+2αn)⊕C−(βj +2αn+βn−1)⊕· · ·⊕C−(βj +2αn+βj+1))⊕C−(βn−1+2αn). (4.1.4) The roots of (4.1.4) are {−(β1 + αn), −(β1 + 2αn), −(β1 + 2αn + βn−1), . . . , −(β1 + 2αn + β3), −(β1 + 2αn + β2)}, {−(β2 + 2αn), −(β2 + 2αn + βn−1), . . . , −(β1 + 2αn + β3)} and −(β1 + 2αn + β2). Since −(βn−1 + 2αn) is orthogonal to α1, by Lemma 2.3(2) we have H 0( Lα1/ Bα1, C−(βn−1+αn)) = C−(βn−1+αn). Since h−(β1 + αn), α1i = −1, by Lemma 2.3(4) we have H 0( Lα1/ Bα1, C−(β1+αn)) = 0. RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 11 Since C−(β1+2αn) ⊕ C−(β2+2αn) is the irreducible two dimensional Lα1-module, by Lemma 2.3(2) we have H 0( Lα1/ Bα1, C−(β1+2αn) ⊕ C−(β2+2αn)) = C−(β1+2αn) ⊕ C−(β2+2αn). Similarly for each 3 ≤ j ≤ n − 1, C−(β1+2αn+βj) ⊕ C−(β2+2αn+βj) is the irreducible two dimensional Lα1−module. Therefore by Lemma 2.3(2) we have H 0( Lα1/ Bα1, C−(β1+2αn+βj) ⊕ C−(β2+2αn+βj)) = C−(β1+2αn+βj) ⊕ C−(β2+2αn+βj)) for each 3 ≤ j ≤ n − 1. Since h−(β1 + 2αn + β2), α1i = 0, by Lemma 2.3(2) we have H 0( Lα1/ Bα1, C−(β1+2αn+β2)) = C−(β1+2αn+β2). From the above discussion, we have H 0(v ′, αn−1) = n−2L i=1 (C−(βi+2αn)⊕C−(βi+2αn+βn−1)⊕· · ·⊕C−(βi+2αn+βi+1))⊕C−(βn−1+2αn). (cid:3) Lemma 4.2. Let 3 ≤ r ≤ k and let v = sar−1sar−1+1 · · · snsar sar+1 · · · sn−1. Then we have (1) H 0(sar−2 · · · snv, αn−1) = ar−1−1L i=ar (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βar−1 )) ar−2−2L i=ar−1 (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). (2) H 0(wr, αn−1) = ar−1−1L i=ar (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βar−1 )) ar−2−2L i=ar−1 (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). (3) Let u1 = (sa1 · · · sn)(sa2 · · · sn) · · · (sak−1 · · · sn)v ′, where v ′ is defined as in Lemma 4.1. Then we have H 0(u1, αn−1) = ak−1−2L i=1 (C−(βi+2αn+βak−1−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). Proof. Proof of (1): Since r ≥ 3, we have ar−1 < n. By Lemma 4.1(1), we have ar−1−1L i=ar (C−(βi+αn) ⊕ C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βar−1)) H 0(v, αn−1) = n−2L Since {−(βi + 2αn + βn−1), . . . , −(βi + 2αn + βar−1) : ar ≤ i ≤ ar−1 − 1} are orthogonal (C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)) ⊕ C−(βn−1+2αn). i=ar−1 to αn, by Lemma 2.3(2), we have H 0( Lαn/ Bαn, C−(βi+2αn+βt)) = C−(βi+2αn+βt) for all ar ≤ i ≤ ar−1 − 1 and ar−1 ≤ t ≤ n − 1. Since {−(βi + 2αn + βn−1), . . . , −(βi + 2αn + βi+1) : ar−1 ≤ i ≤ n − 2} are orthogonal to αn, by Lemma 2.3(2) we have H 0( Lαn/ Bαn, C−(βi+2αn+βl)) = C−(βi+2αn+βl) for all i + 1 ≤ l ≤ n − 1, where ar−1 ≤ i ≤ n − 2. Since h−(βi + 2αn), αni = −2 for all ar ≤ i ≤ n − 1, by Lemma 2.3(3) we have 12 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA H 0( Lαn/ Bαn, C−(βn−1+2αn)) = 0 for all ar−1 ≤ i ≤ n − 1. Moreover, for each ar ≤ i ≤ ar−1 − 1, C−(βi+αn) ⊕ C−(βi+2αn) is an indecomposable two dimensioal Bαn-module. Therefore by Lemma 2.4, we have C−(βi+αn) ⊕ C−(βi+2αn) = Vi ⊗ C−ωn, where Vi is the irreducible two dimensional representation of Lαn. By Lemma 2.3(4) we have H 0( Lαn/ Bαn, C−(βi+αn) ⊕ C−(βi+2αn)) = 0 for each ar ≤ i ≤ ar−1 − 1. From the above discussion, we have H 0(snvr, αn−1) = ar−1−1L i=ar (C−(βi+2αn+βn−1) ⊕· · ·⊕ C−(βi+2αn+βar−1)) n−2L i=ar−1 (C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). Since h−(βi + 2αn + βn−1), αn−1i = −1 for each ar ≤ i ≤ n − 2, by Lemma 2.3(4) we have H 0( Lαn−1/ Bαn−1, C−(βi+αn+βn−1)) = 0 for each ar ≤ i ≤ n − 2. Moreover, {−(βi + 2αn + βn−2), . . . , −(βi + 2αn + βar−1) : ar ≤ i ≤ ar−1 − 1}, {−(βi + 2αn + βn−2), . . . , −(βi + 2αn + βi+1) : ar−1 ≤ i ≤ n − 3}, and βn−1 + 2αn are orthogonal to αn−1. Therefore we have H 0(sn−1snv, αn−1) = ar−1−1L i=ar (C−(βi+2αn+βn−2)⊕· · ·⊕C−(βi+2αn+βar−1)) n−3L i=ar−1 (C−(βi+2αn+βn−2)⊕ · · · ⊕ C−(βi+2αn+βi+1)). Proceeding recursively, we have H 0(sar−2sar−2+1 · · · snv, αn−1) = ar−2−2L (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). i=ar−1 ar−1−1L i=ar (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βar−1 ))L Proof of (2): Since {−(βi + 2αn + βar−2−1), . . . , −(βi + 2αn + βar−1) : ar ≤ i ≤ ar−1 − 1}, {−(βi + 2αn + βar−2−1), . . . , −(βi + 2αn + βi+1) : ar−1 ≤ i ≤ ar−2 − 2} are orthogonal to αj for all ar−3 ≤ j ≤ n, by Lemma 2.3(2) we have H 0(sar−3 · · · snsar−2 · · · snv, αn−1) = H 0(sar−2 · · · snv, αn−1). Proceeding recursively we have H 0(wr, αn−1) = H 0(sar−2 · · · snv, αn−1) = ar−2−2L (C−(βi+2αn+βar−2−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). i=ar−1 ar−1−1L i=ar (C−(βi+2αn+βar−2−1)⊕· · ·⊕C−(βi+2αn+βar−1)) Proof of (3): By Lemma 4.1(2) we have n−2L i=1 H 0(v ′, αn−1) = (C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)) ⊕ C−(βn−1+2αn). Since {−(βi + 2αn + βn−1), . . . , −(βi + 2αn + βi+1) : 1 ≤ i ≤ n − 2} are orthogonal to αn, by Lemma 2.3(2) we have H 0( Lαn/ Bαn, C−(βi+2αn+βl)) = C−(βi+2αn+βl) for all i+1 ≤ l ≤ n−1, where 1 ≤ i ≤ n−2. RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 13 Since h−(βi + 2αn), αni = −2 for all 1 ≤ i ≤ n − 1, by Lemma 2.3(3) we have H 0( Lαn/ Bαn, C−(βn−1+2αn)) = 0 for all 1 ≤ i ≤ n − 1. From the above discussion, we have H 0(snv ′, αn−1) = (C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). n−2L i=1 Since h−(βi + 2αn + βn−1), αn−1i = −1 for each 1 ≤ i ≤ n − 2, by Lemma 2.3(4) we have H 0( Lαn−1/ Bαn−1, C−(βi+2αn+βn−1)) = 0 for each 1 ≤ i ≤ n − 2. Moreover, {−(βi + 2αn + βn−2), . . . , −(βi + 2αn + βi+1) : 1 ≤ i ≤ n − 3}, and βn−1 + 2αn are orthogonal to αn−1. Therefore we have H 0(sn−1snv ′, αn−1) = n−3L i=1 Proceeding recursively we have (C−(βi+2αn+βn−2) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). ak−1−2L i=1 H 0(sak−1sak−1+1 · · · snv ′, αn−1) = (C−(βi+2αn+βak−1−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). Since {−(βi + 2αn + βak−2−1), . . . , −(βi + 2αn + βi+1) : 1 ≤ i ≤ ak−1 − 2} are orthogonal to αj for all ak−2 ≤ j ≤ n, we have H 0(sak−2 · · · snsak−1 · · · snv ′, αn−1) = H 0(sak−1 · · · snv ′, αn−1). Proceeding recursively we have H 0(u1, αn−1) = ak−1−2L i=1 (C−(βi+2αn+βak−1−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)). (cid:3) Lemma 4.3. Let 3 ≤ r ≤ k. Then H 0(wr−2snsar−1sar−1+1 · · · sn+2−r, αn+2−r)µ 6= 0 if µ is of the form µ = −(βj + αn) for some ar−1 ≤ j ≤ ar−2 − 1. Proof. By applying SES repeatedly, it is easy to see that H 0(snsar−1sar−1+1 · · · sn+2−r, αn+2−r) = Ch(αn+2−r) ⊕ ( n+2−rL j=ar−1 C−γj,n+2−r ), where γj,j ′ = (αj + · · · + αj ′) for j ′ ≥ j. Let V1 = H 0(snsar−1sar−1+1 · · · sn+2−r, αn+2−r). We next calculate H 0(sar−2 · · · sn−1, V1). Since n ≥ a1 > a2 > · · · > ak = 1, we have ai ≤ n + 1 − i for all 1 ≤ i ≤ k. Assume l ≥ n + 4 − r, then hγj,n+2−r, αli = 0 for all ar−1 ≤ j ≤ n + 2 − r. By Lemma 2.3(2) we have H 0( Lαl/ Bαl, V1) = V1 for all l ≥ n + 4 − r. Therefore we have H 0(sar−2 · · · sn−1, V1) = H 0(sar−2 · · · sn+2−rsn+3−r, V1). Note that, since h−γj,n+2−r, αn+3−ri = 1 for all ar−1 ≤ j ≤ n + 2 − r, by Lemma 2.3(2) we have H 0(sn+3−r, V1) = Ch(αn+2−r) ⊕ ( n+2−rL j=ar−1 (C−γj,n+2−r ⊕ C−γj,n+3−r )). Since Ch(αn+2−r)⊕C−γn+2−r,n+2−r is an indecomposable two dimensional Bαn+2−r -module, by Lemma 2.4, Ch(αn+2−r) ⊕ C−γn+2−r,n+2−r = V ⊗ C−ωn+2−r , where V is the irreducible two dimensional Lαn+2−r -module. Since hγj,n+2−r, αn+2−ri = −1, for all ar−1 ≤ j ≤ n + 1 − r, by Lemma 2.3(4) we have H 0( Lαn+2−r / Bαn+2−r , Cγj,n+2−r ) = 0 for all ar−1 ≤ j ≤ n + 1 − r. From the above discussion, we have 14 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA H 0(sn+2−r, H 0(sn+3−r, V1)) = n+1−rL j=ar−1 C−γj,n+3−r . Since hγn+1−r,n+3−r, αn+1−ri = −1, by Lemma 2.3(4), we have H 0( Lαn+1−r / Bαn+1−r , Cγn+1−r,n+3−r ) = 0. Moreover, hγj,n+3−r, αn+1−ri = 0 for all ar−1 ≤ j ≤ n − r. Therefore we have H 0(sn+1−r, H 0(sn+2−r, H 0(sn+3−r, V1))) = Proceeding recursively we have n−rL j=ar−1 C−γj,n+3−r . H 0(sar−2 · · · sn−1, V1) = ar−2−1L j=ar−1 C−γj,n+3−r and H 0(snsar−2 · · · sn−1, V1) = ar−2−1L j=ar−1 C−γj,n+3−r . ar−2−1L j=ar−1 C−γj,n+4−r Let V2 = H 0(snsar−2 · · · sn−1, V1). Similarly, we have H 0(sar−3 · · · sn−1, V2) = and H 0(snsar−3 · · · sn−1, V2) = ar−2−1L j=ar−1 C−γj,n+4−r . Proceeding recursively we have Vr−2 = H 0(sa2 · · · sn−1, Vr−3) = ar−2−1L j=ar−1 C−γj,n−1, where Vr−3 = H 0((sns3 · · · sn) · · · (sar−2 · · · sn+2−r), αn+2−r). Note that γj,n−1 = βj. Since h−αn−1, αni = 2, by Lemma 2.3(2) and Lemma 2.4, we have H 0(sn, Vr−2) = ar−2−1M (C−βj ⊕ C−(βj+αn) ⊕ C−(βj +2αn)). j=ar−1 Moreover, h−βj, αn−1i = −1, h−(βj + αn), αn−1i = 0 and h−(βj + 2αn), αn−1i = 1 for all ar−1 ≤ j ≤ ar−2 − 1. Therefore by Lemma 2.3(2), Lemma 2.3(4) and Lemma 2.4 we have H 0(sn−1, H 0(sn, Vr−2)) = ar−2−1M (C−(βj +αn) ⊕ C−(βj+2αn) ⊕ C−(βj +2αn+βn−1)). j=ar−1 Proceeding recursively we have H 0(sa1, H 0(sa1+1 · · · sn, Vr−2)) = ar−2−1L j=ar−1 C−(βj+2αn+βa1 )). Hence the proof of the lemma follows. (C−(βj+αn) ⊕ C−(βj+2αn) ⊕ C−(βj+2αn+βn−1) ⊕ · · · ⊕ Lemma 4.4. Let 3 ≤ r ≤ k. Then H 0(wr−1sn, αn)µ 6= 0 if and only if µ is of the form µ = −(βj + αn), for some ar−1 ≤ j ≤ ar−2 − 1. Proof. By applying SES repeatedly, it is easy to see that (cid:3) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 15 H 0(sar−1 · · · sn−1sn, αn) = n−1L j=ar−1 C−(βj+αn). Let V1 = H 0(sar−1 · · · sn−1sn, αn). Since h−(βj + αn), αni = 0, we have H 0(sn, V1) = V1. Since h−(βn−1 + αn), αn−1i = −1 and h−(βj + αn, αn−1)i = 0, for all ar−1 ≤ j ≤ n − 2, by Lemma 2.3(2), Lemma 2.3(4) and Lemma 2.4 we have H 0(sn−1, V1) = n−2M j=ar−1 C−(βj+αn). Proceeding recursively we have H 0(sar−2 · · · sn−1sn, V1) = ar−2−1M j=ar−1 C−(βj+αn). Since n ≥ a1 > a2 > · · · > ak = 1, we see that h−(βj + αn), αti = 0, for all ar−1 ≤ j ≤ ar−2 − 1 and for all ar−3 ≤ t ≤ n, therefore by Lemma 2.3(2) and Lemma 2.4, we have H 0(sar−3 · · · snsar−2 · · · snsar−1 · · · sn−1sn, αn) = ar−2−1M j=ar−1 C−(βj +αn). Since n ≥ a1 > a2 > · · · > ak = 1, we see that h−(βj + αn), αti = 0 for all ar−1 ≤ j ≤ ar−2 − 1 and for all ar−4 ≤ t ≤ n. By Lemma 2.3(2) and Lemma 2.4, we have H 0(sar−4 · · · snsar−3 · · · snsar−2 · · · snsar−1 · · · sn−1sn, αn) = ar−2−1M j=ar−1 C−(βj+αn). Proceeding recursively we have H 0(wr−1sn, αn) = ar−2−1M j=ar−1 C−(βj+αn). Hence the lemma follows. (cid:3) Lemma 4.5. If µ is of the form µ = −(βj + αn) for some 1 ≤ j ≤ ak−1 − 1, then we have H 0(wk−1sns1s2 · · · sn+1−k, αn+1−k)µ 6= 0. Proof. By applying SES repeatedly, it is easy to see that H 0(snsa1s2 · · · sn+1−k, αn+1−k) = Ch(αn+1−k) ⊕ ( n+1−kL j=1 C−γj,n+1−k), where γj,j ′ = (αj + · · · + αj ′) for j ′ ≥ j. Let V1 = H 0(sns1s2 · · · sn+1−k, αn+1−k). We next calculate H 0(sak−1 · · · sn−1, V1). Since n ≥ a1 > a2 > · · · > ak = 1, we have ai ≤ n + 1 − i for all 1 ≤ i ≤ k. Moreover, hγj,n+1−k, αli = 0 for all 1 ≤ j ≤ n + 1 − k and for all l ≥ n + 3 − k. Therefore by using Lemma 2.3(2) and Lemma 2.4, we have 16 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA H 0(sak−1 · · · sn−1, V1) = H 0(sak−1 · · · sn+1−ksn+2−k, V1). Since h−γj,n+1−k, αn+2−ki = 1 for all 1 ≤ j ≤ n + 1 − k, by using Lemma 2.3(2) and Lemma 2.4, we have H 0(sn+2−k, V1) = Ch(αn+1−k) ⊕ ( n+1−kL j=1 (C−γj,n+1−k ⊕ C−γj,n+2−k )). Since Ch(αn+1−k)⊕C−γn+1−k,n+1−k is an indecomposable two dimensional Bαn+1−k -module, by Lemma 2.4 we have Ch(αn+1−k) ⊕ C−γn+1−k,n+1−k = V ⊗ C−ωn+1−k , where V is the irreducible two dimensional Lαn+1−k -module. By Lemma 2.3(4) we have H 0( Lαn+1−k / Bαn+1−k , Ch(αn+1−k) ⊕ C−γn+1−k,n+1−k ) = 0. Since hγj,n+1−k, αn+1−ki = −1 for all 1 ≤ j ≤ n − k, by Lemma 2.3(4) we have H 0( Lαn+1−k / Bαn+1−k , Cγj,n+1−k) = 0 for all 1 ≤ j ≤ n − k. From the above discussion, we have H 0(sn+1−k, H 0(sn+2−k, V1)) = n−kL j=1 C−γj,n+2−k . Since hγj,n+2−k, αn−ki = 0 for all 1 ≤ j ≤ n − k − 1, and hγn−k,n+2−k, αn−ki = −1, by using Lemma 2.3(2), Lemma 2.3(2)(4) and Lemma 2.4, we have H 0(sn−k, H 0(sn+1−k, H 0(sn+2−k, V1))) = n−k−1M j=1 C−γj,n+2−k . Proceeding recursively we have H 0(snsak−1 · · · sn−1, V1) = ak−1−1M j=1 C−γj,n+2−k . Let V2 = H 0(snsak−1 · · · sn−1, V1). Then similarly, we have H 0(sak−2 · · · sn−1, V2) = ak−1−1M j=1 C−γj,n+3−k . Proceeding recursively we have Vk−1 = H 0(sa2 · · · sn−1, Vk−2) = ak−1−1M j=1 C−γj,n−1 where Vk−2 = H 0((sns3 · · · sn) · · · (sak−1 · · · sn+1−k), αn+1−k). Note that γj,n−1 = βj. Since h−αn−1, αni = 2, by using Lemma 2.3(2) and Lemma 2.4 we have H 0(sn, Vk−1) = ak−1−1L j=1 (C−βj ⊕ C−(βj+αn) ⊕ C−(βj +2αn)). RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 17 Moreover, h−βj, αn−1i = −1, h−(βj + αn), αn−1i = 0 and h−(βj + 2αn), αn−1i = 1, for all 1 ≤ j ≤ ak−1 − 1. Therefore by using Lemma 2.3(2), Lemma 2.3(4) and Lemma 2.4 we have H 0(sn−1, H 0(sn, Vk−1)) = ak−1−1M (C−(βj +αn) ⊕ C−(βj+2αn) ⊕ C−(βj +2αn+βn−1)). Proceeding recursively we have H 0(sa1, H 0(sa1+1 · · · sn, Vk−1)) = j=1 ak−1−1L j=1 C−(βj+2αn+βa1 )). Hence the proof of the lemma follows. (C−(βj+αn) ⊕ C−(βj+2αn) ⊕ C−(βj+2αn+βn−1) ⊕ · · · ⊕ Lemma 4.6. H 0(wksn, αn)µ 6= 0 if and only if µ is of the form µ = −(βj + αn) for some 1 ≤ j ≤ ak−1 − 1. Proof. By applying SES repeatedly it is easy to see that (cid:3) H 0(s1 · · · sn−1sn, αn) = n−1M j=1 C−(βj +αn). Let V1 = H 0(s1 · · · sn−1sn, αn). Since h−(βj + αn), αni = 0, for all 1 ≤ j ≤ n, by Lemma 2.3(2) and Lemma 2.4 we have H 0(sn, V1) = V1. Moreover, h−(βn−1 + αn), αn−1i = −1 and h−(βj + αn, αn−1)i = 0 for all 1 ≤ j ≤ n − 2. Therefore by using Lemma 2.3(4), Lemma 2.3(2) and Lemma 2.4, we have H 0(sn−1, V1) = n−2M j=1 C−(βj +αn). Proceeding recursively we have H 0(sak−1 · · · sn−1sn, V1) = ak−1−1M j=1 C−(βj +αn). Since n ≥ a1 > a2 > · · · > ak = 1, we have h−(βj + αn), αti = 0 for all 1 ≤ j ≤ ak−1 − 1 and for all ak−2 ≤ t ≤ n. By using Lemma 2.3(2), Lemma 2.4 we have H 0(sak−2 · · · snsak−1 · · · sns1 · · · sn−1sn, αn) = ak−1−1M j=1 C−(βj+αn). Similarly, since n ≥ a1 > a2 > · · · > ak = 1, we have h−(βj + αn), αti = 0 for all 1 ≤ j ≤ ak−1 − 1 and for all ak−3 ≤ t ≤ n, therefore by using Lemma 2.3(2) and Lemma 2.4 we have 18 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA H 0(sak−3 · · · snsak−2 · · · snsak−1 · · · sns1 · · · sn−1sn, αn) = ak−1−1M j=1 C−(βj+αn). Proceeding recursively we have H 0(wksn, αn) = ak−1−1M j=1 C−(βj+αn). Hence the proof of the lemma follows. (cid:3) 5. cohomology module H 1 of the relative tangent bundle In this section, we describe the weights of H 1 of the relative tangent bundle. Let n ≥ a1 > a2 > . . . > ak−1 > ak = 1 be a decreasing sequence of integers such that k ≥ 3. Fix 3 ≤ r ≤ k. Lemma 5.1. Let vr = snsar · · · sn−2, vr−1 = sar−1 · · · sn−1 and vr−2 = sar−2 · · · sn−1sn. Then we have (1) H 1(vr−1vrsn−1, αn−1) = 0. (2) Let w = vr−2vr−1vrsn−1. H 1(w, αn−1)µ 6= 0 if and only if µ is of the form µ = −(βt + αn) for some ar−1 ≤ t ≤ ar−2 − 1. In such a case, dim(H 1(w, αn−1)µ) = 1. Proof. Proof of (1): Note that H 0(sn−1, αn−1) = C−αn−1 ⊕ Ch(αn−1) ⊕ Cαn−1 (see [5, Corol- lary 2.5]). (5.1.1) Since hαn−1, αn−1i = 2, we have H 1(sn−1, αn−1) = 0. Since hαn−1, αn−2i = −1, we have (5.1.2) (5.1.3) (5.1.4) (5.1.5) (5.1.6) (5.1.7) (5.1.8) H 1(sn−2, H 0(sn−1, αn−1)) = 0. Therefore by using SES we have H 1(sn−2sn−1, αn−1) = 0. Since h−αn−1, αn−2i = 1, by using (5.1.1) we have H 0(sn−2sn−1, αn−1) = Ch(αn−1) ⊕ C−αn−1 ⊕ C−βn−2. Since h−αn−1, αn−3i = 0 and h−βn−2, αn−3i = 1 we have H 1(sn−3, H 0(sn−2sn−1, αn−1)) = 0. Therefore by using SES together with (5.1.2), (5.1.3) we have H 1(sn−3sn−2sn−1, αn−1) = 0. Proceeding in this way we have H 1(sar · · · sn−2sn−1, αn−1) = 0 and H 0(sar · · · sn−2sn−1, αn−1) = Ch(αn−1) ⊕ n−1L j=ar C−βj . Since h−βj, αni > 0 for all ar ≤ j ≤ n − 1, by using (5.1.6) we have H 1(sn, H 0(sar · · · sn−2sn−1, αn−1)) = 0. Therefore by using SES, (5.1.5) and (5.1.7) together we have H 1(vrsn−1, αn−1) = 0. In the proof of Lemma 4.1(1) we notice that H 0(vrsn−1, αn−1) = Ch(αn−1) ⊕ ( n−1L j=ar Cβj ⊕ C−(βj+αn) ⊕ C−(βj+2αn)). RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 19 Thus we have H 1(sn−1, H 0(vrsn−1, αn−1)) = 0 (see lines from (4.1.1) to (4.1.2)). Proceeding by similar arguments and using (5.1.8), (5.1.9) we have H 1(vr−1vrsn−1, αn−1) = 0. Proof of (2) From the Lemma 4.1(1) we have (5.1.9) ar−1−1L i=ar (C−(βi+αn)⊕C−(βi+2αn)⊕C−(βi+2αn+βn−1)⊕· · ·⊕C−(βi+2αn+βar−1)) H 0(vr−1vrsn−1, αn−1) = n−2L i=ar−1 Notice that for each ar ≤ i ≤ ar−1 − 1, C−(βi+αn) ⊕ C−(βi+2αn) forms an indecomposable (C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕ · · · ⊕ C−(βi+2αn+βi+1)) ⊕ C−(αn−1+2αn). two dimensional Bαn-module. Since h−(βi + 2αn), αni = −2, by Lemma 2.4 we have C−(βi+αn) ⊕ C−(βi+2αn) = Vi ⊗ C−ωn, where Vi is the irreducible two dimensional Lαn- module. By Lemma 2.3(4) we have H j( Lαn/ Bαn, C−(βi+αn) ⊕ C−(βi+2αn)) = 0 for all j ≥ 0 and for all ar ≤ i ≤ ar−1 − 1. Since h−(βi + 2αn + βt), αni = 0 for each ar ≤ i ≤ ar−1 − 1, and ar−1 ≤ t ≤ n − 1, we have H 1( Lαn/ Bαn, C−(βi+2αn+βt)) = 0 for all ar ≤ i ≤ ar−1 − 1 and ar−1 ≤ t ≤ n − 1. Moreover, we have h−(βi + 2αn + βi+1), αni = 0 for each ar−1 ≤ i ≤ n − 2 and h−(βi + 2αn), αni = −2 for each ar−1 ≤ i ≤ n − 1. Therefore we have H 1( Lαn/ Bαn, C−(βi+2αn+βi+1)) = 0 for all ar−1 ≤ i ≤ n − 2. By Lemma 2.3(3) we have H 1( Lαn/ Bαn, C−(βi+2αn)) = H 0( Lαn/ Bαn, sn−1 · −(βi + 2αn)) = C−(βi+αn) for all ar−1 ≤ i ≤ n − 1. From the above discussion, we have H 1(sn, H 0(vr−1vrsn−1, αn−1)) = n−1L i=ar−1 C−(βi+αn). By (1) and using SES we have H 1(sn, H 0(vr−1vrsn−1, αn−1)) = H 1(snvr−1vrsn−1, αn−1). From (5.2.1) and (5.2.2) we have H 1(snvr−1vrsn−1, αn−1) = n−1M i=ar−1 C−(βi+αn). By Lemma 3.1 we have H 1(sn−1, H 0(snvr−1vrsn−1, αn−1)) = 0. Let v = vr−1vrsn−1. Therefore by using SES and (5.2.4) we have H 1(sn−1snv, αn−1) = H 0(sn−1, H 1(snv, αn−1)). (5.2.1) (5.2.2) (5.2.3) (5.2.4) 20 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Since h−(βn−1 + αn), αn−1i = −1 and h−(βi + αn), αn−1i = 0 for all ar−1 ≤ i ≤ n − 2, by using (5.2.3) we have H 1(sn−1snv, αn−1) = n−2L i=ar−1 C−(βi+αn). Proceeding recursively we have H 1(vr−2vr−1vrsn−1, αn−1) = H 1(sar−2 · · · sn−1snv, αn−1) = ar−2−1L i=ar−1 C−(βi+αn). (cid:3) Recall that wr = [a1, n][a2, n] · · · [ar−1, n][ar, n − 1], where 1 ≤ r ≤ k and n ≥ a1 > a2 > . . . > ak−1 > ak = 1. Lemma 5.2. (1) H 1(w1, αn−1) = 0. (2) If a2 6= n − 1, then H 1(w2, αn−1) = 0. (3) Let 3 ≤ r ≤ k. Then, H 1(wr, αn−1)µ 6= 0 if and only if µ =-(βj + αn) for some j such that ar−1 ≤ j ≤ ar−2 − 1. In such case dim(H 1(wr, αn−1))µ = 1. Proof. Proof of (1): Follows from proof of Lemma 3.1 and using SES. Proof of (2): By proof of (1), we have H 1(sa2 · · · sn−1, αn−1) = 0. Since a2 6= n − 1, H 0(sa2 · · · sn−1, αn−1) = Ch(αn−1) ⊕ ( n−1L j=a2 C−(βj )). (5.2.1) Since h−βj, αni ≥ 1 for all a2 ≤ j ≤ n − 1, we have H 1(sn, H 0(sa2 · · · sn−1, αn−1)) = 0. By SES, (5.2.1), (5.2.2) we have H 1(snsa2 · · · sn−1, αn−1) = 0. By using Lemma 3.1 and SES respeatedly, we see that H 1(w2, αn−1) = H 1(sa1 · · · snsa2 · · · sn−1, αn−1) = 0. Proof of (3): Let vr = snsar · · · sn−2, vr−1 = sar−1 · · · sn−1 and vr−2 = sar−2 · · · sn−1sn. Then by Lemma (5.2.2) 5.1(2) we have H 1(vr−2vr−1vrsn−1, αn−1) = ar−2−1M j=ar−1 C−(βj+αn). By Lemma 4.2(1) if (H 0(vr−2vr−1vrsn−1, αn−1))µ 6= 0 then hµ, αni ≥ 0. Therefore we have H 1(sn, H 0(vr−2vr−1vrsn−1, αn−1)) = 0. By using SES and (5.3.1) we have H 1(snvr−2vr−1vrsn−1, αn−1) = H 0(sn, H 1(vr−2vr−1vr, αn−1)). Since h−(βj + αn), αni = 0 for all ar−1 ≤ j ≤ ar−2 − 1, we have from (5.3.2) H 1(snvr−2vr−1vrsn−1, αn−1) = ar−2−1L j=ar−1 C−(βj+αn). By Lemma 3.1, we have H 1(sn−1, H 0(snvr−2vr−1vrsn−1, αn−1)) = 0. Therefore by using SES together with (5.3.4) we have H 1(sn−1snvr−2vr−1vr, αn−1) = H 0(sn−1, H 1(snvr−2vr−1vrsn−1, αn−1)). (5.3.1) (5.3.2) (5.3.4) (5.3.5) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 21 Since h−(βj + αn), αn−1i = 0 for all ar−1 ≤ j ≤ ar−2 − 1, using (5.3.5) we have H 1(sn−1snvr−2vr−1vrsn−1, αn−1) = Proceeding recursively we have ar−2−1L j=ar−1 C−(βj+αn). H 1(sar−3 · · · sn−1snvr−2vr−1vrsn−1, αn−1) = ar−2−1M j=ar−1 C−(βj+αn). Since h−(βj + αn), αti for all ar−1 ≤ j ≤ ar−2 − 1 and ar−4 ≤ t ≤ n, using similar arguments as above we have H 1(wr, αn−1) = ar−2−1M j=ar−1 C−(βj+αn). Corollary 5.3. Let 3 ≤ r ≤ k. If H 1(wr, αn−1)µ 6= 0, then H 0(wr−2snsar−1 · · · sn+2−r, αn+2−r)µ 6= 0. Proof. Corollary follows from Lemma 4.3 and Lemma 5.2(2). Corollary 5.4. Let 3 ≤ r ≤ k. If H 1(wr, αn−1)µ 6= 0, then H 0(wr−1sn, αn)µ 6= 0. Proof. Corollary follows from Lemma 4.4 and Lemma 5.2(2). (cid:3) (cid:3) (cid:3) Lemma 5.5. Let u1 = wksn[ak, n − 1]. Then H 1(u1, αn−1)µ 6= 0 if and only if µ is of the form, µ = −(βj + αn) for some 1 ≤ j ≤ ak−1 − 1. Proof. Let vk+1 = sns1 · · · sn−2, vk = s1 · · · sn−1 and vk−1 = sak−1 · · · sn−1sn. Step 1: H 1(vk−1vkvk+1sn−1, αn−1) = ak−1−1M j=1 C−(βj +αn). Proof of Step 1: By Lemma 4.1(2), we have H 0(s1s2 · · · sns1 · · · sn−1, αn−1) = (C−(βi+2αn) ⊕ C−(βi+2αn+βn−1) ⊕· · ·⊕ C−(βi+2αn+βi+1)) ⊕C−(βn−1+2αn). By Lemma 5.2(1), H 1(s2 · · · sns1 · · · sn−1, αn−1) = 0. Therefore by SES, we have H 1(vkvk+1sn−1, αn−1) = H 1(s1, H 0(s2 · · · sns1 · · · sn−1, αn−1)) = 0 (by Lemma 3.1). (5.5.0) n−2L i=1 By SES and (5.5.0) we have H 1(snvkvk+1sn−1, αn−1) = H 1(sn, H 0(vkvk+1sn−1, αn−1)) = n−1L i=1 C−(βi+αn). By Lemma 3.1, we have H 1(sn−1, H 0(snvkvk+1sn−1, αn−1)) = 0. Therefore by SES, we have H 1(sn−1snvkvk+1sn−1, αn−1) = H 0(sn−1, H 1(snvkvk+1sn−1, αn−1)). (5.5.1) (5.5.2) (5.2.3) (5.2.4) 22 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Since h−(βn−1 + αn), αn−1i = −1 and h−(βi + αn), αn−1i = 0 for all 1 ≤ i ≤ n − 2, we have H 1(sn−1snvkvk+1sn−1, αn−1) = C−(βi+αn). Proceeding in this way recurssively, we i=1 n−2L ak−1−1L i=1 see that H 1(vk−1vkvk+1sn−1, αn−1) = C−(βi+αn). Hence Step 1 follows. By Lemma 4.2, we have H 1(sn, H 0(vk−1vkvk+1sn−1, αn−1)) = 0. Therefore by using SES and (5.5.1) we have H 1(snvk−1vkvk+1sn−1, αn−1) = H 0(sn, H 1(vk−1vkvk+1sn−1, αn−1)). Since h−(βj + αn), αni = 0 for all 1 ≤ j ≤ ak−1 − 1, by using (5.5.2) we have H 1(snvk−1vkvk+1sn−1, αn−1) = ak−1−1L j=1 C−(βj +αn). By Lemma 3.1, we have H 1(sn−1, H 0(snvk−1vkvk+1sn−1, αn−1)) = 0. Therefore by using SES and (5.2.3) we have H 1(sn−1snvk−1vkvk+1sn−1, αn−1) = H 0(sn−1, H 1(snvk−1vkvk+1sn−1, αn−1)). Since h−(βj + αn), αn−1i = 0 for all 1 ≤ j ≤ ak−1 − 1, by using (6.2.4) we have H 1(sn−1snvk−1vkvk+1sn−1, αn−1) = Proceeding recursively we have ak−1−1L j=1 C−(βj +αn). H 1(sak−2 · · · sn−1snvk−1vkvk+1sn−1, αn−1) = ak−1−1M j=1 C−(βj +αn). Using the similar arguments as above we have H 1(u1, αn−1) = ak−1−1M j=1 C−(βj +αn). Corollary 5.6. If H 1(u1, αn−1)µ 6= 0, then H 0(wk−1sns1s2 · · · sn+1−k, αn+1−k)µ 6= 0. Proof. Corollary follows from Lemma 4.5 and Lemma 5.5. Corollary 5.7. If H 1(u1, αn−1)µ 6= 0, then H 0(wksn, αn)µ 6= 0. Proof. Corollary follows from Lemma 4.6 and Lemma 5.5. (cid:3) (cid:3) (cid:3) Let 3 ≤ r ≤ k. Let Mr := {µ ∈ X(T ) : H 1(wr, αn−1)µ 6= 0} and M0 := {µ ∈ X(T ) : H 1(u1, αn−1)µ 6= 0}. Then we have Corollary 5.8. (1) Mr ∩ Mr′ = ∅ whenever r 6= r′. (2) M0 ∩ Mr = ∅ for every 1 ≤ r ≤ k. Proof. Proof of (1) and (2) follow from Lemma 5.2 and Lemma 5.5. (cid:3) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 23 Lemma 5.9. Let c = s1s2 · · · sn. Let Tr = cr−1s1s2 · · · sn−1, for all 2 ≤ r ≤ n. Then, Tr(αj) < 0 for all n + 1 − r ≤ j ≤ n − 1. Proof. Assume r = 2. Then we see that Tr(αn−1) = −α0. We assume that for 2 < l < n, we have Tl(αj) < 0 for n + 1 − l ≤ j ≤ n − 1. Note that Tl+1 = Tlsns1 · · · sn−1. Then for all n − l ≤ i ≤ n − 3, we have s1 · · · sn−1(αi) = αi+1. Since n − (i + 1) ≥ 2 and n + 1 − l ≤ i + 1 ≤ n − 1, we have Tl+1(αi) = Tl(αi+1) < 0 (by assumption). Since s1 · · · sn−1(αn−2) = αn−1, we have Tl+1(αn−2) = Tl(sn(−αn−1)). Since sn(αn−1) = αn−1 +2αn, we have Tlsn(αn−1) = Tl(αn−1 +2αn). As s1 · · · sn−1(αn−1 + 2αn) = β1 + 2αn, we have Tl(αn−1 + 2αn) = Tl−1sn(β1 + 2αn). Since sns1 · · · sn−1sn(β1 + 2αn) = −α1, we have Tl−1sn(β1 + 2αn) = Tl−2(−α1). Since sns1 · · · sn−1(−α1) = −α2, we have Tl−2(−α1) = Tl−3(−α2). Therefore by recursively we have Tl−3(−α2) = −αl−1. Hence Tl+1(αn−2) = −αl−1 < 0. Also it is clear that Tl+1(αn−1) < 0. Therefore we have Tl+1(αj) < 0 for all n − l ≤ j ≤ n − 1. Hence the result follows. (cid:3) Lemma 5.10. Let u, v ∈ W, let v := ( nQ j=1 sj)l−1s1s2 · · · sn−1 for some positive integer l ≤ n, such that l(uv) = l(u) + l(v). Let w = uv. If l ≥ 3, then H i(w, αn−1) = 0 for all i ≥ 0. Proof. We note that by SES, we have H 0(w, αn−1) = H 0(u, H 0(v, αn−1)). We show that H 0(v, αn−1) = 0. By Lemma 5.9 we have for each 1 ≤ r ≤ n − 1, cr−1s1 · · · sn−1(αj) < 0 for all n + 1 − r ≤ j ≤ n − 1. In particular, we have l(vsn−2) = l(v) − 1. Therefore, by Lemma 2.2(4) and using SES, we have H 0(v, αn−1) = H 0(vsn−2, H 0(sn−2, αn−1)) = 0. (5.10.1) Next we show that H 1(w, αn−1) = H 0(u, H 1(v, αn−1)). We will prove by induction l(u). If l(u) = 0 then it follows trivially. Next suppose that l(u) > 1. Then there exists a simple root γ such that l(sγu) = l(u) − 1. By using SES and (5.10.1) we have H 0(sγuv, αn−1) = 0. Again, by induction hypotheses H 1(sγuv, αn−1) = H 0(sγu, H 1(v, αn−1)). Therefore by SES, we have H 1(w, αn−1) = H 0(sγ, H 1(sγuv, αn−1)) = H 0(sγ, H 0(sγu, H 1(v, αn−1))) = H 0(u, H 1(v, αn−1)). H 1(v, αn−1) = 0, follows from the fact that l(vsn−2) = l(v) − 1 and Lemma 2.2(4). Thus, we have H 1(w, αn−1) = 0. Therefore by [14, Corollary 6.4, p.780] we have H i(w, αn−1) = 0 for all i ≥ 0. (cid:3) 6. cohomology modules of the tangent bundle of Z(w, i) Let w ∈ W and let w = si1si2 · · · sir be a reduced expression for w and let i = (i1, i2, . . . , ir). Let τ = si1si2 · · · sir−1 and i′ = (i1, i2, . . . , ir−1). Recall the following long exact sequence of B-modules from [5] (see [5, Proposition 3.1, p.673]): 24 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA 0 → H 0(w, αir) → H 0(Z(w, i), T(w,i)) → H 0(Z(τ, i′), T(τ,i′)) → H 1(w, αir) → H 1(Z(w, i), T(w,i)) → H 1(Z(τ, i′), T(τ,i′)) → H 2(w, αir) → H 2(Z(w, i), T(w,i)) → H 2(Z(τ, i′), T(τ,i′)) → H 3(w, αir) → · · · By [14, Corollary 6.4, p.780], we have H j(w, αir) = 0 for every j ≥ 2. Thus we have the following exact sequence of B-modules: 0 → H 0(w, αir) → H 0(Z(w, i), T(w,i)) → H 0(Z(τ, i′), T(τ,i′)) → H 1(w, αir) → H 1(Z(w, i), T(w,i)) → H 1(Z(τ, i′), T(τ,i′)) → 0 Now onwards we call this exact sequence by LES. Let w0 = sj1sj2 · · · sjN be a reduced expression of w0 such that i = (j1, j2, . . . , jr) and let and j = (j1, j2, . . . , jN ). Lemma 6.1. The natural homomorphism f : H 1(Z(w0, j), T(w0,j)) −→ H 1(Z(w, i), T(w,i)) of B- modules is surjective. Proof. (see [4, Lemma 7.1, p.459]). (cid:3) Lemma 6.2. Let J = S \ {αn−1}. Let v ∈ WJ and u ∈ W be such that l(uv) = l(u) + l(v). Let u = si1 · · · sir and v = sir+1 · · · sit be reduced expressions of u and v respectively. Let i = (i1, i2, . . . , ir) and j = (i1, i2, . . . , ir, ir+1, . . . , it). Then we have (1) The natural homomorphism H 0(Z(uv, j), T(uv,j)) −→ H 0(Z(u, i), T(u,i)) of B-modules is surjective. (2) The natural homomorphism H 1(Z(uv, j), T(uv,j)) −→ H 1(Z(u, i), T(u,i)) of B-modules is an isomorphism. Proof. Let r + 1 ≤ l ≤ t. Let vl = usir+1 · · · sil and il = (i, ir+1, . . . , il). Proof of (1): By Lemma 3.3 we have H 1(vt, αit) = 0. Therefore, using LES the natural homomorphism is surjective. H 0(Z(vt, it), T(vt,it)) −→ H 0(Z(vt−1, it−1), T(vt−1,it−1)) By induction on l(v), the natural homomorphism H 0(Z(vt−1, it−1), T(vt−1,it−1)) −→ H 0(Z(u, i), T(u,i)) is surjective. Hence we conclude that the natural homomorphism H 0(Z(uv, j), T(uv,j)) −→ H 0(Z(u, i), T(u,i)) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 25 is surjective. Proof of (2): We will prove by induction on l(v). By LES, we have the following exact sequence of B-modules: 0 −→ H 0(uv, αit) −→ H 0(Z(uv, j), T(uv,j )) −→ H 0(Z(vt−1, it−1), T(vt−1,it−1)) −→ H 1(uv, αit) −→ H 1(Z(uv, j), T(uv,j)) −→ H 1(Z(vt−1, it−1), T(vt−1,it−1)) −→ 0. By induction on l(v), the natural homomorphism H 1(Z(vt−1, it−1), T(vt−1,it−1)) −→ H 1(Z(u, i), T(u,i)) is an isomorphism. By Lemma 3.3, H 1(uv, αit) = 0. Therefore, by the above exact sequence we have H 1(Z(uv, j), T(uv,j)) −→ H 1(Z(vt−1, it−1), T(vt−1,it−1)) is an isomorphism. Hence we con- clude that the homomorphism H 1(Z(uv, j), T(uv,j)) −→ H 1(Z(u, i), T(u,i)) of B-modules is an isomorphism. (cid:3) Recall that by Lemma 2.6(1) and Lemma 2.8(2) we have w0 = ( k−1Q l1=1 [al1, n])([ak, n]n+1−k)( k−1Q l2=1 [ak, al2 − 1]) is a reduced expression for w0. Let i be the tuple corresponding to this reduced of w0. Let u1 = wksn[ak, n − 1] and i1 be the tuple corresponding to the reduced expression kQ [al1, n]([ak, n − 1]). Note that ak = 1. With this notation, we have l1=1 Lemma 6.3. (1) The natural homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. (2) The natural homomorphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. Proof. For 1 ≤ n − k, let uj = wksn[ak, n]j−1s1s2 · · · sn−1 and ij be the tuple corresponding to the reduced expression uj = ( kQ l1=1 [al1, n])([ak, n])j−1[ak, n − 1] (see Lemma 2.8(2)). Note that w0 = un−ksn( k−1Q l2=1 [ak, al2 − 1]). Case 1: a1 6= n. In this case, we have sn( k−1Q l2=1 [ak, al2−1]) ∈ WJ , where J = S \ {αn−1}. Then by Lemma 6.2, the natural homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(un−k, in−k), T(un−k,in−k)) (6.3.1) 26 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA is surjectie and the natural homomorphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(un−k, in−k), T(un−k,in−k)) of B-modules is an isomorphism. (6.3.2) j be j, n − 1). Hence by LES , we observe that j = ujsn−1 and let i′ If j ≥ 2, then by Lemma 5.10, we have H 1(uj, αn−1) = 0. Let u′ j such that ij = (i′ the partial subsequence of u′ the natural homomorphism H 0(Z(uj, ij), T(uj ,ij)) −→ H 0(Z(u′ is surjective and H 1(Z(uj, ij), T(uj ,ij)) −→ H 1(Z(u′ is an isomorphism. Therefore by Lemma 6.2 we have the natural homomorphism H 0(Z(u′ j )) −→ H 0(Z(uj−1, ij−1), T(uj−1,ij−1)) j), T(u′ j), T(u′ j, i′ j, i′ j)) j ,i′ j)) j ,i′ (6.3.3) (6.3.4) (6.3.5) (6.3.6) j ,i′ j, i′ j), T(u′ is surjective and H 1(Z(u′ j), T(u′ j, i′ j )) −→ H 1(Z(uj−1, ij−1), T(uj−1,ij−1)) j ,i′ is an isomorphism. Therefore, by combining (6.3.3), (6.3.5) together and (6.3.4), (6.3.6) together we see that the homomorphism H 0(Z(uj, ij), T(uj ,ij)) −→ H 0(Z(uj−1, ij−1), T(uj−1,ij−1)) is surjective and H 1(Z(uj, ij), T(uj ,ij)) −→ H 1(Z(uj−1, ij−1), T(uj−1,ij−1)) is an isomorphism. Proceeding recursively we get that the homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(u1, i1), T(u1,i1)) of B-modules is surjective and homomorphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. Further since u−1 1 (α0) < 0, by [5, Lemma 6.2, p.667], we have H 0(Z(u1, i1), T(u1,i1))−α0 6= 0. By [5, Theorem 7.1], H 0(Z(w0, i), T(w0,i)) is a parabolic subalgebra of g and hence there is a unique B-stable line in H 0(Z(w0, i), T(w0,i)), namely g−α0. Thereore we conclude that the natural homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. Case 2 : a1 = n Then by Lemma 6.2, the natural morphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(un+1−k, in+1−k), T(un+1−k,in+1−k)) is surjectie and the natural homomorphism RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 27 H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(un+1−k, in+1−k), T(un+1−k,in+1−k)) of B-modules is an isomorphism. If j ≥ 2, then by Lemma 5.10, we have H 1(uj, αn−1) = 0. Hence by LES, for each 2 ≤ j ≤ n + 1 − k, we observe that the natural homomorphism H 0(Z(uj, ij), T(uj ,ij)) −→ H 0(Z(uj−1, ij−1), T(uj−1,ij−1)) is surjective and H 1(Z(uj, ij), T(uj ,ij)) −→ H 1(Z(uj−1, ij−1), T(uj−1,ij−1)) is an isomorphism. Therefore, the homomorphism of B-modules is surjective and the homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(u1, i1), T(u1,i1)) H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. Further since u−1 1 (α0) < 0, by [5, Lemma 6.2, p.667], we have H 0(Z(u1, i1), T(u1,i1))−α0 6= 0. By [5, Theorem 7.1], H 0(Z(w0, i), T(w0,i)) is a parabolic subalgebra of g and hence there is a unique B-stable line in H 0(Z(w0, i), T(w0,i)), namely g−α0. Thereore we conclude that the natural homomorphism H 0(Z(w0, i), T(w0,i)) −→ H 0(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. The following is a useful Corollary. Corollary 6.4. If µ ∈ X(T ) \ {0}, then, we have dim(H 0(Z(u1, i1), T(u1,i1))µ) ≤ 1. (cid:3) Proof. By [5, Theorem 7.1], H 0(Z(w0, i), T(w0,i)) is a parabolic subalgebra of g. By Lemma 6.3(1) we have H 0(Z(w0, i), T(w0,i)) ≃ H 0(Z(u1, i1), T(u1,i1)). (as B-modules). Hence for any µ ∈ X(T ) \ {0}, we have dim(H 0(Z(u1, i1), T(u1,i1))µ) ≤ 1. Let u′ 1 = wksn[ak, n − 2] and let i′ 1 be the tuple corresponding to the reduced expression (cid:3) [al1, n][ak, n−2]. Let 3 ≤ r ≤ k, and let j = (a1, . . . , n; a2, . . . , n; . . . ; ar−1, . . . , n; ar, . . . , n− r kQ l1=1 1) and j ′ r = (a1, . . . , n; a2, . . . , n; ar, . . . , n − 2). We now prove 28 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Lemma 6.5. Let µ ∈ X(T ) \ {0}. (1) If H 1(u1, αn−1)µ = 0, then dim(H 0(Z(u′ (2) If H 1(u1, αn−1)µ 6= 0, then dim(H 0(Z(u′ 1, i′ 1, i′ 1), T(u′ 1), T(u′ 1,i′ 1,i′ 1))µ)≤ 1. 1))µ)= 2, and the natural homo- morphism is surjective. H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1))µ −→ H 1(u1, αn−1)µ Proof. By LES, we have the following long exact sequence of B-modules: 0 −→ H 0(u1, αn−1) −→ H 0(Z(u1, i1), T(u1,i1)) −→ 1)) −→ H 1(u1, αn−1) −→ · · · 1,i′ 1, i′ 1), T(u′ H 0(Z(u′ Proof of (1): Assume that H 1(u1, αn−1)µ = 0 and µ ∈ X(T ) \ {0}, then by the 1,i1))µ −→ 1))µ is surjective. By Corollary 6.4, we have dim(H 0(Z(u1, i1), T(u1,i1))µ)≤ above exact sequence the natural homoorphism (of T -modules) H 0(Z(u1, i1), T(u′ H 0(Z(u′ 1. 1), T(u′ (6.5.1) 1, i′ 1,i′ Proof of (2): Assume that H 1(u1, αn−1)µ 6= 0. Then by Lemma 5.5, µ is of the form −(βj +αn) for some 1 ≤ j ≤ ak−1 −1 and dim(H 1(u1, αn−1)µ)=1. Hence by using Corollary 6.4, we see that if H 1(u1, αn−1)µ 6= 0, then dim(H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1))µ) ≤ 2. (6.5.2) Let j ′′ = (j k k , n) be the tuple corresponding to the reduced expression wksn = Then by Lemma 6.2 the natural homomorphism H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1)) −→ H 0(Z(wksn, j ′′ ), T(wksn,j ′′ )) k k kQ l1=1 [al1, n]. is surjective. By (6.5.2), we have dim(H 0(Z(wksn, j ′′ Again by the Lemma 6.2 the natural homomorphism ), T(wksn,j ′′ ))µ)≤ 2. k (6.5.3) k H 0(Z(wksn, j ′′ ), T(wksn,j ′′ k k )) −→ H 0(Z(wk, j ), T(wk,j )) k k (6.5.4) is surjective. Recall that u′ of B-modules: t = uksn−1. By LES we have the following long exact sequence 0 −→ H 0(wk, αn−1) −→ H 0(Z(wk, j H 0(Z(u′ k), T(u′ k, i′ k)) −→ H 1(wk, αn−1) −→ · · · ), T(wk,j )) −→ k k k,i′ Since H 1(u1, αn−1)µ 6= 0, by Corollary 5.8 we have H 1(wk, αn−1)µ = 0. Therefore we have an exact sequence 0 −→ H 0(wk, αn−1)µ −→ H 0(Z(wk, j ), T(wk,j k k ))µ −→ H 0(Z(τk, j ′ ), T(τk,j ′ k k ))µ −→ 0. Let σk = wk−1sns1 · · · sn+1−k and j ∗ , n, 1, 2, . . . , n+1−k) be the tuple correspond- ing to this reduced expression of σk. Then by using Lemma 6.2, the natural homomorphism = (j k−1 k RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 29 H 0(Z(τk, j ′ ), T(τk ,j ′ k k )) −→ H 0(Z(σk, j ∗ ), T(σk ,j ∗ )) k k is surjective. Therefore the natural map ))µ −→ H 0(Z(σk, j ∗ H 0(Z(wk, j ), T(wk,j k k ), T(σk ,j ∗ ))µ k k (6.5.5) (6.5.6) is surjective. Since H 1(u1, αn−1)µ 6= 0, by Corollary 5.6 we have H 0(σk, αn+1−k)µ 6= 0. Therefore, ))µ 6= 0. ))µ 6= 0. Thus from (6.5.6) we have H 0(Z(wk, j ), T(wk,j H 0(Z(σk, j ∗ Then we have an exact sequence of T -modules ), T(σk,j ∗ k k k k 0 −→ H 0(wksn, αn)µ −→ H 0(Z(wksn, j ′′ ), T(wksn,j ′′ k k ))µ −→ H 0(Z(wk, j ), T(wk,j k k ))µ −→ 0 Since H 1(u1, αn−1)µ 6= 0, by Corollary 5.7 we have H 0(wksn, αn)µ 6= 0. Therefore dim(H 0(Z(wksn, j ′′ dim(H 0(Z(wksn, j ′′ 2. k ), T(wksn,j ′′ k ), T(wksn,j ′′ k ))µ)≤ 2. Hence we have dim(H 0(Z(wksn, j ′′ ))µ)≥ 2. On the other hand by (6.5.3) we have ))µ)= ), T(wksn,j ′′ k k k Since the natural homomorphism H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1)) −→ H 0(Z(wksn, j ′′ ), T(wksn,j ′′ )) k k is surjective, we have dim(H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1)µ))= 2. It is clear from (6.5.1),(6.5.7), and Corollary 6.4, that H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1))µ −→ H 1(u1, αn−1)µ is surjective. (6.5.7) (cid:3) Corollary 6.6. The natural homomorphism H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1)) −→ H 1(u1, αn−1) is surjective. Proof. Note that by Lemma 5.5, if H 1(u1, αn−1)µ 6= 0 then µ ∈ X(T ) \ {0}. Now the proof follows from Lemma 6.5. (cid:3) Lemma 6.7. (1) If H 1(wm, αn−1)µ = 0 for all r ≤ m ≤ k and H 1(u1, αn−1)µ = 0, then (2) Let 3 ≤ r ≤ k. If H 1(wr, αn−1)µ 6= 0, then dim(H 0(Z(τr, j ′ ), T(τr ,j ′ r r ))µ) = 2, and dim(H 0(Z(τr, j ′ ), T(τr ,j ′ ))µ) ≤ 1. r r the natural homomorphism is surjective. H 0(Z(τr, j ′ ), T(τr ,j ′ r r ))µ −→ H 1(wr, αn−1)µ Proof. Proof of (1): If H 1(u1, αn−1)µ = 0, then by Lemma 6.5, we have dim(H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1))µ)≤ 1. By Lemma 6.2, the natural homomorphism H 0(Z(u′ 1, i′ 1), T(u′ 1,i′ 1))µ −→ H 0(Z(wk, j ), T(wk,j ))µ k k 30 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA is surjective. natural homomorphism If H 1(wm, αn−1)µ = 0 for all r ≤ m ≤ k, by using LES, we see that the H 0(Z(wk, j ), T(wk,j k k ))µ −→ H 0(Z(τr, j ′ is surjective. Therefore, we have dim(H 0(Z(τr, j ′ ), T(τr ,j ′ r r ))µ r ), T(τr ,j ′ r ))µ)≤ 1. Proof of (2): Assume that H 1(wr, αn−1)µ 6= 0. Then by Corollary 5.8, we have H 1(wm, αn−1)µ = 0 for all r + 1 ≤ m ≤ k and H 1(u1, αn−1)µ = 0. Then by (1), we have By Lemma 6.2, the natural homomorphism dim(H 0(Z(τr+1, j ′ ), T(τr+1,j ′ r+1 r+1 ))µ)≤ 1. H 0(Z(τr+1, j ′ ), T(τr+1,j ′ r+1 r+1 ))µ −→ H 0(Z(wr, j ), T(wr ,j ))µ r r is surjective. Hence, we have dim(H 0(Z(wr, j ), T(wr ,j ))µ)≤ 1. (6.7.1) r r By LES we have the following long exact sequence of B-modules: 0 −→ H 0(wr, αn−1) −→ H 0(Z(wr, j H 0(Z(τr, j ′ ), T(τr ,j ′ r )) −→ H 1(wr, αn−1) −→ · · · ), T(wr,j )) −→ r r r ))µ)≤ 1 and dim(H 1(wr, αn−1)µ)= 1, we see that , n) be the tuple corresponding to the reduced expression wr−1sn = (6.7.2) Since dim(H 0(Z(wr, j dim(H 0(Z(τr, j ′ Let j ′′ = (j r r−1 r−1 ), T(wr ,j r r ))µ)≤ 2. r−1Q l1=1 [al1, n]. Then by Lemma 6.2 the natural homomorphism H 0(Z(τr, j ′ ), T(τr ,j ′ r r )) −→ H 0(Z(wr−1sn, j ′′ ), T(wr−1sn,j ′′ )) r−1 r−1 is surjective. By (6.7.2), we have dim(H 0(Z(wr−1sn, j ′′ r−1 Again by the Lemma 6.2 the natural homomorphism ), T(wr−1sn,j ′′ )µ))≤ 2. r−1 (6.7.3) H 0(Z(wr−1sn, j ′′ ), T(wr−1sn,j ′′ r−1 r−1 )) −→ H 0(Z(wr−1, j ), T(wr−1,j )) r−1 r−1 (6.7.4) is surjective. By LES we have the following long exact sequence of B-modules: 0 −→ H 0(wr−1, αn−1) −→ H 0(Z(wr−1, j H 0(Z(τr−1, j ′ ), T(τr−1,j ′ r−1 r−1 ), T(wr−1,j )) −→ r−1 r−1 )) −→ H 1(wr−1, αn−1) −→ · · · Since H 1(wr, αn−1)µ 6= 0, then by Corollary 5.8 we have H 1(wr−1, αn−1)µ = 0. Therefore we have an exact sequence 0 −→ H 0(wr−1, αn−1)µ −→ H 0(Z(wr−1, j ), T(wr−1,j ))µ −→ r−1 r−1 H 0(Z(τr−1, j ′ ), T(τr−1,j ′ r−1 r−1 ))µ −→ 0 Let σr−1 = wr−2snsar−1···sn+2−r , and let j ∗ the reduced expression of σr−1. = (j r−2 r−1 , n, ar−1, ar−1 + 1, . . . , n + 2 − r) be (6.7.5) RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 31 Then by using Lemma 6.2 the natural homomorphism H 0(Z(τr−1, j ′ ), T(τr−1,j ′ r−1 r−1 )) −→ H 0(Z(σr−1, j ∗ ), T(σr−1,j ∗ )) r−1 r−1 is surjective. Therefore, by (6.7.5) and (6.7.6) we have H 0(Z(wr−1, j ), T(wr−1,j r−1 r−1 ))µ −→ H 0(Z(σr−1, j ∗ ), T(σr−1,j ∗ r−1 r−1 (6.7.6) (6.7.7) ))µ is surjective. Since H 1(wr, αn−1)µ 6= 0, by Corollary 5.3 we have H 0(σr−1, αn+2−r)µ 6= 0. Therefore, H 0(Z(σr−1, j ∗ ), Tσr−1,j ∗ r−1))µ 6= 0. r−1 Thus from (6.7.7) we have H 0(Z(wr−1, j ), T(wr−1,j r−1 r−1 ))µ 6= 0. Then we have an exact sequence of T -modules 0 −→ H 0(wr−1sn, αn)µ −→ H 0(Z(wr−1sn, j ′′ H 0(Z(wr−1, j ), T(wr−1,j r−1 r−1 ), T(wr−1sn,j ′′ ))µ −→ r−1 r−1 ))µ −→ 0 Since H 1(wr, αn−1)µ 6= 0, by Corollary 5.4 we have H 0(wr−1sn, αn)µ 6= 0. Therefore dim(H 0(Z(wr−1sn, j ′′ Hence, by (6.7.3) we have dim(H 0(Z(wr−1sn, j ′′ Since the natural homomorphism ), T(wr−1sn,j ′′ ), T(wr−1sn,j ′′ )µ))≥ 2. ))µ)= 2. r−1 r−1 r−1 r−1 H 0(Z(τr, j ′ ), T(τr ,j ′ r r )) −→ H 0(Z(wr−1sn, j ′′ ), T(wr−1sn,j ′′ )) r−1 r−1 is surjective, we have dim(H 0(Z(τr, j ′ Therefore by (6.7.1), ), T(τr ,j ′ r r ))µ)= 2. H 0(Z(τr, j ′ ), T(τr ,j ′ r r ))µ −→ H 1(wr, αn−1)µ is surjective. 7. main theorem (cid:3) In this section we prove the main theorem. Recall that G = P SO(2n + 1, C)(n ≥ 3), and let c be a Coxeter element of W. Then there exists a decreasing sequence n ≥ a1 > a2 > · · · > ak = 1 of positive integers such that c = [a1, n][a2, a1 − 1] · · · [ak, ak−1 − 1], where [i, j] for i ≤ j denotes sisi+1 · · · sj. Let i = (i1, i2, . . . , in) be a sequence corresponding to a reduced expression of w0, where ir (1 ≤ r ≤ n) is a sequence of reduced expressions of c (see Lemma 2.8). Theorem 7.1. H j(Z(w0, i), T(w0,i)) = 0 for all j ≥ 1 if and only if a2 6= n − 1. Proof. From [5, Proposition 3.1, p. 673], we have H j(Z(w0, i), T(w0,i)) = 0 for all j ≥ 2. It is enough to prove the following: H 1(Z(w0, i), T(w0,i)) = 0 if and only if c is of the form [a1, n][a2, a1 − 1] · · · [ak, ak−1 − 1] with a2 6= n − 1. Proof of ( =⇒ ): If a2 = n − 1, then a1 = n and c = snsn−1v, where v ∈ WJ and J = S \ {αn−1, αn}. Let u = snsn−1. Then c = uv. Let j = (n, n − 1) be the sequence corresponding to u. Then using LES, we have: 32 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA 0 −→ H 0(u, αn−1) −→ H 0(Z(u, j), T(u,j)) −→ H 0(sn, αn) −→ H 1(u, αn−1) f−→ H 1(Z(u, j), T(u,j)) −→ 0. We see that H 1(snsn−1, αn−1) = Cαn+αn−1 and H 0(sn, αn)αn+αn−1 = 0. Hence f is non zero homomorphism. Hence H 1(Z(u, j), T(u,j))) 6= 0. By Lemma 6.1, the natural homo- morphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(u, j), T(u,j)) is surjective. Hence we have Proof of ( ⇐= ) : Assume that a2 6= n − 1. Recall that wk = [a1, n] · · · [ak−1, n][ak, n − 1], H 1(Z(w0, i), T(w0,i)) 6= 0. u1 = wksn[ak, n − 1]. By Lemma 6.3(2), the natural homomorphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(u1, i1), T(u1,i1)) of B-modules is an isomorphism. By using LES, we have the following exact sequence of B-modules: (7.1.1) 1)) −→ (7.1.2) (7.1.3) · · · −→ H 0(Z(u1, i1), T(u1,i1)) −→ H 0(Z(u′ 1, i′ H 1(u1, αn−1) −→ H 1(Z(u1, i1), T(u1,i1)) −→ H 1(Z(u′ 1), T(u′ 1, i′ 1,i′ By Corollary 6.6, we see that the natural homomorphism h1 : H 0(Z(u′ 1), T(u′ 1,i′ 1)) h1−→ 1)) −→ 0. 1, i′ 1), T(u′ 1,i′ H 1(u1, αn−1) is surjective. Therefore, the natural homomorphism H 1(Z(u1, i1), T(u1,i1)) −→ H 1(Z(u′ is an isomorphism. 1, i′ 1), T(u′ 1,i′ 1)) By Lemma 6.2(2), the natural homomorphism H 1(Z(u′ 1, i′ 1), T(u′ 1)) −→ H 1(Z(wk, j ), T(wk,j )) 1,i′ is an isomorphism. Therefore, by (7.1.1), (7.1.2) and (7.1.3) the natural homomorphism k k H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(wk, j is an isomorphism. Recall that τk = [a1, n][a2, n] · · · [ak−1, n][ak, n − 2]. By using LES, we have the following ), T(wk,j (7.1.4) )) k k exact sequence of B-modules: · · · −→ H 0(Z(wk, j ), T(wk,j k k )) −→ H 0(Z(τk, j ′ ), T(τk ,j ′ k k )) h2−→ RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 33 H 1(wk, αn−1) −→ H 1(Z(wk, j ), T(wk,j k k )) h3−→ H 1(Z(τk, j ′ ), T(τk,j ′ k k )) −→ 0. By Lemma 6.7(2), we see that the map h2 : H 0(Z(τk, j ′ surjective. ), T(τk ,j ′ k k )) −→ H 1(wk, αn−1) is Therefore, the map h3 : H 1(Z(wk, j is an isomorphism. By using Lemma 6.2(2) we see that the natural map ), T(wk,j k k )) −→ H 1(Z(τk, j ′ ), T(τk,j ′ )) k k (7.1.5) H 1(Z(τk, j ′ ), T(τk ,j ′ k k )) −→ H 1(Z(wk−1, j ), T(wk−1,j )) k−1 k−1 (7.1.6) is an isomorphism. By using LES, we have the following exact sequence of B-modules: · · · −→ H 0(Z(wk−1, j ), T(wk−1,j k−1 k−1 )) −→ H 0(Z(τk−1, j ′ ), T(τk−1,j ′ )) → k−1 k−1 H 1(wk−1, αn−1) −→ H 1(Z(wk−1, j ), T(wk−1,j k−1 k−1 )) → H 1(Z(τk−1, j ′ ), T(τk−1,j ′ k−1 k−1 )) −→ 0. By Lemma 6.7(2), we see that the natural map H 0(Z(τk−1, j ′ ), T(τk−1,j ′ k−1 k−1 )) −→ H 1(wk−1, αn−1) is surjective. Therefore, the natural map H 1(Z(wk−1, j ), T(wk−1,j k−1 is an isomorphism. )) −→ H 1(Z(τk−1, j ′ ), T(τk−1,j ′ )) k−1 k−1 (7.1.7) k−1 By using Lemma 6.2(2) and 6.7(2) repeatedly, we see that the natural map H 1(Z(τk−1, j ′ ), T(τk−1,j ′ k−1 k−1 )) −→ H 1(Z(τr, j ′ ), T(τr ,j ′ )) r r (7.1.8) is an isomorphism for all 3 ≤ r ≤ k − 2. Therefore by ( (7.1.4), (7.1.5), (7.1.6), (7.1.7), (7.1.8) we have the natural homomorphism H 1(Z(w0, i), T(w0,i)) −→ H 1(Z(τ3, j ′ ), T(τ3,j ′ )) 3 3 is an isomorphism. Again from Lemma 6.2(2), we see that the natural map (7.1.9) H 1(Z(τ3, j ′ ), T(τ3,j ′ 3 3 )) −→ H 1(Z(w2, j ), T(w2,j )) 2 2 is an isomorphism. Since a2 6= n − 1, by Lemma 5.2(2) we have H 1(w2, αn−1) = 0. Note that by Lemma 5.2(1) we have H 1(w1, αn−1) = 0. By using Lemma 3.3 and using LES, we have H 1(Z(w2, j ), T(w2,j )) = 0. Hence we 2 2 conclude that H 1(Z(w0, i), T(w0,i)) = 0. This completes the proof of the theorem. (cid:3) 34 S.SENTHAMARAI KANNAN AND PINAKINATH SAHA Corollary 7.2. Let c be a Coxeter element such that c is of the form [a1, n][a2, a1 − 1] · · · [ak, ak−1 − 1] with a2 6= n − 1 and ak = 1. Let (w0, i) be a reduced expression of w0 in terms of c as in Theorem 7.1. Then, Z(w0, i) has no deformations. Proof. By Theorem 7.1 and by [5, Proposition 3.1, p.673], we have H i(Z(w0, i), T(w0,i)) = 0 for all i > 0. Hence, by [12, Proposition 6.2.10, p.272], we see that Z(w0, i) has no deformations. (cid:3) Remark 7.3. Theorem 7.1 does not hold for P SO(5, C). Proof. We take c = s1s2. Here a2 6= 1. Further, we have w0 = c2 = s1s2s1s2. Let i = (1, 2, 1, 2). It is easy to see by using SES repeatedly that H 1(s1s2s1, α1) = Cα1+α2 ⊕Cα2. Further, note that H 0(Z(s1s2, (1, 2)), T(s1s2,(1,2)))α1+α2 = 0 (see [5, Proposition 6.3(1), p.688]). Hence by using LES we have H 1(Z(s1s2s1, (1, 2, 1)), T(s1s2s1,(1,2,1))) 6= 0. Therefore by using Lemma 6.1 we have H 1(Z(w0, i), T(w0,i)) 6= 0. (cid:3) Acknowledgements We thank the referee for the useful comments and suggestions. We thank the Infosys Foundation for the partial finance support. References [1] V. Balaji, S. Senthamarai Kannan, K.V. Subrahmanyam, Cohomology of line bundles on Schubert varieties-I, Transformation Groups 9 (2004), no.2, 105-131. [2] R. Bott and H. Samelson, Applications of the theory of Morse to symmetric spaces, Amer. J. Math. 80 (1958), 964-1029. [3] M. Brion, S. Kumar, Frobenius Splitting Methods in Geometry and Representation theory, Progress in Mathematics, Vol. 231, Birkhauser, Boston, Inc., Boston, MA, 2005. [4] B.Narasimha Chary, S.Senthamarai Kannan, Rigidity of Bott-Samelson-Demazure-Hansen variety for P Sp(2n, C), Journal of Lie Theory 27 (2017), no. 2, 435-468. [5] B.Narasimha Chary, S.Senthamarai Kannan, A.J.Parameswaran, Automorphism group of a Bott- Samelson-Demazure-Hansen variety, Transformation Groups 20 (2015), no.3, 665-698. [6] M. Demazure, Desingularisation des varieties de schubert generalisees. Ann. Sci. Ecole Norm. Sup 7(1974), 53-88. [7] M. Demazure, A very simple proof of Bott's theorem, Invent. Math. 33 (1976), 271-272. [8] H.C. Hansen, On cycles on flag manifolds, Math. Scand. 33 (1973), 269-274. [9] J.E. Humphreys, Introduction to Lie algebras and Representation theory, Springer-Verlag, Berlin Heidelberg, New York, 1972. [10] J.E. Humphreys, Linear Algebraic Groups, Springer-Verlag, Berlin Heidelberg, New York, 1975. [11] J.E. Humphreys, Conjugacy classes in semisimple algebraic groups, Math. Surveys Monographs, vol.43, Amer. Math. Soc., 1995. [12] D. Huybrechts, "Complex Geometry: An Introduction", Springer-Verlag, Berlin Heidelberg, New York, 2005. [13] J.C. Jantzen, Representations of Algebraic Groups, (Second Edition ), Mathematical Surveys and Monographs, Vol.107, 2003. [14] S. Senthamarai Kannan, On the automorphism group of a smooth Schubert variety. Algebr. Repres- ent. Theory 19 (2016), no.4, 761-782. [15] S.W. Yang and A.Zelevinsky, Cluster algebras of finite type via Coxeter elements and principal minors, Transformation Groups 13 (2008), no.3-4, 855-895. RIGIDITY OF BOTT-SAMELSON-DEMAZURE-HANSEN VARIETY FOR P SO(2n + 1, C) 35 Chennai Mathematical Institute, Plot H1, SIPCOT IT Park, Siruseri, Kelambakkam, 603103, India. E-mail address: [email protected]. Chennai Mathematical Institute, Plot H1, SIPCOT IT Park, Siruseri, Kelambakkam, 603103, India. E-mail address: [email protected].
0905.2330	3	0905	2010-03-04T16:22:52	On the second gaussian map for curves on a K3 surface	[ "math.AG" ]	By a theorem of Wahl, for canonically embedded curves which are hyperplane sections of K3 surfaces, the first gaussian map is not surjective. In this paper we prove that if C is a general hyperplane section of high genus (greater than 280) of a general polarized K3 surface, then the second gaussian map of C is surjective. The resulting bound for the genus g of a general curve with surjective second gaussian map is decreased to g >152.	math.AG	math	1 ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE ELISABETTA COLOMBO AND PAOLA FREDIANI Abstract. By a theorem of Wahl, for canonically embedded curves which are hyperplane sections of K3 surfaces, the first Gaussian map is not surjective. In this paper we prove that if C is a general hyperplane section of high genus (greater than 280) of a general polarized K3 surface, then the second Gaussian map of C is surjective. The resulting bound for the genus g of a general curve with surjective second Gaussian map is decreased to g > 152. 1. Introduction The first Gaussian map, or Wahl map, for the canonical series has been extensively studied, and it has been shown that for a general curve of genus ≥ 10, different from 11, it is surjective ([4], [18]). Wahl ([21]) has given a deformation theoretic interpretation of the first Gaussian map, showing that if a canonical curve can be extended in projec- tive space as a hyperplane section of a surface which is not a cone, then the first Gaussian map is not surjective. In particular in [21] it is proven that if a curve lies on a K3 surface, the first Gaussian map can't be surjective (see also [3]). The obstruction to the surjectivity of the first Gaussian map for a curve in a K3 surface is given by the extension class of the cotangent sequence 0 → K −1 C → Ω1 X C → KC → 0, which is a non trivial element in the kernel of the dual of the first Gaussian map (see [3]). This paper is concerned with the second Gaussian map, γ2 C : I2(KC ) → H 0(C, 4KC ). In fact also the second Gaussian map encodes some interesting geometry. Our geometrical motivation to study it comes from its relation with the curvature of the moduli space Mg of curves of genus g with the Siegel metric induced by the period map j : Mg → Ag, that we analyzed in [9]. There the curvature was computed using a formula for the associated second fundamental form given in [11]. In particular in [11] it is proven that the second fundamental form lifts the second Gaussian map as stated in an unpublished paper of Green and Griffiths (cf. [12]). 1 Partially supported by PRIN 2007 MIUR: "Moduli, strutture geometriche e loro applicazioni " and PRIN 2006 of MIUR "Geometry on algebraic varieties". AMS Subject classification: 14H10, 14J28. 1 2 E. COLOMBO AND P. FREDIANI In [9], (3.8) we gave a formula for the holomorphic sectional curvature of Mg along a Schiffer variation ξP , for P a point on the curve C, in terms of the holomorphic sectional curvature of Ag and the second Gaussian map. The relation of the second Gaussian map with curvature properties of Mg in Ag suggested that its rank could give information on the geometry of Mg and of some sublocus of it. Another interesting problem is to understand whether the surjectivity of second Gaussian map provides an obstruction to embed a curve in a surface as a hyperplane section. In this paper we address this problem for curves in a K3 surface and we deduce results for the general curve. Using cohomological techniques in the study of γ2 C for a curve in a K3 surface X it is natural to consider the "symmetric square" of the cotangent extension 0 → Ω1 X C ⊗ K −1 C → S2Ω1 X C → K 2 C → 0. This does not give any obstruction to the surjectivity of γ2 C for the general curve in a general K3 surface, while it gives an obstruction if C is any curve in an abelian surface (cf. [10]). In fact in [10] it is shown that if C is a curve in an abelian surface X, then the corank of γ2 C is at least two. In this paper (theorem (3.1)) we prove surjectivity of the second Gaussian map for a general curve C of high genus (for all g > 280) on a general polarized K3 surface. This clearly implies surjectivity for the general curve of in the moduli space of curves of any genus g > 280. In corollary (3.5) we decrease the lower bound for the genus of the general curve with surjective second Gaussian map till 152, using examples given [8]. Note that, for dimensional reasons, surjectivity can be expected for a general curve of genus at least 18, therefore our bound is far from being optimal, nevertheless, it is the first known lower bound. We recall that examples of curves whose second Gaussian map is surjec- tive were already given by in [8] (for curves lying on the product of two curves) and in [1] (for complete intersections). Note that using complete intersections it is not possible to deduce surjectivity for the general curve of any sufficiently high genus, due to restrictions on the genus. Moreover the first of such examples has much higher genus. The examples of curves in a product of two curves are used in this paper to decrease the bound. On the other hand, theorem (3.1) shows that general curves on K3 surfaces of sufficiently high genus behave as general curves in the moduli space, with respect to the second Gaussian map. To prove our theorem we first show that sufficient conditions for the sur- C for C a curve in a K3 surface X are given by the surjectivity X ⊗ OX(2C)) jectivity of γ2 of the second Gaussian map γ2 and the vanishing of H 1(S2Ω1 OX (C) : I2(OX (C)) → H 0(S2Ω1 X ⊗ OX (C)). ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE 3 To prove surjectivity of γ2 OX (C) we adapt the ideas used in [7] to prove surjectivity of first Gaussian map γ1 OX (C). More precisely, observe that a sufficient condition for the surjectivity of γ2 ∆X ⊗ p∗(OX (C)) ⊗ q∗(OX (C))), where ∆X OX (C) is the vanishing of H 1(I 3 is the diagonal in X × X and p, q are the two projections to X. The idea is to consider the blow-up Y of X × X along the diagonal ∆X and to use Kawamata-Viehweg vanishing theorem ([13], [17]) as follows. Let E be the exceptional divisor and denote by π : Y → X × X the natural morphism and by f := p ◦ π, g := q ◦ π. Then H 1(I 3 ∼= H 1(Y, f ∗(OX (C)) ⊗ g∗(OX (C)) ⊗ KY (−4E)), since KY = OY (E). ∆X ⊗p∗(OX (C))⊗q∗(OX (C))) ∼= H 1(Y, f ∗(OX (C))⊗g∗(OX (C))(−3E)) So by Kawamata-Viehweg vanishing theorem, it suffices to prove that the line bundle L := f ∗(OX (C)) ⊗ g∗(OX (C))(−4E) is big and nef. Now notice that if one decomposes OX(C) as ⊗4 i=1Ai, where Ai are line i=1(f ∗(Ai) ⊗ g∗(Ai)(−E)). To obtain that L is bundles on X, then L = ⊗4 big and nef, we ask suitable conditions on the line bundles Ai, and we study the sublinear system of f ∗(Ai) ⊗ g∗(Ai)(−E) given by P(Λ2(H 0(Ai))) (cf. lemma (3.3)). The vanishing of H 1(S2Ω1 X ⊗ OX (C)) relies on a similar argument, but it requires a more refined version of it. In fact, given a decomposition of OX(C) as OX (D) ⊗ OX (D′), we have H 1(S2Ω1 X ⊗ OX(C)) = H 1(X × X, I 2 ∆X ⊗ p∗(OX (D)) ⊗ q∗(OX (D′))), ∆X /I 3 hence its vanishing is implied by the vanishing of H 1(X×X, I 2 q∗(OX (D′))) and of H 2(X × X, I 3 ∆X ⊗ p∗(OX (D)) ⊗ q∗(OX (D′))). ∆X ⊗p∗(OX (D))⊗ So, with the same argument as above, it suffices to show that f ∗(OX (D))⊗ g∗(OX (D′))(−4E) is big and nef. The strategy is now to choose OX (D) = i=1Ai and D′ = D + B with B nef and effective, so we take C ∈ 2D + B. ⊗4 The above decompositions are shown on concrete examples of K3 surfaces X and of curves C in X, which are explicitely constructed via their Picard lattices (cf proposition 3.4). Finally, regardless the examples that we give, note that the conditions of the line bundles Ai as in lemma (3.3) and the decomposition OX (C) = OX(2D + B) force the genus of C to be far from the optimal lower bound. X ⊗ OX (C)) itself, already implies that the curve C must be of genus at least 31, as one can check looking at the restriction of Ω1 X(C) to C and the induced cohomology exact sequence. Anyhow, observe that the vanishing of H 1(S2Ω1 X ⊗ Ω1 2. Preliminaries on Gaussian maps Let Y be a smooth complex projective variety and let ∆Y ⊂ Y × Y be the diagonal. Let L and M be line bundles on Y . For a non-negative integer k, 4 E. COLOMBO AND P. FREDIANI the k-th Gaussian map associated to these data is the restriction to diagonal map (1) L,M : H 0(Y ×Y, I k γk ∆Y ⊗L⊠M ) → H 0(Y, I k ⊗L⊗M ) ∼= H 0(Y, SkΩ1 ∆Y ∆Y Y ⊗L⊗M ). Usually first Gaussian maps are simply referred to as Gaussian maps. The exact sequence (2) 0 → I k+1 ∆Y → I k ∆Y → SkΩ1 Y → 0, (where SkΩ1 Y is identified to its image via the diagonal map), twisted by L ⊠ M , shows that the domain of the k-th Gaussian map is the kernel of the previous one: L,M : ker γk−1 γk L,M → H 0(SkΩ1 Y ⊗ L ⊗ M ). In our applications, we will exclusively deal with Gaussian maps of order two, assuming also that the two line bundles L and M coincide. For the reader's convenience, we spell out these maps. The map γ0 L is the multiplication map of global sections (3) H 0(Y, L) ⊗ H 0(Y, L) → H 0(Y, L2) which obviously vanishes identically on ∧2H 0(L). Consequently, H 0(Y × Y, I∆Y ⊗ L ⊠ L) decomposes as ∧2H 0(L) ⊕ I2(L), where I2(L) is the kernel of S2H 0(Y, L) → H 0(Y, L2). Since γ1 L vanishes on symmetric tensors, one writes (4) γ1 L : ∧2H 0(L) → H 0(Ω1 Y ⊗ L2). Again, H 0(Y × Y, I 2 kernel of (4). Since γ2 usually writes ∆Y ⊗ L ⊠ L) decomposes as the sum of I2(L) and the L vanishes identically on skew-symmetric tensors, one (5) γ2 L : I2(L) → H 0(S2Ω1 Y ⊗ L2) The primary object of this paper will be the second Gaussian map of the canonical line bundle KC on a curve C: C : I2(KC) → H 0(K 4 γ2 C) In our situation Y will be either a K3 surface X, or a smooth irreducible projective curve C on X. 3. Main theorem Theorem 3.1. If X is a general polarized K3 surface of degree 2g − 2 with g > 280 and C is a general hyperplane section of X, then γ2 C is surjective. Let us explain the strategy of the proof of theorem (3.1). We have the following commutative diagram ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE 5 (6) I2(OX (C)) γ2 OX (C) / H 0(S2Ω1 X ⊗ OX(2C)) )TTTTTTTTTTTTTTT p1 r H 0(S2Ω1 XC ⊗ K 2 C ) I2(KC ) / H 0(K 4 C ) γ2 C ujjjjjjjjjjjjjjjj p2 where r and p1 are restriction maps, and p2 comes from the conormal exten- sion. More precisely, consider the exact sequence coming from the conormal extension 0 → Ω1 XC ⊗ KC → S2Ω1 XC ⊗ K 2 C → K 4 C → 0, then we have H 0(S2Ω1 XC ⊗ K 2 C ) p2→ H 0(K 4 C ) → H 1(Ω1 XC ⊗ KC) ∼= H 0(TXC )∗, hence p2 is surjective by the following lemma. Lemma 3.2. If X is a general K3 surface and C a general curve of genus at least 13 in the very ample linear system OX (C) then H 0(TXC) = 0. Proof. By the exact sequence given by restriction of TX to C, H 0(TXC) injects in H 1(TX (−C)), which vanishes by lemma (2.3) of [6]. (cid:3) The theorem will follow if we prove that also the maps γ2 OX (C) and p1 are surjective. In fact it suffices to exhibit examples of pairs (X, C) where X is a K3 and C is a very ample curve in X of any genus g sufficiently high (g ≥ 281) for which γ2 OX (C) and p1 are surjective. To do this we follow the strategy used in [7] to study the first Wahl map. More precisely, from the exact sequence (7) 0 → I 3 → I 2 ∆X ⊗ p∗(OX (C)) ⊗ q∗(OX (C)) → I 2 ∆X /I 3 ∆X ⊗ p∗(OX (C)) ⊗ q∗(OX (C)) → 0 ∆X ⊗ p∗(OX (C)) ⊗ q∗(OX (C)) and taking global sections, we see that γ2 p∗(OX (C)) ⊗ q∗(OX (C))) = 0. OX (C) is surjective if H 1(I 3 ∆X ⊗ The idea used in [7] is to consider the blow-up Y of X × X along the diagonal ∆X and to use Kawamata-Viehweg vanishing theorem. Let E be the exceptional divisor and denote by π : Y → X × X the natural morphism and by f := p ◦ π, g := q ◦ π. Then H 1(I 3 ∆X ⊗p∗(OX (C))⊗q∗(OX (C))) ∼= H 1(Y, f ∗(OX (C))⊗g∗(OX (C))(−3E)) / ) u / 6 E. COLOMBO AND P. FREDIANI ∼= H 1(Y, f ∗(OX (C)) ⊗ g∗(OX (C)) ⊗ KY (−4E)), since KY = OY (E). So by Kawamata-Viehweg vanishing theorem, it suffices to prove that f ∗(OX (C)) ⊗ g∗(OX (C))(−4E) is big and nef. Consider now the map p1 : H 0(S2Ω1 Clearly p1 is surjective if H 1(S2Ω1 X ⊗ OX(2C)) → H 0(S2Ω1 X ⊗ OX (C)) = 0. XC ⊗ K 2 C). Our strategy to prove the surjectivity of p1 is to adapt the above idea for the vanishing of H 1(Y, f ∗(OX (C)) ⊗ g∗(OX (C)) ⊗ (−3E)) in order to show that also H 1(S2Ω1 X ⊗ OX (C)) vanishes. To this end, let H be a very ample divisor and assume that C ∈ 2H + B, ∆X /I 3 ∆X /I 3 where B is nef and effective. Then X ⊗ OX (C)) ∼= H 1(X, q∗(I 2 ∆X ⊗ p∗(OX (H)) ⊗ q∗(OX (H + B)))) H 1(S2Ω1 ∼= H 1(X × X, I 2 ∆X ⊗ p∗(OX (H)) ⊗ q∗(OX (H + B))), where the last iso- morphism comes from Leray spectral sequence. So, by (7), to prove surjec- ∆X ⊗p∗(OX (H))⊗q∗(OX (H +B))) = tivity of p1 it suffices to show that H 1(I 2 ∆X ⊗p∗(OX (H))⊗q∗(OX (H +B))) = 0. Using again the blow-up 0 and H 2(I 3 Y , this is true if the line bundles f ∗(OX (H)) ⊗ g∗(OX (H + B))(−4E) and f ∗(OX (H)) ⊗ g∗(OX (H + B))(−3E) are big and nef. In conclusion, if we prove that f ∗(OX (H)) ⊗ g∗(OX (H + B))(−4E) and f ∗(OX (H)) ⊗ g∗(OX (H + B))(−3E) are big and nef, then p1 is surjective. Moreover also f ∗(OX (C)) ⊗ g∗(OX (C))(−4E) is big and nef and therefore γ2 OX (C) is surjective. Following [7] we will exhibit pairs (X, C) as above (where C ∈ 2H + B) for which f ∗(OX (H))⊗g∗(OX (H +B))(−4E) and f ∗(OX (H))⊗g∗(OX (H + B))(−3E) are big and nef. First of all observe that if there exist four line bundles Ai i = 1, 2, 3, 4 on X such that OX(H) ∼= A1 ⊗ A2 ⊗ A3 ⊗ A4, then we have f ∗(OX (H)) ⊗ g∗(OX (H + B))(−4E) ∼= ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E)) ⊗ g∗(OX (B)), and f ∗(OX (H)) ⊗ g∗(OX (H + B))(−3E) ∼= ⊗i=1,2,3(f ∗(Ai) ⊗ g∗(Ai)(−E)) ⊗ (f ∗(A4) ⊗ g∗(A4 ⊗ OX (B)). These are big and nef under the conditions given in the following lemma. Lemma 3.3. Let A1, A2, A3, A4 be four base point free line bundles on a K3 surface X with A2 j ≥ 2, j = 1, 2, 3, 4 and such that A1 is very ample. Assume that either A2, A3, A4 are very ample, or they define (2 : 1) finite morphisms onto P2 and that if A2 j = 2, we have (A1 ⊗A2⊗A3 ⊗A4)·Aj ≥ 12, then ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E)) is big and nef. ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE 7 Proof. The proof is almost the same as the proof of lemma (2.2) of [7], but we reproduce it here for the reader's convenience. If Ai is very ample the linear system f ∗(Ai) ⊗ g∗(Ai)(−E) on Y has a sublinear system defining the morphism F : Y → Gr(1, PH 0(Ai)∗), associ- ating to (x, y) ∈ Y the line between φAi(x) and φAi(y), composed with the Plucker embedding. Notice that if (x, y) ∈ E, we can think of (x, y) as a pair where x ∈ X, y ∈ PTX,x, hence F (x, y) is the line generated by (dφAi)x(y). Therefore f ∗(Ai) ⊗ g∗(Ai)(−E) is nef and it is also big, since the image of X in PH 0(Ai)∗ is not ruled. Therefore, since A1 is very ample, (f ∗(A1) ⊗ g∗(A1))(−E) is big, hence ⊗i=1,2,3,4(f ∗(Ai)⊗g∗(Ai)(−E)) is big and it can fail to be nef only on a curve Z contained in the indeterminacy locus of the maps Y → Gr(1, PH 0(Ai)∗), i = 2, 3, 4 if Ai is not very ample. Notice that Z is a curve contained in {(x, y) ∈ Y − E φAi(x) = φAi(y)} ∪ {(x, y) ∈ E (dφAi)x(y) = 0}. Assume that Ai is not very ample, hence, by assumption, it gives a (2 : 1) morphism to P2. If Z 6⊂ E, let τ : X × X → X × X be the involution τ (x, y) = (y, x), then we can assume that the image Z of Z in X × X is such that τ (Z) = Z, because ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E)) is invariant under τ . Then the first (or second) projection Z ′ of Z in X is φ∗ Ai(Z1) for some curve Z1 in P2. If L is a line in P2 and Z1 ∼ mL we have ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E)) · Z = 2(A1 ⊗ A2 ⊗ A3 ⊗ A4) · Z ′ − 4E · Z = 2(A1 ⊗A2 ⊗A3 ⊗A4)·mφ∗ Ai(L)−4E ·Z = 2m(A1 ⊗A2 ⊗A3 ⊗A4)·Ai −4E ·Z, therefore we are done if we show that E · Z = 6m. Let B be the ramification divisor of φAi, then B is a smooth plane sextic and E · Z = mB · L = 6m, if the intersection of E and Z is transverse. This can be checked directly as in lemma (2.2) of [7]. If Z ⊂ E, then it is the strict transform of the ramification divisor R on X of φAi, hence Z · E = −c1(OPTX (1)) · Z = −degTR = 18. Therefore ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E)) · Z = 6(A1 ⊗ A2 ⊗ A3 ⊗ A4) · Ai − 4E · Z ≥ 0. (cid:3) Let us now show the construction of the examples. Proposition 3.4. There exist smooth K3 surfaces X with Picard lattice Γ = ZD⊕ZL⊕ZR⊕ZS⊕ZT with intersection matrix diag(2h, −2k, −2j, −2l, −2m) with j, k, l, m ≥ 2, h ≥ k + 1, j + 1, l + 1, m + 1 and D very ample. Moreover D + L, D + R, D + S and D + T are base point free, and they are either very ample, or they define (2 : 1) morphisms to P2. Proof. Observe that the lattice Γ is even, nondegenerate and of signature (1, 4), hence it occurs as the Neron-Severi group of some algebraic K3 surface (cf. [15], corollary 2.9). We will show that there does not exist a class F ∈ Γ such that F 2 = −2, [2]), D · F = 0. By well known results on periods of K3 surfaces (see e.g. 8 E. COLOMBO AND P. FREDIANI this implies that there exists a K3 surface with Picard lattice Γ and such that D is ample. Assume that F = aD + bL + cR + dS + eT (a, b, c, d, e ∈ Z) is such that F · D = 0, and F 2 = −2. The first equality implies a = 0, and the second one yields 1 = kb2 + jc2 + ld2 + me2, which is absurd, since k, j, l, m ≥ 2. So D is ample and D2 ≥ 4, hence D is very ample provided that there does not exist an irreducible curve F such that F 2 = 0, F · D = 1, 2 (cf. [16], or [14] theorem 5). But this cannot happen, because if we write F = aD + bL + cR + dS + eT , a, b, c, d, e ∈ Z, F · D = 2ha 6= 1, 2 since h ≥ 2. So D is very ample. Note that (D + L)2 = 2h − 2k ≥ 2, since h ≥ k + 1. First of all we show that for any (−2)-curve F , F · (D + L) > 0, hence D + L is ample and it is base point free, provided that there does not exist irreducible curves F, G and an integer a ≥ 2 such that D + L ∼ aF + G, with F 2 = 0, G2 = −2, F · G = 1 (cf. [16], or [14] theorem 5). This clearly cannot happen, since the product of two classes is always even. Set F = aD + bL + cR + dS + eT , a, b, c, d, e ∈ Z with F 2 = −2. Since D is ample, D · F = 2ha > 0, hence a > 0. F 2 = −2 yields a2 = kb2+jc2+ld2+me2−1 . h If (D +L)(aD +bL+cR+dS +eT ) = 2ah−2kb ≤ 0, we have bk ≥ ah > 0, so b > 0 and b2k2 ≥ h2a2 = h(b2k + c2j + d2l + e2m − 1). Thus we get b2(k2 − hk) − hc2j − hd2l − he2m + h ≥ 0, so if we set h = k + 1 + t, t ≥ 0, we obtain 0 ≤ b2(k2 − hk) − hc2j − hd2l − he2m + h ≤ b2(k2 − hk) + h = t(1 − b2k) + k(1 − b2) + 1 ≤ −t + 1 hence we must have either t = 0 or t = 1. But if t = 0, h = k + 1, so 0 ≤ b2(k2 − hk) − hc2j − hd2l − he2m + h = k(1 − b2) − (k + 1)(c2j + d2l + e2m) + 1 ≤ k(1 − b2) + 1, so b = 1, but then −(k + 1)(c2j + d2l + e2m) + 1 ≥ 0, which is absurd. If t = 1, h = k + 2, so we have 0 ≤ −2b2k + k + 2 − (k + 2)(c2j + d2l + e2m) ≤ k(1 − 2b2) + 2, thus we must have b = 1, hence 0 ≤ −k + 2 − (k + 2)(c2j + d2l + e2m) ≤ −(k + 2)(c2j + d2l + e2m), which implies c = d = e = 0. But then a2 = k−1 k+2 , which is absurd. So D + L is ample and base point free. If (D + L)2 ≥ 4 a similar computation shows that there does not exists a curve F such that F 2 = 0 and F · (D + L) = 1, 2, therefore D + L is very ample. If (D + L)2 = 2, it clearly defines a (2:1) morphism to P2. The same holds for D + R, D + S and D + T . (cid:3) Proof of (3.1). Consider the K3 surfaces constructed in (3.4). ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE 9 Let us set A1 = D, Ai i = 2, 3, 4 be either equal to D, or to D + L, or to D + R, or to D + S, or to D + T . Setting H = A1 + A2 + A3 + A4 = 4D+aL+bR+cS+dT , with a, b, c, d ≥ 0, a+b+c+d ≤ 3, H = 2H +B, where B = nD + m(D + L) + r(D + R) + s(D + S) + t(D + T ), with m, n, r, s, t ≥ 0, lemma (3.3) applies. In fact it suffices to check that H · (D + L) ≥ 12, H · (D + R) ≥ 12, and H · (D + S) ≥ 12, H · (D + T ) ≥ 12, which is true for our choices of H. Hence ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E) ⊗ g∗(OX (B)) is big and nef and (⊗i=1,2,3(f ∗(Ai)⊗g∗(Ai)(−E)))⊗(f ∗(A4)⊗g∗(A4)⊗(OX (B)) is also big and nef. If we take C ∈ H, we also have H 1(S2Ω1 X ⊗ OX (C)) = 0, thus γ2 C is surjective. Now we have to check that with our choices of a non divisible H, we H 2 ≥ 281 for curves C ∈ H. To this obtain all the genera g(C) = 1 + 1 2 end, it suffices to take H = aD + sL + tR + vS + rT, where s, t, v, r ≥ 0, relatively prime, s + t + v + r ≤ a − 2, 9 ≤ a ≤ 14 and, if a = 9, at most one among s, t, v, r is odd, if a = 10, at most two among s, t, v, r are odd, if a = 11, at most three among s, t, v, r are odd. In particular, let us start with H = 9D + 6L + R, for C ∈ H, we have g(C) = 1 + 1 2 H 2 = 1 + 81h − 36k − j. Let us set n = k − 2, m = j − 2, h = n + 3 + t, where n, m, t ≥ 0, then we have h ≥ m + 3, so t ≥ m − n, g(C) = 1 + 81(n + 3 + t) − 36(n + 2) − (m + 2), and we have two cases: (1) ρ := n − m ≥ 0, t ≥ 0, then g(C) = 1+81(m+ρ+3+t)−36(m+ρ+2)−(m+2) = 170+45ρ+44m+81t, with t, m, ρ ∈ Z, t, m, ρ ≥ 0. (2) α := m − n ≥ 0, t = α + β, with β ≥ 0, then g(C) = 1+81(n+3+α+β)−36(n+2)−(n+2+α) = 170+44n+80α+81β, with n, α, β ∈ Z, n, α, β ≥ 0. Since 44 and 45 are relatively prime it is clear that with g(C) = 170 + 45ρ + 44m + 81t one gets all sufficiently high genera. Using (1) and (2), one can simply check that g(C) runs through all the integers greater than 620 and with the other choices of H one gets all genera g greater than 280 except for g = 321. For g = 321 we consider the K3 surface contructed in [7] proposition (3.2) with Picard lattice given by Γ = ZD ⊕ ZL with D2 = 4, L2 = 2, D · L = 7. In [7] it is proven that D is very ample and L defines a 2 : 1 finite morphism onto P2. So if we set Ai = D, i = 1, 2, 3, 4, H = 4D, B = 3D + L, H = 2H + B = 11D + L, since H · L = 4D · L = 28, lemma (3.3) applies. Hence ⊗i=1,2,3,4(f ∗(Ai) ⊗ g∗(Ai)(−E) ⊗ g∗(OX (B)) is big and nef and (⊗i=1,2,3(f ∗(Ai) ⊗ g∗(Ai)(−E))) ⊗ (f ∗(A4) ⊗ g∗(A4) ⊗ (OX (B)) is also big and nef. So, as above, if we take C ∈ H, we also have H 1(S2Ω1 X ⊗ 10 E. COLOMBO AND P. FREDIANI OX(C)) = 0, thus γ2 H 2 = 321. g = 1 + 1 2 C is surjective. Now it suffices to check that C has genus (cid:3) Corollary 3.5. For the general curve of genus greater than 152, the second Gaussian map γ2 C is surjective. Proof. By theorem (3.1) and semicontinuity of the corank of γ2 C , for a general curve of genus greater than 280 γ2 C is surjective. Surjectivity for the general curve of genus 153 ≤ g ≤ 280 can be proved exhibiting examples of curves of genus g with surjective second Gaussian map, which are either hyperplane sections of a polarized K3 surface as in the proof of (3.1), or in the product of two curves as in [8] theorem 3.1. More precisely let C1, C2 be two smooth curves of respective genera g1, g2, choose divisors Di on Ci of degree di, i = 1, 2. Set X = C1 × C2, let ∗(D2) be a smooth curve, where pi is the projection from C ∈ p1 C1 × C2 on Ci, then g(C) = 1 + (g2 − 1)d1 + (g1 − 1)d2 + d1d2. ∗(D1) ⊗ p2 In [8] we proved that if either g1, g2 ≥ 2, di ≥ 2gi + 5, i = 1, 2, or g1 ≥ 2, g2 = 1, d1 ≥ 2g1 + 5, d2 ≥ 7, or g2 = 0, d2 ≥ 7, d2(g1 − 1) > 2d1 ≥ 4g1 + 10, then γ2 C is surjective. Then one has to check directly that these values of g(C) cover all the (cid:3) remaining integers between 153 and 280. References [1] Ballico, E., Fontanari, C., On the surjectivity of higher Gaussian maps for complete intersection curves. Ricerche Mat. 53 (2004), no. 1, 79 -- 85 (2005). [2] Beauville, Preliminaires sur les periodes des surfaces K3. Asterisque No. 126 (1985), 91-97. [3] Beauville, A., Merindol, J.-Y., Sections hyperplanes des surfaces K3, Duke Math. Jour., 55 no. 4, (1987), 873 -- 878. [4] Ciliberto, C., Harris, J., Miranda, R. On the surjectivity of the Wahl map, Duke Math. Jour., 57 , (1988), 829 -- 858. [5] Ciliberto, C., Lopez, A. F., Miranda, R. Projective degenerations of K3 surfaces, Gaussian maps, and Fano threefolds. Invent. Math. 114 (1993), no. 3, 641 -- 667. [6] Ciliberto, C., Lopez, A. F., Miranda, R. Classification of varieties with canonical curve section via Gaussian maps on canonical curves. Amer. J. Math. 120 (1998), no. 1, 1 -- 21. [7] Ciliberto, C., Lopez, A. F., Miranda, R. On the corank of Gaussian maps for general embedded K3 surfaces. Proceedings of the Hirzebruch 65 Conference on Algebraic Geometry (Ramat Gan, 1993), 141 -- 157, Israel Math. Conf. Proc., 9, Bar-Ilan Univ., Ramat Gan, 1996. [8] Colombo, E., Frediani, P., Some results on the second Gaussian map for curves, arXiv:0805.3422, to appear on Michigan Journal of Mathematics. [9] Colombo, E., Frediani, P., Siegel metric and curvature of the moduli space of curves, arXiv:0805.3425, to appear on Transaction of the AMS. [10] Colombo, E., Frediani, P., Pareschi, G. Hyperplane sections of abelian surfaces, arXiv:0903.2781. [11] Colombo, E., Pirola, G.P., Tortora, A., Hodge-Gaussian maps, Ann. Scuola Normale Sup. Pisa Cl. Sci. (4) 30 (2001), no. 1, 125 -- 146. ON THE SECOND GAUSSIAN MAP FOR CURVES ON A K3 SURFACE 11 [12] Green, M. L., Infinitesimal methods in Hodge theory, in Algebraic Cycles and Hodge Theory, Torino 1993, Lecture Notes in Mathematics, 1594. Springer, Berlin, (1994), 1 -- 92. [13] Kawamata, Y., A generalization of Kodaira - Ramanujam's vanishing theorem. Math. Ann. 261 (1982), n.1, 43-46. [14] Mori, S., On degrees and genera of curves on smooth quartic surfaces in P 3. Nagoya Math. J. 96 (1984), 127-132. [15] Morrison, D. R., On K3 surfaces with large Picard number. Invent. Math. 75 (1984), no. 1, 105-121. [16] Saint-Donat, B., Projective models of K3 surfaces. Amer. J. Math. 96 (1974), 602-639. [17] Viehweg, E., Vanishing theorems. J.Reine Angew. Math. 335, (1982), 1-8. [18] Voisin, C., Sur l'application de Wahl des courbes satisfaisant la condition de Brill- Noether-Petri, Acta Math. 168 (1992), 249-272. [19] Wahl, J., Gaussian maps on algebraic curves, J. Diff. Geom. 32 (1990), no. 1, 77 -- 98. [20] Wahl, J., Introduction to Gaussian maps on an algebraic curve, Complex projective geometry (Trieste, 1989/Bergen, 1989), London Math. Soc. Lecture Note Ser., 179, Cambridge Univ. Press, Cambridge, (1992), 304 -- 323. [21] Wahl, J., The Jacobian algebra of a graded Gorenstein singularity, Duke Math. Jour. 55 (1987), 843 -- 871. Dipartimento di Matematica, Universit`a di Milano, via Saldini 50, I-20133, Milano, Italy E-mail address: [email protected], Fax:+39-02-50316090 Dipartimento di Matematica, Universit`a di Pavia, via Ferrata 1, I-27100 Pavia, Italy E-mail address: [email protected], Fax:+39-0382-985602
1303.6113	1	1303	2013-03-25T12:51:38	\`A propos des vari\'et\'es de Poncelet	[ "math.AG" ]	We recall the definition of Poncelet varieties that generalize the celebrated Poncelet curves introduced by Darboux. We show that any quadric and any smooth cubic in the projective space of dimension three is a Poncelet surface but that a general surface of degree greater than four is not.	math.AG	math	`A PROPOS DES VARI´ET´ES DE PONCELET JEAN VALL`ES R´esum´e. On d´efinit les vari´et´es de Poncelet. Celles-ci g´en´eralisent tr`es naturellement les c´el`ebres courbes de Poncelet, qui, conform´ement `a l'axiome d'Arnold, ne furent pas introduites par Poncelet mais par Darboux. On montre ensuite qu'une surface g´en´erale de degr´e ≥ 4 dans P3 n'est pas une surface de Poncelet mais, par contre, que toutes les quadriques et toutes les cubiques lisses de P3 sont des surfaces de Poncelet. 1. Introduction La petite histoire des math´ematiques croise souvent l'histoire plus rocambolesque des livres scolaires. Marie Stuart fut d´ecapit´ee pour son gout du chiffre, gout qu'elle acquit `a la cour du roi Charles IX aupr`es de Vi`ete, Pascal d´ecrit l'esprit g´eom´etrique pour instruire les petits ´el`eves jans´enistes avant que Louis XIV ne condamne Port-Royal, et, plus pr`es de nous, Poncelet, jeune lieutenant de la grande arm´ee de Napol´eon posa les bases de la g´eom´etrie projective moderne dans une prison russe o`u il fut retenu prisonnier apr`es la bataille de Krasnoie, bataille rest´ee c´el`ebre car l'empereur des fran¸cais r´eussit `a passer entre les mailles serr´ees du filet tendu par Koutouzov. Dans les M´emoires d'outre-tombe Chateaubriand (cf. [2], page 211, tome 2) relate l'´episode avec emphase : " Les hauteurs environnantes au pied desquelles marchait Napol´eon, se chargeaient d'artillerie et pouvaient `a chaque instant le foudroyer ; il y jette un coup d'oeil et dit : "Qu'un escadron de mes chasseurs s'en empare !" Les Russes n'avaient qu'`a se laisser rouler en bas, leur masse l'eut ´ecras´e ; mais, `a la vue de ce grand homme et des d´ebris de la garde serr´ee en bataillon carr´e, ils demeur`erent immobiles, comme fascin´es : son regard arreta cent mille hommes sur les collines." Poncelet, quant `a lui, resta sur le champ de bataille, "[...] d'o`u il n'est sorti vivant que par une faveur sp´eciale de Dieu, au milieu de ses chefs, de ses camarades tu´es ou atteints de blessures, toutes mortelles dans ce pernicieux climat" (cf. [11], Pr´eface), et fut fait prisonnier le lendemain de la fuite de Napol´eon. Dans le manuscrit r´edig´e lors de son s´ejour russe [11], il d´emontre qu'une conique n-circonscrite `a une conique lisse fix´ee lui est infiniment circonscrite. C'est le th´eor`eme appel´e th´eor`eme de cloture. Quelques d´ecennies plus tard, Darboux a montr´e qu'une (cid:1) sommets d'un polygone dont les n cot´es sont courbe de degr´e n− 1 passant par les(cid:0)n 2 tangents `a une conique lisse C passe par une infinit´e de sommets de tels polygones (cf. [3], p. 248). C'est aussi dire que tout point de la courbe est le sommet d'une configuration de n droites tangentes `a C. Ces courbes sont appel´ees courbes de Poncelet. Pour les vari´et´es de dimensions plus grandes introduites par Trautmann dans [12], et que l'on red´efinit ci-dessous comme d´eterminant de sections d'un certain fibr´e vectoriel (voir la d´efinition 1), le th´eor`eme de Darboux ne s'applique plus ; plus pr´ecis´ement, il ne suffit plus de passer par les sommets d'un polytope (n − 1) tangents `a la courbe rationnelle normale sous-jacente pour que chacun de ses points soit le sommet d'un tel 2000 Mathematics Subject Classification. 14N15, 14N20, 14L35, 14L30. Key words and phrases. Fibr´es de Schwarzenberger, Courbes et surfaces de Poncelet. L'auteur a ´et´e partiellement financ´e par les ANR-09-JCJC-0097-0 INTERLOW et ANR GEOLMI.. 1 2 JEAN VALL`ES Figure 1. Un quadrilat`ere complet circonscrit et une cubique de Poncelet (cid:1) sommets 2 L'espace projectif des courbes planes de degr´e n − 1 passant par les (cid:0)n polytope. On v´erifie ci-dessous que pour qu'une hypersurface soit une hypersurface de Poncelet, il lui faudra, passer par les sommets d'une infinit´e de dimension n− 1 de tels polytopes (voir la proposition 1). de n droites fix´ees est de dimension n − 1. La donn´ee de n droites tangentes `a une conique lisse ´equivaut `a la donn´ee d'un diviseur de degr´e n sur P1. Sa dimension projective est n. Enfin l'espace des coniques du plan est de dimension projective 5. Ainsi un d´ecompte naıf de dimension du sch´ema des courbes planes de degr´e n− 1 qui sont de Poncelet donne (n − 1) + n + 5 = 2n + 4. Mais ce d´ecompte est faux d'apr`es le th´eor`eme de Darboux. On v´erifie en effet que cette dimension est 2n + 3, ce qui correspond au choix d'une conique (dimension 5) et d'un pinceau de formes binaires de degr´e n (dimension de G(1, n) ´egale 2n− 2) sur cette conique. Ainsi, contrairement aux cubiques planes et aux coniques (cf. [1], p. 90 et p. 85), la vari´et´e des quartiques de Poncelet (appel´ees quartiques de Luroth) ne remplit pas l'espace des quartiques planes et forme un diviseur. Morley a prouv´e que le degr´e de la sous-vari´et´e des quartiques de Luroth ´etait ´egal `a 54 (cf. [9]). Barth, dans son article fondateur sur les fibr´es vectoriels de rang deux sur le plan (cit´e ci-dessus), montre que cette hypersurface est l'image de l'espace de modules MP2(0, 4) (des faisceaux localement libres et semistables de premi`ere et seconde classes de Chern ´egales `a 0 et 4) par l'application qui `a un fibr´e associe sa courbe de droites de saut. Qui plus est, le degr´e de l'image est le nombre de Donaldson 1 q13, ce qui accroıt encore l'int´eret pour le diviseur des quartiques de Luroth. En effet, comme le morphisme de Barth est g´en´eriquement injectif, ce degr´e coıncide avec q13 (cf. [8] et [10], ainsi bien-sur que [9]). Le meme probl`eme se pose en dimension plus grande. `A un couple (Cn, f1∧···∧fn)) form´e d'une courbe rationnelle normale et de n formes binaires de degr´e n+k on associe une hypersurface de Poncelet de degr´e k + 1 sur Pn. Un d´ecompte des dimensions attendues montre que cette application ne peut pas etre dominante pour n = 3 d`es que k ≥ 3. Par contre elle peut l'etre lorsque n = 3 pour k = 1 et k = 2. Et en effet, on montre qu'elle l'est, plus pr´ecis´ement, on montre qu'une surface quadrique quelconque 1. Ces nombres, et plus g´en´eralement les invariants de Donaldson, ont ´et´e introduits par Donaldson lui-meme, en 1983, un an apr`es que la classification des vari´et´es topologiques compactes simplement connexes de dimension 4 a ´et´e ´etablie par Freedman [5]. Donaldson, en introduisant ces invariants, a montr´e qu'il existe des surfaces complexes lisses hom´eomorphes qui ne sont pas diff´eomorphes [4]. `A PROPOS DES VARI´ET´ES DE PONCELET 3 ou une surface cubique lisse de P3 sont des surfaces de Poncelet (voir les th´eor`emes 2 et 3). 2. G´en´eralit´es et notations Soit V un espace vectoriel complexe de dimension 2. On note Si les espaces vectoriels i = Symi(V ∨)) invariants sous l'action de SL(V). La multiplication des Symi(V ) (et S∨ formes binaires : n ⊗ S∨ S∨ k −−−−→ S∨ n+k permet de d´efinir deux fibr´es, dits de Schwarzenberger, Ek,n+k sur P(S∨ k ) et En,n+k sur P(S∨ n ) ainsi que la vari´et´e des espaces lin´eaires N s´ecants (avec N = inf(k, n) + 1) `a la courbe Cn+k, image de P(V ) dans P(Sn+k) par le morphisme de Veronese. Les fibr´es projectivis´es sont, pour l'un, l'´eclatement de P(Sn+k) le long de cette vari´et´e et pour l'autre, le mod`ele lisse de cette vari´et´e de s´ecantes, obtenu en l'´eclatant le long de son lieu singulier. Tout ceci est connu et r´edig´e en plusieurs endroits (cf. par exemple [7]). Dans l'article [13] consacr´e aux fibr´es de Schwarzenberger, ces courbes sont interpr´et´ees comme des pinceaux de sections de ces fibr´es. C'est cette interpr´etation que nous reprenons ci-dessous en la g´en´eralisant aux dimensions plus grandes. Consid´erons le fibr´e de Schwarzenberger suivant : 0 −−−−→ Sk ⊗ OP(S∨ n )(−1) −−−−→ Sn+k ⊗ OP(S∨ n ) −−−−→ En,n+k −−−−→ 0. Nous associons, `a un syst`eme lin´eaire Λ =< f0,··· , fr >⊂ Sn+k de dimension r+1 ≤ n, n ) d´efini par les ´equations f0∧···∧fr = 0 (l'´ecriture est abusive un sous-sch´ema de P(S∨ mais nous l'expliciterons plus loin). En effet, comme H0(En,n+k) ∼= H0(OP1(n + k)), les formes binaires fi s'interpr`etent comme des sections de En,n+k. Le diagramme commutatif suivant l'explique : Λ ⊗ OP(S∨ n ) Λ ⊗ OP(S∨ n ) 0 −−−−→ Sk ⊗ OP(S∨ n )(−1) −−−−→ Sn+k ⊗ OP(S∨ n ) −−−−→ En,n+k −−−−→ 0 −−−−→ 0. 0 −−−−→ Sk ⊗ OP(S∨ n )(−1) −−−−→ Sn+k Λ ⊗ OP(S∨ n ) −−−−→ n ), cesse d'etre localement libre le long de la sous-vari´et´e Le faisceau L, d´efini sur P(S∨ ferm´ee d'´equation f0 ∧ ··· ∧ fr = 0. D´efinition 1. Les vari´et´es d´efinies par la donn´ee de r ≤ n sections lin´eairement ind´ependantes d'un fibr´e de Schwarzenberger En,n+k au dessus de P(S∨ n ) sont appel´ees Vari´et´es de Poncelet. la courbe rationnelle normale Cn ⊂ P(Sn) ; cette section s'annule le long des (cid:0)n+k Une section de En,n+k correspond `a la donn´ee de n + k points sur la courbe ra- tionnelle normale Cn+k et, par l'isomorphisme canonique Cn+k (cid:39) Cn (cid:39) P(S1), sur points d'intersection de n hyperplans qui ont un contact d'ordre n − 1 avec Cn le long de n points choisis parmi les n + k points. Ces vari´et´es de Poncelet sont ainsi caract´eris´ees par le fait de passer par les sommets d'un polytope `a n + k faces ; celles-ci sont des hyperplans tangents d'ordre n − 1 `a Cn. Le cas des courbes est bien connu (cf. par exemple [3] et [12]). En dimension sup´erieure, hormis quelques lignes `a la fin de l'article [12] et quelques exemples dans [7], elles n'ont pas ´et´e ´etudi´ees. (cid:1) n (cid:121) (cid:121) (cid:13)(cid:13)(cid:13) (cid:121) (cid:121) L On−1 P(S∨ n ) (cid:121) (cid:121) On−1 P(S∨ n ) (cid:121) (cid:121) (cid:13)(cid:13)(cid:13) 4 JEAN VALL`ES La proposition suivante g´en´eralise le th´eor`eme de Darboux concernant les courbes. Proposition 1. Une hypersurface de degr´e k + 1 est une hypersurface de Poncelet sur Pn si et seulement si elle contient une vari´et´e de Poncelet de codimension 2 d´efinie par (n − 1) sections lin´eairement ind´ependantes du fibr´e En,n+k. D´emonstration. Consid´erons le diagramme commutatif suivant : 0 −−−−→ Sk ⊗ OP(S∨ n )(−1) −−−−→ Sn+k ⊗ OP(S∨ n ) −−−−→ En,n+k −−−−→ 0 0 −−−−→ Sk ⊗ OP(S∨ n )(−1) −−−−→ Sn+kCn−1 ⊗ OP(S∨ n ) −−−−→ IX (k + 1) −−−−→ 0. Le sch´ema X est de dimension n− 2 et il est d´efini par n− 1 sections ind´ependantes de En,n+k. Une section non nulle de H0(IX (k + 1)), c'est-`a-dire une hypersurface de degr´e k + 1 passant par les sommets des n − 1 polytopes d´etermin´es par les n − 1 sections, provient d'une section suppl´ementaire de En,n+k. En effet cela r´esulte de la surjectivit´e de l'application H0(En,n+k) → H0(IX (k + 1)). L'hypersurface est alors d´efinie par n sections, autrement dit, son ´equation est le (cid:3) d´eterminant de ces n sections. 3. Surfaces de Poncelet Les cubiques planes sont toutes de Poncelet tandis que les quartiques planes de Poncelet ne forment qu'un diviseur. Plus g´en´eralement, les courbes de Poncelet de degr´e k forment une sous-vari´et´e de dimension 2(k + 1) + 3 de l'espace des courbes de degr´e k. Pour les hypersurfaces de Poncelet de Pn de degr´e k la dimension attendue est dimG(k, n + k − 1) + dim PGL(n + 1) PGL(2) . Ainsi la dimension attendue des surfaces de Poncelet de degr´e k est 3k + 12. Ce qui fait 18 pour les quadriques, 21 pour les cubiques, 24 pour les quartiques. L'espace de toutes les quadriques et de dimension 9, celui des cubiques de dimension 19 et celui des quartiques 34. Pour k ≥ 4, pour des raisons ´evidentes de dimension, une surface de degr´e k g´en´erale n'est pas de Poncelet. Par contre on attend a priori que toute surface quadrique et toute surface cubique soient de Poncelet (et plutot deux fois qu'une). C'est ce que nous v´erifions maintenant. Le cas des quadriques est ´el´ementaire. Th´eor`eme 2. Toute quadrique de P3 est une quadrique de Poncelet. D´emonstration. Une quadrique de Poncelet est une section du fibr´e de Schwarzenberger d´efini par la suite exacte : 0 −−−−→ S1 ⊗ OP(S∨ 3 ) −−−−→ E3,4 −−−−→ 0. Quitte `a noter P(S3) = proj(C[x0, x1, x2, x3]) la matrice M est bien d´ecrite : 3 )(−1) M−−−−→ S4 ⊗ OP(S∨  x0 0 x1 x0 x2 x1 x3 x2 x3 0  . M = `A PROPOS DES VARI´ET´ES DE PONCELET (cid:12)(cid:12)(cid:12)(cid:12) x1 x0 x3 x2 (cid:12)(cid:12)(cid:12)(cid:12), (cid:12)(cid:12)(cid:12)(cid:12) x1 x0 x2 x1 (cid:12)(cid:12)(cid:12)(cid:12), (cid:12)(cid:12)(cid:12)(cid:12) x1 x0 x3 0 (cid:12)(cid:12)(cid:12)(cid:12) et (cid:12)(cid:12)(cid:12)(cid:12) x0 (cid:12)(cid:12)(cid:12)(cid:12) extraits de M d´efinissent 5 (cid:3) 0 x1 x0 Les mineurs des quadriques de rangs respectivement ´egaux `a 4, 3, 2 et 1. Le cas des cubiques est un peu plus difficile. Th´eor`eme 3. Toute surface cubique lisse de P3 est une surface de Poncelet. Remarque 1. Il est certainement int´eressant de d´eterminer les cubiques singuli`eres qui ne sont pas de Poncelet. Par exemple, je pense que trois plans poss´edant une droite commune ne sont pas une surface de Poncelet. D´emonstration. Soit S une surface cubique lisse. Elle contient 27 droites exactement et on peut en choisir 6 droites parmi elles qui soient deux `a deux disjointes. En contractant ces 6 droites on obtient un P2 et chacune des 6 droites a pour image un point ; de plus les 6 points ainsi obtenus, que l'on note Z, sont en position g´en´erale. C'est classique. Les six points Z (et meme le double six) et la surface cubique S proviennent d'un meme 3-tenseur φ ∈ A ⊗ B ⊗ C avec dimC(A) = dimC(B) = 3 et dimC(C) = 4. Sur P(C) = P3, ce tenseur devient une matrice de formes lin´eaires A∨ ⊗ OP(C)(−1) φ−−−−→ B ⊗ OP(C), dont le d´eterminant est une ´equation de S. Sur P(B) = P2, ce meme tenseur devient une matrice de formes lin´eaires A∨ ⊗ OP(B)(−1) φ−−−−→ C ⊗ OP(B), dont les mineurs maximaux engendrent H0(IZ(3)), o`u IZ est l'id´eal des six points. L'´eclatement du plan le long de Z, `a savoir P(IZ(3)) ⊂ P(B) × P(C), s'envoie isomor- phiquement sur S par la seconde projection. Pour montrer que S est une surface de Poncelet, il suffit de montrer que S est d´efinie par un tenseur φ ∈ S2 ⊗ S3 ⊗ ( S5 A )∨ o`u A = C3 est un sous espace vectoriel de S5. Remarquons tout d'abord que si l'on consid`ere le groupe de 6 points Z associ´es au tenseur : 0 −−−−→ S2 ⊗ OP( S5 A ) (−1) −−−−→ S3 ⊗ OP( S5 A ) −−−−→ IZ(3) −−−−→ 0, la matrice des relations est persym´etrique (ou de Haenkel). En effet, c'est la restriction `a P( S5 sur P(S5). A ) de la matrice persym´etrique canonique (cf. par exemple [6]) de dimension 3× 4 R´eciproquement une matrice 3 × 4 persym´etrique de formes lin´eaires sur P2 est toujours la restriction de la matrice persym´etrique canonique sur P5 (puisque d`es que les nombres de lignes et de colonnes sont fix´es, il n'existe qu'une seule matrice per- sym´etrique g´en´erique). Or les mineurs 2 × 2 de cette matrice persym´etrique g´en´erique d´efinissent une quintique rationnelle normale C5 ⊂ P5, et via l'association d´ecrite ci-dessus, une surface cubique de Poncelet. Le th´eor`eme sera donc prouv´e lorsque le lemme suivant le sera. Lemme 1. Six points en position lin´eaire g´en´erale sont r´esolus par une matrice per- sym´etrique. D´emonstration du lemme. Consid´erons six points en position g´en´erale sur le plan. Le syst`eme lin´eaire des courbes de degr´e 5 passant doublement par chacun des points a dimension projective 2 ((21 − 6 × 3) − 1). Une courbe g´en´erale Γ5 de ce syst`eme est lisse hors des points bases et elle a genre 0. Elle est donc la projection d'une quintique rationnelle lisse de C5 ⊂ P5 sur la quintique Γ5 ⊂ P2 = P(A∨). En notant Si l'ensemble 6 JEAN VALL`ES des formes binaires de degr´e i sur C5 (cid:39) P1, on a C5 ⊂ P(S5) et il existe un sous espace vectoriel A ⊂ S5 de dimension 3 qui d´efinit le centre de projection. Plus pr´ecis´ement, la suite exacte suivante d'espaces vectoriels : 0 −−−−→ A −−−−→ S5 −−−−→ S5 A −−−−→ 0, induit la morphisme de projection : π : P(S5)\P( S5 A ) → P(A∨). La projection restreinte πC5 : C5 −→ Γ5 est ramifi´ee au-dessus des six points doubles de Γ5. (Les 6 points doubles de la projection sont les images des 6 points d'intersection de P( S5 A ) avec la vari´et´e des droites bis´ecantes `a C5.) Les points doubles de la quintique Γ5 sont d´efinis par le lieu de d´eg´en´erescence de la matrice de multiplication (par un ´el´ement variable de A )∨. Il est persym´etrique. (cid:3) S2) de S3 dans S5 A )∨ d´efinit aussi la surface cubique S ⊂ P(S3) (plus exactement un faisceau LS de rang 1 sur S) associ´ee aux six points, ainsi que le fibr´e de Schwarzenberger dont elle est le d´eterminant de trois sections ind´ependantes, comme le montre le diagramme commutatif suivant : Reprise de la d´emonstration du th´eor`eme. Le tenseur S2 ⊗ S3 ⊗ ( S5 A , ou encore un tenseur dans S2⊗ S3⊗ ( S5 A ⊗ OP(S∨ 3 ) A ⊗ OP(S∨ 3 ) (cid:121) = 0 −−−−→ S2 ⊗ OP(S∨ 3 )(−1) −−−−→ S5 ⊗ OP(S∨ 3 ) −−−−→ E3,5 0 −−−−→ S2 ⊗ OP(S∨ 3 )(−1) −−−−→ S5 A ⊗ OP(S∨ 3 ) −−−−→ −−−−→ 0 −−−−→ 0. (cid:3) (cid:121) (cid:121) (cid:121) (cid:121) LS Remarque 2. Pour ´etablir ce dernier th´eor`eme, on peut aussi montrer que la surface cubique S contient une courbe de Poncelet de genre 3 et de degr´e 6. Cette courbe est le plongement dans P3 d'une quartique plane lisse ; est-elle une quartique sp´eciale, une quartique de Poncelet par exemple ? Je remercie le rapporteur anonyme pour cette derni`ere remarque et Laurent Gruson qui m'a donn´e la preuve du lemme 1. R´ef´erences [1] Barth, W. Moduli of vector bundles on the projective plane. Invent. Math. 42 (1977), 63 -- 91. [2] Chateaubriand, F. R. M´emoires d'outre tombe. Livre de poche. [3] Darboux, G. Le¸cons sur la th´eorie g´en´erale des surfaces. III, IV. Les Grands Classiques Gauthier- Villars. [Gauthier-Villars Great Classics]. ´Editions Jacques Gabay, Sceaux, 1993. [4] Donaldson, S. K. An application of gauge theory to four-dimensional topology. J. Differential Geom. 18, 2 (1983), 279 -- 315. [5] Freedman, M. H. The topology of four-dimensional manifolds. J. Differential Geom. 17, 3 (1982), 357 -- 453. [6] Harris, J. Algebraic geometry, vol. 133 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995. A first course, Corrected reprint of the 1992 original. [7] Ilardi, G., Supino, P., and Vall`es, J. Geometry of syzygies via Poncelet varieties. Boll. Unione Mat. Ital. (9) 2, 3 (2009), 579 -- 589. [8] Le Potier, J., and Tikhomirov, A. Sur le morphisme de Barth. Ann. Sci. ´Ecole Norm. Sup. (4) 34, 4 (2001), 573 -- 629. [9] Morley, F. On the Luroth Quartic Curve. Amer. J. Math. 41, 4 (1919), 279 -- 282. [10] Ottaviani, G., and Sernesi, E. On the hypersurface of Luroth quartics. Michigan Math. J. 59, 2 (2010), 365 -- 394. `A PROPOS DES VARI´ET´ES DE PONCELET 7 [11] Poncelet, J.-V. Trait´e des propri´et´es projectives des figures. Tome I. Les Grands Classiques Gauthier-Villars. [Gauthier-Villars Great Classics]. ´Editions Jacques Gabay, Sceaux, 1995. Reprint of the second (1865) edition. [12] Trautmann, G. Poncelet curves and associated theta characteristics. Exposition. Math. 6, 1 (1988), 29 -- 64. [13] Vall`es, J. Fibr´es de Schwarzenberger et coniques de droites sauteuses. Bull. Soc. Math. France 128, 3 (2000), 433 -- 449. E-mail address: [email protected] Universit´e de Pau et des Pays de l'Adour, Avenue de l'Universit´e - BP 576 - 64012 PAU Cedex - France
1412.7204	4	1412	2016-03-28T18:57:19	Scaling of conformal blocks and generalized theta functions over $\bar{M}_{g,n}$	[ "math.AG" ]	By way of intersection theory on $\bar M_{g,n}$, we show that geometric interpretations for conformal blocks, as sections of ample line bundles over projective varieties, do not have to hold at points on the boundary. We show such a translation would imply certain recursion relations for first Chern classes of these bundles. While recursions can fail, geometric interpretations are shown to hold under certain conditions.	math.AG	math	SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA Abstract. By way of intersection theory on Mg,n, we show that geometric interpretations for conformal blocks, as sections of ample line bundles over projective varieties, do not have to hold at points on the boundary. We show such a translation would imply certain recursion relations for first Chern classes of these bundles. While recursions can fail, geometric interpretations are shown to hold under certain conditions. 1. Introduction Conformal blocks are vector spaces associated to stable curves together with certain Lie theoretic data. These form vector bundles on the moduli stacks Mg,n parametrizing stable n-pointed curves of genus g. For such a nontrivial bundle V, if x is a smooth n-pointed curve, then by [BL94,Fal94,KNR94,Pau96,LS97], x (cid:27) H0(Xx, Lx), where Xx is a moduli space determined by the Lie there is a canonical isomorphism V∗ data, and Lx is a canonical ample line bundle on it. The H0(Xx, Lx) are called generalized theta functions. These isomorphisms at interior points x respect both multiplication operations on global sections and the algebra structure on conformal blocks. For example, in type A, for x ∈ Mg,n, one has a natural identification of graded algebras (1.1) Mm≥0 V[m]∗ x (cid:27)Mm≥0 H0(Xx, L⊗m x ), where the V[m], for m ∈ N, are bundles obtained from V under an operation called scaling (Def 2.5). Since these results were proved, it has been an open question as to whether such canonical isomor- phisms exist at all points x in the moduli space. This is natural, and also interesting, as each side of Equation (1.1) has an advantage over the other. Conformal blocks give vector bundles on the entire moduli stack Mg,n, and can be used to study its geometry. For instance, first Chern classes are base point free divisors on M0,n [Fak12]. On the other hand, for a polarized projective variety (X, L), the algebra Lm≥0 H0(X, L⊗m) is finitely generated. Existence of isomorphisms (1.1) at all boundary points x would imply that the algebra of conformal blocksLm≥0 x is finitely generated as well, an open question. Our main result is the following: V[m]∗ Theorem 1.1. There are stable n-pointed curves x = (C; p1, . . . , pn) and vector bundles of conformal blocks V, for which there are no polarized pairs (Xx, Lx) such that Equation 1.1 holds. To prove Thm 1.1, we give an obstruction to the existence of a geometric interpretation of conformal blocks at points x on the boundary of Mg,n for bundles for which (Xx, Lx) are well understood. The ∆-invariant of (Xx, Lx) is the quantity ∆(Xx, Lx) = dim(Xx) + Ldim(Xx) − h0(X, Lx) (see Def 2.9). We say that V has ∆-invariant zero rank scaling if ∆(Xx, Lx) = 0 for x ∈ Mg,n such that V∗ x (cid:27) H0(Xx, Lx). Such pairs can be realized as projective varieties of minimal degree (see Section 2.3). We show the following. Theorem 1.2. Suppose that V has ∆-invariant zero rank scaling, and there are polarized pairs (Xx, Lx) such that Equation 1.1 holds for all points x in Mg,n. Then for m ≥ 1: (1.2) αi(m) c1(V[i]), where D = Ldim Xx x c1(V[m]) = X1≤i≤D and where the coefficients αi(m) are polynomials in m (see Theorem 3.1 for a precise statement). 1 2 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA In Example 4.2, and its generalization 4.4, we give bundles V for which Equation (1.2) fails. Here we use formulas of [Fak12] which compute the Chern classes of V. The remainder of the paper is devoted to supporting a conjectural description for the locus Z on Mg,n for which there are always geometric interpretations for conformal blocks at every point x ∈ Z. Based on the many examples for g ≤ 3, including bundles that do not have ∆-invariant zero rank scaling, we make following assertion. Conjecture 1.3. Given a vector bundle of conformal blocks V of type A on Mg,n, there are polarized pairs (Xx, Lx) such that Equation 1.1 holds for all x ∈ Z = Mrt g,n ∪ ∆0 irr. In the statement of Conjecture 1.3, the notation Mrt g,n stands for the set of points in Mg,n corresponding to stable curves in the locus of rational tails. These are curves with one irreducible component having genus g. By ∆0 irr we mean the set of curves in the interior of the component of the boundary whose generic point has a non-separating node. We note that for g = 0, the locus Z is equal to M0,n. In Section 5 we illustrate how our techniques are not limited to bundles of conformal blocks with ∆-invariant rank zero, and can be applied as long as one has sufficient information about polarized pairs (Xx, Lx) at points x on the interior Mg,n. First, we treat the case where Xx is embedded by Lx as Coble's quartic hypersurface Z4 in P7. Second, we consider a cubic hypersurface also studied by Coble: Z3 ֒→ P8 = P(H0(Jg−1C, OJC(3θ)), for C a smooth curve of genus 2. It is known that (cid:27) H0(SUC(3), L), where SUC(3) is a degree 2 cover of P8 branched over a sextic dual to V(sl3, 1)∗ Z3 [Ort05, Ngu07]. Finally, we consider an intersection of two quadrics in P5 using the explicit formulas for first Chern classes in genus > 1 given in [MOP15]. [C] 1.1. History. Our results can be seen to originate from the Verlinde formula, which gives a closed expres- sion for the dimension of spaces of generalized theta functions. Although there are finite dimensional proofs in special cases, see [Ber93, Tha94, Zag95], the theory of conformal blocks plays an essential role in the general result. We recommend the survey article of [Sor96] for a good account. While admittedly leaving out important contributions, one can summarize the two crucial aspects: (1) There is a canonical identification between conformal blocks and generalized theta functions over smooth curves [BL94, Fal94, KNR94, Pau96, LS97]. (2) The factorization theorem of Tsuchiya-Ueno-Yamada (see Theorem 2.2) allows for a decomposi- tion of conformal blocks at stable pointed curves x ∈ Mg,n \ Mg,n. Several interesting questions arising from the above picture have been investigated. One theme, originating in [NR93, DW93, Ram96], is to try to remove the reference to conformal blocks. The idea is that if the space of generalized theta functions could be factorized geometrically, then one would obtain a "finite dimensional" proof of the Verlinde formula. The goal is then to find pairs (Xx, Lx) for points x ∈ Mg,n \ Mg,n such that global sections have suitable factorization properties. This motif has been continued in the work of Kausz and Sun [Sun00, Sun03, Kau05]. To the best of our knowledge, these factorizations of generalized theta functions have not been related to conformal blocks, also see [Fal96, Tel98]. Our outlook is to suppose that there is an extension of the identification of conformal blocks and generalized thetas at points on the boundary, and to ask what consequences may follow. In type A, there are usually no descent problems, and so it seems reasonable to ask for an interpretation over moduli spaces rather than stacks. Projective varieties of minimal degree have the advantage that they will degenerate to other projective varieties of minimal degree. The unique resolutions associated to their ideal sheaves is our main tool. We also find, in special cases, geometric interpretations over the boundary by studying factorization properties of conformal blocks, and their algebras [TUY89, Man09]. SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 3 Remark 1.4. There has been recent work on realizing the first Chern classes of conformal blocks over M0,n (at a fixed level) geometrically as pull backs of classes under natural morphisms [Fak12, Gia13, AGS14, Fed11, GG12, GJMS13, BG15]. The problem considered here in our paper, precisely stated in Question 2.6, is about realizing the conformal blocks themselves geometrically as sections of ample line bundles on polarized varieties (for a fixed curve). These problems are distinct from one another. Acknowledgments. P.B. was supported on NSF grant DMS-0901249, and A.G. on DMS-1201268 and in part by DMS-1344994 (RTG in Algebra, Algebraic Geometry, and Number Theory, at UGA). We thank the anonymous referee for their time and thoughtful feedback. 2. Notation and basic definitions 2.1. Conformal blocks. For a positive integer ℓ (called the level), we let Pℓ(slr+1) denote the set of dominant integral weights λ with (λ, θ) ≤ ℓ. Here θ is the highest root, and ( , ) is the Killing form, normalized so that (θ, θ) = 2. To a triple (slr+1, ~λ, ℓ), such that ~λ ∈ Pℓ(slr+1)n, there corresponds a vector bundle V of conformal blocks on the stack Mg,n [TUY89, Fak12]. Throughout the paper, as our notation is fixed, we usually just refer to such bundles as V. We recommend the Bourbaki article of Sorger [Sor96] and the survey article of Beauville [Bea96] for some of the background on conformal blocks. By writing V∗ x (cid:27) H0(Xx, Lx), for x ∈ Mg,n, we mean to indicate that Xx = Xx(slr+1, ~λ, ℓ) and Lx = Lx(slr+1, ~λ, ℓ). Definition 2.1. Given a weight µ ∈ Pℓ(g), let Vµ denote the corresponding irreducible representation of g. By µ⋆ ∈ Pℓ(g) we mean the highest weight in (Vµ)∗. Theorem 2.2. [TUY89, Factorization] Let (C0; ~p) be a stable n-pointed curve of genus g where C0 has a node x0. (1) If xo is a non-separating node, ν : C → C0 the normalization of C0 at x0, and ν−1(x0) = {x1, x2}, then V(g, ~λ, ℓ)(C0;~p) (cid:27) Mµ∈Pℓ(g) V(g, ~λ ∪ µ ∪ µ⋆}, ℓ)(C;~p∪{x1,x2}). (2) If x0 is separating, ν : C1 ∪ C2 → C0 the normalization at x0, ν−1(x0) = {x1, x2}, with xi ∈ Ci, then V(g, λ(C1) ∪ {µ}, ℓ)(C1;{pi∈C1}∪{x1}) ⊗ V(g, λ(C2) ∪ {µ⋆}, ℓ)(C2;{pi∈C2}∪{x2}), V(g, ~λ, ℓ)(C0;~p) (cid:27) Mµ∈Pℓ(g) where λ(Ci) = {λ jpj ∈ Ci}. Definition 2.3. The collection of weights {(µ, µ∗) : µ ∈ Pℓ(g)} which appear in Theorem 2.2, and which give nonzero summands are called the restriction data for a bundle V(g, ~λ, ℓ) at (C0; ~p). In many of the examples and computations done here, we find the ranks of conformal blocks bundles using the cohomological form of Witten's Dictionary, which expresses these ranks as the intersection numbers of particular classes (depending on the bundle) in the small quantum cohomology ring of certain Grassmannian varieties. Theorem 2.4. [Bel08, Cohomological version of Witten's Dictionary] ForP λi = (r + 1)(ℓ + s). • If s < 0, then the rank of V(slr+1, ~λ, ℓ) coincides with the rank of classical coinvariants A(slr+1, ~λ). • If s ≥ 0, let λ = (ℓ, 0, . . . , 0). The rank of V(slr+1, ~λ, ℓ) is the coefficient of qs[pt] = qsσ(ℓ,ℓ,...,ℓ) in σλ1 ⋆ · · · ⋆ σλn ⋆ σs λ ∈ QH∗(Y), Y = Gr(r + 1, r + 1 + ℓ), λ is the s-fold quantum ⋆ product of σλ. One can write the above multiplicity also as the where σs coefficient of qsσλc n (λc n is the complement of λn in a (r + 1) × ℓ box) in σλ1 ⋆ σλ2 ⋆ · · · ⋆ σλn−1 ⋆ σs λ ∈ QH∗(Y). 4 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA The relation to quantum cohomology follows from [Wit95] and the twisting procedure of [Bel08], see Eq (3.10) from [Bel08]. Examples of such rank computations were done using Witten's Dictionary in [BGM15a, Kaz16, BGM15b], and [Hob15]. 2.2. V[m] and the algebra of conformal blocks. Definition 2.5. For λi = Pr V = V(slr+1, ~λ, ℓ), set j=1 c jω j ∈ Pℓ(slr+1), and m ∈ N, set mλi = Pr j=1(mc j)ω j ∈ Pmℓ(slr+1). Given where m~λ = (mλ1, . . . , mλn) ∈ Pmℓ(slr+1)n. We often refer to the new bundles V[m] as multiples of V, and we say they are obtained by scaling the Lie data used to form V. V[m] = V(slr+1, m~λ, mℓ), Using the V[m], one can form a flat sheaf of algebras (see [Fal94, p. 368], [Man09]): (2.1) A = Mm∈Z≥0 Am =Mm≥0 V[m]∗, over Mg,n. At any interior point x ∈ Mg,n, one has a natural identification of graded algebras (2.2) Ax = Mm∈Z≥0 (Am)x =Mm≥0 V[m]∗x (cid:27)Mm≥0 H0(Xx, L⊗m x ). Here, for x = (C, p1, . . . , pn) ∈ Mg,n, one has that Xx = Xx(slr+1, ~λ, ℓ) is a moduli space, constructed using GIT, of parabolic bundles of rank r + 1 with trivial determinant on the curve C with parabolic structures at p1, . . . , pn, and Lx = Lx(slr+1, ~λ, ℓ) an ample line bundle. We ask if a similar description for the algebra of conformal blocks holds at points x ∈ Mg,n \ Mg,n: Question 2.6. Given a vector bundle of conformal blocks V on Mg,n in type A, and x ∈ Mg,n \ Mg,n is there a polarized scheme (Xx, Lx) such that Eq (2.2) holds as graded algebras? Definition 2.7. Given a vector bundle of conformal blocks V on Mg,n, if the answer to Question 2.6 is yes for V, then we will say that V has geometric interpretations at boundary points x ∈ Mg,n \ Mg,n. Remark 2.8. (1) We note that Ax for x ∈ Mg,n is an integral domain, since it is a subalgebra (formed by suitable Lie-algebra invariants) of the algebra of sections of a line bundle on the ind-integral affine Grassmannian times an n-fold product of complete flag varieties (see e.g., [LS97, Section 10]). Therefore, the Xx are necessarily irreducible varieties when they exist. (2) The algebras Ax are finitely generated for x ∈ Mg,n, since the coordinate ring of a polarized variety is finitely generated. We do not know if the Ax are finitely generated for x ∈ Mg,n \ Mg,n. If Ax is finitely generated for x ∈ Mg,n \ Mg,n then, Xx (if it exists) will coincide with Proj(Ax). However Proj(Ax) need not carry an ample line bundle Lx whose section ring equals Ax. Note that by Remark 2.8, such Xx (if they exist) are necessarily reduced and irreducible. Moreover, in this question we do not require that (Xx, Lx) fit together into a flat family. Question 2.6 is a point-wise version of the question of existence of a family π : X → Mg,n with relatively ample bundles L such that π∗L⊗m = V[m]∗ consistent with multiplication operations. One could then hope to (recursively) control the Chern classes of V[m] by applications of the Grothendieck-Riemann-Roch formula. In the situation when ∆(Xx, Lx) = 0 for x ∈ Mg,n (described next in Def 2.9), the point-wise question is implied by the family question by cohomology vanishing ([Fuj90, Chapter 1, (5.1)] and Remark 3.4). SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 5 2.3. The ∆ -- invariant and projective varieties of minimal degree. Definition 2.9. Let X be an irreducible projective variety, and L an ample line bundle on X. (1) The self intersection Ldim(X) is dim(X)! times the leading coefficient of the Hilbert polynomial f (m) = χ(X, L⊗m) (see [Fuj90, 3.11], for example). The self intersection D = Ldim(X) is also denoted called the degree of the polarized variety (X, L). (2) The ∆-invariant of the polarized variety (X, L) is given by the formula We refer to ∆(X, L) as the ∆-genus or as the ∆-invariant of the pair (X, L). ∆(X, L) = dim(X) + Ldim(X) − h0(X, L). We note that by [Fuj90, Chapter 1, (4.2), (4.12)], and [BS95, Theorem 3.1.1], ∆(X, L) ≥ 0, and if ∆(X, L) = 0, then H0(X, L) generates the algebraLm≥0 H0(X, L⊗m) and hence L is very ample, giving rise to an embedding into a projective space X ֒→ Proj(B•) = PN, B• = Mm∈Z≥0 Symm(H0(X, L)), so that its image is a (nondegenerate) variety of degree equal to Ldim(X) = 1 + codim(X). Definition 2.10. A nondegenerate variety X ⊂ PN always has deg(X) ≥ 1 + codim(X), and is said to be of minimal degree if deg(X) = 1 + codim(X). Therefore, polarized varieties (X, L) with ∆(X, L) = 0 correspond to projective varieties of minimal degree. A description of these varieties has been given by many authors including the classification of minimal surfaces [DP86] (also [Nag60]), and the higher dimensional cases by Bertini in 1907 [Ber07], and subsequent treatments by Harris, Xamb ´o, Griffiths and Harris, Fujita, and Beltrametti and Sommese [Har81, Xam81, GH78, Fuj90, BS95]. We recommend [EH87] for a good historical account. Proposition 2.11. [Fuj90, Chapter 1, (5.10), (5.15)], [BS95, Proposition 3.1.2] We suppose that ∆(X, L) = 0 and d = dim X ≥ 2: (1) (X, L) (cid:27) (Pd, OPd(1)) if Ld = 1; (2) (X, L) (cid:27) (Q, OQ(1)), where Q is a not necessarily smooth quadric in Pd+1 if Ld = 2; (3) (X, L) is a Pd−1 bundle over P1, X (cid:27) P(E), for a vector bundle E on P1 which is a direct sum of line bundles of positive degrees; (4) (X, L) (cid:27) (P2, OP2(2)); or (5) (X, L) is a (normalized) generalized cone over a smooth subvariety V ⊂ X with ∆(V, LV) = 0, where LV denotes the restriction of L to V. For (1) in Proposition 2.11, we note the related paper [Gor68]. Definition 2.12. We say that V has ∆-invariant zero rank scaling if ∆(Xx, Lx) = 0 for x ∈ Mg,n such that V∗ x (cid:27) H0(Xx, Lx). Remark 2.13. The function f (m) = rk V[m] determines ∆(Xx, Lx): (1) f (1) = h0(Xx, Lx); (2) dim(Xx) = deg( f (m)); and (3) Ldim Xx x = c × dim Xx!, where c is the leading coefficient of f (m). x Moreover, Hi(Xx, L⊗m ) = 0 for i > 0, m ≥ 0 by [Tel00, Theorem 9.6], so the Hilbert polynomial is determined by the first few values of rk V[m]. The dimension of Xx can be bounded above, in genus 0 it is no more than the dimension of the corresponding flag variety. There is also an explicit formula for Ldim Xx , the degree of Lx due to Witten [Wit91] (also see Sections 3, 4 in [BBV15] for examples). x Definition 2.14. We refer to Ldim Xx x as the degree of the block V. 6 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA Example 2.15. If X has dimension one and ∆(X, L) = 0 then (X, L) = (P1, O(D)) for some D > 0. Indeed, ∆(P1, O(D)) = 0. Conversely, 0 = ∆(X, L) = 1 + deg(L) − h0(L). Riemann-Roch gives that χ(X, L) = 1 − ga(X) + deg(L). Hence h1(X, L) = −ga(X) = 0, where ga(X) is the arithmetic genus of X. Definition 2.16. We refer to the various types of ∆-invariant zero rank scaling according to the classifi- cation of projective varieties of minimal degree described in Proposition 2.11. For example: (1) rk(V[m]) =(cid:0)d+m (2) rk(V[m]) = 2(cid:0)m+d−1 m (cid:1), then V has projective rank scaling; d−1 (cid:1), then V has quadric rank scaling; (cid:1) +(cid:0)m+d−1 (3) rk(V[m]) = (m + 1)(1 + m(a+b) (4) rk(V[m]) = (m + 1)(2m + 1), then V has Veronese surface scaling; and (5) rk(V[m]) = dm + 1, then V has (P1, O(d)), or rational normal curve scaling. ), then V has (S(a, b), O(1)) scaling1; 2 d 3. Divisor class formulas in case extensions exist Here we prove the following result, which is the full statement of Theorem 1.2, from the introduction. Theorem 3.1. Suppose that V has ∆-invariant zero rank scaling, and assume that geometric interpreta- tions exist for V at all points (see Definition 2.7). Let D = deg(Lx) for x ∈ Mg,n, and R = rk V. Then there exist vector bundles Wi, i = 1, . . . , D on Mg,n such that for all integers m ≥ 1, (3.1) c1(V[m]) = m + R − 1 R ! + (−1)i−1(i − 1) D DXi=2 i! m − i + R − 1 R !!c1(V) + (−1)i m − i + R − 1 R − 1 !c1(Wi). DXi=2 The first Chern class c1(Wi) can be written in terms of c1(V[m]), m = 1, . . . , i using Equation (3.1). Therefore Equation (3.1) expresses c1(V[m]) in terms of c1(V[1]), . . . , c1(V[D]) for all m. Remark 3.2. (1) Crucially, if V has ∆-invariant zero rank scaling at some x, and if geometric inter- pretations exist for V at all points (see Definition 2.7), then ∆(Xx, Lx) = 0 for all points x ∈ Mg,n. (2) More generally, in any family of polarized varieties, the ∆-invariant is upper semicontinuous (i.e., can only go down at special points). 3.1. Outline of proof. The short alternate proof in the case of quadrics in Corollary 6.10 contains all the essential ideas. The proof of Theorem 3.1 proceeds in two steps. (1) Suppose (X, L) is a polarized variety of ∆-invariant zero. Then we have a canonical embedding of Symm(H0(X, L)). In Section 3.2, we recall X in a projective space X ֒→ Proj(B•) with B• =Lm∈Z≥0 the canonical resolution of the ideal sheaf IX. (2) In Section 3.3, we glue these resolutions of the ideal sheaves corresponding to (Xx, Lx) for all x ∈ Mg,n. This will result in an exact sequence (3.6) which yields (3.1), using Remark 3.3. Remark 3.3. If V is a vector bundle of rank R then rk(Symm V) =(cid:0)m+R−1 R−1 (cid:1), and c1(Symm V) =(cid:0)m+R−1 3.2. Resolutions of ideals of projective varieties of minimal degree. If the pair (X, L) is a polarized variety having ∆-invariant zero, then L is very ample on X, then from L one obtains an embedding X ֒→ Proj(B•), as a variety of minimal degree. In particular, the ideal IX, of X is the kernel of the surjective map R (cid:1)c1(V). φ : B• = Mm∈Z≥0 Symm(H0(X, L)) ։ A• = Mm∈Z≥0 H0(X, L⊗m), and IX is known to have a minimal free resolution of the form (3.2) 0 → WD ⊗ O(−D) → · · · → W3 ⊗ O(−3) → W2 ⊗ O(−2) → IX → 0 1One can write down a more general sequence for the scrolls (S(a1, . . . , ad) = P(E), O(1)), discussed in Section 6.4. SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 7 where for i ∈ {1, . . . , D}, the Wi are vector spaces of dimension (i − 1)(cid:0)D [Nag07, EG84]. i(cid:1), and D = LdimX is the degree of L Remark 3.4. Let Jk be the image of the map Wk+1 ⊗ O(−k − 1) → Wk ⊗ O(−k) in (3.2). Then the exactness of (3.2) implies that Hi(PN, Jk(m)) = 0 for all m ≥ 0 and i > 0, and for i = 0 and m < k + 1. This is proved by induction using the exact sequences 0 → Jk → Wk ⊗ O(−k) → Jk−1 → 0. Note that this also implies that Hi(PN, IX(m)) = Hi(X, L⊗m) = 0 for all for i ≥ 1, and m ≥ 0 and H0(X, L⊗m) = H0(PN, O(m))/H0(PN, IX(m)). Minimal free resolutions in the context of graded rings are known to be unique up to unique iso- morphisms. The same technique shows that the complex (3.2) is unique up to a unique isomorphism. Twisting (3.2) by O(k) with 2 ≤ k ≤ D, and then taking global sections gives an exact sequence (see Remark 3.4) (3.3) 0 → Wk → Wk−1 ⊗ Sym1(B1) → . . . → W2 ⊗ Symk−2(B1) → H0(PN, I(k)) → 0. Therefore, Wk can be reconstructed uniquely, as a kernel, by induction. Lemma 3.5. Given a pair (X, L), having ∆-invariant zero, with notation as above, there are canonical vector spaces Wi, for 2 ≤ i ≤ D, and a long exact sequence of the form: (3.4) 0 → WD ⊗ B•(−D) φi→ · · · → W3 ⊗ B•(−3) giving a graded resolution of the graded ideal I• =Lm≥0 φi→ · · · → W3 ⊗ Bm−3(−3) 0 → WD ⊗ Bm−D(−D) (3.5) φ0→ I• → 0, φ1→ W2 ⊗ B•(−2) Im with Im = H0(PN, IX(m)). In particular φ1→ W2 ⊗ Bm−2(−2) φ0→ Im → 0 is a canonical resolution of the degree m part Im. 3.3. Resolutions in families. Suppose that V has ∆-invariant zero rank scaling, and assume that geo- metric interpretations exist for V at all points. Then for each x ∈ Mg,n, we have a resolution of the form 3.5. In this section we will show that we can glue these resolutions, and get an exact sequence over Mg,n of the form: (3.6) 0 → WD ⊗ Symm−D(V) → · · · → W2 ⊗ Symm−2(V) → Symm(V) φm→ V[m]∗ → 0, where Wi are vector bundles on Mg,n. This will finish the proof of Theorem 3.1. For a given V, we form the following sheaves on Mg,n: Let A = Lm≥0 A0 = O. Let B• =Lm∈Z≥0 Am, where Am = V[m]∗ and Bm, where Bm = Symm A1. Consider φm : Bm = Symm(A1) −→ Am, and Im = ker(φm). Clearly I• is a graded B• ideal. Proposition 3.6. There is an exact sequence of the form (3.6) with fibers equal to the resolution of the ker(φm)x, given in Lemma 3.5. Furthermore, (3.6) is exact and Wi are locally free of finite rank. Proof. We first build a resolution (3.7) Let I0 • = I•. We note that I0 · · · −→ Fk → · · · → Fk−1 → · · · → F0 → I•. • is a graded B•-module. Because I0 I0 2 ⊗ Bm−2 α0 m−→ I0 m, and we set I1 m = ker(α0 m). Now we consider the morphism I1 3 and we set I2 m = ker(αi−1 Ii m ). m = ker(α1 m). At the i-th step we consider the morphism Ii−1 i+1 ⊗ Bm−(i+1) • is an ideal, there is a morphism α1 m−→ I1 m, ⊗ Bm−3 αi−1 m−→ Ii−1 m , and set 8 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA The terms in (3.7) are defined as follows. The sheaf F0 isLm F1 isLm ⊗ Bm−3 and Fi isLm ⊗ Bm−i−2. Finally set Wi = Ii (3.6). I0 2 I1 3 Ii i+2 ⊗ Bm−2 with Bi = 0 for i < 0. The sheaf i+2. This is the sheaf that appears in The map F0 → I• is clear because I• is a graded B• module. The sheaf F1 maps to F0 since I1 3 sits inside F0. One can define the maps in (3.7) by continuing this process. The general process is the following: if f : M → N is a map of graded modules for a graded ring R, then for any positive integer s, ker( fs) ⊗ R(−s) → M → N is a complex (possibly non-exact). Working on the (reduced) scheme {x}, we form a complex similar to (3.7). Let A′ m = (Am) ⊗ k(x). (3.8) · · · −→ F ′ k m : B′ → · · · → F ′ k−1 → · · · → F ′ 0 → I′ •. Here I′ • is ker(φ′ m) where φ′ m. From our assumption on ∆-invariants (and m are surjective. Note that for k ≥ 2, I′k−2 Section 3.2) it follows that (3.8) is exact, and that the maps φ′ coincides with Wk in Lemma 3.5 (as a suitable kernel from (3.3)). Now Im is a kernel of a surjective map of vector bundles since φ′ m are surjective, and the map I• ⊗ k(x) → I′ • an isomorphism. k m = Symm(A′ 1) −→ A′ For the inductive step, fix a point x, and suppose ψ : G• → H• is morphism of graded B• modules. Let F• be the kernel of ψ and F ′ • the kernel of ψ ⊗ k(x) and s ∈ Z. Assume • Gm and Hm are vector bundles for all m; • images of ψm are locally free subbundles of Hm in the localization Spec(Ox); and • the three term sequence on fibers is exact: F ′ •(−s) → G• ⊗ k(x) → H• ⊗ k(x). s ⊗ B′ Then Fs ⊗ B•(−s) → G• → H• is exact over Spec(Ox), and hence Fm is the image of Fs ⊗ Bm−s in Gm, and is a subbundle of the latter. The next step in the induction is with ψ the map Fs ⊗ B•[−s] → G•. We work over Spec(Ox). Note that 0 → Fm → Gm → im(ψ)m → 0 is an exact sequence, and therefore, Fm are vector bundles. Now tensoring it by k(x) we obtain another exact sequence. Note that im(ψ)m ⊗ k(x) → Hm ⊗ k(x) is injective since im(ψ)m → Hm is a sub-bundle. This tells us that Fm ⊗ k(x) → F ′ m is an isomorphism. Finally Fs ⊗ Bm−s → Fm is a surjection of vector bundles because it is so after tensoring with k(x) by our third hypothesis. (cid:3) Remark 3.7. For the validity of Equation (3.1), we only need geometric interpretations on the generic points of all boundary divisors, which would imply that Equation (3.1) holds in Pic(Mg,n \ Z) with Z of codimension 2. Since Picard groups are unaffected by removal of codimension 2 substacks, Equation (3.1) would then hold on Mg,n. In this section we prove Theorem 1.1, which we restate below: 4. Cases when obstructions are non-zero Theorem. There are stable n-pointed curves x = (C; p1, . . . , pn) and vector bundles of conformal blocks V, for which there are no polarized pairs (Xx, Lx) such that Mm≥0 V[m]∗ x (cid:27)Mm≥0 H0(Xx, L⊗m x ), holds. Proof. If such an isomorphism were to exist, then if V has ∆-invariant zero rank scaling, by Thm 3.1, c1(V[m]) = m + R − 1 R ! + (−1)i−1(i − 1) D DXi=2 i! m − i + R − 1 R !!c1(V) + (−1)i m − i + R − 1 DXi=2 R − 1 !c1(Wi). One can determine the c1(Wi) in terms of c1(V), . . . , c1(V[i]) by setting m = 2, . . . , D in Equation (3.1). When one substitutes m = i with 2 ≤ i ≤ D, then Equation (3.1) is of the form (−1)ic1(Wi) plus a linear expression in c1(W2), . . . , c1(Wi−1), and c1(V). These identities can be shown to fail, as in Examples 4.2 and 4.4. (cid:3) SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 9 Remark 4.1. Our techniques are not limited to bundles with ∆-invariant zero rank scaling. Identities analogous to Equation (1.2) exist when one has enough knowledge about pairs (Xx, Lx) at points x on the boundary, if they exist. To illustrate, we give identities where Xx could be embedded by Lx as Coble's quartic hypersurface in P7 (Proposition 5.1), Coble's cubic hypersurface in P8 (Proposition 5.2), and where Xx is the intersection of two quadrics in P5 (Proposition 5.3). 4.1. Examples. Example 4.2. Consider V = V(sl2, 1) on M2. By [NR69], for x ∈ M2 one has V∗ x (cid:27) H0(Xx, Lx), where (Xx, Lx) (cid:27) (P3, O(1)). By Corollary 3.1 (also see Proposition 6.6), if V has geometric interpretation at boundary points (see Definition 2.7), then (4.1) c1(V[m]) = m + 3 4 ! c1(V) = (m + 3)(m + 2)(m + 1)m 24 · c1(V) which we can show fails by intersecting with F-curves. Remark 4.3. We will however show in Example 6.8 that geometric interpretations do indeed hold along the divisor ∆0 of M2, and fail along ∆1,∅. There are two types of F-curves on M2: (1) the image of a map from M0,4 for which points are identified in pairs; (2) the image of a map from M1,1 given by attaching a point (E, p) ∈ M1,1, gluing the curves at the marked points. We see a contradiction when we intersect with either type of F-curve. Intersecting both sides of Equation 4.1 with the F-curve defined by M0,4. By the formulas of Fakhruddin [Fak12], we can see that if Dm = 0, then the intersection of Equation 4.1 with a pigtail type F-curve would imply the following identity: (4.2) X0≤λ,µ≤m deg V(sl2, {λ, λ, µ, µ}, m) = (m + 3)(m + 2)(m + 1)m 24 . Here we note λ∗ = λ for sl2. But Equation 4.2 doesn't hold, for example, for m = 2, the right hand side equals 5, whereas there are three non-zero terms on the left hand side: (4.3) deg V(sl2, {ω1, ω1, 2ω1, 2ω1}, 2) = deg V(sl2, {2ω1, 2ω1, ω1, ω1}, 2) = 1, deg V(sl2, {2ω1, 2ω1, 2ω1, 2ω1}, 2) = 2, which add up to 4. We note that using Witten's dictionary, Theorem 2.4, one can verify that X0≤λ,µ≤m rk V(sl2, {λ, λ, µ, µ}, m) = (m + 3)(m + 2)(m + 1) 6 . Intersecting both sides of Equation 4.1 with the F-curve defined by M1,1. If Equation 4.1 held, then by pulling back along the map from M1,1 onto the second type of F-curve, one would have that (4.4) mXµ=0 µ c(m) r(m) µ∗ = (m + 3)(m + 2)(m + 1)m 24 · (r(1) 1 c(1) 1 + r(0) 0 c(0) 0 ) = − (m + 3)(m + 2)(m + 1)m 24 · 2 · 1 6 , µ and c(m) where r(m) know that r(m) [Fak12, Corollary 6.2]: if µ is even, c(m) µ are ranks and first Chern classes of the bundle V(sl2, µ, m) on M1,1, respectively. We is given by µ = 0 if µ is odd and equals m + 1 − µ if µ is even. The degree c(m) µ = c(m) µ∗ µ = −(cid:16) µ2−3mµ+2m2−µ+2m 24 (cid:17). Consider m = 2, then µ c(m) r(m) µ∗ = r(2) 0 c(2) 0 + r(2) 2 c(2) 2 = −3 − 1 = − 1 2 2 24 19 12 . mXµ=0 10 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA The right hand side of Equation (4.4) is − 10 6 . We again see that Equation (4.1) fails. }, 1) on M2,n if n is even has the property that for x ∈ M2,n one x (cid:27) H0(Xx, Lx), where (Xx, Lx) (cid:27) (P3, O(1)). We note that V(sl2, 1) on M2 does not pullback to Example 4.4. The bundle V = V(sl2, {ωn 1 has V∗ V = V(sl2, {ωn 1 }, 1). If V has geometric interpretation at boundary points, then (4.5) c1(V[m]) = m + 3 4 ! c1(V) = (m + 3)(m + 2)(m + 1)m 24 · c1(V) which we show fails by intersecting with the F-curve given by the image of a clutching map from M0,4 for which the first two points are identified, and the second two points are attached to a point in M0,n+2. The calculation is analogous to the one above, the identity failing already at m = 2. Remark 4.5. By [KP95, Theorem E], given V on Mg,n with n = 0, the corresponding polarized varieties (Xx, Lx) cannot be of Delta invariant zero, for genus g ≥ 3, x ∈ Mg. 5. Conjectural formulas in non ∆-invariant 0 examples To illustrate that the techniqes introduced in this work are applicable in a context more general than for the class of bundles with ∆-invariant zero rank scaling, we work in three other situations where there is a conjectural description of the polarized pairs (Xx, Lx) at points x on the boundary of Mg,n. We note that Theorem 1.2 gives a stronger obstruction than what is obtained here because as we have stated earlier, if ∆((Xx, Lx)) = 0, for some x, and if there are geometric interpretations at points on the boundary, then given (Xx, Lx) for any x ∈ Mg,n, the variety Xx will be projective of minimal degree, and hence classified. Therefore, if there is failure of the identities predicted, then one may conclude that geometric interpretations did not actually exist at boundary points. On the other hand, if the identities from Propositions 5.1, 5.2 or 5.3 fail, it could be either because there are no such polarized pairs (Xx, Lx) at boundary points x, or because if (Xx, Lx) exists, assumptions about the variety (Xx, Lx) were incorrect. We also note that for these examples we have used a script CBRestrictor [Kra15], which compute the ranks of conformal blocks bundles over curves of positive genus, using [Swi10]. [C] 5.1. Coble's quartic hypersurface. Given a smooth nonhyperelliptic curve C of genus 3, one has that V(sl2, 1)∗ (cid:27) H0(Z4, L), where Z4 is Coble's quartic hypersurface in P7 [NR87]. Assuming that geometric interpretations hold at all points of M3 − ∆1, in addition that over M3 − (∆1 ∪ H3), the corresponding "moduli space" remains a quartic hypersurface in P7, we give in Proposition 5.1, an identity which must be satisfied by the first Chern classes of multiples V[m] = V(sl2, m). Using [MOP15, Theorem 3], we have checked the predicted relations hold in the first 7 nontrivial cases. Proposition 5.1. Let V[m] = V(sl2, m) on M3. Assuming that for all points [C] ∈ M3 − ∆1, corresponding to nonhyperelliptic curves C, one has that V(sl2, 1)∗ [C] (cid:27) H0(Z4, L), where Z4 is a quartic hypersurface in P7, then c1(V[m]) + Dm = 7 + m m − 1! − m + 3 8 ! − 165 m + 3 7 !! c1(V) + m + 3 7 !c1(V[4]) and where Dm is a linear combination of the class of the hyperelliptic locus H3 and δ1. Proof. We work over X = M3 \ (∆1 ∪ H), and show that the class of Dm restricted to X is zero. Let Im (with m ≥ 2) be the kernel of the surjective map Symm V∗ → V[m]∗ SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 11 which can be interpreted for x ∈ X as the kernel of map H0(P7, O(m)) → H0(Xx, Lm). Therefore (Im)x is the space of sections of the ideal sheaf of Xx. Clearly Im is a vector bundle over X. Note that (5.1) c1(Im) = c1(V[m]) − 7 + m m − 1!c1(V) We also have isomorphisms I4 ⊗Symm−4 V∗ → Im (the ideal of a quartic is generated in degree 4). Taking Chern classes using Equation (5.1), we get the desired vanishing: Dm X= 0. (cid:3) A careful check, using [MOP15, Theorem 3], keeping in mind their result gives the slope of the divisor, rather than its class, one obtains: (1) c1(V[1]) = 4 λ − δirr; (2) c1(V[2]) = 27 λ − 8 δirr − 3 δ1; (3) c1(V[3]) = 108 λ − 37 δirr − 16 δ1; (4) c1(V[4]) = 329 λ − 128 δirr − 64 δ1; (5) c1(V[5]) = 840 λ − 366 δirr − 192 δ1; (6) c1(V[6]) = 1890 λ − 912 δirr − 502 δ1; (7) c1(V[7]) = 3864 λ − 2046 δirr − 1152 δ1; (8) c1(V[8]) = 7326 λ − 4224 δirr − 2438 δ1; (9) c1(V[9]) = 13068 λ − 8151 δirr − 4752 δ1; So that as we expect from the predicted relations: • 9c1(V[1]) − c1(V[2]) = H + 6δ1; • 45c1(V[1]) − c1(V[3]) = 8 (H + δ1); • 826 c1(V[1]) + c1(V[5]) − 8 c1(V[4]) = 168 H + 824 δ1; • 4662 c1(V[1]) + c1(V[6]) − 36 c1(V[4]) = 966 H + 3384 δ1; • 16842 c1(V[1]) + c1(V[7]) − 120 c1(V[4]) = 3528 H + 17112 δ1; • 48180 c1(V[1]) + c1(V[8]) − 330 c1(V[4]) = 10164 H + 49174 δ1; and • 118305 c1(V[1]) + c1(V[9]) − 792 c1(V[4]) = 25080 H + 121176 δ1. Here H = 9λ − δirr − 3δ1 is the class in the Picard group of the stack M3, of the effective divisor H of hyperelliptic curves in M3. The presence of non-zero coefficients of δ1 is evidence of a failure of geometric interpretations on the boundary divisor ∆1. 5.2. Coble's cubic hypersurface. Coble also considered Z3 ֒→ P8 = P(H0(Jg−1C, OJC(3θ)), for C a smooth (cid:27) H0(SUC(3), L), and SUC(3) curve of genus 2, a cubic hypersurface [Cob03]. It is known that V(sl3, 1)∗ [C] is a degree 2 cover of P8 branched over a sextic dual to Z3 [Ort05, Ngu07]. Proposition 5.2. Let V = V{0}(sl3, 1) on M2. If geometric interpretations exist for V at all points of M2, with every Xx isomorphic to some degree 2 cover of P8 branched over a sextic (and the line bundle pulled back from P8), then for all m ≥ 1, one has that c1(V[m]) = ( m + 8 9 ! + m + 5 8 !c1(V[3]) + Dm, 9 ! − 55 m + 5 8 !)c1(V) + m + 5 where Dm is the anomalous divisor supported on the boundary. In particular, if Conjecture 1.3 is sharp, Dm to be supported entirely on ∆1,φ. The proof uses the projection formula, and the following fact: For π : Y → P8 of degree 2 branched along a sextic, one has an isomorphism π∗O = OL(L ⊗ O(−3)) (for some line L). Using [Fak12] and [MOP15] we computed c1(V[1]) = 9λ − 2 δ0, c1(V[2]) = −11δirr + 9δ1, c1(V[3]) = 332 λ − 94 δ0 − 34 δ1, c1(V[4]) = 1152 λ − 361 δ0 − 153 δ1 and c1(V[5]) = 3330 λ − 964 δ0 − 738 δ1. We note that the relations given in Proposition 5.2 hold for the first three (nontrivial) cases: D2 = 9δ1; D4 = c1(V[4]) + 274 c1(V) − 9 c1(V[3]) = 279 δ1; and D5 = c1(V[5]) + 1750 c1(V) − 45 c1(V[3]) = 1020 δ1. The presence of non-zero coefficients of δ1 provides evidence towards failure of geometric interpretations on the boundary divisor ∆1,φ. 12 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA 5.3. An intersection of two quadrics in P5. Here we consider V[m] = V(sl2, {2mω1}, 2m) on M2,1. By x (cid:27) H0(Xx, Lx), where Xx is the intersection of two quadrics in P5, when [DR76] (and Hecke transforms) V∗ x = (C, p) is a smooth pointed genus 2 curve. Here we conjecture that the same geometric interpretation holds for any stable x = (C, p). Proposition 5.3. Suppose that the geometric interpretation of conformal blocks extends to points on the boundary of M2,1, with (Xx, Lx) an intersection of quadrics in P5 for all x ∈ M2,1. Then, (5.2) c1(V[m]) = (m − 4)(m − 2)m(m + 1)(7m − 5) 12 c1(V[1]) + ( m + 3 5 ! − 21 m + 1 5 !)c1(V[2]) + m + 1 5 !c1(V[4]). Proof. If X is a intersection of two quadrics in P = P5 = P(V), we have a map W = Q ⊗ OP(−2) → OP, rk W = 2. which we can view as a section s of W∗. Now X = Z(s) has (pure) dimension 3, so s corresponds to a regular sequence, and we have a Koszul complex 0 → ∧2W → W → OP → OX → 0 and hence 0 → L ⊗ OP(m − 4) → Q ⊗ OP(m − 2) → OP → OX(m) → 0 where L = det Q. Now take global sections and use standard cohomology vanishing to get an exact sequence: 0 → L ⊗ Symm−4 V∗ → Q ⊗ Symm−2 V∗ → Symm V∗ → V[m]∗ → 0 where V[m]∗ = H0(X, OX(m)). Imagining a family of such X's over a base S we see that L and Q form vector bundles over S. Therefore, (5.3) c1(V[m]) = ( m + 5 6 ! − 2 m + 3 6 ! + m + 1 6 !)c1(V[1]) + m + 3 5 !β 5 !α + m + 1 where α = −c1(Q) and β = c1(L) with α + β = 0. It is easy to see that α = c1(V[2]) − 7c1(V) by putting in m = 2 in Equation (5.3). Put m = 4 in the same equation and get c1(V[4]) = ((cid:0)9 6(cid:1) − 2(cid:0)7 6(cid:1))c1(V[1]) +(cid:0)7 5(cid:1)α + β. And hence we obtain formulas for α and β, and hence Equation (5.2). We also verify α + β = 0, as claimed above, using the numerical formulas below. (cid:3) We have checked Equation (5.2) in the first four (nontrivial) cases. To do so, we computed the first Chern classes of the first 7 multiples using the formula for the slope of c1(V) given in [MOP15, Theorem 3]. These are: 2 δ1; 4 δirr − 3 2 λ + 3 ψ1 − 5 2 λ + 66 ψ1 − 91 (1) c1(V[1]) = 9 (2) c1(V[2]) = 19 λ + 19 ψ1 − 7 δirr − 8 δ1; (3) c1(V[3]) = 99 4 δirr − 51 (4) c1(V[4]) = 102 λ + 170 ψ1 − 281 (5) c1(V[5]) = 365 2 λ + 365 ψ1 − 469 (6) c1(V[6]) = 297 λ + 693 ψ1 − 218 δirr − 240 δ1; and (7) c1(V[7]) = 903 With the help of the relation λ = 1 the following identities are satisfied: 2 δ1; 5 δirr − 312 4 δirr − 259 4 δirr − 819 10 δirr + 1 2 λ + 1204 ψ1 − 1491 5 δ1; 2 δ1; 2 δ1. 5 δ1, proved by Mumford in [Mum83], we were able to verify • −16 c1(V[1]) + 6 c1(V[2]) − c1(V[3]) = 0; • 225 c1(V[1]) − 70 c1(V[2]) + 6 c1(V[4]) − c1(V[5]) = 0; • 1036 c1(V[1]) − 315 c1(V[2]) + 21 c1(V[4]) − c1(V[6]) = 0; and • 3080 c1(V[1]) − 924 c1(V[2]) + 56 c1(V[4]) − c1(V[7]) = 0. SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 13 6. Chern class identities for each type of projective variety of minimal degree In this section we make the identities from Theorem 3.1 explicit for each type of bundle with ∆- invariant zero rank scaling. For those bundles with projective rank scaling, we do a little more, giving both necessary and sufficient conditions for extensions. Examples for each type are given in Section 8. 6.1. Projective spaces as moduli. Definition 6.1. There is a natural morphism coming from the algebra structure on the sheaf of conformal blocks Tm : V[m] → Symm(V), after dualizing. Lemma 6.2. Set d = rk V − 1, and suppose that V∗ equivalent: x (cid:27) H0(Xx, Lx) for x ∈ Mg,n. The following are (a) rk(V[m]) =(cid:0)m+d d (cid:1) for all m. (b) (Xx, Lx) = (Pd, O(1)) for all x ∈ Mg,n. (c) The maps Tm are isomorphisms over Mg,n. Remark 6.3. If d = 0, the condition Lemma 6.2 (a) is automatically true; it is a consequence of a quantum generalization of Fulton's conjecture in representation theory (see [BGM15a] and the references therein). Statement (c) in this case was proved in [BGM15a, Corollary 2.2]. Proof. It is easy to see that (a) implies (b) by Proposition 2.11,(1). For (c), note that on fibers over x ∈ Mg,n, Tm, from Definition 6.1, is dual to the maps φm : Symm(H0(Xx, Lx)) → H0(Xx, L⊗m x ) which are easily verified to be isomorphisms under the assumption (b). So (b) implies (c). It follows that (c) implies (a) by counting dimensions. (cid:3) Recall (Def. 2.16, and Proposition 2.11) that V is said to have projective rank scaling if it has ∆-invariant zero rank scaling with degree 1. Proposition 6.4. If V has projective rank scaling, then for m ≥ 1 (6.1) m + d d + 1!c1(V) = c1(V[m]) + Dm, where Dm is an effective Cartier divisor supported on Mg,n \ Mg,n. Note that D1 = 0. Proof. It follows from Lemma 6.2 that if a bundle V has projective rank scaling, the maps Tm are isomorphisms on Mg,n. Taking determinants, we find a map det V[m] → det Symm(V). We obtain a global section of (det V[m])−1 ⊗ det Symm(V), and can write (det V[m])−1 ⊗ det Symm(V) = O(Dm) for an effective Cartier divisor Dm supported on Mg,n \ Mg,n. This gives the assertion. (cid:3) Definition 6.5. We will call the divisor Dm in Equation 6.1 the projective space scaling anomaly. We make the following basic observation: Proposition 6.6. Suppose V has projective rank scaling. The following are equivalent. (a) For each x ∈ Mg,n, there exists a pair (Xx, Lx) of a projective scheme and an ample line bundle such that Ax (cid:27)Lm≥0 H0(Xx, L⊗m ). (b) Dm = 0 ∈ H2(Mg,n, Q) for all m ≥ 2. (c) The maps Tm are isomorphisms, so that V[m] = Symm(V) for all m. x 14 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA Proof. If (a) holds then as in Lemma 6.2, (Xx, Lx) = (Pd, O(1)) for all x ∈ Mg,n and Tm is dual to the maps Symm(H0(Xx, Lx)) → H0(Xx, L⊗m ), which are isomorphisms since (Xx, Lx) = (Pd, O(1)). Therefore Tm is an isomorphism over Mg,n and Dm = 0. Therefore (b) and (c) hold. If (b) holds then since Dm is an effective Cartier divisor, we conclude that O(Dm) = O and hence Tm is an isomorphism. If (c) holds we can take (Xx, Lx) = (P(Vx), O(1)), and therefore (a) holds. (cid:3) x Remark 6.7. Let x ∈ Mg,n. The proof of Proposition 6.6 shows that if V has projective space scaling, and x < Dm for all m ≥ 2, then there exists a pair (Xx, Lx) of a projective scheme and an ample line bundle such that Ax (cid:27)Lm≥0 H0(Xx, L⊗m x ). Example 6.8. In Example 4.2 we can compute the projective scaling anomalies Dm = α(m)∆0 + β(m)∆1,∅. By [Mum83], the intersection numbers of ∆0 and ∆1,∅, with the pig-tail F-curve, are −2 and 1 respectively; and with the other curve, the numbers are 1 and −1/12 respectively. Note that in [Fak12, Corollary 6.2], the degree of the Hodge bundle is 1/12. In Example 4.2, we get the equations −2α(2) + β(2) = 5 − 4 = 1 and α(2) − 1 12 . Therefore α(2) = 0 and β(2) = 1. Using [Fak12, Corollary 6.2] and B. Alexeev's formula [Swi09, Lemma 3.3], we obtain α(m) = 0, and that β(m) is equal to s(s+1)(2s2+2s−1) if m = 2s is even, and s(s+1)2(s+2) (the corresponding factorization is not quasi-rank one (Definition 7.2)). if m = 2s + 1 is odd. Therefore geometric interpretations extend across ∆0 12 β(2) = − 10 6 + 19 12 = − 1 6 3 6.2. Quadric hypersurfaces as moduli. Recall that if (Xx, Lx) is a polarized variety, such that the degree Ldim(Xx) = 2 and ∆(Xx, Lx) = 0, then by [BS95, Prop 3.1.2], (Xx, Lx) = (Q, OQ(1)) where Q is a quadric hypersurface in projective space. In this section we consider rank scaling properties and divisor class identities governing such bundles. Lemma 6.9. Set d = rk V − 2, and suppose that V∗ equivalent: x (cid:27) H0(Xx, Lx) for x ∈ Mg,n. The following are (a) rk(V[m]) = 2(cid:0)m+d−1 d (cid:1) +(cid:0)m+d−1 d−1 (cid:1). (b) (Xx, Lx) = (Q, OQ(1)), where Q is a (not necessarily smooth) quadric hypersurface in Pd+1. Proof. Assume (a). It follows that the dimension of Xx is d and the degree Ld x = 2. Therefore ∆(Xx, Lx) = d + 2 − (d + 2) = 0, and (b) follows from [BS95, Prop 3.1.2]. The implication (b) =⇒ (a) follows from the exact sequence (6.2) 0 → OPd+1(−2) ·Q → OPd+1 → OQ → 0. (cid:3) Recall (Def. 2.16, and Proposition 2.11)) that V is said to have quadric rank scaling if it has ∆-invariant zero rank scaling with degree 2. Corollary 6.10. If V has quadric rank scaling and if a geometric interpretation exists for V on the boundary of Mg,n, then (6.3) c1(V[m]) =(cid:16) m + d + 1 d + 2 ! − m − 2 + d + 1 d + 2 ! − (d + 3) m − 2 + d + 1 d + 1 !(cid:17)c1(V) + m − 2 + d + 1 d + 1 !c1(V[2]). Proof. This follows from Corollary 3.1. We indicate an alternate direct proof here. Define Im to the kernel of µm : Symm V∗ → V[m]∗. There are maps νm : I2 ⊗ Symm−2 V∗ → Im. Note that µm and νm are defined on Mg,n. In general Im may not be vector bundles. If geometric interpretations exist, νm are isomorphisms for all m > 2, and µm are surjections for all m by working over fibers, and using the exact sequence (6.2) (tensored with OPd+1(m)). Furthermore I2 would be a line bundle on Mg,n. Therefore we obtain formulas for m > 2 c1(I2) = c1(Sym2 V∗) − c1(V[2]∗) SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n and c1(I2) rk(Symm−2 V∗) + c1(Symm−2 V∗) = c1(Im) = c1(Symm(V)∗) − c1(V[m]∗) Putting all these together gives us the desired formula for c1V[m]. 15 (cid:3) 6.3. Veronese surfaces as moduli. Lemma 6.11. Suppose that V∗ x (cid:27) H0(Xx, Lx) for x ∈ Mg,n. The following are equivalent: (a) rk(V[m]) =(cid:0)2m+2 2 (cid:1) = (m + 1)(2m + 1); normal curve of degree 4 in P4. (b) Either Xx is smooth and (Xx, Lx) = (P2, O(2)), or (Xx, Lx) is a generalized cone over a rational Proof. It is easy to see that (b) implies (a). Assume (a). It follows that Xx is two dimensional, the degree L2 x = 4, and H0(Xx, Lx) = 6, so ∆(Xx, Lx) = 0. Now (b) follows from Proposition 2.11. (cid:3) Corollary 6.12. If V has Veronese surface rank scaling and if a geometric interpretation exists for V on the boundary of Mg,n, then for all m ≥ 1, c1(V[m]) = A1(m)c1(V) + A2(m)c1(V[2]) + A3(m)c1(V[3]) + A4(m)c1(V[4]), where 6 (cid:1) − 3(cid:0)m+1 5 (cid:1) − 6(cid:0)m+3 6 (cid:1) + 8(cid:0)m+2 6 (cid:1) +(cid:0)m+5 6 (cid:1); (1) A1(m) = −7(cid:0)m+3 5 (cid:1) + 20(cid:0)m+2 5 (cid:1) − 23(cid:0)m+1 (2) A2(m) =(cid:0)m+3 5 (cid:1) − 6(cid:0)m+2 5 (cid:1) + 15(cid:0)m+1 5 (cid:1); (3) A3(m) =(cid:0)m+2 5 (cid:1) − 6(cid:0)m+1 5 (cid:1); and (4) A4(m) =(cid:0)m+1 5 (cid:1). Ld be the natural ample line bundle on X. It is known that ∆(X, L) = 0. Let D =P ai and N = D + d − 1. Set 6.4. Rational normal scrolls as moduli. Suppose a1, . . . , ad are strictly positive integers, and let E = O(ai), a vector bundle on P1. Let X = S(a1, . . . , ad) = P(E), a projective bundle over P1. Let L = O(1) V = H0(X, L)∗, One can show that dim V = N + 1, and clearly X ֒→ PN = P(V). The varieties S(a1, . . . , ad) are called rational normal scrolls (See Section A2H in [Eis05]). i=1 When some ai are zero and the rest positive, O(1) is base point free but not ample on P(E). The image in P(H0(P(E), O(1))) is again denoted by S(a1, . . . , ad). In this case, ∆(S(a1, . . . , ad), O(1)) = 0. bundles on P1. Therefore, the polarized varieties (S(a1, . . . , ad), O(1)) are deformation equivalent for fixed O(ai) with fixed d andP ai lie in the same component of the moduli stack of The vector bundlesLd d andP ai, and such varieties will have the same rank sequence. One can check that for m ≥ 0, one has H0(S(a, b), O(m)) = (m + 1)(1 + m(a+b) The following is a special case of Corollary 3.1: i=1 ). 2 Corollary 6.13. Suppose V satisfies (S(1, 2), O(1)) scaling so that rk V[m] = (m + 1)(1 + 3m 2 ) for all positive integers m. Assume geometric interpretations exist at boundary points. Then for all m ≥ 1, we have c1(V[m]) = A1(m)c1(V) + A2(m)c1(V[2]) + A3(m)c1(V[3]), where (1) A1(m) =(cid:0)m+4 5 (cid:1) − 6(cid:0)m+2 (2) A2(m) =(cid:0)m+2 4 (cid:1) − 5(cid:0)m+1 (3) A3(m) =(cid:0)m+1 4 (cid:1). 4 (cid:1) + 12(cid:0)m+1 4 (cid:1); and 4 (cid:1) − 3(cid:0)m+2 5 (cid:1) + 2(cid:0)m+1 5 (cid:1); One says that V satisfies (P1, O(d)) scaling if rk V[m] = dm + 1 for all positive integers m. Corollary 6.14. If V satisfies (P1, O(3)) scaling, and if geometric interpretations exist at boundary points, (6.4) c1(V[m]) = A1(m)c1(V) + A2(m)c1(V[2]) + A3(m)c1(V[3]), where A1(m) =(cid:0)m+3 4 (cid:1) − 5(cid:0)m+1 3 (cid:1) − 3(cid:0)m+1 4 (cid:1) + 8(cid:0)m 3(cid:1) + 2(cid:0)m 4(cid:1), A2(m) =(cid:0)m+1 3 (cid:1) − 4(cid:0)m 3(cid:1), and A3(m) =(cid:0)m 3(cid:1). 16 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA 7. Towards extension criteria In this section we turn our attention towards finding, in limited cases, conditions which guarantee that geometric interpretations of conformal blocks hold at boundary points. For this we need Definition 2.12 together with a few additional terms. 7.1. Notation for Theorem 7.5. We recall that the weights that are used at attaching points when "factor- izing" a bundle V(g, ~λ, ℓ) at a point (C0; ~p) using Theorem 2.2, are called the restriction data for V(g, ~λ, ℓ) at (C0; ~p) (see Definition 2.3). Definition 7.1. Given a bundle V = V(slr+1, ~λ, ℓ) on Mg,n, we say that the restriction data for V is free, if given any boundary point x ∈ ∆g1,J = ∆g−g1,Jc or x ∈ ∆irr, and α1, . . ., αP the P weights of restriction data for V(slr+1, ~λ, ℓ) at x, then PXi=1 aiαi = 0 =⇒ PXi=1 ai , 0. For Definition 7.2 recall Factorization, stated in Theorem 2.2. Definition 7.2. Suppose that V = V(slr+1, ~λ, ℓ) on Mg,n is a bundle of rank R with ∆-invariant zero rank scaling. We say that V satisfies quasi rank one factorization if given any boundary point x ∈ ∆g1,J = ∆g−g1,Jc, (or x ∈ ∆irr), and α1, . . ., αP the P weights of restriction data for V(slr+1, ~λ, ℓ) at x, then at most one of the P restriction factors rk V(slr+1, λ(J) ∪ {αi}, ℓ) rk V(slr+1, λ(Jc) ∪ {α∗ i }, ℓ), if x ∈ ∆g1,J = ∆g−g1,Jc or is greater than one. rk V(slr+1, ~λ ∪ {αi, α∗ i }, ℓ), if x ∈ ∆irr Definition 7.3. Suppose that V = V(slr+1, ~λ, ℓ) on Mg,n is a bundle of rank R with ∆-invariant zero rank scaling, and such that V satisfies quasi rank one factorization. If for some boundary point x ∈ ∆g1,J = ∆g−g1,Jc, (or x ∈ ∆irr), and if for one of the P weights of restriction data α1, . . ., αP of V(slr+1, ~λ, ℓ) at x, one has that one of the restriction factors rk V(slr+1, λ(J) ∪ {αi}, ℓ) rk V(slr+1, λ(Jc) ∪ {α∗ i }, ℓ), if x ∈ ∆g1,J = ∆g−g1,Jc or rk V(slr+1, ~λ ∪ {αi, α∗ i }, ℓ), if x ∈ ∆irr is greater than one, then that bundle of rank greater than one is called the socle of V. For a description of ∆-invariant zero rank scaling, see Definition 2.12. Definition 7.4. Suppose that V = V(slr+1, ~λ, ℓ) on Mg,n is a bundle of rank R which satisfies ∆-invariant zero rank scaling. We say that each socle satisfies ∆-invariant zero rank scaling with the same degree as that of V, if (1) For any (generic) boundary point x ∈ ∆g1,J = ∆g−g1,Jc, with socle given by restriction data α1, the }, ℓ) with the corresponding product }, ℓ) has ∆-invariant zero with polarized variety Xx(slr+1, λ(J) ∪ {α1}, ℓ) × Xx(slr+1, λ(Jc) ∪ {α∗ 1 of ample line bundles Lx(slr+1, λ(J) ∪ {α1}, ℓ) ⊠ Lx(slr+1, λ(Jc) ∪ {α∗ 1 the same degree as that of V. (2) For any (generic) x ∈ ∆irr, with socle given by restriction data α1, the polarized variety (Xx(slr+1, ~λ∪ {α1, α∗ 1 }, ℓ), Lx(slr+1, ~λ ∪ {α1, α∗ 1 }, ℓ)) has ∆-invariant zero with the same degree as that of V. SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 17 7.2. Extension criteria in the case of projective rank scaling. Our main result is the following: Theorem 7.5. Given a conformal blocks bundle V on Mg,n such that (1) V has ∆-invariant 0 rank scaling (Def 2.12); (2) the restriction data for V is free (Def 7.1); (3) V satisfies quasi rank one factorization (Def 7.2), and (4) each socle satisfies ∆-invariant 0 rank scaling with the same degree as V (Def 7.4). Then V has geometric interpretations at generic boundary points x ∈ Mg,n \ Mg,n, and therefore the corresponding first Chern class scaling identities from Equation (3.1) hold on Mg,n. The proof proceeds by showing that the natural maps Tm : V[m] → Symm(V) are isomorphisms for all m ≥ 0 on fibers over all points x ∈ Mg,n. First, by Lemma 6.2, it suffices to show that Tm is an isomorphism over marked curves which have exactly one node. Second, since the two sides of Tm have the same ranks, we only need to show that the map is an injection. We make the argument for points in ∆g1,J = ∆g−g1,Jc, as the argument for points in ∆0 is analogous. For simplicity we write ∆J for ∆g1,J. If x ∈ ∆J and the curve corresponding to x is C1 ∪ C2 with "normalization" C = C1 ⊔ C2 with two extra marked points a and b we have by Factorization, Theorem 2.2, V[m]x = Mµ∈Pmℓ(slr+1)eV(slr+1, {m~λ, µ, µ∗}, mℓ)y where y is the pointed curve C. The C are blocks for C; since C is disconnected, the C are tensor products of blocks for C1 and C2. There are maps eV(slr+1, {m~λ, µ, µ∗}, mℓ)∗ ⊗eV(slr+1, {m′~λ, ν, ν∗}, m′ℓ)∗ → eV(slr+1, {(m + m′)~λ, (µ + ν), (µ + ν)∗}, (m + m′)ℓ)∗, inducing an algebra structure on L V[m]∗ denote by eAx. Manon showed that the algebra eAx is a degeneration of Ax [Man09, Prop 3.3]. Therefore, it suffices to show that the map Symm(Vx) → V[m]x is an isomorphism in the algebra eAx. Assuming that the socle exists for our stratum (otherwise take B = C below), and thateV(slr+1, {~λ, α1, α∗ of the spaceseV(slr+1, {~λ, αi, α∗ is of rank R1, we let α2, . . . , αP be the other terms in the factorization. Pick non-zero generators y2, . . . , yP }, ℓ)∗, i > 2. Let S be the conformal blocks algebra of the socle x which we 1 i }, ℓ) S is a graded ring, the total section ring of (M, L) ( C is smooth). The socle satisfies projective rank scaling, so (M, L) = (PR1−1, O(1)), and hence S is a polynomial algebra in R1 variables. We form a new graded ring C = S[Y2, . . . , YP] with graded pieces S =Mm≥0 eV(slr+1, {m~λ, mα1, mα∗ 1}, mℓ)∗y. Cm = Mm1+···+mP=m Sm1Ym2 2 . . . YmP P . Therefore C is a polynomial algebra and the maps Symm(C1) → Cm, are isomorphisms. Now note that the natural map of algebras C → eAx is an isomorphism. To see this, write eAx = LeAm. The natural algebra map C → eAx sends: with µ = m1α1 + m2α2 + · · · + mPαP, m =P mi. It follows that different tuples (m1, . . . , mP) map to different direct summands in eAm. The rank of eAm is the same as that of Cm since both satisfy projective space rank P to eV(slr+1, {m~λ, µ, µ∗}, mℓ)∗y is injective scaling. It therefore suffices to note that the map Sm1Ym2 2 (by [BGM15a, Proposition 2.1]), and the surjection part is valid for any genus g, since we may replace the use of invariants by the integral section rings of line bundles over ind-integral affine Grassmannians. 7→ eV(slr+1, {m~λ, µ, µ∗}, mℓ)∗y, Sm1Ym2 2 . . . YmP P . . . YmP 18 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA rank scaling.) algebra of sections of a polarized variety of ∆-invariant 0 when x is the generic point of a boundary divisor. If S is the conformal blocks algebra of the socle we will need to verify that the algebra C has the 7.3. Proof of Theorem 7.5 in the general case. By Lemma 7.6 below, we need to show that eAx is the same graded ranks as eAx. (The map C → eAx is injective for the same reason as in the case of projective The graded algebra C is again the algebra of sections of a polarized variety of ∆-invariant zero (see Prop 2.11, Part (5), and the definition of the normalized generalized cone over a projective variety in [BS95, Section 1.1.8]). We claim that rank, degree and dimension of this polarized variety are the same as for Ax. If the triple is (R1, D, d1) for the socle, then the triple for the cone is (R1 + p − 1, D, d1 + p − 1). Now R1 + p − 1 equals the rank of (Ax)1, and the desired equality holds (using the ∆-invariant zero condition). also the algebra of sections of a (possibly different) polarized variety of ∆-invariant zero. Proof. Manon has introduced [Man09, Section 3] a filtration of the algebra Ax, so that the associated Lemma 7.6. Suppose eAx is the algebra of sections of a polarized variety of ∆-invariant zero. Then Ax is graded algebra equals eAx. Using a Rees algebra construction [Man09, see Eq 45, page 16], he forms a (flat) family of algebras C• parametrized by A1 such that for 0 , t ∈ A1, Ct (cid:27) Ax, and C0 (cid:27) eAx. Since eAx is generated in degree 1, by Nakayama's lemma and the fact that the algebras are constant for t , 0, we see that C• is generated in degree 1. We can therefore form P = Proj(C•) over A1. Since C• is a flat sheaf of algebras over A1, we get a flat family π : P → A1 with a relatively ample line bundle O(1), and maps Cm → π∗O(m). The higher cohomology Hi(π−1(t), O(m)π−1(t)), i > 0, m ≥ 0 vanishes for t = 0 because of our assumption on ∆ invariants (and Remark 3.4) the polarized variety carrying the algebra eAx is necessarily the fiber of π over t = 0, and hence for all t by semi-continuity. The map Cm → π∗O(m) is an isomorphism on fibers at t = 0, and hence in a neighborhood of t = 0, and hence over A1. Therefore Ax which is integral, is the algebra of sections of the polarized variety (Pt, O(m)Pt ) for t , 0. The ∆-invariant is constant in this family, and the desired statement follows. (cid:3) Remark 7.7. If V is a bundle of rank R such that rk V[m] = (cid:0)m+R−1 R−1 (cid:1), then one can use [Fak12] to reduce the proof of Theorem 7.5 to the statement for n = 4 by showing that divisors on both sides of the purported identity intersect all F-curves in the same degree. Example 7.8. We saw in Example 4.2, that for V = V(sl2, 1) on M2, the divisor scaling identity fails, and so the answer to Question 2.6 is no for this bundle. To see that V has a socle which fails to have projective rank scaling, we find the restriction data a generic point x in ∆1,∅, which can be represented by nodal curve where each arm has genus one. There is one piece of valid restriction data at x, given by α = 0, and when restricted to each arm, the bundle V(sl2, {0}, 1) has rank 2 [Fak12]. The polarized variety associated to the socle is (P1 × P1, O(1) ⊠ O(1)) = (Q, O(1)) where Q is a quadric in P3. This polarized variety is of ∆-invariant zero, and degree 2, and not 1 as required for an application of Theorem 7.5. 8. Examples with ∆-invariant zero Here we give examples of families of bundles of ∆-invariant zero of all types. For g = 0, such examples can be found using the Macaulay 2 [GS], with packages [SS07], and [Swi10], the latter of which implements the formulas of Fakhruddin [Fak12]. Often long and combinatorial Schubert calculus arguments were used to show all multiples V[m] had ranks predicted by experimentation. We include these as they are representative. Remark 8.1. In these and other examples we use the operation of plussing, described in Def 8.2, which can often turn a quantum computation of ranks into a classical one. We note that while plussing preserves the rank of a bundle, it does not preserve other invariants such as the first Chern class. SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 19 Definition 8.2. Let σ j : ωi 7→ ωi+j(mod r+1) be a cyclic permutation of the affine Dynkin diagram of slr+1. Then rk V(slr+1, {λ1, . . . , λn}, ℓ) = rk V(slr+1, {σ j1λ1, . . . , σ jnλn}, ℓ) if j1 + · · · + jn is divisible by r + 1 [FS99]. We refer to this operation as plussing. 8.0.1. Families of bundles with (Pd, O(1)) scaling for d = 0. By a quantum generalization of Fulton's con- jecture [Bel07] and [BK13, Remark 8.5], if V is any bundle of rank one, then rk(V[m]) will also be one, and so V will have projective scaling for d = 0. By [BGM15a, Cor. 2.2], in this case, one has that c1(V[m]) = mc1(V), so by Proposition 6.6, will have geometric extensions across the boundary. These include all level one bundles for slr+1, as proved in [GG12]. In [Kaz16] a complete characterization of Sn-invariant bundles of rank one for sln on M0,n was given. In [Hob15] all rank one bundles for sl2 on M0,n were described, as well as generalizations, those bundles for sl2m with so-called rectangular weights. 8.0.2. A family of bundles with (Pd, O(1)) scaling for d = 1. If n > 1 is odd, let ~λ = {2kω1, ((2k + 1)ω1)n−1}, and if n > 1 is even, let ~λ = {ω1, ((2k + 1)ω1)n−1}. Define V = V(sl2, ~λ, 2k + 1). By Remark 8.1, rk V[m] = rk V(sl2, 2kmω1, m(2k + 1)) = m + 1. We note that the restriction data for V at a point x in the boundary ∆irr i , ~λ, 2k + 1) = 1 is given by µ1 = µ∗ 1 for i = 1, 2. At a point x ∈ ∆1,J = ∆0,Jc, one also gets free restriction data (which differs depending on the parity of J), and all restrictions have rank one. Therefore, by Theorem 7.5, V satisfies projective space = (k + 1)ω1, so it is free, and we have rk V(sl2, µi, µ∗ = kω1 and µ2 = µ∗ 2 Chern class scaling, i.e. c1(mV) = (cid:0)m+1 block V(sl2, ~λ, 2k + 1)x for all points x ∈ M1,n. 2 (cid:1)c1(V), and there is a geometric interpretation of the conformal 8.0.3. A family on M0,n, for n ≥ 4 with (Pd, O(1)) scaling for arbitrary d. On M0,4 let V = V(slr+1, {(ωi + m (cid:1), where d = 2i ≤ r + 1 − 2i ωr+1−i)4}, 2). One can check, using Littlewood-Richardson, that rk V[m] =(cid:0)d+m (so i ≤ ⌊ r+1 ⌋), so these vector bundles satisfy projective rank scaling. These bundles are S4-invariant and 4 so there is just one boundary restriction, up to symmetry. The restriction data at the boundary point is free, given by the d + 1 weights α j = ω j + ωr+1−j, where 1 ≤ j ≤ d, and αd+1 = 0. V has quasi rank one factorization, since for all j ∈ {1, . . . , d + 1}, rk V(slr+1, {ωi + ωr+1−i, ωi + ωr+1−i, α j}, 2) = rk V(slr+1, {ωi + ωr+1−i, ωi + ωr+1−i, α∗ }, 2) = 1. We have checked that we may apply Theorem 7.5 to conclude that first j Chern class scaling identities hold for this bundle V, and using Fakhruddin's formula [Fak12], one can see (by a straightforward, but somewhat involved combinatorial argument), that deg(V) = d(d+1) So deg V[m] =(cid:0)m+d 8.1), we conclude from the n = 4 calculation that in both cases, rk V[m] =(cid:0)d+m =(cid:0)d+1 2 (cid:1). 2 (cid:1). For n = 5, we take V(slr+1, {(ωi + ωr+1−i)3, ωi−1 + ωr−i, 2ω1}, 2). For n ≥ 6, we let m (cid:1), where d = 2i ≤ r + 1 − 2i. λ j = ωi + ωr+1−i, where i ≤ ⌊ r+1 ⌋ for j = 1, . . . , 4. Let λi = 2ω1 for j = 5, . . . , n − 1, s ≡ (r − n + 6) mod (r + 1), 4 and λn = 2ωs. We can define V = V(slr+1, {~λ}, 2). Since the bundles are obtained by plussing (see Remark Checking restriction data, it is straightforward to verify that one can apply Theorem 7.5. We outline the approach for n ≥ 6: d+1(cid:1)(cid:0)d+1 2 1 (1) {1, 2, 3, 4} ⊂ I. In this case the restriction data is given by µ = 2ωk, where k ≡ (r + 1 − we have rk V(slr+1, {~λ(I), µ}, 2) = d + 1, and rk V(slr+1, {~λ(IC), µ∗}, 2) = 1. An argument shows that 2Pj∈I λ j) mod (r + 1). There is one piece of restriction data, so it is free and just as before rk V(slr+1, {m~λ(I), mµ}, 2m) =(cid:0)d+m m (cid:1), so that the socle satisfies projective rank scaling. (2) a ∈ I, b, c, d < I, where {a, b, c, d} = {1, 2, 3, 4}. In this case the restriction data is given by µ = ωi+s + ωr+1−i+s, where the indices are modulo (r +1), and s is such that 1 The restriction data is free, moreover, we have rk V(slr+1, {m~λ(I), mµ}, 2m) = 1, and one can 2Pi∈I,i,1 λi ≡ (r +1 − s) mod (r +1). conclude rk V(slr+1, {m~λ(IC), mµ∗}, 2m) =(cid:0)d+m m (cid:1). (3) {a, b} ⊂ I, c, d < I, where {a, b, c, d} = {1, 2, 3, 4}. In this case the restriction data is given by µ j = ω j + s + ωr+1−j+s, where indices are taken modulo (r + 1), 0 ≤ j ≤ d, and s is such that 20 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA 2Pi∈I,i,1 λi ≡ (r + 1 − s) mod (r + 1). Restriction data is free, and for all j ∈ {1, . . . , d + 1} we have 1 rk V(slr+1, ~λ(I), α j}, 2) = rk V(slr+1, ~λ(IC), α∗ j }, 2) = 1. 8.0.4. Quadric scaling on M0,n for n ≥ 5. Let V = V(sl4, {ω1 + 3ω2 + ω3, 3ω1 + ω2 + ω3, 2ω1 + ω2 + 2ω3, (7ω1)n−4, 7ωs}, 7), where s ≡ (−n) mod 4. One computes ranks of V[m] by noticing that V is ob- tained by applying the plussing operation to V = V(sl4, {ω1 + 3ω2 + ω3, 3ω1 + ω2 + ω3, 2ω1 + ω2 + 2ω3}, 7) on M0,3 which satisfies rk(V[m]3) = 2(cid:0)m+2 3 (cid:1) +(cid:0)m+2 2 (cid:1). To verify V has quadric hypersurface rank scaling for d = 3, and that all the conditions of Claim 7.5 were met, we computed ranks using Witten's Dictionary 2.4 together with the plussing operation and Littlewood-Richardson counting arguments. So by Lemma 6.9, one has that V∗ (cid:27) H0(X, L), where L embeds X as a quadric hypersurface in P4. Given any x ∈ δI for any I there will always be a unique socle with quadric rank scaling. For example, it will have restriction weight µ = 0 at x ∈ δ123, and the rank of V(sl4, {λ1, λ2, λ3, 0}, 7) has quadric hypersurface scaling (for d = 3). Therefore V satisfies the hypothesis of Theorem 7.5, so multiples c1(V[m]) are governed by first Chern class scaling identities given in Corollary 3.1. 8.0.5. Rational normal scroll scaling on on M0,n for n ≥ 5. We set V = V(sl2, {(2ω1)4, 4ω1, (5ω1)n−6, 5ωs}, 5) with s ≡ n (mod (2)). One can check that V satisfies (S(1, 2), O(1)) scaling so that rk V[m] = (m + 1)(1 + 3m 2 ). To do so we use Witten's Dictionary, Theorem 2.4 and Quantum Pieri on M0,5 and then apply the plussing procedure. One can use [Swi10] to compute coefficients of the first Chern classes of multiples V[m] in the nonadjacent bases on M0,5 given by {δ13, δ14, δ24, δ25, δ35}: c1(V) = 2(δ14 + δ25 + δ35), c1(V[2]) = 9(δ14 + δ25 + δ35), c1(V[3]) = 24(δ14 + δ25 + δ35), and c1(V[4]) = 50(δ14 + δ25 + δ35). We calculate c1(V[4]) = 10c1(V) − 10c1(V[2]) + 5c1(V[3]), for example. so the predicted divisor identity given in Section 6.4, holds for c1(V[4]). 8.0.6. Veronese surface scaling on M0,n. Let V = V(sl2, {ω1, 3ω1, 4ω1, 4ω1, 6ω1, (8ω1)n−6, 8ωs}, 8), for which s ≡ n (mod (2)). V[m] can be shown to have rank (m + 1)(2m + 1). To do so we work first on M0,5, with the bundle V = V(sl2, {ω1, 3ω1, 4ω1, 4ω1, 6ω1}, 8) and argue using Witten's dictionary, Theorem 2.4. We then use the standard plussing procedure. In the non-adjacent basis for the Picard group of M0,5 consisting of δ13, δ14, δ24, δ25, and δ35, one can use [Swi10] to check that the first five multiples of V have the following expressions: c1(V) = δ14 + 2δ24 + δ25 + 3δ35; c1(V[2]) = 4δ14 + 11δ24 + 4δ25 + 15δ35; c1(V[3]) = 10δ14 + 32δ24 + 10δ25 + 42δ35; c1(V[4]) = 20δ14 + 70δ24 + 20δ25 + 90δ35; and c1(V[5]) = 35δ14 + 130δ24 + 35δ25 + 165δ35, A calculation shows that c1(V[5]) = −15c1(V) + 20c1(V[2]) − 15c1(V[3]) + 6c1(V[4]), as is predicted. While the identity holds, V does not satisfy the hypothesis of Theorem 7.5. At a point of δ123, one finds three pieces of restriction data: µ ∈ {2ω1, 4ω1, 6ω1}, and by [Swi10], rk(V(sl2, {λ1, λ2, λ3, µ}, 8) = 2. = V(sl2, {2kω1, ((d + 2k)ω1)n−2, (d + 8.0.7. Twisted cubic scaling on M1,n, for n ≥ 1. For d even, let Vd k 2k)ωs}, d + 2k), with s ≡ n (mod (2)). First let n = 1. By [Fak12, p. 27], if i is even, the rank of V(sl2, iω1, ℓ), is ℓ + 1 − i, and one has that rk(Vd k[m]) = dm + 1 for all n ≥ 1. Returning to the case n = 1, while ranks of multiples do not depend on k, by [Fak12, Corollary 6.2] degrees do: deg(Vd . This bundle satisfies the identity for d = 3 in Equation 6.4: 48(k−3)−4(k−3) k[3]). There is one necessary boundary k[m]) = dm + 1. Now by plussing, one has that rk(Vd k) = − m(dm+1)(k+d) k[4]) = 4 deg(V3 = deg(V3 12 k) − 6 deg(V3 k[2]) + 4 deg(V3 k has quasi rank one factorization. 12 restriction, and one has that Vd References [AGS14] Valery Alexeev, Angela Gibney, and David Swinarski, Higher-level sl2 conformal blocks divisors on M0,n, Proc. Edinb. Math. Soc. (2) 57 (2014), no. 1, 7 -- 30, DOI 10.1017/S0013091513000941. MR3165010 ↑3 [BBV15] Velleda Baldoni, Arzu Boysal, and Mich`ele Vergne, Multiple Bernoulli series and volumes of moduli spaces of flat bundles over surfaces, J. Symbolic Comput. 68 (2015), no. part 2, 27 -- 60. ↑5 SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 21 [Bea96] Arnaud Beauville, Conformal blocks, fusion rules and the Verlinde formula, Proceedings of the Hirzebruch 65 Conference on Algebraic Geometry (Ramat Gan, 1993), Israel Math. Conf. Proc., vol. 9, Bar-Ilan Univ., Ramat Gan, 1996, pp. 75 -- 96. MR1360497 ↑3 [BL94] Arnaud Beauville and Yves Laszlo, Conformal blocks and generalized theta functions, Comm. Math. Phys. 164 (1994), no. 2, 385 -- 419. ↑1, 2 [Bel08] Prakash Belkale, Quantum generalization of the Horn conjecture, J. Amer. Math. Soc. 21 (2008), no. 2, 365 -- 408. ↑3, 4 [Bel07] , Geometric proof of a conjecture of Fulton, Adv. Math. 216 (2007), no. 1, 346 -- 357. MR2353260 (2009a:20073) ↑19 [BGM15a] Prakash Belkale, Angela Gibney, and Swarnava Mukhopadhyay, Nonvanishing of conformal blocks divisors, Transfor- mation Groups (2015). See arXiv:1410.2459 [math.AG]. ↑4, 13, 17, 19 [BGM15b] , Vanishing and identities of conformal blocks divisors, Algebr. Geom. 2 (2015), no. 1, 62 -- 90. ↑4 [BK13] Prakash Belkale and Shrawan Kumar, The multiplicative eigenvalue problem and deformed quantum cohomology (2013). arXiv:1310.3191 [math.AG]. ↑19 [Ber07] Eugenio Bertini, Introduzione alla geometria proiettiva degli iperspazi, Enrico Spoerri, Pisa (1907). ↑5 [BS95] Mauro C. Beltrametti and Andrew J. Sommese, The adjunction theory of complex projective varieties, de Gruyter Expositions in Mathematics, vol. 16, Walter de Gruyter & Co., Berlin, 1995. ↑5, 14, 18 [Ber93] Aaron Bertram, Generalized SU(2) theta functions, Invent. Math. 113 (1993), no. 2, 351 -- 372. ↑2 [BG15] Michele Bolognesi and Noah Giansiracusa, Factorization of point configurations, cyclic covers, and conformal blocks, J. Eur. Math. Soc. (JEMS) 17 (2015), no. 10, 2453 -- 2471. ↑3 [Cob03] A. B. Coble, The quartic curve as related to conics, Trans. Amer. Math. Soc. 4 (1903), no. 1, 65 -- 85. ↑11 [DW93] Georgios Daskalopoulos and Richard Wentworth, Local degeneration of the moduli space of vector bundles and factorization of rank two theta functions. I, Math. Ann. 297 (1993), no. 3, 417 -- 466. ↑2 [DP86] Pasquale Del Pezzo, Sulle superficie di ordine n immerse nello spazio di n + 1 dimensioni, Rend. Circ. Mat. Palermo 1 (1886). ↑5 [DR76] U. V. Desale and S. Ramanan, Classification of vector bundles of rank 2 on hyperelliptic curves, Invent. Math. 38 (1976/77), no. 2, 161 -- 185. ↑12 [Eis05] David Eisenbud, The geometry of syzygies, Graduate Texts in Mathematics, vol. 229, Springer-Verlag, New York, 2005. A second course in commutative algebra and algebraic geometry. ↑15 [EG84] David Eisenbud and Shiro Goto, Linear free resolutions and minimal multiplicity, J. Algebra 88 (1984), no. 1, 89 -- 133. ↑7 [EH87] David Eisenbud and Joe Harris, On varieties of minimal degree (a centennial account), Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), Proc. Sympos. Pure Math., vol. 46, Amer. Math. Soc., Providence, RI, 1987, pp. 3 -- 13. ↑5 [Fak12] Najmuddin Fakhruddin, Chern classes of conformal blocks, Compact moduli spaces and vector bundles, Contemp. Math., vol. 564, Amer. Math. Soc., Providence, RI, 2012, pp. 145 -- 176. ↑1, 2, 3, 9, 11, 14, 18, 19, 20 [Fal94] Gerd Faltings, A proof for the Verlinde formula, J. Algebraic Geom. 3 (1994), no. 2, 347 -- 374. ↑1, 2, 4 [Fal96] [Fed11] M. Fedorchuk, Cyclic covering morphisms on M0,n (2011). arXiv:1105.0655. ↑3 , Moduli-stacks for bundles on semistable curves, Math. Ann. 304 (1996), no. 3, 489 -- 515. ↑2 [FS99] J urgen Fuchs and Christoph Schweigert, The action of outer automorphisms on bundles of chiral blocks, Comm. Math. Phys. 206 (1999), no. 3, 691 -- 736. ↑19 [Fuj90] Takao Fujita, Classification theories of polarized varieties, London Mathematical Society Lecture Note Series, vol. 155, Cambridge University Press, Cambridge, 1990. ↑4, 5 [Gia13] Noah Giansiracusa, Conformal blocks and rational normal curves, J. Algebraic Geom. 22 (2013), no. 4, 773 -- 793. ↑3 [GG12] Noah Giansiracusa and Angela Gibney, The cone of type A, level 1 conformal block divisors, Adv. Math. 231 (2012). ↑3, 19 [GJMS13] Angela Gibney, David Jensen, Han-Bom Moon, and David Swinarski, Veronese quotient models of M0,n and conformal blocks, Michigan Math. J. 62 (2013), no. 4, 721 -- 751. ↑3 [Gor68] Robert Goren, Characterization and algebraic deformations of projective space, J. Math. Kyoto Univ. 8 (1968), 41 -- 47. ↑5 [GS] Daniel R. Grayson and Michael E. Stillman, Macaulay2, a software system for research in algebraic geometry. ↑18 [GH78] Phillip Griffiths and Joseph Harris, Principles of algebraic geometry, Wiley-Interscience [John Wiley & Sons], New York, 1978. Pure and Applied Mathematics. ↑5 [Har81] Joe Harris, A bound on the geometric genus of projective varieties, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 8 (1981), no. 1, 35 -- 68. ↑5 [Hob15] Natalie Hobson, Quantum Kostka and the rank one problem for sl2m (2015). arXiv:1508.06952 [math.AG]. ↑4, 19 [Kau05] Ivan Kausz, A canonical decomposition of generalized theta functions on the moduli stack of Gieseker vector bundles, J. Algebraic Geom. 14 (2005), no. 3, 439 -- 480. ↑2 [Kaz16] Anna Kazanova, On Sn-invariant conformal blocks vector bundles of rank one on M0,n, Manuscripta Math. 149 (2016), no. 1-2, 107 -- 115. ↑4, 19 22 PRAKASH BELKALE, ANGELA GIBNEY, AND ANNA KAZANOVA [KP95] Alexis Kouvidakis and Tony Pantev, The automorphism group of the moduli space of semistable vector bundles, Math. Ann. 302 (1995), no. 2, 225 -- 268. ↑10 [KNR94] Shrawan Kumar, M. S. Narasimhan, and A. Ramanathan, Infinite Grassmannians and moduli spaces of G-bundles, Math. Ann. 300 (1994), no. 1, 41 -- 75. ↑1, 2 [Kra15] Daniel Krashen, CBRestrictor: a Macaulay2 script for computing ranks of vector bundles of conformal blocks on moduli of higher genus curves (2015). available at https://github.com/dkrashen/conformal-blocks-rational-locus-restrictor. ↑10 [LS97] Yves Laszlo and Christoph Sorger, The line bundles on the moduli of parabolic G-bundles over curves and their sections, Ann. Sci. ´Ecole Norm. Sup. (4) 30 (1997), no. 4, 499 -- 525. ↑1, 2, 4 [Man09] Chris Manon, The Algebra of Conformal blocks (2009). arXiv:0910.0577. ↑2, 4, 17, 18 [MOP15] Alina Marian, Dragos Oprea, and Rahul Pandharipande, The first Chern class of the Verlinde bundles, String-Math 2012, Proc. Sympos. Pure Math., vol. 90, Amer. Math. Soc., Providence, RI, 2015, pp. 87 -- 111. ↑2, 10, 11, 12 [Mum83] David Mumford, Towards an enumerative geometry of the moduli space of curves, Arithmetic and geometry, Vol. II, Progr. Math., vol. 36, Birkhauser Boston, Boston, MA, 1983, pp. 271 -- 328. MR717614 (85j:14046) ↑12, 14 [Nag60] Masayoshi Nagata, On rational surfaces. I. Irreducible curves of arithmetic genus 0 or 1, Mem. Coll. Sci. Univ. Kyoto Ser. A Math. 32 (1960), 351 -- 370. ↑5 [Nag07] Uwe Nagel, Minimal free resolutions of projective subschemes of small degree, Syzygies and Hilbert functions, Lect. Notes Pure Appl. Math., vol. 254, Chapman & Hall/CRC, Boca Raton, FL, 2007, pp. 209 -- 232. ↑7 [NR93] M. S. Narasimhan and T. R. Ramadas, Factorisation of generalised theta functions. I, Invent. Math. 114 (1993), no. 3, 565 -- 623. ↑2 [NR69] M. S. Narasimhan and S. Ramanan, Moduli of vector bundles on a compact Riemann surface, Ann. of Math. (2) 89 (1969), 14 -- 51. ↑9 [NR87] , 2θ-linear systems on abelian varieties, Vector bundles on algebraic varieties (Bombay, 1984), 1987, pp. 415 -- 427. ↑10 [Ngu07] Quang Minh Nguyenn, Vector bundles, dualities and classical geometry on a curve of genus two, Internat. J. Math. 18 (2007), no. 5, 535 -- 558. ↑2, 11 [Ort05] Angela Ortega, On the moduli space of rank 3 vector bundles on a genus 2 curve and the Coble cubic, J. Algebraic Geom. 14 (2005), no. 2, 327 -- 356. ↑2, 11 [Pau96] Christian Pauly, Espaces de modules de fibr´es paraboliques et blocs conformes, Duke Math. J. 84 (1996), no. 1, 217 -- 235 (French). ↑1, 2 [Ram96] T. R. Ramadas, Factorisation of generalised theta functions. II. The Verlinde formula, Topology 35 (1996), no. 3, 641 -- 654. ↑2 [SS07] Hal Schenck and Michael Stillman, SchurRings: a Macaulay2 package make computations in the representation ring of GL(n) possible (2007). Version 0.2, http://www.math.uiuc.edu/Macaulay2/. ↑18 [Sor96] Christoph Sorger, La formule de Verlinde, Ast´erisque 237 (1996), Exp. No. 794, 3, 87 -- 114. S´eminaire Bourbaki, Vol. 1994/95. ↑2, 3 [Sun00] Xiaotao Sun, Degeneration of moduli spaces and generalized theta functions, J. Algebraic Geom. 9 (2000), no. 3, 459 -- 527. ↑2 , Factorization of generalized theta functions in the reducible case, Ark. Mat. 41 (2003), no. 1, 165 -- 202. ↑2 [Sun03] [Swi09] David Swinarski, sl2 conformal block divisors and the nef cone of M0,n (2009). arXiv:1107.5331. ↑14 [Swi10] , ConformalBlocks: a Macaulay2 package for computing conformal block divisors (2010). Version 1.1, http://www.math.uiuc.edu/Macaulay2/. ↑10, 18, 20 [Tel98] Constantin Teleman, Borel-Weil-Bott theory on the moduli stack of G-bundles over a curve, Invent. Math. 134 (1998), no. 1, 1 -- 57. ↑2 , The quantization conjecture revisited, Ann. of Math. (2) 152 (2000), no. 1, 1 -- 43. ↑5 [Tel00] [Tha94] Michael Thaddeus, Stable pairs, linear systems and the Verlinde formula, Invent. Math. 117 (1994), no. 2, 317 -- 353. ↑2 [TUY89] Akihiro Tsuchiya, Kenji Ueno, and Yasuhiko Yamada, Conformal field theory on universal family of stable curves with gauge symmetries, Integrable systems in quantum field theory and statistical mechanics, 1989, pp. 459 -- 566. ↑2, 3 [Wit91] Edward Witten, On quantum gauge theories in two dimensions, Comm. Math. Phys. 141 (1991), no. 1, 153 -- 209. ↑5 [Wit95] , The Verlinde algebra and the cohomology of the Grassmannian, Geometry, topology, & physics, 1995, pp. 357 -- 422. ↑4 [Xam81] S. Xamb ´o, On projective varieties of minimal degree, Collect. Math. 32 (1981), no. 2, 149 -- 163. ↑5 [Zag95] Don Zagier, On the cohomology of moduli spaces of rank two vector bundles over curves, The moduli space of curves (Texel Island, 1994), Progr. Math., vol. 129, Birkhauser Boston, Boston, MA, 1995, pp. 533 -- 563. ↑2 P.B.: Department of Mathematics, University of North Carolina, Chapel Hill, NC 27599, email: [email protected] SCALING OF CONFORMAL BLOCKS AND GENERALIZED THETA FUNCTIONS OVER Mg,n 23 A.G.: Department of Mathematics, University of Georgia, Athens, GA 30602, email: [email protected] A.K.: Department of Mathematics, University of Georgia, Athens, GA 30602, email: [email protected]
1811.12876	2	1811	2019-07-09T15:47:28	The moduli space of real vector bundles of rank two over a real hyperelliptic curve	[ "math.AG" ]	The Desale-Ramanan Theorem is an isomorphism between the moduli space of rank two vector bundles over complex hyperelliptic curve and the variety of linear subspaces in an intersection of two quadrics. We prove a real version of this theorem for the moduli space of real vector bundles over a real hyperelliptic curve. We then apply this result to study the topology of the moduli space, proving that it is relatively spin and identifying the diffeomorphism type for genus two curves. Our results lay the groundwork for future study of the quantum homology of these moduli spaces.	math.AG	math	THE MODULI SPACE OF REAL VECTOR BUNDLES OF RANK TWO OVER A REAL HYPERELLIPTIC CURVE THOMAS JOHN BAIRD, SHENGDA HU Abstract. The Desale-Ramanan Theorem is an isomorphism between the moduli space of rank two vector bundles over complex hyperelliptic curve and the variety of linear subspaces in an intersection of two quadrics. We prove a real version of this theorem for the moduli space of real vector bundles over a real hyperelliptic curve. We then apply this result to study the topology of the moduli space, proving that it is relatively spin and identifying the diffeo- morphism type for genus two curves. Our results lay the groundwork for future study of the quantum homology of these moduli spaces. 1. Introduction Given a Riemann surface Σ of genus g ≥ 2 and a line bundle ξ → Σ of odd degree, the moduli space Uξ of stable rank 2 vector bundles E → Σ with fixed determinant ∧2E ∼= ξ is a non-singular projective variety of dimension 3g − 3 which is naturally endowed with a Kaehler metric. Given an antiholomorphic involution τ : Σ → Σ, there is an induced antiholo- morphic and antisymplectic involution on Uξ. The fixed point set U τ ξ ⊆ Uξ is a totally geodesic real Lagrangian submanifold of Uξ. The points of U τ ξ correspond to holomorphic bundles over Σ which admit a real structure [3, 20]. The topol- ogy of U τ ξ has been investigated using Atiyah-Bott-Kirwan type methods: the mod 2 betti numbers were calculated in [1, 2, 14], and when g ≥ 3 the rational cohomology ring was calculated in [2]. In the current paper, we study U τ ξ using an entirely different method in the special case when Σ is a hyperelliptic curve. Recall that a hyperelliptic curve is a 2-fold ramified cover π : Σ → CP 1. If Σ has genus g ≥ 2, then π has 2g + 2 ramification points W ⊆ CP 1 also known as the Weierstrass points. Choose affine coordinates CP 1 = C ∪ {∞} and let λ1, ..., λ2g+2 ∈ C denote the coordinates of W . The following is due to Desale-Ramanan [8] (see also [16, 17] for the genus 2 case). Theorem 1.1 (Desale-Ramanan). If Σ is hyperelliptic then Uξ is isomorphic to the subvariety of the Grassmannian X ⊆ Grg−1(C2g+2) consisting of (g − 2)-linear subspaces contained in the intersection of the two quadrics Z(q0)∩Z(q1) ⊂ CP 2g+1 where 1 + ... + x2 2g+2, q0 = x2 q1 = λ1x2 1 + .... + λ2g+2x2 2g+2. 1 Our main result (Theorem 4.1) is a real version of the Desale-Ramanan The- orem. Namely, we prove that if (Σ, τ ) is a real hyperelliptic curve, then U τ is ξ isomorphic to the subvariety of Grg−1(R2g+2) consisting of planes contained in the intersection of two explicit real quadrics. For genus g = 2 curves, this result was proven by Shuguang Wang [21]. The rest of the paper is devoted to applications of this result. In §5 we produce formulas for the Stiefel-Whitney classes of U τ ξ is relatively spin as a Lagrangian submanifold of Uξ. This implies that it has a well-defined quantum homology ring with integer coefficients [4]. ξ and use these to show that U τ In §6 we identify the diffeomorphism type of U τ ξ Previously, only the Z2-Betti numbers were known. for all genus two examples. 2. Real hyper-elliptic curves A hyperelliptic curve, π : Σ → CP 1, is a 2-fold ramified cover over the complex projective line. We assume always that Σ has genus g ≥ 2. Such a curve admits an involution ι : Σ → Σ, which interchanges the two sheets of the cover and thus identifies Σ/ι ∼= CP 1. The ramification points W ⊂ CP 1 are called the Weierstrass points. Choosing affine coordinates on CP 1 = C ∪ {∞} we can represent W as a set of 2g + 2 points in the complex plane C (we assume always ∞ 6∈ W ). A real structure on a Riemann surface Σ is an anti-holomorphic involution τ , that is a smooth map τ : Σ → Σ which reverses the complex structure and satisfies τ 2 = IdΣ. The fixed point set of the involution, ΣR := Στ , is diffeomorphic to disjoint union of circles. In this paper, we consider real structures that are compatible with the hyperelliptic projection, i.e. π ◦ τ = τP 1 ◦ π with respect to the standard real involution τP 1 on CP 1, defined by τP 1([z : w]) := [z : w]. A hyperelliptic curve admits a compatible real structure if and only if τP 1(W ) = W . If this happens, then Σ admits a pair of compatible real structures τ and τ ◦ ι = ι ◦ τ , where ι is the hyperelliptic involution. For consistency, we will always choose τ so that ∞ ∈ π(Στ ) and ∞ 6∈ π(Στ ◦ι). A typical situation for genus g = 2 is illustrated in Figures 1 and 2 below. The ramification points W ⊂ C ⊂ CP 1, drawn in orange, are arranged symmetrically about the real axis. The respective images π(Στ ) and π(Στ ◦ι) are drawn in green. R Figure 1. τ Given a real curve (Σ, τ ) the fixed point set Στ is a union of disjoint circles called the real circles. The topological type of a real curve (Σ, τ ) is determined by the genus of g, the number of path components of Στ and whether or not Σ\ Στ is 2 R Figure 2. ι ◦ τ connected (see [10]). The following proposition classifies real hyperelliptic curves topologically. Proposition 2.1. Decompose W = W0∪W+∪W− , where W0, W+, W− are those Weierstrass points in w ∈ C whose imaginary part is respectively zero, positive, negative. Clearly W+ and W− have equal cardinality so W0 has even cardinality. Let 2n = #W0. (i) If n > 0, then π0(Στ ) = π0(Στ ◦ι) = n. (ii) If n = 0 then π0(Στ ) = 1 and π0(Στ ◦ι) = 0. (iii) If 1 ≤ n ≤ g, then both Σ \ Στ and Σ \ Στ ◦ι are connected. If n = g + 1 then both Σ \ Στ and Σ \ Στ ◦ι are disconnected. If n = 0, then Σ \ Στ is disconnected and Σ \ Στ ◦ι is connected. Proof. Statements (i) and (ii) are pretty clear. For statement (iii), observe that Σ \ Στ is connected if and only if it is possible to draw a closed loop in C \ W which encloses an odd number of points in W and crosses the real line. Then it is a simple matter of looking at pictures. (cid:3) Remark 2.2. Proposition 2.1 implies that for genus g ≥ 4, there exist topological types of real curves that cannot be realized as hyperelliptic curves. Namely, those curves for which 1 < π0(Στ ) < g + 1 and Σ \ Στ is disconnected. 2.1. Real line bundles over real hyperelliptic curves. Let (Σ, τ ) be a real curve. A real line bundle ξ over (Σ, τ ) is a line bundle that admits an antiholo- morphic lift ξ eτ ξ Σ τ / Σ such that eτ 2 = Idξ. If eτ exists then by Schur's Lemma it is uniquely determined up to multiplication by a unit scalar. The fixed point set ξ eτ → Στ is an R1-bundle over Στ . The Stiefel-Whitney class w1(ξ eτ ) ∈ H 1(Στ ; Z2) depends only on (Σ, τ, ξ), If C ⊂ Στ is a real circle, then the following are not on the choice of lift eτ . equivalent (i) ξ eτC is non-orientable (i.e. a Moebius band), (ii) w1(ξ eτ )(C) = 1, 3 / / / We call C odd with respect to ξ if it satisfies these equivalent conditions. If k is the number of odd circles with respect to ξ then by ([3] Prop. 4.1) (2.1) deg(ξ) ≡ w1(ξ eτ )(Στ ) ≡ k mod 2. The Desale-Ramanan Theorem requires the degree of ξ to be odd, which implies In that our real line bundle ξ will always have an odd number of odd circles. particular, we need only consider (Σ, τ ) for which Στ is non-empty. By ([3] Prop. 4.1) if Στ 6= ∅, then real line bundles exist over (Σ, τ ) in all degrees. Tensoring by a real line bundle L over (Σ, τ ) determines an isomorphism ξ ∼= U τ U τ ξ⊗L⊗2 and deg(ξ ⊗ L⊗2) = deg(ξ) + 2deg(L). Therefore in studying U τ without loss of generality that ξ has degree 2g + 1. Given a real line bundle ξ with real structure eτ , there always exists a eτ -invariant meromorphic section which determines a real divisor D = Pi mipi where mi ∈ Z is the multiplicity of the zero of s at pi ∈ Σ. Observe that π(D) = D. A real circle in C ⊆ Στ is odd with respect to ξ if and only if Ppi∈C mi is odd. ξ we may assume 3. The Desale-Ramanan Theorem We review what we need from Desale-Ramanan [8]. 3.1. The moduli space of bundles as a subvariety of a Grassmannian. Let π : Σ → CP 1 be a hyperelliptic curve of genus g with hyperelliptic involution ι. Let W be the set of 2g + 2 Weierstrass points for Σ. We abuse notation and identify W = π−1(W ) ⊂ Σ. Fix a line bundle ξ → Σ of degree 2g + 1. natural map Let E ∈ Uξ be a stable bundle with determinant isomorphic to ξ. There is a H 0(Σ, E ⊗ ι∗E) → M Ew ⊗ ι∗Ew = M w∈W w∈W Ew ⊗ Ew defined by restriction and identifying ι∗Ew = Ew. This map is ι-equivariant. Restricting to the −1 eigenspaces, yields a map (3.1) H 0(Σ, E ⊗ ι∗E)−ι → M (Ew ⊗ Ew)−ι = M ξw ∧2Ew ∼= M w∈W w∈W w∈W where the last isomorphism uses ∧2E ∼= ξ, so is only natural up to a non-zero scalar in C∗. Desale-Ramanan prove that (3.1) is injective, and H 0(Σ, E ⊗ ι∗E)−ι has fixed dimension g + 3 for all E. We therefore obtain a morphism where V ∗ = ⊕w∈W ξw. Applying duality, we can think of this as a morphism where V := ⊕w∈W ξ∗ w. Desale and Ramanan prove that φ is an embedding, so Uξ is isomorphic with the image im(φ). Moreover, they prove that im(φ) is equal Uξ → Grg+3(V ∗) φ : Uξ → Grg−1(V ) 4 to the set of (g − 2)-linear subspaces in P (V ) that lie in the intersection of two quadrics Q0 and Q1. We construct these quadrics in the next section. Now suppose we endow Σ with an antiholomorphic involution τ which commutes with ι. This means in particular that τ permutes the Weierstrass points. If ξ is a real line bundle over (Σ, τ ), then V = ⊕w∈W ξ∗ w inherits an anti-linear involution inducing an involution on Grg−1(V ). It is clear from the construction that φ is τ -equivariant. 3.2. Line bundles and quadrics. A pencil of quadratic forms on a vector space V is determined by a surjective linear map Q : S2(V ) → U where U is a vector space of dimension two. In terms of a basis for U, this is defined by a linearly independent pair of quadratic forms Q0, Q1 : S2(V ) → C. to construct a pencil of quadratic forms on V where Given a line bundle ξ of degree 2g + 1 over the hyperelliptic curve Σ, we want (3.2) V := M w∈W ξ∗ w. We do this by constructing a linear map Q : S2(V ) → H 0(h−2g−1(W )) where h is the hyperplane bundle over CP 1 and W is the Weierstrass divisor. Our pencil will be diagonal with respect to the decomposition (3.2) so it is determined by morphisms for each w ∈ W . We define qw to be the composition of morphisms (3.4), (3.5), (3.6) defined below. Fix an isomorphism qw : (ξ∗ w)⊗2 → H 0(h−2g−1(W )) (3.3) At a Weierstrass point w ∈ W , φ restricts to an isomorphism = h−2g−1 (3.4) φ : ξ∗ ⊗ ι∗ξ∗ ∼= π∗(h)−2g−1. (ξ∗ w)⊗2 = ξ∗ w ∼= π∗(h)−2g−1 w w w ⊗ ι∗ξ∗ (recall we abuse notation identifying w and π(w)). Evaluation at w determines an isomorphism (3.5) h−2g−1 w ∼= H 0(h−2g−1(W − w)), and we have the natural inclusion (3.6) H 0(h−2g−1(W − w)) ֒→ H 0(h−2g−1(W )). Choose affine coordinates on CP 1 = C ∪ {∞} with variable t and Weierstrass points {tw ∈ C}. In these coordinates we have where R(t) = t2g+1Qw∈W (t − tw)−1. The quadratic forms Q0, Q1 are defined by the identity Q = (Q0t − Q1)R(t). H 0(h−2g−1(W )) = {(at + b)R(t)a, b ∈ C} 5 Let u be a meromorphic section of ξ which may be chosen to have no poles or zeros on W so that {uww ∈ W} is a basis for V . Let D be the divisor of u and express in affine coordinates αi ∈ C ⊂ CP 1 where P mi = −2g − 1. Then up to a scalar (which we may fix to equal one), u ⊗ ι∗(u) is sent by φ to the pull-back of Qi(t − αi)mi. Tracing through the definition of qw we get qw(u2 w) = (t − tw)Y (tw − αi)mi Y (tw − tw′)R(t). w′6=w π(D) = X miαi Therefore (3.7) (3.8) i i Q0(u2 w) := Y (tw − αi)mi Y w′6=w (tw − tw′) Q1(u2 w) := twQ0(u2 w). We recover the expression in Theorem 1.1 simply by replacing the basis uw with 1√Q0(u2 uw. To get the real version, we must choose a different basis. w) 4. Main Result Suppose now that π : Σ → CP 1 is a real hyperelliptic curve with anti-holomorphic involution τ compatible with the standard involution τP 1 on CP 1. Suppose this lifts to a real structure eτ on ξ and choose a eτ -invariant meromorphic section u. We want to replace the basis {uww ∈ W} of V by a basis whose elements are fixed by τ and with respect to which the expressions for Q0 and Q1 simplify. Given affine coordinates C∪{∞}, recall that if #W0 = 2n > 0, then π(Στ )∩ C is a union of n disjoint closed intervals in R, two of which are half infinite. Declare such an interval I ⊆ R to be odd with respect to π(D) = Pi miαi if Pαi∈I mi is odd. Note that a bounded interval is odd if and only if it is the image of an odd circle for ξ. To simplify what follows we require when n > 0 that the negative half infinite interval be even. This can always be arranged by changing the affine coordinate chart using a Moebius transformation. Theorem 4.1. Suppose that W0 consists of real numbers r1 < r2 < ... < r2n while W+ consists of complex numbers a1+ib1, ...., ag+1−n+ibg+1−n where bi > 0. Assume that ∞ ∈ π(Στ ). There is a choice of coordinates x1..., x2r, , z1, ..., zs, w1, ..., ws on V , in which the quadratic forms Q0 and Q1 are expressed as polynomials (4.1) q0 := (4.2) q1 := i=1 2nX 2nX i=1 2i−1 − x2 ǫi(cid:0)x2 ǫi(cid:0)r2i−1x2 2i(cid:1) + 2i−1 − r2ix2 j=1 sX (cid:0)z2 j(cid:1) j − w2 sX 2i(cid:1) + (cid:0)ajz2 j=1 6 j − ajz2 j + 2bjzjwj(cid:1) where ǫi ∈ {±1} equals (−1)(Ni+1) where Ni is the number of odd intervals to the right of r2i−1. This convention implies ǫ1 = 1. Proof. Clearly τP 1 preserves both π(D) and W . Let ν be the permutation of W such that τP 1(w) = ν(w). The induced anti-linear involution on V = ⊕w∈W ξ∗ sends a linear combination Pw∈W λwuw to Pw∈W λwuν(w). w) where Rw > 0 and eiθw ∈ [0, 2π). This Introduce the notation Rweiθw := Q0(u2 w is well-defined because from (3.7) we see Q0(u2 B := {vw w ∈ W0} ∪ {v′ w) 6= 0. Define the basis w, v′′ w w ∈ W+} where vw := v′ w := uw 1 √Rw e−iθw/2 √2Rw uw + uν(w) eiθw/2 √2Rw v′′ w := −i(cid:16) e−iθw/2 √2Rw uw − eiθw/2 √2Rw uν(w)(cid:17). Clearly each basis vector in B is fixed by conjugation. It is straightforward to check that Q0 is diagonal in this basis, with Q0(v2 w) = eiθw Q0((v′ Q0((v′ w)2) = 1 w)2) = −1 while Q1 satisfies Q1(v2 w) = tweiθw Q1((v′ Q1((v′ Q1(v′ w)2) = (tw + tw)/2 = Re(tw) w)2) = −(tw + tw)/2 = −Re(tw) wv′′ w) = −i(tw − tw)/2 = Im(tw) with all other entries zero. Finally, check that if w ∈ W0, then eiθw = (−1)N where N = #{w′ ∈ W0tw′ > tw} + X mi. αi∈R,αi>tw Therefore eiθw = ǫi if tw = r2i−1 and eiθw = −ǫi if tw = r2i as desired. (cid:3) 5. Stiefel-Whitney classes Let GC := Grg−1(C2g+2) with tautological bundle VC → GC and let G := Grg−1(R2g+2) and V be their real counterparts. We saw in §3.1 that there is a τ -equivariant embedding φ : Uξ ֒→ GC, which restricts to an embedding of U τ ξ into 7 G. The image of φ coincides with the set of (g − 1)-planes on which the quadratic forms Q0, Q1 both vanish. Proposition 5.1. The image of φ equals the zero locus Z(s) of a τ -invariant sec- tion s ∈ H 0(G; S2(V ∗ C )⊕2) which intersects the zero section transversely. Therefore ξ ∼= Z(s)τ = Z(sτ ) where sτ is restricted section of S2(V ∗). U τ Proof. The identification of Z(s) with linear subspaces of Q0, Q1 is explained in Borcea [6]. Transversality follows from ([6] Corollary 2.2) because the pencil of quadrics xQ0 + yQ1 is generic. Invariance under τ is clear. (cid:3) ξ0 and U τ1 ξ1 are cobordant. Corollary 5.2. Let (Σ0, τ0) and (Σ1, τ1) be real hyperelliptic curves of the same genus g ≥ 2 equipped with real line bundles ξ1 and ξ2 respectively of degree 2g + 1. Then U τ0 Proof. There exist real sections s0, s1 ∈ Γ(S2(V ∗)) such that U τi ξi = Z(si) := s−1 i (0). Choose a homotopy s : Grg−1(R2g+2) × I → S2(V ∗) from s0 to s1 which intersects the zero section transversally. Then Z(s) provides a cobordism between Z(s0) = M0 and Z(s1) = M1. (cid:3) Let N := im(φ)τ ∼= U τ (5.1) ξ . We have the following isomorphism of vector bundles T GN ∼= T N ⊕ S2(V ∗N ) ⊕ S2(V ∗N ) which can be used to compute Stiefel-Whitney classes. Proposition 5.3. Let N ∼= U τ tangent bundle of N is equals ξ be as above. The total Stiefel-Whitney class of the (5.2) w(T N) = w(V ∗N )2g+2w(V N ⊗ V ∗N )−1w(S2(V ∗)N )−2. In particular, we have w1(U τ w2(U τ ξ ) = 0 ξ ) = (g + 1)φ∗(w1)2 where wi is the tautological Stiefel-Whitney class in H i(G; Z2). Proof. By (5.1) and the Whitney sum formula, we get w(T N) = w(T GN )w(S2(V ∗)N )−2. We have the well-known isomorphism T G ∼= Hom(V, W ) where V and W are the tautological bundle and its orthogonal complement respectively. In particular, V ⊕ W ∼= G × R2g+2. Therefore T G ⊕ Hom(V, V ) = Hom(V, R2g+2) = V ∗ ⊕ ... ⊕ V ∗, so by the Whitney sum formula w(T G) = w(V ∗)2g+2w(V ⊗ V ∗)−1 8 proving (5.2). By definition w(V ) = w(V ∗) = 1 + w1 + w2 + .... + wg−1. A simple calculation using the splitting principle gives w(V ⊗ V ∗) = 1 + (rk(V ) − 1)w2 1 + O(3) = 1 + gw2 1 + O(3) w(S2(V ∗))2 = 1 + w1(S2(V ∗))2 + O(3) = 1 + gw2 1 + O(3) where O(3) is a sum of terms in degree three or higher. The values of wi(U τ ξ ) follow by direct calculation and functoriality. (cid:3) Before stating the next corollary we to introduce some terminology. Definition 5.4. Let L be a Lagrangian submanifold in a symplectic manifold (M, ω). We call L • relatively pin if w2(L) lies in the image of H 2(M; Z2) → H 2(L; Z2). • relatively spin if it is relatively pin and orientable. • spin if it is orientable and w2(L) = 0. Corollary 5.5. The Lagrangian submanifold U τ g ≥ 2 and is spin if g is odd. Proof. That U τ tion 5.3. Consider the commutative diagram of inclusions ξ is orientable follows from w1(U τ ξ ⊂ Uξ is relatively spin for all ξ ) = 0 which is proven in Proposi- Uξ U τ ξ φ φ / GC . i / G ξ ) = (g + 1)φ∗(w1)2 by Proposition 5.3, and we have w2 We have w2(U τ 1 = i∗(u) where u is the pull-back of the generator of H 2(GC; Z2) ([15] problem 15-A). Commutativity completes the argument. (cid:3) Remark 5.6. We note that the moduli space Uξ is monotone with minimal Chern number 2, which implies that U τ ξ is monotone with minimal Maslov number greater than or equal to two (see [2] Theorem 1.6). It follows from [4] that the quantum homology of the Lagrangian submanifold U τ ξ is well defined over the Novikov ring with integer coefficients. It is well-known that Dehn twist on Σ induces fibered Dehn twist of Uξ (e.g. [22] and references therein). For the real curve (Σ, τ ), we can consider a real Dehn twist. In this case, the corresponding moduli spaces of real bundles can be related by a Lagrangian cobordism in Lefschetz fibration (c.f. Biran-Cornea [5]). We hope to pursue this in future work. 9 / O O / O O 6. The genus 2 case Given a generic intersection of real quadrics X = Z(q0) ∩ Z(q1) ⊆ RP N we can form the double cover eX → X by pulling back the double cover SN → RP N . The diffeomorphism types of such eX were classified by Guti´errez and L´opez de Medrano [11], which we review below. Suppose q0 and q1 are determined by symmetric real matrices A and B. Up to a small perturbation that doesn't affect the diffeomorphism type, we may assume that A−1B is diagonalizable with distinct eigenvalues. Introduce real variables x1, ..., xr, u1, v1, ..., us, vs where each xi corresponds to the eigenspace for a real eigenvalue of A−1B and each pair ui, vi corresponds to the real part of the sum of eigenspaces for a complex conjugate pair of eigenvalues. Then the coefficients of q0, q1 can be continuously varied without changing the diffeomorphism type to a pair of quadratic forms (6.1) (6.2) p0 = p1 = i=1 rX rX i=1 aix2 i + bix2 i + j=1 sX sX j=1 (u2 j − v2 j ) 2ujvj where the bi/ai are the real eigenvalues of A−1B. Consider the set of points λi = (ai, bi) ∈ R2. The fact that the intersection is generic implies that (0, 0) does not lie on the line segment joining λi and λj for any pair i, j ∈ {1, ..., r}. If the coefficients λi are continuously varied without violating this property, then the diffeomorphism type of the intersection doesn't change. The λi can in this way be put into a standard form, such that all of the λi lie on 2l + 1 roots of unity (in R2 = C) for a minimal value l. This determines a cyclically ordered partition r = n1 +...+n2l+1 where ni > 0 counts the multiplicity of λi at the ith root of unity. The diffeomorphism type of eX, hence also X, is determined by s and the partition r = n1 + ..., +n2l+1. Proposition 6.1. Suppose that (Σ, τ ) is a real hyperelliptic curve of genus g = 2 with 2n real Weierstrass points and let ξ be a real line bundle with k many odd circles. Then the quadric intersection type of Z(q0)∩Z(q1) and the diffeomorphism type of the double cover eU τ ξ are as follows. (n,k) quadric intersection type diffeomorphism type of eU τ ξ (0,1) (1,1) (2,1) (3,1) (3,3) s=3, r=0 s=2, r=2 s=1, r=1+1+2 s=0, r=1+1+1+1+2 s=0, r=2+2+2 RP 3 S1 × S2 #3(S1 × S2) #5(S1 × S2) T 3 10 Proof. Comparing to formulas (4.1), (4.2) with (6.1), (6.2), we see that r = 2n is equal to the number of real Weierstrass points and 2s is the number of non-real Weierstrass points. The partitions can be worked out by hand case-by-case. The diffeomorphism types follows from the main theorem of [11]. (cid:3) Guti´errez and L´opez de Medrano do not provide a general formula for the dif- feomorphism type of the intersection of projective quadrics X itself, but it is not hard to determine it for our examples. Theorem 6.2. Suppose that (Σ, τ ) is a real hyperelliptic curve of genus 2 with 2n real Weierstrass points and let ξ be a real line bundle with k many odd circles (as for Theorem 6.1). The diffeomorphism type of U τ ξ is (n,k) diffeomorphism type of U τ ξ (0,1) (1,1) (2,1) (3,1) (3,3) #2(S1 × S2) #3(S1 × S2) L(4, 1) S1 × S2 T 3 Proof. For the case (0, 1) we know that U τ ξ quadric defined by is diffeomorphic to the projective 3X i=1 (cid:0)u2 i − v2 i(cid:1) = 3X i=1 2uivi = 0 If we regard these as affine equations and impose the extra affine condition 3X i=1 (cid:0)u2 i + v2 i(cid:1) = 1 then these define the unit tangent bundle of S2. Taking the projective quotient gives the unit tangent bundle of RP 2 which is diffeomorphic to the lens space L(4, 1) (see [13]). The cases (1,1), (2,1), and (3,1) are each diffeomorphic to an intersection of quadrics of the form 1 + x2 x2 2 + F (x3, x4, x5, x6) = 0 0 + F (x3, x4, x5, x6) = 0 which admits an SO(2)-action defined by rotating the coordinates x1, x2. Note that points in the intersection must either be fixed by this rotation (if x1 = x2 = 0) or have trivial stabilizer (because the first equation implies there are no non-zero solutions of the form (a1, a2, 0, 0, 0, 0)). In all three cases it is easy to verify that the fixed point set is non-empty. Since by Corollary 5.5 we also know that U τ ξ is orientable, a result of Raymond ([18] Theorem 1) implies that U τ ξ is diffeomorphic to a connected sum of copies of S1 × S2. The mod 2 Betti numbers of U τ ξ were 11 calculated in [2], from which we can determine the number of copies of S1 × S2 occurring in each case. The case (3, 3) actually admits an effective action by the 3-torus SO(2)3 (see (cid:3) [11] §4.1) and therefore must be diffeomorphic to a 3-torus. Remark 6.3. The case (3, 1) was considered by Saveliev and Wang [19]. They correctly proved U τ ξ has rational Poincar´e polynomial equal to 1 + 3t + 3t2 + t3, but is not homeomorphic to T 3. Remark 6.4. In the proof of Theorem 6.2, we made use of the existence of torus actions on (a manifold diffeomorphic to) U τ ξ . These circle actions may be related to the Jeffrey-Weitsman torus action [12] obtained from Goldman's integral system [9]. The Narasimhan-Seshadri Theorem determines a diffeomorphism between Uξ and the twisted representation variety M = Hom−1(π1(Σ), SU(2))/SU(2). Jeffrey and Weitsman produced Hamiltonian SO(2) actions defined on dense open subsets of M corresponding to embedded circles in the Riemann surface Σ. These SO(2) actions commute whenever the embedded circles are disjoint. The action is not defined everywhere on M because the Hamiltonian functions are not everywhere differentiable. However, in unpublished work by the first author it is shown that for an embedded circle coinciding with an odd circle for a real curve (Σ, τ ), the corresponding SO(2)-action restricts to a globally defined circle action on the submanifold M τ identified with U τ ξ . Therefore M τ admits a torus action of rank equal to the number of odd circles. This may correspond to the torus actions employed above. Acknowledgements: The authors would like to thank Joel Kamnitzer, Luis Haug and Francois Charette for insightful discussions. Thanks also to Guti´errez and Lopez de Medrano for answering our questions about their work. This research was supported in part by NSERC Discovery grants. References [1] T. J. Baird, Moduli spaces of vector bundles over a real curve: Z/2-Betti numbers, in Canadian Journal of Mathematics, 66(5), 961 -- 992, October, 2014. [2] T. J. Baird, Moduli spaces of vector bundles with fixed determinant over a real curve, arXiv preprint arXiv:1703.00778 (2017). [3] I. Biswas, J. Huisman and J. C. Hurtubise, The moduli space of stable vector bundles over a real algebraic curve, Mathematische Annalen, 347 (2010) no. 1, 201 -- 233. [4] P. Biran and O. Cornea, A Lagrangian quantum homology, in New perspectives and chal- lenges in symplectic field theory, 1 -- 44, CRM Proc. Lecture Notes, 49, Amer. Math. Soc., Providence, RI, 2009. [5] P. Biran and O. Cornea, Lagrangian cobordism in Lefschetz fibrations, arXiv preprint arXiv:1504.00922 (2015). [6] C. Borcea, Deforming varieties of k-planes of projective complete intersections, Pacific Jour- nal of Mathematics 143.1 (1990): 25-36. [7] C. Borcea, Homogeneous Vector Bundles and Families, Several Complex Variables and Complex Geometry 52.Part 2 (1991): 83. 12 [8] U. V. Desale and S. Ramanan, Classification of vector bundles of rank 2 on hyperelliptic curves, Inv. Math. 38 (1976), pp. 161 -- 185. [9] W. Goldman, Invariant functions on Lie groups and Hamiltonian flows of surface group representations, Invent. Math. vol. 85 (1986), 263-302. [10] P. Griffiths and J. Harris, Principles of Algebraic Geometry, John Wiley and Sons. Inc, 1978. [11] V. G. Guti´errez and S. Lopez de Medrano, Topology of the intersections of quadrics II, Boletn de la Sociedad Matemtica Mexicana 20.2 (2014): 237-255. [12] L. C. Jeffrey and J. Weitsman, Bohr-Sommerfeld orbits in the moduli space of flat connec- tions and the Verlinde dimension formula, Commun. Math. Phys. vol. 150 number 3 (1992): 593-630. [13] T. Konno, Unit tangent bundle over two-dimensional real projective space, Nihonkai Math- ematical Journal 13.1 (2002): 57-66. [14] C.-C. M. Liu, and F. Schaffhauser, The Yang-Mills equations over Klein surfaces, Journal of Topology 6.3 (2013): 569-643. [15] J. Milnor and J. D. Stasheff, Characteristic Classes, (AM-76). Vol. 76. Princeton university press, 2016. [16] M. S. Narasimhan and S. Ramanan, Moduli of vector bundles on a compact Riemann surface, Ann. Math. Second Series, 89 (1969) no. 1, pp 14 -- 51. [17] P. E. Newstead, Stable bundles of rank 2 and odd degree over a curve of genus 2, Topology, 7 (1968), 205 -- 215. [18] F. Raymond, Classification of the actions of the circle on 3-manifolds, Transactions of the American Mathematical Society 131.1 (1968): 51-78. [19] N. Saveliev and S. Wang, On real moduli spaces of holomorphic bundles over M -curves, Topology and its Applications 158.3 (2011): 344-351. [20] F. Schaffhauser, Real points of coarse moduli schemes of vector bundles on a real algebraic curve, Journal of Symplectic Geometry 10 (2012), no. 4, 503 -- 534. [21] S. Wang, Classification of real moduli spaces over genus 2 curves, Geom. Dedicata, 57 (1995), 207-215. [22] K. Wehrheim and C. Woodward, Exact triangle for fibered Dehn twists, Res. Math. Sci. (2016) 3:17. 13
1106.4529	1	1106	2011-06-22T18:54:15	A new offspring of PALP	[ "math.AG", "hep-th" ]	We describe the C program mori.x. It is part of PALP, a package for analyzing lattice polytopes. Its main purpose is the construction and analysis of three--dimensional smooth Calabi--Yau hypersurfaces in toric varieties. The ambient toric varieties are given in terms of fans over the facets of reflexive lattice polytopes. The program performs crepant star triangulations of reflexive polytopes and determines the Mori cones of the resulting toric varieties. Furthermore, it computes the intersection rings and characteristic classes of hypersurfaces.	math.AG	math	TUW-11-12 A new offspring of PALP Andreas P. Braun1 and Nils-Ole Walliser2 Institut fur Theoretische Physik, Vienna University of Technology, Wiedner Hauptstrasse 8-10, 1040 Vienna, Austria 1 1 0 2 n u J 2 2 ] . G A h t a m [ 1 v 9 2 5 4 . 6 0 1 1 : v i X r a Abstract We describe the C program mori.x. It is part of PALP, a package for analyzing lattice polytopes. Its main purpose is the construction and analysis of three -- dimensional smooth Calabi -- Yau hypersurfaces in toric varieties. The ambient toric varieties are given in terms of fans over the facets of reflexive lattice polytopes. The program performs crepant star triangulations of reflexive polytopes and determines the Mori cones of the resulting toric varieties. Furthermore, it computes the intersection rings and characteristic classes of hypersurfaces. In memory of Maximilian Kreuzer 1e-mail: abraun AT hep.itp.tuwien.ac.at 2e-mail: walliser AT hep.itp.tuwien.ac.at Contents 1 Introduction 2 I/O formats and options of mori.x 3 Supplementary details 3.1 Triangulations and point -- like singularities . . . . . . . . . . . . . . . . . . . . 3.2 The Kreuzer polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 The Mori cone of the ambient space . . . . . . . . . . . . . . . . . . . . . . . 3.4 The topological data of toric divisors . . . . . . . . . . . . . . . . . . . . . . . 4 Structure of the program 1 Introduction 2 3 7 7 9 9 9 10 PALP [1], a package for analyzing lattice polytopes, has the ability to construct, manipulate and analyze lattice polytopes. The present program mori.x, which is also included in the new version of PALP (starting from release 2.0) available at [2], adds further functionalities con- cerning Calabi -- Yau (CY) three -- fold hypersurfaces. The smoothness of the CY hypersurface is achieved by triangulating the reflexive polytope, i.e. appropriately resolving the ambient space. This functionality is optimized for the CY hypersurface case. Generic CY hypersur- faces avoid point -- like singularities of the ambient space as well as divisors that correspond to interior points of facets. Consequently, our algorithm performs crepant star triangulations only up to such interior points.1 Using these triangulations, the SR -- ideals and the inter- section rings are computed. The latter are determined with the help of SINGULAR [5], a computer algebra system for polynomial computations. Furthermore, we have implemented the Oda -- Park algorithm [6, 7] to find the Mori cone of the ambient space. The program can also analyze arbitrary three -- dimensional hypersurfaces embedded in the ambient toric varieties. It is capable of computing the intersection ring, characteristic classes and Hodge numbers. As we do not triangulate the polyhedron completely, non -- CY hypersurfaces can also contain singularities. Hence, the intersection numbers of non -- CY hypersurfaces determined by mori.x are not integers in general. The triangulations, which are at the heart of this program, are subject to two types of lim- itations. First, mori.x performs crepant star triangulation of the polytope by subdividing its facets into simplices. This procedure is, however, only implemented for those non -- simplicial cones which give rise to at most three -- dimensional secondary fans. Second, the limitations of PALP apply, as mori.x is a part of this package. PALP works with fixed precision; maximal 1Complete triangulations of arbitrary polytopes can be performed with the program TOPCOM [3], which is also included in the open -- source mathematics software system SAGE [4]. SAGE also contains various tools for handling toric varieties. 2 dimensions, maximal number of polytope points, vertices and facets need to be fixed before compiling. By editing the file Global.h, the user can change the default values if needed. See the PALP documentation [1] for further details. This note is meant as a manual for mori.x, an introduction to the mathematical back- ground is beyond its scope. We refer the reader to [8, 9] for introductions to toric geometry and the construction of CY hypersurfaces using polytopes. The structure of the paper is as follows. In section 2 we explain the I/O formats, present the options of mori.x and describe the basic functions of the main program. This section should be used as a quick reference for the various functionalities. Aspects which deserve further explanations are presented in section 3. There, we also discuss the resolution of singular compact toric spaces via (crepant star) triangulations. Section 4 describes the structure of the program. The principal routines are listed and commented on in the header file Mori.h. Acknowledgments: We dedicate this work to Maximilian Kreuzer who passed away be- fore it was completed. He contributed the core routines to the source code of mori.x. The present publication is an occasion to savor his work. We hope that we have been able to realize the original spirit in which he started this project. We would like to thank Harald Skarke for valuable advice in the completion phase of the program, and Christoph Mayrhofer for his comments and contributions to the source code at an early stage. Furthermore, we would like to thank Johanna Knapp, Volker Braun, Maximilian Attems, Ching -- Ming Chen and Michal Michalcik for discussion and comments. The work of N.-O. Walliser was supported by the FWF under grant P21239 and I192. The work of A. P. Braun was supported by the FWF under grant P21239 and P22000. 2 I/O formats and options of mori.x In this section we explain the I/O formats of mori.x and give an overview of its options. This program is part of the latest version of the package PALP that is available at the web page [2] under the GNU license terms. There, the reader can find a detailed installation description. In short, type "make all" in the command line to compile the entire suite of programs. Otherwise, type "make mori" to compile mori.x only. Consult [1] for the documentation of the other functions of the package: poly.x, cws.x, class.x and nef.x. Let us start with the help screen. It provides essential information about all the function- alities of the program (letters in boldface denote the command line input): $ mori.x -h This is ''mori.x'': performing crepant star triangulations of a polytope P* in the N-lattice computing the Mori cone of the corresponding toric ambient spaces determining the intersection rings of embedded (CY) hypersurfaces Usage: Options (concatenate any number of them into <Option-string>): mori.x [-<Option-string>] [in-file [out-file]] -h -f print this information use as filter 3 -g -I -m -P -K -b -i -c -t -d -a -D -H general output: triangulation and Stanley-Reisner ideal incidence information of the facets (ignoring interior points of facets) Mori generators of the ambient space IP-simplices among points of P* (ignoring interior points of facets) points of P* in Kreuzer polynomial form Hodge numbers and Euler number or arithmetic genera and Euler number if combined with -H intersection ring Chern classes of the (CY) hypersurface triple intersection numbers topological information on toric divisors and del Pezzo conditions all of the above except h, f, I and K lattice polytope points of P* as input (default CWS) arbitrary hypersurface class 'H = c1D1 + c2D2 + ...' as input (default CY) Input: coefficients 'c1 c2 ...' Input: 1) standard input format: degrees and weights 'd1 w11 w12 ... d2 w21 w22 ...' 2) alternative (use -D): 'd np' or 'np d' (d=Dimension, np=#[points]) and (after newline) npd coordinates Output: as specified by options As the program works with reflexive polytopes only, the input of the program must specify such a polytope. There are two input formats available: one either gives a combined weight system (CWS) or directly provides a matrix of lattice points whose convex hull is the polytope. The input is checked for reflexivity, see [1, 10] for details on how a CWS is converted into a polytope. The construction of a CY hypersurface uses a dual pair of reflexive lattice polytopes. One polytope determines the CY hypersurface, whereas its dual specifies the ambient toric variety. In this work, we are mainly interested in the N -- lattice polytope which gives rise to the fan of the ambient toric variety. We follow the conventions of the literature and PALP in referring to this polytope as the dual polytope P ∗. Let us consider a P1 fibered over a P3 as an example: $ mori.x -P Degrees and weights 5 1 1 1 1 1 0 2 0 0 0 0 1 1 4 7 points of P and IP-simplices 'd1 w11 w12 ... d2 w21 w22 ...': 0 1 0 0 0 0 0 -1 ------------------------------ 1 1 -1 -1 -1 1 1 0 0 0 1 0 0 0 1 0 1 0 0 0 0 1 0 1 1 0 1 0 0 0 0 0 #IP-simp=2 5=d 2=d codim=0 codim=3 Note that the ordering of the CWS input is not obeyed by the output of lattice points. Once the order is displayed, however, it is fixed and determines the labeling of toric divisors in any further output. An alternative way to provide the input is to type lattice polytope points directly. In this case, one has to use the parameter "-D". Let us reconsider the example above: 4 $ mori.x -DP '#lines #colums' (= 'PolyDim #Points' or '#Points PolyDim'): 4 7 Type the 28 coordinates as dim=4 lines with #pts=7 columns: 1 0 0 -1 0 1 0 -1 0 0 1 -1 0 0 0 4 7 0 0 0 0 0 0 0 0 0 1 -1 1 0 points of P* and IP-simplices -1 -1 -1 1 0 0 0 1 ------------------------------ 0 1 0 0 0 -1 1 0 0 0 1 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 1 1 0 0 0 0 #IP-simp=2 5=d 2=d codim=0 codim=3 The order of the lattice points displayed is the same as the order set in the input. This type of input should be preferred if one wants to control the order of the polytope points. Note that the IP -- simplices among points of an M -- lattice polytope P can be obtained from the lattice points of P ∗ with the option "-P" of poly.x. Since we focus on the N -- lattice polytope P ∗, this option is suppressed in mori.x for the sake of simplicity. The rest of this section contains a detailed description of the options listed in the help screen. If no flag is specified, the program starts with the parameter "-g". By default, the program considers a CY hypersurface embedded in the ambient toric variety. The option "-H" has to be used in order to consider non -- CY hypersurfaces. -h This option prints the help screen. -f -g This parameter suppresses the prompt of the command line. This is useful if one wants to build pipelines. First, the triangulation data of the facets is displayed. The number of triangulated simplices is followed by the incidence structure of the simplices. The incidence infor- mation for each simplex is encoded in terms of a binary number: there is a digit for each polytope point; a 1 denotes that the point belongs to the simplex. Second, the Stanley -- Reisner ideal is displayed: the number of elements of the ideal is followed by its elements. Each element is denoted by a binary number as above. -I The incidence structure of the facets of the polytope P ∗ is displayed. Interior points of the facets are neglected. -m The Mori cone generators of the ambient space are displayed in the form of a matrix. Each row corresponds to a generator. The entries of each row are the intersections of the generator with the toric divisor classes. The Oda -- Park algorithm is used to compute the generators. 5 -P Those IP -- simplices whose vertices are lattice points of the polytope, but not inner points of the facets, are displayed. -K The Kreuzer polynomial is displayed, see section 3.2 for details. The number of points in the interior of facets is shown as intpts. The multiplicities of the toric divisors are displayed as multd if they are greater than one. Furthermore, the Picard number of the CY hypersurface is computed and printed as Pic. -b We distinguish two cases. For the Calabi -- Yau hypersurface (default) case, the Hodge numbers h1,1, h2,1 and the Euler characteristic are displayed. If the input is an arbitrary hypersurface (see option "-H"), the zeroth and first arithmetic genera are displayed instead of the Hodge numbers. As a non -- CY hypersurface can be singular, care is needed for an interpretation of the results; see section 3.1 for more details. -i -c -t -d This option displays the intersection polynomial restricted to a CY hypersurface. The polynomial is displayed in terms of an integral basis of the toric divisors. The coefficients of the monomials are the triple intersection numbers in this basis. This option can also be used together with "-H" to perform this task for non -- CY hypersurfaces. The Chern classes of the hypersurface (CY or non -- CY) are displayed in terms of an integral basis of the toric divisors. The triple intersection numbers of the toric divisors are displayed. When the intersection numbers are equal to zero, they are not shown. This option displays topological data of the toric divisors restricted to the (CY or non -- CY) hypersurface. The Euler characteristics of the toric divisor classes and their arithmetic genera are shown. The toric divisor classes are tested against necessary conditions for del Pezzo surfaces. The following data is listed: the del Pezzo candidates preceded by their number (their type is given in brackets) and those del Pezzo candidates that do not intersect other del Pezzos. -a This is a shortcut for "-gmPbictd". -D This tells the program to expect a matrix of lattice points of the polytope P ∗ as the input. This is useful if one wants to set and control the order of the toric divisors. -H Using this option, one can specify a (non -- CY) hypersurface. The user determines the hypersurface divisor class H =Pi ciDi in terms of the toric divisor classes Di by typing its coefficients ci. The hypersurface can then be analyzed by combining "-H" with other options, as described above. Just using "-H", the program runs "-Hb". See section 3.1 for an example. Note that the options "-b, -i, -c, -t, -d, -a, -H" need SINGULAR to be installed. 6 3 Supplementary details We designed this program with an eye for applications to string theory model building, in particular the analysis of CY three -- folds. In this context, our motivation was to implement the construction of smooth CY hypersurfaces embedded in toric varieties, using the construction of Batyrev [11]. Here, the starting point is a dual pair of reflexive polytopes which determines both the ambient toric variety and a CY hypersurface. 3.1 Triangulations and point -- like singularities Batyrev [11] has shown that any four -- dimensional reflexive polytope gives rise to a smooth CY hypersurface after triangulation. The ambient toric space is not necessarily smooth, as some of the cones might have a volume greater than one even after triangulation. This leads to point -- like singularities, which, however, do not meet a generic CY hypersurface. The reflexivity of the polytopes sensibly simplifies the triangulation procedure. The induced simplices of lower dimension do not contribute any further singularities. mori.x performs crepant star triangulations of a polyhedron ignoring points in the interior of facets. Even though this introduces further point -- like singularities into the ambient toric variety, these are also avoided by generic CY hypersurfaces. Polytopes can be triangulated by subdividing the secondary fans of its non -- simplicial facets [12, 13]. This triangulation algorithm is implemented in mori.x for up to three -- dimensional secondary fans. The program exits with a warning message if the subdivision is not properly completed. Consider the following CWS: $ 4 8 echo -e '8 4 1 1 1 1 0 6 3 1 0 1 0 1' mori.x -fPI points of P* and IP-simplices 1 0 0 0 3 -1 3 -4 -1 0 -1 1 ------------------------------ 4 3 0 0 1 0 0 1 0 0 1 1 0 0 0 1 0 1 1 1 1 0 1 0 0 0 0 0 1 0 1 -1 #IP-simp=2 8=d 6=d codim=0 codim=1 Incidence: 101011 001111 111110 110101 011101 111001 100111 This system describes a four -- dimensional lattice polytope with eight points, six of which are vertices. We label the column vectors with v1, . . . , v8. Point v7 lies in the interior of the third facet. The incidence data show the intersections of the six polytope points with the seven facets. The third facet contains the five points v1, . . . , v5, hence it is not simplicial and we have to triangulate it. $ echo -e '8 4 1 1 1 1 0 9 Triangulation 101011 001111 110101 011101 111001 100111 011110 110110 111010 2 SR-ideal 6 3 1 0 1 0 1' mori.x -fg 7 101100 010011 8 Triangulation 101011 001111 110101 011101 111001 100111 101110 111100 2 SR-ideal 010010 101101 The program performs the two possible crepant triangulations of the facet h12345i. The first result yields the three simplices hd13425i, whereas the second gives the two simplices h1b2534i (in this notation the hat indicates that one of the points is dropped). Nevertheless, the two resolutions give the same CY intersection polynomial: 6 3 1 0 1 0 1' mori.x -fi $ echo -e '8 4 1 1 1 1 0 SINGULAR -> divisor classes (integral basis J1 ... J2): d1=J1, d2=-3J1+J2, d3=J1, d4=4J1-J2, d5=-3J1+J2, d6=J2 SINGULAR -> intersection polynomial: 2J13+108J23+8J12J2+30J22J1 SINGULAR -> divisor classes (integral basis J1 ... J2): d1=J1, d2=-3J1+J2, d3=J1, d4=4J1-J2, d5=-3J1+J2, d6=J2 SINGULAR -> intersection polynomial: 2J13+108J23+8J12J2+30J22J1 d1, . . . , d6 denote the toric divisors corresponding to the lattice points v1, . . . , v6. There are two independent divisor classes. Indeed, mori.x expresses the intersection polynomial in terms of the integral basis J1 = D1 = D3 and J2 = D6. Let us take a closer look at non -- CY hypersurfaces. The reader is warned: for these cases smoothness is not guaranteed anymore, so that the intersection numbers can become fractional. Some choices of the hypersurface equation may intersect point -- like singularities not resolved by the triangulation. Consider e.g. the hypersurface divisor class H = D1 + D6. Remember that the order in which mori.x expects the coefficients of the hypersurface divisor class is fixed by the polytope matrix and not by the CWS input. Hence, the correct input for H is the string 1 0 0 0 0 1. 'd1 w11 w12 ... d2 w21 w22 ...' $ mori.x -H Degrees and weights 8 4 1 1 1 1 0 6 3 1 0 1 0 1 Type the 6 (integer) entries for the hypersurface class: 1 0 0 0 0 1 Hypersurface degrees: ( 5 Hypersurface class: 1d1 1d6 SINGULAR chi_0: 35/32 , chi_1: 143/32 SINGULAR chi_0: 29/27 , chi_1: 128/27 -> Arithmetic genera and Euler number of H: -> Arithmetic genera and Euler number of H: 4 ) [ -27/4 ] [ -22/3 ] To calculate these quantities, the program determines the characteristic classes of the divisors using adjunction. It then performs the appropriate integration with the help of the triple intersection numbers. The fractional results of the arithmetic genera and the Euler number in our example indicate that the intersection polynomial has fractional entries. This happens because the singularity of the ambient toric variety descends to the hypersurface H. In the 8 first triangulation, the simplex h1235i still has volume four. In the second triangulation, there are two simplices with volume three: h1234i and h1345i. Indeed, any hypersurface in the divisor class H = D1 + D6 is forced to pass through the corresponding singularities. 3.2 The Kreuzer polynomial The Kreuzer polynomial encodes lattice polytope points in a compact form. The number of variables equals the dimension of the polytope. Each lattice point gives rise to a Laurent monomial in which the exponents of the variables are the coordinates. Vertices and non -- vertices are distinguished by coefficients "+" and "-" respectively. Points in the interior of facets are ignored. Consider the example presented above. $ echo -e '8 4 1 1 1 1 0 KreuzerPoly=t_13t_33/(t_2t_44)+t_1+t_2+t_4+t_3+t_4/(t_1t_3); intpts=1; 6 3 1 0 1 0 1' ./mori.x -fK Pic=2 Note that negative coordinates are always displayed by putting the variables in the denomi- nator. 3.3 The Mori cone of the ambient space For toric varieties, mori.x uses the algorithm of Oda and Park [6,7] to compute the Mori cone. The generators of the Mori cone are given in terms of their intersections with the toric divisors. For singular toric varieties, the Picard group of Cartier divisors is a non -- trivial subgroup of the Chow group, which contains the Weil divisors. Hence one can consider the Kahler cone, which is dual to the Mori cone, as a cone in the vector space spanned by the elements of either the Picard or the Chow group. The program mori.x only deals with simplicial toric varieties, for which the Picard group is always a finite index subgroup of the Chow group [8]. Hence the Cartier divisors are integer multiples of the Weil divisors and this ambiguity does not arise. 3.4 The topological data of toric divisors Using option "-d", mori.x computes the arithmetic genera of the toric divisors restricted to the embedded hypersurface and determines the del Pezzo candidates among them. The program checks the del Pezzo property against two necessary conditions: First, for a del Pezzo divisor S of type n, the following equations should hold: ZS c1(S)2 = 9 − n , ZS c2(S) = n + 3 =⇒ χ0(S) =ZS Td(S) = 1 . (1) Here, Td(S) denotes the Todd class of S, which gives the zeroth arithmetic genus of S upon integration. This test also allows to determine the type of the del Pezzo surface in question. The second necessary condition comes from the fact that a del Pezzo surface is a two -- dimensional Fano manifold. Hence, the first Chern class of S integrated over all curves on S 9 has to be positive: Di ∩ S ∩ c1(S) > 0 ∀Di : Di 6= S , Di ∩ S 6= 0 . (2) This condition would be sufficient if we were able to access all curves of the hypersurface. In our construction, however, we can only check for curves induced by toric divisors. This functionality was added to carry out the analysis of base manifolds for elliptic fibrations in [14]. 4 Structure of the program mori.x is part of the new releases of PALP (starting from version 2.0). The general structure of the package has not been changed, only some new files have been added. Hence, all general annotations to the package in [1] remain valid. In this section, we provide an overview of the composition of mori.x and discuss its dependencies on pre -- existing files. The source code of mori.x is contained in the program files mori.c, MoriCone.c, Singu- larInput.c, and the header file Mori.h. Makefile reflects the dependencies of the program. The compilation supports 32 as well as 64 bit architectures. One can adjust the compilation parameters according to ones needs. The optimization level is set at -03 by default. mori.c contains the main and the help information routines. Further, basic manipulations (completion, calculation of facet equations,...) of the polyhedron are performed with the help of core routines from Vertex.c. In particular, reflexivity of the polytope is checked. MoriCone.c is at the heart of mori.x. After determining the non -- simplicial facets, their triangulation is performed by the routine GKZSubdivide. This function identifies the maximal dimensional secondary fans of the facets and makes a case -- by -- case triangulation depending on their dimensions. This function is only implemented for secondary fans up to dimension three.2 Once the subdivision is accomplished, the program determines the Stanley -- Reisner ideal (StanleyReisner) and computes the Mori cone (Print_Mori). Furthermore, it finds a basis of the toric divisor classes. SingularInput.c is the interface to SINGULAR [5]. The latter is a very efficient computer algebra system for computations with polynomial rings. In SingularInput.c, the Chow ring is determined from the Stanley -- Reisner ideal, the linear relations among the toric divisors, and a basis of the toric divisors of the triangulated polytope. This data is put together by HyperSurfSingular and redirected to SINGULAR, which then determines the intersection ring restricted to the hypersurface and computes its characteristic classes. 3 The most important routines of mori.x are documented in the file Mori.h. This header file provides a more detailed description of the structure of the program. 2The pre -- compiler command TRACE TRIANGULATION in MoriCone.c enables diagnostic information about the triangulation. This data might be of use for the motivated programmer who wants to extend the subdivision algorithm. 3The input for SINGULAR can be displayed in the standard output of mori.x by turning on the pre -- compiler definition TEST PRINT SINGULAR IO in SingularInput.c. 10 References [1] M. Kreuzer, H. Skarke, "PALP: A Package for analyzing lattice polytopes with applica- tions to toric geometry," Comput. Phys. Commun. 157 (2004) 87-106. [math/0204356 [math-sc]]. [2] http://hep.itp.tuwien.ac.at/kreuzer/CY.html [3] Jrg Rambau, "TOPCOM: Triangulations of Point Configurations and Oriented Ma- troids", Mathematical Software - ICMS 2002 (Cohen, Arjeh M. and Gao, Xiao-Shan and Takayama, Nobuki, eds.), World Scientific (2002), pp. 330-340. [4] W. A. Stein et al., "Sage Mathematics Software", The Sage Development Team, http://www.sagemath.org [5] Decker, W.; Greuel, G.-M.; Pfister, G.; Schonemann, H.: "Singular 3-1-3 -- A computer algebra system for polynomial computations" (2011). http://www.singular.uni-kl.de [6] T. Oda and H. Park, "Linear Gale transforms and Gelfand-Kapranov-Zelevinskij decom- positions," Thoku Math. J. 43 (1991) 375399. [7] P. Berglund, S. H. Katz and A. Klemm, "Mirror symmetry and the moduli space for generic hypersurfaces in toric varieties," Nucl. Phys. B 456 (1995) 153 [arXiv:hep-th/9506091]. [8] W. Fulton, "Introduction to toric varieties", Annals of mathematics studies, Princeton University Press, Princeton (1993). T. Oda, "Convex Bodies and Algebraic Geometry," Springer-Verlag (1988). [9] D. A. Cox, S. Katz, "Mirror symmetry and algebraic geometry," Providence, USA: AMS (2000) 469 p. M. Kreuzer, "Toric geometry and Calabi-Yau compactifications," Ukr. J. Phys. 55 (2010) 613. [hep-th/0612307]. J. Knapp, M. Kreuzer, "Toric Methods in F-theory Model Building," [arXiv:1103.3358 [hep-th]]. [10] M. Kreuzer, H. Skarke, "Reflexive polyhedra, weights and toric Calabi-Yau fibrations," Rev. Math. Phys. 14 (2002) 343-374. [math/0001106 [math-ag]]. [11] V. V. Batyrev, "Dual polyhedra and mirror symmetry for Calabi -- Yau hypersurfaces in toric varieties," J. Alg. Geom. 3 (1994) 493-545. [alg-geom/9310003]. [12] L. J. Billera, P. Filliman, B. Sturmfels, Constructions and complexity of secondary poly- topes, Adv. in Math. 83 (1990) 155. 11 [13] I. M. Gelfand, M. M. Kapranov, A. V. Zelevinsky,Discriminants, Resultants, and Multidi- mensional Determinants, Birkhauser, Boston (1994). [14] J. Knapp, M. Kreuzer, C. Mayrhofer, N.-O. Walliser, "Toric Construction of Global F-Theory GUTs," JHEP 1103 (2011) 138. [arXiv:1101.4908 [hep-th]]. 12
1901.04479	1	1901	2019-01-14T07:56:50	Invariants of the bi-Lipschitz contact equivalence of continuous definable function germs	[ "math.AG" ]	We construct an invariant of the bi-Lipschitz contact equivalence of continuous function germs definable in a polynomially bounded o-minimal structure, such as semialgebraic functions. For a germ $f,$ the invariant is given in terms of the leading coefficients of the asymptotic expansions of $f$ along the connected components of the tangency variety of $f.$	math.AG	math	INVARIANTS OF THE BI-LIPSCHITZ CONTACT EQUIVALENCE OF CONTINUOUS DEFINABLE FUNCTION GERMS . N PHA. M AND NGUYE N THA' O NGUYEN B `UI TIE´N-SO Abstract. We construct an invariant of the bi-Lipschitz contact equivalence of con- tinuous function germs definable in a polynomially bounded o-minimal structure, such as semialgebraic functions. For a germ f, the invariant is given in terms of the leading coefficients of the asymptotic expansions of f along the connected components of the tangency variety of f. 1. Introduction Lipschitz geometry of maps is a rapidly growing subject in contemporary Singularity Theory. Recent progress in this area is due to the tameness theorems proved by several authors (see, for example, [1, 6, 8, 9, 10, 17]). However the description of a set of in- variants is barely developed (see also [2]). This paper presents a numerical invariant of continuous function germs definable in a polynomially bounded o-minimal structure (e.g., semialgebraic functions) with respect to the bi-Lipschitz contact equivalence. The most important ingredient of the invariant constructed here is the so-called tangency variety. More precisely, let f : (Rn, 0) → (R, 0) be a continuous function germ, which is definable in a polynomially bounded o-minimal structure. The tangency variety Γ(f ) of f consists of all points x in some neighborhood of the origin 0 ∈ Rn such that the fiber f −1(f (x)) is tangent to the sphere in Rn centered at 0 with radius kxk. The restriction of f on each connected component of Γ(f ) \ {0} defines a definable function fk of a single variable. Then the invariant of f is given in terms of the leading coefficients of the asymptotic expansions of these functions fk. The rest of the paper is organized as follows. In Section 2, we present some preliminaries which will be used later. The definition and some properties of tangency varieties are given in Section 3. The main result is provided in Section 4. Date: January 16, 2019. 2010 Mathematics Subject Classification. 14P15 · 32S05 · 03C64. Key words and phrases. Bi-Lipschitz contact equivalence, o-minimal structure, tangencies. The authors are partially supported by Vietnam National Foundation for Science and Technology Development (NAFOSTED), grant 101.04-2016.05. 1 2. Preliminaries Throughout this work we shall consider the Euclidean vector space Rn endowed with its canonical scalar product h·, ·i and we shall denote its associated norm k · k. The closed ball (resp., the sphere) centered at the origin 0 ∈ Rn of radius ǫ will be denoted by Bǫ (resp., Sǫ). 2.1. The bi-Lipschitz contact equivalence. The contact equivalence between (smooth) mappings was introduced by J. Mather [14]. The natural extension of Mather's definition to the Lipschitz setting in the function case appeared in [1], and to the general case in [17]. Let us start with the following definition. Definition 2.1. Two map germs f, g : (Rn, 0) → (Rp, 0) are called bi-Lipschitz con- tact equivalent (or K-bi-Lipschitz equivalent) if there exist two germs of bi-Lipschitz homeomorphisms h : (Rn, 0) → (Rn, 0) and H : (Rn × Rp, 0) → (Rn × Rp, 0) such that H(Rn × {0}) = Rn × {0} and the following diagram is commutative: (Rn, 0) (id,f ) −−−→ (Rn × Rp, 0) πn−−−→ (Rn, 0) (Rn, 0) (id,g) −−−→ (Rn × Rp, 0) πn−−−→ (Rn, 0) hy yH hy where id : Rn → Rn is the identity map and πn : Rn × Rp → Rn is the canonical projection. In this paper we consider the case p = 1, thus the maps f, g are functions. There is a more convenient way to work with the bi-Lipschitz contact equivalence of functions, due to the following result: Theorem 2.1 (see [1, Theorem 2.1]). Let f, g : (Rn, 0) → (R, 0) be two continuous func- tion germs. If f and g are bi-Lipschitz contact equivalent, then there exists a bi-Lipschitz homeomorphism germ h : (Rn, 0) → (Rn, 0), there exist positive constants c1, c2 and a sign σ ∈ {−1, 1} such that in a neighbourhood of the origin 0 ∈ Rn the following inequalities hold true c1f (x) ≤ σg(h(x)) ≤ c2f (x). 2.2. O-minimal structures. The notion of o-minimality was developed in the late 1980s after it was noticed that many proofs of analytic and geometric properties of semi-algebraic sets and maps could be carried over verbatim for sub-analytic sets and maps. We refer the reader to [4, 12, 13, 18, 19] for the basic properties of o-minimal structures used in this paper. Definition 2.2. An o-minimal structure on the real field R is a sequence S := (Sn)n∈N such that for each n ∈ N: 2 (a) Sn is a Boolean algebra of subsets of Rn. (b) If A ∈ Sm and B ∈ Sn, then A × B ∈ Sm+n. (c) If A ∈ Sn+1, then p(A) ∈ Sn, where p : Rn+1 → Rn is the projection on the first n coordinates. (d) Sn contains all algebraic subsets of Rn. (e) Each set belonging to S1 is a finite union of points and intervals. A set A ⊂ Rn is said to be a definable set if A ∈ Sn. A map f : A → Rm is said to be a definable map if its graph is definable. The structure S is said to be polynomially bounded if for every definable function f : R → R, there exist d ∈ N and R > 0 (depending on f ) such that f (x) ≤ xd for all x > R, Examples of (polynomially bounded) o-minimal structures are • the semi-linear sets, • the semi-algebraic sets (by the Tarski -- Seidenberg theorem), • the globally sub-analytic sets, i.e., the sub-analytic sets of Rn whose (compact) closures in RPn are sub-analytic (using Gabrielov's complement theorem). 2.3. Normals and subdifferentials. Here we recall the notions of the normal cones to sets and the subdifferentials of real-valued functions used in this paper. For more details we refer the reader to [15, 16]. Definition 2.3. Consider a set Ω ⊂ Rn and a point x ∈ Ω. (i) The regular normal cone (known also as the prenormal or Fr´echet normal cone) bNxΩ to Ω at x consists of all vectors v ∈ Rn satisfying hv, x′ − xi ≤ o(kx′ − xk) x′ → x with as x′ ∈ Ω. (ii) The limiting normal cone (known also as the basic or Mordukhovich normal cone) NxΩ to Ω at x consists of all vectors v ∈ Rn such that there are sequences xk → x with xk ∈ Ω and vk → v with vk ∈ bNxkΩ. NxΩ are equal to the normal space to Ω at x in the sense of differential geometry, i.e., If Ω is a manifold of class C 1, then for every point x ∈ Ω, the normal cones bNxΩ and bNxΩ = NxΩ and v ⊥ TxΩ for all v ∈ bNxΩ, where TxΩ stands for the tangent space of Ω at x; see [16, Example 6.8]. For a function f : Rn → R, we define the epigraph of f to be epif := {(x, y) ∈ Rn × R y ≥ f (x)}. A function f : Rn → R is said to be lower semi-continuous at x if it holds that Functional counterparts of normal cones are subdifferentials. lim inf x′→x f (x′) ≥ f (x). 3 Definition 2.4. Consider a function f : Rn → R and a point x ∈ Rn. The limiting and horizon subdifferentials of f at x are defined respectively by ∂f (x) ∂∞f (x) := (cid:8)v ∈ Rn (cid:12)(cid:12) (v, −1) ∈ N(x,f (x))epif(cid:9), := (cid:8)v ∈ Rn (cid:12)(cid:12) (v, 0) ∈ N(x,f (x))epif(cid:9). The limiting subdifferential ∂f (x) generalizes the classical notion of gradient. In par- ticular, for C 1-smooth functions f on Rn, the subdifferential consists only of the gradient ∇f (x) for each x ∈ Rn. The horizon subdifferential ∂∞f (x) plays an entirely different role -- it detects horizontal "normal" to the epigraph -- and it plays a decisive role in subdif- ferential calculus; see [16, Corollary 10.9] for more details. Theorem 2.2 (Fermat rule). Consider a lower semi-continuous function f : Rn → R and a closed set Ω ⊂ Rn. If ¯x ∈ Ω is a local minimizer of f on Ω and the qualification condition ∂∞f (¯x) ∩ N¯xΩ = {0} is valid, then the inclusion 0 ∈ ∂f (¯x) + N¯xΩ holds. We will also need the following lemma. Lemma 2.1. Consider a lower semi-continuous definable function f : Rn → R and a definable curve φ : [a, b] → Rn. Then for all but finitely many t ∈ [a, b], the mappings φ and f ◦ φ are C 1-smooth at t and satisfy v ∈ ∂f (φ(t)) =⇒ hv, φ′(t)i = (f ◦ φ)′(t), v ∈ ∂∞f (φ(t)) =⇒ hv, φ′(t)i = 0. [3, Proposition 4] and [5, Lemma 2.10]). Without loss of generality, assume Proof. (cf. that the curve φ is non-constant. In light of the monotonicity theorem [19, Theorem 4.1], there exists a real number ǫ ∈ (0, 1) such that on the open interval (0, ǫ) we have the mappings φ and f ◦ φ are C 1-smooth and φ′ is nonzero. Let M := {(φ(t), f (φ(t))) t ∈ (0, ǫ)}, which is a subset of the epigraph of f. Clearly, M is a connected definable C 1-manifold of dimension 1. Taking if necessary a smaller ǫ, we can be sure that there exists a Whitney C 1-stratification W of epif such that M is a stratum of W ; see [19, Theorem 4.8], for example. Take arbitrary (but fixed) t ∈ (0, ǫ) and v ∈ ∂f (φ(t)). By definition, there exist se- quences {xk} ⊂ U and {(vk, tk)} ⊂ bN(xk,f (xk))epif ⊂ Rn × R, such that xk → x := φ(t) and (vk, tk) → (v, −1) as k → ∞. Due to the finiteness property of W , we may suppose that the sequence {(xk, f (xk))} lies entirely in some stratum S ∈ W of dimension d. Using the compactness of the Grassmannian manifold of d-dimensional subspaces of Rn, we may 4 assume that the sequence of tangent spaces T(xk ,f (xk))S converges to some vector space T of dimension d. Then the Whitney-(a) property yields that T(x,f (x))M ⊂ T. By definition, for each k ≥ 1 we have that the vector (vk, tk) is Fr´echet normal to the epigraph epif of f at (xk, f (xk)); hence, it is also normal (in the classical sense) to the tangent space T(xk ,f (xk))S. By a standard continuity argument, the vector (v, −1) = lim k→∞ (vk, tk) must be normal to T and a fortiori to T(x,f (x))M. On the other hand, T(x,f (x))M is the vector space generated by the vector (φ′(t), (f ◦ φ)′(t)) ∈ Rn × R. Consequently, we obtain A similar argument also shows hv, φ′(t)i = (f ◦ φ)′(t). hv, φ′(t)i = 0 for all t ∈ (0, ǫ) and all v ∈ ∂∞(φ(t)). Finally, let c be the supremum of real numbers T ∈ [0, 1] such that for all but finitely many t ∈ [0, T ), we have for all v ∈ ∂f (φ(t)) and all w ∈ ∂∞(φ(t)), hv, φ′(t)i = (f ◦ φ)′(t) and hw, φ′(t)i = 0. Then c ≥ ǫ. We must prove that c = 1. Suppose that this is not the case. Replacing the interval [0, 1) by the interval [c, 1) and repeating the previous argument, we find a small real number ǫ′ > 0 such that for all t ∈ (c, c + ǫ′), all v ∈ ∂f (φ(t)) and all w ∈ ∂∞(φ(t)), hv, φ′(t)i = (f ◦ φ)′(t) and hw, φ′(t)i = 0, thus contradicting the definition of c. The proof is complete. (cid:3) 3. Tangencies Let f : (Rn, 0) → (R, 0) be a continuous definable function germ. Let us begin with the following definition (see also [7]). Definition 3.1. The tangency variety of f (at 0) is defined as follows: Γ(f ) := {x ∈ (Rn, 0) ∃λ ∈ R such that λx ∈ ∂f (x) ∪ ∂(−f )(x)}. Remark 3.1. When f is of class C 1 one has ∂f (x) = −∂(−f )(x) = {∇f (x)}, and so Γ(f ) = {x ∈ (Rn, 0) ∃λ ∈ R such that λx = ∇f (x)}. 5 By definition, it is not hard to check that Γ(f ) is a definable set. Moreover, thanks to the Fermat rule (Theorem 2.2), we can see that for any t > 0, the tangency variety Γ(f ) contains the set of minimizers (and minimizers) of f on the sphere St; in particular, 0 is a cluster point of Γ(f ). Applying the Hardt triviality theorem (see [19, Theorem 4.11]) for the definable function Γ(f ) → R, x 7→ kxk, we find a constant ǫ > 0 such that the restriction of this function on Γ(f ) ∩ Bǫ \ {0} is a topological trivial fibration. Let p be the number of connected components of a fiber of this restriction. Then Γ(f ) ∩ Bǫ \ {0} has exactly p connected components, say Γ1, . . . , Γp, and each such component is a definable set. Moreover, for all t ∈ (0, ǫ) and all k = 1, . . . , p, the sets Γk ∩ St are connected. Corresponding to each Γk, let fk : (0, ǫ) → R, t 7→ fk(t), be the function defined by fk(t) := f (x), where x ∈ Γk ∩ St. Lemma 3.1. For each ǫ > 0 small enough, all the functions fk are well-defined and definable. Proof. Fix k ∈ {1, . . . , p} and take any t ∈ (0, ǫ). We will show that the restriction of f on Γk ∩ St is constant. To see this, let φ : [0, 1] → Rn be a definable C 1-curve such that φ(τ ) ∈ Γk ∩ St for all τ ∈ [0, 1]. By definition, we have kφ(τ )k = t and either λ(τ )φ(τ ) ∈ ∂f (φ(τ )) or λ(τ )φ(τ ) ∈ ∂(−f )(φ(τ )) for some λ(τ ) ∈ R. By replacing f by −f, if necessary, we may assume that λ(τ )φ(τ ) ∈ ∂f (φ(τ )). In view of Lemma 2.1, for all but finitely many τ ∈ [a, b], the mappings φ and f ◦ φ are C 1-smooth at τ and satisfy v ∈ ∂f (φ(τ )) =⇒ hv, φ′(τ )i = (f ◦ φ)′(τ ). Therefore (f ◦ φ)′(τ ) = hλ(τ )φ(τ ), φ′(τ )i = λ(τ ) dkφ(τ )k2 2 dτ = 0. So f is constant on the curve φ. On the other hand, since the set Γk ∩ St is connected definable, it is path connected. Hence, any two points in Γk∩St can be joined by a piecewise C 1-smooth definable curve. It follows that the restriction of f on Γk ∩St is constant and so the function fk is well-defined. Finally, by definition, fk is definable. (cid:3) 6 For each t ∈ (0, ǫ), the sphere St is a nonempty compact definable set. Hence, the functions ψ : (0, ǫ) → R, t 7→ ψ(t) := min x∈St f (x), ψ : (0, ǫ) → R, t 7→ ψ(t) := max x∈St f (x), are well-defined and definable. The following lemma is simple but useful. Lemma 3.2. For ǫ > 0 small enough, the following equalities ψ(t) = min k=1,...,p fk(t) and ψ(t) = max k=1,...,p fk(t) hold for all t ∈ (0, ǫ). Proof. Applying the Curve Selection Lemma (see [19, Property 1.17]) and shrinking ǫ (if necessary), we find a definable C 1-curve φ : (0, ǫ) → Rn such that for all t ∈ (0, ǫ), kφ(t)k = t and (f ◦ φ)(t) = ψ(t). By Lemma 2.1, then we have for any t ∈ (0, ǫ), v ∈ ∂∞f (φ(t)) =⇒ hv, φ′(t)i = 0. Observe hφ(t), φ′(t)i = 1 2 d dt kφ(t)k2, and hence the qualification condition ∂∞f (φ(t)) ∩ Nφ(t)St = {0} holds for all t ∈ (0, ǫ). Consequently, since φ(t) minimizes f subject to kxk = t, applying the Fermat rule (Theorem 2.2), we deduce that φ(t) belongs to Γ(f ). Therefore, ψ(t) = min x∈St f (x) = min x∈Γ(f )∩St f (x) = min k=1,...,p min x∈Γk∩St f (x) = min k=1,...,p fk(t). Using the same argument, we also have ψ(t) = max x∈St f (x) = max x∈Γ(f )∩St f (x) = max k=1,...,p max x∈Γk∩St f (x) = max k=1,...,p fk(t). The lemma is proved. (cid:3) 7 4. The main result In this section, we fix a polynomially bounded o-minimal structure on R. The word "definable" will mean definable in this structure. Let f : (Rn, 0) → (R, 0) be a continuous definable function germ. As in the previous section, we associate to the function f a finite number of (definable) functions f1, . . . , fp of a single variable. Let K0 := {k fk is constant}. By the Growth Dichotomy Lemma (see [19, Theorem 4.12]), we can write for each k ∈ {1, . . . , p} \ K0, fk(t) = aktαk + o(tαk ) as t → 0+, where ak ∈ R, ak 6= 0, and αk ∈ R, αk > 0. Put K− := {k /∈ K0 ak < 0}, K+ := {k /∈ K0 ak > 0}. Finally we let Inv(f ) :=   (0, mink∈K+ αk) (− mink∈K− αk, 0) (− mink∈K− αk, mink∈K+ αk) (mink∈K+ αk, maxk∈K+ αk) (− mink∈K− αk, − maxk∈K− αk) (0, 0) if K0 6= ∅, K− = ∅ and K+ 6= ∅, if K0 6= ∅, K− 6= ∅ and K+ = ∅, if K− 6= ∅ and K+ 6= ∅, if K0 = K− = ∅ and K+ 6= ∅. if K0 = K+ = ∅ and K− 6= ∅. if K− = K+ = ∅. If Inv(f ) = (a, b), we follow the convention that −Inv(f ) := Inv(−f ) = (−b, −a). We now arrive to the main result of this paper. Theorem 4.1. Let f, g : (Rn, 0) → (R, 0) be two continuous definable function germs. If f and g are bi-Lipschitz contact equivalent then Inv(f ) = ±Inv(g). Proof. Since f and g are bi-Lipschitz contact equivalent, it follows from Theorem 2.1 that there exist a bi-Lipschitz homeomorphism germ h : (Rn, 0) → (Rn, 0) and some positive constants c1, c2 and a sign σ ∈ {±1} such that c1f (x) ≤ σ(g ◦ h)(x) ≤ c2f (x) for all kxk ≪ 1. (1) Assume that σ = 1. (The case σ = −1 is proved similarly.) Consider the definable functions ψf : [0, ǫ) → R, t 7→ ψf (t) := min x∈St f (x), ψf : [0, ǫ) → R, ψg : [0, ǫ) → R, t 7→ ψg(t) := min x∈St g(x), ψg : [0, ǫ) → R, t 7→ ψf (t) := max x∈St t 7→ ψg(t) := max x∈St f (x), g(x), 8 where ǫ is a positive number and small enough so that these functions are either constant or strictly monotone. Assume that we have proved the following relations: ψf ≃ ψg and ψf ≃ ψg, (2) where A ≃ B means that A/B lies between two positive constants. These, together with Lemma 3.2, imply easily that Inv(f ) = Inv(g), which is the desired conclusion. So we are left with showing (2). We will prove the first relation; the second one is Indeed, if ψf ≡ 0, then ψg ≡ 0 because of (1) and there is nothing proved similarly. to prove. So assume that ψf 6≡ 0. Since h is a bi-Lipschitz homeomorphism germ, there exists a positive constant L such that L−1kx − x′k ≤ kh(x) − h(x′)k ≤ L kx − x′k for all k(x, x′)k ≪ 1. In particular, we get L−1kxk ≤ kh(x)k ≤ Lkxk for all kxk ≪ 1. This, together with (1), implies that for all sufficiently small t ≥ 0, c2ψf (t) = c2 min x∈St f (x) ≥ min x∈St (g ◦ h)(x) ≥ min L−1t≤kh(x)k≤Lt (g ◦ h)(x) = min L−1t≤kyk≤Lt g(y). (3) (4) Let φ : [0, ǫ) → Rn be a definable curve such that g(φ(t)) = min L−1t≤kyk≤Lt g(y). Reducing ǫ if necessary, we may assume that φ is of class C 1 and that either L−1t < kφ(t)k < Lt or kφ(t)k = L−1t or kφ(t)k = Lt for all t ∈ [0, ǫ). If L−1t < kφ(t)k < Lt, then φ(t) is a local minimizer of the function g on the open set {y ∈ Rn L−1t < kyk < Lt}. By the Fermat rule (Theorem 2.2), we get 0 ∈ ∂g(φ(t)). This, together with Lemma 2.1, implies that for all but finitely many t ∈ [0, ǫ), (g ◦ φ)′(t) = h0, φ′(t)i = 0. Consequently, (g ◦ φ)(t) = (g ◦ φ)(0) = 0 for all t ∈ [0, ǫ), which is a contradiction. Therefore, we have kφ(t)k ≡ rt, where either r = L−1 or r = L. Moreover, it holds that min L−1t≤kyk≤Lt g(y) = min y∈Srt g(y) = ψg(rt) ≃ ψg(t). Combining this with (3) and (4), we can find a constant c > 0 such that c ψf (t) ≥ ψg(t) for all 0 ≤ t ≪ 1. Applying the above argument again and using the first inequality in (1), we also obtain c′ψg(t) ≥ ψf (t) for all 0 ≤ t ≪ 1 for some c′ > 0. Therefore, ψf ≃ ψg. (cid:3) 9 Remark 4.1. (i) Notice that, in the above proof, we do not assume that the homeomor- phism h is definable. (ii) When f is of class C 1, it is not hard to see that the exponents αk belong to the set of characteristic exponents defined by Kurdyka, Mostowski, and Parusi´nski [11], and moreover, the latter set is preserved by bi-Lipschitz homeomorphisms (see [9]). On the other hand, we do not know whether the set of the exponents αk is an invariant of the bi-Lipschitz contact equivalence or not. We conclude the paper with some examples illustrating our results. For simplicity we consider the case where f is a C 1-function in two variables (x, y) ∈ R2. By definition, then Γ(f ) := (cid:26)(x, y) ∈ R2 y ∂f ∂x − x ∂f ∂y = 0(cid:27) . In view of Theorem 4.1, the four functions given below are not bi-Lipschitz contact equiv- alent to each other. Example 4.1. (i) Let f (x, y) := x3 + y6. The tangency variety Γ(f ) is given by the equation: Hence, for ǫ > 0 the set (Γ(f ) ∩ Bǫ) \ {0} has six connected components: 3x2y − 6xy5 = 0. Γ±1 Γ±2 Γ±3 Consequently, := {(0, ±t) 0 < t < ǫ} , := (cid:8)(2t4, ±t) 0 < t < ǫ(cid:9) , := {(±t, 0) 0 < t < ǫ} . f Γ±1 = t6, f Γ±2 = t6 + 8t12, f Γ±3 = ±t3. It follows that K0 = ∅, K− = {−3}, K+ = {±1, ±2, 3} and Inv(f ) = {−3, 3}. (ii) Let f (x, y) := (x2 − y3)2. The tangency variety Γ(f ) is given by the equation: 2yx(3y − 2)(x2 − y3) = 0. Hence, for 0 < ǫ < 2 3, the set (Γ(f ) ∩ Bǫ) \ {0} has six connected components: Γ±1 Γ±2 Γ±3 := {(0, ±t) 0 < t < ǫ} , := (cid:8)(±t3, t2) 0 < t < ǫ(cid:9) , := {(±t, 0) 0 < t < ǫ} . 10 Consequently, f Γ±1 = t6, f Γ±2 = 0, f Γ±3 = t4. It follows that K0 = {±2}, K− = ∅, K+ = {±1, ±3} and Inv(f ) = {0, 4}. (iii) Let f (x, y) := x2 + y4. The tangency variety Γ(f ) is given by the equation: 2 , the set (Γ(f ) ∩ Bǫ) \ {0} has four connected components: Hence, for 0 < ǫ <q 1 Consequently, 2xy − 4xy4 = 0. Γ±1 Γ±2 := {(0, ±t) 0 < t < ǫ} , := {(±t, 0) 0 < t < ǫ} . f Γ±1 = t4, f Γ±2 = t2. It follows that K0 = K− = ∅, K+ = {±1, ±2} and Inv(f ) = {2, 4}. (iv) Let f (x, y) := −x2 − 2y6. The tangency variety Γ(f ) is given by the equation: 6, the set (Γ(f ) ∩ Bǫ) \ {0} has four connected components: Hence, for 0 < ǫ < 4q 1 Consequently, −2xy + 6xy5 = 0. Γ±1 Γ±2 := {(0, ±t) 0 < t < ǫ} , := {(±t, 0) 0 < t < ǫ} . f Γ±1 = −2t6, f Γ±2 = −t2. It follows that K0 = K+ = ∅, K− = {±1, ±2} and Inv(f ) = {−2, −6}. References [1] L. Birbrair, J. C. Costa, F. A. Fernandes, and M. A. S. Ruas. K-bi-lipschitz equivalence of real function-germs. Proc. Amer. Math. Soc., 135(4):1089 -- 1095, 2007. [2] L. Birbrair, A. Fernandes, A. Gabrielov, and V. Grandjean. Lipschitz contact equivalence of function germs in R2. Ann. Sc. Norm. Super. Pisa Cl. Sci., XVII(5):81 -- 92, 2017. [3] J. Bolte, A. Daniilidis, A. S. Lewis, and M. Shiota. Clarke subgradients of stratifiable functions. SIAM J. Optim., 18(2):556 -- 572, 2007. [4] M. Coste. An introduction to o-minimal geometry. Dip. Mat. Univ. Pisa, Dottorato di Ricerca in Matematica. Istituti Editoriali e Poligrafici Internazionali, Pisa, 2000. 11 [5] D. Drusvyatskiy and A. D. Ioffe. Quadratic growth and critical point stability of semi-algebraic functions. Math. Program. Ser. A, 153(2):635 -- 653, 2015. [6] T. Fukuda. Types topologiques des polynomes. Publ. Math. Inst. Hautes ´Etudes Sci., 46:87 -- 106, 1976. [7] H. V. H`a and T. S. Pha. m. Genericity in polynomial optimization, volume 3 of Series on Optimization and Its Applications. World Scientific, Singapore, 2017. [8] J.-P. Henry and A. Parusi´nski. Existence of moduli for bi-Lipschitz equivalence of analytic functions. Compositio Math., 136(2):217 -- 235, 2003. [9] J.-P. Henry and A. Parusi´nski. Invariants of bi-Lipschitz equivalence of real analytic functions. In Geometric singularity theory, volume 65, pages 67 -- 75. Banach Center Publ. Polish Acad. Sci. Inst. Math. Warsaw, 2004. [10] T.-C. Kuo. On classification of real singularities. Invent. Math., 82(2):257 -- 262, 1985. [11] K. Kurdyka, T. Mostowski, and A. Parusi´nski. Proof of the gradient conjecture of R. Thom. Ann. of Math. (2), 152(3):763 -- 792, 2000. [12] T. L. Loi. Lecture 1: O-minimal structures. In The Japanese-Australian Workshop on Real and Complex Singularities -- JARCS III, volume 43, pages 19 -- 30, Proc. Centre Math. Appl. Austral. Nat. Univ., Austral. Nat. Univ., Canberra, 2010. [13] T. L. Loi. Lecture 2: Stratifications in o-minimal structures. In The Japanese-Australian Workshop on Real and Complex Singularities -- JARCS III, volume 43, pages 31 -- 39, Proc. Centre Math. Appl. Austral. Nat. Univ., Austral. Nat. Univ., Canberra, 2010. [14] J. Mather. Stability of C 1 mappings III: Finitely determined map-germs. Inst. Hautes ´Etudes Sci. Publ. Math., 35:279 -- 308, 1968. [15] B. S. Mordukhovich. Variational analysis and generalized differentiation, I: Basic theory; II: Appli- cations. Springer, Berlin, 2006. [16] R. T. Rockafellar and R. Wets. Variational analysis, volume 317 of Grundlehren Math. Wiss. Springer-Verlag, Berlin, 1998. [17] M. Ruas and G. Valette. C 0 and bi-Lipschitz K-equivalence of mappings. Math. Z., 269(1 -- 2):293 -- 308, 2011. [18] L. van den Dries. Tame topology and o-minimal structures, volume 248 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 1998. [19] L. van den Dries and C. Miller. Geometric categories and o-minimal structures. Duke Math. J., 84:497 -- 540, 1996. Department of Mathematics, University of Dalat, 1 Phu Dong Thien Vuong, Dalat, Vietnam E-mail address: [email protected] Department of Pedagogy, University of Dalat, 1 Phu Dong Thien Vuong, Dalat, Viet- nam E-mail address: [email protected] 12
0910.0105	2	0910	2010-05-19T09:06:34	Generalized Donaldson-Thomas invariants	[ "math.AG", "hep-th" ]	This is a survey of the book arXiv:0810.5645 with Yinan Song. Let X be a Calabi-Yau 3-fold over C. The Donaldson-Thomas invariants of X are integers DT^a(t) which count stable sheaves with Chern character a on X, with respect to a Gieseker stability condition t. They are defined only for Chern characters a for which there are no strictly semistable sheaves on X. They have the good property that they are unchanged under deformations of X. Their behaviour under change of stability condition t was not understood until now. We discuss "generalized Donaldson-Thomas invariants" \bar{DT}^a(t). These are rational numbers, defined for all Chern characters a, and are equal to DT^a(t) if there are no strictly semistable sheaves in class a. They are deformation-invariant, and have a known transformation law under change of stability condition. We conjecture they can be written in terms of integral "BPS invariants" \hat{DT}^a(t) when the stability condition t is "generic". We extend the theory to abelian categories of representations of a quiver with relations coming from a superpotential, and connect our ideas with Szendroi's "noncommutative Donaldson-Thomas invariants" and work by Reineke and others. There is significant overlap between arXiv:0810.5645 and the independent paper arXiv:0811.2435 by Kontsevich and Soibelman.	math.AG	math	Generalized Donaldson -- Thomas invariants Dominic Joyce Abstract This is a survey of the book [16] with Yinan Song. Donaldson -- Thomas invariants DT α(τ ) ∈ Z 'count' τ -(semi)stable coherent sheaves with Chern character α on a Calabi -- Yau 3-fold X. They are unchanged under defor- mations of X. The conventional definition works only for classes α with no strictly τ -semistable sheaves. Behrend showed that DT α(τ ) can be st(τ )(cid:1) of the sta- written as a weighted Euler characteristic χ(cid:0)Mα ble moduli scheme Mα st(τ ) we call the 'Behrend function'. st(τ ), νMα st(τ ) by a constructible function νMα We discuss generalized Donaldson -- Thomas invariants ¯DT α(τ ) ∈ Q. These are defined for all classes α, and are equal to DT α(τ ) when it is defined. They are unchanged under deformations of X, and transform according to a known wall-crossing formula under change of stability con- dition τ . We conjecture that they can be written in terms of integral BPS invariants DT α(τ ) ∈ Z when the stability condition τ is 'generic'. We extend the theory to abelian categories mod-CQ/I of representa- tions of a quiver Q with relations I coming from a superpotential W on Q, and connect our ideas with Szendroi's noncommutative Donaldson -- Thomas invariants, and work by Reineke and others on invariants count- ing quiver representations. The book [16] has significant overlap with a recent, independent paper of Kontsevich and Soibelman [18]. 1 Introduction This is a survey of the book [16] by the author and Yinan Song. Let X be a Calabi -- Yau 3-fold over C, and OX (1) a very ample line bundle on X. Our definition of Calabi -- Yau 3-fold requires X to be projective, with H 1(OX ) = 0. Write coh(X) for the abelian category of coherent sheaves on X, and K(X) for the numerical Grothendieck group of coh(X). Let τ denote Gieseker stability of coherent sheaves w.r.t. OX (1). If E is a coherent sheaf on X then [E] ∈ K(X) is in effect the Chern character ch(E) of E in H even(X; Q). For α ∈ K(X) we can form the coarse moduli schemes Mα st(τ ) of τ -(semi)stable sheaves E with [E] = α. Then Mα ss(τ ) is a projective C-scheme whose points correspond to S-equivalence classes of τ -semistable sheaves, and Mα ss(τ ) whose points correspond to isomor- phism classes of τ -stable sheaves. st(τ ) is an open subscheme of Mα ss(τ ), Mα For Chern characters α with Mα st(τ ), following Donaldson and Thomas [4, §3], Thomas [33] constructed a symmetric obstruction theory on ss(τ ) = Mα 1 Mα st(τ ) and defined the Donaldson -- Thomas invariant to be the virtual class DT α(τ ) =R[Mα st(τ )]vir 1 ∈ Z, an integer which 'counts' τ -semistable sheaves in class α. Thomas' main re- sult [33, §3] is that DT α(τ ) is unchanged under deformations of the underlying Calabi -- Yau 3-fold X. Later, Behrend [1] showed that Donaldson -- Thomas in- variants can be written as a weighted Euler characteristic DT α(τ ) = χ(cid:0)Mα st(τ ), νMα st(τ )(cid:1), st(τ ) is the Behrend function, a constructible function on Mα st(τ ) de- where νMα pending only on Mα st(τ ) as a C-scheme. Conventional Donaldson -- Thomas invariants DT α(τ ) are only defined for classes α with Mα st(τ ), that is, when there are no strictly τ -semistable sheaves. Also, although DT α(τ ) depends on the stability condition τ , that is, on the choice of very ample line bundle OX (1) on X, this dependence was not understood until now. The main goal of [16] is to address these two issues. ss(τ ) = Mα For a Calabi -- Yau 3-fold X over C we will define generalized Donaldson -- Thomas invariants ¯DT α(τ ) ∈ Q for all α ∈ K(X), which 'count' τ -semistable sheaves in class α. These have the following important properties: • • If Mα ¯DT α(τ ) ∈ Q is unchanged by deformations of the Calabi -- Yau 3-fold X. st(τ ) then ¯DT α(τ ) lies in Z and equals the conventional ss(τ ) = Mα Donaldson -- Thomas invariant DT α(τ ) defined by Thomas [33]. • If Mα ss(τ ) 6= Mα st(τ ) then conventional Donaldson -- Thomas invariants DT α(τ ) are not defined for class α. Our generalized invariant ¯DT α(τ ) may lie in Q because strictly semistable sheaves E make (complicated) Q- valued contributions to ¯DT α(τ ). For 'generic' τ we have a conjecture that writes the ¯DT α(τ ) in terms of other, integer-valued invariants DT α(τ ). • If τ, τ are two stability conditions on coh(X), there is an explicit change of stability condition formula giving ¯DT α(τ ) in terms of the ¯DT β(τ ). These invariants are a continuation of the author's programme [9 -- 15]. We begin in §2 with some background material on constructible functions and stack functions on Artin stacks, taken from [9, 10]. Then §3 summarizes ideas from [11 -- 14] on Euler-characteristic type invariants J α(τ ) counting sheaves on Calabi -- Yau 3-folds and their wall-crossing under change of stability condition, and facts on Donaldson -- Thomas invariants from Thomas [33] and Behrend [1]. Section 4 summarizes [16, §5 -- §6], and is the heart of the paper. Let X be a Calabi -- Yau 3-fold, and M the moduli stack of coherent sheaves on X. Write ¯χ : K(X) × K(X) → Z for the Euler form of coh(X). We will explain that the 2 Behrend function νM of M satisfies two important identities νM(E1 ⊕ E2) = (−1) ¯χ([E1],[E2])νM(E1)νM(E2), Z[λ]∈P(Ext1(E2,E1)): λ ⇔ 0→E1→F →E2→0 νM(F )dχ −Z[λ′]∈P(Ext1(E1,E2)): λ′ ⇔ 0→E2→F ′→E1→0 νM(F ′)dχ =(cid:0)dim Ext1(E2, E1) − dim Ext1(E1, E2)(cid:1)νM(E1 ⊕ E2). al (M) → L(X), where We use these to define a Lie algebra morphism Ψ : SFind SFind al (M) is a special Lie subalgebra of the Ringel -- Hall algebra SFal(M) of X, a large algebra with a universal construction, and L(X) is a much smaller explicit Lie algebra, the Q-vector space with basis λα for α ∈ K(X), and Lie bracket [λα, λβ] = (−1) ¯χ(α,β) ¯χ(α, β)λα+β . If τ is Gieseker stability in coh(X) and α ∈ K(X), we define an element ¯ǫα(τ ) in SFind al (M) which 'counts' τ -semistable sheaves in class α in a special way. We define the generalized Donaldson -- Thomas invariant ¯DT α(τ ) ∈ Q by Ψ(cid:0)¯ǫα(τ )(cid:1) = − ¯DT α(τ )λα. By results in [14], the ¯ǫα(τ ) transform according to a universal transformation law in the Lie algebra SFind al (M) under change of stability condition. Applying Ψ shows that − ¯DT α(τ )λα transform according to the same law in L(X). This yields a wall-crossing formula for two stability conditions τ, τ on coh(X): ¯DT α(τ ) = Xiso. classes of finite sets I κ(i)=α Xconnected, Xκ:I→C(X): P i∈I simply- connected digraphs Γ, vertices I (−1)I−1V (I, Γ, κ; τ, τ ) ·Yi∈I 2 Pi,j∈I ¯χ(κ(i),κ(j)) ·Yedges ·(−1) 1 ¯DT κ(i)(τ ) (1) ¯χ(κ(i), κ(j)), i • → j • in Γ where V (I, Γ, κ; τ, τ ) ∈ Q are combinatorial coefficients, and there are only finitely many nonzero terms. To prove that ¯DT α(τ ) is unchanged under deformations of X, we introduce auxiliary invariants P I α,n(τ ′) ∈ Z counting 'stable pairs' s : O(−n) → E, for n ≫ 0 and E ∈ coh(X) τ -semistable in class α ∈ K(X). The moduli space Mα,n stp (τ ′) of such stable pairs is a proper fine moduli C-scheme with a symmetric obstruction theory, so by the same proof as for Donaldson -- Thomas invariants [33], the virtual count P I α,n(τ ′) of Mα,n stp (τ ′) is deformation-invariant. By a wall-crossing proof similar to that for (1) we find that P I α,n(τ ′) = Xα1,...,αl∈C(X), l>1: α1+···+αl=α, τ (αi)=τ (α), all i (−1)l l! l Yi=1(cid:2)(−1) ¯χ([OX (−n)]−α1−···−αi−1,αi) ¯χ(cid:0)[OX (−n)]−α1−· · ·−αi−1, αi(cid:1) ¯DT αi(τ )(cid:3). 3 Using deformation-invariance of the P I α,n(τ ′) and induction on rank α we find that ¯DT α(τ ) is deformation-invariant. Examples show that in general the ¯DT α(τ ) lie in Q rather than Z. So it is an interesting question whether we can rewrite the ¯DT α(τ ) in terms of some system of Z-valued invariants, just as Q-valued Gromov -- Witten invariants of Calabi -- Yau 3-folds are (conjecturally) written in terms of Z-valued Gopakumar -- Vafa invariants [7]. We define new BPS invariants DT α(τ ) for α ∈ C(X) to satisfy ¯DT α(τ ) = Xm>1, mα 1 m2 DT α/m(τ ), and we conjecture that DT α(τ ) ∈ Z for all α if the stability condition τ is 'generic'. Evidence for this conjecture is given in [16, §6.1 -- §6.5 & §7.6]. Section 5 summarizes [16, §7], which develops an analogue of Donaldson -- Thomas theory for representations of quivers with relations coming from a su- perpotential. This provides a kind of toy model for Donaldson -- Thomas in- variants using only polynomials and finite-dimensional algebra, and is a source of many simple, explicit examples. Counting invariants for quivers with su- perpotential have been studied by Nakajima, Reineke, Szendroi and other au- thors for some years [5, 24 -- 27, 29, 30, 32], under the general name of 'noncom- mutative Donaldson -- Thomas invariants'. Curiously, the invariants studied so far are the analogues of our pair invariants P I α,n(τ ′), and the analogues of ¯DT α(τ ), DT α(τ ) seem to have received no attention, although they appear to the author to be more fundamental. A recent paper by Kontsevich and Soibelman [18], summarized in [19], has considerable overlap with both [16] and the already published [9 -- 15]. The two were completed largely independently, and the first versions of [16,18] appeared on the arXiv within a few days of each other. Kontsevich and Soibelman are far more ambitious than us, working in triangulated categories rather than abelian categories, over general fields K rather than C, and with general motivic in- variants rather than the Euler characteristic. But for this reason, almost every major result in [18] depends explicitly or implicitly on conjectures. The author would like to acknowledge the contribution of [18] to the ideas on DT α(τ ) and integrality in §4.4 below, and to the material on quivers with superpotential in §5. The relationship between [16] and [18] is discussed in detail in [16, §1.6]. Acknowledgements. The author would like to thank Tom Bridgeland, Richard Thomas, Bal´azs Szendroi, and his co-author Yinan Song. This research was supported by EPSRC grant EP/D077990/1. 2 Constructible functions and stack functions We begin with some background material on Artin stacks, constructible func- tions, and stack functions, drawn from [9, 10]. We restrict to the field K = C. 4 2.1 Artin stacks and constructible functions Artin stacks are a class of geometric spaces, generalizing schemes and algebraic spaces. For a good introduction to Artin stacks see G´omez [6], and for a thor- ough treatment see Laumon and Moret-Bailly [20]. We work throughout over the field C. We make the convention that all Artin stacks in this paper are lo- cally of finite type, with affine geometric stabilizers, that is, all stabilizer groups IsoF(x) are affine algebraic C-groups, and substacks are locally closed. Artin C-stacks form a 2-category. That is, we have objects which are C-stacks F, G, and also two kinds of morphisms, 1-morphisms φ, ψ : F → G between C- stacks, and 2-morphisms A : φ → ψ between 1-morphisms. Definition 2.1. Let F be a C-stack. Write F(C) for the set of 2-isomorphism classes [x] of 1-morphisms x : Spec C → F. Elements of F(C) are called C-points of F. If φ : F → G is a 1-morphism then composition with φ induces a map of sets φ∗ : F(C) → G(C). For a 1-morphism x : Spec C → F, the stabilizer group IsoF(x) is the group of 2-morphisms A : x → x. When F is an Artin C-stack, IsoF(x) is an algebraic C-group, which we assume is affine. If φ : F → G is a 1-morphism, composition induces a morphism of C-groups φ∗ : IsoF([x]) → IsoG(cid:0)φ∗([x])(cid:1), for [x] ∈ F(C). We discuss constructible functions on C-stacks, following [9]. Definition 2.2. Let F be an Artin C-stack. We call C ⊆ F(C) constructible if C = Si∈I Fi(C), where {Fi : i ∈ I} is a finite collection of finite type Artin C-substacks Fi of F. We call S ⊆ F(C) locally constructible if S ∩ C is con- structible for all constructible C ⊆ F(C). A function f : F(C) → Q is called constructible if f (F(C)) is finite and f −1(c) is a constructible set in F(C) for each c ∈ f (F(C)) \ {0}. A function f : F(C) → Q is called locally constructible if f · δC is constructible for all constructible C ⊆ F(C), where δC is the char- acteristic function of C. Write CF(F) and LCF(F) for the Q-vector spaces of Q-valued constructible and locally constructible functions on F. Following [9, §4 -- §5] we define pushforwards and pullbacks of constructible functions along 1-morphisms. Definition 2.3. Let F, G be Artin C-stacks and φ : F → G a representable 1-morphism. For f ∈ CF(F), define CFstk(φ)f : G(C) → Q by CFstk(φ)f (y) = χ(cid:0)F ×φ,G,y Spec C, π∗ F(f )(cid:1) for y ∈ G(C), where F ×φ,G,y Spec C is a C-scheme (or algebraic space) as φ is representable, and χ(· · · ) is the Euler characteristic of this C-scheme weighted by π∗ F(f ). Then CFstk(φ) : CF(F) → CF(G) is a Q-linear map called the stack pushforward. Let θ : F → G be a finite type 1-morphism. The pullback θ∗ : CF(G) → CF(F) is given by θ∗(f ) = f ◦ θ∗. It is a Q-linear map. Here [9, §4 -- §5] are some properties of these. 5 Theorem 2.4. Let E, F, G, H be Artin C-stacks and β : F → G, γ : G → H be 1-morphisms. Then CFstk(γ ◦ β) = CFstk(γ) ◦ CFstk(β) : CF(F) → CF(H), (γ ◦ β)∗ = β∗ ◦ γ∗ : CF(H) → CF(F), (2) (3) supposing β, γ representable in (2), and of finite type in (3). If E η G θ ψ F φ / H is a Cartesian square with η, φ representable and θ, ψ of finite type, then the following commutes: CF(E) / CF(G) CFstk(η) θ∗ ψ∗ CF(F) CFstk(φ) CF(H). 2.2 Stack functions Stack functions are a universal generalization of constructible functions intro- duced in [10, §3]. Here [10, Def. 3.1] is the basic definition. Definition 2.5. Let F be an Artin C-stack. Consider pairs (R, ρ), where R is a finite type Artin C-stack and ρ : R → F is a representable 1-morphism. We call two pairs (R, ρ), (R′, ρ′) equivalent if there exists a 1-isomorphism ι : R → R′ such that ρ′ ◦ ι and ρ are 2-isomorphic 1-morphisms R → F. Write [(R, ρ)] for the equivalence class of (R, ρ). If (R, ρ) is such a pair and S is a closed C-substack of R then (S, ρS), (R \ S, ρR\S) are pairs of the same kind. Define SF(F) to be the Q-vector space generated by equivalence classes [(R, ρ)] as above, with for each closed C-substack S of R a relation [(R, ρ)] = [(S, ρS)] + [(R \ S, ρR\S)]. (4) Elements of SF(F) will be called stack functions. We relate CF(F) and SF(F). Definition 2.6. Let F be an Artin C-stack and C ⊆ F(C) be constructible. i=1 Ri(C), for R1, . . . , Rn finite type C-substacks of F. Let ρi : Ri → F be the inclusion 1-morphism. Then [(Ri, ρi)] ∈ SF(F). Define ¯δC = i=1[(Ri, ρi)] ∈ SF(F). We think of this as the analogue of the characteristic function δC ∈ CF(F) of C. Define a Q-linear map ιF : CF(F) → SF(F) by Then C = `n Pn ιF(f ) =P06=c∈f (F(C)) c · ¯δf −1(c). Define Q-linear πstk F : SF(F) → CF(F) by i=1 ci CFstk(ρi)1Ri, πstk F (cid:0)Pn i=1 ci[(Ri, ρi)](cid:1) =Pn where 1Ri is the function 1 ∈ CF(Ri). Then πstk F ◦ ιF is the identity on CF(F). The operations on constructible functions in §2.1 extend to stack functions. Definition 2.7. Let φ : F → G be a representable 1-morphism of Artin C-stacks. Define the pushforward φ∗ : SF(F) → SF(G) by φ∗ :Pm i=1 ci[(Ri, ρi)] 7−→Pm i=1 ci[(Ri, φ ◦ ρi)]. 6 (5) / / / / / / O O O O Let φ : F → G be of finite type. Define the pullback φ∗ : SF(G) → SF(F) by φ∗ :Pm i=1 ci[(Ri, ρi)] 7−→Pm The tensor product ⊗ : SF(F) × SF(G) → SF(F × G) is i=1 ci[(Ri ×ρi,G,φ F, πF)]. (cid:0)Pm i=1 ci[(Ri, ρi)](cid:1)⊗(cid:0)Pn j=1 dj[(Sj, σj)](cid:1) =Pi,j cidj[(Ri ×Sj, ρi ×σj)]. Here [10, Th. 3.5] is the analogue of Theorem 2.4. (6) (7) Theorem 2.8. Let E, F, G, H be Artin C-stacks and β : F → G, γ : G → H be 1-morphisms. Then (γ ◦β)∗ = γ∗◦β∗ : SF(F) → SF(H), (γ ◦β)∗= β∗ ◦γ∗ : SF(H) → SF(F), for β, γ representable in the first equation, and of finite type in the second. If E η G θ ψ F φ / H is a Cartesian square with θ, ψ of finite type and η, φ representable, then the following commutes: SF(E) θ∗ SF(F) η∗ φ∗ / SF(G) ψ∗ O SF(H). In [10, §3] we relate pushforwards and pullbacks of stack and constructible functions using ιF, πstk F . Theorem 2.9. Let φ : F → G be a 1-morphism of Artin C-stacks. Then (a) φ∗ ◦ιG = ιF ◦φ∗ : CF(G) → SF(F) if φ is of finite type; (b) πstk F : SF(F) → CF(G) if φ is representable; and G ◦ φ∗ = CFstk(φ) ◦ πstk F ◦ φ∗ = φ∗ ◦ πstk (c) πstk G : SF(G) → CF(F) if φ is of finite type. We define some projections Πvi n : SF(F) → SF(F), [10, §5]. Definition 2.10. For any Artin C-stack F we will define linear maps Πvi n : SF(F) → SF(F) for n > 0. Now SF(F) is generated by [(R, ρ)] with R 1- isomorphic to a quotient [X/G], for X a quasiprojective C-variety and G a special algebraic C-group, with maximal torus T G. Let S(T G) be the set of subsets of T G defined by Boolean operations upon closed C-subgroups L of T G. Define a measure dµn : S(T G) → Z to be additive upon disjoint unions of sets in S(T G), and to satisfy dµn(L) = 1 if dim L = n and dµn(L) = 0 if dim L 6= 0 for all algebraic C-subgroups L of T G. Define Πvi n(cid:0)[(R, ρ)](cid:1) = Zt∈T G {w ∈ W (G, T G) : w · t = t} W (G, T G) 7 (cid:2)(cid:0)[X {t}/CG({t})], ρ ◦ ι{t}(cid:1)(cid:3)dµn. (8) / / / / / / O O O Here X {t} is the subscheme of X fixed by t, and CG({t}) is the centralizer of t in G, and ι{t} : [X {t}/CG({t})] → [X/G] is the obvious 1-morphism. The integrand in (8), regarded as a function of t ∈ T G, is a constructible function taking only finitely many values. The level sets of the function lie in S(T G), so they are measurable w.r.t. dµn, and the integral is well-defined. In [10, §5] we show (8) induces a unique linear map Πvi n : SF(F) → SF(F). Here [10, §5] are some properties of the Πvi n . Theorem 2.11. In the situation above, we have: (i) (Πvi n )2 = Πvi n , so that Πvi n is a projection, and Πvi m ◦ Πvi n = 0 for m 6= n. (ii) For all f ∈ SF(F) we have f =Pn>0 Πvi n (f ) = 0 for n ≫ 0. as Πvi n (f ), where the sum makes sense (iii) If φ : F → G is a 1-morphism of Artin C-stacks then Πvi n ◦ φ∗ = φ∗ ◦ Πvi n : SF(F) → SF(G). (iv) If f ∈ SF(F), g ∈ SF(G) then Πvi m=0 Πvi m(f ) ⊗ Πvi n−m(g). n (f ⊗ g) =Pn Roughly speaking, Πvi n projects [(R, ρ)] ∈ SF(F) to [(Rn, ρ)], where Rn is the substack of points r ∈ R(C) whose stabilizer groups IsoR(r) have rank n. 2.3 Stack function spaces ¯SF(F, χ, Q) We will also need another family of spaces ¯SF(F, χ, Q), from [10, §5 -- §6]. Definition 2.12. Let F be an Artin C-stack. Consider pairs (R, ρ), where R is a finite type Artin C-stack and ρ : R → F is a representable 1-morphism, with equivalence as in Definition 2.5. Define ¯SF(F, χ, Q) to be the Q-vector space generated by equivalence classes [(R, ρ)], with the following relations: (i) Given [(R, ρ)] as above and S a closed C-substack of R we have [(R, ρ)] = [(S, ρS)] + [(R \ S, ρR\S)], as in (4). (ii) Let R be a finite type Artin C-stack, U a quasiprojective C-variety, πR : R × U → R the natural projection, and ρ : R → F a 1-morphism. Then [(R × U, ρ ◦ πR)] = χ([U ])[(R, ρ)]. Here χ(U ) ∈ Z is the Euler characteristic of U . It is a motivic invariant of C-schemes, that is, χ(U ) = χ(V ) + χ(U \ V ) for V ⊂ U closed. (iii) Given [(R, ρ)] as above and a 1-isomorphism R ∼= [X/G] for X a quasipro- jective C-variety and G a very special algebraic C-group acting on X with maximal torus T G, we have [(R, ρ)] =PQ∈Q(G,T G) F (G, T G, Q)(cid:2)(cid:0)[X/Q], ρ ◦ ιQ(cid:1)(cid:3), where ιQ : [X/Q] → R ∼= [X/G] is the natural projection 1-morphism. 8 Here Q(G, T G) is a certain finite set of C-subgroups of T G, and F (G, T G, Q) ∈ Q are a system of rational coefficients defined in [10, §6.2]. Define ¯Πχ,Q : SF(F) → ¯SF(F, χ, Q) by ¯Πχ,Q n on the spaces ¯SF(∗, χ, Q) φ∗, pullbacks φ∗, tensor products ⊗ and projections Πvi as in §2.2. The important point is that (5) -- (8) are compatible with the rela- tions defining ¯SF(∗, χ, Q), or they would not be well-defined. The analogues of Theorems 2.8, 2.9 and 2.11 hold for ¯SF(∗, χ, Q). :Pi∈I ci[(Ri, ρi)] 7→Pi∈I ci[(Ri, ρi)]. Define pushforwards F F Here [10, §5 -- §6] is a useful way to represent these spaces. It means that by working in ¯SF(F, χ, Q), we can treat all stabilizer groups as if they are abelian. Proposition 2.13. ¯SF(F, χ, Q) is spanned over Q by [(U × [Spec C/T ], ρ)], for U a quasiprojective C-variety and T an algebraic C-group isomorphic to Gk m×K for k > 0 and K finite abelian. Moreover Πvi n(cid:0)[(U × [Spec C/T ], ρ)](cid:1) =([(U × [Spec C/T ], ρ)], dim T = n, otherwise. 0, 3 Background material on Calabi -- Yau 3-folds We now summarize some facts on Donaldson -- Thomas invariants and other sheaf-counting invariants on Calabi -- Yau 3-folds prior to our book [16]. Sections 3.1 -- 3.3 review material from the author's series of papers [11 -- 14], and §3.4 ex- plains results on Donaldson -- Thomas theory from Thomas [33] and Behrend [1]. For simplicity we restrict to Calabi -- Yau 3-folds and to the field K = C, although much of [1, 11 -- 14, 33] works in greater generality. 3.1 The Ringel -- Hall algebra of a Calabi -- Yau 3-fold We will use the following notation for the rest of the paper. Definition 3.1. A Calabi -- Yau 3-fold is a smooth projective 3-fold X over C, with trivial canonical bundle KX . In §4 we will also assume that H 1(OX ) = 0. The Grothendieck group K0(X) of coh(X) is the abelian group generated by all isomorphism classes [E] of objects E in coh(X), with the relations [E] + [G] = [F ] for each short exact sequence 0 → E → F → G → 0. The Euler form ¯χ : K0(X) × K0(X) → Z is a biadditive map satisfying ¯χ(cid:0)[E], [F ](cid:1) =Pi>0(−1)i dim Exti(E, F ) for all E, F ∈ coh(X). As X is a Calabi -- Yau 3-fold, Serre duality gives Exti(F, E) ∼= Ext3−i(E, F )∗, so dim Exti(F, E) = dim Ext3−i(E, F ) for all E, F ∈ coh(X). Therefore ¯χ is also given by (9) ¯χ(cid:0)[E], [F ](cid:1) =(cid:0)dim Hom(E, F ) − dim Ext1(E, F )(cid:1)− (cid:0)dim Hom(F, E) − dim Ext1(F, E)(cid:1). 9 (10) Thus the Euler form ¯χ on K0(X) is antisymmetric. The numerical Grothendieck group K(X) is the quotient of K0(X) by the kernel of ¯χ. Then ¯χ on K0(X) descends to a nondegenerate, biadditive Euler form ¯χ : K(X) × K(X) → Z. Define the 'positive cone' C(X) in K(X) to be C(X) =(cid:8)[E] ∈ K(X) : 0 6∼= E ∈ coh(X)(cid:9) ⊂ K(X). Write M for the moduli stack of objects in coh(X). It is an Artin C-stack, locally of finite type. Points of M(C) correspond to isomorphism classes [E] of objects E in coh(X), and the stabilizer group IsoM([E]) in M is isomorphic as an algebraic C-group to the automorphism group Aut(E). For α ∈ C(X), write Mα for the substack of objects E ∈ coh(X) in class α in K(X). It is an open and closed C-substack of M. Write Exact for the moduli stack of short exact sequences 0 → E1 → E2 → E3 → 0 in coh(X). It is an Artin C-stack, locally of finite type. For j = 1, 2, 3 write πj : Exact → M for the 1-morphism projecting 0 → E1 → E2 → E3 → 0 to Ej. Then π2 is representable, and π1 × π3 : Exact → M × M is of finite type. In [12] we define Ringel -- Hall algebras, using stack functions. Definition 3.2. Define bilinear operations ∗ on SF(M), ¯SF(M, χ, Q) by f ∗ g = (π2)∗(cid:0)(π1 × π3)∗(f ⊗ g)(cid:1), using pushforwards, pullbacks and tensor products in Definition 2.7. They are well-defined as π2 is representable, and π1 × π3 is of finite type. By [12, Th. 5.2], whose proof uses Theorem 2.8, this * is associative, and makes SF(M), ¯SF(M, χ, Q) into noncommutative Q-algebras, called Ringel -- Hall algebras, with identity ¯δ[0], where [0] ∈ M is the zero object. The projection ¯Πχ,Q M : SF(M) → ¯SF(M, χ, Q) is an algebra morphism. As these algebras are inconveniently large for some purposes, in [12, Def. 5.5] we define subalgebras SFal(M), ¯SFal(M, χ, Q) using the algebra structure on stabilizer groups in M. Suppose [(R, ρ)] is a generator of SF(M). Let r ∈ R(C) with ρ∗(r) = [E] ∈ M(C), for some E ∈ coh(X). Then ρ induces a morphism of stabilizer C-groups ρ∗ : IsoR(r) → IsoM([E]) ∼= Aut(E). As ρ is representable this is injective, and induces an isomorphism of IsoR(r) with a C-subgroup of Aut(E). Now Aut(E) = End(E)× is the C-group of invertible elements in a finite-dimensional C-algebra End(E) = Hom(E, E). We say that [(R, ρ)] has algebra stabilizers if whenever r ∈ R(C) with ρ∗(r) = [E], the C- subgroup ρ∗(cid:0)IsoR(r)(cid:1) in Aut(E) is the C-group A× of invertible elements in a C-subalgebra A in End(E). Write SFal(M), ¯SFal(M, χ, Q) for the subspaces of SF(M), ¯SF(M, χ, Q) spanned over Q by [(R, ρ)] with algebra stabilizers. Then [12, Prop. 5.7] shows that SFal(M), ¯SFal(M, χ, Q) are subalgebras of the Ringel -- Hall algebras SF(M), ¯SF(M, χ, Q). Now [12, Cor. 5.10] shows that SFal(M), ¯SFal(M, χ, Q) are closed under the n on SF(M), ¯SF(M, χ, Q) defined in §2.2. In [12, Def. 5.14] we define operators Πvi 10 al (M), ¯SFind SFind 1 (f ) = f . We think of SFind with Πvi 'supported on virtual indecomposables'. al (M, χ, Q) to be the subspaces of f in SFal(M) and ¯SFal(M, χ, Q) al (M, χ, Q) as stack functions al (M), ¯SFind In [12, Th. 5.18] we show that SFind al (M, χ, Q) are closed un- der the Lie bracket [f, g] = f ∗ g − g ∗ f on SFal(M), ¯SFal(M, χ, Q). Thus, SFind al (M, χ, Q) are Lie subalgebras of SFal(M), ¯SFal(M, χ, Q). al (M), ¯SFind al (M), ¯SFind As in [12, Cor. 5.11], Proposition 2.13 simplifies to give: Proposition 3.3. ¯SFal(M, χ, Q) is spanned over Q by elements of the form [(U × [Spec C/Gk m], ρ)] with algebra stabilizers, for U a quasiprojective C-variety and k > 0. Also ¯SFind al (M, χ, Q) is spanned over Q by [(U × [Spec C/Gm], ρ)] with algebra stabilizers, for U a quasiprojective C-variety. All the above except (10) works for X an arbitrary smooth projective C- scheme, but our next result uses the Calabi -- Yau 3-fold assumption on X in an essential way. We follow [12, §6.5 -- §6.6], but use the notation of [16, §3.4]. Definition 3.4. Define an explicit Lie algebra L(X) over Q to be the Q-vector space with basis of symbols λα for α ∈ K(X), with Lie bracket [λα, λβ] = ¯χ(α, β)λα+β (11) for α, β ∈ K(X). As ¯χ is antisymmetric, (11) satisfies the Jacobi identity and makes L(X) into an infinite-dimensional Lie algebra over Q. al (M, χ, Q) → L(X) by Define a Q-linear map Ψχ,Q : ¯SFind Ψχ,Q(f ) =Pα∈K(X) γαλα, where γα ∈ Q is defined as follows. Proposition 3.3 says ¯SFind spanned by elements [(U × [Spec C/Gm], ρ)]. We may write al (M, χ, Q) is (12) (13) where δi ∈ Q and Ui is a quasiprojective C-variety. We set f Mα =Pn i=1 δi[(Ui × [Spec C/Gm], ρi)], This is independent of the choices in (13). Now define Ψ : SFind by Ψ = Ψχ,Q ◦ ¯Πχ,Q M . al (M) → L(X) i=1 δiχ(Ui). γα =Pn In [12, Th. 6.12], using equation (10), we prove: Theorem 3.5. Ψ : SFind Lie algebra morphisms. al (M) → L(X) and Ψχ,Q : ¯SFind al (M, χ, Q) → L(X) are 11 3.2 Stability conditions on coh(X) and invariants J α(τ ) Next we discuss material in [13] on stability conditions. We continue to use the notation of §3.1, with X a Calabi -- Yau 3-fold. Definition 3.6. Suppose (T, 6) is a totally ordered set, and τ : C(X) → T a map. We call (τ, T, 6) a stability condition on coh(X) if whenever α, β, γ ∈ C(X) with β = α + γ then either τ (α) < τ (β) < τ (γ), or τ (α) > τ (β) > τ (γ), or τ (α) = τ (β) = τ (γ). We call (τ, T, 6) a weak stability condition on coh(X) if whenever α, β, γ ∈ C(X) with β = α + γ then either τ (α) 6 τ (β) 6 τ (γ), or τ (α) > τ (β) > τ (γ). For such (τ, T, 6), we call a nonzero sheaf E in coh(X) (i) τ -stable if for all S ⊂ E with S 6∼= 0, E we have τ ([S]) < τ ([E/S]); and (ii) τ -semistable if for all S ⊂ E with S 6∼= 0, E we have τ ([S]) 6 τ ([E/S]). For α ∈ C(X), write Mα st(τ ) for the moduli stacks of τ -(semi)stable E ∈ A with class [E] = α in K(X). They are open C-substacks of Mα. We call (τ, T, 6) permissible if: ss(τ ), Mα (a) coh(X) is τ -artinian, that is, there exist no infinite chains of subobjects · · · ( E2 ( E1 ( E0 = X in A and τ ([En+1]) > τ ([En/En+1]) for all n; and (b) Mα ss(τ ) is a finite type substack of Mα for all α ∈ C(X). Here are two important examples: Example 3.7. Define G to be the set of monic rational polynomials in t of degree at most 3: G =(cid:8)p(t) = td + ad−1td−1 + · · · + a0 : d = 0, 1, 2, 3, a0, . . . , ad−1 ∈ Q(cid:9). Define a total order '6' on G by p 6 p′ for p, p′ ∈ G if either (a) deg p > deg p′, or (b) deg p = deg p′ and p(t) 6 p′(t) for all t ≫ 0. We write p < q if p 6 q and p 6= q. Fix a very ample line bundle OX (1) on X. For E ∈ coh(X), the Hilbert poly- nomial PE is the unique polynomial in Q[t] such that PE(n) = dim H 0(E(n)) for all n ≫ 0. Equivalently, PE(n) = ¯χ(cid:0)[OX (−n)], [E](cid:1) for all n ∈ Z. Thus, PE depends only on the class α ∈ K(X) of E, and we may write Pα instead of PE. Define τ : C(X) → G by τ (α) = Pα/rα, where Pα is the Hilbert polyno- mial of α, and rα is the (positive) leading coefficient of Pα. Then (τ, G, 6) is a permissible stability condition on coh(X) [13, Ex. 4.16], called Gieseker stability. st(τ ) for the coarse moduli schemes of τ -(semi)stable sheaves E with class [E] = α in K(X). By [8, Th. 4.3.4], Mα ss(τ ) is a projective C-scheme whose C-points correspond to S-equivalence classes of Gieseker semistable sheaves in class α, and Mα st(τ ) is an open C-subscheme whose C-points correspond to isomorphism classes of Gieseker stable sheaves in class α. Gieseker stability is studied in [8, §1.2]. Write Mα ss(τ ), Mα 12 Example 3.8. In the situation of Example 3.7, define M =(cid:8)p(t) = td + ad−1td−1 : d = 0, 1, 2, 3, ad−1 ∈ Q, a−1 = 0(cid:9) ⊂ G and restrict the total order 6 on G to M . Define µ : C(X) → M by µ(α) = td + ad−1td−1 when τ (α) = Pα/rα = td + ad−1td−1 + · · · + a0, that is, µ(α) is the truncation of the polynomial τ (α) in Example 3.7 at its second term. Then as in [13, Ex. 4.17], (µ, M, 6) is a permissible weak stability condition on coh(X). It is called µ-stability, and is studied in [8, §1.6]. In [13, §8] we define interesting stack functions ¯δα ss(τ ), ¯ǫα(τ ) in SFal(M). Definition 3.9. Let (τ, T, 6) be a permissible weak stability condition on coh(X). Define stack functions ¯δα ss(τ ) in SFal(M) for α ∈ C(X). That is, ¯δα ss(τ ) is the characteristic function, in the sense of Definition 2.6, of the moduli substack Mα ss(τ ) of τ -semistable sheaves in M. In [13, Def. 8.1] we define elements ¯ǫα(τ ) in SFal(M) by ss(τ ) = ¯δMα (−1)n−1 n ss (τ ) ∗ ¯δα2 ¯δα1 ss (τ ) ∗ · · · ∗ ¯δαn ss (τ ), (14) n>1, α1,...,αn∈C(X): α1+···+αn=α, τ (αi)=τ (α), all i ¯ǫα(τ ) = ¯δα ss(τ ) = X X where ∗ is the Ringel -- Hall multiplication in SFal(M). Then [13, Th. 8.2] proves n>1, α1,...,αn∈C(X): α1+···+αn=α, τ (αi)=τ (α), all i 1 n! ¯ǫα1 (τ ) ∗ ¯ǫα2 (τ ) ∗ · · · ∗ ¯ǫαn (τ ). (15) There are only finitely many nonzero terms in (14) -- (15). ss(τ ) = Mα ss(τ ). If Mα Equations (14) and (15) are inverse, so that knowing the ¯ǫα(τ ) is equivalent to knowing the ¯δα ss(τ ). The difference between ¯ǫα(τ ) and ¯δα ss(τ ) is that ¯ǫα(τ ) 'counts' strictly semistable sheaves in a special, complicated way. Here [13, Th. 8.7] is an important property of the ¯ǫα(τ ), which does not hold for the ¯δα ss(τ ). The proof is highly nontrivial, using the full power of the configurations formalism of [11 -- 14]. Theorem 3.10. ¯ǫα(τ ) lies in the Lie subalgebra SFind st(τ ) then ¯ǫα(τ ) = ¯δα al (M) in SFal(M). In [14, §6.6] we define invariants J α(τ ) ∈ Q for all α ∈ C(X) by This is valid by Theorem 3.10. These J α(τ ) are rational numbers 'counting' τ -semistable sheaves E in class α. When Mα ss(τ ) = Mα st(τ ) we have that is, J α(τ ) is the Euler characteristic of the moduli space Mα st(τ ). In the notation of §3.4, this is not weighted by the Behrend function νMα st(τ ), and so is not the Donaldson -- Thomas invariant DT α(τ ). As in [16, Ex. 6.9], the J α(τ ) are in general not unchanged under deformations of X. (16) (17) Ψ(cid:0)¯ǫα(τ )(cid:1) = J α(τ )λα. J α(τ ) = χ(cid:0)Mα st(τ )(cid:1), 13 3.3 Changing stability conditions and algebra identities In [14] we prove transformation laws for the ¯δα ss(τ ), ¯ǫα(τ ) under change of sta- bility condition. These involve combinatorial coefficients S(∗; τ, τ ) ∈ Z and U (∗; τ, τ ) ∈ Q defined in [14, §4.1]. Definition 3.11. Let (τ, T, 6),(τ , T , 6) be weak stability conditions on coh(X). Let n > 1 and α1, . . . , αn ∈ C(X). If for all i = 1, . . . , n − 1 we have either (a) τ (αi) 6 τ (αi+1) and τ (α1 + · · · + αi) > τ (αi+1 + · · · + αn) or (b) τ (αi) > τ (αi+1) and τ (α1 + · · · + αi) 6 τ (αi+1 + · · · + αn), then define S(α1, . . . , αn; τ, τ ) = (−1)r, where r is the number of i = 1, . . . , n−1 satisfying (a). Otherwise define S(α1, . . . , αn; τ, τ ) = 0. Now define U (α1, . . . , αn; τ, τ ) = X (−1)l−1 l ·Yl i=1 16l6m6n, 0=a0<a1<···<am=n, 0=b0<b1<···<bl=m: Define β1, . . . , βm ∈ C(X) by βi = αai−1 +1 + · · · + αai . Define γ1, . . . , γl ∈ C(X) by γi = βbi−1+1 + · · · + βbi . Then τ (βi) = τ (αj), i = 1, . . . , m, ai−1 < j 6 ai, and τ (γi) = τ (α1 + · · · + αn), i = 1, . . . , l S(βbi−1+1, βbi−1+2, . . . , βbi ; τ, τ ) · m Yi=1 1 (ai − ai−1)! . Then in [14, §5] we derive wall-crossing formulae for the ¯δα ss(τ ), ¯ǫα(τ ) under change of stability condition from (τ, T, 6) to (τ , T , 6): Theorem 3.12. Let (τ, T, 6), (τ , T , 6) be permissible weak stability conditions on coh(X). Then under some mild extra conditions, for all α ∈ C(X) we have ¯δα α1+···+αn=α ss(τ ) = Xn>1, α1,...,αn∈C(X): ¯ǫα(τ ) = Xn>1, α1,...,αn∈C(X): α1+···+αn=α S(α1, . . . , αn; τ, τ )· ss (τ ) ∗ ¯δα2 ¯δα1 ss (τ ) ∗ · · · ∗ ¯δαn ss (τ ), U (α1, . . . , αn; τ, τ )· ¯ǫα1(τ ) ∗ ¯ǫα2(τ ) ∗ · · · ∗ ¯ǫαn (τ ), (18) (19) where there are only finitely many nonzero terms in (18) -- (19). The 'mild extra conditions' in the theorem are required to ensure that there are only finitely many nonzero terms in (18) -- (19). In fact the author expects that this always holds when (τ, T, 6), (τ , T , 6) are of Gieseker or µ-stability type, but for irritating technical reasons has not been able to prove this. As in [14, §5.1], the author can show that one can go between any two (weak) stability conditions on coh(X) of Gieseker or µ-stability type by finitely many applications of Theorem 3.12. In [14, Th. 5.4] we prove: Theorem 3.13. Equation (19) may be rewritten as an equation in SFind using the Lie bracket [ , ] on SFind using the Ringel -- Hall product ∗. al (M) al (M), rather than as an equation in SFal(M) 14 Therefore we may apply the Lie algebra morphism Ψ of §3.1 to equation (19). As (19) is not expressed explicitly in terms of Lie brackets, it is helpful to write this in the universal enveloping algebra U (L(X)). This gives J α(τ )λα = Xn>1, α1,...,αn∈C(X): U (α1, . . . , αn; τ, τ ) · Qn α1+···+αn=α i=1 J αi (τ ) · λα1 ⋆ λα2 ⋆ · · · ⋆ λαn , (20) where ⋆ is the product in U (L(X)). Now in [12, §6.5], a basis is given for U (L(X)) in terms of symbols λ[I,κ], and multiplication ⋆ in U (L(X)) is written in terms of the λ[I,κ] as a sum over graphs. Here I is a finite set, κ maps I → C(X), and when I = 1, so that I = {i}, we have λ[I,κ] = λκ(i). Then [14, eq. (127)] gives an expression for λα1 ⋆ · · · ⋆ λαn in U (L(X)), in terms of sums over directed graphs (digraphs): λα1 ⋆ · · · ⋆ λαn = terms in λ[I,κ], I > 1, (21) +(cid:20) 1 2n−1 connected, simply-connected digraphs Γ: vertices {1, . . . , n}, edge i • → j • implies i < j X ¯χ(αi, αj)(cid:21)λα1+···+αn . Yedges j i • → • in Γ Substitute (21) into (20). The terms in λ[I,κ] for I > 1 all cancel, as (20) lies in L(X) ⊂ U (L(X)). So equating coefficients of λα yields J α(τ ) = Xn>1, α1,...,αn∈C(X): α1+···+αn=α connected, simply-connected digraphs Γ: vertices {1, . . . , n}, edge i • → j • implies i < j X 1 2n−1 U (α1, . . . , αn; τ, τ ) Yedges i • → ¯χ(αi, αj) j • in Γ n Yi=1 J αi(τ ). (22) Following [14, Def. 6.27], we define combinatorial coefficients V (I, Γ, κ; τ, τ ): Definition 3.14. In the situation above, let Γ be a connected, simply-connected digraph with finite vertex set I, where I = n, and κ : I → C(X) be a map. Define V (I, Γ, κ; τ, τ ) ∈ Q by V (I, Γ, κ; τ, τ ) = 1 2n−1n! Xorderings i1, . . . , in of I: edge ia• → ib• in Γ implies a < b U (κ(i1), κ(i2), . . . , κ(in); τ, τ ). (23) Then as in [14, Th. 6.28], using (23) to rewrite (22) yields a transformation law for the J α(τ ) under change of stability condition: J α(τ ) =Xiso. classes of finite sets I Pi∈I κ(i)=α Xconnected, Xκ:I→C(X): V (I, Γ, κ; τ, τ ) ·Yedges ·Yi∈I simply-connected digraphs Γ, vertices I ¯χ(κ(i), κ(j)) j i • → • in Γ J κ(i)(τ ). (24) 15 3.4 Donaldson -- Thomas invariants of Calabi -- Yau 3-folds Donaldson -- Thomas invariants DT α(τ ) were defined by Richard Thomas [33], following a proposal of Donaldson and Thomas [4, §3]. Definition 3.15. Let X be a Calabi -- Yau 3-fold. Fix a very ample line bundle OX (1) on X, and let (τ, G, 6) be Gieseker stability on coh(X) w.r.t. OX (1), as in Example 3.7. For α ∈ K(X), write Mα st(τ ) for the coarse moduli schemes of τ -(semi)stable sheaves E with class [E] = α. Then Mα ss(τ ) is a projective C-scheme, and Mα st(τ ) an open subscheme. ss(τ ), Mα Thomas [33] constructs a symmetric obstruction theory on Mα ss(τ ) = Mα that Mα Behrend and Fantechi [2] define a virtual class [Mα Donaldson -- Thomas invariant [33] is defined to be st(τ ). Then Mα st(τ ). Suppose st(τ ) is proper, so using the obstruction theory st(τ )). The st(τ )]vir ∈ A0(Mα DT α(τ ) =R[Mα st(τ )]vir 1. (25) Note that DT α(τ ) is defined only when Mα st(τ ), that is, there are no strictly semistable sheaves E in class α. One of our main goals is to extend the definition to all α ∈ K(X). Thomas' main result [33, §3] is that ss(τ ) = Mα Theorem 3.16. DT α(τ ) is unchanged by continuous deformations of the un- derlying Calabi -- Yau 3-fold X. An important advance in Donaldson -- Thomas theory was made by Behrend [1], who found a way to rewrite the definition (25) of Donaldson -- Thomas invari- ants as a weighted Euler characteristic. Let F be an Artin C-stack, locally of finite type. Then F has a unique Behrend function νF : F(C) → Z, a Z-valued locally constructible function on F. The definition, which we do not give, can be found in [1, §1] when F is a finite type C-scheme, and in [16, §4.1] in the general case. Here are some important properties of Behrend functions from [1, 16]. Theorem 3.17. Let F, G be Artin C-stacks locally of finite type. Then: (i) If F is a smooth of dimension n then νF ≡ (−1)n. (ii) If ϕ : F → G is smooth with relative dimension n then νF ≡ (−1)nϕ∗(νG). (iii) νF×G = νF ⊡ νG in LCF(F × G), where (νF ⊡ νG)(x, y) = νF(x)νG(y). (iv) Suppose M is a proper C-scheme and has a symmetric obstruction theory, and [M]vir ∈ A0(M) is the corresponding virtual class from [2]. Then R[M]vir 1 = χ(M, νM) ∈ Z, where χ(M, νM) = RM(C) νMdχ is the weighted Euler characteristic of M, weighted by the constructible function νM. In particular, R[M]vir 1 depends only on the C-scheme structure of M, not on the choice of sym- metric obstruction theory. 16 (v) Let M be a C-scheme, let x ∈ M(C), and suppose there exist a complex manifold U, a holomorphic function f : U → C, and a point u ∈ Crit(f ) ⊆ U such that locally in the analytic topology, M(C) near x is isomorphic as a complex analytic space to Crit(f ) near u. Then νM(x) = (−1)dim U(cid:0)1 − χ(M Ff (u))(cid:1), where χ(M Ff (u)) is the Euler characteristic of the Milnor fibre M Ff (u). Here the Milnor fibre in (v) is defined as follows: Definition 3.18. Let U be a complex analytic space, locally of finite type, f : U → C a holomorphic function, and u ∈ U . Let d( , ) be a metric on U near u induced by a local embedding of U in some CN . For u ∈ U and δ, ǫ > 0, consider the holomorphic map Φf,u :(cid:8)v ∈ U : d(u, v) < δ, 0 < f (v)−f (u) < ǫ(cid:9) −→(cid:8)z ∈ C : 0 < z < ǫ(cid:9) given by Φf,u(v) = f (v) − f (u). Then Φf,u is a smooth locally trivial fibration provided 0 < ǫ ≪ δ ≪ 1. The Milnor fibre M Ff (u) is the fibre of Φf,u. It is independent of the choice of 0 < ǫ ≪ δ ≪ 1. Theorem 3.17(iv) implies that DT α(τ ) in (25) is given by DT α(τ ) = χ(cid:0)Mα st(τ ), νMα st(τ )(cid:1). (26) ss(τ ) = Mα This is similar to the expression (17) for J α(τ ) when Mα st(τ ). There is a big difference between the two equations (25) and (26) defining Donaldson -- Thomas invariants. Equation (25) is non-local, and non-motivic, and makes sense only if Mα st(τ ) is a proper C-scheme. But (26) is local, and (in a sense) motivic, and makes sense for arbitrary finite type C-schemes Mα It is tempting to take (26) to be the definition of Donaldson -- Thomas invariants even when Mα st(τ ), but in [16, §6.5] we show that this is not a good idea, as then DT α(τ ) would not be unchanged under deformations of X. ss(τ ) 6= Mα st(τ ). Equation (26) was the inspiration for [16]. It shows that Donaldson -- Thomas invariants DT α(τ ) can be written as motivic invariants, like those studied in [11 -- 15], and suggests extending the results of [11 -- 15] to Donaldson -- Thomas invariants by including Behrend functions as weights. 4 Generalized Donaldson -- Thomas invariants We now summarize [16, §5 -- §6]. All this section is joint work with Yinan Song. Let X be a Calabi -- Yau 3-fold over C, and OX (1) a very ample line bundle over X. We now assume that H 1(OX ) = 0, which was not needed in §3. We use the notation of §3, with M the moduli stack of coherent sheaves on X, and so on. 17 4.1 Local description of the moduli of coherent sheaves In [16, Th. 5.5] we give a local characterization of an atlas for the moduli stack M as the critical points of a holomorphic function on a complex manifold. Theorem 4.1. Let X be a Calabi -- Yau 3-fold over C, and M the moduli stack of coherent sheaves on X. Suppose E is a coherent sheaf on X, so that [E] ∈ M(C). Let G be a maximal compact subgroup in Aut(E), and GC its complexification. Then GC is an algebraic C-subgroup of Aut(E), a maximal reductive subgroup, and GC = Aut(E) if and only if Aut(E) is reductive. There exists a quasiprojective C-scheme S, an action of GC on S, a point s ∈ S(C) fixed by GC, and a 1-morphism of Artin C-stacks Φ : [S/GC] → M, which is smooth of relative dimension dim Aut(E) − dim GC, where [S/GC] is the quotient stack, such that Φ(s GC) = [E], the induced morphism on stabi- lizer groups Φ∗ : Iso[S/GC](s GC) → IsoM([E]) is the natural morphism GC ֒→ Aut(E) ∼= IsoM([E]), and dΦs GC : TsS ∼= Ts GC[S/GC] → T[E]M ∼= Ext1(E, E) is an isomorphism. Furthermore, S parametrizes a formally versal family (S, D) of coherent sheaves on X, equivariant under the action of GC on S, with fibre Ds ∼= E at s. If Aut(E) is reductive then Φ is ´etale. Write San for the complex analytic space underlying the C-scheme S. Then there exists an open neighbourhood U of 0 in Ext1(E, E) in the analytic topol- ogy, a holomorphic function f : U → C with f (0) = df 0 = 0, an open neighbourhood V of s in San, and an isomorphism of complex analytic spaces Ξ : Crit(f ) → V, such that Ξ(0) = s and dΞ0 : T0 Crit(f ) → TsV is the in- verse of dΦs GC : TsS → Ext1(E, E). Moreover we can choose U, f, V to be GC-invariant, and Ξ to be GC-equivariant. The proof of Theorem 4.1 comes in two parts. First we show in [16, §8] that M near [E] is locally isomorphic, as an Artin C-stack, to the moduli stack Vect of algebraic vector bundles on X near [E′] for some vector bundle E′ → X. The proof uses algebraic geometry, and is valid for X an Calabi -- Yau m-fold for any m > 0 over any algebraically closed field K. The local morphism M → Vect is the composition of shifts and m Seidel -- Thomas twists by OX (−n) for n ≫ 0. Thus, it is enough to prove Theorem 4.1 with Vect in place of M. We do this in [16, §9] using gauge theory on vector bundles over X, motivated by an idea of Donaldson and Thomas [4, §3], [33, §2], and results of Miyajima [21]. Let E → X be a fixed complex (not holomorphic) vector bundle over X. Write A for the infinite-dimensional affine space of smooth semiconnections ( ¯∂-operators) on E, and G for the infinite-dimensional Lie group of smooth gauge transformations of E. Then G acts on A , and B = A /G is the space of gauge-equivalence classes of semiconnections on E. We fix ¯∂E in A coming from a holomorphic vector bundle structure on E. ¯∂E +A makes E into a holomorphic vector bundle if F 0,2 Then points in A are of the form ¯∂E +A for A ∈ C∞(cid:0)End(E)⊗C Λ0,1T ∗X(cid:1), and C∞(cid:0)End(E) ⊗C Λ0,2T ∗X(cid:1). Thus, the moduli space (stack) of holomorphic vec- tor bundle structures on E is isomorphic to { ¯∂E+A ∈ A : F 0,2 A = 0}/G . Thomas observes that when X is a Calabi -- Yau 3-fold, there is a natural holomorphic A = ¯∂EA+A∧A is zero in 18 function CS : A → C called the holomorphic Chern -- Simons functional, invari- ant under G up to addition of constants, such that { ¯∂E + A ∈ A : F 0,2 A = 0} is the critical locus of CS. Thus, Vect is (informally) locally the critical points of a holomorphic function CS on an infinite-dimensional complex stack B = A /G . To prove Theorem 4.1 we show that we can find a finite-dimensional complex submanifold U in A and a finite-dimensional complex Lie subgroup GC in G preserving U such that the theorem holds with f = CSU . In [16, Th. 5.11] we prove identities on the Behrend function of M, as in §3.4. Theorem 4.2. Let X be a Calabi -- Yau 3-fold over C, and M the moduli stack of coherent sheaves on X. The Behrend function νM : M(C) → Z is a natural locally constructible function on M. For all E1, E2 ∈ coh(X), it satisfies: νM(E1 ⊕ E2) = (−1) ¯χ([E1],[E2])νM(E1)νM(E2), Z[λ]∈P(Ext1(E2,E1)): λ ⇔ 0→E1→F →E2→0 νM(F ) dχ −Z[λ′]∈P(Ext1(E1,E2)): λ′ ⇔ 0→E2→F ′→E1→0 νM(F ′) dχ (27) (28) =(cid:0)dim Ext1(E2, E1) − dim Ext1(E1, E2)(cid:1)νM(E1 ⊕ E2). Here ¯χ([E1], [E2]) in (27) is defined in (9), and in (28) the correspondence be- tween [λ] ∈ P(Ext1(E2, E1)) and F ∈ coh(X) is that [λ] ∈ P(Ext1(E2, E1)) lifts to some 0 6= λ ∈ Ext1(E2, E1), which corresponds to a short exact sequence 0 → E1 → F → E2 → 0 in coh(X) in the usual way. The function [λ] 7→ νM(F ) is a constructible function P(Ext1(E2, E1)) → Z, and the integrals in (28) are integrals of constructible functions using the Euler characteristic as measure. We prove Theorem 4.2 using Theorem 4.1 and the Milnor fibre description of Behrend functions from Theorem 3.17(v). We apply Theorem 4.1 to E = E1 ⊕ E2, and we take the maximal compact subgroup G of Aut(E) to contain the subgroup (cid:8)idE1 +λ idE2 : λ ∈ U(1)(cid:9), so that GC contains (cid:8)idE1 +λ idE2 : λ ∈ Gm(cid:9). Equations (27) and (28) are proved by a kind of localization using this Gm-action on Ext1(E1 ⊕ E2, E1 ⊕ E2). Note that Theorem 4.2 makes sense as a statement in algebraic geometry, for Calabi -- Yau 3-folds over an algebraically closed field K of characteristic zero, and the author expects it to be true in this generality. However, our proof of Theorem 4.2 uses gauge theory, and transcendental complex analytic geometry methods, and is valid only over K = C. 4.2 A Lie algebra morphism Ψ : SFind al (M) → L(X), and generalized Donaldson -- Thomas invariants ¯DT α(τ ) al (M) → L(X) and Ψχ,Q : ¯SFind In §3.1 we defined an explicit Lie algebra L(X) and Lie algebra morphisms Ψ : SFind al (M, χ, Q) → L(X). We now define modified versions L(X), Ψ, Ψχ,Q, with Ψ, Ψχ,Q weighted by the Behrend function νM of M. We continue to use the notation of §2 -- §3. 19 Definition 4.3. Define a Lie algebra L(X) to be the Q-vector space with basis of symbols λα for α ∈ K(X), with Lie bracket [λα, λβ] = (−1) ¯χ(α,β) ¯χ(α, β)λα+β , (29) which is (11) with a sign change. As ¯χ is antisymmetric, (29) satisfies the Jacobi identity, and makes L(X) into an infinite-dimensional Lie algebra over Q. Define a Q-linear map Ψχ,Q : ¯SFind al (M, χ, Q) → L(X) by as in (12), where γα ∈ Q is defined as follows. Write f Mα in terms of δi, Ui, ρi as in (13), and set (30) Ψχ,Q(f ) =Pα∈K(X) γαλα, γα =Pn i=1 δiχ(cid:0)Ui, ρ∗ i (νM)(cid:1), where ρ∗ i (νM) is the pullback of the Behrend function νM to a constructible function on Ui × [Spec C/Gm], or equivalently on Ui, and χ(cid:0)Ui, ρ∗ Euler characteristic of Ui weighted by ρ∗ (13) to (30) is compatible with the relations in ¯SFind well-defined. Define Ψ : SFind al (M) → L(X) by Ψ = Ψχ,Q ◦ ¯Πχ,Q M . i (νM). One can show that the map from al (Mα, χ, Q), and so Ψχ,Q is i (νM)(cid:1) is the The reason for the sign change between (11) and (29) is the signs involved in Behrend functions, in particular, the (−1)n in Theorem 3.17(ii), which is responsible for the factor (−1) ¯χ([E1],[E2]) in (27). Here [16, Th. 5.14] is the analogue of Theorem 3.5. Theorem 4.4. Ψ : SFind Lie algebra morphisms. al (M) → L(X) and Ψχ,Q : ¯SFind al (M, χ, Q) → L(X) are We can now define generalized Donaldson -- Thomas invariants. Definition 4.5. Let X be a projective Calabi -- Yau 3-fold over C, let OX (1) be a very ample line bundle on X, and let (τ, G, 6) be Gieseker stability and (µ, M, 6) be µ-stability on coh(X) w.r.t. OX (1), as in Examples 3.7 and 3.8. ¯DT α(τ ) ∈ Q and As in (16), define generalized Donaldson -- Thomas invariants ¯DT α(µ) ∈ Q for all α ∈ C(X) by Ψ(cid:0)¯ǫα(τ )(cid:1) = − ¯DT α(τ )λα and Ψ(cid:0)¯ǫα(µ)(cid:1) = − ¯DT α(µ)λα. Here ¯ǫα(τ ), ¯ǫα(µ) are defined in (14), and lie in SFind al (M) by Theorem 3.10, so ¯DT α(τ ), ¯DT α(µ) are well-defined. In [16, Prop. 5.17] we show that if Mα ss(τ ) = st(τ ) then ¯DT α(τ ) = DT α(τ ). That is, our new generalized Donaldson -- Mα Thomas invariants ¯DT α(τ ) are equal to the original Donaldson -- Thomas invari- ants DT α(τ ) of [33] whenever the DT α(τ ) are defined. (31) We can now repeat the argument of §3.3 to deduce transformation laws for generalized Donaldson -- Thomas invariants under change of stability condition. In the situation of Theorem 3.12, equation (19) is an identity in the Lie algebra al (M), so we can apply the Lie algebra morphism Ψ to transform (19) into SFind 20 an identity in the Lie algebra L(X), and use (31) to write this in terms of generalized Donaldson -- Thomas invariants. As for (20), this gives an equation in the universal enveloping algebra U ( L(X)): ¯DT α(τ )λα = Xn>1, α1,...,αn∈C(X): U (α1, . . . , αn; τ, τ ) · (−1)n−1Qn α1+···+αn=α λα1 ⋆ λα2 ⋆ · · · ⋆ λαn . ¯DT αi(τ )· i=1 Following the proof of (24) in §3.3 with sign changes, in [16, Th. 5.18] we obtain: Theorem 4.6. In the situation of Theorem 3.12, for all α ∈ C(X) we have ¯DT α(τ ) = Xiso. classes of finite sets I κ(i)=α Xconnected, Xκ:I→C(X): P i∈I simply- connected digraphs Γ, vertices I (−1)I−1V (I, Γ, κ; τ, τ ) ·Yi∈I 2 Pi,j∈I ¯χ(κ(i),κ(j)) · Yedges ·(−1) 1 ¯DT κ(i)(τ ) (32) ¯χ(κ(i), κ(j)), i • → j • in Γ with only finitely many nonzero terms. The discussion after Theorem 3.12 implies [16, Cor. 5.19]: Corollary 4.7. Let (τ, T, 6), (τ , T , 6) be two permissible weak stability condi- tions on coh(X) of Gieseker or µ-stability type, as in Examples 3.7 and 3.8. Then the ¯DT α(τ ) for all α ∈ C(X) completely determine the ¯DT α(τ ) for all α ∈ C(X), and vice versa, through finitely many applications of (32). 4.3 Invariants P I α,n(τ ′) counting stable pairs, and deformation-invariance of the ¯DT α(τ ) We wish to prove that our invariants ¯DT α(τ ) are unchanged under deformations of X. We do this indirectly: we first define another family of auxiliary invariants P I α,n(τ ′) counting stable pairs on X, and show that P I α,n(τ ′) are unchanged under deformations of X. Then we prove an identity (35) expressing P I α,n(τ ′) in terms of the ¯DT β(τ ), and use it to show ¯DT α(τ ) is deformation-invariant. This approach was inspired by Pandharipande and Thomas [28], who use invariants counting pairs to study curve counting in Calabi -- Yau 3-folds. Definition 4.8. Let X be a Calabi -- Yau 3-fold over C, with H 1(OX ) = 0. Choose a very ample line bundle OX (1) on X, and write (τ, G, 6) for Gieseker stability w.r.t. OX (1), as in Example 3.7. Fix n ≫ 0 in Z. A pair is a nonzero morphism of sheaves s : OX (−n) → E, where E is a nonzero sheaf. A morphism between two pairs s : OX (−n) → E and t : OX (−n) → F is a morphism of OX -modules f : E → F , with f ◦ s = t. A pair s : OX (−n) → E is called stable if: (i) τ ([E′]) 6 τ ([E]) for all subsheaves E′ of E with 0 6= E′ 6= E; and (ii) If also s factors through E′, then τ ([E′]) < τ ([E]). 21 Note that (i) implies that if s : OX (−n) → E is stable then E is τ -semistable. The class of a pair s : OX (−n) → E is the numerical class [E] in K(X). We will use τ ′ to denote stability of pairs, defined using OX (1). In [16, Th.s 5.22 & 5.23] we use results of Le Potier to prove: Theorem 4.9. If n is sufficiently large then the moduli functor of stable pairs has a fine moduli scheme, a projective C-scheme Mα,n stp (τ ′), with a symmetric obstruction theory. Definition 4.10. In the situation above, for α ∈ K(X) and n ≫ 0, define stable pair invariants P I α,n(τ ′) in Z by P I α,n(τ ′) =R[Mα,n stp (τ ′)]vir 1, (33) stp (τ ′)]vir ∈ A0(Mα,n where [Mα,n stp (τ ′)) is the virtual class constructed by Behrend and Fantechi [2] using the symmetric obstruction theory from Theorem 4.9. Theorem 3.17(iv) implies that the stable pair invariants may also be written P I α,n(τ ′) = χ(cid:0)Mα,n stp (τ ′), νMα,n stp (τ ′)(cid:1). (34) In [16, Cor. 5.26] we prove an analogue of Theorem 3.16: Theorem 4.11. P I α,n(τ ′) is unchanged by continuous deformations of the un- derlying Calabi -- Yau 3-fold X. In [16, Th. 5.27] we express the pair invariants P I α,n(τ ′) above in terms of the generalized Donaldson -- Thomas invariants ¯DT β(τ ) of §4.2. Equation (35) is a wall-crossing formula similar to (32), and we prove it by change of stability condition in an auxiliary abelian category. Theorem 4.12. For α ∈ C(X) and n ≫ 0 we have P I α,n(τ ′) = Xα1,...,αl∈C(X), l>1: α1+···+αl=α, τ (αi)=τ (α), all i l (−1)l l! Yi=1(cid:2)(−1) ¯χ([OX (−n)]−α1−···−αi−1,αi) ¯χ(cid:0)[OX (−n)]−α1−· · ·−αi−1, αi(cid:1) ¯DT αi(τ )(cid:3), (35) where there are only finitely many nonzero terms in the sum. Equation (35) is useful for computing invariants ¯DT α(τ ) in examples. By combining Theorems 4.11 and 4.12 and using induction on the leading coefficient of the Hilbert polynomial of α, we deduce [16, Cor. 5.28]: Corollary 4.13. The generalized Donaldson -- Thomas invariants ¯DT α(τ ) de- fined in §4.2 are unchanged under continuous deformations of the underlying Calabi -- Yau 3-fold X. 22 4.4 Integrality properties of the ¯DT α(τ ) This subsection is based on ideas in Kontsevich and Soibelman [18, §2.5 & §7.1]. The following example is taken from [16, Ex.s 6.1 & 6.2]. Example 4.14. Let X be a Calabi -- Yau 3-fold over C equipped with a very ample line bundle OX (1). Suppose α ∈ C(X), and that E ∈ coh(X) with p m copies q [E] = α is τ -stable and rigid, so that Ext1(E, E) = 0. Then mE = E ⊕ · · · ⊕ E for m > 2 is a strictly τ -semistable sheaf of class mα, which is also rigid. For simplicity, assume that mE is the only τ -semistable sheaf of class mα for all m > 1, up to isomorphism, so that Mmα ss (τ ) = {[mE]}. A pair s : O(−n) → mE may be regarded as m elements s1, . . . , sm of H 0(E(n)) ∼= CPα(n), where Pα is the Hilbert polynomial of E. Such a pair turns out to be stable if and only if s1, . . . , sm are linearly independent in H 0(E(n)). Two such pairs are equivalent if they are identified under the action of Aut(mE) ∼= GL(m, C), acting in the obvious way on (s1, . . . , sm). Thus, equivalence classes of stable pairs correspond to linear subspaces of dimension m in H 0(E(n)), so the moduli space Mmα,n (τ ′) is isomorphic as a C-scheme to the Grassmannian Gr(Cm, CPα(n)). This is smooth of dimension m(Pα(n) − m), (τ ′) ≡ (−1)m(Pα(n)−m) by Theorem 3.17(i). Also Gr(Cm, CPα(n)) so that νMmα,n stp has Euler characteristic the binomial coefficient (cid:0)Pα(n) m (cid:1). Therefore (34) gives P I mα,n(τ ′) = (−1)m(Pα(n)−m)(cid:0)Pα(n) m (cid:1). Consider (35) with mα in place of α. If α1, . . . , αl give a nonzero term on the right hand side of (35) then mα = α1 + · · · + αl, and ¯DT αi (τ ) 6= 0, so there exists a τ -semistable Ei in class αi. Thus E1 ⊕ · · · ⊕ El lies in class mα, and is τ -semistable as τ (αi) = τ (α) for all i. Hence E1 ⊕ · · · ⊕ El ∼= mE, which implies that Ei ∼= kiE for some k1, . . . , kl > 1 with k1 + · · · + kl = m, and αi = kiα. (36) Setting αi = kiα, we see that ¯χ(αj , αi) = 0 and ¯χ([OX (−n)], αi) = kiPα(n), where Pα is the Hilbert polynomial of E. Thus in (35) we have ¯χ([OX (−n)] − α1 − · · · − αi−1, αi) = kiPα(n). Combining (36), and (35) with these substitu- tions, and cancelling a factor of (−1)mPα(n) on both sides, yields (−1)m(cid:18)Pα(n) m (cid:19) = Xl,k1,...,kl>1: k1+···+kl=m (−1)l l! l Yi=1 kiPα(n) ¯DT kiα(τ ). stp Regarding each side as a polynomial in Pα(n) and taking the linear term in Pα(n) we see that ¯DT mα(τ ) = for all m > 1. 1 m2 Example 4.14 shows that given a rigid τ -stable sheaf E in class α, the sheaves mE contribute 1/m2 to ¯DT mα(τ ) for all m > 1. We can regard this as a kind of 'multiple cover formula', analogous to the well known Aspinwall -- Morrison 23 computation for a Calabi -- Yau 3-fold X that a rigid embedded CP1 in class α ∈ H2(X; Z) contributes 1/m3 to the genus zero Gromov -- Witten invariant of X in class mα for all m > 1. So we can define new invariants DT α(τ ) which subtract out these contributions from mE for m > 1. Definition 4.15. Let X be a projective Calabi -- Yau 3-fold over C, let OX (1) be a very ample line bundle on X, and let (τ, T, 6) be a weak stability condition on coh(X) of Gieseker or µ-stability type. Then Definition 4.5 defines generalized Donaldson -- Thomas invariants ¯DT α(τ ) ∈ Q for α ∈ C(X). Let us define new invariants DT α(τ ) for α ∈ C(X) to satisfy ¯DT α(τ ) = Xm>1, mα 1 m2 DT α/m(τ ). By the Mobius inversion formula, the inverse of (37) is DT α(τ ) = Xm>1, mα Mo(m) m2 ¯DT α/m(τ ), (37) (38) where the Mobius function Mo : N → {−1, 0, 1} is Mo(n) = (−1)d if n = 1, 2, . . . is square-free and has d prime factors, and Mo(n) = 0 otherwise. We take (38) to be the definition of DT α(τ ), and then reversing the argument shows that (37) holds. We call DT α(τ ) the BPS invariants of X, as Kontsevich and Soibelman suggest their analogous invariants Ω(α) count BPS states. If Mα ss(τ ) = Mα st(τ ) then Mα/m DT α(τ ) = ¯DT α(τ ) = DT α(τ ), as in Definition 4.5. (τ ) = ∅ for all m > 2 dividing α, and so ss We make a conjecture [16, Conj. 6.12], based on [18, Conj. 6]. Conjecture 4.16. Let X be a Calabi -- Yau 3-fold over C, and (τ, T, 6) a weak stability condition on coh(X) of Gieseker or µ-stability type. Call (τ, T, 6) generic if for all α, β ∈ C(X) with τ (α) = τ (β) we have ¯χ(α, β) = 0. If (τ, T, 6) is generic, then DT α(τ ) ∈ Z for all α ∈ C(X). The author, Martijn Kool and Sven Meinhardt are working on a proof of Conjecture 4.16. In [16, §6] we prove that Conjecture 4.16 holds in a number of examples, and give an example [16, Ex. 6.8] in which (τ, T, 6) is not generic and DT α(τ ) /∈ Z. In [16, Th. 7.29] we prove the analogue of Conjecture 4.16 for invariants counting representations of quivers without relations. 4.5 Counting dimension 0 and 1 sheaves Let X be a Calabi -- Yau 3-fold over C with H 1(OX ) = 0, let OX (1) be a very ample line bundle on X, and (τ, G, 6) the associated Gieseker stability condition on coh(X). The Chern character gives an injective group homomorphism ch : K(X) → H even(X; Q). So we can regard K(X) as a subgroup of H even(X; Q), and write α ∈ K(X) as (α0, α2, α4, α6) with α2j ∈ H 2j(X; Q). If E → X is a vector bundle with [E] = α then α0 = rank E ∈ Z. 24 We will consider invariants ¯DT α(τ ), DT α(τ ) counting pure sheaves E of dimensions 0 and 1 on X, following [16, §6.3 -- §6.4]. For sheaves E of dimension zero ch E = (0, 0, 0, d) where d > 1 is the length of E. In [16, §6.3] we observe that for dimension 0 sheaves the moduli scheme M(0,0,0,d),n (τ ′) is independent of n, and is isomorphic to the Hilbert scheme Hilbd X. Therefore (34) gives stp P I (0,0,0,d),n(τ ′) = χ(cid:0)Hilbd X, νHilbd X(cid:1), for all n ∈ Z and d > 0. Values for χ(Hilbd X, νHilbd X ) were conjectured by Maulik et al. [22, Conj. 1], and proved by Behrend and Fantechi [3, Th. 4.12] and others. These yield a generating function for the P I (0,0,0,d),n(τ ′): 1 +Pd>1 P I (0,0,0,d),n(τ ′)sd =(cid:2)Qk>1(1−sk)−k(cid:3)χ(X) Computing using (35) then shows that . ¯DT (0,0,0,d)(τ ) = −χ(X) Xl>1, ld 1 l2 . So from (37) -- (38) we deduce that DT (0,0,0,d)(τ ) = −χ(X), all d > 1. This confirms Conjecture 4.16 for dimension 0 sheaves. It is one of several examples in [16] in which the values of the P I α,n(τ ′) are complex, the values of the ¯DT α(τ ) are simpler, and the values of the DT α(τ ) are simpler still, which suggests that of the three the invariants DT α(τ ) are the most fundamental. Now let β ∈ H 4(X; Z) and k ∈ Z. In [16, §6.4] we study invariants ¯DT (0,0,β,k)(τ ), DT (0,0,β,k)(τ ) counting semistable dimension 1 sheaves, that is, sheaves E supported on curves C in X. One expects these to be related to curve-counting invariants like Gromov -- Witten invariants, as in the MNOP Con- jecture [22, 23]. Here is a summary of our results: ¯DT (0,0,β,k)(τ ), DT (0,0,β,k)(τ ) are independent of the choice of (τ, T, 6). (a) (b) Assume Conjecture 4.16 holds. Then DT (0,0,β,k)(τ ) ∈ Z. (c) For any l ∈ β ∪ H 2(X; Z) ⊆ Z we have ¯DT (0,0,β,k)(τ ) = ¯DT (0,0,β,k+l)(τ ) and DT (0,0,β,k)(τ ) = DT (0,0,β,k+l)(τ ). (d) Let C be an embedded rational curve in X with normal bundle O(−1) ⊕ O(−1), and β ∈ H 4(X; Z) be Poincar´e dual to [C] ∈ H2(X; Z). Then sheaves supported on C contribute 1/m2 to ¯DT (0,0,mβ,k)(τ ) if m > 1 and m k, and contribute 0 to ¯DT (0,0,mβ,k)(τ ) if m > 1 and m ∤ k. They contribute 1 to DT (0,0,β,k)(τ ), and 0 to DT (0,0,mβ,k)(τ ) if m > 1. (e) Let C be a nonsingular embedded curve in X of genus g > 1, and let β ∈ H 4(X; Z) be Poincar´e dual to [C] ∈ H2(X; Z). Then sheaves supported on C contribute 0 to ¯DT (0,0,mβ,k)(τ ), DT (0,0,mβ,k)(τ ) for all m > 1 and k ∈ Z. 25 Motivated by these and by Katz [17, Conj. 2.3], we conjecture [16, Conj. 6.20]: Conjecture 4.17. Let X be a Calabi -- Yau 3-fold over C, and (τ, T, 6) a weak stability condition on coh(X) of Gieseker or µ-stability type. Then for γ ∈ H2(X; Z) with β ∈ H 4(X; Z) Poincar´e dual to γ and all k ∈ Z we have DT (0,0,β,k)(τ ) = GV0(γ). In particular, DT (0,0,β,k)(τ ) is independent of k, τ . Here GV0(γ) is the genus zero Gopakumar -- Vafa invariant, given in terms of the genus zero Gromov -- Witten invariants GW0(γ) ∈ Q of X by GW0(γ) =Xmγ 1 m3 GV0(γ/m). A priori we have GV0(γ) ∈ Q, but Gopakumar and Vafa [7] conjecture that the GV0(γ) are integers, and count something meaningful in String Theory. 5 Quivers with superpotentials We now summarize [16, §7], which develops an analogue of the results of §4 for representations of a quiver Q with relations I coming from a superpotential W . In the quiver case we have no analogue of ¯DT α(τ ), DT α(τ ), P I α,n(τ ) in §4 being deformation-invariant, since the proof of deformation-invariance uses the fact that the moduli scheme Mα,n stp (τ ′) is proper (i.e. compact), but the analogous moduli schemes Md,e stf Q,I (µ′) in the quiver case need not be proper. However, all the other important aspects of the sheaf case transfer to the quiver case. 5.1 Background on quivers Here are the basic definitions in quiver theory. Definition 5.1. A quiver Q is a finite directed graph. That is, Q is a quadruple (Q0, Q1, h, t), where Q0 is a finite set of vertices, Q1 is a finite set of edges, and h, t : Q1 → Q0 are maps giving the head and tail of each edge. The path algebra CQ is an associative algebra over C with basis all paths of length k > 0, that is, sequences of the form v0 e1−→ v1 → · · · → vk−1 ek−→ vk, where v0, . . . , vk ∈ Q0, e1, . . . , ek ∈ Q1, t(ai) = vi−1 and h(ai) = vi. Multiplica- tion is given by composition of paths in reverse order. For n > 0, write CQ(n) for the vector subspace of CQ with basis all paths of length k > n. It is an ideal in CQ. A quiver with relations (Q, I) is defined to be a quiver Q together with a two-sided ideal I in CQ with I ⊆ CQ(2). Then CQ/I is an associative C-algebra. For v ∈ Q0, write iv ∈ CQ for the path of length 0 at v. The image of iv in CQ/I is also written iv. Then i2 v = iv, iviw = 0 if v 6= w ∈ Q0, and Pv∈Q0 26 iv = 1 in CQ or CQ/I. (39) implies a decomposition of complex vector spaces E = Lv∈Q0 Write mod-CQ or mod-CQ/I for the abelian categories of finite-dimensional left CQ or CQ/I-modules, respectively. If E ∈ mod-CQ or mod-CQ/I then (39) iv(E). Define the dimension vector dimE ∈ ZQ0 If 0 → E → F → G → 0 is an exact sequence in mod-CQ or mod-CQ/I then dimF = dimE + dim G. Hence dim induces surjective morphisms dim : K0(mod-CQ) → ZQ0 and dim : K0(mod-CQ/I) → ZQ0 . >0 ⊂ ZQ0 by dim E : v 7→ dimC(ivE). Write K(mod-CQ) = K(mod-CQ/I) = ZQ0 , regarded as quotients of the Grothendieck groups K0(mod-CQ), K0(mod-CQ/I) induced by dim . Write C(mod-CQ) = C(mod-CQ/I) = ZQ0 >0 \{0}, the subsets of classes in K(mod-CQ), K(mod-CQ/I) of nonzero objects in mod-CQ, mod-CQ/I. Here K(mod-CQ), K(mod-CQ/I) are our substitutes for K(X) = K num(coh(X)) in §3 -- §4. We do not use the numerical Grothendieck groups K num(mod-CQ), K num(mod-CQ/I), as these may be zero in interesting cases. Definition 5.2. Let Q be a quiver. A superpotential W for Q over C is an element of CQ/[CQ, CQ]. The cycles in Q up to cyclic permutation form a basis for CQ/[CQ, CQ] over C, so we can think of W as a finite C-linear combination of cycles up to cyclic permutation. We call W minimal if all cycles in W have length at least 3. We will consider only minimal superpotentials W . Define I to be the two-sided ideal in CQ generated by ∂eW for all edges e ∈ Q1, where if C is a cycle in Q, we define ∂eC to be the sum over all occurrences of the edge e in C of the path obtained by cyclically permuting C until e is in first position, and then deleting it. Since W is minimal, I ⊆ CQ(2), and (Q, I) is a quiver with relations. We allow W ≡ 0, so that I = 0. Here is [16, Th. 7.6], which gives an analogue of equation (10) for quivers with superpotentials. Now (10) depended crucially on X being a Calabi -- Yau 3-fold, which implies that coh(X) has Serre duality in dimension 3. In general the categories mod-CQ/I coming from quivers with superpotentials do not have Serre duality in dimension 3. However, as explained in [16, §7.2], if (Q, I) comes from a quiver with superpotential then we can embed mod-CQ/I as the heart of a t-structure in a 3-Calabi -- Yau triangulated category T (which is usually not Dbmod-CQ/I), and Serre duality in dimension 3 holds in T . This is why quivers with superpotentials are algebraic analogues of Calabi -- Yau 3-folds, and have a version of Donaldson -- Thomas theory. Theorem 5.3. Let Q = (Q0, Q1, h, t) be a quiver with relations I coming from a minimal superpotential W on Q over C. Define ¯χ : ZQ0 × ZQ0 → Z by ¯χ(d, e) =Pe∈Q1(cid:0)d(h(e))e(t(e)) − d(t(e))e(h(e))(cid:1). Then for any D, E ∈ mod-CQ/I we have (40) ¯χ(cid:0)dim D, dimE(cid:1) =(cid:0)dim Hom(D, E) − dim Ext1(D, E)(cid:1)− (cid:0)dim Hom(E, D) − dim Ext1(E, D)(cid:1). 27 If Q is a quiver, the moduli stack MQ of objects E in mod-CQ is an Artin Q of E with dim E = d has a very C-stack. For d ∈ ZQ0 explicit description: as a quotient C-stack we have >0 , the open substack Md Md Q ∼=(cid:2)Qe∈Q1 Hom(Cd(t(e)), Cd(h(e)))/Qv∈Q0 GL(d(v))(cid:3). If (Q, I) is a quiver with relations, the moduli stack MQ,I of objects E in mod-CQ/I is a substack of MQ, and for d ∈ ZQ0 >0 we may write where V d (Cd(t(e)), Cd(h(e))) defined using the relations I. Q,I is a closed Qv∈Q0 When I comes from a superpotential W , we can improve the description GL(d(v))-invariant polynomial (41) of the moduli stacks Md Md Q,I (41) Hom GL(d(v))(cid:3), ∼=(cid:2)V d Q,I /Qv∈Q0 GL(d(v))-invariant C-subscheme of Qe∈Q1 Q,I . Define aQv∈Q0 Hom(cid:0)Cd(t(e)), Cd(h(e))(cid:1) −→ C W d :Qe∈Q1 as follows. Write W as a finite sum Pi γiCi,where γi ∈ C and Ci is a cycle ei ki−→ vi ei 1−→ vi 1 → · · · → vi 0 in Q. Set ki = vi ki −1 vi 0 W d(cid:0)Ae : e ∈ Q1(cid:1) =Pi γi Tr(cid:0)Aei ki Then V d Q,I = Crit(W d) in (41), so that ◦ Aei ki −1 ◦ · · · ◦ Aei 1(cid:1). Md Q,I ∼=(cid:2)Crit(W d)/Qv∈Q0 GL(d(v))(cid:3). (42) Equation (42) is an analogue of Theorem 4.1 for categories mod-CQ/I coming from a superpotential W on Q. We define a class of stability conditions on mod-CQ/I, [15, Ex. 4.14]. Example 5.4. Let (Q, I) be a quiver with relations. Let c : Q0 → R and r : Q0 → (0, ∞) be maps. Define µ : C(mod-CQ/I) → R by µ(d) = Pv∈Q0 Pv∈Q0 c(v)d(v) r(v)d(v) . Note that Pv∈Q0 r(v)d(v) > 0 as r(v) > 0 for all v ∈ Q0, and d(v) > 0 for all v with d(v) > 0 for some v. Then [15, Ex. 4.14] shows that (µ, R, 6) is a permissible stability condition on mod-CQ, which we call slope stability. Write Md ss(µ) for the open C-substack of µ-semistable objects in class d in Md A simple case is to take c ≡ 0 and r ≡ 1, so that µ ≡ 0. Then (0, R, 6) is a trivial stability condition on mod-CQ or mod-CQ/I, and every nonzero object in mod-CQ or mod-CQ/I is 0-semistable, so that Md Q,I . ss(0) = Md Q,I . 28 5.2 Behrend function identities, Lie algebra morphisms, and Donaldson -- Thomas type invariants Let Q be a quiver with relations I coming from a minimal superpotential W on Q over C. We now generalize §4 from coh(X) to mod-CQ/I. The proof of Theorem 4.2 depends on two things: the description of M in terms of Crit(f ) in Theorem 4.1, and equation (10). For mod-CQ/I equation (42) provides an analogue of Theorem 4.1, and Theorem 5.3 an analogue of (10). Thus, the proof of Theorem 4.2 also yields [16, Th. 7.11]: Theorem 5.5. In the situation above, with MQ,I the moduli stack of objects in a category mod-CQ/I coming from a quiver Q with superpotential W, and ¯χ defined in (40), the Behrend function νMQ,I of MQ,I satisfies the identities (27) -- (28) for all E1, E2 ∈ mod-CQ/I. Here is the analogue of Definition 4.3. Definition 5.6. Define a Lie algebra L(Q) to be the Q-vector space with basis of symbols λd for d ∈ ZQ0 , with Lie bracket [λd, λe] = (−1) ¯χ(d,e) ¯χ(d, e)λd+e, as for (29), with ¯χ given in (40). This makes L(Q) into an infinite-dimensional Lie algebra over Q. Define Q-linear maps Ψχ,Q al (MQ,I , χ, Q) → L(Q) and ΨQ,I : SFind al (MQ,I ) → L(Q) exactly as for Ψχ,Q, Ψ in Definition 4.3. Q,I : ¯SFind The proof of Theorem 4.4 has two ingredients: equation (10) and Theorem 4.2. Theorems 5.3 and 5.5 are analogues of these for quivers with superpoten- tials. So the proof of Theorem 4.4 also yields [16, Th. 7.14]: Theorem 5.7. ΨQ,I : SFind → L(Q) are Lie algebra morphisms. al (MQ,I ) → L(Q) and Ψχ,Q Q,I : ¯SFind al (MQ,I , χ, Q) Here is the analogue of Definitions 4.5 and 4.15. Definition 5.8. Let (µ, R, 6) be a slope stability condition on mod-CQ/I as in Example 5.4. As in §3.2 we have elements ¯δd ss(µ) ∈ SFal(MQ,I ) and ¯ǫd(µ) ∈ SFind al (MQ,I ) for all d ∈ C(mod-CQ/I). As in (31), define quiver generalized Donaldson -- Thomas invariants ¯DT d Q,I (µ) ∈ Q for all d ∈ C(mod-CQ/I) by As in (38), define quiver BPS invariants DT d Q,I (µ) ∈ Q by ΨQ,I(cid:0)¯ǫd(µ)(cid:1) = − ¯DT d Q,I (µ)λd. DT d Q,I (µ) = Xm>1, md Mo(m) m2 ¯DT d/m Q,I (µ), (43) where Mo : N → Q is the Mobius function. As for (37), the inverse of (43) is ¯DT d Q,I (µ) = Xm>1, md 1 m2 DT d/m Q,I (µ). (44) 29 If W ≡ 0, so that mod-CQ/I = mod-CQ, we write ¯DT d Q,I (µ), DT d Q(µ) for ¯DT d Q,I (µ). Note that µ ≡ 0 is allowed as a slope stability condition, with every object in mod-CQ/I 0-semistable, and is a natural choice. So we have invariants ¯DT d Q(µ), DT d Q,I (0) and ¯DT d Q(0), DT d Q,I (0), DT d Q(0). Here is the analogue of the integrality conjecture, Conjecture 4.16. Conjecture 5.9. Call (µ, R, 6) generic if for all d, e ∈ C(mod-CQ/I) with µ(d) = µ(e) we have ¯χ(d, e) = 0. If (µ, R, 6) is generic, then DT d Q,I (µ) ∈ Z for all d ∈ C(mod-CQ/I). In [16, Th. 7.29] we prove Conjecture 5.9 when W ≡ 0, using results of Q(µ) ∈ Z for all d. In [16, It holds for arbitrary µ, µ, Reineke [31]. That is, if µ is generic we show DT d Th. 7.17] we prove an analogue of Theorem 4.6. without requiring extra technical conditions as in Theorem 3.12. Theorem 5.10. Let (µ, R, 6) and (µ, R, 6) be any two slope stability conditions on mod-CQ/I, and ¯χ be as in (40). Then for all d ∈ C(mod-CQ/I) we have ¯DT d Q,I (µ) = Xiso. classes of finite sets I Xκ:I→C(mod-CQ/I): Pi∈I κ(i)=d Xconnected, simply- connected digraphs Γ, vertices I (−1)I−1V (I, Γ, κ; µ, µ) ·Yi∈I 2 Pi,j∈I ¯χ(κ(i),κ(j)) · Yedges · (−1) 1 ¯DT κ(i) Q,I (µ) ¯χ(κ(i), κ(j)), i • → j • in Γ with only finitely many nonzero terms. 5.3 Pair invariants for quivers We now discuss analogues for quivers of the moduli spaces of stable pairs Mα,n stp (τ ′) and stable pair invariants P I α,n(τ ′) in §4.3, and the identity (35) in Theorem 4.12 relating P I α,n(τ ′) and the ¯DT β(τ ). Definition 5.11. Let Q be a quiver with relations I coming from a super- potential W on Q over C. Suppose (µ, R, 6) is a slope stability condition on mod-CQ/I, as in Example 5.4. Let d, e ∈ ZQ0 >0 be dimension vectors. A framed representation (E, σ) of (Q, I) of type (d, e) consists of a representation E of mod-CQ/I with dimE = d, together with linear maps σv : Ce(v) → iv(E) for all v ∈ Q0. We call a framed representation (E, σ) stable if (i) µ([E′]) 6 µ([E]) for all subobjects 0 6= E′ ⊂ E in mod-CQ/I; and (ii) If also σ factors through E′, that is, σv(Ce(v)) ⊆ iv(E′) ⊆ iv(E) for all v ∈ Q0, then µ([E′]) < µ([E]). We will use µ′ to denote stability of framed representations, defined using µ. 30 Following Engel and Reineke [5, §3] or Szendroi [32, §1.2], we can in a stan- dard way define moduli problems for all framed representations, and for stable framed representations. The moduli space of all framed representations of type d,e (d, e) is an Artin C-stack M fr Q,I with an explicit description similar to (42), and the moduli space of stable framed representations of type (d, e) is a fine moduli C-scheme Md,e stf Q,I (µ′), an open C-substack of M d,e fr Q,I . We can now define our analogues of invariants P I α,n(τ ′) for quivers. Definition 5.12. In the situation above, define N DT d,e Q,I (µ′) = χ(cid:0)Md,e stf Q,I (µ′), νMd,e stf Q,I (µ′)(cid:1). (45) When W ≡ 0, so that mod-CQ/I = mod-CQ, we also write N DT d,e N DT d,e commutative Donaldson -- Thomas invariants. Q,I (µ′). Following Szendroi [32] we call N DT d,e Q,I (µ′), N DT d,e Q (µ′) = Q (µ′) non- Here (45) is the analogue of (34) in the sheaf case. We have no analogue of (33), since in general Md,e stf Q,I (µ′) is not proper, and so does not have a funda- mental class. These quiver analogues of Mα,n stp (τ ′), P I α,n(τ ′) are not new, sim- ilar things have been studied in quiver theory by Nakajima, Reineke, Szendroi and other authors for some years [5, 24 -- 27, 29, 30, 32]. Here [16, Th. 7.23] is the analogue of Theorem 4.12 for quivers. Theorem 5.13. Suppose Q is a quiver with relations I coming from a minimal superpotential W on Q over C. Let (µ, R, 6) be a slope stability condition on mod-CQ/I, as in Example 5.4, and ¯χ be as in (40). Then for all d, e in C(mod-CQ/I) = ZQ0 >0 \ {0} ⊂ ZQ0 , we have (−1)l l N DT d,e Q,I (µ′) = Xd1,...,dl∈C(mod-CQ/I), l>1: d1+···+dl=d, µ(di)=µ(d), all i l! Yi=1(cid:2)(−1)e·di− ¯χ(d1+···+di−1,di) (cid:0)e · di − ¯χ(d1 +· · ·+di−1, di)(cid:1) ¯DT di Q,I (µ)(cid:3), (46) 5.8, 5.12. When W ≡ 0, the same equation holds for N DT d,e Q (µ′), ¯DT d Q(µ). e(v)di(v), and ¯DT di Q,I (µ), N DT d,e Q,I (µ′) as in Definitions with e · di = Pv∈Q0 For Donaldson -- Thomas invariants in §4, we regarded the invariants ¯DT α(τ ), DT α(τ ) as our primary objects of study, and the pair invariants P I α,n(τ ′) as secondary, not of that much interest in themselves. In contrast, in the quiver lit- erature to date the invariants ¯DT d Q(µ) have not been seriously considered even in the stable = semistable case, and the ana- logues N DT d,e Q (µ′) of pair invariants P I α,n(τ ′) have been the cen- tral object of study. Q,I (µ′), N DT d,e Q(µ) and DT d Q,I (µ), ¯DT d Q,I (µ), DT d We argue that the invariants ¯DT d Q,I (µ), . . . , DT d garded as more fundamental and more interesting than the N DT d,e (46) the N DT d,e Q,I (µ′) can be written in terms of the ¯DT d Q(µ) should actually be re- Q,I (µ′). By Q,I (µ), and hence by 31 Q,I (µ) are simpler than the N DT d,e (44) in terms of the DT d Q,I (µ), so the pair invariants contain no more informa- tion. The ¯DT d Q,I (µ′) as they depend only on d rather than on d, e. In examples in [16, §7.5 -- §7.6] we find that the values of the ¯DT d Q,I (µ) may be much simpler and more illuminating than the values of the N DT d,e Q,I (µ′), as in (47) -- (49) below. Q,I (µ) and especially of the DT d Here is an example taken from [16, §7.5.2]. Example 5.14. Following Szendroi [32, §2.1], let Q = (Q0, Q1, h, t) have two vertices Q0 = {v0, v1} and edges e1, e2 : v0 → v1 and f1, f2 : v1 → v0, as below: • v0 e1 e2 f1 f2 . • v1 Define a superpotential W on Q by W = e1f1e2f2 − e1f2e2f1, and let I be the associated relations. Then mod-CQ/I is a 3-Calabi -- Yau abelian category. Theorem 5.3 shows that the Euler form ¯χ on mod-CQ/I is zero. Write elements d of C(mod-CQ/I) as (d0, d1) where dj = d(vj ). Szendroi [32, Th. 2.7.1] computed the noncommutative Donaldson -- Thomas invariants N DT (d0,d1),(1,0) (0′) for mod-CQ/I as combinatorial sums, and using work of Young [34] wrote their generating function as a product [32, Th. 2.7.2], giving Q,I Q,I N DT (d0,d1),(1,0) 1 + X(0,0)6=(d0,d1)∈N2 = Yk>1(cid:0)1 − (−q0q1)k)(cid:1)−2k(cid:0)1 − (−q0)kqk−1 (0′)qd0 1 0 qd1 1 (cid:1)k(cid:0)1 − (−q0)kqk+1 1 (cid:1)k . (47) Computing using (46) and (47) shows that ¯DT (d0,d1) Q,I (0) =  −2 Xl>1, ld 1 l2 , 1  l2 , 0, (0) =  DT (d0,d1) Q,I 1 l2 , d0 = d1 = d > 1, d0 = kl, d1 = (k − 1)l, k, l > 1, (48) d0 = kl, d1 = (k + 1)l, k > 0, l > 1, otherwise. (d0, d1) = (k, k), k > 1, (d0, d1) = (k, k − 1), k > 1, (d0, d1) = (k − 1, k), k > 1, −2, 1, 1, 0, otherwise. (49) Combining (44) and (48) we see that Note that the values of the DT (d0,d1) and are far simpler than those of the N DT (d0,d1),(1,0) Q,I Q,I (0′) in (47). (0) in (49) lie in Z, as in Conjecture 5.9, 32 * * * * j j j j This example is connected to Donaldson -- Thomas theory for (noncompact) Calabi -- Yau 3-folds as follows. We have equivalences of derived categories Db(mod-CQ/I) ∼ Db(cohcs(X)) ∼ Db(cohcs(X+)), (50) where π : X → Y and π+ : X+ → Y are the two crepant resolutions of the conifold Y = (cid:8)(z1, z2, z3, z4) ∈ C4 : z2 1 + · · · + z2 4 = 0(cid:9), and X, X+ are related One idea in [32] is that counting invariants N DT d,e by a flop. Here X, X+ are regarded as 'commutative' crepant resolutions of Y , and mod-CQ/I as a 'noncommutative' resolution of Y , in the sense that mod-CQ/I can be regarded as the coherent sheaves on the 'noncommutative scheme' Spec(CQ/I) constructed from the noncommutative C-algebra CQ/I. Q,I (µ′) for mod-CQ/I should be related to Donaldson -- Thomas type invariants counting sheaves on X, X+ by some kind of wall-crossing formula under change of stability condi- tion in the derived categories, using the equivalences (50). This picture has In [16, §7.5.2] we been worked out further by Nagao and Nakajima [25, 26]. show that in this case the situation for invariants ¯DT d Q,I (µ) is actu- ally much simpler, because they are unchanged by wall-crossing as ¯χ ≡ 0, so we can identify the invariants ¯DT d Q,I (µ), DT d Q,I (µ) in (48) -- (49) directly with Donaldson -- Thomas invariants for X and X+. Q,I (µ), DT d References [1] K. Behrend, Donaldson -- Thomas type invariants via microlocal geometry, to appear in Annals of Mathematics. math.AG/0507523, 2005. [2] K. Behrend and B. Fantechi, The intrinsic normal cone, Invent. Math. 128 (1997), 45 -- 88. [3] K. Behrend and B. Fantechi, Symmetric obstruction theories and Hilbert schemes of points on threefolds, Algebra and Number Theory 2 (2008), 313 -- 345. math.AG/0512556. [4] S.K. Donaldson and R.P. Thomas, Gauge Theory in Higher Dimensions, Chapter 3 in S.A. Huggett, L.J. Mason, K.P. Tod, S.T. Tsou and N.M.J. Woodhouse, editors, The Geometric Universe, Oxford University Press, Oxford, 1998. [5] J. Engel and M. Reineke, Smooth models of quiver moduli, Math. Z. 262 (2009), 817 -- 848. arXiv:0706.4306. [6] T.L. G´omez, Algebraic stacks, Proc. Indian Acad. Sci. Math. Sci. 111 (2001), 1 -- 31. math.AG/9911199. [7] R. Gopakumar and C. Vafa, M-theory and topological strings. II, hep-th/9812127, 1998. [8] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves, Aspects of Math. E31, Vieweg, Braunschweig/Wiesbaden, 1997. 33 [9] D. Joyce, Constructible functions on Artin stacks, J. London Math. Soc. 74 (2006), 583 -- 606. math.AG/0403305. [10] D. Joyce, Motivic invariants of Artin stacks and 'stack functions', Quart. J. Math. 58 (2007), 345 -- 392. math.AG/0509722. [11] D. Joyce, Configurations in abelian categories. I. Basic properties and mod- uli stacks, Adv. Math. 203 (2006), 194 -- 255. math.AG/0312190. [12] D. Joyce, Configurations in abelian categories. II. Ringel -- Hall algebras, Adv. Math. 210 (2007), 635 -- 706. math.AG/0503029. [13] D. Joyce, Configurations in abelian categories. III. Stability conditions and identities, Adv. Math. 215 (2007), 153 -- 219. math.AG/0410267. [14] D. Joyce, Configurations in abelian categories. IV. Invariants and changing stability conditions, Adv. Math. 217 (2008), 125 -- 204. math.AG/0410268. [15] D. Joyce, Holomorphic generating functions for invariants counting coher- ent sheaves on Calabi -- Yau 3-folds, Geometry and Topology 11 (2007), 667 -- 725. hep-th/0607039. [16] D. Joyce and Y. Song, A theory of generalized Donaldson -- Thomas invari- ants, arXiv:0810.5645v5, May 2010. 211 pages. [17] S. Katz, Genus zero Gopakumar -- Vafa invariants of contractible curves, J. Diff. Geom. 79 (2008), 185-195. math.AG/0601193. [18] M. Kontsevich and Y. Soibelman, Stability structures, motivic Donaldson -- Thomas invariants and cluster transformations, arXiv:0811.2435, 2008. [19] M. Kontsevich and Y. Soibelman, Motivic Donaldson -- Thomas invariants: summary of results, arXiv:0910.4315, 2009. [20] G. Laumon and L. Moret-Bailly, Champs alg´ebriques, Ergeb. der Math. und ihrer Grenzgebiete 39, Springer-Verlag, Berlin, 2000. [21] K. Miyajima, Kuranishi family of vector bundles and algebraic description of Einstein -- Hermitian connections, Publ. RIMS, Kyoto Univ. 25 (1989), 301 -- 320. [22] D. Maulik, N. Nekrasov, A. Okounkov, and R. Pandharipande, Gromov- Witten theory and Donaldson -- Thomas theory. I, Compos. Math. 142 (2006), 1263 -- 1285. math.AG/0312059. [23] D. Maulik, N. Nekrasov, A. Okounkov, and R. Pandharipande, Gromov- Witten theory and Donaldson -- Thomas theory. II, Compos. Math. 142 (2006), 1286 -- 1304. math.AG/0406092. [24] S. Mozgovoy and M. Reineke, On the noncommutative Donaldson-Thomas invariants arising from brane tilings, arXiv:0809.0117, 2008. 34 [25] K. Nagao, Derived categories of small toric Calabi -- Yau 3-folds and curve counting invariants, arXiv:0809.2994, 2008. [26] K. Nagao and H. Nakajima, Counting invariant of perverse coherent sheaves and its wall-crossing, arXiv:0809.2992, 2008. [27] H. Nakajima, Varieties associated with quivers, pages 139 -- 157 in R. Bautista et al., editors, Representation theory of algebras and related topics, C.M.S. Conf. Proc. 19, A.M.S., Providence, RI, 1996. [28] R. Pandharipande and R.P. Thomas, Curve counting via stable pairs in the derived category, arXiv:0707.2348, 2007. [29] M. Reineke, Cohomology of noncommutative Hilbert schemes, Algebr. Rep- resent. Theory 8 (2005), 541 -- 561. math.AG/0306185. [30] M. Reineke, Framed quiver moduli, cohomology, and quantum groups, J. Algebra 320 (2008), 94 -- 115. math.AG/0411101. [31] M. Reineke, Cohomology of quiver moduli, functional equations, and inte- grality of Donaldson -- Thomas type invariants, arXiv:0903.0261, 2009. [32] B. Szendroi, Non-commutative Donaldson -- Thomas theory and the conifold, Geom. Topol. 12 (2008), 1171 -- 1202. arXiv:0705.3419. [33] R.P. Thomas, A holomorphic Casson invariant for Calabi -- Yau 3-folds, and bundles on K3 fibrations, J. Diff. Geom. 54 (2000), 367 -- 438. math.AG/9806111. [34] B. Young, Computing a pyramid partition generating function with dimer shuffling, J. Combin. Theory Ser. A 116 (2009), 334 -- 350. arXiv:0709.3079. The Mathematical Institute, 24-29 St. Giles, Oxford, OX1 3LB, U.K. E-mail: [email protected] 35
1102.0032	1	1102	2011-01-31T23:08:03	The integrability of the 2-Toda lattice on a simple Lie algebra	[ "math.AG" ]	We define the 2-Toda lattice on every simple Lie algebra g, and we show its Liouville integrability. We show that this lattice is given by a pair of Hamiltonian vector fields, associated with a Poisson bracket which results from an R-matrix of the underlying Lie algebra. We construct a big family of constants of motion which we use to prove the Liouville integrability of the system. We achieve the proof of their integrability by using several results on simple Lie algebras, R-matrices, invariant functions and root systems.	math.AG	math	The integrability of the 2-Toda lattice on a simple Lie algebra Khaoula Ben Abdeljelil Abstract We define the 2-Toda lattice on every simple Lie algebra g, and we show its Liouville integrability. We show that this lattice is given by a pair of Hamiltonian vector fields, associated with a Poisson bracket which results from an R-matrix of the underlying Lie algebra. We construct a big family of constants of motion which we use to prove the Liouville integrability of the system. We achieve the proof of their integrability by using several results on simple Lie algebras, R-matrices, invariant functions and root systems. Contents 1 Introduction 2 R-matrices, Poisson structures and functions in involution on the square of a Lie algebra 2.1 Construction of R-matrices on g2 . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Linear Poisson structures on g∗ × g∗ . . . . . . . . . . . . . . . . . . . . . . 2.3 Casimirs and functions in involution on g∗ × g∗ . . . . . . . . . . . . . . . . 3 The Liouville integrability of the 2-Toda lattice 3.1 Definition of the 2-Toda lattice . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The 2-Toda lattice is a Hamiltonian system . . . . . . . . . . . . . . . . . . 3.3 The integrability of the 2-Toda lattice . . . . . . . . . . . . . . . . . . . . . 3.3.1 The restriction of F to T 2 is involutive for {· , ·}R . . . . . . . . . . 3.3.2 The restriction of F to T 2 is independent . . . . . . . . . . . . . . . 3.3.3 The exact number of functions . . . . . . . . . . . . . . . . . . . . . 3.4 The integrability for the quadratic Poisson R-bracket . . . . . . . . . . . . . 3.4.1 The Liouville integrability of the 2-Toda lattice on gln(C) for the linear Poisson R-bracket . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 The integrability of the 2-Toda lattice on gln(C) for the quadratic . . . . . . . . . . . . . . . . . . . . . . . . . . . . Poisson R-bracket 1 2 4 5 7 7 12 12 12 15 16 17 20 22 22 23 3.5 The relation between the 2-Toda lattice and the Toda lattice . . . . . . . . 3.5.1 The Toda lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.2 Restriction of the 2-Toda lattice and construction of the Toda lattice 27 27 28 1 Introduction The 2-Toda lattice associated with sln(C) is the pair of differential equations given by the Lax equations ∂(L, M ) ∂t ∂(L, M ) ∂s = [(Lu, Lu), (L, M )], = [(Mℓ, Mℓ), (L, M )], (1)   where (L, M ) are traceless matrices of the form (L, M ) =     a11 1 0 . . . . . . a21 a22 ... an1 1 · · · an,n−1 ann ,     b11 b21 0 · · · . . . . . . · · · . . . bn,n−1 b1n ... ... bnn     , (2) and where Lu is the upper triangular part of L and Mℓ is the strictly lower triangular part of M . 2 Trace L2 and 1 The 2-Toda lattice was first introduced in the context of infinite-dimensional matrices [1], [3], [13]. This system is Hamiltonian with respect to a Poisson structure, associated with an R-matrix with 1 2 Trace M 2 as Hamiltonians. These Hamiltonians admit the Ad-invariant functions on gl((∞)) × gl((∞)) as Poisson commuting constants of motion. In the finite-dimensional setting, i.e., on sln(C), this system is still Hamiltonian with respect to a Poisson structure associated with an R-matrix similar to that of the infinite-dimensional setting. The family of Ad-invariant functions are again in involution with respect to the Poisson structure associated with the R-matrix but their number of independent Ad-invariant functions, which is 2n − 2, is much too small compared with the dimension of the phase space of the 2-Toda lattice, which is n2 + 2n − 3. The main purpose of the present article is to introduce a family of functions which contains the Ad-invariant functions and is large enough to give the Liouville integrability of the 2-Toda lattice. This family is F := (Fj,i, 1 6 i 6 n − 1 and 0 6 j 6 i + 1), where Fj,i is the function on sln(C) constructed from the the relation Trace(λL − M )i+1 = i+1 (−1)mi+1−jλjFj,i(L, M ). Xj=0 2 The Toda lattice can be defined for every simple Lie algebra [7]. In the same spirit we intro- duce the 2-Toda system for every simple Lie algebra and we show its Liouville integrability. We recall that, according to [2, Definition 4.13], in the general case a system (M, {· , ·} , F) is Liouville integrable if (M, {· , ·}) is a Poisson manifold of rank 2r and F = (F1, . . . , Fs) is involutive and independent, with s = dim M − r. The main result of this paper is given in Section 3: we show the Liouville integrability of the 2-Toda lattice, not only on sln(C), but also on an arbitrary simple Lie algebra g. We define this lattice in this general context. We choose a simple Lie algebra g; then we denote by h a Cartan subalgebra, by α1, . . . , αℓ the simple roots of g with respect to h, by e1, . . . , eℓ the corresponding eigenvectors, and by gi the subspace spanned by the eigenvectors associated with roots of length i ∈ Z. The 2-Toda lattice, associated with g, is the pair of differential equations on T 2 given by the following Lax pair equations ∂(L, M ) ∂t ∂(L, M ) ∂s = [(L+, L+), (L, M )], = [(M−, M−), (L, M )], (3)   where (L, M ) belongs to the phase space T 2 := Pi60 gi × Pi>−1 gi + (Pℓ i=1 ei, 0), and here, L+ (resp. M−) stands for the projection of L ∈ g (resp. M ∈ g) on Pi>0 gi (resp. Pi<0 gi). We construct a linear Poisson structure with respect to which the 2-Toda lattice is Hamiltonian. This linear Poisson structure comes from an R-matrix of the type studied in Section 2, namely from the endomorphism of g2 given by R(x, y) = (R(x − y) + y, R(x − y) + x), for all x, y ∈ g, where R := P+ − P− is the classical R-matrix on a simple Lie algebra, associated with the Lie algebra splitting g = Pi>0 gi ⊕Pi<0 gi. Using the Killing form of g, the product Lie algebra g2 is equipped with a bilinear, symmetric, Ad-invariant, non-degenerate form, which allows us to identify g2 with its dual and hence to equip g2 with a linear Poisson structure that we denote by {· , ·}R. We then establish that the phase space T 2 of the 2-Toda lattice is a Poisson submanifold of (g2, {· , ·}R) (see Propositions 8) and the Hamiltonian vector fields of functions H(x, y) := 1 2 hy yi are tangent to T 2 and describe on T 2 the equations of motion (3) for the 2-Toda lattice (see Proposition 10). 2 hx xi and H(x, y) := 1 We construct the integrable system. Let (P1, . . . , Pℓ) be a generating family of the algebra of Ad-invariant functions, chosen to be of respective degrees m1 + 1, . . . , mℓ + 1. For all 1 6 i 6 ℓ and 0 6 j 6 mi + 1, we define Fj,i ∈ F(g2) to be the coefficient in λj of the polynomial (x, y) 7→ Pi(λx − y), and we consider the following family of functions on g2: F = (Fj,i, 1 6 i 6 ℓ and 0 6 j 6 mi + 1). 3 Our main result is the following theorem: Theorem The triplet (T 2, FT 2, {· , ·}R) is an integrable system. To prove this result, we proceed as follows: • We show in Proposition 14 that F is involutive for {· , ·}R: this result is a particular case of the second point of Theorem 4 that we prove in Section 2. • We show in Proposition 17 the independence of F restricted to T 2 by using a theorem of Raïs [10]. • We compute the rank of the restriction of the Poisson R-bracket to T 2. This is done by establishing a Poisson isomorphism between (T 2, {· , ·}R) and the product Poisson manifold (g∗, {· , ·}) × (g0 ⊕ g1, {· , ·}R), where {· , ·} is the Lie-Poisson bracket on g∗, and {· , ·}R is the above R-bracket (restricted to g0 ⊕ g1; see Proposition 18). • We check that card FT 2 = dim T 2 − 1 2 Rk(T 2, {· , ·}R). We also define the 2-Toda from on gln(C) and we show its Liouville integrability with respect to the linear and quadratic Poisson R-bracket. This result is mainly based on the fact that the Hamiltonian vector fields of the linear and quadratic Poisson structure are essentially identical. In the last subsection, by restricting the 2-Toda lattice on g to a well chosen affine subspace, we find again the usual Toda lattice. 2 R-matrices, Poisson structures and functions in involution on the square of a Lie algebra In this section we fix (g, [· , ·]) a Lie algebra over F, with F = R or F = C. The vector space g2 := g × g, endowed with the Lie bracket [(x, y), (z, s)] := ([x, z], [y, s]), ∀x, y, z, s ∈ g, (4) is a Lie algebra. We construct an R-matrix of g2 with the help of an endomorphism R of g satisfying some conditions. With this R-matrix, when g is finite dimensional, we construct a linear Poisson bracket on the dual g2∗ of g2, we explain the construction of a large family of functions on g2∗ which commute for this linear Poisson structure and we spell out the expressions of their Hamiltonian vector fields. We also give some Casimir functions. 4 2.1 Construction of R-matrices on g2 We begin by recalling some properties and definitions of R-matrices (see [2, Section 4.4]). Let g be a Lie algebra. A (vector space) endomorphism R of g is called an R-matrix of g if the bilinear map [· , ·]R : g × g → g, defined for all x, y ∈ g by [x, y]R := 1 2 ([Rx, y] + [x, Ry]), (5) defines a (second) Lie bracket on g, which is then called a Lie R-bracket. Let BR : g× g → g be defined, for all x, y ∈ g, by BR(x, y) := [Rx, Ry] − R([Rx, y] + [x, Ry]). The bracket [· , ·]R satisfies the Jacobi identity if and only if [BR(x, y), z] + [BR(y, z), x] + [BR(z, x), y] = 0, ∀x, y, z ∈ g. (6) (7) A sufficient condition for (7) to be satisfied is that there exists a constant c ∈ F such that BR(x, y) = −c2[x, y], ∀x, y ∈ g. (8) We call (8) the modified classical Yang-Baxter equation (mCYBE) of g of constant c. We construct in the following proposition an R-matrix of g2 by using an endomorphism of g which satisfies some condition which generalizes the modified classical Yang-Baxter equation of g2. Proposition 1 Let g be a Lie algebra, z(g) its center and c ∈ F a constant. Let R be an endomorphism of g and R the endomorphism of g2 defined, for every (x, y) ∈ g2, by R(x, y) := (R(x − y) + cy, R(x − y) + cx) . The endomorphism R is an R-matrix of g2 if and only if, for every x, y ∈ g, BR(x, y) + c2[x, y] ∈ z(g). (9) (10) In the particular case where g is a complex semi-simple Lie algebra, this is in turn equivalent to R being a solution of (mCYBE) of constant c. Proof. Using Definition (6) of BR and replacing R by its Expression (9), we obtain BR((x, y), (z, s)) = (BR(x − y, z − s) + c2([x − y, z − s] − [x, z]), BR(x − y, z − s) + c2([x − y, z − s] − [y, s])). (11) According to Formula (7), R is an R-matrix of g2 if and only if, for every (x, y), (x′, y′) and (x′′, y′′) in g2, [BR((x, y), (x′, y′)), (x′′, y′′)]+ (cid:9)= (0, 0), (12) 5 where (cid:9)= cycl((x, y), (x′, y′), (x′′, y′′)). By Formula (11), the left hand side of Equation (12) is equal to (cid:0)[BR(x − y, x′ − y′) + c2[x − y, x′ − y′], x′′], [BR(x − y, x′ − y′) + c2[x − y, x′ − y′], y′′](cid:1) + (cid:9), (13) where we used the Jacobi identities [[x, x′], x′′]+ (cid:9)= 0 and [[y, y′], y′′]+ (cid:9)= 0. Now, if R is an R-matrix of g2, then (13) is equal to zero (even without circular permutation), so Formula (10) holds. If Formula (10) holds then the first component of the left hand side of (13) is equal to zero: [BR(x − y, x′ − y′) + c2[x − y, x′ − y′], x′′] + cycl((x, y), (x′, y′), (x′′, y′′)) = 0, (14) for every x, y, x′, y′, x′′, y′′ ∈ g; in particular, for x′ = x′′ = 0, we obtain [BR(y′, y′′) + c2[y′, y′′], x] = 0, ∀y′, y′′, x ∈ g, hence the endomorphism R is an R-matrix of g2 holds. (cid:3) For future use, we also state the following corollary, which is a consequence of Formula (11) and already appeared in [11] (with c = 1). Corollary 2 Let g be a Lie algebra and let R be an endomorphism of g. If R is a solution of (mCYBE) of g of constant c, then the endomorphism R of g2, defined, for all (x, y) ∈ g2, by R(x, y) := (R(x − y) + cy, R(x − y) + cx), is a solution of (mCYBE) of g2 of constant c. Example 3 Let g := g+ ⊕ g− be a Lie algebra splitting (i.e., g is, as a vector space, the direct sum of Lie subalgebras g+ and g−). Let P± be the projections of g on g± and let R := P+ − P−. Since R is a solution of (mCYBE) of g of c = 1, Corollary 2 implies that the endomorphism R, defined for every x, y ∈ g by R(x, y) = (R(x − y) + y, R(x − y) + x) is a solution of (mCYBE) of g2 of c = 1. Replacing R by its expression P+ − P−, we verify that R(x, y) = (x+ − x− + 2y−, y− − y+ + 2x+) = (x+ + y−, x+ + y−) − (x− − y−, y+ − x+), (15) where x = x+ + x− and y = y+ + y− are the decompositions of x and y with respect to the splitting g = g+ ⊕ g−. Since the Lie subalgebras g2 + := {(x, x) x ∈ g} and g2 − := {(x, y) x ∈ g− and y ∈ g+} + ⊕ g2 have a trivial intersection, (15) implies that g2 = g2 + and g2 R is the difference of the projections on g2 −. − is a Lie algebra splitting and that 6 Note that independently of Proposition 1 or Corollary 2 the above example shows that from an R-matrix of g, which is the difference of the projections on two supplementary Lie subalgebras of g, the endomorphism R given by the Formula (9) is automatically an R-matrix of g2, which is the difference of projections on two supplementary Lie subalgebras of g2. 2.2 Linear Poisson structures on g∗ × g∗ From now and until the end of the article, we assume that g is a finite-dimensional Lie algebra and we denote by F(g) the algebra of smooth functions on g (if g is a Lie algebra over R) or holomorphic functions on g (if g is a Lie algebra over C). Let R be an R-matrix of g2. The dual (g2)∗ of the Lie algebra (g2, [· , ·]R) admits a Lie-Poisson structure, defined for every F, G ∈ F((g2)∗) at ϕ ∈ (g2)∗ by {F, G}R(ϕ) : = (cid:10)ϕ, [dϕF, dϕG]R(cid:11) 1 2 = hϕ, [RdϕF, dϕG] + [dϕF, RdϕG]i , (16) where dϕF ∈ ((g2)∗)∗ with g∗ × g∗ as follows: ⋍ g2 is the differential of F at the point ϕ ∈ (g2)∗. We identify (g2)∗ Φ : g∗ × g∗ → (g2)∗ (ξ, η) 7→ ((x, y) 7→ h(ξ, η), (x, y)i) := hξ, xi − hη, yi . By means of this identification, the Poisson structure (16) can be transported to the vector space g∗×g∗. This transported Poisson structure is given, for all functions F, G ∈ F(g∗ ×g∗) at (ξ, η) ∈ g∗ × g∗, by {F, G}R(ξ, η) := D(ξ, η), [d(ξ,η)F, d(ξ,η)G]RE , (17) (ξ,η)F ∈ (g∗ × g∗)∗ ≃ g × g . We call this Poisson structure the Poisson R-bracket where d on g∗ × g∗. 2.3 Casimirs and functions in involution on g∗ × g∗ Let g be a finite-dimensional Lie algebra. We denote by G a connected Lie group whose Lie algebra is g. We denote by F(g∗)G the algebra of Ad∗-invariant functions on g∗. According to the Adler-Kostant-Symes theorem [2, Theorem 4.37], the Ad∗-invariant functions on g∗ × g∗ (i.e., functions in F(g∗ × g∗)G×G) are in involution for the Poisson structure associated with an R-matrix. But, when g is a semi-simple Lie algebra, this family of functions is too small to insure the Liouville integrability. We construct a larger family1 of functions in involution. 1Which contains the Ad∗-invariant functions on g∗ × g∗. 7 Theorem 4 Let λ ∈ F be a constant and let ψλ be the map ψλ : g∗ × g∗ → g∗ (ξ, η) 7→ λξ − η. (18) Let R be an endomorphism2 of g and c ∈ F a constant. Assume that the endomorphism R of g2 defined for every (x, y) ∈ g2 by R(x, y) := (R(x − y) + cy, R(x − y) + cx) (19) is an R-matrix of g2, and denote by {· , ·}R the Poisson R-bracket on g∗ × g∗. Then: (1) For every F ∈ F(g∗)G, the function F ◦ ψ1 is a Casimir for {· , ·}R. (2) For every F, G ∈ F(g∗)G and every λ, γ ∈ F, the functions F ◦ ψλ and G ◦ ψγ are in involution for {· , ·}R. In particular, if F and G are polynomials of degree respectively l and k, then the functions F0, . . . , Fl, G0, . . . , Gk, defined, for every (ξ, η) ∈ g∗ × g∗, by F (ψλ(ξ, η)) = Pl i=0 λiFi(ξ, η) are in involution for {· , ·}R. and G(ψγ(ξ, η)) = Pk j=0 γjGj(ξ, η), (3) If c = 1, the map ψ1 : (g∗ × g∗, {· , ·}R) → (g∗, {· , ·}) is a Poisson morphism. (4) For every H ∈ F(g∗)G, the Hamiltonian vector field XH◦ψλ := {· , H ◦ ψλ}R is given at (ξ, µ) ∈ g∗ × g∗ by XH◦ψλ (ξ, η) = 1 2 (1 − λ) ad∗ ((R−cI)dλξ−η H,(R+cI)dλξ−η H)(ξ, η). (20) In particular, if g := g+ ⊕ g− is a Lie algebra splitting and R := P+ − P− is the difference of the projections on g+ and g−, we have XH◦φλ(ξ, η) = (1 − λ) ad∗ ((−dλξ−η H)−,(dλξ−η H)+)(ξ, η). (21) To prove this theorem, we need the following lemmas, the proof of the first of which is left to the reader. Lemma 5 Let F ∈ F(g∗), (ξ, η) ∈ g∗ × g∗ and λ ∈ F be a constant. (1) The differential of F ◦ ψλ at (ξ, η) ∈ g∗ × g∗ is given by d (ξ,η)(F ◦ ψλ) = (λdλξ−ηF, dλξ−ηF ). (22) 2R is not necessarily an R-matrix. 8 (2) Let R be an endomorphism of g and let R be the endomorphism of g2 defined in (19). Then R d (ξ,η)(F ◦ ψλ) = (λ − 1)(R dλξ−ηF, R dλξ−ηF ) + c(dλξ−ηF, λdλξ−ηF ). (23) Lemma 6 Let F, G ∈ F(g∗)G. (1) The differentials of F and G commute at every point of g∗, i.e., [dξF, dξG] = 0, ∀ξ ∈ g∗. (2) For every ξ, η ∈ g∗ and every a, b ∈ F, haξ + bη, [dξF, dηG]i = 0. Proof. Using the Ad∗-invariance of F and G, we obtain ad∗ dξF ξ = 0 and ad∗ dξG ξ = 0. (24) (25) (26) Stated differently, dξF ∈ gξ and dξG ∈ gξ where gξ is the centralizer of ξ. For ξ regular, according to [12, Proposition 19.7.5], the centralizer gξ is abelian, so Formula (24) holds at least for every regular point ξ of g∗. Since the set of regular elements of g∗ is dense in g∗ (see [12, Proposition 19.7.5]), the continuous map ξ → [dξF, dξG] is zero at all points. This establishes (24). The second point of the lemma follows directly from (26). (cid:3) Let us now prove Theorem 4. Proof. (1) Let K ∈ F(g∗ × g∗) and let (ξ, η) ∈ g∗ × g∗. The Poisson R-bracket between ψ∗ 1F = F ◦ ψ1 and K at (ξ, η) is given by {ψ∗ 1F, K}R(ξ, η) = 1 2 (cid:10)(ξ, η), [R d(ξ,η)(F ◦ ψ1), d(ξ,η)K](cid:11) 2 (cid:10)(ξ, η), [d (ξ,η)(F ◦ ψ1), R d (ξ,η)K](cid:11) . 1 + (ξ,η)(F ◦ψ1) = (dξ−ηF, dξ−ηF ) and R d (ξ,η)(F ◦ψ1) = According to (22) and (23), we obtain d c(dξ−ηF, dξ−ηF ), so {ψ∗ 1F, K}R(ξ, η) = 1 Using Formula (19), we have that R d the explicit expression of which is not needed in this proof. Then (27) becomes (ξ,η)K is of the form (x, x) for some x ∈ g, (ξ,η)K + cd 2 (cid:10)(ξ, η), [(dξ−ηF, dξ−ηF ), R d (ξ,η)K + cd (ξ,η)K](cid:11) . (27) {ψ∗ 1F, K}R(ξ, η) = 1 2 1 2 = 0, = h(ξ, η), [(dξ−ηF, dξ−ηF ), (x, x)]i hξ − η, [dξ−ηF, x]i 9 where we used that F is an Ad∗-invariant function on g∗ in the last line. This shows that ψ∗ 1F is a Casimir on ((g2)∗, {· , ·}R). (2) Let us prove that the Poisson R-bracket between F ◦ ψλ and G ◦ ψγ at an arbitrary point (ξ, η) ∈ g∗ × g∗ is equal to zero. According to (17), λF, ψ∗ (cid:8)ψ∗ γG(cid:9)R (ξ, η) = 1 2 (cid:10)(ξ, η), [R d 2 (cid:10)(ξ, η), [d + 1 (ξ,η)(F ◦ ψλ), d (ξ,η)(F ◦ ψλ), R d (ξ,η)(G ◦ ψγ)](cid:11) (ξ,η)(G ◦ ψγ)](cid:11) . γG(cid:9)R λF, ψ∗ (ξ, η) of (28) becomes (28) Using Formulae (22) and (23), the Poisson R-bracket (cid:8)ψ∗ λF, ψ∗ (cid:8)ψ∗ γG(cid:9)R (ξ, η) = 1 2 h(ξ, η), [(λ − 1)(R dλξ−ηF, R dλξ−ηF ), (γdγξ−ηG, dγξ−ηG)]i 1 2 2 cγ 2 + = = + h(ξ, η), [(cdλξ−ηF, cλdλξ−ηF ), (γdγξ−ηG, dγξ−ηG)]i −((F, λ) ↔ (G, γ)) (λ − 1) hγξ − η, [R dλξ−ηF, dγξ+ηG]i hξ, [dλξ−ηF, dγξ−ηG]i − cλ 2 hη, [dλξ−ηF, dγξ−ηG]i −((F, λ) ↔ (G, γ)) cγ 2 −((F, λ) ↔ (G, γ)), hξ, [dλξ−ηF, dγξ−ηG]i − cλ 2 hη, [dλξ−ηF, dγξ−ηG]i where we have used the Ad∗-invariance of F and G on g∗ to simplify the expression. Ac- cording to Formula (25), all four terms of the previous expression vanish if λ 6= γ. If λ = γ, the bracket (cid:8)ψ∗ λF, ψ∗ γG(cid:9)R becomes λF, ψ∗ {ψ∗ λG}R(ξ, η) = cλ hξ − η, [dλξ−ηF, dλξ−ηG]i , which is zero in view of Formula (24). We now suppose that F and G are polynomials of degrees l and k, so for all (ξ, η) ∈ g × × g∗, F ◦ ψλ(ξ, η) = Pl i=0 λiFi(ξ, η) and G ◦ ψγ(ξ, η) = Pk j=0 γjGj(ξ, η); hence l k {F ◦ ψλ, G ◦ ψγ}R = Xi=0 Xj=0 λiγj{Fi, Gj }R. (29) As we just showed, the term on the left hand side of Equation (29) is zero, for all λ, γ ∈ F. Therefore, all the coefficients of Pl i=0Pk j=0 λiγj{Fi, Gj }R are zero. 10 (3) We assume that c = 1. Let F, G ∈ F(g∗) and let (ξ, η) be a point in g∗ × g∗. From (28), (23) and (22), it follows that the Poisson R-bracket between F ◦ ψ1 and G ◦ ψ1 at (ξ, η) is given by {F ◦ ψ1, G ◦ ψ1}R(ξ, η) = hξ − η, [dξ−ηF, dξ−ηG]i = {F, G} (ψ1(ξ, η)). Hence ψ∗ 1 is a Poisson map. (4) Let K be a function on g∗ × g∗ and (ξ, η) be an element of g∗ × g∗. A direct computation gives XH◦ψλ(ξ, η)[K] = {K, H ◦ ψλ}R(ξ, η) = 1 2 (cid:10)(ξ, η), [R d (ξ,η)K, d (ξ,η)(H ◦ ψλ)] + [d (ξ,η)K, R d (ξ,η)(H ◦ ψλ)](cid:11) . (30) In order to rewrite this formula, we temporarily use (x, y) := d(ξ,η)K. According to (19), R d(ξ,η)K = R(x, y) is given by R d (ξ,η)K = (R(x − y), R(x − y)) + c(y, x). (31) Formulae (31) and (22) allow one to rewrite the first term of (30) in the following manner: (cid:10)(ξ, η), [Rd(ξ,η)K, d(ξ,η)(H ◦ ψλ)](cid:11) = h(ξ, η), [(R(x − y), R(x − y)), (λdλξ−ηH, dλξ−ηH)]i +c h(ξ, η), [(y, x), (λdλξ−η H, dλξ−ηH)]i = hλξ − η, [R(x − y), dλξ−ηH]i + c hλξ, [y, dλξ−ηH]i − c hη, [x, dλξ−ηH]i = c hη, [y, dλξ−ηH]i − c hλξ, [x, dλξ−ηH]i , where, in the second line, we used, three times, the Ad∗-invariance of H. Thus (cid:10)(ξ, η), [R d(ξ,η)K, d(ξ,η)(H ◦ ψλ)](cid:11) = h(ξ, η), [(cλdλξ−η H, cdλξ−ηH), (x, y)]i . By replacing the first term of (30) by its expression given in (33) and by using Formula (23) to express the second term of (30), we rewrite XH◦ψλ(ξ, η)[F ] as follows: (32) (33) XH◦ψλ(ξ, η)[K] = 1 2 − − h(ξ, η), [(cλdλξ−ηH, cdλξ−ηH), (x, y)]i (λ − 1) 2 h(ξ, η), [(R dλξ−ηH, R dλξ−ηH), (x, y)]i 1 2 h(ξ, η), [(cdλξ−η H, cλdλξ−ηH), (x, y)]i = (λ − 1) 2 (cid:10)(ξ, η), [((cI − R)dλξ−ηH, −(R + cI)dλξ−ηH), d(ξ,η)K](cid:11) . We then deduce Formula (20). When g is a Lie algebra splitting and R = P+ − P−, we have c = 1, R − cI = −2P− and R + cI = 2P+. This implies that (20) gives (21). (cid:3) 11 3 The Liouville integrability of the 2-Toda lattice In this section we define the 2-Toda lattice for every complex simple Lie algebra and we prove its Liouville integrability. In order to define the 2-Toda lattice for every simple Lie algebra, we need some notation. Let g be a simple Lie algebra of rank ℓ, with Killing form h· · i. We choose h, a Cartan subalgebra with roots system Φ, and Π = (α1, . . . , αℓ), a system of simple roots with respect to h. For every α in Φ\{−Π, Π}, we denote by eα a non-zero eigenvector associated with eigenvalue α, and, for every 1 6 i 6 ℓ, we denote by ei and e−i non-zero eigenvectors associated respectively with αi and −αi. The Lie algebra g = Pk∈Z gk is endowed with the natural grading (i.e., for every k, l ∈ Z, [gk, gl] ⊂ gk+l) defined by g0 := h and, for every k ∈ Z, gk := heα α ∈ Φ, α = ki, for α the length of the root α, i.e., α is Pℓ i=1 ai for α = Pℓ i=1 aiαi. In the sequel, we shall use the following property: hgk gli = 0 if k + l 6= 0. We introduce the following notation g<k := Pi<k gi, g>k := Pi>k gi, Also, g+ := g>0 and g− := g<0. g6k := Pi6k gi, g>k := Pi>k gi. 3.1 Definition of the 2-Toda lattice The next definition gives again the definition given in (1) when specialized to the case g = sln(C), taking for h the Lie subalgebra of diagonal matrices. Definition 7 The 2-Toda lattice associated with a simple Lie algebra g is the system of differential equations given by the following Lax equations: ∂(L, M ) ∂t ∂(L, M ) ∂s = [(L+, L+), (L, M )], = [(M−, M−), (L, M )], (34) (35) where (L, M ) is an element of the phase space of the 2-Toda lattice T 2 := g60 × g>−1 + i=1 ei, 0), where L+ := P+(L), M− := P−(M ), and where P± is the projection of g on (Pℓ g±. 3.2 The 2-Toda lattice is a Hamiltonian system We recall from Example 3 that when g = g+ ⊕g− is a Lie algebra splitting then g2 = g2 is also a Lie algebra splitting, where + ⊕g2 − g2 + := {(x, x) x ∈ g} et g2 − := {(x, y) x ∈ g− et y ∈ g+}. (36) 12 Also, for every (x, y) ∈ g2 we have (x, y) = (x, y)+ + (x, y)−, where (cid:26) (x, y)+ = (x+ + y−, x+ + y−) ∈ g2 +, (x, y)− = (x− − y−, y+ − x+) ∈ g2 −. (37) Let R be the difference of the projections of g2 on the Lie subalgebras g2 to Example 3, we have + and g2 −. According R(x, y) = (x+ − x− + 2y−, y− − y+ + 2x+) = (R(x − y) + y, R(x − y) + x), (38) (39) where R is the difference of the projections of g on g+ and on g−. We provide g2 with the following Ad-invariant, non-degenerate symmetric bilinear form: h· · i2 : g2 × g2 ((x1, y1), (x2, y2)) → 7→ hx1 x2i − hy1 y2i . C (40) We use it to identify g2 with its dual and we obtain according to (16) a linear Poisson structure on g2, defined for every F, G ∈ F(g2) at (x, y) ∈ g2 by {F, G}R(x, y) = 1 2(cid:10)(x, y) [R∇(x,y)F, ∇(x,y)G] + [∇(x,y)F, R∇(x,y)G](cid:11)2, (41) where ∇(x,y)F is the gradient of F at (x, y) ∈ g2 (with respect to h· · i2), i.e., (cid:10)∇(x,y)F (z, s)(cid:11)2 = (cid:10)d(x,y)F, (z, s)(cid:11) , ∀(z, s) ∈ g2. We show that the phase space T 2 is equipped with a Poisson structure and the equations of motion of the 2-Toda lattice are Hamiltonian. Proposition 8 T 2 is a Poisson submanifold of (g2, {· , ·}R). We use the following lemma to show the above proposition. Lemma 9 Let g be a Lie algebra equipped with a non-degenerate, symmetric, bilinear form h· · i. Let a ∈ g and let E be a subspace of g. We suppose that: (1) The orthogonal3 E⊥ of E is a Lie ideal of (g, [· , ·]). (2) For every x, y ∈ g, we have ha [x, y]i = 0. Then a + E and E are Poisson submanifolds of (g, {· , ·}), equipped with its linear Poisson structure {· , ·}. The map x → a + x is a Poisson isomorphism between them. Moreover E is Poisson isomorphic to the Lie-Poisson manifold (g/E⊥)∗, also equipped with its linear Poisson structure. 3Here the orthogonality is with respect to the form h· · i. 13 Proof. We start to show that E is a Poisson submanifold of (g, {· , ·}). Let J := hF ∈ F(g) F ≡ 0 on Ei. To show that E is a Poisson submanifold is equivalent to show that J is Poisson ideal. Let F ∈ J and let x ∈ E, we notice that ∇xF ∈ E⊥. According to the first condition of the proposition, for every G ∈ F(g), we have [∇xf, ∇xG] ∈ E⊥. This implies that {F, G} (x) = hx [∇xf, ∇xG]i = 0. Then J is a Poisson ideal. We now show that E + a is a Poisson submanifold. Let I := hF ∈ F(g2) F ≡ 0 on E + ai, F ∈ I and let x + a ∈ E + a. Notice that ∇x+aF is an element of E⊥, which is according to first condition of these proposition is a Lie ideal of (g, [· , ·]). Then, for every G ∈ F(g), [∇x+aF, ∇x+aG] ∈ E⊥. Let now compute the Poisson bracket between F and G ∈ F(g) at a point x + a. {F, G} (x + a) = hx + a [∇x+aF, ∇x+aG]i = hx [∇x+aF, ∇x+aG]i = 0, where we have used the second condition of proposition to justify the transition from first to second line and we have zero in the last line because x ∈ E and [∇x+aF, ∇x+aG] ∈ E⊥. Let now show the translation by a is a Poisson isomorphism between g and g. We denote by Ta the translation by a, defined for every x ∈ g, by Ta(x) = x + a. For every F, G ∈ F(g) and every x ∈ g, we have {F ◦ Ta, G ◦ Ta} (x) = hx [∇x(F ◦ Ta), ∇x(G ◦ Ta)]i = (cid:10)x [∇Ta(x)F, ∇Ta(x)G](cid:11) = (cid:10)x + a [∇Ta(x)F, ∇Ta(x)G](cid:11) = {F, G} (Ta(x)), where we have used Condition (2) to justify the transition from second to third line. Since E is a Poisson submanifold of (g, {· , ·}) the restriction of the Poisson isomorphism Ta to E is also a Poisson isomorphism on its image. (cid:3) We now prove Proposition 8. Its easy to verify that T 2 admits the following description as an affine subspace of Proof. g2: T 2 = (e, e) + ∆(g0 ⊕ g−1) ⊕ g2 −, where ∆(g0 ⊕ g−1) := {(x, x) x ∈ g0 ⊕ g−1} and e = Pℓ assumptions of Lemma 9 are satisfied, with a = (e, e), E = ∆(g0 ⊕ g−1) ⊕ g2 [· , ·] = [· , ·]R. (1) It is clear that E = g60 × g>−1. The orthogonal of E is E⊥ = g<0 × g>1, which is a − and it is an ideal of (g2, [· , ·]R), because, for every (x, y) ∈ g<0 × g>1 and subspace of g2 (z, s) ∈ g2, (42) i=1 ei. We check that the two −, g = g2 and [(x, y), (z, s)]R ∈ [g<0 × g>1, g<0 × g>0] ⊂ g<−1 × g>1 ⊂ g<0 × g>1. 14 (2) For every (x, y), (x′, y′) ∈ g2, we have (cid:10)(e, e) [(x, y), (x′, y′)]R(cid:11) = (cid:10)(e, e) [(x, y)+, (x′, y′)+] − [(x, y)−, (x′, y′)−](cid:11)2. Since [(x, y)+, (x′, y′)+] = (u, u) for some u ∈ g it follows that the first term on the right side of Equation (43) is zero. Also [(x, y)−, (x′, y′)−] ∈ g<−1 × g>0, which is orthogonal to (e, e) ∈ g1 × g1; hence the second term on the right hand side of Equation (43) is zero. (43) Proposition 10 Let H, H ∈ F(g2) be defined at every point (x, y) of g2 by H(x, y) := 1 2 hx xi and H(x, y) := 1 2 hy yi . (cid:3) (44) The Hamiltonian vector field XH := {· , H}R (resp. X H := n· , HoR describes on T 2 the equation of motion (34) (resp. (35)) for the 2-Toda lattice. ) is tangent to T 2 and Proof. Let F ∈ F(g2) and let (L, M ) ∈ T 2. We have XH[F ](L, M ) = {F, H}R(L, M ) = 1 + 1 2(cid:10)(L, M ) (cid:2)R∇(L,M )F, ∇(L,M )H(cid:3)(cid:11)2 2(cid:10)(L, M ) (cid:2)∇(L,M )F, R∇(L,M )H(cid:3)(cid:11)2 2(cid:10)(cid:2)(L, M ), R∇(L,M )H(cid:3) ∇(L,M )F(cid:11)2, 1 = − (45) where we have used the Ad-invariance on g2 of H to justify the transition from second to third equality. We deduce from Equation (45) that XH(L, M ) = 1 Hence according to Formula (37), we have 2 (cid:2)R∇(L,M )H, (L, M )(cid:3). XH(L, M ) = [(∇(L,M )H)+, (L, M )] = [(L, 0)+, (L, M )] = [(L+, L+), (L, M )]. To show that X H describes Equation (35) it suffices to repeat the same reasoning of XH. Since T 2 is a Poisson submanifold of (g2, {· , ·}), the Hamiltonian vector fields XH and (cid:3) X H are tangent to T 2. 3.3 The integrability of the 2-Toda lattice According to [5, Theorem 7.3.8], for every simple Lie algebra g of rank ℓ, there exist ℓ homogeneous, independent, Ad-invariant polynomials P1, . . . , Pℓ which generate the al- gebra of Ad-invariant polynomial functions on g and which are of degree, respectively, m1 + 1, . . . , mℓ + 1, where m1, . . . , mℓ are the exponents of g (we note that m1 6 . . . 6 mℓ). Each Pi induces mi + 2 functions Fj,i ∈ F(g2), as follows: 15 Pi(λx − y) = X06j6mi+1 (−1)mi+1−jλjFj,i(x, y). (46) Every function Fj,i for 1 6 i 6 ℓ and 0 6 j 6 mi + 1 is homogeneous of degree j with respect to its first variable and of degree mi + 1 − j with respect to its second variable. Notation 11 We denote by F the family of functions on g2 given by F := (Fj,i, 1 6 i 6 ℓ and 0 6 j 6 mi + 1). (47) Remark 12 (1) The functions F0,1 and F2,1 are the Hamiltonians of the 2-Toda lattice introduced in Proposition 10. (2) The functions F0,i and Fmi+1,i for 1 6 i 6 ℓ are Ad-invariant functions on g2. According to the Adler-Kostant-Symes theorem [2, Theorem 4.37] they are in involution with respect to the bracket {· , ·}R. Also they are independent on g2. They are therefore a good candidates for giving the Liouville integrability of the 2-Toda lattice. However their cardinal 2ℓ is very small compared to dim T 2 − 1 2 Rk(T 2, {· , ·}R). Since, as we noticed in Remark 12, the Hamiltonians of the 2-Toda lattice appear among the functions composing F, the next theorem gives the Liouville integrability of the 2-Toda lattice. Theorem 13 The triplet (T 2, FT 2, {· , ·}R) is an integrable system. Proof. According to the definition of integrability in the sense of Liouville (see [2, Defini- tion 4.13]), to prove Theorem 4 we must show that (1) FT 2 is involutive for the Poisson R-bracket {· , ·}R. (2) FT 2 is independent. (3) The cardinal of the restriction of F to T 2 satisfies card FT 2 = dim T 2 − 1 2 Rk(T 2, {· , ·}R). (48) The proofs of these three points are given in respectively Proposition 14, Proposition 17 and Proposition 19, which are given in the next three subsections. (cid:3) 3.3.1 The restriction of F to T 2 is involutive for {· , ·}R Unlike in the Toda lattice case, it is not the Adler-Kostant-Symes theorem that gives us the commutativity of the considered family of functions F, which is not formed only by the Ad-invariant functions. This commutativity arises from the fact our phase space is a Poisson submanifold of (g2, {· , ·}R), and from the second item of Theorem 4. 16 Proposition 14 The family of functions FT 2 is involutive for the Poisson structure {· , ·}R. Proof. Since the polynomials P1, . . . , Pℓ are Ad-invariant on g (and hence Ad∗-invariant upon identifying g with g∗), according to the second point of Theorem 4 the family F is involutive on (g2, {· , ·}R). Furthermore, T 2 being a Poisson submanifold of (g2, {· , ·}R) (see Proposition 8), the restriction of F to T 2 is involutive. (cid:3) 3.3.2 The restriction of F to T 2 is independent We use an unpublished result of Raïs [10], which establishes the independence of a large family of functions on g2. We state this result below and the proof is in [4, Section 1]. Theorem 15 Let P1, . . . , Pℓ be a generating family of homogeneous polynomials of the algebra of Ad-invariant polynomial functions on g. Let e and h be two elements of g such that e is regular and [h, e] = 2e. For every F ∈ F(g), and every y ∈ g, we denote by dk yF the differential of order k of F at y. Denote by Vk,i, for every 1 6 i 6 ℓ and 0 6 k 6 mi, the element of g defined by hVk,i zi = Ddk+1 h Pi, (ek, z)E , ∀z ∈ g, (49) where, for every x ∈ g and k ∈ N, xk is shorthand for (x, . . . , x) (k times). (1) The family F1 := (Vk,i, 1 6 i 6 ℓ and 0 6 k 6 mi) is linearly independent. (2) The subspace generated by F1 is the Lie subalgebra formed by the sum of the all eigenspaces of adh associated with positive or zero eigenvalues. We first prove, using the first point of Theorem 15, the independence of F (which is a family F(g2)) at a well chosen point (e, h) ∈ T 2 ∩ g2. Afterwards, using the second point of Theorem 15, we show that the restriction of F to the phase space T 2 of the 2-Toda lattice is also an independent family of functions. Proposition 16 Let h ∈ h be such that [h, e] = 2e. (1) The polynomial functions F0,1(x, y), . . . , F0,ℓ(x, y) only depend on the second variable y. Their differentials at the point (e, h) are independent; (2) We denote by ∂ partial derivatives ∂Fj,i (3) The family (d 2 (dim g + ℓ) ∂x , 1 6 i 6 ℓ and 1 6 j 6 mi + 1 are independent at the point (e, h); ∂x the partial differential with respect to the variable x. The 1 (e,h)Fj,i, 1 6 i 6 ℓ, 0 6 j 6 mi + 1) is independent. Proof. (1) For every 1 6 i 6 ℓ and every (x, y) ∈ g2, F0,i(x, y) is the term of degree 0 in λ of Pi(λx − y), that is Pi(y). Hence, the function F0,i, for 1 6 i 6 ℓ, is a homogeneous polynomial of degree mi + 1 which only depends on the second variable y. Moreover, according to two theorems of Kostant [6, Theorem 9] and [7, Theorem 5.2], the differentials of the polynomials P1, . . . , Pℓ are independent at every regular point of g. In particular, 17 they are independent at the regular4 point h. In conclusion the differentials of the functions F0,1, . . . , F0,ℓ are independent at (e, h). (2) For 1 6 i 6 ℓ, the Taylor Formula applied to the polynomial Pi, 1 6 i 6 ℓ, at λx − y, yields Pi(λx − y) = Xk=0 mi+1 (−1)mi+1−k λk k! Ddk yPi, xkE . (50) By identifying the coefficients of Equations (50) and (46), we obtain, for every 1 6 i 6 ℓ and 1 6 k 6 mi + 1, the equality Fk,i(x, y) = 1 k! Ddk yPi, xkE . (51) By differentiating Fk+1,i with respect to the variable x, we obtain (cid:28) ∂Fk+1,i ∂x (x, y), z(cid:29) = 1 k! Ddk+1 y Pi, (xk, z)E , ∀k = 0, . . . , mi, and ∀z ∈ g. (52) In particular, according to Theorem 15, when (x, y) = (e, h), Equation (52) becomes (cid:28) ∂Fk+1,i ∂x (e, h), z(cid:29) = 1 k! Ddk+1 h Pi, (ek, z)E = 1 k! hVk,i zi , ∀z ∈ g, (53) for every 1 6 i 6 ℓ and 1 6 k 6 mi + 1. According to Theorem 15, the family (Vk,i, 1 6 i 6 ℓ and 0 6 k 6 ℓ) is independent. This implies the independence of the family ( ∂Fk,i (3) Denote by M the matrix ∂x (e, h), 1 6 i 6 ℓ and 1 6 k 6 mi + 1). M = (d(e,h)Fk,i, 0 6 k 6 mi + 1 and 1 6 i 6 ℓ) =   ... ∂Fmℓ+1,ℓ ∂x ∂F0,1 ∂x (e, h) ... ∂F0,1 ∂y (e, h) ... ∂F0,ℓ ∂x (e, h) ∂F1,1 ∂x (e, h) ∂F0,ℓ ∂y (e, h) ∂F1,1 ∂y (e, h) ... ∂Fmℓ+1,ℓ ∂y (e, h) (e, h) .   According to item (1), the form of the matrix M is M = (cid:18) 0 A B ∗ (cid:19) , 4The element h of g is regular. Indeed, since it verifies [h, e] = 2e with e regular, it belongs to a principal sl(2) triple (see [12, Theorem 32.1.5]). 18 ∂y (e, h), 1 6 i 6 ℓ) and B = ( ∂Fk,i where A = ( ∂F0,i ∂x (e, h), 1 6 k 6 mi + 1 and 1 6 i 6 ℓ). According to (1) and (2), the ℓ rows of the matrix A and the 1 2 (dim g + ℓ) rows of the matrix B are independent. The independence of the rows of M yields the independence of the differentials of the family of functions F at (e, h) ∈ T 2. (cid:3) Proposition 17 The restriction of the family of functions F to the phase space T 2 of the 2-Toda lattice is an independent family. Proof. differentials It suffices to show that the restrictions on the tangent space T(e,h)T 2 of the (d(e,h)Fj,i, 1 6 i 6 ℓ et 0 6 j 6 mi + 1) are independent. Since the tangent space of T 2 is g60 ×g>−1 and since (d vanish on g60 × {0}, it suffices to show that: (a) The restriction to the space {0} × h of the linear forms (e,h)F0,i, 1 6 i 6 ℓ) (d(e,h)F0,i, 1 6 i 6 ℓ ) is an independent family. (b) The restriction to the space g60 × {0} of the linear independent forms (d (e,h)Fj,i, 1 6 i 6 ℓ and 1 6 j 6 mi + 1) is an independent family. According to Definition (46), F0,i(x, y) = Pi(y) for every 1 6 i 6 ℓ and every (x, y) ∈ g2. Then ∂F0,i ∂x (e, h) = 0 and (cid:28) ∂F0,i ∂y (e, h), z(cid:29) = h∇hPi zi , ∀z ∈ g. (54) Since ∇hPi ∈ h (because [h, ∇hPi] = 0 and h ∈ h is regular), the restriction to h of the family ( ∂F0,1 ∂y (e, h)) is an independent family of linear forms. This implies the independence of the restriction to {0} × h of d ∂y (e, h), . . . , ∂F0,ℓ (e,h)F0,1, . . . , d (e,h)F0,ℓ. According to equation (53), for every 1 6 i 6 ℓ and 0 6 k 6 ℓ, hVk,i zi = k!(cid:28) ∂Fk+1,i ∂x (e, h), z(cid:29) . (55) The second point of Theorem 15 shows that the subspace generated by the family (Vk,i, 1 6 i 6 ℓ and 0 6 k 6 mi) is contained in the Lie subalgebra obtained by summing the eigenspaces of adh, associated with the nonnegative eigenvalues, which is in our case g>0. This proves, using Equation (55), that the restriction to g60 of the family of linear forms ( ∂Fk,i ∂x (e, h), 1 6 i 6 ℓ and 1 6 k 6 mi + 1) is an independent family. Therefore the restriction to g60 × {0} of (d (e,h)Fk,i, 1 6 i 6 ℓ and 1 6 i 6 mi + 1) is an independent family. (cid:3) 19 3.3.3 The exact number of functions According to Equation (47), the cardinal of F is related to the exponents mi, 1 6 i 6 ℓ, as follows ℓ ℓ card F = Xi=1 (mi + 2) = mi + 2ℓ. Xi=1 Since Pℓ i=1 mi = 1 have card FT 2 = 1 the relation below is satisfied: 2 (dim g−ℓ) (see [5, Theorem 7.3.8]) and FT 2 is an independent family, we 2 (dim g + 3ℓ). The dimension of T 2 is equal to dim g + 2ℓ. In conclusion, card FT 2 = dim T 2 − 1 2 Rk(T 2, {· , ·}R) if and only if Rk(T 2, {· , ·}R) = dim g + ℓ. We need therefore to prove this last result, which will be done in Proposition 19 below. The rank of the restriction of {· , ·}R to T 2: The purpose of this part is to compute the rank of the Poisson manifold (T 2, {· , ·}R), i.e., the maximum of the rank at x of the Poisson R-bracket for every x ∈ T 2. We begin by establishing an isomorphism between the Poisson manifold (T 2, {· , ·}R) and a product Poisson manifold. Before doing this, notice that it follows from Lemma 9 applied to a = 0 and E = g0 ⊕ g1, g = g and [· , ·] = [· , ·]R that (the subspace) g0 ⊕ g1 is a Poisson submanifold of (g, {· , ·}R), since the orthogonal of g0 ⊕ g1 is a Lie ideal of (g, [· , ·]R). Proposition 18 The Poisson manifold (T 2, {· , ·}R) is isomorphic to the product Poisson manifold (g∗, {· , ·}) × (g0 ⊕ g1, {· , ·}R), where {· , ·} is the Lie-Poisson bracket on g∗, R is the difference of projections on g+ and g−, and {· , ·}R is the Poisson R-bracket on g (restricted to g0 ⊕ g1). Proof. As we saw in the proof of Proposition 8, a := (e, e) and E := g2 −⊕∆(g0⊕g−1) satisfy Conditions (1) and (2) of Lemma 9; hence T 2 = (e, e) + E and E are isomorphic Poisson submanifolds of (g2, {· , ·}R). According to Lemma 9 again, E is Poisson isomorphic to (g2/E⊥)∗, where g2/E⊥ is equipped with the quotient of the Lie bracket [· , ·]R with respect to the Lie ideal E⊥. In conclusion, (T 2, {· , ·}R) is Poisson isomorphic to ((g2/E⊥)∗, {· , ·}′), where {· , ·}′ is the Lie Poisson bracket of (g2/E⊥)∗. Let us determine the quotient Lie algebra g2/E⊥. The Lie algebra g2 (endowed with the bracket [· , ·]R) is isomorphic to the direct sum of the Lie algebras g, g6−1 and g>0, the isomorphism being given, for every x ∈ g, y− ∈ g−, y+ ∈ g+, by g × g− × g+ ≃ (x, y−, y+) → (x, x) + (y−, y+). g2 (56) 20 The Lie algebra isomorphism (56) identifies the subspace E⊥ = g− × g>1 ⊂ g2 with (0, g−, g>2). Hence the quotient g2/E⊥ is the direct sum of the Lie algebra g (endowed with the usual bracket) and the quotient Lie algebra g+/g>2. We conclude that T 2 is Poisson isomorphic to the dual of the direct sum Lie algebra g ⊕ (g+/g>2), endowed with the Lie-Poisson structure, and then is isomorphic to the product of the Poisson manifold g∗ (endowed with the Lie-Poisson structure) with (g+/g>2)∗ (endowed with the Lie-Poisson structure). To complete the proof, it suffices to recall (see Lemma 9) that g0 ⊕ g1 is a Poisson submanifold of (g, {· , ·}R), isomorphic to the Lie-Poisson structure on the dual of the quotient Lie algebra g/(g0 ⊕ g1)⊥ = g+/g>2. (cid:3) Proposition 19 The rank of the restriction of the Poisson R-bracket {· , ·}R to the mani- fold T 2 is dim g + ℓ. As a consequence, the following relation is satisfied: card FT 2 = dim T 2 − 1 2 Rk(T 2, {· , ·}R). Proof. According to Proposition 18 the Poisson submanifold (T 2, {· , ·}R) is isomorphic to the product manifold (g∗, {· , ·}) × (g0 ⊕ g1, {· , ·}R). This result proves that the restriction of the rank of Poisson R-bracket {· , ·}R to the manifold T 2 is the sum of the rank of the Lie-Poisson structure on g∗, which is dim g − ℓ (see [12, Proposition 29.3.2]), and the rank of the Poisson R-bracket on g0 ⊕ g1. We calculate the latter rank and show that it is 2ℓ, which finishes the proof. Let (z1, . . . , z2ℓ) be the coordinate system on g0 ⊕ g1, defined by (cid:26) zi(x) = hhi xi , zi+ℓ(x) = he−i xi , ∀i ∈ {1 . . . , ℓ}, where as before, for 1 6 i 6 ℓ, the element e−i is a non-zero eigenvector associated with the root −αi. The Lie-Poisson brackets between these coordinate functions are given by the following formulae: (cid:26) {zi, zj}R = {zi+ℓ, zj+ℓ}R = 0, {zi+ℓ, zj}R = cijzi+ℓ, ∀i, j ∈ {1, . . . , ℓ}, where C := (cij)16i,j6ℓ is the Cartan matrix of g. Then the Poisson matrix M = ({zi, zj}R)16i,j62ℓ is equal to M = (cid:18) 0 −AT 0 (cid:19) , A where A := vC and v := diag(z1+ℓ, . . . , z2ℓ). Since C is invertible, the rank of A at any point for which (z1+ℓ, . . . , z2ℓ) = (1, . . . , 1) is ℓ; therefore the rank of M is 2ℓ, which implies that the rank of the Poisson R-structure on g0 ⊕ g1 is 2ℓ. (cid:3) 21 3.4 The integrability for the quadratic Poisson R-bracket In this subsection we study the integrability of the 2-Toda lattice on some Lie algebra g with respect to a quadratic Poisson R-bracket. To construct a quadratic Poisson R-bracket on g2 it is necessary that g is an associative algebra of finite dimensional and R and its antisymmetric part R− are solutions of (mCYBE). For this it suffices to choose g = gln(C) and R(x, y) = (R(x − y) + y, R(x − y) + x), where where R = P+ − P−, P+ (resp. P−) being the projection of gln(C) on the subalgebra of upper (resp. strictly lower) triangular matrices (we will show again later that R− is a solution of (mCYBE) of g2). In this subsection we show the integrability of a system of equations which is exactly the system (1), up to the fact that we do no longer assume the matrices L, M to be traceless. We denote by T 2′ the phase space of the 2-Toda lattice, i.e., (L, M ) =     a11 1 a21 a22 ... an1   ,  T 2′ := {(L, M ) ∈ gln(C) × gln(C) · · · . . . . . . . . . . . .  b11 b21 0 1 · · · . . . · · · an,n−1 ann 0 bn,n−1 b1n ... ... bnn   }   and we name as the 2-Toda lattice on gln(C) the system of differential equations (1) with the constraint (L, M ) ∈ T 2′. Before studying the integrability of the 2-Toda lattice with respect to the quadratic Poisson R-bracket we will study the integrability of the latter system with respect to the linear Poisson R-bracket. 3.4.1 The Liouville integrability of the 2-Toda lattice on gln(C) for the linear Poisson R-bracket We equip gln(C) × gln(C) with the Ad-invariant, non-degenerate, symmetric, bilinear form h· · i2, defined for every (x, y), (x′, y′) ∈ gln(C) × gln(C) by (cid:10)(x, y) (x′, y′)(cid:11)2 := (cid:10)x x′(cid:11) −(cid:10)y y′(cid:11) = Trace(xx′) − Trace(yy′). (57) As in Subsection 3.2 we consider R(x, y) = (R(x − y) + y, R(x − y) + x) and we consider the linear Poisson R-bracket5, defined for every F, G ∈ F(gln(C) × gln(C)) at (x, y) ∈ gln(C) × gln(C), by {F, G}R(x, y) = 1 2 ((cid:10)(x, y) [R∇(x,y)F, ∇(x,y)G](cid:11)2 +(cid:10)(x, y) [∇(x,y)F, R∇(x,y)G](cid:11)2). 5According to Corollary 2, the endomorphism R is an R-matrix on gln(C) × gln(C). 22 By using the proofs of Propositions 8 and 10 we show that the phase space T 2′ is a Poisson submanifold of (gln(C) × gln(C), {· , ·}R) and XH := {· , H}R and XH ′ := {· , H ′}R where 2 Trace y2 describes on T 2′ the equations of motion of H(x, y) = 1 the 2-Toda lattice. 2 Trace x2 and H ′(x, y) = 1 Let us now study the integrability of the 2-Toda lattice on gln(C). Let Pi, for ev- ery i ∈ N be the Ad-invariant function of gln(C) defined for all x ∈ gln(C) by Pi(x) = 1 i+1 Trace(xi+1). We define Fj,i ∈ F(gln(C)×gln(C)) by Pi(λx−y) = Pi+1 and we define j=0(−1)mi+1−jλjFj,i(x, y) F ′ = (Fj,i, 0 6 i 6 n − 1 and 0 6 j 6 i + 1). Note that the functions that make up F ′ are the functions that make up F together with the functions F1,0(x, y) := Trace(x) and F0,0(x, y) := Trace(y). We have the following proposition. Proposition 20 The triplet (T 2′, F ′ T 2 ′ , {· , ·}R) is an integrable system. Proof. According to the item (2) of Theorem 4 the family F ′ is involutive for the Pois- son bracket {· , ·}R. Since F ′ is an independent family on T 2′ (this follows from the in- dependence of the differentials dF0,0 and dF1,0 at all points of gln(C) × gln(C) and the independence of F on the submanifold T 2, which is the submanifold of T 2′ defined by F1,0 = F0,0 = 0), and since the cardinal of F ′ is n(n+3) we have, according to [2, Proposi- tion 4.12], the inequality 2 Rk(T 2′ , {· , ·}R) 6 2(dim(T 2′ ) − Card(F ′)) = n2 + n − 2. (58) Furthermore Rk(T 2′, {· , ·}R) > Rk(T 2, {· , ·}R) = n2 + n − 2. This implies that the rank T 2′ = dim T 2′ − of the restriction to T 2′ of {· , ·}Q 2 Rk(T 2′, {· , ·}Q 1 (cid:3) R is exactly n2 + n − 2. Then card F ′ R). This completes the proof. 3.4.2 The integrability of the 2-Toda lattice on gln(C) for the quadratic Poisson R-bracket Let the quadratic R-bracket be defined for every F, G ∈ F(gln(C) × gln(C)) at (x, y) ∈ gln(C) × gln(C), by {F, G}Q R(x, y) := 1 2(cid:10)[(x, y), ∇(x,y)F ] R((x, y)∇(x,y)G + ∇(x,y)G(x, y))(cid:11)2 −(F ↔ G). (59) Since R−(x, y) = 1 2 (R − R∗)(x, y) = (R−(x − y) + y, R−(x − y) + x) and since both the endomorphism R = P+ − P− and its antisymmetric part R− = 1 2 (R − R∗) are solutions of (mCYBE) of gln(C) of constant c = 1, it follows from Corollary 2 that R and R− are solutions of (mCYBE) of gln(C) × gln(C) of constant c = 1. According to [8, Section 4] the 23 R-bracket (59) is indeed a Poisson bracket, that we call the quadratic Poisson R-bracket (on gln(C) × gln(C)). By a direct computation in coordinates we can prove the following proposition. Proposition 21 The phase space T 2′ of the 2-Toda lattice on gln(C) is a Poisson sub- manifold of (gln(C) × gln(C), {· , ·}Q We show the following proposition. R). Proposition 22 Let Pi, for every i ∈ N be the Ad-invariant functions of gln(C) defined for all x ∈ gln(C) by Pi(x) = 1 i+1 Trace(xi+1) and let φλ : gln(C) × gln(C) → gln(C), for every λ ∈ F, be defined by φλ : (x, y) → λx − y. (1) For every i, j ∈ N and every λ, γ ∈ C, the functions Pi ◦φλ and Pj ◦φγ are in involution for {· , ·}Q R. (2) The Hamiltonian vector field X Q := {· , Pi ◦ φλ}Q R is given by Pi◦φλ X Q Pi◦φλ = −[(x, y), ((R − I)(λx − y)i+1, (R + I)(λx − y)i+1)]. (60) Proof. (1) According to (59) and (22), for every x, y ∈ g, {Pi ◦ φλ, Pj ◦ φµ}Q R(x, y) 1 = 2(cid:10)[(x, y), (λ(λx − y)i, (λx − y)i)] R(µx(µx − y)j, y(µx − y)j)(cid:11)2 2(cid:10)[(x, y), (λ(λx − y)i, (λx − y)i)] R(µ(µx − y)jx, (µx − y)jy)(cid:11)2 1 + −(j, µ) ←→ (i, λ). By replacing R by its expression R(x, y) = (R(x − y) + y, R(x − y) + x), we obtain {Pi ◦ φλ, Pj ◦ φµ}Q = (cid:10)[(x, y), (λ(λx − y)i, (λx − y)i)] (R((µx − y)j+1), R((µx − y)j+1))(cid:11)2 R(x, y) 1 + + 1 2(cid:10)[(x, y), (λ(λx − y)i, (λx − µ)i)] (y(µx − y)j, µx(µx − y)j)(cid:11)2 2(cid:10)[(x, y), (λ(λx − y)i, (λx − µ)i)] ((µx − y)jy, µ(µx − y)jx)(cid:11)2 −(j, µ) ←→ (i, λ). By using Formula (57) we obtain {Pi ◦ φλ, Pj ◦ φµ}Q R(x, y) = (cid:10)[λx − y, (λx − y)i] R((µx − y)j+1)(cid:11) 1 + − 1 2 (cid:10)[λx, (λx − y)i] y(µx − y)j + (µx − y)jy(cid:11) 2 (cid:10)[y, (λx − y)i] µx(µx − y)j + µ(µx − y)jx)(cid:11) −(j, µ) ←→ (i, λ). 24 Since [λx, (λx − y)i] = [y, (λx − y)i], {Pi ◦ φλ, Pj ◦ φµ}Q R(x, y) = − 1 2 (cid:10)[y, (λx − y)i] 2(µx − y)j+1(cid:11) − (j, µ) ←→ (i, λ) = (cid:10)y [(λx − y)i, (µx − y)j+1](cid:11) − (j, µ) ←→ (i, λ) = 0, where we used Lemma 6 to provide the last line. (2) Let K be a function of g2; we denote by (a, b) = ∇(x,y)K, according to (59) and (22), X Q (x, y)[K] Pi◦φλ 1 = 2 (cid:10)[(x, y), (a, b)] R(λx(λx − y)i + λ(λx − y)ix, y(λx − y)i + (λx − y)iy)(cid:11) 2 (cid:10)[(x, y), (λ(λx − y)i, (λx − y)i)] R(xa + ax, yb + by)(cid:11) . 1 − (61) We introduce the shorthand B = xa + ax − yb − by. By using the expression for R the Hamiltonian vector field (61) becomes X Q Pi◦φλ (x, y)[K] = (cid:10)[(x, y), (a, b)] (R((λx − y)i+1), R((λx − y)i+1))(cid:11)2 1 + 1 − 1 2(cid:10)[(x, y), (a, b)] (y(λx − y)i + (λx − y)iy, λx(λx − y)i + λ(λx − y)ix)(cid:11)2 2(cid:10)[(x, y), (λ(λx − y)i, (λx − y)i)] (R(B), R(B))(cid:11)2 2(cid:10)[(x, y), (λ( x − y)i, (λx − y)i)] (yb + by, xa + ax)(cid:11)2 1 − − = −(cid:10)[(x, y), (R((λx − y)i+1), R((λx − y)i+1))] (a, b)(cid:11)2 1 2(cid:10)[(x, y), (y(λx − y)i + (λx − y)iy, λx(λx − y)i + λ(λx − y)ix)] (a, b)(cid:11)2 2 (cid:10)[λx − y, (λx − y)i] R(B)(cid:11) 2 (cid:10)[x, λ(λx − y)i] yb + by(cid:11) + 2 (cid:10)[y, (λx − y)i] xa + ax(cid:11) , 1 1 − − Moreover according to the Ad-invariance of Pi and the property hx yzi = hxy zi , ∀ x,y,z ∈ g, we obtain X Q Pi◦φλ (x, y)[K] = −(cid:10)[(x, y), (R((λx − y)i+1), R((λx − y)i+1))] (a, b)(cid:11)2 1 − 1 2(cid:10)[(x, y), (y(λx − y)i + (λx − y)iy, λx(λx − y)i + λ(λx − y)ix)] (a, b)(cid:11)2 2 (cid:10)y[x, λ(λx − y)i] + [x, λ(λx − y)i]y b(cid:11) 2 (cid:10)x[y, (λx − y)i] + [y, (λx − y)i]x a(cid:11) . 1 − + 25 We then deduce that X Q Pi◦φλ (x, y) = −[(x, y), (R((λx − y)i+1), R((λx − y)i+1))] − + 1 2 1 2 [(x, y), (y(λx − y)i + (λx − y)iy, λx(λx − y)i + λ(λx − y)ix)] (x[y, (λx − y)i] + [y, (λx − y)i]x, y[x, λ(λx − y)i] + [x, λ(λx − y)i]y). Let us compute separately the two last lines: − 1 2 [(x, y), (y(λx − y)i + (λx − y)iy, λx(λx − y)i + λ(λx − y)ix)] + 1 2 (x[y, (λx − y)i] + [y, (λx − y)i]x, y[x, λ(λx − y)i] + [x, λ(λx − y)i]y) = −(x(λx − y)iy − y(λx − y)ix, λy(λx − y)ix − λx(λx − y)iy) = ([x, (λx − y)i+1], −λ[x, (λx − y)i+1]) = [(x, y), ((λx − y)i+1, −(λx − y)i+1)]. Then we have Formula (60). (cid:3) Let us compare the Hamiltonian vector fields for the quadratic R-Poisson bracket and the linear R-Poisson bracket. Formula (20) gives, in our case, the following expression for the Hamiltonian vector field of Pi+1 ◦ φλ with respect to the linear R-Poisson structure: XPi+1◦φλ = 1 2 (λ − 1)[(x, y), ((R − I)(λx − y)i+1, (R + I)(λx − y)i+1)]. (62) Comparing Formulae (62) and (60), we obtain, for λ 6= 1, X Q Pi◦φλ (x, y) = 2 1 − λ XPi+1◦φλ(x, y). The relation (63) implies that for every 0 6 i 6 n − 1, we have (63) (64) X Q F0,i = 2XF0,i+1, X Q Fj,i − X Q Fj−1,i = 2XFj,i+1, 1 6 j 6 i + 1, X Q Fi+1,i = −2XFi+2,i+1.   Notice also that the Hamiltonian vector fields X Q and XF1,0 are precisely the equations F0,0 of the 2-Toda lattice on gln(C) as can be shown by specializing (64) to i = 0 and the first item of Remark 12. 26 Theorem 23 The triplet (T 2′, F ′ T 2′ , {· , ·}Q R) is an integrable system. Proof. The involutivity of the family F ′ on (g2, {· , ·}R) follows from item (1) in Propo- sition 22. Since F ′ is an independent family on T 2′ (see the proof of Proposition 20), and since the cardinal of F ′ is n(n+3) , we have, according to [2, Proposition 4.12], the inequality 2 Rk(T 2′ , {· , ·}Q R) 6 2(dim(T 2′ ) − Card(F ′)) = n2 + n − 2. (65) Moreover, according to Formula (64), the family of vector fields XF ′ := (X Q , 0 6 i 6 n − Fj,i 1 and 0 6 j 6 i) and the family of vector fields XF := (XFj,i, 1 6 i 6 n−1 and 1 6 j 6 i+1) have the same rank at all points. By choosing a point in T 2, we can deduce from the fact that F is an integrable system on T 2 (Theorem 13) that this rank is at least the cardinal of F (= n(n+3) 2 − 2) minus the number of independent Casimir functions on T for the linear bracket {· , ·}R ( = n − 1). Since F ′ is involutive, we have therefore the inequality Rk {· , ·}Q R > 2(Rk(XF ′)) = 2(Rk(XF )) > 2( n(n + 3) 2 − 2 − n + 1) = n2 + n − 2. Together with (65), this implies that the rank of the restriction to T 2′ of {· , ·}Q n2 + n − 2. The identity card F ′ the proof. R is exactly R) follows and completes (cid:3) T 2 ′ = dim T 2′ − 1 2 Rk(T 2′, {· , ·}Q 3.5 The relation between the 2-Toda lattice and the Toda lattice In this section we show that the Toda lattice is a restriction of the 2-Toda lattice. We begin by recalling the Liouville integrable system of the Toda lattice. 3.5.1 The Toda lattice We give some definitions and properties of the Toda lattice, which will be useful afterwards in this section. Definition 24 (1) The phase space T of the Toda lattice is the affine subspace of g given by T := g−1 ⊕ g0 + e, (66) where e = Pℓ i=1 ei. (2) The Toda lattice is the system of differential equations on T given by the Lax equation A = [A+, A], (67) where A+ is the projection of A on g+. 27 We consider the endomorphism R := P+ − P− of g, the difference of the projections on g+ and g−. The Poisson R-bracket on g is defined, for every F, G ∈ F(g) and every x ∈ g, by {F, G}R := 1 2 hx [R∇xF, ∇xG] + [∇xF, R∇xG]i . (68) Theorem 25 [9, Section 4.1] (1) The affine subspace T of g is a Poisson submanifold of (g, {· , ·}R). (2) Let H ∈ F(g), defined for every x ∈ g by H(x) = 1 Hamiltonian field XH := {· , H}R is the equation of motion (67) of the Toda lattice. (3) Let P1, . . . , Pℓ be a generating family of homogeneous polynomials of the algebra of Ad-invariant functions on g of degree respectively, m1 + 1, . . . , mℓ + 1. We define F0 := (P1, . . . , Pℓ). The triplet (F0, {· , ·}R, T ) is Liouville integrable system and the equation of motion of the Toda lattice is 2 hx xi. The equation of the A := {· , P1}R(A) = [(∇AP1)+, A]. (69) 3.5.2 Restriction of the 2-Toda lattice and construction of the Toda lattice The phase space T 2 of the 2-Toda lattice decomposes as T 2 := g6−1 × g>0 ⊕ ∆(g−1 ⊕ g0) + (e, e), (70) where ∆(g−1 ⊕ g0) := {(x, x) x ∈ g−1 ⊕ g0}. Theorem 26 Let g2 be the Lie algebra g × g with {· , ·}R the R-Poisson bracket, T 2 the phase space of the 2-Toda lattice and T ′ := ∆(g−1 ⊕ g0) + (e, e). (1) The submanifold T ′ is a Poisson submanifold of (g2, {· , ·}R). (2) Let (T , {· , ·}R) be the phase space of the Toda lattice, equipped with the Poisson R- bracket given as in (68). The map ϕ : (T , {· , ·}R) → (T ′, {· , ·}R) x 7→ (x, x) (71) is an isomorphism of integrable systems. Proof. (1) It is clear that T ′ = ∆(g−1 ⊕ g0 + e) = T 2 ∩ g2 is a Poisson submanifold of (g2, {· , ·}R) while g2 because the orthogonal of g2, which is g2 section T ′ is then a Poison submanifold of (g2, {· , ·}R). (2) By using the coordinate functions on T and on T ′ we show that ϕ is a Poisson isomor- phism. Furthermore, the functions of the integrable system F = (Fk,i, 1 6 i 6 ℓ and 0 6 +. According to Proposition 8, T 2 + is a Poisson submanifold of (g2, {· , ·}R) +, is a Lie ideal for the bracket [· , ·]R; their inter- 28 k 6 mi + 1), restricted to T ′, and pulled back on T by ϕ, give again the functions of the family F0 of the Toda lattice. Specifically, for every i in 1, . . . , ℓ and every 0 6 k 6 mi + 1, the functions Fk,i are all equal (up to multiplicative constants) to the function Pi. In fact, for every x ∈ g0 ⊕ g−1, Pi ◦ φλ(ϕ(x)) = Pi(φλ(x, −x)) = Pi(λx − x) = (λ − 1)mi+1Pi(x), and so mi+1 (−1)mi+1−kλkFk,i(ϕ(x)) = Xk=0 mi+1 (−1)mi+1−kλkC k mi+1Pi(x) Xk=0 In conclusion, ϕ : T → T ′ is Poisson isomorphic and and Fk,i(ϕ(x)) = C k ϕ∗F = F0 (see item (3) of Theorem 25 for the definition of F0). This proves the claim. (cid:3) mi+1Pi(x). References [1] Mark Adler and Pierre van Moerbeke. String-orthogonal polynomials, string equations, and 2-Toda symmetries. Comm. Pure Appl. Math., 50(3):241 -- 290, 1997. [2] Mark Adler, Pierre van Moerbeke, and Pol Vanhaecke. Algebraic integrability, Painlevé geometry and Lie algebras, volume 47 of Ergebnisse der Mathematik und ihrer Grenzge- biete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics]. Springer- Verlag, Berlin, 2004. [3] Guido Carlet. The Hamiltonian structures of the two-dimensional Toda lattice and R-matrices. Lett. Math. Phys., 71(3):209 -- 226, 2005. [4] Dennis DeTurck, Hubert Goldschmidt, and Janet Talvacchia. Connections with pre- scribed curvature and Yang-Mills currents: the semi-simple case. Ann. Sci. École Norm. Sup. (4), 24(1):57 -- 112, 1991. [5] Jacques Dixmier. Enveloping algebras, volume 11 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 1996. Revised reprint of the 1977 translation. [6] Bertram Kostant. Lie group representations on polynomial rings. Bull. Amer. Math. Soc., 69:518 -- 526, 1963. [7] Bertram Kostant. The solution to a generalized Toda lattice and representation theory. Adv. in Math., 34(3):195 -- 338, 1979. 29 [8] Luen Chau Li and Serge Parmentier. Nonlinear Poisson structures and r-matrices. Comm. Math. Phys., 125(4):545 -- 563, 1989. [9] A. M. Perelomov. Integrable systems of classical mechanics and Lie algebras. Vol. I. Birkhäuser Verlag, Basel, 1990. Translated from the Russian by A. G. Reyman. [10] Mustapha Rais. Sur les données des polynômes invariants sur une algèbre de Lie semi-simple complexe (lettre à Hubert Goldschmidt,1988). [11] Michael A. Semenov-Tian-Shansky. Dressing transformations and Poisson group ac- tions. Publ. Res. Inst. Math. Sci., 21(6):1237 -- 1260, 1985. [12] Patrice Tauvel and Rupert W. T. Yu. Lie algebras and algebraic groups. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2005. [13] Kimio Ueno and Kanehisa Takasaki. Toda lattice hierarchy. II. Proc. Japan Acad. Ser. A Math. Sci., 59(6):215 -- 218, 1983. 30
1107.5890	1	1107	2011-07-29T07:50:06	Non-commutative Hodge structures	[ "math.AG", "math.CV" ]	This article gives a survey of recent results on a generalization of the notion of a Hodge structure. The main example is related to the Fourier-Laplace transform of a variation of polarizable Hodge structure on the punctured affine line, like the Gauss-Manin systems of a proper or tame algebraic function on a smooth quasi-projective variety. Variations of non-commutative Hodge structures often occur on the tangent bundle of Frobenius manifolds, giving rise to a tt* geometry.	math.AG	math	NON-COMMUTATIVE HODGE STRUCTURES by Claude Sabbah Abstract. This article gives a survey of recent results on a generalization of the notion of a Hodge structure. The main example is related to the Fourier-Laplace transform of a variation of polarizable Hodge structure on the punctured affine line, like the Gauss-Manin systems of a proper or tame algebraic function on a smooth quasi-projective variety. Variations of non-commutative Hodge structures often occur on the tangent bundle of Frobenius manifolds, giving rise to a tt* geometry. R´esum´e (Structures de Hodge non commutatives). Nous donnons un panorama des r´esultats r´ecents concernant une g´en´eralisation de la notion de structure de Hodge. L’exemple principal est celui produit par la transformation de Fourier-Laplace d’une variation de structure de Hodge polarisable sur la droite affine ´epoint´ee, comme les syst`emes de Gauss-Manin de fonctions alg´ebriques propres ou mod´er´ees sur une vari´et´e quasi-projective lisse complexe. Le fibr´e tangent d’une vari´et´e de Frobenius peut souvent etre muni d’une variation de structures de Hodge non-commutatives polarisables, d’o`u l’on d´eduit une g´eom´etrie sp´eciale du type tt. 1. Introduction Notation 1.1. All along this article, the abbreviation “nc” means “non-commutative”. We will consider the Riemann sphere P1 equipped with two affine charts U0 = Spec C[z], U∞ = Spec C[z′], and U0 ∩ U∞ = Spec C[z, z−1] = Spec C[z′, z′−1] where the identification is given by z′ = z−1. We will consider the involution ι : P1 → P1 defined by ι(z) = −z, and the anti-holomorphic involution γ : P1 → P1 defined by γ(z) = 1/z. The composed involution γ ◦ ι will be denoted by σ. When restricted to S := {z ∈ U0 z = 1}, ι and σ coincide, since γS = IdS. 2000 Mathematics Subject Classification. 14D07, 34M40. Key words and phrases. Non-commutative Hodge structure, Fourier-Laplace transformation, Brieskorn lattice. This research was supported by the grant ANR-08-BLAN-0317-01 of the Agence nationale de la recherche. 2 C. SABBAH 1.2. The terminology “non-commutative Hodge structure” (which should not be con- fused with that of non-abelian Hodge theory developed by C. Simpson [49, 52, 53]) has been introduced by Katzarkov-Kontsevich-Pantev [26] to cover the kind of Hodge structure one should expect on the periodic cyclic cohomology of smooth compact non-commutative spaces (although we will keep the setting of standard commutative algebraic geometry in this article and we will not provide any example in the setting of non-commutative spaces envisioned in loc. cit., which we refer to for details, as well as to [19] and the recent preprints [20, 47]). Generalizations of the notion of a Hodge structure already occur for usual algebraic geometric objects, like polynomial functions, where such a structure is related to oscillating integrals. In this context, such a structure has been brought to light by Cecotti, Vafa et al. in various articles [5, 4, 6], and formalized by C. Hertling in [10] under the name of pure (and polarized) TERP structure, so as to treat variations of such objects. When such a structure occurs on the tangent bundle of a manifold, this produces a tt geometry on this manifold. By mirror symmetry, such a structure is expected on the quantum cohomology of some algebraic varieties, in the sense that the associated Frobenius manifold should underlie a variation of polarized non-commutative Hodge structures on its tangent bundle. Hertling (cf. loc. cit.) has also formalized the compatibility relations be- tween the Frobenius manifold relations and the tt* geometry under the name of a C(ecotti)D(ubrovin)V(afa)-structure. Explicit formulas from the quantum cohomol- ogy point of view have been obtained by H. Iritani [17, 18] (cf. also [26] for the projective space). We will not go further in this direction. The purpose of this article is to survey recent developments concerning non- commutative Hodge structures by themselves. We will only mention some results concerning their variations and the corresponding limit theorems. One can already emphasize that these structures allow limiting behaviour with irregular singularities (wild behaviour), while usual variations of Hodge structures only allow limiting be- haviour with regular singularities (tame behaviour). As a consequence, they fit with transformations which do not preserve the tame behaviour, like the Fourier-Laplace transformation. The first attempt to provide a good Hodge theory in such a context seems to be the notes [8] by P. Deligne (the relation with nc. Hodge structures is briefly discussed in §6 and with more details in [41, §6]), and the expectation of a behaviour by Fourier transformation analogous to the case of Fourier-Deligne transformation in ℓ- adic theory is clearly stated in [25, Rem. 7.3.3.3]. In some sense, Theorem 5.6 answers this remark, at least in dimension one. On the other hand, the frame of such a theory, restricted to the regular singularity case, was in germ in the work of A.N. Varchenko [55] in Singularity theory, as well as in various later works concerning Hodge theory for isolated singularity of holomorphic functions. 1.3. The underlying analytic theorems in this non-commutative Hodge theory ap- ply to a wider context, that of twistor structures. The main guiding article in this NON-COMMUTATIVE HODGE STRUCTURES 3 direction is due to C. Simpson [51], which unifies under the notion of twistor struc- ture (and variation of such) various objects considered in nonabelian Hodge theory. The theory of variations of pure and polarized twistor structures has been completed by T. Mochizuki [30] in the tame case, extending the previous work of Simpson in dimension one [48], and this work culminates with [33], where T. Mochizuki extends his previous results to the wild setting (see also [37, 40]). Giving a pure and polarized twistor structure of weight w ∈ Z is equivalent to giving a complex vector space H and a positive definite Hermitian form h on it, together with an integer w (although giving a variation of such objects is more subtle, since it involves the notion of a harmonic metric [50]). This can be encoded in a vector bundle of pure slope w on P1 (which is now called a pure twistor structure of weight w), together with the right replacement of a Hermitian form, called a polarization of this twistor structure. 1.4. The next step consists in generalizing the notion of a polarized complex Hodge structure. The bigrading is replaced with a meromorphic connection on the vector bundle on P1 considered above, having a pole of order at most two at 0 and ∞, and no other pole. The reason for restricting to poles of order two is explained in §2.a. A vector bundle on P1 with such a connection is called an integrable twistor structure (integrable because, when considering variations, it consists to adding an integrability condition to variations of twistor structures). It can also be described by linear algebra objects on H (cf. §3.a), that is, endomorphisms U , Q of H. However, variations of such objects produce non-trivial integrability conditions on these endomorphisms (cf. [10], [37, Chap. 7]). The endomorphism Q is self-adjoint with respect to the metric h and its eigenvalues play the role of the Hodge exponent p (more precisely, p − w/2) in H p,w−p. However, these eigenvalues may vary in variations of pure polarizable integrable twistor structure (that we now call polarized pure complex nc. Hodge structures). 1.5. The main object in this article is the notion of a polarized nc. k-Hodge structure, when k is a subfield of R, e.g. k = Q. When k = R, it corresponds to the notion of a pure polarized TERP structure as defined in [10], if one moreover takes care of a R-structure on the Stokes data of the connection, in case the singularity at 0 and ∞ is irregular. We will recall the well-known analysis that we need of connections with a pole of order at most two in §2. The k-structure is included at the level of generalized monodromy data, i.e., monodromy and (in the wild case) Stokes data. This way of treating the k-structure has been much generalized by T. Mochizuki in [31] for arbitrary holonomic D-modules. While one can define the notion of a pure nc. Hodge structure without introducing a polarization (cf. §3.6), proving that a given set of data form a pure nc. Hodge structure often uses the supplementary existence of a polarization. A simple, but nontrivial, example is given in §4, confirming a conjecture of C. Hertling and Ch. Sevenheck in [13], that they proved in some special cases. The main argument (cf. [12]) relies on a general way to produce polarized complex Hodge structures, which is given in 4 C. SABBAH [39]. Namely, given a variation of polarized complex Hodge structure on the complex punctured affine line (a finite number of points deleted), one can associate, by Fourier- Laplace transformation, a polarized complex nc. Hodge structure. This is reviewed in §5, where we also give some complements for pure polarized nc. k-Hodge structures. The same idea is used (cf. [39]) to prove that the Brieskorn lattice of a regular function on an affine manifold with isolated singularities and a tame behaviour of the fibres at infinity (e.g. a convenient and non-degenerate (Laurent) polynomial) underlies a rational nc. Hodge structure. Results of this kind are reviewed in §6. Meanwhile, we compare in §3.12 the notion of nc. Hodge structure with that of an exponential pure Hodge structure, introduced in [27]. 1.6. Lastly, we consider numerical invariants of nc. Hodge structures. There are various ways to replace the exponent p in H p,q (Hodge structure of weight w, so that q = w − p). These are called the spectral numbers at z = 0, the spectral numbers at z = ∞, and the supersymmetric index. The latter may vary in a real analytic way in variations of nc. Hodge structures, hence is difficult to compute in general, being of a very transcendental nature. On the other hand, the definition of spectral numbers at z = 0 goes back to Varchenko [55] and Steenbrink (cf. [54]) in Singularity theory, and the spectral numbers at z = ∞ were introduced in [35] (cf. also [38]). Many authors have considered these invariants for local or global singularities. For TERP(w) structures, they should remain constant when defining classifying spaces (cf. [14, 15, 16]), an important question when considering associated period mappings. 1.7. In order to keep reasonable length to this paper, we do not treat with details variations of nc. Hodge structures and their limits. We would like to emphasize, however, that the many-variable nilpotent orbit theorem of [3] and [21] has now a “wild twistor” counterpart [33], as well as a TERP counterpart [14], [32]. On the other hand, the Q-structure can also been considered, according to [31] (see also [42]). Acknowledgements. I thank Claus Hertling for reading a preliminary version of the manuscript and useful comments. The content of this survey article owes much to discussions and collaboration with him. I also thank Takuro Mochizuki, Christian Sevenheck and Jean-Baptiste Teyssier for their questions and suggestions, and for the various discussions we had. 2. Connections with a pole of order two 2.a. Filtered vector spaces with automorphism. We will regard connections with a pole of order (at most) two as a suitable generalization of the notion of a filtered complex vector space equipped with an automorphism in the following sense. Let V be a finite dimensional complex vector space, and let T be an automorphism of V . We denote by Ts its semi-simple part and by Tu its unipotent part, we set N = i 2π log Ts. By a filtration of (V, T) we will mean an exhaustive decreasing filtration F •V of V indexed by Z which is stable 2π log Tu and we choose a logarithm Ds = i NON-COMMUTATIVE HODGE STRUCTURES 5 by Ts and such that N(F pV ) ⊂ F p−1V for each p ∈ Z. In particular, (V, T, F •V ) decomposes according to the eigenvalue decomposition of (V, Ts). We also call (V, T) the associated Betti structure of (V, T, F •V ) and, if k is a subfield of C (e.g. k = Q, R or C), we say that the Betti structure is defined over k if (V, T) = C ⊗k (Vk, Tk). The objects (Vk, Tk, F •V ) obviously form a category, with a duality functor and a tensor product. We associate to these data a connection with a pole of order two, that is, the free C[z]-module RF V := Lp F pV z−p equipped with the connection ∇ = d + (Ds + N/z)dz/z. This connection has a pole of order two at z = 0, but has regular singularity there, and the monodromy is given by (V, T) (choose a semi-simple endomorphism H with half-integral eigenvalues which commutes with Ds and such that [H, N] = N, and apply the base change zH after a possible ramification of order two). In the following, we will restrict to the case where the eigenvalues of T have absolute value equal to 1. 2.b. Connections with a pole of order two (regular singularity case). By a connection with a pole of order two (H ,∇) we mean a free C{z}-module H of finite rank equipped with a connection ∇ having a pole of order at most two. We set G = C({z})⊗C{z}H (where C({z}) denotes the field C{z}[z−1] of convergent Laurent series in the variable z), equipped with the induced connection ∇. We assume in this subsection that (G ,∇) has a regular singularity. There exists then a canonical decreasing filtration V •G indexed by R, but with a finite number of jumps modulo Z, such that each V aG is a free C{z}-submodule of G of rank equal to dimC({z}) G , on which the connection has a pole of order at most one whose residue has eigenvalues with real part in [a, a + 1), and such that zkV aG = V a+kG for each k ∈ Z and each a ∈ R. In the following, we will assume that the eigenvalues of the residue are real. We will also denote by (H ,∇) the unique extension of the germ (H ,∇) as a free OU0 -module (cf. Notation 1.1) equipped with a meromorphic connection hav- ing a pole of order two at the origin and no other pole. If needed, we will use the unique C[z]-submodule of Γ(U0, H ) (Birkhoff extension) on which the connection has a regular singularity at infinity (by algebraizing the Deligne meromorphic exten- sion of (H ,∇) at infinity). Recall that the functor C({z})⊗C[z,z−1] (resp. C{z}⊗C[z]) induces an equivalence between the category of free C[z, z−1]-modules (resp. free C[z]- modules) with a connection ∇ having poles at 0 and ∞ only, the pole at ∞ being a regular singularity, and C({z})-vector spaces (resp. free C{z}-modules) with a con- nection. We will implicitly use this functor. 2.1. The Betti structure. By using the OU0 -module approach, we consider the local system L on C∗ defined as Ker∇C∗ , that we call the Betti structure of the connection with a pole of order two. Our previous assumption amounts to assuming that the eigenvalues of the monodromy of L have absolute value equal to 1. A k-structure of the connection with a pole of order two (H ,∇) consists by definition of a k-structure 6 C. SABBAH of the corresponding local system L . The objects (H ,∇, Lk) of connections with a pole of order two with a k-Betti structure obviously form a category with a duality functor and a tensor product. 2.2. V -gradation. The gradation functor replaces (G ,∇) with (grV we have set grV G =La∈R(V aG /V >aG ) and where grV ∇ is the naturally induced connection: for a fixed a ∈ R,Lk∈Z(V a+kG /V >(a+k)G ) is isomorphic to C[z, z−1]⊗C (V aG /V >aG ) and the connection is defined as d + (Resa ∇)dz/z, with Resa ∇ = a Id +Na with Na nilpotent. G , grV ∇), where That ∇ has a regular singularity implies that (G ,∇) ≃ (grV H ,∇) in a natural way, although it is not isomorphic to (H ,∇) in general. G , grV ∇). The con- nection with a pole of order two induces a graded connection with a pole of order two (grV Lemma 2.3. The functor (H ,∇) 7→ (V, T, F •V ), where T = La∈(−1,0] V = La∈(−1,0] exp(−2πi Resa ∇), gra V G , F pV = La∈(−1,0] gra V (zpH ), induces an equivalence between the subcategory of graded connections with a pole of order two and that of filtered vector spaces with an automorphism as in §2.a. This functor extends naturally to objects with a k-structure. Proof. For the first part, cf. e.g. [36, Chap. III]. Let us make precise that the gradation functor is the identity at the Betti level. This will imply that the k-structure on (G ,∇) induces a k-structure on (grV G , grV ∇). Recall that the functor of taking global multi-valued sections induces an equivalence between the category of local systems L to the category of vector spaces (L, T) with an automorphism. The composed functor (G ,∇) 7→ L 7→ (L, T) is identified with (G ,∇) 7−→ Ker[∇∂z : eO ⊗C({z}) G → eO ⊗C({z}) G ], where eO denotes the space of germs at the origin of multi-valued holomorphic func- tions on C∗, equipped with its natural monodromy operator. Let N−a ⊂ eO be the z−a(log z)ℓ/ℓ!, with its natural connection, and set N = La∈(−1,0] Ker[∇∂z : eO ⊗C({z}) G → eO ⊗C({z}) G ] is an isomorphism compatible with monodromy, space of linear combinations with coefficients in C({z}) of the multi-valued functions N−a. Then on the one hand, the natural inclusion of Ker[∇∂z : N ⊗C({z}) G → N ⊗C({z}) G ] into and on the other hand, the first term is canonically identified to (V, T) as defined in the lemma. Since the gradation functor is the identity at the (V, T) level, it remains the identity through the previous canonical functors at the (L, T) level, hence at the Betti level. 2.c. Connections with a pole of order two (nr. exponential case). We now relax the condition that (G ,∇) has a regular singularity. Since ∇ has a pole of order at most two on H , the connection has slopes 6 1 (cf. e.g. [28]) and we say, following [28, Chap. XII], that (G ,∇) has exponential type. We will moreover assume that the NON-COMMUTATIVE HODGE STRUCTURES 7 slopes are either 0 or 1. This is equivalent to the property that the Levelt-Turrittin decomposition of (bG ,b∇) := C((z))⊗C({z}) (G ,∇) needs no ramification, that is, (bG ,b∇) is isomorphic to a finite sum indexed by a finite subset C ⊂ C: (2.4) (bG ,b∇) ≃ Lc∈C (E −c/z ⊗ bGc), (2.5) where (bGc,∇c) is a connection with regular singularities and E −c/z ⊗ bGc is the C((z))- vector space bGc equipped with the connection ∇c +c Id dz/z2. It is known (cf. e.g. [36, Rem. II.5.8]) that, in such a case, ( cH ,b∇) decomposes accordingly, as (E −c/z ⊗ cHc), ( cH ,b∇) := C[[z]] ⊗C{z} (H ,∇) ≃ Lc∈C where (Hc,∇c) is a connection with a pole of order two with regular singularities as in §2.b. All over this article, nr. exponential type will be a shortcut for exponential type with no ramification. Example 2.6. If the leading matrix of ∇ in some C{z}-basis of H is semi-simple, then (H ,∇) has nr. exponential type. On the other hand, consider the connection with matrix with dz z , z + Id(cid:17)P (z)−1 · A(z)dz := P (z)(cid:16) Y  , P (z) = Id +zZ, Z = 0 0 0 1 0 0 0 1 0 Y = 0 0 1 0 0 0 0 0 0  . Then the matrix z2A(z) = P (Y + z Id)P −1 has characteristic polynomial equal to χ(λ) = (λ − z)3, but one can check that it is of exponential type but needs ramification (compare with [26, Rem. 2.13]). 2.7. The Betti structure (Stokes filtration). The local system L attached to (G ,∇) comes equipped with a family of pairs of subsheaves L<c ⊂ L6c for each c ∈ C, which satisfies the properties below (cf. [7], [28], [12], [42, Lect. 2]). For a fixed z ∈ C∗, define a partial order 6z on C compatible with addition by setting c 6z 0 iff c = 0 or Re(c/z) < 0 (and c <z 0 iff c 6= 0 and Re(c/z) < 0). This partial order on C only depends on z/z. The required properties are as follows. • For each z ∈ C∗, the germs L6c,z form an exhaustive increasing filtration of Lz, compatible with the order 6z. • The graded sheaves L6c/L<c are local systems on C∗. • For each z ∈ C∗, the germ L<c,z can be recovered asPc′<zc • The rank of Lc∈C L6c/L<c is equal to the rank of L , so that both local systems are locally isomorphic, and there is only a finite set C ⊂ C of jumping indices. L6c′,z. We will say that (L , L•) is a Stokes-filtered local system of nr. exponential type , or simply a Stokes-filtered local system, as we will only consider those of nr. exponential type in this article. A k-structure consists of a Stokes-filtered local system (Lk, Lk,•) 8 C. SABBAH such that (L , L•) = C ⊗k (Lk, Lk,•) (cf. [26, Def. 2.14] and [42, Prop. 2.26] for an equivalent definition). In particular, the monodromy of L is defined over k and, if k ⊂ R, this implies that Re Tr(Res ∇) ∈ 1 2 Z. On the other hand, recall that the Riemann-Hilbert functor (G ,∇) 7→ (L , L•) is an equivalence of categories (cf. [7] or [28, p. 58]) compatible with duality and tensor product. functor (L , L•) 7→Lc∈C The decomposition (2.4) or (2.5) is unique, and the formalization functor (G ,∇) corresponds, via the Riemann-Hilbert functor (G ,∇) 7→ (L , L•) to the Stokes grading L6c/L<c, so that the local system associated to Gc is grc L (cf. loc. cit.). As a consequence, a k-structure on (H ,∇) induces a k-structure on each (Hc,∇c). 2.8. The Betti structure (Stokes data). The previous description of the Betti structure is independent of any choice. On the other hand, the description with Stokes data below depends on some choices (cf. e.g. [12] for details). Let C be a non-empty finite subset of C. We say that θo ∈ R/2πZ is generic with respect to C if the set C is totally ordered with respect to 6z when z = eiθo. Once θo generic with respect to C is chosen, there is a unique numbering {c1, . . . , cn} of the set C in strictly increasing order. We will set θ′ o, so that −C is numbered as {−c1, . . . ,−cn} by θ′ o. of two families of k-vector spaces (Lc,1, Lc,2)c∈C and a diagram of morphisms The category of Stokes data of type (C, θo) defined over k has objects consisting o = θo + π. Note that the order is exactly reversed at θ′ L1 = Lci,1 nLi=1 S S′ L2 = Lci,2 nLi=1 (2.9) such that 1. S = (Sij )i,j=1,...,n is block-upper triangular, i.e., Sij : Lci,1 → Lcj,2 is zero unless i 6 j, and Sii is invertible (so dim Lci,1 = dim Lci,2, and S itself is invertible), 2. S′ = (S′ ij )i,j=1,...,n is block-lower triangular, i.e., S′ unless i > j, and S′ ii is invertible (so S′ itself is invertible). ij : Lci,1 → Lcj ,2 is zero A morphism of Stokes data of type (C, θo) consists of morphisms of k-vector spaces λc,ℓ : Lc,ℓ → L′ c,ℓ, c ∈ C, ℓ = 1, 2, which are compatible with the corresponding dia- grams (2.9). This allows one to classify Stokes data of type (C, θo) up to isomorphism. The monodromy T1 on L1 is defined by T1 = S−1S′. Grading the Stokes data means replacing (S, S′) with their block diagonal parts. There is a natural notion of tensor product in the category of Stokes data of type (C, θo), and a duality from Stokes data of type (C, θo) to Stokes data of type (−C, θo). by matrices (Σ, Σ′) where Σ = (Σij)i,j=1,...,n (resp. Σ′ = (Σ′ ii) is invertible. The matrix Σ−1 (resp. -upper) triangular and each Σii (resp. Σ′ ii Σ′ the matrix of monodromy of Lci,1, while Σ−1Σ′ is that of the monodromy of L1. Fixing bases in the spaces Lc,ℓ, c ∈ C, ℓ = 1, 2, allows one to present Stokes data ij )i,j=1,...,n) is block-lower ii is ) ) 5 5 NON-COMMUTATIVE HODGE STRUCTURES 9 Given θo generic with respect to C, there is an equivalence (depending on θo) between the category of Stokes filtered local systems (Lk, Lk,•) defined over k with jumping indices in C and that of Stokes data of type (C, θo) defined over k, which is compatible with grading, duality and tensor product (cf. e.g. [12, §2]). 2.d. Connections with a pole of order two of nr. exponential type obtained by Laplace transformation. 2.10. Inverse Laplace transformation. The condition for a meromorphic connection (G ,∇) to have only slopes 0 and 1 is equivalent to the property that (G ,∇) is obtained by a Laplace transformation procedure from a meromorphic connection with regular singularity. Let us make this precise. Recall that, by extending (G ,∇) as a free OC(∗0)-module with connection having a pole at z = 0 only, then choosing the Deligne extension as a OP1 (∗{0,∞})-module on which the connection has a regular singularity at infinity, and then taking global sections, we find a free C[z, z−1]-module with connection (G,∇). Then G is a left C[z′]h∂z′i-module (z′ := z−1). If we identify the ring C[z′]h∂z′i to C[t]h∂ti by the isomorphism z′ = ∂t, ∂z′ = −t, then G is a C[t]h∂ti- module that we denote by FG. The condition that (G ,∇) is of nr. exponential type is equivalent to the condition that FG has only regular singularities, at finite distance and at infinity, on the complex t-line (cf. e.g. [41, Lemma 1.5]). We call FG the inverse Laplace transform of (G,∇). Equivalently, (G,∇) is the Laplace transform of FG with kernel e−tz′ , and we use the notation G = F(FG). Note that the action of ∂t on FG is bijective. 2.11. Minimal extension and Brieskorn lattice. We will have to consider the minimal extension M of FG. By definition, this is the unique C[t]h∂ti-submodule of FG such that M has neither sub- nor quotient C[t]h∂ti-module supported on a point. The Laplace transform FM of M satisfies in turn C[z′, z′−1] ⊗C[z′] Any good filtration of M (in the sense of C[t]h∂ti-modules) gives rise to a vector bundle (H ,∇) in (G ,∇) whose connection has a pole of order two. We will not recall its construction here, cf. [39, §1.d], [41, §1.c] for details. It is called the Brieskorn lattice of the good filtration. FM = G. 2.12. Betti structure. We can pass from the Betti structure of M to that of FM by the topological Laplace transformation, and the way back by the inverse topological Laplace transformation (cf. [28]). Let V be the local system of horizontal sections of M away from its singularities C ⊂ A1. Assume that it is defined over k, that is, V = C ⊗k Vk. Let j : X = A1 r C ֒→ A1 denote the inclusion. The analytic de Rham complex DRan M on A1 has cohomology in degree zero only, equal to j∗V, hence has a natural k-structure given by j∗Vk. The topological Laplace transform of the k-perverse sheaf F := j∗Vk[1] is defined in [28, Chap. VI,§2] (cf. also [42, §7.d], and [26] for a different approach), as a per- verse sheaf on A1 z′ with a Stokes structure at infinity, that we denote by (FF , FF•). 10 C. SABBAH Forgetting the behaviour of this object near z′ = 0 (this corresponds to tensoring FM with C[z′, z′−1], that is, to considering G) allows one to describe it as a Stokes-filtered local system (Lk, Lk,•). One could also use classical integral formulas for the Stokes matrices (cf. [1]) to describe the k-structure. Notice that in [12, Prop. 4.7] one finds conversely the description of Vk in terms of (Lk, Lk,•). Let now QB be a nondegenerate pairing Vk ⊗ Vk → kX . It extends as a nonde- generate pairing j∗QB : j∗Vk ⊗ j∗Vk → kA1. By topological Laplace transformation, we get a pairing [j∗QB : (Lk, Lk,•) ⊗ ι−1(Lk, Lk,•) → kS. Its germ on Lk ⊗ ι−1Lk (that is, forgetting the Stokes filtration) at zo ∈ S is described as follows. LeteP1 be oriented real blow-up space of P1 at t = ∞. This is topologically a disc, obtained by adding a boundary S1 to A1, with coordinate (∞, eiθ). For each zo ∈ S, let Φ±zo denote the family of closed sets of A1 whose closure in eP1 does not cut the closed set {(∞, eiθ) Re(eiθ/ ± zo) > 0}. Then [j∗QBzo is the pairing induced by the cup product followed by QB: H 1 Φzo (A1, j∗Vk) ⊗ H 1 Φ−zo (A1, j∗Vk) −→ H 2 c (A1, Q) ≃ Q, where we remark that the intersection family Φzo ∩ Φ−zo is that of compact sets in A1. Proposition 2.13. The Laplace transform [j∗QB induces a nondegenerate pairing (Lk, Lk,•) ⊗ ι−1(Lk, Lk,•) → kS. Sketch of proof. This follows from Poincar´e duality if we forget the Stokes filtration, and we have to check that the Stokes filtration behaves correctly. Let us denote by D the Poincar´e-Verdier duality functor. Then one shows for any c ∈ C, the existence of isomorphisms (2.14) F(DF )<c ≃ ι−1D(FF6c), F(DF )6c ≃ ι−1D(FF<c) which are exchanged by duality up to bi-duality isomorphisms. In order to do so, one uses an integral formula for the topological Laplace transformation (cf. [28, §VI.2], see also [42, Lect. 7]). These isomorphisms are obtained by a local duality statement on the two-dimensional space of variables t, z′ suitably blown-up, and then by using the commutation (up to a suitable shift) of Poincar´e-Verdier duality with smooth pull-back and proper push-forward which enter in the definition of the topological Laplace transformation. 2.e. Deligne-Malgrange lattices. We explain here another example of pairs It can be obtained as the Brieskorn lattice of a (H ,∇) of nr. exponential type. natural filtration of M , namely the filtration by Deligne lattices. Let (G ,∇) be a meromorphic connection of nr. exponential type, that is, satisfying (2.4). The functor which associates to any lattice H of G (i.e., a C{z}-free submodule such that C({z}) ⊗C{z} H = G ) its formalization C[[z]] ⊗C{z} H is an equivalence between the full subcategory of lattices of G and that of lattices of bG (cf. [29]). In particular, let us consider for each c ∈ C and a ∈ R the Deligne lattices bGc a NON-COMMUTATIVE HODGE STRUCTURES 11 c has a simple pole and the real parts of the a >a) of the regular connection (bGc,b∇c) considered in (2.4), characterized by (resp. bGc the property that the connection b∇c on bG a eigenvalues of its residue belong to [a, a+1) (resp. to (a, a+1]). Clearly, bGc a+1 = z bGc >a+1 = z bGc and bGc According to the previous equivalence, there exist unique lattices of G , denoted by DMa(G ,∇) (resp. DM>a(G ,∇), which induce, by formalization, the decomposed lat- ticeLc E −c/z ⊗ bGc >a). They are called the Deligne-Malgrange lattices of (G ,∇). We regard them as defining a decreasing filtration of G . Lemma 2.15. Any morphism (G ,∇) → (G ′,∇′) of meromorphic connections of nr. exponential type is strictly compatible with the filtration by Deligne-Malgrange lattices. a (resp.Lc >a. E −c/z ⊗ bGc Sketch of proof. The associated formal morphism is block-diagonal with respect to the decomposition (2.4), and each diagonal block induces a morphism between the corresponding regular parts, which is known to be strict with respect to the filtration by the Deligne lattices. The behaviour by duality below is proved similarly by reducing to the regular singularity case (cf. e.g. [36, §III.1.b] or [38, Lem. 3.2]). Lemma 2.16. Let (G ,∇) be as above and let (G ,∇)∨ be the dual meromorphic connec- tion. Then, there are canonical isomorphisms [DMa(G ,∇)]∨ ≃ DM>−a−1[(G ,∇)∨], [DM>a(G ,∇)]∨ ≃ DM−a−1[(G ,∇)∨]. We will use this lemma as follows (ι∗, Notation 1.1, will be needed later). Corollary 2.17. Let (G ,∇) be of nr. exponential type, with associated Stokes structure (L , L•). Let QB : (L , L•) ⊗C ι−1(L , L•) → C be a nondegenerate bilinear pairing. Let Q : (G ,∇)⊗C({z}) ι∗(G ,∇) → (C({z}), d) be the nondegenerate pairing correspond- ing to QB via the Riemann-Hilbert correspondence. Then, for each a ∈ R, Q extends in a unique way as a nondegenerate pairing DMa(G ,∇) ⊗C{z} ι∗ DM>−a−1(G ,∇) → (C{z}, d). Corollary 2.18. With the assumptions of Corollary 2.17, assume moreover that a is an integer. Then, 1. if none of the monodromies of the bGc has 1 as an eigenvalue, then DMa(G ,∇) = DM>a(G ,∇) for each integer a, and Q induces a nondegenerate pairing DMa(G ,∇) ⊗C{z} ι∗ DMa(G ,∇) → (z2a+1C{z}, d), the bGc has −1 as an eigenvalue, then DMa−1/2(G ,∇) = DM>a−1/2(G ,∇) for each integer a, and Q induces a nonde- generate pairing DMa−1/2(G ,∇) ⊗C{z} ι∗ DMa−1/2(G ,∇) → (z2aC{z}, d). the monodromies of 2. if none of 12 C. SABBAH 3. Non-commutative Hodge structures In this section, we fix a subfield k of R, e.g. k = Q or R. The presentation given below owes much to various sources: [10, 11], [13, 15], [26], [30, 32] and [37]. 3.a. Non-commutative Hodge structures via linear algebra. 3.1. A reminder on Hodge structures. Let H be a finite dimensional complex vector space. Recall that a complex Hodge structure of weight w ∈ Z consists in a decompo- sition H =Lp∈Z H p,w−p. Equivalently, it consists of a semi-simple endomorphism Q of H with half-integral eigenvalues. The eigenspace of Q corresponding to the eigen- value p − w/2, p ∈ Z, is H p,w−p. The role of the weight only consists in fixing the bigrading. A real structure is a R-vector space HR such that H = C ⊗R HR, with respect to which H w−p,p = H p,w−p. Then the matrix of Q in any basis of HR is purely imaginary. On the other hand, a polarization of a complex Hodge structure is a nondegenerate (−1)w-Hermitian pairing k on H such that the decomposition is k-orthogonal and such that the Hermitian form h on H defined by hHp,w−p = ip−(w−p)kHp,w−p = i−w(−1)pkHp,w−p is positive definite, in other words, defining the Weil operator C by eπiQ, h = k(C •, •). For a real Hodge structure, the real polarization Q is then defined as the real part of k, and it is (−1)w-symmetric. 3.2. Complex nc. Hodge structures. By a complex nc. Hodge structure of weight w ∈ Z, we mean the data (H, U , U †, Q, w), where U , U †, Q are endomorphisms of H. When w is fixed, these data form a category, where morphisms are linear morphisms H → H ′ commuting with the endomorphisms U , U †, Q. For a complex Hodge struc- ture, we have U = U † = 0 and Q is as above. The category of complex nc. Hodge structures of weight w ∈ Z is abelian. Example. Assume U = U † = 0 and Q is semi-simple. One can decompose H = Lλ∈C∗ Hλ, where Hλ is the λ-eigenspace of e−2πiQ. Then each (Hλ, 0, 0, Q, w) is a Hodge structure of weight w and we can regard (H, 0, 0, Q, w) as a Hodge structure of weight w equipped with a semi-simple automorphism, with eigenvalue λ on Hλ. In order to understand various operations on complex nc. Hodge structures, we associate to (H, U , U †, Q, w) the C[z]-module H = C[z] ⊗C H, with the connection (3.2 ∗) This connection has a (possibly irregular) singularity at z = 0 and z = ∞, and no other singularity. Duality and tensor product are defined in a natural way, according to the rules for connections. Hence ∇ = d +(cid:0)z−1U − (Q + (w/2) Id) − zU †(cid:1) dz z . (H, U , U †, Q, w)∨ = (H ∨,−tU ,−tU †,−tQ,−w), NON-COMMUTATIVE HODGE STRUCTURES 13 1 , Q1, w1) ⊗ (H2, U2, U † and (H1, U1, U † 2 , Q2, w2) has weight w1 + w2 and the endo- morphisms are defined by formulas like U1⊗Id2 + Id1 ⊗U2. The involution ι : z 7→ −z induces a functor ι∗, with ι∗(H, U , U †, Q, w) = (H,−U ,−U †, Q, w). 3.3. Real nc. Hodge structures. The complex conjugate of the complex nc. Hodge structure is defined as (H, U , U †, Q, w) := (H, U †, U ,−Q, w), where H is the R-vector space H together with the conjugate complex structure. A real structure κ on (H, U , U †, Q, w) is an isomorphism from it to its conjugate, such that κ ◦ κ = Id. A real structure consists therefore in giving a real structure HR on H, with respect to which U † = U and Q + Q = 0. We denote such a structure as (HR, U , Q, w). Morphisms are R-linear morphisms compatible with U and Q. Real nc. Hodge structures (HR, U , Q, w) satisfy properties similar to that of complex nc. Hodge structures and we have similar operations defined in a natural way. 3.4. Polarization of a complex nc. Hodge structure. A polarization of (H, U , U †, Q, w) is a nondegenerate Hermitian form h on H such that • h is positive definite, • U † is the h-adjoint of U and Q is self-adjoint with respect to h. It is useful here to introduce the complex Tate object TC(ℓ) defined as (C, 0, 0, 0,−2ℓ) for ℓ ∈ Z, corresponding to the Hodge structure C−ℓ,−ℓ. The Tate twist by TC(ℓ) is simply denoted by (ℓ). The last condition is equivalent to asking that h defines an isomorphism (H, U , U †, Q, w) ∼−→ ι∗(H, U , U †, Q, w)∨(−w). The tensor product (H1, U1, U † 1 , Q1, h1, w1) ⊗ (H2, U2, U † 2 , Q2, h2, w2) of polarized complex nc. Hodge structures is defined by the supplementary relation h = h1 ⊗ h2, and is also polarized. 3.5. Polarization of a real nc. Hodge structure and the Betti structure. Although the notion of a real nc. Hodge structure seems to be defined over R, the real vector space HR does not contain all the possible “real” information on the structure, in cases more general than that of a Hodge structure. The Weil operator is not defined in this setting. The formula C = eπiQ exhibits the Weil operator as a square root of the monodromy of the connection d − Qdz/z. This suggests that the monodromy of the connection (3.2 ∗) should be taken into account in order to properly define the notion of a real nc. Hodge structure, and further, that of a nc. k-Hodge structure. Even further, if ∇ has an irregular singularity, the Betti real structure is encoded in the Stokes data attached to the connection, not only in the monodromy, as ex- plained in §2.7, together with the notion of a k-Betti structure, for any subfield k of R (e.g. k = Q). 14 C. SABBAH Here is another drawback of the presentation of a complex nc. Hodge structure as a vector space with endomorphisms: the notion of a variation of such objects is not defined in a holomorphic way, exactly as H p,w−p do not vary holomorphically in classical Hodge theory. Good variations are characterized by the property of the Hermitian metric to be harmonic in the sense of [50], and the endomorphisms U , Q satisfy relations encoded in the notion of a CV structure [10]. The correct generalization of the Hodge filtration is that of a vector bundle on the complex affine line together with a meromorphic connection. This motivates the definition of a nc. Hodge structure in §3.b by taking integrable twistor structures (i.e., vector bundles on U0 with a meromorphic connection having a pole of order two at zero and no other pole, plus gluing data with the “twistor conjugate” γ∗H or “twistor adjoint” σ∗H ∨ object) as the starting point. 3.b. Non-commutative Hodge structures via integrable twistor structures. 3.6. Non-commutative k-Hodge structures. Let (H ,∇) be a connection with a pole of order two of rank d, and let L be the corresponding local system on S. A real structure LR on L allows one to produce a holomorphic vector bundle with connection ( fH ,e∇) on P1, by gluing (H ,∇) (chart U0) with γ∗(H ,∇) (chart U∞, cf. Notation 1.1) uniquely through the flat gluing isomorphism g : (H ,∇)U0∩U∞ defined as follows (cf. [10]): ∼−→ γ∗(H ,∇)U0∩U∞ • g is uniquely defined from g∇ : H ∇ • g∇ is uniquely defined from its restriction g∇ U0∩U∞ ∼−→ γ−1H ∇ U0∩U∞ , S : L → L to S (recall that γS = Id), S is defined to be the isomorphism induced by the real structure on L . • g∇ We say that (H ,∇, LR) is pure of weight w if the bundle fH is isomorphic to OP1(w)d. Notice that, for k ∈ Z and for any λ ∈ C, (zkH ,∇, λLR) is then pure of weight w − 2k. Definition (cf. [10, Rem. 2.13], [26]). Let (H ,∇, (Lk, Lk,•)) be a connection with a pole of order two and k-Betti structure. We say that it is a pure nc. k-Hodge structure of weight w if the underlying triple (H ,∇, LR), with LR := R ⊗k Lk, is pure of weight w (if k = R and forgetting the Stokes filtration Lk,•, we recover the notion of TER structure of [10, Rem. 2.13]). the standard differential on H ⊗C C[z, z−1], that we restrict to the C[z]-submodule Example. Let Hk be a k-vector space and set H = C ⊗k Hk. A k-Hodge structure of weight w consists of the data (Hk, F •H) such that the filtration F •H and its conjugate are w-opposed. Set (H ,∇) = (RF H, d) := (Lp F pHz−p, d) where d is Lp∈Z F pHz−p (we work here with the algebraic version of H ). The local system LR is the constant local system HR and γ∗(H ,∇) =Lq F qz′−q. The condition that (H ,∇, LR) is pure of weight w is equivalent to the w-opposedness property. Indeed, let us show one direction for instance. If both filtrations F • and F • are w-opposed, NON-COMMUTATIVE HODGE STRUCTURES 15 we have a bigrading H = Lp H p,w−p with H q,p = H p,q and F • defined as usual. Then we have decompositions into finite sums (H p,w−p ⊗ z−pC[z]), (H q,w−q ⊗ z′−qC[z′]) =Lp RF H =Lp γ∗RF H =Lq (H p,w−p ⊗ z′p−wC[z′]), and the gluing morphism is the identity C[z, z−1] ⊗C[z] RF H = H[z, z−1] = H[z′, z′−1] = C[z′, z′−1] ⊗C[z′] γ∗RF H. For each p, the line bundle defined by the gluing data (z−pC[z], z′p−wC[z′], Id) is nothing but OP1(w). Example (The Tate object TQ(ℓ) with ℓ ∈ 1 2 Z). For non-commutative Hodge struc- tures, we can take the opportunity of having monodromy 6= Id to define the Tate nc. k-Hodge structure Tk(ℓ) for ℓ ∈ 1 (3.6 ∗) where (2πi/z)ℓkS denotes the rank-one local system on S generated by the (possibly) multivalued function (2πi/z)ℓ. It has monodromy equal to (−1)2ℓ Id. By the residue theorem, TQ(ℓ) has weight −2ℓ. Tk(ℓ) = (OU0 , d + ℓdz/z, (2πi/z)ℓkS), 2 Z. We set If ℓ ∈ Z, then we have an isomorphism Tk(ℓ) (3.6 ∗∗) where we regard zℓOU0 as included in OU0 (1/z), with the induced differential d, and the latter object corresponds to the object constructed in the previous example for the Tate Hodge structure k(ℓ), with Hk = (2πi)ℓk and H = H −ℓ,−ℓ. zℓ −−−→ (zℓOU0 , d, (2πi)ℓkS), ∼−→ (H ,∇)∨ 3.7. Polarized complex nc. Hodge structures. Let now (H ,∇) be a connection with a pole of order two equipped with a nondegenerate ι-Hermitian pairing C∇ S : L ⊗C ι−1L → CS. Recalling that γS = Id, we regard C∇ S as a nondegenerate σ-Hermitian pairing L ⊗σ−1L → CS, and we extend it in a unique way as a nondegenerate pairing C∇ : H ∇ U0∩U∞ → CU0∩U∞ . This pairing in turn defines in a unique way a flat isomorphism C : σ∗(H ,∇)U0∩U∞ . By gluing with this isomorphism the dual bundle H ∨ (chart U0) with the σ-conjugate bundle σ∗(H ,∇) U0∩U∞ ⊗ σ−1H ∇ compatible with the connections (it is induced by Id on each chart) and which is σ-Hermitian. We say that (H ,∇, C) is a pure complex nc. Hodge structure (of weight (chart U∞), we obtain a bundle cH on P1, which has degree zero by the residue formula. Since C is σ-Hermitian, we obtain a natural morphism S : cH → σ∗ cH ∨ 0) if cH is the trivial bundle on P1. global sections. Let us set H = Γ(P1, cH ). Identifying in a natural way Γ(P1, σ∗ cH ) with Γ(P1, cH ) = H, the isomorphism h := Γ(P1, S ) : H → H ∨ is a nondegenerate Hermitian pairing h : H⊗H → C. We then say that (H ,∇, C) is a complex nc. Hodge structure, pure (of weight 0) and polarized if h is positive definite. If (H ,∇, C) is pure of weight 0, the isomorphism S induces an isomorphism on U0∩U∞ 16 C. SABBAH Remark. The following criterion can be used for the purity and polarizability: (H ,∇, C) is pure (of weight 0) and polarized if and only if there exists a OU0 -basis ε of H such that the matrix C(ε, σ∗ε) is the identity (cf. [37, Rem. 2.2.3]). In this way, one checks that if (H ,∇, C) is pure (of weight 0) and polarized, then so is (zkH ,∇, (−1)kC) for every k ∈ Z. Example. Let H = Lp∈Z H p,w−p be a grading of H, and let h be a positive defi- nite Hermitian pairing on H such that the decomposition is h-orthogonal. Let k be the (−1)w-Hermitian pairing on H such that the decomposition is k-orthogonal and h(•, •) = k(C •, •), where C is the standard Weil operator ip−q Id on H p,q. Define the Hodge filtration F •H as usual, and (H ,∇) as in Example of §3.6. In the al- S := i−wk by gebraic setting, HU0∩U∞ = H ⊗C C[z, z−1]. Let C be defined from C∇ sesquilinearity, that is, C(Pp vpz−p, σ∗Pq wqz−q) :=Pp,q i−wk(vp, wq)z−p(−1/z)−q ∈ C[z, z−1]. For vp ∈ H p,w−p and wq ∈ H q,w−q, we have C(vpz−p, σ∗wqz−q) = 0 unless p = q, in which case it is equal to h(vp, wp), showing that a h-orthonormal basis ofL H p,w−p induces a basis of H which is C-orthonormal in the sense of the previous remark, so (H ,∇, C) is pure (of weight 0) and polarized. 3.8. Polarized nc. k-Hodge structures. For any integer w, we regard z−wOU0 as con- tained in OU0 [z−1] and we consider on it the connection induced by d, that we still denote by d. The corresponding local system is the constant sheaf C on U0∩ U∞, and we endow it with the usual Q-structure (hence k-structure). The Stokes filtration is the trivial one. We now consider a connection with a pole of order two (H ,∇) with k-Betti struc- ture (Lk, Lk,•) together with a nondegenerate (−1)w-ι-symmetric pairing (3.8 ∗) (Q, QB) : ((H ,∇), (Lk, Lk,•)) ⊗ (ι∗(H ,∇), ι−1(Lk, Lk,•)) −→ ((z−wOU0 , d), kS). Here we mean that QB induces, by the Riemann-Hilbert correspondence, a nondegen- erate (−1)w-ι-symmetric pairing Q on (G ,∇) with values in OU0 (∗0), whose restric- tion to H takes values in z−wOU0 and is non-degenerate as such (and automatically (−1)w-ι-symmetric). We will write QB instead of (Q, QB) when this causes no confu- sion. On the one hand, (H ,∇, LR) defines a vector bundle fH . On the other hand, QB S := i−wQ∇ induces a (−1)w-ι-symmetric pairing Q∇ S , made sesquilinear according to the isomorphism κ : L → L induced by the real structure, defines a ι-Hermitian nondegenerate pairing C∇ S : L ⊗C ι−1L → C, with S : LR⊗Cι−1LR → R, and thus P∇ S (•, ι−1 •) := P∇ C∇ S (•, ι−1κ(•)) = i−wQ∇ S (•, ι−1κ(•)). (3.8 ∗∗) Extending C∇ vector bundle cH on P1. S as a flat σ-Hermitian pairing C on HU0∩U∞, we define as above a NON-COMMUTATIVE HODGE STRUCTURES 17 Remarks. • The data (H ,∇, LR, P) as above (forgetting the Stokes structure, if any, and setting k = R) is called a TERP(−w)-structure in [10]. • Let us set S = (z/2πi)wQ. Then we can regard S as a morphism of nc. k-Hodge structures with values in Tk(−w). Lemma (cf. [10, Th. 2.19]). The pairing Q induces an isomorphism (ι∗H ∨ ∼−→ ι∗ cH ⊗ OP1(w). fH Proof. We have an isomorphism between the bundle fH defined by the gluing data ∼−→ γ∗H U0∩U∞ ) and the bundle defined by the gluing U0 ⊗ z−wOU0 , (ι∗σ∗H )U∞ , ι∗C), (HU0 , (γ∗H )U∞ , κ : HU0∩U∞ data U0 , (ι∗σ∗H )U∞ , ι∗C), hence the assertion. which is given by P on U0 and Id on U∞ (since γ = σ ◦ ι). On the other hand, ι∗ cH is defined by the gluing data (ι∗H ∨ Definition. A connection (H ,∇) with a pole of order two equipped with a k-Betti structure (Lk, Lk,•) and a nondegenerate (−1)w-ι-symmetric k-pairing QB as in (3.8 ∗) is called a pure and polarized nc. k-Hodge structure of weight w if the asso- ciated triple (H ,∇, C) is a pure complex nc. Hodge structure (of weight 0) which is polarized (cf. §3.7). A consequence of the lemma above is that, if ((H ,∇), (Lk, Lk,•), QB) is a pure and polarized nc. k-Hodge structure of weight w, then it is a pure nc. k-Hodge structure of weight w (as defined in §3.6). Notice also that ((zkH ,∇), (Lk, Lk,•), QB) is then pure and polarized of weight w − 2k. Example. Let (Hk, F •H) be a pure k-Hodge structure of weight w and let QB : Hk⊗ Hk → k be a polarization, in particular a (−1)w-symmetric nondegenerate pair- ing. Recall also that Q(H p,q, H p′,q′ ) = 0 for p′ 6= w − p, so Q(F pHz−p, F qHz−q) ⊂ z−wC[z]. Let Hk,S be the constant local system on S. We have a canonical iden- tification Hk,S = ι−1Hk,S. Then QB induces a (−1)w-ι-symmetric nondegenerate pairing QB on Hk,S in a natural way. The associated pairing Q : RF H ⊗C[z] ι∗RF H → (z−wC[z], d, k) is the pairing generated over C[z]⊗ι∗C[z] by Q. Then (H ,∇, Hk,S, Q) is a pure and polarized nc. k-Hodge structure of weight w. Indeed, for each p, let εp be a basis of H p,w−p which is orthonormal for Q(C •, •), where C is the Weil operator. Recall that RF H =Lp H p,w−pz−pC[z]. The family (εpz−p)p is a C[z]-basis of RF H. We have, by definition and because σ∗z = −1/z, C(εpz−p, σ∗εqz−q) = i−wQ(εp, εq)z−pσ∗z−q =(0 Q(Cεp, εp) if p 6= q, if p = q, so the family (εpz−p)p is an orthonormal basis for C, hence the polarization property, according to the remark in §3.7. 18 C. SABBAH Example (Polarization of Tk(ℓ) for ℓ ∈ Z, cf. [32]). We will use the expression (3.6 ∗∗) and the previous example. We have a natural isomorphism ι∗Tk(ℓ) = Tk(ℓ) and a natural nondegenerate symmetric pairing S : Tk(ℓ) ⊗ Tk(ℓ) −→ Tk(2ℓ) induced by the natural product pairing kS ⊗ kS → kS. With this in mind, QB : (2πi)ℓkS ⊗ (2πi)ℓkS → kS is defined as (2πi)−2ℓ times the product. For ℓ ∈ 1 2 Z, the polarization will be indicated in Remark 4.4 below. 3.9. Duality and tensor products. Duality and tensor product are well defined both for connections with a pole of order two and for k-Stokes-filtered local systems, and are compatible via the Riemann-Hilbert correspondence (cf. [28], cf. also [42]). These functors preserve the category of nc. k-Hodge structures, since they correspond to the corresponding functors for the bundles ( fH ,e∇), and thus transform weights as expected. Similarly, polarization behaves in a natural way. For instance, for ℓ ∈ Z, the Tate twist ((H ,∇), (Lk, Lk,•), QB)(ℓ) := ((H ,∇), (Lk, Lk,•), QB) ⊗ Tk(ℓ) consists in replacing (H ,∇) with (zℓH ,∇), (Lk, Lk,•) with (2πi)ℓ(Lk, Lk,•) and QB with (2πi)−2ℓQB. 3.10. Comparison with the linear algebra approach of §3.a. We now forget about the Stokes structure. Tensoring (H ,∇, Lk) with Tk(ℓ) (ℓ ∈ 1 2 Z) allows one to only treat nc. k-Hodge structures of weight 0. A similar reduction can be done for the case of complex nc. Hodge structures, as well as for that of polarized nc. k-Hodge structures (by using Remark 4.4 below). In the linear algebra approach of §3.a, the corresponding reduction simply amounts to setting w = 0 in the objects. We will also assume that k = R, since only this case can occur in §3.a. Lemma (cf. [10, Th. 2.19]). The functor which associates to (H ,∇, LR) pure of weight 0 (resp. to a pure complex nc. Hodge structure (H ,∇, C) of weight 0, resp. to a pure polarized TERP structure (H ,∇, LR, P) of weight 0) the object (Γ(P1, fH ),e∇) (resp. (Γ(P1, cH ),b∇), resp. ...) is an equivalence with the corresponding linear algebra data (H,∇) with a connection (3.2 ∗). 3.11. Rescaling. C. Hertling [10, Th. 7.20] has considered the action of C∗ on the category of connections (H ,∇) with a pole of order two obtained by rescaling the variable z. For x ∈ C∗, consider the map µx : U0 → U0 defined by µx(z) = xz. The rescaled connection is µ∗ x (S) = {z xz = 1} 6= S if x 6= 1, we define the pull-back local system µ−1 L by working with local systems on U0 ∩ U∞ (recall that the inclusion S ֒→ U0 ∩ U∞ induces an equivalence of categories of local systems on the corresponding spaces). This approach is taken up by C. Hertling and Ch. Sevenheck in [13]. x(H ,∇). Since µ−1 x NON-COMMUTATIVE HODGE STRUCTURES 19 The rescaling acts on the category of objects (H ,∇, LR) (by the same procedure as above), on the category of objects (H ,∇, C) since ι commutes with µx, and sim- ilarly on the category of objects (H ,∇, Lk, QB) as in §3 (without paying attention to the nc. Hodge property at the moment). It also acts on Stokes-filtered local sys- tems (Lk, Lk,•) in a way compatible, by the Riemann-Hilbert correspondence, to the action on the meromorphic bundles (G ,∇): the subsheaf (µ−1 L )k,6c is defined as µ−1 x (Lk,6c/x). If (H ,∇, LR) is pure of weight w (resp. if (H ,∇, C) is pure of weight 0 and polarized, resp. if (H ,∇, Lk, QB) is pure of weight w and polarized) then, provided x − 1 is small enough, the corresponding rescaled object remains pure (resp. pure and polarized) of the same weight: this follows from the rigidity of trivial bundles on P1. On the other hand, this may not remain true for all values of the rescaling parameter x. x The subcategory of pure polarized nc. Hodge structures which remain so by rescal- ing by any x ∈ C∗ is a global analogue of a nilpotent orbit in the theory of variation of polarized Hodge structures. It has been extensively studied in [13, 15] and [32]. On the other hand, on this subcategory, the no ramification condition for the connec- tion is a consequence of an expected nice behaviour at the limiting values x = 0,∞ (this is discussed in [41, App. B]). This would lead to the definition of the category of rescalable polarizable pure nc. Hodge structures, which would be particular cases of variations of polarizable pure nc. Hodge structures on C∗ with a wild (resp. tame) behaviour at x = 0 (resp. x = ∞). 3.12. Exponential Hodge structures. In a similar spirit, Kontsevich and Soibelman [27] develop the notion of exponential mixed Hodge structure. Given a holonomic C[t]h∂ti-module M , the C[t]h∂ti-module C[t]h∂t, ∂−1 i⊗C[t]h∂ti M is still holonomic and its Laplace transform is equal to the localization C[τ, τ −1] ⊗C[τ ] FM . If M is regular holonomic, the corresponding operation on its de Rham complex, via the Riemann- Hilbert correspondence, is the convolution with j!QA1r{0}, where j : A1 r {0} ֒→ A1 denotes the inclusion, as defined in [24]. The convolution with j!QA1r{0} can be done within the frame of mixed Hodge modules defined in [46]. Then, given a pure Hodge module on A1, its convolution with j!QA1r{0} is mixed, but is declared to be “exponentially pure” of the same weight. Theorem 5.6 below and the remark which follows imply that the Fourier-Laplace transform of an exponential mixed Hodge structure (i.e., a mixed Hodge D-module on A1 whose de Rham hypercohomology spaces are zero) is the localization at τ = 0 of a rescalable polarized pure nc. Hodge structure. t 4. Non-commutative Hodge structures from Deligne-Malgrange lattices 4.1. The basic construction. Let C ⊂ C be a finite set, let θo ∈ R/2πZ be generic with respect to C (cf. §2.8) defining thus a numbering {c1, . . . , cn} of the set C in strictly increasing order, and let Σ be a block-lower triangular invertible square matrix of 20 C. SABBAH size d with entries in k ⊂ R, the blocks being indexed by C ordered by θo. Under some assumptions on Σ, we will associate to these data and to each integer w a connection with a pole of order two (H ,∇) with k-Betti structure (Lk, Lk,•) and a nondegenerate (−1)w-ι-symmetric nondegenerate k-pairing Q as in §3.8, giving rise in particular to a TERP(−w)-structure. We will denote these data by ncH(C, θo, Σ, w). The matrix Σ determines Stokes data ((Lc,1, Lc,2), S, S′) of type (C, θo) (cf. (2.9)) by setting L1 = L2 = kd, and Lc,j (c ∈ C, j = 1, 2) correspond to the blocks of Σ, which defines a linear morphism S : L1 → L2, and we define S′ as the linear morphism attached to Σ′ := (−1)w· tΣ. These Stokes data in turn correspond to a Stokes filtered local system (Lk, Lk,•). The underlying local system Lk is completely determined by the k-vector space L1 = kd together with (the conjugacy class of) its monodromy, whose matrix is (−1)wΣ−1· tΣ. On the other hand, each diagonal block Σci of Σ gives · tΣci, which represents the monodromy of the rise to an invertible matrix (−1)wΣ−1 ci meromorphic connection corresponding to bGci in the decomposition (2.4). A nondegenerate ι-pairing QB on (Lk, Lk,•) with values in k is determined by a pair of nondegenerate pairings Q12 : L1,k ⊗ L2,k → k and Q21 : L2,k ⊗ L1,k → k which satisfy Q21(x2, x1) = Q12(S−1x2, S′x1), and the (−1)w-ι-symmetry amounts to Q21(x2, x1) = (−1)wQ12(x1, x2) (cf. [12, (3.3) & (3.4)]). In the fixed bases of L1 and L2, we define(1) Q21(x2, x1) = tx2 · x1, so that Q12(x1, x2) = tx1 tΣ · Σ′−1x2; the (−1)w-symmetry follows from Σ′ = (−1)w · tΣ. From the Riemann-Hilbert correspon- dence we finally obtain a nondegenerate (−1)w-ι-symmetric pairing (4.1 ∗) QB : ((G ,∇), (Lk, Lk,•)) ⊗ ι∗((G ,∇), (Lk, Lk,•)) −→ (C({z}), d, kS). We will set (using the notation of §2.e, and stressing upon the fact that the con- struction of (G ,∇) and (L , L•) above depends on the parity of w): (4.1 ∗∗) Let us note that Corollary 2.18 reads as follows: ncH(C, θo, Σ, w) = (DM−(w+1)/2(G ,∇), (Lk, Lk,•), QB). Corollary 4.2. Let (C, θo, Σ, w) be as above. Assume that Ker(Σci + tΣci) = 0 for all i. Then DM>−(w+1)/2 = DM−(w+1)/2 and the pairing QB given by (4.1 ∗) induces a nondegenerate (−1)w-ι-symmetric pairing, also denoted by QB: DM−(w+1)/2(G ,∇) ⊗C{z} ι∗ DM−(w+1)/2(G ,∇) −→ (z−wC{z}, d). Theorem 4.3 (cf. [12]). Let (C, θo, Σ, w) be as in §4.1. We moreover assume the fol- lowing: 1. for each c∈ C, the diagonal block Σc of Σ satisfies Ker(Σc + tΣc) = 0, 2. the quadratic form Σ + tΣ is positive semi-definite. Then ncH(C, θo, Σ, w) is a pure and polarized nc. k-Hodge structure of weight w. (1)Compared with the definition of h 21 in [12, §3c & 3.e], which is the case where w is odd, we changed the sign in order to have a better correspondence with the k-structure and the example of §3.8. 12 and h NON-COMMUTATIVE HODGE STRUCTURES 21 Proof. The condition 4.3(1) implies that the assumption of Corollary 4.2 holds. The case where w = −1 is contained in the statement of [12, Th. 5.9], where QB above corresponds to −hB there (cf. Footnote (1)), and where the assumption 4.3(1) implies Kc = 0 for each c ∈ C in [12, (5.9)(∗∗)]. The proof of [12, Th. 5.9] consists in ex- pressing ncH(C, θo, Σ,−1) as the Fourier-Laplace transform of a variation of complex polarized Hodge structure of type (0, 0) on C r C, and to use Theorem 5.5 explained in §5 below. In order to treat the general case, we remark that (cf. (4.1 ∗∗)): ncH(C, θo, Σ, w − k) = ncH(C, θo, Σ, w) ⊗ ncH(C = {0}, θo = 0, Σ = 1, w = −k), for each k ∈ Z. According to §3.9, it is thus enough to prove that the rank-one object ncH(C = {0}, θo = 0, Σ = 1, w = −k) is pure and polarized of weight −k and, by iterating tensor products, it is enough to consider k = 1. Let us compute the polarization Q(1/2) for the object (OU0 , d+(1/2)dz/z, z−1/2kS), which is seen to correspond to C = {0}, θo = 0, w = −1. Then L1 = k · e1 is the space of sections of Lk on (0, π) and L2 = k · e2 on (−π, 0). Moreover, Q(1/2) (e1, e2) = −1. The sections exp(iθ/2)e1 ∈ OU0(0,π) and exp(iθ/2)e2 ∈ OU0(−π,0) are the restriction of the OU0 -basis vo = 1 ∈ OU0 : indeed, vo restricted to U0 r {z ∈ R} is written z1/2ei, i = 1, 2. We have P(1/2),∇ = iQ(1/2) and C(1/2) is the σ-sesquilinear extension of P(1/2). The map z 7→ σ(z) = −1/z restricts to R/2πZ as (e2, e1) = 1 and Q(1/2) 12 21 (0, π) −→ (0, π) θ 7−→ −(θ − π) (−π, 0) −→ (−π, 0) θ 7−→ −(θ + π) We then have C(1/2) S (vo, σ∗vo) =(eiθ/2 · e−i(θ−π)/2 · i · Q(1/2) eiθ/2 · e−i(θ+π)/2 · i · Q(1/2) 12 21 (e1, e2) = 1 (e2, e1) = 1 for θ ∈ (0, π), for θ ∈ (−π, 0). This shows that (vo, σ∗vo) is an orthonormal basis for C(1/2), hence (cf. the remark in §3.7) the corresponding (OU0 , d + 1 Remark 4.4. By rotating the Q-structure, the last argument can be used similarly to show that the Tate pure nc. k-Hodge structure Tk(ℓ) of weight −2ℓ is polarized, and make explicit its polarization. 2 dz/z, C(1/2)) is pure of weight 0 and polarized. 5. Non-commutative k-Hodge structures by Fourier-Laplace transformation 5.1. Let C ⊂ A1 be a finite set of points on the complex affine line with coordinate t. Let (Vk, F •V,∇, QB) be a variation of polarized Hodge structure of weight w ∈ Z on X := A1 r C. Namely, • (V,∇) is a holomorphic vector bundle with connection on X, 22 C. SABBAH • F •V is a finite decreasing filtration of V by holomorphic sub-bundles satisfying the Griffiths transversality property: ∇F pV ⊂ F p−1V ⊗OX Ω1 X , • Vk is a k-local system on X with Vk ⊗k C = V ∇, • QB : Vk ⊗k Vk → k is a nondegenerate (−1)w-symmetric pairing, all these data being such that the restriction at each x ∈ X is a polarized Hodge structure of weight w (cf. e.g. [34] or the example in §3.8). We denote by Q the non- degenerate pairing (V,∇) ⊗ (V,∇) → OX that we get from QB through the canonical isomorphism OX ⊗k Vk = V . The associated nondegenerate sesquilinear pairing is denoted by k : (V,∇) ⊗C (V,∇) → C ∞ X , which is obtained from kB : V ⊗C V → C similarly. It is (−1)w-Hermitian and i−wk induces a flat Hermitian pairing on the X ⊗OX V,∇ + ∂). We can regard (V,∇, F •V, i−wk) as a variation of C∞-bundle (C ∞ polarized complex Hodge structure, pure of weight 0. 5.2. The middle (or minimal) extension. This procedure transforms the previous data, defined on A1 r C, into similar data defined on A1. Let j : X ֒→ A1 denote the inclusion. Betti side: The pairing QB (resp. kB) extends in a unique way as a nondegener- ate (−1)w-symmetric (resp. (−1)w-Hermitian) pairing j∗QB : j∗Vk ⊗k j∗Vk → kA1 (resp. j∗kB : j∗VC ⊗C j∗VC → CA1 ). De Rham side: The bundle (V,∇) can be extended in a unique way as a free OP1(∗C ∪ {∞})-module with a connection ∇ having a regular singularity at C ∪ {∞} (Deligne meromorphic extension). Taking global sections on P1 produces a left mod- and which has no quotient submodule supported in C. The pairing k extends first ule fM on the Weyl algebra C[t]h∂ti. The minimal extension (along C) of fM is the unique submodule M of fM which coincides with fM after tensoring both by C(t), (due to the regularity of the connection) as a pairing ek : fM ⊗C fM → S ′(A1 r C), and S ′(A1 r C) := C(cid:2)t,Qc∈C(t − c)−1(cid:3) ⊗C[t] S ′(A1). Then one shows that, when restricted to M ⊗C M ,ek takes values in S ′(A1), and we denote it by k (cf. [12, §1.a], where no distinction is made between h and k since we only deal there with variations of Hodge structures of type (p, p) for some p). where S ′(A1) denotes the Schwartz space of temperate distributions on A1 = R2, z′ ) with kernel exp(tz′ − tz′) i : S ′(A1 Hodge side: The Hodge filtration F •V extends, according to a procedure due to M. Saito [43, §3.2], to a good filtration F •M of M as a C[t]h∂ti-module (cf. [39, §3.d]). 5.3. Fourier-Laplace transformation. The Fourier transformation Ft t) → S ′(A1 2π dt ∧ dt is an isomorphism from the Schwartz space S ′(A1 z′ ) considered as a C[z′]h∂z′i ⊗C C[z′]h∂z ′i-module. z′ ), where FM is the Laplace transform of M as in §2.10, and where ι+ denotes the pull- back by ι in the sense of C[z′]h∂z′i-modules (or C[z′]-modules with connection). See [39, §1.a] for the need of ι+. Composing k with Ft defines a sesquilinear pairing Fk : FM ⊗C ι+FM → S ′(A1 t) considered as a C[t]h∂ti ⊗C C[t]h∂ti-module, to S ′(A1 NON-COMMUTATIVE HODGE STRUCTURES 23 Restricting to C∗ produces a sesquilinear pairing Fk : (G ,∇)⊗ ι∗(G ,∇) → (C ∞ C∗ , d), S : L ⊗ ι−1L → CS (we use whose horizontal part restricted to S defines a pairing C∇ here the notation of §2.10). We then define C as in §3.7. The pairing Fk restricts to horizontal sections of (G ,∇) to produce a Betti ι-sesquilinear pairing (Fk)B on L . It is defined only over C∗. On the other hand, in a way similar to Proposition 2.13, there is a topological Fourier-Laplace transform [j∗kB, which is compatible with the Stokes filtration. The comparison between both is given by: Lemma 5.4 (cf. [39, Prop. 1.18] & [12, Appendix]). Over C∗ we have (Fk)B = i 2π dj∗kB. Lastly, we denote by H ⊂ G the Brieskorn lattice of the good filtration F •M (cf. §2.11), and we recall that the connection has a pole of order two at most on H . Theorem 5.5 ([39, Cor. 3.15]). Let (Vk, F •V,∇, QB) be a variation of polarized Hodge structure of weight w ∈ Z on X := A1 r C. Then (H ,∇, i−wC) defined as above is a pure polarized complex nc. Hodge structure of weight 0. We now make more precise Theorem 5.5, which only produces a polarized complex nc. Hodge structure, in order to get a polarized nc. k-Hodge structure. Recall that the pairing [j∗QB has been considered in Proposition 2.13. We notice that the topological Laplace transform [j∗kB is nothing but the ι-sesquilinear pairing associated with the ι-pairing [j∗QB on Lk. Theorem 5.6. Let (Vk, F •V,∇, QB) be a variation of polarized k-Hodge structure of weight w ∈ Z on X := A1 r C. Then ((H ,∇), (Lk, Lk,•),− [j∗QB) is a pure polarized nc. k-Hodge structure of weight w + 1. Remark. One can show (see [39, Rem. 2.5]) that the pure polarized nc. k-Hodge struc- tures that one gets by Fourier-Laplace transformation are rescalable, in the sense given in §3.11. Remark. In order to go from Theorem 5.5 to Theorem 5.6, we need some more work on bilinear pairings (while Lemma 5.4 and Theorem 5.5 only use sesquilinear pairings at the topological or analytical level). The pair (Q, QB) of (3.8 ∗) that we wish to use consists of QB = − [j∗QB and of a constant multiple of the Laplace transform of the algebraic duality isomorphism of the C[t]h∂ti-module M associated with (V,∇) as in §5.2. We first check that this algebraic duality is compatible with filtrations (Lemma 5.7), hence satisfies the holomorphic part of (3.8 ∗), by using properties of Hodge D- modules (Corollary 5.8). We then show that the pairing corresponding to − [j∗QB by the Riemann-Hilbert correspondence essentially coincides with this algebraic pairing (Lemma 5.9). Moreover, notice that Lemma 5.4 only holds over C∗, while Lemma 5.9 holds including at z = 0. In particular, Lemma 5.9 and Lemma 5.4 are not of the same nature and are independent one from the other. 24 C. SABBAH Proof of Theorem 5.6. Let (Vk, F •V,∇, QB) be a variation of polarized Hodge struc- ture of weight w on X. According to [43, Th. 2], it corresponds to a polarized Hodge module which is pure of weight w + 1. It extends in a unique way as a polarized Hodge module on A1, with underlying filtered C[t]h∂ti-module (M, F •M ) as con- sidered in §5.2 above. The polarization induces a (−1)w+1-symmetric isomorphism Q : (M, F •M ) ∼−→ D(M, F •M )(−(w + 1)) (cf. loc. cit. for the duality of filtered D-modules, and Lemma 5.2.12 there for the polarization, which is denoted by S′ and corresponds to (2πi)−wQ), and where, for a filtered module (N, F •N ), we set (N, F •N )(k) = (N, F •+kN ). It is easy to check, from the very definition of the Brieskorn lattice of a good filtration (cf. [39, §1.d]), that the Brieskorn lattice of (N, F •N )(k) is equal to zkH , where H is the Brieskorn lattice of (N, F •N ). On the other hand, according to the relation between duality and Laplace trans- forms of C[t]h∂ti-modules (cf. [28, Lem. 3.6, p. 86], see also [36, §V.2.b]), we have F(DM ) = ι+DFM (where i+ denotes the pull-back of C[z′]h∂z′i-modules and D is the duality of holonomic C[t]h∂ti-modules), and thus, according to [45, §2.7] (cf. also [38, Lem. 3.8]), the localized Laplace transform of DM is identified with ι∗G ∨ (cf. Lemma 2.16 for the notation). Given a holonomic C[t]h∂ti-module with a good filtration (M, F •M ), we say that D(M, F •M ) is strict (cf. [43]) if the dual complex DRF M (as RF C[t]h∂ti-modules, cf. §A.1) has cohomology in degree one only, without C[z]-torsion. This cohomology can then be written in a unique way as RF DM for some good filtration F •DM . For a polarizable Hodge module, D(M, F •M ) is strict (cf. [43]), and the polariza- tion gives a filtered isomorphism between M and DM as above. The proof of the following lemma will be sketched in Appendix A.a. Lemma 5.7. Assume that D(M, F •M ) is strict. Then, through the previous identifi- cation of the localized Laplace transform of DM with ι∗G ∨, the Brieskorn lattice of (DM, F •DM ) is identified with ι∗H ∨. Corollary 5.8. The morphism FQ induces an isomorphism (G ,∇) ∼−→ ι∗(G ,∇)∨ which sends H onto z−(w+1)ι∗H ∨, and thus induces a nondegenerate (−1)w+1-ι-symmetric pairing Q on H . by [j∗QB through the inverse RH correspondence. The proof of the following lemma will be sketched in Appendix A.b. Let us now denote by dj∗Q the nondegenerate pairing (G ,∇) ∼−→ ι∗(G ,∇)∨ induced Lemma 5.9. The pairings FQ and dj∗Q coincide up to a multiplicative constant. Let us set QB = −[j∗QB. Then, according to Lemma 5.9 and Corollary 5.8, (Q, QB) S := S = −i−(w+1)[j∗kB = 2π · i−w(Fk)B, satisfies (3.8 ∗) with w+1 instead of w. For the polarizability property, let us set C∇ i−(w+1)Q∇ according to Lemma 5.4. Hence Theorem 5.5 gives the polarizability. S made sesquilinear. We thus have C∇ NON-COMMUTATIVE HODGE STRUCTURES 25 6. The nc. Q-Hodge structure attached to a tame function 6.1. Let X be a complex smooth quasi-projective variety and let f : X → A1 be a regular function on it, that we regard as a morphism to the affine line A1 with coordinate t. For each k ∈ Z, the perverse cohomology sheaf pH k(Rf∗QX ) underlies a mixed Hodge module (cf. [46]). The fibre at z = 1 of its topological Laplace transform (cf. [28, Chap. VI,§2]) is the k-th exponential cohomology space of X with respect to f (or simply of (X, f )). algebraic differential form of degree k on X and γ is a k-cycle in the Borel-Moore homology of X, such that Re f > ε > 0 on the support of γ and away from a compact set in X. The exponential periods attached to f are the integrals Rγ e−f ω, where ω is an Formulas likeRR e−x2 dx = √π suggest (cf. [8, p. 118]) to produce a “Hodge filtra- tion” with rational or real indices (here 1/2) on the exponential cohomology of (X, f ). This is developed in loc. cit. for particular examples. On the other hand, when f is proper, pH k(Rf∗QX ) ⋆ j!QA1r{0} is exponentially pure in the sense of Kontsevich-Soibelman [27] (cf. §3.12), but so is also the case when all the graded object, except one, with respect to the weight filtration are constant, hence killed by the convolution operation. The cohomologically tame case considered below enters this frame. nential cohomology space of (X, f ) as Γ(P1, cH ), where cH is defined in §3.7. The non-commutative Hodge structure approach that we explain below consists, when pH k(Rf∗QX ) ⋆ j!QA1r{0} is exponentially pure, in considering the k-th expo- In particular, the non-commutative Hodge structure is defined on H . The relation with the construction of Deligne quoted above is explained in [41, §6]. 6.2. For the sake of simplicity, we will only consider the case of a cohomologically tame function f : U → A1 on a smooth affine complex manifold U , for which there is only one non-zero exponential cohomology space. We will constantly refer to [35, 38] and [39]. Recall (cf. loc. cit.) that cohomological tameness implies that there exists a dia- gram κ / X U @ @ f @ @ @ @ @ F A1 where X is quasi-projective and F is projective, such that the cone of natural mor- phism κ!QU → Rκ∗QU has no vanishing cycle with respect to F − c for any c ∈ C (cf. also [23, Th. 14.13.3]). We will use the perverse shift convention by setting pQU = Q[dim U ]. By Poincar´e- Verdier duality, we have a natural pairing QB : Rf! pQU ⊗Q Rf∗ pQU −→ QA1[2]. / 26 C. SABBAH Considering the Q-perverse sheaf F = pH 0(Rf∗ pQU ), we therefore get a morphism DF → F , whose kernel and cokernel (in the perverse sense) are constant sheaves up to a shift. Let (LQ, LQ,•) be the Stokes-filtered local system on S deduced from the topological Laplace transform of F (cf. §2.12). According to Lemma 2.13, it comes equipped with a nondegenerate pairing On the other hand, let G0 denote the Brieskorn lattice of f . By definition, QB := −[j∗QB : (LQ, LQ,•) ⊗ ι−1(LQ, LQ,•) −→ QS. G0 = Ωdim U (U )[z](cid:14)(zd − df∧)Ωdim U−1(U )[z], and set G = C[z, z−1] ⊗C[z] G0, with the action of ∇∂z induced by ∂z + f /z2 = ef /z ◦ ∂z ◦ e−f /z on Ωdim U (U )[z]. We also set Gk = z−kG0. For ℓ ∈ Z we set ε(ℓ) = (−1)ℓ(ℓ−1)/2. Theorem 6.3. The data ((Gdim U ,∇), (LQ, LQ,•), ε(dim U − 1)QB) is a polarized nc. Q-Hodge structure which is pure of weight dim U . Sketch of proof. We refer to [39, Proof of Th. 4.10]. We first replace the perverse sheaf F defined above with F!∗ := pH 0(RF∗κ!∗ pQU ), which generically is the local system of intersection cohomology of the fibres of F , and we have a corresponding Poincar´e-Verdier duality pairing QB,!∗, whose topological Laplace transform − \j∗QB,!∗ coincides with QB. By applying M. Saito’s results on polarizable Hodge D-modules, together with Theorem 5.6, we find that ((GH 0 ,∇), (LQ, LQ,•), ε(dim U − 1)QB) is a pure polarized nc. Q-Hodge structure of weight dim U , where GH 0 is the Brieskorn lat- tice of the Hodge filtration of the Hodge module corresponding to F!∗. By [39, Lem. 4.7], taking also into account the shift between the standard filtration and M. Saito’s Hodge filtration, we have GH 0 = Gdim U . Corollary 6.4. The data ((G0,∇), (LQ, LQ,•), ε(dim U − 1)QB) is a pure polarized nc. Q-Hodge structure of weight − dim U , and the corresponding P can be written as i− dim U ε(dim U )[j∗QB. 7. Numerical invariants of nc. Hodge structures 7.1. Spectrum at z = ∞. To any germ (H ,∇) consisting of a free C{z}-module of finite rank equipped with a meromorphic connection (with pole of arbitrary order at z = 0) for which the eigenvalues of the monodromy have absolute value equal to one, is attached a numerical invariant called its spectrum at infinity (cf. [36, §III.2.b] or [41, §1.a]), and encoded as a polynomial Sp∞ in R (and more precisely, e2πiγ is an eigenvalue of the monodromy), and νγ ∈ N. Its behaviour by duality is described in [36, Prop. III.2.7], and the behaviour with respect to tensor product is better described in terms of the divisor of the polynomial Sp∞ H (T ) = Qγ(T − γ)νγ , where γ varies H (T ), an element of the ring Z[R] which is sometimes written as P νγuγ. Then NON-COMMUTATIVE HODGE STRUCTURES 27 H (T ). It has properties similar to that of Sp∞ this divisor behaves in a multiplicative way with respect to tensor product, provided that both terms of the tensor product are essentially self-dual (cf. [36, Ex. III.2.9]). 7.2. Spectrum at z = 0. Assume now that (H ,∇) is of nr. exponential type, so has a formal decomposition (2.5). Assume also that the eigenvalues of the monodromy of each Hi have absolute value equal to one. Then each (Hi,∇i) has a spectrum at the origin, defined with the help of the V -filtration (cf. §2.b) at z = 0. We use the convention of [41, Def. 1.7]. The product of the spectral polynomials Sp0 Hi(T ) for all i is denoted Sp0 H (T ) with respect to various operations (cf. [36, §III.1.c] and [41, 1.b]). 7.3. The “new supersymmetric index” (cf. [10]). Let T = (H ,∇, C) be a pure com- plex nc. Hodge structure of weight 0 (cf. §3.7) and let Q be the “new supersym- Its character- metric index” associated to it through the correspondence of §3.10. istic polynomial will be denoted by SusyT (T ). If (H ,∇, C) is polarized, then Q is self-adjoint with respect to the corresponding positive definite Hermitian form h, hence is semi-simple with real eigenvalues, so the roots of SusyT (T ) are real. If ((H ,∇), (LQ, LQ,•), QB) is a polarized nc. Hodge structure which is pure of weight w, then the “new supersymmetric index” of the corresponding pure complex nc. Hodge structure of weight 0 is purely imaginary, hence its (real) eigenvalues are symmetric with respect to the origin. Example (cf. [41, Lemma 5.4]). For a polarized Hodge structure of weight w and Hodge . On the other hand, numbers hp,w−p, we have SusyT (T ) =Qp(T − p + w/2)hp,w−p . Sp∞ H (T ) = Sp0 H (T ) =Qp(T − p)hp,w−p Rescaling. This relationship between Sp0, Sp∞ and Susy can be generalized by con- sidering the action of the rescaling (cf. §3.11), when the nc. Hodge structure is stable by rescaling, and has a good limiting behaviour, as in the case of a Fourier-Laplace transform of a variation of Hodge structure. More precisely, one gets: Theorem 7.4 (cf. [41, Th. 7.1]). Let ((H ,∇), (LQ, LQ,•), QB) be the nc. Q-Hodge struc- ture obtained by Fourier-Laplace transformation from a variation of polarized Hodge structure of weight w on A1 r C (cf. §5) and let T be the associated polarized complex nc. Hodge structure of weight 0. Then Sp0 H (T ) = lim x→0 H (T ) = lim x→∞ Sp∞ Susyµ∗ x Susyµ∗ T (T − w/2), T (T − w/2). x 7.5. Limit theorems. The previous theorem is proved by showing that the rescaled objects µ∗ x((H ,∇), (LQ, LQ,•), QB) form a variation of nc. Q-Hodge structures parametrized by x ∈ C∗, and µ∗ T extends as a pure polarized wild twistor D-module (cf. [40]) on the projective completion P1 ⊃ C∗ (wildness only occurs at x = 0). One proves limit theorems in such a setting (cf. [41]), showing that the limiting twistor structure, when x → ∞, is a mixed Hodge structure polarized by x 28 C. SABBAH a nilpotent endomorphism, giving the first equality, and, when x → 0, this limit- ing structure decomposes as the direct sum of exponentially twisted mixed Hodge structures polarized by a nilpotent endomorphism, giving the second equality. A more general approach to these limit theorems, more in the spirit of Schmid’s nilpotent orbit theorem, but in both tame and wild cases, and with many param- eters, has been obtained by T. Mochizuki in [32], proving thereby a conjecture of C. Hertling and Ch. Sevenheck [13, Conj. 9.2] on nilpotent orbits of pure polarized TERP structures. Appendix increasing and decreasing filtrations). The Rees module RF M = Lℓ FℓM zℓ is a A.a. Proof of Lemma 5.7. We first make more precise the notion of Brieskorn lattice of a filtered C[t]h∂ti-module, in order to manipulate it more easily. We mainly refer to [39, §2.d]. A.1. The Brieskorn lattice. Let (M, F•M ) be a holonomic C[t]h∂ti-module equipped with a good filtration (here we use the increasing version of a filtration, in order to be compatible with [39, §2.d]; recall the standard convention Fp = F −p relating module of finite type over the Rees ring RF C[t]h∂ti, that we identify with the ring C[t, z]hðti by setting ðt = z∂t. The Laplace transform F(RF M ) is the C[z]-module RF M equipped with the structure of a C[τ, z]hðτi-module, where τ acts as ðt and ðτ as −t. It is also of finite type. Moreover, RF M is also equipped with an action of z2∂z (i.e., is a C[t, z]hðt, z2∂zi-module), which is the natural one on RF M , defined as z2∂z(mℓ ⊗ zℓ) = ℓmℓ⊗ zℓ+1. The action of z2∂z on F(RF M ) (i.e., its C[τ, z]hðτ , z2∂zi- F(mℓ ⊗ zℓ) = (∂tt + ℓ)mℓ ⊗ zℓ+1 (cf. [39, Rem. 2.2]). The structure) is twisted as z2∂z Brieskorn lattice G(F•) of (M, F•M ) is the restriction to τ = 1 of F(RF M ), with the induced C[z]hz2∂zi structure. Moreover, let ε : C[τ, z]hðτi → C[τ, z]hðτi denote the involution τ 7→ −τ , ðτ 7→ −ðτ . The restriction at τ = 1 of ε∗F(RF M ) is equal to the restriction at τ = −1 of F(RF M ), and the formulas given in [39, Lem. 2.1 & Rem. 2.2] identify it with ι∗G(F•) . A.2. Duality and Laplace transformation. Exactly as in the case of C[t]h∂ti-modules (cf. [28, Lem. V.3.6], see also [36, §V.2.b]), the relation between duality and Laplace transformation of RF C[t]h∂ti-modules is given by DF(RF M ) = ε∗F(DRF M ). Since D(M, F•M ) is strict, i.e., DRF M has cohomology in degree one only, without C[z]- torsion, then F(DRF M ) is also strict, hence so is DF(RF M ). Moreover, the unique cohomology of F(DRF M ) is F(RF DM ), and we denote by DF(RF M ) the unique coho- mology of DF(RF M ), so DF(RF M ) = ε∗F(RF DM ). Let us now localize with respect to τ as in [39, Lem. 2.1], i.e., apply C[τ, τ −1, z]⊗C[τ,z] to both terms, which then become C[τ, τ −1, z]-free of finite type. After loc. cit., the right-hand term is identi- fied with C[τ, τ −1]⊗ ι∗H ′, where H ′ denotes the Brieskorn lattice of (DM, F•DM ). On the other hand, arguing as in [45, §2.7] (cf. also [38, Lem. 3.8]), the left-hand 0 0 NON-COMMUTATIVE HODGE STRUCTURES 29 term is identified with the dual module (C[τ, τ −1, z] ⊗C[τ,z] RF M )∨ with its natural connection. Restricting to τ = 1 gives H ∨. Therefore, ι∗H ′ = H ∨, as wanted. A.b. Sketch of proof of Lemma 5.9. This kind of comparison result goes back at least to the notion of higher residue pairings, due to K. Saito, in the theory of singularities of complex hypersurfaces. One finds in [45, §2.7] a similar result, proved by using a universal unfolding of a holomorphic function having an isolated singularity. Another geometric approach, in the present setting, has been proposed by C. Hertling (unpublished notes) following the geometric construction, due to F. Pham, of the intersection form on Lefschetz thimbles (cf. [9, Th. 10.28] for the analogous result in Singularity theory). We will sketch a sheaf-theoretic proof, by following the Riemann- Hilbert correspondence all along the Fourier-Laplace transformation. t, A1 t × A1 z′ → A1 A.3. The Laplace transform FM of M can be obtained by an “integral formula” z′, and FM =bp+(p+M ⊗ E−tz′ E−tz′ A.4. The duality isomorphism F(DM ) ∼−→ ι+D FM mentioned in the proof of Theo- ), where p,bp denote the projections A1 = (C[t, z′], d − d(tz′)). rem 5.6 can be obtained by applying standard isomorphisms of commutation between the duality functor and the proper direct image functor one the one hand, the smooth inverse image functor on the other hand, in the realm of D-module theory (cf. e.g. [2, VII.9.6 & VII.9.13] or [22, Th. 4.33 & Th. 4.12]). In order to do so, it is convenient to extend the previous setting to P1 a way, the duality isomorphism at the level of D-modules and the one at the level of Stokes-filtered local systems are constructed in parallel ways. z′ in order to work with a proper mapbp. In such t × P1 A.5. The correspondence between the duality isomorphism for proper maps, in both settings, is proved in [44]. A similar statement for pull-back by a smooth morphism can be proved similarly. t × C∗ z′ . On the one hand, we have D(p+M ⊗ E−tz′ A.6. This reduces the problem to a comparison between both isomorphisms on t × P1 P1 z′ . For our purpose, one can show that it is enough to compare both isomor- ) ≃ p+(DM )⊗ Etz′ phisms over P1 (the change of sign explaining the need of ι). On the other hand, let Z = P1 z′ and z′. If F is the perverse sheaf t via the de Rham functor, the perverse sheaf on Z associated via the de Rham functor can be written as R∗β!Rα∗(p◦ j)−1F [1], : eZ → Z be the oriented real blowing-up along {∞}×C∗ associated to M on A1 to p+M ⊗ E−tz′ where, setting t × C∗ L′ 60 = {(t, z′) t = 0 or arg t + arg z′ ∈ [π/2, 3π/2] mod 2π} ⊂ eZ, α, β are the inclusions α t × C∗ A1 −−→ L′ z′ ֒ and j = ◦ β ◦ α is the inclusion A1 t × C∗ z′ ֒→ Z. The question is then reduced to a local duality theorem, comparing the duality isomorphism for D-modules and the Poincar´e-Verdier duality for perverse sheaves. −−→ eZ. 60 ֒ β 30 C. SABBAH References [1] W. Balser, W.B. Jurkat & D.A. Lutz – “On the reduction of connection problems for differential equations with an irregular singular point to ones with only regular singularities. I”, SIAM J. Math. Anal. 12 (1981), no. 5, p. 691–721. [2] A. Borel – “Chap. VI-IX”, in Algebraic D-modules, Perspectives in Math., vol. 2, Aca- demic Press, Boston, 1987, p. 207–352. [3] E. Cattani, A. Kaplan & W. Schmid – “Degeneration of Hodge structures”, Ann. of Math. 123 (1986), p. 457–535. [4] S. Cecotti, P. Fendley, K. Intriligator & C. Vafa – “A new supersymmetric index”, Nuclear Phys. B 386 (1992), p. 405–452. [5] S. Cecotti & C. Vafa – “Topological-antitopological fusion”, Nuclear Phys. B 367 (1991), p. 359–461. [6] , “On classification of N = 2 supersymmetric theories”, Comm. Math. Phys. 158 (1993), p. 569–644. [7] P. Deligne – “Lettre `a B. Malgrange du 19/4/1978”, in Singularit´es irr´eguli`eres, Cor- respondance et documents, Documents math´ematiques, vol. 5, Soci´et´e Math´ematique de France, Paris, 2007, p. 25–26. [8] , “Th´eorie de Hodge irr´eguli`ere (mars 1984 & aout 2006)”, in Singularit´es irr´eguli`eres, Correspondance et documents, Documents math´ematiques, vol. 5, Soci´et´e Math´ematique de France, Paris, 2007, p. 109–114 & 115–128. [9] C. Hertling – Frobenius manifolds and moduli spaces for singularities, Cambridge Tracts in Mathematics, vol. 151, Cambridge University Press, 2002. [10] , “tt∗ geometry, Frobenius manifolds, their connections, and the construction for singularities”, J. reine angew. Math. 555 (2003), p. 77–161. [11] , “tt∗ geometry and mixed Hodge structures”, in Singularity theory and its ap- plications, Adv. Stud. Pure Math., vol. 43, Math. Soc. Japan, Tokyo, 2006, p. 73–84. [12] C. Hertling & C. Sabbah – “Examples of non-commutative Hodge structures”, Jour- nal de l’Institut math´ematique de Jussieu 10 (2011), no. 3, p. 635–674. [13] C. Hertling & Ch. Sevenheck – “Nilpotent orbits of a generalization of Hodge structures”, J. reine angew. Math. 609 (2007), p. 23–80, arXiv: math.AG/0603564. [14] , “Curvature of classifying spaces for Brieskorn lattices”, J. Geom. Phys. 58 (2008), no. 11, p. 1591–1606, arXiv: 0712.3691. [15] , “Twistor structures, tt-geometry and singularity theory”, in From Hodge the- ory to integrability and TQFT: tt-geometry (R. Donagi & K. Wendland, eds.), Proc. Sym- posia in Pure Math., vol. 78, American Mathematical Society, Providence, RI, 2008, p. 49– 73, arXiv: 0807.2199. [16] , “Limits of families of Brieskorn lattices and compactified classifying spaces”, Adv. in Math. 223 (2010), p. 1155–1224, arXiv: 0805.4777. [17] H. Iritani – “An integral structure in quantum cohomology and mirror symmetry for toric orbifolds”, Adv. in Math. 222 (2009), no. 3, p. 1016–1079, arXiv: 0903.1463v3. , “tt-geometry in quantum cohomology”, arXiv: 0906.1307, 2009. [18] [19] D. Kaledin – “Non-commutative Hodge-to-de Rham degeneration via the method of Deligne-Illusie”, Pure Appl. Math. Q. 4 (2008), no. 3, Special issue in honor of F. Bogo- molov, Part 2, p. 785–875. [20] , “Motivic structures in non-commutative geometry”, arXiv: 1003.3210; to ap- pear in Proc. ICM, 2010. [21] M. Kashiwara – “The asymptotic behaviour of a variation of polarized Hodge struc- ture”, Publ. RIMS, Kyoto Univ. 21 (1985), p. 853–875. NON-COMMUTATIVE HODGE STRUCTURES 31 [22] , D-modules and microlocal calculus, Translations of Mathematical Monographs, vol. 217, American Mathematical Society, Providence, RI, 2003. [23] N. Katz – Exponential sums and differential equations, Ann. of Math. studies, vol. 124, Princeton University Press, Princeton, NJ, 1990. [24] , Rigid local systems, Ann. of Math. studies, vol. 139, Princeton University Press, Princeton, NJ, 1996. [25] N. Katz & G. Laumon – “Transformation de Fourier et majoration de sommes expo- nentielles”, Publ. Math. Inst. Hautes ´Etudes Sci. 62 (1985), p. 145–202. [26] L. Katzarkov, M. Kontsevich & T. Pantev – “Hodge theoretic aspects of mirror symmetry”, in From Hodge theory to integrability and TQFT: tt-geometry (R. Donagi & K. Wendland, eds.), Proc. Symposia in Pure Math., vol. 78, American Mathematical Society, Providence, RI, 2008, arXiv: 0806.0107, p. 87–174. [27] M. Kontsevich & Y. Soibelman – “Cohomological Hall algebra, exponential Hodge structures and motivic Donaldson-Thomas invariants”, arXiv: 1006.2706, 2010. [28] B. Malgrange – ´Equations diff´erentielles `a coefficients polynomiaux, Progress in Math., vol. 96, Birkhauser, Basel, Boston, 1991. [29] , “Connexions m´eromorphes, II: le r´eseau canonique”, Invent. Math. 124 (1996), p. 367–387. [30] T. Mochizuki – Asymptotic behaviour of tame harmonic bundles and an application to pure twistor D-modules, vol. 185, Mem. Amer. Math. Soc., no. 869-870, American Mathe- matical Society, Providence, RI, 2007, arXiv: math.DG/0312230 & math.DG/0402122. [31] [32] , “Holonomic D-modules with Betti structure”, arXiv: 1001.2336, 2010. , “Asymptotic behaviour of variation of pure polarized TERP structures”, Publ. RIMS, Kyoto Univ. 47 (2011), no. 2, p. 419–534, arXiv: 0811.1384. [33] , Wild harmonic bundles and wild pure twistor D-modules, Ast´erisque, Soci´et´e Math´ematique de France, Paris, 2011, to appear; arXiv: 0803.1344. [34] C. Peters & J.H.M. Steenbrink – Mixed Hodge Structures, Ergebnisse der Mathe- matik und ihrer Grenzgebiete. 3. Folge, vol. 52, Springer-Verlag, Berlin, 2008. [35] C. Sabbah – “Hypergeometric period for a tame polynomial”, C. R. Acad. Sci. Paris S´er. I Math. 328 (1999), p. 603–608, & math.AG/9805077. , D´eformations isomonodromiques et vari´et´es de Frobenius, Savoirs Actuels, [36] CNRS ´Editions & EDP Sciences, Paris, 2002, English Transl.: Universitext, Springer & EDP Sciences, 2007. [37] , Polarizable twistor D-modules, Ast´erisque, vol. 300, Soci´et´e Math´ematique de France, Paris, 2005. [38] , “Hypergeometric periods for a tame polynomial”, Portugal. Math. 63 (2006), no. 2, p. 173–226, arXiv: math.AG/9805077. [39] , “Fourier-Laplace transform of a variation of polarized complex Hodge struc- ture”, J. reine angew. Math. 621 (2008), p. 123–158, arXiv: math.AG/0508551. [40] , “Wild twistor D-modules”, in Algebraic Analysis and Around: In Honor of Professor Masaki Kashiwara’s 60th Birthday (Kyoto, June 2007), Advanced Studies in Pure Math., vol. 54, Math. Soc. Japan, Tokyo, 2009, p. 293–353, arXiv: 0803.0287. [41] , “Fourier-Laplace transform of a variation of polarized complex Hodge structure, II”, in New developments in Algebraic Geometry, Integrable Systems and Mirror symmetry (Kyoto, January 2008), Advanced Studies in Pure Math., vol. 59, Math. Soc. Japan, Tokyo, 2010, p. 289–347, arXiv: 0804.4328. [42] , “Introduction to Stokes structures”, Lecture Notes (Lisboa, January 2009) 200 pages, arXiv: 0912.2762, 2010. 32 C. SABBAH [43] M. Saito – “Modules de Hodge polarisables”, Publ. RIMS, Kyoto Univ. 24 (1988), p. 849–995. [44] , “Induced D-modules and differential complexes”, Bull. Soc. math. France 117 (1989), p. 361–387. [45] , “On the structure of Brieskorn lattices”, Ann. Inst. Fourier (Grenoble) 39 (1989), p. 27–72. [46] [47] D. Shklyarov – “Non-commutative hodge structures: towards matching categorical , “Mixed Hodge Modules”, Publ. RIMS, Kyoto Univ. 26 (1990), p. 221–333. and geometric examples”, arXiv: 1107.3156, 2011. [48] C. Simpson – “Harmonic bundles on noncompact curves”, J. Amer. Math. Soc. 3 (1990), p. 713–770. [49] , “Nonabelian Hodge theory”, in Proceedings of the International Congress of Mathematicians, (Kyoto, 1990), Math. Soc. Japan, Tokyo, 1991, p. 747–756. [50] , “Higgs bundles and local systems”, Publ. Math. Inst. Hautes ´Etudes Sci. 75 (1992), p. 5–95. [51] , “Mixed twistor structures”, Pr´epublication Universit´e de Toulouse & arXiv: math.AG/9705006, 1997. [52] , “The Hodge filtration on nonabelian cohomology”, in Algebraic geometry (Santa Cruz, 1995), Proc. of AMS summer conferences, American Mathematical Society, 1997, p. 217–281. [53] , “Algebraic aspects of higher nonabelian Hodge theory”, in Motives, polylog- arithms and Hodge theory, Part II (Irvine, CA, 1998), Int. Press Lect. Ser., vol. 3, Int. Press, Somerville, MA, 2002, p. 417–604. [54] J.H.M. Steenbrink – “The spectrum of hypersurface singularities”, in Th´eorie de Hodge (Luminy, 1987), Ast´erisque, vol. 179-180, Soci´et´e Math´ematique de France, 1989, p. 163–184. [55] A.N. Varchenko – “Asymptotic Hodge structure on the cohomology of the Milnor fiber”, Math. USSR Izv. 18 (1982), p. 469–512. C. Sabbah, UMR 7640 du CNRS, Centre de Math´ematiques Laurent Schwartz, ´Ecole polytechnique, F–91128 Palaiseau cedex, France Url : http://www.math.polytechnique.fr/sabbah • E-mail : [email protected]
1303.0115	2	1303	2013-12-24T07:47:06	Bruhat strata for Shimura varieties of PEL type	[ "math.AG", "math.NT" ]	The Bruhat stratification for Shimura varieties of PEL type is studied. In the Siegel case this stratification is a scheme-theoretic variant of the stratification by the a-number. We show that all Bruhat strata are smooth and determine their dimensions. We also prove that the closure of a Bruhat stratum is a union of Bruhat strata and describe which Bruhat strata are contained in the closure of a given Bruhat stratum.	math.AG	math	Bruhat strata for Shimura varieties of PEL type Torsten Wedhorn∗ July 4, 2018 Abstract. We study the Bruhat stratification for Shimura varieties of PEL type. In the Siegel case this stratification is a scheme-theoretic variant of the stratification by the a-number. We show that all Bruhat strata are smooth and determine their dimensions. We also prove that the closure of a Bruhat stratum is a union of Bruhat strata and describe which Bruhat strata are contained in the closure of a given Bruhat stratum. MSC (2010): 14G35, 14K10, 20G40 Key words: Shimura varieties, Bruhat decomposition, F -zips, Newton stratification Introduction Background Let X be an abelian variety over a perfect field k of characteristic p > 0. An important invariant of X is its a-number a(X) := dimk Hom(αp, X), where αp := Ker(Frob : Ga → Ga). One always has 0 ≤ a(X) ≤ dim(X) with a(X) = 0 (resp. a(X) = dim(X)) if and only if X is an ordinary abelian variety (resp. X is a superspecial abelian variety). The stratification by a-number on moduli spaces of polarized abelian varieties has been studied intensively for instance by G. van der Geer in [vdG]. F. Oort has shown that in every non-ordinary Newton stratum of the moduli spaces of principally polarized abelian varieties the stratum where the a-number is equal to 1 is open and dense in that Newton stratum. He then used this fact to prove a conjecture by Grothendieck which says that the closure of Newton stratum corresponding to a (concave) Newton polygon ν is the union of those Newton strata corresponding to the Newton polygons lying below ν ([Oo]). Main results In this paper we study the Bruhat stratification introduced in [Wd2] for good reductions of Shimura varieties of PEL type. These good reductions are still ∗Dept. of Mathematics, University of Paderborn, Warburger Str. 100, D-33098 Pader- born, Germany, [email protected] 1 moduli spaces of abelian varieties with additional structures. We show in Example 2.4 that for the Siegel case (i.e. for the moduli space of principally polarized abelian varieties in characterstic p > 0) the Bruhat stratification is a scheme-theoretic variant of the stratification by the a-number. 1 In general, the Bruhat stratification is a decomposition of the special fiber A0 of the PEL-moduli space into locally closed subspaces. These sub- spaces can roughly be described as follows. The first de Rham homology H DR (X /A0) of the universal abelian scheme X over A0 with all its ad- ditional structures is endowed with two local direct summands, the Hodge filtration and the conjugate filtration (see (2.2) for a precise definition). The Bruhat strata are then the loci, where these two filtrations with all their ad- ditional structures are in a fixed relative position. In the case of the moduli space of principally polarized abelian varieties of dimension g, H DR (X /A0) is a locally free module of rank 2g endowed with a symplectic pairing given by the polarization, and the Hodge C and the conjugate filtration D are totally isotropic local direct summands of H DR (X /A0) of rank g. Then the Bruhat strata are simply the locally closed subspaces aA , where C + D is a local direct summand of H DR (X /A0) of some fixed rank 2g − a. Moreover aA (Fp) consists of those principally polarized abelian varieties (X, λ) over Fp with a(X) = a. 1 1 1 In general, the Bruhat stratification is indexed as follows. Let G be the reductive group over Q of the PEL-Shimura datum. As we assume that the Shimura variety has good reduction at p, the group GQp has a reductive model G over Zp. Let G/Fp be its special fiber and denote by (W, I) its Weyl group together with its set of simple reflections. Let [µ] be given by the Shimura the minuscule conjugacy class of cocharacters of GFp and datum. It determines a conjugacy class of parabolic subgroups of GFp hence a subset J ⊆ I (see (1.2) for its precise definition). The (geometric) Frobenius acts on (W, I) via an automorphism ¯ϕ of Coxeter systems. We set K := ¯ϕ(J)opp, where ( )opp denotes the opposite type. The field of definition of J is the finite extension κ of Fp over which the special fiber A0 of the PEL-moduli space is defined. Let ΓJ be the Galois group of κ. We let J W K be the set of elements in w ∈ W that are of minimal length in their double coset WJ wWK, where WJ and WK are the subgroups of W generated by J resp. K. Then the Galois group ΓJ acts on J W K and the Bruhat stratification is a decomposition A0 = [ [x]A [x]∈ΓJ \J W K into locally closed subspaces. Our main result is the following (Corol- lary 4.2): Theorem 1. The Bruhat strata [x]A are smooth of dimension ℓ(xJ,K) (see (4.1) for the definition of xJ,K). In particular all Bruhat strata are non- 2 empty. The closure of a Bruhat stratum is given by [x]A = [ [x′]A , [x′]≤[x] where ≤ denotes the partial order on ΓJ \J W K induced by the Bruhat order on the Coxeter group W . In the Siegel case we show in Example 4.3 that one reobtains the known formulas for the dimension of the a-number strata (e.g., see [vdG]). There are two essential tools to prove Theorem 1. The first is the descrip- tion of the Bruhat stack BJ,K (2.1) obtained in [Wd2] and briefly recalled in the beginning of Section 2. The second tool is to show that the morphism a : A0 → BJ,K, which sends an abelian variety X with additional structure to the triple consisting of H DR (X) (with its additional structure), the Hodge filtration, and the conjugate filtration, is a smooth morphism (Theorem 4.1). 1 In [ViWd] Viehmann and the author defined and studied the Ekedahl- Oort stratification for Shimura varieties of PEL type. By construction every Bruhat stratum is a union of Ekedahl-Oort strata. Using the results of [Wd2] we describe in Proposition 3.1 which Ekedahl-Oort strata are contained in a given Bruhat stratum. We deduce the following result (Theorem 3.5): Theorem 2. There exists a unique open Bruhat stratum [x]A . It is dense in A0. Moreover the following assertions are equivalent. (i) [x]A is equal to the generic Newton stratum. (ii) The field of definition κ of J is equal to Fp. (iii) There exists an abelian variety X with additional structure in A0(Fp) which is ordinary. Contents of the paper In Section 1 we introduce the PEL-moduli spaces and all other notations. We also very briefly recall the Newton stratification. In Section 2 we define the Bruhat stratification of the special fiber of the PEL-Moduli space and show that it generalizes the a-number stratification in the Siegel case. We compare the Ekedahl-Oort and the Bruhat stratification and deduce Theorem 2 in Section 3. Section 4 contains Theorem 1 and its proof using a general result about deformations of opposite parabolic subgroups. This result is shown in Section 5. Acknowledgements. I am grateful to the referee for his/her careful read- ing and for the suggestion of Remark 3.4. 3 1 Moduli spaces of PEL type with good reduction Let S be a scheme over Fp and let q be a power of p. The pullback of a scheme or a sheaf or a morphism over S under the qth-power Frobenius morphism S → S is denoted by ( )(q). In this section we recall the notion of Shimura-PEL-data and their at- tached moduli spaces. Our main references are Kottwitz [Kot] and Rapoport and Zink [RaZi]. The PEL moduli space A Let D = (cid:0)B, ∗, V, h , i, OB, Λ, h(cid:1) denote an integral Shimura-PEL-datum that is unramified at a prime p > 0 in the sense of [ViWd] Section 1.1. Let G be the associated reductive group over Q, and denote by [µ] the associated conjugacy class of cocharacters of G. We assume that G is connected, i.e., we exclude the case (D) of [ViWd] Remark 1.1. Let ¯Q be the algebraic closure of Q in C. Then [µ] is already defined over ¯Q. Let E be the reflex field associated with D, i.e. the field of definition of [µ]. It is a finite extension of Q contained in ¯Q. Let ¯Qp be an algebraic closure and fix an embedding ¯Q ֒→ ¯Qp. This determines a place v of E over p. Let Ev ⊆ Qp be the v-adic completion of E, OEv its ring of integers, and let κ be its residue field. The assumption on D to be unramified implies that Ev is an unramified extension of Qp. Let ¯κ be the residue field of the ring of integers of Qp. This is an algebraic closure of κ. f be the ring of finite adeles over Q with trivial p-th component and let C p ⊂ G(Ap f ) be a compact open subgroup. We denote by A = AD,C p the moduli space defined by Kottwitz [Kot] §5 (see also [ViWd] Section 1). Then A is an algebraic Deligne-Mumford stack which is smooth over If C p is sufficiently small, A is representable by a smooth quasi- OEv . projective scheme over OEv (see loc. cit. or [Lan] 1.4.1.11 and 1.4.1.13). We denote its special fiber by Let Ap A0 = AD,C p,0 := AD,C p ⊗OEv κ. Group theoretical notation Let G be the Zp-group scheme of OB-linear symplectic similitudes of Λ. This is a reductive group scheme over Zp whose generic fiber is GQp. We denote by G its special fiber. This is a connected reductive group over Fp and hence quasi-split (as any reductive group over a finite field). We fix a maximal torus T of G and a Borel subgroup B of G containing T (both defined over Fp). For every Fp-algebra R we set GR := G ⊗Fp R, TR := T ⊗Fp R, BR := B ⊗Fp R 4 We denote by X ∗(T ) (resp. X∗(T )) the group of characters (resp. of cochar- acters) of T¯κ. Let (W, I) be the Weyl group together with its set of simple reflections of (G, B, T ). We denote by w0 ∈ W the element of maximal length. If Φ ⊂ X ∗(T ) is the set of roots of (G, T ) and ∆ ⊂ Φ the root basis corresponding to B, then there exists a unique involution ι of Φ such that ι(α) = −w0(α) for α ∈ Φ. It preserves ∆. The induced involution of the Coxeter system (W, I), given by sα → sι(α), where sα ∈ W is the reflection corresponding to α ∈ Φ, is the conjugation with w0. If S ⊆ W is a subset, we denote by Sopp := w0S the image of S under this involution. The Galois group Γ := Gal(¯κ/Fp) acts on X ∗(T ), on X∗(T ), and on the Coxeter system (W, I). It is topologically generated by the geometric Frobenius automorphism σ ∈ Γ (i.e. σ−1 is the automorphism x 7→ xp of ¯κ). The automorphism of the Coxeter systems (W, I) induced by σ is denoted by ¯ϕ. Note that ¯ϕ is also the automorphism induced by the geometric Frobenius ϕ : GFp → GFp. For any subsets J, K ⊆ I, we denote by WJ the subgroup of W generated by J and by J W (resp. W K, resp. J W K) the set of w ∈ W that are of minimal length in the left coset WJ w (resp. in the right coset wWK , resp. in the double coset WJ wWK ). Then J W K = J W ∩ W K. The Coxeter group is endowed with the Bruhat order which we denote "≤": For x, w ∈ W we have x ≤ w if and only if for some (or, equivalently, any) reduced expression w = s1 · · · sn as a product of simple reflections si ∈ I, one gets a reduced expression for x by removing certain si from this product. The minimal number ℓ needed to express w ∈ W as product of ℓ simple reflections is the length of w and denoted ℓ(w). For J ⊆ I let κ(J) be its field of definition, i.e. κ(J) is the finite extension of Fp in ¯κ such that (1.1) ΓJ := Gal(¯κ/κ(J)) = { γ ∈ Γ ; γ(J) = J } Let PJ ⊆ Gκ(J) be the unique parabolic subgroup of type J containing B, and let ParJ = Gκ(J)/PJ be the projective κ(J)-scheme of parabolics of type J of G. Let T be a maximal torus of G such that T ⊗Zp Fp = T (such a T always exists because the scheme of maximal tori of G is smooth, [SGA3] Exp. XV, 8.15) and let T be its generic fiber. Let Qnr p denote the maximal unram- ified extension of Qp in Qp and identify Gal(Qnr p /Qp) with Γ. Then the cocharacter group X∗(T ⊗Qp Qnr p ) is isomorphic to X∗(T ) as Γ-modules. Every element in the conjugacy class [µ] is conjugate via some element in G(Qnr p ) which is unique up to conjugation with an element of the normalizer of T ⊗Qp Qnr p . Thus via the identifica- tion X∗(T ⊗Qp Qnr p ) = X∗(T ) we may consider the conjugacy class [µ] as an element of X∗(T )/W . p ) to an element of X∗(T ⊗Qp Qnr 5 Let µ be the B-dominant representative in X∗(T ) in the conjugacy class [µ]. As G is quasi-split, it is defined over the field of definition of its conju- gacy class, i.e. over κ. We set (1.2) J := { i ∈ I ; hµ, αii = 0 }, where αi ∈ X ∗(T ) is the simple root corresponding to i ∈ I. We also set (1.3) K := w0 ¯ϕ(J) = ¯ϕ(J)opp. As κ is the field of definition of [µ], one has for the fields of definitions of J and K κ(J) = κ(K) = κ. The Newton stratification and the µ-ordinary locus As in [ViWd] Section 7.2 we denote by Nt : A0 → B(GQp, µ) the map that sends each point s of A0 to the isogeny class of the p-divisible group with D-structure given by a geometric point lying over s. For b ∈ B(GQp, µ) we denote by Nb := Nt−1(b) the corresponding Newton stratum. The set B(GQp, µ) is finite and partially ordered. It contains a unique maximal element bµ and the corresponding Newton stratum Nµ := Nbµ is called the µ-ordinary Newton stratum. It is open and dense in A0 by the main result of [Wd1]. 2 Definition of the Bruhat stratification In [ViWd] 2.1 Viehmann and the author constructed a morphism ζ from A0 to the stack of so-called D-zips which we identified with the stack of G-zips of type µ (in the sense of [PWZ2]) in [ViWd] 4. The corresponding zip stratification ([PWZ2] (3.27) or [Wd2] Definition 2.8) is nothing but the Ekedahl-Oort stratification. As explained in [Wd2] we also obtain a Bruhat stratification by composing ζ with the canonical morphism from the stack of G-zips of type µ into the Bruhat stack (2.1) BJ,K := [Gκ\(ParJ × ParK)] defined in [Wd2]. Here [G\X] denotes the quotient stack over a scheme S if a group scheme G over S acts from the left on a scheme X over S. We denote this composition by a : A0 → BJ,K. Below we will give a more direct description of the morphism a. We recall the properties of BJ,K shown in [Wd2] (note that the notation here is considerably simplier because G is by hypothesis connected). 6 (1) BJ,K is a smooth algebraic stack of finite type over κ. It is of dimension dim(G) − dim(PJ ) − dim(PK ) over κ. Note that with K defined as in (1.3) we have dim PJ = dim PK and hence dim BJ,K = − dim(LJ ), where LJ is a Levi subgroup of PJ . (2) To describe its underlying topological space recall that one may endow any partially ordered set X with a topology by defining the open sets as those subsets U of X such that for all u ∈ U and x ∈ X with x ≥ u one has x ∈ U . This defines a fully faithful functor from the category of partially open sets (morphisms are order preserving maps) to the category of topological spaces. Then the underlying topological space of BJ,K can be identified with the space attached to the partially ordered set ΓJ \J W K. Here J W K carries the Bruhat order which is preserved by the action by ΓJ , and ΓJ \J W K is endowed with the induced order. Over the algebraic closure, the underlying topological space of BJ,K⊗κ ¯κ is given by the partially ordered set J W K and the projection BJ,K ⊗κ ¯κ → BJ,K induces the canonical map J W K → ΓJ \J W K on underlying topological spaces. (3) For each x ∈ J W K consider the reduced locally closed substack Gx of BJ,K ⊗ ¯κ consisting only of the point x. It is already defined over κ(x) and it is smooth of dimension dim(PJ xPK ) − dim PK − dim PJ = ℓJ,K(x) − dim PK, where ℓJ,K(x) := ℓ(x) + ℓ(w0,K) − ℓ(w0,Jw). Here w0,K denotes the element of maximal length in WK and Jw := K ∩ w−1Jw. D-module structure on the De Rham homology Let S be a κ-scheme and let (A, ι, λ, η) be an S-valued point of A0. Let D(A[p]) be the covariant Dieudonn´e crystal of the p-torsion and let M (A) be its evaluation in the trivial PD-thickening (S, S, 0). Then there is an OS-linear functorial isomorphism M (A) ∼= H DR (A/S), where the De Rham homology is defined as the OS-linear dual of the De Rham cohomology H 1 DR(A/S) ([BBM] Prop. 2.5.8). Thus M (A) is a locally free OS-module of rank dimQ(V ). Via functoriality it is endowed with an OB-action (induced by ι) and with a similitude class of perfect alternating forms (induced by λ), i.e. M (A) is equipped with the structure of a D-module in the following sense. 1 Definition 2.1. Let T be a Zp-scheme. A D-module over T is a locally free OT -module M of rank dimQ(V ) endowed with an OB-action and the similitude class of a symplectic form h , i such that hbm, m′i = hm, b∗m′i for all b ∈ OB and local sections m, m′ of M . 7 An isomorphism M1 ∼→ M2 of D-modules over T is an OT ⊗ OB-linear symplectic similitude. Here we call two perfect pairings h , i1 and h , i2 on a finite locally free OT -module M similar if there exists an open affine covering T = Sj Vj and for all j a unit cj ∈ Γ(Vj, O × ) such that hm, m′i2 = cjhm, m′i1 for all Vj m, m′ ∈ Γ(Vj, M ). For a morphism f : T ′ → T of Zp-schemes and a D-module M there is the obvious notion of a pull back f ∗M = MT ′ of M to a D-module on T ′. Altogether we obtain the category ^D−Mod of D-modules fibered over the category of Zp-schemes. As descent for finite locally free modules is effective for fpqc-morphisms, ^D−Mod is a stack for the fpqc topology. In fact it is the classifying stack of G: Proposition 2.2. ^D−Mod = [ G\ Spec(Zp)]. Here we endow Spec(Zp) with the trivial G-action. Proof. The Zp-module Λ together with the induced OB-action and the in- duced pairing is a D-module over Zp. Its automorphism group scheme is G. Moreover, [RaZi] Theorem 3.16 shows that all D-modules are ´etale locally isomorphic to the D-module ΛT . This shows the claim. We set D−Mod := ^D−Mod ⊗Zp Fp = [G\ Spec(Fp)]. We will also use the following torsor A # over A . For every OEv -scheme S the S-valued points of A # are given by quintuples (A, ι, λ, η, α), where (A, ι, λ, η) ∈ A (S) and where α is an OB-linear symplectic similitude H DR (A/S) ∼→ Λ ⊗Zp OS. 1 By Proposition 2.2, A # is a GOEv -torsor over A for the ´etale topology. We also set A # 0 := A # ⊗OE,v κ. Hodge filtration and conjugate filtration on the De Rham homol- ogy Now let (A, ι, λ, η, α) be an S-valued point of A # homology M (A) = H DR homology of the relative de Rham complex Ω• isotropic locally direct summands, namely 0 . Then the first de Rham (A/S) (i.e., the OS-linear dual of the first hyperco- A/S) comes with two totally 1 (2.2) C := ωA∨ := f ∨ ∗ Ω1 A∨/S, D := R1f∗(H 0(Ω• A/S))⊥. Here f : A → S (resp. f ∨ : A∨ → S) denotes the structure morphism of A (resp. of the dual abelian scheme A∨). 8 We define P (resp. Q) to be the stabilizer of α(C) (resp. α(D)) in GS. Then P (resp. Q) is a parabolic subgroup of type J (resp. K = ¯ϕ(J)opp) of GS by [ViWd] Prop. 1.13. As the formation of C and D is functorial, we obtain a morphism a# : A # 0 → ParJ × ParK . We endow ParJ × ParK with the obvious diagonal left action of Gκ and denote by BJ,K := BJ,K(G) := [Gκ\(ParJ × ParK)] the quotient stack. Then a# is Gκ-equivariant and hence it induces the morphism (2.3) a : A0 → BJ,K. Applying [Wd2] Definition 1.13, the morphism a yields the Bruhat stratifi- cation of A0, indexed by ΓJ \J W K, i.e. we obtain a decomposition (2.4) A0 = [ [x]A [x]∈ΓJ \J W K of A0 into locally closed substacks. Definition 2.3. The locally closed substacks [x]A for [x] ∈ ΓJ \K W J are called the Bruhat strata of A0. We also define geometric Bruhat strata as follows. Let κ ⊆ k ⊂ ¯κ be the splitting field of the reductive group G. Then Gal(¯κ/k) acts trivially on (W, I) and the underlying topological space of BJ,K ⊗κk is KW J . The residue gerbe Gx attached to x ∈ K W J is a locally closed substack of BJ,K ⊗κ κ(x), where κ ⊆ κ(x) ⊆ k is the field of definition of x. We set (2.5) xA := a−1(Gx ⊗κ(x) k) ⊆ A0 ⊗κ k. This is a locally closed substack of A0 ⊗κ k, called a geometric Bruhat stra- tum. Then one has for [x] = ΓJ · x ∈ ΓJ \J W K a decomposition into open and closed substacks [x]A ⊗κ k = a z∈ΓJ ·x zA . Altogether we obtain a disjoint decomposition into locally closed substacks A0 ⊗κ k = [ xA , x∈K W J which we call the geometric Bruhat stratification of A0. 9 i.e., we choose B = Q in the Example 2.4. Consider the Siegel case, Shimura-PEL-datum D, and hence G = GSp(Λ, h , i). Set g := dimQ(V )/2. Let us explain that in this case the Bruhat stratification is nothing but a scheme-theoretic version of the stratification by the a-number defined by Oort in [Oo]. The reflex field E is equal to Q and hence κ = Fp. Then a D-module over an Zp-scheme S is a locally free OS-module of rank 2g endowed with the similitude class of a symplectic pairing. We have J = K and attaching to a Lagrangian C in the free D-module ΛS := Λ ⊗Zp OS (i.e., C is a totally isotropic locally direct summand of rank g) its stabilizer in GS yields a bijection between the set of Lagrangians in ΛS and set of parabolic subgroups of type J of GS. Let k be a field extension of κ, Ci, Di ⊂ Λk Lagrangians (i = 1, 2) and let Pi = StabGk (Ci) and Qi = StabGk (Di) be the correpsonding subgroups of type J. Then two pairs (P1, Q1) and (P2, Q2) are in the same G(k)- orbit if and only if dimk(C1 ∩ D1) = dimk(C2 ∩ D2) ∈ {0, . . . , g}. This shows that in this case we can identify J W J with {0, . . . , g} via a map w 7→ n(w). Moreover we have w ≤ w′ (w.r.t. the Bruhat order) if and only if n(w) ≥ n(w′). Now assume that k is perfect and let (A, λ, η, α) be a k-valued point of A # 0 . Denote by σ the absolute Frobenius on k. Thus α is a symplectic simil- itude M (A) ∼= Λk := k⊗Zp Λ, where the symplectic form on M (A) is induced by λ and on Λk by h , i. Set C := α(ωA∨) and D := α(R1f∗(H 0(Ω• A/S))⊥). These are totally isotropic g-dimensional subspaces of Λk. Recall that for any k-vector space W we set W (p) = k ⊗σ,k W . Let F : M (A)(p) → M (A), V : M (A) → M (A)(p) be the k-linear maps induced by Verschiebung and Frobenius, respectively. By transport of structure via α we obtain k-linear k → Λk and V : Λk → Λ(p) maps F : Λ(p) such that C (p) = V (Λk) and D = F (Λ(p) k ) (this is the dual version of [Od] 5.11, where De Rham co- homology and contravariant Dieudonn´e theory are considered). k Let F ♭ : Λk → Λk be the σ-linear map corresponding to F , i.e. F ♭(x) = F (1 ⊗ x) for x ∈ Λk. As σ is bijective, the image of F and of F ♭ coincide. We also denote by V ♭ : Λk → Λk the σ−1-linear map corresponding to V , i.e., V ♭ = τ ◦ V with τ : Λ(p) k → Λk, (a ⊗ x) = σ−1(a)x. On the other hand, as Λk has an Fp-rational stucture Λ ⊗Zp Fp, there are a canonical k-linear ∼→ Λk and a σ-linear map σΛ : Λk → Λk, α ⊗ z 7→ αp ⊗ z isomorphism γ : Λ(p) k for α ∈ k and z ∈ Λ. Finally let β : Λk → Λ(p) the σ-linear map x 7→ 1 ⊗ x k for x ∈ Λk. Then (2.6) (2.7) σΛ ◦ τ = γ, γ ◦ β = σΛ. 10 Thus we find σΛ(C) (2.7) = γ(β(C)) = γ(C (p)) = γ(V (Λk)) (2.6) = σΛ(V ♭(Λk)) and hence V ♭(Λk) = C. Therefore (A, λ, η) is in the Bruhat stratum corresponding to w ∈ J W J if and only if dimk(F ♭M (A) ∩ V ♭M (A)) = n(w), i.e. if and only if the point (A, λ, η) has a-number equal to n(w) in the sense of [Oo]. The morphism a induces a continuous map A0(¯κ) → J W K. This can also be expressed as the following semi-continuity result. Remark 2.5. As usual endow J W K with the Bruhat order. For all x ∈ J W K the set { s ∈ A0(¯κ) ; a(s) ≤ x } is closed. Below (Corollary 4.2) we will show that this set is the closure of { s ∈ A0(¯κ) ; a(s) = x }. For instance we can apply this in the Siegel case (Example 2.4). There we identified the partially ordered sets (J W K, ≤) and ({0, . . . , g}, ≥). Hence for d ∈ {0, . . . , g} the locus where the a-number ≥ d is always a closed subset. Moreover, it is the closure of the locus of points where the a-number is equal to d. 3 Ekedahl-Oort strata and Bruhat strata The Ekedahl-Oort stratification defined in [ViWd] is a special case of the zip stratification introduced in [PWZ2]. It is a refinement of the Bruhat stratification and the results of [Wd2] allow to give a simple description of the Ekedahl-Oort strata that are contained in a given Bruhat stratum. For this recall that the Ekedahl-Oort stratification is a decomposition into locally closed substacks (3.1) A0 = [ A [w]. [w]∈ΓJ \J W As above we also have geometric Ekedahl-Oort strata A w for w ∈ J W defined over a splitting field k of G. By [ViWd] Section 6 and Section 9, the A w are quasi-affine, smooth of dimension ℓ(w), and the closure of an Ekedahl-Oort stratum (resp. a geometric an Ekedahl-Oort stratum) is the union of Ekedahl-Oort strata (resp. of geometric Ekedahl-Oort stratum). It suffices to compare geometric Ekedahl-Oort strata and geometric Bruhat strata. Proposition 3.1. For x ∈ J W K one has xA = [ A xy y∈K∩x−1J xWK 11 Proof. In [Wd2] Proposition 2.10 it is shown that the canonical morphism from the stack of G-zips of type µ to the Bruhat stack BJ,K is given on the underlying topological spaces by attaching to w ∈ J W the unique element of minimal length in wWK . This is a surjective map Hence for x ∈ J W K one has (3.2) π : J W → J W K. xA = [ A w, w∈π−1(x) But by a corollary of a result of Howlett on Coxeter groups (e.g., [PWZ1] Prop. 2.8) one has π−1(x) = { xy ; y ∈ K∩x−1JxWK }. Corollary 3.2. For x ∈ J W K the corresponding Bruhat stratum xA con- sists of a single Ekedahl-Oort stratum if and only if K = x−1Jx. Proof. By Proposition 3.1, xA consists of a single Ekedahl-Oort stratum if and only K ∩ x−1Jx = K. As J and K have the same number of elements, this condition is equivalent to K = x−1Jx. Example 3.3. Assume that J = K (e.g., if G is a group of Dynkin type C, because then the element w0 of maximal length is central and ¯ϕ(J) = J). Let 1 ∈ K W J be the element of minimal length. Then π−1(1) = {1} and 1A is the superspecial Ekedahl-Oort stratum (in the sense of [ViWd] Exam- ple 4.16). For J 6= K, 1A is the union of more than one Ekedahl-Oort stratum. The maximal Bruhat stratum Let x ∈ J W K be the element of maximal length. The corresponding geo- metric Bruhat stratum xA is called the maximal Bruhat stratum. Remark 3.4. As the Galois action by ΓJ on J W K preserves the length of elements, we have xA = [x]A , i.e., the maximal Bruhat stratum is already defined over κ. The maximal Bruhat stratum is open and dense in A0: If we consider J W K as a topological space, the maximality of x means that {x} is open in J W K. Thus xA is an open in A0. Moreover it contains by Proposition 3.1 the µ-ordinary Ekedahl-Oort stratum, i.e. the Ekedahl-Oort stratum cor- responding to the maximal element wµ in J W ([ViWd] Example 4.16). By [ViWd] Theorem 6.1 the µ-ordinary Ekedahl-Oort stratum is dense in A0 12 and thus xA is dense. The density of xA will also follow from Corollary 4.2 below. In general, xA is not equal to µ-ordinary Ekedahl-Oort stratum (which is also the µ-ordinary Newton stratum by the first main result of [Moo]; see also [Wo] for a purely group-theoretical proof). Theorem 3.5. The following assertions are equivalent. (i) The generic Bruhat stratum xA is equal to the µ-ordinary Newton stra- tum. (ii) ¯ϕ(J) = J. (iii) Ev = Qp. (iv) The ordinary locus of A0 (i.e. the locus of points (A, ι, λ, η) where the underlying abelian scheme A is ordinary) is non-empty. (v) The ordinary locus of A0 is equal to the µ-ordinary locus. Proof. Assertion (ii) means that κ(J) = Fp and thus is equivalent to (iii). The equivalence of (iii), (iv) and (v) is [Wd1] (1.6.3). Thus it remains to show that (ii) is equivalent to the equality of xA and the µ-ordinary Ekedahl- Oort stratum (by the above mentioned result of Moonen). By Corollary 3.2 this equality holds if and only if K = x−1J x = J opp. But by definition K = ¯ϕ(J)opp. This shows the equivalence of (i) and (ii). Example 3.6. The equivalent conditions of Proposition 3.5 are satisfied in the following cases. (1) All connected components of the Dynkin diagram of G are of Dynkin type C. (2) G is split. 4 Properties of the Bruhat strata Theorem 4.1. The morphism a (2.3) is smooth and surjective. Before giving the proof of this theorem we deduce some properties of the geometric Bruhat strata. For this we first introduce the following no- tation. For x ∈ J W K let xJ,K be the element of maximal length in J W ∩ WJ xWK. Then xJ,K = xxJ,K, where xJ,K is the element of maximal length in K∩x−1JxWK and we have (4.1) ℓ(xJ,K) = ℓJ,K(x) = ℓ(x) + ℓ(xJ,K) by a result of Howlett ([PWZ1] 2.7 and 2.8). Corollary 4.2. Let x ∈ J W K. (1) The corresponding geometric Bruhat stratum xA is smooth of pure di- mension ℓ(xJ,K) (4.1). In particular, all Bruhat strata are non-empty. 13 (2) The closure of xA is given by xA = [ x′≤x x′ A , where ≤ denotes the Bruhat order on J W K. Proof. As the morphism a is surjective, all Bruhat strata are non-empty. Let Gx ⊂ BJ,K ⊗ κ(x) be the residue gerbe of x. Then the cartesian diagram xA xa Gx A0 ⊗ κ(x) aκ(x) / BJ,K ⊗ κ(x) shows that the smoothness of a implies the smoothness of xa. As Gx is smooth over κ(x), xA is smooth over κ(x). Finally, as a is smooth, it preserves codimension. By [Wd2] Proposi- tion 1.12, Gx has in BJ,K ⊗ κ(x) pure codimension dim G − dim PJ − ℓ(xJ,K). As dim A0 = dim G − dim PJ , we obtain that xA is of pure dimension ℓ(xJ,K). As a is smooth, it is an open morphism. Hence (2) follows from the description of the underlying topological space of BJ,K ([Wd2] Proposi- tion 1.10) and the fact that for open morphisms taking closures commutes with taking inverse images. Proof of Theorem 4.1. The morphism a is the composition of the morphism ζ, which attaches to every point of A0 its G-zip of type µ, and the mor- phism from the stack of G-zips of type µ to the Bruhat stack BJ,K. This morphism is surjective by Theorem 9.1 of [ViWd]. As explained in the proof of Proposition 3.1, the morphism from the stack of G-zips of type µ to the Bruhat stack BJ,K is also surjective. Therefore a is surjective. Moreover, source and target of a are both smooth over Spec κ. Therefore it suffices to show that a is surjective on tangent spaces. By this we mean the following. Let k be an algebraically closed extension of κ, let k[ε] be the ring of dual numbers over k, and let (4.2) Spec k Spec k[ε] x x A0 a / BJ,K be a 2-commutative diagram. Then we have to show that there exists a morphism Spec k[ε] → A0 which 2-commutes with (4.2). For this we use Dieudonn´e displays defined by Zink and Lau ([Zi], [Lau1], [Lau2]; see also [ViWd] 3.1 for a short reminder). We denote the Zink 14 / / / / / / rings of k and k[ε] by W(k) and W(k[ε]), respectively. Then W(k) is equal to the Witt ring W (k) and there is a surjective ring homomorphism W(k[ε]) → W(k) whose kernel consists of nilpotent elements. Let Ik = pW (k) (resp. Ik[ε]) be the kernel of W(k) → k (resp. of W(k[ε]) → k[ε]). We denote the Frobenius on the Zink ring by σ and the pull back via σ by ( )(σ). The k-valued point x is given by a tuple (A, λ, ι, η). Let (P, Q, F, F1) be the Dieudonn´e display attached to its p-divisible group. It is endowed with a similitude class of perfect alternating forms h , i induced by λ and an OB- action induced by ι. In other words, P is a D-module over W(k). As W (k) is strictly henselian, we may choose by Proposition 2.2 an isomorphism of P with the D-module Λ ⊗Zp W(k). This defines in particular a Zp-rational structure on the D-module P and an identification P/IkP = Λk. 1 The Hodge filtration is given by C = Q/IkP ⊂ H DR (A/k) = P/IkP . Choose a totally isotropic OB-invariant direct summand C ′ of P lifting C and let T ′ ⊂ P (σ) = P be an OB-invariant totally isotropic complement of (C ′)(σ). Then the composition g : P ∼= P (σ) = (C ′)(σ) ⊕ T ′ F1(C′)(σ) ⊕F T ′ −−−−→ P is an automorphism of D-modules, i.e., g ∈ G(W(k)). The image of g(T ′) in P/IkP is the conjugate filtration D. The morphism x : Spec k[ε] → BJ,K then corresponds to totally isotropic OB-invariant direct summands C and D of Λk[ε] lifting C and D, respec- tively. Applying Lemma 5.1 below to the stabilizers of D and C (p), there exists g ∈ G(k[ε]) with g ≡ g mod ε such that T := g−1( D) is a complement of C (p). As explained in [ViWd] Section 4.1, triples (C ′, T ′, g) as above are parametrized by a smooth scheme and thus we can apply [ViWd] Lemma 3.1 to see that there exists a totally isotropic OB-invariant direct summand C ′ of ΛW(k[ε]), a totally isotropic OB-invariant complement T ′ of ( C ′)(σ) and g′ ∈ G(W(k[ε])) whose pullback to W(k) is (C ′, T ′, g) and whose pullback to k[ε] is ( C, T , g). We obtain a Dieudonn´e display ( P , Q, F , F1) as follows. We define P := ΛW(k[ε]) and denote by Q the inverse image of C under P → P /Ik[ε] P = Λk[ε]. Finally let F : P (σ) → P and F1 : Q(σ) → P be the unique W(k[ε])-linear maps making ( P , Q, F , F1) into a Dieudonn´e display such that the direct sum of the restriction of F1 to ( C ′)(σ) and the restric- tion of F to T ′ is given by g. Then ( P , Q, F , F1) is a Dieudonn´e display with symplectic form and OB-action lifting (P, Q, F, F1) and thus defining a morphism Spec k[ε] → A0 which 2-commutes with (4.2). Example 4.3. We consider again the Siegel case, i.e., B = Q and therefore G = GSp(Λ, h , i). Set g := dimQ(V )/2. We have seen in Example 2.4 that the Bruhat stratification on A (¯κ) can also be described by A (¯κ) = [ 0≤i≤m Ai, 15 where Ai denotes the locally closed subvariety of A (¯κ) consisting of prin- cipally polarized abelian varieties whose a-number is equal to i. By Corol- lary 4.2 we have Ai = [ j≥i Aj and Ai is smooth and equi-dimensional of dimension d(i) with d(i) := g X j=i+1 j = g(g + 1) − i(i + 1) 2 . This explicit description of ℓ(xJ,J ) for x ∈ J W J ∼= {0, . . . , g} follows from the explicit calculation of the length in [ViWd] Appendix A.7, in particular the formulas (12.3) and (12.4) there. Note that for the identification of J W J and {0, . . . , g} made here in Example 2.4, the map in loc. cit. (12.4) should be given by (εi)i 7→ #{ i ; εi = 0 }. 5 Deforming parabolics in opposite position Here we prove a group-theoretical lemma used in the proof of Theorem 4.1. Let k be any field, and let G be a reductive group over k. Let (W, I) be the Weyl group of G together with its set of simple reflections. Recall that for any k-scheme S, two parabolic subgroups P and Q of GS are called opposite if P ∩ Q is a Levi subgroup of P and of Q. We indicate this by writing P ⊲⊳ Q. Moreover for every Levi subgroup L of a parabolic subgroup P there exists a unique parabolic subgroup Q of GS such that P ∩ Q = L and then P ⊲⊳ Q ([SGA3] Exp.XXVI Th´eor`eme 4.3.2). This implies that if J ⊆ I is the type of P , the type J opp of a parabolic subgroup Q opposite to P depends only on J. Lemma 5.1. Let ZJ be the k-scheme whose S-valued points are the set of triples (P, Q, g), where P is a parabolic subgroup of GS of type J, Q is a parabolic subgroup of GS of type J opp, and g ∈ G(S) such that gQ ⊲⊳ P . Then the canonical morphism ZJ → ParJ × ParJ opp, (P, Q, g) 7→ (P, Q) is smooth. Proof. We set X := ParJ × ParJ opp. Let PJ ⊂ GParJ be the universal parabolic subgroup of GParJ of type J and let P ′ J be its pull back to X under the first projection. This is a parabolic subgroup of GX . Define similarly a parabolic subgroup P ′ J opp of GX . Then the X-scheme ZJ is isomorphic to the subscheme of GX whose S-valued points are given by { g ∈ GX (S) ; gP ′ J }. This is an open subscheme of GX by [SGA3] Exp.XXVI Th´eor`eme 4.3.2. As GX is smooth over X, this implies that ZJ is smooth over X. J opp ⊲⊳ P ′ 16 References [BBM] P. Berthelot, L. Breen, W. Messing: Th´eorie de Dieudonn´e cristalline II, LNM 930, Springer-Verlag (1982). [Kot] [Lan] [Lau1] [Lau2] [Moo] [Od] [Oo] R. E. Kottwitz, Points on some Shimura varieties over finite fields, J. AMS 5 (1992), 373 -- 444. K.-W. Lan, Arithmetic compactification of PEL type Shimura va- rieties, PhD thesis, Harvard 2008, revised version July 2010, to appear in London Mathematical Society Monographs. E. Lau, A duality theorem for Dieudonn´e displays, Ann. Sci. ´Ec. Norm. Sup. (4) 42 (2009), 241 -- 259. E. Lau, A relation between Dieudonn´e displays and crystalline Dieudonn´e theory, arXiv:1006.2720v1. B. Moonen, Serre-Tate theory for moduli spaces of PEL type, Ann. Sci. ´Ec. Norm. Sup. (4) 37 (2004), 223 -- 269. T. Oda, The first de Rham cohomology group and Dieudonn´e mod- ules, Ann. Sci. de l'ENS. 2 (1) (1969), 63 -- 135. F. Oort, Newton polygon strata in the moduli space of abelian vari- eties, in "Moduli of Abelian Varieties" (ed. by C. Faber, G. van der Geer, F. Oort), Progress in Mathematics 195, Birkhauser (2001), 417 -- 440. [PWZ1] R. Pink, T. Wedhorn, P. Ziegler, Algebraic Zip data, Documenta Math. 16 (2011), 253 -- 300. [PWZ2] R. Pink, T. Wedhorn, P. Ziegler, F -zips with additional structure, preprint August 2012, 49 pages, arXiv:1208.3547 [RaZi] M. Rapoport, T. Zink, Period spaces for p-divisible groups, Annals of Math. Studies 141, Princeton Univ. Press (1996). [SGA3] A. Grothendieck et al., Sch´emas en groupes, LNM 151, 152, 153, Springer 1970. [vdG] G. van der Geer, Cycles on the moduli space of abelian vari- eties, Moduli of curves and abelian varieties, Aspects Math., E33, Vieweg, Braunschweig (1999), 65 -- 89. [ViWd] E. Viehmann, T. Wedhorn, Ekedahl-Oort and Newton strata for Shimura varieties of PEL type, Math. Ann. 356 (4) (2013), 1493 -- 1550. 17 [Wd1] [Wd2] [Wo] [Zi] T. Wedhorn, Ordinariness in good reductions of Shimura varieties of PEL-type, Ann. sci. ´Ec. Norm. Sup. (4), 32, (1999), 575 -- 618. T. Wedhorn, Bruhat strata and F -zips with additional structures, preprint 2013, 31 pages, arXiv:1302.6715, to appear in Munster Journal of Mathematics. D. Wortmann, The µ-ordinary locus for Shimura varieties of Hodge type, preprint 2013, 34 pages, arXiv:1310.6444. T. Zink, A Dieudonn´e Theory for p-divisible groups, in: Class Field Theory, Its Centenary and Prospect pp. 1 -- 22, Advanced Studies in Pure Mathematics, Tokyo 2000. 18
1802.04463	4	1802	2019-10-29T03:02:55	qKZ/tRS Duality via Quantum K-Theoretic Counts	[ "math.AG", "hep-th", "math-ph", "math-ph", "math.QA", "math.RT" ]	We show that normalized quantum K-theoretic vertex functions for cotangent bundles of partial flag varieties are the eigenfunctions of quantum trigonometric Ruijsenaars-Schneider (tRS) Hamiltonians. Using recently observed relations between quantum Knizhnik-Zamolodchikov (qKZ) equations and tRS integrable system we derive a nontrivial identity for vertex functions with relative insertions.	math.AG	math	QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS PETER KOROTEEV AND ANTON M. ZEITLIN Abstract. We show that normalized quantum K-theoretic vertex functions for cotan- gent bundles of partial flag varieties are the eigenfunctions of quantum trigonometric Ruijsenaars-Schneider (tRS) Hamiltonians. Using recently observed relations between quantum Knizhnik-Zamolodchikov (qKZ) equations and tRS integrable system we derive a nontrivial identity for vertex functions with relative insertions. Contents Introduction tRS Difference Equations 1. 2. Quantum Knizhnik-Zamolodchikov equations and Nakajima varieties 2.1. Nakajima quiver varieties and their equivariant K-theory 2.2. Quasimaps, nonsingular and relative conditions 2.3. Vertices with Descendants, Fusion Operator and qKZ Equation 2.4. Relative vs Descendant Insertions 3. Vertex Functions 4. Trigonometric RS Difference Operators 4.1. 4.2. Vertex Functions from tRS Eigenfunctions 4.3. Example for T ∗P1 4.4. Quantum Toda Chain from Trigonometric RS Model 5. qKZ versus tRS 5.1. Notations and Conventions 5.2. Transfer matrix, Bethe ansatz and Hamiltonians 5.3. qKZ vs tRS 5.4. Geometric Meaning of the qKZ/tRS correspondence References 1 3 3 4 6 9 9 11 14 16 16 17 18 18 19 21 24 25 1. Introduction The fundamental work of Givental and Lee [GL01] on quantum K-theory opened a path to study this subject from the perspective of quantum integrable systems. The example they studied in detail is quantum K-theory of flag varieties, which turned out to be related to the many-body system known as relativistic Toda chain. They introduced certain gen- erating K-theory-valued functions, containing the corresponding enumerative data, known 1 2 PETER KOROTEEV AND ANTON M. ZEITLIN as Givental's J-functions, which were found to be eigenfunctions for the first nontrivial relativistic Toda Hamiltonian. Recent progress in understanding the relationship between enumerative geometry and integrability, inspired by the physics papers of Nekrasov and Shatashvili [NS09a, NS09b], was achieved by Okounkov and his collaborators [BMO11, MO12, Okob], in the study of enumerative geometry of symplectic resolutions (see e.g. [Okoc, Okoa]). For a wide class of symplectic resolutions, known as Nakajima quiver varieties, an ex- plicit relation between quantum K-theoretic counts, based on a theory of quasimaps devel- oped in [CFKM14], and q-difference equations, known to representation theory community as quantum Knizhnik-Zamolodchikov (qKZ) equations [FR92] and the related dynamical equations was discovered in [Okob, OS16]. We would like to mention that the Okounkov's approach to quantum K-theory and enumerative computations is substantially different from the original Givental's approach and while in this paper we will establish correspondences in certain explicit examples of K-theoretic counts, as we have done already on the level of quantum rings [KPSZ], the precise relation between the two approaches still requires further understanding. The relationship between quantum K-theoretic counts and quantum difference equations lead to several new results, in particular to proper understanding of the ring structure of K-theory of Nakajima variety [PSZ], [KPSZ] as Bethe algebra for various kinds of spin chain models, explicit combinatorial formulae for quantum multiplication and many other observations. A very important recent discovery of Aganagic and Okounkov [AO] is the explicit construction and geometric interpretation of solutions of qKZ equations. At the same time, close kinship between integrable models constructed using quantum groups and many-body systems of Toda, Ruijsenaars-Schneider and Calogero-Moser type was inspired by work of Cherednik [Che94] and Matsuo [Mat92] resulting in numerous follow-up contributions, see e.g. [MTV14, MTV12, Zab15, BLZZ16, ZZ17] which focused their study on integrable models as well as in physics literature [GK13, BKK15]. Recently the quantum K-theory rings of cotangent bundles to flag varieties, as well as flag varieties themselves, were reinterpreted as algebras of functions on the Lagrangian subvari- eties in the phase spaces of classical trigonometric Ruijsenaars-Schneider (tRS) integrable system and relativistic Toda lattice correspondingly [KPSZ]. Motivated by the renewed interest to difference equations in quantum K-theory (see e.g. [ACT]), this paper is aiming to provide a proper analogue of Givental's J-functions for Nakajima quiver varieties (we shall focus on cotangent bundles to partial flag vari- eties of type A) and establish their relation to solutions of quantum difference Knizhnik- Zamolodchikov equations. We would like to reiterate that K-theory vertex functions which will be computed in this paper are defined differently than Givental's J-functions (quasimaps vs. stable maps). The structure of this paper is as follows. In Section 2 we discuss necessary background in quantum K-theory, in particular in quantum K-theory of Nakajima varieties, bare and capped vertex functions, fusion matrix, and the relation to qKZ equations along the lines of [Okob, OS16, AO], providing detailed references to the corresponding sections and state- ments in loc. cit. QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 3 In Section 3 we discuss a particular integral representation for bare vertex functions of cotangent bundles for partial flag varieties, which were studied in [KPSZ] (see [PSZ] for the detailed study of cotangent bundles to Grassmannians). In Section 4 we prove that the normalized vertex function without insertions is the eigenfunction of trigonometric Ruijsenaars-Schneider difference operators (also known as Macdonald operators). To the best of our knowledge, Section 4 of this manuscript contains the first mathematical proof of this fact. The paper by Givental and Lee on quantum K-theory of complete flag varieties states that their J-functions satisfy quantum difference Toda (qToda) relations. We shall discuss the connection between the tRS and qToda systems. Finally, in Section 5 we return to qKZ equations. First we shall discuss the trigonomet- ric version of the duality between solutions of qKZ equations and the tRS eigenfunctions, thereby generalizing [ZZ17]. Then, by combining that with the results of Section 4, we obtain a nontrivial identity between normalized vertex function and vertex functions with relative insertions. This provides a geometric meaning for the qKZ/tRS correspondence. Acknowledgements. We would like to thank P. Pushkar and A. Okounkov for helpful discussions. The work of A.M.Zeitlin is partially supported by Simons Collaboration Grant, Award ID: 578501. P.K. acknowledges support of IH´ES and funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (QUASIFT grant agreement 677368). 2. Quantum Knizhnik-Zamolodchikov equations and Nakajima varieties In this section we provide a brief review the Okounkov's quasimap approach1 to the quantum equivariant K-theory of Nakajima quiver varieties. One of the important advan- tages of this approach is the emergence of quantum Knizhnik-Zamolodchikov equation and its dynamical counterpart. In this section we present a short review of quantum K-theory, where we will introduce all necessary objects to guide the reader towards qKZ equations, which will be our main focus. This is unlike our previous work [PSZ], [KPSZ], where the emphasis was made on the dual dynamical equation. For the comfort of the reader, we provide detailed references to the appropriate sections of Okounkov's lectures [Okob] as well as to recent paper by Aganagic and Okounkov [AO]. We also encourage the reader to consult introductory part of [PSZ]. 2.1. Nakajima quiver varieties and their equivariant K-theory. A quiver is a col- lection of vertices and oriented edges connecting them. A framed quiver is a quiver, where the set of vertices is doubled, and each of the vertices in the added set has an edge going from it to the vertex, whose copy it is. 1We note that quasimap approach to quantum K-theory introduced by Okounkov is different from the [KPSZ]), the exact original approach of Givental. While certain K-theoretic counts do coincide (see e.g. relationship between such approaches is still unknown. 4 PETER KOROTEEV AND ANTON M. ZEITLIN A representation of a framed quiver is a set of vector spaces Vi, Wi, where Vi correspond to original vertices, and Wi correspond to their copies, together with a set of morphisms between these vertices, corresponding to edges of the quiver. Hom(Vi, Vj) denote the linear space of quiver representation with dimension vectors v and w, where vi = dim Vi, wi = dim Wi and Qij stands for the incidence matrix of the quiver, For a given framed quiver, let M := Rep(v, w) = Pi∈I Hom(Wi, Vi) ⊕Pi,j∈I Qij ⊗ i.e. the number of edges between vertices i and j. Then the group G =Qi GL(Vi) acts on this space in an obvious way. Thus we have µ : T ∗M → Lie(G)∗, to be the moment map and let L(v, w) = µ−1(0) be the zero locus of the moment map. The Nakajima variety X corresponding to the quiver is an algebraic symplectic reduction X = N (v, w) = L(v, w)//θG = L(v, w)ss/G, depending on a choice of stablity parameter θ ∈ ZI (see [Gin] for a detailed definition). The group Yi.j GL(Qij) ×Yi GL(Wi) × C× acts as automorphisms of X, coming form its action on Rep(v, w). Here C× scales cotan- gent directions with weight and therefore symplectic form with weight −1. Let us denote by T the maximal torus of this group. Our main example will be the space of partial flags, i.e. we are considering the following quiver of type An with single framing vertex: wn−1 v1 v2 . . . vn−1 The stability condition is chosen so that maps Wn−1 → Vn−1 and Vi → Vi−1 are sur- jective. For the variety to be non-empty the sequence v1, . . . , vn−1, wn−1 must be non- decreasing. The fixed points of this Nakajima quiver variety and the stability conditions are classified by chains of subsets V1 ⊂ . . . ⊂ Vn−1 ⊂ Wn−1. For a Nakajima quiver variety X one can define a set of tautological bundles Vi = L(v, w)ss × Vi/G, Wi = L(v, w)ss × Wi/G. From this construction it follows that all bundles Wi are topologically trivial. It is known that tensorial polynomials of these bundles and their duals generate KT(X). For more details regarding quiver varieties, one can consult with e.g. [Gin], introduction to [MO12] or Section 4 of [Okob]. 2.2. Quasimaps, nonsingular and relative conditions. The proper reference for this subsection would be Sections 4.3. and 6 of [Okob]. QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 5 Definition 2.1. A quasimap f from C ∼= P1 to X f : C 99K X Is a collection of vector bundles Vi on C of ranks vi together with a section of the bundle f ∈ H 0(C, M ⊕ M ∗ ⊗ ), satisfying µ = 0, where M =Xi∈I Hom(Wi, Vi) ⊕ Xi,j∈I Qij ⊗ Hom(Vi, Vj), so that Wi and Qij are trivial bundles of rank wi and µ is the moment map. Here is a trivial line bundle with weight introduced to have the action of T on the space of quasimaps. The degree d of a quasimap is a the vector of degrees of bundles Vi. We say that a quasimap is stable if f (p) is also a stable (belongs to the orbits giving rise to Nakajima variety) for all but finitely many points of C. If f (p) is not stable, we will say that the quasimap f is singular at p. For a point on the curve p ∈ C we have an evaluation map from the moduli space of stable quasimaps to the quotient stack evp : QMd → L(v, w)/G defined by evp(f ) = f (p). Note that the quotient stack contains X as an open subset corresponding to locus of semistable points: X = µ−1 ss (0)/G ⊂ L(v, w)/G. A quasimap f is called nonsingular at p if f (p) ⊂ X. In short, we conclude that the open subset QMd nonsing p ⊂ QMd of stable quasimaps of degree d2, nonsingular at the given point p, is endowed with a natural evaluation map: (2.1) QMd nonsing p evp−→ X which sends a quasimap to its value at p. The moduli space of relative quasimaps QMd is a resolution of evp (or compactification), meaning we have a commutative diagram: relative p QMd relative p evp eevp X QMd nonsing p with a proper evaluation map eevp from QMd relative p to X. The construction of this reso- lution and the moduli space of relative quasimaps is explained in [Okob]. The main idea of this construction is to allow the base curve to change in cases, when the relative point becomes singular. When this happens we replace the relative point by a chain of non-rigid projective lines, such that the endpoint and all the nodes are not singular. Similarly, for nodal curves, we do not allow the singularity to be at the node, and if that happens we instead paste in a chain of non-rigid projective lines. 2Degree d of a quasimap is understood as the vector of degrees of bundles Vi, and thus can be considered as an element of H2(X, Z). 6 PETER KOROTEEV AND ANTON M. ZEITLIN These moduli spaces have a natural action of maximal torus T, lifting its action from X. When there are at most two special (relative or marked) points and the original curve is P1 we extend T by additional torus C× q , which scales P1 such that the tangent space T0P1 has character denoted by q. We call the full torus by Tq = T × C× q . The moduli spaces of quasimaps constructed above have a perfect deformation-obstruction theory [CFKM14]. This allows one to construct a tangent virtual bundle T vir, a virtual structure sheaf bOvir and a virtual canonical bundle [Okob]. curve C = P1 keeping two points p1 and p2 fixed. The spaces QMd nonsing p2 and QMd relative p2 admit the action of extra torus C× q which scales One can also define the twisted version of quasimaps. Let σ : C× → A be a cocharacter of subtorus A = Ker ⊂ T , preserving symplectic form, which determines the twist of a quasimap to X. As T is acting on Wi, σ determines Wi over the curve C as bundles associated to O(1). For every fixed point x ∈ X σ there is a constant twisted quasimap f (c) = x for any c ∈ C. There is an important formula defining the nontrivial degree of such quasimap (see Section 4.3.12 of [Okob]). Namely, given a fixed point of subtorus of a ∈ A, means that there is an action on line bundles Li = det(Vi) (where the tautological bundles Vi on X correspond to the spaces Vi), which produces a locally constant map µ : X A → P ic(X)∨ ⊗ A∨, which is compatible with the restrictions to the subgroups of A. Then where the pairing is between characters and cocharacters. deg(f ≡ x) := hµ(x), − ⊗ σi, 2.3. Vertices with Descendants, Fusion Operator and qKZ Equation. The topics in this subsection are spread over Sections 7-10 of [Okob], so we will provide more detailed references throughout. The primary interest of this work is to study certain virtual quasimap counts, which are pushforwards of the twisted virtual structure sheaf bOvir on the moduli spaces of stable quasimaps accompanied by extra insertions. Given a point p on the curve, one can pick a fiber Vip. The collection of all such fibers forms a sheaf over the corresponding QM space, as quasimap varies over this moduli space. Let τ be the tensor polynomial of V{i}, V ∗ {i}, representing the K-theoretic class τ (V{i}) over X. We denote by τ ( V{i} ) the corresponding class over the quasimap moduli space, obtained via the application of τ to the collection V{i}p. With that in mind, let us make the following definition (see Sections 7.2 and 7.4 of [Okob] as well as [PSZ], [KPSZ] for details). Definition 2.2. The elements V (τ )(z) = ∞Xd=~0 zdevp2,∗(cid:16)QMd nonsing p2,bOvir ⊗ τ ( V{i}(cid:12)(cid:12)p1 )(cid:17) ∈ KTq (X)loc[[z]] QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 7 bV (τ )(z) = ∞Xd=~0 zdevp2,∗(cid:16)QMd relative p2,bOvir ⊗ τ ( V{i}(cid:12)(cid:12)p1 )(cid:17) ∈ KTq (X)[[z]] are called bare and capped vertex with descendants τ correspondingly. The space QMd nonsing p2 is not proper (the condition of non-singularity at a point is an open condition), but the set of Tq-fixed points is, hence the bare vertex is singular at q = 1. The following statement is known for capped vertices [Smi]. quantum parameters z. Theorem 2.3. The power series bV (τ )(z) is a Taylor expansion of a rational function in The operator which relates capped and bare vertices, is known as capping operator (see Section 7.4.4. of [Okob]) and is defined as the following class in the localized K-theory: (2.2) Ψ(z) = (X)loc[[z]] . Note that one can define the following bilinear form (·, ·) on KT(X) via the equivariant Euler characteristic ∞Xd=0 relative p1 nonsing p2 zdevp1,∗ ⊗ evp2,∗(cid:16)QMd ,bOvir(cid:17) ∈ K ⊗2 2(cid:17) , (F, G) = χ(cid:16)F ⊗ G ⊗ K − 1 Tq where map ξ : H 2(X, C) ⊗ Lie(T ) → End K(X T ) ⊗ C, extends by linearity the map of the logarithm of the tensor multiplication by a line bundle: L → log(L ⊗ −) to H 2(X, C), while the shifted Kahler parameters are defined as follows: shifted = zd(−−1/2q)−hdet T 1/2,di, where d ∈ H2(X, Z). zd (2.3) (2.4) twisted by a root of the canonical class (the latter exists for Nakajima quiver varieties). This twisting is introduced due to the similar twisting of the structure sheaf bOvir (see [Okob]). Because of this pairing, Ψ(z) can be considered as an operator acting from the second to the first copy of KTq (X)loc[[z]]. In other words, the capped vertex with descendent τ is a result of applying of the capping operator to the bare vertex bV (τ )(z) = Ψ(z)V (τ )(z). The above operator satisfies quantum difference equations. For convenience one has to normalize the fusion operator as follows (see section 8.2 of [Okob]): (2.5) bΦ(a, z) = Ψ(a, z)Θ(a), Θ(a) = ϕ((1 − q−1)T 1/2X)Ξ . Here T 1/2X = M −Pi∈I End(Vi, Vi) is the polarization bundle for the Nakajima variety and ϕ : K(X) → K(X) ⊗ Z[[q]] defined as ϕ(F) =Q(fi)∞ = Λ•(cid:16)F ⊗Pi≥0 qi(cid:17), where fi are Chern roots of F. The operator Ξ is defined as follows (see subsection 8.2.3 of [Okob]): Ξ = exp(cid:18) ξ(log zshifted, log t) log q (cid:19) , 8 PETER KOROTEEV AND ANTON M. ZEITLIN The operator Ξ satisfies the following difference equation: Ξ(qσa) = (zshifted)hµ(·),−⊗σiΞ, where we recognize the term with a degree of twisted constant quasimap from the end of Section 2.2. We can now write down the following important Theorem (see Proposition 6.5.30, Section 8.2.17, Theorem 8.2.20 and Corollary 8.2.24 of [Okob]): Theorem 2.4. The normalized capping operator bΦ(z) is the solution of the quantum dif- ference equations: where Sσ = bΦ(qσa) = SσbΦ(a) zd evp1,∗ ⊗ evp2,∗(cid:18)QM σ,d relative p1 relative p2 ∞Xd=0 where QM σ,d stands for the moduli space of σ-twisted quasimaps of degree d and G is the gluing operator defined as follows: (X)loc[[z]], Tq ,bOvir(cid:19) G−1 ∈ K ⊗2 ,bOvir(cid:17) ∈ K ⊗2 Tq (X)[[z]]. G = ∞Xd=0 zd evp1,∗ ⊗ evp2,∗(cid:16)QMd relative p1 relative p2 Remark. Note, that along with the difference operators in equivariant parameters there is an additional commuting set of difference operators for bΦ in z-variables (see second equation in Theorem 8.2.20 of [Okob]), i.e. which played a pivotal role in [PSZ, KPSZ]. bΦ(qLz) = MLbΦ(z), Assume the action σ : C× → A is such that w = aw′ + w′′, u ∈ C×. Then X σ = Gv′+v′′=v N (v′, w′) × N (v′′, w′′) . In section 9 of [Okob] the following maps in localized K-theory Stab±(a) : KT (X σ) → (2.6) given Nakajima variety is the composition of these maps3 KT (X) were introduced, such that the R-matrix for affine Lie algebra bg associated to a In the case of An quiver varieties,bg = bgl(n) and the R-matrix is the product of trigono- metric R-matrices (as it is proven in [OS16]), associated with bgl(n) which we will discuss Moreover, the following theorem (see Theorem 9.3.1 from [Okob] as well as Theorem 3.1 in detail in Sec. 5. R(a) = Stab−1 − Stab+ . of [OS16]) is true 3See [MO12] for the cohomological version. QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 9 Theorem 2.5. The shift operator Sσ is related to the R-matrix R(a) in the following way σ zv′ τ −1 R(a) = Stab−1 + τ −1 σ Sσ(a, z)Stab+, where τσf (a) = f (qσa), identifying the difference equation from Theorem 2.4 with the quantum Knizhnik-Zamolodchikov equation. Remark. We note, that here we use a modified z-variables as in [OS16], namely, we should use the shifted variable zv′ (−1)codim/2, i.e. in components, zi → (−1)2κi zi, so that 2κi = w′ − Cv′, where C is the Cartan matrix of quiver. → zv′ 2.4. Relative vs Descendant Insertions. Finally, the last important topic we want to discuss is how the matrix elements of the fusion matrix can be expressed using vertices with descendants. This problem was recently solved by Aganagic and Okounkov [AO]. Namely, for every element α ∈ KT (X) which serves as a relative insertion at the vertex, one constructs a localized descendent insertion, i.e. an element fα ∈ Q(T × TG/WG), where TG and WG are correspondingly maximal torus and Weyl group of G =Qi GL(Vi). While being a rational function, fα admits only specific kinds of denominators: fα = sα ∆ , such that sα ∈ Z(T × TG/WG) and (−)−kΛkLie(G) =Yi Yk,l(cid:16)1 − so that xi,r are grouped accordingly to G =Qi∈I GL(Vi). Therefore the following theorem of [AO] holds. ∆ =Xk xi,k xi,l(cid:17) , Theorem 2.6. The matrix elements of Ψ can be expressed as follows: Ψp,q = V (fq) , where p stands for p-th fixed point component of bare vertex with the descendant fq ∈ Q(T × TG/WG) as described above. p In this section we describe vertex functions for partial flag varieties which were recently computed using localization technique in [KPSZ]. 3. Vertex Functions It will be convenient to introduce the following notation: v′ i = vi+1 − vi−1, for i = 2, . . . , n − 2, v′ n−1 = wn−1 − vn−2, v′ 1 = v2. To describe the expression for the vertex one needs to take into account the fixed points of QMd nonsing p2. Each such a point is described by the data ({Vi}, {Wn−1}), where degVi = di, degWn−1 = 0. Each bundle Vi can be decomposed into a sum of line bundles Vi = O(di,1) ⊕ . . . ⊕ O(di,vi) (here di = di,1 + . . . + di,vi). For a stable quasimap with such data to exist the collection of di,j must satisfy the following conditions • di,j ≥ 0, • for each i = 1, . . . , n − 2 there should exist a subset in {di+1,1, . . . di+1,vi+1} of cardinality vi {di+1,j1, . . . di+1,jvi }, such that di,k ≥ di+1,jk . 10 PETER KOROTEEV AND ANTON M. ZEITLIN In the following we will denote the chamber containing such collections {di,j} as C. Now we are ready to write down the contributions for the entire vertex function: Theorem 3.1. Let p = V1 ⊂ . . . ⊂ Vn−1 ⊂ {a1, · · · , awn−1} (Vi = {xi,1, . . . xi,vi}) be a chain of subsets defining a torus fixed point p ∈ X T. Then the coefficient of the vertex function for this point is given by: V (τ ) (z) = Xdi,j ∈C where d = (d1, . . . , dn−1), di =Pvi p j=1 di,j, N (d) = v′ idi, zdqN (d)/2 EHG τ (xi,jq−di,j ), E = n−1Yi=1 viYj,k=1 {xi,j/xi,k}−1 di,j −di,k , G = vn−1Yj=1 wn−1Yk=1 {xn−1,j/ak}dn−1,j , H = n−2Yi=1 viYj=1 vi+1Yk=1 {xi,j/xi+1,k}di,j −di+1,k . Here {x}d = (/x, q)d (q/x, q)d (−q1/2−1/2)d, where (x, q)d = (1 − qix). d−1Yi=0 The same formula for the vertex can an be obtained using the following integral repre- sentation [AO, AFO]. Proposition 3.2. The p-th component of a bare vertex function is given by EintGintHintτ (s1, · · · , sk), 1 2πiαpZCp (3.1) V (τ ) p (z) = where Eint = n−1Yi=1 αp = where (3.2) log(z si,j e− dsi,j si,j ♯ ) log(si,j ) i log(q) n−1Yi=1 viYj=1 ϕ(cid:16) si,j si,k(cid:17) wn−1Yj=1 vn−1Yk=1 viYj,k=1 si,k(cid:17) , Gint = ϕ(cid:16) q si+1,j(cid:17) ϕ(cid:16) q vi+1Yj=1 n−2Yi=1 viYk=1 si+1,j(cid:17) , ϕ(cid:16) si,k EintGintHint(cid:12)(cid:12)(cid:12)si,j=xi,j ♯ ) log(si,j ) i log(q) viYj=1 n−1Yi=1 Hint = e− si,k log(z , sn−1,k ϕ(cid:16) q aj (cid:17) aj (cid:17) , ϕ(cid:16) sn−1,k ϕ(x) = (1 − qix) , ∞Yi=0 QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 11 and the contour Cp runs around points corresponding to chamber C and the shifted variable z♯ = z(− 1/2)v′ i. i=1 z♯ i , so that z♯ i = zi(− 1/2)det(P). Here z♯ =Qn−1 In [PSZ], [KPSZ] we found these formulas to be useful to study their asymptotics at q → 1 which lead to Bethe ansatz equations, producing the relations for the quantum K-theory ring. In this article, we however will leave parameter q intact. 4. Trigonometric RS Difference Operators Proposition 3.2 provides integral formulas for vertex functions Vp of X which depend on the choice of the contour Cp. In this section we study properties of integral (3.1) without explicitly specifying the contour. In particular, we shall demonstrate that for a properly chosen contour (3.1) solves quantum difference equations of the trigonometric Ruijsenaars- Schneider model. In this work we shall only study difference equations in equivariant parameters of X, see [Kor18] (Theorem 2.6). In full generality tRS Hamiltonians read4 (4.1) Tr(a) = XI⊂{1,...,n} I=r Yi∈I j /∈I t ai − aj ai − aj Yi∈I pi , where a = {a1, . . . , awn−1}, the shift operator pif (ai) = f (qai) and we denoted t = q . In order to understand how the above difference operators act on integrals of the form (3.1) we need to study in detail how they act on the ingredients of the integrand. In what follows we shall describe these actions for vertex functions of quiver variety X in question. The analysis for cotangent bundles to complete flag varieties was performed in [HR12] and in [BKK15]. Consider the following function (4.2) Hvn,vn+1(sn, sn+1) = vnYk=1 vn+1Yj=1 sn,k ϕ(cid:16) q sn+1,j(cid:17) ϕ(cid:16) sn,k sn+1,j(cid:17) , were sn = {sn,1, . . . , sn,vn} and sn+1 = {sn+1,1, . . . , sn+1,vn+1}. The following lemma de- scribes action of the difference operator pn,k (4.3) pn,kf (sn,1, . . . , sn,k, . . . sn,vn) = f (sn,1, . . . , qsn,k, . . . sn,vn) . on this function. Lemma 4.1. Let H be given in (4.2) then (4.4) pn,kHvn,vn+1(sn, sn+1) = vn+1Yj=1 sn+1,j − sn,k sn+1,j − q sn,k · Hvn,vn+1(sn, sn+1) . 4In this section we use slightly different normalization of the tRS operators than in [KPSZ]. 12 PETER KOROTEEV AND ANTON M. ZEITLIN Proof. From the definition of the function ϕ(x) (3.2) we get the identity pn,kϕ(sn,k) = (1 − sn,k)−1ϕ(sn,k), which we need to apply twice. (cid:3) Note another useful identity (4.5) p−1 n,kϕ(sn,k) := ϕ(q−1sn,k) = (1 − q−1sn,k)ϕ(sn,k) . Let us also define tRS operators with the opposite shift t ai − aj (4.6) T ′ r(a) = XI⊂{1,...,n} I=r Yi∈I j /∈I ai − aj Yi∈I p−1 i , where p−1 i f (ai) = f (q−1ai). Next, we need to prove another lemma which directly follows from Lemma 4.1. Lemma 4.2. Consider function (4.2) with vn+1 = vn. Then (4.7) Tr(sn)Hvn,vn (sn, sn+1) = T ′ r (sn+1) Hvn,vn(sn, sn+1) . Proof. The lemma follows from the direct computation after using the identity which was proven in [HR12] (4.8) t sn,i − sn,j sn,i − sn,j · vnYj=1 sn+1,i − sn,j sn+1,i − t sn,j j /∈I XI⊂{1,...,vn} I=r Yi∈I = XI⊂{1,...,vn} I=r Yi∈I j /∈I t sn+1,i − sn+1,j sn+1,i − sn+1,j · sn+1,i − sn,j sn+1,i − t sn,j . vnYj=1 (cid:3) Using the above lemma we can find generalizations of (4.7) to the cases when tRS operators act on function (4.2) with different labels vn and vn+1. The action can be expressed in terms of quantum dimensions of the irreducible representations of glvn+1−vn. Definition 4.3. Let ΛN resentation of glN . Then its quantum dimension is given by s be the s-th antisymmetric tensor power of the fundamental rep- (4.9) qdim(ΛN s ) = sΛN s (tN −1, . . . , t1−N ) , where t is the quantum parameter and sρ(x1, . . . , x2N −1) is symmetric Schur polynomial for partition ρ. Note that in the above definition the quantum dimensions of ΛN s are nothing but ele- mentary symmetric polynomial qdim(ΛN s ) = es(tN −1, . . . , t1−N ) which satisfies (4.10) es(tN , . . . , t−N ) = tses(tN −1, . . . , t1−N ) + ts−1−N es−1(tN −1, . . . , t1−N ) . We can now formulate QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 13 Lemma 4.4. Assume that vn+1 − vn ≥ 0. Then (4.11) Tr(sn) · Hvn,vn+1 (sn, sn+1) = where it is assumed that T ′ 0(sn+1) = 1. min(r,vn+1−vn)Xs=0 qdim(Λvn+1−vn s ) T ′ r−s (sn+1) · Hvn,vn+1(sn, sn+1) , Proof. We start with (4.8) where we replace vn by vn+1. Then we take the limit sn+1,j → ∞, where j = vn+1 − vn, vn+1 − vn + 1, . . . , vn+1 . Analogously to the previous lemma the desired identity follows after acting with the shift operators and applying (4.10). (cid:3) For the first tRS Hamiltonian T1 identity (4.11) simplifies as follows. Corollary 4.5. (4.12) T1(sn) · Hvn,vn+1 (sn, sn+1) =(cid:20)T ′ 1 (sn+1) + tvn+1−vn − t−vn+1+vn t − t−1 (cid:21) · Hvn,vn+1(sn, sn+1) . Note that (4.11) can be proven using this corollary by induction using (4.10). Equation (4.12) relates first tRS Hamiltonian with its conjugate of its decomposition into represen- tations of U (N − 1) to ΛN r into ΛN −1 Let us now study the action of tRS operators on function E. The following lemma can r ⊕ ΛN −1 r−1 . be easily proven using properties of function ϕ. Lemma 4.6. Let E be defined as (4.13) E(sn) = vnYj,k=1 sn,k(cid:17) ϕ(cid:16) sn,j ϕ(cid:16)t sn,j sn,k(cid:17) , then it satisfies the following difference relation for the inverse shift (4.14) p−1 n,kE(sn) = q−1sn,i − sn,k q−1tsn,i − sn,k t sn,i − sn,k sn,i − sn,k · E(sn) . vnYi,k=1 The important property of the tRS difference operators is that they are self-adjoint with respect to the measure dsn sn · E(sn) := dsn,i sn,i E(sn,i) on the Cartan subalgebra of U (vn). We shall prove this property in the following lemma. vnQi=1 Lemma 4.7. Let f and g be meromorphic functions of their arguments. Then, provided that contour C does not encounter any poles of these functions upon shift sn,k → q−1sn,k, the following identity holds (4.15) ZC dsn sn · E(sn)f (sn) [Tr(sn) · g(sn)] =ZC dsn sn · E(sn)(cid:2)T ′ r(sn) · f (sn)(cid:3) g(sn) . 14 PETER KOROTEEV AND ANTON M. ZEITLIN Proof. Consider sn,k where k ∈ I from the definition of difference operators Tr (4.1). Assuming that we do not hit any poles, we shift the contour of integration by sn,k → q−1sn,k only for k ∈ I. This operation can be expressed via acting with the inverse shift p−1 n,k on the integrand of the left hand side of (4.15) (4.16) ZC vnYi=1 dsn,i sn,i XI⊂{1,...,n} I=r "Yk∈I p−1 n,k · E(sn,i)f (sn)# ·Yi∈I j /∈I tq−1 sn,i − sn,j q−1sn,i − sn,j · g(sn) . Using (4.5) and (4.14) we arrive to the right hand side of (4.15). (cid:3) 4.1. tRS Difference Equations. Now we shall use the lemmas which we have just proven to construct a solution for the quantum difference tRS equations. First, let us change quantum parameters in K-theory as follows z♯ 1 = z♯ i = z♯ n−1 = , ζ1 ζ2 ζi ζi+1 ζn−1 ζn , i = 2, . . . , n − 2 . (4.17) Theorem 4.8. The following function constructed for the cotangent bundle to the partial flag variety X labelled by v1, . . . , vn−1, wn−1 (4.18) V(a, ~ζ) = log ζn Pn−1 i=1 log q log ai e 2πi ZC n−1Ym=1 vmYi=1 dsm,i sm,i E(sm,i) e− log ζm/ζm+1 ·log sm,i log q · vm+1Yj=1 Hvm,vm+1 (sm,i, sm+1,j) , where contour C is chosen in such a way that shifts of the contour s → q±1 encounter any poles, satisfies tRS difference relations s do not (4.19) Tr(a)V(a, ~ζ) = Sr(~ζ, t)V(a, ~ζ) , r = 1, . . . , wn−1 where function Sr is r-symmetric polynomial of the following Pn n−1−1ζn, . . . , t−v 1+1ζ1, . . . . . . , tv 1−1ζ1, . . . , t−v (4.20) ′ ′ {tv ′ k=1 ksk variables n−1+1ζn} , ′ ′ where v i = vi+1 − vi for i = 1, . . . n − 2 and v ′ n−1 = wn−1 − vn−1. Proof. First, we need to justify that contour C can be always chosen in such a way that its shifts do not result in any additional residues and that Lemma 4.7 can be applied. Indeed, in any given complex plane sn,i poles in the integrand of (4.18) are located at sn,i = σq−dn,i for some σ (different for each plane). The contour can be safely chosen to avoid the collision with poles. Indeed, suppose q is real so various poles of the integrand are located on lines which are parallel to the real axis. The contour is therefore chosen to go above and below the above string of poles. Since the contour is parallel to the real axis it won't be affected by the q-shift. QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 15 We prove the theorem by induction. The base of induction is A1 quiver with no Vi bundles, but merely constant bundle W1 or rank v1 (in other words, Grassmannian Gr0,v1). For this quiver the function Vp has no integration V(s1, ζ1) = e log ζ1 P v1 i=1 log q log s1,i , (4.21) which satisfies (4.22) Tr(s1)V(s1, ζ1) = er(cid:16)t 1−v1 2 , . . . , t 1 · V(s1, ζ1) , v1 −1 2 (cid:17) ζ r where er is elementary r-th symmetric polynomial of its arguments. This follows directly from the structure of tRS operators. Now let us assume that we for the quiver variety X with n − 2 nodes labelled by v1, . . . vn−2 relation (4.19) holds. Let us add another node of rank vn−1. Using this decomposition we can rewrite the following function (4.23) V(a, ~ζ (n)) = e log ζn P wn−1 i=1 log q log ai ZC(n−1) dE(sn−1,i) Hvn−1,wn−1 (sn−1,i, a) · V(sn−1, ~ζ (n−1)) , vn−1Yi=1 for the proper choice of the contour C (n−1) and where we indicate different numbers of parameters ζ in the arguments of the vertex functions by the corresponding superscripts. By acting on the integral with the tRS operator Tr(sn) (here we identify equivariant parameters a with Bethe roots sn) we get the following expression Tr(a)V(a, ~ζ (n)) = er(cid:16)t ZC(n−1) vn−1Yi=1 1−wn−1 2 , . . . , t wn−1−1 2 log ζn P wn−1 i=1 log q log ai (cid:17) ζ r n · e dE(sn−1,i)(cid:2)Tr(a) · Hvn−1,wn−1(sn−1,i, a)(cid:3) · V(sn−1, ~ζ (n−1)/ζn) . (4.24) Using Lemma 4.4 we can replace the action of the operator Tr(a) on variables aj by the right hand side of (4.11). Then, by employing Lemma 4.7 we can 'integrate by parts' each term in the resulting sum. Finally we use the inductive assumption about the eigenvalues Sr−s(~ζ (n−1)/ζn, t) of Tr−s(sn−1) to get the following formula for the eigenvalue of Tr(a): 2 2 min(r,v′ 1−wn−1 wn−1−1 (4.25) , . . . , ζnt Sr(cid:16)ζnt n)Xs=0 where for the first polynomial we used that Sr(cid:16)ζnt ner(cid:16)t (cid:17) · 2 1−wn−1 wn−1−1 ζ r that the above expression is equal to an r-symmetric polynomial of variables listed in (4.20). We shall illustrate this fact for the case when v′ n = 1, then the expression in (4.25) (cid:17) . It can be shown using properties of symmetric polynomials , . . . , t 2 qdim(cid:16)Λv′ s (cid:17) · Sr−s(~ζ (n−1)/ζn, t) , n 1−wn−1 2 , . . . , ζnt wn−1−1 2 (cid:17) = 16 reads (4.26) PETER KOROTEEV AND ANTON M. ZEITLIN ζ r n(cid:16)Sr(~ζ (n−1)/ζn, t) + Sr−1(~ζ (n−1)/ζn, t)(cid:17) = Sr(ζ (n)) , where we used (4.10) at the last step. The analogue of the above equality for higher v′ n can be shown by induction. (cid:3) 4.2. Vertex Functions from tRS Eigenfunctions. We can now demonstrate how to compute vertex functions from Proposition 3.1 using the general tRS solution (4.18) by properly specifying the contour of integration. In order to do that we need to understand how to identify each chamber C by choosing contour C in (4.18). The prescription goes as follows. We shall only pick poles of functions H in the inte- grand and ignore poles of E functions. It can be argued (see Section 3 of [BKK15]) that for a contour which encircles all poles of φ functions of the integrand only poles of H functions survive, whereas poles of E functions are cancelled by zeroes of H's. Poles of Hvn,vn+1(sn, sn+1) have the form sn,i/sn+1,k = q−dn,i for some nonnegative degrees dn,i. The contour can then be chosen to go only around the poles whose degrees dn,i satisfy the inequalities in the definition of chambers described in the beginning of Sec. 3. Once the integration contour C is properly chosen for a given fixed point p the integral from 3.1 can be readily evaluated. Theorem 4.9. Consider αp and V (1) as defined previously in Theorem 3.1. Then for each fixed point p of the maximal torus of X there is a contour C for which integral (4.18) evaluates to p (4.27) log ζn Pn−1 i=1 log q log ai V = e αpV (1) p . Let us illustrate this statement on a simple example. 4.3. Example for T ∗P1. The vertex function (3.1) for T ∗P1 for a trivial class τ = 1 reads (4.28) V (1) p = 1 2πiαpZCp ds s (z♯)− log s log q 2Yi=1 ϕ(cid:16)t s ai(cid:17) ϕ(cid:16) s ai(cid:17) , for the two fixed points p = {a1} and p = {a2}. The poles are given by s = apq−d for nonnegative d. By taking the residues we arrive to the q-hypergeometric function (4.29) V (1) p =Xd>0 (z♯)d 2Yi=1 =2φ1(cid:18)t, t ap a¯p , ap a¯p ; q; z♯(cid:19) . ap ai (cid:16) q (cid:16) ap ai ; q(cid:17)d ; q(cid:17)d Here a¯1 = a2 and a¯2 = a1. One can easily see that Weyl reflection of gl2 interchanges V (1) and V (1) 2 . 1 QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 17 The integral in (4.18) for T ∗P1 reads as follows (4.30) V = log ζ2 log a1a2 log q e 2πi ZC ds s (cid:18) ζ1 ζ2(cid:19)− log s log q 2Yi=1 s ϕ(cid:16) q ai(cid:17) ϕ(cid:16) s ai(cid:17) . The denominator has two semi-infinite strings of poles at s = aiq−d for i = 1, 2 and d ∈ Z≥0. In order to reproduce vertex functions we can pick the contour such that it encloses only the poles which start at a1 or the one which starts at a2 contour. After a straightforward calculation we find that (4.31) for each fixed point. V = e log ζ2 log a1a2 log q αpV (1) p Remark. Theorem 4.8 proves that the K-theory vertex function of the cotangent bundle to the partial flag variety in type A is an eigenfunction of the tRS difference operator. Using similar techniques, in particular Lemmas 4.1 -- 4.6, with slight modifications, it will be possible to demonstrate that K-theory vertex functions for a more generic quiver varieties of type A (so that each node of the quiver may have framing) describe the spectrum of a certain restrictions of the quantum tRS integrable systems (the construction involves nilpotent orbits of glN and Slodowy slices, see [GK13] Sections 2.7, 4). Some calculations in this direction were done in the physics literature (see [BKK15], Section 3). 4.4. Quantum Toda Chain from Trigonometric RS Model. The construction de- scribed in this section provides explicit construction for the spectrum of the tRS model in terms of integrals of the Mellin-Barnes type. Prior to our construction similar ideas were developed for difference Toda chains [KLSTS02, GLO10], see also geometric approach of [BF05, BF14, BF17]. Here we would like to comment on how to obtain the results in loc. cit. from our construction. It was shown in [KPSZ] that the integrals of motion of classical tRS integrable system played a role of relations in the quantum equivariant K-theory ring of the cotangent bundle to (complete) flag variety. It was also proven in the same paper that in the → ∞ limit the tRS integrals of motion become the q-Toda integrals of motion. Thus, the version of the quantum equivariant K-theory of flag varieties emerges in this limit, which magically repeats the results of [GL01], where a different approach to K-theoretic counts was used. Based on the computations of the current manuscript we can make a quantum analogue of that statement5. Indeed, (4.19) are quantum (read q-difference) equations of the tRS model and function V is their formal solution, which can be used to derive K-theoretic vertex functions for X (4.27). Therefore we expect that in the limit → ∞ K-theoretic vertex functions for the cotangent bundle to the complete flag variety T ∗G/P will transform 5Word 'quantum' is used here in two different contexts -- quantum K-theory in the previous paragraph and quantum integrable system. 18 PETER KOROTEEV AND ANTON M. ZEITLIN into Givental J-functions for the corresponding flags G/P 6. A similar statement is expected in cohomology, however, in K-theory it demonstrates a non-trivial connection between the theory of quasimaps which we used to study T ∗G/P and theory of stable maps which was used by Givental et al. As an illustration let us take X = T ∗P1 again. Taking the limit → ∞7 (or t → 0) in (4.29) we obtain (4.32) V (1) p →2φ1(cid:18)0, 0, ap a¯p ; q; z♯(cid:19) =:1φ0(cid:18) ap a¯p ; q; z♯(cid:19) = ∞Xk=0 (z♯)k (cid:16) ap a¯p , q(cid:17)k , (q, q)k where p and ¯p denote two fixed points, which up to a constant coincides with the Givental J-function for P1 from [GL01]. We would like to emphasize that, despite the coincidences above, Givental's and Ok- ounkov's approaches to quantum K-theoretic counts are conceptually different and more works needs to be done in order to understand the exact relationship. 5. qKZ versus tRS In this final section we shall study in detail qKZ equations for the cotangent bundle to partial flag variety and their solutions. After reminding the reader about the exact cor- respondences between K-theoretic and representation-theoretic data, we shall discuss the derivation of the relationship between the qKZ equations and the tRS eigenvalue problem along the lines of the paper by Zabrodin and Zotov [ZZ17]. We shall provide the gener- alization of their argument in application to the trigonometric RS model (The loc. cit. offers a proof only for the rational case). In the end we show that this analysis leads to a nontrivial relation for K-theoretic vertex functions with relative insertions, which is the second main result of this paper. 5.1. Notations and Conventions. In Sec. 2 we described Nakajima quiver varieties N (v, w) of type A. Let us now consider the union of the cotangent bundles for partial flag varietiesFv N (v, w) with fixed framing. Its localized quantum K-theory, as a vector space, is spanned by K-theory classes corresponding to fixed points of the maximal torus. The corresponding vector spaces can be identified with the standard weight subspaces in (5.1) V = V (a1) ⊗ · · · ⊗ V (awn−1) = ⊕{sa}V({sa}), of the evaluation parameter [CP94] and the weight parameters sa are the eigenvalues of where V (a) is an n-dimensional evaluation representation of U(bgln) where a plays a role Sa =Pn The identification of the weight subspaces V({sa}) with K-theoretic data is as follows: k=1 e(k) aa . (5.2) V({sa}) = K loc T (N (v, w)) , 6Here P is a parabolic subgroup of G. In this paper G = U (wn−1) and P = U (wn−1 − vn−1) × · · · × U (v2 − v1) × U (v1) 7The so-called Inosemtsev limit [Ino89] in the literature on integrable systems (see more on this in [GKKV19]). QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 19 provided that sa = va − va−1, where vn ≡ wn−1 and s1 = v1. Let R be a trigonometric R-matrix: R(x) : V (xy) ⊗ V (y) → V (xy) ⊗ V (y) , R(x) = eaa ⊗ eaa + nXa=1 2 x−1Xa<b 2 1 2 − − 1 2 x − − 1 1 x − x−1 2 x − − 1 1 2 x−1 eaa ⊗ ebb + nXa6=b (xeab ⊗ eba − x−1eba ⊗ eab). (5.3) Let Rij(x) be the notation for the R-matrix operator acting in the i-th and j-th cites of V. Then let us define qKZ operator as follows: (5.4) K (q) i : V → V , K (q) i = Ri i−1(cid:18) aiq ai−1(cid:19) . . . Ri 1(cid:18) aiq a1(cid:19) Z (i)Ri wn−1(cid:18) ai awn−1(cid:19) . . . Ri i+1(cid:18) ai ai+1(cid:19) , where Z (i) is an auxiliary diagonal matrix diag(ζ1, . . . , ζn) acting on i-th cite, so that shifted as in the Remark after Theorem 2.5 parameters zi = ζi ζi+1 , i = 1, . . . , n − 1. The qKZ equations which we have described from geometric perspective in Sec. 2.3 represent a family of difference equations of the following kind (5.5) piΦ = K (q) i Φ, Φ ∈ V , where as before pi is a shift operator, so that piΦ(. . . , ai, . . . ) = Φ(. . . , qai . . . ). We refer the reader to [Okob, OS16, AFO] for additional technical details on the map Stab, which we briefly reviewed in Sec. 2.3. 5.2. Transfer matrix, Bethe ansatz and Hamiltonians. Proposition 5.1. The permutation form of the R-matrix is as follows: (5.6) where (5.7) R(x) = P + x − x−1 2 x − − 1 1 2 x−1 (I − P ), P = nXa=1 eaa ⊗ eaa + 1 2 nXa>b eab ⊗ eba + − 1 2 eab ⊗ eba nXa<b and P = P 1 is a permutation operator. It is very useful to rescale the R-matrix by introducing another notation (5.8) eR(x) = 1 2 x − − 1 x − x−1 2 x−1 R = I − P + 1 2 x − − 1 x − x−1 2 x−1 P . 20 PETER KOROTEEV AND ANTON M. ZEITLIN (5.9) Let us define the transfer matrix T (x), acting on V, obtained by tensoring it with auxiliary cite V0: and then taking the trace of the product of the R-matrices: T (x) = TrV0(cid:16)eR0wn−1(cid:0)x/awn−1(cid:1) . . .eR01 (x/a1) Z (0)(cid:17) , where V0 is the n-dimensional evaluation representation (auxiliary space). The following results are known since 1980s using techniques usually referred to as algebraic Bethe ansatz, see e.g. [BLZZ16]. Theorem 5.2. The eigenvalues Λ(x) of matrix T (x) are given by the following formula (5.10) Λ(x) = ζ1 x 1 2 − ak− 1 x − ak 2 wn−1Yk=1 v1Yα=1 x− 1 α 2 − σ1 x − σ1 α 1 2 + ζi nXi=2 vi−1Yα=1 x 1 α − 1 2 2 − σi−1 x − σi−1 α x α− 1 2 1 2 − σi x − σi α , viYα=1 where vi are integers (so that v0 = vwn−1 = 0) which denote the number of Bethe roots, i.e. the solutions of the following equations: ζ1 ζi (5.11) wn−1Yi=1 vi−1Yβ=1 σ1 α − ai σ1 α − ai = ζ2 σ1 α − σ1 β α − σ1 σ1 β σ1 α − σ2 β α − σ2 σ1 β , α − σi−1 σi β α − σi−1 σ1 β = ζi+1 α − σi σi β α − σi σi β α − σi+1 σi β α − σi+1 σi β v1Yβ6=α viYβ6=α r2Yβ=1 ri+1Yβ=1 One can re-expand the transfer matrix using quantum nonlocal Hamiltonians, which play an important role in the construction (cf. formulae (3.7)-(3.9) of [BLZZ16]): Proposition 5.3. T (x) has the following simple pole expansion: (5.12) T (x) = C + 1 2 − − 1 2 2 wn−1Xk=1 x2 + a2 k x2 − a2 k Hk, where C, Hi are the operators on V and the eigenvalues hi of Hi are given by the following formula: (5.13) hi = ζ1 1 2 − − 1 2 2 ai wn−1Yk6=i 1 2 − ak− 1 ai − ak 2 v1Yα=1 ai− 1 α 2 − σ1 ai − σ1 α 1 2 . Proof. The first part of the proposition follows from the asymptotic behavior of the transfer matrix T (x), when x → ∞, 0 based on the behavior of the R-matrices as well as its trigonometric structure. The second part follows directly from Theorem 5.2. The operators Hi are known as nonlocal Hamiltonians of the XXZ model. (cid:3) Proposition 5.4. i) The relation between the qKZ operators and the nonlocal Hamiltonians is as follows: (5.14) Hi = ai wn−1Yj6=i 1 2 − aj− 1 ai − aj 2 K (1) i . QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 21 ii)The sum of all hamiltonians can be expressed in the following way using Cartan gener- ators acting on V: (5.15) wn−1Xk=1 Hk = ζa nXa=1 Sa 2 − − Sa 2 − − 1 1 2 2 . Proof. Based on the polar structure of the expansion of T (x) in Proposition 5.3 we conclude that Hi are proportional to residues at points z = a2 k, where z = x2. In the definition of the transfer matrix (5.9) the only term in the product of R-matrices containing such a pole is R0,k(x/ak) (this is in accord with the polar structure of the normalized R-matrix (5.8)). The corresponding coefficient of the polar term is proportional to the permutation matrix. Computing the residue and inserting it into the trace of (5.9) in place of R0,k(x/ak) we obtain that Hi = eRi i−1(cid:18) ai ai−1(cid:19) . . .eRi 1(cid:18) ai By removing normalization we obtain (5.14), thereby proving i). The proof of ii) goes as follows. Notice that the asymptotic expansions of the eigenvalues of the transfer matrix are such that awn−1(cid:19) . . .eRi i+1(cid:18) ai a1(cid:19) Z (i)eRi wn−1(cid:18) ai ai+1(cid:19) . 2 )Xk 2 − − 1 hk 1 Λ(∞) − Λ(0) = ( to be a solution of the qKZ equation of a given weight parametrized by {sa}. Let us define ℓ(J) as the minimal number of permutations required to put tuple J = (j1, . . . .jwn−1 ) into 1 ≤ j1 ≤ j2 ≤ · · · ≤ jwn−1 ≤ n. The following proposition holds. Proposition 5.5. The vector E, such that E =PJ erties: ℓ(J ) 2 EJ satisfies the following prop- i−1 iE = E, P Ri−1 i(x)E = Pi−1 iE . Proof. One can show that vector E is obtained as a sum over all possible applications of P to the standard vector e⊗s1 . This proposition immediately follows. (cid:3) ⊗swn−1 wn−1 1 ⊗ · · · ⊗ e from Proposition 5.3 and at the same from Bethe ansatz equations Λ(∞) − Λ(0) = ζa( Sa 2 − − Sa 2 ) , nXa=1 thus proving the second part of the proposition. (cid:3) 5.3. qKZ vs tRS. Let EJ = ej1 ⊗ · · · ⊗ ejwn−1 Consider , where J = (j1, . . . , jwn−1) is multi-index. (5.16) ΦJ EJ ∈ V({sa}) Φ =XJ 22 PETER KOROTEEV AND ANTON M. ZEITLIN Let us define (5.17) Υ = hE, Φi =XJ ℓ(J ) 2 ΦJ . We will now show that Υ is the eigenfunction of the tRS Hamiltonians. Theorem 5.6. The following formulae hold (5.18) (5.19) i) ii) piΥ =DE, K (1) i ΦE , wn−1Xi=1 wn−1Yi6=j ai 2 − aj − 1 ai − aj 2 1 piΥ = nXa=1 ζa sa 2 − − sa 2 − − 1 1 2 2 ! Υ . Proof. To show i) one needs to prove that hE, K (1) dependence on q is involved in the expression for K (q) before multiplication by Z (i). Also, one can check that Proposition 5.5 gives: i Φi is independent on q. Notice that the only in the product of R-matrices i DE, P i i−1ρE =DE, ρE for any vector ρ ∈ V({sa}). Property Pi i−1P i−1 i−2Pi i−1 together with the sec- ond part of Proposition 5.5 reduce that product of R-matrices in K (q) to the product of permutation matrices P , i.e. i i−2 = P (cid:28)E, Ri i−1(cid:18) aiq ai−1(cid:19) . . . Ri 1(cid:18) aiq a1(cid:19) ρ(cid:29) =DE, Pi i−1 . . . Pi 1ρE for any vector ρ from V({sa}). In order to prove ii), we multiply i) by ai−aj ai−aj Proposition 5.4. and sum over i. After that we use (cid:3) The operator, whose eigenfunction we calculated in ii) of the above theorem is the first tRS Hamiltonian. To show that Υ is indeed an eigenfunction of the entire family of tRS Hamiltonians we need the following statement, which is a generalization of i) of Theorem 5.6. Proposition 5.7. Action of the products of difference operators T (i) as follows: q on Υ can be expressed (5.20) dYk=1 wn−1Yr6=ik aik − ar aik − ar dYk=1 pik Υ =DE, Hi1 . . . HidΦE Proof. In order to prove that one just has to use the same principle as in Theorem 5.6 and to prove this identity: (5.21) dYk=1 pik Υ =DE, K (1) i1 . . . K (1) id ΦE , QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 23 namely, use the properties from Proposition 5.5 when moving q-shifted R-operators to the left of twisted matrices Z (i). Then multiplying on the appropriate coefficients as in Proposition 5.4 we obtain the statement of the theorem. (cid:3) Let us use now an important relation proven in [BLZZ16]: Proposition 5.8. The following combinatorial formula holds for the sums of products of Hamiltonians: (5.22) X1≤i1<···<ik≤wn−1 Hi1 . . . Hik Y1≤α<β≤k where C(aiα/aiβ ) = 1 2 − − 1 2 2 !k X1≤i1<···<ik≤N λi1 . . . λik , C(x) = (x x − x−1 1 2 − x−1− 1 2 )(x− 1 2 − x−1 1 2 ) and λim are eigenvalues of a certain operator which depend only on and {zi}. From now on, to align the notations with the previous section, we rescale parameters in tRS Hamiltonians (4.1) as t → −1 and multiply each Hamiltonian Td by a prefactor (5.23) d 2 Td . bHd = In order to put these Hamiltonians in touch with Proposition 5.7, we need the following statement, which can be proved by direct calculation. Proposition 5.9. The ordered expression for tRS Hamiltonians is given by the following formula: (5.24) bHd = X1≤i1<···<id≤n dYk=1Yj6=ik aik 1 2 − aj − 1 aik − aj 2 Y1≤m<n≤d C(aim/ain) pik dYk=1 Using this expression and then combining Proposition 5.8 with Proposition 5.7 we arrive to the main theorem. Theorem 5.10. Function Υ, associated to the solution of qKZ equation via formula (5.17) is an eigenfunction of tRS Hamiltonians Hd. Remark. We would like to make a comment about the new role which the qKZ equation plays for integrable systems. The above theorem relates solutions of the qKZ equation with solutions of the system of tRS equations. The qKZ equation (5.5) represents a mod- ern viewpoint on diagonalization of the XXZ spin chain Hamiltonians. The connection between tRS model and the qKZ equation/XXZ chain was first observed in Cherednik [Che94] and Matsuo [Mat92]. Such correspondence naturally appears on the level of gauge theory, see [GK13]. In loc. cit. the tRS model was considered classical (hence the name quantum/classical duality), whereas in this paper we study quantum tRS Hamiltonians, thereby promoting it to quantum/quantum duality between tRS and XXZ models. 24 PETER KOROTEEV AND ANTON M. ZEITLIN 5.4. Geometric Meaning of the qKZ/tRS correspondence. Finally we shall discuss the geometric outcome of formula (5.17). First, let us recall the geometric meaning of as follows equations according to Theorem 2.5. In order to establish a relationship with solutions Φ + operator in order to put the shift operator solutions of qKZ equations. The columns of matrix bΦ (2.5) are the solutions of the qKZ from (5.16) we need to precomposebΦ with Stab−1 + bΦ reads in its R-matrix form according to Theorem 2.4. The resulting operatoreΦ = Stab−1 eΦ =neΦp,qo =(cid:26)V where we denoted the corresponding insertions of stable basis elements in vertex functions as (Stab−1 + f )q and Θp are the eigenvalues of matrix Θ (2.5) at fixed points p. · Θp(cid:27) , ((Stab−1 p (5.25) + f )q) By combining the results of the previous two sections, we can now formulate the following theorem which establishes a geometrical relationship between solutions of qKZ equations eΦ and eigenfunctions of tRS Hamiltonians V(a, ~ζ) (4.19). we can state the following theorem. Recall that after formula (5.16) we defined permutation length ℓ(J). Using this notation Theorem 5.11. Let X be a cotangent bundle to partial flag variety of type A. Then its non-normalized vertex function is given by the following weighted sum of vertices with insertions (5.26) V(a, ~ζ) = c · Θ ·Xq ℓ(q) 2 V ((Stab−1 + f )q) , where c is a constant which does not depend on the equivariant parameters and ℓ(q) is defined using the natural identification of fixed points with the points in weight subspaces of representations of gl(n) and shifting the sign of z-variables as prescribed in Sec. 2.3. Proof. We can identify the components of the solution of the qKZ equations with nor- malized vertices with insertions as in (5.25) and Υ (5.17), and the eigenfunctions of tRS Hamiltonians, with the normalized vertices of Theorem 4.9. One can then see that the normalization coefficients are identical. Thus (5.17) is equivalent to (5.26) for each choice of the contour C defining the vertex. (cid:3) Example. Let us look at T ∗P1 again. The normalized vertex function which is the eigenvalue of the tRS operator T1V = (ζ1 + ζ2)V reads (5.27) V = log ζ2 log a1a2 log q e 2πi ZC ds s (cid:18) ζ1 ζ2(cid:19)− log s log q Let us analyze how the two part of T1 act on V (5.28) T1V = ζ2 log ζ2 log a1a2 log q e 2πi ZC ds s (cid:20) ta1 − a2 a1 − a2 a1 − s a1 − s + ta2 − a1 a2 − a1 s 2Yi=1 ϕ(cid:16) q ai(cid:17) ϕ(cid:16) s ai(cid:17) . ζ2(cid:19)− log s a2 − s (cid:21)(cid:18) ζ1 log q a2 − s 2Yi=1 s ϕ(cid:16) q ai(cid:17) ϕ(cid:16) s ai(cid:17) . log ζ2 log a1a2 e T1V = ζ2 1(s) + T ′ 0(s)](cid:18) ζ1 ζ2(cid:19)− log s log q s 2πi log q ZC ds s 2Yi=1 ai(cid:17) ϕ(cid:16) q ai(cid:17) [T ′ ϕ(cid:16) s + 1(cid:19) V = (ζ1 + ζ2)V , = ζ2(cid:18) ζ1 ζ2 (5.29) QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 25 We observe in the square brackets the sum of two terms, which up to a normalization, reproduce stable envelopes for T ∗P1 . On the other hand, using the Corollary 4.5 we get which illustrates that the elements of the stable basis are summed to a constant. References [ACT] David Anderson, Linda Chen, and Hsian-Hua Tseng, On the quantum K-ring of the flag man- ifold, 1711.08414. [AFO] Mina Aganagic, Edward Frenkel, and Andrei Okounkov, Quantum q-Langlands Correspondence, 1701.03146. [AO] Mina Aganagic and Andrei Okounkov, Quasimap counts and Bethe eigenfunctions, arXiv:1704.08746. [BF05] Alexander Braverman and Michael Finkelberg, Finite difference quantum Toda lattice via equi- variant K-theory, Transform. Groups 10 (2005), no. 3-4, 363 -- 386, math/0503456. [BF14] , Semi-infinite Schubert varieties and quantum K-theory of flag manifolds, JAMS 27 (2014), 1147 -- 1168, 1111.2266. [BF17] A. Braverman and M. Finkelberg, Twisted zastava and q-Whittaker functions, J. Lond. Math. Soc. 96 (2017), 309 -- 325, 1410.2365. [BKK15] Mathew Bullimore, Hee-Cheol Kim, and Peter Koroteev, Defects and quantum seiberg-witten geometry, JHEP 05 (2015), 095, 1412.6081. [BLZZ16] M. Beketov, A. Liashuk, A. Zabrodin, and A. Zotov, Trigonometric version of quantum-classical duality, Nuclear Physics B 903 (2016), 1201.3990. [BMO11] Alexander Braverman, Davesh Maulik, and Andrei Okounkov, Quantum cohomology of the Springer resolution, Adv. Math. 227 (2011), 421 -- 458, 1001.0056. [CFKM14] Ionut Ciocan-Fontanine, Bumsig Kim, and Davesh Maulik, Stable quasimaps to GIT quotients, J. Geom. Phys. 75 (2014), 17 -- 47, 1106.3724. [Che94] Ivan Cherednik, Integration of quantum many body problems by Affine Knizhnik-Zamolodchikov equations, Adv. Math. 106 (1994), 65 -- 95. [CP94] V. Chari and A. Pressley, A guide to quantum groups, CUP (1994). [FR92] Igor Frenkel and Nicolai Reshetikhin, Quantum affine algebras and holonomic difference equa- tions, Communications in Mathematical Physics 146 (1992). [Gin] Victor Ginzburg, Lectures on Nakajima's Quiver Varieties, 0905.0686. [GK13] Davide Gaiotto and Peter Koroteev, On Three Dimensional Quiver Gauge Theories and Inte- grability, JHEP 1305 (2013), 126, 1304.0779. [GKKV19] Alexander Gorsky, Olesya Koroteeva, Peter Koroteev, and Arkady Vainshtein, Dimensional transmutation in 1d quantum hydrodynamics, to appear (2019). [GL01] A. Givental and Y. P. Lee, Quantum k-theory on flag manifolds, finite-difference toda lattices and quantum groups (2001), math/0108105. [GLO10] Anton Gerasimov, Dimitri Lebedev, and Sergey Oblezin, On q-deformed gl(l+1)-Whittaker function, Comm. Math. Phys. 294 (2010), 97 -- 119, 0803.0145. [HR12] M. Hallnas and S. Ruijsenaars, Kernel functions and Backlund transformations for relativis- tic Calogero-Moser and Toda systems, Journal of Mathematical Physics 53 (December 2012), no. 12, 123512, 1206.3786. 26 PETER KOROTEEV AND ANTON M. ZEITLIN [Ino89] V. I. Inozemtsev, The finite toda lattices, Comm. Math. Phys. 121 (1989), no. 4, 629 -- 638. [KLSTS02] S. Kharchev, D. Lebedev, and M. Semenov-Tian-Shansky, Unitary representations of the modular double, and the multiparticle q-deformed Toda chains, Com- Uq(sl(2, R)), mun.Math.Phys. 225 (2002), 573 -- 609, hep-th/0102180. [Kor18] Peter Koroteev, A-type Quiver Varieties and ADHM Moduli Spaces (2018), 1805.00986. [KPSZ] Peter Koroteev, Petr P. Pushkar, Andrey Smirnov, and Anton M. Zeitlin, Quantum K-theory of Quiver Varieties and Many-Body Systems, 1705.10419. [Mat92] Atushi Matsuo, Integrable connections related to zonal spherical functions., Inventiones math- ematicae 110 (1992), no. 1, 95 -- 122. [MO12] D. Maulik and A. Okounkov, Quantum Groups and Quantum Cohomology (2012), 1211.1287. [MTV12] Evgeny Mukhin, Vitaly Tarasov, and Alexander Varchenko, KZ Characteristic Variety as the Zero Set of Classical Calogero-Moser Hamiltonians, SIGMA 072 (2012), 1201.3990. [MTV14] E. Mukhin, V. Tarasov, and A. Varchenko, Bethe algebra of Gaudin model, Calogero-Moser space and Cherednik algebra, IMRN 2014 (2014), 1174 -- 1204, 0906.5185. [NS09a] Nikita A. Nekrasov and Samson L. Shatashvili, Quantum integrability and supersymmetric vacua, Prog.Theor.Phys.Suppl. 177 (2009), 105 -- 119, 0901.4748. 21 pp., short version II, con- ference in honour of T.Eguchi's 60th anniversary. [NS09b] , Supersymmetric vacua and Bethe ansatz, Nucl.Phys.Proc.Suppl. 192-193 (2009), 91 -- 112, 0901.4744. [Okoa] Andrei Okounkov, Enumerative geometry and geometric representation theory, 1701.00713. [Okob] [Okoc] , Lectures on K-theoretic computations in enumerative geometry, 1512.07363. , On the crossroads of enumerative geometry and geometric representation theory, 1801.09818. [OS16] Andrei Okounkov and Andrey Smirnov, Quantum difference equation for Nakajima varieties (2016), 1602.09007. [PSZ] Petr P. Pushkar, Andrey Smirnov, and Anton M. Zeitlin, Baxter Q-operator from quantum K-theory, 1612.08723. [Smi] Andrey Smirnov, Rationality of capped descendent vertex in K-theory, 1612.01048. [Zab15] A. Zabrodin, Quantum spin chains and integrable many-body systems of classical mechanics, Springer Proceedings in Physics (2015), 29 -- 48, 1409.4099. [ZZ17] A. Zabrodin and A. Zotov, QKZ-Ruijsenaars correspondence revisited, Nucl. Phys. B 922 (2017), 113 -- 125, arXiv:1711.01036. Peter Koroteev Department of Mathematics, University of California at Davis, Mathematical Sciences Building, One Shields Ave, University of California, Davis, CA 95616; Department of Mathematics, University of California Berkeley, 970 Evans Hall, University of California, Berkeley, CA 94720-3840, [email protected] https://math.berkeley.edu/~pkoroteev QKZ/TRS DUALITY VIA QUANTUM K-THEORETIC COUNTS 27 Anton M. Zeitlin, Department of Mathematics, Louisiana State University, Baton Rouge LA 70803-4918; IPME RAS, V.O. Bolshoj pr., 61, 199178, St. Petersburg [email protected], http://math.lsu.edu/~zeitlin
1909.04543	1	1909	2019-09-10T14:57:43	Moduli Problem of Hitchin Pairs over Deligne-Mumford Stack	[ "math.AG" ]	We define the moduli problem of Hitchin pairs over Deligne-Mumford Stack and prove this moduli problem is represented by a separated and locally finitely presented algebraic space, which is considered as the moduli space of Hitchin pairs over Deligne-Mumford stack.	math.AG	math	MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK HAO SUN Abstract. We define the moduli problem of Hitchin pairs over Deligne-Mumford Stack and prove this moduli problem is represented by a separated and locally finitely presented algebraic space, which is considered as the moduli space of Hitchin pairs over Deligne-Mumford stack. 1. Introduction The Hitchin pair was introduced by Hitchin in 1987 [Hit87], which is also known as the Higgs bundle. Numerous mathematicians made great contributions to the construction of the moduli space of Hitchin pairs in the past thirty years. After Hitchin's paper, Niture constructed the moduli space of semistable Hitchin pairs over smooth curves [Nit91], and Simpson constructed the moduli space of semistable Hitchin pairs over smooth projective varieties [Sim94]. Combined with the study of parabolic bundle [MS80], people constructed the moduli space of parabolic Higgs bundle [Yok93]. It is well-known that the construction of the parabolic bundle comes from the bundle over orbifold. If we consider the orbifold as stacks (or as a special type of Deligne-Mumford stack), a natural idea is how to construct the moduli space of Hitchin pairs over Deligne-Mumford stack. In this paper, we partially answer this question by proving the existence of the moduli space of Hitchin pairs over a Deligne-Mumford stack. Let X → S be a separated, locally finitely-presented morphism from a Deligne-Mumford stack X to an algebraic space S. Let G be a coherent OX -module. Olsson and Starr defined the quotient functor Q(G/X /S) := Quot(G, X , S) and proved that the quotient functor is represented by an algebraic space which is separated and locally finitely presented over S [OS03]. Let F be a locally finitely-presented quasi-coherent OX -module in Q(G/X /S), and we fix a line bundle (locally free sheaf with rank one) L over X , which is considered as the twisted bundle. An L-twisted Higgs field Φ on the quasi-coherent sheaf F is a homomorphism Φ : F → F ⊗ L. An L-twisted Hitchin pair over X is a pair (F , Φ), where F is a locally finitely-presented quasi-coherent sheaf over X and Φ is an L-twisted Higgs field. We consider the following moduli problem (functor) MX ,L,Higgs : (Sch/S)op → Set for L-twisted Hitchin pairs over X as follows. For each T ∈ Sch/S, define MX ,L,Higgs(T ) = {(FT , ΦT ) FT ∈ Q(G/X /S)(T ), ΦT : FT → FT ⊗ p∗ X L}, where pX : X ×S T → X is the natural projection. The main result of this paper is the following theorem. Theorem. 2.2 Let X → S be a separated, locally finitely-presented morphism from a Deligne-Mumford stack X to an algebraic space S. Fix a line bundle L on X . The moduli problem MX ,L,Higgs is represented by an algebraic space which is separated and locally finitely presented. ∗MSC2010 Class: 14A20 †Key words: Deligne-Mumford stack, Hitchin pair, moduli problem 1 2 HAO SUN To prove the main result, we use a theorem by Artin in 1968 [Art69, Theorem 5.3] (see Theorem In fact, Olsson and Starr also used this theorem to prove that the quotient functor is 3.1 in §3). representable by an algebraic space [OS03]. Artin's theorem is the key point to prove Theorem 2.2. In §2, we review some basic definitions and properties about Deligne-Mumford stack, define the moduli problem for Hitchin pairs and state our main result Theorem 2.2. §3 is devoted to the proof of Theorem 2.2. In §3.1, we review Artin's theorem [Art69, Theorem 5.3] and give a brief overview of the proof. In §3.2 and §3.3, we prove the main theorem in this paper. Moreover, we construct a well- defined deformation and construction for the moduli problem MX ,L,Higgs, which is a generalization of the infinitesimal deformation of Higgs bundle [BR94]. Acknowledgments. The author would like to thank Georgios Kydonakis and Lutian Zhao for helpful discussions. 2. Definition and Statement of the Main Result 2.1. Definition. In this section, we review the definition of Deligne-Mumford stack and coherent sheaves over an algebraic stack. Details can be found in [Ols16]. Let S0 be a scheme, which is the spectrum of an algebraically closed field. Let Sch/S0 be the category of schemes over S0. Let f : E → F be a morphism of sheaves on Sch/S0 with respect to the ´etale topology. The morphism f is representable by schemes if for every S0-scheme T and morphism T → F , the fiber product E ×F T is a scheme. An algebraic space X over S0 is a functor X : (Sch/S0)op → Set such that (1) X is a sheaf with repsect to the ´etale topology. (2) ∆ : X → X ×S0 X is representable by schemes. (3) There exists an S0-scheme U and a surjective ´etale morphism U → X . . Sheaves on an algebraic space X is defined locally with respect to the ´etale topology of Sch/S0. Consider a local chart U → X , an ´etale morphism with U a scheme. Let RU = U ×X U , which is a scheme by the representability of the diagonal morphism. Denote by s, t : RU → U the source and target map. Let (RU ⇒ U )´et be the set of pairs (FU , ǫU ), where FU is a sheaf on U and is an isomorphism of sheaves on RU . Now let U, V be two schemes, and let fU : U → X , fV : V → X ǫU : s∗FU → t∗FU be two ´etale morphisms. Let h : U → V be a morphism of schemes such that the following diagram commutes U fU h X V fV A sheaf F on X consists of a collection of pairs {(FU , ǫU )} for each ´etale morphism fU : U → X along with isomorphisms ah : FU → h∗FV , where FU , FV are sheaves over U , V respectively. A sheaf F is coherent (resp. quasi-coherent, locally free) sheaf if the local data FU is coherent (resp. quasi-coherent, locally free) for each ´etale morphism U → X . A sheaf F = {(FU , ǫU )} is an OX -module, if FU is an OU -module for each ´etale morphism U → X . It is easy to check that the category of sheaves over an algebraic space X is equivalent to the category of pairs (FU , ǫU ), where U is a scheme and U → X is a surjective ´etale morphism. Such a surjective ´etale morphism always exists by the definition of algebraic space. Under this definition, the category of sheaves over X is independent of the choice of the surjective ´etale morphism U → X (see [Ols16, §7]). MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK 3 A stack X : (Sch/S0)op → Set is a functor with groupoid as fiber (fibered category) satisfying the descent condition ([Ols16, §4.6]). A morphism of stacks f : X → Y is representable if for every S-scheme T and morphism T → Y, the fiber product is an algebraic space. An algebraic stack X over scheme S0 is a stack satisfying the following conditions: X ×Y T (1) The diagonal is representable. ∆ : X → X ×S0 X (2) There exists a smooth surjective morphism U → X with U a scheme. Moreover, if there exists a ´etale surjection U → X , the algebraic stack X is a Deligne-Mumford stack. 2.2. Moduli Space of Twisted Hitchin Pairs over Deligne-Mumform Stack. In this section, we give the definition of the moduli problem of twisted Hitchin pairs over Deligne-Mumford stack and state the main result, Theorem 2.2 of this paper. Let S be an algebraic space, which is locally of finite type over S0, where S0 is the spectrum of an algebraically closed field as we defined in §2.1. Locally, the algebraic space S is a scheme of finite type over Z with respect to the ´etale topology. Denote by (Sch/S)´et the category of S-schemes with respect to the ´etale topology. In other words, the objects in (Sch/S)´et are ´etale morphism from schemes to S. In the rest of the paper, we use the notation Sch/S instead of (Sch/S)´et. Let X → S be a separated, locally finitely-presented morphism from a Deligne-Mumford stack X to S. Let G be a coherent OX -module. We define the functor Q(G/X /S) := Quot(G, X , S) : (Sch/S)op → Set as follows. For each S-scheme T , define XT as X ×S T and GT the pullback of G to XT . Define Q(G/X /S)(T ) to be the set of OXT -module quotients GT → FT such that (1) FT is a locally finitely-presneted quasi-coherent OXT -module; (2) FT is flat over T ; (3) the support of FT is proper over T . The functor Q(G/X /S) is called the quotient functor. Artin proved that the quotient functor Q(G/X /S) is represented by a separated and locally finitely- presented algebraic space over S when X is an algebraic space [Art69]. Olsson and Starr generalized this result to Deligne-Mumford stack [OS03]. Theorem 2.1 (Theorem 1.1 in [OS03]). With respect to the above notation, the functor Q(G/X /S) = Quot(G, X , S) is represented by an algebraic space which is separated and locally finitely presented over S. Let F be a locally finitely-presented quasi-coherent OX -module in Q(G/X /S), and we fix a line bundle (locally free sheaf with rank one) L over X , which is considered as the twisted bundle. An L-twisted Higgs field Φ on the quasi-coherent sheaf F is a homomorphism Φ : F → F ⊗ L. An L-twisted Hitchin pair over X is a pair (F , Φ), where F is a locally finitely-presented quasi-coherent sheaf over X and Φ is an L-twisted Higgs field. We consider the following moduli problem (functor) MX ,L,Higgs : (Sch/S)op → Set for L-twisted Hitchin pairs over X as follows. For each T ∈ Sch/S, define MX ,L,Higgs(T ) = {(FT , ΦT ) FT ∈ Q(G/X /S)(T ), ΦT : FT → FT ⊗ p∗ X L}, where pX : X ×S T → X is the natural projection. Note that the definition of the moduli problem MX ,L,Higgs depends on the choice of the quotient functor Q(G/X /S). But, this definition can be extended to the category, more precisely the stack, of 4 HAO SUN coherent sheaves CohX over X . Nironi proved that a certain family of quotient functors form a smooth open atlas of the coherent sheaves CohX [Nir09, §2]. Thus the moduli problem MX ,L,Higgs can be naturally extended to the stack of coherent sheaves CohX over X . Based on this property, we neglect the quotient functor in the definition of MX ,L,Higgs. Now we are ready to state the main theorem in this paper. Theorem 2.2. Let X → S be a separated, locally finitely-presented morphism from a Deligne-Mumford stack X to an algebraic space S. Fix a line bundle L on X . The moduli problem MX ,L,Higgs is represented by an algebraic space which is separated and locally finitely presented. This theorem will be proved in the next section. 3. Proof of Theorem 2.2 3.1. Background and Overview of the Proof. We first review the deformation theory defined in [Art69] and give the statement of Theorem 5.3 in [Art69], which is the key to prove Theorem 2.2. Then, we go over some basic ideas of the proof of Theorem 2.1 that the quotient functor Q(G/X /S) is represented by a separated and locally finitely presented algebraic space [OS03, Theorem 1.1]. Finally, we give a brief overview of the proof of Theorem 2.2 at the end of this subsection. Let S be an algebraic space. An infinitesimal extension of an OS-algebra A is a surjective map of OS-algebras A′ ։ A such that the kernel M = ker(A′ → A) is a finitely generated nilpotent ideal. Let F be a contravariant functor from Sch/S to sets (moduli problem). Let A0 be a noetherian OS-domain. We prefer to use the notation F (A0) instead of F (Spec A0). The deformation situation is defined as a triple (A′ → A → A0, M, ξ) where A′ → A → A0 is a diagram of infinitesimal extension, M = ker(A′ → A) a finite A0-module and ξ ∈ F (A0). Let ξ be an element in F (A0). As a contravariant functor (for S-schemes), we have a natural map F (A) → F (A0). Denote by Fξ(A) the set of elements in F (A) whose image is ξ ∈ F (A0). The deformation theory we consider in this paper is described in [Art69, Definition 5.2]. A defor- mation theory for F consists of the following data and conditions (1) A functor associates to every triple (A0, M, ξ) an A0-module D = D(A0, M, ξ), and to every map of triples (A0, M, ξ) → (B0, N, η) an linear map D(A0, M, ξ) → D(B0, N, η). (2) For every deformation situation, there is an operation of the additive group of D(A0, M, ξ) on Fξ(A′) such that two elements are in the same orbit under the operation if and only if they have the same image in Fξ(A), where Fξ(A′) is the subset of F (A′) of elements whose image in F (A0) is ξ. Theorem 3.1 (Theorem 5.3 in [Art69]). Let F be a functor on (Sch/S)op. Given a deformation theory for F , then F is represented by a separated and locally of finite type algebraic space over S, if the following conditions hold: (1) F is a sheaf for the f ppf -topology and F is locally of finite presentatiion. (2) (Inverse Limits) Let ¯A be a complete noetherian local OS-algebra with residue field of finite type F ( ¯A/mn) is over S and let m be the maximal idea of ¯A. Then the canonical map F ( ¯A) → lim injective, and its image is dense in lim ← F ( ¯A/mn). ← (3) (Seperation) (a) Let A0 be a geometric discrete valuation ring, which is a localization of a finite type OS- algebra with residue field of finite type over OS. Let K, k be its fraction field and residue field respectively. If ξ, η ∈ F (A0) induce the same element in F (K) and F (k), then ξ = η. (b) Let A0 be an OS-integral domain of finite type. Let ξ, η ∈ F (A0). Suppose that there is a dense set S in Spec(A0) such that ξ = η in F (k(s)) for all s ∈ S. Then ξ = η on a non-empty open subset of Spec(A0). (4) (Deformation) MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK 5 (a) The module D = D(A0, M, ξ) commutes with localization in A0 and is a finite module when M is free of rank one. (b) The module operates freely on Fξ(A′) when M is of length one. (c) Let A0 be an OS-integral domain of finite type. There is a non-empty open set U of Spec(A0) such that for every closed point s ∈ U , we have D ⊗A0 k(s) = D(k, M ⊗A0 k(s), ξs) . (5) (Obstruction) Suppose we have a deformation situation (A′ → A → A0, M, ξ). (a) Let A0 be of finite type and M of length one. Let B′ A′ B A A0 A0 be a diagram of infinitesimal extensions of A0 with B′ = A′ ×A B. If b ∈ F (B) is an element lying over ξ whose image a ∈ F (A) can be lifted to F (A′), then b can be lifted to F (B′). (b) A0 is a geometric discrete valuation ring with fraction field K and M free of rank one. K the localizations of A, A′ respectively. If the image of ξ in F (AK ) can K ), then its image in F (A0 ×K AK) can be lifted to F (A0 ×K A′ Denote by AK , A′ be lifted to F (A′ K). (c) With the same notation in 5(b). Let M be a free module of rank n and ξ ∈ F (A). Suppose that for every one-dimensional quotient M ∗ K ) is obstructed, where A′ K. Then there is a non-empty open set U of Spec(A0) such that for every quotient ǫ : M → M ∗ of length one with support in U , the lifting of ξ to F (A∗) is obstructed, where A′ → A∗ → A denotes the resulting extension. K → AK is the extension determined by M ∗ K of MK the lifting of ξK to F (A∗ K → A∗ Olsson and Starr used Theorem 3.1 to prove that the quotient functor Q(G/X /S) is represented In other words, the quotient functor by a separated and locally finitely presented algebraic space. Q(G/X /S) satisfies conditions (1) to (5) in Theorem 3.1. Now we go back to the moduli problem for L-twisted Higgs bundle MX ,L,Higgs. In this section, we use M := MX ,L,Higgs to simplify the notation for the moduli problem we are interested in. Note that the problem of representability of M is ´etale local on S. Therefore we may assume that S is an affine scheme and of finite type over Spec(Z) as we explained at the beginning of §2.2. It is easy to check that the functor M satisfies the first condition. We will prove conditions (2) and (3) in §3.2. The deformation and obstruction theory will be discussed in §3.3. Condition (4) will be proved in §3.3.1 by constructing a well-defined deformation theory for M. The obstruction property will be discussed in §3.3.2. 3.2. Inverse Limit and Separation. We prove the inverse limit condition and separation condition in this subsection. Olsson and Starr proved that the quotient functor Q(G/X /S) is represented by pa separated, locally finitely-presented algebraic space over S [OS03, Theorem 1.1]. In other words, the functor Q(G/X /S) preserves the inverse limit and satisfies the separation condition. We use these properties of the quotient functor Q(G/X /S) to prove the properties of inverse limit and separation of the functor M. To prove a functor F satisfying the inverse limit condition, we have to prove that the map F ( ¯A) → F ( ¯A/mn) is injective and for any (ξn) ∈ lim F ( ¯A/mn) there is an element ξ′ ∈ F ( ¯A) which induces lim ← ξ1 ∈ F ( ¯A/m2). Let X = lim(X ⊗A A/mn). There is a natural morphism j : X → X . This morphism induces the following map ← j : Q(G/X /S)( ¯A) → lim ← Q(G/X /S)( ¯A/mn), F → j∗F . 6 HAO SUN In other words, let (Fn) This map is injective and has a dense image by [OS03, Theorem 1.1]. Q(G/X /S)( ¯A/mn). There is an element F ′ ∈ Q(G/X /S)( ¯A) such that F ′ in- be an element in lim ← duces F1 ∈ Q(G/X /S)( ¯A/m2). The same argument holds for the sheaf End(F ). Given an element Q(G/X /S)( ¯A/mn), there is an element End(F ′) ∈ Q(G/X /S)( ¯A) such that End(F ′) End(Fn) ∈ lim ← induces End(F1) ∈ Q(G/X /S)( ¯A/m2). Note that the morphism j : X → X also induces the following map j : M( ¯A) → lim ← M( ¯A/mn), (F , Φ) → (j∗F , j∗Φ), where F ∈ Q(G/X /S)( ¯A) and Φ ∈ End(F ) ⊗ L. The map j is clearly injective. Given any element M( ¯A/mn), where Fn ∈ Q(G/X /S)( ¯A/mn) and Φn ∈ End(Fn) ⊗ L, there exists ((Fn, Φn))n≥1 ∈ lim ← F ′ ∈ Q(G/X /S)( ¯A) such that F ′ induces F1. Thus we can find an element Φ′ ∈ End(F ′) ⊗ L such that Φ′ induces Φ1. Therefore the map j : M( ¯A) → lim M( ¯A/mn) has a dense image and the functor M satisfies the inverse limit condition. ← Now we are going to prove the separation property. With the same notations as in Theorem 3.1 (3), let ξ = (Fξ, Φξ) and η = (Fη, Φη) be two elements in M(A0) inducing the same element in M(K) and M(k). By Theorem 2.1, the quotient functor Q(G/X /S) satisfies the separation condition. Thus we have Fξ = Fη, which also implies that End(Fξ) ⊗ L = End(Fη) ⊗ L. Therefore we have ξ = η. The same argument works for the condition (3b). This finishes the proof of the separation condition. 3.3. Deformation and Obstruction Theory. With respect to the same notation as in §3.1, let ξ = (F , Φ) ∈ M(A0). The key object we will study in §3.3 about the deformation and obstruction theory is Mξ(A0[M ]), where Mξ(A0[M ]) is the set of elements in M(A0[M ]) whose restriction to A0 is ξ. The multiplication of the ring A0[M ] is defined as (a, m)(a′, m′) = (aa′, am′ + a′m). Clearly, M is a nilpotent ideal in A0[M ]. Note that the conditions of the deformation theory and obstruction theory ((4) and (5) in Theorem 3.1) are local property on both X and S. Thus we can take a local chart UX → X of X and a local chart US → S of S, where UX and US are schemes. We have the following diagram UX ×S US X ×S US UX X US S Thus we work on the separated and locally finitely-presented morphism UX ×S US → US locally. By the definition of Deligne-Mumford stack, UX ×S US is an algebraic space. Thus we may assume that S is an affine scheme and X is an algebraic space when we do the deformation (§3.3.1) and obstruction theory (§3.3.2). At the end of the setup of the deformation and obstruction theory, we review the statement of the well-known five lemma, which will be used frequently in this subsection. Lemma 3.2 (Five Lemma). Assume that all objects below are in an abelian category. A a A′ B b B′ C c C ′ D d D′ E e E′ If the rows are exact, b and d are isomorphisms, a is an epimorphism and e is a monomorphism, then c is an isomorphism. MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK 7 3.3.1. Deformation Theory. By the definition of the deformation theory in §3.1, we have to construct an A0-module D(A0, M, ξ) for each triple (A0, M, ξ), where ξ ∈ M(A0). In this section, we calculate the A0-module Mξ(A0[M ]) and prove that this module is the correct deformation theory for M. Biswas and Ramanan calculated the infinitesimal deformation theory for M, i.e. Mξ(C[ε]), ξ ∈ M(C) [BR94]. We generalize their approach to the deformation theory in this paper. Let us consider a special case first. Let A′ = A0[M ] := A0 ⊕M . Let (F , Φ) be an element in M(A0). Define F ′ = F ×Spec A0 Spec A′. Abusing the notation, we consider F ′ as F ⊕ F [M ]. For a section s of End(F ) ⊗ M , the corresponding automorphism of F ′ is denoted by 1 + s. Moreover, if v + w is a section of End(F ′) ⊗ L′, where L′ is the pull-back of L under the projection Spec A′ → Spec A0, we have ρ(1 + s)(v + w) = v + w + ρ(s)(v), where ρ is the natural action of End(F ) on itself. The deformation complex C • follows M (F , Φ) is defined as C • M (F , Φ) : C0 M (F ) = End(F ) ⊗ M e(Φ) −−−→ C1 M (F ) = End(F ) ⊗ L ⊗ M, where the map e(Φ) is given by e(Φ)(s) = −ρ(s)(Φ). If there is no ambiguity, we omit the notations M , F , Φ and use the following notation C • : C0 = End(F ) ⊗ M e(Φ) −−−→ C1 = End(F ) ⊗ L ⊗ M for the deformation complex. Now we are ready to calculate Mξ(A0[M ]). The following proposition is a generalization of Theorem 2.3 in [BR94]. Proposition 3.3. Let ξ = (F , Φ) be an L-twisted Hitchin pair in M(A0). The set Mξ(A0[M ]) is isomorphic to the hypercohomology group H1(C •), where C • is the complex C • : C0 = End(F ) ⊗ M e(Φ) −−−→ C1 = End(F ) ⊗ L ⊗ M, where e(Φ)(s) = −ρ(s)(Φ) is defined as above. Proof. Let U = {Ui = Spec(Ai)} be an ´etale covering of X by open affine schemes. The covering U of X also gives an ´etale covering {Ui × Spec(A0)} of X × Spec(A0). To be precise, the product Ui × Spec(A0) is taken over S, i.e. Ui ×S Spec(A0). We omit the base scheme S to simplify the notation. Define Ui[M ] = Ui × Spec(A0[M ]). Set End(F ) ⊗ M Ui[M] = C0 i , End(F ) ⊗ L ⊗ M Ui[M] = C1 i , i and C1 where C0 C1) to Uij = Ui T Uj. We consider the following Cech resolution of C •: i are A0-modules. Similarly, modules C0 ij (resp. C1 ij ) are resctrictions of C0 (resp. 8 0 0 0 HAO SUN 0 0 C0 d0 0 P C0 i d0 1 P C0 ij d0 2 ... e(Φ) e(Φ) e(Φ) C1 d1 0 P C1 i d1 1 P C1 ij d1 2 ... 0 0 0 We calculate the first hypercohomology H1(C •) from the above diagram. Let Z be the set of pairs (sij, ti), where sij ∈ C0 i satisfying the following conditions: ij and ti ∈ C1 (1) sij + sjk = sik as elements of C0 (2) ti − tj = e(Φ)(sij ) as elements of C1 ij . ijk. Let B be the subset of Z consisting of elements (si − sj, e(Φ)(si)), where si ∈ C0 Z/B. i . Clearly, H1(C •) = Given an element (sij , ti) ∈ Z, we shall construct a L-twisted Higgs bundle (F ′, Φ′) on X × Spec(A0[M ]) such that F ′X ×Spec(A0) ∼= F and Φ′X ×Spec(A0) ∼= Φ. For each Ui[M ], there is a natural projection π : Ui[M ] → Ui × Spec(A0). Take the sheaf F ′ i and F ′ i = j to ij. Therefore we get a well-defined quasi-coherent sheaf F ′ on π∗(F Ui×Spec(A0)). By the first condition of Z, we can identify the restrictions of F ′ Uij[M ] by the isomorphism 1 + sij of F ′ X ⊗ Spec(A0[M ]). On each affine set Ui[M ], we have Φi + ti : End(F ′ i) ⊗ L′. It is easy to check e(Φi + ti)(1 + sij) = Φj + tj by the second condition of Z. Therefore {Φi +ti} can be glued together to give a global homomorphism Φ′ : F ′ → F ′ ⊗ L′. For the element (sij , ti) in Z, we construct an element (F ′, Φ′) in Mξ(A0[M ]). Let (sij , ti) be an element in B. In other words, sij = si − sj and ti = e(Φ)(si). The identification of F ′ i ∼= F ′ j on Uij[M ] is given by the isomorphism Consider the following diagram 1 + sij = 1 + (si − sj). 1+si F ′ ij 1+sij 1+sj F ′ ij F ′ ij Id F ′ ij The commutativity of the above diagram implies that E′ is trivial. Similarly, we have e(Φi + ti)(1 + si) = Φi. Therefore the associated Hitchin pair (F ′, Φ′) is isomorphic to (π∗F , π∗Φ). The above construction gives us a well-defined map from H1(C •) to Mξ(A0[M ]). Now we have to construct the inverse map from Mξ(A0[M ]) to H1(C •). Let (F ′, Φ′) ∈ Mξ(A0[M ]) be a L-twisted Hitchin pairs over X × Spec(A0[M ]) such that (F ′X ×Spec(A0), Φ′X ×Spec(A0)) = ξ = (F , Φ). We still use the covering {Ui[M ]} of X × Spec(A0[M ]) to work on this problem locally. MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK 9 Clearly, F ′ i = F ′Ui[M] is the pull-back of F X ×Spec(A0). We obtain F ′ by gluing F ′ i together. Thus the autormorphism 1 + sij of F ′ ij over the intersection Uij[M ] should satisfy the condition sij + sjk = sik on Uijk[M ]. Similarly, Φ′ is given by Φi + ti locally, where Φi ∈ End(F ) ⊗ LUi and ti ∈ End(F ) ⊗ L ⊗ M Ui[M]. By the compatbility condition of Φi + ti on Uij[M ], we have e(Φi + ti)(1 + sij ) = Φj + tj, which gives us e(Φ)(sij ) = ti − tj. Therefore, (sij , ti) ∈ Z. The above discussion gives us a map from Mξ(A0[M ]) to H1(C •). It is easy to check that these (cid:3) two maps are inverse to each other. We finish the proof of this proposition. Corollary 3.4. We have the following long exact sequence 0 → H0(C •) → H 0(X , C0) → H 0(X , C1) → H1(C •) → H 1(X , C0) → H 1(X , C1) → H2(C •) → . . . . Proof. This long exact sequence follows directly from the definition of hypercohomology (see [BR94]). (cid:3) Corollary 3.5. Let 0 → M1 → M2 → M3 → 0 be a short exact sequence for finitely generated A0-modules. We have a long exact sequence for hypercohomology · · · → Hi(C • M1 ) → Hi(C • M2 ) → Hi(C • M3 ) → Hi+1(C • M1 ) → . . . . Now we are ready to check the conditions on the deformation theory D = D(A0, M, ξ) = Mξ(A0[M ]) (condition (4) in Theorem 3.1). Note that the cohomology H i(X , C j ) commutes with localization in A0. Applying the well-known five lemma to the long exact sequence in Corollary 3.4, the module Mξ(A0[M ]) ∼= H1(C •) also commutes with localization. Now let M be a free A0-module of rank one. We use the notation A0[ε] := A0[M ], if M is a free module with rank one. By the finiteness theorem of cohomology over algebraic spaces [Ols16, §7.5], the modules H i(X , C j ) are finitely generated for 0 ≤ i, j ≤ 1. Thus Mξ(A0[ε]) is a finitely generated module by the long exact sequence in Corollary 3.4. For condition (4b), it is enough to check the case A = A0 and A′ = A0[M ]. In other words, we have to define an action D = Mξ(A0[M ]) on itself and show that this action is free. By Proposition 3.3, we know that Mξ(A0[M ]) ∼= H1(C •) = Z/B, where Z is the set of pairs (sij , ti), where sij ∈ C0 ij and ti ∈ C1 i satisfying the following conditions (1) sij + sjk = sik as elements of C0 (2) ti − tj = e(Φ)(sij ) as elements of C1 ij , ijk; and B is the subset of Z consisting of elements (si − sj, e(Φ)(si)), where si ∈ C0 action of Z on itself i . There is a natural (s′ ij , t′ i)(sij , ti) := (s′ ij + sij, t′ i + ti), where (s′ itself. Thus we define an action D = Mξ(A0[M ]) on itself. It is easy to check this action is free. i), (sij, ti) ∈ Z. This action can be naturally extended to a well-defined action of Z/B on ij , t′ Now we will discuss the condition (4c). Note that the condition of (4c) is a local property. We may assume that X is an algebraic space and S is an affine scheme as we mentioned as the beginning of §3.3. To prove (4c), we need the following lemma. Lemma 3.6 (Lemma 6.8, 6.9 in [Art69]). Let X be an algebraic space of finite type over an affine scheme S = Spec B, where B is an integral domain. Let F , G be two coherent sheaves on X . Then there is a non-empty open set S′ of S such that for each s ∈ S′, the canonical map is an isomorphism Extq X (F , G)s ∼=−→ Extq Xs (Fs, Gs), q ≥ 0, 10 and HAO SUN H q(X , F ) ⊗A0 k(s) ∼=−→ H q(Xs, Fs), q ≥ 0. By the above lemma, we can find a non-empty open set S′ such that Extq X (F , G)s ∼=−→ Extq Xs (Fs, Gs), and for s ∈ S′. Thus we have H i(X , C j )s ∼= H i(Xs, C j s ), i ≥ 0, j = 1, 2, H1(C •)s ∼= H1(C • s ) by applying five lemma to the long exact sequence in Lemma 3.4, where C • complex C • to the point s. We finish the proof of the condition (4c). s is the restriction of the 3.3.2. Obstruction. Fix a deformation situation (A′ → A → A0, M, ξ), where M is a free A0-module of rank n. For any quotient ǫ : M → M ∗, let A′ → A∗ be the quotient of A′ defined by M ∗. M ′ ǫ M ∗ A′ A∗ A A We can define the deformation situation (A∗ → A → A0, M ∗, ξ). For any element (F ∗, Φ∗) ∈ Mξ(A∗), we want to lift it to a well-defined element in Mξ(A′). We claim that the obstruction for this lifting comes from the vanishing of the second hypercohomology group H2(C • kerǫ). By Corollary 3.5, there is a long exact sequence for the hypercohomology groups · · · → H1(C • kerǫ) → H1(C • M ) → H1(C • M ∗ ) → H2(C • kerǫ) → . . . . Such a lifting exists if and only if the morphism H1(C • of the second hypercohomology H2(C • M ) → H1(C • M ∗ ) is surjective. Thus the vanishing kerǫ) is necessary and sufficient for the existence of the lifting. For (5a), we have to show that there exists a lifting (FB ′, ΦB ′ ) of (FB, ΦB) with respect to the lifting (FA′, ΦA′ ) of (FA, ΦA), where B′ = B ×A A′. If we forget the morphism part Φ and only consider the coherent sheaf F , such a lifting FB ′ exists by [Art69, OS03]. More generally, the map Q(B′) → Q(A′) ×Q(A) Q(B) is bijective, where Q(A) := Q(GA/XA/S) and so are Q(A′) and Q(B) [OS03]. We do the same argument for End(F ). The lifting End(FB ′) of End(FB) exists by the same reason. Thus the lifting ΦB ′ of ΦB also exists. The proof of (5b) is similar to [Art69, (5b') page 65]. There is an inclusion map i : XA′ → XA0×K A′ K , K which induces the isomorphism Extq A′ K (i∗F1, F2) ∼= Extq A0×K A′ K (F1, i∗F2), q ≥ 0, for coherent sheaves F1 and F2. Then we have H q A′ Thus we have H2 A′ Corollary 3.4. It follows that the obstruction to lift element to M(A0 ×K A′ same. (X , F ) for coherent sheaf F . M ) by applying five lemma to the long exact sequence in K) are the K) and to M(A′ (X , F ) ∼= H q ) ∼= H2 A0×K A′ K A0×K A′ K (C • MK (C • K K The proof of (5c) is similar to that of (4c). The difference is that (4c) is working on the deformation H1(C •), while (5c) is working on the obstruction H2(C •). We use the same notation as in the statement of (5c). Let ξ ∈ M(A). Suppost that for every one-dimensional quotient MK → M ∗ K, there is a non- trivial obstruction to lift ξK ∈ M(AK) to M(A∗ K. We have to prove that there exists an open subset S′ ⊆ S such that ξ cannot be lifted to M(A∗). K is the extension defined by M ∗ K), where A∗ MODULI PROBLEM OF HITCHIN PAIRS OVER DELIGNE-MUMFORD STACK 11 By the discussion at the beginning of §3.3.2, we know that the vanishing of the second hyperco- N ) is necessary and sufficient for the lifting of ξ ∈ M(A) in M(A∗), where N is any homology H2(C • submodule of M . Thus we have to show H2(C • N )s ∼= H2((C • N )s). This isomorphism implies that there is a non-trivial obstruction to lift ξK to M(A∗ K) if and only there exists a non-trivial obstruction to lift ξ to M(A∗). By Lemma 3.6, we can choose an open set S′ of S such that Extq Xs ((F1)s, (F2)s) ∼= Extq X (F1, F2)s for s ∈ S′. By the spectral sequence relating local and global Ext functor, we have for s ∈ S′. Taking F1 = OX and F2 = C0, we have Extq Xs ((F1)s, (F2)s) ∼= Extq X (F1, F2)s H q(X , C0 N )s ∼= H q(Xs, (C0 N )s), for s ∈ S′. Similarly, we have H q(X , C1 H2((C • N )s) holds. This finishes the proof of (5c). N )s ∼= H q(Xs, (C1 N )s). Thus the isomorphism H2(C • N )s ∼= References [Art69] M. Artin. Algebraization of formal moduli. I. In Global Analysis (Papers in Honor of K. Kodaira), pages 21 -- 71. [BR94] Univ. Tokyo Press, Tokyo, 1969. I. Biswas and S. Ramanan. An infinitesimal study of the moduli of Hitchin pairs. J. London Math. Soc. (2), 49(2):219 -- 231, 1994. [Hit87] N. J. Hitchin. The self-duality equations on a Riemann surface. Proc. London Math. Soc. (3), 55(1):59 -- 126, 1987. [MS80] V. B. Mehta and C. S. Seshadri. Moduli of vector bundles on curves with parabolic structures. Math. Ann., 248(3):205 -- 239, 1980. [Nir09] Fabio Nironi. Moduli spaces of semistable sheaves on projective deligne-mumford stacks. arXiv preprint arXiv:0811.0949, 2009. [Nit91] Nitin Nitsure. Moduli space of semistable pairs on a curve. Proc. London Math. Soc. (3), 62(2):275 -- 300, 1991. [Ols16] Martin Olsson. Algebraic spaces and stacks, volume 62 of American Mathematical Society Colloquium Publica- tions. American Mathematical Society, Providence, RI, 2016. [OS03] Martin Olsson and Jason Starr. Quot functors for Deligne-Mumford stacks. Comm. Algebra, 31(8):4069 -- 4096, 2003. Special issue in honor of Steven L. Kleiman. [Sim94] Carlos T. Simpson. Moduli of representations of the fundamental group of a smooth projective variety. I. Inst. Hautes ´Etudes Sci. Publ. Math., (79):47 -- 129, 1994. [Yok93] Koji Yokogawa. Compactification of moduli of parabolic sheaves and moduli of parabolic Higgs sheaves. Journal of Mathematics of Kyoto University, 33(2):451 -- 504, 1993. [Yok95] Koji Yokogawa. Infinitesimal deformation of parabolic Higgs sheaves. International Journal of Mathematics, 6(1):125 -- 148, 1995. Department of Mathematics, Sun Yat-Sen University 135 Xingang W Rd, BinJiang Lu, Haizhu Qu, Guangzhou Shi, Guangdong Sheng, China E-mail address: [email protected]
1108.6209	1	1108	2011-08-31T12:44:17	Complements of hyperplane sub-bundles in projective space bundles over the projective line	[ "math.AG" ]	We establish that the isomorphy type as an abstract algebraic variety of the complement of an ample hyperplane sub-bundle H of a projective space bundle of rank r-1 over the projective line depends only on the the r-fold self-intersection of H . In particular it depends neither on the ambient bundle nor on a particular ample hyperplane sub-bundle with given r-fold self-intersection. Our proof exploits the unexpected property that every such complement comes equipped with the structure of a non trivial torsor under a vector bundle on the affine line with a double origin.	math.AG	math	COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 ADRIEN DUBOULOZ Abstract. We establish that the isomorphy type as an abstract algebraic variety of the complement of an ample hyperplane sub-bundle H of a Pr−1-bundle P(E) → P1 depends only on the the r-fold self-intersection (Hr ) ∈ Z of H. In particular it depends neither on the ambient bundle P(E) nor on the choice of a particular ample sub-bundle with given r-fold self-intersection. Our proof exploits the unexpected property that every such complement comes equipped with the structure of a non trivial torsor under a vector bundle over the affine line with a double origin. Introduction The Danilov-Gizatullin Isomorphism Theorem [5, Theorem 5.8.1] (see also [2, 4] for short self-contained proofs) is a surprising result which asserts that the isomorphy type as an abstract algebraic variety of the complement of an ample section C of a P1-bundle ν : P(E) → P1 C over the complex projective line depends only on the self-intersection (C 2) ≥ 2 of C. In particular, it depends neither on the ambient bundle nor on the chosen ample section with fixed self-intersection d. For such a section, the locally trivial fibration ν : P(E) \ C → P1 induced by the restriction of the structure morphism ν is homeomorphic in the euclidean topology to the complex line bundle OP1 (−d) → P1. However the non vanishing of H 1(P1, OP1 (−d)) for d ≥ 2 implies that ν : P(E) \ C → P1 is in general a non trivial algebraic OP1 (−d)-torsor, and the ampleness of C is in fact precisely equivalent to its non triviality. So the Danilov-Gizatullin Theorem can be rephrased as the fact that the isomorphy type as an abstract algebraic variety of the total space of a non trivial OP1 (−d)-torsor is uniquely determined by its underlying structure of topological complex line bundle over P1. More generally, given a vector bundle E → P1 of rank r ≥ 3 and a sub-vector bundle F ⊂ E of corank one with quotient line bundle L, the complement in the projective bundle ν : P (E) → P1 of lines in E of the hyperplane sub-bundle H = P (F ) inherits the structure of an F ⊗ L−1-torsor ν : P (E) \ H → P1. Similarly as above, the latter is homeomorphic in the euclidean topology to the complex vector bundle F ⊗ L−1 → P1. Furthermore, F ⊗ L−1 is homeomorphic as a complex vector bundle to det(F ⊗ L−1) ⊕ Ar−2 P1 , where (H r) ∈ Z denotes the r-fold intersection product of H with itself. So one may ask by analogy with the one dimensional case if for an ample H, the integer (H r) ≥ r uniquely determines the isomorphy type of P (E) \ H as an algebraic variety. However, since the ampleness of H is in general no longer equivalent to the non triviality of the torsor ν : P (E) \ H → P1, the following problem seems more natural: P1 ≃ OP1 (−(H r)) ⊕ Ar−2 Question. Is the isomorphy type as an abstract algebraic variety of the total space of a nontrivial torsor under an algebraic vector bundle G → P1 uniquely determined by the isomorphy type of G as a topological complex vector bundle over P1, that is, by the rank and the degree of G ? While our results imply in particular that non trivial torsors under homeomorphic complex vector bundles do in- deed have isomorphic total spaces, the answer to this question is negative in general: there exists torsors under non homeomorphic vector bundles or rank r ≥ 2 which have isomorphic total spaces as algebraic varieties. For instance, the complement of the diagonal in P1 × P1 is an affine surface S which inherits two structures of non trivial OP1 (−2)-torsor via the first and the second projections pri : S → P1, i = 1, 2. The Picard group of S is isomorphic to Z and for every 2OP1 (−k) are isomorphic. It follows that for every k ∈ Z, the total space of k ∈ Z the line bundles pr∗ 1OP1 (k) is simultaneously the total space of an OP1 (−2) ⊕ OP1 (k)-torsor and of an OP1 (−2) ⊕ OP1 (−k)-torsor over pr∗ P1 via the first and the second projection respectively. In particular, for every k 6= 0, we obtain an affine variety which is simultaneously the total space of non trivial torsors under complex vector bundles over P1 with different topological types. 1OP1 (k) and pr∗ So the degree of the complex vector bundle G → P1 is not the appropriate numerical invariant to classify isomorphy types of total spaces of non trivial algebraic G-torsors. In contrast, our main result can be summarized as follows: Theorem. The total space of a torsor ν : V → P1 under a vector bundle G → P1 is an affine variety if and only if ν : V → P1 is a non trivial torsor. If so, the isomorphy type of V as an abstract algebraic variety is uniquely determined by the rank of G and the absolute value of deg G + 2. The role of the integer deg G + 2 may look surprising but the latter is intimately related to the subgroup of the Picard V ) of V . Indeed, for a G-torsor ν : V → P1, the group Pic (V ) ≃ Z of V generated by the canonical bundle KV = det(Ω1 relative cotangent bundle Ω1 V /P1 is isomorphic to ν ∗G∨ and so it follows from the relative cotangent exact sequence 0 → ν ∗Ω1 P1 → Ω1 V → Ω1 V /P1 ≃ ν ∗G∨ → 0 that KV ≃ ν ∗(det G∨ ⊗ Ω1 P1 ) whence that the subgroup of Pic(V ) generated by KV is isomorphic to deg G + 2 Z ⊂ Z. 2000 Mathematics Subject Classification. 14L30, 14R05, 14R25 . Key words and phrases. Hyperplane sub-bundles, affine-linear bundles, Danilov-Gizatullin Theorem. 1 COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 2 In the particular case where ν : V → P1 is a torsor arising as the complement P (E) \ H of an ample hyperplane sub-bundle H, one has deg G = − (H r) ≤ −r and so the above characterization specializes to the following generalization of the geometric form of the Danilov-Gizatullin Theorem: Corollary. The isomorphy type as an abstract algebraic variety of the complement of an ample hyperplane sub-bundle H of a Pr−1-bundle ρ : P (E) → P1 depends only on the r-fold self-intersection (H r) ≥ r of H. The proof of the above results exploits a hidden and unexpected feature of total spaces of non trivial torsors ν : V → P1 under a vector bundle G → P1 of rank r ≥ 1, namely the existence on every such V of the structure ρ : V → X of a non trivial torsor under a vector bundle G → X of the same rank r, on a non separated scheme X, isomorphic to the affine line with a double origin. The structure of these bundles G → X is very similar to that of vector bundles on P1: in particular the existence of a covering of X by two open subsets isomorphic to C and intersecting along C∗ implies that as a topological complex vector bundle, G → X is uniquely determined by a homotopy class of maps S1 → GLr (C), whence simply by the "degree" of its determinant. In this setting, we establish that the total space of a non trivial torsor ν : V → P1 under a vector bundle G → P1 of degree d carries the structure of a non trivial torsor ρ : V → X under a vector bundle G → X of degree d + 2 uniquely determined by G. While there exists infinite moduli for isomorphy type of total spaces of non trivial torsors under a line bundle on X, the Danilov-Gizatullin Theorem can be re-interpreted as the fact that the total spaces of non trivial torsors ν : V → P1 under G = OP1 (−d) are all isomorphic as torsors ρ : V → X under the corresponding line bundle G. For higher dimensional vector bundles G → X, we establish in contrast that the isomorphy type as an abstract variety of the total space of a non trivial G-torsor ρ : V → X is uniquely determined by the absolute value of the degree of G. The article is organized as follows: the first section recalls basic properties of torsors under vector bundles. Then section two is devoted to the study of isomorphy types of total spaces of non trivial torsors under vector bundles on the affine line with a double origin. These results are applied in the third section to the case of torsors under vector bundles on P1. Here, to fix the notation and convention that will be used in the sequel, we briefly recall classical facts about vector 1. Recollection on affine-linear bundles bundles, projective bundles and affine-linear bundles. 1.1. Vector bundles and projective bundles. 1.1.1. A vector bundle of rank r ≥ 1 on a scheme X is the relative spectrum p : F = SpecX (Sym(F ∨)) → X of the symmetric algebra of the dual of a locally free coherent OX-module F of rank r. The X-scheme F represents the contravariant functor from the category of schemes over X to the category of abelian groups which associates to an X-scheme f : Y → X the group Γ (Y, f ∗F) of global sections of f ∗F over Y . It follows in particular that p : F → X is a locally constant commutative affine group scheme over X, with group law F ×X F → F induced by the diagonal homomorphism of OX-module F ∨ → F ∨ ⊕ F ∨. 1.1.2. Given a vector bundle p : E = SpecX(Sym(E ∨)) → X, the projective bundle of lines in E is the relative proj ν : P (E) = ProjX (Sym(E ∨)) → X of the symmetric algebra of E ∨ considered as a graded quasi-quoherent OX- algebra Sym(E ∨) = Ln≥0 Symn(E ∨). There is a canonical surjection ν ∗E∨ → OP(E)(1) on P (E) which yields dually a closed embedding OP(E)(−1) ֒→ ν ∗E of the tautogical line bundle as a line sub-bundle of ν ∗E. More generally, if f : Y → X is a scheme over X then an X-morphism f : Y → P (E) is uniquely determined by a sub-line bundle of f ∗E, which then coincides with f ∗OP(E)(−1). In other word, P (E) represents the functor which associates to an X-scheme f : Y → X the set of sub-line bundles of f ∗E. Recall that if L = SpecX (Sym(L∨)) → X is a line bundle on X, then the canonical isomorphism Sym ((E ⊗ L)∨) ≃ Ln≥0 Symn(E ∨)⊗(L∨)⊗n of graded OX-algebras yields an isomorphism ϕ : P (E) ∼→ P (E ⊗ L) of schemes over X with ϕ∗OP(E⊗L) (−1) ≃ OP(E) (−1) ⊗ ν ∗L. 1.2. Torsors under vector bundles. 1.2.1. Torsors under a vector bundle on a scheme are the analogues in a relative setting of affine spaces attached to a vector space. Namely, given a vector bundle p : F = SpecX (Sym(F ∨)) → X, a principal homogeneous F -bundle, or an F -torsor, is a scheme ν : V → X equipped with an action µ : F ×X V → V of the group scheme F for which there exists a covering of X by open subsets {Ui}i∈I such that for every i ∈ I, V Ui = ν −1(Ui) is equivariantly isomorphic to F Ui ∼→ F Ui , i ∈ I, it follows that acting on itself by translations. Given a collection of equivariant trivializations τi : V Ui for every i, j ∈ I, τi ◦ τ −1 Ui ∩Uj is an equivariant automorphism of F Ui∩Uj whence is a translation determined by a section gij ∈ Γ(Ui ∩ Uj, F ) = Γ(Ui ∩ Uj , F) of F Ui ∩Uj . Clearly, gik Ui∩Uj ∩Uk = gij Ui∩Uj ∩Uk +gjk Ui∩Uj ∩Uk for every i, j, k ∈ I, that is, (gij )i,j∈I is a Čech 1-cocycle with value in the sheaf F for the open covering {Ui}i∈I. Changing the trivializations τi replaces (gij)i,j∈I by a cohomologous cocycle and the cohomology class is unaltered if V is replaced by an isomorphic torsor. Thus ν : V → X defines a class c(V ) ∈ H 1({Ui}i∈I , F ) = H 1({Ui}i∈I , F) and a standard argument eventually shows that there is a one-to-one correspondence between isomorphy classes of F -torsors and elements of the cohomology group H 1(X, F ) ≃ H 1(X, F ), with 0 ∈ H 1(X, F ) corresponding the the trivial F -torsor p : F → X (see e.g. [7, 16.4.9]). This implies in particular that every F -torsor on an affine scheme X is isomorphic to the trivial one. j 1.2.2. Recall that the relative cotangent bundle Ω1 F /X of a vector bundle p : F → X is canonicaly isomorphic to p∗F ∨ [7, 16.5.15]. One checks using the above local description that this holds more generally for any F -torsor ν : V → X, providing a canonical short exact sequence 0 → ν ∗Ω1 If X V /X ≃ ν ∗F ∨ → 0 of vector bundles on V . X → Ω1 V → Ω1 COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 3 is normal then the natural homomorphism ν ∗ : Pic (X) ∼→ Pic (V ) is an isomorphism; the relative canonical bundle V ) of V then coincide respectively with the images by ν ∗ of KV /X = det(Ω1 the line bundles det(F ∨) and KX ⊗ det(F ∨) on X. V /X ) and the canonical bundle KV = det(Ω1 Ext1(cid:0)A1 X , F(cid:1) with the isomorphy class of an extension 0 → F → E → A1 1.2.3. Given a vector bundle p : F → X, every class c ∈ H 1(X, F ) coincides via the canonical isomorphism H 1(X, F ) ≃ X → 0 of vector bundles on X, where X = X × A1 denotes the trivial line bundle on X. The inclusion F ֒→ E induces a closed immersion of P (F ) into A1 ν : P (E) → X as the zero locus of the regular section of OP(E)(1) deduced from the composition OP(E)(−1) ֒→ ν ∗E → X, and the complement P (E) \ P (F ) is then isomorphic as a scheme over X to the total space of an F -torsor ν ∗A1 ν : V → X with isomorphy class c. X corresponds to the canonical X ). More generally, given a line bundle L → X on X and a short exact sequence open immersion of F into P(F ⊕ A1 of vector bundles 0 → F → E → L → 0, the complement of P (F ) ≃ P(F ⊗ L−1) in P (E) ≃ P(E ⊗ L−1) inherits the structure of an F ⊗ L−1-torsor with isomorphy class in H 1(X, F ⊗ L−1) ≃ Ext1 (L, F ) given by the class of the extension 0 → F → E → L → 0. In particular, the trivial extension E = F ⊕ A1 1.3. Affine-linear bundles. 1.3.1. Affine-linear bundles over a scheme X from a sub-class of the class of locally trivial An-bundle over X, namely, an affine-linear bundle of rank r ≥ 1 over X is an X-scheme ν : V → X for which there exists a open covering {Ui}i∈I of X and a collection of isomorphisms τi Ui of schemes over Ui such that for every i, j ∈ I, (OUi∩Uj [x1, . . . , xr]). This means that there τij = τi ◦ τ −1 exists (Aij , Tij) ∈ Aff r(Ui ∩ Uj) = GLr(Ui ∩ Uj ) ⋊ Gr a(Ui ∩ Uj ) such that τij (x1, . . . , xr) = Aij(x1, . . . , xr) + Tij for every i, j ∈ I. It follows that isomorphy classes of affine-linear bundles of rank r are in one-to-one correspondence with that of principal homogeneous bundles under the affine group Aff r = GLr ⋊ Gr a. Ui∩Uj is an affine automorphism of Ar Ui∩Uj = SpecUi∩Uj : V Ui ∼→ Ar j 1.3.2. Of course, every torsor under a vector bundle of rank r ≥ 1 is an affine linear bundle of rank r. Conversely, let ν : V → X be an affine-linear bundle of rank r ≥ 1 with trivializations τi : V Ui Ui. Then for every triple of indices i, j, k ∈ I, the identities ∼→ Ar (Aik = AjkAij GLr+1(Ui ∩ Uj ) consisting of matrices of the form Aij =(cid:18) Aij hold in GLr(Ui ∩ Uj ∩ Uk) and Gr a(Ui ∩ Uj ∩ Uk) respectively. 0 tTik = Ajk ·t Tij +t Tjk If we identify Aff r(Ui ∩ Uj ) with the sub-group of tTij 1 (cid:19) these relations say equivalently that ( Aij)i,j∈I and (Aij)i,j∈I are Čech cocycles with value in GLr+1 and GLr respectively for the open covering {Ui}i∈I of X. These define respectively a vector bundle E → X of rank r + 1 and a sub-vector bundle F of E fitting in a short exact sequence X → 0 of vector bundles on X. By construction, ν : V → X is isomorphic as a scheme over X 0 → F → E → A1 to the complement of P (F ) in ν : P (E) → X, whence can be equipped with the structure of an F -torsor. Changing the trivializations τi by means of affine automorphisms changes the cocycle (Aij)i,j∈I for a cohomologous one and the cohomology class in H 1(X, GLr) is unaltered if we replace ν : V → X by an isomorphic affine-linear bundle. Therefore, the vector bundle F for which an affine-linear bundle can be equipped with the structure of an F -torsor is uniquely X, F ) defined by ( Aij )i,j∈I is unaltered if we change the determined up to isomorphism. Similarly, the class in Ext1(A1 τi's or replace ν : V → X by an isomorphic affine-linear bundle, and so the isomorphy class in H 1(X, F ) ≃ Ext(A1 X, F ) of ν : V → X as an F -torsor is also uniquely determined. 2. Affine-linear bundles over the affine with a double origin The affine line with a double origin is the scheme δ : X → A1 = Spec(C [x]) obtained by gluing two copies X± of ± = X± \ {o±}. the affine line A1 = Spec(C [x]), with respective origins o±, by the identity along the open subsets X ∗ It comes equipped with a canonical covering U by the open subsets X+ and X−. The morphism δ is induced by the identity morphism on X± and restricts to an isomorphism X \ {o±} ≃ Spec(cid:0)C[x±1](cid:1). Since every automorphism of X is induced by an automorphism of X+ ⊔ X− of the form X± ∋ x 7→ ax ∈ Xε·±, where a ∈ C∗ and ε = ±1, the automorphism group Aut (X) of X is isomorphic to Gm × Z2. In what follows, we denote by θ = (1, −1) ∈ Gm × Z2 the automorphism which exchanges the open subsets X± of X. 2.1. Vector bundles on the affine line with a double origin. 2.1.1. Since every line bundle on X becomes trivial on the canonical covering U, the Picard group Pic(X) of X is isomorphic to H 1(U, O∗ X ) ≃ C[x±1]∗/C∗ ≃ Z. In what follows we fix as a generator for Pic (X) the class of the line bundle p : L → X with trivializations L X± ≃ Spec (C [x] [u±]) and transition isomorphism τ± : X ∗ − × A1, (x, u+) 7→ (x, xu+). The pull-back of L by the automorphism θ of X which exchanges the two open subsets X± of X is isomorphic to the dual L−1 of L. A line bundle L → X isomorphic to Lk for some k ∈ Z is said to be of degree −k. + × A1 ∼−→ X ∗ More generally, every vector bundle E → X of rank r ≥ 2 becomes trivial on the canonical covering U of X, H 1(U, GLr) ≃ GLr (C [x]) \GLr(C[x±1])/GLr (C [x]). Since for a suitable n ≥ 0, E ⊗ Ln is determined by a matrix whence is determined up to isomorphism by the equivalence class of a matrix M ∈ GLr(cid:0)C[x±1](cid:1) in the double quotient M ∈ Mr (C [x]) ∩ GLr(cid:0)C[x±1](cid:1) equivalent in the double quotient GLr (C [x]) \Mr (C [x]) /GLr (C [x]) to its Smith diagonal normal form, it follows that E splits into a direct sum of line bundles, i.e., is isomorphic to Lr suitable k1, . . . , kr ∈ Z. The Grothendieck group K0 (X) of vector bundles on X is described as follows: Lki for i=1 Lemma 2.1. The map K0 (X) → Pic (X) ⊕ Z, [E] 7→ (det E, rk (E)) is an isomorphism of groups. COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 4 Proof. The only non trivial part is to show that this map is injective, or equivalently, that if E → X is a vector bundle of rank r ≥ 2 then E and det (E) ⊕ Ar−1 X have the same class in K0 (X). We proceed by induction on r ≥ 2. Since every vector bundle on X is decomposable, we may assume that E = Lm ⊕ E′ where k ∈ Z and E′ is a vector bundle of rank r − 1. By induction hypothesis, E′ has the same class in K0 (X) as det (E′) = Ln ⊕ Ar−2 X where n = deg (E′), and so it X have the same class in K0 (X). If either m or n is is enough to check that for every m, n ∈ Z, Lm ⊕ Ln and Lm+n ⊕ A1 equal to zero then we are done. Otherwise, up to changing Lm ⊕ Ln for its dual and exchanging the roles of m and n, we may assume that either 0 < m ≤ n or m < 0 < n. In the first case, the matrix M =(cid:18) xm 0 1 xn (cid:19) =(cid:18) 0 −1 xn−1 (cid:19)(cid:18) xm+n 1 0 0 1 (cid:19)(cid:18) 1 xm−1 0 1 (cid:19) ∈ GL2(cid:0)C(cid:2)x±1(cid:3)(cid:1) Lm ⊕ Ln and Lm+n ⊕ A1 fact that the matrix X → Ln → 0. Hence X have the same class in K0 (X). The second case follows from the same argument using the is equivalent in H 1(U, GL2) to diag(cid:0)xm+n, 1(cid:1) and defines an extension 0 → Lm → Lm+n ⊕ A1 0 (cid:19) ∈ GL2(cid:0)C(cid:2)x±1(cid:3)(cid:1) 0 xm+n (cid:19) =(cid:18) 1 xn (cid:19)(cid:18) x−m 1 1 (cid:19)(cid:18) xm N =(cid:18) 1 is equivalent in H 1(U, GL2) to diag (xm, xn) and defines an extension 0 → A1 X → Lm ⊕ Ln → Lm+n → 0. xm −1 xn (cid:3) 0 0 0 2.2. Affine-linear bundles of rank one. Here we review the classification of affine-linear bundle of rank one over X following [1, 3]. 2.2.1. In view of the above description of Pic(X), every affine-linear bundle ρ : S → X of rank one over X is an Lk-torsor nontrivial Lk-torsor ρ : S → X is isomorphic to a one obtain by gluing X+ × A1 and X− × A1 over X+ ∩ X− by an for a certain k ∈ Z. We deduce from the isomorphism H 1(X, Lk) ≃ H 1(U, Lk) ≃ C(cid:2)x±1(cid:3) /hxkC [x] + C [x]i that every isomorphism of the form (x, u+) 7→ (x, xku+ + g (x)) for a Laurent polynomial g (x) ∈ C(cid:2)x±1(cid:3) with non zero residue class in C(cid:2)x±1(cid:3) /hxkC [x] + C [x]i. This implies in turn that the total space of a nontrivial Lk-torsor is an affine surface. Indeed, writing g = x−lh (x) where h ∈ C [x] \ xC [x] and l > min (0, −k), the local regular functions ϕ+ = xk+lu+ + h (x) ∈ Γ(S X+ , OS) and ϕ− = xlu− ∈ Γ(S X− , OS) glue to a global one ϕ ∈ Γ(S, OS) for which the morphism π = (δ ◦ ρ, ϕ) : S → A2 = Spec (C [x, y]) maps the fibers ρ−1 (o±) to the distinct points (0, h (0)) and (0, 0) respectively and restricts to an isomorphism S \ ρ−1 ({o±}) ≃ Spec(cid:0)C(cid:2)x±1, y(cid:3)(cid:1). Since the inverse images by π of the principal affine open subsets y 6= h (0) and y 6= 0 of A2 are principal open subsets of S \ ρ−1 (o+) ≃ A2 and S \ ρ−1 (o−) ≃ A2 respectively, it follows that π : S → A2 is an affine morphism whence that S is an affine scheme. Note that conversely the total space of a trivial Lk-torsor cannot be affine since it is not even separated. Example 2.2. For every d ∈ Z, we let ζd : S (d) → X be the nontrivial Ld-torsor with gluing isomorphism X+ × A1 ⊃ X ∗ + × A1 ∼→ X ∗ − × A1 ⊂ X− × A1, (x, u+) 7→(cid:16)x, xdu+ + xmin(−1,d−1)(cid:17) . One checks easily that ζ−d : S (−d) → X is isomorphic to the pull-back S (d)×X X of ζd : S (d) → X by the automorphism θ of X which exchanges the open subsets X±. Since Ω1 L−d whence that Pic(S(d))/hKS(d)i ≃ Z/ d Z. Therefore, the surfaces S(d) are pairwise non isomorphic as schemes over X while S(d) and S(d′) are isomorphic as abstract schemes if and only if d = ±d′. Note that since Pic (S(d)) ≃ Pic(S(d) ×X Ar X ) for every r ≥ 1, the same argument shows more generally that S(d) ×X Ar X as a scheme over X if and only if d = d′, and as an abstract scheme if and only if d = ±d′. X is trivial it follows that KS(d) ≃ Ω1 X is isomorphic to S(d′) ×X Ar S(d)/X ≃ ζ ∗ d If d ≥ 0 then, letting ϕ ∈ Γ(S (d) , OS(d)) be defined locally by (ϕ+, ϕ−) = (xd+1u+ + 1, xu−) as in 2.2.1, one checks that the rational functions ψ = x−1ϕ (ϕ − 1) and ξ = x−dϕdψ on S (d) are regular and that the morphism (δ ◦ ζd, ϕ, ψ, ξ) : S (d) → A4 = Spec (C [x, y, z, u]) is a closed embedding of S (d) as the surface defined by the equations xz = y (y − 1) , (y − 1)d u = zd+1, xdu = ydz. The following result shows that for a non trivial affine-linear bundle of rank one ρ : S → X, the isomorphy type of S as an abstract scheme is essentially uniquely determined by its one as an affine-linear bundle over X: Theorem 2.3. Two non trivial affine-linear bundles or rank one ρi : Si → X, i = 1, 2, have isomorphic total spaces if and only if their isomorphy classes in H 1(X, Aff 1) belong to the same orbit of the action of Aut (X). Proof. The condition is clearly sufficient. Conversely, if either S1 or S2 admits a unique affine-linear bundle structure over X up to composition by automorphisms of X then both admit a unique such structure and so every isomorphism ∼→ S1 descends to an automorphism ϕ of X such that ρ1 ◦ Φ = ϕ ◦ ρ2. This implies in turn that Φ factors through Φ : S2 an isomorphism of affine-linear bundles Φ′ : S2 → S1 ×X X whence that the isomorphy classes in H 1(X, Aff 1) of the Aff 1-bundles associated to S1 and S2 belong to a same orbit of Aut (X). Otherwise, if S1 and S2 both admit at least two affine-linear bundle structures over X with distinct general fibers then, by combining Theorem 3.11 and 5.3 in [1], if the canonical bundles KS1 and KS2 are both trivial then ρi : Si → X, i = 1, 2, are both we obtain the following : isomorphic to ζ0 : S (0) → X. Otherwise, for d = ord(Pic(S1)/hKS1 i) = ord(Pic(S2)/hKS2 i), ρi : Si → X is isomorphic as an affine-linear bundle either to the one ζd : S (d) → X or to the one ζ−d : S (−d) → X. This completes the proof since the latters are obtained from each others via the base change by the automorphism θ of X (see Example 2.2 above). (cid:3) COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 5 2.3. Affine-linear bundles of higher ranks. In this subsection, we consider affine-linear bundles ρ : V → X of rank r ≥ 2. Recall that to every such bundle is associated a vector bundle E → X unique up to isomorphism for which ρ : V → X inherits the structure of an E-torsor. In contrast with the case of affine-linear bundles of rank one, we have the following characterization: Theorem 2.4. Let pi : Ei → X, i = 1, 2 be vector bundles of the same rank r ≥ 2 and let ρi : Vi → X be non trivial Ei-torsors, i = 1, 2. Then the following holds: 1) V1 and V2 are isomorphic as schemes over X if and only if deg(E1) = deg(E2), 2) V1 and V2 are isomorphic as abstract schemes if and only if deg(E1) = ± deg(E2). 2.3.1. Theorem 2.4 is a consequence of Lemmas 2.5 and Proposition 2.6 below which, combined with Lemma 2.1, imply that the total space of non trivial E-torsor ρ : V → X or rank r ≥ 2 is isomorphic as a scheme over X to S (d) ×X Ar−1 X , where d = − deg (E) and where ζd : S(d) → X is the non trivial Ld-torsor defined in Example 2.2 above. Lemma 2.5. The total spaces of all non trivial torsors under a fixed vector bundle p : E → X of rank r ≥ 2 are affine and isomorphic as schemes over X. Lki , k1, . . . , kr ∈ Z. Given a non trivial E-torsor ρ : V → X there exists an index i such that the i-th component of the isomorphy class (v1, . . . , vr) of V in H 1(X, E) ≃ i=1 H 1(X, Lki ) is not zero. Up to a permutation, we may assume from now on that v1 6= 0. The actions of Lk1 and Proof. By virtue of 2.1.1 above, we may assume that E = Lr Lr E1 =Lr Lki ≃ E/Lk1 on V commute and we have a cartesian square i=2 i=1 π1 V S1 = V /E1 V /Lk1 ρ1 / X where ρ1 : S1 → X is an Lk1-torsor with isomorphy class v1 ∈ H 1(X, Lk1 ) and where π1 : V → S1 = V /E1 is a 1E1-torsor. Since ρ1 : S1 → X is a non trivial torsor, S1 is an affine scheme by virtue of 2.2.1 and so π1 : V → S1 is ρ∗ isomorphic as a scheme over X to the total space of the trivial ρ∗ 1E1-torsor p1 : S1 ×X E1 → S1. In particular, V is an affine scheme. With the notation of Example 2.2 above, we claim that S1 ×X E1 is isomorphic as a scheme over X to the r-fold fiber product S(k1) ×X · · · ×X S(kr). Indeed, since for every k ∈ Z, ζk : S (k) → X is an Lk-torsor, the fiber product S1 ×X S (k) is simultaneously the total space of a ρ∗ Lk1-torsor over S (k) via the first and the second projection respectively. The fact that S1 and S (k) are both affine implies that the latter are both trivial torsors, which yields isomorphisms S1 ×X Lk ≃ S1 ×X S (k) ≃ Lk1 ×X S (k) of schemes over X. The same argument applied to the non trivial Lk1-torsor ζk1 : S(k1) → X provides isomorphisms S(k1) ×X Lk ≃ S(k1) ×X S (k) ≃ Lk1 ×X S (k) of 1 Lk-torsor over S1 and of a ζ ∗ k schemes over X. Letting E2 = E1/Lk2 ≃Lr i=3 Lki, we finally obtain isomorphisms S1 ×X E1 ≃ S(k1) ×X S(k2) ×X E2 ≃ S(k1) ×X S(k2) ×X (S(k3) ×X · · · ×X S(kr)) where the last isomorphism follows from the fact that the fiber product of the affine scheme q : S(k1) ×X S(k2) → X with the E2-torsor S(k3)×X · · ·×X S(kr) is isomorphic to the trivial q∗E2-torsor S(k1)×X S(k2)×X E2 over S(k1)×X S(k2). (cid:3) Proposition 2.6. The isomorphy type of the total space of a non trivial affine-linear bundle ρ : V → X or rank r ≥ 2 as a scheme over X depends only on the class in K0 (X) of its associated vector bundle. on the rank of E. By combining 2.1.1 and Lemma 2.5 above, we may assume that E =Lr Proof. Given a vector bundle E → X of rank r ≥ 2, we will show more precisely that the total space of a non trivial E-torsor ρ : V → X is isomorphic as a scheme over X to S (d) ×X Ar−1 X , where d = − deg (E). We proceed by induction Lki, where k1, . . . , kr ∈ Z and −k1 − · · · − kr = −d and that V = S(k1) ×X · · · ×X S(kr). Furthermore, since the induction hypothesis implies that S(k2) ×X · · · ×X S(kr) ≃ S(d − k1) ×X Ar−1 X as schemes over X, it is enough to show that for every m, n ∈ Z, S(m) ×X S(n) and S(m + n) ×X A1 X are isomorphic as schemes over X. If m or n is equal to zero then we are done. Otherwise, up to taking the pull-back of S(m) ×X S(n) by the automorphism θ of X and exchanging the roles of m and n, we may assume similarly as in the proof of Lemma 2.1 above that either 0 < m ≤ n or m < 0 < n. In the first case, letting E ≃ Lm+n ⊕ A1 X be the vector bundle on X defined by the matrix i=1 it follows from Lemma 2.5 that the total space of the non trivial E-torsor ρ : V → X with gluing isomorphism M =(cid:18) xm 0 1 xn (cid:19) ∈ GL2(cid:0)C(cid:2)x±1(cid:3)(cid:1) , X+ × A2 ⊃ X ∗ + × A2 (x, (v+, u+)) − × A2 ⊂ X− × A2, ∼→ X ∗ 7→ (x, (xmv+ + u+, xnu+ + x−1)) is isomorphic to S(m + n) ×X A1 X. On the other hand, since E is an extension of Ln by Lm, V inherits a free action of Lm whose quotient V /Lm coincides with the total space of the Ln-torsor ζn : S(n) → X with gluing isomorphism (x, u+) 7→ (x, xnu+ + x−1). Furthermore, the quotient morphism V → V /Lm ≃ S(n) inherits the structure of a ζ ∗ nLm- torsor whence is isomorphic to the trivial one S(n) ×X Lm as S(n) is affine. Summing up, we obtain isomorphisms S(m + n) ×X A1 X ≃ V ≃ S(n) ×X Lm ≃ S(n) ×X S (m) of schemes over X. / / / COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 6 The case m < 0 < n follows from a similar argument starting from the vector bundle E ≃ Lm ⊕ Ln defined by the matrix and the non trivial E-torsor ρ : V → X with gluing isomorphism N =(cid:18) 1 xm 0 xm+n (cid:19) ∈ GL2(cid:0)C(cid:2)x±1(cid:3)(cid:1) , X+ × A2 ⊃ X ∗ + × A2 (x, (v+, u+)) − × A2 ⊂ X− × A2 ∼→ X ∗ 7→ (x, (v+ + xmu+, xm+nu+ + xmin(−1,m+n−1))). (cid:3) 3. Isomorphy types of complements of hyperplane sub-bundles In this section, we consider total spaces of non trivial affine-linear bundles ν : V → P1 over the projective line. We first review the case of affine-linear bundles of rank one: the crucial observation there is the fact that the total space of a non trivial OP1 (−d)-torsor ν : V → P1, where d ≥ 2, is isomorphic to the affine surface ζd−2 : S (d − 2) → X of Example 2.2, whence admits the structure of a non trivial Ld−2-torsor over the affine line with a double origin. This enables to consider total spaces of non trivial affine-linear bundles ν : V → P1 of higher ranks as being simultaneously that of certain non trivial affine-linear bundles over X, and to deduce the classification of total spaces of such bundles as a particular case of the results established in the first section. 3.1. Affine linear bundles of rank one and the Danilov-Gizatullin Theorem. The Danilov-Gizatullin Theorem [5, Theorem 5.8.1 ] asserts that the isomorphy type of the complement of an ample section C in a Hirzebruch surface πn : Fn = P (OP1 ⊕ OP1 (n)) → P1, n ≥ 0, depends only on the self-intersection (C 2) ≥ 2 of C. Since there is a one-to-one correspondence between non trivial OP1 (−d)-torsors ν : V → P1 and complements of ample sections C with self-intersection (C 2) = d in Hirzebruch surfaces [5, Remark 4.8.6], the Danilov-Gizatullin Theorem can be rephrased as the fact that the isomorphy type of the total space of a non trivial OP1 (−d)-torsor ν : V → P1 depends only on d and not on its isomorphy class as a torsor in H 1(P1, OP1 (−d)). We have the following more effective result: Proposition 3.1. The total space of a non trivial OP1 (−d)-torsor ν : V → P1, d ≥ 2, is isomorphic to the surface ζd−2 : S (d − 2) → X of Example 2.2. Furthermore the isomorphism V ≃ S(d − 2) can be chosen in such a way that ν ∗OP1 (1) = ζ ∗ L−1 in Pic (V ) ≃ Z. d−2 Proof. Recall that for every d ≥ 2, ζd−2 : S(d − 2) → X is isomorphic to the surface in A4 = Spec(C [x, y, z, u]) defined by the equations xz = y (y − 1) , (y − 1)d−2 u = zd−1, xd−2u = yd−2z. Letting P1 = Proj(C[w0, w1]), U0 = P1 \ {[1 : 0]} = Spec (C [w]) and U∞ = P1 \ {[0 : 1]} = Spec (C [w′]) where w = w0/w1 and w′ = w1/w0, one checks that the morphism νd−2 : S(d − 2) → P1, (x, y, z, u) 7→ [x : y] = [y − 1 : z] defines an OP1 (−d)-torsor with local trivializations τ0 : ν −1 d−2(U0) ∼→ Spec (C [w] [u]) , τ∞ : ν −1 d−2(U∞) ∼→ Spec(cid:0)C(cid:2)w′(cid:3) [x](cid:1) 0 and transition isomorphism τ∞ ◦ τ −1 from the construction of ζd−2 : S (d − 2) → X and νd−2 : S(d − 2) → P1 that ν −1 classes of these divisors in Cl (S(d − 2)) ≃ Pic (S(d − 2)) coincide respectively with the line bundles ν ∗ ζ ∗ d−2 νi : Vi → P1, i = 1, 2, are non trivial OP1 (−d)-torsors, then there exists an isomorphism f : V1 f ∗(ν ∗ d−2(o+). Since the d−2OP1 (1) and L−1, the assertion follows from the "refined" Danilov-Gizatullin Theorem [2, Theorem 3.1] which asserts that if ∼→ V2 such that (cid:3) U0∩U∞ given by (w, u) 7→ (w′, x) = (cid:0)w−1, wdu + w(cid:1). Furthermore, it follows d−2([0 : 1]) = ζ −1 2 OP1 (1)) ≃ ν ∗ 1 OP1 (1). 3.2. Affine-linear bundles of arbitrary ranks. By combining our previous results, we obtain the following characterization: Theorem 3.2. Let pi : Ei → P1, i = 1, 2 be vector bundles of the same rank r ≥ 1 and let νi : Vi → P1 be non trivial Ei-torsors, i = 1, 2. Then V1 and V2 are affine, and isomorphic as abstract varieties if and only if deg(det E∨ P1 ) = ± deg(det E∨ 1 ⊗ Ω1 2 ⊗ Ω1 P1 ). Proof. The argument is very similar to the one used in the proof of Lemma 2.5 above. Recall that every vector bundle i=1 OP1 (ki), k1, . . . , kr ∈ Z, of line bundles [6]. Therefore, if ν : V → P1 is a non trivial E-torsor, then there exists an index i ∈ {1, . . . , r} such that the i-th component of its E → P1 of rank r ≥ 2 splits into a direct sum E = Lr isomorphy class (a1, . . . , ar) ∈ H 1(P1, E) ≃Lr i=1 H 1(P1, OP1 (ki)) is non zero. Letting Ei =Lj6=i OP1 (kj) ≃ E/OP1 (ki), the quotient of V by the induced action of Ei inherits the structure of a non trivial OP1 (ki)-torsor νi : Si → P1 with isomorphy class ai ∈ H 1(P1, OP1 (ki)). Furthermore, the quotient morphism V → Si = V /Ei has the structure of a i Ei-torsor. Proposition 3.1 above implies that ki = −d for some d ≥ 2 and that Si is isomorphic to the surface ν ∗ i Ei-torsor Si ×P1 Ei. ζd−2 : S(d − 2) → X. In particular, Si ≃ S(d − 2) is affine and so V is isomorphic to the trivial ν ∗ L−1, we obtain that Si ×P1 Ei Moreover, by choosing the isomorphism Si ≃ S(d − 2) in such a way that ν ∗ is an affine variety, isomorphic as a scheme over X to the total space of a non trivial torsor under the vector bundle E = Ld−2 ⊕ L−k2 ⊕ · · · ⊕ L−kr . Since deg E = − deg(det E∨ ⊗ Ω1 P1 ), the assertion follows from Theorem 2.4 above. (cid:3) i OP1 (1) ≃ ζ ∗ d−2 COMPLEMENTS OF HYPERPLANE SUB-BUNDLES IN PROJECTIVE SPACE BUNDLES OVER P1 7 Example 3.3. Let us consider again the example given in the introduction of an affine variety which is simultaneously the total space of non trivial torsors under complex vector bundles of different topological types. The Euler exact sequence P1-torsor v : V → P1 with total space isomorphic 0 → Ω1 to the complement of the diagonal ∆ ≃ P(Ω1 P1 ) in P(OP1 (−1) ⊕ OP1 (−1)) ≃ P1 × P1. For every k ∈ Z, the variety Vk = V ×P1 OP1 (k) → P1 is then a torsor under the vector bundle Fk = Ω1 P1 ) = k = − deg(det(F ∨ P1 ), Theorem 3.2 implies that V−k and Vk are isomorphic affine varieties. The exact sequence j → OP1 (−1)⊕2 → OP1 → 0 on P1 defines a non trivial Ω1 P1 ⊕ OP1 (k). Since deg(det(F ∨ −k) ⊗ Ω1 k ) ⊗ Ω1 P1 0 → Fk = Ω1 P1 ⊕ OP1 (k) j⊕id → Ek = OP1 (−1)⊕2 ⊕ OP1 (k) → OP1 → 0 provides an open embedding of Vk ≃ V−k into P(Ek) as the complement of the hyperplane sub-bundle Hk = P(Fk). Note that if k = 0 then H0 is nef but not ample and that if k > 2 then H−k is ample whereas Hk has negative self-intersection (H 3 k) = 2 − k. More generally, for any n ≥ 2, the complement V in Pn × Pn = Proj(C[x0, . . . , xn]) × Proj(C[y0, . . . , yn]) of the ample i=0 xiyi = 0 inherits simultaneously via the first and the second projection the structure Pn → OPn (−1)⊕n+1 → OPn → 0 on 2Pic (Pn), we have for every k ∈ Z, 1 OPn (−k) in Pic (V ) ≃ Z. Therefore, similarly as in the previous case, we may interpret V ×ν1,Pn OPn (−k) ≃ Pn ⊕ OPn (k)-torsor over of an Ω1 each factor in Pn × Pn. Since D is of type (1, 1) in Pic (Pn × Pn) ≃ p∗ ν ∗ 2 OPn (k) = ν ∗ V ×ν2,Pn OPn (k) as being simultaneously the total space of an Ω1 Pn via the first and the second projection respectively. divisor D with equation Pn Pn -torsor νi : V → Pn associated with the Euler exact sequence 0 → Ω1 Pn ⊕ OPn (−k)-torsor and of an Ω1 1Pic(Pn) ⊕ p∗ References 1. A. Dubouloz, Danielewski-Fieseler Surfaces, Transformation Groups 10 (2005), no. 2, 139 -- 162. 2. A. Dubouloz and D. Finston, On exotic affine 3-spheres, Preprint arXiv:1106.2900v1 (2011). 3. K.-H. Fieseler, On complex affine surfaces with C+-action, Comment. Math. Helv. 69 (1994), no. 1, 5 -- 27. 4. H. Flenner, S. Kaliman, and M. Zaidenberg, On the Danilov-Gizatullin Isomorphism Theorem, L'Enseignement Math ´matique 55 (2009), no. 2, 1 -- 9. 5. M. H. Gizatullin and V. I. Danilov, Automorphisms of affine surfaces II, Izv. Akad. Nauk SSSR Ser. Mat. 41 (1977), no. 1, 54 -- 103. 6. A. Grothendieck, Sur la classification des fibrés holomorphes sur la sphère de Riemann, American Journal of Maths. 79 (1957), no. 1, 121 -- 138. 7. A. Grothendieck and J. Dieudonné, EGA IV. Étude locale des schémas et des morphismes de schémas, Publ. Math. IHES, vol. 20 (1964), 24 (1965), 28 (1966), 32 (1967). CNRS, Institut de Mathématiques de Bourgogne, Université de Bourgogne, 9 avenue Alain Savary - BP 47870, 21078 Dijon cedex, France E-mail address: [email protected]
1009.3219	1	1009	2010-09-16T16:49:04	Flash Geometry of Algebraic Curves	[ "math.AG" ]	In this paper, we develop the theory of flashes of an algebraic curve. We show that the theory is birationally invariant in a sense which we will make more precise below. We also show how the theory provides a foundation for the method of asymptotic degenerations, a particular class of degenerations of plane projective algebraic curves. In particular, we consider the geometrical technique in relation to the Severi problem of degenerating nodal curves to lines in general position, and correct Severi's original proof of his conjecture.	math.AG	math	FLASH GEOMETRY OF ALGEBRAIC CURVES TRISTRAM DE PIRO Abstract. In this paper, we develop the theory of flashes of an algebraic curve. We show that the theory is birationally invariant in a sense which we will make more precise below. We also show how the theory provides a foundation for the method of asymptotic degenerations, a particular class of degenerations of plane projec- tive algebraic curves. In particular, we consider the geometrical technique in relation to the Severi problem of degenerating nodal curves to lines in general position, and correct Severi's original proof of his conjecture. 1. Degenerating Nodal Curves In this section, we return to the convention in [6] that a plane alge- braic curve C is an irreducible subvariety of P 2, of dimension 1. By a node, we mean an ordinary double point, as in Definition 1.2. of [8]. By a plane nodal curve, we will mean a plane algebraic curve which has at most ordinary double points as singularities, as defined in [8]. In particular, this convention is different from the use of the term "node" in [6], see also the explanation in [8]. We will also occasionally make use of the further assumption on C, see [6] and [7]; A generic point of C has character (1, 1) (†) This is to exclude certain exceptional curves in non-zero character- istic. We will return to these exceptional cases in the final section. An interesting geometric problem in the theory of nodal curves is to provide a general method of "degenerating" a plane nodal curve C of degree m into a union of m lines, having transverse intersections. More specifically, if C is a plane nodal curve of degree m, one can consider degenerations, over an irreducible parameter space, which, for conve- nience of terminology, we denote by P 1, of the form {Ct}t∈P 1 with the following properties; 1 2 TRISTRAM DE PIRO (i). The family {Ct : t ∈ P 1} consists of plane reduced algebraic curves. (ii). C0 = C. (iii). For any t ∈ P 1, all the singularities of Ct are nodes. (iv). For any t ∈ (P 1 \ ∞), Ct is a nodal curve (in particularly irre- ducible) and has the same number of singular points as C. (v) C∞ is a union of m lines {l1, . . . , lm}. if m ≥ 3, in general position, that is, the lines are distinct and, there does not exist a triple (j1, j2, j3), with j1 < j2 < j3 such that (lj1 ∩ lj2) = (lj2 ∩ lj3). Severi referred to a plane nodal curve, having such a degeneration, as a "nodal curve of the main stream". He conjectured that any nodal curve is a nodal curve of the main stream. However, his proof of the result is erroneous. We do not attempt to answer this conjecture fully in the following paper. However, it is hoped that the general theory of flash geometry, which we will develop here, can help to resolve the fallacious steps in Severi's argument. We will return to this question in a later section. The following lemma will be used repeatedly in arguments concern- ing nodal curves. The reader should look at [6] or [7] for relevant terminology. Lemma 1.1. Nodal Curve Presentation Let C be a plane nodal curve of degree m, satisfying (†). Then there exists an affine coordinate system (x, y) for P 2, such that, in this co- ordinate system; (i). The line at ∞ cuts C transversely in m distinct non-singular points. (ii). The tangent lines to C parallel to the y-axis all have 2-fold con- tact(contatto), and are based at non-singular points of C, which are in finite position. FLASH GEOMETRY OF ALGEBRAIC CURVES 3 (iii). Considering x as a rational function on C, in the terminology of [7], the weighted set G = (x = 0) consists of m distinct branches, each counted once, lying inside NonSing(C). Moreover, these branches are all in finite position, with base points distinct from the points of con- tact in (ii). (iv). There is no line l, parallel to the y-axis, intersecting C in more than one point of the set consisting of its finitely many ordinary double points or the finitely many points named in (ii). Proof. By the assumption (†) and Remarks 6.6 of [6], there exist only finitely many non-ordinary branches. Hence, there exist finitely many tangent lines {lγ1, . . . , lγr}, based at {p1, . . . , pr}, (possibly with repe- titions), such that; Iγj (C, pj, lγj ) ≥ 3, (for 1 ≤ j ≤ r) , lγq2 s s , lγ2 q1 , . . . , lγq1 Moreover, C has finitely many ordinary double points {q1, . . . , qs} as } be the 2s tangent lines, (pos- singularities. Let {lγ1 q1 sibly with repetitions), corresponding to these ordinary double points. Let {lqiqj : 1 ≤ i < j ≤ s} be the finitely many lines, (possibly with repetitions), passing through at least 2 of these ordinary double points. Let {l1, . . . , lt} be the finitely many lines which are tangent to at least 2 distinct branches of C. That there exist finitely many such lines fol- lows from duality arguments, the reader should look at Lemma 5.11 of [10]. Finally, let {l1, . . . , lw} be the finitely many lines, which have the property that, they pass through one of the ordinary double points {q1, . . . , qs}, and are also tangent to a branch, centred at a nonsingular point. That there are finitely many such lines follows from considera- tion of each of the finite number of pencils centred at {q1, . . . , qs}, and results from [6]. Now choose a point P not lying on C or any of the above defined lines. Let Σ = {lP λ }λ∈P 1 be the pencil defined by all lines passing through the point P . Then Σ defines a g1 m on C without fixed branches. By Lemma 2.17 of [6], for generic λ, lP λ intersects C transversely in m distinct branches, based at non singular points of C. By construction, we also have that, if lP x belongs to Σ and defines the tangent line lγx to a branch γx based at x, then x ∈ NonSing(C), lP x has 2-fold contact (contatto) with the branch γx, and lP x cannot pass through either an ordinary double point or be tangent to another branch γx′ of C, for x′ distinct from x. (∗). Moreover, if lP x belongs to Σ, it can pass through 4 TRISTRAM DE PIRO Z ′ and y′ = Y ′ at most one ordinary double point of C, (∗∗). Now choose a homogra- phy, sending the point [0 : 1 : 0] and the line Z ′′ = 0, in the original coordinates [X ′′ : Y ′′ : Z ′′] on P 2, to P and lP λ . Let [X ′ : Y ′ : Z ′] be the new coordinate system defined by this homography. For the affine coordinate system (x′, y′), defined by x′ = X ′ Z ′ , we have that the line at ∞ has the property (i). The lines parallel to the y′- axis correspond to the lines, excluding lP λ , in the pencil defined by Σ, in this new coordinate system. Hence, (ii) and (iv) follows immediately from (∗) and (∗∗). Now, considering x′ as a rational function on C, by the same argument as above, for generic µ 6= ∞, the line defined by (x′ = µ), intersects C transversely in m distinct non-singular branches. Moreover, as P ([0 : 1 : 0] in this new coordinate system) does not lie on C, it follows that all these intersections lie in finite position, (∗ ∗ ∗). Now, let θ be the homography of P 2 defined by the affine translation (x′, y′) 7→ (x′ − µ, y′) and let (x, y) be the new coordinates defined by this map. As P and the line at ∞ are fixed by θ, and, moreover, the tangent lines parallel to the y′-axis are translated to tangent lines par- allel to the y-axis, conditions (i), (ii) and (iv) are preserved. By (∗ ∗ ∗), we then have that (iii) holds as well. (cid:3) 2. Newtonian Methods We now discuss some geometric methods, originally developed by Isaac Newton, that constitute the main ideas behind the technique of flash geometry. Definition 2.1. Asymptotes, Hyperbolic and Parabolic Branches Let C be a plane irreducible algebraic curve of degree m, defined in the coordinate system (x, y), not equal to a line. We define a line l(a,b,t), given by the equation ax + by = t, to be an asymptote if; (i). l(a,b,t) passes through one of the intersections p ∈ C ∩ l∞ (ii). l(a,b,t) is tangent to at least one of the branches γp centred at p. We define a branch γp, centred at one of the intersections p ∈ C ∩l∞, to be an infinite branch, and a finite branch otherwise. We also define; γp to be parabolic if the line l∞ is the tangent line to the branch. FLASH GEOMETRY OF ALGEBRAIC CURVES 5 γp to be hyperbolic otherwise. If C is equal to a line l, and this line does not coincide with the line l∞, we define l to be its asymptote, and its one infinite branch γ∞ to be hyperbolic. If C is the line l∞, we say that it has no asymptotes and all its infinite branches are parabolic. If C is not irreducible, we consider its distinct irreducible factors {C1, . . . , Ck}. The asymptotes of C are defined as the union of the asymptotes of each factor Cj, for 1 ≤ j ≤ k. Similarly, its branches are classified according to the definition above, for each irreducible fac- tor. As a simple consequence of this definition, we have; Lemma 2.2. Let notation be as in Definition 2.1 and let C be any plane algebraic curve (possibly not irreducible) of degree m, having fi- nite intersection with l∞. Then C has r1 ≤ m parabolic branches and r2 ≤ m hyperbolic branches with r1 + r2 ≤ m. In particular there exist r3 ≤ r2 ≤ m distinct asymptotes to the curve C. If C is a plane nodal curve and the coordinate system is chosen as in Lemma 1.1, then there exist exactly m asymptotes to the curve C. Proof. Suppose first that C is irreducible. As we may assume that C has finite intersection with l∞, by the Hyperspatial Bezout Theorem (Theorem 2.3 of [6]), we have that; Pinf inite branches γp Iγp(C, l∞) = m (∗) It follows trivially that r1 + r2 ≤ m. If l(a,b,t) is an asymptote, then it defines the tangent line to an infinite branch γp. As l(a,b,t) is dis- tinct from the line l∞, such a branch γp must be hyperbolic. Any (hyperbolic) branch has a unique tangent line, hence distinct asymp- totes must define tangent lines to distinct hyperbolic branches. There- fore, r3 ≤ r2 ≤ m. If C is not irreducible, let {C1, . . . , Cj, . . . , Ck} be its distinct irreducible factors, let r1,j be the number of parabolic branches of Cj and let r2,j be the number of hyperbolic branches of Cj. If deg(Cj) = mj, we clearly have that m1 + . . . + mj + . . . + mk ≤ m, (∗∗). Applying the previous considerations to each irreducible factor Cj, we have that r1,j +r2,j ≤ mj. Applying the above Definition 2.1 and (∗∗), we have that r1 + r2 ≤ m. Again, by a similar argument to the above, we have that r3 ≤ r2 ≤ m. For the final part of the lemma, by 6 TRISTRAM DE PIRO (i) of Lemma 1.1 and (∗), we have that there exist m distinct infinite hyperbolic branches. The tangent lines of these hyperbolic branches must be distinct, otherwise at least one would coincide with the line l∞, hence there exist exactly m asymptotes as required. (cid:3) Remarks 2.3. An elegant method of determining the real asymptotes to a given plane real irreducible algebraic curve is given in [25]. I hope the author will not mind me essentially restating his results in the con- text of algebraically closed fields. Theorem. (adapted almost verbatim from Nunemacher) Let C be any any plane algebraic curve and let C be defined in the coordinate system (x, y) by; P (x, y) = Pm k=0 Pk(x, y) where Pk(x, y) is a homogeneous polynomial of degree k. Suppose that (ax + by) is a factor of the top degree form Pm(x, y) of multiplicity r. Let s ≤ r denote the largest integer with the property that there ex- ist polynomials Qj(x, y) for m−s+1 ≤ j ≤ m satisfying the conditions; Pm(x, y) = (ax + by)sQm(x, y) Pm−1(x, y) = (ax + by)s−1Qm−1(x, y) . . . . . . Pm−s+1(x, y) = (ax + by)Qm−s+1(x, y) (∗) Then associated with the factor (ax+ by) is a set of at most s asymp- totes ax + by = t0, where t0 is a root of the equation; tsQm(b, −a)+ts−1Qm−1(b, −a)+. . .+tQm−s+1(x, y)+Pm−s(b, −a) = 0 All the asymptotes to the curve C in the coordinate system (x, y) arise in this way as (ax + by) ranges over the linear factors of Pm(x, y) Corollary. (adapted almost verbatim from Nunemacher) With the notation and hypotheses of the above theorem, if (ax + by) is a simple factor of Pm(x, y), then associated with this factor is the FLASH GEOMETRY OF ALGEBRAIC CURVES 7 single asymptote to C, defined by the equation; (ax + by)Qm(b, −a) + Pm−1(b, −a) = 0 The proof of the Theorem and its Corollary follow easily from the paper [25]. We have strengthened the hypotheses of the Theorem to include any plane algebraic curve, the corresponding Theorem in [25] excludes the degenerate case that C contains a line as one of its ir- reducible factors. However, given Definition 2.1, which includes this case, one can check that his proof is unaffected in this greater degree of generality. In order to give an illustration of the theorem and its corollary, con- sider the following curves; (i). The hyperbola defined by xy = λ. In this case, we have that P2(x, y) = xy, whose simple linear factors are x and y, and P1(x, y) = 0. By the corollary, the asymptotes are given by; x.y(0, −1) + 0(0, −1) = 0 that is x = 0 y.x(1, 0) + 0(1, 0) = 0 that is y = 0 (Observe that this gives the correct result even in the degenerate case when λ = 0) (ii). The parabola defined by x2 − y = 0. In this case, we have that P2(x, y) = x2, which has x as a linear factor of multiplicity 2, and P1(x, y) = −y. We have that x does not divide P1(x, y), hence, by the theorem, the asymptotes are given by x = t0, where t0 is a root of; t.x(0, −1) − y(0, −1) = 0 that is 1 = 0 As this has no solutions, we conclude that the parabola has no asymp- totes. (iii). The curve defined by; 8 TRISTRAM DE PIRO xy2 + ey = a2x3 + bx2 + cx + d, see (1) of Remarks 2.4. Case 1. a 6= 0. In this case, P3(x, y) = xy2−a2x3, which, assuming char(L) 6= 2, has three distinct linear factors {x, (y +ax), (y −ax)}, and P2(x, y) = −bx2. By the corollary, the asymptotes are given by; x.(y2 − a2x2)(0, −1) + (−bx2)(0, −1) = 0 that is x = 0 (y − ax)(xy + ax2)(1, a) + (−bx2)(1, a) that is y − ax = b 2a (y + ax)(xy − ax2)(1, −a) + (−bx2)(1, −a) that is y + ax = −b 2a The curve therefore has three distinct asymptotes, see (2) of Remarks 2.4. If b = 0, these asymptotes intersect at the origin (0, 0), see (3) of Remarks 2.5. If b 6= 0, these asymptotes intersect in a triangle with vertices centred at {(0, b 2a2 , 0)}, see (4) of Remarks 2.4. 2a ), (0, −b 2a ), ( −b Case 2. a = 0, b 6= 0. In this case, P3(x, y) = xy2, which has x as a simple factor and y as a linear factor of multiplicity 2, again P2(x, y) = −bx2. By the corol- lary, one asymptote is given by; x.y2(0, −1) + (−bx2)(0, −1) = 0 that is x = 0. As y does not divide P2(x, y), the theorem gives the other asymptotes as y = t0, where t0 is a root of; t.(xy)(1, 0) + (−bx2)(1, 0) = 0 that is b = 0 As this has no solutions, we conclude there are no further asymp- totes, see (5) of Remarks 2.4. Case 3. a = 0, b = 0. By the same reasoning as Case 2, we obtain x = 0 as an asymptote. In this case, however, y divides P2(x, y) = 0, so we have to consider P1(x, y) = ey −cx. By the theorem, the other asymptotes are then given by y = t0, where t0 is a root of; FLASH GEOMETRY OF ALGEBRAIC CURVES 9 t2.x(1, 0) + t.0(1, 0) + (ey − cx)(1, 0), that is t2 − c = 0 In the case when c = 0, we obtain one further asymptote y = 0, oth- erwise, assuming char(L) 6= 2, we obtain 2 further asymptotes y = c and y = −c 2 , see (6) of Remarks 2.4. 1 1 2 (iv). The curve defined by; xy = ax3 + bx2 + cx + d = 0, see (7) of Remarks 2.4. We have that P3(x, y) = ax3, which has x as a linear factor of mul- tiplicity 3, P2(x, y) = bx2 − xy. By the theorem, the asymptotes are given by x = t0, where t0 is a root of; t2.x(0, −1) + t.(bx − y)(0, −1) + cx(0, −1) = 0 that is t = 0 Hence, x = 0 is the only asymptote to this curve. (v). The curve defined by; y2 = ax3 + bx2 + cx + d, with a 6= 0, see (8) of Remarks 2.4. We have that P3(x, y) = ax3, which has x as a linear factor of multi- plicity 3, P2(x, y) = bx2 − y2. By the theorem, the asymptotes are given by x = t0, where t0 is a root of; t.x2(0, −1) + (bx2 − y2)(0, −1) = 0, that is −1 = 0 This has no solutions, hence the curve has no asymptotes. (vi). The curve defined by; y = ax3 + bx2 + cx + d, with a 6= 0, see (9) of Remarks 2.4 We have that P3(x, y) = ax3, which has x as a linear factor of mul- tiplicity 3, P2(x, y) = bx2, and P1(x, y) = cx − y. By the theorem, the asymptotes are given by x = t0, where t0 is a root of; t2.x(0, −1) + t.bx(0, −1) + (cx − y)(0, −1) = 0, that is 1 = 0 10 TRISTRAM DE PIRO This has no solutions, hence the curve has no asymptotes. Remarks 2.4. Isaac Newton gives the first effective method of deter- mining the asymptotes of a real algebraic curve C in [22]. The technique is part of a more general construction called "Newton's parallelogram". An excellent account of this construction is given in [27]. Moreover, Newton makes a systematic use of the method of asymptotes in his classification of real cubic curves, see [23] and [27]. His analysis of such curves in [23] begins with the observation; "All lines of the 1st, 3rd, 5th, 7th, or odd orders, have at least two infinite branches extending in opposite directions" In the case of a real cubic curve C, this translates to the fact that the line l∞ intersects C in at least 1 point, counted with multiplicity (appropriately defined). The observation is easily seen to be true from the fact that any real polynomial p ∈ R[y] of degree 3 has at least 1 real solution and consideration the limiting behavior of such a solution as x → ∞ or x → −∞ in the defining equation P (x, y) of C. Exclud- ing the degenerate case when C has a parabolic branch, that is the line l∞ intersects C in a point with multiplicity 2, we obtain a hyperbolic branch and at least 1 asymptote to the curve C. In this case, by taking such an asymptote to be the coordinate axis x = 0, the general equation of C reduces to the form; xy2 + (bx2 + cx + d)y = cx3 + dx2 + ex + f (∗) Such a quadratic equation can be analysed using the simple method of completing the square. This construction forms the basis for Newton's division of real cubics into four Cases. These Cases are then further analysed according to the behaviour of their asymptotes and the exis- tence of a diameter (see [23] and [27] for a precise definition.) This leads to Newton's division of real cubics into a total of 72 species. The curves considered as examples in the previous remark constitute a num- ber of these species; (1). Newton refers to this equation as The Case I cubic, see [23]. (2). Newton refers to such a curve, in [23], as a redundant or triple hyperbola (if a2 > 0), and as a defective hyperbola (if a2 < 0). FLASH GEOMETRY OF ALGEBRAIC CURVES 11 (3). The triple hyperbolas with this property are considered as Species 24-32 in Newton's classification of cubics, see [23]. (4). The triple hyperbolas with this property are considered as Species 1-23 in Newton's classification of cubics (and are subdivided further according to the number of diameters). The defective hyperbolas are considered as Species 33-45 in Newton's classification, see [23]. (5). Newton refers to such curves as parabolic hyperbolas, they ac- count for Species 46-56 of his classification, see [23]. (6). Newton refers to such curves as the hyperbolisms of the hyper- bola ,(c > 0), ellipse, (c < 0), or the parabola, (c = 0), they account for species 57-60, species 61-63 and species 64-65 respectively, see [23]. (7). Newton calls this equation The Case II cubic, see [23], also known as Newton's trident. (8). Newton calls this equation The Case III cubic, see [23], referred to there as the Divergent Parabola. (9). Newton calls this equation The Case IV cubic, see [23], referred to there as the Cubic Parabola. Perhaps the most important ingredient in Newton's analysis of cubics is the use that he makes of asymptotes as a means of representing cubic curves. A purely algebraic proof of his result would be inconceivable without the supporting intuition of such a representation. The reader is strongly encouraged to look at the illustrations that Newton originally gave of each species in his classification, see [27]. The fact that Newton referred to a cubic curve (curve of degree m) as a "line of the third order" (a "line of the m'th order"), clearly shows that he thought of such curves in terms of a system of lines, (†), (1). I discuss this idea in greater detail in my forthcoming book "Christian Geometry", where I consider the aesthetics ideas underlying Newton's work. The principle (†) is also illustrated by the problem of degenerating curves of degree m to a union of m lines, which we discussed in the first section. 1In the following section, this viewpoint will become clearer when we introduce the notion of flashes, as a means of representing an algebraic curve. 12 TRISTRAM DE PIRO Using the coordinate system of Lemma 1.1, we can find the following decomposition of C; Lemma 2.5. Newton's Theorem Let hypotheses be as in Lemma 1.1, and let F (x, y) define C in the coordinate system (x, y) given by Lemma 1.1. Then we can find alge- braic power series {η1(x), . . . , ηm(x)} in L[[x]], with {η1(0), . . . , ηm(0)} distinct, such that; F (x, y) = (y − η1(x)) . . . (y − ηm(x)) as an identity in the ring L[[x]][y]. Proof. Let {a1, . . . , am} enumerate the m distinct solutions to F (0, y) = 0. Making the linear change of coordinates y′ = y − aj, the equation for C in this system is given by; Fj(x, y′) = F (x, y′ + aj) = 0 (†) By construction, we have that Fj(0, 0) = 0 and, moreover, by (iii) of Lemma 1.1, ∂Fj ∂y (0, 0) 6= 0. It follows, applying the implicit function theorem, that we can find an algebraic power series δj(x) ∈ L[[x]], with δj(0) = 0, such that Fj(x, δj(x)) = 0, (∗), see especially the following remark. Now let ηj(x) = aj + δj(x). By (†), it follows immediately that F (x, ηj(x)) = 0, for 1 ≤ j ≤ m. Hence, we have that; (y − ηj(x))F (x, y), for 1 ≤ j ≤ m, in L((x))[y] using the fact that the division property holds in the polynomial ring L((x))[y], as L((x)) is a field. We clearly have that each (y − ηj(x)) is irreducible in L((x))[y], hence prime. Moreover, as the power series {η1(x), . . . , ηm(x)} are distinct, the factors (y − ηj(x)) are coprime in L((x))[y], for 1 ≤ j ≤ m. It, therefore, follows that; Q1≤j≤m(y − ηj(x))F (x, y) in L((x))[y] (∗∗) Now, using the fact that both sides of the relation have degree m, we can find g(x) ∈ L((x)) such that; F (x, y) = g(x)(y − η1(x)) . . . (y − ηm(x)) in L((x))[y] FLASH GEOMETRY OF ALGEBRAIC CURVES 13 As F (0, y) = 0 has m distinct solutions in L, we can write F in the form; F (x, y) = ym + p1(x)ym−1 + . . . + pj(x)ym−j + . . . + pm(x) with pj ∈ L[x] and deg(pj) = j. It follows that we can take g(x) = 1 in (∗∗), so that the identity holds in L[[x]][y]. This shows the lemma. (cid:3) Remarks 2.6. The existence of power series solutions to polynomial equations G(x, y) = 0, satisfying (∗) of the previous lemma, was first shown by Isaac Newton. He gives the following method of constructing a solution by successive approximations, in "The Praxis of Resolution", Paragraph 36 of [24]; The assumption that G(0, 0) = 0 and ∂G ∂y (0, 0) 6= 0 allows us to write; G(x, y) = pm(x)ym + . . . + p1(x)y + p0(x) (∗) with p0(0) = 0 and p1(0) 6= 0. For (x, y) "small", Newton observes that it, therefore, makes sense to take as a first approximate solution; y0 = −λ0xi0 p1(0) where λ0xi0, (i0 ≥ 1), is the first term in the expression for p0(x). Now, Newton makes the substitution y = (y′ + y0) in (∗), this results in a further polynomial equation of the same form; qm(x)y′m + . . . + q1(x)y′ + q0(x) = 0 (∗∗) By a straightforward algebraic calculation, using the fact that i0 ≥ 1, one checks that q1(0) 6= 0 and ord(q0(x)) > ord(p0(x)). Hence, one can take as the second quote; y1 = −λ1xi1 q1(0) where λ1xi1, (i1 > i0), is the first term in the expression for q0(x). Continuing in this way, one obtains a sequence of approximate solu- tions; sn(x) = y0(x) + y1(x) + . . . + yn(x), for n ≥ 0 14 TRISTRAM DE PIRO Either this process terminates after a finite number of approxima- tions, giving a polynomial solution to (∗), or one obtains an infinite power series; s(x) = Pi≥0 yi(x) A straightforward algebraic calculation shows that; ord(G(x, sn+1(x))) ≥ ord(G(x, sn(x))) + 1 Hence, by elementary completeness arguments for the power series ring L[[x]], we are guaranteed that G(x, s(x)) = 0. This method of constructing a solution to the polynomial equation G(x, y) = 0 is closely related to the possibly more familiar Newton- Raphson method. Namely, one considers the function; G : L[x] → L[x], G(y) = pm(x)ym + . . . + p1(x)y + p0(x) (†) Having obtained a first approximation y0 = s0 to the equation G(y) = 0, the Newton-Raphson method gives the further approximation; s1 = y0 − G(y0) G′(y0) = y0(x) − q0(x) q1(x) where q0(x) and q1(x) are obtained from the transformed polynomial (∗∗) above. By a similar argument to the above, replacing the succes- sive approximations − λ1xi1 q1(x) and noting that; q1(0) by − q0(x) q0(x) q1(x) = λ1xi1 q1(0) u(x) for a unit u(x) ∈ L[[x]] one is similarly guaranteed that this method also yields the same power series solution s(x) to the equation G(y) = 0 in (†). The Newton-Raphson method is usually applied to functions of a real vari- able, rather than the rings L[x] or L[[x]]. However, that Newton in- tended his method of finding power series solutions to polynomial equa- tions of the form G(x, y) = 0, ("species"), to be a partial generalisation of this method is borne out by his consideration of the case of "affected equations", at the beginning of [24]. An affected equation is just the case where the indeterminate x is replaced by an explicit numerical value. Newton also gives a general method for finding power series solutions to polynomial equations of the form G(x, y) = 0, without the simplifying FLASH GEOMETRY OF ALGEBRAIC CURVES 15 1 1 assumption that ∂G ∂y (0, 0) 6= 0. This is done by the introduction of "New- ton's parallelogram", in Paragraph 29 of [24], and fractional exponents t , for t ≥ 1. The parallelogram method is further explained in [22]. x This leads to a general method of finding n solutions yj(x) ∈ L((x t )), for 1 ≤ j ≤ n, of a polynomial equation G(x, y) = 0 of degree n, see [1] for a more precise statement of "Newton's Theorem". However, it seems clear from Newton's highly geometric approach, that he intended the theorem to provide a method of understanding the global structure of an algebraic curve, rather than as a means of analysing the nature of local singularities. This view is supported by consideration of his work on the classification of plane cubics, in [23]. Here, his power series method is used to find the asymptotes of such curves, which he refers to as "infinite branches", and branches at finite distances (those crossing the axis x = 0), also referred to in the literature as "satellite branches". Much of the work in [23] is concerned with "filling in" the rest of the curve, from this initial information. This is also the approach taken by a number of Newton's followers in England, see, for example, the papers [26],[21] and the commentary on [23], in [27]. As one can al- ways choose the "satellite branches" to be non-singular, see Lemma 1.1, one can obtain a solution to "Newton's Theorem" without introducing "Puiseux Series", as, indeed, is done in the previous Lemma 2.5. It is extremely unclear how to give a geometric interpretation of "Puiseux Series". It seems, therefore, to be a rather unfortunate historical ac- cident that these were introduced, as a response to Newton's work, in order to analyse algebraic curves. The proofs that we give in the fol- lowing section, which develop flash geometry, closely follows Newton's original approach. I also discuss the general aesthetics behind Newton's representation of algebraic curves in "Christian Geometry" (in preparation). As already mentioned, the reader would benefit greatly from looking at Newton's original "sketches" of cubic curves in [23]. 3. Flash Geometry of Nodal Curves Definition 3.1. We will denote the coordinates of P 2 by {X, Y, W } or {X, Z, W }, the coordinates of P 3 by {X, Y, Z, W } and the points {[0 : 0 : 1 : 0], [0 : 1 : 0 : 0]} by {Q1, Q2}. We let {pr1, pr2} be the canonical conic projection morphisms defined by; pr1 : (P 3 \ Q1) → P 2 1 ; pr1([X : Y : Z : W ]) = [X : Y : 0 : W ] 16 TRISTRAM DE PIRO pr2 : (P 3 \ Q2) → P 2 2 ; pr2([X : Y : Z : W ]) = [X : 0 : Z : W ] 1 ⊂ P 2 We let A2 W , and l1,∞ ⊂ P 2 1 . Similarly, we let A2 1 be the open subset defined by the coordinate system W and y = Y (x, y), where x = X 1 be the line at infinity 2 ⊂ P 2 defined by (W = 0) ∩ P 2 2 be the open sub- W and z = Z set defined by the coordinate system (x, z), where x = X W , and l2,∞ ⊂ P 2 2 . We let A3 ⊂ P 3 be the open subset defined by the coordinate system (x, y, z), where x = X W . The restriction of the projection morphisms {pr1, pr2} are then given in affine coordinates by; 2 the line at infinity defined by (W = 0) ∩ P 2 W and z = Z W , y = Y pr1 : A3 → A2 1; pr1(x, y, z) = (x, y) pr2 : A3 → A2 2; pr2(x, y, z) = (x, z) We will also refer to the projected points {pr2(Q1), pr1(Q2)} as {Q1, Q2}, and the restriction of the projections {pr1, pr2} to {A2 2, A2 1} by pr. We observe the following straightforward property; Lemma 3.2. There does not exist a pair {a, b}, distinct from {Q1, Q2}, with pr−1 1 (a) = pr−1 2 (b). Proof. Observe that a fibre of pr1 consists of a line passing through Q1, with the point Q1 removed. Similar considerations apply to pr2, (†). Suppose a pair, as in the lemma, existed. If {a, b} are distinct, then, by (†), we would clearly have that pr−1 2 (b) = lab = lQ1Q2. As the image of pr1◦pr2 is P 2 2 is fixed both by pr1 and pr2, (∗), it follows that a = pr2(a) = pr1(b) = b. Hence, we have that pr−1 2 , which clearly contradicts (∗). 2 , this implies that lab = P 2 1 (a) = pr−1 1 (a) = pr−1 2 (a) = P 2 2 . As P 2 1 ∩ P 2 1 ∩P 2 1 ∩P 2 1 ∩P 2 We now show the following; (cid:3) Lemma 3.3. Let hypotheses be as in Lemma 2.5 and notation as in Definition 3.1, then we can find an irreducible plane projective curve C ′ ⊂ P 2, relative to the coordinate system (x, z), such that; (i). There exists an open V ′ ⊂ C ′, with (0, 0) ∈ V ′, such that; FLASH GEOMETRY OF ALGEBRAIC CURVES 17 pr : (V ′, (00)) → (A1, 0), (∗) is an etale morphism. (ii). {δ1(x), . . . , δm(x)} ⊂ R(V ′), for a given embedding of R(V ′) in L[[x]] ∩ L(x)alg. (iii). The function field L(V ′) is generated over L(x) by the power series {δ1(x), . . . , δm(x)}. (iv). The cover defined by (∗) is Galois, with the extension L(V ′)/L(x) equal to the Galois closure of the extension L(C)/L(x). Proof. From the previous construction of Lemma 2.5, the power se- ries {δ1(x), . . . , δm(x)} belong to L[[x]] ∩ L(x)alg. By Theorem 1.3 and subsequent remarks at the beginning of Section 3 of [5], we have that L[[x]] ∩ L(x)alg is isomorphic to the local ring for the etale topology Oet (A1,0). By the definition of this local ring, we can find irreducible varieties; {(U1, (0)lif t 1 ), . . . , (Ui, (0)lif t i ), . . . , (Um, (0)lif t m )} and etale morphisms; φi : (Ui, (0)lif t i ) → (A1, 0), for 1 ≤ i ≤ m with δi(x) ∈ R(Ui). It follows, from algebraic considerations, see, for example, [17], that the fibre product; (U1...m, (0)lif t) = (U1 ×A1 × . . . ×A1 Um, (0)lif t 1 ×(0) . . . ×(0) (0)lif t m ) is irreducible, nonsingular, and defines etale covers; ψ1...m : (U1...m, (0lif t)) → (A1, 0) ψi : (U1...m, (0lif t)) → (Ui, (0)lif t i ), for 1 ≤ i ≤ m such that φi ◦ ψi = ψ1...m. By the definition of the etale local ring, in terms of maps and equivalence relations, we can assume that all the power series {δ1(x), . . . , δm(x)} are represented in the coordinate ring R(U1...m). 18 TRISTRAM DE PIRO The dominant morphism ψ1...m induces an inclusion; L(x) ⊂ L(U1...m) and a corresponding factorisation; L(x) ⊂i1 L(x, δ1(x), . . . , δm(x)) ⊂i2 L(U1...m),(2) Let C ′′ be an irreducible nonsingular model of L(x, δ1(x), . . . , δm(x)), with corresponding dominant morphisms; α2 : U1...m → C ′′, α1 : C ′′ → A1 corresponding to the inclusions {i1, i2}, such that α1 ◦ α2 = ψ1...m. Let (0)lif t ∈ C ′′ also denote the image α2((0)lif t), for (0)lif t ∈ U1...m. As the composition α1 ◦ α2 is etale, it follows, using an elementary Zariski structure argument, that the morphism α1 is Zariski unrami- fied at (0)lif t ∈ C ′′. By nonsingularity of C ′′, Footnote 2 and Theo- rems (6.10,6.11) of [7], there exists an open subset W ⊂ C ′′, containing (0)lif t, such that the restriction; α1 : (W, (0)lif t) → (A1, 0) is etale. We now use the local presentation lemma for etale morphisms, given in Fact 1.5 of [5], that is we can find an open subset U ′ ⊂ A1, contain- ing (0), and a monic polynomial F (z) ∈ R(U ′)[z], such that α1 can be presented for an open subset V ′ ⊂ W ⊂ C ′′ in the form; Spec(( R(U ′)[z] F (z) )d) → Spec(R(U ′)) (∗∗) with F ′(z) invertible in ( R(U ′)[z] F (z) )d. Now let p(x) ∈ L[x] be a polyno- mial, with p(0) 6= 0, such that R(U ′) = L[x]p. Then we can write the equation for F (z) in the form; F (z) = zn + c1(x) prn (x) = 0 pr1 (x)zn−1 + . . . + cn−1(x) prn−1 (x) z + cn(x) 2When char(L) 6= 0, using the fact that the roots η1(x), . . . , ηm(x), found in Lemma 2.5, are distinct, it is straightforward to check that L(x, δ1(x), . . . , δm(x)) is a seperable extension of L(x). FLASH GEOMETRY OF ALGEBRAIC CURVES 19 for some n ≥ 1 and 0 ≤ rj ≤ r, with cj(x) coprime to p(x) for 1 ≤ j ≤ n. Clearing coefficients and using the fact that V ′ is irre- ducible, the polynomial; G(x, z) = pr(x)zn + c1(x)pr−r1(x)zn−1 + . . . + cn(x)pr−rn(x) defines an irreducible curve C ′ in A2 2, whose restriction to V ′ = (pr−1 1 (U ′) ∩ (G = 0) ∩ (d 6= 0)) corresponds to the cover defined in (∗∗). By making a vertical translation of the curve, we can assume that G(0, 0) = 0 and (0, 0) corresponds to the point (0)lif t of V ′ ⊂ W ⊂ C ′′. We finally check the properties (i) to (iv) for (V ′, C ′, (0, 0)). (i). Follows immediately from the presentation (∗∗). (ii). As there exists a birational map between (V ′, C ′′) and (V ′, C ′) over (A1, 0), we can assume that the power series {δ1(x), . . . , δm(x)} ⊂ R(V ′) ⊂ L(C ′). That δ1(0, 0) = . . . = δm(0, 0) = 0, follows from the definition of the maximal ideal met (A1,0) ⊂ Oet (A1,0). (iii). Follows from the construction of C ′′ and the fact that {C ′, C ′′} are birational. (iv). Follows from (iii) and the fact that L(x, δ1(x), . . . , δm(x)) = L(x, η1(x), . . . , ηm(x)) is a splitting field for the polynomial F (x, y), given in Lemma 2.5. The existence of a curve C ′ ⊂ P 2, satisfying the conditions of the lemma, is far from unique. More precisely, we have the following; (cid:3) Lemma 3.4. Let C1 satisfy the conditions of Lemma 3.3 and let; g(x, z) = a(x)z+b(x) c(x)z+d(x) with {a(x), b(x), c(x), d(x)} polynomials in L[x], such that b(0) = 0 and a(0)d(0) 6= 0 Then g(x, z) determines a birational,(3), morphism θg : C1 ! C2, with C2 satisfying the conditions of Lemma 3.3. 3Unless otherwise stated, when speaking of a birational morphism θ between curves C1 and C2, satisfying the conditions of Lemma 3.3, we will always assume 20 TRISTRAM DE PIRO Proof. As a(0)d(0) 6= b(0)c(0), there exists an open subset U ⊂ A1, containing 0, such that, for all x ∈ U, g(x, z) = a(x)z+b(x) c(x)z+d(x) defines an invertible Mobius transformation. Let; V1 = {(x, z) ∈ U × A1 : c(x)z + d(x) 6= 0} V2 = {(x, z) ∈ U × A1 : c(x)z − a(x) 6= 0} Then the morphism; θg : V1 → V2; (x, z) 7→ (x, g(x, z)) has an inverse provided by the transformation; θ−1 g : V2 → V1; (x, z) 7→ (x, g−1(x, z)) where g−1 is the inverse Mobius transformation to g. Moreover, we have that pr1 ◦ θg = Id(U \c(x)=0). The image of the irreducible open subset (C1 ∩ V1) ⊂ C1 is con- structible, hence, we can form its projective closure C2. It follows immediately, that θg defines an isomorphism between the open sub- sets (C1 ∩ V1) and (C2 ∩ V2), (†), therefore, defines a birational map θg : C1 ! C2. We check conditions (i) to (iv) for the curve C2. (i). We have that (0, 0) ∈ (V1 ∩ V2) and g(0, 0) = 0. By (†), (0, 0) defines a nonsingular point of the curve C2. In order to ver- ify that pr1 defines an etale morphism in an open neighborhood of (0, 0) ∈ C2, it is then sufficient to verify that the morphism is Zariski unramified at (0, 0), by Theorems (6.10,6.11) of [7]. Suppose not, then we can find ǫ ∈ (V0 ∩ A1), and distinct {ǫ1, ǫ2} ⊂ V0 such that {(ǫ, ǫ1), (ǫ, ǫ2)} ⊂ V(0,0) ∩ C2 ∩ V1 ∩ V2. It follows immediately that {(ǫ, g−1(ǫ, ǫ1)), (ǫ, g−1(ǫ, ǫ2))} ⊂ V(0,0) ∩ C1 ∩ V1 ∩ V2. Moreover, the pair that θ is an isomorphism in the etale topology with (0, 0) as labelled points, and (pr ◦ θ) = pr. Obvious examples of birational morphisms which do not satisfy this requirement are non-identity elements of the automorphism group (L(C1)/L(x)). It is not immediately obvious that, for any two curves C1 and C2, satisfying the conditions of Lemma 3.3, there does exist a birational morphism between C1 and C2, with the above extra requirement. However, this will be shown later in the paper. FLASH GEOMETRY OF ALGEBRAIC CURVES 21 {(ǫ, g−1(ǫ, ǫ1)), (ǫ, g−1(ǫ, ǫ2))} is distinct, by birationality. This contra- dicts the fact that pr1 defines an etale morphism in an open neighbor- hood of (0, 0) ∈ C1. The remaining conditions (ii), (iii), (iv) are straightforward to check, using the fact that; θg : (C1 ∩ V1, (0, 0)) → (C2 ∩ V2, (0, 0)) is an isomorphism, with the property that (pr1 ◦ θg) = pr1. (cid:3) The verticalised, polar geometry of such birational transformations, suggests that we should give special consideration to the effect on branches, centred at the point Q1, determined by the coordinate sys- tem (x, z); Lemma 3.5. Effect on hyperbolic branches at Q1 Let C1, C2 and θg satisfy the conditions of the previous lemma and suppose that γ is a hyperbolic branch of C1, centred at Q1, with tangent line x = α. Then, the corresponding branch θg(γ) of C2, determined by the birational map θg, is centred at Q1, if c(α) = 0 and a(α) 6= 0, and at (α, a(α) c(α) ), if c(α) 6= 0. Proof. As x = α is tangent to the branch γ, and not equal to l∞, by results of [6], given an infinitesimal ǫ ∈ V0, there exists z(ǫ) such that; (α + ǫ, z(ǫ)) ∈ (VQ1 ∩ C1 ∩ γ) By the explicit construction of specialisations, given in [3], we may r ]], (r ≥ 1), and i > 0, ǫi , for a unit u(x) ∈ L[[x assume that z(ǫ) = u(ǫ) i ∈ Q, (4) 1 If θg(γ) is centred at P , then, using results of [6]; 4The use of Puiseux series, here, is necessary, due to the existence of blue branches (Definition 3.11). We will find a more effective method of understand- ing the geometry of such branches, later in the paper. 22 TRISTRAM DE PIRO θg((α + ǫ, z(ǫ))) ∈ (VP ∩ C2 ∩ θg(γ)) We explicitly determine P by specialisation in the following cases; Case 1. c(α) = 0 and a(α) 6= 0 We have that; g(α + ǫ, z(ǫ)) = a(α+ǫ)z(ǫ)+b(α+ǫ) c(α+ǫ)z(ǫ)+d(α+ǫ) = a(α+ǫ)u(ǫ)+b(α+ǫ)ǫi c(α+ǫ)u(ǫ)+d(α+ǫ)ǫi hence, the projective coordinates of θg(ǫ, z(ǫ)), see Definition 3.1, are given by; X = (α + ǫ)(c(α + ǫ)u(ǫ) + d(α + ǫ)ǫi) Z = a(α + ǫ)u(ǫ) + b(α + ǫ)ǫi W = c(α + ǫ)u(ǫ) + d(α + ǫ)ǫi Using the fact that c(α) = 0 and a(α) 6= 0, the specialised coordi- nates are given by; X = 0, Z = 1, W = 0 hence, the result follows. Case 2. c(α) 6= 0. Writing z(ǫ) = 1 t(ǫ) , where t(ǫ) = ǫiu(ǫ), we have that; θg(α + ǫ, z(ǫ)) = (α + ǫ, a(α+ǫ)+b(α+ǫ)t(ǫ) c(α+ǫ)+d(α+ǫ)t(ǫ) ) which, clearly specialises to (α, a(α) c(α) ), using the fact that c(α) 6= 0. The lemma is then shown. (cid:3) Lemma 3.6. Effect on parabolic branches at Q1 Let C1, C2 and θg satisfy the conditions of the previous lemma and suppose that γ is a parabolic branch of C1, centred at Q1. Then the FLASH GEOMETRY OF ALGEBRAIC CURVES 23 corresponding branch θg(γ) is centred at; [1 : 0 : 0] if max{ord(a(x)), ord(b(x))} << min{ord(c(x)), ord(d(x))} [0 : 1 : 0] if max{ord(c(x)), ord(d(x))} << min{ord(a(x)), ord(b(x))} Proof. As γ is parabolic and centred at Q1, given an infinitesimal ǫ ∈ V0, there exists z(ǫ) such that ( 1 ǫ , z(ǫ)) ∈ (VQ1 ∩ C1 ∩ γ). By a straightforward calculation, we can assume that z(ǫ) is of the form u(ǫ) ǫi , for a unit u(x) ∈ L[[x r ]], r ≥ 1, and 1 < i ≤ N, i ∈ Q, where N depends only on the degree of C1. We have that; 1 g( 1 ǫ , z(ǫ)) = a( 1 c( 1 ǫ )u(ǫ)+b( 1 ǫ )u(ǫ)+d( 1 ǫ )ǫi ǫ )ǫi = q(ǫ) and θg( 1 ǫ , z(ǫ)) = [ 1 ǫ : q(ǫ) : 1] = [1 : ǫq(ǫ) : ǫ], (⋆) In the case that ord(q(ǫ)) > −1, the specialisation, determined from (⋆), is given by [1 : 0 : 0], and, in the case that ord(q(ǫ)) < −1, the specialisation is given by [0 : 1 : 0]. By inspection of q(ǫ), these conditions are clearly met, given the corresponding conditions of the lemma. (cid:3) It is convenient, in what follows, to consider the effect of the follow- ing transformation; Lemma 3.7. Let C1 satisfy the conditions of the previous lemma, and let α 6= 0 be chosen so that x = α does not correspond to the tangent line of any hyperbolic branch of C1, centred at Q1. Then the transfor- mation; θg : (x, z) 7→ (x, (x − α)z) has the following properties; (i). θg defines a birational map θg : C1 ! C2, with C2 satisfying the conditions of Lemma 3.3. 24 TRISTRAM DE PIRO (ii). If γ is a branch of C1, centred along (l∞ \ [1 : 0 : 0]), then θg(γ) is centred at Q1 ∈ C2. Moreover, (iii). If γ is a branch of C1, centred at [1 : 0 : 0], then, after finitely many repetitions of transformations, having the form θg, we obtain a birational map θ : C1 ! C2, such that θ(γ) is centred at Q1 ∈ C2. (iv). In particular, C2 intersects the line l∞ only at Q1. Proof. The property (i) follows immediately from Lemma 3.4 and the choice of α 6= 0. By explicit calculation, the projective formulation of θg is given by; θg : [X : Z : W ] 7→ [XW : XZ − αW Z : W 2] (⋆⋆) Any point P , lying on (l∞ \ (Q1 ∪ [1 : 0 : 0]), satisfies W = 0 and XZ 6= 0, hence, by (⋆⋆), is mapped to Q1. In particular, if γ is a branch of C1, centred along (l∞ \ (Q1 ∪ [1 : 0 : 0]), then θg(γ) is centred at Q1. By the choice of α in relation to tangent lines, and Lemma 3.5, the hyperbolic branches of C1, centred at Q1, are fixed. One finds the effect on the parabolic branches of C1, centred at Q1, by explicit calculation. Using the notation of Lemma 3.6, for z(ǫ) = u(ǫ) ǫi , with i > 1, we have that; θg( 1 ǫ , z(ǫ)) = [ǫi : u(ǫ) − αu(ǫ)ǫ : ǫi+1], (⋆ ⋆ ⋆) Clearly, the specialisation of this point is Q1 = [0 : 1 : 0]. Hence, the parabolic branches of C1, centred at Q1, are also fixed. (ii) is then shown. Again, we find the effect on a branch γ, centred at [1 : 0 : 0], by explicit calculation. Using the notation of Lemma 3.6, for z(ǫ) = u(ǫ) ǫi , with i < 1, by considering (⋆ ⋆ ⋆), we have that θg(γ) is centred along (l∞ \ [1 : 0 : 0]) iff 0 ≤ i < 1. Otherwise, we obtain a new z1(ǫ) = u1(ǫ) ǫi1 , with i1 = i + 1. Repeating the calculation, for new transformations of the form θg, satisfying the previous conditions of the lemma, we are guaranteed that this process eventually produces a birational map θ : C1 ! C2, with θ(γ) centred along (l∞ \ [1 : 0 : 0]). Using the previous properties (i) and (ii), we can assume that θ(γ) is centred at FLASH GEOMETRY OF ALGEBRAIC CURVES 25 Q1. Hence, (iii) is shown. The final statement follows straightforwardly from (i), (ii) and (iii). We now consider the following method of moving branches in finite position, not situated along the axis x = 0, to Q1; (cid:3) Lemma 3.8. Let C1, C2, g and θg satisfy the conditions of Lemma 3.4, with the additional property that, for a given α 6= 0, c(α) = d(α) = 0, a(α) 6= 0 and a(α)β + b(α) 6= 0, for any of the finitely many branches of C1 in finite position, centred along the axis x = α, with coordinates (α, β). Then, if γ is a branch of C1, centred in finite position along the axis x = α, θg(γ) is centred at Q1 ∈ C2, and, any hyperbolic branch of C1, centred at Q1 with tangent line x = α, is fixed. Proof. The case when γ is centred in finite position, with coordinates (α, β), such that a(α)β + b(α) 6= 0 follows by a direct calculation. The statement, concerning hyperbolic branches, follows immediately from Lemma 3.5. (cid:3) As a consequence of the previous arguments, we have the following; Lemma 3.9. There exists an irreducible plane projective curve C ′ ⊂ P 2, satisfying the conditions of Lemma 3.3, with the additional prop- erties; (i). (C ′ ∩ l∞) = Q1. (ii). All the singularities of C ′ lie along the axis x = 0. Proof. By Lemma 3.7(ii), we can assume there exists a curve C1, sat- isfying the conditions of Lemma 3.3, with the additional property that (C1 ∩ l∞) = {Q1}. In particular, all the singularities of (C1 \ {Q1}) are in finite position. Let A = {α1, . . . , αj, . . . , αr} enumerate the axes x = αj, containing the singularities in finite position, away from x = 0. Let B = {αr+1, . . . , αk, . . . , αs} enumerate the axes x = αk, corresponding to tangent lines of hyperbolic branches, centred at Q1, away from x = 0, not contained in A. Let Cj = {βj1, . . . , βji, . . . , βjm}, for 1 ≤ j ≤ r, 1 ≤ i ≤ m, where deg(C1) = m, enumerate the z- coordinates of the intersections in finite position of C1 with the axis 26 TRISTRAM DE PIRO It is an elementary exercise to show, using the fact that x = αj. 0 /∈ (A∪B), that there exist polynomials {a(x), b(x), c(x), d(x)} ∈ L[x], with the properties; (a). a(α1) . . . a(αr) . . . a(αs) 6= 0. (b). c(α1) = . . . = c(αr) = . . . = c(αs) = 0. (c). d(α1) = . . . = d(αr) = 0. (d). b(αj) 6= −a(αj)βji, 1 ≤ j ≤ r, 1 ≤ i ≤ m. (e). a(0)d(0) 6= 0 and b(0) = 0. (f). max{ord(c(x)), ord(d(x))} << min{ord(a(x)), ord(b(x))}. By (e) and Lemma 3.4, these polynomials determine a birational morphism θg : C1 ! C2, such that C2 also satisfies the conditions of Lemma 3.3. By (a),(b),(c),(d) and Lemma 3.8, all the singularities of C1, in finite position, away from the axis x = 0, are moved to Q1 ∈ C2. By (a),(b) and Lemma 3.5, the position of all the hyperbolic branches of C1 at Q1 is fixed. By (f) and Lemma 3.6, the position of all the parabolic branches of C1 at Q1 is fixed. Denoting C2 by C ′, it is then clear that conditions (i) and (ii) of the lemma are satisfied for C ′. (cid:3) We recall from [5] that the notion of a branch is invariant under bi- rational transformations. The following result allows us to move any finite number of branches away from the critical point Q1. Lemma 3.10. Let C1 satisfy the conditions of Lemma 3.3 and let {γ1, . . . , γt} enumerate a finite subset of its branches. Then there exists a birational map θg : C1 ! C2, with C2 also satisfying the conditions of Lemma 3.3, such that the corresponding branches {θg(γ1), . . . , θg(γt)} are either in finite position or centred at the point [1 : 0 : 0] along the axis l∞. In particular, none of the branches are centred at the point Q1. Proof. By birationality, we can assume that C1 satisfies the conditions of Lemma 3.9. Given the above enumeration, let; FLASH GEOMETRY OF ALGEBRAIC CURVES 27 (i). {γ1, . . . , γr} consist of those in finite position. (ii). {γr+1, . . . , γs} consist of the hyperbolic branches, centred at Q1. (iii). {γs+1, . . . , γt} consist of the parabolic branches, centred at Q1. Let {(α1, β1), . . . , (αj, βj), . . . , (αr, βr)} enumerate the coordinates of the branches given in (i). Let {δr+1, . . . , δk, . . . , δs} enumerate the axes x = δk, appearing as tangent lines to the branches given in (ii). It is an elementary exercise to show that there exist polynomials {a(x), b(x), c(x), d(x)} ∈ L[x], with the properties; (a). c(α1) . . . c(αj) . . . c(αr)c(δr+1) . . . c(δk) . . . c(δs) 6= 0. (b). a(αj)βj + b(αj) = βj, c(αj)βj + d(αj) = 1, (1 ≤ j ≤ r) (c). a(0)d(0) 6= 0 and b(0) = 0. (d).max{ord(a(x)), ord(b(x))} << min{ord(c(x)), ord(d(x))}. By (c) and Lemma 3.4, there exists a corresponding birational map θg : C1 ! C2, with C2 also satisfying the conditions of Lemma 3.3. By (b), the branches {θg(γ1), . . . , θg(γr)} are all centred at the same points as {γ1, . . . , γr} respectively. In particular, the branches from (i) remain in finite position. By (a) and Lemma 3.5, the branches {θg(γr+1), . . . , θg(γs)} are centred at the points in finite position; {(δr+1, a(δr+1) c(δr+1) ), . . . , (δs, a(δs) c(δs) )} respectively. In particular, the branches from (ii) are moved to finite position. By (d) and Lemma 3.6, the branches {θg(γs+1), . . . , θg(γt)} are centred at the point [1 : 0 : 0]. Hence, the lemma is shown. (cid:3) We now make the following definition; Definition 3.11. Fixing an enumeration of branches for some curve C ′, satisfying the conditions of Lemma 3.3, we define a branch γ to be blue, if there exists a birational map θ : C ! C1, with the curve C1 also satisfying the conditions of Lemma 3.3, such that; 28 TRISTRAM DE PIRO Either (i). θ(γ) is centred in finite position (a, b), with Iγ(C1, x = a) ≥ 2. Or (ii). θ(γ) is centred at [1 : 0 : 0], with Iγ(C1, l∞) ≥ 2. where Iγ is defined as in [6]. We define a branch γ to be silver otherwise. We observe the following; Remarks 3.12. (a). The choice of colours is partly motivated by con- sideration of the optical spectrum of visible light, as well as theoretical views of the author on chemical processes that occur in the production of light by stars. I hope to explain my views on the connections between geometry and a 4-fold model of the theory of light in a forthcoming pa- per. (b). On a geometrical level, the above construction is well-defined. For suppose that C1 and C2 are both curves, for which Definition 3.11 is made, relative to C ′. Then there exists a birational map θ1 : C1 ! C2, for which (pr1 ◦ θ1) = pr1, (⋆). It is a simple exercise to check, by (⋆), the definition of Iγ using infinitesimals, see [6], and birationality of the map θ1, that, if γ is a branch of C1, centred at (a, b), with θ(γ) centred at (a, b′), then Iγ(C1, x = a) = Iθ(γ)(C2, x = a). Similarly, if γ is a branch of C1, centred at [1 : 0 : 0], with θ(γ) centred at [1 : 0 : 0], then Iγ(C1, l∞) = Iθ(γ)(C2, l∞). (c). Using Lemmas 3.5, 3.6 and 3.9, any branch γ of C ′ may be centred in finite position or at [1 : 0 : 0], by an appropriate birational map θ. Hence, we can give an equivalent characterisation of a silver branch as in Definition 3.11, replacing the either/or clause by; Either (i). θ(γ) is centred in finite position (a, b), with Iγ(C1, x = a) = 1. Or FLASH GEOMETRY OF ALGEBRAIC CURVES 29 (ii). θ(γ) is centred at [1 : 0 : 0], with Iγ(C1, l∞) = 1. On an intuitive level, this means that a silver branch is characterised by the condition that it is non-singular and transverse to a vertical axis or to the line l∞, when centred away from Q1. Similarly, a blue branch is characterised by the condition that it is either singular or tangent to a vertical axis/the line l∞, when centred away from Q1. (d). There exist infinitely many silver branches and finitely many blue branches belonging to C ′. This follows immediately from the above characterisation and the fact that, there exists a unique non-singular branch, transverse to the axis x = a, for a generic point (a, b) belong- ing to C ′. Clearly, there exist finitely many silver branches and finitely many blue branches, belonging to C ′, centred along the axis x = 0. We now show the following; Lemma 3.13. Uniformity of branches Let C ′ satisfy the conditions of Lemma 3.3. Then the distribution of silver and blue branches along any axis is uniform. More precisely; (i). If x = α is a vertical axis, then, either, every branch centred in finite position along x = α is silver, or, every branch centred in finite position along x = α is blue. (ii). Either, every branch centred along l∞ \ {Q1} is silver, or, every branch centred along l∞ \ {Q1} is blue. (iii). If x = α is a vertical axis, then, either, every hyperbolic branch, centred at Q1, tangent to the line x = α, is silver, or, every hyperbolic branch, centred at Q1, tangent to the line x = α, is blue. (iv). Either, every parabolic branch, centred at Q1, is silver, or, ev- ery parabolic branch, centred at Q1, is blue. Proof. It is elementary that, for generic δ, the curve C ′ intersects the axis x = δ transversely at t distinct points in finite position, where t = deg(L(C ′)/L(x)). As, by Lemma 3.3(iv), the extension L(C ′)/L(x) is Galois, we have that G = t, where G = Gal(L(C ′)/L(x)), and G acts transitively on the fibre (C ′ ∩ (x = α)) \ {Q1}, (⋆). As this 30 TRISTRAM DE PIRO 1) ∈ γ1, (α + ǫ, β′ condition is definable, we can find an open subset U ⊂ A1, with the property that, for any δ ∈ U, (⋆) holds for the fibre x = δ. Sup- pose that there exists a silver branch γ1 and a blue branch γ2 situated on the axis x = α, centred at the coordinates (α, β1) and (α, β2) re- spectively. Choose an infinitesimal ǫ ∈ V0, such that α + ǫ ∈ U, and (α + ǫ, β′ 2) ∈ γ2. As (⋆) holds for the axis x = α + ǫ, we can find a birational morphism θh : C ′ ! C ′, representing the action of h ∈ G, with pr◦θh = pr, and θh(α+ǫ, β′ 2). It follows, by construction, that θh(γ1) = γ2. However, by previous remarks 3.12(b), we have that Iθh(γ1)(C ′, x = α) = Iγ1(C ′, x = α) = 1, contradicting the fact that θh(γ1) = γ2 is a blue branch. Hence, (i) is shown. In order to show (ii), one should employ a similar argument, using the fact that, for ǫ ∈ V0, with ǫ 6= 0, x = 1 ǫ lies in U. For (iii) and (iv), one can use Lemmas 3.5, Lemmas 3.6 and Definition 3.11, to reduce to the cases (i) and (ii) respectively. 1) = (α+ǫ, β′ (cid:3) Lemma 3.14. Separation of silver branches Let C ′ satisfy the conditions of Lemma 3.3, then any silver branch γ of C ′, not centred along the axis x = 0, (5), can be isolated away from Q1. That is, there exists a birational map θ : C ′ ! C1, with C1 also satisfying the conditions of Lemma 3.3, such that; Either (i). θ(γ) is centred at a point (α, β) in finite position, with α 6= 0, and is the unique branch of C1 to be centred at this point. Or (ii). θ(γ) is centred at a point of (l∞ \ {Q1}), and is the unique branch of C1 to be centred at this point. Moreover, given any finite set {γ1, . . . , γr} of silver branches, centred at Q1, with distinct tangent directions, not including x = 0, there exists a birational map θ : C ′ ! C1, with C1 also satifying the conditions of Lemma 3.3, such that {θ(γ1), . . . , θ(γr)} are all isolated away from Q1. Proof. By Lemmas 3.5 and 3.6, we can assume that the silver branch γ is centred either in finite position or at [1 : 0 : 0]. Case 1. Assume that γ is centred at (α, β), with α 6= 0. 5We follow the convention that a branch, centred along x = 0, includes hyperbolic branches, centred at Q1, with tangent line x = 0. FLASH GEOMETRY OF ALGEBRAIC CURVES 31 By Lemma 3.13, all the branches situated along the axis x = α are silver. In particular, all the branches centred at (α, β) are silver. Let {γ, γ1, . . . , γi, . . . , γr} be an enumeration of these branches. By results of [6], see also Remarks 2.6, we can find parametrisations of the form; {(α + x, β + η(x)), (α + x, β + η1(x)), . . . , (α + x, β + ηr(x))} for this enumeration, where {η(x), η1(x), . . . , ηi(x), . . . , ηr(x)} are dis- tinct power series, belonging to L[[x]], of the form; η(x) = P∞ j=1 cjxj, ηi(x) = P∞ j=1 cijxj, 1 ≤ i ≤ r Let g(x, z) be defined by; g(x, z) = (z−β) (x−α) − β α = (αz−βx) α(x−α) , (⋆) We have that α.α(x − α)(0,0) = −α3 6= 0 and (−βx)(0,0) = 0, hence, by Lemma 3.4, g(x, z) defines a birational map θg : C ′ ! C ′′, with C ′′ also satisfying the conditions of Lemma 3.3. Using Lemma 3.8, if γ′ is a branch of C ′ in finite position, along the axis x = α, not at (α, β), then θg(γ) is centred at the point Q1 ∈ C ′′. By Lemma 3.5, if γ′ is a hyper- bolic branch of C ′, centred at Q1, with tangent line x = α, then θ(γ′) is also centred at Q1 ∈ C ′′. It follows that the only branches, in finite position along the axis x = α of C ′′, are {θg(γ), θg(γ1), . . . , θg(γr)}. We now consider the positions of these branches; By results of [6] and the definition of g(x, z), the branches {θg(γ), θg(γ1), . . . , θg(γr)} are parametrised by; {(α + x, δ(x) − β α), (α + x, δ1(x) − β α ), . . . , (α + x, δr(x) − β α )} where η(x) = α + xδ(x) and ηi(x) = α + xδi(x), for 1 ≤ i ≤ r. It follows that θg(γ) is centred in finite position, at (α, β1), where β1 depends only on the first coefficient c1 of the power series η(x). In particularly, a branch θg(γi), for 1 ≤ i ≤ r, is also centred at (α, β1), if and only if, ci1 = c1. We then repeat the above construction for the new centre (α, β1), obtaining a new birational map θg1. The only branches remaining in finite position along the axis x = α after the composition (θg1 ◦ θg), being those for which ci1 = c1, and, the only branches whose centre coincides with that of (θg1 ◦ θg)(γ), being those for which ci1 = c1 and ci2 = c2. As the power series η(x) is dis- tinct from {η1(x), . . . , ηr(x)}, clearly the construction terminates after 32 TRISTRAM DE PIRO a finite number of steps, and, we obtain a curve C1, satisfying the con- ditions of Lemma 3.3, together with a birational map θ : C ! C1, such that θ(γ) is the only branch, centred in finite position, along the axis x = α. Hence, (i) is shown for this case. Case 2. Assume that γ is centred at [1 : 0 : 0]. Choosing α 6= 0, we consider the rotation ν about Q1, defined by; ν : (x, z) 7→ ( 1 x+α − 1 α , z) = ( −x α(x+α) , z) By construction, we have that ν(0, 0) = (0, 0) and ν fixes the axis x = 0. It is clear that ν determines a birational morphism θν : C ′ ! C ′′, which is etale in an open neighborhood of (0, 0), (6). It is a straight- forward exercise, using infinitesimals, and the fact that ν permutes the vertical axes, to check that θν interchanges; (a). The silver branches of C ′, centred along (l∞ \ ([1 : 0 : 0] ∪ Q1)), and the silver parabolic branches of C ′, centred at Q1, with the silver hyperbolic branches of C ′′, centred at Q1, with tangent line x = − 1 α. (b). The silver branches of C ′, centred at [1 : 0 : 0], with the silver α . (⋆⋆) branches of C ′′, centred in finite position along the axis x = − 1 We now apply Case 1, to obtain a birational morphism θ1 : C ′′ ! C ′′′, in the stronger sense of Lemma 3.4, such that the silver branch θν(γ) is isolated at (− 1 α , 0), that is (θ1 ◦ θν)(γ) is the unique branch of C ′′′ to be centred at (− 1 α , 0), (⋆ ⋆ ⋆). Then, we apply the inverse rotation ν−1 about Q1, defined by; ν−1 : (x, z) 7→ ( 1 x+ 1 α − α, z) = ( −α2x αx+1, z) which determines a birational map θν−1 : C ′′′ ! C ′′′′. It is sim- ple to check that the composition (θν−1 ◦ θ1 ◦ θν) : C ′ ! C ′′′′ is a birational map, in the stronger sense of Lemma 3.4. Using (⋆⋆), we have that (θν−1 ◦ θ1 ◦ θν)(γ) is centred at [1 : 0 : 0]. If ( −1 α + x, δ(x)), with δ(x) ∈ L[[x]] and ord(δ(x)) ≥ 1, is a parametrisation of the silver branch θ1 ◦ θν)(γ), then the corresponding parametrisation of 6This is not a morphism in the sense of Lemma 3.4 and footnote 3, as, clearly, (pr ◦ θ) 6= pr FLASH GEOMETRY OF ALGEBRAIC CURVES 33 (θν−1 ◦ θ1 ◦ θν)(γ) is given by ( 1 x − α, δ(x)). Moreover, using (⋆⋆)(b) and (⋆ ⋆ ⋆), if ( 1 x − α, δ1(x)) is a parametrisation of any other sil- ver branch of C ′′′′, centred at [1 : 0 : 0], then we must have that ord(δ1(x)) = 0, (⋆ ⋆ ⋆⋆). We, finally, apply a transformation of the form θg : C ′′′′ ! C ′′′′′, considered in Lemma 3.7. By (⋆⋆⋆⋆), it is elementary to check, using infinitesimals, that the branch (θg ◦ θν−1 ◦ θ1 ◦ θν)(γ) is fixed at [1 : 0 : 0] and the other branches of C ′′′′, centred at [1 : 0 : 0], are moved to positions along the axis (l∞ \ (Q1 ∪ [1 : 0 : 0])), see also the proof of Lemma 3.7. By the properties of θg, considered in Lemma 3.7, (θg ◦ θν−1 ◦ θ1 ◦ θν)(γ) is the unique branch of C ′′′′′ to be centred at [1 : 0 : 0]. Letting θ = (θg ◦ θν−1 ◦ θ1 ◦ θν) and C1 = C ′′′′′, we have that (ii) is shown for this case. For the final part of the lemma, assume, first, that all of the branches {γ1, . . . , γr} are hyperbolic. By Lemma 3.10, we can assume that they are all in finite position. Let {α1, . . . , αr} enumerate the distinct x- coordinates of these branches. Using Case 1 above, we can isolate the branch γ1, away from Q1, by applying finitely many transforma- tions of the form, given in (⋆), with α = α1. Clearly, the branches {γ2, . . . , γr} remain in finite position after these transformations. Sup- pose, inductively, that we have isolated the branches {γ1, . . . , γi}, with 1 ≤ i < r, away from Q1. We can then apply finitely many trans- formations of the form, given in (⋆), with α = αi+1, in order to isolate the branch γi+1, away from Q1. As the restriction of these transformations to the axes {x = α1, . . . , x = αi} is of the form a(αj)z + b(αj), with a(αj) = (αi+1 − αj)−1 6= 0, for 1 ≤ j ≤ i, the branches {γ1, . . . , γi} also remain isolated away from Q1. Hence, we have isolated the branches {γ1, . . . , γi, γi+1}. By induction, we can, therefore, isolate all the branches {γ1, . . . , γr}, away from Q1, as re- quired, (⋆ ⋆ ⋆ ⋆ ⋆). Assume, now, that {γ1, . . . , γr} are hyperbolic branches and γr+1 is parabolic. By Lemma 3.10, we can assume that {γ1, . . . , γr} are all centred in finite position, and γr+1 is centred at [1 : 0 : 0]. By (⋆ ⋆ ⋆ ⋆ ⋆), we can isolate the branches {γ1, . . . , γr}, away from Q1, using finitely many transformations of the type considered in (⋆). By a simple projective calculation, such transformations fix the point [1 : 0 : 0], in particular the branch γr+1 remains centred at [1 : 0 : 0]. We now follow the method of Case 2, in order to isolate the branch γr+1, away from Q1, while, retaining the property (⋆ ⋆ ⋆ ⋆ ⋆⋆), that the remaining branches {γ1, . . . , γr} are isolated, away from Q1, in finite position. This amounts to showing that (⋆ ⋆ ⋆ ⋆ ⋆⋆) is preserved 34 TRISTRAM DE PIRO in the following steps from Case 2; (1). For the rotation ν, with α 6= αi, for 1 ≤ i ≤ r. (2). For the construction of the birational map θ1, with the addi- tional use of the previous inductive proof. (3). For the inverse rotation ν−1, with α 6= αi, for 1 ≤ i ≤ r. (4). For transformations of the form θg, considered in Lemma 3.7, with α 6= αi, for 1 ≤ i ≤ r. Hence, the final part of the lemma is shown. (cid:3) Remarks 3.15. This seems to be a reasonable geometrical property of a theory of light, motivated by theoretical considerations of the author on the chemical production of energy by fission in stars, see also Remarks 3.12(a). The increased level of radioactivity along the axis x = 0 is, in fact, an observable phenomenon of pulsars, a type of neutron star, see [11]. The following result is straightforward, see the previous footnote 3; Lemma 3.16. Let C1 and C2 satisfy the conditions of Lemma 3.3, then there exists a birational map θ : C1 ! C2, which is an isomorphism in the etale topology with (0, 0) as labelled points, and (pr ◦ θ) = θ. Proof. By the definition of the etale topology, there exists an irre- ducible projective algebraic curve (C3, ∗), together with an open subset U3 ⊂ C3, containing the marked point ∗, and open subsets U1 ⊂ C1, U2 ⊂ C2, containing (0, 0), with etale morphisms; i13 : (C3, U3, ∗) → (C1, U1, (0, 0)) i23 : (C3, U3, ∗) → (C2, U2, (0, 0)) such that i∗ 23(δi(x)), for 1 ≤ i ≤ m, (∗), and (pr ◦ i13) = (pr ◦i23), (∗∗). Using (∗∗), the morphisms i13 and i23 induce inclusions; 13(δi(x)) = i∗ i∗ 13 : L(C1) → L(C3) FLASH GEOMETRY OF ALGEBRAIC CURVES 35 i∗ 23 : L(C2) → L(C3) with (i∗ 13 ◦ pr∗) = (i∗ 23 ◦ pr∗) : L(x) → L(C3), (∗ ∗ ∗). By (∗), (∗ ∗ ∗), and Lemma 3.3(iii), we can define an isomorphism φ∗ : L(C2) → L(C1), such that (φ∗ ◦ pr∗) = pr∗ : L(x) → L(C1). Hence, there exists a birational map φ : C1 ! C2, with (pr ◦ φ) = pr, (∗ ∗ ∗∗). By Lemma 3.3(i), there exists a unique silver branch γ1 of C1, centred at the origin (0, 0). Using (∗ ∗ ∗∗), it is straightforward that θ(γ1) is a silver branch of C2, centred along the axis x = 0. Using Lemma 3.3(iv), and the method of Lemma 3.13, we can find a birational map θh : C2 ! C2, with (pr ◦ θh) = pr, (∗ ∗ ∗ ∗ ∗) representing h ∈ Gal(L(C2)/L(x)), such that (θh◦φ)(γ1) = γ2, where γ2 is the unique silver branch of C2, centred at (0, 0). It is a simple calculation, using infinitesimals, Theorems (6.10,6.11) of [4] and the fact that the branches {γ1, γ2} are nonsingular, to show that (θh ◦ φ) is etale in an open neighborhood of C1, containing (0, 0). Similar considerations apply to the morphism (θh ◦ φ)−1, hence, (θh ◦ θ) induces an isomorphism in the etale topology, with (0, 0) as labelled points. Letting θ = (θh ◦ φ), the fact that (pr ◦ θ) = pr follows immediately from (∗ ∗ ∗∗) and (∗ ∗ ∗ ∗ ∗). Hence, the lemma is shown. (cid:3) Remarks 3.17. It is a straightforward exercise, to show that a mor- phism satisfying the conditions of Lemma 3.17 is unique. Hence, we can unambigiously identify the branches of any two curves, satisfying the conditions of Lemma 3.3. The previous Lemmas 3.5 to 3.15, sup- port the intuition that the orbits of each branch resemble silver or blue flashes of light, emanating from and towards a central point Q1. Given the high gravitational field of a neutron star, this is, possibly, a good description of the behaviour of light in certain situations. Remarks 3.18. Rose windows are another geometrically interesting source of examples, that exhibit both radial and rotational symmetry, as well as intricate light effects. Some of the best can be found in the Rayonnant designs of French medieval cathedrals. An excellent account of the historical development of rose windows is given in [2]. For the reader, interested in making rose windows, one should look at [16]. A number of pictures of rose windows should be available soon on the website, http://www.magneticstrix.net, where you can find some good examples of neutron star style windows in the Rayonnant Rose of Saint Denis, and the North and South Roses of Notres Dame cathedral. We now make the following refinement of the branch terminology that we have used; 36 TRISTRAM DE PIRO Definition 3.19. Let conditions be as in Definition 3.11, then we de- fine a branch γ to be violet, if; γ is a non-singular blue branch. and we define a branch γ to be green, if; γ is a silver branch, centred along the axis x = 0, (see also footnote 5), or γ is a blue branch, but not a violet branch. Remarks 3.20. In order to see that this is a good definition, We then have the following geometric interpretation of Lemma 2.5. Lemma 3.21. Let hypotheses be as in Lemma 2.5 and notation as in Definition 3.1, then we can find an irreducible closed algebraic sur- face S ⊂ P 3, such that the restriction of pr1 to (S \ Q1) defines a quasi-finite morphism, pr1 : S → P 2 1 , (♥), with the following further additional properties; (i). There exist open subsets U ⊂ A2 1 ⊂ P 2 1 and V ⊂ S∩A3 such that; pr1 : (V, (00)lif t) → (U, (00)) is etale. (ii). There exist distinct irreducible closed algebraic curves {C1, . . . , Cm} contained in S, such that; a. pr1 : (Cj \ {Q1}) → C is a finite cover, for 1 ≤ j ≤ m, (7) b. (pr−1 1 (C) ∩ S) = ∪1≤j≤mCj. c. The coordinate ring R(V ) embeds in L[[x]][y] ∩ L(x, y)alg and the defining equations of Cj ∩ V are given by; y − ηj(x) = 0, for 1 ≤ j ≤ m. (iii). There exists an irreducible algebraic curve, C ′ ⊂ P 2 2 , satisfy- ing the conditions of Lemma 3.3, with respect to the coordinate system 7We use the term cover, in the sense that pr1 defines a dominant morphism from (Cj \ {Q1}) onto an open subset of C. The projection naturally extends to define a surjective map from the branches of Cj, including those centred at Q1, to the branches of C. The reader should look at the paper [6] FLASH GEOMETRY OF ALGEBRAIC CURVES 37 (x, z), such that; a. pr2 : Cj → C ′ is a finite cover, for 1 ≤ j ≤ m. b. pr2 : Cj ! C ′ is a birational map, with inverse ηj : C ′ ! Cj and an open subset V ′ ⊂ C ′, such that; c. pr1 : (V ′, (00)) → (A1, 0) is a Galois cover, with the extension L(V ′)/L(x) equal to the Galois closure of the extension L(C)/L(x). Proof. From the previous construction of Lemma 2.5, the power series {δ1(x), . . . , δm(x)} belong to L[[x]] ∩ L(x)alg. By Theorem 1.3 and re- marks at the beginning of Section 3 of [5], we have that L[[x]] ∩ L(x)alg is isomorphic to the local ring for the etale topology Oet (A1,0). By the construction in Lemma 3.3, we can find an irreducible plane projective curve C ′ ⊂ P 2, such that; (i). There exists an open subset U1 ⊂ C ′, such that (0, 0) ∈ U1 and; pr : (U1, (0, 0)) → (A1, 0) is an etale morphism, (†) (ii). {δ1(x), . . . , δm(x)} all belong to R(U1), for a given embedding of R(U1) in L[[x]] ∩ L(x)alg, and δ1(0, 0) = . . . = δm(0, 0) = 0, (†′) (iii). The function field L(U1) is generated over L(x) by the power series {δ1(x), . . . , δm(x)}, (8). (†′′). By Fact 1.5 of [5], we can find an open subset U ′ ⊂ A1, containing (0), and a monic polynomial F (z) ∈ R(U ′)[z] such that the etale mor- phism pr can be presented, for an open neighborhood V ′ ⊂ U1 ⊂ C ′ of (0)lif t = (0, 0) in the form; Spec(( R(U ′)[z] F (z) )d) → Spec(R(U ′)) (∗) with F ′(z) invertible in ( R(U ′)[z] F (z) )d. As (V ′, (0, 0)) is a localisation of (U1, (0, 0)), the coordinate ring R(U1) ⊂ R(V ′) ⊂ L[[x]] ∩ L(x)alg and 8In non-zero characteristic, one should take L(U1) to be the seperable closure of L(x, δ1(x), . . . , δm(x)) 38 TRISTRAM DE PIRO the conditions (†), (†′) and (†′′) are preserved, replacing U1 by V ′. Let G(x, z) define the projective curve C ′, restricted to A2 2, see Def- inition 3.1, which we also denote by C ′. The restriction of C ′ to V ′ = (pr−1 1 (U ′) ∩ (G = 0) ∩ (d 6= 0)) corresponds to the cover de- fined in (∗). We also consider G(x, z) as defining an irreducible alge- braic surface C ′ × A1 in A3, using the coordinates (x, y, z). We now projectivize the affine polynomial G(x, z), by making the substitutions {x = X W }, and obtain an irreducible homogeneous polynomial R(X, Z, W ). This defines an irreducible algebraic surface S ⊂ P 3. Ob- serve that, as C ′ passes through (0, 0), the surface S contains the point It is clear from the construction that pr2(S \ {Q2}) = C ′ and Q2. pr1(S \ Q1) ⊂ P 2 1 . It is straightforward to see that the fibres of pr1 restricted to (S \ Q1) are all finite, showing (♥) in the statement of the lemma. W , z = Z We now set U = U ′ × A1 ⊂ A2 1 and V = V ′ × A1 ⊂ S ∩ A3. By the fact that R(V ′) embeds in L[[x]] ∩ L(x)alg, we have that R(V ) embeds in L[[x]][y] ∩ L(x, y)alg, (∗∗). The etale morphism pr1 : (V ′, (00)) → (U ′, (0)) may be presented in the form; R(U ′) → R(U ′)[z,w] <F (z),wδ(z)−1> with ∂(F (z),wδ(z)−1) ∂(z,w) p 6= 0, for p ∈ V ′ The morphism pr1 : (V, (000) → (U, (00)) may then be presented in the same form, replacing the ring R(U ′) by the ring R(U) = R(U ′)[y]. By the definition of an etale morphism in Definition 1.1 of [5], we ob- tain immediately that this morphism is etale, hence (i) of the Lemma is shown. By (†′) and (∗∗), we now have that the algebraic power series {y−δ1(x), . . . , y−δm(x)} define irreducible algebraic curves {D1, . . . , Dm} on V , passing through (00)lif t = (000). By the construction of V , we have that the algebraic power series {y −a1 −δ1(x), . . . , y −am −δm(x)} define distinct irreducible algebraic curves {D′ j, . . . , D′ m} on V ⊂ S, passing through (0, aj, 0), for 1 ≤ j ≤ m. We let Cj = D′ j, for 1 ≤ j ≤ m. By elementary facts on Zariski closure, each Cj ⊂ S is irreducible and (ii)(c) of the lemma holds, by the definition of ηj(x) = aj + δj(x) in Lemma 2.5, and the fact that D′ j = Cj ∩ V . Again, by Lemma 2.5, (pr−1 1 (C)∩V ) = ∪1≤j≤m(Cj ∩V ), (∗∗∗). It follows immedi- ately, from elementary facts about quasi-finite morphisms, that (ii)(a) of the lemma holds and ∪1≤j≤mCj ⊆ (pr−1 1 (C) ∩ S). Now suppose that 1, . . . , D′ FLASH GEOMETRY OF ALGEBRAIC CURVES 39 C0 ⊂ S is an irreducible component of (pr−1 1 (C) ∩ S). By elementary dimension theory, C0 is an algebraic curve and must define a finite cover of C. We may suppose that (C0 ∩ A3) 6= ∅, otherwise C0 is contained in the plane (W = 0) and, therefore, pr1(C0) = C is contained in the line l1,∞, contradicting the presentation of C in Lemma 1.1. Suppose that (C0 ∩ V ) = ∅, then C0 ∩ A3 must be a line of the form {p} × A1, where p = (p1, p2) is one of the finitely many exceptional points on the affine curve C ′ ⊂ A2 2, obtained by removing V ′. It follows that pr1(C0) = C is contained in the closure of the line x = p1, in the coordinate system (x, y) of A2 1, again contradicting the presentation of C in Lemma 1.1. Hence, we may assume that (C0 ∩ V ) 6= ∅. It follows that (C0 ∩ V ) defines an irreducible component curve of (pr−1(C) ∩ V ), hence, by (∗ ∗ ∗), coincides with (Cj ∩ V ), for some 1 ≤ j ≤ m. It follows, by elementary facts on Zariski closure, that C0 = Cj. Therefore, (ii)(b) of the lemma holds and (ii) of the lemma is shown. By the presentation given in Lemma 1.1, the point Q2 does not lie on C. By (ii)(a) of the lemma, and the fact that Q2 is fixed by the projection pr1, Q2 cannot lie on Cj for any 1 ≤ j ≤ m either. It follows that pr2 is defined on all of Cj, for 1 ≤ j ≤ m. As we observed earlier, pr2(S \ {Q2}) = C ′, hence, as each curve Cj belongs to (S \ {Q2}), and, using the argument above to exclude the exceptional case that pr2(Cj) is a point, we obtain (iii)(a) of the lemma. Now, observe that the algebraic power series {η1(x), . . . , ηm(x)} define rational functions in L(C ′), and belong to the coordinate ring R(V ′), for the open subset V ′ ⊂ C ′. We may, therefore, define a morphism, θj : V ′ → V θj(x, z) = (x, ηj(x, z), z), for (x, z) ∈ V ′ The defining equation of the image of this morphism in R(V ) is given by y − ηj(x, z) = 0, hence, by (ii)(c) of the lemma, corresponds to (Cj ∩ V ). It follows that θj : V ′ → Cj must define a morphism as well. By the explicit definition of pr2 in the affine coordinates (x, y, z), we have that pr2 ◦ θj = IdV ′. It follows immediately that pr2 defines an isomorphism between the open subsets (Cj ∩ V ) ⊂ Cj and V ′ ⊂ C ′. Hence, (iii)(b) of the lemma is shown. The property (iii)(c) follows immediately from the construction of V ′, by observing that L(V ′) = L(x, δ1(x), . . . , δm(x)) = L(x, η1(x), . . . , ηm(x)) is a splitting field for the polynomial F (x, y) over L(x), hence, the cover defined in (iii)(c) is Galois and the extension L(V ′)/L(x) is equal to the Galois closure of 40 TRISTRAM DE PIRO L(C)/L(x), see also footnote 1. Therefore, (iii) of the lemma, is shown. This completes the proof. (cid:3) Definition 3.22. We define the set of curves {C1, . . . , Cj, . . . , Cm}, found in the previous Lemma 3.21, as flashes of the original algebraic curve C, (9) We then have; Lemma 3.23. There exists a finite group G and an algebraically defin- able action of G on an open subvariety W ′ ⊂ V ′, with pr1 : W ′ → A1 a quasi-finite morphism, such that; (i). For generic x ∈ A1, G acts sharply transitively on the fibre 1 (x). pr−1 (ii). G induces an algebraically definable, transitive action on the set of flashes {C1, . . . , Cm}. Proof. We let G = Gal(L(C ′)/L(x)). By standard considerations, a given g ∈ G induces a birational morphism θg : C ′ ! C ′, with the property that pr1 ◦ θg = pr1 : C ′ P 1, as rational maps. More specif- ically, we can find polynomials {pg(x, z), qg(x, z)} with the property that θg is defined, in affine coordinates, by; θg : (x, z) 7→ (x, pg(x,z) qg(x,z) ) = (x, g (cid:5) z) (∗) Letting G(x, z) define the curve C ′, restricted to the coordinate sys- tem (x, z), set Rg = ((G = 0) ∩ (qg = 0)), and Sg = pr1(Rg). Let C ′′ define the curve obtained by removing the finitely many coordinate lines {x = s : s ∈ Sg}, from the curve C ′, realised by G(x, z). It is straightforward to check that C ′′ is invariant under the action of θg. By repeating the construction, we can choose C ′′ to be invariant under 9More accurately, a flash Cj should be considered as representing the class of curves, corresponding to each power series ηj(x). That is, given two curves {C1, C2}, satisfying the conditions of Lemma 3.3, and a birational map α12 : C1 ! C2, as in Lemma 3.16, if {C1 m} denote the sets of curves obtained in Lemma 3.21, then we obtain naturally defined bira- tional maps θj = (ηj ◦ α12 ◦ (pr2)−1) : C1 j . We can then define a flash as the j equivalence class of a given curve Cj under these birational maps; by Lemma 3.16, this accounts for all such curves. m} and {C2 1 , . . . , C1 ! C2 j , . . . , C1 1 , . . . , C2 j , . . . , C2 FLASH GEOMETRY OF ALGEBRAIC CURVES 41 the action of θg, for any g ∈ G, (∗∗). If g ∈ G, then C ′′ is invariant under the action of both θg and θg−1 = (θg)−1, it follows that θg maps C ′′ isomorphically onto itself, (∗ ∗ ∗). By a similar argument, removing also coordinate lines of the form {x = t : t ∈ Tg}, from V ′, where; Tg = {t ∈ A1 : Card(V ′(t)) < Card(V ′(x)), f or generic x ∈ A1} we can find an open subvariety W ′ ⊂ V ′, with the same properties (∗∗) and (∗ ∗ ∗). We consider the action of G on a generic fibre W ′(x). Assuming degzG(x, z) = l, this consists of l distinct points, in finite position. It is also straightforward to check that deg(L(W ′)/L(x)) = l. As the extension L(W ′)/L(x) is Galois, Card(G) = l. If the action fails to be transitive, then, by a simple counting argument, we can find a non-trivial g0 ∈ G and x0 ∈ W ′(x), with g0 (cid:5) x0 = θg0(x0) = x0. As the morphisms θg0 and θId = Id are etale, it follows that θg0 = Id, see Proposition 3.15 of [17], which is a contradiction. Sharp transitivity is obtained by a similar argument. This proves (i). By Lemma 2.5, we have that the power series {η1(x), . . . , ηm(x)} constitute a complete set of distinct roots for the polynomial F (x, y), considered as belonging to L(x)[y]. By (†) of Lemma 3.3, and the later remark there, the power series {η1(x), . . . , ηm(x)} also belong to the coordinate ring R(V ′) ⊂ L(C ′). In particular, we have that the function field L(C ′) is a splitting field for the polynomial F (x, y), over the subfield L(x), and G acts transitively on the power series {η1(x), . . . , ηm(x)} ⊂ R(V ′) ⊂ R(W ′), (♯). We can naturally extend the action of G to the coordinate ring R(W ) = R(W ′)[y]. By (♯), this defines a transitive action on the irreducible curves {D′ m}, which extends to an action on the flashes {C1, . . . , Cm}, using ele- mentary facts on Zariski closure and the definition of Cj = D′ j, for 1 ≤ j ≤ m. (This action clearly also respects the equivalence relation, defined in footnote 9.) Hence, (ii) is shown. Algebraic definability of the group action follows easily, by enumerating the parameters involved in the action defined by (∗), and restricting to the variety W ′. (cid:3) 1, . . . , D′ We now give a more refined version of Lemma 3.3, by analysing the intersections and singularities of the flashes {Cj : 1 ≤ j ≤ m}. We first observe the following, we recall the notion of valγ, for a birational morphism, from [7]. 42 TRISTRAM DE PIRO Lemma 3.24. Let C ′ satisfy the conditions of Lemmas 3.3 and 3.21, then, if a branch γ of C ′ is centred in finite position along the axis x = 0, the values of the birational morphisms valγ(ηj) are distinct, belonging to the set {a1, . . . , am}. Proof. Suppose that γ is a branch, centred along x = 0. If bj de- notes the value valγ(ηj), and γ is centred at (0, c), then (0, bj, c) be- longs to the curve Cj. By (ii)(a) of Lemma 3.21, the projection (0, bj) belongs to C, hence, bj belongs to the set {a1, . . . , am}. Sup- pose that valγ(ηj0) = valγ(ηj1) = aj, for j0 6= j1. The birational morphisms {ηj0, ηj1} are distinct, by Lemma 3.21, hence, there ex- ists an open subset U ⊂ C ′, for which (ηj0 − ηj1) is non-zero. Let (x, z) ∈ (γ ∩ U), then the values ηj0(x, z) and ηj1(x, z) are distinct and the points {(x, ηj0(x, z), z), (x, ηj1(x, z), z)} belong to the curves (Cj0 ∩ V(0,aj ,c)) and (Cj1 ∩ V(0,aj ,c)) respectively. Applying Lemma 3.21, the projected points {(x, ηj0(x, z)), (x, ηj0(x, z))} belong to C ∩ V(0,aj ). As x ∈ V0, it follows that I(0,aj )(C, x = 0) ≥ 2. This clearly contradicts the presentation of C, given in Lemma 1.1(iii). (cid:3) We recall that C is a nodal curve, satisfying the assumptions of Lemma 1.1. Definition 3.25. Let ℘ = {ρ1, . . . , ρi, . . . , ρr} enumerate the branches of C with the following properties, relative to the coordinate system (x, y); Either (i). The tangent line of ρi to C is vertical. Or (ii). ρi is centred at a node, belonging to C. By Lemma 1.1, all such branches are in finite position, and the con- ditions are mutually exclusive. In the terminology of Lemma 3.21, we observe that the projection pr1 defines a quasi finite map from (Cj \ {Q1}) to C, which extends to give a well-defined map from the branches γ of Cj to C, and that the branches of Cj can be naturally identified with the branches of C ′, using the birational map ηj. This motivates the following definition; Definition 3.26. For the branches in Definition 3.25, we define, Γj = {γ1,j, . . . , γt,j} to be the lifting of these branches to the flash Cj, in Lemma 3.21, and also of the projective curve C ′, given in Lemma 3.24. We define Γ = {γ1, . . . , γt} = S1≤j≤m Γj. FLASH GEOMETRY OF ALGEBRAIC CURVES 43 We have the following birational invariance; Lemma 3.27. Let α12 : C 1 ! C 2 satisfy the conditions of Lemma 3.3 and Lemma 3.16, with equivalent flashes θj : C 1 j , defined j by Lemma 3.21, see also footnote 9. Then the branches {γ, ηj(γ)} of {C 1, C 1 j } lift the same branch ρ of C, as {α12(γ), θj(ηj(γ))}. In par- ticular, the sets Γj and Γ are birationally invariant, and valγ(ηj) = valα12(γ)(ηj). ! C 2 Proof. We denote the representatives of the power series ηj by the ra- tional functions f ∈ L(C 1) and g ∈ L(C 2). By the definition of α12, we have that (g ◦ α12) = f , and (pr1 ◦ α12) = pr1, as rational functions on C 1. It is sufficient to show that (pr1 ◦ ηj) = (pr1 ◦ ηj ◦ α12) : C 1 → C, as rational maps. For a generic point (x, z) ∈ C 1, we have that; (pr1 ◦ ηj)(x, z) = pr1(x, f (x, z), z) = (x, f (x, z)) (pr1 ◦ ηj ◦ α12)(x, z) = (pr1 ◦ ηj)(x, (pr ◦ α12)(x, z)) = pr1(x, g(x, (pr ◦ α12)(x, z)), z) = pr1(x, (g ◦ α12)(x, z), z) = (x, f (x, z)) (10) Hence, the result is shown. We make the following definition; (cid:3) Definition 3.28. Let C ′ satisfy the conditions of Lemmas 3.3 and 3.21, then, we define a branch to be geometric, if it is either finite, or, hyperbolic and centred at Q1. We define a branch to be perpendicular, if it is infinite, and also parabolic, if centred at Q1. See Definition 2.1 for relevant terminology. Remarks 3.29. A branch is geometric if and only if it is not perpen- dicular. This follows immediately from Definition 2.1. The notions of geometric and perpendicular are also birationally invariant, in the sense of Lemma 3.16. Lemma 3.30. Let C ′ satisfy the conditions of Lemmas 3.3 and 3.21, then, for a geometric branch γ of C ′, we have that; 10We have temporarily used the notation pr for the projection onto the z- coordinate. 44 TRISTRAM DE PIRO valγ(ηj0) = valγ(ηj1), for j0 6= j1, iff γ ∈ (Γj0 ∩ Γj1) There does not exist a geometric branch γ of C ′, on which {ηj0, ηj1, ηj2} take the same valγ, for a distinct triple (j0, j1, j2) Proof. By Lemma 3.10 and Remarks 3.29, we can assume that γ is in finite position. If valγ(ηj) = ∞, then the point [0 : 1 : 0 : 0] belongs to the flash Cj, hence, the point [0 : 1 : 0] belongs to the curve C. This contradicts the presentation of C in Lemma 1.1(iii). Suppose γ is centred at (x0, z0), with valγ(ηj0) = valγ(ηj1) = y0 < ∞. Then the point (x0, y0, z0) belongs to the intersection Cj0 ∩ Cj1, and projects to a point (x0, y0) on C. We can choose (x, z) ∈ V(x0,z0), such that ηj0(x, z) 6= ηj1(x, z), and, therefore, find distinct points {y, y′}, with (x, y, z) ∈ (V(x0,y0,z0) ∩ Cj0) and (x, y′, z) ∈ (V(x0,y0,z0) ∩ Cj1). The dis- tinct projected points (x, y) and (x, y′) belong to (V(x0,y0) ∩ C). It follows that I(x0,y0)(C, x = x0) ≥ 2. By the presentation of C in Lemma 1.1, this can only occur if the branch(es), centred at (x0, y0), It follows, from Definition 3.26, that the branch γ ∈ belong to ℘. (Γj0 ∩ Γj1). Conversely, suppose that γ ∈ (Γj0 ∩ Γj1), so the projected branches pr1(ηj0(γ)) and pr1(ηj1(γ)), (∗), belongs to ℘. Supposing that valγ(ηj0) 6= valγ(ηj1), then the projections, (∗), are centred at distinct points (x0, y0), (x0, y1) of C. This contradicts Lemma 1.1(iv). Finally, suppose that valγ(ηj0) = valγ(ηj1) = valγ(ηj2) = y0 < ∞. By a sim- ilar argument to the above, we would find a point (x0, y0) ∈ C, with I(x0,y0)(C, x = x0) ≥ 3. This contradicts Lemma 1.1(ii). (cid:3) As a consequence of the preceding lemma, we obtain; Lemma 3.31. Let C ′ satisfy the conditions of Lemma 3.30, then we have that; Γ = F1≤j0<j1≤m(Γj0 ∩ Γj1) is a disjoint union of pairwise intersections. Proof. Suppose that γ ∈ Γ, therefore, by Definition 3.26, belongs to Γj0, for some 1 ≤ j0 ≤ m. We claim that γ is geometric. For suppose that γ is perpendicular, then, by Lemma 3.10, it can be centred at [1 : 0 : 0]. One can verify, that the projected branch pr1(ηj0(γ)) is centred on FLASH GEOMETRY OF ALGEBRAIC CURVES 45 the line l∞, (11). However, (℘ ∩ l∞) = ∅, by Lemma 1.1(i),(ii). By Lemma 3.10 and Remarks 3.29, the branch γ can be centred in finite position, (x0, z0), and the projected branch pr1(ηj0(γ)) is centred, in finite position, along the axis x = x0. Suppose that the values valγ(ηj), (∗), are distinct from valγ(ηj0), for j 6= j0. If these values, (∗), are, themselves, different, then, using the observation that the values are finite, we clearly obtain m seperate points of C, along the axis x = x0. However, the projected branch ρ0 = pr1(ηj0(γ)) belongs to ℘, and, therefore, Iρ0(C, x = x0) = 2. It follows, by Bezout's theorem, that there can only exist m − 1 different points along the axis x = x0. It follows that there must exist {ηj1, ηj2}, with (j0, j1, j2) a distinct triple, such that valγ(ηj1) = valγ(ηj2). Now, by Lemma 3.30, the projected branch ρ1 = pr1(ηj1(γ)) belongs to ℘. It follows that we obtain two different branches {ρ0, ρ1} ⊂ ℘, centred at seperate points along the axis x = x0. This contradicts Lemma 1.1(iv). Hence, there must exist j0 6= j1, with valγ(ηj0) = valγ(ηj1). In particular, pr1(ηj1(γ)) belongs to ℘. Therefore, γ ∈ (Γj0 ∩ Γj1). If the intersection of two of the sets in S1≤j0<j1≤m(Γj0 ∩ Γj1), (∗∗), were non-empty, then, using the first part of Lemma 3.30, we would obtain a distinct triple (j0, j1, j2), with valγ(ηj0) = valγ(ηj1) = valγ(ηj2). This contradicts the second part of Lemma 3.30. Hence, the union (∗∗) is disjoint. (cid:3) We make the following definition; Definition 3.32. Let γ be a blue branch of C ′, in finite position, cen- tred at (x0, z0). We define a function h ∈ L(C ′) to be symmetric at γ, if; For generic x1 ∈ Vx0, and {(x1, y1), . . . , (x1, yt)} = (γ ∩ x = x1), we have that; h(x1, y1) = . . . = h(x1, yt), (∗). Remarks 3.33. This is a good definition. Without loss of generality, assume that valγ(h) = y0, and let γ′ be the corresponding branch, cen- tred at (x0, y0, z0), of the image curve C ′ h, and γ′′ the corresponding branch, centred at (x0, y0), of the projected curve h ⊂ P 3, with h : C ′ ! C ′ 11The case that valγ(ηj0 ) is infinite, being the more difficult calculation. We approximate the point [1 : 0 : 0], by [ 1 ǫ : 1 ǫj u(ǫ) : 1 : 1] for the centre of the branch ηj0 (γ). This specialises to [1 : u(0) : 0 : 0] and projects to [1 : u(0) : 0], on l∞. See similar calculations, previously in the paper. ǫ : 1 : 1] and obtain the approximation [ 1 46 TRISTRAM DE PIRO h), with pr1 : C ′ pr1(C ′ Card(γ′′ ∩ (x = x1)) = 1, witnessed by the projections; h). We are then able to observe that pr1(C ′ h (x1, h(x1, y1)) = . . . = (x1, h(x1, yt)) It follows that Iγ ′′(pr1(C ′ h), x = x0) = 1, by definitions in [6]. In particular, the branch γ′′ is non-singular, and cuts the axis x = x0 transversely. By reversing the argument, we deduce easily that, for any x1 ∈ Vx0, the property (∗) of Definition 3.32 holds. We observe the following symmetry property; Lemma 3.34. Let C ′ satisfy the above conditions and let γ be a blue branch of C ′ in finite position. Then, it is not the case that every function {η1(x), . . . , ηm(x)} is symmetric at γ. Proof. Suppose, for contradiction, that each function {η1(x), . . . , ηm(x)} is symmetric at γ. Let γ be centred at (x0, y0), and choose x1 ∈ Vx0 generically, with {(x1, y1), . . . , (x1, yt)} = (γ ∩(x = x1)). As the branch γ is blue, we have that t ≥ 2. By Lemma 3.23(i), we can find a bira- tional morphism θg : C ′ ! C ′, with θg(x1, y1) = (x1, y2). θ∗ g defines an isomorphism of the function field L(C ′)/L(x), in particular it permutes the set of roots {η1(x), . . . , ηm(x)}. The permutation is non-trivial, as L(C ′) is generated by these roots, over L(x), and θ∗ g 6= Id. Without loss of generality, we can, therefore, assume that θ∗ g(η1) = η2. As both (x1, y1) and θg(x1, y1) belong to γ, we must have that γ is invariant under the action of θg. In particular, for any x1 ∈ Vx0, with x1 6= x0, θg defines a non trivial permutation of the fibre γ ∩ (x = x1) Now, for any point (x1, y1) on the branch γ, distinct from (x0, y0), using the assumption of symmetry, we have that; η2(x1, y1) = θ∗ g(η1)(x1, y1) = η1(θg(x1, y1)) = η1(x1, y1) This implies that the functions η1 and η2 are identical, which is not the case. Hence, one of the functions {η1(x), . . . , ηm(x)} is not symmetric at γ. (cid:3) Remarks 3.35. Definition 3.32 can be easily extended to include the case of a blue branch centred at [1 : 0 : 0]. Namely, we define a function h ∈ L(C ′) to be symmetric at γ, if (∗) holds, with the requirement that x1 ∈ V∞ is generic, (in which case the intersections γ ∩ x = x1 are all FLASH GEOMETRY OF ALGEBRAIC CURVES 47 in finite position). It is a straightforward exercise, left to the reader, to check that Remarks 3.33, and Lemma 3.34 hold for this definition. We can now specify the positions of the blue and silver branches of C ′; Lemma 3.36. Suppose that the r vertical tangents of C are situated along the axes {x = a1, . . . , x = ar}, in the coordinate system (x, y), then the geometric blue branches of C ′, up to birationality, see Defini- tions 3.11 and 3.28, are situated exactly along the same axes, in the coordinate system (x, z), and the geometric silver branches of C ′, up to birationality, are situated exactly along the complementary axes, that is axes of the form x = b, for b distinct from {a1, . . . , ar}. In particular, the geometric branches, which, up to birationality, are situated along the line x = 0 are silver. Proof. Suppose, for contradiction, that a blue branch γ, in finite posi- tion, is situated along a complementary axis x = b. By Lemma 3.34, we can assume that the function η1(x) is not symmetric at γ. As we saw in Lemma 3.30, valγ(η1) = d < ∞. We, therefore, obtain, by the standard method, a corresponding branch γ′ of C, centred at (b, d). As η1 is not symmetric, it is a straightforward calculation, to show that Iγ ′(C, x = b) ≥ 2. This implies that γ′ is a vertical tangent of C, con- tradicting the supposition of the lemma. Any geometric blue branch can be centred in finite position, using Lemma 3.10 and Definition 3.11. If γ′ defines a vertical tangent of C, centred along x = aj, for a given 1 ≤ j ≤ r, then, using Lemma 3.30, we can find a corresponding branch γ, up to birationality, in finite position, along the same axis x = aj. It is a simple calculation, to show that γ must be a blue branch. This show the first part of the lemma. The second part follows immediately from Lemma 3.13. The final part follows from Lemma 1.1(iii). (cid:3) Lemma 3.37. All the perpendicular branches of C ′ are silver. Proof. Suppose, for contradiction, that γ is a perpendicular blue branch. By Definition 3.28, Lemma 3.10 and Definition 3.11, we can assume that γ is centred at [1 : 0 : 0]. As Iγ(C ′, l∞) ≥ 2, by Definition 3.11, and, without loss of generality, the function η1(x) is not symmetric at γ, by Remarks 3.35, we can find distinct points [ 1 : z1 : 1] and ǫ ǫ : z2 : 1], belonging to γ ∩ x = 1 [ 1 ǫ , z2). The ǫ , z1), : 1], [ 1 distinct points {[ 1 ǫ , z1), : 1]} belong to the pro- jected image branch γ′ of C ′, centred along the line l∞. It follows that Iγ ′(C ′, l∞) ≥ 2. This contradicts Lemma 1.1(i). (cid:3) ǫ , with η1( 1 ǫ : η1( 1 ǫ : η1( 1 ǫ , z1) 6= η1( 1 48 TRISTRAM DE PIRO We can use the above specification to simplify the analysis of the nodes of C. We make the following refinement of terminology; Definition 3.38. Let ℘ν enumerate the branches of C, satisfying Def- inition 3.25(ii), Γj,ν the lifting of these branches to Cj and C ′, as in Definition 3.26, and Γν = S1≤j≤m Γj,ν. It is a straightforward exercise to refine Lemma 3.27 and Lemma 3.31, namely; Lemma 3.39. The sets Γj,ν and Γν are birationally invariant, and Γν = F1≤j0<j1≤m(Γj0,ν ∩ Γj1,ν) is a disjoint union of pairwise intersec- tions. Definition 3.40. If γ ∈ Γν, we say that γ is associated to a node ν0 of C, if, pr1(ηj0(γ)) and pr1(ηj1(γ)), determined uniquely by Lemma 3.39, form the branches of ν0. We can define an equivalence relation on Γν as follows; Definition 3.41. Let γ1 and γ2 belong to Γν. We define γ1 ∼ γ2 if they are associated to the same node of C. Remarks 3.42. It is an interesting exercise to show that, if the s nodes of C are situated along the axes {x = b1, . . . , x = bs}, in the coordinate system (x, y), then any 2 branches of C ′, which, up to birationality, can be centred along a given axis x = bj, for some 1 ≤ j ≤ s, in the coordinate system (x, z), belong to Γν and are equivalent. It is also clear, from the above, that the equivalence relation on Γν has exactly s classes. With these definitions, we have; Lemma 3.43. There exists C ′, satisfying the conditions of Lemma 3.3 and Lemma 3.21, with s branches {γ1, . . . , γs} belonging to Γν, representing each equivalence classes in Definition 3.41, such that each branch is isolated and in finite position. Proof. By Lemma 3.36, and Lemma 1.1(iv), all the branches belonging to Γν are silver. We can represent each equivalence class by s distinct silver branches, which, up to birationality, can be centred along the distinct coordinate axes {x = b1, . . . , x = bs}. By Lemma 3.14, we can simultaneously isolate these branches in finite position. (cid:3) FLASH GEOMETRY OF ALGEBRAIC CURVES 49 The following result is geometrically clarifying; Lemma 3.44. Flash Intersections Let hypotheses and notation be as in the preceding lemmas, then; (i). All the branches of the flashes {C1, . . . , Cm} are non-singular. (ii). No three flashes intersect in a point. That is there is no point p ∈ S and distinct flashes {Cj1, Cj2, Cj3}, with j1 < j2 < j3, such that p ∈ (Cj1 ∩ Cj2 ∩ Cj3). (iii). There exists a mapping from the intersections of flashes to the vertical tangent points and nodes of C. That is; a. For every point p ∈ S and distinct flashes {Cj1, Cj2}, with j1 < j2, such that p ∈ (Cj1 ∩ Cj2), we have that pr1(p) defines a vertical tangent point or a node of C in the sense of Lemma 1.1(ii). b. For every p1 ∈ C, defining a vertical tangent point or a node, there exists a p ∈ S and distinct flashes {Cj1, Cj2}, with j1 < j2, such that p ∈ (Cj1 ∩ Cj2) and pr1(p) = p1. (iv). There exists a bijection between the intersections of flashes and the spirit lines. That is; a. Every intersection between two distinct flashes {Cj1, Cj2}, with j1 < j2, lies on a unique spirit line l ⊂ S. b. For every spirit line l ⊂ S, there exists a unique point p ∈ S and distinct flashes {Cj1, Cj2}, with j1 < j2, such that p ∈ (l ∩ Cj1 ∩ Cj2). (v). In particular, there exists a mapping from the spirit lines to the vertical tangent lines of C. Proof. We adopt the coordinate of Definition 3.1. The coordinate pro- jection pr1 defines a quasi finite map from (Ci \ {Q1}) to C, which extends to give a well-defined map from the branches γ of Ci to C. This follows from the observation that there exists a unique morphism to C ns, for which (pr1 ◦ θi) = (θ ◦ φ), φ from the desingularisation C ns i 50 TRISTRAM DE PIRO l , infinitely close to Pl, with Ci∩P ′ where {θ, θi} are the desingularization morphisms of {C, Ci} respec- tively. We claim that if γ is a singular branch of Ci, then pr1(γ) is a singular branch of C, (∗). Let pr1(γ) be centred at A ∈ P 2, and let γ be centred at B ∈ P 3. Let l be a line passing through A. Let Pl be the plane spanned by the lines {l, lAQ1}, which must pass through the point B. If γ is singular, then Iitalian,γ(Ci, Pl) ≥ 2, hence, we can choose a plane P ′ l ∩VB, containing two distinct points {B1, B2}. By an appropriate choice of P ′ l , we can assume that B2 /∈ lB1Q1. Let l′ = P 2∩P ′ l , then {pr1(B1), pr1(B2)} are distinct points belonging to C ∩ l′ ∩ VA. It follows that pr1(γ) must be singular, show- ing (∗). By the hypothesis that C is non-singular, it follows that all the branches of Ci are non-singular, giving (i). Suppose, for contradiction, that three flashes intersect in a point p ∈ S. Without loss of general- ity, assume that q = pr2(p) ∈ C ′ is in finite position. Choosing generic q′ ∈ Vq, we can assume that the functions {ηj1, ηj2, ηj3} are defined at q′, with {ηj1(q′), ηj2(q′), ηj3(q′)} ⊂ Vp. As the flashes {Cj1, Cj2, Cj3} are distinct, and therefore intersect in finitely many points, by assum- ing q′ is generic, we must have that the y-coordinates {yjl, yj2, yj3} of the triple {ηj1(q′), ηj2(q′), ηj3(q′)} are also distinct, (†). Suppose that pr1(p) ∈ C has coordinates (a, b) in the system (x, y), then, the coor- dinates of {pr1(ηj1(q′)), pr1(ηj2(q′)), pr1(ηj3(q′))} ⊂ C are of the form {(a′yjl), (a′, yj2), (a′, yj3)}, with a′ ∈ Va, and {yjl, yj2, yj3} ⊂ Vb. It fol- lows that I(a,b)(C, x = a) = 3, contradicting non-singularity of C or Lemma 1.1(ii). This shows (ii). The remaining cases (iii) and (iv) are left as an exercise for the reader, It is mainly a question of converting the results of Lemmas 3.30 and Lemma 3.31. One should be careful, to exclude exceptional cases, arising from the images of perpendicular branches. (cid:3) 4. Asymptotic Degenerations The flashes model of a non-singular algebraic curve C is useful for studying the problem of degenerations of curves to lines in general po- sition, as, such degenerations can be viewed as a continuous family of flashes, (♯). (12) This is made clearer by the following; Definition 4.1. Asymptotic Degenerations 12The modern catastrophe theory approach to the problem seems, at least from an aesthetic and practical point of view, to have certain deficiencies. FLASH GEOMETRY OF ALGEBRAIC CURVES 51 Let {Ct}t∈P 1 be a family of plane projective algebraic curves of de- gree m ≥ 2, with the following properties, relative to a fixed coordinate system {x, y}; (i). The intersection of each curve Ct in the family with the axis x = 0 consists of a fixed set of m distinct points {(0, a1), . . . , (0, aj), . . . , (0, am)} in finite position, distinct from (0, 0). (13). (ii). For generic t ∈ P 1, Ct is irreducible. (iii). Each curve Ct has m distinct non-singular branches, centred at the points {(0, a1), . . . , (0, am)}, and, the tangent lines {la1, . . . , lam} at these branches are all fixed and distinct from the axis x = 0. We call such a family of curves an asymptotic degeneration. Lemma 4.2. Existence of Asymptotic Degenerations Any nodal curve C is part of a non-trivial asymptotic degeneration. Proof. Given a plane nodal curve C, the existence of asymptotic de- generation of C, follows from; (i). Lemma 1.1. (ii). Dimension calculations for the space of such curves, with fixed degree and number of nodes. The reader should look at [19] and [28] for more details about the calculations in (ii). Any such curve C is part of an irreducible space of dimension 3m − 1 + k, where m = deg(C), and k is a non-negative integer depending on the number of nodes. The condition on asymp- totes imposes at most m + 1 ≤ 3m − 1 ≤ 3m − 1 + k conditions on such a space, (∗). This follows from the observation that the condition on tangent lines in 3.28(iii) is a consequence, (∗∗), of 3.28(i), 3.28(ii), 13It is, on occasion, useful to impose the following additional assumptions; (a). Using projective coordinates {X, Y, W }, the intersection of each curve Ct in the family, with the axis W = 0, consists of a fixed set of m different points, distinct from [1 : 0 : 0]. (b). The tangent line to one of the branches, centred at a point in (b), is fixed, and distinct from the axis W = 0. This is technically convenient in terms of later terminology, and is required to exclude extremely exceptional cases, which will be mentioned later. 52 TRISTRAM DE PIRO fy(x,y;t) (0,aj ). Lemma 1.1 and the extra single condition that the tangent line to one branch, centred along the axis x = 0 is fixed. The proof of (∗∗) is straightforward, suppose that we have a family of curves {Ct}t∈P 1, satisfying the conditions 3.28(i),(ii), with the branches of the generic curve, centred at {(0, a1), . . . , (0, am)}, transverse to x = 0. Using an analytic representation of a given branch, at (0, aj), see [6], the gra- dient of the tangent line at (0, aj) is given by the rational function θj(t) = − fx(x,y;t) If the condition on tangent lines in 3.28(iii) fails, then θj0(t) is non-constant, for some 1 ≤ j0 ≤ m. In this case, by compactness of the parameter space P 1, it attains the value ∞, for finitely many {t1, . . . , tr}. This implies that fy(x, y; tj) = 0, for 1 ≤ j ≤ r, and the branch, at (0, aj0) is either tangent to the axis x = 0, or singular for the curve Ctj . By Bezout's theorem, it follows that Ctj either has at most m − 1 distinct points of intersection with the axis x = 0, contradicting the condition 3.28(i), or contains the axis x = 0 as a component. This last case is excluded by the condition that the tangent line to one branch, centred along the axis x = 0 is fixed. The stipulation that the curve does not contain the point (0, 0) is simply achieved by a translation.(14) (cid:3) Lemma 4.3. Uniformity of Asymptotic Degenerations Let {Ct}t∈P 1 be an asymptotic degeneration, of fixed degree, then, there exists a corresponding algebraic family of plane curves, {Dr : r ∈ R}, in a fixed coordinate system {x, z} with R → P 1 a finite cover, such that; (i). For generic r ∈ R, Dr is irreducible, (0, 0) ∈ Dr, and (Dr, (0, 0)) is an etale cover of (A1, (0, 0)). (ii). For non-generic r∞ ∈ R, corresponding to t∞, with Ct∞ hav- t∞}, Dr∞ is a union of component curves r∞, . . . , Dl ing components {C 1 {D1 t∞, . . . , C k r∞}, with l ≥ k. There exists an algebraic family of parameters, s ∈ S, with S → R → P 1 a finite cover, and a sequence of morphisms {η1,s, . . . , ηj,s, . . . , ηm,s : 14To obtain the extra exceptional requirements in footnote 11, one needs to impose a further m + 1 conditions. Lemma 1.1 allows one to find two distinct axes, which cut the given curve transversely. The existence of such strongly asymptotic degenerations follows again from the fact that 2(m + 1) ≤ 3m − 1 ≤ 3m − 1 + k, if m ≥ 2. The case m = 1 is just a line. FLASH GEOMETRY OF ALGEBRAIC CURVES 53 s ∈ S}, with the properties; (iii). For corresponding generic s ∈ S, r ∈ R, t ∈ P 1; (a). ηj,sDr is birational, of fixed degree, and ηj,s(0, 0) = (0, aj, 0), relative to the coordinate systems (x, z) ⊂ (x, y, z). (b). The sequence of image curves {C1,r,s, . . . , Cj,r,s, . . . , Cm,r,s} are distinct irreducible covers of Ct, using the projection pr1 from the co- ordinate system {x, y, z} to {x, y}. (iv). For corresponding non-generic (s∞, r∞, ∞) ∈ S × R × P 1, the reduced, (15), functions {η1,s∞, . . . , ηj,s∞, . . . , ηm,s∞} restrict to mor- phisms on each component {D1 r∞} of Dr∞, with the properties that; r∞, . . . , Dt r∞, . . . , Di (a). ηj,s∞Di r∞ is birational. (b). Each image curve Ci,j,∞ is a cover of an irreducible component of C∞. (c). The intersections (Ci0,j0,∞ ∩ Ci1,j1,∞) are finite for (i0, j0) 6= (i1, j1). (d). For any given j, every irreducible component of C∞ is covered by an image curve Ci,j,∞. (v). For 1 ≤ j ≤ m, the family of image curves {Cj,r,s}, including the non-generic limit curves {Ci,j,∞}, form an irreducible closed variety over a 1-parameter space Sj, with S → Sj → R → P 1 finite covers. Proof. Let P N denote the parameter space for plane curves of degree n, in the coordinate system {X, Z, W }, where N = n(n+3) . We, first, observe that there exists a constructible set V ⊂ P N , with the proper- ties that, for r ∈ V ; 2 (i). The corresponding curve Dr is irreducible. 15By which we mean the functions contain no factors, vanishing or taking the value ∞, entirely on a component of Dr∞ ; removing such factors may lower their degree. 54 TRISTRAM DE PIRO (ii). Dr passes through (0, 0), in the coordinate system (x, z), where {x = X W , z = Z W }. (iii). (Dr, (0, 0)) defines an etale cover of (A1, 0). (∗) Let Fn(X, Z, W, r) = Pi+j+k=n rijkX iY jW k enumerate all plane curves of degree n, and let Fp be similar polynomials, enumerating all plane curves of degree 1 ≤ p < n. Let Φ1(r) ≡ ¬∃1≤p<nx1 . . . xp . . . xn−1∃1≤q<ny1 . . . yq . . . yn−1 [Wp+q=n Fn(x, z, r) = Fp(x, z, xp)Fq(x, z, yq)] By standard properties, Φ1 defines a constructible subset V1 ⊂ P N , such that, (i) of (), holds for r ∈ V1. Let Φ2(r) ≡ [Φ1(r) ∧ Fn(0, 0, r) = 0] Then Φ2 defines a constructible set V2 ⊂ V1 ⊂ P N , for which (ii) of (), holds as well. Let Φ(r) ≡ [Φ2(r) ∧ ∂F ∂Z (X, Z, W, r)[0:0:1] 6= 0] It is an easy exercise, using the definition of an etale morphism, to see that Φ defines a constructible set V ⊂ V2, for which all the conditions of () are satisfied. We adopt the coordinate notation of Definition 3.1. Let P 2Nj be a parameter space for the set of rational functions of degree nj on P 2, where Nj = nj(nj +3) . We observe, that, for a pair (t, r) ∈ (P 1 × V ), there exists at most one rational function ηj(x, z), of fixed degree nj,(16), with the properties that; 2 (i). ηj(x, z) is defined at (0, 0) and ηj(0, 0) = aj (ii). The image curve Cj,t,r, defined as the projective closure of; {(x, y, z) : y = ηj(x, z), (x, z) ∈ Dr} 16Having a uniquely determined parameter sj ∈ P 2Nj FLASH GEOMETRY OF ALGEBRAIC CURVES 55 is a cover of Ct, using the projection pr1, (), (17) For a pair (x, z) belonging to the infinitesimal neighborhood (V(0,0) ∩ Dr), there can clearly exist at most one possible point (x, y, z), for which (x, y) belongs to (V(0,aj ) ∩ Ct), by the definition 3.28(iii). Hence, ηj is uniquely determined on this neighborhood, and, therefore, on the curve Dr, showing (). By Lemma 3.21, for generic t ∈ P 1, (†), and appropriate {n, nj}, there exists r ∈ V and sj ∈ P 2Ni, such that () is satisfied for the triple (t, r, sj). Let Gnj (X, Z, W, zi) enumerate all rational functions of degree nj, and let {r, s1, . . . , sj, . . . , sm} denote new parameters, with each sj abbreviating sj, and r abbreviating r. Let Ψj(t, r, sj) ≡ [Φ(r) ∧ Gnj (0, 0, sj) = aj ∧ ((Fn(x, z, r) = 0) → Ct(x, Gnj (x, z, sj)))] and; Ψ(t, r, s1, . . . , sj, . . . sm) ≡ ∧1≤j≤mΨj(t, r, sj) By elementary considerations, Ψ defines a constructible subset U of (P 1 × V × P 2N1 × . . . P 2Ni × . . . P 2Nm). We let Θ(t) ≡ ∃r∃s1 . . . ∃sj . . . ∃smΨ(t, r, s1, . . . , sj, . . . , sm). By (†), Θ defines an open subset of P 1. We define S = U to be the closure of U in (P 1 × V × P 2N1 × . . . P 2Nj × . . . P 2Nm), and also R ⊂ (P 1 × V ) to be the projection pr(S). Without loss of generality, we can assume that S and R are irreducible, with the properties; (i). pr : S → R → P 1 is onto. (ii). For a generic pair (t, r) ∈ R, there exists a unique corresponding triple (t, r, s) ∈ S (iii).For a generic pair (t, r) ∈ R, r ∈ V . () 17The conic projection on P 3 is defined, excluding the exceptional point [0 : 0 : 1 : 0], in the case that a curve passes through this point, one still obtains a well defined projected curve, see the paper [6] for the explicit construction. 56 TRISTRAM DE PIRO where s = (s1, . . . , sj, . . . , sm). This follows from elementary consid- erations on dimension, the above mentioned property of Θ, and (∗∗). We assume that dim(S) ≥ 2. We then make the simple observation that, for a suitable embedding of S in P L, it is possible to find a hy- perplane H ⊂ P L, with the properties that; (a). (H ∩ S) is irreducible and (dim(H) ∩ S) = dim(S) − 1. (b). pr(H ∩ S) = P 1 (c). H passes through a generic triple (t, r, s) ∈ S. This follows from a generalised version of Bertini's Theorem, see [14] or [15]. It follows, by induction, that we can obtain the statement (), with the additional property that dim(S) = 1. We then rephrase the property (iii) of (∗); (iii)'. For generic t ∈ P 1, the fibre R(t) ⊂ V . The statements (i) and (iii)(a), (b) of the theorem now follow imme- diately from ()(i),(ii),(iii), the distinctness claim following immedi- ately from the fact that the functions ηj,s are different on the curve Dr, by (∗∗)(i). We now consider the non-generic cases in the theorem. We abbreviate the tuple s to s and let; Sj(t, r, sj) ≡ ∃s1 . . . ∃si6=j . . . ∃smS(t, r, s1, . . . , si, . . . , sm) and Uj = prj(U), so that Uj ⊆ Sj. We consider the following vari- eties Wj, Tj ⊆ Sj × P 3 defined by; Wj(t, r, sj, x, y, z) ≡ Uj(t, r, sj) ∧ Dr(x, z) = 0 ∧ gs2 j (x, z) 6= 0 ∧ gs1 j g s2 j (x,z) (x,z) = y Tj(t, r, sj, x, y, z) ≡ Uj(t, r, sj) ∧ gs2 j (x, z) 6= 0 ∧ (x,z) (x,z) = y g g s1 j s2 j , gs2 j j where {gs1 } are homogeneous polynomials, of degree nj, rep- resented by the parameter sj. Tj is irreducible, as Uj × A2 is irre- ducible and a generic fibre Tj(t, r, sj, x, z) consists of a single point. Wj is irreducible, as Uj is irreducible, and a generic fibre Wj(t, r, sj) corresponds to the irreducible image curve Cj,r,s considered in case FLASH GEOMETRY OF ALGEBRAIC CURVES 57 (iii)(b). It follows that the Zariski closures Wj, Tj, which we again de- note by Wj, Tj ⊆ Sj × P 3, are also irreducible. We clearly have that Wj ⊂ Tj. It follows, by dimension considerations, that, for non-generic (t∞, r∞, sj,∞), the fibre Wj(t∞, r∞, sj,∞) defines a closed, possibly not irreducible, variety of dimension 1 Cj,t∞,r∞ ⊂ P 3, (18). The aim of case (iv) in the theorem, is to describe these limit curves Cj,t∞,r∞ more ex- plicitly, this will incidentally exclude the scenario of footnote 18. We now examine the behaviour of the morphism ηj,∞(x, z), deter- mined by the parameter sj,∞, on the curve Dt∞,r∞, and aim to show that the curve Cj,t∞,r∞ is obtained as the image of Dt∞,r∞, after making certain reductions in the factorisation of ηj,∞, see footnote 15, (19). We consider the closed variety Vj ⊂ (Sj × P 2) defined by; Vj(t, r, sj, x, z) ≡ Sj(t, r, sj) ∧ Dr(x, z) Vj is equidimensional, as the space Qn of degree n curves is irre- ducible, a proof is given in the paper [5], the variety (Sj × P 2) is irre- ducible, and Vj is obtained as the intersection of these varieties within a non-singular total projective space. Irreducibility follows easily from the fact that the generic curve Dr is irreducible, see the similar ar- gument in the just quoted proof. Now suppose that (x0, z0) belongs to some component of the curve Dt∞,r∞, then we can find sj,∞ with Vj(t∞, r∞, sj,∞, x0, z0). By Lemma 3.2 of [4], we can find a generic (t, r, sj, x1, z1), belonging to Vj, specialising to (t∞, r∞, sj,∞, x0, z0). As dim(Vj) = 2, dim(x1, z1/t, r, sj) = 1, in particular, we may assume that (x, z) = 0, using also (x1, z1) does not belong to the curve defined by gs2 cannot vanish identically on the irreducible curve Dt,r, the fact that gs2 by the description (iii)(a) of the theorem. It follows that we can find y1, with Wj(t, r, sj, x1, y1, z1), and, by specialisation, there exists y0, with Wj(t∞, r∞, sj,∞, x0, y0, z0). This shows that the "projection" of the curve Cj,t∞,r∞ contains the curve Dt∞,r∞, (). j j As the variety Tj is irreducible and Wj ⊂ Tj, by Lemma 3.2 of [4], if a point (x0, y0, z0) lies on the limit curve Cj,t∞,r∞, there exists 18It is conceivable that the variety could have extraneous points, that is irre- ducible components of dimension 0. We will refer to the variety as a curve in the course of the proof. 19Until (!), we make the assumption that Dt∞,r∞ does not contain the line W = 0 as a component in the projective coordinate system {X, Z, W } 58 TRISTRAM DE PIRO a generic, (*), (t, r, sj, x1, y1, z1) of Tj, specialising, (**), to (t∞, r∞, sj,∞, x0, y0, z0). We assume that the field < t∞, r∞, sj,∞, x0, y0, z0 > is contained in the base field L. By (), the facts that dim(Sj) = 1 and dim(Tj) = 3, we can assume that dim(t, r, sj/L) = 1, and dim(x1, z1/L) = dim(x1, z1/L < t, r, sj >) = 2. We pick indepen- dent infinitesimals {ǫ, δ1, δ2}, such that {t, r, sj} ⊂ L(ǫ)alg, {x1, z1} ⊂ L(δ1, δ2)alg, and the fields L(ǫ)alg, L(δ1, δ2)alg are algebraically disjoint over L. By results of [3] or [4], we can construct a sequence of special- isations; π1 : P (L(ǫ, δ1, δ2)alg) = P (K(ǫ)alg) → P (L(δ1, δ2)alg) = P (K) π2 : P (L(δ1, δ2)alg) → P (L) with π1 : P (L(ǫ)alg) → P (L) By the construction of a universal specialisation, see [3], we can take the composition π = π2 ◦ π1 to satisfy (**). By the definition of Tj, we have that (x1,z1) (x1,z1) = z1, we wish to compute the specialisation g g s1 j s2 j π(z1). We first show that; π1( g s1 j gs2 j (x1,z1) (x1,z1) ) = π1( g s1 j,∞ gs2 j,∞ (x1,z1) (x1,z1)) (***) By a straightforward computation, abbreviating the values {gs1 j,∞ (x1, z1), gs2 j,∞ (x1, z1)} by {k1, k2} ⊂ K, with k2 6= 0, we have; (x1,z1) (x1,z1) − gs1 j gs2 j gs1 j,∞ gs2 j,∞ (x1,z1) (x1,z1) = k1+ǫ1 k2+ǫ2 − k1 k2 = k2ǫ1−k1ǫ2 k2 2+ǫ2k2 where {ǫ1, ǫ2} are infinitesimals in K(ǫ)alg. Clearly, the last term in the above expression specialises to zero, showing (***). We now show that; π2( gs1 j,∞ gs2 j,∞ (x1,z1) (x1,z1)) = g g s 1,red j,∞ s 2,red j,∞ (x0,z0) (x0,z0) (****) where the red terminology is introduced to indicate that the homoge- } are reduced by any homogeneous fac- neous polynomials {gs1 tors, vanishing entirely on a component of the projective curve Dt∞,r∞. By (), we can pick (x0, z0) to lie on a unique component of Dt∞,r∞. , gs2 j,∞ j,∞ FLASH GEOMETRY OF ALGEBRAIC CURVES 59 j,∞ j,∞ j,∞ and gs2 = hnws1 Let this component be defined by the homogeneous polynomial h, sup- , for some integers pose that gs1 m, n ≥ 0. We can also assume that (x0, z0) does not lie on the in- tersection Dt∞,r∞ ∩ (ws2 = 0), as this consists of finitely many points on the given component. By the above construction, (x1, z1) does not lie on h = 0, as it does not lie on the curve Dt∞,r∞. It must, therefore, follow that; = hmws2 j,∞ j,∞ π2( g s1 j,∞ gs2 j,∞ (x1,z1) (x1,z1)) = π2[hn−m(x1, z1)]π2( w s1 j,∞ ws2 j,∞ (x1,z1) (x1,z1) ) = w s1 j,∞ ws2 j,∞ (x0,z0) (x0,z0) where, we have used the fact that n = m and, therefore, hn−m(x1, z1) = 1; otherwise, by a simple calculation, using the property (iii)(b), the specialised curve Ct∞ would contain either one of the points [0 : 0 : 1] or [0 : 1 : 0], contradicting Defn 3.28(i). As (x0, z0) does not lie on any other component of Dt∞,r∞, we have that; ws1 j,∞ w s2 j,∞ (x0,z0) (x0,z0) = r(x0,z0)g r(x0,z0)g s 1,red j,∞ s 2,red j,∞ (x0,z0) (x0,z0) = g g s 1,red j,∞ s 2,red j,∞ (x0,z0) (x0,z0) where r is the maximal homogeneous factor, of both gs1 , vanishing entirely on the remaining components of Dt∞,r∞. Hence, (****) is shown. Combining (***) and (****), we then obtain; and gs2 j,∞ j,∞ π( gs1 j gs2 j (x1,z1) (x1,z1) ) = g g s 1,red j,∞ s 2,red j,∞ (x0,z0) (x0,z0) Therefore, the limit curve Cj,t∞,r∞ is contained in the union of image curves, defined in (iv)(b). By (), it consists exactly of the union of these image curves. This show (iv)(a) and (v). Now, suppose that (x0, y0, z0) belongs to the image curve Ci,j,∞, therefore to the limit curve Cj,t∞,r∞, this means that we can find a generic (t, r, sj, x1, y1, z1) ∈ Wj, specialising to (t∞, r∞, sj,∞, x0, y0, z0). As (x1, y1, z1) belongs to Cj,r,s, by (iii)(b), the projection pr1(x1, y1, z1) = (x1, y1) belongs to the generic curve Ct. We consider the following closed variety Zj ⊂ (Sj × P 2) defined by; Zj(t, r, sj, x, y) ≡ Ct(x, y) We have that Zj(t, r, sj, x1, y1), hence, by specialisation, we must have Zj(t∞, r∞, sj,∞, x0, y0), that is, (x0, y0) belongs to the degenerate 60 TRISTRAM DE PIRO curve C∞. It follows that the projection, pr1 of the image curve Ci,j,∞ is contained in C∞, (*****). Conversely, let us suppose that (x0, y0) belongs to the degenerate curve C∞, then, as Zj is irreducible, by a similar argument to the above, we can find a generic (t, r, sj, x1, y1) ∈ Zj, specialising to (t∞, r∞, sj,∞, x0, y0). As (x1, y1) belongs to a generic curve Ct, by (iii)(b), we can find a lift- ing (x1, y1, z1), belonging to Cj,r,s. It follows that Wj(t, r, sj, x1, y1, z1), and, by specialisation, we must have Wj(t∞, r∞, sj,∞, x0, y0, z0), that is (x0, y0, z0) belongs to the limit curve Cj,t∞,r∞. It follows that C∞ is covered by the limit curve Cj,t∞,r∞, (******). We can assume that the reduced function ηj,∞ is non-constant, in which case, we have, by (******), that the degenerate curve C∞ is contained in a line, contradicting the assumption on degree in Definition 3.28. It follows that the projection pr1 of an image curve Ci,j,∞ has dimension 1. As Ci,j,∞ is irreducible, then, using (*****), we obtain (iv)(b). By similar reasoning to the previous argument , using (*******), we obtain (iv)(d), (20). Now, suppose that Ci0,j0,∞ and Ci1,j1,∞ are distinct image curves, we aim to show (iv)(c). Suppose that i0 6= i1. Then, there exist distinct components {Di0 r∞} of the curve Dr∞, and (reduced) birational maps; r∞, Di1 ηj0,s∞ : Di0 r∞ ! Ci0,j0,∞ ηj1,s∞ : Di1 r∞ ! Ci1,j1,∞ (♯) If (xyz) ∈ (Ci0,j0,∞ ∩ Ci1,j1,∞), then, in particularly, by (♯), (xz) ∈ r∞ ∩ Di1 r∞). We, therefore, obtain a map; (Di0 pr2 : (Ci0,j0,∞ ∩ Ci1,j1,∞ ∩ A3) → (Di0 r∞ ∩ Di1 r∞ ∩ A2) (♯♯) The fibres of pr2 are finite, by the birationality claim in (♯), and the image of pr2 is finite, by the distinctness of components claim in (♯). 20(!) The exceptional case when the limit curve Dt∞,r∞ contains the line W = 0 needs to be considered separately. For a general point [1 : λ : 0] on this line, we can find infinitesimals {ǫ1, ǫ2, ǫ3}, with [1+ǫ1 : λ+ǫ2 : ǫ3], defining a general point on the generic curve Dt,r. The image of this point [1 + ǫ1 : ǫ3ηj,s( 1+ǫ1 ) : λ + ǫ2 : ǫ3], ǫ3 on Cj,r,s, projects to [1 + ǫ1 : ǫ3ηj,s( 1+ǫ1 ) : ǫ3], on the generic curve Ct, and ǫ3 specialises to [1 : 0 : 0] on the limit curve Ct∞ . We can avoid this degenerate scenario, by imposing the extra conditions specified in footnote 12. , λ+ǫ2 ǫ3 , λ+ǫ2 ǫ3 FLASH GEOMETRY OF ALGEBRAIC CURVES 61 The result (iv)(c), then follows, by observing, (using footnote 20 to exclude the degenerate case), that the intersection of the curve Ci0,j0,∞ with the projective plane W = 0, is finite. r∞. Suppose that Di0 Now, suppose that j0 6= j1, it is sufficient to show that the inter- section of the image curves {Ci0,j0,∞, Ci0,j1,∞} is finite. Equivalently, that the reduced functions ηj0,s∞ and ηj1,s∞ define distinct birational morphisms on the component Di0 r∞ intersects the axis x = 0 at a point (0, c). Let U ⊂ Di0 r∞ denote an open sub- set on which the reduced factors {g1,red sj0,∞, g2,red sj1,∞} are non- vanishing. As Di0 r∞ is irreducible, we can find (x0, z0) ∈ (U ∩ V(0,c)). We can assume that dim(x0, z0/L) = 1, and find an infinitesimal ǫ such that {x0, z0} ⊂ L(ǫ)alg. In particular, there exist infinitesimals {ǫ1, ǫ2} ⊂ L(ǫ)alg, with (x0, z0) = (ǫ1, c + ǫ2). It is sufficient to show that the unique values of the reduced functions {ηj0,s∞, ηj1,s∞} are dis- tinct at (x0, z0). We consider the irreducible variety H ⊂ (Sj0,j1 × P 2) defined by; sj1 ,∞, g2,red sj0,∞, g1,red H(t, r, sj0, sj1, x, z) ≡ Sj0,j1(t, r, sj1) ∧ Dr(x, z) where Sj0,j1 is the irreducible variety, defined by projection of S, sim- ilarly to the construction of Sj. The intersection Hǫ1 ⊂ (Sj0,j1 × P 2) is defined by; Hǫ1(t, r, sj0, sj1, x, z) ≡ H(t, r, sj0, sj1, x, z) ∧ x = ǫ1 We have that Hǫ1(t∞, r∞, sj0,∞, sj1,∞, x0, z0) and claim that, for (t, r, sj0, sj1) ∈ (Sj0,j1 ∩ V(t∞,r∞,sj0 ,∞,sj1 ,∞)), there exists (x1, z1) ∈ P 2, with (x1, z1) ∈ V(x0,z0) and Hǫ1(t, r, sj0, sj1, x1, z1), (♯♯♯). In particular, it follows that Hǫ1 consists of a finite union {H 1 ǫ1} of irreducible curves, each of which projects onto Sj0,j1. ǫ1, . . . , H k The proof of (♯♯♯) is straightforward, using methods from the paper [5]. The tuple (r∞, (1, 0, −ǫ1)) belongs to the smooth projective space, parametrising intersections between degree n and degree 1 curves. As the intersection (Dr∞ ∩ (x = ǫ1)) is finite, we can apply Theorem 2.4 of [5]. We choose (t, r, sj0, sj1, x1, z1) generically in Hǫ1. In particular, we have that dim(t, r, sj0, sj1/L(ǫ)alg) = 1. It follows that we can find an infinitesimal δ, with {t, r, sj0, sj1} ⊂ L(δ)alg and dim(δ/L(ǫ)alg) = 62 TRISTRAM DE PIRO dim(ǫ/L(δ)alg) = 1. We employ the notation above, to produce the commuting specialisation maps; (π1 ◦ π2) : P (L(ǫ, δ)alg) → P (L(ǫ)alg) → P (L). (π2 ◦ π1) : P (L(ǫ, δ)alg) → P (L(δ)alg) → P (L). We have that π2(x1, z1) = (x0, z0) and π1(x0, z0) = (0, c). Hence, (x1, z1) = (ǫ1, c + ν1), where ν1 is an infinitesimal in L(ǫ, δ)alg. As the specialisations commute, we have that (π2 ◦ π1)(x1, z1) = (π1 ◦ π2)(x1, z1) = (0, c). It follows that π1(x1, z1) = (0, c + δ1) ∈ Dt,r, where δ1 is an infinitesimal in L(δ)alg. Let π1 define an infinitesimal neigh- borhood V(0,c+δ1), and let (x1, z1) belong to the branch γ, centred at (0, c + δ1), relative to this neighborhood. Using the condition (iii)(a) and the property of Definition 3.30(iii), by a straightforward adapta- tion of Lemma 3.27, we have that the values valγ(ηj0,s) and valγ(ηj1,s) are distinct, belonging to the fixed set {a1, . . . , am} ⊂ L. For conve- nience of notation, we will denote them by aj0 and aj1, (21). We have that; ηj0,s(x1, z1) − ηj1,s(x1, z1) = (ηγ j0,s(0, c + δ1) − ηγ j1,s(0, c + δ1)) +(ηj0,s(x1, z1) − ηγ j0,s(0, c + δ1)) −(ηj1,s(x1, z1) − ηγ j1,s(0, c + δ1)) By a simple calculation, the terms (ηj0,s(x1, z1) − ηγ j0,s(0, c + δ1)) and (ηj1,s(x1, z1) − ηγ j1,s(0, c + δ1)) belong to Ker(π1) ∩ L(ǫ, δ)alg. Hence; π1(ηj0,s(x1, z1) − ηj1,s(x1, z1)) = aj0 − aj1 and ηj0,s(x1, z1) − ηj1,s(x1, z1) = aj0 − aj1 + ν with ν an infinitesimal, belonging to L(ǫ, δ)alg. 21Although it is possible that some permutation of values can occur, for a given function ηj,s, at different branches along the axis x = 0. FLASH GEOMETRY OF ALGEBRAIC CURVES 63 It follows, using the limit calculation above, that; π2(ηj0,s(x1, z1) − ηj1,s(x1, z1)) = ηj0,s∞(x0, z0) − ηj1,s∞(x0, z0) = aj0 − aj1 + ǫ3 where ǫ3 is an infinitesimal, belonging to L(ǫ)alg. Therefore; π1(ηj0,s∞ − ηj1,s∞) = aj0 − aj1 6= 0 In particular, it follows that the reduced functions ηj0,s∞ and ηj1,s∞ take distinct values at the point (x0, z0). This shows (iv)(c). The proof of (ii) follows immediately from (iv)(b)(d). (cid:3) We now show the following fundamental properties of asymptotic degenerations; Lemma 4.4. Let G be the group obtained in the last section. Then the action of G on a generic curve Dr, with r ∈ R, extends uniformly to almost all curves of the form Dr′ in the family, with r′ ∈ R. Proof. We can consider G as acting on the curve Dr, for generic r ∈ R, replacing C ′ in Lemma 3.23. We enumerate the parameters needed to define the action in Lemma 3.23(), as a tuple; (g1 1, g2 1, . . . , g1 i , g2 i , . . . , g1 l , g2 l ) ⊂ P 2N l where Card(G) = l, which we abbreviate by g = (g1, . . . , gi, . . . , gl), (22). Let ΨG(t, r, ¯h) ≡ (a).θhi : Dr ! Dr, 1 ≤ i ≤ l (b).(θhi ◦ θhj ) = θhi∗j , 1 ≤ i, j ≤ l (c).θhi = Id if f (d).(θhi)−1 = θhi−1 , 1 ≤ i ≤ l i = 1, 1 ≤ i ≤ l with ∗ and −1 copying the group multiplication and inversion of G onto the labelled set {1, . . . , l}, and ¯h being a new variable, substituting 22Having obtained a set of 2l rational functions by the standard projective method, we can equalise their degrees, by, for example, multiplying through with factors of the form (xsg(x, z) + 1), where g(x, z) is a polynomial vanishing on the curve Dr. This is mainly technically convenient in terms of notation. 64 TRISTRAM DE PIRO g. The conditions (a) and (d) are formulated simultaneously as follows; ∃x1 . . . xp∃z1 . . . zp∀x∀z[Vp Vw i=1(x 6= xi) ∧ Vp i=1 Dt,r(θhi(x, z))∧Vw i=1(z 6= zi) ∧ Dt,r(x, z) → i=1(θhi−1 ◦θhi)(x, z) = (x, z)] The formulation of conditions (b) and (c)is a similar exercise, which we leave to the reader. By Lemma 3.23, the condition ψG holds for the tuple (t, r, g). Imitating the construction above of the variety S, in Lemma 4.3, we can find an irreducible closed variety E ⊂ R × P 2N l, with E → R a finite cover, and open subsets UG ⊂ E and VG ⊂ D, for which the formula ΨG holds and pr−1(VG) ⊂ UG respectively. If (t′, r′) ∈ VG, then the above construction provides a Galois group action on the curve Dr′. By imitating the proof of Lemma 3.23, and using conditions (c) and (d), one can assume that such an action has the generic transitivity property. (cid:3) As before, we consider the non-generic case, adopting the terminol- ogy from Lemma 4.4; Lemma 4.5. Let (t∞, r∞, ¯h∞) be a non-generic element of E, then, the reduced functions {θred hl,∞} restrict to morphisms from the curve Dt∞,r∞ to itself, satisfying conditions (a) − (d) of the previous lemma, and the generic transitivity property. Proof. We consider the closed variety J ⊂ (E × P 2), defined by; h1,∞, . . . , θred J(t, r, ¯h, x, z) ≡ E(t, r, ¯h) ∧ Dt,r(x, z) By similar arguments to the above, J is irreducible. We also consider the variety JG ⊂ (E × P 2), defined by; JG(t, r, ¯h, x, z) ≡ VG(t, r) ∧ J(t, r, ¯h, x, z) ∧ ∃z1 . . . zi . . . zl(z = hi (cid:5) zi) (23) We let J G be the Zariski closure of JG in (E × P 2). By the construc- tion of VG, a generic fibre J G(t, r, ¯h) is exactly the curve Dt,r ⊂ P 2. As VG and Dt,r are both irreducible, it follows that JG, and, therefore, J G are both irreducible. Hence, J = J G, (∗). Now suppose, for contradic- tion, that the reduced function θred h1,∞ fails to define a morphism from 23The reader should supply the extra conditions necessary to exclude the finite sets of points for which the action (cid:5) of G is undefined, a similar argument being used in Lemma 4.3 FLASH GEOMETRY OF ALGEBRAIC CURVES 65 h1,∞ = 0, (∗ ∗ ∗), where qred the curve Dt∞,r∞ to itself. Then the intersection Dt∞,r∞ ∩θred h1,∞(Dt∞,r∞) defines a proper closed subset of Dt∞,r∞. We can choose (x0, z0) from Dt∞,r∞, not lying on this intersection, (∗∗). By (∗), we have that JG(t∞, r∞, ¯h∞, x0, z0). As JG is irreducible, we can find a generic (t, r, ¯h, x1, z1), belonging to JG, specialising to (t∞, r∞, ¯h∞, x0, z0). By the definition of VG, we can find (x1, z2), belonging to Dt,r, with h1 (cid:5)z2 = z1. We can assume that (x1, z2) does not specialise to a point belonging to Dt∞,r∞ ∩ qred h1,∞ is the reduced denominator term, used in the definition of θred h1,∞, see Lemma 3.23. This follows by observing that there are only a finite number of points on the curve Dt∞,r∞, for which the reduced denominator term is zero. We can choose the coordinate x0 so that the line x = x0 has an empty intersection with this finite number of points. Therefore, we may assume (∗ ∗ ∗). It follows, by the limit calculation used in Lemma 4.3, that the spe- cialisation (x0, z0) of θh1(x1, z2) is θred h1,∞(x0, z3), with (x0, z3) ∈ Dt∞,r∞ . This contradicts (∗∗). Hence, the first part of the lemma is shown. We now check conditions (a) − (d). Condition (a) follows from condition (d) and the fact that the set of reduced functions {θred hl,∞} de- fine morphisms on the curve Dt∞,r∞. Conditions (b) and (d) amount to checking that for a triple (t, r, ¯g), specialising to (t∞, r∞, ¯g∞), for a generic point (x1, z1) of Dt,r, specialising to (x0, z0) of Dt∞,r∞, we have that; h1,∞, . . . , θred sp(θhi ◦ θhj (x1, z1)) = θred hi,∞ ◦ θred hj,∞(x0, z0) (∗ ∗ ∗∗) We assume that the three points {(x1, z1), θhj (x1, z1), θhi ◦θhj (x1, z1)} satisfy (∗ ∗ ∗), by excluding a finite number of the coordinates x0. In which case, we have, by repeating the limit calculation in Lemma 4.3, that; sp(θhi ◦ θhj (x1, z1)) = θred = θred = θred hi,∞(sp(θhj (x1, z1))) hi,∞(θred hi,∞ ◦ θred hj,∞(sp(x1, z1))) hj,∞(x0, z0) as required in (∗ ∗ ∗∗). One direction of condition (c) is straightfor- ward and left to the reader. We show the converse direction, (∗ ∗ ∗ ∗ ∗). We let Ei, for 1 ≤ i ≤ l, be the projections of E onto the l factors of G. As before, the covers Ei → R are irreducible and finite. We also introduce the combined parameter spaces; 66 TRISTRAM DE PIRO Sj,k(t, r, sj, sk) ≡ Sj(t, r, sj) ∧ Sk(t, r, sk), for 1 ≤ j, k ≤ m Fi,j,k(t, r, hi, sj, sk) ≡ Ei(t, r, hi) ∧ Sj,k(t, r, sj, sk), for 1 ≤ i ≤ l, 1 ≤ j, k ≤ m Again, the covers Sj,k → R and Fi,j,k → R are irreducible and finite. We recall, from Lemma 3.23, that, for a generic parameter (t, r) ∈ R, any g ∈ G, acting on the curve Dt,r, induces an action on the asso- ciated set of flashes {C1,r,s, . . . , Cj,r,s, . . . Cm,r,s}. If θg 6= Id, then, by definition of the Galois group, this action is non-trivial. By the enu- meration in the previous lemma, we denote g by g2 6= g1, and ,without loss of generality, we can suppose that (g2 (cid:5) C1,r,s) = C2,r,s. We recall the definition of the variety T1 ⊂ F2,1,2 × P 3, considered in Lemma 4.3; T1(t, r, h2, s1, s2, x, y, z) ≡ F2,1,2(t, r, h2, s1, s2)∧Dt,r(x, z)∧y = η1,s1(x, z) and define the following variety T2,g2 ⊂ F2,1,2 × P 3; T2,g2(t, r, h2, s1, s2, x, y, z) ≡ F2,1,2(t, r, h2, s1, s2) ∧ Dt,r(x, z) ∧ y = (η2,s2 ◦ θ2,h2)(x, z) (24) As we showed in Lemma 4.3, the Zariski closure T1 is irreducible. By the condition that (θ2,h2)∗(η2,s2) = η1,s1 on the curve Dt,r, for a generic choice of (t, r, h2, s1, s2), we clearly have that T1 = T2,g2, and the closure T2,g2 is irreducible. By results in Lemma 4.3, a degener- ate fibre T1(t∞, r∞, h2,∞, s1,∞, s2,∞) is exactly the limit curve, C1,t∞,r∞ . Suppose (x0, y0, z0) ∈ C1,t∞,r∞ , then we can choose a generic point (t, r, h2, s1, s2, x1, y1, z1) on the variety T2,g2, specialising to the tuple (t∞, r∞, h2,∞, s1,∞, s2,∞, x0, y0, z0). By the definition of T2,g2, we have that y1 = (η2,s2 ◦ θ2,h2)(x1, z1). Repeating the above calculation; y0 = sp(y1) = sp((η2,s2 ◦ θ2,h2)(x1, z1)) 2,s2,∞(sp(θ2,h2(x1, z1))) 2,s2,∞(θred 2,s2,∞ ◦ θred = ηred = ηred = (ηred 2,h2,∞)(x0, z0) 2,h2,∞(sp(x1, z1))) 24It is necessary to supply extra conditions, in order to ensure both varieties are well defined. This was done in Lemma 4.3, for the variety T1. For the variety T2,g2, one should restrict the parameter x, by adding the conditions that x /∈ Discx′(Dt,r(x′, z), qh2 (x′, z) = 0) and x /∈ Discx′(Dt,r(x′, z), gs2 (x′, z) = 0) 2 2 FLASH GEOMETRY OF ALGEBRAIC CURVES 67 where, we take the usual precaution, of restricting the range of the coordinate x0, in order to apply the calculation of Lemma 4.3. The 2,h2,∞)∗(η2,s2,∞) = η1,s1,∞ on the result implies that, curve Dt∞,r∞. By Lemma 4.3(c), we have that the rational functions {η1,s1,∞, η2,s2,∞} are distinct. Hence, we conclude, that θred 2,h2,∞ 6= Id. This shows (∗ ∗ ∗ ∗ ∗). in the limit, (θred We finally show the generic transitivity property, for the curve Dt∞,r∞. Consider the variety F ⊂ R × P 1 × P 2; F (t, r, x0, x1, X, Z, W ) ≡ Dt,r(X, Z, W ) ∧ x0X = x1W which defines a finite cover F → (R × P 1). Let Q1 = [0 : 1 : 0], and let k = mult(t∞,r∞,a,Q1)(F/(R × P 1), where a is generic over the param- eter (t∞, r∞), and the parameters defining R. For generic (t, r, a) ∈ (R × P 1), we have that IQ1(Dt,r, x = a) = n − l, where Card(G) = l and deg(Dt,r) = n. This follows, from a simple application of Be- zout's theorem, and the previously observed fact, that the curve Dt,r has l points in finite position, along the axis x = a. It follows that IQ1(Dt∞,r∞, x = a) = n − l + (k − 1) = n − (l − k + 1). This follows from summability of specialisation, see [5]. Applying Bezout's theorem again, the curve Dt∞,r∞ has exactly (l − k + 1) points in finite position along the generic axis x = a. If k > 1, we obtain l1 < l points in finite position along a generic axis. By the counting argument, used in Lemma 3.23, properties (a) − (d) shown above, and the fact that Card(G) = l, we can find a reduced function θred hi,∞, for some (i > 2), acting as the identity on Dt∞,r∞, this contradicts property (c). Hence, k = 1, and the curve Dt∞,r∞ has exactly l points in finite position along a generic axis. Transitivity of the G-action, then follows by applying the methods of Lemma 3.23, (25) (cid:3) We now observe the following uniformity in results from the previous section. Lemma 4.6. Let g(x, z) be chosen, satisfying the conditions of Lemma 3.4, with corresponding θg, then θg defines a birational morphism on each curve in the family Dr, for r ∈ R. Moreover, we can find a 25For a general 1-dimensional family of curves of fixed projective degree, all of which are irreducible, it is not necessarily true, that deg(Ct0 ) = deg(Ct1 ), for a pair (t1, t2), where deg(C) denotes deg(L(C)/L(x)). It is an interesting question to determine the conditions for which this property holds. The family parametrised by R is close to having this property, depending on whether the degenerate fibres can be chosen as irreducible. 68 TRISTRAM DE PIRO g )∗(ηj,r,s), for generic (r, s), ηred new space of parameters R′, satisfying conditions 4.3(i),(ii), S′ sat- isfying conditions 4.3(iii),(iv),(v), such that the algebraic family Dr′, for r′ ∈ R, consists of the images of θg : Dr ! Dr′, the functions ηj,r′,s′ = (θ−1 j,r∞,s∞), for degenerate (r∞, s∞) and E′, satisfying Lemma 4.4, with θr′,¯h′ = (θ−1 ), for de- generate (r∞, ¯h∞). There exists a naturally defined total morphism from the unprimed to the primed parameter spaces, Γg : (R, S, E) → (R′, S′, E′), such that all relevant diagrams commute. g )∗(θr,¯h), for generic (r, ¯h), and θred ∞,¯h′ r′ g )∗(ηred = (θ−1 = (θ−1 g )∗(θred r∞,¯h∞ j,r′ ∞,s′ ∞ ∞ Proof. The proof is straightforward but lengthy. That θg defines a bi- rational morphism on each curve in the family Dr, for r ∈ R, follows easily from the proof of Lemma 3.4, (26). In order to construct the pa- rameter space R′, observe that we can write the equation of a general member of the family Dr, in the form; P0≤i+j≤n rijxizj = 0 (∗) Now make the substitution z = g(x, z′) in (∗) and clear the denomi- nator term, to obtain the expression; P0≤i+j≤n rijxi[a(x)z′ + b(x)]j[c(x)z′ + d(x)]n−j = 0 (∗∗) We can then expand (∗∗), obtaining an expression of the form; P0≤i+j≤n′ r′ ijxiz′j (∗ ∗ ∗) in which each r′ ij term is a linear combination of terms rij, with coeffi- cients depending on the those of the polynomials {a(x), b(x), c(x), d(x)}. Comparing (∗) and (∗ ∗ ∗), we obtain a linear projective morphism Γ′ from R ⊂ P N to R′ ⊂ P N ′, which parametrises an irreducible 1- dimensional family of curves, with degree n′. By construction, we have that θg : Dr ! Dr′. The family R′ satisfies condition 4.3(i), by the proof of Lemma 3.4. The construction of S′ is similar, by substitu- tion of the polynomial g−1(x, z′), defining the inverse transformation θ−1 (x, z)}. The statement g , for z, in the rational functions {gs1 (x, z), gs2 j j 26One can easily exclude the exceptional case that θg is not uniformly and bi- rationally defined, namely that a degenerate curve Dr∞ has a component l∞, by imposing the extra conditions in footnote 12 FLASH GEOMETRY OF ALGEBRAIC CURVES 69 concerning the transformation of ηj,r,s, for generic (r, s) follows imme- diately from this method, while the statement concerning the transfor- mation of ηred j∞,r∞,s∞, for degenerate (r∞, s∞), is a limit calculation, of a form we have considered above. We leave the construction of E′ and the associated transformations of the Galois action to the reader. It is clear how the total morphism Γg is obtained from the partial linear projective morphisms on the spaces {R, S, E}. (cid:3) Remarks 4.7. The above construction can be repeated for any abstract birational map θg, which is uniformly birational on the curves in the family Dr, for r ∈ R, and preserves the conditions of Lemma 3.3, for a generic curve, see also footnote 3. When referring to birational maps on a degenerate fibre, we will assume that these extra conditions are satisfied. In particular, we always have that (pr ◦ θg) = pr uniformly. As a consequence of the lemma, we are able to extend Definition 3.11, to the case of a degenerate curve Dt∞,r∞; Definition 4.8. We define a branch γ of Dt∞,r∞ to be blue, if there ∞, see previous remark, exists a birational map θ : Dt∞,r∞ such that the either/or clause of Definition 3.11 holds, for Dt∞,r′ ∞ and θ(γ). We define a branch γ to be silver otherwise. ! Dt∞,r′ Remarks 4.9. This is a good definition, using the same argument as in Remarks 3.12(b). We can also give an equivalent definition of a silver branch, using Remarks 3.12(c). We can now use uniformity of the G-action, to simplify the geometry of the degenerate curve. Lemma 4.10. Lemma 3.13 holds for Dt∞,r∞, that is the distribution of silver and blue branches along any axis is uniform. Proof. The result follows immediately from the proof of Lemma 3.13 and uniformity of the G-action, Lemma 4.5. (cid:3) and to exclude symmetry properties; Lemma 4.11. Lemma 3.34 holds for Dt∞,r∞. That is, for a blue branch γ in finite position, it is not the case that every function {ηred is symmetric at γ. 1,t∞,r∞, . . . , ηred m,t∞,r∞} Proof. Using Lemma 3.34, Lemma 4.3(iv)(c) and Lemma 4.5. (cid:3) 70 TRISTRAM DE PIRO 5. Asymptotic Degenerations to Lines Passing through a Single Point We consider a special case of the families considered in the previous section, for which a degenerate fibre Ct∞ of {Ct}t∈P 1, consists of m lines, passing through a single point O. It is convenient to assign the point O with the coordinates (k, 0), for k 6= 0, in the (x, y) plane. The m distinct lines must then correspond to l[(0,aj ),(k,0)], for 1 ≤ j ≤ m, with equations (aj − 1)x + ky = 0. The analysis of degenerate flash intersections is easier due to the simplified geometry of this configu- ration. In particular, the image curves Ci,j,∞ of Lemma 4.3, are all contained in the union of planes formed by the m distinct lines and the point Q1 ∈ P 3. We make the following definition; Definition 5.1. Let ℓ = {l1, . . . , lm} enumerate the branches of Ct∞, which are centred at O. Definition 5.2. For the branches in Definition 5.1, we define Υj = {γ1,j, . . . , γs,j} to be the lifting of these branches to the flash Cj,∞ = S1≤i≤t Ci,j,∞, and also to the curve Dt∞,r∞. We define Υ = S1≤j≤m Υj. Remarks 5.3. We emphasise the point that, for any j, every branch in ℓ lifts to a branch in Υj, this follows immediately from Lemma 4.3(iv)(d) We have the same birational invariance of such branches, as in Lemma 3.27; Lemma 5.4. Let θ : Dt∞,r∞ sense of Remarks 4.7, then the branches {γ, ηred lift the same branch l on Ct∞ as {θ(γ), (ηred In particular, the sets Υj and Υ are birationally invariant. ! Dt∞,r′ j,s∞,t∞(γ)} of {Dt∞,r∞, Cj,s∞} ∞, Cj,s′ ∞,t∞◦θ◦pr2)(γ)} of Dt∞,r′ ∞ be a birational map, in the j,s′ ∞. Proof. The proof is the same as Lemma 3.27. (cid:3) We have a stronger version of Lemmas 3.30 and 3.31; Lemma 5.5. Let γ be a geometric branch of Dt∞,r∞. Then; valγ(ηred j0,s∞,t∞) = valγ(ηred j1,s∞,t∞), for j0 6= j1, iff γ ∈ (Υj0 ∩ Υj1) iff γ ∈ T1≤j≤m Υj Moreover, Υ = Υj, for 1 ≤ j ≤ m. FLASH GEOMETRY OF ALGEBRAIC CURVES 71 j1,s∞,r∞) = 0. Suppose that valγ(ηred Proof. The first two equivalences follow by using the proof of Lemma 3.30, together with the fact that Dt∞,r∞ has a unique singularity at O. In order to show the third equivalence, suppose that the branch γ is in finite position along the axis x = k, with valγ(ηred j0,s∞,r∞) = valγ(ηred j2,s∞,r∞) = a 6= 0, for some j2 /∈ {j0, j1}. We can assume that this value is finite, in which case we obtain that (k, a) belongs to the curve Dt∞,r∞. This is clearly impos- sible. Hence, valγ(ηred j2,s∞,r∞) = 0, and, γ belongs to Υj2. For the final statement, simply observe that if, without loss of generality, γ belongs to Υ1, then γ is centred along the axis x = k, and valγ(ηred 1,s∞,r∞) = 0. If valγ(ηred j,s∞,r∞) = a 6= 0, for j 6= 1, we again obtain a point (k, a) belonging to the curve Dt∞,r∞, which is a contradiction. Therefore γ ∈ T1≤j≤m Υj, and Υ = Υj, for 1 ≤ j ≤ m. (cid:3) We obtain the following characterisation of branches for the degen- erate curve; Lemma 5.6. All the branches of Dt∞,r∞ are silver. Proof. Using the proofs of Lemma 3.36 and Lemma 3.37. (cid:3) We now show that we can simultaneously isolate silver branches, and refer to the previous section for notation; Lemma 5.7. Suppose that γ is a silver branch, in finite position, of the generic curve Dt,r, centred at (k′, a′), specialising to the point (k, a) of the degenerate curve Dt∞,r∞. Then, after a sequence of birational transformations, we can assume that γ is isolated and there exists a unique silver branch of Dt∞,r∞, centred at (k, a). Proof. We follow the proof of Lemma 3.14. Let (k′ + x, a′ + η1(x)) be a parametrisation of γ, where η(x) is a power series belonging to K[[x]], with sp(P (K)) = P (L), of the form; η(x) = P∞ j=1 a′ jxj Suppose that the sequence (a′, a′ quence (a, a1, . . . , aj, . . .). We define; 1, . . . , a′ j, . . .) specialises to the se- g(x, z) = (z−a) (x−k) − a k and apply the birational map θg, satisfying Lemma 4.6. Let { γ1, . . . , γr} enumerate the branches of Dt∞,r∞, centred at (k, a), then, using Lemma 72 TRISTRAM DE PIRO in finite position along the axis 5.6, the only remaining branches, x = k, after applying this transformation, are {θg( γ1), . . . , θg( γr)}. Their centres depend only on the first coefficient b1 of the power series {η1(x), . . . , ηr(x)} defining these branches. We consider the effect of θg on γ. The power series representation of θg(γ) is given by; (k′ + x, (a′−a)+η(x) (k′−k)+x − a k ) = (k′ + x, ǫ1+η(x) ǫ2+x − a k ) so, the centre of γ1 = θg(γ), specialises to (k, a1 − a k ). By elemen- tary properties of specialisations, one of the branches { γ1, . . . , γr} must therefore be centred at (k, a1 − a k ) = (k, c1), requiring that b1 = a1. We repeat the process for the new function; g1(x, z) = (z−c1) (x−k) − c1 k Now the centre of θg1(γ1), specialises to (k, a2 − c1 k ). Again, one of the remaining branches in { γ1, . . . , γr} must be centred at (k, a2 − c1 k ) = (k, c2), requiring that, the second coefficient of the defining series b2 = a2. We repeat the process, until we are left with a single silver branch γ1 say, along the axis x = k, for the curve Dt∞,r∞. The original branch γ is also isolated, using, for example, preservation of singularities, see Remarks 5.11. Hence, the lemma is shown (cid:3) Remarks 5.8. As an immediate consequence of the proof, we see that for the remaining silver branch γ, the original power series representa- tion is given by; a + η(x), where η(x) = P∞ j=1 ajxj By applying the finite number of birational transformations, in the previous lemma, to the series (a′ + η(x)) and (a + η(x)), we obtain, in the final situation, that the defining series of {γ, γ} are related by specialisation. We make the following definition; Definition 5.9. By a general nodal degeneration, we mean a family of curves {Ct : t ∈ P 1}, satisfying (i) − (iv) of the conditions at the beginning of Section 1, together with (v′). C∞ is a union of m distinct lines {l1, . . . , lm}. FLASH GEOMETRY OF ALGEBRAIC CURVES 73 A desirable property of general nodal degenerations is the following; Definition 5.10. We say that a general nodal degeneration has good specialisation, if, for any given node of the generic curve Ct, centred at (a′, b′), and specialising to a point (a, b), formed by the intersection of two lines {l1, l2} of the degenerate curve C∞; if {c′, d′} denote the gradients of the tangent lines to the two branches of the node on Ct, then the specialisations {c, d} are the gradients of the two lines {l1, l2}. Remarks 5.11. It is always true that the point (a′, b′) specialises to a point (a, b) formed by the intersection of two lines. This is simply a consequence of the preservation of singularities. More specifcally, if the family of curves is determined by a polynomial; F (X, Y, W ; t) = Pi+j+k=m aijk(t)X iY jZ k ∂Y = ∂F ∂X = ∂F Then the nodes of Ct are determined by the intersections of the vari- eties ∂F ∂Z = 0. For a point (a′, b′) belonging to this intersec- tion, the specialisation also belongs to this intersection at t = ∞. This is clearly a singular point of the variety C∞, hence, a point formed by the intersection of two lines. However, the stronger property concern- ing specialisation of gradients is not necessarily true; Zariski gives a counterexample in his paper [28], refuting Severi's original proof of his conjecture. We show how to correct this error; Theorem 5.12. An asymptotic general nodal degeneration has good specialisation. Proof. We will consider only the special case when the degenerate fi- bre consists of m distinct lines passing through a point O. The more general case of m distinct lines is left as an exercise for the reader. Let {ρ1, ρ2} be branches of a given node of the generic curve Ct. By Lemmas 3.36 and 3.43, we can find a silver branch γ ∈ Dt,r, which is isolated in finite position, and such that, without loss of generality, η1,r,s(γ) and η2,r,s(γ) lift the branches ρ1 and ρ2 respectively, with value ǫ. We have, using, for example, Remarks 5.11, that γ is centred at a point (k′, a′), specialising to (k, a). By Lemma 5.7, we can assume that there exists a unique silver branch γ of the degenerate curve Dt∞,r∞, centred at (k, a). The point (t, r, s, k′, ǫ, a′) belongs to the closed vari- ety W1,2, where Wj is defined in Lemma 4.3; 74 TRISTRAM DE PIRO W1,2(t, r, s, x, y, z) ≡ W1(t, r, s, x, y, z) ∧ W2(t, r, s, x, y, z) Hence, so does the specialisation (t∞, r∞, s∞, k, 0, a). It follows from the description in Lemma 4.3(iv), that; valγ(ηred 1,r∞,s∞) = valγ(ηred 2,r∞,s∞) = 0 By Lemma 5.5, it follows that ηred 2,r∞,s∞(γ) lift the branches l1 and l2 respectively, where { l1, l2} are the branches formed by the intersection of a pair of lines at the origin O. 1,r∞,s∞(γ) and ηred We now compute the slopes of the branches ρ1 and ρ2, denoted {c′, d′}. Recall that the plane curves Dr(x, z) are parametrised by; F (x, z; r) = Pi+j≤k rijxizj and the rational function η1,r,s(x, z) = c′ = D(η1,r,s)(k′,a′) (cid:5) D(Fr)(k′,a′) = ∂η1,r,s ∂x d′ = D(η2,r,s)(k′,a′) (cid:5) D(Fr)(k′,a′) = ∂η2,r,s ∂x 1 ∂Fr ∂y (k′,a′) ∂Fr ∂y (k′,a′) (x,z) (x,z). We have that; gs1 gs2 ∂x (k′,a′) + ∂η1,r,s ∂y ∂Fr 1 ∂Fr ∂x (k′,a′) + ∂η2,r,s ∂y Using a similar limit calculation to the above, (27), we have; sp(c′) = sp(D(η1,r,s)(k′,a′) (cid:5) D(Fr)(k′,a′)) = D(ηred 1,r∞,s∞)(k,a) (cid:5) D(Fr∞)(k,a) sp(d′) = sp(D(η2,r,s)(k′,a′) (cid:5)D(Fr)(k′,a′)) = D(ηred 2,r∞,s∞)(k,a) (cid:5)D(Fr∞)(k,a) By the same reasoning as above, we see that sp(c′) = c and sp(d′) = d (cid:3) as required. It is clearly impossible to obtain an asymptototic degeneration, in the sense of Definition 4.1, to a set of lines passing through a single point O, without imposing the additional assumption on C, that all of the tangent lines to the branches along the axis x = 0 pass through the point O. However, we can find a weaker version of Definition 4.1(iii), 27The differential calculation should be checked carefully, one should compute the differential of (x,z) (x,z) by parts, first, and then, as was done above, find the specialisation of this function, without specialising the corresponding point. It is easily checked that this specialisation is the same as the differential of ηred gs1 gs2 1 1 1,r∞,s∞ FLASH GEOMETRY OF ALGEBRAIC CURVES 75 for which the results we have shown still hold, namely; (iii)'. Almost every curve Ct, has m distinct non-singular branches, centred at the points {(0, a1), . . . , (0, am)}. We will also refer to this condition as asymptotic. The specific construction of asymptotic degenerations to lines through a given point, follows using a projective method, generally referred to as Severi's cone construction. We consider a class of degenerations, obtained by projecting a fixed irreducible curve C ⊂ P 3, onto a plane H ⊂ P 3, from a variable point Q. We first show the following lemma; Lemma 5.13. Let C ⊂ P 3 be an irreducible projective curve, not con- tained in a hyperplane section. Let ω, l be a hyperplane and a line, re- spectively, with l ω, and (l ∩ ω) = Q. Let P arm = P N denote the pa- rameter space for plane projective curves of degree m, where N =m(m+3) . Then, there exist morphisms {θ, φ}; 2 θ : P 1 → P art, Ct def ined by θ(t) φ : P 1 ∼= l, φ(t) = Qt, Q = Q0 with the property that, for t 6= 0, Ct is the plane projective curve, obtained as the projection of C from Qt to ω, see the remarks at the beginning of Section 4 of [6]. Proof. Without loss of generality, by making a linear change of vari- ables, we can suppose that the coordinates of P 3 are given by {X, Y, Z, W } and {X, Y, W } denote the coordinates of the hyperplane ω, embedded into P 3 by the map; i : P 2 → P 3 [X : Y : W ] 7→ [X : Y : 0 : W ] We denote the restriction of these coordinates to A2 and A3 by (x, y) and (x, y, z) respectively, and we have a corresponding inclusion; i : A2 → A3 (x, y) 7→ (x, y, 0) 76 TRISTRAM DE PIRO We can assume that Q corresponds to the origin (0, 0, 0) of the co- ordinate system (x, y, z), with projective coordinates [0 : 0 : 0 : 1], and l corresponds to the line x = y = 0. Let C be defined by the affine equations F1(x, y, z) = F2(x, y, z) = 0, and let (0, 0, t) parametrise a point Qt, along the line l. Let; Ct(x, y) ≡ ∃λ∃z1∃z2∃z3[F1(z1, z2, z3) = F2(z1, z2, z3) = 0 ∧λ(z1, z2, z3)+(1−λ)(0, 0, t) = (x, y, 0)] (∗) By the assumptions on C, the construction in Section 4 of [6], and the definition (∗), for t 6= 0, Ct(x, y) = prQt(C), where the closure is taken inside the projective plane ω or P 2. We let C ⊂ A3 denote the closure of C(x, y, t). By elementary dimension considerations, and the fact that the generic fibre Ct(x, y) is irreducible, being an open subset of the irreducible curve prQt(C), C is also irreducible. One can, therefore, find a single equation G(x, y, t) = 0, defining the surface C. By making the substitutions x = X W , we can find a homogeneous poly- nomial H(X, Y, W, t), of degree m, in the variables {X, Y, W }, whose restriction to A3, corresponds to G(x, y, t). We define θ, as a mor- phism to P arm, by taking the coefficients of the polynomial H. It is a simple exercise to check that the conditions of the lemma are satisfied. W and y = Y (cid:3) We observe the following further property of the degenerations, given in Lemma 5.13. Lemma 5.14. Projective Degenerations Let C ⊂ P 3 be an irreducible projective curve and ω a hyperplane, intersecting C transversely in m distinct points {P1, . . . , Pm}, where m = deg(C). Let l be a line, intersecting ω in a new point Q, then the curve C0 consists of the union of lines S1≤j≤m lPj Q. Proof. We show that the curves Ct are obtained as the intersection of ω, with the cone ConeQt(C), for t ∈ P 1, (∗). We, first, consider the case, that t 6= 0, and, without loss of generality, assume that Qt /∈ C, the case when Qt ∈ C being left to the reader. It is a straightforward exercise, to check that; ConeQt(C) = Sw∈C lQtw FLASH GEOMETRY OF ALGEBRAIC CURVES 77 is a closed, irreducible,subvariety of P 3. The fact that prQt(C) = (ConeQt(C) ∩ ω) follows immediately from the definitions of the pro- jection prQt, and the cone ConeQt(C). As before, we assume that C is defined by the affine equations F1(x, y, z) = F2(x, y, z) = 0, Qt is parametrised by (0, 0, t) and ω is defined by z = 0, in the coordinate system (x, y, z). Let; Dt(x, y, z) ≡ ∃λ∃z1∃z2∃z3[F1(z1, z2, z3) = F2(z1, z2, z3) = 0 ∧ λ(z1, z2, z3) + (1 − λ)(0, 0, t) = (x, y, z)] Let D ⊂ A4 denote the closure of D(x, y, z, t). As before, using the fact that the generic fibre Dt(x, y, z) is an open subset of the irreducible variety ConeQt(C), D is irreducible and we can find a single equation G(x, y, z, t) = 0, defining the surface D. By projectivizing the polyno- mial G, we can find a single homogeneous polynomial G(X, Y, Z, W, t), parametrising the cones ConeQt(C). We let; H(X, Y, W, t) ≡ G(X, Y, 0, W, t) By definition, H parametrises the intersections (ConeQt(C)∩ω). It is a simple exercise, to check that this is a different definition of the same variety, given in Lemma 8.1, hence (∗) is shown. In order to complete the proof, it is sufficient to compute the intersection (ConeQ0(C)∩ω). If w ∈ (C \ ω), then (lwQ0 ∩ω) = Q0, whereas, if w ∈ (C ∩ω), then (lwQ0 ∩ ω) = lwQ0, using the fact that Q0 ∈ ω. The result follows immediately, from the fact that (C ∩ω) = {P1, . . . , Pm}, and distinctness of the point Q = Q0. (cid:3) The above lemmas allow us to easily construct degenerations to lines through a given point. In order to obtain the extra asymptotic condi- tions, we require the following variant; Lemma 5.15. Mirror Construction Let {ω1, ω2} ⊂ P 3 be two planes intersecting in a line l. Let C ⊂ ω1, with the property that prP (C) = C, relative to the hyperplane ω2 ⊂ P 3. Let l′ be a line, passing through P , and intersecting ω in Q. The set of projections; prx(C) : x ∈ (l′ \ Q)} 78 TRISTRAM DE PIRO together with the limit curve "prQ(C)", forms a projective degenera- tion, in the sense of the previous lemma and remarks at the beginning of Section 1. Moreover, if {P1, . . . , Pm} denote the points of intersec- tions (C ∩ l) = (C ∩ l), then the degeneration is asymptotic relative to the axis l. Proof. The proof that this gives a degeneration, satisfying the previous lemma and remarks at the beginning of Section 1, is given in the pre- vious lemma and the paper [6], using Lemma 8.1 to show that degree is preserved, and Lemma 8.3, to show that the genus may be preserved for a suitable choice of P (a birational projection will preserve the genus of the original curve). The limit curve "prQ(C)" is not formally defined in terms of projections, however, is easily shown to be a union Sy∈(C∩ω1) lQy, by the previous lemma. Again, the reader should look at the paper [6], Lemma 8.2. The asymptotic condition follows straight- forwardly from the fact that each curve in the degeneration contains the points {P1, . . . , Pm}, which are fixed by each projection, being in the plane ω2. (cid:3) The above results correct the erroneous step of Severi's conjecture, found in [20], as pointed out by Zariski in [28]. Zariski's counterexample fails to apply to Severi's cone construction, as it is not asymptotic. We will leave a closer examination of Severi's proof to another paper. References [1] S.S. Abhyankar, Algebraic Geometry for Scientists and Engineers, AMS Mathematical Surveys 35, (1990) [2] Painton Cowen, "The Rose Window, Splendour and Symbol", Thames and Hudson, (2005) [3] T. de Piro, Infinitesimals in a Recursively Enumerable Prime Model, available at http://www.magneticstrix.net [4] T. de Piro, Zariski Structures and Algebraic Curves, available at http://www.magneticstrix.net [5] T. de Piro, A Non-Standard Bezout Theorem for Curves, available at http://www.magneticstrix.net [6] T. de Piro, A Theory of Branches for Algebraic Curves, available at http://www.magneticstrix.net FLASH GEOMETRY OF ALGEBRAIC CURVES 79 [7] T. de Piro, A Theory of Divisors for Algebraic Curves, available at http://www.magneticstrix.net [8] T. de Piro, Some Geometry of Nodal Curves, available at http://www.magneticstrix.net [9] T. de Piro, A Note on the Etale Topology, available at http://www.magneticstrix.net [10] T.de Piro, A Theory of Duality for Algebraic Curves, available at http://www.magneticstrix.net [11] John Gribbin and Mary Gribbin, From Here to Infinity, National Maritime Museum, Royal Observatory, Greenwich, (2008) [12] R. Guralnick, Monodromy Groups of Coverings of Curves, Galois Groups and Fundamental Groups, 1-46, Math. Sci. Res. Inst. Publ., 41, Cambridge University Press, Cambridge, (2003). [13] , J. Harris, On the Severi Problem, [14] R. Hartshorne, Algebraic Geometry, Springer (1977) [15] S. Kleiman, Bertini and his Two Fundamental Theorems, AG arXiv, 9704018. [16] Helga Meyerbroker, "Rose Windows and How to Make Them", Coloured tissue paper crafts, Floris Books, (1989) [17] J.S Milne, Etale http://www.math.lsa.umich.edu/∼jmilne/ Lectures on Cohomology, available at [18] D. Mumford, Red Book of Varieties and Schemes, Springer (1999) [19] E. Sernesi, Deformations of Algebraic Schemes, Volume 334, Springer, (2006) [20] F. Severi, Vorlesungen uber algebraische Geometrie, Anhang F, Teubner, Leipzig, 1921. [21] E.H Neville, The Tracing of Cubic Curves, The Mathematical Gazette, Vol 18, No. 230, pp 258-266, (1934) [22] Isaac Newton, Analysis by Equations of an Infinite Number of Terms, (written in 1669, a published version (in Latin) appearing in 1711, and a translated published version appearing in 1745) [23] Isaac Newton, Enumeration of Lines of the Third Order. 80 TRISTRAM DE PIRO [24] Isaac Newton, The Method of Fluxions and Infinite Series. [25] J. Nunemacher, Asymptotes, Cubic Curves, and the Projective Plane, Mathematics Magazine, Volume 72, No. 3, June 1999. [26] E.H Smart, The Nature of an Algebraic Curve at Infinity and the Conic of Closest Contact at Infinity, The Mathematical Gazette, Vol 13, No. 188, pp 357-361, (1927) [27] C.R.M Talbot, Notes and Examples on Sir Isaac Newton's Enumeration of Lines of the Third Order, (H.G Bohn, Covent Garden, London), (1861) [28] Dimension Theoretic Characterization of Maximal Irreducible Algebraic Systems, American Journal of Mathematics, Vol 104, No 1, pp 209-226, (1979) Flat 1, 98 Prestbury Road, Cheltenham, GL52 2DJ E-mail address: [email protected]
1509.08284	2	1509	2016-02-25T06:03:49	Genus one enumerative invariants in del-Pezzo surfaces with a fixed complex structure	[ "math.AG", "math.SG" ]	We obtain a formula for the number of genus one curves with a fixed complex structure of a given degree on a del-Pezzo surface that pass through an appropriate number of generic points of the surface. This enumerative problem is expressed as the difference between the symplectic invariant and an intersection number on the moduli space of rational curves.	math.AG	math	GENUS ONE ENUMERATIVE INVARIANTS IN DEL-PEZZO SURFACES WITH A FIXED COMPLEX STRUCTURE INDRANIL BISWAS, RITWIK MUKHERJEE, AND VARUN THAKRE Abstract. We obtain a formula for the number of genus one curves with a fixed complex structure of a given degree on a del-Pezzo surface that pass through an appropriate number of generic points of the surface. This enumerative problem is expressed as the difference between the symplectic invariant and an intersection number on the moduli space of rational curves. R´esum´e. Invariants ´enum´eratifs de genre un avec une structure complexe fix´ee pour des surfaces de del Pezzo. Nous obtenons une formule pour le nombre de courbes de genre un avec une structure complexe fix´ee, de degr´e donn´e, et passant par un nombre appropri´e de points g´en´eriques de la surface. La solution est exprim´ee comme la diff´erence entre l'invariant symplectique et un nombre d'intersection sur l'espace de modules de courbes rationnelles. 1. Introduction Enumerative Geometry of rational curves in P2 C is a classical question in algebraic geometry. A natural generalization is to ask how many elliptic curves, with a fixed j-invariant, are there of a given degree that pass through the right number of generic points. In [8] and [4], using methods of algebraic and symplectic geometry respectively, Pandharipande and Ionel obtain a formula for the number of degree d genus one curves with a fixed complex structure in P2 C that pass through 3d − 1 generic points. In this paper, we extend their result to del-Pezzo surfaces. Let X be a complex del-Pezzo surface and β ∈ H2(X, Z) a given homology class. Let n0,β denote the number of rational curves of degree β in X that pass through δβ generic points, where δβ := hc1(T X), βi − 1. We prove the following: Theorem 1.1. Let X and β be as above. Let nj with fixed j invariant of degree β in X that pass through δβ generic points. Then 1,β denote the number of elliptic curves nj 1,β = 2gβ Aut(Σ1, j) n0,β where gβ := β · β − c1(T X) · β + 2 2 , Aut(Σ1, j) denotes the number of automorphisms of a genus one Riemann surface with fixed j invariant that fixes a point and "·" denotes topological intersection. Note that gβ in Theorem 1.1 coincides with the genus of a smooth degree β curve on X. The numbers n0,β are computed in [6, p. 29] and [2, Theorem 3.6] using a recursive formula. When X := P2, our formula for nj 1,β is consistent with the formula of Pandharipande and Ionel in [8] and [4]. In [6], the authors actually give a formula to compute the 2010 Mathematics Subject Classification. 53D45, 14N35, 14J45. Key words and phrases. Elliptic curve, del-Pezzo surface, symplectic invariant, enumerative invariant. 1 2 I. BISWAS, R. MUKHERJEE, AND V. THAKRE genus 0 Gromov -- Witten invariants of the del-Pezzo surfaces, which a priori need not be enumerative. It is shown in [2, page 63, last paragraph] that the numbers obtained in [6] are actually equal to n0,β. The results of Pandharipande and Ionel generalize the result of P. Aluffi; in [1], he computes the number of genus one cubics with a fixed complex structure in P2 through 8 generic points. The problem of enumerating elliptic curves with a fixed j-invariant has also been studied by tropical geometers. In [5], Kerber and Markwig compute the number of tropical elliptic curves in P2 with a fixed j-invariant. Combined with the correspondence theorem [7, Theorem A], one can conclude that the number computed is indeed the same as the number of plane elliptic curves with a fixed j-invariant. Currently, this question is also being studied for other surfaces. In [7], Len and Ranganathan obtain a formula for the number of elliptic curves with a fixed j-invariant of a given degree for Hirzebruch surfaces, using methods from tropical geometry. 2. Enumerative versus symplectic invariant We now explain the basic idea to compute nj 1,β. Let (X, J, ω) be a compact semi-positive symplectic manifold, with a compatible almost complex structure J of dimension 2m and β ∈ H2(X, Z) be a homology class. Let k be a nonnegative integer such that k + 2g ≥ 3. Let α1, · · · , αk and γ1, · · · , γl be integral homology classes in H∗(X, Z) such that kX i=1 2m − deg(αi) + lX j=1 (2m − 2 − deg(γj)) = 2m(1 − g) + 2hc1(T X), βi . (2.1) Fix pseudocycles Ai, 1 ≤ i ≤ k, and Bj, 1 ≤ j ≤ l, on X representing the homology classes αi and γj. Fix a compact Riemann surface Σg of genus g; its complex structure will be denoted by j. Define Mν,j g,k(X, β; α1, · · · , αk; γ1, · · · , γk) := {(u, y1, · · · , yk) ∈ C∞(Σg, X) × X k u∗[Σg] = β, ∂j,Ju = ν, u(yi) ∈ Ai ∀ i = 1, · · · , k, Im(u) ∩ Bj 6= ∅ ∀j = 1, · · · , l} , where ν : Σg × X −→ T ∗Σg ⊗ T X is a generic smooth perturbation and ∂j,Ju := 1 2(cid:16)du + J ◦ du ◦ j(cid:17). The symplectic invariant (or the Ruan -- Tian invariant) is defined to be the signed cardi- nality of the above set, i.e., RTg,β(α1, · · · , αk; γ1, · · · , γl) := ±Mν,j g,k(X, β; α1, · · · , αk; γ1, · · · , γk) . When k = 0, we denote the invariant as RTg,β(∅; γ1, · · · , γl). Furthermore, when γ1, . . . , γl all denote the class of a point, then we abbreviate the invariant as RTg,β. Similarly, when l = 0 we denote the invariant as If (2.1) is not satisfied, then we formally define the invariant to be zero. RTg,β(α1, · · · , αk; ∅). GENUS ONE ENUMERATIVE INVARIANTS IN DEL-PEZZO SURFACES 3 A natural question is to ask whether the symplectic invariant RTg,β is equal to the enu- merative invariant nj g,β. For P2, and more generally for del-Pezzo surfaces, the genus zero symplectic invariant is equal to the enumerative invariant [9, page 267]. However, even for P2, the genus one symplectic invariant is not enumerative. In general, the following fact is true ([11, Theorem 1.1]) RT1,β = Aut(Σ1, j)nj 1,β + CR , (2.2) where CR denotes a correction term. Let us explain what this term means. First we note that the factor of Aut(Σ1, j) is there because we do not mod out by automorphisms in the definition of Mν,j g,k. Hence, if u : (Σ1, j) −→ X is a solution to the ∂ -- equation and the complex structure on X is genus one regular, then there will be Aut(Σ1, j) new solutions close to u to the perturbed ∂ -- equation. Next, we note that as ν → 0, a sequence of (J, ν)-holomorphic maps can also converge to a bubble tree whose base (the torus) is a constant (ghost) map [4, page 2]. These maps will also contribute to the computation of RT1,β invariant. This extra contribution is defined to be the correction term CR. We now explain how to compute the correction term. Let M0,n(X, β) denote the moduli space of rational degree β curves on X that represent the class β ∈ H2(X, Z) and are equipped with n ordered marked points, modulo equivalence. In other words, M0,n(X, β) := {(u, y1, · · · , yn) ∈ C∞(P1, X)×(P1)n ∂u = 0, u∗[P1] = β}/PSL(2, C) , with PSL(2, C) acting diagonally on P1 × (P1)n. Let M0,n(X, β) denote the stable map compactification of M0,n(X, β). Let us now focus on M0,1(X, β), the moduli space of curves with one marked point. Let H be the divisor in M0,1(X, β) corresponding to the extra condition that the curve passes through a given point. Let L −→ M0,1(X, β) and ev : M0,1(X, β) −→ X be the universal tangent bundle and the evaluation map at the marked point. Following the same argument as in [4, Lemma 1.23], we conclude that the bundle ev∗T X −→ M0,1(X, β)∩Hδβ admits a nowhere vanishing section ν. This is because the rank of ev∗T X is two, while the dimension of the variety M0,1(X, β)∩Hδβ is one. Hence ev∗T X −→ M0,1(X, β)∩Hδβ admits a trivial sub bundle spanned by ν, which we denote as Chνi. When X := P2, it is shown in [4, Lemma 1.25], that the correction term is given by CR = hc1(L∗ ⊗ ev∗T X/Chνi), [M0,1(X, β)] ∩ Hδβ i. (2.3) A more detailed justification of (2.3) is given in [11], by using the results of [12]. Further- more, the gluing construction in [12] is valid in general for Kahler manifolds [12, page 8]. Hence, we conclude that (2.3) holds for del-Pezzo surfaces as well. Zinger also pointed out this fact to the second author of this paper in a personal communication ([13]). In the next section we will obtain a formula for c1(L∗) and compute the right hand side of (2.3). The left hand side of (2.2) is computed using the formula given in [9]. Hence, we obtain nj 1,β. 3. Computation of the correction term We will now give a self contained proof of obtaining a formula for c1(L∗) and hence computing the correction term. Alternatively, one can also compute the Chern classes by using the dilation equation and the divisor equation as given in [3, Section 26.3]. 4 I. BISWAS, R. MUKHERJEE, AND V. THAKRE Lemma 3.1. On M0,1(X, β), the following equality of divisors holds: c1(L∗) = 1 (β · x1)2(cid:16)(x1 · x1)H − 2(β · x1)ev∗(x1) + X Bβ1,β2(β2 · x1)2(cid:17) , β1+β2=β, β1,β26=0 (3.1) where H is the locus satisfying the extra condition that the curve passes through a given point, Bβ1,β2 denotes the boundary stratum corresponding to the splitting into a degree β1 curve and degree β2 curve with the last marked point lying on the degree β1 component and xi := ci(T X). Proof. The proof is similar to the one given in [4]. Let µ1 , µ2 ∈ X be two generic pseudocycles in X that represent the class Poincar´e dual to x1. Let fM be a cover of M0,1(X, β) with two additional marked points with the last two marked points lying on µ1 and µ2 respectively. More precisely, fM := ev−1 2 (µ1) ∩ ev−1 3 (µ2) ⊂ M0,3(X, β) where ev2 and ev3 denote the evaluation maps at the second and third marked points respectively. Note that the projection π : fM −→ M0,1(X, β) that forgets the last two marked points is a (β · x1)2 -- to -- one map. We now construct a meromorphic section φ : fM −→ π∗L∗ given by φ([u, y1; y2, y3]) := (y2 − y3)dy1 (y1 − y2)(y1 − y3) . (3.2) The right -- hand side of (3.2) involves an abuse of notation: it is to be interpreted in an affine coordinate chart and then extended as a meromorphic section on the whole of P1. Note that on (P1)3, the holomorphic line bundle η := q∗ 1KP1 ⊗ O(P1)3(∆12 + ∆13 − ∆23) is trivial, where q1 : (P1)3 −→ P1 is the projection to the first factor and ∆jk ⊂ (P1)3 is the divisor consisting of all points (zi , z2 , z3) such that zj = zk. The diagonal action of PSL(2, C) on (P1)3 lifts to η preserving its trivialization. The section φ in (3.2) is given by this trivialization of η. Since c1(π∗L∗) is the zero divisor minus the pole divisor of φ, we gather that c1(π∗L∗) = {y2 = y3} − {y1 = y2} − {y1 = y3} . When projected down to M0,1(X, β), the divisor {y2 = y3} becomes (x1 · x1)H + (β2 · x1)2Bβ1,β2, while both the divisors {y1 = y2} and {y1 = y3} become (β · x1)ev∗(x1). Since fM is a (β · x1)2 -- to -- one cover of M0,1(X, β), we obtain (3.1). (cid:3) Using Lemma 3.1, we conclude that hc1(L∗), [M0,1(X, β)] ∩ Hδβ i = −2n0,β. (3.3) To see why this is so, we first note that M0,1(X, β) ∩ Hδβ +1 is zero. This is because the number of rational curves through δβ + 1 generic points is zero. Next, we note that M0,1(X, β) ∩ Hδβ ∩ Bβ1,β2 is also zero. This is because the number of β curves which pass through δβ points can not split into a degree β1 curve and a degree β2 curve. This is because such a split curve will pass through δβ1 + δβ2 points, which is one less than GENUS ONE ENUMERATIVE INVARIANTS IN DEL-PEZZO SURFACES 5 δβ. So a split curve can not pass through δβ generic points. Finally we note that for any homology class µ ∈ H2(X, Z), the following is true [M0,1(X, β)] ∩ Hδβ ∩ ev∗[µ] = n0,β(β · µ). (3.4) To see why this is so, we note that the left hand side of (3.4) counts the number of degree β rational curves through δβ points and one marked point, such that the marked point lies on some cycle representing the class β. There are β · µ choices for that marked point to lie, which gives us the right hand side of (3.4). These three facts give us (3.3). Note that, when we say ev∗[µ], we mean the pullback of the cohomology class Poincar´e dual to µ (inside X). Using (3.4) (with µ := c1(T X)), we conclude that (3.5) (3.6) hc1(ev∗T X), [M0,1(X, β)] ∩ Hδβ i = (cid:16)β · c1(T X)(cid:17)n0,β . From (3.3), (3.5) and (2.3) it follows that CR = (cid:16)β · c1(T X) − 2(cid:17)n0,β . 4. Computation of the symplectic invariant We now compute the symplectic invariant RT1,β := RT1,β(∅; p1, . . . , pδβ ) using the for- mula [9, page 263, (1.2)]. Let e1 , e2 , · · · , ek be a basis for H∗(X, Z). Let gij := ei · ej and gij := (cid:16)g−1(cid:17)ij . If the degrees of ei and ej do not add up to be the dimension of X then define gij to be zero. Using [9, page 263, (1.2)] we conclude that RT1,β(∅; p1, . . . , pδβ ) = X i,j gijRT0,β(ei, ej; p1, . . . , pδβ ) = X i,j gijn0,β(β · ei)(β · ej) = (β · β)n0,β. (4.1) The last equality follows by writing β in the given basis ei and using the definition of gij; the second equality follows from the same we justify (3.4). Equations (4.1), (3.6) and (2.2) give us the formula of Theorem 1.1. 5. Regularity of the complex structure for del-Pezzo surfaces We now show that the complex structure on the del-Pezzo surfaces is genus one regular for immersion. In the statement of Theorem 1.1 the curve u passes passes through δβ generic points. Hence the curve is going to be an immersion and hence it suffices to prove regularity for immersions. Lemma 5.1. Let X be P2 blown up at k points and (Σ1, j) a compact genus 1 Riemann surface with a complex structure j. Let u : Σ1 −→ X be a holomorphic map representing the class β := dL − m1E1 − . . . − mkEk ∈ H2(X, Z), where L and Ei denote the class of a line and the exceptional divisors respectively. Then Du, the linearization of the ∂j,J at u is surjective, provided d > 0 and u is an immersion. In particular, the complex structure on the del-Pezzo surface is genus 1 regular for immersions. 6 I. BISWAS, R. MUKHERJEE, AND V. THAKRE Proof. We first note that if L is a holomorphic line bundle on (Σ1, j) of positive degree, then the cup product H 0(Σ1, L) ⊗ H 1(Σ1, L∗) −→ H 1(Σ1, L) ⊗ L∗) = C is nondegenerate. Indeed, this coincides with the Serre duality pairing because the canon- ical line bundle of Σ1 is trivial, and hence the pairing is nondegenerate. Next consider the short exact sequence of vector bundles on Σ1 given by the differential of u 0 −→ T Σ1 du−→ u∗T X −→ Q := (u∗T X)/T Σ1 −→ 0 . (5.1) Let H 0(Σ1, Q) ρ−→ H 1(Σ1, T Σ1) −→ H 1(Σ1, u∗T X) −→ H 1(Σ1, Q) (5.2) be the long exact sequence of cohomologies associated to it. We have H 1(Σ1, Q) = 0 because degree(Q) > 0 (note that degree(u∗T X) > 0 and degree(T Σ1) = 0). The exact sequence in (5.1) does not split; for the corresponding extension class ψ ∈ H 1(Σ1, (T Σ1)⊗ Q∗) = H 1(Σ1, Q∗), as observed before, there is ψ′ ∈ H 0(Σ1, Q) such that ψ ∪ ψ′ 6= 0. Hence ρ in (5.2) is nonzero. This implies that ρ is surjective because dim H 1(Σ1, u∗T X) = 1. hence from (5.2) we conclude that H 1(Σ1, u∗T X) = 0, which proves that the cokernel of Du is zero (i.e., Du is surjective). (cid:3) When X := P1 × P1, the complex structure is genus one regular; that is because for P1 the complex structure is genus one regular (by [10, Corollary 6.5]). Hence, the same fact holds for products of P1. Acknowledgements We are very grateful to the two referees for detailed comments. The second author is grateful to Hannah Markwig and Yoav Len for some fruitful discussions on this subject and informing us about the ongoing work [7]. We thank them for letting us know that this question is of interest to tropical geometers. References [1] P. Aluffi, How many smooth plane cubics with given j-invariant are tangent to 8 lines in general position?, Contemp. Math. 123 (1991), 15 -- 29. [2] L. Gottsche and R. Pandharipande, The quantum cohomology of blow-ups of P2 and enumer- ative geometry, Jour. Diff. Geom. 48 (1998), 61 -- 90. [3] K. Hori, S. Katz, A. Kelmm, R. Randharipande, R. Thomas, C. Vafa, R. Vakil, and E. Zaslow, Mirror Symmetry, Clay Math. Inst. AMS 2003. [4] E. Ionel, Genus-one enumerative invariants in Pn with fixed j-invariant, Duke Math Jour. 94 (1998), 279 -- 324. [5] M. Kerber and H. Markwig, Counting tropical elliptic plane curves with fixed j-invariant, Comment. Math. Helv. 84 (2009), 387 -- 427. [6] M. Kontsevich and Y. Manin, Gromov -- Witten classes, quantum cohomology, and enumerative geometry, in Mirror symmetry, II, 607 -- 653, vol. 1 of AMS/IP Stud. Adv. Math., Amer. Math. Soc., Providence, RI, 1997. [7] Y. Len and D. Ranganathan, Enumerative geometry of elliptic curves on toric surfaces https://arxiv.org/abs/1510.08556. [8] R. Pandharipande, Counting elliptic plane curves with fixed j-invariant, Proc. Amer. Math. Soc. 125 (1997), 3471 -- 3479. GENUS ONE ENUMERATIVE INVARIANTS IN DEL-PEZZO SURFACES 7 [9] Y. Ruan and G. Tian, A mathematical theory of quantum cohomology, Jour. Diff. Geom. 42 (1995), 259 -- 367. [10] A. Zinger, Enumeration of genus-two curves with a fixed complex structure in P2 and P3, Jour. Diff. Geom. 65 (2003), 341 -- 467. [11] A. Zinger, Enumeration of One-Nodal Rational Curves in Projective Spaces, Topology 4 (2004), 793 -- 829. [12] A. Zinger, Enumerative versus symplectic invariants, J. Symp Geom. 2 (2004), 445 -- 543. [13] A. Zinger, Personal communication. School of Mathematics, Tata Institute of fundamental research, Homi Bhabha road, Mumbai 400005, India E-mail address: [email protected] School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, India E-mail address: [email protected] Department of Mathematics, Harish Chandra Research Institute, Allahabad 211019, India E-mail address: [email protected]
1607.08383	1	1607	2016-07-28T09:45:17	Symmetric noncommutative birational transformations	[ "math.AG" ]	In a previous paper (arXiv:1410.5207) certain birational transformations were constructed between the noncommutative schemes associated to quadratic and cubic three dimensional Sklyanin algebras. In the current paper we consider the inverse birational transformations and show that they are of the same type. Moreover we extend everything to the $\mathbb{Z}$-algebras context, which allows us to incorporate the noncommutative quadrics introduced by Van den Bergh.	math.AG	math	SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. DENNIS PRESOTTO Abstract. In [10] certain birational transformations were constructed be- tween the noncommutative schemes associated to quadratic and cubic three dimensional Sklyanin algebras. In the current paper we consider the inverse birational transformations and show that they are of the same type. Moreover we extend everything to the Z-algebras context, which allows us to incorporate the noncommutative quadrics introduced by Van den Bergh in [17]. Contents Introduction Inner morphisms Inverting quadratic transforms between quadratic Sklyanin algebras and 1. 2. Acknowledgements 3. Z-algebras 3.1. AS-regular Z-algebras 3.2. Z-domains and Z-fields of fractions 4. Summary of the results in [10] 5. Addendum to [10] 5.1. Noncommutative P1 P1 (cid:57)(cid:57)(cid:75) P2. (Proof of Theorem 5.1) 5.2. Noncommutative P2 (cid:57)(cid:57)(cid:75) P1 P1 (Proof of Theorem 5.4) 6. Cremona Transformations and quadratic transforms 7. 8. 8.1. The Z2-algebra associated to a noncommutative P2 (cid:57)(cid:57)(cid:75) P1 P1 8.2. The Z2 algebra associated to a noncommutative P1 P1 (cid:57)(cid:57)(cid:75) P2 8.3. Appendix A. Quadrics admit Z-fields of fractions Appendix B. References Invertability of the quadratic transforms cubic Sklyanin Z-algebras I-bases for quadratic Sklyanin algebras and Lemma 8.3 1 4 4 5 6 7 8 9 11 13 16 18 18 22 25 27 32 37 1. Introduction Throughout this paper k will be an algebraically closed field and all rings will be algebras over k. Following tradition [5] we associate a noncommutative projective scheme X " ProjpAq to a connected graded k-algebra A " k ' A1 ' . . . which is The author was supported by a Ph.D. fellowship of the Research Foundation Flanders (FWO). 1 2 DENNIS PRESOTTO generated by A1. ProjpAq is defined via its category of "quasicoherent sheaves": QcohpXq :" QGrpAq " GrpAq{ TorspAq where GrpAq is the category of graded right A-modules and TorspAq is the full subcategory of torsion A-modules, i.e. those graded right A-modules that have locally right bounded grading [5]. A non-trivial class of noncommutative surfaces is given by three dimensional Artin-Schelter regular algebras as defined in [1]. Recall that a connected graded algebra A is called AS-regular of dimension d if it satisfies the following conditions (i) The Hilbert series hnpAq :" dimkpAnq is bounded by a polynomial in n. (ii) A has finite global dimension d (iii) A has the Gorenstein property with respect to d Such algebras with d " 3 and which are generated in degree 1 are classified in [1, 2]. There are two possibilities for the number of generators and relations: (i) A is generated by three elements satisfying three quadratic relations (the "quadratic case"). In this case A has Hilbert series 1{p1 ´ tq3, i.e. the same Hilbert series as a polynomial ring in three variables. Therefore we think of ProjpAq as being a noncommutative P2. (ii) A is generated by two elements satisfying two cubic relations (the "cubic case"). In this case A has Hilbert series 1{p1 ´ tq2p1 ´ t2q and we can think of ProjpAq as being a noncommutative P1 P1. (The rationale for this is explained in [17].) We write pr, sq for the number of generators of A and the degrees of the relations. Thus pr, sq " p3, 2q or p2, 3q depending on whether A is quadratic or cubic. The classification of three-dimensional AS-regular algebras A is in terms of suitable geometric data pY,L, σq where Y is a k-scheme, σ is an automorphism of Y and L is a line bundle on Y . More precisely: starting from pY,L, σq one can construct a twisted homogeneous coordinate ring BpY,L, σq (see for example [4]). This is a connected graded k-algebra with r :" dimkpB1q equal to 2 or 3. Setting s " 5 ´ r as above, we can then find the AS-regular algebra ApY,L, σq by dropping all relations in degree s ` 1 and higher. In particular there is a surjective morphism ApY,L, σq Ñ BpY,L, σq giving rise to an inclusion QGrpBq ãÑ QGrpAq (1) Moreover there is an equivalence of categories QGrpBq -- QcohpY q (see for example [4]). Therefore one often says there is a commutative curve Y contained inside the noncommutative surface X. Below we say that A is a (quadratic or cubic) Sklyanin algebra if Y is smooth and σ is a translation. Inspired by the commutative case it makes sense to expect that in a suitable sense noncommutative P2s are birationally equivalent to noncommutative P1P1s. Such birational equivalences were constructed in [11] and [10]. In [10] we provide a noncommutative version of the standard birational trans- formation P1 P1 (cid:57)(cid:57)(cid:75) P2 by showing that for each cubic Sklyanin algebra A there exists a quadratic Sklyanin algebra A1 and an inclusion (2) A1 ãÑ Ap2q SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 3 where A and A1 are the associated Z-algebras. (See §3 for more on Z-algebras.) We then check that this inclusion gives rise to an isomorphism of the function fields Frac0pA1q -- Frac0pAq, a result which was already announced in [14] and [13]. In [10] we also provide a noncommutative version of the Cremona transform P2 (cid:57)(cid:57)(cid:75) P2 by showing that for each quadratic Sklyanin algebra A there exists a quadratic Sklyanin algebra A1 and an inclusion (3) A1 ãÑ Ap2q The construction was based on the choice of 3 non-collinear points p, q, r on Y and if A " ApY,L, ψq then A1 " ApY,L b ψL b OY p´p ´ q ´ rq, ψq. In §4 we recapitulate the construction of (2) and (3). In §5 we extend the above to the level of Z-algebras (see §3 and Remark 5.2 for the appropriate definitions) and prove the following: Theorem (Theorem 5.1). Let A be a cubic Sklyanin Z-algebra. Then there is a quadratic Sklyanin Z-algebra A1 and an inclusion A1 ãÑ Ap2q inducing an iso- morphism between their function fields. This inclusion is constructed with respect to a point p on Y such that if A " ApY,L0,L1,L2q then A1 " ApY,L0 b L1 b OY p´pq,L2 b L3 b OY p´τ´1pq. Remark 1.1. The existence of the function field a cubic Sklyanin Z-algebra (or more generally a "quadric") is proven in Appendix A. In §5 we also provide noncommutative versions of the inverse birational trans- formation P2 (cid:57)(cid:57)(cid:75) P1 P1 as follows: Theorem (Theorem 5.4). Let A be a quadratic Sklyanin Z-algebra, then there is a cubic Sklyanin Z-algebra A1 and an inclusion A1 ãÑ A inducing an isomor- phism between their function fields. This inclusion is constructed with respect to points p, q on Y such that if A " ApY,L, ψq then A1 " ApY,L b OY p´pq, ψL b OY p´qq, ψ2L b OY p´ψ´3pqq. In §6 we show (Theorem 6.2) that, modulo some technical hypothesis, the non- commutative Cremona as in (3) factors through the noncommutative P2 (cid:57)(cid:57)(cid:75) P1P1 and P1 P1 (cid:57)(cid:57)(cid:75) P2. As such all of these are examples of "quadratic transforms", a more general type of noncommutative birational transformations introduced in §6. In the last sections we show that quadratic transforms are invertible in the following sense (for simplicity we omit some technical hypotheses): Theorem (Theorem 7.4). Let γ : A1 ãÑ Apwq be a quadratic transform. Then there exists a quadratic transform δ : A ãÑ A1pvq such that the compositions γ δ : A ãÑ Apvwq and δ γ : A1 ãÑ A1pvwq induce the identity map on the function fields. Remark 1.2. In the above theorem the possible values for w are determined by the definition of a quadratic transform. Moreover v can be chosen as a function of w w P and v In §7 we show that it suffices to prove this theorem in case γ is as in Theorem 5.1 or Theorem 5.4. Both of these cases are covered in §8. The proof is quite technical and uses a Z2-algebra which in a certain sense glues A and A1. On the other hand 1 2 , 1, 2 (cid:32) ( . à R " Rm,n pm,nqPZ2 4 DENNIS PRESOTTO the geometric picture is rather simple. For example if A " ApY,L, ψq is a quadratic Sklyanin algebra and γ : A1 ãÑ A is constructed with respect to points p, q on Y as in Theorem 5.4, then the inverse δ : A ãÑ A1p2q is constructed with respect to p1 where p ` q ` p1 „ rψLs in the Picard group of Y . Conversely if A " ApY,pLiqiq is a cubic Sklyanin Z-algebra and γ : A1 ãÑ Ap2q is constructed with respect to a point p on Y as in Theorem 5.1, then the inverse δ : A ãÑ A1 is constructed with respect to p1, q1 where p ` τ q1 „ rL0s and p ` p1 „ rL1s in the Picard group of Y . 2. Acknowledgements The author wishes to thank Michel Van den Bergh for providing many interest- ing ideas and for reading through the results multiple times. The author is also very grateful to Susan J. Sierra for inviting him to the University of Edinburgh and for posing the question about the nature of the inverses to the birational transfor- mations considered in [10]. This question directly lead to the current paper. 3. Z-algebras In this section we recall some definitions and facts on Z-algebras. We refer the reader to [12] or sections 3 and 4 of [17] for a more thorough introduction. Recall that a Z-algebra is defined as an algebra R (without unit) with a decomposition such that addition is degree-wise and multiplication satisfies Rm,nRn,j Ă Rm,j and Rm,nRi,j " 0 if n ‰ i. Moreover there are local units en P Rn,n such that for each x P Rm,n : emx " x " xen. Notation . If A is a graded algebra, then it gives rise to a Z-algebra A via Am,n " An´m. In particular the notion of a Z-algebra is a generalization of a (Z)-graded alge- bra. Based on this, most graded notions have a natural Z-algebra counterpart. For example we say that a Z-algebra R is positively graded if Rm,n " 0 for m ą n. A graded R-module is an R-module M together with a decomposition M " 'nMn such that the R-action on M satisfies MmRm,n Ă Mn and MmRi,n " 0 if i ‰ m. The category of graded R-modules is denoted GrpRq and similar to the graded case we use the notation QGrpRq :" GrpRq{ TorspRq. Remark 3.1. If A is a graded algebra, then obviously GrpAq " Grp Aq by identifying Apnq P GrpAq with e´n A P Grp Aq. Similarly QGrpAq " QGrp Aq. Definition 3.2. Let R be a Z-algebra. Then for each n P Z we define Rpnq by setting pRpnqqi,j " Ri`n,j`n with obvious multiplication. We say R is n-periodic if there is a Z-algebra-isomorphism R -- Rpnq. Lemma 3.3. Let R be a Z-algebra. Then there exists a graded algebra A such that R " A if and only if R is 1-periodic. Proof. The "only if" part is obvious from the definition of A. The "if" part is (cid:3) proven in [17, Lemma 3.4]. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 5 Remark 3.4. From a categorical point of view a Z-algebra R is nothing but a k-linear category R whose objects are given by the integers. The homogeneous elements of the algebra then correspond to morphisms between two such integers via HomRp´j,´iq " Ri,j and multiplication in R corresponds to composition of morphisms in R. 3.1. AS-regular Z-algebras. Definition 3.5. Let R be a Z-algebra, then R is said to be connected, if it is positively graded, dimkpRm,nq ă 8 for each m, n and Rm,m -- k for all m. We say R is generated in degree 1 if Rm,m`1Rm`1,n " Rm,n holds for all m ă n. If R is a connected Z-algebra, generated in degree 1, then we denote Sn,R " enR{penRqěn`1. I.e. Sn,R is the unique R-module concentrated in degree n where it is equal to the base field k. We can now give the definition of an AS-regular Z-algebra as in [17] Definition 3.6. A Z-algebra R over k is said to be AS-regular if the following conditions are satisfied: (1) R is connected and generated in degree 1 (2) dimkpRm,nq is bounded by a polynomial in n ´ m (3) The projective dimension of Sn,R is finite and bounded by a number inde- (4) @n P N : " 1 (the "Gorenstein condition") ¯ GrpRqpSj,R, enRq pendent of n ÿ dimk Exti ´ i,j It is immediate that if a graded algebra A is AS-regular, then A is AS-regular in the above sense. Z-algebra analogues of three dimensional quadratic and cubic AS-regular alge- bras were classified in [17], it is shown in loc.sit. that every quadratic AS-regular Z-algebra is of the form A for some quadratic AS-regular algebra A. However most cubic AS-regular Z-algebras are not 1-periodic. To distinguish cubic AS-regular algebras from the more general cubic AS-regular Z-algebras, one often refers to the latter as quadrics. Similar to the graded case, the classification of three-dimensional quadratic and cubic AS-regular Z-algebras in terms of geometric data pY,pLiqiPZq where Y is a k-scheme and pLiqiPZ is an elliptic helix of line bundles on Y (see [6] for more information on helices). Starting from the geometric data one first constructs a Z-algebra analogue B " BpY,pLiqiq of the twisted homogeneous coordinate ring: # ΓpY,Li b Li`1 b . . .Lj´1q 0 Bi,j :" if i ď j if i ą j à Again there is an equivalence of categories. QcohpY q -- QGrpBq : F ÞÑ ΓpY,F b L0 b L1 b . . .Li´1q iě0 (4) (see [17, Corollary 5.5.9]). r " dimkpBi,i`1q does not depend on i and equals 3 (in the quadratic case) or 2 (in the cubic case). ApY,pLiqiq is then obtained from B by only preserving the relations in degree pi, i ` sq for s " 5 ´ r. It is shown that dimkpAi,i`nq does not depend on i; it hence makes sense to write hpnq :" dimkpAi,i`nq and one checks that hpnq coincides with the Hilbert series in the graded case (even thought AS-regular Z-algebras need not be 1-periodic). 6 DENNIS PRESOTTO 3.2. Z-domains and Z-fields of fractions. In this section we give the natural generalizations of "domain" and "field of fractions" for Z-algebras. Among other generalizations, these notions can also be found in [7, §2]. Definition 3.7. Let R be a Z-algebra. Then we say that R is a Z-domain if the following condition is satisfied: @i, j, k P Z,@x P Ri,j,@y P Rj,k : xy " 0 ñ x " 0 _ y " 0 It is known that three dimensional AS-regular algebras are domains ([3, Theorem 3.9] and [3, Theorem 8.1]). We extend this result to the level of quadrics and show: Theorem 3.8. Let A " ApY,pLiqiq be a quadric. Then A is a Z-domain. Proof. The proof of this theorem is postponed to Appendix A. (cid:3) Let R be a Z-algebra and R the associated category as in Remark 3.4. Let W be a collection of homogeneous elements and let W be the corresponding collection of morphisms. We then say that R is localizable at W if pR,Wq admits a calculus of fractions (see for example [8]). In this case we define RrW ´1s as the Z-algebra associated to RrW´1s. Using the theory of (right) fractions for a category one easily checks that the following definition makes sense. Definition 3.9. Let R be a Z-domain, then R admits a Z-field of (right) fractions if the following condition is satisfied: @r P Rl,i, s P Rl,jzt0u : Dn P Z : Dr1 P Rj,n, s1 P Ri,nzt0u : rs1 " sr1 The elements of FracpRqi,j are equivalence classes of couples pr, sq where r P Ri,l, s P Rj,lzt0u for some l P Z. The equivalence relation is given by @l1, l2 P Z,@r1 P Ri,l1 , r2 P Ri,l2, s1 P Rj,l1zt0u, s2 P Rj,l2zt0u : pr1, s1q „ pr2, s2q õ Dl3 P Z,Dx P Rl1,l3zt0u,Dy P Rl2,l3zt0u : r1x " r2y and s1x " s2y The following is obvious from the definition: Proposition 3.10. Let A be a graded domain. Then A admits a (graded) field of (right) fractions ô A admits a Z-field of (right) fractions Moreover in this case - F racpAq " F racp Aq Recall that graded domains admit fields of (right-)fractions when they are graded (right-)noetherian or when they have subexponential growth. A similar result can be found in [7]: Proposition 3.11. Let R be a Z-domain such that each eiR is a uniform module (i.e. for all nonzero M, N Ă eiR : M X N ‰ 0). Then R admits a Z-field of (right) fractions. Proof. We need to show that for all r P Rl,i, s P Rl,jzt0u there exists an n P Z and elements r1 P Rj,n, s1 P Ri,nzt0u such that rs1 " sr1. If r " 0 then it suffices to take r1 " 0. If r ‰ 0 the existence of r1 and s1 follows from the fact that rR and sR are (cid:3) nonzero submodules of the uniform module elR. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 7 Similarly the following will be shown in Appendix A: Theorem 3.12. Let A be a quadric. Then A admits a Z-field of fractions. Remark 3.13. As three dimensional, quadratic AS-regular Z-algebras are 1-periodic, the existence of their Z-field of fractions is automatic from Proposition 3.10. Remark 3.14. When Q " FracpAq it is customary to refer to Q0,0 as the function field of QGrpAq. We will do so throughout this paper. 4. Summary of the results in [10] In [10] we construct noncommutative versions of the birational transformation P1 P1 (cid:57)(cid:57)(cid:75) P2 and the Cremona Transform P2 (cid:57)(cid:57)(cid:75) P2 using the following recipe: Step 1) Let A be a 3-dimensional Sklyanin algebra. Let pY,L, σq be the associated geometric data, B be the associated twisted homogeneous coordinate ring and QcohpXq " QGrpAq. Denote BimodpY ´ Y q ãÑ BimodpX ´ Xq for the categories of bimodules (recall: BimodpX1 ´ X2q is the category of right exact functors QcohpX1q Ñ QcohpX2q commuting with direct lim- its) with oY , oX corresponding to the identity functors. Let oXp´Y q " kerpoX (cid:16) oY q1. Step 2) Choose a divisor d on Y in the following way: ‚ d " p`q`r for p, q, r distinct non-collinear points in case A is quadratic ‚ d " p in case A is cubic Let Od be it's structure sheaf and denote od P BimodpY ´Y q Ă BimodpX´ Xq for the bimodule od :" p´ bY Odq and (5) md :" kerpoX (cid:16) odq md,Y :" kerpoY (cid:16) odq Step 3) Define Z-algebras DY and D as follows pDY qm,n " OY p´2nq,OY p´2mq boY mτ´md,Y . . . mτ´n`1d,Y if n ě m if n ă m ` Hom 0 # # (6) (7) Dm,n " HompOXp´2nq,OXp´2mq boX mτ´md . . . mτ´n`1dq 0 if n ě m if n ă m where τ " σs`1, OXpiq " πpApiqq, OY piq " πpBpiqq and π is the quotient functor GrpAq Ñ QGrpAq. By [16, Lemma 8.2.1] the inclusions mτ´id ãÑ oX give rise to an inclusion D ãÑ Ap2q. 1Unfortunately BimodpX1´X2q appears not to be an abelian category.This technical difficulty is solved in [16] by embedding BimodpX ´ Y q into a larger abelian category BIMODpX1 ´ X2q consisting of "weak bimodules". In particular we will always construct kernels and cokernels as elements in BIMODpX1´X2q. This being said, these technical complication will be invisible in this paper as all bimodules we construct will be elements of the (smaller) category BimodpX1 ´ X2q. Step 4) One shows that DY is a twisted homogeneous coordinate ring, i.e. there is a sequence of line bundles pGiqiPZ such that Moreover the Gi must form an "elliptic helix" (for a quadratic AS regular Z-algebra), i.e. ‚ deg Gi " 3 for all i ‚ G0 fl G1 ‚ Gi b Gb´2 i`1 b Gi`2 " OY 8 DENNIS PRESOTTO (8) pDY qm,n " ΓpY,Gm b . . . b Gn´1q Step 5) One shows that D is generated in degree 1. For this we need some natural vanishing results on OXpaq b mτ´md . . . mτ´n`1d (We will recapitulate and slightly extend these vanishing results in Lemmas 5.8 and 5.10.) Step 6) One shows that the canonical map D Ñ DY is surjective and that D has the correct Hilbert series. It then follows that D is the AS-regular Z-algebra corresponding to the elliptic helix pGiqiPZ on Y . As D is quadratic, there is an AS-regular quadratic algebra A1 such that D "A1 (see [17, Theorem 4.2.2]). Step 7) Check that the induced morphism Frac0pA1q " Frac0,0pDq Ñ Frac0,0p Aq " Frac0pAq is an isomorphism. This check is immediate as injectivity comes for free and surjectivity follows by a Hilbert series argument. 5. Addendum to [10] As explained in the previous section, in [10] we construct noncommutative ver- sions of the birational transformation P1 P1 (cid:57)(cid:57)(cid:75) P2. In this section we slightly generalize this (Theorem 5.1) and in addition we discuss a noncommutative version of the inverse birational transformation P2 (cid:57)(cid:57)(cid:75) P1 P1 (Theorem 5.4). Theorem 5.1. Let A " ApY,pLiqiPZq be a quadric such that ‚ Y is a smooth elliptic curve ‚ L2 -- αL0 for some α P AutpY q which is given by translation by a point then there exists a quadratic Sklanin algebra A1 and an inclusionA1 ãÑ Ap2q induc- of order at least 3 (i.e. α2 ‰ Id). ing an isomorphism of function fields. Remark 5.2. We refer to a quadric as in Theorem 5.1 as a cubic Sklyanin Z-algebra. Remark 5.3. Theorem 5.1 extends [10] in the sense that it does not require A to be 1-periodic. Theorem 5.4. Let A " ApY, ψ,Lq be a quadratic Sklyanin algebra. Then there exists a cubic Sklyanin Z-algebra D and an inclusion D ãÑ A inducing an isomor- phism of function fields. The construction of D depends on two points, p, q P Y that σ2 " ψ´3. and D is 1-periodic, i.e. of the form D " A1, if q " σp for σ a translation such As both Theorem 5.1 and 5.4 contain statements on AS-regular Z-algebras which need not be 1-periodic we introduce some notation for an AS-regular Z-algebra A: Inspired by p7q and Remark 3.1 we define (9) OXpiq " πpe´iAq, OY piq " πpe´iBq SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 9 Under the equivalence of categories QGrpBq -- QcohpY q (see (4)) OY piq can be identified with $'&'%L0L1 . . .Li´1 (10) OY pL0L1 . . .L´i´1q´1 Similar to the graded case [10, §4] we have if i ą 0 if i " 0 if i ă 0 HompOXp´jq,OXp´iqq " HomQGrpAqpπpejAq, πpeiAqq nÑ8 HomGrpAqpejAěn, eiAq " lim " HomGrpAqpejA, eiAq " Ai,j where the second equality follows from the fact that ejA and eiA are finitely gen- erated A-modules (see [5, Proposition 7.2] for the graded case). The third equality follows from the Gorenstein condition as this implies HomGrpAqpejA{ejAěn, eiAq " Ext1 GrpAqpejA{ejAěn, eiAq " 0. We also have the following standard vanishing result Lemma 5.5. Let A be an AS-regular Z-algebra of dimension 3 with Hilbert function h and let X " QGrpAq and OXpiq be as above, then we have ` dimk ExtnpOXpiq,OXpjqq " 0 for n ‰ 0, 2 Ext2pOXpiq,OXpjqq " 0 hpi ´ s ´ jq # if j ě i ´ s if j ď i ´ s ´ 1 (cid:3) Proof. Similar to [5, Theorem 8.1]. Remark 5.6. Throughout the rest of the paper X and OXpiq will be defined as in Lemma 5.5. We now give an overview of both proofs separately. The recipe is as in §4 and we only highlight the steps that need to be adapted. 5.1. Noncommutative P1 P1 (cid:57)(cid:57)(cid:75) P2. (Proof of Theorem 5.1). Step 1) Let A " ApY,pLiqiPZq be a cubic Sklyanin Z-algebra as in the statement of Steps 2) and 3) Let p be a point on Y and set τ " α2 ‰ Id and d " p in the definition of Step 4) Computations similar to the ones in [10, §5] show that DY is a twisted D and DY as in (6). Theorem 5.1. homogeneous coordinate ring, i.e. DY,m,n " ΓpY,Gm b . . . b Gn´1q (11) for some new collection of line bundles on Y : Gi " OY p´τ´ipq b L2i b L2i`1 It follows easily that these line bundles satisfy (a) deg Gi " 3 (b) G0 fl G1. (c) Gi b Gb´2 (d) ψpGiq -- Gi`1 where ψ is an arbitrary translation satisfying ψ3 " τ . i`1 b Gi`2 -- OY . 10 DENNIS PRESOTTO Steps 5), 6) and 7) We need the following generalizations of [10, Lemma 6.2, Lemma 6.4 and (12) (13) Lemma 6.5]: Lemma 5.7. Let A be as above and let p be a point in Y . Let i P Z and let Pi be the point module generated in degree i corresponding to the point p (i.e. the point module truncated at degree i). Then there is a complex of the following form: 0 Ñ ei`5A where ζ is part of the minimal resolution of Si`1 as given in [16, Definition 4.1.1.] pζ,0qÝÝÝÑ ei`4A'2 ' ei`3A Ñ ei`2A'3 Ñ eiA Ñ Pi Ñ 0 0 Ñ ei`5A ζÝÑ ei`4A'2 εÝÑ ei`2A'2 δ0ÝÑ ei`1A γÝÑ Si`1 Ñ 0 Moreover the complex (12) is exact everywhere except at eiA where it has one-dimensional cohomology, concentrated in degree i ` 1. Proof. Similar to the proof of [10, Lemma 6.2] this is based on fact that the minimal projective resolution for Pi is given by 0 Ñ ei`3A Ñ ei`1A ' ei`2A Ñ eiA Ñ Pi Ñ 0 The latter is a Z-algebra analogue of the projective resolution of a point module as in [3, Proposition 6.7.i]. The proof in loc. cit. can be adapted to a Z-algebra version as follows: First we note that by [16, §5] there is a 1-1-correspondence between point modules (truncated in degree i) and points on Y , so the correspondence between p P Y and Pi is well defined. Next we generalize line modules to modules of the form ejA{aA where a P Aj,j`1. As A is a Z-domain (Theorem 3.8) we know that line modules have the desired Hilbert series and have projective dimension 1. The char- acterization of line modules as in [3, Corollary 2.43] generalizes to this new definition of line modules if we replace the modules Ap´iq in the graded case by eiA. Finally we find p13q by generalizing the proof of [3, Proposition 6.7.i]. This proof uses the fact that point modules in the graded case have projective dimension 2. A careful observation however shows that projective dimen- sion ď 2 suffices for the proof. This in turn can be shown by a variation of (cid:3) [3, Proposition 2.46.i]. In particular [10, Lemma 6.4, Lemma 6.5] generalize to the following Lemma 5.8. With the notations as above we have for i ´ j ď 3: Ext2pOXpiq,OXpjq b mτ´md . . . mτ´n`1dq " 0 and for i ´ j ď 2m ´ 2n ` 2: Ext1pOXpiq,OXpjq b mτ´md . . . mτ´n`1dq " 0 Proof. Similar to [10, Lemma 6.4, Lemma 6.5] by using Lemmas 5.7 and (cid:3) 5.5. One can now copy the proofs in [10, §6] to find that D and DY are generated in degree 1, that the canonical map D Ñ DY is surjective and an isomorphism in degree 0,1 and 2, that D has the correct Hilbert series SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. and hence D -- ApY,pGiqiPZq "A1 for the graded algebra A1 " A1pY,G0, ψq. Similarly the results in [10, §7] extend to an isomorphism of function fields Frac0,0pAq -- Frac0pA1q. 11 5.2. Noncommutative P2 (cid:57)(cid:57)(cid:75) P1 P1 (Proof of Theorem 5.4). Step 1) Throughout this section A " ApY,L, ψq is a quadratic Sklyanin algebra, hence Y is a smooth elliptic curve, L is a degree 3 line bundle on Y and ψ P AutpY q is given by translation by a point with τ " ψ3 ‰ Id. Steps 2) and 3) The construction of D is slightly different from the one in (6): Let p, q be points on Y and denote # di " τ´jp if i " 2j τ´jq if i " 2j ` 1 (14) (15) (16) (17) where τ " ψ3. We define mdi,Y and mdi as in p5q and introduce Z-algebra DY and D via pDY qm,n " HompOY p´nq,OY p´mq boY mdm,Y . . . mdn´1,Y q if n ě m if n ă m # # 0 HompOXp´nq,OXp´mq boX mdm . . . mdn´1q Dm,n " Again using [16, Lemma 8.2.1], there is an inclusion D ãÑ A. 0 if n ě m if n ă m Step 4) Computations similar to the ones in [10, §5] show DY,m,n " ΓpY,Gm b . . . b Gn´1q for some collection of line bundles on Y given by: Gi " OY p´diq b ψiL and these line bundles satisfy (a) deg Gi " 2 (b) G0 fl G2. (c) Gi b G´1 (d) αpGiq -- Gi`2 where α is an arbitrary translation satisfying α2 " τ . Moreover if q " α´1p then Gi`1 -- σGi for an arbitrary translation σ : Y Ñ Y such that σ2 " α. Steps 5) and 6) This is completely analogous to [10, §6]. The only difference lies in the resolution of Op´iq b mdn : this time it is of the form: i`2 b Gi`3 -- OY . i`1 b G´1 0 Ñ Op´i ´ 3q Ñ Op´i ´ 2q'2 Ñ Op´iq b mdn Ñ 0 This is based on the following: Lemma 5.9. Let A " ApY,L, ψq be a quadratic AS-regular algebra of dimension 3, let p P Y be a point and let P be the corresponding point module. Then the minimal resolution of P has the following form 0 Ñ Ap´2q Ñ Ap´1q'2 Ñ A Ñ P Ñ 0 Proof. This follows immediately from [3, Proposition 6.7]. (cid:3) From this we find the following standard vanishing results: 12 DENNIS PRESOTTO Lemma 5.10. With the notations as above we have for i ´ j ď 2: Ext2pOXpiq,OXpjq b mmd . . . mdn´1q " 0 and for i ´ j ď m ´ n ` 1: Ext1pOXpiq,OXpjq b mdm . . . mdn´1q " 0 Proof. Similar to [10, Lemma 6.4, Lemma 6.5] by using Lemmas 5.9 and (cid:3) 5.5. For checking that D has the correct Hilbert series we see that (using com- putation similar to [16, Corollary 5.2.4]) the colength of mdm . . . mdm`i´1 inside oX is (18) if i " 2a if i " 2a ` 1 " pa ` 1q2 pa ` 2qpa ` 1q 2 " apa ` 1q ` apa ` 1q $'&'%2 apa ` 1q $'&'%p2a ` 1qp2a ` 2q p2a ` 2qp2a ` 3q 2 2 Using Lemma 5.10 this implies dim Dm,m`i " 2 2 ´ apa ` 1q " pa ` 1q2 ´ pa ` 1q2 " pa ` 1qpa ` 2q if i " 2a if i " 2a ` 1 i.e. D has the Hilbert series of a cubic AS-regular Z ´ algebra and hence we can conclude that D is AS-regular. Step 7) Injectivity of Frac0,0pDq Ñ Frac0pAq is again immediate. For surjectivity we need to prove that for any fixed n P N there is some N P N such that HompOXp´i ´ n ´ Nq,OXp´i ´ nq b mdi . . . mdi`n`N´1q ‰ 0 holds for all i P Z. Now note that the colength of mdi . . . mdi`n`N´1 inside oX is given by a2 or apa` 1q if n` N " 2a or 2a` 1 respectively. In particular this colenth for N large. Hence the codimension of HompOXp´i ´ n ´ grows like N 2 Nq,OXp´i´ nqb mdi . . . mdi`n`N´1q inside Ai`n,i`n`N grows at most like 4 it hence follows that ΓpOXp´i ´ 4 . As dimk Ai`n,i`n`N grows like N 2 N 2 n ´ Nq,OXp´i ´ nq b mdi . . . mdi`n`N´1q ‰ 0 for large N . 2 SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 13 6. Cremona Transformations and quadratic transforms Convention 6.1. In the following sections we will always identify graded algebras with their associated Z-algebras. The Z-algebra D which was introduced in sections 4 and 5 will now be denoted by A1. Moreover we will use the terminology "(three dimensional) Sklyanin Z-algebras" for both quadratic and cubic Sklyanin Z-algebras. The goal of this section is to link the noncommutative version of the Cremona transform P2 (cid:57)(cid:57)(cid:75) P2 as constructed in [10] (see also §4 for a reminder of this construction) to the noncommutative versions of P1P1 (cid:57)(cid:57)(cid:75) P2 and P2 (cid:57)(cid:57)(cid:75) P1P1 as in Theorems 5.1, 5.4. This is inspired by the following commutative picture: Figure 1. The commutative Cremona transform i.e. the (commutative) Cremona transform γ : P2 (cid:57)(cid:57)(cid:75) P2 (given by blowing up three non-collinear points p1, q1, r1, see for example [9, Example V.4.2.3.]) factors as γ " γ1 γ2 where γ1 : P1 P1 (cid:57)(cid:57)(cid:75) P2 and γ2 : P2 (cid:57)(cid:57)(cid:75) P1 P1 are the standard birational transformations obtained by blowing up one point in P1 P1 and two points in P2. We prove that (modulo some technical assumptions) the same holds in the non- commutative setting: 14 DENNIS PRESOTTO Theorem 6.2. Let A " ApY,L, ψq and A1 be quadratic Sklyanin algebras and let γ : A1 ãÑ Ap2q be a noncommutative Cremona transformation as in [10]. Denote d " p` q ` r as in loc. cit. and assume p, q and r are non-collinear points lying in different τ -orbits where τ " ψ3. Then there exists a cubic Sklyanin Z-algebra A2 and inclusions γ1 : A1 ãÑ A2p2q, γ2 : A2 ãÑ A such that γ1 is as in Theorem 5.1, γ2 is as in Theorem 5.4 and γ " γ2 γ1. Before we can prove this we need the following technical result: Lemma 6.3. Let p and q be points such that p ‰ τ iq for i P t´1, 0, 1u. Then mp`q " mpmq. Proof. Consider the following diagram: mp b oq a1 g1 mp f1 a2 oq g2 oX f2 b2 mp b mq mq a3 b3 op f3 op b mq The middle row and column are the exact sequences as in (5). The last row is obtained by applying ´ b mq to the middle row, it hence is automatically right exact. Similarly the first column is right exact. We can identify mpmq both with the image of f1 and a3 (and hence the kernel of g1 and b3). To see this recall that mpmq is defined as the image of f : mp b mq Ñ oX b oX Ñ oX The identification then follows as f " f2 a3 " a2 f1 and as a2 and f2 are monomorphisms. Next we claim that a1 and f3 are in fact isomorphisms. The proof follows from this claim as it implies mpmq serves both as a kernel for b2 f2 and g2 a2. As such it is the pullback of mp a2 oX f2 mq On the other hand op`q " op'oq such that mp`q, being the kernel of oX also is a pullback of the above diagram. In particular mp`q -- mpmq. b2'g2ÝÝÝÝÑ op`q, # where di " τ´jp if i " 2j τ´jq if i " 2j ` 1 SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 15 It hence remains to prove the above claim. As the argument is the same for both morphisms we only explain this for a1. Note that this map is obtained by tensoring mp ãÑ oX with oq. As such the cokernel of a1 is given by opb oq whereas the kernel is a subobject of T or1pop, oqq (see [16, §3] for the definition of T ori). In particular it suffices to show op b oq " 0 " T or1pop, oqq. By [16, Lemma 5.5.1] we need to show HomXpOp,Oqq " 0 " Ext1 XpOp,Oqq. For HomX this follows from the fact that Op and Oq are simple. For Ext1, we can use [16, Proposition 5.1.2 and (5.3)] Y pOp,Oqq " 0 " HomY pOp,Oτ qq. The to reduce the computations to showing Ext1 latter follows p, q and τ q are different points by assumption. (cid:3) We can now finish the main result of this section: Proof of Theorem 6.2. Recall from §4 that γ : A1 ãÑ Ap2q was constructed as fol- lows: pA1qm,n " HomXpOXp´2nq,OXp´2mq boX mτ´md . . . mτ´n`1dq (19) We construct γ2 : A2 ãÑ A with respect to the points p, q as in Theorem 5.4. I.e. A2m,2n " HomXpOXp´2nq,OXp´2mqq pA2qm,n " HomXpOXp´nq,OXp´mq boX mdm . . . mdn´1q ãÑ By Lemma 6.3 for each i we can write md2imd2i`1 " mτ´ipp`qq. In particular the inclusions mτ´ipp`q`rq ãÑ mτ´ipp`qq give rise to an inclusion γ1 : A1 ãÑ A2p2q such that γ " γ2 γ1. It hence remains to show that γ1 is in fact an inclusion as in Theorem 5.1. For this we need to prove the existence of a point p1 such that τ´n`1p1q (20) where m1 (20) for n " m ` 1 in which case the left hand side of (20) equals p1 " kerpoX1 Ñ op1q. As A1 is generated in degree 1, it suffices to check γ1ppA1qm,nq " HomX1pOX1p´2nq,OX1p´2mq boX1 m1 τ´mp1 . . . m1 ΓpY, ψ2mL b ψ2m`1L b OY p´τ´mp ´ τ´mq ´ τ´mrqq and the right hand side equals ΓpY, ψ2mL b OY pτ´mpq b ψ2m`1L b OY p´τ´mqq b OY pτ´mp1qq . Hence the theorem is proven by choosing p1 " r. (cid:3) Inspired by Theorem 6.2 we make the following definition: Definition 6.4. Let A, A1 be three dimensional Sklyanin algebras. An inclusion A1 ãÑ Apvq is called a quadratic transform if it can be written as a composition of inclusions as in Theorems 5.1 and 5.4. Remark 6.5. By construction a quadratic transform always induces an isomorphism of function fields. 16 DENNIS PRESOTTO It immediately follows from the definition that if A1 ãÑ Apvq is a quadratic trans- form, then v " 2n for some nonnegative integer n. By construction our noncom- mutative versions of P1 P1 (cid:57)(cid:57)(cid:75) P2 and P2 (cid:57)(cid:57)(cid:75) P1 P1 as in Theorems 5.1, 5.4 are quadratic transforms. Theorem 6.2 implies the noncommutative Cremona trans- form as in §4 is a quadratic transform as well. Similar to the construction in Figure 1 one can introduce a (commutative) cubic Cremona transform P1P1 (cid:57)(cid:57)(cid:75) P1P1 by blowing up two points in each copy of P1P1, provided that these points do not lie on one ruling, see for example Figure 2. As this cubic Cremona factors through P2 using the classical birational transformation, there is an obvious noncommu- tative generalization A1 ãÑ Ap2q where A and A1 are cubic Sklyanin Z-algebras. By definition this noncommutative cubic Cremona transformation is a quadratic transform as well. Figure 2. The commutative cubic Cremona transform 7. Inner morphisms The main goal of the following sections is to prove that in a suitable sense quadratic transforms are invertible: Definition 7.1. Let A be a Z-algebra such that Q :" FracpAq exists. We say that an injective morphism of Z-algebras φ : A ãÑ Apvq is inner if there exist SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 17 zm P Qvm,m ´ t0u such that for a P Am,n we have φpaq " zmaz´1 require n . Moreover we (21) zm P Avm,m z0 " 1 m P Am,vm z´1 The following is clear if m ă 0 if m ą 0 Proposition 7.2. If A Ñ Apvq is inner then the induced map Q0,0 Ñ pQpvqq0,0 " Q0,0 is the identity. Definition 7.3. Let A and A1 be Z-algebras such that FracpAq and FracpA1q exist. An inclusion γ : A1 ãÑ Apwq is said to be invertible if there exists an inclusion δ : A ãÑ A1pvq such that δ γ : A1 ãÑ A1pvwq and γ δ : A ãÑ Apvwq are inner. We can now state the main result of this paper: Theorem 7.4. Assume that γ : A1 ãÑ Ap2mq is a quadratic transform between (three dimensional) Sklyanin Z-algebras. Write γ " γ1 . . . γt with the γj as p2q in Theorems 5.1, 5.4. Moreover we assume that for each factor γi : Ai ãÑ A i`1 (with Ai`1 necessarily cubic) the points pi, qi used in the construction of γi lie in different τ -orbits. Then γ is invertible and the "inverse" δ can be chosen as a quadratic transform δ : A Ñ A1p2nq with n ´ m ď 1. We will call δ as in the previous theorem an inverse quadratic transform to γ. The following reduces the amount of work for proving Theorem 7.4 dramatically. ãÑ Apw1q and γ2 : A2 ãÑ A1pw2q be invertible as in Lemma 7.5. Let γ1 : A1 definition 7.3. Then γ1 γ2 is invertible as well. Proof. By assumption there exist inclusion δ1 : A ãÑ A1pv1q and δ2 : A1 ãÑ A2pv2q such that δ1γ1, δ2γ2, γ1δ1 and γ2δ2 are inner. We now claim pδ2δ1qpγ1γ2q and pγ1 γ2q pδ2 δ1q are inner as well. As both proofs are analogous, we only prove the latter. Let z1,m P Frac Aw1v1m,m ´ t0u and z2,m P Frac A1 w2v2m,m ´ t0u be as in Definition 7.1. Then for each a P Am,n: ppγ1 γ2q pδ2 δ1qqpaq " γ1 ppγ2 δ2qpδ1pxqqq z2,v1mδ1pxqz´1 ` 2,v1n " γ1 " γ1pz2,v1mqpγ1 δ1qpxqγ1pz2,v1nq´1 " γ1pz2,v1mqpz1,mxz´1 1,nqγ1pz2,v1nq´1 " pγ1pz2,v1mqz1,mq xpγ1pz2,v1nqz1,nq´1 Moreover obviously γ1pz2,v2mqz1,m P FracpAqw1v1w2v2m,mzt0u γ1pz2,v2mqz1,m P A if m ă 0 γ1pz2,0qqz1,0 " 1 pγ1pz2,v2mqz1,mq´1 P A if m ą 0 (cid:3) The proof of Theorem 7.4 then follows from the following 2 theorems which are proven in the next section: 18 DENNIS PRESOTTO Theorem 7.6. Let γ : A1 ãÑ A be a quadratic transform as in Theorem 5.4. Then γ is invertible and the inverse can be chosen as δ : A ãÑ A1p2q as in Theorem 5.1. Theorem 7.7. Let γ : A1 ãÑ Ap2q be a quadratic transform as in Theorem 5.1 and assume that the points p, q used in the construction of γ lie in different τ -orbits. Then γ is invertible and the inverse can be chosen as δ : A ãÑ A1 as in Theorem 5.4. Remark 7.8. Our approach to proving Theorems 7.6 and 7.7 is as follows: we first construct an inclusion of Z-algebras δ : A ãÑ A 1pvq such that the composition γ δ : A Ñ Apwvq is inner. Afterwards we will show that δ is in fact a quadratic transform and that δ γ is inner as well. 8. Inverting quadratic transforms between quadratic Sklyanin algebras and cubic Sklyanin Z-algebras In this section we finish the proof of Theorem 7.4 by proving Theorem 7.6 and Theorem 7.7. The proofs of these theorems are intertwined: ãÑ A is as in Theorem 5.4, then there is a In §8.1 we prove that if γ : A1 quadratic transform δ : A ãÑ A1p2q such that γ δ is inner. In §8.2 we prove that if γ1 : A1 ãÑ Ap2q is as in Theorem 5.1, then there is a quadratic transform δ1 : A ãÑ A1 such that γ1 δ1 is inner. Finally in §8.3 we prove that these constructions are each if we were to construct δ out of γ as in §8.1, set γ1 " δ and others inverses. I.e. compute δ1 as in §8.2, then δ1 " γ. This allows us to conclude that not only γ δ, but also δ γ is inner (as it is equal to γ1 δ1). The analogous results are true if we were to start from γ1. 8.1. The Z2-algebra associated to a noncommutative P2 (cid:57)(cid:57)(cid:75) P1P1. Through- out this subsection γ : A1 ãÑ A will be a quadratic transform between a quadratic Sklyanin algebra A " ApY,L, ψq and a cubic Sklyanin Z-algebra A1. Recall from §5.2 that the construction of γ is based on the choice of two points p, q P Y . We will use the notation from this section, in particular di is defined as in p14q. Moreover we define τ " ψ3, Li " σiL and assume p and q lie different τ -orbits. We "glue" the algebras A and A1 into a single Z2-algebra: # HomXpOXp´m ´ nq,OXp´i ´ jqmdj . . . mdn´1 HomXpOXp´m ´ nq,OXp´i ´ jqq if n ą j if n ď j Api,jq,pm,nq :" With X and OXpiq as in Lemma 5.5. Note that and # Ap0,jq,p0,nq " such that Api,0q,pm,0q " Ai,m HomXpOXp´nq,OXp´jqmdj mdn´1q 0 if n ě j if n ă j Ap0,jq,p0,nq " A1 j,n In other words A "contains" both A and A1. Remark 8.1. By construction Api,jq,pm,nq Ă Ai`j,m`n and as such A contains no nontrivial zero divisors. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 19 We now give some results on the dimensions of certain Api,jq,pm,nq. For this let hpnq be the Hilbert function of A and h1pnq be the Hilbert function of A1. I.e. (22) @i P Z : dimkpAi,i`nq " hpnq and dimkpA1 i,i`nq " h1pnq The following easy properties of A are immediate from the definition: Proposition 8.2. Let A be as above then (1) dimk Api,jq,pi,j`bq " h1pbq holds for all b, i, j P Z with b ě 0 (2) dimk Api,jq,pi`a,j`bq " hpa ` bq holds for all a, b, i, j P Z with a ě 0, b ď 0 More interesting is the following: Lemma 8.3. Let a ď 0, then: dimk Api,jq,pi`a,j`bq " dimk Api,jq,pi,j`b`2aq " h1pb ` 2 aq (23) Proof. The case a " ´1 can be done analogously to §5.2 Step 5) and 6). For a ď ´2 we can no longer use Lemma 5.10 and the proof is based on the existence of an "I-basis" (see [15] for the definition and construction of an I-basis). As we will only use the case a " ´1, we refer the interested reader to Appendix B for the details of the proof for a ď ´2. (cid:3) As a result of Lemma 8.3 both Api,jq,pi´1,j`2q and Api,jq,pi`1,j´1q are one dimen- sional. Let δi,j and γi,j be nonzero elements in these spaces. We can then visualise A on a 2-dimensional square grid. p3, 1q p3, 3q p3, 0q p3, 2q δ3,0 γ2,1 γ2,2 δ3,1 γ2,3 p2, 1q p2, 2q δ2,0 γ1,1 δ2,1 γ1,2 p1, 1q p1, 2q δ1,0 γ0,1 δ1,1 γ0,2 p0, 1q p0, 2q γ1,3 γ0,3 p2, 0q p1, 0q p0, 0q p2, 3q p1, 3q p0, 3q The vertical arrows represent three dimensional vector spaces, whereas the hor- izontal arrows represent two dimensional vector spaces and the dotted arrows rep- resent one dimensional vector spaces (labeled by γi,j and δi,j). Now consider the following diagram 20 pi, jq DENNIS PRESOTTO pi, j ` bq δi,j pi ´ 1, j ` 2q δi,j`b pi ´ 1, j ` b ` 2q From Lemma 8.3 we conclude that the vector spaces on the solid arrows all have the same dimension. Hence since A is a domain we have an isomorphism of vector spaces: i,j δi,j`b : Api,jq,pi,j`bq Ñ Api´1,j`2q,pi´1,j`b`2q δ´1 (24) Whenever 2i ` j " 2m ` n and i ě m we write (25) when i ă m we define (26) δpi,jq,pm,nq " δi,jδi´1,j`2 . . . δm`1,n´2 P Api,jq,pm,nq δpi,jq,pm,nq :" δ´1pm,nq,pi,jq P FracpAqi`2j,m`2n In particular we always have δpi,jq,pk,lqδpk,lq,pm,nq " δpi,jq,pm,nq From (24) we obtain an isomorphism δp0,2i`jq,pi,jq δpi,j`bq,p0,2i`j`bq : Api,jq,pi,j`bq Ñ Ap0,2i`jq,p0,2i`j`bq Now note that there is always an inclusion δpm,nq,pi,n`2m´2iq : Api,jq,pm,nq Ñ Api,jq,pi,n`2m´2iq (27) If m ě i this follows from the fact that A is a domain. Lemma 8.3 tells us the map is also well defined if m ă i, in which case case (27) even is an isomorphism. Summarizing we obtain an inclusion δp0,2i`jq,pi,jq δpm,nq,p0,2m`nq : Api,jq,pm,nq Ñ Ap0,2i`jq,p0,2m`nq And hence an inclusion δp0,2iq,pi,0q δpm,0q,p0,2mq : Ai,m " Api,0q,pm,0q Ñ Ap0,2iq,p0,2mq " A1 2i,2m One easily checks that these inclusions are compatible with multiplication on A and A1 such that we get an inclusion of algebras δ : A ãÑ A1p2q Our goal is to show that δ is in fact a quadratic transform as in Definition 6.4. Moreover we want γ δ to be inner (as stated in the introduction of this section, the proof to show that δ γ is inner is postponed to §8.3). We first prove the latter: Let γpi,jq,pm,nq be defined like δpi,jq,pm,nq but using γi,j instead of δi,j (and i` j " m ` n in stead of 2i ` j " 2m ` n; recall γi,j lies in the 1-dimensional space Api,jq,pi`1,j´1q). It is easy to see that the quadratic transform γ : A1 Ñ A we started with is given by γpj,0q,p0,jq γp0,nq,pn,0q : A1 j,n " Ap0,jq,p0,nq Ñ Ap2j,0q,p2n,0q " A2j,2n So the composition is given by (28) γp2i,0q,p0,2iqδp0,2iq,pi,0q δpm,0q,p0,2mqγp0,2mq,p2m,0q : Ai,m Ñ A2i,2m SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 21 This composition is inner with zi " γp2i,0q,p0,2iqδp0,2iq,pi,0q P FracpAq2i,i One easily checks that the elements zi indeed satisfy the conditions in (21). We now prove that δ is a quadratic transform. For this we need to show the existence of a point p1 P Y such that δpAi,i`1q " ΓpY,G2iG2i`1p´τ´ip1q " ΓpY,L2iL2i`1p´d2i ´ d2i`1 ´ τ´ip1qq We first define B like A but starting from B instead of from A. We find (using ΓpY,Li`jLi`j`1 . . .Lm`np´dj ´ . . . ´ dn´1qq ΓpY,Li`jLi`j`1 . . .Lm`nq if n ą j if n ď j Similar to [10, Lemma 6.7] one can show Lemma 8.4. The canonical map Api,jq,pm,nq Ñ Bpi,jq,pm,nq is an epimorphism in the first quadrant (i.e. m ě i, n ě j). Recall that by the definition of δ we have for each x P Ai,i`1 δi,0δi´1,2 δ1,2i´2δpxq " xδi`1,0δi,2 δ1,2i (30) when i ě 0 and (31) when i ă 0. Hence in order to prove the existence of p1 in (29) we have to understand (the product of) the image(s) δi,j of δi,j in δpxqδ0,2i`2δ´1,2i`4 . . . δi`2,´2 " δ0,2iδ´1,2i`2 . . . δi`1,´2x Bpi,jq,pi´1,j`2q " ΓpY,Li`jp´dj ´ dj`1qq (32) As Li`jp´dj ´ dj`1q has degree 1 on Y we can choose a point p1 (33) dj ` dj`1 ` p1 i,j „ rLi`js i,j defined by such that Bpi,jq,pi´1,j`2q " ΓpY,Li`jp´dj ´ dj`1qq " ΓpY,Li`jp´dj ´ dj`1 ´ p1 i,jqq (34) Lemma 8.5. Define p1 by the following identitiy p ` q ` τ p1 „ rL0s i,j " τ´i`1p1. Then p1 Proof. As ψ is a translation such that Li`1 -- ψLi, there is an invertible sheaf N of degree zero (see for example [17, Theorem 4.2.3]) such that (29) # (9) and (10)) Bpi,jq,pm,nq :" and rLns " rL0s ` 3nrNs τ´ip „ p ` 3irNs As p1 i,j is uniquely defined by p33q this proves the lemma. (cid:3) 22 DENNIS PRESOTTO In particular if p1 is as in the above lemma, then (34) gives rise to (35) Bpi,jq,pi´1,j`2q " ΓpY,Li`jp´dj ´ dj`1qq " ΓpY,Li`jp´dj ´ dj`1 ´ τ´i`1p1qq In particular δi,j is a non-zero section of Li`jp´dj ´ dj`1 ´ τ´i`1p1q. As the In particular, going back to (30) (and hence assuming i ě 0, the case i ă 0 being latter has degree zero on Y δi,j is everywhere non-zero on Y . completely similar) we see that δi,0δi´1,2 . . . δ1,2i´2 is an everywhere non-zero section of Li . . .L2i´1p´d0 ´ d1 ´ τ´i`1p1 ´ d2 ´ d3 ´ τ´i`2p1 ´ . . . ´ d2i´2 ´ d2i´1 ´ p1q Likewise x δi`1,0δi,2 . . . δ1,2i is a section of Li . . .L2i`1p´d0 ´ d1 ´ τ´ip1 ´ d2 ´ d3 ´ τ´i`1p1 ´ . . . ´ d2i ´ d2i`1 ´ p1q so that δpxq is a section of L2iL2i`1p´d2i´ d2i`1´ τ´ip1q. This is precisely what we had to show according to (29). 8.2. The Z2 algebra associated to a noncommutative P1P1 (cid:57)(cid:57)(cid:75) P2. Through- out this subsection γ : A1 ãÑ Ap2q will be a quadratic transform between a cubic Sklyanin Z-algebra A " ApY,pLiqiPZq and a quadratic Sklyanin algebra A1. Recall from §5.1 that the construction of γ is based on the choice of a points p P Y . We will use the notation from this section, in particular τ " α2. We define the Z2-algebra A as follows: # Api,jq,pm,nq :" if n ą j if n ď j As in the previous section the following easy properties of A are immediate from HomXpOXp´m ´ 2nq,OXp´i ´ 2jqmτ´j p . . . mτ´n`1p HomXpOXp´m ´ 2nq,OXp´i ´ 2jqq the definition.: Proposition 8.6. Let A be as above then j,n (1) Api,0q,pm,0q " Ai,m (2) Ap0,jq,p0,nq " A1 (3) A contains no nontrivial zero divisors (4) dimk Api,jq,pi,j`bq " h1pbq holds for all b, i, j P Z with b ě 0 and h1 the (5) dimk Api,jq,pi`a,j`bq " hpa ` 2bq holds for all a, b, i, j P Z with a ě 0, b ď 0 Where for p3q we used the fact that A is a Z-domain as in Theorem 3.8. We also have the following partial analogue of Lemma 8.3: Hilbert series of A1 and h the Hilbert series of A Lemma 8.7. dimk Api,jq,pi´1,j`bq " dimk Api,jq,pi,j`b´1q " h1pb ´ 1q Proof. The computation is completely similar to the a " ´1 case of Lemma 8.3 (cid:3) using Lemma 5.10 and [10, Lemma 6.5] SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 23 Remark 8.8. Although one cannot use an I-basis in the classical sense we expect dimk Api,jq,pi`a,j`bq " h1pb ` 2 aq to hold for all a ď 0. As a corollary of Lemma 8.7 and Proposition 8.6(3) we know Api,jq,pi´1,j`2q and Api,jq,pi`1,j´1q are one dimensional. Let δi,j and γi,j be nonzero elements in these spaces. We can then visualise A on a 2-dimensional square grid. p3, 0q p2, 0q p1, 0q p0, 0q p3, 1q p2, 1q δ3,1 γ1,2 p3, 2q p2, 2q δ3,2 γ1,3 δ3,0 γ1,1 δ2,0 δ2,1 δ2,2 γ0,1 δ1,0 p1, 1q p0, 1q γ0,2 δ1,1 p1, 2q p0, 2q γ0,3 δ1,2 p3, 3q p2, 3q p1, 3q p0, 3q All horizontal arrows represent three dimensional vector spaces whereas the vertical arrows represent two dimensional vector spaces and dotted arrows represent one dimensional vector spaces (labeled by γi,j and δi,j). Completely identical to previous sections there is an inclusion δp0,iq,pi,0q δpm,0q,p0,mq : Ai,m " Api,0q,pm,0q Ñ Ap0,iq,p0,mq " A1 i,m (where the elements δpi,jq,pm,nq are defined as in p25q and p26q. The only thing which essentially changed is that δpi,jq,pm,nq is now only defined when i` j " m` n in stead of i ` 2j " m ` 2n.) The induced inclusion of algebras is such that the composition γ δ is inner with δ : A ãÑ A1 zi " γpi,0q,p0,2iqδp0,iq,pi,0q # τ´lp1 τ´lq1 if i " 2l if i " 2l ` 1 Our next aim is to show that δ is a quadratic transform. For this we need to show the existence of two points p1, q1 P Y such that if we define d1 i as (36) i " d1 then we have (37) δpAi,i`1q " ΓpY,Gip´d1 iqq " ΓpY,L2iL2i`1p´τ´ip ´ d1 iqq We again start by defining a Z2-algebra B. This time it takes the following form: if n ą j Bpi,jq,pm,nq :" if n ď j ΓpY,Li`2jLi`2j`1 . . .Lm`2np´τ´jp ´ . . . ´ τ´n`1pqq ΓpY,Li`2jLi`2j`1 . . .Lm`2nq # 24 DENNIS PRESOTTO Similar to [10, Lemma 6.7] one can show Lemma 8.9. The canonical map Api,jq,pm,nq Ñ Bpi,jq,pm,nq is an epimorphism in the first quadrant (i.e. m ě i, n ě j). Recall that for each x P Ai,i`1, δpxq is related to x and elements δpi,jq,pm,nq via δi,0δi´1,1 δ1,i´1δpxq " xδi`1,0δi,1 δ1,i (38) when i ě 0 and δpxqδ0,i`1δ´1,i`2 . . . δi`2,´1 " δ0,iδ´1,i`1 . . . δi`1,´1x 2i,2i`2 " B1 (39) when i ă 0. Hence in order to understand δpxq in A1 2i,2i`2 we need to understand (the product of) the image(s) δi,j of δi,j in B. First remark that if we choose p1, q1 P Y such that (40) p ` τ q1 „ rL0s p ` p1 „ rL1s then similar to Lemma 8.5 we then have for all i, j: τ´jp ` d1 i´1 „ rLi`2js with d1 i as in (36). δi,j is a nonzero element of ΓpY,Li`2jp´τ´jp ´ d1 i´1qq As Li`2jp´τ´jp ´ d1 i´1q has degree zero on Y , δi,j is everywhere non-zero on Y . In particular, going back to (30) (and hence assuming i ě 0, the case i ă 0 being completely similar) we see that is an everywhere non-zero section of δi,0δi´1,1 . . . δ1,i´1 Li . . .L2i´1p´p ´ d1 i´1 ´ τ´1p ´ d1 i´2 ´ . . . ´ τ´i`1p ´ d1 0q Likewise is a section of x δi`1,0δi,1 . . . δ1,i Li . . .L2i`1p´p ´ d1 i ´ τ´1p ´ d1 i´1 ´ . . . ´ τ´ip ´ d1 0q so that δpxq is a section of L2iL2i`1p´τ´ip ´ d1 iq. This is precisely what we had to show according to (29). SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 25 8.3. Invertability of the quadratic transforms. We now show that if γ and δ are as in §8.1 or §8.2, then δ γ is inner. This boils down to computations on the geometric data associated to γ and δ. First assume A " ApY,L, ψq is quadratic and γ : A1 ãÑ A is constructed with respect to p, q P Y . Then according to Lemma 8.5 the quadratic transform δ : A ãÑ A1p2q is constructed with respect to a point p1 P Y satisfying p ` q ` τ p1 „ rL0s Using the techniques in §8.2 we find a quadratic transform γ : A1 Ñ A such that gamma δ is inner. By (40) we know γ is constructed with respect to points p2, q2 P Y satisfying p1 ` τ q2 „ rG0s p1 ` p2 „ rG1s with Gi as in (17). Moreover §8.1 constructs A out of an elliptic helix pL2 by iqiPZ given i " G2i b G2i`1 b OY p´τ´ipq L2 We need to check that p2 " p, q2 " q and L2 i -- Li. For this recall from [17, Theorem 4.2.3] that there exists a linebundle N of degree zero on Y such that for each linebundle M we have rψMs " rMs ` degpMq rNs. Using this we find: p1 ` τ q2 „ rG0s " rL0 b OY p´pqs p1 ` q2 ´ 3rNs ` p „ rL0s p ` q2 ` τ p1 „ p ` q ` τ p1 ó ó ó q2 " q Similarly p2 " p. Next we show that the elliptic helix pL2 iqiPZ coincides with pLiqiPZ: rL2 is " rG2is ` rG2i`1s ` rOY p´τ´ip1qs " rL2is ` rL2i`1s ` rOY p´d2i ´ d2i`1qs ` rOY p´τ´ip1qs " rL2is ` rL2i`1s ` rOY p´di ´ di`1qs ´ 3i rNs ` rOY p´τ´ip1qs " prL2is ´ 3i rNsq ` " rψiL2is " rLis rL2i`1s ` rOY p´di ´ di`1 ´ τ´ip1qs ` Next we do similar computations in case A " ApY,pLiqiPZ is a quadric and γ : A1 ãÑ Ap2q in constructed with respect to a point p P Y as in §8.1. Similar to the above it suffices to prove p2 " p and L2 i " Li where p1 ` q1 ` τ p2 „ rG0s i " Gi b OY p´d1 iq L2 i are as in (40) and (36). First we prove p2 " p, for where Gi is as in (11) and p1, q1, d1 this we take N a degree zero linebundle on Y such that rαMs " rMs`degpMqrNs with α as in Theorem 5.1 26 DENNIS PRESOTTO p1 ` q1 ` τ p2 „ rG0s " rL0 b L1 b OY p´pqs p1 ` q1 ` p2 ´ 2rNs ` p „ rL0s ` rL1s p1 ` τ q1 ` p2 ` p „ p ` τ q1 ` p ` p1 ó ó ó We now that the elliptic helix pL2 rL2 is " rGis ` rOY p´diqs p2 " p iqiPZ coincides with pLiqiPZ: " rL2is ` rL2i`1s ` rOY p´diqs ` rOY p´τ´ip1qs " rLis ` rL3i`1s ` rOY p´diqs ` rOY p´τ´ip1qs " rLis Finishing the proof of Theorem 7.4. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 27 Appendix A. Quadrics admit Z-fields of fractions In this appendix we prove the following Theorem (Theorem 3.8 and 3.12). Let A be a quadric, then A is a Z-domain and A admits a Z-field of fractions. The proof of this theorem is based on several preliminary results. Notation . Throughout this appendix A will always be a quadric. Moreover for any A-module M we let pdpMq and GKdimpMq denote the projective and Gelfand- Kirillov dimension respectively. A.1. Preliminary results. A.1.1. Some lemmas. Lemma A.1. Let M be a finitely generated left- or right-A-module and assume pdpMq ď 1, then GKdimpMq ě 2. Proof. Upon replacing the projective modules Ap´iq by eiA or Aei one can copy (cid:3) the proof of [3, Proposition 2.41] Lemma A.2. Let M be a finitely generated right-A-module and let Si denote the simple module eiA{eiAąi, then pdpMq ď 2 ñ @i P Z : HomApSi, Mq " 0 Proof. Upon replacing the projective modules Ap´iq by eiA one can copy the proof (cid:3) of [3, Proposition 2.46 (i)] Lemma A.3. Let i P Z be any integer and M be some graded submodule of eiA then GKdimpMq " 3 ô GKdimpeiA{Mq ă 3. Proof. GKdimpeiA{Mq ă 3 ñ gkdimpMq " 3 is trivial. Let us prove the other direction. Assume by way of contradiction that GKdimpMq " GKdimpeiA{Mq " 3. As both M and eiA{M are nonzero we have epMq ą 0 and epeiA{Mq ą 0. However as they have equal GKdim, we have epeiAq " epMq ` epeiA{Mq. A direct computation shows that epejAq " 1 2 holds for all j. Similar to the proof of [3, Proposition 2.21 (iii)] we then know that epMq and epeiA{Mq must be a nonnegative multiple of 1 2 . (cid:3) Contradiction! in [3]: We now introduce a homogeneous ideal N of A in a similar fashion as was done (1) eiA is a Noetherian object in GrpAq. In particular any ascending chain of submodules of eiA must stabilize. This allows us to set Ni to be the largest submodule of eiA of GKdim ď 2. à (2) Define N as iPZ Ni. Then N is a homogeneous two-sided ideal of A. To see why N also has the structure of a left ideal, note that if a P Ai,j then aNj is a submodule of eiA of GKdim ď 2. This implies that aNj Ă Ni for otherwise aNj ` Ni would be a strictly larger submodule than Ni but it still has GKdim ď 2. 28 DENNIS PRESOTTO Remark A.4. Recall that A, being a quadric, is 2-periodic [17, Proposition 5.6.1]. I.e. there is an isomorphism A -- Ap2q. This isomorphism induces an isomorphism N -- Np2q. To see this, fix any i P Z and let fi : eiA Ñ ei`2Ap2q be the induced isomorphism. Then fipeiNq has GKdim " 2 in particular, being an Ap2q-submodule of ei`2Ap2q we must have fipeiNq Ă ei`2Np2q. By considering f´1 i we see that this must in fact be an equality. Lemma A.5. Let N be as above, then A :" A{N is a Z-domain. Proof. Let b P Ai,jzNi,j, we then need to show that the induced morphism is injective. For this consider the commutative diagram ejA bÝÑ eiA 0 0 kerpbq kerpbq ejA ejA bA bA 0 0 ` kerpbq bA ` Now suppose by way of contradiction that kerpbq ‰ 0. By construction ejA " ejA{Nj does not contain submodules of GKdim ď 2, hence GKdim " 3. This implies that GKdimpkerpbqq " 3 as well. By Lemma A.3 we must have GKdimpbAq ă 3, hence also GKdim ă 3. As bA Ă eiA and eiA does not contain submodules of GKdim ď 2 we must have bA " 0, contradicting the fact that b ‰ 0. (cid:3) A.1.2. Dual modules. Next we introduce the notion of dualization of (right-)A- modules. Throughout this section we will use Aop to denote the opposite algebra of A. Aop is a Z-algebra by setting pAopqi,j " pA´j,´iqop. With this Z-algebra structure graded right-Aop-modules can be identified with graded left-A-modules; for example eiAop naturally corresponds to Ae´i. It hence makes sense to let GrpAopq denote the categories of graded left A-modules. HomGrpAqpM, eiAq naturally has the Let M be a graded right-A-module, then à structure of a graded left-A-module via: iPZ Aij b HomGrpAqpM, ejAq Ñ HomGrpAqpM, eiAq : x b f ÞÑ x f px fqpmq :" x fpmq where We denote this graded left-A-module by M or HomGrpAqpM, Aq and it is called the dual of M . One easily checks that this induces a left-exact functor HomGrpAqp´, Aq " p´q : GrpAq Ñ GrpAopq. Note that as HomGrpAqpeiA, ejAq -- Aji we naturally have HomGrpAqpeiA, Aq " peiAq " Aei. This allows us to define the right derived functors RHomGrpAqp´, Aq : fpGrpAopqq. If C‚ is some object in Db fpGrpAqq Ñ Db fpGrpAqq which is repre- à à Db sented by a bounded exact complex of finitely generated projectives, say eiA'li,n MnÝÝÝÑ eiA'li,m Ñ 0 eiA'li,n´1 Ñ . . . Ñ à 0 Ñ iPZ iPZ iPZ SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 29 à iPZ à iPZ à iPZ where Mn is some matrix whose entries are homogeneous elements in A, then RHomGrpAqpC‚, Aq is represented by the complex Ae Ae Ae MnÐÝÝÝ 'li,n i 'li,m i 'li,n´1 i Ð 0 0 Ð Ð . . . Ð (where each term in position j in the original complex gives rise to a term in position ´j in the new complex) Similar to the graded case we use the shorthand notation pC‚qD :" RHomGrpAqpC‚, Aq If M is some graded right-A-module, then we denote Exti Remark A.6. If we introduce RHomGrpAopqp´, Aq : Db fpGrpAqq in an analogous way, then pp´qDqD -- Id holds, giving rise to a biduality spectral sequence as in the graded case. GrpAqpM, Aq :" Ri HomGrpAqpM, Aq " hi fpGrpAopqq Ñ Db M D . ` For a bounded complex C‚ of (finitely generated, graded right-)A-modules (or Aop-modules) we define the Hilbert series of C‚ as ` hC‚ptq " hipC‚qti with hipC‚q " p´1qj dimk pC jqi ÿ jPZ ÿ iPZ and we denote epC‚q to be the leading coefficient of the series expension of hC‚ptq in terms of p1 ´ tq´1 and GKdimpC‚q as the highest power of p1 ´ tq´1 in this expansion, i.e. the order of pole of hC‚ptq We then have the following: Lemma A.7. Let C‚ P Db functions: fpGrpAqq, then we have the following equality of rational hpC‚qDptq " ´t4 hC‚pt´1q Proof. By linearity of the definition of hC‚, it suffices to prove the equality in case C‚ is given by some projective eiA concentrated in position j. In this case pC‚qD is given by Aei (hence e´iAop) concentrated in position p´jq such that pt´1qi t´i hpC‚qDptq " p´1q´j p1 ´ t2qp1 ´ tq2 " p´1qj " ´t´4 " ´t´4 hC‚pt´1q ´t4p1 ´ pt´1q2qp1 ´ t´1q2 pt´1qi p1 ´ pt´1q2qp1 ´ t´1q2 (cid:3) Corollary A.8. Let C‚ be a bounded complex of (finitely generated) right A- modules. Let m " GKdimpC‚q, then (1) GKdimppC‚qDq " m (2) eppC‚qDq " p´1qm`1epC‚q Proof. Suppose ř8 i"0 αip1 ´ tqi p1 ´ tqm ř8 rαip1 ´ tqi with α0 " epC‚q ‰ 0. Then we need to show that hC‚ptq " hpC‚qDptq " i"0 p1 ´ tqm 30 with Ăα0 " p´1qm`1α0. DENNIS PRESOTTO First note that for each n P Z we can write: 8ÿ j"1 tn " 1 ` βn,jp1 ´ tqj with βn,j P k. Then by Lemma A.7 we have hpC‚qDptq " " " " where rαi " p´1qm`i`1 p1 ´ t´1qm ř8 i"0 αip´t´4qp1 ´ t´1qi ř8 i"0 ´αip´1qit´i´4p1 ´ tqi ř8 p´1qmt´mp1 ´ tqm i"0 αip´1qm`i`1tm´i´4p1 ´ tqi ř8 rαip1 ´ tqi αi ` iÿ p´1qjβm´i`j´4,jαi´j i"0 p1 ´ tqm p1 ´ tqm ¸ j"1 (cid:3) GrpAqpM, Aq Lemma A.9. Let M be a finitely generated right-A-module, then Ext3 is a finite dimensional k-vectorspace. Proof. This is an immediate generalization of [3, Proposition 2.46(ii)]. (cid:3) We now prove some more results on the homogeneous ideal N as above: Lemma A.10. For each i P Z we have GKdimpN eiq ď 2. Proof. By construction we know that for each i we have GKdimpeiNq ď 2. In particular there is for each i a degree 2 polynomial Pi such that dimkpNi,i`lq ď Piplq holds for all l sufficiently large. Now fix some i P Z, we must show that there is a degree 2 polynomial Q such that dimkpNi´l,iq ď Qplq. For this recall that the 2-periodicity of A descends to N (see Remark A.4). In particular we have # dimkpNi,i`lq dimkpNi`1,i`l`1q if l is even if l is odd dimkpNi´l,iq " Without loss of generality we can now assume that Piplq ě Pi`1plq holds for all l sufficiently large. We can finish the proof by setting Q " Pi. (cid:3) Lemma A.11. Fix some i P Z and let I Ă Aei be the left-annihilator of eiN , then GKdimpIq " 3. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 31 Proof. As eiN is a submodule of the noetherian right A-module eiA, it is finitely generated. I.e. there are elements xi,i P Ai,i, xi,i`1 P Ai,i`1, . . . , xi,n P Ai,n such that eiN " nÿ xi,jA " nÿ j"i xi,jejA j"i Let Ij be the left annihilator of xi,j, i.e. Ij " ta P Aei axi,j " 0 Then there is an exact sequence of left A-modules Ş 0 Ñ Ij Ñ Aei xi,jÝÝÝÑ Aej j Ij such that such that Aei{Ij -- Axi,j. Moreover I " GKdimpA{Iq ď maxj GKdimpAxi,jq ď maxj GKdimpAejq " 2 (cid:3) (cid:3) The result now follows from Lemma A.3. Lemma A.12. For each i P Z we have: (i) eiN is a second syzygy (ii) pdpeiNq ď 1 Proof. piiq obviously follows from piq, so we only need to prove eiN is a second syzygy. By Lemmas A.11 and A.10 we know there exists an element b P Aji such that bN " 0 while b R N ei. Hence eiN Ă kerpbq while eiA bÝÑ ejA is injective by Lemma A.5. This implies that we have a left exact sequence 0 Ñ eiN Ñ eiA bÝÑ ejA finishing the proof. A.2. Proof of Theorem 3.8. Let N be as above. By Lemma A.5 it suffices to prove that N " 0. Suppose by way of contradiction that this is not the case. Without loss of generality we can assume e0N ‰ 0. Then by Lemma A.12 pdpe0Nq ď 1 which by Lemma A.1 implies GKdimpe0Nq ě 2. As by construction GKdimpe0Nq ď 2, we have GKdimpe0Nq " 2. Let pe0NqD " RHompe0N, Aq de- note the dual complex as above, by the projective dimension of e0N , this complex only has homology at position 0 and 1. By Lemma A.12 e0N is a second syzygy and hence we have h1ppe0NqDq " Ext1pe0N, Aq -- Ext3pM, Aq for some module M . Lemma A.9 then implies that h1ppe0NqDq is finite dimensional. In particular the Gelfand-Kirillov dimension and multiplicity of pe0NqD are solely determined by pe0Nq. Corollary A.8 gives eppe0Nqq " eppe0NqDq " ´epe0Nq. A contradiction! A.3. Proof of Theorem 3.12. By Theorem 3.8 and Proposition 3.11 it suffices to prove that all eiA are uniform modules. For this fix any i and nonzero M Ă eiA. Then GKdimpMq " 3. To see this let x be any nonzero element in Mj Ă Ai,j, then by Theorem 3.8 we have 3 " GKdimpejAq " GKdimpxejAq ď GKdimpMq ď GKdimpeiAq " 3 Now let N be any other nonzero submodule of eiA. Then obviously GKdimpNq " 3 as well. Suppose by way of contradiction that M X N " 0, then the following composition is a monomorphism: N ãÑ eiA Ñ eiA{M 32 DENNIS PRESOTTO such that GKdimpeiA{Mq " 3. This gives a contradiction with Lemma A.3. Hence for any nonzero M, N Ă eA we must have M X N ‰ 0, s that eiA is a uniform module. ` " dimk (41) Appendix B. I-bases for quadratic Sklyanin algebras and Lemma 8.3 Throughout this section we assume A " ApY,L, ψq is a quadratic Sklyanin alge- bra with Hilbert series h. p and q are points lying in different τ -orbits with τ " ψ3. Our goal is to prove that for a ď ´2 : HompOXp´i ´ a ´ j ´ bq,OXp´i ´ jq b mdj . . . mdj`b´1q if 2a ` b " 2n ě 0 if 2a ` b " 2n ` 1 ą 0 ` if 2a ` b ă 0 OX b mψ´np where di is as in (14). Using pOXpnq b mpqpmq " pn ` mq (see for example [10, §6]) and replacing p and q by ψxp, ψyq for the appropriate values of x and y this is equivalent to proving $'&'%pn ` 1q2 pn ` 1qpn ` 2q 0 h1p2a ` bq " (42) dimk HompOX , OX b md0 . . . mdb´1 pa ` bqq " h1p2a ` bq ` ` We will prove this using I-bases. B.1. I-bases. In this subsection we recall the definition and construction of an I- basis for a quadratic Sklyanin algebra. For a more thorough introduction to I-bases we refer the reader to [15]. Definition B.1. Let A " ApY,L, ψq be a quadratic Sklyanin algebra and let G denote the monoid of monomials in x, y, z. Let Gn denote the subset of all degree n monomials. An I-basis for A is then given by a map v : G Ñ A satisfying the following properties: i) vpGnq is a k-basis for An ii) for any g P G there are elements xg, yg, zg P A1 such that vpgxq " vpgqxg, vpgyq " vpgqyg and vpgzq " vpgqzg Remark B.2. Note that vpxq " x1, vpyq " y1, vpzq " z1. An I basis can hence alter- natively be given by a collection of txg, yg, zgugPG satisfying xgyxg " ygxyg, xgzxg " zgxzg, ygzyg " zgyzg In [15, §4] Tate and Van den Bergh give a construction for an I-basis for a In the case of a quadratic Sklyanin algebra this construction Sklyanin algebra. depends on the choice of a rational point o " po1, o2, o3q P Y 3. For each g P G one defines og by setting ox " pψo1, ψ´2o2, ψ´2o3q oy " pψ´2o1, σo2, ψ´2o3q oz " pψ´2o1, ψ´2o2, ψo3q such that if g " xαyβzλ then og " pψα´2β´2λo1, ψβ´2α´2λo2, ψλ´2α´2βo3q. SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 33 We then define xg, yg, zg P A1 " ΓpY,Lq (up to a scalar multiple) by setting xgppogq1q ‰ 0 xgppogq2q " 0 xgppogq3q " 0 ygppogq1q " 0 ygppogq3q " 0 zgppogq1q " 0 zgppogq3q ‰ 0 ygppogq2q ‰ 0 zgppogq2q " 0 (the scalar multiples are then chosen such that the relations in Remark B.2 hold) In particular vpxq " x1 P ΓpY,Lp´o2 ´ o3qq " HomXpOX ,OXp1q b mo2`o3q " HomXpOX ,pOX b mσ´1o2`σ´1o3qp1qq and analogously vpyq P HomXpOX ,pOX b mψ´1o1`ψ´1o3qp1qq vpzq P HomXpOX ,pOX b mψ´1o1`ψ´1o2qp1qq Similar computations are possible for monomials of higher degree, for example: vpxyq " x1 yx lies in the image of HomXpOX ,pOXbmψ´1o2`ψ´1o3qp1qqbHomXpOX ,pOXbmψ´1ox1`ψ´1ox3qp1qq ãÑ A1bA1 Ñ A2 This image is given by ` ` ` OX b mψ´1o2`ψ´1o3mψ´1o1`ψ´4o3 OX b mψ´1o1`ψ´1o2`ψ´1o3mτ´1ψ´1o3 p2q p2q ` OX , Hom -- Hom OX , Where we used the fact that o1, o2, o3 lie in different τ -orbits. Inspired by the above results we make the following choices for o1, o2, o3: (43) o1 " σp and o2 " σq and o3 " σr Where r is some point on Y lying in a different τ -orbit than p and q. (recall that we required p and q to lie in different τ -orbits). We can can inductively show the following: assume α, β, λ are nonnegative inte- ´ ` OX , gers then: vpxαyβzλq P Hom (44) OX b mpmτ´1p . . . mτ´β´γ`1p mqmτ´1q . . . mτ´α´γ`1q mrmτ´1r . . . mτ´α´β`1r ¯ pα ` β ` γq B.2. Proof of Lemma 8.3. Using the above language of I-bases the following ´ proposition reduces the proof of (42) to a combinatorial problem Proposition B.3. Let α, β, γ ě 0 then OX , OX b md0 . . . mdh´1 pα ` β ` γq ¯ Hom ` (45) vpxαyβzλq V ď β ` γ R h 2 P õ and Z ^ h 2 ď α ` γ Proof. ò follows from p44q. For ó we need some more computations ... (cid:3) 34 DENNIS PRESOTTO ` ` pnq Fix h P N and define the following right submodules M and M1 of AA: Mn " HomX n " Spantvpxαyβzλq α ` β ` λ " n, rh{2s ď β ` γ and th{2u ď α ` γu M1 OX b md0 . . . mdh´1 OX , To see that M1 is a right A-module, recall that for each g " xαyβzλ, the elements xg, yg, zg give a k-basis for A1 hence the image of k vpxαyβzλq b A1 Ñ An b A1 Ñ An`1 lies inside Spantvpxα`1yβzλq, vpxαyβ`1zλq, vpxαyβzλ`1qu. We then have the following lemmas. Lemma B.4. Let M and M1 be as above then for all n sufficiently large we have dimkpMnq " dimkpM1 nq n0`1 is a nonzero element in Nn0`1zM1 (cid:3) continuation of Proposition B.3. We have already proven ó which is equivalent to M1 Ă M . It then immediately follows from Lemmas B.4 and B.5 that M " M1, (cid:3) finishing the proof of the proposition. X \ \ We can now prove p42q. By Proposition B.3 it suffices to count the number of ď β ` γ and triples of natural numbers α, β, λ satisfying α` β ` λ " a` b and ď α ` γ. The latter is equivalent to α ď a ` b ´ such that P , β ď a ` b ´ b 2 our problem has turned into a combinatorial problem: we need to show T X T P b 2 b 2 b 2 Proof. We prove this for h even. The case h odd is completely similar. Let h " 2a, then we have for n ě 2a ´ 1 we have dimkpMnq " pn ` 1qpn ` 2q ´ 2 apa ` 1q " dimkpM1 nq 2 2 for M this follows from Lemma 5.10 and [16, Corollary 5.2.4]. For M1 this follows from the fact that for n ě 2a´1 at most one of the inequalities α`λ ě a, β`λ ě au can fail when α ` β ` λ " n. (cid:3) Lemma B.5. Let N be a right A-modules such that M1 Ă N Ă A and suppose there is an n0 P N such that M1 Proof. By induction it suffices to show M1 nonzero element in Nn0zM1 n Ĺ Nn for all n ě n0 n0`1 Ĺ Nn0`1. For this choose some tα,β,λvpxαyβzλq. R V n0 Ĺ Nn0 then we have M1 . This element can be written as α`β`λ"n ÿ α,β,λ n0 Without loss of generality we can assume there is a tα0,β0,λ0 ‰ 0 with β0`λ0 ă Choose such a t with α0 maximal and let g " xα0yβ0 zλ0 then h 2 . ‹‹‚xg " tα0,β0,λ0 vpxα0`1yβ0zλ0q` ÿ tα,β,λvpxαyβzλq ÿ α,β,λ α`β`λ"n α1,β1,λ1vpxα1 t1 yβ1 zλ1q α1`β1`λ1"n`1 α1,β1,λ1 α1ďα0 Lemma B.6. # (46) ! pα, β, λq P N3 P T V α ` β ` λ " a ` b, R α ď a ` b ´ ^) " h1p2a ` bq Z T b 2 b 2 , β ď a ` b ´ P SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. 35 We first show that the left hand side equals zero when 2a ` b ă 0. Note that in contradicts α P N such this case a`b ă b that the left hand side of (46) is zero. Hence from now on we can assume 2a`b ě 0. , hence the condition α ď a`b´ 2 ď b 2 b 2 This combinatorial problem has a graphical interpretation: it asks for count- ing the number of dots in Figure 3 whose coefficients pα, β, λq satisfy the above inequalities. Figure 3. Combinatorial problem # R V In order to compute this number of dots we can rewrite p46q as: ! pα, β, λq P N3 α ` β ` λ " a ` b, α ď a ` b ´ ^) Z ! ) , β ď a ` b ´ b V) R 2 pα, β, λq P N3 α ` β ` λ " a ` b ´ ^) Z ´ b 2 ^) Z ` b 2 ! ! pα, β, λq P N3 α ` β ` λ " a ` b, α ě a ` b ` 1 ´ pα, β, λq P N3 α ` β ` λ " a ` b, β ě a ` b ` 1 ´ , β ě a ` b ` 1 ´ ! pα, β, λq P N3 α ` β ` λ " a ` b, α ě a ` b ` 1 ´ (47) # V R b 2 # # # b 2 b 2 " The green, red and blue numbers can be visualized as in Figure 4. The reason for writing our combinatorial problem as in p47q is the existence of the following bijection: tpα, β, λq P N3 α ` β ` λ " n1, α ě n2u Ñ tpα, β, λq P N3 α ` β ` λ " n1 ´ n2u : pα, β, λq ÞÑ pα ´ n2, β, λq 36 DENNIS PRESOTTO Figure 4. Visualizing formula (47) ! Using similar bijections for the other sets we can write p47q as pα, β, λq P N3 α ` β ` λ " a ` b, α ď a ` b ´ # " b 2 V R Z ^) ! ) , β ď a ` b ´ b V R 2 ) ´ pα, β, λq P N3 α ` β ` λ " a ` b ^ Z ) ´ 1 ´ ) ` ´ 1 ´ a ´ b ´ 2 " ´a ´ 2 ! # ! pα, β, λq P N3 α ` β ` λ " pα, β, λq P N3 α ` β ` λ " b 2 b 2 ^ R Z V ` b 2 #tpα, βq P N2 α ` β " iu (48) # ! pα, β, λq P N3 α ` β ` λ " # # b 2 Now we can use the following: for all n ě 0: #tpα, β, λq P N3 α ` β ` λ " nu " nÿ " nÿ " pn ` 2qpn ` 1q i ` 1 i"0 i"0 2 (49) (recall that we assumed a ď ´2 and a`b ě 0 such that ´a´2, V Hence combining (48) and (49) we find that ! pα, β, λq P N3 α ` β ` λ " a ` b, α ď a ` b ´ R # Z V R b 2 ´1, ^ Z b 2 b 2 ´1 ě 0) ^) , β ď a ` b ´ b 2 SYMMETRIC NONCOMMUTATIVE BIRATIONAL TRANSFORMATIONS. equals pa ` b ` 2qpa ` b ` 1q 2 a2 ` 2ab ` b2 ` 3a ` 3b ` 2 ´ (50) P T`P b 2 ´ T ` 1 b 2 2 \`X X X T2 ´ b 2 P ` 1 \ \2 ´ b 2 b 2 2 b 2 P ´ b 2 2 2a2 ` 2ab ` b2 ` 4a ` 2b ` 2 ´ 2 \ T X ` p´1 ´ aqp´aq 2 \2 X T2 ´ P ´ ` a2 ` a b 2 b 2 b 2 37 " " We now treat the cases b even and b odd separately. First assume b " 2r for some r P N. Then (50) equals 2a2 ` 4ar ` 4r2 ` 4a ` 4r ` 2 ´ 2r2 Next assume b " 2r ` 1. Then (50) equals " a2 ` 2ar ` r2 ` 2a ` 2r ` 1 " pa ` r ` 1q2 2 2a2 ` 4ar ` 2a ` 4r2 ` 4r ` 1 ` 4a ` 4r ` 2 ` 2 ´ pr ` 1q2 ´ r2 2a2 ` 4ar ` 2r2 ` 6r ` 6a ` 4 " " a2 ` 2ar ` r2 ` 3r ` 3a ` 2 " 2 2 pa ` r ` 1qpa ` r ` 2q Finally letting n " a ` r we see that this agrees with (41) References 1. M. Artin and W.F. Schelter, Graded algebras of global dimension 3, Adv.Math 66 (1987), 171 -- 216. 2. M. Artin, J. Tate, and M. Van den Bergh, Some algebras associated to automorphisms of el- liptic curves, The Grothendieck Festschrift (P. et al. Cartier, ed.), Modern Birkhuser Classics, vol. 1, Birkhuser Boston, 1990, pp. 33 -- 85. 3. , Modules over regular algebras of dimension 3, Inventiones mathematicae 106 (1991), no. 1, 335 -- 388. 4. M. Artin and M. Van den Bergh, Twisted homogeneous coordinate rings, Journal of Algebra 133 (1990), no. 2, 249 -- 271. 5. M. Artin and J.J. Zhang, Noncommutative projective schemes, Advances in Mathematics 109 (1994), no. 2, 228 -- 287. 6. A. Bondal and A. Polishchuk, Homological properties of associative algebras: the method of helices, Russian Acad. Sci. Izv. Math. 42 (1994), 219 -- 260. 7. D. Chan and A. Nyman, Species and non-commutative P1's over non-algebraic bimodules, ArXiv e-prints (2015). 8. P. Gabriel and M. Zisman, Calculus of fractions and homotopy theory, Ergebnisse der Math- ematik und ihrer Grenzgebiete, vol. 35, Springer, 1967. 9. R. Hartshorne, Algebraic geometry, 8 ed., Graduate Texts in Mathematics, Springer-Verslag, 1997. 10. D. Presotto and M. Van den Bergh, Noncommutative versions of some classical birational transformations, Journal of Noncommutative Geometry 10 (2016), no. 1, 221 -- 244. 11. D. Rogalski, S.J. Sierra, and J.T. Stafford, Noncommutative blowups of elliptic algebras, Algebras and Representation Theory (2014), 1 -- 39. 12. S. J. Sierra, G-algebras, twistings, and equivalences of graded categories, Algebr. Represent. Theory 14 (2011), no. 2, 377 -- 390. 13. S.J. Sierra, Talk: Ring-theoretic blowing down (joint work with Rogalski, D. and Stafford, J.T.), Workshop Interactions between Algebraic Geometry and Noncommutative Algebra, 2014. 38 DENNIS PRESOTTO 14. J. T. Stafford and M. Van den Bergh, Noncommutative curves and noncommutative surfaces, Bull. Amer. Math. Soc. (N.S.) 38 (2001), no. 2, 171 -- 216. 15. J. Tate and M. Van den Bergh, Homological properties of sklyanin algebras, Inventiones math- ematicae 124 (1996), no. 1, 619 -- 647. 16. M. Van den Bergh, Blowing up non-commutative smooth surfaces, Mem. Amer. Math. Soc. 154 (2001), no. 734. 17. , Non-commutative quadrics, ArXiv e-prints (2008).
1701.02573	2	1701	2018-01-18T08:01:42	Relative singular locus and Balmer spectrum of matrix factorizations	[ "math.AG", "math.CT", "math.RT" ]	For a separated Noetherian scheme $X$ with an ample family of line bundles and a non-zero-divisor $W\in\Gamma(X,L)$ of a line bundle $L$ on $X$, we classify certain thick subcategories of the derived matrix factorization category ${\rm DMF}(X,L,W)$ of the Landau-Ginzburg model $(X,L,W)$. Furthermore, by using the classification result and the theory of Balmer's tensor triangular geometry, we show that the spectrum of the tensor triangulated category $({\rm DMF}(X,L,W), \otimes^{\frac{1}{2}})$ is homeomorphic to the relative singular locus ${\rm Sing}(X_0/X)$, introduced in this paper, of the zero scheme $X_0\subset X$ of $W$.	math.AG	math	RELATIVE SINGULAR LOCUS AND BALMER SPECTRUM OF MATRIX FACTORIZATIONS YUKI HIRANO Abstract. For a separated Noetherian scheme X with an ample family of line bundles and a non-zero-divisor W ∈ Γ(X, L) of a line bundle L on X, we classify certain thick subcat- egories of the derived matrix factorization category DMF(X, L, W ) of the Landau-Ginzburg model (X, L, W ). Furthermore, by using the classification result and the theory of Balmer's tensor triangular geometry, we show that the spectrum of the tensor triangulated category 1 2 ) is homeomorphic to the relative singular locus Sing(X0/X), introduced (DMF(X, L, W ), ⊗ in this paper, of the zero scheme X0 ⊂ X of W . 1. Introduction 1.1. Background. For a given category of algebraic objects associated to a scheme, it is ex- pected that we can extract geometric information of the scheme or the scheme itself from the category. Gabriel reconstructed a Noetherian scheme X from the abelian category Qcoh X of quasi-coherent sheaves on X [Gab], and later Rosenberg generalized the reconstruction theorem for arbitrary schemes [Ros]. Although we can't reconstruct a smooth variety from the derived cat- egory of coherent sheaves in general, Balmer reconstructed arbitrary Noetherian scheme X from the tensor triangulated category (Perf X,⊗) of perfect complexes on X with the natural tensor structure ⊗ [Bal]. Balmer's idea is to associate to any tensor triangulated category (T ,⊗) a ringed space Spec(T ,⊗) = (Spc(T ,⊗),O), and he proved an isomorphism X ∼= Spec(Perf X,⊗) by using Thomason's result of classification of thick subcategories of perfect complexes Perf X which are closed under ⊗-action of Perf X. In addition to the Thomason's result, classifications of thick subcategories of triangulated cat- egories are studied in many articles. For example, Takahashi classified thick subcategories of the stable category CM(R) of maximal Cohen-Macaulay modules over an abstract hypersurface local ring R [Tak]. Stevenson proved a classification of certain thick subcategories of the singularity category Dsg(X) of a hypersurface singularity X [Ste]. 1.2. Relative singular locus. To state our main results, we introduce a new notion of relative singular locus. Let i : T ֒→ S be a closed immersion of Noetherian schemes. We define the relative singular locus, denoted by Sing(T /S), of i as the following subset of T ; Sing(T /S) := { p ∈ T ∃F ∈ coh T such that Fp /∈ Perf OT,p and i∗(F ) ∈ Perf S }. We also consider the locally relative singular locus Singloc(T /S) of i defined by Singloc(T /S) := { p ∈ T ∃M ∈ modOT,p such that M /∈ Perf OT,p and ip∗(M ) ∈ Perf OS,p }. By definition, we have the inclusions Sing(T /S) ⊆ Singloc(T /S) ⊆ Sing(T ), where Sing(T ) is the usual singular locus of T . These loci can be different to each other, but, if S is regular, these loci are equal; Sing(T /S) = Singloc(T /S) = Sing(T ). Roughly speaking, the relative singular locus Sing(T /S) is a set of points p in T such that the mildness of the singularity of p in T is worse than the mildness of the singularity of p in S. In fact, for a quasi-projective variety X over C and a regular function f ∈ Γ(X,OX ) which is non-zero-divisor, we have the following equality of subsets of the associated complex analytic space (X an,Oan X ); Singloc(f −1(0)/X) ∩ X an = Crit(f an) ∩ Zero(f an), 2010 Mathematics Subject Classification. Primary 14F05, 18E30; Secondary 14B05, 32S05. Key words and phrases. relative singular locus, matrix factorizations, tensor triangular geometry. 1 2 Y. HIRANO where Crit(f an) denotes the critical locus of the associated function f an ∈ Γ(X an,Oan defined by X ), which is and Zero(f an) is the zero locus of f an. Crit(f an) := { p ∈ X an (f an)p − f an(p) ∈ m2 p}, 1.3. Main results. A data (X, L, W ) is called Landau-Ginzburg model, or just LG-model, if X is a scheme, L is a line bundle on X, and W ∈ Γ(X, L) is a section of L. To a LG-model (X, L, W ) we associate a triangulated category DMF(X, L, W ), called the derived matrix factorization category, introduced by Positselski [Pos, EP]. Tensor products of matrix factorizations defines the bifunctor; (−) ⊗ (−) : DMF(X, L, W1) × DMF(X, L, W2) → DMF(X, L, W1 + W2). In particular, DMF(X, L, W ) has a tensor action from DMF(X, L, 0). The following is our main result of classification of thick subcategories of derived matrix factor- ization categories. Theorem 1.1 (Theorem 5.6). Let X be a separated Noetherian scheme with an ample family of line bundles, L be a line bundle on X, and let W ∈ Γ(X, L) be a non-zero-divisor. Denote by X0 the zero scheme of W . Then there is a bijective correspondence (cid:26) unions of closed subsets of Sing(X0/X)(cid:27) σ ←−(cid:26) thick subcategories of DMF(X, L, W ) that are closed under tensor action f rom DMF(X, L, 0)(cid:27) −→ τ The bijective map σ sends Y to the thick subcategory consisting of matrix factorizations F ∈ DMF(X, L, W ) with Supp(F ) ⊆ Y . The inverse bijection τ sends T to the unionSF ∈T Supp(F ). If X is a regular separated Noetherian scheme, then X has an ample family of line bundles and DMF(X, L, W ) is equivalent to the singularity category Dsg(X0). Furthermore, if X = Spec R is affine with R regular local, DMF(X, L, W ) is equivalent to the stable category CM(R/W ) of max- imal Cohen-Macaulay modules over the hypersurface R/W . Hence Theorem 1.1 can be considered as a simultaneous generalization of Stevenson's result in [Ste] and Takahashi's result in [Tak]. As an application of the above main result, we see that the closedness of the relative singular locus Sing(X0/X) is related to the existence of a ⊗-generator of DMF(X, L, W ), where we say that an object G ∈ DMF(X, L, W ) is a ⊗-generator if the smallest thick subcategory that is closed under tensor action from DMF(X, L, 0) and contains G is DMF(X, L, W ). Corollary 1.2 (Corollary 5.8). Notation is same as in Theorem 1.1. Then the subset Sing(X0/X) of X0 is closed if and only if DMF(X, L, W ) has a ⊗-generator. Furthermore, we construct the relative singular loci from the derived matrix factorization cate- gories. If 2 ∈ Γ(X,OX ) is a unit in the ring Γ(X,OX ), the derived matrix factorization category DMF(X, L, W ) has a natural (pseudo) tensor triangulated structure ⊗ 2 on it. Using Theorem 1.1 and the theory of Balmer's tensor triangular geometry, we prove that the spectrum of the (pseudo) tensor triangulated category (DMF(X, L, W ),⊗ 1 Theorem 1.3 (Corollary 6.10). Let X be a separated Noetherian scheme with an ample ample family of line bundles, and let W ∈ Γ(X, L) be a non-zero-divisor of a line bundle L. Assume that 2 ∈ Γ(X,OX ) is a unit. Then we have a homeomorphism 2 ) is the relative singular locus Sing(X0/X). 1 Spc(DMF(X, L, W ),⊗ 1 2 ) ∼= Sing(X0/X). This result is a generalization of Yu's result [Yu2, Theorem 1.2], where he proved Theorem 1.3 in the case when X is an affine regular scheme of finite Krull dimension by using the classification result due to Walker. 1.4. Plan of the paper. In section 2 we provide basic definitions and properties about derived matrix factorization categories. In section 3 we give the definitions of globally/locally relative singular loci and prove some properties about relative singular loci for zero schemes of regular sections of line bundles. In section 4 we prove tensor nilpotence properties of matrix factorizations which are key properties for our classification result. In section 5 we prove the main result Theorem 1.1. In section 6 we recall the theory of Balmer's tensor triangular geometry, and we study the natural tensor triangulated structure on derived matrix factorization categories. 3 1.5. Acknowledgements. The author would like to thank Hokuto Uehara for many useful com- ments and his continuous support. The author also thank Michael Wemyss and Shinnosuke Okawa for valuable discussions and comments on a draft version of the paper. The author was a Research Fellow of Japan Society for the Promotion of Science. He was partially supported by Grant-in-Aid for JSPS Fellows No.26-6240. 2. Derived matrix factorizations 2.1. Derived matrix factorization categories. In the first subsection, we recall the definition of the derived matrix factorization category of a Landau-Ginzburg model, which is introduced by Positselski (cf. [Pos], [EP]), and provide its basic properties. Definition 2.1. A Landau-Ginzburg model, or LG model, is data (X, L, W ) consisting of a scheme X, an invertible sheaf L on X, and a section W ∈ Γ(X, L) of L. Notation 2.2. If L is isomorphic to the structure sheaf OX , we denote the LG model by (X, W ). If X = Spec R is an affine scheme, we denote the LG model by (R, L, W ), where L is considered as an invertible R-module and W ∈ L. For a LG model, we consider its factorizations which are twisted" complexes. " Definition 2.3. Let (X, L, W ) be a LG model. A factorization F of (X, L, W ) is a sequence ϕF F =(cid:16)F1 ϕF 0−−→ F1 ⊗ L(cid:17), 1 ⊗ L) ◦ ϕF 1−−→ F0 where each Fi is a coherent sheaf on X and each ϕF 1 = W·idF1 i and (ϕF 0 = W · idF0. Coherent sheaves F0 and F1 in the above sequence are called components of the factorization F . If the components Fi of F are locally free sheaves, we call F a matrix factorization of (X, L, W ). Notation 2.4. We can consider any coherent sheaf F ∈ coh X as a factorization of (X, L, 0) of the following form is a homomorphism such that ϕF 0 ◦ϕF (cid:16)0 −→ F −→ 0(cid:17). By abuse of notation, we will often denote the above factorization by the same notation F . Definition 2.5. For a LG model (X, L, W ), we define an exact category coh(X, L, W ) whose objects are factorizations of (X, L, W ), and whose set of morphisms are defined as follows: For two objects E, F ∈ coh(X, L, W ), we define Hom(E, F ) as the set of pairs (f1, f0) of fi ∈ HomcohX (Ei, Fi) such that the following diagram is commutative; E1 f1 F1 ϕE 1 ϕF 1 E0 f0 / F0 ϕE 0 ϕF 0 E1 ⊗ L f1⊗L / F1 ⊗ L. Note that if X is Noetherian, coh(X, L, W ) is an abelian category. We define a full additive subcategory MF(X, L, W ) of coh(X, L, W ) whose objects are matrix factorizations. By construction, MF(X, L, W ) is also an exact category. Since factorizations are " tions. twisted" complexes, we can consider homotopy category of factoriza- Definition 2.6. Two morphisms f = (f1, f0) and g = (g1, g0) in Homcoh(X,L,W )(E, F ) are ho- motopy equivalent, denoted by f ∼ g, if there exist two homomorphisms in coh X such that f0 − g0 = ϕF 1 h0 + h1ϕE 0 and f1 ⊗ L − g1 ⊗ L = ϕF 0 h1 + (h0 ⊗ L)(ϕE 1 ⊗ L). h0 : E0 → F1 and h1 : E1 ⊗ L → F0 / / / / / / 4 Y. HIRANO The homotopy category of factorizations Kcoh(X, L, W ) is defined as the category whose objects are same as coh(X, L, W ), and the set of morphisms are defined as the set of homotopy equivalence classes; Similarly, we define the homotopy category of matrix factorizations KMF(X, L, W ), i.e. HomKcoh(X,L,W )(E, F ) := Homcoh(X,L,W )(E, F )/ ∼ . Ob(KMF(X, L, W )) := MF(X, L, W ) HomKMF(X,L,W )(E, F ) := HomMF(X,L,W )(E, F )/ ∼ . Next we define the totalization of a bounded complex of factorizations, which is an analogy of the total complex of a double complex. Definition 2.7. Let F • = (··· → F i δi l = 0, 1, set −→ F i+1 → ···) be a bounded complex of coh(X, L, W ). For and let be a homomorphism given by Tl := Mi+j=−l j ⊗ L⊗⌈j/2⌉, F i tl : Tl → Tl+1 tlF i j ⊗L⊗⌈j/2⌉ := δi j ⊗ L⊗⌈j/2⌉ + (−1)iϕF i j ⊗ L⊗⌈j/2⌉, where n is n modulo 2, and ⌈m⌉ is the minimum integer which is greater than or equal to a real number m. We define the totalization Tot(F •) ∈ coh(X, L, W )) of F • as Tot(F •) :=(cid:16)T1 t1−→ T0 t0−→ T1 ⊗ L(cid:17). In what follows, we will recall that the homotopy categories Kcoh(X, L, W ) and KMF(X, L, W ) have structures of triangulated categories. Definition 2.8. We define an automorphism T on Kcoh(X, L, W )), which is called shift functor, as follows. For an object F ∈ Kcoh(X, L, W ), we define an object T (F ) as T (F ) :=(cid:16)F0 −ϕF 0−−−→ F1 ⊗ L −ϕF 1 ⊗L −−−−−→ F0 ⊗ L(cid:17), and for a morphism f = (f1, f0) ∈ Hom(E, F ) we set T (f ) := (f0, f1 ⊗ L) ∈ Hom(T (E), T (F )). For any integer n ∈ Z, denote by (−)[n] the functor T n(−). Definition 2.9. Let f : E → F be a morphism in coh(X, L, W ). We define its mapping cone Cone(f ) to be the totalization of the complex with F in degree zero. A distinguished triangle is a sequence in Kcoh(X, L, W ) which is isomorphic to a sequence (· · · → 0 → E f −→ F → 0 → · · ·) of the form −→ E[1], where i and p are natural injection and projection respectively. −→ F i−→ Cone(f ) E p f The following proposition is well known to experts. Proposition 2.10. The homotopy categories Kcoh(X, L, W ) and KMF(X, L, W ) are triangulated categories with respect to the above shift functor and the above distinguished triangles. Following Positselski ([Pos], [EP]), we define derived factorization categories. 5 Definition 2.11. Denote by Acoh(X, L, W ) the smallest thick subcategory of Kcoh(X, L, W ) containing all totalizations of short exact sequences in coh(X, L, W ). We define the derived factorization category of (X, L, W ) as the Verdier quotient Dcoh(X, L, W ) := Kcoh(X, L, W )/Acoh(X, L, W ). Similarly, we consider the thick subcategory AMF(X, L, W ) containing all totalizations of short exact sequences in the exact category MF(X, L, W ), and define the derived matrix factorization category by DMF(X, L, W ) := KMF(X, L, W )/AMF(X, L, W ). The following proposition is a special case of [BDFIK, Lemma 2.24]. Proposition 2.12 (cf. For P ∈ KMF(R, L, W ) and A ∈ Acoh(R, L, W ), we have [BDFIK, Lemma 2.24]). Assume that X = Spec R is an affine scheme. HomKcoh(R,L,W )(P, A) = 0. In particular, the Verdier localizing functor is an equivalence. KMF(R, L, W ) ∼−→ DMF(R, L, W ) For later use, we consider larger categories of factorizations. Denote by Sh(X, L, W ) the abelian category whose objects are factorizations whose components are OX -modules. More precisely, objects of Sh(X, L, W ) are sequences of the following form F =(cid:16)F1 ϕF 1−−→ F0 ϕF 0−−→ F1 ⊗ L(cid:17), i are homomorphisms such that ϕF where Fi are OX -modules and ϕF 1 ⊗ L ◦ ϕF 0 = W · idF0. Denote by Qcoh(X, L, W ), InjSh(X, L, W ), and InjQcoh(X, L, W ) the full subcategories of Sh(X, L, W ) consisting of factorizations whose components are quasi-coherent sheaves, injective OX -modules, and injective quasi-coherent sheaves respectively. Then, similarly to Kcoh(X, L, W ), we can consider their homotopy categories KSh(X, L, W ), KQcoh(X, L, W ), KInjSh(X, L, W ), KInjQcoh(X, L, W ) respectively, and these homotopy cate- gories have natural triangulated structures similar to Kcoh(X, L, W ). 0 ◦ ϕF 1 = W · idF1 and ϕF Definition 2.13. Denote by AcoSh(X, L, W ) (resp. AcoQcoh(X, L, W )) the smallest thick subcat- egory of KSh(X, L, W ) (resp. KQcoh(X, L, W )) containing all totalizations of short exact sequences in Sh(X, L, W ) (resp. Qcoh(X, L, W )) and closed under arbitrary direct sums. Following [Pos], [EP], we define the coderived factorization categories DcoSh(X, L, W ) and DcoQcoh(X, L, W ) as the following Verdier quotients DcoSh(X, L, W ) := KSh(X, L, W )/AcoSh(X, L, W ) DcoQcoh(X, L, W ) := KQcoh(X, L, W )/AcoQcoh(X, L, W ). Lemma 2.14 ([BDFIK], [EP]). Assume that X is Noetherian. (1) The natural functor KInjSh(X, L, W ) → DcoSh(X, L, W ) is an equivalence. (2) The natural functor KInjQcoh(X, L, W ) → DcoQcoh(X, L, W ) is an equivalence. (3) The natural functor DcoQcoh(X, L, W ) → DcoSh(X, L, W ) is fully faithful. (4) The natural functor Dcoh(X, L, W ) → DcoQcoh(X, L, W ) is fully faithful. (5) The natural functor DMF(X, L, W ) → Dcoh(X, L, W ) is fully faithful. Proof. (1) and (2) follow from [BDFIK, Corollary 2.25]. (3) follows from (1) and (2). (4) and (5) are [EP, Propostion 1.5.(d)] and [EP, Corollary 2.3.(i)] respectively. (cid:3) 6 Y. HIRANO 2.2. Case when W = 0. In this section, we consider cases when W = 0. Firstly, we will define cohomologies of factorizations of (X, L, 0). Definition 2.15. For an object F ∈ Qcoh(X, L, 0), we define its cohomologies Hi(F ) ∈ Qcoh X as Hi(F ) := Ker(ϕF i )/Im(ϕF i−1) for i ∈ Z/2Z Lemma 2.16. Let k be any field. Then, for any object F ∈ KMF(k, 0), there are two finite dimensional k-vector spaces V1 and V2 such that F is isomorphic to V1 ⊕ V2[1] in KMF(k, 0), where Vi denotes the factorization of the form (0 → Vi → 0) by Notation 2.4. Proof. By [BDFIK, Lemma 2.26], there are two finite dimensional k-vactor spaces V and V ′, and a triangle of the following form in Dcoh(k, 0) = DMF(k, 0) V → V ′ → F → V [1]. But DMF(k, 0) = KMF(k, 0) by Proposition 2.12, so we have a k-linear homomorphism f : V → V ′ such that F is isomorphic to C(f ), where f is the morphism in KMF(k, 0) represented by the following morphism in MF(k, 0) 0 0 V f / V ′ / 0 / 0 By construction of mapping cones, C(f ) is isomorphic to the following matrix factorization (cid:16)V f −→ V ′ 0−→ V(cid:17). Let I := Im(f ) be the image of f , and let K := Ker(f ) be the kernel of f . Then there is a k-vector space J such that V ′ = I ⊕ J. Since V = K ⊕ I, we have the following isomorphism in MF(k, 0) F ∼=(cid:16)K 0−→ 0 0−→ K(cid:17) ⊕(cid:16)I ∼−→ I 0−→ I(cid:17) ⊕(cid:16)0 0−→ J 0−→ 0(cid:17) But the object(cid:16)I ∼−→ I 0−→ I(cid:17) is zero in KMF(k, 0). Hence F ∼= J ⊕ K[1] in KMF(k, 0). Corollary 2.17. Let k be a field. Any non-zero morphism f : (0 → k → 0) → E in KMF(k, 0) is a split mono. (cid:3) Proof. This follows from Lemma 2.16. (cid:3) 2.3. Tensor products and sheaf Homs functors. In this subsection, we recall tensor products and local homs on derived matrix factorization categories. Let (X, L, W ) be a LG model, and V ∈ Γ(X, L) be another global section. For E ∈ MF(X, L, V ) and F ∈ MF(X, L, W ), we define the tensor product E ⊗ F ∈ MF(X, L, V + W ) of E and F as and (E ⊗ F )1 :=(cid:0)E1 ⊗ F0(cid:1) ⊕(cid:0)E0 ⊗ F1(cid:1), (E ⊗ F )0 :=(cid:0)E0 ⊗ F0(cid:1) ⊕(cid:0)E1 ⊗ F1 ⊗ L(cid:1), 0 ⊗ 1(cid:19) , 1 ⊗ 1 −1 ⊗ ϕF 1 ⊗ 1 ! . 0 ⊗ 1 −1 ⊗ ϕF 1 ⊗ ϕF :=(cid:18) ϕE := ϕE 1 ⊗ ϕF 0 ϕE ϕE⊗F ϕE⊗F 1 1 1 ϕE 0 0 This defines an additive functor (−) ⊗ (−) : MF(X, L, V ) × MF(X, L, W ) → MF(X, L, V + W ), and it naturally induces an exact functor (−) ⊗ (−) : DMF(X, L, V ) × DMF(X, L, W ) → DMF(X, L, V + W ). / / / / / We define the sheaf Hom from E to F as Hom(E, F ) ∈ MF(X, L, W − V ) 7 Hom(E, F )1 := (Hom(E1, F0) ⊗ L−1) ⊕ Hom(E0, F1), Hom(E, F )0 := Hom(E0, F0) ⊕ Hom(E1, F1), ϕHom(E,F ) 1 (ϕF :=(cid:18) :=(cid:18)−(∗) ◦ ϕE 1 0 ϕF (∗) ◦ ϕE 1 ◦ (∗) (∗) ◦ ϕE 0 ⊗ L−1) ◦ (∗) ϕF 1 ◦ (∗) 1(cid:19) , 0 ⊗ L−1)(cid:19) . ϕF 0 ◦ (∗) −(∗) ◦ (ϕE and ϕHom(E,F ) 0 This defines an additive functor Hom(−,−) : MF(X, L, V )op × MF(X, L, W ) → MF(X, L, W − V ), and it induces an exact functor Hom(−,−) : DMF(X, L, V )op × DMF(X, L, W ) → DMF(X, L, W − V ). The following is standard, so we skip the proof (see [BFK] or [LS] for details) . Proposition 2.18. Let E ∈ MF(X, L, V ), F ∈ MF(X, L, W ), and G ∈ MF(X, L, V + W ). (1) We have a natural isomorphism HomDMF(X,L,V +W )(E ⊗ F, G) ∼= HomDMF(X,L,V )(E,Hom(F, G)). (2) There is a natural isomorphism in MF(X, L, V + W ) Hom(G, E ⊗ F ) ∼= Hom(G, E) ⊗ F. Recall that OX ∈ MF(X, L, 0) denotes the matrix factorization of the form(cid:16)0 → OX → 0(cid:17) by Notation 2.4. For any object F ∈ MF(X, L, W ), we define the dual F ∨ := Hom(F,OX ) ∈ MF(X, L,−W ) of F . By Proposition 2.18, the functors (−) ⊗ F : DMF(X, L, V ) → DMF(X, L, V + W ) and (−) ⊗ F ∨ : DMF(X, L, V + W ) → DMF(X, L, V ) are adjoint; (−) ⊗ F ⊣ (−) ⊗ F ∨. 2.4. Supports of matrix factorizations. We study the supports of objects in derived matrix factorization categories. Let (X, L, W ) be a LG model. For any point p ∈ X, we denote by Xp := Spec(OX,p) the stalk of X at p, and let vect X be the category of locally free sheaves of finite ranks on X. Taking the stalk (−)p : vect X → vect Xp at p induces the functor (−)p : KMF(X, L, W ) → KMF(Xp, Wp) defined by Fp :=(cid:16)(F1)p (ϕ1)p−−−→ (F0)p (ϕ0)p−−−→ (F1)p(cid:17). Since the functor (−)p : vect X → vect Xp preserves short exact sequences, the induced functor (−)p : KMF(X, L, W ) → KMF(Xp, Wp) maps AMF(X, L, W ) to AMF(Xp, Wp). Hence it defines the following functor; (−)p : DMF(X, L, W ) → KMF(Xp, Wp), since we have the natural equivalence KMF(Xp, Wp) ∼−→ DMF(Xp, Wp) by Proposition 2.12. Definition 2.19. For an object F ∈ DMF(X, L, W ), we define its support as Supp(F ) := { p ∈ X Fp 6= 0 ∈ KMF(Xp, Wp)}. Proposition 2.20. Let F ∈ DMF(X, L, W ) be an object. (1) If X =Si∈I Ui is an open covering of X, we have the equality of subsets of X Supp(F ) =[i∈I Supp(FUi ), where FUi is the restriction of F to DMF(Ui, LUi , WUi). (2) Supp(F ) is a closed subset of X. 8 Y. HIRANO Proof. (1) This follows from isomorphisms Fp ∼= (FUi )p for any p ∈ Ui. (2) We show the following equality Supp(F )c = [U∈U U, inclusion Supp(F )c ⊃SU∈U U is obvious. We verify that Supp(F )c ⊆SU∈U U . By definition, for where U := { U U is an open subscheme of X such that FU = 0 in DMF(U, LU , WU )}. The any p ∈ Supp(F )c, Fp = 0 in KMF(Xp, Wp). Let h = (h1, h0) be a homotopy giving the homotopy equivalence idFp ∼ 0. Then there is a neighborhood U of p such that there exist morphisms h1 : F1U ⊗ LU → F0U and h0 : F0U → F1U in coh U with h1p = h1 and h0p = h0. Furthermore, since id(F0)p − (ϕ1)ph0− h1(ϕ0)p = 0 in coh Up, there exists an open neighborhood V ⊆ U of p such that idF0V − ϕ1V h1V − ϕ0V h0V = 0. Then hV = (h1V , h0V ) gives a homotopy equivalence idF V ∼ 0. Hence FV = 0 in KMF(V, LV , WV ), in particular, so is in DMF(V, LV , WV ). Therefore, we have V ∈ U, which implies that p ∈SU∈U U . Definition 2.21. Let F ∈ DMF(X, L, W ) be an object. For any point p ∈ X, let ιp : Spec k(p) → X be a natural morphism, where k(p) := OX,p/mp is the residue field of the local ring (OX,p, mp). Then we denote by W ⊗ k(p) the pull-back ι∗ −−−→ ι∗ pF0 F ⊗ k(p) := ι∗ pF1(cid:17) ∈ KMF(k(p), W ⊗ k(p)). pF =(cid:16)ι∗ pW , and we set ι∗ pϕ0 −−−→ ι∗ ι∗ pϕ1 pF1 (cid:3) Then (−) ⊗ k(p) defines an exact functor (−) ⊗ k(p) : DMF(X, L, W ) → KMF(k(p), W ⊗ k(p)). The following lemma is a version of Nakayama's lemma for matrix factorizations. Lemma 2.22. Let F ∈ DMF(X, L, W ) be an object, and let p ∈ X be a point. Then if and only if F ⊗ k(p) 6= 0 in KMF(k(p), W ⊗ k(p)). p ∈ Supp(F ) Proof. Recall that we have natural equivalences KMF(k(p), W ⊗ k(p)) ∼−→ DMF(k(p), W ⊗ k(p)) and KMF(Xp, Wp) ∼−→ DMF(Xp, Wp). Note that the following diagram of functors is commutative; DMF(X, L, W ) (−)⊗k(p) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ (−)p 4❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ (−)⊗k(mp) KMF(Xp, Wp) KMF(k(p), W ⊗ k(p)) where mp ∈ Xp is the unique closed point. Hence, if Fp = 0 in KMF(Xp, Wp), then F ⊗ k(p) = 0 in KMF(k(p), W ⊗ k(p)). For the other implication, it suffices to show that for a local ring (R, m), an element w ∈ R, and an object E ∈ KMF(R, w), if E ⊗R R/m = 0 in KMF(R/m, w ⊗ R/m), then E = 0 in KMF(R, w). Since R is local, any locally free modules are free. Hence, the object E can be represented by some matrix factorization of the following form (cid:16)R⊕n1 ϕ1−→ R⊕n0 ϕ0−→ R⊕n1(cid:17). If E ⊗R R/m = 0, there exist homotopies h0 : (R/m)⊕n0 → (R/m)⊕n1 and h1 : (R/m)⊕n1 → (R/m)⊕n0 such that id(R/m)⊕n0 = (ϕ1⊗ R/m)h0 + h1(ϕ0 ⊗ R/m) and id(R/m)⊕n1 = (ϕ0 ⊗ R/m)h1 + h0(ϕ1 ⊗ R/m). Since h0 and h1 can be represented by a matrix of units in R, there exist homo- morphisms h0 : R⊕n0 → R⊕n1 and h1 : R⊕n1 → R⊕n0 such that hi ⊗R R/m = hi for i = 0, 1. Set α1 := ϕ0h1 + h0ϕ1 : R⊕n1 −→ R⊕n1 α0 := ϕ1h0 + h1ϕ0 : R⊕n0 −→ R⊕n0. Then the pair α := (α1, α0) defines an endomorphism of E in the exact category MF(R, w). By construction, α = 0 in KMF(R, w). To show that E = 0 in KMF(R, w), it is enough to show that α : E → E is an automorphism in MF(R, w). For each i ∈ {0, 1}, we only need to show / / 4 that αi is an automorphism. Since the tensor product (−) ⊗R R/m is a right exact functor and αi ⊗R R/m = id, we have Cok(αi) ⊗R R/m ∼= Cok(αi ⊗R R/m) = 0. By Nakayama's lemma, the above implies that Cok(αi) = 0. Hence αi is an automorphism by [Mat2, Theorem 2.4]. (cid:3) 9 For later use, we provide the following lemma. Lemma 2.23. Let R be a ring, and let F =(cid:16)R we have the following equality of subsets of Spec R: f0−→ R(cid:17) ∈ KMF(R, f0f1) be an object. Then Z(fi), f1−→ R Supp(F ) = \i=0,1 where Z(fi) := { p ∈ Spec R fi ⊗ k(p) = 0 }. Proof. (⊆) If p ∈ Supp(F ), then F ⊗k(p) 6= 0 in KMF(k(p), (f0f1)⊗k(p)) by Lemma 2.22. Suppose p /∈ Z(fi) for some i. Then fi ⊗ k(p) is a unit in k(p), and hence F ⊗ k(p) is homotopic to zero. This contradicts to F ⊗ k(p) 6= 0. Hence p ∈Ti=0,1 Z(fi). (⊇) If p ∈Ti=0,1 Z(fi). Then fi ⊗ k(p) = 0 for i = 0, 1, and so Hi(F ⊗ k(p)) = k(p) 6= 0. Hence by [LS, Proposition 2.30], F ⊗ k(p) 6= 0. Again by Lemma 2.22, p ∈ Supp(F ). (cid:3) The following lemma is useful to compute the support of tensor products of matrix factorizations. Lemma 2.24. Let V, W ∈ Γ(X, L) be any global sections of L, and let E ∈ DMF(X, L, V ) and F ∈ DMF(X, L, W ) be objects. Let p ∈ X be a point such that V ⊗ k(p) = W ⊗ k(p) = 0. Then p ∈ Supp(E ⊗ F ) if and only if p ∈ Supp(E) ∩ Supp(F ). Proof. We have (E ⊗ F ) ⊗ k(p) ∼= (E ⊗ k(p)) ⊗ (F ⊗ k(p)) in KMF(k(p), 0). By Lemma 2.22, it is enough to show that for a field k, and for objects M, N ∈ KMF(k, 0), M ⊗ N 6= 0 if and only if M 6= 0 and N 6= 0. By Lemma 2.16, for i = 0, 1, we may assume ϕM i = 0, and then Hi(M ) = Mi and Hi(N ) = Ni. Then, since ϕM⊗N = 0 for i = 0, 1, we have i = ϕN i H1(M ⊗ N ) =(cid:16)H1(M ) ⊗ H0(N )(cid:17) ⊕(cid:16)H0(M ) ⊗ H1(N )(cid:17) H0(M ⊗ N ) =(cid:16)H0(M ) ⊗ H0(N )(cid:17) ⊕(cid:16)H1(M ) ⊗ H1(N )(cid:17). Hence, by [LS, Proposition 2.30], we see that M ⊗ N 6= 0 if and only if M 6= 0 and N 6= 0. (cid:3) At the end of this section, we organize fundamental properties of supports of matrix factoriza- tions. Lemma 2.25. Let E, F, G ∈ DMF(X, L, W ) be objects. We have the following. (1) Supp(E ⊕ F ) = Supp(E) ∪ Supp(F ). (2) Supp(F [1]) = Supp(F ). (3) Supp(E) ⊆ Supp(F ) ∪ Supp(G) for any distinguished triangle E → F → G → E[1]. (4) Supp(E ⊗ F ) = Supp(E) ∩ Supp(F ). Proof. (1), (2), and (3) are obvious. If p ∈ Supp(M ) for some object M ∈ DMF(X, L, W ), then W ⊗ k(p) = 0, since KMF(k(p), W ⊗ k(p)) = 0 if W ⊗ k(p) 6= 0. Hence (4) follows from Lemma 2.24. (cid:3) 3. Relative singular locus and singularity category In this section, we define relative singular loci and prove some properties about it. Let S be a Noetherian scheme and let F ∈ Db(coh S) be a bounded complex of coherent sheaves. The complex F is called perfect if it is locally quasi-isomorphic to a bounded complex of locally free sheaves of finite rank. Perf S ⊂ Db(coh S) denotes the thick subcategory of perfect complexes. point p ∈ S. Definition 3.1. Let S be a Noetherian scheme, and let i : T ֒→ S be a closed immersion. We define globally/locally relative singular locus. Recall our notation Sp := Spec(OS,p) for any 10 Y. HIRANO (1) The subset Sing(T /S) ⊂ T , called the singular locus of T globally relative (or just relative) to S, is defined by Sing(T /S) := { p ∈ T ∃F ∈ coh T such that Fp /∈ Perf Tp and i∗(F ) ∈ Perf S } (2) The subset Singloc(T /S) ⊂ T , called the singular locus of T locally relative to S, is defined by Singloc(T /S) := { p ∈ T ∃F ∈ coh Tp such that F /∈ Perf Tp and ip∗(F ) ∈ Perf Sp } where ip : Tp ֒→ Sp is the closed immersion induced by i : T ֒→ S Proposition 3.2. Let S be a Noetherian scheme, and let i : T ֒→ S be a closed immersion. Then we have Furthermore, if S is regular, globally and locally relative singular loci coincide with usual singular locus; Sing(T /S) ⊆ Singloc(T /S) ⊆ Sing(T ). Sing(T /S) = Singloc(T /S) = Sing(T ). Proof. The first assertion Sing(T /S) ⊆ Singloc(T /S) ⊆ Sing(T ) is obvious. For the latter assertion, assume that S is regular. If Sing(T ) = ∅, Sing(T /S) = Singloc(T /S) = Sing(T ) = ∅ by the former assertion. Assume that Sing(T ) 6= ∅, and let p ∈ Sing(T ) be a singular point. It is enough to show that p ∈ Sing(T /S). Since the projective dimension, denoted by pdOT ,p k(p), of k(p) as OT,p-module coincides with the global dimension of OT,p, we have pdOT ,p (k(p)) = ∞. This implies that k(p) /∈ Perf Tp. Let {a1, . . . , ar} ⊂ OT,p be generators of the maximal ideal mp of the local ring OT,p. Then there exist a small open affine neighborhood U = Spec R ⊂ T of p and elements b1, . . . , br ∈ R such that (bi)p = ai. Let I := hb1, . . . , bri be the ideal of R generated by bi. Then Ip ∼= mp and (R/I)p ∼= k(p). Take an extension F ∈ coh T of the coherent sheaf gR/I ∈ coh U , i.e. FU ∼= gR/I. Then Fp ∼= k(p) /∈ Perf Tp. Since S is regular, we have Db(coh S) = Perf S, and so i∗(F ) ∈ Perf S. Hence p ∈ Sing(T /S). (cid:3) The locally relative singular locus has a local property. Lemma 3.3. Let i : T ֒→ S be a closed immersion of Noetherian schemes. Then we have Singloc(T /S) = [p∈T Singloc(Tp/Sp), where the sets Singloc(Tp/Sp) on the right hand side are considered as the subsets of T via the natural injective maps jp : Tp ֒→ T . Proof. If p ∈ Singloc(T /S), then mp ∈ Singloc(Tp/Sp), and jp(mp) = p. This means that p ∈ Singloc(Tp/Sp), and so Singloc(T /S) ⊆Sp∈T Singloc(Tp/Sp). If q ∈ Singloc(Tp/Sp) for some p ∈ T , then, since (Tp)q ∼= Tjp(q), jp(q) ∈ Singloc(T /S). HenceSp∈T Singloc(Tp/Sp) ⊆ Singloc(T /S). (cid:3) Next we recall singularity categories. Let X be a separated Noetherian scheme with resolution property, i.e. for any F ∈ coh X, there exist a locally free coherent sheaf E and a surjective homomorphism E ։ F . Following [Orl1], we define the triangulated category of singularities Dsg(X) as the Verdier quotient Dsg(X) = Db(coh X)/ Perf X. In our assumption, Perf X coincides with thick subcategory of complexes which are quasi-isomorphic to a bounded complex of locally free sheaves of finite rank. We recall that derived matrix factorization categories can be embedded into singularity cate- gories. Let L be a line bundle on X, and W ∈ Γ(X, L) be a non-zero-divisor, i.e. the induced homomorphism W : OX → L is injective, and denote by X0 the zero scheme of W . Denote by j : X0 ֒→ X the closed immersion. Since the direct image j∗ : Db(coh X0) → Db(coh X) preserves perfect complexes by [TT, Proposition 2.7.(a)], it induces an exact functor j◦ : Dsg(X0) → Dsg(X). As in [Orl2], the cokernel functor Σ : MF(X, L, W ) → coh X0 defined by Σ(F ) := Cok(ϕF an exact functor 1 ) induces Σ : DMF(X, L, W ) → Dsg(X0). Theorem 3.4 ([Orl2, Theorem 1], [EP, Theorem 2.7]). The functor Σ : DMF(X, L, W ) → Dsg(X0) is fully faithful, and the essential image of Σ is the thick subcategory consisting of objects F such that j◦(F ) = 0 ∈ Dsg(X). In particular, if X is regular, Σ is an equivalence. The following result is the key motivation for our definitions of relative singular loci. 11 Proposition 3.5. (1) We have an equality of subsets of X Sing(X0/X) = Supp(F ). [F ∈DMF(X,L,W ) (2) We have an equality of subsets of X Singloc(X0/X) = { p ∈ X KMF(Xp, Wp) 6= 0} Proof. Since (2) follows from a similar proof of (1), we prove only (1). i=1 mini. Set Ki := (R W =Pr If p ∈ Sing(X0/X), by definition, there exists A ∈ Dsg(X0) such that Ap 6= 0 in Dsg((X0)p) and j◦(A) = 0 in Dsg(X). Then by Theorem 3.4, there is an object F ∈ DMF(X, L, W ) such that Σ(F ) ∼= A. Then we have Ap ∼= Σ(F )p ∼= Σp(Fp), where Σp : KMF(Xp, Wp) → Dsg((X0)p) is the exact functor defined as above. Since Ap 6= 0 and Σp is fully faithful, Fp 6= 0 and hence p ∈ Supp(F ). Conversely, if p ∈ Supp(F ) for some F ∈ DMF(X, L, W ), then Fp 6= 0 ∈ KMF(Xp, Wp). Since Σp : KMF(Xp, Wp) → Dsg((X0)p) is fully faithful, Σ(F )p ∼= Σp(Fp) 6= 0, and so Σ(F )p /∈ Perf(X0)p. Furthermore, j∗(Σ(F )) ∈ Perf X by Theorem 3.4. Hence p ∈ Sing(X0/X). Lemma 3.6. Let (R, m) be a local ring, and let W ∈ R be an element. The category KMF(R, W ) has a non-zero object if and only if W ∈ m2. Proof. Assume that W ∈ m2. Then there exist non-units mi, ni ∈ m for 1 ≤ i ≤ r such that i=1 Ki ∈ KMF(R, W ). Then we claim that K 6= 0 in KMF(R, W ). Indeed, if K = 0 in KMF(R, W ), there are morphisms h0 : K0 → K1, h1 : K1 → K0 such that idK0 = ϕK is a matrix whose entries are non-units in R, the equation implies that 1R ∈ m, which is a contradiction. For the converse, let F ∈ KMF(R, W ) is a non-zero object. Since R is local, every locally free modules are free modules. Hence each ϕF m,n)1≤m,n≤r in elements in R. i We claim that F is isomorphic to a matrix factorization F ′ such that all entries of matrices ϕF ′ i are non-units. Indeed, if there is a unit entry u ∈ R in the square matrix ϕF m,n)1≤m,n≤r, applying elementary row/column operations, we may assume that u = f 1 1,m = 0 1,m = 0 for m 6= 1. Hence the object(cid:16)R u−→ R u−1W−−−−→ R(cid:17) is for m 6= 1. Then we see that f 0 a direct summand of F , but this is isomorphic to the zero object in KMF(R, W ). Since the rank of matrices ϕF i are finite and F 6= 0, repeating this process, we may assume that all entries of ϕF are non-units, and hence W ∈ m2 since W = ϕF ni−→ R) ∈ KMF(R, mini) and K :=Nr 1 h0 + h1ϕK 0 . Since each ϕK i is a r-square matrix (f i mi−−→ R 1 = (f 1 1,1 and f 1 m,1 = f 1 m,1 = f 0 0 ϕF 1 . (cid:3) i (cid:3) The following result is useful to compute the relative singular loci of zero schemes of regular sections of line bundles. Proposition 3.7. Notation is same as above. We have Singloc(X0/X) = { p ∈ X Wp ∈ m2 p } Proof. This follows from Proposition 3.5.(2) and Lemma 3.6. Remark 3.8. Recall that the critical locus Crit(ϕ) of a function ϕ ∈ Γ(Y,OY ) on a complex analytic space (Y,OY ) is defined by (cid:3) Crit(ϕ) := { p ∈ Y ϕp − ϕ(p) ∈ m2 p }, where mp is the maximal ideal of the local ring OY,p. Assume that X is a quasi-projective variety over C, and let f ∈ Γ(X,OX ) be a regular function which is a non-zero-divisor. Denote by 12 Y. HIRANO X ) the complex analytic space associated to X, and let f an ∈ Γ(X an,Oan (X an,Oan associated to f . For any p ∈ X an, since the morphism of local rings ϕp : OX,p → Oan ϕp induces the isomorphism mp/m2 p )2. Hence Proposition 3.7 implies the following equality of sets; and only if f an p )2 (see [Mat1, Theorem 49]), and so fp ∈ m2 X ) is the function X,p is flat [Ser], p if p ∼= man p /(man p ∈ (man where Zero(f an) is the zero locus of f an. Singloc(f −1(0)/X) ∩ X an = Crit(f an) ∩ Zero(f an), Recall that the codimension of a Noetherian local ring (R, m) is defined by codim(R) := emb. dim(R) − dim(R), where emb. dim(R) := dimR/m(m/m2) is the dimension of R/m-vector space of m/m2, which is called the embedding dimension of (R, m). The following result provides a numerical charac- terization of Sing(X0/X). Proposition 3.9. Assume that for any p ∈ X0 we have dim(OX,p) − dim(OX0,p) = 1. Then we have the following equality of sets; Singloc(X0/X) = { p ∈ X0 codim(OX0,p) > codim(OX,p)}. Proof. Let p ∈ X0 be a point, and denote by mp ⊂ OX,p and np ⊂ OX0,p the maximal ideals of OX,p and OX0,p respectively. The surjective homomorphism π : OX,p ։ OX0,p induces a surjective map π : mp/m2 p ։ np/n2 p. Hence we have Moreover, since dim(OX,p) − dim(OX0,p) = 1, we have codim(OX0,p) > codim(OX,p) if and only if emb. dim(OX0,p) ≥ emb. dim(OX,p). Hence, by Proposition 3.7, it is enough to show that emb. dim(OX0,p) ≤ emb. dim(OX,p). For (⇒), assume that Wp ∈ m2 there exists y ∈ OX,p such that Wp ∈ m2 p ⇔ π is injective. p, and let x ∈ mp with π(x) ∈ n2 p. Then, since OX0,p ∼= OX,p/hWpi, Hence x ∈ m2 π is injective. Since p ∈ X0, Wp ∈ mp. Since π(Wp) = 0 and π is injective, we have Wp ∈ m2 x + yWp ∈ m2 p. p. This means that π is injective. For (⇐), assume that p. (cid:3) p by the assumption Wp ∈ m2 Next, we show some properties describing relationships between relative singular loci and locally relative singular loci. Proposition 3.10. Let R be a Noetherian ring, and let W ∈ R is a non-zero-divisor. Set X := Spec R and X0 := Spec(R/W ). (1) We have Sing(X0/X) ∩ Max(R/W ) = Singloc(X0/X) ∩ Max(R/W ). (2) Let p ∈ X0 be a point with p /∈ Singloc(X0/X). Let q ∈ X0 be a point and denote by {q} the closure of q. If p ∈ {q}, then q /∈ Sing(X0/X). Proof. (1) The inclusion Sing(X0/X) ∩ Max(R/W ) ⊆ Singloc(X0/X) ∩ Max(R/W ) follows from Proposition 3.2. To show the opposite inclusion, let m ∈ Singloc(X0/X)∩Max(R/W ) be a maximal ideal of R/W which is contained in Singloc(X0/X), and let m ∈ Max R be the maximal ideal of R such that i(m) = m, where i : X0 ֒→ X is the closed immersion. By Proposition 3.7, there exist elements mi, ni, r ∈ R (1 ≤ i ≤ r) such that mi ∈ m, ni ∈ m, r /∈ m, and rW =Pr i=1 mini. Since r /∈ m and m is maximal, we have hri + m = R, and so there exists an element a ∈ R such that 1 − ar ∈ m. Then we have W = (1 − ar)W + amini. rXi=1 13 Consider the following matrix factorizations K0 :=(cid:16)R W−→ R 1−ar−−−→ R(cid:17) ∈ KMF(R, (1 − ar)W ) ni−→ R(cid:17) ∈ KMF(R, amini) Ki :=(cid:16)R rOi=0 Ki ∈ KMF(R, W ). ami−−→ R K := Then, by Lemma 2.23 and Lemma 2.24, we see that m ∈Tr i=0 Supp(Ki) = Supp(K). Proposition 3.5.(1) implies that m ∈ Sing(X0/X). (2) Assume that q ∈ Sing(X0/X). Since the relative singular locus Sing(X0/X) is a union of closed subsets by Proposition 3.5.(1), we have {q} ⊆ Sing(X0/X). Since p ∈ {q} and Sing(X0/X) ⊆ Singloc(X0/X), p ∈ Singloc(X0/X), which contradicts to the assumption of p. (cid:3) At the end of this section, we compute examples of relative singular loci using the above results. Example 3.11. We give two examples of the relative singular loci which are not equal to the usual singular loci. (1) Let R := C[x, y]/hxni for n > 1, and let W := y ∈ R. Set X := Spec R and X0 := Spec(R/W ). Although Sing(X0) = {pt} 6= ∅, by Proposition 3.7, we have Sing(X0/X) = Singloc(X0/X) = ∅. (2) Let R := C[x, y, z, w]/hxy − zwi, and let W := w ∈ R. Set X := Spec R and X0 := Spec(R/W ). Then we have By Proposition 3.7, we see that Sing(X0) = { hx, yi,hx, y, z − ai a ∈ C}. and, by Proposition 3.10, we have Singloc(X0/X) = Sing(X0) \ {hx, y, zi}, In this example, all kinds of singular loci are different; Sing(X0/X) = Singloc(X0/X) \ {hx, yi}. Sing(X0/X) ( Singloc(X0/X) ( Sing(X0). 4. Tensor nilpotence properties In this section, we prove the tensor nilpotent properties, which will be necessary for our main result. The properties are analogous to [Tho, Theorem 3.6, 3.8], and the strategy of the proof is similar to loc. cit. Let X be a Noetherian scheme, and let W ∈ Γ(X, L) be a global section of a line bundle L on X. 4.1. Mayer-Vietoris sequence. We provide a Mayer-Vietoris sequence for factorizations for the proof of the tensor nilpotence properties in the next section. For an open immersion i : U ֒→ X, consider an induced LG model (U, LU , WU ). Let i! : ModOU → ModOX be the extension by ϕF 1−−→ F0 ϕF 0−−→ F1 ⊗ LU(cid:17) ∈ Sh(U, LU , WU ), we define an object zero. For an object F = (cid:16)F1 i!(F ) ∈ Sh(X, L, W ) by where σ : i!(F1 ⊗ LU ) i!(F ) :=(cid:16)i!(F1) ∼−→ i!(F1) ⊗ L is a natural isomorphism. This defines an exact functor −−−−−−→ i!(F1) ⊗ L(cid:17), i!(ϕF −−−−→ i!(F0) σ◦i!(ϕF 0 ) 1 ) Similarly, the inverse image functor i∗ : ModOX → ModOU defines an exact functor These functors induce an exact functors between homotopy categories; i! : Sh(U, LU , WU ) → Sh(X, L, W ). i∗ : Sh(X, L, W ) → Sh(U, LU , WU ). i! : KSh(U, LU , WU ) → KSh(X, L, W ) 14 Y. HIRANO Since both of i! and i∗ are exact functors and preserve arbitrary direct sums, these functors defines exact functors between coderived categories i∗ : KSh(X, L, W ) → KSh(U, LU , WU ). i! : DcoSh(U, LU , WU ) → DcoSh(X, L, W ) i∗ : DcoSh(X, L, W ) → DcoSh(U, LU , WU ). Adjunction i! ⊣ i∗ of functors between ModOU and ModOX naturally induces adjunction i! ⊣ i∗ of functors between coderived categories DcoSh(U, LU , WU ) and DcoSh(X, L, W ). Lemma 4.1 (Mayer-Vietoris). Let U1 and U2 be open subschemas of X, and suppose X = U1∪U2. Denote by U1,2 the intersection U1 ∩ U2. Let il : Ul → X and i1,2 : U1,2 → X be open immersions. Then, for any objects E, F ∈ DcoSh(X, L, W ) and for any k ∈ Z, we have the following long exact sequence: HomU1,2 (i∗ 1,2E, i∗ 1,2F [k − 1]) HomU1,2 (i∗ 1,2E, i∗ 1,2F [k]) δk−1 HomX (E, F [k]) δk HomX (E, F [k + 1]) l=1 HomUl(i∗ l E, i∗ l F [k]) l=1 HomUl(i∗ l E, i∗ l F [k + 1]) L2 L2 where Hom(−) denotes the set of morphisms in DcoSh(−, L(−), W(−)). Proof. For each components Ej (j = 0, 1), we have the following exact sequence in ModOX : 0 → i1,2!i∗ 1,2Ej → i1!i∗ 1Ej ⊕ i2!i∗ 2Ej → Ej → 0. These exact sequences induce the following exact sequence in the abelian category Sh(X, L, W ): By the same argument as in [LS, Lemma 2.7.(a)], the above exact sequence gives rise to a triangle 0 → i1,2!i∗ 1,2E → i1!i∗ 1E ⊕ i2!i∗ 2E → E → 0. (∗) i1,2!i∗ 1,2E → i1!i∗ 1E ⊕ i2!i∗ 2E → E → i1,2!i∗ 1,2E[1] ∗ for open immersions i(−), we obtain the result. in DcoSh(X, L, W ). By applying HomDcoSh(X,L,W )(−, F [k]) to the triangle and using the adjunction i(−)! ⊣ i(−) 4.2. Tensor nilpotence properties. The following lemma is an analogy of [Tho, Theorem 3.6], and we show it by a similar argument in the proof of loc. cit. Lemma 4.2. Let f : E → F be a morphism in DMF(X, L, W ). If f ⊗ k(p) = 0 in DMF(k(p), W ⊗ k(p)) for any p ∈ X, then there is an integer n such that f ⊗n = 0 in DMF(X, L, nW ). Proof. The proof will be divided into 5 steps. (cid:3) have a finite number of open affine covering Sk for k − 1. Set V :=Sk Step 1. In the first step, we will reduce to the case that X is affine. Since X is Noetherian, we i=1 Ui of X, where Ui = Spec Ri. Set Li := LUi and Wi := WUi . We will show that, if there are positive integers ni such that (fUi )⊗ni = 0 ∈ KMF(Ri, Li, niWi) for every i, there is a positive integer n such that f ⊗n = 0 ∈ DMF(X, L, nW ). We will do this by induction on k. If k = 1, then f ⊗n1 = 0. For k > 1, suppose that the result is true = 0 in DMF(V, LV , n′WV ). If we set m := max{n1, n′}, we have f ⊗mV = 0 and f ⊗mU1 = 0. By Lemma 4.1, there exists a morphism g : j∗E⊗m[1] → j∗F ⊗m in DcoSh(U1 ∩ V, LU1∩V , WU1∩V ) such that the following diagram in DcoSh(X, L, W ) is commutative: i=2 Ui. Then by induction hypothesis, there exists n′ such that (fV )⊗n′ E⊗m δ f ⊗m F ⊗m ε j!(j∗E⊗m[1]) j!(g) / j!j∗F ⊗m where j : U1 ∩ V → X is the open immersion, δ is the morphism corresponding to the third morphism in the above triangle (∗) in the proof of Lemma 4.1, and ε is the adjunction morphism of j! ⊣ j∗. Since f ⊗m ⊗ j!(g) is identified with j!(j∗(f ⊗m) ⊗ g) via natural isomorphisms, we have f ⊗m ⊗ j!(g) = 0. Hence for n := 2m, it follows that f ⊗n = 0 in DcoSh(X, L, nW ), as f ⊗n factors through f ⊗m ⊗ j!(g). By Lemma 2.14, we obtain f ⊗n = 0 in DMF(X, L, nW ). / / / O O 15 Step 2. By the above step, we may assume that X = Spec R is affine and L is an invertible In this step, we R-module. Recall that DMF(R, L, W ) ∼= KMF(R, L, W ) by Proposition 2.12. reduce to the case when W = 0 and E = (0 → R → 0). We have the following adjunction Φ : HomKMF(R,L,W )(E, F ) ∼−→ HomKMF(R,L,0)(R, E∨ ⊗ F ). Via natural isomorphisms, we can identify Φ(f ⊗n) with Φ(f )⊗n and Φ(f⊗k(p)) with Φ(f ⊗n)⊗k(p) respectively. Therefore, we may assume that W = 0 and E = R. Step 3. Since the components Fi of F and L are locally free, by the above first step, shrinking X = Spec R if necessary, we may assume that X = Spec R is affine scheme such that each Fi is free R-module and L ∼= R. Furthermore, by the second step, we may assume that E = R and W = 0. Hence, since the natural isomorphism HomMF(R,0)(R, F ) ∼= Ker(ϕF 0 ) induces the isomorphism HomKMF(R,0)(R, F ) ∼= H0(F ), we only need to show that if f ∈ H0(F ) satisfies f ⊗ k(p) = 0 ∈ H0(F ⊗ k(p)) for any p ∈ Spec R, then f ⊗n = 0 ∈ H0(F ⊗n) for some n > 0. In this step, we reduce to the case when the Noetherian ring R is of finite Krull dimension. Since the components Fi of F are free R-modules, the morphisms ϕF i can be represented by a matrices whose entries are elements in R. Let {Rα}α∈A be the family of all subrings Rα ⊂ R of R such that dim Rα < ∞ and Rα contains all entries of matrices ϕ1 and ϕ0. Then for any α ∈ A, there is the natural object Fα ∈ KMF(Rα, 0) such that its components (Fα)i are free Rα-modules and α(Fα) ∼= F in the additive category MF(R, 0), where πα : Spec R → Spec Rα is the morphism π∗ induced by the inclusion Rα ⊂ R. Let F • := (··· 0 → F1 α := (··· 0 → (Fα)1 F • ϕF ϕF 1−−→ F0 1−−−→ (Fα)0 0−−→ F1 → 0 ··· ) 0−−−→ (Fα)1 → 0 ··· ) ϕFα ϕFα be the complexes of free modules such that the term F0 and (Fα)0 are of degree 0. Then we have H 0(F •) = H0(F ) and H 0(F •⊗k(p)) = H0(F⊗k(p)) for any p ∈ Spec R. Since R is the direct colimit of the system {Rα}α∈A; R = lim−→ Rα, by the same argument as in the step (3.6.4) in the proof of [Tho, Theirem 3.6], if f ⊗ k(p) = 0 in H0(F ⊗ k(p)) for any p ∈ X, there exist β ∈ A and an element fβ ∈ H 0(F • β ⊗ k(p)) = H0(Fβ ⊗ k(p)) for any p ∈ Spec Rβ. Therefore, if the assertion is true for X = Spec Rβ, there exists n > 0 such that f ⊗n β ) = 0 in H0(F ⊗n). This completes the reduction to the case when the ring R is of finite Krull dimension. β(fβ) = f and fβ ⊗ k(p) = 0 in H 0(F • β ) = H0(Fβ) such that π∗ β ), and then f ⊗n = π∗ β = 0 in H0(F ⊗n β(f ⊗n ) = F ⊗n′ = 0 ∈ H0((h∗F )⊗n′ ). Since (h∗F )⊗n′ ∼= h∗(F ⊗n′ Step 4. In this step, we reduce to the case when the Noetherian ring R of finite Krull dimension is reduced. Let N ⊂ R be the ideal of nilpotent elements in R, and denote by h : Spec R/N → Spec R If f ⊗ k(p) = 0 for any p ∈ Spec R, then h∗(f ) ⊗ k(q) = 0 for any be the closed immersion. q ∈ Spec R/N. Hence, for the reduction, we claim that, if h∗(f )⊗n′ ) for some n′ > 0, then there exists n > 0 such that f ⊗n = 0 ∈ H0(F ⊗n). Assume that h∗(f )⊗n′ = 0 ∈ H0((h∗F )⊗n′ ⊗R R/N, the assumption implies that there exist elements x ∈ (F ⊗n′ (x) + y in (F ⊗n′ R-module(cid:0)(F ⊗n′ )0(cid:1)⊗m )0(cid:1), there is a positive integer m such that y⊗m = 0 in the free Let S be a ring, and let E ∈ MF(S, 0) be an object such that its components Ei are free S- modules. For e ∈ Ker(ϕE 0 ) ⊂ E0, suppose that there are elements u ∈ E1 and v ∈ E0 such that 1 (u) + v and v⊗n = 0 in the S-free module (E0)⊗n for some n > 0. Then, considering e = ϕE e⊗n as an element in (E⊗n)0 via the natural split mono (E0)⊗n ֒→ (E⊗n)0, there is an element w ∈ (E⊗n)1 such that ϕE⊗n The element e⊗n = (ϕE (w) = e⊗n in (E⊗n)0. In particular, e⊗n = 0 ∈ H0(E⊗n). )0. Since y ∈ N(cid:0)(F ⊗n′ )1 and y ∈ N(cid:0)(F ⊗n′ . Therefore, it is enough to show the following claim: )0(cid:1) ⊂ (F ⊗n′ )0 such that f ⊗n′ = ϕF ⊗n′ 1 1 (u) + v)⊗n ∈ (E0)⊗n can be decomposed into the following form 1 (u) ⊗ w1 + v⊗n−1 ⊗ ϕE 1 (u) ⊗ wn−2 + ··· + v⊗n−2 ⊗ ϕE 1 (u), e⊗n = ϕE 1 (u) ⊗ wn−1 + v ⊗ ϕE where wi is an element in (E0)⊗i. For an ordered sequence (i1, i2, ..., in) of ik ∈ {0, 1}, set E(i1,i2,...,in) := Ei1 ⊗ Ei2 ⊗ ··· ⊗ Ein , and set 1 eE := E(1,0,0,...,0) ⊕ E(0,1,0,...,0) ⊕ ··· ⊕ E(0,...,0,1). 16 Y. HIRANO Let ew := (u ⊗ wn−1) ⊕ (v ⊗ u ⊗ wn−2) ⊕ ··· ⊕ (v⊗n−2 ⊗ u ⊗ w1) ⊕ (v⊗n−1 ⊗ u) ∈ eE, and let w := ι(ew) ∈ (E⊗n)1 be the image of ew under the natural split mono ι : eE ֒→ (E⊗n)1. Since 1 (u)) = 0, and each wi ∈ (E0)⊗i is a summation of elements of (w) = e⊗n 1 (u) or v, we obtain an equality ϕE⊗n 0 (v) = ϕE 1 (u)) = 0, ϕE ϕE 0 (ϕE the form a1 ⊗ a2 ⊗ ··· ⊗ ai where ak are either ϕE in (E⊗n)0. This completes the proof of the claim. 0 (e − ϕE 1 H0(F ⊗ k(p)) for any p ∈ X, then f = 0 in H0(F ). Consider a case when d > 0, and assume that the result holds for Noetherian rings of dimension less than d. Denote by Min R the finite set of all prime ideals of R of height zero. Then the Step 5. Now we may assume that X = Spec R is reduced affine scheme of finite Krull dimension, the components Fi of F are free R-modules, L ∼= R, W = 0, and E = R ∈ KMF(R, 0). In this step, we finish the proof by induction on the Krull dimension d := dim R of R. If d = 0, then R ∼= Lp∈X k(p) and H0(F ) ∼= Lp∈X H0(F ⊗ k(p)). Hence, if f ⊗ k(p) = 0 ∈ productQp∈Min R k(p) of residue fields is isomorphic to the localization S−1R for the set S of all non zero-divisors in R, as the residue fields k(p) is equal to the local ring Rp for any p ∈ Min R. By hypothesis, f⊗k(p) = 0 ∈ H0(F ⊗k(p)) for any p ∈ Min R, hence f⊗RS−1R = 0 ∈ H0(F ⊗RS−1R). This means that, for a representative f ∈ F0 of the equivalence class f ∈ H0(F ), there exist 1 (y) in F0. Set Ks := (R s−→ R 0−→ R) ∈ MF(R, 0) elements y ∈ F1 and s ∈ S such that sf = ϕF and let γ := (y, f ) : Ks → F be the morphism defined as the pair of morphisms y : R → F1 and f : R → F0 of R-modules. Denote by i : Spec R/s → Spec R the natural closed immersion. The canonical quotient R → i∗(R/s) naturally defines morphisms δ : Ks → (0 → i∗(R/s) → 0) and α : (0 → R → 0) → (0 → i∗(R/s) → 0) in coh(R, 0). Since s : R → R is injective, we have an exact sequence 0 →(cid:16)R id−→ R 0−→ R(cid:17) (id,s) −−−→ Ks δ−→(cid:16)0 → i∗(R/s) → 0(cid:17) → 0 in coh(R, 0). Since (R id−→ R 0−→ R) is zero in Dcoh(R, 0), δ is an isomorphism in Dcoh(R, 0) by [LS, Lemma 2.7.(a)]. Set β : (0 → i∗(R/s) → 0) → F be the composition γ ◦ δ−1 in Dcoh(R, 0). Then the composition β ◦ α : (0 → R → 0) → F is equal to f : (0 → R → 0) → F in Dcoh(R, 0), since α = δ ◦ ι and f = δ ◦ ι in coh(R, 0), where ι : (0 → R → 0) → Ks is the morphism such that ι1 = 0 and ι0 = idR. Hence, for any n > 0, we have the following commutative diagram in Dcoh(R, 0): ∼ R R⊗n ⊗ R 1⊗α / R⊗n ⊗ i∗(R/s) ∼ / / i∗(R/s⊗n) f ⊗n⊗1 &▼▼▼▼▼▼▼▼▼▼ xqqqqqqqqqqq 1⊗f F ⊗n ⊗ R 1⊗α 'PPPPPPPPPPPP f ⊗n+1 f ⊗n⊗f f ⊗n⊗1 i∗(i∗f ⊗n) F ⊗n+1 ∼ / / F ⊗n ⊗ F 1⊗β F ⊗n ⊗ i∗(R/s) ∼ / / i∗(i∗F ⊗n) where R = (0 → R → 0) and i∗(R/s) = (0 → i∗(R/s) → 0) by abuse of notation. Since dim R/s < dim R and i∗f ⊗R/s k(p) = f ⊗R k(p) = 0 in H0(i∗F ⊗k(p)), by the induction hypothesis, there exists m > 0 such that i∗f ⊗m = 0 in KMF(R, 0), in particular i∗f ⊗m = 0 in Dcoh(R, 0). By the above commutative diagram, we see that f ⊗m+1 factors through i∗(i∗f ⊗m) in Dcoh(R, 0), and hence f ⊗m+1 = 0 in Dcoh(R, 0). Then, by Lemma 2.14.(5), f ⊗m+1 = 0 in KMF(R, 0), and this completes the proof. (cid:3) The following lemma is a consequence of the above lemma. Lemma 4.3. Let a : E → F be a morphism in DMF(X, L, 0), and let G ∈ DMF(X, L, W ) be an object. If a ⊗ k(p) = 0 in DMF(k(p), 0) for all p ∈ Supp(G), then there is an integer n > 0 such that G ⊗ (a⊗n) = 0 in DMF(X, L, W ). Proof. Since (G ⊗ a) ⊗ k(p) = 0 in DMF(k(p), W ⊗ k(p)) for any p ∈ X, by Lemma 4.2, there is a positive integer n > 0 such that (G ⊗ a)⊗n = G⊗n ⊗ a⊗n = 0 in DMF(X, L, nW ). Hence ((G∨)⊗n−1 ⊗ G⊗n) ⊗ a⊗n = 0, and so it suffices to show that G ⊗ a⊗n is a retract of ((G∨)⊗n−1 ⊗ G⊗n) ⊗ a⊗n in DMF(X, L, W ). / / & / x ' o o 17 We will show it by proving that G is a direct summand of (G∨)⊗n−1 ⊗ G⊗n in DMF(X, L, W ) by induction on n. The n = 1 case is trivial. For n ≥ 2, assume that G is a direct summand It is enough to show that (G∨)⊗n−2 ⊗ G⊗n−1 is a direct summand of of (G∨)⊗n−2 ⊗ G⊗n−1. (G∨)⊗n−1⊗G⊗n. But for n ≥ 3 case, this follows from n = 2 case by tensoring (G∨)⊗n−3⊗G⊗n−2. Therefore it suffices to prove that G is a direct summand of G∨⊗G⊗2. The tensor product (−)⊗G : DMF(X, L, 0) → DMF(X, L, W ) is left adjoint to (−) ⊗ G∨ : DMF(X, L, W ) → DMF(X, L, 0). Let η : id → (−) ⊗ G ⊗ G∨ be its adjunction morphism. Then the functor morphism is a split mono. Evaluating OX ∈ DMF(X, L, 0), we see that G is a direct summand of G∨ ⊗ G⊗2. (cid:3) (−) ⊗ G η⊗G −−−→ (−) ⊗ G ⊗ G∨ ⊗ G 5. Classification of thick subcategories of DMF(X, L, W ) In this section, we prove our main result. Let X be a separated Noetherian scheme, and let W ∈ Γ(X, L) be any section of a line bundle L on X. At first, following [TT], we recall the definition of ample families of line bundles. Definition 5.1. We say a quasi-compact and quasi-separated scheme S has an ample family of line bundles if there exists a family L := {Lα}α∈A of line bundles on S such that the family {Sf f ∈ Γ(S, L⊗n α ), Lα ∈ L, n > 0} of open subsets form an open basis of S, where Sf := { p ∈ S f (p) 6= 0}. Remark 5.2. (1) Any scheme with an ample line bundle has an ample family of line bundles. In particular, any affine scheme has an ample family of line bundles. Any separated regular Noetherian scheme has an ample family of line bundles. See [TT, Example 2.1.2] for more examples of schemes with ample families of line bundles. (2) If S has an ample family of line bundles, then S satisfies the resolution property by [TT, Lemma 2.1.3]. Proposition 5.3. Let Z ⊆ X be a closed subset of X. Assume that X has an ample family of line bundles. Then, the following holds. (1) There exists a matrix factorization K ∈ DMF(X, L, 0) such that Supp(K) = Z. (2) If W is a non-zero-divisor and Z is contained in Sing(X0/X), then there exists a matrix factorization F ∈ DMF(X, L, W ) such that Supp(F ) = Z. Ki :=(cid:16)OX Proof. (1) Since X has an ample family of line bundles, there are finitely many sections fi ∈ i=1 Z(fi) as closed subsets of X, where Z(fi) is the zero scheme of fi. Let Ki be the object in DMF(X, L, 0) of the following form Γ(X, Li) of line bundles Li (1 ≤ i ≤ l) such that Z =Tl 0−→ L(cid:17). fi−→ Li By Proposition 2.20.(1) and Lemma 2.23, we have Supp(Ki) = Z(fi). If we set K := Nl then K ∈ DMF(X, L, 0) and Supp(K) =T Supp(Ki) = Z. Z = Sr Supp(Ki ⊗ Ei) = Supp(Ki) ∩ Supp(Ei). If we set F :=Lr (2) Since X is Noetherian, we can decompose Z into finitely many irreducible components I=1 Zi. Then, for each 1 ≤ i ≤ r, there is a unique generic point pi ∈ Zi of Zi. By Proposition 3.5, there exists a matrix factorization Ei ∈ DMF(X, L, W ) such that pi ∈ Supp(Ei). Then Zi ⊆ Supp(Ei), since pi is a generic point of Zi. By (1), there are matrix factorizations Ki ∈ DMF(X, L, 0), for 1 ≤ i ≤ r, such that Supp(Ki) = Zi. Note that Lemma 2.24 implies the equality i=1(Ki ⊗ Ei), then F ∈ DMF(X, L, W ) and we have i=1 Ki, Supp(F ) = This completes the proof. r[i=1 Supp(Ki ⊗ Ei) = (Supp(Ki) ∩ Supp(Ei)) = Z. r[i=1 (cid:3) 18 Y. HIRANO Definition 5.4. Let T ⊂ DMF(X, L, W ) be a triangulated full subcategory. We say that T is ⊗-submodule if it is closed under tensor action of DMF(X, L, 0), i.e. for any F ∈ DMF(X, L, 0) and any T ∈ T , we have F ⊗ T ∈ T . For an object F ∈ DMF(X, L, W ), we denote by hFi⊗ ⊂ DMF(X, L, W ) the smallest thick ⊗-submodule containing F . We prove the following proposition by using tensor nilpotence properties in the previous section. Proposition 5.5. For E, F ∈ DMF(X, L, W ), if Supp(E) ⊆ Supp(F ), then E ∈ hFi⊗. Proof. Let f : OX → F ∨ ⊗ F be the adjunction morphism in DMF(X, L, 0) induced by the adjoint pair (−) ⊗ F ⊣ Hom(F,−). Set G := C(f )[−1], and let a : G → OX be a morphism which completes the following distinguished triangle Then, since hFi⊗ is closed under DMF(X, L, 0)-action, E ⊗ C(a) ∼= (E ⊗ F ∨) ⊗ F ∈ hFi⊗. We claim that for any n > 0, E ⊗ C(a⊗n) ∈ hFi⊗. Indeed, consider the following diagram G a−→ OX f −→ F ∨ ⊗ F → G[1]. G⊗n+1 G⊗(a⊗n) (cid:8) a⊗n+1 G⊗n+1 G a / OX C(a) G[1] G⊗bn G ⊗ C(a⊗n) / G⊗n+1[1] bn+1 C(a⊗n+1) / G⊗n+1 C(a) G⊗bn[1] / G ⊗ C(a⊗n)[1] where the top horizontal sequence is the distinguished triangle obtained by tensoring G with the following triangle (∗) G⊗n a⊗n −−→ OX bn−→ C(a⊗n) → G⊗n[1]. Then, by the octahedral axiom, we obtain the triangle completing the vertical sequence on the right side in the above diagram G ⊗ C(a⊗n) → C(a⊗n+1) → C(a) → G ⊗ C(a⊗n)[1]. Considering the triangle obtained by tensoring this triangle with E, we can prove that E⊗C(a⊗n) ∈ hFi⊗ by induction on n. Tensoring E with the above triangle (∗) for any n > 0, we have a triangle −−−−−→ E −→ E ⊗ C(a⊗n) −→ E ⊗ G⊗n[1]. E ⊗ G⊗n E⊗a⊗n If E ⊗ a⊗n = 0 for some n > 0, then E is a direct summand of E ⊗ C(a⊗n) ∈ hFi⊗, which implies that E ∈ hFi⊗. Hence it suffices to show that there is an integer n > 0 such that E ⊗ a⊗n = 0. By Lemma 4.3, it is enough to show that for any p ∈ Supp(E), a ⊗ k(p) = 0 in KMF(k(p), 0). Since p ∈ Supp(E) ⊆ Supp(F ), p ∈ Supp(F ∨⊗F ). By Lemma 2.22, F ∨⊗F⊗k(p) ∼= Homk(p)(F ⊗ k(p), F ⊗ k(p)) 6= 0 in KMF(k(p), 0). Hence the natural map g k(p) −→ Homk(p)(F ⊗ k(p), F ⊗ k(p)) is a split mono by Corollary 2.17, since it is non-zero map. Since f ⊗ k(p) : k(p) → F ∨⊗ F ⊗ k(p) is equal to the composition of g : k(p) → Homk(p)(F ⊗ k(p), F ⊗ k(p)) and the natural isomorphism Homk(p)(F ⊗ k(p), F ⊗ k(p)) ∼= F ∨ ⊗ F ⊗ k(p), f ⊗ k(p) is also a split mono. Hence the triangle G ⊗ k(p) a⊗k(p) −−−−→ k(p) f ⊗k(p) −−−−→ F ∨ ⊗ F ⊗ k(p) → G ⊗ k(p)[1] implies that a ⊗ k(p) = 0. (cid:3) / / / / ✤ ✤ ✤ / / / / ✤ ✤ ✤ / ✤ ✤ ✤ / 19 Now we are ready to prove the following main result. Recall that a subset S ⊆ T of a topological space T is called specialization-closed if it is a union of closed subsets of T . We easily see that S is specialization-closed if and only if s ∈ S implies {s} ⊆ S. Theorem 5.6. Let X be a separated Noetherian scheme with an ample family of line bundles, L be a line bundle on X, and W ∈ Γ(X, L) be a non-zero-divisor. There is one-to-one correspondence: subsets of Sing(X0/X)(cid:27) σ−→ (cid:26) specialization-closed of DMF(X, L, W ) (cid:27) τ←−(cid:26) thick ⊗ -submodules The bijective map σ sends Y to the thick subcategory consisting of matrix factorizations F ∈ DMF(X, L, W ) with Supp(F ) ⊆ Y . The inverse bijection τ sends T to the unionSF ∈T Supp(F ). Proof. The map σ is well defined since for any E ∈ DMF(X, L, 0) and F ∈ DMF(X, L, W ), we have Supp(E ⊗ F ) ⊆ Supp(E) ∩ Supp(F ) by Lemma 2.24. The map τ is also well defined by Proposition 2.20.(2) and Proposition 3.5.(1). We show that σ and τ are mutually inverse. Let Y be a specialization-closed subset of Sing(X0/X), and let T be a thick ⊗-submodule of DMF(X, L, W ). By construction, we have τ (σ(Y )) ⊆ Y and T ⊆ σ(τ (T )). Hence it is enough to show the inclusions Y ⊆ τ (σ(Y )) and σ(τ (T )) ⊆ T . Since Sing(X0/X) is a specialization-closed subset of X by Proposition 3.5.(1), Y is specialization- Proposition 5.3.(2), for each Yλ, there exists Fλ ∈ DMF(X, L, W ) with Supp(Fλ) = Yλ. Since Fλ ∈ σ(Y ), we have Yλ = Supp(Fλ) ⊆ τ (σ(Y )). Hence Y ⊆ τ (σ(Y )). To finish the proof, we show that σ(τ (T )) ⊆ T . Let F ∈ σ(τ (T )) be an object. Then, by closed in X. Hence Y can be described as a union of closed subsets Yλ of X; Y = S Yλ. By construction, we have Supp(F ) ⊆ STλ∈T Supp(Tλ). Then, as in the proof of [Tho, Theorem 3.15], there is a finite set {Tλ}λ∈Λ of objects in T such that Supp(F ) ⊆Sλ∈Λ Supp(Tλ). Hence Supp(F ) ⊆ Supp(⊕λ∈ΛTλ). Since ⊕λ∈ΛTλ ∈ T , it follows from Proposition 5.5 that F ∈ T . Definition 5.7. We say that an object G ∈ DMF(X, L, W ) is a ⊗-generator of DMF(X, L, W ) if hGi⊗ = DMF(X, L, W ). (cid:3) The following corollary says that the closedness of relative singular locus Sing(X0/X) in X0 is related to the existence of a ⊗-generator of DMF(X, L, W ). Corollary 5.8. Notation is same as in Theorem 5.6. The relative singular locus Sing(X0/X) is closed in X0 if and only if DMF(X, L, W ) has a ⊗-generator. Proof. Assume that the subset Sing(X0/X) is closed in X0. Since the relative singular locus Sing(X0/X) is the union of supports of all objects in DMF(X, L, W ) and X is Noetherian, there is a finite subset {Fi} of objects Fi ∈ DMF(X, L, W ) such that Sing(X0/X) = Si Supp(Fi) = Supp(⊕iFi). By Theorem 5.6, there is a specialization-closed subset Y ⊆ Sing(X0/X) such that σ(Y ) = h⊕iFii⊗. Since Sing(X0/X) = Supp(⊕iFi) ⊆ Y , we have Y = Sing(X0/X). Hence h⊕iFii⊗ = σ(Sing(X0/X)) = DMF(X, L, W ). If DMF(X, L, W ) has a ⊗-generator G, for any object F ∈ DMF(X, L, W ), we have Supp(F ) ⊆ Supp(G) by Lemma 2.25. Hence Proposition 3.5.(1) implies that Sing(X0/X) = Supp(G), and so Sing(X0/X) is closed in X0. (cid:3) Remark 5.9. Let (X, L, W ) be the same LG model as in Theorem 5.6. If X is regular and L is am- ple, any thick subcategory of DMF(X, L, W ) is automatically ⊗-submodule. In particular, the set on the right-hand side in Theorem 5.6 is equal to the set of thick subcategories of DMF(X, L, W ). Since this fact is proved by Stevenson in a different context [Ste], we do not include the proof here. Example 5.10. Let (X, W ) and X0 be same as in the Example 3.11.(2). Then Sing(X0/X) = {hx, y, z − ai a ∈ C \ {0}}. In this case, using the above results we see the following: (1) Since any point in Sing(X0/X) is a closed point, by Theorem 5.6 the set of thick ⊗-submodules of DMF(X, W ) is bijective to the set of arbitrary subsets of C \ {0}. (2) Since Sing(X0/X) is not a closed subset of X0, by Corollary 5.8 DMF(X, W ) does not have a ⊗-generator. In particular, DMF(X, W ) does not have a classical generator, i.e. there is no object F ∈ DMF(X, W ) such that hFi = DMF(X, W ). This was proved in [EP, Section 3.3] by a different argument. 20 Y. HIRANO 6. Tensor structures on matrix factorizations and its spectrum Using the classification result in the previous section, we will construct the relative singular loci from derived matrix factorization categories by considering tensor structures induced by tensor products. 6.1. Balmer's tensor triangular geometry. Following [Bal] and [Yu1, Chapter 4], we will recall some basic definitions and results of the theory of tensor triangular geometry. Definition 6.1. A pseudo tensor triangulated category (T ,⊗) consists of a triangulated category T and symmetric associated bifunctor ⊗ : T × T → T which is exact in each variable. For the precise definition, see [Yu1, Definition 4.1.1]. Remark 6.2. We don't assume that a pseudo tensor triangulated category has a unit 1⊗, and this is the only difference from the original definition of tensor triangulated categories in [Bal]. Definition 6.3. Let (T ,⊗) be a pseudo tensor triangulated category. (1) A thick subcategory I ⊂ T is called ⊗-ideal if the following implication holds: A ∈ T and B ∈ I ⇒ A ⊗ B ∈ I. (2) A ⊗-ideal P is called prime if the following holds (3) A ⊗-ideal I is called radical if √I = I, where √I is the radical of I, i.e. A /∈ P and B /∈ P ⇒ A ⊗ B /∈ P. √ I := { A ∈ T ∃ n ≥ 1 such that A⊗n ∈ I } For a pseudo tensor triangulated category (T ,⊗), we can consider the Zariski topology on the set of all prime ideals of (T ,⊗). Definition 6.4. The spectrum, denoted by Spc(T ,⊗), of (T ,⊗) is defined as the set of all prime ⊗-ideals Spc(T ,⊗) := { P P is a prime ⊗-ideal } The Zariski topology on Spc(T ,⊗) is defined by the collection of closed subsets of the form Z(S) := {P ∈ Spc(T ,⊗)S ∩ P = ∅} for any family of objects S ⊆ T . Definition 6.5. A support data on a pseudo tensor triangulated category (T ,⊗) is a pair (X, σ) of a topological space X and an assignment σ : ObT → {closed subsets of X} satisfying the following conditions: (1) σ(0) = ∅ andSA∈T σ(A) = X. (2) σ(A ⊕ B) = σ(A) ∪ σ(B). (3) σ(A[1]) = σ(A). (4) σ(A) ⊆ σ(B) ∪ σ(C) for any triangle A → B → C → A[1]. (5) σ(A ⊗ B) = σ(A) ∩ σ(B). defined by f (x) := {A ∈ T x /∈ σ(A)}. f : X ∼−→ Spc((T ,⊗)), We say that a support data (X, σ) is classifying if the following properties hold: (a) The topological space X is Noetherian and any non-empty irreducible closed subset has a unique generic point. (b) There is the following bijective correspondence: Θ :(cid:8) specialization-closed subsets of X(cid:9) ∼−→(cid:8) radical thick ⊗ -ideals(cid:9) defined by Θ(Y ) := {A ∈ T σ(A) ⊆ X}, with Θ−1(I) =SA∈I σ(A). Remark 6.6. Because of the lack of the unit 1⊗ in (T ,⊗) in our setting, we replace the con- dition σ(1⊗) = X in the original definition of support data in [Bal, Definition 3.1 (SD1)] with SA∈T σ(A) = X. The following result is essentially due to Balmer [Bal], and it is the key result for the result in the next subsection. See also [Yu1, Theorem 4.1.16]. Theorem 6.7 ([Bal, Theorem 5.2]). Assume that (X, σ) is a classifying support data on a pseudo tensor triangulated category (T ,⊗). Then we have the canonical homeomorphism 6.2. Construction of relative singular loci from matrix factorizations. In this section, using our classification result, we construct relative singular loci from pseudo tensor triangulated structures on derived matrix factorization categories. This kind of observation is also discussed in [Yu2]. Following [Yu2], we consider the natural pseudo tensor triangulated structure on derived matrix factorization categories. 21 Throughout this section, X is a separated Noetherian scheme with an ample family of line bundles, L is a line bundle on X, and W ∈ Γ(X, L) is a non zero-divisor. Denote by X0 the zero scheme of W , and assume that 2 ∈ Γ(X,OX ) is a unit of the ring Γ(X,OX ). For any unit λ ∈ Γ(X,OX )× in the ring Γ(X,OX ), we have a natural functor λ : DMF(X, L, W ) → DMF(X, L, λW ) defined by λ(cid:0)F1 ϕ1−→ F0 ϕ0−→ F1(cid:1) :=(cid:0)F1 ϕ1−→ F0 λϕ0−−→ F1(cid:1), and it is a triangulated equivalence (see [Yu1, Proposition 4.1.19]). Definition 6.8. Suppose that 2 ∈ Γ(X,OX ) is a unit. Then we define a bifunctor 1 ⊗ 2 : DMF(X, L, W ) × DMF(X, L, W ) → DMF(X, L, W ) 1 as the composition DMF(X, L, W ) × DMF(X, L, W ) ⊗−→ DMF(X, L, 2W ) 2−→ DMF(X, L, W ). triangulated category. As in [Yu1, Proposition 4.1.22], we see that the pair (DMF(X, L, W ),⊗ Consider an assignment 1 2 ) is a pseudo tensor Supp(−) : Ob(DMF(X, L, W )) → {specialization-closed subsets of Sing(X0/X)} defined by the supports of objects in DMF(X, L, W ). Theorem 6.9. (cid:0)Supp, Sing(X0/X)(cid:1) is a classifying support data on(cid:0)DMF(X, L, W ),⊗ 1 2(cid:1). 2 : DMF(X, L, 2W ) → DMF(X, L, W ) is an equivalence and commutes Proof. Since the functor 1 with taking stalks, we have Supp( (F )) = Supp(F ). 1 2 Therefore, it follows from Lemma 2.25 and Proposition 3.5 that(cid:0)Supp, Sing(X0/X)(cid:1) is a support data. We will show that the support data satisfies the conditions (a) and (b) in Definition 6.5. We check the condition (a). Note that X satisfies the condition (a). It follows that Sing(X0/X) is a Noetherian topological space, since so is X. Note that any irreducible closed subset Z of Sing(X0/X) is closed in X. Indeed, since Z is closed in a specialization-closed subset of X, Z is specialization-closed in X. Hence Z is a union of irreducible closed subsets Zλ of X; Z =Sλ Zλ. Since Z is irreducible in Sing(X0/X), there is an irreducible closed subset Zλ′ of X such that Z = Zλ′ . Hence Z is an irreducible closed subset of X, and it has a unique generic point. Next, we verify the condition (b). By Theorem 5.6, it is enough to show that for a thick subcategory T of DMF(X, L, W ) T is ⊗-submodule ⇔ T is radical ⊗ 1 2 -ideal. The implication (⇒) follows immediately from Theorem 5.6 and Lemma 2.25.(4). We show the other implication (⇐). For this, let I ⊂ DMF(X, L, W ) be a radical thick ⊗ 2 -ideal. For objects E ∈ DMF(X, L, 0) and F ∈ I, it suffices to prove that E ⊗ F ∈ I. We have 1 (E ⊗ F ) ⊗ 1 1 2 (E ⊗ F ) = ∼= = (E ⊗ F ⊗ E) ⊗ 2(cid:0)(E ⊗ F ) ⊗ (E ⊗ F )(cid:1) 2(cid:0)(E ⊗ F ⊗ E) ⊗ F ) 2 F, 1 1 and the object in the bottom line is in I since E ⊗ F ⊗ E ∈ DMF(X, L, W ) and I is ⊗ Hence E ⊗ F ∈ I as I is radical. 1 2 -ideal. (cid:3) Theorem 6.7 and Theorem 6.9 imply the following result. 22 Y. HIRANO Corollary 6.10. There is a homeomorphism Spc(DMF(X, L, W ),⊗ 1 2 ) ∼= Sing(X0/X) Remark 6.11. By Proposition 2.12 and Proposition 3.2, we see that Corollary 6.10 is a gener- alization of [Yu2, Theorem 1.2], where Yu consider the case when X is a regular affine scheme of finite Krull dimension. References [BFK] M. Ballard, D. Favero and L. Katzarkov, A category of kernels for equivariant factorizations and its implications for Hodge theory, Publ. Math. Inst. Hautes ´Etudes Sci. 120 (2014), 1-111. [BDFIK] M. Ballard, D. Deliu, D. Favero, M. U. Isik, and L. Katzarkov, Resolutions in factorization categories, [Bal] [EP] [Gab] [LS] [Mat1] [Mat2] [Orl1] [Orl2] [Pos] [Ros] [Ser] [Ste] [Tak] [Tho] [TT] [Yu1] [Yu2] Adv. in Math. 295 (2016), 195-249. P. Balmer, The spectrum of prime ideals in tensor triangulated categories, J. Reine Angew. Math. 588 (2005), 149-168. A. I. Efimov and L. Positselski, Coherent analogues of matrix factorizations and relative singularity categories, Algebra Number Theory 9 (2015), no. 5, 1159-1292. P. Gabriel, Des cat´egories ab´elinnes, Bull. Soc. Math. France, 90 (1962), 323-448. V. A. Lunts and O. M. Schnurer, Matrix factorizations and semi-orthogonal decompositions for blowing- ups, J. Noncommut. Geom. 10 (2016), no. 3, 907-979. Commutative algebra, Second edition, Mathematics Lecture Note Series, 56. Benjamin/Cummings Pub- lishing Co., Inc., Reading, Mass., (1980). H. Matsumura, Commutative ring theory, second ed. Cambridge Studies in Advanced Mathematics, vol. 8. Cambridge University Press, Cambridge (1989). D. O. Orlov, Triangulated categories of singularities and D-branes in Landau-Ginzburg models, Tr. Mat. Inst. Steklova, 246 (2004), 240-262. In Russian; translated in Proc. Steklov Math. Inst. 246, (2004), 227-248. D. Orlov, Matrix factorizations for nonaffine LG-models, Math. Ann. 353 (2012), no. 1, 95-108. L. Positselski, Two kinds of derived categories, Koszul duality, and comodule-contramodule correspon- dence, Mem. Amer. Math. Soc. 212 (2011), no. 966. A. L. Rosenberg, The spectrum of abelian categories and reconstruction of schemes, Rings, Hopf algebras, and Brauer groups (Antwerp/Brussels, 1996), 257-274, Lecture Notes in Pure and Appl. Math., 197, Dekker, New York, 1998. J.-P. Serre, G´eom´etrie alg´ebrique et g´eom´etrie analytique, Ann. Inst. Fourier, Grenoble, 6 (1956), 1-42. G. Stevenson, Subcategories of singularity categories via tensor actions, Compos. Math. 150 (2014), no. 2, 229-272. R. Takahashi, Classifying thick subcategories of the stable category of Cohen-Macaulay modules, Adv. Math. 225 (2010), no. 4, 2076-2116. R. W. Thomason, The classification of triangulated subcategories, Compos. Math. 105 (1997), 1-27. R. W. Thomason and T. Trobaugh, Higher algebraic K-theory of schemes and of derived categories, The Grothendieck Festschrift, Vol. III, 247-435, Progr. Math., vol. 88, Birkhauser Boston, Boston, MA 1990. Xuan Yu, Geometric study of the category of matrix factorizations, ProQuest LLC, Ann Arbor, MI, 2013. Thesis (Ph.D.)-The University of Nebraska-Lincoln. MR3187316. Xuan Yu, The triangular spectrum of matrix factorizations is the singular locus, Proc. Amer. Math. Soc., 144 (2016), no. 8, 3283-3290. Department of Mathematics, Kyoto University, Kitashirakawa-Oiwake-cho, Sakyo-ku, Kyoto, 606- 8502, Japan E-mail address: [email protected]
1212.4912	1	1212	2012-12-20T03:35:29	A Torelli-like Theorem for Smooth Plane Curves	[ "math.AG", "cs.IT", "cs.IT" ]	The Information-Theoretic Schottky Problem treats the period matrix of a compact Riemann Surface as a compressible signal. In this case, the period matrix of a smooth plane curve is characterized by only 4 of its columns, a significant compression.	math.AG	math	A TORELLI -- LIKE THEOREM FOR SMOOTH PLANE CURVES JAMES S. WOLPER Abstract. The Information-Theoretic Schottky Problem treats the period matrix of a compact Riemann Surface as a compressible signal. In this case, the period matrix of a smooth plane curve is characterized by only 4 of its columns, a significant compression. 1. Introduction Begin by fixing notation; consult [5] as a general reference. Let X be a compact Riemann Surface of genus g > 1; equivalently, X is a non- singular complex algebraic curve. Choose a basis ω1, . . . , ωg for the space H 1,0(X) of holomorphic differentials on X, and a symplectic basis α1, . . . , αg, β1, . . . , βg for the singular homology H1(X, Z), normalized so Rαi ωj = δij , the Dirac delta. The matrix Ωij := Rβi ωj is the period matrix; Riemann proved that it is symmetric with positive definite imaginary part. The torus Cg/[IΩ] is the Jacobian of X. Torelli's Theorem asserts that the Jacobian determines all of the properties of X. In practice deciding which properties apply is seldom successful (but see [7]). The period matrix is symmetric with positive-definite imaginary part, and the space of such matrices forms the Siegel upper half-space Hg. Its dimension is g(g + 1)/2, while the dimension of the moduli space of curves of degree g has dimension 3g − 3. Distinguishing the period matrices from arbitrary elements of Hg is the Schottky Problem. See [3] for details on the problem and some of its previous solutions. Now, recast the problem in terms of communication. Suppose that Alice wants to tell Bob about a curve. By Torelli's Theorem, she can do so by telling him the period matrix, but this means transmitting O(g2) complex numbers in order to describe something that depends on O(g) parameters. In other words, the period matrix is sparse in the sense of [2], and should therefore be compressible. The perspective that the period matrix is a compressible signal is the central idea of the Information -- Theoretic Schottky Problem. The attempt to apply ideas from Compressed Sensing [2] to the Schottky problem has led to many interesting experiments, conjectures, and theorems [8]. The result described here is purely mathematical, rather than computational; however, it was inspired by an attempt to implement ideas in blind Compressed Sensing, as described in [4]. Date: 17 December, 2012. 2010 Mathematics Subject Classification. 94A15, 65D32. 1 2 JAMES S. WOLPER 2. Plane Curves Shift the focus to a smooth plane curve whose affine equation f (x, y) = 0 has Its genus is g = (d−1)(d−2) degree d > 4. , and its holomorphic differentials are given by h(x, y) dx ∂f /∂y , where h is a so -- called adjoint polynomial of degree d − 3. Fix an order for the monomials of degree d − 3, eg the usual lexicographic order x0y0 < x1y0 < x0y1 < · · · < x0yd−3, thus forming a proxy basis for H (1,0)(X). 2 The main result is Theorem 2.1. There is a set of 4 columns of the period matrix of a smooth plane curve that characterize the curve; in other words, if X ′ is another plane curve whose period matrix includes these four columns, then X and X ′ are holomorphically equivalent. The four columns involved have 4g entries, so constitute a rather small superset of "moduli." Thus, this is a significant loss -- less compression of the period matrix. The number 4 seems rather arbitrary, but the condition that a curve have a smooth planar representation is strong; one would not expect such a strong result from weaker hypotheses. 3. Period Matrices and Moduli The primary tool relating period matrices to moduli is the following theorem of Rauch. Let K denote the canonical divisor on X. Theorem 3.1. [Rauch] Let {ζ1, . . . ζg} be a normalized basis for H (1,0)(X) of a non-hyperelliptic Riemann surface X, and suppose that {ζiζj : (i, j) ∈ (I, J)} form a basis for the quadratic differentials H 0(X, 2K). If another Riemann surface X ′ has the same entries as X in the (I, J) positions of its period matrix then X and X ′ are holomorphically equivalent. The proof, while not strictly relevant here, may be of interest for further ITSP investigations. It chooses the minimal member of the homotopy class of maps from the underlying surface of X to the underlying surface of X ′ with respect to the Douglas -- Dirichlet energy, and proceeds by a delicate argument using infinitessimal quadratic differentials to show that this map is holomorphic. In principle, then, Alice can send Bob 3g − 3 entries of the period matrix, and he can then verify that his period matrix is the same. However, there is no canonical way to choose which 3g − 3 elements to send, and there are many choices of 3g − 3 elements that do not form moduli. The point of Theorem A, then, is that in the case of a smooth plane curve Alice can canonically choose a slightly larger set of periods to send. Returning to plane curves, the strategy is to choose a basis for H 0(2K) carefully; in the end, this will involve only 4 columns of the period matrix. 4. Proof of the Theorem Recall the theorem of Noether (quoted in [6]; also see [1]) that every quadratic differential is a product of ordinary differentials. To determine (I, J), define a g ×g matrix Q whose rows and columns are indexed ∂y is omitted, In writing the matrix the factor dx/ ∂f by the adjoint monomials. A TORELLI -- LIKE THEOREM FOR SMOOTH PLANE CURVES 3 and in considering quadratic differentials one only need to look at products of the monomials. In the case of degree d = 6, the curve has genus g = 10 and, filling only the top row and leftmost column, Q =   1 x y x2 xy y2 x3 x2y xy2 y3 x y x2 xy y2 x3 x2y xy2 y3 . . .   In this case, the columns beginning with 1, x3, and y3 must be included to form a basis for H 0(X, 2K). More generally, the xd−3 and yd−3 columns must be included in order to get all of the monomials with x -- degree (resp. y -- degree) greater than d − 3. Note that these three columns contain duplicate entries, for example, xd−3yd−3 appears in both the xd−3 -- and yd−3 -- columns. The entries in these three columns do not constitute a basis, since they omit the monomials of degree greater than d − 3 but of x -- or y -- degree ≤ d − 4, but all of these monomials are in the x2y2 column. To see this, let xryd−2−r be a monomial of degree d − 2; here r ≤ d − 4. This monomial factors as x2y2 · xr−2yd−4−r, and xr−2yd−4−r is a monomial in the first column. Thus, xryd−2−r appears in the x2y2 column; the same applies to xd−2−ryr. Similarly, monomials of degree d − 1 can be written xryd−1−r, which factors as x2y2 · xr−2yd−3−r. Clearly d − 3 − r ≤ d − 3, so again such a monomial is a product of x2y2 and a monomial from the first column. The largest -- degree monomial satisfying the conditions is xd−4yd−4 = x2y2 · xd−6yd−6. Thus, every "missing" monomial appears in the x2y2 column. Now consider the corresponding entries in the period matrix. Since the differ- entials of the chosen basis are not normalized, multiply the right half of the period matrix by the inverse of the left half. Each entry from (I, J) in the normalized pe- riod matrix is a thus linear combination of entries from the corresponding column in the matrix associated with Q. But these entries still correspond to a (superset) of a basis for H 0(2K), and thus by the Rauch Theorem determine the curve up to isomorphism. 5. Complements • The four columns contain no more than 4g entries, which is a substantial compression of the period matrix. Even removing the duplicates, there are other relations among the quadratic differentials and thus more relations between the periods. This is easiest to see in degree 6, where removing duplicate entries leaves 28 positions, while the number of moduli is 27. The missing relation occurs in degree 6, and is, in fact, the equation of the curve. In other words, some of the redundancy from the superset of periods used to determine the curve come from the equation itself. 4 JAMES S. WOLPER In higher degrees, many of the redundancies are in the ideal generated by the equation. • Since the columns of the normalized period matrix each correspond to integrals over a cycle, it appears that only four of the generators of the first singular homology group determine the whole topology of the curve, but this is not the case because of the symmetry of the normalized period matrix. • It is neither true nor claimed that every set of four columns determines the curve; using the Alice -- Bob scenario, Alice, knowing that she has a plane curve, chooses the columns used in the proof and sends them to Bob. Bob also has a curve, or perhaps a period matrix, but may or may not know a priori whether his period matrix is a plane curve, but if it contains the four columns then he has determined that the curve that Alice sent is the one he has. In this sense the theorem provides more of a verification than an actual communication. • D. Litt points out that it may not be possible to transmit periods in a fi- nite message, although many complex numbers do have compact descriptions (eg Gaussian rationals, surds). In other cases it may only be possible to transmit an approximation of the periods. If this is so, then Bob knows that his curve is close (in an analytic sense) to the plane curve locus, which is already significant. References [1] Arbarello, E., M. Cornalba, P. Griffiths, and J. Harris, Geometry of Algebraic Curves, Volume I. Grundlehren der mathematische Wissenschaften 267. Berlin, Heidelberg, NY: Springer -- Verlag (1985). [2] Donoho, D., "Compressed Sensing," IEEE Transactions on Information Theory, 52 (April, 2006), no. 3, 1289 -- 1306. [3] Grushevsky, Samuel, "The Schottky Problem," arXiv:1009.0369v1, 2 September 2010. [4] Gleicham, S., and Y. Eldar, "Blind Compressed Sensing," arXiv:1002.2586v2, 28 April 2010. [5] Griffiths, P., and J. Harris, Princioples of Algebraic Geometry. NY: Wiley (1979). [6] Rauch, H., "On the Transcendental Moduli of Algebraic Riemann Surfaces," Proceedings of the National Academy of Sciences, 41 (1955), 42 -- 9. [7] Wolper, J., "Analytic Computation of some Automorphism groups of Riemann Surfaces," Kodai Mathematical Journal, 30 (2007), 394 -- 408. [8] Wolper, J., ¡D2¿Numerical experiments related to the information-theoretic Schottky and Torelli problems,¡D3¿ Albanian Journal of Mathematics [Special issue on computational alge- braic geometry], 6, No. 1, (2012). Dept. of Mathematics, Idaho State University, 921 S. 8th Ave., Mail Stop 8085, Pocatello, ID 83209 USA E-mail address: [email protected]
1303.2457	1	1303	2013-03-11T09:02:39	Real and complex rank for real symmetric tensors with low ranks	[ "math.AG" ]	We study the case of a real homogeneous polynomial $P$ whose minimal real and complex decompositions in terms of powers of linear forms are different. We prove that, if the sum of the complex and the real ranks of $P$ is at most $ 3\deg(P)-1$, then the difference of the two decompositions is completely determined either on a line or on a conic or two disjoint lines.	math.AG	math	REAL AND COMPLEX RANK FOR REAL SYMMETRIC TENSORS WITH LOW RANKS EDOARDO BALLICO, ALESSANDRA BERNARDI Abstract. We study the case of a real homogeneous polynomial P whose minimal real and complex decompositions in terms of powers of linear forms are different. We prove that, if the sum of the complex and the real ranks of P is at most 3 deg(P ) − 1, then the difference of the two decompositions is completely determined either on a line or on a conic or two disjoint lines. Introduction The problem of decomposing a tensor into a minimal sum of rank-1 terms, is raising interest and attention from many applied areas as signal processing for telecommunications [15], independent component analysis [11], complexity of ma- trix multiplication [20], complexity problem of P versus NP [21], quantum physics [16], [5] and phylogenetic [1]. The particular instance in which the tensor is sym- metric and hence representable by a homogeneous polynomial, is one of the most studied and developed ones (cfr. In this last case we say that the rank of a homogeneous polynomial P of degree d is the minimum integer r needed to write it as a linear combination of pure powers of linear forms L1, . . . , Lr: [17] and references therein). (1) P = c1Ld 1 + · · · + cdLd r. with ci 6= 0. Most of the papers concerning the abstract theory of the symmetric tensor rank, require that the base field is algebraically closed. In this case we may take ci = 1 for all i without loss of generality. However, for the applications, it is very important to consider the case of real polynomials and look at their real decomposition. Namely, one can study separately the case in which the linear forms appearing in (1) are complex or reals. In the real case we may take ci = 1 for all i if d is odd, while we take ci ∈ {−1, 1} if d is even. When we look for a minimal complex (resp. real) decomposition as in (1) we say that we are computing the complex symmetric rank (resp. real symmetric rank ) of P and we will indicate it rC(P ) (resp. rR(P )). Obviously and in many cases such an equality is strict. rC(P ) ≤ rR(P ), 1991 Mathematics Subject Classification. 14N05; 15A69. Key words and phrases. symmetric tensor rank; Veronese embedding; real symmetric tensor rank. The authors were partially supported by CIRM of FBK Trento (Italy), Project Galaad of INRIA Sophia Antipolis M´editerran´ee (France), Institut Mittag-Leffler (Sweden), Marie Curie: Promoting science (FP7-PEOPLE-2009-IEF), MIUR and GNSAGA of INdAM (Italy). 1 2 EDOARDO BALLICO, ALESSANDRA BERNARDI In [12] P. Comon and G. Ottaviani studied the real case for bivariate symmetric tensors. Even in this case there are many open conjectures and, up to now, few cases are completely settled ([12], [9], [2], [7]). In this paper we want to study the relation among rC(P ) and rR(P ) in the special circumstance in which rC(P ) < rR(P ). In particular we will show that, in a certain range (say, rC(P ) + rR(P ) ≤ 3 deg(P ) − 1), all homogeneous polynomials P of that degree with rR(P ) 6= rC(P ) are characterized by the existence of a curve with the property that the sets evincing the real and the complex ranks coincide out of it (see Theorem 1 for the precise statement). More precisely: let P ∈ SdRm+1 be a real homogeneous polynomial of degree d in m + 1 variables such that rC(P ) < rR(P ) and rC(P ) + rR(P ) ≤ 3 deg(P ) − 1; therefore its real and complex decomposition are P = a1Ld P = N d 1 + · · · + akLd 1 + · · · + N d k + c1M d h + c1M d 1 + · · · + csM d s , 1 + · · · + csM d s respectively, with k > h, ai ∈ {−1, 1}, cj ∈ {−1, 1}, M1, . . . , Ms ∈ S1Rm+1, h+k > d + 2. Moreover there exists a curve C′ ⊂ Pm such that [L1], . . . , [Lk], [N1], . . . , [Nh] " depends only from the variables of C′ " and C′ is either a line or a reduced conic or a disjoint union of two lines. If C′ is a line (item (a) in Theorem 1) then both the Li's and the Ni's are linear forms in the same two "variables". If C′ is a conic, then Li's and Ni's depend on 3 "variables" and their projectivizations lie on C′. See item (c) of Theorem 1 for the geometric interpretation of the reduction of L1, . . . , Lk and N1, . . . , Nh to bivariate forms involved with C′ when C′ = l ⊔ r is a disjoint union of two lines l and r (we have two sets of bivariant forms, one for the variables of l and one for the variables of r). 1. Notation and statements Before giving the precise statement of Theorem 1 we need to introduce the main algebraic geometric tools that we will use all along the paper. Let νd : Pm → PN , N := (cid:0)m+d d (cid:1) − 1, denote the degree d Veronese embedding of Pm (say, defined over C). Set Xm,d := νd(Pm). For any P ∈ PN , the symmetric rank or symmetric tensor rank or, just, the rank rC(P ) of P is the minimal cardinality of a finite set S ⊂ Pm(C) such that P ∈ hνd(S)i, where h i denote the linear span (here the linear span is with respect to complex coefficients), and we will say that S evinces rC(P ). Notice that the Veronese embedding νd is defined over R, i.e. νd(Pm(R)) ⊂ PN (R). For each P ∈ PN (R) the real symmetric rank rR(P ) of P is the minimal cardinality of a finite set S ⊂ Pm(R) such that P ∈ hνd(S)iR, where h iR means the linear span with real coefficients, and we will say that S evinces rR(P ). The integer rR(P ) is well-defined, because νd(Pm(R)) spans PN (R). Let us fix some notation: If C ⊂ Pm is either a curve or a subspace and S ⊂ Pm is a finite set, we will use the following abbreviations: SC := S ∩ C, S C := S \ (S ∩ C). Theorem 1. Let P ∈ PN (R) be such that rC(P ) + rR(P ) ≤ 3d − 1 and rC(P ) 6= rR(P ). Fix any set SC ⊂ Pm(C) and SR ⊂ Pm(R) evincing rC(P ) and rR(P ) respectively. Then one of the following cases (a), (b), (c) occurs: (a) There is a line l ⊂ Pm defined over R and with the following properties: SYMMETRIC TENSOR RANK 3 (i) SC and SR coincide out of the line l in a set Sl: SC \ SC ∩ l = SR \ SR ∩ l =: Sl; (ii) there is a point Pl ∈ hνd(SC,l)i ∩ hνd(SR,l)i such that SC,l evinces rC(Pl) and SR,l evinces rR(Pl); (iii) ♯(SC,l ∪ SR,l) ≥ d + 2 and ♯(SC,l) < ♯(SR,l). (b) There is a conic C ⊂ Pm defined over R and with the following properties: (i) SC and SR coincide out of the conic C in a set S C : SC \ SC,C = SR \ SR ∩ C =: S C ; (ii) there is a point PC ∈ hνd(SC,C )i ∩ hνd(SR,C )i such that SC,C evinces rC(PC ) and SR,C evinces rR(PC ); (iii) ♯(SC,C ∪ SR,C) ≥ 2d + 2 and ♯(SC,C ) < ♯(SR,C ); (iv) if C is reducible, say C = l1 ∪ l2 with Q = l1 ∩ l2, then ♯((SC ∪ SR) ∩ (li \ Q)) ≥ d + 1 for i ∈ {1, 2}. (c) m ≥ 3 and there are 2 disjoint lines l, r ⊂ Pm defined over R with the following properties: (i) SC and SR coincide out of the union Γ := l ∪ r in a set SΓ: SC \ SC ∩ (l ∪ r) = SR \ SR ∩ (l ∪ r) := SΓ; (ii) ♯(SC,l ∪ SR,l) ≥ d + 2 and ♯(SC,r ∪ SR,r) ≥ d + 2; (iii) the set hνd(SΓ)i ∩ hνd(Γ)i is a single point, OΓ ∈ PN (R); SC,Γ evinces rC(OΓ) and SR,Γ evinces OΓ; (iv) the set h{OΓ} ∪ νd(l)i ∩ hνd(l)i (resp. h{OΓ} ∪ νd(r)i ∩ hνd(r)i) is formed by a unique point Ol ∈ PN (R) (resp. Or ∈ PN (R)); SC,l (resp. SC,r) evinces rC(Ol) (resp. rC(Or)); SR,l (resp. SR,r) evinces rR(Ol) (resp rR(Or)). 2. The proof Remark 1. Let S ⊂ PN (R). It will be noteworthy in the sequel that S can be used to span both a real space hSiR ⊂ PN (R) and a complex space hSiC ⊂ PN (C) of the same dimension and hSiC ∩ PN (R) = hSiR. In the following we will always use h i to denote h iC. Remark 2. Fix P ∈ PN and a finite set S ⊂ PN such that S evinces rC(P ). Fix any E ( S. Then the set h{P } ∪ Ei ∩ hS \ Ei is a single point (call it P1) and S \ E evinces rC(P1). Now assume P ∈ PN (R) and S ⊂ PN (R). Then P1 ∈ PN (R). If S evinces rR(P ), then S \ E evinces rR(P1). Lemma 1. Let C ⊂ Pm be a reduced curve of degree t with t = 1, 2. Fix finite sets A, B ⊂ Pm. Fix an integer d > t such that: h1(I(A∪B) C (d − t)) = 0. Assume the existence of P ∈ hνd(A)i ∩ hνd(B)i and P /∈ hνd(S′)i for any S′ ( A and any S′ ( B. Then A C = B C . Proof. The case t = 1 is [4, Lemma 8]. If t = 2, then either C is a conic or m ≥ 3 and C is a disjoint union of 2 lines. In both cases we have h0(IC (t)) > 0 and the linear system IC (t) has no base points outside C. Since A ∪ B is a finite set, there 4 EDOARDO BALLICO, ALESSANDRA BERNARDI is M ∈ IC (t) such that M ∩ (A ∪ B) = C ∩ (A ∪ B). Look at the residual exact sequence (also called the Castelnuovo's exact sequence): (2) 0 → I(A∪B) C (d − t) → IA∪B(d) → I(A∪B)∩M,M (d) → 0 We can now repeat the same proof of [4, Lemma 8] but starting with (2) instead of the exact sequence used there (cfr. first displayed formula in the proof of [4, Lemma 8]). We will therefore get A M = B M . Now, since M ∩ (A ∪ B) = C ∩ (A ∪ B), we (cid:3) are done. We are now going to prove Theorem 1. Proof of Theorem 1 Fix P ∈ PN (R) such that rC(P ) + rR(P ) ≤ 3d − 1 and rC(P ) 6= rR(P ). Fix any set SC ⊂ Pm(C) evincing rC(P ) and any SR ⊂ Pm(R) evincing rR(P ). By applying [3], Lemma 1, we immediately get that h1(ISC∪SR(d)) > 0. Since ♯(SC) + ♯(SR) ≤ 3d − 1, either there is a line l ⊂ Pm such that ♯(SC,l ∪ SR,l) ≥ d + 2 or there is a conic C such that ♯(SC,C ∪ SR,C ) ≥ 2d + 2 ([14], Theorem 3.8). We are going to study separately these two cases in items (1) and (2) below. (1) In this step we assume the existence of a line l ⊂ Pm such that ♯(SC,l ∪ SR,l) ≥ d + 2. This hypothesis, together with rR(P ) 6= rC(P ), immediately implies property (aiii) of the statement of the theorem. We are now going to distinguish the case h1(ISC∪SR(d − 1)) = 0 (item (1.1) below) from the case h1(ISC∪SR(d − 1)) > 0 (item (1.2) below). (1.1) Assume h1(ISC∪SR(d − 1)) = 0. First of all, observe that the line l ⊂ Pm is well defined over R since it contains at least 2 points of SR (Remark 1). Then, by Lemma 1, we have that SC and SR have to coincide out of the line l: and this proves (ai) of the statement of the theorem in this case (1.1). SC \ SC,l = SR \ SR,l := Sl, The fact that ♯(SC) < ♯(SR) implies that ♯(SR,l) > (d + 2)/2 and ♯(Sl) ≤ d, hence h1(ISl∪l(d)) = 0. Therefore we have that dim(hνd(Sl ∪l)i) = ♯(Sl)+d+1, dim(hνd(Sl)i) = ♯(Sl)−1 and dim(hνd(l)i) = d + 1, and Grassmann's formula gives hνd(Sl)i ∩ hνd(l)i = ∅. Since P ∈ hνd(Sl ∪ SC)i and SC,l ⊂ l, the set hνd(Sl) ∪ {P }i ∩ hνd(l)i is a single point, Pl ∈ PN (R). Since P ∈ hνd(SC)i and P /∈ hνd(S′ C)i for any S′ ( SC, the set hνd(Sl) ∪ {P }i ∩ hνd(SC,l)i is a single point, PC (Remark 2). Then obviously: PC = Pl ∈ PN (R). Since SC evinces rC(P ), then SC,l evinces Pl (Remark 2). In the same way we see that hνd(Sl) ∪ {P }i ∩ hνd(SR,l)i = {Pl} and that SR,l evinces rR(Pl). This proves (aii) of Theorem 1 in this case (1.1). SYMMETRIC TENSOR RANK 5 (1.2) Assume h1(ISC∪SR(d − 1)) > 0. First of all, observe that there exists a line r ⊂ Pm such that ♯(r ∩ (SC ∪ SR)l) ≥ d + 1, because ♯(SC ∪ SR)l ≤ 3d − 1 − d − 2 ≤ 2(d − 1) + 1. By the same reason, if we write: C := l ∪ r, we get that ♯(SC ∪ SR) C ≤ 3d − 1 − d − 2 − d − 1 ≤ d − 2 and hence h1(I(SC∪SR) C (d − 2)) = 0 (e.g. by [6], Lemma 34, or by [14], Theorem 3.8). Lemma 1 gives: (3) SC, C = SR, C. -- Assume for the moment l ∩ r 6= ∅. In this case, Remark 2 indicates that we can consider case (b) of the statement of the theorem. Therefore (3) proves (bi) in the case that the conic C in (b) in the statement of the theorem is reduced. Moreover, condition (biv) is satisfied, because ♯(r ∩ (SC ∪ SR)l) ≥ d + 1. -- Now assume l ∩ r = ∅. We will check that we are in case (a) with respect to the line l if ♯(SC,r ∪ SR,r) = d + 1, while we are in case (c) with respect to the lines l and r if ♯(SC,r ∪ SR,r) ≥ d + 2, and the case ♯(SC,l ∪ SR,l) = ♯(SC,r ∪ SR,r) = d + 1 cannot occur. Set Γ := l ∪ r. Since r ∩ l = ∅, we have dimhΓi = 3 and hence m ≥ 3. -- Assume for the moment m ≥ 4. Hence ♯(SC∪SR) chΓi ≤ d and h1(I(SC∪SR) dhΓi (d− 1)) = 0. Therefore: (4) SC, chΓi = SR, chΓi and the set h{P } ∪ νd(SC, chΓi)i ∩ hνd(hΓi)i is a single real point: (5) O := h{P } ∪ νd(SC, chΓi)i ∩ hνd(hΓi)i ∈ PN (R), SC,Γ evinces rC(O) and SR,Γ evinces rR(O). Now, (4) implies that if we are either in case (a) or in case (c) of the theorem, we can simply study what happens at SC,hΓi and at SR,hΓi, which means that we can reduce our study to the case m = 3, since hΓi = P3. -- Until step (2) below, we will assume m = 3. The linear system IΓ(2) on hΓi has no base points outside Γ itself. Since SC ∪SR is finite, there is a smooth quadric surface W containing Γ such that SC,W ∪ SR,W = SC,Γ ∪ SR,Γ. Moreover, such a W can be found among the real smooth quadrics, since l and r are real lines. Since ♯(SC,hΓi ∪ SR,hΓi) W ≤ d − 1, we have h1(I(SC∪SR) W (d − 2)) = 0. Hence Lemma 1 applied to the point O defined in (5), gives: (SC,hΓi) W = (SR,hΓi) W , h{O} ∪ νd(SC,hΓ,i) W i ∩ hνd(W )i is a single real point (6) O′ = h{O} ∪ νd(SC,hΓ,i) W i ∩ hνd(W )i ∈ PN (R), and SC,W evinces rC(O′). If (O′, SC,W , SR,W ) is either as in case (a) or in case (c) of the statement of the theorem, then (O, SC,hΓi, SR,hΓi) is in the same case. Consider the system (1, 0) of lines on the smooth quadric surface W containing Γ. We have 6 EDOARDO BALLICO, ALESSANDRA BERNARDI that h1(W, OW (d − 2, d)) = 0, and hence the restriction map H 0(W, OW (d)) → H 0(Γ, OΓ(d)) is surjective. Therefore h1(W, IW ∩(SC∪SR)(d)) = h1(l, Il∩(SC∪SR),l(d)) + h1(r, Ir∩(SC∪SR),r(d)). Now, this last equality, together with the facts that νd(SC,W ) and νd(SR,W ) are linearly independent and (SC ∪ SR)W ⊂ Γ, gives: dim(hνd(SC,W )i ∩ hνd(SR,W )i) = ♯(SC ∩ SR)l + ♯(SC ∩ SR)r+ (7) h1(l, Il∩(SC∪SR),l(d)) + h1(r, Ir∩(SC∪SR),r(d)). (1.2.1) Observe that (7) implies that the case ♯(SC ∪ SR)r = ♯(SC ∪ SR)l = d + 1 cannot happen because there is no contribution from h1(l, Il∩(SC∪SR),l(d)) + h1(r, Ir∩(SC∪SR),r(d)) since both terms, in this case, are equal to 0. So, we can assume that at least ♯(SC ∪ SR)l > d + 1. (1.2.2) Assume ♯(SC ∪ SR)r = d + 1 and ♯(SC ∪ SR)l > d + 1. To prove that we are in case (a) with respect to l it is sufficient to prove SC,r = SR,r. We have dimhνd(SC∪SR)i = ♯(SC∪SR)−1−h1(ISC∪SR(d)). Since νd(SC) and νd(SR) are linearly independent, ♯(SC ∪ SR) = ♯(SC) + ♯(SR) − ♯(SC ∩ SR), ♯(SC,l ∪ SR,l) = ♯(SC,l) + ♯(SR,l) − ♯(SC,l ∩ SR,l) and h1(ISC∪SR(d)) = h1(l, ISC,l∪SR,l), Grassmann's formula gives that hνd(SC)i ∩ hνd(SR)i is generated by hνd(SC,l)i ∩ hνd(SR,l)i. Since P /∈ hνd(S′)i for any S′ ( SC, we get SC = SC,l ∪ (SC ∩ SR)l, i.e. Sl = SR \ SC,l. Hence we can consider case (a), and (3) proves property (ai) also for the case (1.2) that we are treating. The point Pl that we need to get (aii) can be identified with the point O′ defined in (6) while (aiii) comes from our hypotheses. This gives all cases (a) of Theorem 1. (1.2.3) Assume that both ♯(SC ∪ SR)r ≥ d + 2 and ♯(SC ∪ SR)l ≥ d + 2. We need to prove that we are in case (c). Recall that SC,Γ = SR,Γ and that h1(ISC, Γ∪Γ(d)) = 0. The latter equality implies, as in Remark 2, that h{P } ∪ νd(SC,Γ)i ∩ hνd(Γ)i is a single real point O1, that SC,Γ evinces rC(O1) and that SR,Γ evinces O1. Now O1 plays the role of OΓ of case (ciii) in Theorem 1. Since hνd(l)i∩hνd(r)i = ∅ and O1 ∈ hνd(Γ)i, the sets h{O1}∪νd(l)i∩hνd(l)i (resp. h{O1}∪νd(r)i∩hνd(r)i) are formed by a unique point O2 (resp. O3). Remark 2 gives that Oi ∈ PN (R), i = 1, 2, SC,l evinces rC(O2), SR,l evinces rR(O2), SC,r evinces rC(O3) and SR,r evinces rR(O3). The hypotheses of the case (1.2.3) coincide with (cii) of the statement of the theorem, while (3) gives also property (ci). Moreover O2 and O3 defined above coincide with Ol and Or in (civ) of Theorem 1, therefore we have also proved case (c) of Theorem 1. (2) Now assume the existence of a conic C ⊂ Pm such that (8) deg(SC ∪ SR)C ≥ 2d + 2. Since ♯(SC ∪ SR)C ≤ 3d − 1 − 2d − 2 ≤ d − 1, we have h1(I(SC∪SR) C By Lemma 1 we have (d − 2)) = 0. (9) SC, C = SR, C, the set h{P } ∪ νd(SC, C )i ∩ hνd(hCi)i is a single point: (10) P ′ := h{P } ∪ νd(SC, C )i ∩ hνd(hCi)i and SC,C evinces rC(P ′). Moreover, if C is defined over R, then P ′ ∈ PN (R) and SR,C evinces rR(P ′). Hence ♯(SC,C ) < ♯(SC ∩ SR). SYMMETRIC TENSOR RANK 7 (2.1) Assume that C is smooth. Therefore (9) proves (bi) of the statement of the theorem in the case where C is smooth. Since the reduced case is proved above (immediately after the displayed formula (3)), we have concluded the proof of (bi). Moreover the hypothesis (8) coincides with (biii) of the statement of the theorem since ♯(SC,C ) is obviously strictly smaller than ♯(SR,C ). This concludes (biii). The fact that ♯(SC,C ) < ♯(SR,C ) also implies that ♯(SR,C ) ≥ 5. Since each point of SR is real, C is real. Remark 2 gives that νd(SR,C ) evinces rR(P ′). Since SR,C ⊂ C, SR,C also evinces the real symmetric tensor rank of P ′ with respect to the degree 2d rational normal curve νd(C). The point P ′ defined in (10) plays the role of the point PC appearing in (bii) of the statement of the theorem. Therefore, we have just proved (bii) of Theorem 1. We treat the case (2.2) below for sake of completeness, but we can observe that this concludes the proof of Theorem 1. (2.2) Assume that C is reducible, say C = L1 ∪ L2 with L1 and L2 lines and ♯((SC ∪ SR)L1) ≥ ♯((SC ∩ SR)L2). If ♯((SC ∪ SR) ∩ (L2 \ L2 ∩ L1)) ≤ d, then we proved in step (1) that we are in case (a) with respect to the line L1. Hence we may assume ♯((SC ∪ SR) ∩ (L2 \ L2 ∩ L1)) ≥ d + 1. Thus even condition (biv) is satisfied as already remarked above after the displayed formula (3). References [1] E. S. Allman and J. A. Rhodes, Phylogenetic ideals and varieties for the general Markov model, Adv. in Appl. Math. 40 (2008), no. 2, 127 -- 148, [2] E. Ballico, On the typical rank of real bivariate polynomials, arXiv:1204.3161 [math.AG]. [3] E. Ballico and A. Bernardi, Decomposition of homogeneous polynomials with low rank, Math. Z. 271 (2012), 1141 -- 1149. [4] E. Ballico and A. Bernardi, A partial stratification of secant varieties of Veronese varieties via curvilinear subschemes, Sarajevo Journal of Mathematics. 8 (20), 33 -- 52 (2012). [5] A. Bernardi and I. Carusotto, Algebraic geometry tools for the study of entanglement: an application to spin squeezed states, J. Phys. A: Math. Theor. 45 (2012) 105304 (13pp). DOI 10.1088/1751-8113/45/10/105304. [6] A. Bernardi, A. Gimigliano and M. Id`a, Computing symmetric rank for symmetric tensors, J. Symbolic. Comput. 46 (2011), no. 1, 34 -- 53. [7] G. Blekherman, Typical real ranks of binary forms, arXiv:1205.3257 [math.AG]. [8] J. Brachat, P. Comon, B. Mourrain and E. P. Tsigaridas, Symmetric tensor decomposition, Linear Algebra Appl. 433 (2010), no. 11 -- 12, 1851 -- 1872. [9] A. Causa and R. Re, On the maximum rank of a real binary form, Ann. Mat. Pura Appl. (4) 190 (2011), 55 -- 59. [10] G. Comas and M. Seiguer, On the rank of a binary form, Found. Comp. Math. 11 (2011), no. 1, 65 -- 78. [11] P. Comon, Independent Component Analysis, Higher Order Statistics, Elsevier, J-L. La- coume, Amsterdam, London, (1992), 29 -- 38. [12] P. Comon and G. Ottaviani, On the typical rank of real binary forms, Linear Multilinear Algebra 60 (2012), no. 6, 657 -- 667. [13] P. Comon, G. H. Golub, L.-H. Lim and B. Mourrain, Symmetric tensors and symmetric tensor rank, SIAM J. Matrix Anal. 30 (2008) 1254 -- 1279. [14] A. Couvreur, The dual minimum distance of arbitrary dimensional algebraic-geometric codes, J. Algebra 350 (2012), no. 1, 84 -- 107. [15] L. de Lathauwer and J. Castaing, Tensor-Based Techniques for the Blind Separation of DS- CDMA Signals, Signal Processing, 87 (2007), 322 -- 336. [16] J. Eisert and D. Gross, Multiparticle entanglement, Bruss, Dagmar (ed.) et al., Lectures on quantum information. Weinheim: Wiley-VCH. Physics Textbook, 237 -- 252 (2007), (2007). [17] J. M. Landsberg, Tensors: Geometry and Applications, Graduate Studies in Mathematics, Vol. 128, Amer. Math. Soc. Providence, 2012. 8 EDOARDO BALLICO, ALESSANDRA BERNARDI [18] J. M. Landsberg and Z. Teitler, On the ranks and border ranks of symmetric tensors, Found. Comput. Math. 10 (2010), no. 3, 339 -- 366. [19] L.-H. Lim and V. de Silva, Tensor rank and the ill-posedness of the best low-rank approxi- mation problem, SIAM J. Matrix Anal. Appl. 30 (2008), no. 3, 1084 -- 1127. [20] V. Strassen, Rank and Optimal Computation of Generic Tensors, Linear Algebra Appl., (1983), 645 -- 685. [21] L. G. Valiant, Quantum computers that can be simulated classically in polynomial time, Proceedings of the Thirty-Third Annual ACM Symposium on Theory of Computing, 114 -- 123 (electronic), ACM, New York, (2001). University of Trento, Dept. of Mathematics, 38123 Povo (TN), Italy University of Torino, Dept. of Mathematics "Giuseppe Peano", 10123 Torino, Italy E-mail address: [email protected], [email protected]
1310.3569	1	1310	2013-10-14T06:10:06	Unirationality of Ueno-Campana's threefold	[ "math.AG" ]	We shall prove that the threefold studied in the paper "Remarks on an Example of K. Ueno" by F. Campana is unirational. This gives an affirmative answer to a question posed in the paper above and also in the book by K. Ueno, "Classification theory of algebraic varieties and compact complex spaces".	math.AG	math	UNIRATIONALITY OF UENO-CAMPANA'S THREEFOLD FABRIZIO CATANESE, KEIJI OGUISO AND TUYEN TRUNG TRUONG Abstract. We shall prove that the threefold studied in the paper " Remarks on an Example of K. Ueno" by F. Campana is unirational. This gives an affirmative answer to a question posed in the paper above and also in the book by K. Ueno, "Classification theory of algebraic varieties and compact complex spaces". Let k be any field of characteristic 6= 2 containing a primitive fourth root of unity √−1. We shall work over k unless otherwise stated. Let [x : y : z] be the homogeneous coordinates of P2 and let 1. Introduction be the harmonic elliptic curve, having an automorphism g of order 4 defined by C := (y2z = x(x2 − z2)) ⊂ P2 g∗(x : y : z) = (−x : √−1y : z) (C, g) ≃ (E√ −1,√−1) , whose quotient is P1. When k is the complex number field C, we have where E√ −1 = C/(Z + √−1Z), the elliptic curve of period √−1 and √−1 is the automor- phism induced by multiplication by √−1 on C. This is because the complex elliptic curve with an automorphism of order 4 acting on the space of global holomorphic 1-forms as √−1 is unique up to isomorphism. Let (Cj, gi) (j = 1, 2, 3) be three copies of (C, g). Let Z = C1 × C2 × C3 . For simplicity, we denote the automorphism of Z defined by (g1, g2, g3) by the same letter g. Then g is an automorphism of Z of order 4 and the quotient threefold Y := (C1 × C2 × C3)/hgi has 8 singular points of type (1, 1, 1)/4 and 28 singular points of type (1, 1, 1)/2. Let X be the blow up of Y at the maximal ideals of these singular points. Then X is a smooth projective threefold defined over k. In his paper [Ca12], F. Campana proved that X is a rationally connected threefold when k = C. We shall call X the Ueno-Campana's threefold. In [Ca12, Question 4], F. Campana asked if X is rational or unirational (at least over C)? See also [Ue75, Page 208] for this Question and [OT13] for a relevant example and application to complex dynamics. The aim of this short note is to give an affirmative answer to this question: 1 2 FABRIZIO CATANESE, KEIJI OGUISO AND TUYEN TRUNG TRUONG Theorem 1.1. Ueno-Campana's threefold X is unirational, i.e., there is a dominant ra- tional map P3 ··· → X. We shall show that X is birationally equivalent to the Galois quotient of a conic bundle over P2 with a rational section, while X itself is birationally equivalent to a conic bundle over P2 without any rational section. We are still working on the question whether X is a rational variety. Aknowledgement. We would like to express our thanks to Professor De-Qi Zhang for his invitation to Singapore where the initial idea of this note grew up. 2. Proof of Theorem (1.1) The curves (Ci, gi) (i = 1, 2, 3) are birationally equivalent to (C 0 curve in the affine space A2 = Spec k[Xi, Yi], and gi is the automorphism of C 0 i , gi), where C 0 is the i , defined by i Y 2 i = Xi(X 2 The affine coordinate ring k[C 0 k[C 0 i − Xi(X 2 i − 1) , g∗i Yi = √−1Yi , g∗i Xi = −Xi . i ] of C 0 i ] = k[Xi, Yi]/(Y 2 is i i − Xi(X 2 i − 1)). , WE set xi := Xi mod (Y 2 i − 1)). We note that y2 i = xi(x2 Then (Z = C1 × C2 × C3, g = (g1, g2, g3)) is birationally equivalent to the affine threefold i − 1), g∗yi = √−1yi, g∗xi = −xi in k[C 0 1 × C 0 2 × C 0 i − 1)), yi := Yi mod (Y 2 i − Xi(X 2 V := C 0 with automorphism (g1, g2, g3), which we denote by the same letter g, and with affine coordinate ring i ]. 3 k[V ] = k[C 0 i ] ⊗ k[C 0 2 ] ⊗ k[C 0 3 ] generated by x1, x2, x3, y1, y2, y3 . The rational function field k(Z) of Z is In both k[V ] and k(Z), we have k(Z) = k(V ) = k(x1, x2, x3, y1, y2, y3) . (I) y2 i = xi(x2 i − 1) , (II) g∗yi = √−1yi , g∗xi = −xi . Since X is birationally equivalent to V /hgi, the rational function field K(X) of X is iden- tified with the invariant subfield k(Z)g of k(Z), i.e., Consider the following elements in k(Z): k(X) = k(Z)g = {f ∈ k(Z)g∗f = f} . (III) b2 := x2 x1 , b3 := x3 x1 , a2 := y2 y1 , a3 := y3 y1 , (IV) u1 := x2 1 , w1 := y4 1 , λ1 := x1y2 1 , and define the subfield L of k(Z) by Here we used the fact that x1 6= 0, y1 6= 0 in k(Z). L := k(b2, b3, a2, a3, u1, w1, λ1) . UNIRATIONALITY 3 Lemma 2.1. k(X) = L in k(Z). Proof. By (II) and (III), b2, b3, a2, a3, u1, w1, λ1 are g-invariant. Hence Note that k(Z) = L(y1). This is because (V) L ⊂ k(X) ⊂ k(V ) . x1 = λ1 y2 1 , x2 = b2x1, x3 = b3x1 , y2 = a2y1 , y3 = a3y1 , by (III) and (IV). Since y4 1 = w1 and w1 ∈ k(Z), it follows that (VI) [k(Z) : L] ≤ 4 , where [k(Z) : L] is the degree of the field extension L ⊂ k(Z), i.e., the dimension of k(Z) being naturally regarded as the vector space over L. On the other hand, the group hgi ⊂ Gal(k(Z)/k(X)) is of order 4. Thus, by the funda- mental theorem of Galois theory, we have that (VII) [k(Z) : k(X)] = [K(Z) : k(Z)g] = ord (g) = 4 . The result now follows from (V), (VI), (VII). Indeed, by (V), we have [k(Z) : L] = [k(Z) : k(X)][k(X) : L] . On the other hand, [k(Z) : L] ≤ 4 by (VI), and [k(X) : L] ≥ 1. Hence [k(X) : L] = 1 by (VII). This means that L = k(X) in k(Z), as claimed. (cid:3) Lemma 2.2. L = k(u1, b2, b3, a2, a3) in k(Z). Proof. Since u1, b2, b3, a2, a3 ∈ L, it follows that k(u1, b2, b3, a2, a3) ⊂ L. Let us show L ⊂ k(u1, b2, b3, a2, a3). For this, it suffices to show that w1, λ1 ∈ k(u1, b2, b3, a2, a3). 1 − 1), Hence taking the square and using (VI), we obtain that Recall by (I), y2 1 = x1(x2 Hence w1 ∈ k(u1, b2, b3, a2, a3). From y2 (IX) λ1 = x1y2 Hence λ1 ∈ k(u1, b2, b3, a2, a3) as well. Lemma 2.3. Let j = 2, 3. Then, a2 Proof. By using (I), we obtain that (VIII) w1 = y4 1 = x2 1(x2 1 − 1)2 = u1(u1 − 1)2 . 1 = x1(x2 1 = x2 1(x2 1 − 1) again, we have that 1 − 1) = u1(u1 − 1) . j − bj 6= 0 in both k(Z) and k(X). xj x1 = xj x1 (X) a2 y2 j 1 − y2 j − bj = xj(x2 x1(x2 j − 1) 1 − 1) − in k(Z). Recall that xi 6= 0 in k(Z). Thus, if a2 j − bj = 0 in k(Z), then we would have (x2 1 − 1) = 1 in K(Z) = k(V ) from the equality above, and therefore, xj = ±x1 in k[V ]. However, this contradicts to the fact that x1 is identically 0 on the set of k-valued points ({0} × C2 × C3)(k) but ±xj (j = 2, 3) are not identically 0 on it. This contradiction implies that a2 j − bj ∈ k(Z)g = k(X) and k(X) is a subfield of k(Z), it follows that a2 j − bj 6= 0 in k(Z). Since a2 x2 j − 1 1 − 1 − 1) , x2 j − 1)/(x2 xj x1 ( = j − bj 6= 0 in k(X) as well. (cid:3) (cid:3) 4 FABRIZIO CATANESE, KEIJI OGUISO AND TUYEN TRUNG TRUONG Proposition 2.4. k(X) = L = k(b2, b3, a2, a3) in k(Z). More precisely, in k(Z), we have (XI) u1 = a2 2 − b2 a2 2 − b3 2 = a2 3 − b3 a2 3 − b3 3 . Proof. By Lemma (2.1, 2.2), it suffices to show the equality (X) in k(Z). Observe that, for j = 2, 3: y2 j = xj(x2 1a2 j − 1) ⇔ y2 j = x1bj(x2 1b2 j − 1) hence multiplication by x1 yields 1bj(x2 x2 and dividing by x2 1(x2 j − 1) = x1y2 1a2 1b2 1 and observing that u1 = x2 1 we obtain j = x2 1 − 1)a2 j , i.e., bj(u1b2 j − 1) = (u1 − 1)a2 j (∗∗) u1(a2 Using the previous lemma we obtain (a2 j − b3 j − b3 j − bj. j ) = a2 j ) 6= 0, so we can divide and obtain (XI). (cid:3) Proposition 2.5. X is birationally equivalent to the affine hypersurface H in A4 = Spec k[a, b, α, β], defined by or equivalently defined by (a2 − b)(α2 − β3) = (α2 − β)(a2 − b3) , a2β(1 − β2) = α2b(1 − b2) + bβ(b2 − β2) , Proof. By Lemma (2.1) and Proposition (2.4), k(X) = k(a2, a3, b2, b3) in k(Z), with a relation (XII) (a2 2 − b3 2) . 2a2 Expanding both sides and subtracting then the common term a2 2 + b3b3 2 . 3 − b3)(a2 2 − b2)(a2 3) = (a2 3 − b3 2b3 −a2 3 − b2a2 3b3 3 = −a2 2 − b3a2 Solving this relation in terms of a2, we obtain that 3b2(1 − b2 2 − b2 Since b3 = x3/x1 is not a constant in k(Z), it follows that b3(1 − b2 also not 0 in k(X). Thus 2b3(1 − b2 2) + b2b3(b2 (XIII) a2 3 + b2b3 3) = a2 3, we obtain 3) . 3) 6= 0 in k(Z), whence (XIV) a2 2 = 3b2(1 − b2 a2 2) + b2b3(b2 b3(1 − b2 3) 2 − b2 3) . Therefore a2 is algebraic over k(a3, b2, b3) of degree at most 2. Since X is of dimension 3 over k, it follows that a3, b2, b3 form a transcendence basis of k(X) over k. Thus, the subring k[a3, b2, b3] of k(X) is isomorphic to the polynomial ring over k of Krull-dimension 3. Moreover, the right hand side of (XIV) is not a square in k(a3, b2, b3). Indeed, the multiplicity of b3 in the denominator is 1 while the numerator is not in k and the multiplicity UNIRATIONALITY 5 of b3 in the numerator is 0. Thus the equation (XIV) is the minimal polynomial of a2 over k(a3, b2, b3). Hence X is birationally equivalent to the double cover of A3 = Spec k[a3, b2, b3], defined by (XIV). This means that X is birationally equivalent to the hypersurface in the affine space A4 = Spec k[a, α, b, β], defined by (XIV) or equivalently defined by (XIII) or by (XII), in which (a2, a3, b2, b3) are replaced by (a, α, b, β). (cid:3) Corollary 2.6. Let H ⊂ A4 = Spec k[a, α, b, β] be the same as in Proposition (2.5). Con- sider the affine plane A2 = Spec k[b, β] and the natural projection defined by Then the natural restriction map π : A4 → A2 (a, b, α, β) 7→ (b, β) . p := πH : H → A2 is a conic bundle over A2. In particular, the graph Γ of the rational map p : X ··· → P2 naturally induced by p forms a conic bundle on Γ over P2. We note that Γ is projective and birationally equivalent to X over k. Proof. The fibre π−1(η) of π over the generic point η ∈ A2 = Spec k[b, β] is the affine space A2 η = Spec k(b, β)[a, α] defined over κ(η) = k(b, β). Thus by the second equation in Proposition (2.5), the generic fibre Xη := (πH)−1(η) is the conic in A2 η, defined by a2β(1 − β2) = α2b(1 − b2) + bβ(b2 − β2) . This implies the result. (cid:3) Remark 2.7. The conic Xη in the proof of Proposition (2.6) has no rational point over κ(η) = k(b, β), i.e., the set Xη(k(b, β)) is empty. Proof. Suppose to the contrary that (a(b, β), α(b, β)) ∈ Xη(k(b, β)). We can write a(b, β) = P (b, β) Q(b, β) , α(b, β) = R(b, β) Q(b, β) , where P (b, β), Q(b, β), R(b, β) ∈ k[b, β] with no non-constant common factor, possibly after replacing the denominators by their product. Then substituting the above into the equation of Xη and clearing the denominator, we would have the following identity in k[b, β]: P (b, β)2β(1 − β2) = R(b, β)2b(1 − b2) + Q(b, β)2bβ(b2 − β2) . Since k[b, β] is a polynomial ring, in particular, it is a UFD, it would follow that P (b, β) is divisible by b and R(b, β) is divisible by β in k[b, β]. Thus P (b, β) = P1(b, β)b and R(b, β) = R1(b, β)β for some P1(b, β), R1(b, β) ∈ k[b, β]. Substituting these two into the equality above and dividing by bβ 6= 0, it follows that P1(b, β)2b(1 − β2) = R1(b, β)2β(1 − b2) + Q(b, β)2(b2 − β2) . Substitute b = 0 into this equation: we obtain R1(0, β)2β +Q(0, β)2β2 = 0, which implies that R1(0, β) = Q(0, β) = 0. This means that both R1(b, β) and Q(b, β) are divisible by 6 FABRIZIO CATANESE, KEIJI OGUISO AND TUYEN TRUNG TRUONG b. Similarly, if we substitute β = 0 into the above equation we find that both P1(b, β) and Q(b, β) are divisible by β. Thus we can write P1(b, β) = βP2(b, β), R1(b, β) = bR2(b, β), Q(b, β) = bβQ2(b, β), where P2(b, β), R2(b, β), Q(b, β) ∈ k[b, β]. But this implies that all P (b, β), Q(b, β), R(b, β) are divisible by bβ, a contradiction. (cid:3) The next corollary completes the proof of Theorem (1.1): Corollary 2.8. Let H ⊂ A4 = Spec k[a, α, b, β], p : H → A2 = Spec k[b, β] be the same as in Proposition 2.5 and Corollary 2.6. Consider another affine space Spec k[s, t] and the (finite Galois) morphism of degree 4 defined by Consider then the fibre product f : Spec k[s, t] → Spec k[b, β] f ∗b = s2 , f ∗β = t2 . Q := H ×Spec k[b,β] Spec k[s, t] and the natural second projection p2 : Q → Spec k[s, t]. Then p2 is a conic bundle with a rational section and Q is a rational threefold. In particular, H, hence X, is unirational. Proof. Recall that H is the hypersurface in Spec k[a, b, α, β] defined by or equivalently by a2β(1 − β2) = α2b(1 − b2) + bβ(b2 − β2) , Thus, by definition of the fibre product, Q is a hypersurface in the affine space A4 = Spec k[a, α, s, t], defined by (a2 − b)(α2 − β3) = (α2 − β)(a2 − b3) . or equivalently by a2t2(1 − t4) = α2s2(1 − s4) + s2t2(s4 − t4) , (a2 − s2)(α2 − t6) = (α2 − t2)(a2 − s6) . Then the natural projection p2 : Q → Spec k[s, t] is a conic bundle with generic fibre Qη′ = (a2t2(1 − t4) = α2s2(1 − s2) + s2t2(s4 − t4)) ⊂ Spec k(s, t)[a, α] = A2 η′ , where η′ is the generic point of Spec k[s, t]. Then Qη′ has a rational point (a, α) = (s, t) ∈ Q(k(s, t)) over κ(η′) = k(s, t). Hence Qη′ is isomorphic to P1 η′ over k(s, t). Thus, denoting the affine coordinate of P1 η′ by v, we obtain that k(Q) = k(s, t)(Qη′ ) ≃ k(s, t)(P1 η′ ) = k(s, t)(v) = k(s, t, v) . Since Q is of dimension 3 over k, it follows that s, t, v are algebraically independent over k. Hence, k(Q) is isomorphic to the rational function field of P3 over k. Hence Q is a rational threefold over k, i.e., birationally equivalent to P3 over k. Since the natural morphism p1 : Q → H, i.e., the first projection morphism in the fibre product, is a finite dominant morphism of degree 4, Q is birational to P3 and H is birationally equivalent to X, all over UNIRATIONALITY 7 k, we obtain a rational dominant map q : P3 ··· → X over k, from the natural projection p1 : Q → H. Hence X is unirational. (cid:3) References [Ca12] F. Campana, Remarks on an example of K. Ueno, Series of congress reports, Classification of algebraic varieties, Editors: C. Faber, G. van der Geer, and E. Looijenga, European Mathematical Society, 2011, 115 -- 121. [OT13] K. Oguiso and T. T. Truong, Explicit Examples of rational and Calabi-Yau threefolds with primitive automorphisms of positive entropy, arXiv:1306.1590. [Ue75] K. Ueno, Classification theory of algebraic varieties and compact complex spaces. Notes written in collaboration with P. Cherenack, Lecture Notes in Mathematics, 439 Springer-Verlag, Berlin-New York, 1975.
1801.03192	3	1801	2018-12-06T00:58:44	Nonvanishing for threefolds in characteristic $p>5$	[ "math.AG" ]	We prove the Nonvanishing Theorem for threefolds over an algebraically closed field $k$ of characteristic $p >5$.	math.AG	math	NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 CHENYANG XU AND LEI ZHANG Abstract. We prove the Nonvanishing Theorem for threefolds over an alge- braically closed field k of characteristic p > 5. Contents 1-foliation Introduction 1. 2. The case ρ∗KX · c2(Y ) < 0 2.1. Stabilities in positive characteristic 2.2. Numerical and nef dimension 2.3. 2.4. Proof of Theorem 2.1 3. Fundamental group and numerically flat vector bundles 3.1. Local fundamental groups 3.2. Numerically flat vector bundles 4. Nonvanishing 4.1. Preliminaries 4.2. Proof of Theorem 4.1 References 1 3 4 5 5 6 8 9 11 13 13 15 22 1. Introduction Starting from [HX15], the minimal model program for threefolds in positive char- acteristic p > 5 has moved forward quickly. Running a minimal model program in this setting is established by [HX15, CTX15, Bir16, BW17, HNT17] in increasing generalities. The aim of this paper is to attack the abundance conjecture for min- imal threefolds over characteristic p > 5. Previously, this result was known for X of general type [Bir16, Xu15] or satisfying dim Pic0(X) > 0 [DW16, Zha16, Zha17]. The main result in this paper is the following nonvanishing theorem. Theorem 1.1 (Non-vanishing theorem). Let X be a terminal projective mini- mal threefold over an algebraically closed field k of characteristic p > 5. Then κ(X, KX ) ≥ 0, that is, nonvanishing holds for X. Let us recall the proof of this theorem in characteristic zero which is the main theorem in [Miy88] (see also [Kol92, Section 9]). It has a different flavour from the proof of the results for running the minimal model program, namely, it involves more investigation of higher rank bundles (e.g. the tangent bundle), as opposed to focusing on line bundles (e.g. the canonical bundle). Two of the main tools used in 1 2 CHENYANG XU AND LEI ZHANG the proof are the pseudo-effectivity of c2 and Donaldson's theorem on stable bundles on smooth complex projective surfaces. As well known, these results usually do not hold in in positive characteristics. Luckily, the general understanding of the stability of sheaves in positive characteristics has provided us with enough tools to deal with the new positive characteristic phenomenon. Let us discuss how to treat the first issue: Let ρ : Y → X be a smooth resolution of singularities, i.e., Y is smooth and ρ is a birational morphism isomorphic over the smooth locus X sm. The pseudo-effectivity of c2 means ρ∗KX · c2(Y ) ≥ 0, which is a starting point in [Miy88] (clearly the intersection number does not depend on the choice of the smooth resolutions). As we mentioned, this kind of results may not hold in positive characteristics. Nevertheless, it was understood before that the failure of this type of inequalities will usually imply a special geometric structure on Y , e.g., a foliation by rational curves. Using this, we can indeed show the full abundance under the assumption that ρ∗KX · c2(Y ) < 0. Theorem 1.2 (=Theorem 2.1). Let X be a terminal projective minimal 3-fold over an algebraically closed field k of characteristic p ≥ 5. Let ρ : Y → X be a smooth resolution of singularities. Assume that ν(X, KX ) < 3 and ρ∗KX · c2(Y ) < 0. Then κ(KX ) = 1 and KX is semiample. We want to note that in the proof of Theorem 1.2, we do not need to run any minimal model program. This is the reason why we can treat the char k = 5 case. More precisely, to prove Theorem 1.2, we start with the assumption ρ∗KX · c2(Y ) < 0, which is equivalent to ∆(TY ) · ρ∗KX < 0 as ρ∗KX · K 2 Y = K 3 X = 0 since ρ is isomorphic outside isolated points on X. Using Bogomolov inequal- ity [Lan04, Theorem 3.2], we immediately get that the tangent bundle TY is not strongly semistable, and in our situation we can show that TY is unstable. There- fore, we can use the results in [Lan15] to get a foliation F on Y , and a family of ρ∗KX -trivial rational curves covering X, hence the nef dimension of KX satis- fies n(X, KX ) ≤ 2. Under the assumption that n(X, KX ) ≤ 2, the nonvanishing is recently confirmed in [Wit17] in a more general setting. However, in our case, we can verify it by a straightforward argument: Consider the nef reduction map f : X (cid:57)(cid:57)(cid:75) Z, if n(X, KX ) = 2, then by the adjunction formula the generic fiber F of f has pa(F ) = 1, so the geometric fiber ¯F must be smooth since char k ≥ 5 by [Zha17, Prop. 2.9] (or [PW17, Cor. 1.8]), which contradicts that the reduction of general fibers of f are rational curves; if n(X, KX ) = 1, then f is a morphism, and KX is the pull-back of an ample divisor on Z hence is semiample. Now let us discuss the proof of Theorem 1.1. By [Bir16, Xu15, Zha17], it suffices to prove Theorem 1.1 for 0 ≤ ν(KX ) ≤ 2 and dim Pic0(X) = 0, and by Theorem 1.2 we can assume ρ∗KX · c2(Y ) ≥ 0. In this case, we closely follow the ideas in characteristic 0 firstly developed in [Miy88] (see also[Kol92, Secion 9]). However, the proof in [Miy88] used many ingredients from characteristic zero. Here we sketch our idea on how to find a positive characteristic replacement of Donaldson's theorem of stable vector bundles on smooth complex projective surfaces, and apply it to our proof: To start, for separating Case (2) and (3) in the proof of [Kol92, Theorem 9.0.6], the finiteness of the local algebraic fundamental group for a klt singularity x ∈ X NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 3 is needed in characteristic zero for [Kol92, Theorem 9.0.6, Case (3)]. We cannot prove this in full generality in characteristic p > 0 because of the possible wild ramifications. Nevertheless, the argument in [Xu14] implies that the tame funda- (x ∈ X) is always finite (see Theorem 3.4). Therefore, in positive mental group πloc,t characteristics, we separate the cases by considering whether the tame fundamental group πt 1(X sm) is finite. 1 1(X sm) is finite), we reduce to the case that πet Then in the case similar to [Kol92, Theorem 9.0.6, Case (2)] in characteristic zero (i.e., when πt 1 (X sm) is a pro-p-group. We prove Theorem 3.6 (generalizing a result of [Lan12, Lemma 8.1] which itself is built on [EM10]). This is precisely what we use to replace Donaldson's theorem on stable vector bundles on surfaces over C (see [Kol92, Page 113]). Then we can mimic the characteristic zero argument by applying it to the sheaf E obtained by an extension induced by some element in H 2(X, nKX )∗ and show that E (up to some covers) is not strongly stable. Eventually, by considering the destablizing X E, we construct sections of nKX . We note that the extension exact sequence of F e∗ involved may get split after being pulled back by some absolute Frobenius iterations, so further arguments using the construction originally from [Eke88] (see [Kol96, Section II. 6]) are needed. There are other smaller technical issues related to the differences between char- acteristic zero and characteristic p. • For the case ν(KX ) = 1 and when X sm has an infinite tame fundamental group as in [Kol92, Proof of Theorem 9.0.6, Case (3)], to follow Miyaoka's proof, we still need to deal with the failure of Hodge symmetry. Instead we will use Hodge-Witt numbers and the symmetry of such numbers (cf. [Ill83, Section 6.3] and [Eke86]). • Since it is not known whether terminal singularities are Cohen-Macaulay (this is only proved in large characteristics by [HW17]), dualizing complexes are used. • For the case ν(KX ) = 2, though Kawamata-Viehweg vanishing may fail in positive characteristic, a weaker vanishing theorem for surfaces (see [Szp79, Lan09, Tan15a]) is enough for our case. Notation and conventions: we follow [KM98] and [HL10] for the standard ter- minologies. Acknowledgement: We would like to thank H´el`ene Esnault, C.D. Hacon, Adrian Langer, Hiromu Tanaka, Liang Xiao and Weizhe Zheng for helpful conversations. We are also grateful to the referees for many valuable suggestions. Part of this work was done when the second author visited Fudan University Key Laboratory this summer, he thanks Meng Chen for his invitation. The first author was partially supported by NSFC Fund for Distinguished Young Scholars (No. 11425101). The second author was partially supported by NSFC (No. 11771260). 2. The case ρ∗KX · c2(Y ) < 0 In this section, we treat the case ρ∗KX · c2(Y ) < 0. This phenomenon only appears in positive characteristic. Recall the aimed theorem 4 CHENYANG XU AND LEI ZHANG Theorem 2.1. Let X be a minimal projective 3-fold over an algebraically closed field k with char k = p ≥ 5, and let ρ : Y → X be a smooth resolution of singu- larities. Assume that 0 < ν(X, KX ) < 3 and ρ∗KX · c2(Y ) < 0. Then the Kodaira dimension κ(X, KX ) = 1 and KX is semiample. Remark 2.2. Under the situation of Theorem 2.1, the case n(KX ) = 2 does not happen. This case is excluded at the end of the proof, using a property of the fibration of curves, which relies on the condition char k ≥ 5. We also remind that, our arguments can prove abundance under the situation that TY is not strongly semistable w.r.t. (D1, D2) where D1, D2 are introduced at the beginning of Sec. 2.4. This is a mildly stronger result than Theorem 2.1. Indeed, by Bogomolov inequality (Theorem 2.5), the condition ρ∗KX · c2(Y ) < 0 implies that TY is not strongly semistable w.r.t. (D1, D2). 2.1. Stabilities in positive characteristic. Let X be a smooth projective va- riety of dimension n over an algebraically closed field k with char k = p > 0, let D1, ..., Dn−1 be nef R-Cartier divisors, and let E be a torsion free coherent sheaf on X. The slope of E with respect to (D1, ..., Dn−1) is defined as µ(E) = c1(E) · D1 ··· Dn−1 . rk E The sheaf E is semistable (resp. stable) w.r.t. (D1, ..., Dn−1) if for every nontrivial subsheaf E(cid:48) ⊂ E µ(E(cid:48)) ≤ µ(E) (resp. < ). Recall that E has Harder-Narasimhan filtration ([HL10, Sec. I.1.3]) 0 = E0 ⊂ E1 ⊂ E2 ⊂ ··· ⊂ Em = E Denote by F : X ∼= X1 → X the Frobenius map. Recall that F ∗E has a such that the (Ei/Ei−1)s are semistable and µ(Ei/Ei−1) > µ(Ei+1/Ei). canonical connection ∇can : F ∗E → F ∗E ⊗ Ω1 which is determined by X1 for a ∈ OX1 and e ∈ E, ∇can : a ⊗ e (cid:55)→ da ⊗ e, and it yields the Cartier equivalence of categories between the category of quasi- coherent sheaves on X and the category of quasi-coherent OX1-modules with inte- grable k-connections, whose p-curvature is zero (cf. [Kat70, Theorem 5.1]). It is well known that in positive characteristic the stability is not preserved by the Frobenius pullback. We say that E is strongly semistable (resp. strongly stable) w.r.t. (D1, ..., Dn−1) if for every integer e ≥ 0 the pullback F e∗ X E is semistable (resp. stable). Let us recall some results on semistable sheaves in positive characteristic. Lemma 2.3. Let D1, ..., Dn−1 be ample Q-Cartier divisors. If E1 and E2 are (D1,··· , Dn−1)-semistable sheaves and p > rkE1 + rkE2 − 2, then E1 ⊗ E2 is semistable. If E is semistable and p > i(rkE − i) then ∧iE is semistable. (cid:3) Proof. See e.g. [Lan15, Corollary 1.2 and Remark 1.3]. Theorem 2.4. Let E be a semistable torsion-free sheaf such that F ∗E is unstable. Let be the Harder-Narasimhan filtration. Then the canonical connection ∇can induces non-trivial OX -homomorphisms Ei → (F ∗E/Ei) ⊗ Ω1 X . 0 = E1 ⊂ E2 ⊂ ··· ⊂ Em = F ∗E NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 Proof. See e.g. [Lan04, Corollary 2.4]. 5 (cid:3) Theorem 2.5 ([Lan04, Theorem 3.2]). Let E be a strongly (D1, ..., Dn−1)-semistable torsion-free sheaf. Then ∆(E) · D2 ··· Dn−1 ≥ 0 where ∆(E) = (rkE − 1)c1(E)2 − 2rkE · c2(E) denotes the discriminant. 2.2. Numerical and nef dimension. Let X be a normal projective variety over an algebraically closed field k and let D be a nef divisor on X. The numerical dimension ν(D) is defined as ν(D) = max{k ∈ NDk · An−k > 0 for an ample divisor A on X}. It is easy to see the definition does not depend on the choice of the ample divisor A. By the main result of [BCE+02] (see also [CTX15, Theorem 2.9]), if k is un- countable, then there exists a nonempty Zariski open set U ⊂ X and a proper morphism f : U → V , namely, the nef reduction map, such that D is numerically trivial on a very general fibre F of f , and for a very general point x ∈ U and C a curve containing x, we have that D · C = 0 if and only if C is contained in the fibre of f . The nef dimension of D is defined as n(D) = dim V . 2.3. 1-foliation. Let X be a smooth variety over an algebraically closed field k with char k = p > 0. Recall that a 1-foliation is a saturated subsheaf F ⊂ TX which is involutive (i.e., [F,F] ⊂ F) and p-closed (i.e., ξp ∈ F,∀ξ ∈ F). A 1-foliation F induces a finite purely inseparable morphism (cf. [Eke87]) π : X → Y = X/F = Spec(Ann(F) =defn {a ∈ OXξ(a) = 0,∀ξ ∈ F}), and if F is locally free then Y is smooth and KX ∼ π∗KY + (p − 1) detFX . We can often get 1-foliations when the tangent bundle is unstable by using the following criterion. Lemma 2.6. Let F be a saturated OX -submodule of TX . If HomX (∧2F, TX /F) = HomX (F ∗ XF, TX /F) = 0 then F is a 1-foliation. Proof. See e.g. [Kol92, 9.1.2.2], [Eke87, Lemma 4.2] or [Lan15, Lemma 1.5]. (cid:3) By studying the purely inseparable morphisms induced from 1-foliations, one can prove bend-and-break type results for foliations. Theorem 2.7. ([Lan15, Theorem 2.1]) Let L be a nef R-divisor on X. Let f : C → X be a non-constant morphism from a smooth projective curve C such that X is smooth along f (C). Let F ⊆ TX be a 1-foliation, smooth along f (C). Assume that c1(F) · C > KX · C p − 1 . Then for every x ∈ f (C) there is a rational curve Bx ⊆ X passing through x such that L · Bx ≤ 2 dim X pL · C (p − 1)c1(F) · C − KX · C . 6 CHENYANG XU AND LEI ZHANG 2.4. Proof of Theorem 2.1. Let H be an ample divisor on Y . We set the nef divisors D1 = ρ∗A if ν(KX ) = 1 or D1 = ρ∗KX if ν(KX ) = 2, and D2 = ρ∗KX where A is an ample divisor on X. Then D1 · D2 is nontrivial. Denote by µ(G) the slope of a coherent sheaf G on Y with respect to (D1, D2) and by µ(G) the slope with respect to (D1 + H, D2 + H). Before giving the proof let us recall some well-known facts: • the Harder-Narasimhan filtration of a coherent sheaf with respect to (D1 + H, D2 + H) is independent of sufficiently small > 0 ([Lan04, p. 263]); • if a sheaf E is µ-semistable for sufficiently small > 0 then it is µ-semistable and µ(E) = lim→0 µ(E); and • for two torsion free sheave E1,E2 on Y , if µmin(E1) > µmax(E2) then HomY (E1,E2) = 0. Since KY · D1 · D2 = 0 and the exceptional locus is contracted to those isolated singularities on X, it follows that µ(TY ) = µ(Ω1 Y ) = 0 and ∆(TY ) · ρ∗KX = 6c2(Y ) · ρ∗KX . Lemma 2.8. We have TY is not µ-semistable. Proof. Applying Theorem 2.5, the assumption c2(Y )· ρ∗KX < 0 implies that TY is not strongly µ-semistable. e > 0, F e∗ F e∗TY and let G(cid:48) = F e∗TY /G. Then We argue by contradiction. Assume TY is µ-semistable, and for some smallest Y TY is not µ-semistable. Let G be the maximal destablizing subsheaf of µ(G) > 0, µ(G(cid:48)) < 0 and µmax(G(cid:48)) < µ(G). Applying Lemma 2.4, there exists a nontrivial OY -module homomorphism η : G → G(cid:48) ⊗ Ω1 Y . By the definition of G, it is semistable, thus we have µ(G) = µmin(G) ≤ µmax(G(cid:48) ⊗ Ω1 Case rk G = 2: In this case rk G(cid:48) = 1, thus G(cid:48) ⊗ Ω1 contradiction by Y ). Y is µ-semistable. We get a Y ) = µ(G(cid:48)) + µ(Ω1 Y ) = µ(G(cid:48)) < µ(G). Y ) = µ(G(cid:48) ⊗ Ω1 Y is µ-semistable, then as µmax(G(cid:48) ⊗ Ω1 µ(G) ≤ µmax(G(cid:48) ⊗ Ω1 Case rk G = 1: If G(cid:48) ⊗ Ω1 Y ) = µ(G(cid:48) ⊗ Ω1 Y ), we get the same contradiction as above. Thus we may assume that G(cid:48) ⊗ Ω1 Y is not µ-semistable, hence for sufficiently small > 0, G(cid:48) ⊗ Ω1 Y is not (D1 + H, D2 + H)- semistable. Applying Lemma 2.3, since p ≥ 5, we have that at least one of G(cid:48) and Ω1 Y is not (D1 + H, D2 + H)-semistable. If G(cid:48) is not (D1 + H, D2 + H)-semistable, consider the Harder-Narasimhan filtration for sufficiently small > 0 0 → G(cid:48) By µ(G(cid:48) 1) > µ(G(cid:48) 2), taking limit we get 1 → G(cid:48) → G(cid:48) 2 → 0. µmax(G(cid:48)) ≥ µ(G(cid:48) 1) ≥ µ(G(cid:48) 2). NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 7 Since rk G(cid:48) i = 1, the sheaves G(cid:48) µ(G(cid:48) i ⊗ Ω1 It follows that HomY (G,G(cid:48) which is a contradiction. i ⊗ Ω1 Y ) = µ(G(cid:48) i⊗Ω1 Y are µ-semistable, and i) + µ(Ω1 Y ) = µ(G(cid:48) i) < µ(G). Y ) = 0. We can conclude that HomY (G,G(cid:48)⊗Ω1 Y ) = 0, For the remaining case, we assume G(cid:48) is semistable while Ω1 Y is not semistable with respect to (D1 + H, D2 + H). Consider the Harder-Narasimhan filtration of Ω1 Y with respect to (D1 + H, D2 + H) 0 = E0 ⊂ E1 ⊂ ··· ⊂ Ek = Ω1 Y where k = 2 or 3. Let Gi = Ei/Ei−1. Applying Lemma 2.3, the sheaf Gi ⊗ G(cid:48) is (D1 + H, D2 + H) (hence (D1, D2))-semistable. Since Ω1 Y is semistable with respect to (D1, D2), we in fact can show that µ(Gi) = µ(Ω1 Y ) = 0, thus µmax(Gi ⊗ G(cid:48)) = µ(Gi ⊗ G(cid:48)) = µ(G(cid:48)) < µ(G). Therefore HomY (G,Gi ⊗ G(cid:48)) = 0, and this implies HomY (G,G(cid:48) ⊗ Ω1 a contradiction again. Y ) = 0. We get (cid:3) Let F ⊂ TY be the maximal destablizing subsheaf, and let G = TY /F. Then c1(F) · D1 · D2 > 0. Lemma 2.9. The subsheaf F ⊂ TY yields a 1-foliation. Proof. If rk F = 1, then ∧2F = 0. And since µmax(G) < µ(F) < µ(F ∗ HomY (F ∗ Y F,G) = 0. It follows that F is a foliation by Lemma 2.6. Y F), we have If rk F = 2, then rk G = rk ∧2 F = 1 and thus HomY (∧2F,G) = 0. µ(F ∗ may assume that F ∗ Y F) = p· µ(F) > µ(G), if F ∗ µ(∧2F) = c1(F) · D1 · D2 > c1(G) · D1 · D2, It remains to prove that HomY (F ∗ Y F,G) = 0. Since Y F is also µ-semistable, then we are done. Now we Y F has a nontrivial Harder-Narasimhan filtration 0 → F1 → F ∗ Y F → F2 → 0. Therefore, there exists a nonzero homomorphism F1 → F2 ⊗ Ω1 Y , thus µ(F1) ≤ µmax(F2 ⊗ Ω1 Y ) = µ(F2) + µmax(Ω1 Y ) = µ(F2) + µ(G∗), i.e., µ(F2) ≥ µ(F1) + µ(G). Therefore, Since µ(F1) > µ(F ∗ Y F) = p · µ(F) > 0, we also have µ(F1) > µ(F2) > µ(G). HomY (F1,G) = HomY (F2,G) = 0 Y F,G) = 0. Lemma 2.10. The nef dimension which implies HomY (F ∗ (cid:3) n(Y, ρ∗KX ) = n(X, KX ) ≤ 2. 8 CHENYANG XU AND LEI ZHANG Proof. Suppose that A is sufficiently ample. Then for any sufficiently divisible m the divisor mKX + A is also very ample (see [Kee08]) on X. Take a smooth curve Cm as the intersection of two general divisors in ρ∗mKX + A. We can assume F is smooth along Cm. Then KY · Cm = KY · (mρ∗KX + ρ∗A)2 = 2mKY · ρ∗KX · ρ∗A + KY · (ρ∗A)2 = 2m(KX )2 · A + KX · A2 and c1(F) · Cm = c1(F) · (mρ∗KX + ρ∗A)2 = m2c1(F) · ρ∗K 2 X + 2mc1(F) · ρ∗KX · ρ∗A + c1(F) · (ρ∗A)2. For m (cid:29) 0, by the construction of F, in either case ν(X, KX ) = 1 or 2 we have c1(F) · Cm (cid:29) KY · Cm. In fact, when ν(X, KX ) = 1, the right hand side is a constant while the left hand side increases linearly with m; when ν(X, KX ) = 2, the right hand side is a linear term while the left hand side has a quadratic term in m. Set L = ρ∗r0KX where r0 is the Cartier index of KX . Applying Theorem 2.7, for general y ∈ Cm, there exists a rational curve By such that L · By ≤ 6 pL · Cm (p − 1)c1(F) · Cm − KY · Cm . Fix m (cid:29) 0. Since L · Cm ≤ r0KY · Cm, the intersection number L · By is a non- negative integer less than one, hence L· By = 0. As Cm moves, we can find a family of µ∗KX -trivial curves covering Y , which concludes that the nef dimension is at (cid:3) most 2. Proof of Theorem 2.1. Since it suffices to show the result after a base field exten- sion, we can assume k is uncountable. Consider the nef reduction map. If necessary, by blowing up Y we get a fibration f : Y → Z. We first exclude the case that Z is a surface. Otherwise denote by F a general fiber of f . By the construction, F does not intersect with exceptional divisors over X. We have that KY · F = ρ∗KX · F = 0, thus pa(F ) = 1. Since char k ≥ 5, by [Zha17, Prop. 2.9] (or [PW17, Cor. 1.8]) F is a smooth elliptic curve. However, by the construction in Step 3 the fiber F contains a rational curve as a component. We have that ν(KX ) = n(KX ) = 1, thus ρ∗KX ≡ f∗D for an effective divisor D > 0 on the curve Z by [BW17, Lemma 5.2]. This implies that ρ∗KX ∼Q f∗D(cid:48) by [Tan15b, Theorem 1.1] for some D(cid:48) ≡ D. Therefore, ρ∗KX is semiample, and (cid:3) the proof is completed. 3. Fundamental group and numerically flat vector bundles To mimic the proof from characteristic zero, which uses the dual relation be- tween the fundamental groups and the flat sheaves, we need to understand the corresponding picture in positive characteristics. NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 9 3.1. Local fundamental groups. In this section we will study the local funda- mental groups of three dimensional klt singularities in characteristic p > 5. In general, the local algebraic fundamental groups are finite for all klt singularities in characteristic zero (cf. [Xu14, BGO17]), and for strongly F -regular singularities (cf. [CRST16]) in all dimensions. In our situation, since wild ramification may happen, using the strategy in [Xu14], we can only prove that the local tame fundamental group is finite in the sense of Theorem 3.4. Definition 3.1 ([GM71, 2.1.2]). Let L/K be a finite separable extension of two fields, let v be a discrete valuation of K, and denote by Av the valuation ring, by k(v) the residue field and by Γv the valuation group. We say L is tamely ramified over K w.r.t. v if for each extension w of v (1)p (cid:45) [Γw : Γv]; and (2) the residue field extension k(w) is separable over k(v). Let π : X(cid:48) → X be a finite morphism between normal varieties and let D be a prime divisor on X. We say π is tamely ramified along D if the extension K(X(cid:48))/K(X) is tame w.r.t. vD. Geometrically this means that for any divisor D(cid:48) ⊂ X(cid:48) over D ⊂ X, the branch degree along D(cid:48) is not divisible by p and the residue field extension k(D) ⊂ k(D(cid:48)) is separable. Let X be a smooth variety, let D be a simple normal crossing divisor, and denote by RevD(X) the category of the Galois covers of X unramified over X \ D and tamely ramified along D. Corresponding to RevD(X), one can define the tame fundamental group w.r.t. D, denoted by πt 1(X, D) (see [GM71, Sec.4.2,4.3]). Lemma 3.2. Let π : X(cid:48) → X be a finite separable morphism between normal vari- eties over an algebraically closed field of characteristic p > 0. Let P be a codimen- sion one point on X and DP the corresponding prime divisor. Let D(cid:48) be the reduced divisor supported on π−1(DP ). We have KX(cid:48) + D(cid:48) − π∗(KX + D) ≥ 0 and the left hand side is positive around D(cid:48) if and only if the branch degree along a component DQ of D(cid:48) over DP is divisible by p. Proof. Let u and u(cid:48) be uniformizers of DP and a component DQ over it, and u → f · u(cid:48)r where f is a unit in OX(cid:48),Q. Thus du . So it is a (cid:3) generator if and only if r is not divisible by p. u is sent to f−1df + rdu(cid:48) u(cid:48) As an immediate consequence we get the following result. Corollary 3.3. Let π : X(cid:48) → X be a finite morphism between normal varieties over an algebraically closed field of characteristic p > 0. Assume that π is unramified over the smooth locus X sm. If X has at most terminal singularities, then X(cid:48) is terminal. Proof. We apply the calculation from characteristic zero as in [KM98, 5.20]. Lemma 3.2 implies that on the covering X(cid:48), the log discrepancy must be larger than or equal to the formula obtained in characteristic zero. In particular, if X is terminal, then (cid:3) X(cid:48) is terminal. Theorem 3.4. Let (X, x) be a germ of a three dimensional algebraic klt singularity over an algebraically closed field k with char k = p > 5. If ··· → (Xn, xn) ηn−→ (Xn−1, xn−1) → ··· → (X1, x1) η1−→ (X0, x0) = (X, x) 10 CHENYANG XU AND LEI ZHANG is any sequence of finite covers ´etale over X \{x} and tamely ramified around every divisor over x, then there exists some n0 such that for all n > n0 the covers ηn are isomorphisms. Proof. The proof follows closely to the argument in [Xu14], but we replace the fundamental group in characteristic zero by the tame fundamental group, since the latter behaves similarly as the ´etale fundamental group in characteristic zero (see e.g. [GM71]) By the standard construction of the Koll´ar component ([Xu14, Lemma 1]), we can find a divisor ∆ on X, an ample divisor H on X and a Koll´ar component E of (X, ∆) such that E is the unique divisor such that the discrepancy a(E, X, ∆+H) = −1. Let f : Y → X be the morphism which only extracts the Koll´ar component E. Then (Y, E) is plt and (KY + E + f−1∗ ∆)E = KE + Γ is antiample. Let Yi be the normalization of Xi ×X Y , and denote by fi : Yi → Xi and ψi : Yi → Y the natural projections. Since πi : Xi → X is tamely ramified, applying Lemma 3.2 we conclude that an exceptional divisor Fi over Xi has the discrepancy i (∆ + H)) = −1 if and only if it is dominant over E. Let Ei be one a(Fi, Xi, π∗ i (KY + E + f−1∗ ∆)Ei = KEi + Γi, which such a component, then we define Γi by ψ∗ is just the pull back of KE + Γ under the finite morphism γi : Ei → E. Since γi is separable by Definition 3.1(2), Lemma 3.2 implies that (Ei, Γi) is klt. Thus Ei yields the unique irreducible fi-exceptional component and it is a Koll´ar component of (Xi, ∆i = π∗ i ∆). Then as in characteristic zero case, we get a sequence of separable morphisms between klt pairs ··· → (Ei, Γi) → ··· → (E1, Γ1) → (E0, Γ0) with KEi + Γi = γ∗ i (KE + Γ). Fix an integer N such that N (KE + Γ) is Cartier. Then N (KEi + Γi) is Cartier. Since (Ei, Γi) are all 1 N )-lc log Fano surfaces [Ale94, Theorem 6.9], {Ei} are contained in a bounded family. As deg(γi)(−KE − Γ)2 = (−KEi − Γi)2 ≤ vol(−KEi) N -lc, by the boundedness of (= 1 Zl[1]) of Ohs is bounded from above, we conclude that for sufficiently big i and j > i, Ej → Ei is an isomorphism. be the generator of the tame fundamental group (∼=(cid:81) We complete the argument as in [Xu14, p.413]. By the above discussion, we may fix i∗ as above such that for any j > i∗, Ej → Ei∗ is isomorphic. Let γ PEi∗ ,Yi∗ (see [Gro03, XIII, 5.3]), where PEi∗ is the generic point of Ei∗ and Ohs PEi∗ ,Yi∗ is the strict henselization of the local ring OPEi∗ ,Yi∗ . Denote by PEi∗ ,Yi∗ ) → πloc,t the natural homomorphism. Since Ej → Ei∗ is an isomorphism for any j ≥ i∗, we (xi∗ , Xi∗ ) → Z/(dj) know Yj → Yi∗ is a totally ramified tame cover induced by πloc,t for some positive integer dj with (dj, p) = 1, which sends ρ(γ) ∈ πloc,t (xi∗ , Xi∗ ) to the generator [1] of Z/(dj). Therefore, it suffices to show that the image of γ in πloc,t In fact, to see this we can cut Yi∗ by a general hypersurface Ti∗ ⊂ Yi∗ to get 1 a surface Ti∗ . For the Z/(dj) quotient of πloc,t (xi∗ , Xi∗ ) which corresponds to Yj → Yi∗ , it induces a degree dj morphism η : Tj := Yj ×Yi∗ Ti∗ → Ti∗ which is degree dj totally ramified around the curve Ci∗ := Ei∗ ∩ Ti∗ . For this cyclic cover, (xi∗ , Xi∗ ) is a torsion, which we will verify in the next paragraph. (P hs Ei∗ ,Ohs ρ : πloc,t 1 (xi∗ , Xi∗ ) 1 p(cid:54)=l 1 1 1 NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 11 by considering the Galois group (∼= Z/(dj))-action, as gcd(dj, p) = 1, we obtain a decomposition of η∗OTj corresponding to the eigenvalues of the Galois action, which can be written as η∗OTjT sm k=0 L⊗k, restricting on the smooth locus T sm i∗ i∗ where L is a line bundle corresponding to the sections with eigenvalue being some i∗ (−Ci∗ ). Thus dj-th primitive root ξ. Moreover, we have the relation L⊗dj ∼ OT sm [Ci∗ ] = djαj for some αj ∈ Cl(Ti∗ ) corresponding to the line bundle L−1 on T sm i∗ Taking the intersection with [Ci∗ ] on Ti∗ , we have ∼= (cid:76)dj−1 . 0 > C 2 i∗ = dj(Ci∗ · αj), and [Ci∗ ]2 is a fixed rational number. Assume [Ci∗ ]2 = − s r where r, s are two positive integers prime to each other. As αj ∈ Cl(Ti∗ ), we know [Ci∗ ] · αj = Z, where e is the Cartier index of [Ci∗ ]. Thus dj ≤ es. (cid:3) i∗ ∈ 1 C 2 e 1 dj 3.2. Numerically flat vector bundles. A key tool used in the proof of nonvan- ishing in characteristic zero is Donaldson's theorem on the equivalence between flat bundles and semistable bundles with trivial Chern classes, which is a special case of Simpson's correspondence of stable sheaves. We note that in the proof in [Miy88], only the case when the variety has a trivial fundamental group is needed. In this section, we recall the definition and basic results on numerically flat vector bundles. They form a neutral Tannaka category whose Tannaka dual is used to define the S-fundamental group scheme in [Lan11]. Using such a theory, part of Simpson's correspondence can be recovered in characteristic p (see [Lan11, Lan12]). In partic- ular, together with the arguments in [EM10], we can obtain a similar description of numerically flat vector bundles on a variety with the fundamental group being 'small', i.e., only consisting of the p-part. This is precisely Theorem 3.6, which plays an analogous role as Donaldson's Theorem in Miyaoka's proof. It is a gen- 1 (X, x) = {1} is studied. eralization of [Lan12, Proposition 8.2], where the case πet For our purpose, we need to treat the case that πet 1 (X, x) is a pro-p-group. Definition 3.5. Let X be a smooth projective variety over an algebraically closed field k of characteristic p > 0. A vector bundle E on X is said to be numerically flat if both E and its dual E∗ are nef. There are several other equivalent characterizations of numerically flat sheaves (see [Lan12, Theorem 2.2]). For example, E is numerically flat if and only if E is a strongly H-semistable torsion free sheaf for some (hence for all) ample divisor H and ci(E) = 0 for every i ∈ Z. Theorem 3.6. Let X be a smooth projective variety over an algebraically closed field k of characteristic p > 0 and x ∈ X a closed point. Assume that πet 1 (X, x) is a pro-p-group. If E is a numerically flat vector bundle of rank r on X, then there Y π∗E ∼= ⊕rOY . exists an integer e > 0 and an ´etale cover π : Y → X such that F e∗ Proof. First we prove the following lemma, which says that torsion points in the moduli space of stable sheaves Mr ([EM10, Definition 3.12]) correspond to exactly OX . Lemma 3.7. If E(cid:48) is a stable vector bundle on X as in Theorem 3.6 and there exists some e > 0 such that F e∗E(cid:48) ∼= E(cid:48), then E(cid:48) ∼= OX . ´etale cover η : Z → X trivializing E(cid:48), i.e., η∗E(cid:48) ∼= Or Z, and E(cid:48) ∼= (η∗Or invariant isomorphism η∗E(cid:48) ∼= Or Proof. Applying Lange-Stuhler's theorem [LS77, Satz 1.4], there exists a Galois Z. We have a G(:= π1(Z/X))- Z)G, which corresponds to a 12 CHENYANG XU AND LEI ZHANG representation via the action of G on V = H 0(Z, η∗E(cid:48)) ∼= kr. ρ : G → GL(r, k) Since G is a p-group and char k = p, we have the following two immediate facts: (1) for g ∈ G, every eigenvalue of the action of g on V is equal to 1, thus the g-invariant subspace V g has rank > 0; (2) the center H (cid:67) G is a nontrivial p-group, and for every h ∈ H the h-invariant subspace V h is G-equi-invariant, i.e., gV h ⊆ V h, thus there exists a natural action of the quotient group G/ < h > on V h. By induction on the order of G we can conclude that rk V G > 0. From this we see ∼= OX . Therefore, if E(cid:48) is stable then E(cid:48) ∼= OX as that E(cid:48) contains a subsheaf E(cid:48) (cid:3) µ(E(cid:48)) = 0. 0 Granted the above lemma, then we can show the following lemma by an appli- cation of the proof of [EM10, Theorem 3.15] (see also [Lan12, Lemma 8.1]). We copy them here for readers' convenience. Lemma 3.8. If E(cid:48) is a strongly stable numerically flat vector bundle on X as above, then there exists e > 0 such that F e∗E(cid:48) ∼= OX . X )∗E(cid:48) are stable. Let N be Proof. By assumption all vector bundles En = (F n the Zariski closure of the set {[E0], [E1], ...,} in the moduli space Mr. If N has X )∗E(cid:48) is a torsion point dimension zero then some Frobenius pullback Ee = (F e of Mr. Since πet 1 (X, x) is a pro-p-group, we can apply Lemma 3.7 to show that Ee = OX . Hence we can assume that N has dimension at least one. Note that the set N(cid:48) of irreducible components of N of dimension greater than or equal to 1 is Verschiebung divisible (see [EM10, Definition 3.6]), because V N is defined at points En for n ≥ 1. Let N(cid:48) S by [EM10, Theorem 3.14]. Since the specialization morphism of the ´etale fundamental ¯Fp) is surjective for closed s ∈ S (see [Gro03, Chapter X groups πet Theorem 3.8], the latter is also a pro-p-group, which implies that the torsion points correspond to the trivial bundle by Lemma 3.7. Therefore, the trivial bundle is (cid:3) dense in N(cid:48), which is a contradiction. S be a good model of N(cid:48), then the torsion points of N(cid:48) S are dense in N(cid:48) 1 (X) → πet 1 (Xs×S Recall the following well known result which in fact holds for general projective normal varieties. Lemma 3.9. For an extension 0 → OX → V → OX → 0, there exists η : Z → X composed of a sequence of Z/(p)-´etale covers and Frobenius iterations of X such that the pull back extension 0 → OZ → η∗V → OZ → 0 splits. Proof. This is proved by the well known idea of 'killing cohomology'. By [Ser58, Proposition 12 and Section 9] (see also [HPZ17, end of the proof of Lemma 1.3]), the morphism η can be chosen to be a composite of Z/(p)-´etale covers and Frobenius (cid:3) morphisms. NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 13 Now let us finish the proof of the theorem. If e is sufficiently large, then the Jordan-Holder filtration 0 = E0 ⊂ E1 ⊂ E2 ⊂ ··· ⊂ Er = F e∗ X E has all the graded pieces being strongly stable (cf. [Lan04, Theorem 2.7]), and so they satisfy that Ei/Ei−1 By induction, applying Lemma 3.9 we can get an ´etale cover π : Y → X such that for e (cid:29) 0, π∗F e∗ (cid:3) ∼= OX by Lemma 3.8. Y π∗E ∼= Or Y . X E ∼= F e∗ 4. Nonvanishing In this section we aim to prove nonvanishing Theorem 1.1. Abundance has been proved when KX is big in [Bir16] [Xu15], dim Pic0(X) > 0 in [Zha16, Zha17], or ρ∗KX · c2(Y ) < 0 by Theorem 1.2. We only need to prove Theorem 4.1. Let X be a terminal projective minimal 3-fold over an algebraically closed field k of characteristic p > 5. Let ρ : Y → X be a smooth resolution of singularities. Assume that dim Pic0(Y ) = 0, ν(X, KX ) < 3 and ρ∗KX · c2(Y ) ≥ 0. Then κ(X, KX ) ≥ 0. Here we note that dim Pic0(X) = dim Pic0(Y ) as the fibers of ρ : Y → X are covered by rational curves (see [BW17, Theorem 1.1]). 4.1. Preliminaries. For readers' convenience, we recall some known results. The results in this section hold for any algebraically closed field of characteristic p > 0, i.e., we do not need p > 5 here. First recall the following well known result about Picard schemes. Theorem 4.2. Let X be a normal projective variety. Then the reduction Pic0(X)red of Pic0(X) is an abelian variety, and dim Pic0(X) ≥ h1(OX ) − h2(OX ). Proof. See e.g. [Fan05, Remark 9.5.15, 9.5.25]. (cid:3) Kawamata-Viehweg vanishing does not hold in positive characteristic, instead we will use a weaker vanishing on surfaces. Theorem 4.3. Let X be a smooth projective surface and L a nef and big line bundle on X. Then for i > 0 and sufficiently large integer n H i(X, KX + nL) = 0. Proof. This is first proved in [Szp79, Proposition 2.1]. For more general statements, (cid:3) see [Lan09, Proposition 2.24] and [Tan15a, Theorem 0.3]. The following result on Hodge-Witt numbers plays an analogous role as Hodge symmetry of Hodge numbers in dimension three. For a general background, see [Ill83, Section 6.3]. Theorem 4.4. Let Y be a smooth projective variety. j(−1)jhi,j Y ); j(−1)jhj(Ωi Y ) = (cid:80) Y ) = (cid:80) W (Y ) ≤ hj(Ωi (i) hi,j (ii) χ(Ωi (iii) if dim Y ≤ 3 then hi,j Hodge-Witt numbers of Y ; W (Y ) = hj,i W (Y ). W (Y ) where hi,j W (Y ) denote the 14 CHENYANG XU AND LEI ZHANG Proof. The inequality (i) is called Ekedahl's inequality (see [Ill83, Theorem 6.3.10] or [Eke86, p.86 Theorem 3.3]). The formula (ii) is known as Crew's formula (see e.g., [Ill83, (6.3.5)] or or [Eke86, p.85 Theorem 3.2]). For (iii) refer to [Eke86, p.113, (cid:3) Corollary 3.3 (iii)]. In the proof we come up with a non-split extension but whose pull-back via some absolute Frobenius iterations gets split. To treat this phenomenon, the idea is to use Ekedahl's construction of torsors. Theorem 4.5. Let L be a line bundle on a smooth projective variety X. Elements of the kernel of the Frobenius action X : H 1(X, L) → H 1(X, pL) F ∗ give rise to αL-torsors. Furthermore, the torsor arising from any non-trivial ele- ment of the kernel is non-trivial, that is a purely inseparable cover π(cid:48) : X(cid:48) → X of degree p, and KX(cid:48) ∼ π(cid:48)∗(KX + (p − 1)L). Proof. The construction is due to Ekedahl [Eke88]. See [Kol96, Sec.II.6] for more (cid:3) details. Corollary 4.6. Let X be a smooth projective variety of dimension d, L a line X : H 1(X, L) → H 1(X, pL) is nontrivial, M bundle on X such that the kernel of F ∗ a nef Cartier divisor and H an ample divisor on X. If −(KX + (p − 1)L) · H d−1 > 2dM · H d−1 then the nef dimension n(M ) ≤ d − 1. Proof. Let X(cid:48) be the variety induced by a nonzero element of the kernel of F ∗ X : H 1(X, L) → H 1(X, pL) as in Theorem 4.5. Denote by Z → X(cid:48) the normalization of X(cid:48) and by π : Z → X the natural composition morphism. There exists an effective divisor ∆ on Z, which arises from the conductor of the normalization Z → X(cid:48), such that KZ + ∆ ∼Q π∗(KX + (p − 1)L). For general z ∈ Z, there exists a curve C arising from the intersections of d − 1 divisors in kπ∗H, such that z ∈ C and that Z is smooth along C. Applying bend-and-break [Kol96, II Theorem 5.8], there exists a rational curve Γz passing through z such that π∗M · Γz ≤ 2d ≤ 2d π∗M · (kπ∗H)d−1 −KZ · (kπ∗H)d−1 π∗M · (kπ∗H)d−1 −π∗(KX + (p − 1)L) · (kπ∗H)d−1 ≤ 2dpkd−1M · H d−1 −pkd−1(KX + (p − 1)L) · H d−1 < 1, where the ratio being smaller than 1 is from the assumption. Since M is Cartier the intersection number π∗M ·Γz = 0. We conclude that the nef dimension n(M ) ≤ d − 1. (cid:3) In the same spirit, one can show the following vanishing result. NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 15 Corollary 4.7. Let X be a smooth projective variety and L an ample line bundle on X. If for every curve C the inequality ((p − 1)L − KX ) · C > dim X + 1 holds, then H 1(X,−L) = 0. Proof. This follows from [Kol96, II. Theorem 6.2]. (cid:3) The hard Lefschetz theorem will be used to construct global sections of a line bundle by considering its restriction on a sufficiently ample divisor. Theorem 4.8. Let X be a projective normal variety of dimension d ≥ 3. Let H be a smooth ample divisor contained in the smooth locus X sm of X such that, H 1(H,OH (−nHH )) = 0, for any n ≥ 1. Then Pic(X) (cid:44)→ Pic(H). Proof. This is [Gro05, XII, Corollary 3.6]. (cid:3) We will need the following simple vanishing of H 1(E ⊗ L−1) for a reflexive sheaf E and a sufficiently ample line bundle L. Lemma 4.9. Let X be a projective normal variety of dimension with dim X ≥ 2. Let E be a torsion free reflexive sheaf and H a very ample Cartier divisor on X. Then H 1(X, E(−nH)) = 0 for n (cid:29) 0. Proof. This is well known when E is a vector bundle, see [Har77, III 7.8]. For a general torsion free reflexive sheaf E on a normal projective variety, we know that there exists an exact sequence 0 → E → V1 → V2, where V1 and V2 are vector bundles. Then we conclude by a simple diagram chase. (cid:3) 4.2. Proof of Theorem 4.1. In this section, we aim to finish proving Theorem 4.1. We can assume the smooth resolution ρ : Y → X is isomorphic over X sm and KX · ρ∗c2(Y ) ≥ 0. Since dim Pic0(X) = 0 and h3(OX ) = h0(X, ωX ) ([Har77, p.241 Lemma 7.4]), by Theorem 4.2 χ(OX ) = h0(OX ) − (h1(OX ) − h2(OX )) − h3(OX ) ≥ h0(OX ) − h0(X, ωX ). We only need to consider the case h0(X, ωX ) = 0, thus we can assume χ(OX ) > 0. Lemma 4.10. For any integer n > 0 divisible by the Cartier index r0 of KX , χ(X,OX (nKX )) > 0. 16 CHENYANG XU AND LEI ZHANG Proof. Applying Riemann-Roch formula, we have χ(Y, ρ∗OX (nKX )) (ρ∗K 3 2n3 − 3n2 2n3 − 3n2 12 = = n 12 KX · (K 2 K 3 X + n 12 12 KX · ρ∗c2(Y ) + χ(OY ) n 12 = ≥ χ(OY ). X ) + (ρ∗KX ) · (K 2 Y + c2(Y )) + χ(OY ) X + ρ∗c2(Y )) + χ(OY ) For i > 0, the sheaves Riρ∗OY are supported at the union of those isolated irrational singularities on X. By χ(Y, ρ∗OX (nKX )) = χ(X, Rρ∗OY ⊗ OX (nKX )) = χ(X,OX (nKX )) − length(R1ρ∗OY ) + length(R2ρ∗OY ) = χ(X,OX (nKX )) + χ(OY ) − χ(OX ) we conclude that χ(X,OX (nKX )) = (χ(Y, ρ∗OX (nKX )) − χ(OY )) + χ(OX ) ≥ χ(OX ) > 0. (cid:3) If ν(X, KX ) = 0, then KX ≡ 0, and since dim Pic0(X) = 0, it follows that KX ∼Q 0. Thus following the proof of [Kol92, Theorem 9.0.6], we shall divide the argument into three cases. Case (1): ν(KX ) = 2. Let H be a smooth ample divisor on X contained in the smooth locus X sm of X. The exact sequence below 0 → OX (nKX ) → OX (nKX + H) → OH (nKX + H) → 0, induces a cohomological long exact sequence by taking cohomology → H 1(OH (nKX + H)) → H 2(OX (nKX )) → H 2(OX (nKX + H)) → ··· . Since KX is nef, applying Fujita vanishing (see [Kee03] or [Laz04, 1.4.35-36]) we can assume H is ample enough so that for any n > 0 divisible by r0, H 2(X,OX (nKX + H)) = 0. Note that KXH is a nef big Cartier divisor. Applying Theorem 4.3 for n0 (cid:29) 0 divisible by r0 H 1(OH (n0KX + H)) ∼= H 1(OH (KH + (n0 − 1)KXH )) = 0. Therefore H 2(OX (n0KX )) = 0, and in turn we have h0(OX (n0KX )) ≥ χ(OX (n0KX )) > 0. This finishes the proof of Case (1). (1) n > (4r0 + 2)KX · H 2 + H 3 KX · H 2 . NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 17 Remark 4.11. In the following we will focus on the case ν(KX ) = 1. As before, we can assume h2(OX (nKX )) > 0. We will split the case by considering whether πt 1(X sm) is finite or not. Here πt 1(X sm) is defined as the tame fundamental group 1(X, D) where X sm ⊂ X is a compactification with the boundary D = X \ πt X being simple normal crossing (see Sec. 3.1). For other equivalent definitions, see [KS10, Theorem 1.1]. In particular, it does not depend on the choice of the compactification X. Case (2): ν(X) = 1 and the tame fundamental group πt 1(X sm) is finite. 1(X sm 1 ) is trivial. This implies that πet By our assumption, we can take a finite cover π : X1 → X ´etale over X sm such 1 ) is a pro-p-group. Then π∗KX = that πt KX1 , and by Corollary 3.3 X1 also has terminal singularities, thus κ(X1) = κ(X). If dim Pic0(X1) = 0 then we have been done by [Zha17]. We can replace X with X1, and assume that πet 1 (X sm) is a pro-p-group. Note that if A is a smooth ample divisor on X contained in X sm, then πet 1 (X sm) by [Gro05, X. Example 2.2, Theorem 3.10]. 1 (A) ∼= πet 1 (X sm To start, we fix a smooth ample divisor H contained in X sm. Since KX is nef, by Fujita vanishing ([Kee03] or [Laz04, 1.4.35-36]) we can assume H is ample enough that H i(X, mKX + H) = 0 for i > 0 and every integer m > 0 divisible by r0. We fix an integer n divisible by r0 and such that Denote by ω• (OX (nKX ), ω• Take a nonzero element α ∈ Ext1OX non-split triangle X the dualizing complex. Applying Grothendieck duality, we get Ext1OX X [−3]) ∼= H 2(OX (nKX ))∗ (cid:54)= 0. (OX (nKX ), ω• X [−3]). Then α corresponds to a X [−3] → E• → OX (nKX ) → ω• ω• X [−2]. Taking the 0th cohomology gives the extension (∗) 0 → ωX → E → IT · OX (nKX ) → 0, where T is a closed subscheme of X supported at the union of those singularities of X and IT is its ideal sheaf. Lemma 4.12. The restriction of (∗) on H (2) 0 → OH (KX ) → F = EH → OH (nKX ) → 0 does not split. Proof. Since Ext1OX (OX (nKX ),OH (KX )) ∼= Ext1OH (OH (nKX ),OH (KX )), it suffices to prove that the restriction of α on H αH ∈ Im(Ext1OX (OX (nKX ), ω• X [−3]) rH−−→ Ext1OX (OX (nKX ),OH (KX ))) is nonzero. Applying RHomX (OX (nKX ),−) to the following triangle X [−2](−H) X [−3] → OH (KX ) → ω• X [−3](−H) → ω• ω• 18 CHENYANG XU AND LEI ZHANG gives the exact sequence Ext1OX (OX (nKX ),ω X [−3](−H)) → Ext1OX (OX (nKX ), ω • rH−−→ Ext1OX (OX (nKX ),OH (KX )) → Ext2OX (OX (nKX ), ω X [−3](−H)). • Applying Grothendieck duality, by the construction of H we have that for i = 1, 2 X [−3]) • ExtiOX (OX (nKX ), ω• X [−3](−H)) ∼= H 3−i(OX (nKX + H))∗ = 0. In turn we conclude the restriction map rH is an isomorphism. (cid:3) We need the following weak version of Theorem 3.6 for X which is not necessarily smooth. Lemma 4.13. There exists a finite cover η : Z → X ´etale over X sm such that η∗E is not strongly η∗H-stable. Proof. We may assume that E is strongly H-stable. Since K 2 X E) · H = 0. c1(E)2 · H = c2(E) · H = ∆(F e∗ X · H = 0, we have Take a smooth divisor A ∈ kH contained in X sm. Applying [Lan04, Theorem X EA is stable, 5.2], we can fix a sufficiently large k (independent of e) such that F e∗ hence the restriction F = EA is strongly H-stable. We note that [Lan04, Theorem 5.2] holds in the smooth setting, but in our case X has isolated singularities, so we can verify the assertion on a smooth resolution, as the reflexive hull of the pull back of E is strongly stable with respect to the pull back of H. Let G = End(F) ∼= F ⊗ F∗. Then G is strongly semistable by Lemma 2.3. We have c1(G) = 0 and c2(G) = 4c2(F) − c1(F)2 = 0. The sheaf G is numerically flat by [Lan12, Theorem 2.2]. Since πet 1 (A) is a pro-p- group, applying Theorem 3.6, we can find an ´etale cover ηA : A(cid:48) → A and e > 0 such that F e∗ A(cid:48) η∗ AG ∼= η∗ AF e∗ A G ∼= ⊕4OA(cid:48). It follows that dim HomA(cid:48)(F e∗ A(cid:48) η∗ AF, F e∗ A(cid:48) η∗ AF) = h0(End(F e∗ A(cid:48) η∗ AF)) = 4. 1 (A) ∼= πet By πet 1 (X sm), the cover ηA : A(cid:48) → A extends to a finite cover η : Z → X ´etale over X sm, where Z is assumed to be normal. We will show that η∗E is not strongly η∗H-stable. Since stable sheaves do not admit nontrivial endomorphisms, it suffices to prove that the sheaf Ge =defn EndZ(F e∗ Z η∗E) satisfies h0(Z, Ge) > 1. We fix an integer m such that m(p − 2)H − KX is ample, and at the beginning we can choose k > m + 4, then (p − 1)AA − KA = (p − 1)AA − (KX + A)A = ((p − 2)kH − KX )A satisfies the condition in Corollary 4.7. Obviously this property holds for A(cid:48) on Z. Applying Corollary 4.7 gives H 1(A(cid:48),OA(cid:48)(−lA(cid:48))) = 0 for every l > 0. Then by GeA(cid:48)(−lA(cid:48)) ∼= ⊕4OA(cid:48)(−lA(cid:48)), it follows immediately that H 1(A(cid:48), GeA(cid:48)(−lA(cid:48))) = 0 for every l > 0. NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 19 Consider the long exact sequence induced by taking cohomology of the short exact sequence 0 → GeA(cid:48)(−lA(cid:48)) → Ge(l+1)A(cid:48) → GelA(cid:48) → 0. We can show the restriction map H 0(A(cid:48), Ge(l+1)A(cid:48)) → H 0(A(cid:48), GelA(cid:48)) is surjective. By induction on l, we conclude h0( GelA(cid:48)) ≥ 4 for every l. Consider the exact sequence 0 → Ge(−lA(cid:48)) → Ge → GelA(cid:48) → 0. Then for a sufficiently large l, Lemma 4.9 tells that H 1(Z, Ge(−lA(cid:48))) = 0. In turn we conclude that dim H 0(Z, Ge) ≥ 4. (cid:3) Denote the pull back H(cid:48) = η−1(H) and ηH : H(cid:48) → H. We claim that the pull-back of the extension (2) via ηH (3) 0 → OH(cid:48)(η∗KX ) → η∗ HF = η∗EH(cid:48) → OH(cid:48)(η∗nKX ) → 0 does not split. For this we only need to show that the pull-back homomorphism γ : H 1(H,OH ((1 − n)KX )) → H 1(H(cid:48), η∗ HOH ((1 − n)KX )) is injective. Up to the isomorphism below H 1(H(cid:48), η∗ HOH ((1 − n)KX )) ∼= H 1(H, ηH∗OH(cid:48) ⊗ OH ((1 − n)KX )) the map γ fits into the following exact sequence H 0(H,C⊗OH ((1−n)KX )) → H 1(H,OH ((1−n)KX )) → H 1(H, ηH∗OH(cid:48)⊗OH ((1−n)KX )) which is induced by tensoring O((1 − n)KX ) with the natural exact sequence 0 → OH → ηH∗OH(cid:48) → C → 0 and taking the cohomology. To show the injectivity of γ, we only need to prove the vanishing H 0(H,C ⊗ OH ((1 − n)KX )) = 0. Denote by F e Since ηH : H(cid:48) → H is ´etale, we have the following commutative diagram e → H(cid:48) the eth absolute Frobenius iterations. H : He → H and F e H(cid:48) : H(cid:48) H(cid:48) ∼= H(cid:48) e ∼= ηH F e H(cid:48) / H(cid:48) ×H He / H(cid:48) ηH H ∼= He F e H / H Then F e∗ H C ∼= C. Since C is torsion free and ν(H, KXH ) = 1, if e (cid:29) 0 then F e∗ H (ηH∗OH(cid:48)) ∼= ηH∗(F e∗ H(cid:48)OH(cid:48)) ∼= ηH∗OH(cid:48) by [Har77, III.9.3], and thus H 0(H, F e∗ H (C ⊗ OH ((1 − n)KX ))) ∼= H 0(H,C ⊗ OH (pe(1 − n)KX )) = 0. We can conclude H 0(H,C ⊗ OH ((1 − n)KX )) = 0 since the map below is injective H : H 0(H,C ⊗ OH ((1 − n)KX )) → H 0(H, F e∗ F e∗ H (C ⊗ OH ((1 − n)KX ))). ( ( ' ' / / / 20 CHENYANG XU AND LEI ZHANG Let us proceed with the proof. In the following we will replace X, H with Z, H(cid:48) introduced in the lemma above, and replace the extension (∗) with its pull-back via η. We note that the inequality (1) still holds on Z and H(cid:48). X E is not H-stable for some e > 0. By Lemma 4.13, we may assume that F e∗ There exists an exact sequence µ-destabilizing F e∗ 0 → A → F e∗ X E. We may write that X E → B → 0 A∗∗ = OX (A) and B∗∗ = OX (B) where A, B are Weil divisors on X. Then we get nonzero homomorphisms η1 : OX (A) → OX (penKX ) and η2 : OX (peKX ) → OX (B). In particular both penKX−A and B−peKX are linearly equivalent to some effective divisors. Lemma 4.14. Let S be a smooth ample divisor on X contained in X sm. Then we have Q · AS = Q · KXS in NS(S) ⊗ Q. Proof. This follows from the same argument as in [Kol92, p.113-114]. For readers' convenience, we give all the details. Put AS = a, BS = b, SS = s and KXS = k. Then a + b = pe(n + 1)k, s · (a − b) ≥ 0 and a · b ≤ c2(F e∗ X ES) = 0. We claim that (a − b)2 = 0. Otherwise, (a − b)2 = (a + b)2 − 4a · b = −4a · b > 0, applying [Har77, V.1.8] shows that a − b ∈ C +(S) (i.e., the positive cone of S), so that h0(OS(m(a − b))) = O(m2) for m (cid:29) 0. Since b − pek and npek − a are both effective and mpe(n − 1)k = m(a − b) + m(b − pek) + m(npek − a), it follows that h0(OS(mpe(n − 1)k)) ≥ h0(OS(m(a − b))) = O(m2) for m (cid:29) 0 which contradicts that ν(S, k) = 1. Since OS(a) (cid:44)→ OS(penk) and OS(pek) (cid:44)→ OS(b), we get OS(a−b) (cid:44)→ OS(pe(n− 1)k). Since k2 = 0 and k is nef, we find that k · (a − b) = 0. Since k and (a − b) both lie in the closure of C +(S), the Hodge index theorem gives a − b ∈ Q · k in N S(S) ⊗ Q, thus a, b ∈ Q · k. (cid:3) If dim Pic0(S) > 0, we can always find an l((cid:54)= p)-torsion in Pic(S), which will induces a Z/(l) quotient of πet 1 (S) is a pro-p- group, Pic0(S) is of dimension zero. We can assume S is sufficiently ample to satisfy Lefschetz Hyperplane Theorem 4.8 Pic(X) (cid:44)→ Pic(S), indeed by 1 (S). Thus it follows from that πet (p − 1)nS − KS = (p − 1)nS − (KX + S)S = (((p − 1)n − 1)S − KX )S, for n > 0 if S is sufficiently ample then H 1(S,−nSS) = 0 by Corollary 4.7. From Lemma 4.14 we conclude that there exist positive integers n1, n2 such that n1A ∼ n2KX . The inclusion η1 : OX (A) (cid:44)→ OX (penKX ) induces another inclusion OX (n1A) ∼= OX (n2KX ) (cid:44)→ OX (n1penKX ). NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 21 We obtain that H 0(X, (n1pen − n2)KX ) (cid:54)= 0. If the pull-back of the extension (2) H OH (KX ) → F e∗ 0 → F e∗ (4) does not split, then OH (AH ) (cid:44)→ OH (penKXH ) is not an isomorphism, thus n1pen − n2 (cid:54)= 0. The theorem follows easily. The above argument applies for every smooth divisor of H contained in X sm. H E → F e∗ H OH (nKX ) → 0 The remaining case follows from the lemma below. Lemma 4.15. We assume for some n > (4r0 + 2)KX · H 2 + H 3 KX · H 2 and general smooth divisors in H, the extension (4) splits. Then KX is semiample. Proof. We have a minimal natural number 1 ≤ a ≤ e such that the class H OH ((1 − n)KX ))), H : H 1(OH ((1 − n)KX )) → H 1(F a∗ α ∈ ker(F a∗ while F (a−1)∗ H α is nonzero in H 1(F (a−1)∗ OH ((1 − n)KX )). Set L = pa−1(1 − n)KXH and M = r0KXH , H then by our assumption on n, we know −(KH + (p − 1)L) · H > 4M · H. Applying Corollary 4.6 shows n(H, KXH ) ≤ 1. As this holds for general mem- bers in H, we conclude n(X, KX ) ≤ 2, and then we can proceed to prove the claim (cid:3) by the same argument as in Theorem 2.1. Case (3): ν(X) = 1 and πt 1(X sm) is infinite. By assumption we have an infinite tower of nontrivial quasi-´etale Galois covers ··· → Xn → Xn−1 → ··· → X1 → X0 = X, which are all tamely ramified over X sm. By Theorem 3.4, there exists n0 such that for n > n0, the covers Xn+1 → Xn are all ´etale. Therefore, we can assume the resolutions Yn+1 → Yn are also finite ´etale for n > n0. Similarly as in the previous case, we have that Xn has terminal singularities and κ(X) = κ(Xn), and if dim Pic0(Xn) = 0 then we have been done by [Zha17]. So we can replace X with Xn, and always assume dim Pic0(X) = 0, thus χ(OY ) > 0 by the same argument as at the beginning of the proof. Up to a finite ´etale cover we can assume that χ(OY ) ≥ 4. Applying the properties of Hodge-Witt numbers summarized in Theorem 4.4, we know (−1)jh0,j W (Y ) = (−1)jhj,0 W (Y ) ≤ h0,0 W (Y ) + h2,0 W (Y ), j j (cid:88) χ(OY ) = which implies (cid:88) The linearly independent global sections of Ω2 Y . Take the saturation E ⊂ Ω2 Ω2 Y of the image of this homomorphism. Y give a OY -homomorphism O⊕3 Y → h0(Ω2 Y ) ≥ h2,0 W (Y ) ≥ 3. 22 CHENYANG XU AND LEI ZHANG If rk E = 3, then Ω2 Y is generically globally generated, thus H 0(Y, 2KY ) ∼= H 0(Y, det Ω2 Y ) (cid:54)= 0. If rk E = 1 or 2, then dim detE ≥ 1. We can write that detE = M + F where M denotes the movable part and F denotes the fixed part. Consider the sheaf F = E ⊗ ω−1 Let H be an ample divisor on X. Then Y ⊂ Ω2 ∼= TY . Y ⊗ ω−1 Y c1(F) · ρ∗KX · ρ∗H = c1(E) · ρ∗KX · ρ∗H + (rk E)(−KY ) · ρ∗KX · ρ∗H = c1(E) · ρ∗KX · ρ∗H ≥ 0. abundance holds in this case by Remark 2.2. If c1(E) · ρ∗KX · ρ∗H > 0, then TY is unstable with respect to (ρ∗KX , ρ∗H). So If c1(E)· ρ∗KX · ρ∗H = 0 then M · ρ∗KX · ρ∗H = 0. Assume H is a smooth ample divisor contained in the smooth locus X sm. By the Hodge index theorem we have Qρ∗MH = QKXH in NS(H) ⊗ Q, and by the above argument we can assume this holds for all sufficiently ample H. We conclude that Qρ∗M = QKX in NS(X) ⊗ Q. There exists some t ∈ Q such that ρ∗M ≡ tKX , which concludes ρ∗M ∼Q tKX since dim Pic0(X) = 0. In conclusion we complete the proof of all cases. References [Ale94] V. Alexeev, Boundedness and K2 for log surfaces, Internat. J. Math. 5 (1994), no. 6, 779 -- 810. [BCE+02] T. Bauer, F. Campana, T. Eckl, S. Kebekus, T. Peternell, S. Rams, T. Szemberg, and L. Wotzlaw, A reduction map for nef line bundles, Complex geometry (Gottingen, 2000), Springer, Berlin, 2002, pp. 27 -- 36. [BGO17] B. Bhatt, O. Gabber, and M. Olsson, Finiteness of ´etale fundamental groups by re- duction modulo p, arXiv:1705.07303 (2017). [Bir16] C. Birkar, Existence of flips and minimal models for 3-folds in char p, Ann. Sci. ´Ec. Norm. Sup´er. (4) 49 (2016), no. 1, 169 -- 212. [BW17] C. Birkar and J. Waldron, Existence of Mori fibre spaces for 3-folds in char p, Adv. Math. 313 (2017), 62 -- 101. [CRST16] J. Carvajal-Rojas, K. Schwede, and K. Tucker, Fundamental groups of F-regular sin- gularities via F-signature, arXiv:1606.04088, to appear in Ann. Sci. ´Ec. Norm. Sup´er. (4) (2016). [CTX15] P. Cascini, H. Tanaka, and C. Xu, On base point freeness in positive characteristic, Ann. Sci. ´Ec. Norm. Sup´er. (4) 48 (2015), no. 5, 1239 -- 1272. [DW16] O. Das and J. Waldron, On the Abundance Problem for 3-folds in characteristic p > 5, arXiv:1610.03403 (2016). [Eke86] T. Ekedahl, Diagonal complexes and F -gauge structures, Travaux en Cours., Hermann, Paris, 1986. [Eke87] , Foliations and inseparable morphisms, Algebraic geometry, Bowdoin, 1985 (Brunswick, Maine, 1985), 1987, pp. 139 -- 149. [Eke88] , Canonical models of surfaces of general type in positive characteristic, Inst. Hautes ´Etudes Sci. Publ. Math. 67 (1988), 97 -- 144. [EM10] H Esnault and V. Mehta, Simply connected projective manifolds in characteristic p > 0 have no nontrivial stratified bundles, Invent. Math. 181 (2010), no. 3, 449-465. [Fan05] B. Fantechi et al., Fundamental algebraic geometry, Mathematical Surveys and Mono- graphs, vol. 123, American Mathematical Society, Providence, RI, 2005. NONVANISHING FOR THREEFOLDS IN CHARACTERISTIC p > 5 23 [GM71] A. Grothendieck and J. Murre, The tame fundamental group of a formal neighbourhood of a divisor with normal crossings on a scheme, Lecture Notes in Mathematics, Vol. 208, Springer-Verlag, Berlin-New York, 1971. [Gro03] A. Grothendieck, Revetements ´etales et groupe fondamental (SGA 1), Docu- ments Math´ematiques (Paris) [Mathematical Documents (Paris)], vol. 3, Soci´et´e Math´ematique de France, Paris, 2003. [Gro05] , Cohomologie locale des faisceaux coh´erents et th´eor`emes de Lefschetz locaux et globaux (SGA 2), Documents Math´ematiques (Paris) [Mathematical Documents (Paris)], vol. 4, Soci´et´e Math´ematique de France, Paris, 2005. [Har77] Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977. Graduate Texts in Mathematics, No. 52. [HNT17] K. Hashizume, Y. Nakamura, and H. Tanaka, Minimal model program for log canonical threefolds in positive characteristic, arXiv:1711.10706 (2017). [HPZ17] C. Hacon, Z. Patakfalvi, and L. Zhang, Birational characterization of abelian varieties and ordinary abelian varieties in characteristic p > 0, arXiv:1703.06631 (2017). [HW17] C. Hacon and J. Witaszek, On the rationality of kawamata log terminal singular- ities in positive characteristic, arXiv:1706.03204 (2017). [HX15] C. Hacon and C. Xu, On the three dimensional minimal model program in positive characteristic, J. Amer. Math. Soc. 28 (2015), no. 3, 711 -- 744. [HL10] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves, Cambridge University Press, The Edinburgh Building, Cambridge, 2010. [Ill83] L. Illusie, Finiteness, duality, and Kunneth theorems in the cohomology of the de Rham-Witt complex, Algebraic geometry (Tokyo/Kyoto, 1982), Lecture Notes in Math., vol. 1016, Springer, Berlin, 1983, pp. 20 -- 72. [Kat70] N. Katz, Nilpotent connections and the monodromy theorem: applications of a result Turrittin, Inst. Hautes ´Etudes Sci. Publ. Math. 39 (1970), 175-232. [Kee03] D. Keeler, Ample filters of invertible sheaves, J. Algebra 259 (2003), no. 1, 243 -- 283. [Kee08] , Fujita's conjecture and Frobenius amplitude, Amer. J. Math. 130 (2008), no. 5, 1327 -- 1336. [KM98] J. Koll´ar and S. Mori, Birational geometry of algebraic varieties, Cambridge Tracts in Mathematics, vol. 134, Cambridge University Press, Cambridge, 1998. With the col- laboration of C. H. Clemens and A. Corti; Translated from the 1998 Japanese original. [Kol92] J. Koll´ar et al., Flips and abundance for algebraic threefolds (1992), 1 -- 258. Ast´erisque No. 211 (1992). [Kol96] J. Koll´ar, Rational curves on algebraic varieties, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics, vol. 32, Springer- Verlag, Berlin, 1996. [Kol13] , Singularities of the minimal model program, Cambridge Tracts in Mathemat- ics, vol. 200, Cambridge University Press, Cambridge, 2013. With a collaboration of S´andor Kov´acs. [KS10] M. Kerz and A. Schmidt, On different notions of tameness in arithmetic geometry, Math. Ann. 346 (2010), no. 3, 641 -- 668. [Laz04] R. Lazarsfeld, Positivity in algebraic geometry. I, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Math- ematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], vol. 48, Springer-Verlag, Berlin, 2004. Classical setting: line bundles and linear series. [LS77] H. Lange and U Stuhler, Vektorbundel auf Kurven und Darstellungen der algebraischen Fundamentalgruppe, Math. Z. 156 (1977), no. 1, 73 -- 83. [Lan04] A. Langer, Semistable sheaves in positive characteristic, Ann. of Math. (2) 159 (2004), no. 1, 251 -- 276. [Lan09] , Moduli spaces of sheaves and principal G-bundles, Algebraic geometry -- Seattle 2005. Part 1, 2009, pp. 273 -- 308. [Lan11] , On the S-fundamental group scheme, Ann. Inst. Fourier (Grenoble) 61 (2011), no. 5, 2077 -- 2119 (2012). [Lan12] , On the S-fundamental group scheme. II, J. Inst. Math. Jussieu 11 (2012), no. 4, 835 -- 854. [Lan15] , Generic positivity and foliations in positive characteristic, Adv. Math. 277 (2015), 1 -- 23. 24 CHENYANG XU AND LEI ZHANG [Miy88] Y. Miyaoka, On the Kodaira dimension of minimal threefolds, Math. Ann. 281 (1988), no. 2, 325 -- 332. [PW17] Z. Patakfalvi and J. Waldron, Singularities of general fibers and the LMMP, arXiv:1708.04268 (2017). [Ser58] J. Serre, Sur la topologie des vari´et´es alg´ebriques en caract´eristique p, Symposium internacional de topolog´ıa algebraica International symposium on algebraic topology, 1958, pp. 24 -- 53. [Szp79] L. Szpiro, Sur le th´eor`eme de rigidit´e de Parsin et Arakelov, Journ´ees de G´eom´etrie Alg´ebrique de Rennes (Rennes, 1978), Vol. II, 1979, pp. 169 -- 202. [Tan15a] H. Tanaka, The X-method for klt surfaces in positive characteristic, J. Algebraic Geom. 24 (2015), no. 4, 605 -- 628. [Tan15b] , Abundance theorem for surfaces over imperfect fields, arXiv:1502.01383 (2015). [Wit17] J. Witaszek, On the canonical bundle formula for fibrations of relative dimension one in positive characteristic, arXiv:1711.04380 (2017). [Xu14] C. Xu, Finiteness of algebraic fundamental groups, Compos. Math. 150 (2014), no. 3, 409 -- 414. [Xu15] , On the base-point-free theorem of 3-folds in positive characteristic, J. Inst. Math. Jussieu 14 (2015), no. 3, 577 -- 588. [Zha16] L. Zhang, Abundance for non-uniruled 3-folds with non-trivial Albanese maps in positive characteristics, to appear in J. Lond. Math. Soc., arXiv:1610.03637 (2016). [Zha17] , Abundance for 3-folds with non-trivial Albanese maps in positive character- istic, arXiv: 1705.00847 (2017). Current: MIT, Cambridge, MA 02139, USA Former: Beijing International Center for Mathematical Research, Beijing 100871, China E-mail address: [email protected], [email protected] Current: School of Mathematical Science, University of Science and Technology of China, Hefei 230026, P.R.China Former: School of Mathematics and Information Sciences, Shaanxi Normal Univer- sity, Xi'an 710062, China E-mail address: [email protected], [email protected]
1207.3684	1	1207	2012-07-16T14:04:25	On Canonical Homomorphisms of Tensor Sheaves	[ "math.AG" ]	In this paper we define tensor modules(sheaves) of Schur type,or of generalized Schur type associated with the give module(sheaf), using the so-called Schur functors. Then using global method we construct canonical homomorphisms between these modules(sheaves). We will get canonical isomorphisms if the original sheaf is locally free using idea of algebraic geometry, which is in fact a generalization of result in linear algebra. In the final section, we give canonical complexes using homomorphisms constructed before, and these complexes will become split exact sequence if further condition holds. And we could use local method to give concrete descriptions of these canonical homomorphisms.	math.AG	math	On Canonical Homomorphisms of Tensor Sheaves Jianke Chen∗ Abstract In this paper we define tensor modules(sheaves) of Schur type, or of generalized Schur type associated with the give module (sheaf), using the so-called Schurfunctors. Then using global method we construct canonical homomorphisms between these modules(sheaves). We will get canonical isomorphisms if the original sheaf is locally free using idea of algebraic geometry, which is in fact a generalization of result in linear algebra. In the final section, we give canonical complexes using homomorphisms con- structed before, and these complexes will become split exact sequence if further condition holds. And we could use local method to give concrete descriptions of these canonical homomorphisms. Keywords: Tensor Sheaves of Schur Types, Generalized isomorphism on determinant, Canonical Homomorphisms,Canonical Split Sequences. AMS Subject Classifications: 13A05, 1400, 15A15, 18E10, Acknowledgements: I wish to thank Prof.Ke-zheng Li, Prof. Zi-feng Yang for their heuristic dis- cussions with me on this topic. 1 Introduction Tensor Power, Symmetric Power and Wedge Power are basic tools used in mathe- matics. We are going to list its applications in linear algebra, representation theory and algebraic geometry. ∗Department of Mathematics, Capital Normal University, Beijing 100048,China. Email:[email protected] Tel:+86-13811897120 1 For linear algebra, let V be a n dimensional vector space over k, and f ∈ Endk(V), then using functorial property of tensor product, we may get the follow- ing homomorphism T r k(V). Similar homomorphisms can also be induced for symmetric power and wedge power. Then there exists the following equality det(T r k( f )) = (r−1)!(n−r)! . A special case of the statements before is the following equality: (det( f )) k( f )) = (det( f ))rnr−1, det(S r k( f )) = (det( f )) n!(r−1)! and det(∧r k(V) → T r k( f ) : T r (n+r−1)! (n−1)! 3ab2 3a2b a3 b3 a2c a2d + 2abc b2c + 2abd b2d ac2 bc2 + 2acd ad2 + 2bcd bd2 c3 d3 3c2d 3cd2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 6 a b c d (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) where a, b, c, d are elements of commutative ring R, see [5] or [6] for details. In these references, the proof is an algebraic proof, in this place we are going to prove similar results into a more general fact using geometric method. Next we are going to state some applications of these operations in representa- tion theory. For a given representation, we may use these operations to construct new representations, such as the basic example that if V is a finite dimensional complex vector space, then we have a canonical GL(V)−invariant decomposi- tion V ⊗ V (cid:27) S ym2(V) ⊕ ∧2(V), and more generally, we may write V ⊗ V ⊗ V (cid:27) S ym3(V) ⊕ ∧3(V) ⊕ another spaces, see [3] or [10] for details. Schur functors are generalized symmetric powers and wedge powers used in construction of new rep- resentations. As a general concrete example,there admits the following decompo- [ n 2 ]Li=0 sition, S ymn(S ym2V) (cid:27) S ym2n−4iV One basic fact for complex representation theory is that C is an algebraic closed field, so if we want to extend these re- sults into general base, similar results may NOT globally hold, which means these properties(isomorphisms, or equalities) hold if we add some localization property. Finally these operations occur in algebraic geometry, such as considering the Grassmanian space G(n, m), which means the set of m dimensional subspaces of a given n dimensional vector space. The standard procedure is taking r-th wedge of V, we denote this space by ∧r(V). In this method, every m dimensional subspace becomes 1 dimensional vector space, then as well known, the set of these 1 dimensional subspaces becomes a geometric object, namely P(∧r(V)), or equivalently Pro j(S (∧r(V))). In this construction, symmetric power and wedge power occurs naturally. In this paper, we will define tensor modules (resp. sheaves) of schur type, and more generally, generalized Schur type associated with general R−modules (resp. OX−modules), and investigate canonical homomorphisms between these kind of modules. In section 2, using the idea of algebraic geometry, we will give a geometric global canonical isomorphism of such kind of modules, if the origi- 2 nal module is locally free. In section 3, we will study several kinds of canonical homomorphisms of tensor sheaves. Note that for general tensor modules of Schur type, we can not use local coordinates as for the locally free case. The reason is for locally free sheaves, the gluing data information is reflected by a n × n in- vertible matrix, but for general sheaves, we know nothing about the gluing data information. The main idea for our method is that tensor sheaves of Schur type admit canonical permutation group's action. In fact, this method is already used in Prof. ke-zheng Li's lectures on de Rham Complexes. For details, see [7]. In the final section, we will give two concrete examples of general theorems, in which case that these canonical homomorphisms will make complexes. And un- der some further assumption, these complexes will become split exact sequences. We may use permutations to write these canonical homomorphisms clearly, and we will use local method (locally defined homomorphisms, then check these ho- momorphisms are independent of the given basis), and check that it it the same with global method. 2 SOME PREPARATIONS Let R be a commutative ring with 1, and M be an R-module. Using tensor product one can construct R-modules with M in the following ways: R(M) for any n ∈ N ); i) symmetric product ( S n ii) wedge product ( ∧n iii) tensor product over R. One can use the above constructions repeatedly, for a finite number of (but R(M) for any n ∈ N ); arbitrarily many) times. For example, S(M) = S r R(S n R(M) ⊗R ∧m R (M))⊗R2 (1) Such a construction procedure is called a Shur type. In many cases (e.g in al- gebraic geometry or representation theory) such kind of complex construction procedures of modules appear. Note that any Shur type is canonical, i.e. for any Shur type S, any R-module homomorphism f : M → N induces an R-module homomorphism S( f ) : S(M) → S(N) canonically. For any Shur type S, the construction procedure gives an epimorphism M⊗Rd → S(M). We call d the degreeof S, denoted by d(S). For example, it is not hard to see the Shur type in (1) has d(S) = 2r(n + m). In some cases one also uses the following way of construction in addition to i-iii): iv) homomorphism module HomR(·, ·), 3 especially for locally free modules of finite rank. For example, S′(M) = S r R(S n R(M) ⊗R HomR(∧m R (M), M⊗R2) (2) Such a construction procedure is called a generalizedShurtype. Let X be a scheme and F be a quasi-coherent sheaf on X. Then any Shur type S defines a quasi-coherent sheaf S(F ), because S is canonical. For example, if the Shur type S is as in (1), then S(F ) = S r OX(S n OX(F ) ⊗OX ∧m OX(F ))⊗OX 2) Furthermore, if f : F → F ′ is a homomorphism of quasi-coherent sheaves on X, then f induces a homomorphism S( f ) : S(F ) → S(F ′) canonically. If S′ is a generalized Shur type and F is locally free of finite rank, one can also define S′(F ). For example, if the Shur type S′ is as in (2), then S′(F ) = S r OX(S n OX(F ) ⊗OX H omOX(∧m OX(F ), F ⊗OX 2) Definition 2.1. Let R be a commutative ring with 1, and M be an R-module. By a tensormodule(resp. generalizedtensormodule) of M over R we mean a direct sum S(M) = S1(M) ⊕ · · · ⊕ Sn(M) (resp. S′(M) = S′ n(M)), for some Shur types S1,..., Sn (resp. generalized Shur types S′ 1(M) ⊕ · · · ⊕ S′ 1,..., S′ n). Let X be a scheme and F be a quasi-coherent sheaf on X. By a tensorsheaf of F we mean a direct sum S(F ) = S1(F ) ⊕ · · · ⊕ Sn(F ), for some Shur types S1,..., Sn. Furthermore, if F is locally free of finite rank, then by a generalized tensorsheafof F we mean a direct sum S′(F ) = S′ n(F ), for some generalized Shur types S′ 1(F ) ⊕ · · · ⊕ S′ 1,..., S′ n. By the above argument, we see that any S in Definition 1 gives a covariant functor MR → MR, where MR is the category of R-modules; and a covariant functor CohX → CohX, where CohX is the category of quasi-coherent sheaves over X. 3 CANONICAL ISOMORPHISM In this section, we are going to prove a global isomorphism of tensor sheaves of Schurtypeassociated with locally free sheave F of finite rank. Theorem 3.1. Let (X, OX) be a scheme and F be a locally free OX−module sheaf of rank n, E is a direct sum of tensor sheaves of schur type associated to F , E = 4 Li Ei, with each Ei is of rank mi and of degree di. Then we have the following canonical isomorphism of invertible sheaves: (detE ) (cid:27) (detF )⊗Pi midi n . For canonical, we mean if we set f : X′ → X be a morphism of schemes, set E ′ = f ∗E and set F ′ = f ∗F , then apply f ∗ to the above isomorphism, we get (detE ′) (cid:27) (detF ′)⊗Pi midi n . Proof. Our proof will be divided into several steps. Note that E = Li the wedge functor, we have: Ei, using detE (cid:27) (det(Li Ei)) (cid:27)Li1+i2 +···=d1 +d2+··· (cid:27) ∧m1(E1) ⊗OX ∧m2(E2) ⊗OX · · · (cid:27) det(E1) ⊗OX det(E2) ⊗OX · · · ∧i1(E1) ⊗OX ∧i2(E2) ⊗ · · · Hence we only need to prove for each i, the following equality holds: detEi (cid:27) (detF )⊗ midi n . So we may reduce the question to the case where E is a tensor sheaf of schur type of rank m and of degree d associated to free OX−sheaf F of rank n. setp1. First assume md n is an integer, We will prove this fact in the end of this step. Then detE and (detF )⊗ md n are all invertible sheaves, so locally checking, there always admits an isomorphism, the main problem is how to glue these local data to get a global isomorphism. Our method is to prove this iso- morphism is canonical. Take an affine open neighborhood U = Spec(R) of X, then F U (cid:27) eM, where M is free R−module of rank n and E U (cid:27) fM′ is a ten- sor module of schur type associated to M. Suppose M admits a basis mi such that ϕ mi = λimi for some ϕ ∈ GL(n, R) and λi ∈ R , then according to our def- inition of tensor modules of schur type, which is in fact the functorial property of these operations(symmetric products, for example), M′ has a canonical basis which is induced from mi, say m′ i, and ϕ induces an automorphism ϕ′ acts on M′, hence ϕ′m′ i for each i, and each µi is a product of λis of length d, so the i sum of these eigenvalues is dm(with multiplications). Also note that GL(n, R)'s diagonal action is also kept to M′, so all the multiplications of λi in µi should be equal, which means the multiplication is just md n . step2. Using notations as above, S r RM has a canonical basis inherited from M, say mi1 ⊗ · · · ⊗ mir with i1 ≤ i2 ≤ · · · ≤ ir and ∧r RM has a canonical basis inherited from M, which is mi1 ∧ · · · ∧ mir with i1 < i2 < · · · < ir. According to our definitions, detM′ have a canonical basis inherited from given basis of M. Hence = µim′ 5 we have as (detM)⊗R an isomorphism from one generator to the other generator, n and detM′ are free R−modules of rank 1, we may define md φ : (detM)⊗R md n −→ detM′ (m1 ∧ · · · ∧ mn)⊗ md n 7→ m′ 1 ∧ · · · ∧ m′ m . Next we need to prove this is a canonical isomorphism, which means if we take α ∈ GL(n, R) ,then α also induces an automorphism on M′ ,say α′ , then there exists the following commutative diagram: (detM)⊗R md n −−−−−→ detM′ φ φ detα′y md n y(detα) md n (detM)⊗R −−−−−→ detM′ So we only need to check detα′ = (detα) md n for all α ∈ GL(n, R) .Note that accord- ing to step1, if α admits a basis such that α is diagonal, then the equality already holds! As an element α = (αi j) ∈ GL(n, R) corresponds to an algebraic homomor- phism (Z[xi j]1≤ i, j≤ n)det(xi j) → R , which is defined by αi j 7→ xi j. In fact this is a morphism from Spec(R) to Spec((Z[xi j]1≤ i, j≤ n)det(xi j)) , so in order to check the equality over R , we only need to check it over X = S pec((Z[xi j]1≤ i, j≤ n)det(xi j)). step3. As detα′ = (detα) md n defines a closed subset of X, say V. We are going to prove it contains a nonempty open subset U , hence the taking closure we have ¯U ⊂ ¯V ⊂ ¯V. As X is irreducible, every nonempty open subset is dense, hence we have V = X. Set U = {ϕ χϕ has n different eigenvalues}, according to step1, we have U ⊂ V. Also note that χϕ has n different eigenvalues if and only if △ϕ , 0, so U is an open subset of X. Set R0 = (Z[xi j]1≤ i, j≤ n)det(xi j), K = q.f.(R0) and ϕ′ = (xi j). Then χϕ′ = λ I − (xi j) be the characteristic polynomial of (xi j), and L/K be the splitting filed of χϕ′ over K, then ϕ′ has n different eigenvalues λi ∈ L such that ϕ′vi = λivi for vi ∈ Ln for all i = 1, · · · , n. We may take a common de- nominator f such that these equalities hold in a open neighborhood of the generic point of R0, hence U , ∅, taking closure we have V = X. As the isomorphism is canonical, hence we may glue together to get a global isomorphism. (cid:3) 4 Canonical Homomorphisms In this section,we are going to study canonical homomorphisms of tensor sheaves. More precisely,we will construct canonical homomorphisms for OX-module F . 6 In order to define homomorphisms of tensor sheaves of F , note that if F is locally free of rank n, we can use local generators of F (U) for open subset U ⊂ X. The reason is that for locally free sheaves, when we take local generators, any two different choices of generators are differed by a n × n invertible matrix, and the gluing data is also reflected by n × n invertible matrix, so in order to check the morphism is well defined, we just need to fix a basis, and prove the morphism is independent of the basis we choose before. But for general OX−module F , we don't know any information of local generators and information of gluing data. But for general tensor sheaves, it admits a canonical permutation group's action, this is the point we use in this paper. In fact the main idea is already occurred in the de Rham Complex in Prof. Ke-Zheng Li's lectures in "Moduli Space And Its Applications". For details, see [7]. Let k and n be two integers, (X, OX) be a scheme and F be an OX−module. Theorem 4.1. There exists canonical homomorphism ϕn,k : S n OX (S k OX F ) −→ S k OX (∧n OX F ). if k is an even integer. There exists canonical homomorphism φk,n : S k F ) −→ S n OX(S k OX F ). OX OX(∧n (S k OX (∧n OX F ) are all quotient sheaves of F ⊗(n+k), so in Proof. As S k OX order to define ϕn,k and φk,n, we only need to define endomorphisms of F ⊗(n+k) such that ϕn,k and φk,n can be induced from these endomorphisms. F ) and S n OX Note F ⊗(n+k) admits a canonical permutation group Sn+k's action, so in order to define global homomorphisms, we could use these permutations. Take σ2, · · · , σn ∈ Sk, define ϕn,k as follows: F ⊗(n+k) (a11 ⊗ · · · ⊗ a1k) ⊗ · · · ⊗ (an1 ⊗ · · · ⊗ ank) −→ 7→ F ⊗(n+k) (a11 ⊗ a2σ2(1) ⊗ · · · ⊗ anσn(1))⊗ Xσ2,··· ,σn∈Sk (a12 ⊗ a2σ2(2) ⊗ · · · ⊗ anσn(2)) ⊗ · · · ⊗(a1k ⊗ a2σ2(k) ⊗ · · · ⊗ anσn(k)) Or equivalently, we may write this homomorphism as ⊗n i=1(⊗k j=1ai j) 7→ Xσ2,··· ,σn∈Sk ⊗k i=1(a1i ⊗ (⊗n j=2a jσ j(i))). It is not difficult to verify on ideal of definitions for these 2 tensor sheaves F ). We denote it that ϕn,k induces a homomorphism from S n OX F ) to S k OX (∧n OX (S k OX 7 by ϕn,k. If we write this homomorphism in local coordinates, it is ⊗n i=1(⊗k j=1ai j) 7→ Pσ2,··· ,σn∈Sk kQi=1 (a1i ∧ (∧n j=2a jσ j(i))). In order to define φk,n, we still use the fact F ⊗(n+k) have permutation group's canonical action. Take τ2, · · · , τk ∈ Sn, we define φk,n as follows: F ⊗(n+k) (a11 ⊗ · · · ⊗ a1n) ⊗ · · · ⊗ (ak1 ⊗ · · · ⊗ akn) −→ 7→ Xτ2,··· ,τk∈Sn F ⊗(n+k) (−1)τ(τ2···τk)(a11 ⊗ a2τ2(1) ⊗ · · · ⊗ akτk(1)) ⊗(a12 ⊗ a2τ2(2) ⊗ · · · ⊗ akτk(2))⊗ · · · ⊗ (a1n ⊗ a2τ2(n) ⊗ · · · ⊗ akτk(n)) Or equivalently we may write this homomorphism as ⊗k i=1(⊗n j=1ai j) 7→ Xτ2,··· ,τn∈Sn (−1)τ(τ2 ···τk) ⊗n i=1 (a1i ⊗ (⊗k j=2a jσ j(i))). It is easy to verify that φk,n induces a homomorphism from S k OX (S k OX (∧n OX F ), which means we have the following commutative diagram: S n OX F ) to F ⊗(k+n) φk,n F ⊗(k+n) S k OX (∧n OX F ) / S n OX (S k OX F ) We denote this homomorphism by φk,n. If we write this homomorphism in local coordinates, it isQk This finishes the proof. i=1(∧n j=1ai j) 7→ Pτ2,··· ,τk∈Sn ((−1)τ2···τk ⊗n i=1 (a1i ⊗ (⊗k j=2a jτ j(i)))). (cid:3) But unfortunately for most cases we can not get the composition properties for ϕn,k ◦ φk,n and φk,n ◦ ϕn,k, even when F is locally free. The main problem is although ϕn,k and φk,n are globally well defined, it still has too many additional combinational terms if we consider compositions for these homomorphisms. A special case is ϕ2,k ◦ φk,2 when F is locally free OX−module of rank 2. In this case, we have φk,2((x ∧ y)k) = (k−i)!(i−1)! xi ⊗ yk−i · xk−i ⊗ yi). ϕ2,k(xi ⊗ yk−i · xk−i ⊗ yi) = (−1)k−i(k − i)! i!(x ∧ y)k. ((−1)k−i (k−1)! kPi=1 So the composition is a scalar multiplication by i · (k − 1)! = (k+1)! 2 kPi=1 . So we have ϕ2,k ◦ φk,2 = (k+1)! kPi=1 2 (−1)k−i (k−1)! (k−i)!(i−1)! (−1)k−i(k−i)! i! = · idS k OX (∧2 OX F ). We also compute 8 / / / the case ϕ3,2 ◦ φ2,3 when F is locally free of rank 3, it is a scalar multiplication by 12, which is equal to (2+3−1)! . Hence we think the general case is still true. Namely the equality 2 ϕn,k ◦ φk,n = if F is locally free of rank n. (k + n − 1)! 2 · idS k OX F ) . (∧n OX Next we are going to use this theorem to get canonical homomorphisms of S 2 (S n F ), where F is an OX−module. Our aim to study this kind of sheaf OX OX comes from [3], as there exists a decomposition of this kind of module. But in there, they deal with representation theory, the base is C, we want to get more general result over general base X and the main idea is directed by [3]. First we will state the two steps handled in [3] if F is a locally free OX−module F ) F ) which makes the kernel for the upper canonical surjective homo- of rank 2. There exists a canonical homomorphism from (∧2 OX to S 2 OX morphism. If so we could get the short exact sequence: F )⊗2 ⊗S 2 OX (S n−2 OX (S n OX 0 → (∧2 OX F )⊗2 ⊗ S 2 OX(S n−2 OX F ) −→ S 2 OX(S n OX F ) −→ S 2n OX F → 0. (S n OX Then the problem is whether we can construct a canonical homomorphism from S 2 F ) which forms a canonical retraction for the OX upper exact sequence. Of course this canonical retraction should be constructed from the following commutative diagram: F ) to (∧2 OX F )⊗2⊗S 2 OX (S n−2 OX F ⊗2n unknown F ⊗2n S 2 OX (S n OX F ) / (∧2 OX F )⊗2 ⊗ S 2 OX (S n−2 OX F ). If these 2 steps are correct, we may write S 2 OX Hence we may decompose the locally free sheaf S 2 OX (S n OX (S n−2 OX F ). F ) as a direct sum by induction. But unfortunately what we have done now is not as above. We can construct global homomorphisms. The main problem is although these homomorphisms are globally well defined, as before they have too many combinational terms. Hence for composition they don't have good properties, even when F is locally free OX−module of rank 2. Let n ≥ 3 be an integer. Theorem 4.2. There exists the following canonical homomorphisms S 2 OX (∧2 OX F ) ⊗ S 2 OX (S n−2 OX F ) i / S 2 OX (S n OX F ) q / S 2n OX F 9 / / / / / and S 2 OX (∧2 OX F ) ⊗ S 2 OX (S n−2 OX F ) j S 2 OX (S n OX F ) ϕ S 2n OX F such that q ◦ i = 0 , q ◦ ϕ is a scalar multiplication of S 2 OX F by (2n−1)···(n+1) (n−1)! . Proof. It is easy to see there exists a canonical surjective homomorphism S 2 OX(S n OX F ) −→ S 2n OX F → 0. which is induced by the identity map on F ⊗2n → F ⊗2n. We denote this quotient homomorphism by q. Next set N be a subset of the set {1, 2, · · · , 2n} such that 1 ∈ N and N = n. Define a homomorphism ϕ as follows: ϕ : F ⊗(2n) a1 ⊗ · · · ⊗ a2n F ⊗(2n) ((⊗i∈ Nai) ⊗ (⊗ j¯∈Na j)) . −→ 7→ PN It is not difficult to check that ϕ induces a homomorphism from S 2n OX Namely we have the following commutative diagram: F to S 2 OX (S n OX F ). F ⊗(2n) ϕ F ⊗(2n) S 2n OX F / S 2 OX (S n OX F ) We denote this homomorphism by ϕ. For q ◦ ϕ, note that for a general term α ∈ S 2n OX F , ϕ(α) is just proper per- mutations of coordinates of α, it does not have any other terms, so if mapped into S 2n F again, they are all the same element. Hence we only need to compute OX the multiplication. According to the definition of the subset N, we have q ◦ ϕ = (2n−1)···(n+1) (n−1)! Next we will define i and j. First note that according to the theorem before, OX · idS 2n F . we have canonical homomorphisms φ2,2 : S 2 OX (∧2 OX F ) −→ (a11 ∧ a12) · (a21 ∧ a22) 7→ and ϕ2,2 : S 2 OX (S 2 OX F ) −→ (a11 ⊗ a12) · (a21 ⊗ a22) 7→ 10 S 2 OX (S 2 OX F ) (a11 ⊗ a21) · (a12 ⊗ a22)− (a11 ⊗ a22) · (a12 ⊗ a21) S 2 OX (∧2 OX F ) (a11 ∧ a21) · (a12 ∧ a22)+ (a11 ∧ a22) · (a12 ∧ a21) . . o o o o / / / Where ai j are local coordinates of F for i, j = 1, 2. So in order to define i and j, we only need to define homomorphisms f : S 2 OX(S 2 OX F ) ⊗ S 2 OX(S n−2 OX F ) → S 2 OX(S n OX F ). and g : S 2 OX(S n F ) → S 2 OX(S 2 OX F )) and j = (ϕ2,2 ⊗ idS 2 OX F ) ⊗ S 2 OX F ). OX(S n−2 F )) ◦ g. In order to check set i = f ◦ (φ2,2 ⊗ idS 2 the composition properties, we just need to check it on local generators. (S n−2 OX (S n−2 OX OX OX Define f and g as follows: f : F ⊗(2n) (t1 ⊗ t2 ⊗ t3 ⊗ t4) ⊗ ((a1 ⊗ · · · ⊗ an−2) · (b1 ⊗ · · · ⊗ bn−2)) −→ 7→ F ⊗(2n) (t1t3a1 · · · an−2) ⊗ (t2t4b1 · · · bn−2) +(t1t4a1 · · · an−2) ⊗ (t2t3b1 · · · bn−2) +(t2t4a1 · · · an−2) ⊗ (t1t3b1 · · · bn−2) +(t2t3a1 · · · an−2) ⊗ (t1t4b1 · · · bn−2) g : F ⊗(2n) (a1 · · · an)⊗ (b1 · · · bn) −→ 7→ Xi< j Xk<l F ⊗(2n) (ai ⊗ bk · a j ⊗ bl + ai ⊗ bl · a j ⊗ bk)· It is not difficult to check on ideal of definitions that f and g induce homomor- phisms f and g. In order to check the computation properties we write i and j in local coordinates. We have (a1 · · ·bai · · ·ba j · · · an) ⊗ (b1 · · ·bbk · · ·bbl · · · bn) i((x1 ∧ y1 · x2 ∧ y2) · (a1 ⊗ · · · ⊗ an−2 · b1 ⊗ · · · ⊗ bn−2)) = (x1y2a1 · · · an−2) ⊗ (x2y1b1 · · · bn−2) + (x2y1a1 · · · an−2) ⊗ (x1y2b1 · · · bn−2)− (x1x2a1 · · · an−2) ⊗ (y1y2b1 · · · bn−2) − (y1y2a1 · · · an−2) ⊗ (x1x2b1 · · · bn−2) And j ((a1 ⊗ · · · ⊗ an) · (b1 ⊗ · · · ⊗ bn)) −(ai ∧ bl · a j ∧ bk + ai ∧ bk · a j ⊗ bl)· = Pi< j Pk<l (a1 · · ·bai · · ·ba j · · · an) · (b1 · · ·bbk · · ·bbl · · · bn) Hence we define all the homomorphisms. It is not difficult to see that for a general element α ∈ S 2 F ), i(α) is just permutations of coordinates OX of α,when mapped into S 2n F , they are the same element. Also note that these OX terms have different signs, hence q ◦ i = 0. F ) ⊗ S 2 OX (S n−2 OX (∧2 OX This finishes the proof. (cid:3) 11 5 Canonical Decompositions In this final section, we are going to give 2 concrete examples of the theorems in section2. First we are going to prove the following lemma, which is an exercise in [4], which generalized the exercise in[6]. Lemma 5.1. Let (X, OX) be a scheme, suppose 0 → F ′ → F → F ′′ → 0 is an OX−module's exact sequence, F ′, F and F ′′ are locally free of rank m, m + n and n . Then we have ∧m+n OX F (cid:27) ∧m OX F ′ ⊗OX ∧n OX F ′′. (3) Proof. It is easy to see that ∧m+n F ′′ are both locally OX free R−modules of rank 1. Use the functorial property of ∧ and the exact se- quence of 0 → F ′ → F → F ′′ → 0 we may get canonical surjective homomor- phism ∧n F ′ ⊗OX − for the upper sequence and taking OX the kernel we get exact sequence F ′′ → 0 , ∧m OX F ′ ⊗OX ∧n OX F and ∧m OX F → ∧n OX 0 → Ker(α) → ∧m OX F ′ ⊗OX ∧n OX F α −→ ∧m OX F ′ ⊗OX ∧n OX F ′′ → 0. (4) F which are locally generated by sections of the form (a′ From the definition it is easy to see that Ker(α) is a subsheaf of ∧m F ′ ⊗OX OX ∧n m) ⊗ OX (a1 ∧ · · · ∧ an) with ai ∈ M′ for some i = 1, · · · , n. Also note that there ex- ists a canonical homomorphism from ∧m F by map- OX m ∧ a1 ∧ · · · ∧ an, in fact one only ping (a′ need to compute there exists the following commutative diagram: m, a1 ∧ · · · ∧ an) to a′ F → ∧m+n OX F ⊗OX ∧n OX 1 ∧ · · · ∧ a′ 1 ∧ · · · ∧ a′ 1 ∧ · · · ∧ a′ F ⊗OX (m+n) id F ⊗OX (m+n) ∧m OX F ⊗ ∧n OX F / ∧m+n OX F . Also we have a canonical homomorphism ∧m OX Taking composition we get F ′⊗OX ∧n OX F → ∧m OX F ⊗OX ∧n OX F . ∧m OX F ′ ⊗OX ∧n OX F → ∧m+n OX F . Denote this morphism by β . Taking the kernel we have 0 → Ker(β) → ∧m OX F ′ ⊗OX ∧n OX F → ∧m+n OX F . (5) From the construction it is easy to see that Ker(β) is subsheaf of ∧m OX ∧n OX F locally generated by sections of the form (a′ F ′ ⊗OX m)⊗(a1 ∧· · ·∧ an) with 1 ∧· · ·∧ a′ 12 / / / some a′ i of Ker(α) . So we may get the following commutative diagram = a j for some pair i , j . Then it is easy to see that Ker(β) is a submodule 0 → Ker(β) → ∧m OX F ′ ⊗OX ∧n OX F ↓ ↓ β −→ (∧m OX F ′ ⊗OX ∧n OX ↓ F )/Ker(β) → 0 0 → Ker(α) → ∧m OX F ′ ⊗OX ∧n OX F α −→ ∧m OX F ′ ⊗OX ∧n OX F ′′ → 0 F ′⊗OX ∧n OX F )/Ker(β) to ∧m OX According the snake lemma we get a canonical surjective homomorphism from (∧m OX locally free OX−module of rank ≥ 1. But in fact (∧m OX a submodule of ∧m+n OX have (∧m F ′ ⊗OX ∧n OX OX the former to the latter. Thus this is an isomorphism. This finishes the proof. F ′⊗OX ∧n OX F )/Ker(β) is F , which is OX−module locally free of rank 1. Thus we F )/Ker(β) = ∧m+n F and a surjective homomorphism from OX F ′′ . So (∧m OX F ′ ⊗OX ∧n OX F ′⊗OX ∧n OX F )/Ker(β) is (cid:3) Remark:In here we cannot use the formula of Theorem1.17. As in here F ′, F and F ′′ are locally free OX−modules, so there doesn't exist a canonical split exact sequence. Consider the special case where F is a locally free OX-module of rank 2,we have the following theorem. Theorem 5.2. There exists a canonical exact sequence: 0 −→ S 2 OX(∧2 OX F ) −→ S 2 OX(S 2 OX F ) −→ S 4 OX F −→ 0. There also admits a canonical homomorphism τ : S 2 F ), OX composite τ with the canonical inclusion, it it multiplication by 3. So if X is an Spec(Z[ 1 3]) scheme, which means if 3 is an invertible element in OX, the canon- ical exact sequence splits, hence we have canonical direct sum F ) −→ S 2 OX (∧2 OX (S 2 OX S 2 OX(S 2 OX F ) (cid:27) S 2 OX(∧2 OX F ) ⊕ S 4 OX F . Proof. As in this case F is locally free of rank 2, we are going to prove this theorem on local constructions. First it's easy to compute S 2 OX (∧2 OX F ) , S 2 OX modules of rank 1,6 and 5. Also we have the following commutative diagram: (S 2 OX F ) and S 4 OX F are free OX- F ⊗OX 4 id F ⊗OX 4 S 2 OX (S 2 OX F ) / S 4 OX F . Hence there exists a canonical homomorphism f : S 2 F ) −→ S 4 OX OX is defined by mapping local sections of the form m1⊗ m2⊗ m3⊗ m4 ∈ S 2 OX (S 2 OX F which (S 2 F ) to OX 13 / / / its image in S 4 OX short exact sequence F , whose kernel Ker( f ) is free of rank 1, and we may get the 0 −→ Ker( f ) −→ S 2 OX(S 2 OX F ) −→ S 4 OX F −→ 0. Using the upper remark and lemma, we have ∧6 OX (S 2 OX(S 2 OX F )) (cid:27) (∧1 OX Ker( f )) ⊗OX ∧5 OX (S 4 OX F ) (cid:27) Ker( f ) ⊗OX ∧5 OX (S 4 OX F ) and F )⊗ 2·2·6 Hence we may get Ker( f ) (cid:27) (∧2 OX in the theorem. (∧2 OX 2 (cid:27) Ker( f ) ⊗OX (∧2 OX F ) and the exact sequence as F )⊗2 (cid:27) S 2 OX (∧2 OX F )⊗ 5·4 2 Suppose x and y form a local basis for F U = eM on some affine open neigh- borhood of U ⊂ X, it is easy to check that Ker( f )U is generated by the local sections of the form (x⊗2 ⊗ y⊗2 − x ⊗ y ⊗ x ⊗ y), the inclusion of (∧2 RM)⊗2 −→ S 2 RM) is defined by sending (x ∧ y) ⊗ (x ∧ y) to (x⊗2 ⊗ y⊗2 − x ⊗ y ⊗ x ⊗ y). R(S 2 Next we need to check this inclusion is independent of the choice of basis, then we may glue together to get a canonical global homomorphism. More precisely, if x′ and y′ form another basis of M, then the gluing data is reflected by a 2 × 2 in- vertible matrix, say x′ y′ ! = a b c d ! x y !,where a b c d ! ∈ GL(2, R). We need to check there exists the following equality: (det( f ))2(x⊗2 ⊗ y⊗2 − x ⊗ y ⊗ x ⊗ y) = (x′⊗2 ⊗ y′⊗2 − x′ ⊗ y′ ⊗ x′ ⊗ y′) = ((ax + by)⊗2 ⊗ (cx + dy)⊗2− (ax + by) ⊗ (cx + dy) ⊗ (ax + by)⊗ (cx + dy)). This can be proved by direct computation. Hence we may glue together this ho- momorphism to get the exact sequence as described in the theorem. Our next aim is to construct "retraction" of the canonical inclusion, which means we are going to construct a global homomorphism from S 2 F ) to OX (∧2 F )⊗2. Note that they are all quotient sheaf of F ⊗4, so we may use this OX original tensor sheaf F ⊗4 to induce the retraction. As F ⊗4 admits permutation group S 4's canonical action, we can define a homomorphism φ of F ⊗4 as follows (S 2 OX φ : F ⊗4 −→ F ⊗4 f1 ⊗ f2 ⊗ f3 ⊗ f4 7→ σ(23)( f1 ⊗ f2 ⊗ f3 ⊗ f4) +σ(234)( f1 ⊗ f2 ⊗ f3 ⊗ f4) 14 Where fi are local sections of F and σ ∈ S 4 permutate the coordinates of F ⊗4. Next we need to check the following commutative diagram: F ⊗4 φ F ⊗4 S 2 OX (S 2 OX F ) / S 2 OX (∧2 OX F ) By a direct computation on local generators we know this is a commutative dia- gram. Hence we get well defined homomorphism from S 2 F )⊗2, OX which we denote it by τ. F ) to (∧2 OX (S 2 OX As these homomorphisms are globally well defined, so in order to check the composition property, we could fix a given basis, as x and y we give before. In this case, we could give concrete forms of τ. By a direct computation on local generators of these sheaves, we have τ ◦ i = 3 as τ(x⊗2 ⊗ y⊗2 − x ⊗ y ⊗ x ⊗ y) = x ∧ y ⊗ x ∧ y + x ∧ y ⊗ x ∧ y −x ∧ x ⊗ y ∧ y − x ∧ y ⊗ y ∧ x = 3x ∧ y ⊗ x ∧ y. So if 3 is an invertible element in OX, which means if X is an Spec(Z[ 1 then the inclusion i has retraction by making τ′ = 1 split exact sequence. This finishes the proof. 3]) scheme, 3 τ and the sequence is a natural (cid:3) Remark: Compare this theorem with theorem2.1, one can check these 2 ways, globally define by permutation group's action and locally define by local generators then glue together, are the same. Before state the following example, we introduce some notations. Note that F ⊗4 admits S 4's permutation action by permutate F ⊗4's coordinates, and this ac- tion is canonical. Set σ = (12) ∈ S 4, we will denote the canonical automorphism of F ⊗4 by permutating the 1st and 2nd coordinate by σ(12). The identity element's action on F ⊗4 will be denoted by σid. Define the following homomorphism of F ⊗4 as follows: F ⊗4 f1 / F ⊗4 f2 / F ⊗4 f3 / F ⊗4 where f1 = σid − σ(23) + σ(234), f2 = σ(23) − σ(234) and f3 is identity map. By a direct computation, we may get the following homomorphism of tensor sheaves: ∧4 OX F α1 / S 2 OX (∧2 OX F ) α2 / S 2 OX (S 2 OX F ) α3 / S 4 OX F . We denote the induced morphism by α1, α2 and α3. 15 / / / / / / / / / We can also define homomorphisms of F ⊗4 as follows: F ⊗4 g3 / F ⊗4 g2 / F ⊗4 g1 / F ⊗4 where g3 = σid + σ(23) + σ(234), g2 = σ(23) + σ(234) and f3 is identity map. By a direct computation, we may get the following homomorphism of tensor sheaves: S 4 OX F β3 / S 2 OX (S 2 OX F ) β2 / S 2 OX (∧2 OX F ) β1 / ∧4 OX F . We denote the induced homomorphism of tensor sheaves by β3, β2 and β1. Next we are going to prove the following theorem. Theorem 5.3. There exists a canonical OX-module complex: 0 / ∧4 OX F α1 / S 2 OX (∧2 OX F ) α2 / S 2 OX (S 2 OX F ) α3 / S 4 OX F / / 0. (6) There also exists a canonical OX-module complex: 0 / S 4 OX F β3 / S 2 OX (S 2 OX F ) β2 / S 2 OX (∧2 OX F ) β1 / ∧4 OX F / / 0 (7) We have β1 ◦ α1 and α3 ◦ β3 are equal to multiplication by 3. Furthermore if X is an Spec(Z[ 1 3]) scheme, which means if 3 is an invertible element in OX , the up- per 2 complexes are both split exact sequences. Hence there exists a canonical isomorphism ∧4 OX F ) (cid:27) S 2 OX F ) ⊕ S 4 OX (F ) 1. F ⊕ S 2 OX (S 2 OX (∧2 OX Proof. First we will check the complex properties of (6) and (7). Note that all αi and βi are induced from homomorphisms of F ⊗4, so in order to check the complex property of (6) and (7). We don't have to check on αi or βi, we just need to compute the composition of f or g, and prove the image in the corresponding sheaf is trivial. Take α2 ◦ α1 as example. We have f2 ◦ f1 = f2(σid − σ(23) + σ(234)) = (σ(23) − σ(234)) − (σid − σ(24)) + (σ(34) − σ(243)) 1If F is free modules,then it can be computed by hand that the upper modules are of the same rank although it looks complicated.Namely we need to prove n + 1 2 2 ! + 1 4 ! + n This can be proved by easy computation.  = n 2 ! + 1 2  + n + 3 4 ! . 16 / / / / / / / / / / / / / / Consider its image in the sheaf S 2 OX also compute β2 ◦ β2. As (S 2 OX F ), it is easy to check it is 0. We may g2 ◦ g3 = (σ(23) + σ(234)) ◦ (σid + σ(23) + σ(234)) = (σ(23) + σ(234)) + (σid + σ(24)) + (σ(34) + σ(243)) Consider its image in the sheaf S 2 OX f3 and g1 are induced from identity, the left can be proved by easy computation. F ), we also have β2 ◦ β3 = 0. As (∧2 OX Next we need to check the composition properties of these homomorphisms. Take β1 ◦ α1 as an example. We have g1 ◦ f1 = σid ◦ (σid − σ(23) + σ(234)) = σid − σ(23) + σ(234). Note that in ∧4 OX is multiply by 3. Similar result can be proved for α3 ◦ β3, as F , all these terms is equal to σid's action, hence the composition f3 ◦ g3 = σid ◦ (σid + σ(23) + σ(234)) = σid + σ(23) + σ(234) Note that the image is in the sheaf S 4 OX which means the composition is multiply by 3. F , hence they are all equal to σid's action, When X is an Spec(Z[ 1 3]) scheme, which means if 3 is an invertible element in OX, we have α1 and β3 are injective, α3 and β1 are surjective. This proves the exactness of (6) and (7) at the point of ∧4 OX F and S 4 OX F . As g2 ◦ f2 = (σ(23) + σ(234)) ◦ (σ(23) − σ(234)) = (σid − σ(34)) + (σ(24) − σ(243)) If we consider its image in the sheaf S 2 OX (∧2 OX F ), we have2 (σid − σ(34)) + (σ(24) − σ(243)) = 3σid − (σid − σ(24) + σ(243)) Which is just 3σid − α1, hence we get β2 ◦ α2 = 3id − α1. Hence for any local section f of the subsheaf Ker(α2) , we have local section f ′ ∈ ∧4 F such that 3 f − OX α1( f ′) = 0 , namely we have f = 1 3 α1( f ′) . This means Ker(α2) = im(α1) . So the (∧2 exactness at S 2 F ) in the complex (6) is exact. Similar result can be proved OX OX for exactness at S 2 (S 2 F ) for the complex (7). At last we need to check the OX OX exactness property at S 2 (S 2 OX OX F ) in the complex (6) . As g3 ◦ f3 = (σid + σ(23) + σ(234)) ◦ σid = σid + σ(23) + σ(234). 17 Consider its image in the sheaf S 2 OX σ(23) + σ(24), also note that (S 2 OX F ), we have the image is equal to σid + α2 = σ(23) − σ(234) α2 ◦ σ(24) = σ(243) − σ(34) α2 ◦ σ(243) = σ(24) − σid (S 2 OX So in the sheaf S 2 F ), there images are equal to σ(23) − σ(24), σ(23) − σid and OX σ(24) −σid. Hence we have β3 ◦α3 = 3σid +α2(σ(24) +σ(243)). As σ(24) and σ(243) can be regarded as automorphisms, hence we may write β3 ◦ α3 = 3σid + α2, so the ex- actness of (6) at S 2 F ) OX in the complex (7) . This ends the proof. F ) is proved. Similar result can be proved at S 2 OX (S 2 OX (∧2 OX (cid:3) 18 References [1] N.Bourbaki: Algebre. N.Bourbaki et Springer-Verlag Berlin Heidelberg (2006). [2] N.Bourbaki: Algebre Commtative. N.Bourbaki et Springer-Verlag Berlin Heidelberg (2006). [3] Willam Fulton. Joe Harris: Representation Theory(A First Course). GTM129. Springer-Verlag(1991) [4] R.Hartshorn: Algebraic Geometry. GTM52. Springer-Verlag(1977). [5] Ke-Zheng Li: Abstract Algebra(in Chinese). [6] Ke-Zheng Li: Commutative Algebra and Homological Algebra(in Chi- nese). Sci.Pub.(1998),2nd print(1999). [7] Ke-zheng Li: Moduli Space And Its Applications(in Chinese). MCM- CNU Lectures(2010-2011). [8] Hideyuki Matsumura: Commutative Ring Theory(English Version trans- lated by M. Reid). Cambridge University Press [9] William C.Waterhouse: Introduction to Affine Group Schemes. GTM66. Springer-Verlag(1979). [10] Jean-Pieere Serre: Linear Representations of Finite Groups. GTM42. Springer(1966). 19
1901.04848	1	1901	2019-01-13T12:51:07	MMP Via Wall-crossing for Moduli Spaces of Stable Sheaves on an Enriques surface	[ "math.AG" ]	We use wall-crossing in the Bridgeland stability manifold to systematically study the birational geometry of the moduli space $M_\sigma(\mathbf{v})$ of $\sigma$-semistable objects of class $\mathbf{v}$ for a generic stability condition $\sigma$ on an arbitrary Enriques surface $X$. In particular, we show that for any other generic stability condition $\tau$, the two moduli spaces $M_\tau(\mathbf{v})$ and $M_\sigma(\mathbf{v})$ are birational. As a consequence, we show that for primitive $\mathbf{v}$ of odd rank $M_\sigma(\mathbf{v})$ is birational to a Hilbert scheme of points. Similarly, in even rank we show that $M_\sigma(\mathbf{v})$ is birational to a moduli space of torsion sheaves supported on a hyperelliptic curve when $\ell(\mathbf{v})=1$. As an added bonus of our work, we prove that the Donaldson-Mukai map $\theta_{\mathbf{v},\sigma}:\mathbf{v}^\perp\to\mathrm{Pic}(M_\sigma(\mathbf{v}))$ is an isomorphism for these classes. Finally, we use our classification to fully describe the geometry of the only two examples of moduli of stable sheaves on $X$ that are uniruled (and thus not K-trivial).	math.AG	math	MMP VIA WALL-CROSSING FOR MODULI SPACES OF STABLES SHEAVES ON AN ENRIQUES SURFACE HOWARD NUER AND K ¯OTA YOSHIOKA 9 1 0 2 n a J 3 1 ] Abstract. We use wall-crossing in the Bridgeland stability manifold to systematically study the birational geometry of the moduli space Mσpvq of σ-semistable objects of class v for a generic stability condition σ on an arbitrary Enriques surface X. In particular, we show that for any other generic stability condition τ , the two moduli spaces Mτ pvq and Mσpvq are birational. As a consequence, we show that for primitive v of odd rank Mσpvq is birational to a Hilbert scheme of points. Similarly, in even rank we show that Mσpvq is birational to a moduli space of torsion sheaves supported on a hyperelliptic curve when ℓpvq " 1. As an added bonus of our work, we K Ñ PicpMσpvqq is an isomorphism for these classes. prove that the Donaldson-Mukai map θv,σ : v Finally, we use our classification to fully describe the geometry of the only two examples of moduli of stable sheaves on X that are uniruled (and thus not K-trivial). . G A h t a m [ 1 v 8 4 8 4 0 . 1 0 9 1 : v i X r a Contents Introduction 1. 2. Review: Bridgeland stability conditions 3. Review: Stability conditions on Enriques surfaces, K3 surfaces, and moduli spaces 4. Dimension estimates of substacks of Harder-Narasimhan filtrations 5. The hyperbolic lattice associated to a wall 6. Totally semistable non-isotropic walls 7. Divisorial contractions in the non-isotropic case 8. Isotropic walls 9. Flopping walls 10. LGU on the covering K3 surface 11. Main theorems 12. Picard groups of moduli spaces 13. Appendix: Some non-normal moduli spaces of sheaves References 1 5 6 14 17 21 35 39 56 60 67 69 72 82 1. Introduction For almost forty years, moduli spaces of stable sheaves have attracted great interest from math- ematicians and physicists alike. They have been studied using vastly different mathematical dis- ciplines, but more recently the development and application of Bridgeland stability conditions to their study has unified most of these tools. Introduced by Bridgeland [12] to formulate a rigorous definition of Douglas's π-stability for branes in string theory, stability conditions on the derived category DbpXq of a smooth projective variety X provide an adequately robust arena in which to study moduli spaces of sheaves using tools such as Fourier-Mukai transforms, enumerative and mo- tivic invariants, and the minimal model program. In particular, as stability conditions move around in a complex manifold StabpXq which admits a wall-and-chamber for any given Chern character, there is a strong connection between wall-crossing in StabpXq on the one hand, and wall-crossing formulae for enumerative invariants and birational transformations on moduli spaces on the other. 1 2 HOWARD NUER AND K ¯OTA YOSHIOKA In this paper, we bring this toolbox to bear on the study of moduli spaces of sheaves on an Enriques surface X. In previous work [39, 40, 48, 52, 51, 53], we had shown that for a generic stability condition σ in a certain distinguished connected component Stab:pXq Ă StabpXq there exist projective coarse moduli spaces Mσpvq parametrizing (S-equivalence classes) of σ-semistable objects of Mukai vector v (see Section 3 for definitions). We also classified precisely for which Mukai vectors v (or equivalently Chern character) the moduli space Mσpvq is nonempty, and we studied some coarse geometric and topological invariants of Mσpvq (refer again to Section 3 for a brief recap of these results). In particular, for the Mukai vectors (or Chern characters) of stable sheaves, we described some of these invariants for the moduli space of stable sheaves using a combination of modern and classical techniques in conjunction with Bridgeland stability. We continue our investigation of these moduli spaces in this paper with two more specific goals in mind. The first is to intimately study the effect on moduli spaces of crossing a wall W. More specifically, given a Mukai vector v, a wall W Ă Stab:pXq for v, and two stability conditions σ in the opposite and adjacent chambers separated by W, we seek a precise answer to the question: how are Mσ`pvq and Mσ´pvq related? The basic answer is given by Theorem 1.1 which says that Mσ`pvq and Mσ´pvq are birational. This question, and others associated with it, are motivated by a larger trend in moduli theory, wherein minimal models of a given moduli space are shown to be moduli spaces in and of themselves, just of slightly different objects. In the case of K3 surfaces, such a result, stating that all minimal models of Mσpvq are isomorphic to some Mτpvq for a different τ P Stab:pXq, was shown to be true by Bayer and Macr`ı in [6]. While we had hoped to prove such a result in the case of Enriques surfaces, our investigation instead led to a possible counterexample, see Example 8.23. Our second goal is to use wall-crossing, Fourier-Mukai transforms, and Theorem 1.1 to pin- point precisely where the moduli spaces Mσpvq live in the classification of algebraic varieties. We accomplish this goal with success for almost all Mukai vectors in Theorems 1.2 and 1.3. It has become more apparent with these results and similar results for other surfaces that Bridgeland stability conditions are crucial tool not only for studying the birational geometry of moduli spaces, but also more intrinsic and more classical questions about the geometry of moduli spaces. Summary of Results and Techniques. Let us turn now to stating our main results and the main tool we use to prove them. While we briefly introduce notation, the reader is invited to see Section 3 for more details. For an Enriques surface X, denote by : rX Ñ X the two-to-one K3 universal covering. The topological invariants of a coherent sheaf or object E in the derived category DbpXq are encoded in its Mukai vector We consider the Mukai lattice vpEq :" chpEqatdpXq P HpX, Qq. HalgpX, Zq :" vpKpXqq, where KpXq is the Grothendieck group, along with the induced pairing xvpEq, vpFqy " ´χpE, Fq. For any primitive v P HalgpX, Zq, if we write v " ℓpvqw P HalgprX, Zq with w primitive, then ℓpvq " 1 or 2 (see Lemma 3.2). The stability conditions that we consider are all contained in the distinguished connected component Stab:pXq of StabpXq containing those stability conditions σ such that skyscraper sheaves of points are σ-stable. For σ P Stab:pXq, we denote by Mσpvq the moduli space of (S-equivalence classes of) σ-semistable objects E P DbpXq with vpEq " v. These moduli spaces admit a decomposition Mσpvq " Mσpv, Lqğ Mσpv, L ` KXq, where L P PicpXq satisfies c1pLq " c1pvq in H 2pX, Qq and Mσpv, Lq parametrizes E P Mσpvq such that detpEq " L. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 3 Our first main result states that any two Bridgeland moduli spaces are birational. Theorem 1.1. Let v P HalgpX, Zq satisfy v2 ą 0, and let σ, τ P Stab:pXq be generic stability conditions with respect to v (that is, they are not contained on any wall for v). to each other. (1) The two moduli spaces Mσpvq and Mτpvq of Bridgeland-semistable objects are birational (2) More precisely, there is a birational map induced by a derived (anti-)autoequivalence Φ of DbpXq in the following sense: there exists a common open subset U Ă Mσpvq, U Ă Mτpvq such that for any u P U , the corresponding objects Eu P Mσpvq and Fu P Mτpvq satisfy have codimension at least two. Fu " ΦpEuq. If PicprXq " PicpXq, then the complements of U in Mσpvq and Mτpvq We remark that the generic Enriques surface X satisfies PicpXq " PicprXq. The proof of the analogous statement to Theorem 1.1 for K3 surfaces [6, Theorem 1.1] relies heavily on the fact that Mσpvq is a projective hyperkahler manifold in the K3 case. Our proof is based on studying stacks of Harder-Narasimhan filtrations and is thus more universal. In fact, the first author has already used this tool to obtain analogous results for bielliptic surfaces [38]. Let Y be a smooth projective variety, and suppose we are given a stability condition σ P Stab:pY q on a wall W for v and a sequence of Mukai vectors (or Chern characters) v1, . . . , vs of descending slope with respect to a nearby generic stability condition σ´. Then under some mild assumptions on Y , we prove in Theorem 4.2 that the substack of σ-semistable objects whose Harder-Narasimhan filtration with respect to σ´ has i-th σ´-semistable factor of class vi has dimension sÿi"1 dim Mσ´pviq `ÿiăjxvi, vjy. See Section 4 for more details. In particular, in the case of an Enriques surface, Theorem 4.2 allows us to determine what objects in Mσ`pv, Lq are destabilized when crossing the wall W, and most importantly the dimension of this locus, purely in terms of a hyperpolic lattice HW associated with the wall. This is the content of Theorem 5.8 which gives a much more refined and detailed classification of the type of birational map in Theorem 1.1 in terms of the lattices HW associated to the walls W that are crossed on along a path from σ to τ . One of the most powerful uses of a result like Theorem 1.1 appears in the next three results. They all follow a similar pattern. To study a given moduli space Mσpv, Lq, we apply a Fourier- Mukai transform Φ : DbpXq Ñ DbpXq inducing an isomorphism Mσpv, Lq „ÝÑ MΦpσqpΦpvq, L1q, where Φpvq has a more familiar form, say that of a well-understand type of coherent sheaf. Using Theorem 1.1, we know that there is a birational map MΦpσqpΦpvq, L1q 99K MτpΦpvq, L1q, where τ is now in the so-called Gieseker chamber so that MτpΦpvq, L1q is simply the moduli space of stable sheaves of Mukai vector Φpvq and determinant L1. An argument exactly along these lines gives the next two results: Theorem 1.2. Let v be a primitive Mukai vector such that v2 ą 0 is odd. Then for a general σ, there is an (anti-)autoequivalence Φ of DbpXq which induces an isomorphism Φ : U Ñ U1 where U Ă Mσpv, Lq and U1 Ă Hilb pXq are dense open subsets. In particular, Mσpv, Lq is birationally Mσpv,Lq -- OMσpv,Lq equivalent to Hilb for v2 ě 1. Moreover, if PicprXq " PicpXq, then the complements of U, U1 have codimension at least two. Theorem 1.3. Let v be a primitive Mukai vector such that v2 is even and ℓpvq " 1. Then there is an elliptic fibration π : X Ñ P1, a curve C Ă X, and a Mukai vector w " p0, C, χq with χ ‰ 0 such that (1) πC : C Ñ P1 is a double cover, and (2) there is an (anti-)autoequivalence Φ of pXq, π1pMσpv, Lqq -- Z{2Z, KMσpv,Lq fl OMσpv,Lq and Kb2 v2`1 v2`1 2 2 4 HOWARD NUER AND K ¯OTA YOSHIOKA DbpXq which induces an isomorphism Φ : U Ñ U1 where U Ă Mσpv, Lq and U1 Ă MHpw, L1q are open subsets. In particular, Mσpv, Lq is birationally equivalent to MHpw, L1q for a general σ. If PicprXq " PicpXq and v2 ě 2, then the complements of U and U1 are of codimension at least two, π1pMσpv, Lqq -- Z{2Z, KMσpv,Lq -- OMσpv,Lq, and hp,0pMσpv, Lqq " 0 for p ‰ 0, v2 ` 1. The remaining case, when v2 is even and ℓpvq " 2, can still be attacked in the same way, but the corresponding moduli space of torsion sheaves is not well understood due to the presence of non-reduced curves of larger multiplicity. Our third use of Theorem 1.1 goes in a slightly different direction. To prove that all minimal mod- els of Mσpv, Lq come from Bridgeland wall-crossing, an important step, arguably a necessary one, is to show that the Donaldson-Mukai map is surjective. Set KpXqv :" t x P KpXq xvpxq, vy " 0 u. For a universal family E on Mσpv, Lq X, the Donaldson-Mukai map is defined by (1) θv,σ : KpXqv Ñ PicpMσpv, Lqq x ÞÑ detppMσpv,Lq!pE b pXpx_qqq. Then using Theorem 1.1, we obtain the following result. Corollary 1.4. Let v be a primitive Mukai vector with v2 ě 3 and σ P Stab:pXq a generic stability condition with respect to v. If either (1) v2 is odd, or then θv,σ is an isomorphism. (2) v2 is even, ℓpvq " 1, and PicprXq " PicpXq, In an appendix, we use our results to describe the geometry of the two examples of moduli spaces of sheaves on an Enriques surface X that are uniruled. These constitute the only counterexamples to the fact that moduli spaces of stable sheaves on an Enriques surface have numerically trivial canonical divisors. One of these examples only exists on nodal Enriques surfaces, while the other is ubiquitous, but only applies for Mukai vectors of the form v " 2v0 where v2 0 " 1. In each case we show that there is always a wall W Ă Stab:pXq that induces the structure of a P1-fibration on Mσpvq. Correspondingly, these are the only counterexamples to the conjecture of the first author in [39] that the Bayer-Macr`ı divisor at the wall is big. Relation to other work. The relation between wall-crossing and the minimal model program studied here has been explored for many other surfaces. We refer the reader to [7, 6, 36, 37] for K3 surfaces, to [3, 10, 17, 18, 29] for P2, to [9, 4] for Hirzebruch and del Pezzo surfaces, and to [34, 47, 50] for abelian surfaces. During the writing phase of this project, the authors became aware of the recent preprint [8]. In that paper, the author proves part (1) of Theorem 1.1 as well as Theorems 1.2 and 1.3 for generic Enriques surfaces (that is, when PicprXq " PicpXq) and for Mukai vectors v such that v is primitive (see Theorem 4.4, Proposition 4.6, and Proposition 4.7, respectively, in [8] for the precise results). The approach is entirely different from ours and is again based on hyperkahler geometry, specifically the concept of a constant cycle subvariety of a hyperkahler manifold. Nevertheless, our approach has numerous advantages. The first is that we do not assume in Theorem 1.1 that v is primitive or that X is generic.1 The second advantage, which is a consequence of the first, is that we obtain (in Theorem 5.8) a complete classification of the birational behavior induced by any wall W without any assumptions on X or the divisibility of v. Even when v is primitive, our approach has the added advantage of giving necessary and sufficient conditions for a potential numerical wall 1In the case of [8, Proposition 4.6], the author actually does not need to assume that X is generic, but instead uses a clever deformation argument with relative moduli spaces of stable sheaves to prove his analogue of Theorem 1.2 only for moduli of Gieseker stable sheaves. This is another benefit of our method; although, as the author points out, one can obtain the same generality as in Theorem 1.2 using the construction of relative moduli spaces of Bridgeland stable objects in the forthcoming article [5] MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 5 W to actually be a wall, which is not guaranteed by just using the corresponding condition on the covering K3 surface rX. Acknowledgements. The authors would like to thank Arend Bayer, Izzet Coskun, Emanuele Macr`ı, Benjamin Schmidt, and Xiaolei Zhao for very helpful discussions related to this article. This project originally began while the authors were attending the workshop, "Derived categories and Chow groups of Hyperkaehler and Calabi-Yau varieties," at the Simons Center in fall of 2016, and we would like to thank the center for its warm hospitality and stimulating environment. The collaboration was aided by the second author's visits to the University of Illinois at Chicago, and we would like to thank the mathematics department for providing a productive working atmosphere. H.N. was partially supported by the NSF postdoctoral fellowship DMS-1606283, by the NSF RTG grant DMS-1246844, and by the NSF FRG grant DMS-1664215. K.Y. was partially supported by the Grant-in-aid for Scientific Research (No. 18H01113, 17H06127, 26287007), JSPS. Notation. For a complex number z P C we denote its real and imaginary parts by ℜz and ℑz, respectively. We will denote by DbpXq the bounded derived category of coherent sheaves on a smooth projec- tive variety X. On occasion we will consider the bounded derived category DbpX, αq of the abelian category CohpX, αq of α-twisted coherent sheaves, where α P BrpXq is a Brauer class. See [6, pp. 515-516] and the references contained therein for more background on twisted sheaves. We will use non-script letters (E, F, G, . . . ) for objects on a fixed scheme and reserve curly letters (E, F , G, . . . ) for families of such objects. For a vector v in a lattice H with pairing x , y, we abuse notation and write v2 :" xv, vy. The intersection pairing on a smooth surface X will be denoted by p , For a given lattice H and integer k P Z, we denote by Hpkq the same underlying lattice with pairing multiplied by the integer k. q and the self-intersection of a divisor D by pD2q. The fundamental class of a smooth projective variety X will be denoted by X . By an irreducible object of an abelian category, we mean an object that has no non-trivial subobjects. These are sometimes called simple objects in the literature. Recall that an object S in the derived category of a K3 or Enriques surface is called spherical if RHompS, Sq " C ' Cr´2s. We denote the associated spherical reflection by RS; it is defined by (2) RSpEq :" ConepRHompS, Eq b S Ñ Eq. Similarly, an object E0 in the derived category of an Enriques surface is called exceptional if RHompE0, E0q " C. In analogy to the spherical case, we denote the associated exceptional , or weakly spherical, reflection by RE0; it is defined by RE0pEq :" ConepRHompE0, Eq b E0 ' RHompE0pKXq, Eq b E0pKXq Ñ Eq. (3) This has also been called the Fourier-Mukai transform associated to p´1q-reflection in the literature. 2. Review: Bridgeland stability conditions In this section, we summarize the notion of Bridgeland stability conditions on a triangulated category D. The main reference is [12]. Definition 2.1. A slicing P of the category D is a collection of full extension-closed subcategories Ppφq for φ P R with the following properties : (1) Ppφ ` 1q " Ppφqr1s. (2) If φ1 ą φ2, then HompPpφ1q, Ppφ2qq " 0. 6 (4) HOWARD NUER AND K ¯OTA YOSHIOKA (3) For any E P D, there exists a collection of real numbers φ1 ą φ2 ą ą φn and a sequence of triangles 0 " E0 with Ai P Ppφiq. / E1 ⑧ ⑧ ⑧ ⑧ ⑧ A1 ⑤ ⑤ ⑤ }⑤ E2 / En´1 / En " E yssssss An A2 The subcategory Ppφq is abelian; its nonzero objects are called semistable of phase φ, and its simple objects are called stable. Appropriately, the collection of triangles in (4) is called the Harder- Narasimhan (HN) filtration of E, and we define φmaxpEq :" φ1 and φminpEq :" φn. For any φ P R, we denote by Ppφ ´ 1, φs the full subcategory of objects with φmin ą φ ´ 1 and φmax ď φ. This is the heart of a bounded t-structure. We usually consider A " Pp0, 1s. Let us fix a lattice of finite rank Λ and a surjective map v : KpDq ։ Λ. Definition 2.2. A Bridgeland stability condition on D is a pair σ " pZ, Pq, where satisfying the following compatibility conditions: ‚ the central charge Z : Λ Ñ C is a group homomorphism, and ‚ P is a slicing of Z, (1) For all non-zero E P Ppφq, 1 (2) For a fixed norm π arg ZpvpEqq " φ; on ΛR, there exists a constant C ą 0 such that for all semistable E. ZpvpEqq ě CvpEq We will write ZpEq for ZpvpEqq from here on. Furthermore, when we wish to refer to the central charge, the heart, or the slicing of a stability condition σ, we will denote it by Zσ, Aσ, and Pσ, respectively. It is worth noting that giving a stability condition pZ, Pq is equivalent to giving a pair pZ, Aq where Z : Λ Ñ C is a stability function with the HN-property in the sense of [13]. See [13, Proposition 3.5] specifically for this equivalence. The main theorem in [12] asserts that the set StabpDq of stability conditions is a complex manifold of dimension rkpΛq. The manifold StabpDq carries two group actions: the group AutpDq of autoequivalences acts on the left by ΦpZ, Pq " pZ Φ´1 , ΦpPqq, where Φ P AutpDq and Φ is the 2 pRq of matrices in GL2pRq with positive determinant acts on the right. This second action lifts the action of GL2pRq on HompKpDq, Cq " HompKpDq, R2q. 3. Review: Stability conditions on Enriques surfaces, K3 surfaces, and moduli action KpDq, and the universal cover ĂGL` spaces We give here a review of Bridgeland stability conditions on Enriques surfaces and K3 surfaces and their moduli spaces of stable complexes. The main references are [39, 51]. Throughout this section, Y will denote a K3 or Enriques surface. 3.1. The algebraic Mukai lattice. Let X be an Enriques surface over an algebraically closed field k of charpkq ‰ 2, and let rX be its covering K3 surface with covering map : rX Ñ X and covering involution ι. We denote by HalgprX, Zq the algebraic part of the whole cohomology of rX, namely (5) HalgprX, Zq " H 0prX, Zq ' NSprXq ' H 4prX, Zq. Similarly, for the Enriques surface X, we define (6) where NumpXq " NSpXq{xKXy. HalgpX, Zq :" pr, D, s 2q r, s P Z, r " s mod 2, D P NumpXq( Ă HpX, Qq, / / / / / } / / y c c a a c c MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 7 Definition 3.1. Let Y " X or rX. (1) We denote by v : KpY q ։ HalgpY, Zq, the Mukai vector vpEq :" chpEqatdpY q. When Y " rX, the Mukai vector takes the form vpEq " prpEq, c1pEq, rpEq ` ch2pEqq, in the decomposition (5), and when Y " X it takes the form rpEq 2 ` ch2pEqq, vpEq " prpEq, c1pEq, in the decomposition (6). (2) The Mukai pairing x , y is defined on HalgpY, Zq by xpr, c, sq,pr1, c1, s1qy :" pc, c1q ´ rs1 ´ r1s P Z, where p , DbpY q. q is the intersection pairing on H 2pY, Zq. The Mukai pairing has signature p2, ρpY qq and satisfies xvpEq, vpFqy " ´χpE, Fq " ´řip´1qi extipE, Fq for all E, F P (3) The algebraic Mukai lattice is defined to be the pair pHalgpY, Zq,x , yq. Given a Mukai vector v P HalgpY, Zq, we denote its orthogonal complement by such that xv, wy " 2xv, wy and identifies HalgpX, Zq with an index 2 sublattice of the ι- invariant component of HalgprX, Zq. The following lemma makes this precise: Lemma 3.2 ([39, Lem. 2.1]). A Mukai vector v " pr, c1, s r ` s 2 q " 1. 2q P HalgpX, Zq is primitive if and only if gcdpr, c1, For primitive v, we define ℓpvq by v " ℓpvqw, where w is primitive in HalgprX, Zq. Then ℓpvq " gcdpr, c1, sq and can be either 1 or 2. Moreover, ‚ if ℓpvq " 1, then either r or c1 is not divisible by 2; ‚ if ℓpvq " 2, then c2 must be odd and r ` s " 2 pmod 4q. In particular, for odd rank Mukai vectors or Mukai vectors with c1 primitive, v is still primi- tive, while primitive Mukai vectors with gcdpr, c1q " 2 (and thus necessarily gcdpr, c1, sq " 2) must satisfy v2 " 0 pmod 8q, as can be easily seen. 3.2. Stability conditions on Enriques and K3 surfaces. We continue to let Y " X or rX. Definition 3.3. A (full, numerical) stability condition on Y is a Bridgeland stability condition on DbpY q with Λ " HalgpY, Zq and v defined as in Section 3.1. In particular, for a stability condition σ " pZ, Pq on Y , the category Ppφq has finite length for φ P R, so any σ-semistable object E P Ppφq admits a filtration with σ-stable objects Ei P Ppφq. While the filtration itself, called a Jordan-Holder (JH) filtration, is not unique, the σ-stable factors Ei are unique, up to reordering. We call a Mukai vector v primitive if it is not divisible in HalgpY, Zq. The covering map induces an embedding vK :" w P HalgpY, Zq xv, wy " 0( . : HalgpX, Zq ãÑ HalgprX, Zq 8 HOWARD NUER AND K ¯OTA YOSHIOKA A connected component Stab:pY q of the space of full numerical stability conditions on DbpY q is described in [13, 51]. Let β, ω P NSpY qR be two real divisor classes, with ω ample. For E P DbpY q, define and consider the heart Aω,β defined by Zω,βpEq :" xeβ`iω, vpEqy, Aω,β :"$&% E P DbpY q ‚ HppEq " 0 for p R t´1, 0u, ‚ H´1pEq P Fω,β, ‚ H0pEq P Tω,β , ,.- where Fω,β and Tω,β are defined by (1) Fω,β is the set of torsion-free sheaves F such that every subsheaf F 1 Ď F satisfies ℑZω,βpF 1q ď (2) Tω,β is the set of sheaves T such that, for every non-zero torsion-free quotient T ։ Q, we 0; have ℑZω,βpQq ą 0. We have the following result: ĂGL` Theorem 3.4 ([13, Prop. 10.3],[51]). Let σ be a stability condition such that all skyscraper sheaves kpyq of points y P Y are σ-stable. Then there are ω, β P NSpY qR with ω ample, such that, up to the 2 pRq-action, Zσ " Zω,β and Pσp0, 1s " Aω,β. We call such stability conditions geometric, and we let UpY q be the open subset of StabpY q consisting of geometric stability conditions and denote by Stab:pY q the connected component of StabpY q containing UpY q. Using the Mukai pairing, for any stability condition σ " pZσ, Pσq, we can find ✵σ P HalgpY, Cq such that Zσp q " x✵σ, vp qy. (7) When Y " X, we define the subsets For σ P Stab:pY q, ℜ✵σ and ℑ✵σ span a positive definite 2-plane in HalgpY, Rq. In fact, we can say more. Let PpY q be the set of ✵ P HalgpY, Cq with components spanning a positive definite 2-plane, which consists of two connected components, and denote by P `pY q the component containing vectors of the form eβ`iω. When Y " rX, define the subset ∆pY q´1 :" w P HalgpY, Zq w2 " ´1( ∆pY q´2 :"" w P HalgpY, Zq ∆pY q :" w P HalgpY, Zq w2 " ´2( . and take their union w2 " ´2 and c1pvq " D pmod 2q, where D is a nodal cycle * , (8) (9) ∆pY q :" ∆pY q´1 Y ∆pY q´2. In either case, we consider the subset P `0 pY q :" P `pY qz ďwP∆pY q wK. Then we have the following fundamental theorem: Theorem 3.5 ([13, Prop. 8.3,Thm. 13.2],[35, Cor. 3.8]). The map Z : Stab:pY q Ñ HalgpY, Cq, σ ÞÑ ✵σ MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 9 is a covering map of the open subset P `0 pY q, where Zσp q " x✵σ, vp qy. If we let Aut:0pY q :"! Φ P AutpDbpY qq Φ " idH algpY,Zq, ΦpStab:pY qq " Stab:pY q) , then the group of deck transformations for the covering map Z is precisely Aut:0pY q. Finally, let T be the subgroup of AutpDbpY qq generated by ROCpkq, where C is a p´2q-curve, and R2 T , where T is a spherical or exceptional object (of nonvanishing rank). Then Stab:pY q " ďΦPT ΦpUpY qq. Here RT is the spherical or exceptional twist. 3.2.1. Inducing stability conditions. The natural pull-back and push-forward functors relate stabil- ity conditions on rX to those on X and vice-versa. Indeed, a stability condition σ " pZσ, Pσq P Stab:pXq induces a stability condition pσq " pZpσq, Ppσqq P Stab:prXq by defining Zpσq :" Zσ , Ppσqpφq :"! E P DbprXq pEq P Pσpφq) . In the opposite direction, for a stability condition σ1 " pZσ1 , Pσ1q P Stab:prXq, we can induce a stability condition pσ1q " pZpσ1q, Ppσ1qq P Stab:pXq via the definition Zpσ1q :" Zσ1 , Ppσ1qpφq :"! E P DbpXq pEq P Pσ1pφq) . It was shown in [35] that : Stab:pXq Ñ Stab:prXq is a closed embedding onto the submanifold of Stab:prXq consisting of ι-invariant stability conditions, albeit using different notation. 3.3. Walls. Of paramount importance to our investigation here, the space of Bridgeland sta- bility conditions admits a well-behaved wall and chamber structure. For a fixed Mukai vector v P HalgpY, Zq, there exists a locally finite set of walls (real codimension one submanifolds with boundary) in Stab:pY q, depending only on v, with the following properties: (1) When σ varies in a chamber, that is, a connected component of the complement of the union of walls, the sets of σ-semistable and σ-stable objects of class v do not change. (2) When σ lies on a single wall W Ă Stab:pY q, there is a σ-semistable object that is unstable (3) These same properties remain true for the wall and chamber structure on W1 X X Wk in one of the adjacent chambers and semistable in the other adjacent chamber. whose walls are W1 X X Wk X W for an additional wall W for v. These walls were originally defined in [13, Prop. 2.3] for K3 surfaces, and more generally in [45]. By the construction of these walls, it follows that, for primitive v and σ in a chamber for v, a JH- filtration factor of a σ-semistable object of class mv (m P N) must have class m1v for 1 ď m1 ă m. In particular, for σ in a chamber for v, σ-stability coincides with σ-semistability. Definition 3.6. Let v P HalgpY, Zq. A stability condition σ P Stab:pY q is called generic with respect to v if it does not lie on any wall for v. It is worth recalling that, given a polarization H P AmppY q and the Mukai vector v of an H- Gieseker semistable sheaf, there exists a chamber C for v, the Gieseker chamber, where the set of σ-semistable objects of class v coincides with the set of H-Gieseker semistable sheaves [13, Prop. 14.2]. 10 HOWARD NUER AND K ¯OTA YOSHIOKA 3.4. Moduli stacks and moduli spaces. For σ P Stab:pY q, let Mσpvq be the moduli stack of σ-semistable objects E with vpEq " v and Ms σpvq the open substack of σ-stable objects. That is, for a scheme T , MσpvqpTq is the category of E P DbpY Tq such that E is relatively perfect over T ([31, Def. 2.1.1]) and Et are σ-semistable objects with vpEtq " v for all t P T . By (the proof of) [45, Thm. 4.12], Mσpvq is an Artin stack of finite type which is an open substack of Lieblich's "mother of all moduli spaces" M which parametrizes families E P DbpY Tq with E relatively perfect over T and such that ExtipEt, Etq " 0 for all i ă 0 and t P T (see [31]). We say two objects E1 and E2 in Mσpvqpkq are S-equivalent if they have the same JH-filtration factors. For σ P Stab:pY q generic with respect to v, Mσpvq (resp. Ms σpvq) admits a projec- tive coarse moduli scheme Mσpvq (resp. M s σpvq), which parametrizes S-equivalence classes of σ- semistable (resp. σ-stable) objects E with vpEq " v (see [7] for the K3 case and [40, sect. 9],[51] for the Enriques case). It was shown in [7] and [51] that there exists an autoequivalence Φ P AutpDbpY qq inducing an isomorphism between Mσpvq and the stack MHpv1q of H-Gieseker semistable sheaves of Mukai vector v1 " Φpvq for some polarization H generic with respect to v1. Thus where Qpv1q is the open subscheme of the Quot scheme parametrizing quotients Mσpvq -- rQpv1qss{ GLNs, such that λ : OY p´mHq'N ։ E (1) vpEq " v1; (2) λ induces an isomorphism H 0pY, O'N (3) H ipY, EpmHqq " 0 for i ą 0, Y q -- H 0pY, EpmHqq; where m " 0 is fixed and Qpv1qss (resp. Qpv1qs) is the open sublocus where E is semistable (resp. stable). It follows that dim Ms 3.5. Some properties of moduli spaces. We recall here what is known about the moduli spaces Mσpvq and their coarse moduli spaces Mσpvq. Before we get into details for K3 surfaces and Enriques surfaces individually, let us point out that using the definition of inducing stability con- ditions in Section 3.2.1 we can relate their respective moduli spaces. Indeed, for σ P Stab:pXq, v P HalgpX, Zq, and w P HalgprX, Zq, there are morphisms of stacks Mpσqpwq Ñ Mσppwqq, E ÞÑ pEq Mσpvq Ñ Mpσqpvq, E ÞÑ pEq. Only the second of these requires comment. For the stability condition ppσqq, we have Zppσqq " 2Zσ and Pppσqqpφq " Pσpφq, so in particular E P DbpXq is σ-semistable if and only if pEq is pσq-semistable. σpvq " dim Qpv1qs ´ dim GLN " pdim Qpv1qs ´ dim PGLNq ´ 1 " dim M s (10) since PGLN acts freely on Qpv1qs. define Ms σpv, Lq, Mσpv, Lq and M s from the notation. In particular, as NSprXq " NumprXq, we drop it from the notation in the K3 For L P NSpY q, we let Mσpv, Lq be the substack of Mσpvq consisting of E with c1pEq " L. We σpv, Lq similary. When the determinant is irrelevant, we drop L case. On the other hand, in the Enriques case, when Y " X, σpvq ´ 1, Mσpvq " Mσpv, Lqğ Mσpv, L ` KXq. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 11 moduli spaces Mσpvq in the K3 case and is proven in [7] and [6]. 3.5.1. Y " rX is a K3 surface. The following result gives precise conditions on nonemptiness of the Theorem 3.7 ([13, Thm. 2.15]). Let rX be a K3 surface over k, and let σ P Stab:prXq be a generic stability condition with respect to v " mv0 P HalgprX, Zq, where v0 is primitive and m ą 0. (1) The coarse moduli space Mσpvq is non-empty if and only if v2 (2) Either dim Mσpvq " v2 ` 2 and M s σpvq ‰ ∅, or m ą 1 and v2 (3) When v2 0 ą 0, Mσpvq is a normal irreducible projective variety with Q-factorial singulari- 0 ě ´2. 0 ď 0. ties. 3.5.2. Y " X is an Enriques surface. The following results follow from [39]. Since Mσpvq is iso- morphic to a moduli stack of Gieseker semi-stable sheaves [51], they also follow from corresponding results for Gieseker semistable sheaves [52, 53]. Since the Enriques case is more subtle, we break the statement into smaller pieces, beginning with the primitive case: Theorem 3.8 (cf. [40],[52, Thm. 3.1],[53, Theorem 4.10]). Let X be an Enriques surface over k, and let σ P Stab:pXq be a generic stability condition with respect to primitive v P HalgpX, Zq. Then for L P NSpXq such that rL mod KXs " c1pvq, Mσpv, Lq ‰ ∅ if and only if (1) ℓpvq " 1 and v2 ě ´1 or (2) ℓpvq " 2 and v2 ą 0 or (3) ℓpvq " 2, v2 " 0, and L " r (4) v2 " ´2, L " D ` r H 1pODq " 0. Furthermore, when non-empty, 2 KX pmod 2q or 2 KX pmod 2q, where D is a nodal cycle, that is, pD2q " ´2 and (1) Mσpv, Lq is connected, and (2) if X is unnodal or v2 ě 4, then Mσpv, Lq is irreducible. Case (4) in Theorem 3.8 only occurs when X contains a smooth rational curve C which necessarily satisfies C 2 " ´2, in which case X is called a nodal Enriques surface. An Enriques surfaces containing no smooth rational curve is called unnodal. A Mukai vector v as in (4) and an object E P M s σpvq are called spherical, and it can be shown that the existence of a spherical object on σpvq with v2 " ´1 is called X is equivalent to X being nodal [24]. Similarly, an object E P M s exceptional. For non-primitive Mukai vectors, we phrase the results in terms of the moduli stacks, as it is in this form that we will use them. We state the positive square case first: [53, Lem. 1.5, Cor. 1.6], [40, Thm. 8.2]). Let v P HalgpX, Zq be a Mukai Proposition 3.9 (cf. vector with v2 ą 0, and let σ P Stab:pXq be generic with respect to v. For L P NSpXq such that rL mod KXs " c1pvq, we set Mσpv, Lqpss :" t E P Mσpv, Lq E is properly σ-semistable u Then (1) dim Mσpv, Lqpss ď v2 ´ 1. Moreover dim Mσpv, Lqpss ď v2 ´ 2 unless v " 2v0 with (2) Mσpv, Lqs ‰ ∅, Mσpv, Lq is reduced, and dim Mσpv, Lq " v2. (3) Mσpv, Lq is normal, unless v2 0 " 1. (a) v " 2v0 with v2 (b) v2 " 2. 0 " 1 and L " r 2 KX pmod 2q, or The statements in Proposition 3.9 remain true for the coarse moduli spaces with dimensions adjusted in accordance with (10). In particular, dim Mσpv, Lq " v2 ` 1. dim Mσpmwq "$'&'% ´m2, ´ m2 2 , ´ m2`1 2 , if w2 " ´2, if w2 " ´1, m " 0 pmod 2q, if w2 " ´1, m " 1 pmod 2q. dim Mσpmwq " dim Mσpmwq ´ dim AutpS'mq " ´m2, Proof. By [40, Proposition 9.9], in case w2 " ´2, the coarse moduli space Mσpmwq consists of a single point, S'm, where S is the unique σ-stable spherical object of class w. As AutpS'mq " GLmpkq, we get If w2 " ´1, then by [40, Lemma 9.2] the coarse moduli space Mσpmwq consists of the m ` 1 points E'i ' EpKXq'm´i(m i"0, where E and EpKXq are the two σ-stable exceptional objects of class w. As E and EpKXq are both exceptional, AutpE'i ' EpKXq'm´iq " GLipkq GLm´ipkq of dimension i2 ` pm ´ iq2. But then dim Mσpmwq " max 0ďiďm dimE'i'EpKXq'm´i Mσpmwq ´ AutpE'i ' EpKXq'm´iq (11) which gives the dimension as claimed. " ´ min0ďiďm i2 ` pm ´ iq2, 12 HOWARD NUER AND K ¯OTA YOSHIOKA For Mukai vectors with v2 ď 0 it is particularly useful to use the moduli stacks for dimension estimates as we now see: Proposition 3.10 (cf. Mukai vector, and let σ P Stab:pXq be generic with respect to u. [53, Proposition 1.9]). Let u P HalgpX, Zq be an isotropic and primitive (1) If Ms (2) Ms σpmuq ‰ ∅, then m " 1, 2. σp2u, Lq ‰ ∅ if and only if ℓpuq " 1 and L " 0 pmod 2q. Moreover σp2uq "! pFq F P Ms σp2uq is smooth of dimension 1. pσqpvq, ιpFq fl F) . Ms In particular, Ms (3) dim Mσpmuq ď t mℓpuq2 u. Finally, we consider the negative square case: Lemma 3.11. Let w P HalgpX, Zq be a primitive spherical or exceptional class, and let σ P Stab:pXq be a generic stability condition with respect to w. Then (cid:3) 3.6. Line bundles on moduli spaces. We again let Y " X or rX, and we recall the definition of the Donaldson-Mukai morphism. Fix a Mukai vector v P HalgpY, Zq, a stability condition σ P Stab:pY q, and a universal family E P DbpMσpv, Lq Y q. Then we have the following definition. Definition 3.12. Set KpY qv :" t x P KpY q xvpxq, vy " 0 u. The Donaldson-Mukai morphism from KpY qv to PicpMσpv, Lqq is defined by: θv,σ : KpY qv Ñ (12) PicpMσpv, Lqq x ÞÑ detppMσpv,Lq!pE b pY px_qqq. More generally, for a scheme S and a family E P DbpS Y q over S of objects in Mσpv, Lq, there is a Donaldson-Mukai morphism θE : KpY qv Ñ PicpSq associated to E, defined as in (12), which satisfies θE " λE θv,σ, where λE : S Ñ Mσpv, Lq is the associated classifying map. See [22, Section 8.1] for more details. Setting (13) ξσ :" ℑ ✵σ x✵σ, vy P vK, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 13 we define the numerical divisor class (14) ℓσ :" θv,σpξσq P NumpMσpv, Lqq, where we abuse notation by also using θv,σ for the extension of (12) to the Mukai lattice. From Section 3.3 it follows that the moduli space Mσpv, Lq and E remain constant when varying σ in a chamber for v, so for each chamber C, we get a map ℓC : C Ñ NSpMCpv, Lqq, σ ÞÑ ℓσ, where the notation MCpv, Lq denotes the fixed moduli space. By the proof of the projectivity of Mσpv, Lq in [51] with the argument in [37] or [7], we get a generalization of [40, Theorem 10.3] and can say even more: Theorem 3.13 (cf. [7, Theorem 4.1] in the K3 case). For a generic σ, ℓσ " θv,σpξσq is an ample divisor on Mσpv, Lq. 3.7. Wall-crossing and Birational transformations. In this last subsection, we recall one more result that will be essential for our study of the connection between crossing walls in Stab:pY q and birational transformations of the moduli space Mσpv, Lq. Let v P HalgpY, Zq with v2 ą 0, and let W be a wall for v. We say σ0 P W is generic if it does not belong to any other wall, and we denote by σ` and σ´ two generic stability conditions nearby W in two opposite adjacent chambers. Then all σ-semistable objects are still σ0-semistable, and thus the universal familes E on Mσpvq Y induce nef divisors ℓσ0, on Mσpvq by ℓσ0, :" θv,σpξσ0q. The main result about ℓσ0, is the following: Theorem 3.14 ([7, Thm. 1.4(a)],[40, Thm. 11.3]). Let v P HalgpY, Zq satisfy v2 ą 0, and let σ be two stability conditions in opposite chambers nearby a generic σ0 P W. Then: (1) The divisors ℓσ0, are semiample on Mσpvq. In particular, they induce contractions π : Mσpvq Ñ M, where M are normal projective varieties. When Y " rX, the divisors ℓσ0, are big so that π are birational and M are irreducible. (2) For any curve C Ă Mσpvq, ℓσ0,.C " 0 if and only if the two objects Ec and Ec1 cor- responding to two general points c, c1 P C are S-equivalent. In particular, the curves con- tracted by π are precisely the curves of objects that are S-equivalent with respect to σ0. This theorem leads us to the following definition describing a wall W in terms of the geometry of the induced morphisms π. Definition 3.15. We call a wall W: σ0pvq " ∅; induces a flopping contraction; (1) a fake wall, if there are no curves contracted by π; (2) a totally semistable wall, if M s (3) a flopping wall, if we can identify M` " M´ and the induced map Mσ`pvq 99K Mσ´pvq (4) a divisorial wall, if the morphisms π are both divisorial contractions; (5) a P1-wall, if the morphisms π are both P1-fibrations. A non-fake wall W such that codimpMσpvqzM s σ0pvqq ě 2 is necessarily a flopping wall by [7, Thm. 1.4(b)] and [40, Thm. 11.3]. 14 HOWARD NUER AND K ¯OTA YOSHIOKA 4. Dimension estimates of substacks of Harder-Narasimhan filtrations Having finished our review of known tools for studying wall-crossing, we develop here the first tool we will use to classify the behavior induced by crossing a wall. In this section, we will denote by Y any smooth projective variety satisfying openness of stability and boundedness of Bridgeland semistable objects as in [45, Lemma 3.4]. For Mukai vectors v1, v2, . . . , vs with the same phase φ with respect to σ, let Fpv1, . . . , vsq be the stack of filtrations: for a scheme T , (15) Fpv1, . . . , vsqpTq :" t 0 Ă F1 Ă Ă Fs Fi{Fi´1 P MσpviqpTq, 1 ď i ď s, Fs P MσpvqpTq u , where v " v1 ` ` vs. Then we have the following result. Proposition 4.1. With the notation as above, let Y be a smooth projective variety satisfying boundedness and openness of stability. Then Fpv1, . . . , vsq is an Artin stack of finite type. Proof. We prove the proposition by induction on s. Assuming that Fpv1, . . . , vs´1q is an Artin stack of finite type, we shall prove that Fpv1, . . . , vs´1, vsq is also an Artin stack of finite type. We set v "řs i"1 vi. It is sufficient to show that schemes, and (1) the natural morphism Fpv1, . . . , vs´1, vsq Ñ Fpv1, . . . , vs´1q Mσpvq is representable by (2) the diagonal morphism ∆ : Fpv1, . . . , vsq Ñ Fpv1, . . . , vsqFpv1, . . . , vsq is representable. Indeed, if we take a smooth surjective morphism M Ñ Fpv1, . . . , vs´1q Mσpvq from a scheme M of finite type, then we get a smooth surjective morphism Fpv1, . . . , vsq Fpv1,...,vs´1qMσpvq M Ñ Fpv1, . . . , vsq, where Fpv1, . . . , vsq Fpv1,...,vs´1qMσpvq M is a finite type scheme by (1). The statement in (2) is simply the other condition in the definition of an Artin stack [42, Def. 8.1.4] We prove (1) first. Let T be a scheme and T Ñ Fpv1, . . . , vs´1q Mσpvq a morphism. Then we have a family of filtrations 0 Ă F1 Ă Ă Fs´1 (16) on T Y such that Fi{Fi´1 are relatively perfect over T , pFi{Fi´1qt P Mσpviq for all t P T and a family of objects Fs such that Fs is relatively perfect over T and pFsqt P Mσpvq. By [23, Prop. 1.1], there is a scheme p : Q Ñ T which represents the functor Q : pSch{Tq Ñ pSetsq defined by (17) Let ξ : pp 1Y qpFs´1q Ñ pp 1Y qpFsq be the universal family of homomorphisms. Let Q0 be the subscheme of Q such that Conepξqq " Conepξqq P Aσ for all q P Q0, which is open by the Open Heart Property [45, Theorem 3.8],[1, Theorem 3.3.2]. It follows that ξq is injective in Aσ for all q P Q0. Then on Q0 we have a family of filtrations (18) Ñ Tq " tf f : pϕ 1Y qpFs´1q Ñ pϕ 1Y qpFsqu. QpU ϕ 0 Ă F1 Ă Ă Fs. Therefore Q0 -- Fpv1, . . . , vsq Fpv1,...,vs´1qMσpvq T. (19) In particular, the morphism Fpv1, . . . , vsq Ñ Fpv1, . . . , vs´1qMσpvq is representable by schemes, as claimed. Now let us prove (2). By [42, Lem. 8.1.8], it is equivalent to showing that for every scheme T and two families of filtrations F , F1 P Fpv1, . . . , vsqpTq, the sheaf IsomFpv1,...,vsq{TpF , F1q is an algebraic space. So let T be a scheme and consider two families (20) F :0 Ă F1 Ă Ă Fs F1 :0 Ă F11 Ă Ă F1s MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 15 of relatively perfect filtrations. An isomorphism φ : F Ñ F1 is an isomorphism Fs Ñ F1s as families in MσpvqpTq which preserves the filtration. But φ preserves the filtration if and only if the induced maps Fi Ñ F1s{F1i are the 0-map for all 0 ă i ă s. This is a closed condition by [23, Prop. 1.1], and if φpFiq Ď F1i, then we must in fact have equality as they are both families of σ-semistable objects with the same Mukai vector. Hence the sheaf IsomFpv1,...,vsq{TpF , F1q is parametrized by a closed algebraic subspace of the algebraic space IsomMσpvq{TpFs, F1sq, which shows that IsomFpv1,...,vsq{TpF , F1q is an algebraic space, as required. (cid:3) We have a natural morphism Fpv1, . . . , vsq p0 Ă F1 Ă Ă Fsq and hence a morphism ÝÑ ÞÝÑ pp0 Ă F1 Ă Ă Fs´1q, Fs{Fs´1q Fpv1, . . . , vs´1q Mσpvsq , Π : Fpv1, . . . , vsq Ñ Mσpviq. sźi"1 Let Fpv1, . . . , vsq :" Π´1 sźi"1pMσ´pviq X Mσpviqq¸ Ă Fpv1, . . . , vsq be the open substack of Fpv1, . . . , vsq where each Fi{Fi´1 is σ´-semistable as well, where σ´ is sufficiently close to σ. The intersections are taken within the large moduli space M, and as Mσpvq is an open substack of M for any σ P StabpY q by openness of stability, it follows that Mσ´pviq X Mσpviq is open in Mσpviq. Thus Fpv1, . . . , vsq is indeed well-defined and an open substack of Fpv1, . . . , vsq, as claimed. While we cannot say much more about the stack of filtrations in general, if we assume that v1, . . . , vs are the Mukai vectors of the semistable factors of the Harder-Narasimhan filtration with respect to σ´ of an object E P Mσpvq, with v "řs Fpv1, . . . , vsq (21) p0 Ă F1 Ă Ă Fsq Ñ Mσpvq ÞÑ Fs i"1 vi, then the natural map is injective with image the substack of Mσpvq parameterizing objects with Harder-Narasimhan filtration factors having Mukai vectors v1, . . . , vs. In this case, we can say even more and prove the following theorem, whose proof is similar to that of [14, Prop. 6.2]. Theorem 4.2. As above, suppose that Y satisfies boundedness and openness of stability, and let v1, . . . , vs be the Mukai vectors of the semistable factors of the Harder-Narasimhan filtration with respect to σ´ of some object E P Mσpvq, where v " řs i"1 vi. Suppose further that for any σ P StabpY q and E, E1 P Aσ such that φminpE1q ą φmaxpEq we have HompE, E1rksq " 0 for 2 ď k ď dim Y . Then dim Fpv1, . . . , vsq " dim Fpv1, . . . , vs´1q ` dim Mσ´pvsq ` xv ´ vs, vsy (22) " sÿi"1 dim Mσ´pviq `ÿiăjxvi, vjy. Proof. For an atlas ϕ : T Ñ Fpv1, . . . , vs´1q Mσ´pvsq, we set R :" T Fpv1,...,vs´1qMσ´pvsq T . Let 0 Ă F1 Ă Ă Fs´1 and Es be objects on T Y corresponding to the morphism ϕ. We note that for all t P T , HomppEsqt,pFs´1qtrksq " 0 for k ‰ 0, 1. Indeed, for k R r0, dim Y s, this is clear as pEsqt,pFs´1qt P Aσ´, while for k P r2, dim Y s, this follows from the hypothesis of the theorem. We for i ‰ j. It follows that shall stratify T "Ťi Ti by nptq :" homppEsqt,pFs´1qtq so that nptq is constant on Ti and nTi ‰ nTj χppEsqt,pFs´1qtq " homppEsqt,pFs´1qtq ´ homppEsqt,pFs´1qtr1sq 16 HOWARD NUER AND K ¯OTA YOSHIOKA We set Rij :" R TT Ti Tj. Then Rij " ∅ if i ‰ j and we have a stratification R "Ťi Rii. is constant for all t P T , and thus on each Ti, both nptq and homppEsqt,pFs´1qtr1sq are constant. Let pk i Ñ Ti pk " 0, 1q be the vector bundles associated to HompippEsqTi,pFs´1qTirksq, where pi : Ti Y Ñ Ti is the projection. i : V k As in [14, Lemma 6.1], there is a universal extension iqFs´1 Ñ F Ñ pp1 i , and the family F determines a morphism qi : V 1 0 Ñ pp1 over V 1 i qEs Ñ 0 i Ñ Fpv1, . . . , vsq which factors through i Ñ Ti Fpv1,...,vs´1qMσ´pvsq Fpv1, . . . , vsq. V 1 As in [14, p. 131], one can show that there is an isomorphism V 1 i Fpv1,...,vsq V 1 i -- V 0 i Rii V 1 i with a commutative diagram (23) i Rii V 1 V 0 i ÝÝÝÝÑ V 1 i V 1 i §§đ ÝÝÝÝÑ Ti Ti. §§đ Rii i V 1 i Rii V 1 Mσ´pvsq. Since It follows that in this description V 1 given by V 0 i V 1 i (i provide an atlas for Fpv1, . . . , vsq with relations i (i just as t Ti Ti ui and t Rii ui, respectively, do for Fpv1, . . . , vs´1q i Rii V 1 dim V 1 i ´ dim V 0 " dim Ti Ti ` 2 rk V 1 i ´ pdim Rii ` rk V 1 "pdim Ti Ti ´ dim Riiq ` xv ´ vs, vsy, i ` rk V 0 i q (24) i we get (25) dim Fpv1, . . . , vsq " max " max " dim Fpv1, . . . , vs´1q ` dim Mσ´pvsq ` xv ´ vs, vsy. i tdim V 1 i tpdim Ti Ti ´ dim Riiqu ` xv ´ vs, vsy i ´ dim V 0 i Rii V 1 i u i V 1 This gives the first equation in (22), while the second follows by induction. (cid:3) We can apply the above theorem to study the locus in Mσ`pvq of strictly σ0-semistable objects. Recall our setup: X is an Enriques surface, and for a given v P HalgpX, Zq with v2 ą 0, we take a generic stability condition σ0 on a wall W for v and two generic nearby stability conditions σ in opposite adjacent chambers. By letting σ " σ0 in the above theorem, we get the following result: Proposition 4.3. Let X be an Enriques surface, and suppose that v1, . . . , vs are the Mukai vectors of the semistable factors of the Harder-Narasimhan filtration with respect to σ´ of an object E P i"1 vi satisfies v2 ą 0. Then letting Fpv1, . . . , vsqo :" Fpv1, . . . , vsq X Mσ`pvq, where the intersection is taken in Mσ0pvq, we have Mσ`pvq, where v " řs (26) codim Fpv1, . . . , vsqo ě sÿi"1`v2 i ´ dim Mσ´pviq `ÿiăj xvi, vjy, where the codimension is taken with respect to Mσ`pvq. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 17 Proof. In order to apply Theorem 4.2, we first observe that the hypothesis of the theorem is met where σ " σ0 in this case. Indeed, for E, E1 P Aσ0 such that φminpE1q ą φmaxpEq, Serre duality gives HompE, E1r2sq " HompE1, EpKXqq " 0, where the last equality follows since X is numerically K-trivial so that φmaxpEpKXqq " φmaxpEq. Noting that Fpv1, . . . , vsqo is an open substack of Fpv1, . . . , vsq by openness of stability, we get that dim Fpv1, . . . , vsqo ď dim Fpv1, . . . , vsq, with equality if and only if the component of Fpv1, . . . , vsq of largest dimension contains a σ`-semistable object. As v2 ą 0, dim Mσ`pvq " v2 by Proposition 3.9, so codim Fpv1, . . . , vsqo " v2 ´ dim Fpv1, . . . , vsqo ě v2 ´ dim Fpv1, . . . , vsq " v2 ´ sÿi"1 sÿi"1`v2 dim Mσ´pviq `ÿiăj xvi, vjy¸ i ´ dim Mσ´pviq `ÿiăjxvi, vjy, " (27) as claimed. (cid:3) While Proposition 4.3 is phrased for Enriques surfaces, it applies to any K-trivial surface, with (26) modified appropriately. In particular, the first author uses it in [38] to study wall-crossing for bielliptic surfaces. 5. The hyperbolic lattice associated to a wall In order to effectively use the estimates provided by Proposition 4.3, we need to gain some understanding of the Mukai vectors vi which appear as Harder-Narasimhan factors of an object E P Mσ`pvq when we cross a wall W. From their definition, these walls are associated to the existence of another Mukai vector with the same phase as v, so to any wall W it is natural to consider the set of these "extra" classes, as in the following definition. As it turns out, this set will contain all of the Mukai vectors we are interested in. Proposition and Definition 5.1. To a wall W, let HW Ă HalgpX, Zq be the set of Mukai vectors Then HW has the following properties: HW :"" w P HalgpX, Zq ℑ Zpwq Zpvq " 0 for all σ P W* . tion of the Mukai form). (1) It is a primitive sublattice of rank two and of signature p1,´1q (with respect to the restric- (2) Let σ`, σ´ be two sufficiently close and generic stability conditions on opposite sides of the wall W, and consider any σ`-stable object E P Mσ`pvq. Then any HN-filtration factor Ai of E with respect to σ´ satisfies vpAiq P HW . (3) If σ0 is a generic stability condition on the wall W, the conclusion of the previous claim also holds for any σ0-semistable object E of class v. (4) Similarly, let E be any object with vpEq P HW , and assume that it is σ0-semistable for a generic stability condition σ0 P W. Then every Jordan-Holder factor of E with respect to σ0 will have Mukai vector contained in HW. Proof. The proof of [7, Proposition 5.1] carries over word for word. (cid:3) We would like to characterize the type of the wall W, i.e. the type of birational transformation induced by crossing it, in terms of the lattice HW . We will find it helpful to also go in the opposite direction as in [7, Definition 5.2]: 18 HOWARD NUER AND K ¯OTA YOSHIOKA Definition 5.2. Let H Ă HalgpX, Zq be a primitive rank two hyperbolic sublattice containing v. A potential wall W associated to H is a connected component of the real codimension one submanifold consisting of those stability conditions σ such that ZσpHq is contained in a line. We will also have cause to consider two special convex cones in HR. The first is defined as follows (see [7, Definition 5.4]): Definition 5.3. Given any hyperbolic lattice H Ă HalgpX, Zq of rank two containing v, denote by PH Ă HR the cone generated by classes u P H with u2 ě 0 and xv, uy ą 0. We call PH the positive cone of H, and a class in PH X H a positive class. The next cone, called the effective cone and whose integral classes are effective classes, is classified by the following proposition (see [7, Proposition 5.5] for the analogue in the K3 case): Proposition 5.4. Let W be a potential wall associated to a hyperbolic rank two sublattice H Ă HalgpX, Zq. For any σ P W, let Cσ Ă HR be the cone generated by classes u P H satisfying the two conditions Zσpuq ℜ Zσpvq ą 0 and #u2 ě ´1, or u2 " ´2, c1puq " D pmod 2q, D a nodal cycle . Then this cone does not depend on the choice of σ P W, so we may and will denote it by CW . Moreover, it contains PH. If u P CW , then there exists a σ-semistable object of class u for every σ P W, and if u R CW , then for a generic σ P W there does not exist a σ-semistable object of class u. Proof. The proof is identical to that of [7, Proposition 5.5] except that for the statements about the existence of semistable objects we must use Theorem 3.8. This accounts for the more subtle condition on u2 compared to the corresponding condition u2 ě ´2 for K3 surfaces. (cid:3) We also recall [7, Remark 5.6]: Remark 5.5. From the positivity condition on ℜ Zσpuq , it is clear that CW contains no line through Zσpvq the origin, i.e. if u P CW then ´u R CW . Thus there are only finitely many classes in CW X pv ´ CWq X H. We use this fact to make the following assumption: when we refer to a generic σ0 P W, we mean that σ0 is not in any of the other walls associated to the finitely many classes in CW Xpv´CWqXH. Likewise, σ will refer to stability conditions in adjacent chambers to W in this more refined wall- and-chamber decomposition. Finally, we single-out two types of primitive hyperbolic lattices as the nature of our arguments differ greatly between them: Definition 5.6. We say that W is isotropic if HW contains an isotropic class and non-isotropic otherwise. We begin our investigation by determining precisely the kind of Mukai vectors that can be contained in HW. Proposition 5.7. Let H be the hyperbolic lattice associated to a wall W and σ0 " pZ, P0q P W generic. Then H and σ0 satisfy one of the following mutually exclusive conditions: (1) H contains no effective spherical or exceptional classes. (2) (a) H contains precisely one effective spherical class, and there exists a unique σ0-stable (b) H contains precisely one effective exceptional class, and there exists exactly two σ0- spherical object S with vpSq P H. stable exceptional objects E, EpKXq with vpEq " vpEpKXqq P H. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 19 (3) There are infinitely many effective spherical or exceptional classes in H, and either (a) there exist exactly two σ0-stable spherical objects S, T whose classes are in H; or (b) there exist exactly four σ0-stable exceptional objects E1, E1pKXq, E2, E2pKXq with (c) there exists exactly one σ0-stable spherical object S and exactly two σ0-stable excep- vpE1q " vpE1pKXqq, vpE2q " vpE2pKXqq P H; or tional objects E, EpKXq with vpSq, vpEq " vpEpKXqq P H. In case (3), H is non-isotropic. Proof. Suppose that H contains precisely one effective spherical (resp. exceptional) class w. Then by Theorem 3.8, there exists a unique σ`-stable object S with vpSq " w (resp. precisely two σ`- stable objects E and EpKXq with vpEq " vpEpKXqq " w), which must then be spherical (resp. exceptional) by [51, Lemma 4.3]. Suppose that S (resp. E) is strictly σ0-semistable. Then by [51, Lemma 4.6] every σ0-stable factor F of S (resp. E ' EpKXq) must satisfy Ext1pF, Fq " 0. But then vpFq2 ă 0, so by [51, Lemma 4.3] vpFq2 " ´1 or vpFq2 " ´2 and c1pFq " D pmod 2q for some nodal cylce D, i.e. vpFq is either an effective spherical or exceptional class. But this is a contradiction to the assumption, so S is (resp. E, EpKXq are) σ0-stable, giving Case (2). It remains to consider Case (3). Let φ be the phase of v with respect to σ0. It will suffice for our purposes to show that, up to twisting by KX , there cannot be any combination of three stable spherical or exceptional objects S1, S2, S3 in P0pφq. Since each Si is σ0-stable of the same phase and distinct up to twisting by KX, we must have HompSi, Sjq " HompSj , SipKXqq " 0 for each i ‰ j. Thus if wi " vpSiq, then xwi, wjy " ext1pSi, Sjq ě 0. Now any two of the wi must be linearly independent, and we may choose, say, w1 and w2 to represent either both spherical or both exceptional σ0-stable objects. Denote by m :" xw1, w2y ě 0. Since H has signature p1,´1q, So m ě 2 or 3. We write w3 " xw1 ` yw2 with x, y P Q, and from xw3, w1y,xw3, w2y ě 0, we get that if w2 if w2 1 " w2 1 " w2 2 " ´1, 2 " ´2. if w2 if w2 1 " w2 1 " w2 2 " ´1, 2 " ´2. y x ď m ď m `am2 ´ 1 m ` ?m2 ´ 4 2 y x ď m 2 ď xw1, w2y2 ą w2 1w2 2 "#1, 4, 2 # 1 m ď y x ď m, m ď y x ď m 2 , m ´am2 ´ 1 ď 1 m ď m ´ ?m2 ´ 4 3 " x2#´1 ` 2mp y ´2 ` 2mp y 2 m ď ď 2 (28) But then since for m ě 2 and for m ě 3, we see that w2 xq ´ p y xq2, xq ´ 2p y xq2, must be positive, in contradiction to the fact that w2 3 " ´1 or ´2. Thus, we see that, up to tensoring by KX , there can only be at most two σ0-stable spherical or exceptional objects with Mukai vectors in H. Notice further that if H admits any combination of two linearly independent spherical or exceptional classes, then the group generated by the associated spherical and p´1q reflections is infinite, so the orbit of a spherical or exceptional class gives infinitely many Mukai vectors of the same kind. 2 " ´1, 2 " ´2, 1 " w2 1 " w2 if w2 if w2 Furthermore, we see that solving the quadratic equation 0 " u2 " pxw1 ` yw2q2 20 gives (29) HOWARD NUER AND K ¯OTA YOSHIOKA y x "#m ?m2 ´ 1, m?m2´4 , 2 if w2 if w2 1 " w2 1 " w2 2 " ´1, 2 " ´2, which are irrational. Thus in subcases (3)(a) and (3)(b), there can be no isotropic classes, as these would give rational solutions in (29). Finally, it only remains to show that in subcase (3)(c) H is non-isotropic. Similar to the previous subcases, we write an integral isotropic class u " xw1 ` yw2 with x, y P Q, where w1 " vpSq is a spherical class and w2 " vpEq " vpEpKXqq is an exceptional class. Solving the quadratic equation u2 " 0 gives y x " m am2 ´ 2, where m " xw1, w2y. But S and E are σ0-stable objects of the same phase with classes in a lattice of signature p1,´1q, so we must have m ě 2, as in the arguments for the preceeding cases. But this gives a contradiction as m2 ´ 2 cannot be a square for m ě 2. (cid:3) All of our main results are a consequence of the following classification theorem, which essentially The set W is a totally semistable wall if and only if one of the following conditions hold: says that the birational behavior induced by crossing W is entirely determined by HW and CW . Theorem 5.8. Let H Ă HalgpX, Zq be a primitive hyperbolic rank two sublattice containing v, and let W Ă Stab:pXq be a potential wall associated to H. (TSS1): there exists a spherical or exceptional class w P CW X H such that xv, wy ă 0; (TSS2): there exists an isotropic class u P H with ℓpuq " 2 and xv, uy " 1; or (TSS3): there exists a primitive isotropic class u P H such that xv, uy " ℓpuq and xv, wy " 0 for (TSS4): there exists a primitive isotropic class u P H such that xv, uy " 2 " ℓpuq and xv, wy " 0 a spherical w P CW X H; or for an exceptional w P CW X H. In addition, (1) The set W is a wall inducing a divisorial contraction if one of the following conditions hold: (Brill-Noether): there exists a spherical class w P CW X H such that xw, vy " 0, or there exists an exceptional class w P CW X H such that xw, vy " 0 and v ´ 2w P ∆pXq´2 X CW X H; or 2 " ℓpuq; or 1 " ℓpuq and v2 ě 3. (Hilbert-Chow): there exists an isotropic class u with xv, uy " 1 and ℓpuq " 2; or (Li-Gieseker-Uhlenbeck): there exists a primitive isotropic class u P H with xv, uy " (induced Li-Gieseker-Uhlenbeck): there exists an isotropic class u P H with xv, uy " (2) The set W is a wall inducing a P1-fibration on Mσ`pv, Lq if one of the following conditions hold: (Exceptional case): there exists a primitive isotropic class u with xv, uy " 2 " ℓpuq, an (Spherical case): there exists an isotropic class u with xv, uy " ℓpuq, a spherical class 2 KX pmod 2q, where D is a nodal exceptional class w P CW X H with xv, wy " 0, and L " KX pmod 2q, or w P CW X H with xv, wy " 0, and L " D ` rk v cycle. (3) Otherwise, if v is primitive and either (a) v2 ě 3 and v can be written as the sum v " a1` a2 with ai P PH such that L " r (b) there exists an exceptional class w P H and either pmod 2q if for each i, a2 i " 0 and ℓpaiq " 2; or 2 KX MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 21 2 , or (i) 0 ă xw, vy ď v2 (ii) xv, wy " 0 and v2 ě 3; or (i) 0 ă xw, vy ă v2 (ii) xw, vy " v2 2 , or (c) there exists a spherical class w P pCW X Hq and either then W induces a small contraction. 2 and v ´ w is a spherical class, (4) In all other cases, W is either a fake wall or not a wall at all. The proof of the above theorem will occupy us for the next four sections, but before we enter into a more involved and lattice specific analysis of the wall-crossing behavior, we present a general result on the codimension of the strictly σ0-semistable locus corresponding to the simplest Harder- Narasimhan filtration as above: Proposition 5.9. As above, let Fpa1, . . . , anqo be the substack of Mσ`pvq parametrizing objects with σ´ Harder-Narasimhan filtration factors of classes a1, . . . , an (in order of descending phase with respect to φσ´), and suppose that a2 i ą 0 for all i. Then codim Fpa1, . . . , anqo ě 2. i ą 0 implies that dim Mσ´paiq " a2 i . Thus by Proof. By Theorem 3.8, the assumption that a2 Proposition 4.3, codim Fpa1, . . . , anqo ěÿiăj xai, ajy. xai, ajy ąba2 i a2 j ě 1, codim Fpa1, . . . , anqo ě npn ´ 1q ě 2, But as a2 i ě 1 and H has signature p1,´1q, we must have for i ă j. Thus xai, ajy ě 2. It follows that as n ě 2. (cid:3) It follows from the proposition that in order for there to be more interesting wall-crossing be- havior, H must contain some class w with w2 ď 0. We will begin with the non-isotropic case first in the next section. 6. Totally semistable non-isotropic walls In this section we describe the criterion for a potential non-isotropic wall W to be totally semistable, that is, every E P Mσ`pvq is strictly σ0-semistable. We will see that by applying an appropriate sequence of spherical or weakly spherical twists, we can always reduce to the case of a non-totally semistable wall to study the birational behavior of crossing W. Let us begin with a sufficient condition for W to be totally semistable in general. Lemma 6.1. Let W be a potential wall such that xv, wy ă 0 for an effective spherical or exceptional class w P HW. Then W is totally semistable. Proof. Suppose there were a σ0-stable object E of class v. Let E0 be a σ0-semistable object E0 with vp E0q " w. As all stable factors of E0 are spherical or exceptional [51, Lemma 4.3, Lemma 4.6], we may find a σ0-stable object E0 such that xv, vpE0qy ă 0 and vpE0q2 " ´1 or ´2. As E and E0 (resp. E and E0pKXq) are non-isomorphic σ0-stable objects of the same phase, we must have HompE, E0q " HompE0pKXq, Eq " 0. But then 0 ą xv, vpE0qy " ext1pE, E0q ě 0, a contradiction. (cid:3) For a non-isotropic wall, the condition in Lemma 6.1 is actually necessary as we see now. 22 HOWARD NUER AND K ¯OTA YOSHIOKA Lemma 6.2. Suppose that H is non-isotropic and xv, wy ě 0 for all spherical or exceptional classes w P CW X H. Then W cannot be a totally semistable wall, and if codimpMσ`pvqzM s σ0pvqq " 1 then xv, wy " 0 for some spherical or exceptional class w P CW X H or xv, wy " 1 " xv, v ´ wy for spherical classes w, v ´ w P CW X H. Moreover, the generic member of Mσ`pvqzM s σ0pvq has HN-filtration factors of classes w and v ´ w with respect to σ´-stability. Proof. Consider the stack Fpa1, . . . , anqo of Harder-Narasimhan filtrations with respect to σ´- stability as in Section 4. We wish to estimate the codimension of Fpa1, . . . , anqo. Suppose that I " i a2 If b2 ą 0, ?a2b2 ě 1, so v2 " a2 ` 2xa, by ` b2 ě a2 ` 5. If, instead then we automatically have xa, by ą b2 ă 0, then note that b is the sum of effective spherical and/or exceptional classes. Indeed, for i R I, a2 i ă 0 and Mσ´paiq ‰ ∅, so ai is the sum of effective spherical and/or exceptional classes by [51, Lemma 4.3, Lemma 4.6], so the same holds for b " řiRI ai. By the assumption that v pairs non-negatively with all effective spherical and exceptional classes, we see that xv, by ě 0, so a2 " v2 ´ 2xv, by ` b2 ď v2 ´ 1. In any case we have a2 ă v2, and we write a2 " v2 ´ k with k P Zě0 and k ą 0 if v ‰ a. Expanding the squares on each side gives i ą 0( ‰ ∅ and let a :" řiPI ai. Write b :" v ´ a. and rearranging gives since dim Mσ´paiq " a2 we have (30) (31) (32) By Proposition 4.3, we get that a2 a2 i . k 2 ´ 1 2 ÿiPI c i ` 2 ÿiăj,pi,jqPpI 2qcxai, ajy ´ k, 0 " ÿiPI c ÿiăj,pi,jqPpI 2qcxai, ajy " i ´ dim Mσ´paiqq `ÿiăj codim Fpa1, . . . , anqo ěÿi pa2 " ÿiPI cpa2 i ´ dim Mσ´paiqq ` ÿiăj,pi,jqPI 2xai, ajy ` ÿiăj,pi,jqPpI 2qcxai, ajy, 2 ` ÿiPI c a2 i for i P I. Using (30) and writing ai " miwi with wi primitive for i P I c, codim Fpa1, . . . , anqo ě xai, ajy i 2 ´ dim Mσ´paiq ` ÿiăj,pi,jqPI 2xai, ajy 2 ` ÿiăj,pi,jqPI 2xai, ajy, ÿ i "mi"1 pmod 2q 1 k k 2 ` " iPI c:w2 where the final equality follows from Lemma 3.11. Moreover, for pi, jq P I 2 with i ‰ j, the signature of H forces xai, ajy ě 2. Thus codim Fpa1, . . . , anqo ě k 2 ` IpI ´ 1q ą 0. As this holds true for all possible HN-filtrations of objects in Mσ`pvq with respect to σ´-stability, W cannot be totally semistable. Moreover, note that if I " 0, so that a " 0, the estimate in (32) is still valid. In that case, we must have k " v2 ą 0 from which we see that codim Fpa1, . . . , anqo ą 0. For the second and third claim, we note that if codim Fpa1, . . . , anqo " 1, then from (32) we see that I " 0, 1 and k " 1, 2. If I " 1, then b " v ´ a must satisfy b2 ă 0 (otherwise k ě 5 as we saw above), and solving for xv, by in v2 ´ k " pv ´ bq2 gives 0 ď 2xv, by " b2 ` k ď b2 ` 2 ă 2, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 23 σ0pvq. v. Moreover, as b "řiRI ai, it follows from the assumption on v and xb, vy " 0 that n " 2 and so xv, by " 0 and b2 " ´k. Thus b is an effective spherical or exceptional class orthogonal to b " a1 or a2. Letting w " b, we get the claim about the HN-filtration of the generic member of Mσ`pvqzM s If instead I " 0, then k " v2 " 1, 2, and we rearrange the ai, if necessary, so that xv, wiy ď xv, wi`1y for all i. Let us assume first that v2 " 2, and note that it follows from (32) that mi must be even if w2 1 and H is non-isotropic, we must have w2 1 " ´1. It is easy to see that then H " Zv K Zw1. For wj (j ‰ 1), we set wj " xjv` yjw1 i " ´1. If xv, w1y " 0, then as pv ´ w1q2 " 2 ` w2 for xj, yj P Z. Then 0 ď xv, wjy " 2xj. Since v "řn n " 2 and m2 " x2 " 1. Hence v " m1w1 ` w2. Since 2 ´ m2 j"1 mjwj, it follows that 1 "řj‰ mjxj, so 2, m1 " 2 and w2 2 " ´2. 1 " w2 If xv, wiy ą 0 for all i, then from 2 " v2 " mixv, wiy, 1 " v2 " mixv, wiy, nÿi"1 nÿi"1 we see that n " 2 and xv, w1y " xv, w2y " m1 " m2 " 1. Thus w2 2 " w2 w2 Now assume that v2 " 1. Then from 2. As the mi are odd, we must have w2 i " ´2, which implies xw1, w2y " 3. 1 " pv ´ w2q2 " v2 ´ 2xv, w2y ` we see that n " 2 and v " m1w1 ` m2w2 with xv, w1y " 0 and xv, w2y " m2 " 1. Since pv ´ w1q2 " 1 ` w2 1 and H is not isotropic, we must have w2 1 " ´2. Then implies w2 2 " ´1 and m1 " 1. Hence xw1, w2y " 2. v2 ´ 2m2 1 " pv ´ m1w1q2 " w2 2 (cid:3) The previous lemma justifies singling out Mukai vectors which pair non-negatively with all effec- tive spherical or exceptional classes, and we will spend the remainder of this section using the theory of Pell's equation to reduce to this case. Such a Mukai vector is called minimal in its GH-orbit, or simply minimal for short, because of the following definition. Proposition and Definition 6.3. Let GH Ă AutpHq be the group generated by spherical and exceptional reflections associated to effective spherical and exceptional classes in CW . For a given positive class v P PH X H, the GH-orbit of v contains a unique class v0 such that xv, wy ě 0 for all effective spherical and exceptional classes w P CW . We call v0 the minimal class of the orbit GH v. The proof of the existence of v0 is almost identical to that of [6, Proposition and Definition 6.6], so we omit it. In the remainder of this section we will explore the action of GH more fully. We will consider explicitly only Cases (3)(b) and (3)(c), as Case (3)(a) is covered in detail in [6, Section 6]. Furthermore, our discussion is clearly irrelevant to Case (1), and applies in a much simpler but completely analogous way in Case (2). For simplicity, we observe that up to the action of ĂGL` all objects in CW is 1. We make this assumption throughout the rest of this section. 2 pRq, we may assume that the phase of 6.1. Stability. In order to study the behavior of stability under GH, we must study the interplay between rotating a stability function and tilting. Let B be an abelian category with a stability function Z : DbpBq Ñ C such that ZpBzt0uq " H Y Ră0. We set φpEq :" 1 Definition 6.4. E P B is Z-semistable if φpFq ď φpEq for all subobjects F of E in B. π arg ZpEq. 24 HOWARD NUER AND K ¯OTA YOSHIOKA We assume that Z satisfies the HN-filtration property and that the category t E P B E is Z-semistable of phase φ u is finite so that JH-filtrations exist, see [13, Sections 3 and 4]. Now we recall the definition of a torsion pair: Definition 6.5 ([20]). A torsion pair in an abelian category B is a pair of full subcategories pT , Fq of B which satisfy HompT, Fq " 0 for T P T and F P F, and such that every object E P B fits into a short exact sequence for some pair of objects T P T and F P F. We write T " T pEq and F " FpEq. 0 Ñ T Ñ E Ñ F Ñ 0 One often constructs torsion pairs via HN-filtrations as in the following definition of the rotation of a stability function Z and the tilt of corresponding torsion pair. Definition 6.6. For a real number θ P p´1, 1q, we set Zθ :" eπ?´1θZ. (1) If θ ě 0, then let pT 1, F1q be the torsion pair of B such that (2) If θ ď 0, then let pT 2, F2q be the torsion pair of B such that (a) T 1 is generated by Z-stable objects E P B with φpEq ` θ ą 1. (b) F1 is generated by Z-stable objects E P B with φpEq ` θ ď 1. We set B1 :" xT 1r´1s, F1y and Z1 :" Zθ. We denote the corresponding phase by φ1. (a) T 2 is generated by Z-stable objects E P B with φpEq ` θ ą 0. (b) F2 is generated by Z-stable objects E P B with φpEq ` θ ď 0. We set B2 :" xT 2, F2r1sy and Z2 :" Zθ. We denote the corresponding phase by φ2. We determine now precisely when an object in the tilted category is semistable with respect to the rotated stability function, generalizing [6, Lemma 6.10]. Proposition 6.7. (1) Assume that E P B1. Then E is Z1-semistable if and only if (2) Assume that E P B2. Then E is Z2-semistable if and only if (a) E P F1 and E is Z-semistable or (b) Er1s P T 1 and Er1s is Z-semi-stable. (a) Er´1s P F2 and Er´1s is Z-semistable or (b) E P T 2 and E is Z-semistable. Although this result might be known to experts, we provide a proof for the sake of convenience. Proof. We prove (1) first. Note that φ1pE1q ą φ1pE2q for all E1 P F1 and E2 P T 1r´1s. Then, if E P B1 is Z1-semistable, it follows from the canonical exact sequence in B1, (33) that either H0pEq " 0 or H1pEq " 0. Assume first that H1pEq " 0, that is, E P F1. Let E1 be a subobject of E in B. By considering the HN-filtration of E{E1, we see that there is a subobject E11 Ă E in B such that E1 Ă E11, E11{E1 P T 1 and E{E11 P F1. Since E{E11 is a quotient object of E in B1, Z1-semistability of E implies 0 Ñ H0pEq Ñ E Ñ H1pEqr´1s Ñ 0, φpEq ` θ " φ1pEq ď φ1pE{E11q " φpE{E11q ` θ. In particular, φpEq ď φpE{E11q. From the definitions of T 1 and F1 and our choice of E11, we see that φpE11{E1q ě φpE{E1q ě φpE{E11q, so it follows that φpEq ď φpE{E1q. Therefore E is Z-semistable. Now assume that H0pEq " 0, that is, Er1s P T 1. Let E1 be a subobject of Er1s in B. Then there is a subobject E11 Ă E1 in B such that E11 P T 1 and E1{E11 P F1. Since Er1s{E11 P T 1, E11 is a subobject of Er1s in B1r1s. By the Z1-semi-stability of E, φpE11q ` θ " φ1pE11q ď φ1pEr1sq " φpEr1sq ` θ, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 25 so in particular φpE11q ď φpEr1sq. Since φpE11q ě φpE1q ě φpE1{E11q, from the definition of T 1 and F1, we see that φpE1q ď φpEr1sq. Therefore E is Z-semistable. Next we shall prove the converse direction. Let E be a Z-semistable object of B. We first assume that E P F1. Taking the long exact sequence associated to a given short exact sequence in B1, (34) 0 Ñ E1 Ñ E Ñ E2 Ñ 0, we get Then φpim ϕq ď φpE2q by H1pE1q P T 1. By the Z-semistability of E, φpEq ď φpim ϕq so that 0 Ñ H0pE1q Ñ E Ñ H0pE2q Ñ H1pE1q Ñ 0. ϕ φ1pEq ´ θ " φpEq ď φpE2q " φ1pE2q ´ θ. Therefore E is Z1-semistable. We next assume that E P T 1. Again we take the long exact sequence associated to a short exact sequence in B1, (35) and we get the exact sequence 0 Ñ E1 Ñ Er´1s Ñ E2 Ñ 0, (36) Then φpim ϕq ě φpH 1pE1qq by H0pE2q P F1. From the Z-semistability of E, it follows that φpEq ě φpim ϕq, and thus we have 0 Ñ H0pE2q Ñ H1pE1q Ñ E Ñ H1pE2q Ñ 0. φ1pEr´1sq ´ θ " φ1pEq ´ θ " φpEq ě φpH 1pE1qq " φpE1r1sq " φ1pE1r1sq ´ θ " φ1pE1q ´ θ. ϕ Therefore E is Z1-semi-stable, as required. To prove (2), we note that for the abelian category Br1s with the stability function Z : Br1s r´1sÑ B ZÑ C, we have pBr1sq1 " B2, where Z1`θ " Zθ. Hence the claim follows from (1). (cid:3) We will use Proposition 6.7 to study how stability is effected under the action of GH. To do so, we must investigate Cases (3)(b) and (3)(c) separately. 6.2. (3)(b): Exactly two σ0-stable exceptional objects up to ´ b OXpKXq. Recall from Proposition 5.7 that in this case H contains infinitely many exceptional classes, precisely two of which represent σ0-stable objects (up two tensoring with OXpKXq). Denote one of these classes by w0. We may complete w0 to a basis so that H " Zw0 ` Zz, where xw0, zy " 0 and D :" z2 ą 0. Recalling the definition of ∆pXq´1 in (8), we see that ∆pXq´1X H is described by the Pell equation (37) Note that ?D must be irrational since H would be isotropic otherwise, in contradiction to Case (3) of Proposition 5.7. x2 ´ Dy2 " 1. Recall, for example from [28, Theorem 8.6], that the solutions to Pell's equation (37) form a group isomorphic to Z' Z{2Z. Indeed, let pp1, q1q be the fundamental solution of (37) with p1 ă 0 and q1 ą 0, and define pn, qn P Z by (38) pn ` qn?D "#´p´p1 ´ q1?Dqn, n ą 0 . p´p1 ´ q1?Dqn, n ď 0 Then setting wn :" pnw0 ` qnz, we see that all solutions to (37) are given by ∆pXq´1 X H " t wn n P Z u . 26 HOWARD NUER AND K ¯OTA YOSHIOKA ℜ Zσ0p q Zσ0pvq ą 0 Z´1p0q w2 w1 ´w0 ℜ Zσ0p q Zσ0pvq ă 0 w´2 w´1 w0 ´w1 Zσ0puq Zσ0pvq ą Figure 1. The shaded gray area is the half plane containing u such that ℜ 0, which is bounded by the line Z´1p0q. The hyperbola is defined by u2 " ´1, and the lines by u2 " 0. Recall that to an exceptional object E0 P DbpXq, we get a weakly spherical reflection RE0 as in (3). We abuse notation by using RvpE0q to denote the action on cohomology, which is given by RvpE0qpvq " v ` 2xv, vpE0qyvpE0q. Then it is easy to see that (39) wn`1 " ´ Rwnpwn´1q, pn ě 2q wn´1 " ´ Rwnpwn`1q, pn ď ´1q w2 "Rw1pw0q, w´1 " Rw0pw1q. Since Z´1p0qX HR is negative definite and limnÑ8 qn " ¯?D, it is not difficult to see that for w1 " ´w1, Z´1p0q X pRą0w0 ` Rą0w1q ‰ ∅ and ∆pXq´1 X pRą0w0 ` Rą0w1q " ∅, see Figure 1. Then ℜ Zσ0p´w1q pn Zσ0pvq ą 0, and since w1 " ´w1 we get It follows that wn P CW for all n P Z. Let T0 and T1 be σ0-semistable objects with vpTiq " wi for i " 0 and 1, respectively. By construction, T0 is σ0-stable, and we claim that T1 is as well. Indeed, by [51, Lemma 4.3, Lemma 4.6], any σ0-stable factor T of T0 must be exceptional so that vp Tq P ∆pXq´1 X CW. In particular, by the description in (39), vp Tq " aw0 ` bw1, where a and b are nonnegative integers. This gives a contradiction unless vp Tq " w1 so that T1 is σ0-stable, as claimed. We set (40) Note that for n ă 0, we have (41) Cn :"" xw0 ` yz Dqn`1 pn`1 y ă x ă Dqn pn y, y ą 0* . Cn " t u P CW xu, wn`1y ă 0 ă xu, wny u , It follows that CW " Rą0w0 ` Rą0w1. Furthermore, the positive cone can be described as Zσ0puq " u P H ℜ Zσ0pvq ą 0, u2 ě ´1* Ă Qě0w0 ` Qě0w1. PH " xw0 ` yz P H y2D ´ x2 ą 0, y ě 0( . MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 27 and for n ą 0, (42) while for n " 0, (43) Then tCn n P Zu is the chamber decomposition of PH under the action of GH. Cn " t u P CW xu, wn`1y ą 0 ą xu, wny u , Cn " t u P CW 0 ă xu, wny,xu, wn`1y u . For v0 P C0, we set (44) vn :"#Rwn Rwn´1 Rw1pv0q, wn`2 R´1 R´1 wn`1 R´1 n ą 0 w0pv0q, n ă 0. Then for v P Cn, there is v0 P C0 such that v " vn for this v0. 6.2.1. The abelian categories Ai. Up to reordering, we may assume that φ`pT1q ą φ`pT0q (and hence φ´pT1q ă φ´pT0q), where φ denotes the phase with respect to σ, respectively. For i P Z, let T i P P0p1q be σ-stable objects with vpT i q " wi. Then (45) for any σ`-stable object E with vpEq2 ě 0. We note that T `i " T ´i " Ti pi " 0, 1q are σ0-stable objects. φ`pT `1 q ą φ`pT `2 q ą ą φ`pEq ą ą φ`pT´1q ą φ`pT `0 q We make the following definition which generalizes the approach of [6, Lemma 6.8]. Definition 6.8. Assume that i ě 0. (1) Let pTi, Fiq be the torsion pair of P0p1q such that (2) Let pT i , Fi q be the torsion pair of P0p1q such that erated by σ`-stable objects F with φ`pFq ą φ`pT `i`1q and φ`pT `i`1q. (a) Ti " xT `1 , T `1 pKXq, T `2 , T `2 pKXq, ..., T `i , T `i pKXqy is the subcategory of P0p1q gen- (b) Fi is the subcategory of P0p1q generated by σ`-stable objects F with φ`pFq ď Let Ai :" xTir´1s, Fiy be the tilting. (a) T i is the subcategory of P0p1q generated by σ´-stable objects F with φ´pFq ě (b) Fi " xT ´1 , T ´1 pKXq, T ´2 , T ´2 pKXq, ..., T ´i , T ´i pKXqy is the subcategory of P0p1q gen- Let Ai :" xT i , Fi r1sy be the tilting. φ´pT ´i`1q. erated by σ´-stable objects F with φ´pFq ă φ´pT ´i`1q. Since T0 " 0 and F0 " 0, we have A0 " A0 " P0p1q. Remark 6.9. We note that from the definition of Ai (resp. Ai ), it follows that: (1) T `i`1, T `i`1pKXq, T `i r´1s, T `i pKXqr´1s are irreducible objects of Ai. (2) T ´i`1, T ´i`1pKXq, T ´i r1s, T ´i pKXqr1s are irreducible objects of Ai . With these notions in place, we prove the following stronger form of the induction claim in [6, p. 541]. Proposition 6.10. For i ě 0, RT ` Proof. We set Φ :" RT ` i`1 We first prove that RT ` the claim follows for a general E P Ai from the short exact sequence 0 Ñ H0pEq Ñ E Ñ H1pEqr´1s Ñ 0, (46) as H0pEq P Fi and H1pEq P Ti. induces an equivalence Ai Ñ Ai`1. i`1 and ΦppEq :" HppΦpEqq for E P DbpXq. i`1pAiq Ă Ai`1 by showing that ΦpFiq Ă Ai`1 and ΦpTiq P Ai`1r1s. Then 28 HOWARD NUER AND K ¯OTA YOSHIOKA Observe first that for E P P0p1q, ExtppT `i`1, Eq " ExtppT `i`1pKXq, Eq " 0 for p ‰ 0, 1, 2. Hence, in (3), we have an exact sequence from the definition of RT ` (47) i`1 0 ÝÝÝÝÑ Φ´1pEq ÝÝÝÝÑ HompT `i`1, Eq b T `i`1 ' HompT `i`1pKXq, Eq b T `i`1pKXq ÝÝÝÝÑ E ÝÝÝÝÑ Φ0pEq ÝÝÝÝÑ Ext1pT `i`1, Eq b T `i`1 ' Ext1pT `i`1pKXq, Eq b T `i`1pKXq ÝÝÝÝÑ 0 , ϕ and also an isomorphism (48) Φ1pEq -- Ext2pT `i`1, Eq b T `i`1 ' Ext2pT `i`1pKXq, Eq b T `i`1pKXq P Ti`1. Suppose first that E P Fi so that φ`maxpEq ď φ`pT `i`1q. Then it follows that ϕ is injective (so that Φ´1pEq " 0) and coker ϕ P Fi. Thus Φ0pEq P Fi, but in fact we can say more. Noting that ΦpT `i`1pKXqq " T `i`1r´1s (and similarly that ΦpT `i`1q " T `i`1pKXqr´1s), we get (49) HompT `i`1, ΦpEqq " HompΦpT `i`1pKXqqr1s, ΦpEqq " HompT `i`1pKXq, Er´1sq " 0, HompT `i`1pKXq, ΦpEqq " HompΦpT `i`1qr1s, ΦpEqq " HompT `i`1, Er´1sq " 0, where the final equality follows from φ`maxpEr´1sq " φ`maxpEq ´ 1 ă φ`pT `i`1q " φ`pT `i`1pKXqq. From the triangle Φ0pEq Ñ ΦpEq Ñ Φě1pEqr´1s Ñ Φ0pEqr1s we see that this implies HompT `i`1, Φ0pEqq " HompT `i`1pKXq, Φ0pEqq " 0 so that Φ0pEq P Fi`1. Therefore ΦpEq P Ai`1, as claimed. Now we assume that E P Ti, from which it follows coker ϕ P Ti Ă Ti`1. Since T `i`1, T `i`1pKXq P Ti`1, we get Φ0pEq P Ti`1. By T `i`1, T `i`1pKXq P Fi, we get Φ´1pEq P Fi. Since (50) HompT `i`1, ΦpEqr´1sq " HompΦpT `i`1pKXqq, ΦpEqr´2sq " HompT `i`1pKXq, Er´2sq " 0, HompT `i`1pKXq, ΦpEqr´1sq " HompΦpT `i`1q, ΦpEqr´2sq " HompT `i`1, Er´2sq " 0, for the same reasons as above, we see from the triangle that HompT `i`1, Φ´1pEqq " HompT `i`1pKXq, Φ´1pEqq " 0. Thus we get Φ´1pEq P Fi`1. Finally, by Serre duality we have Φ´1pEq Ñ ΦpEq Ñ Φě0pEq Ñ Φ´1pEqr1s Ext2pT `i`1, Eq_ -- HompE, T `i`1pKXqq " 0, Ext2pT `i`1pKXq, Eq_ -- HompE, T `i`1q " 0, since E P Ti implies that φ`minpEq ą φ`pT `i`1q " φ`pT `i`1pKXqq. Therefore, Φ1pEq " 0, so ΦpEq P Ai`1r1s, as required. i`1pAi`1q Ă Ai. Let Ψ be the inverse of Φ, and set ΨppEq :" HppΨpEqq We next claim that R´1 T ` for any E P DbpXqq. Recall that for any E P DbpXq we have a distinguished triangle (51) ΨpEq Ñ E Ñ T `i`1r2s b RHompT `i`1, Eq ' T `i`1pKXqr2s b RHompT `i`1pKXq, Eq Ñ ΨpEqr1s. Then (52) HompT `i`1, EpKXqq b T `i`1 ' HompT `i`1, Eq b T `i`1pKXq -- Ψ´1pEq and we have an exact sequence (53) 0 ÝÝÝÝÑ Ext1pT `i`1, EpKXqq b T `i`1 ' Ext1pT `i`1, Eq b T `i`1pKXq ÝÝÝÝÑ Ψ0pEq ÝÝÝÝÑ E ÝÝÝÝÑ Ext2pT `i`1, EpKXqq b T `i`1 ' Ext2pT `i`1, Eq b T `i`1pKXq ÝÝÝÝÑ Ψ1pEq ÝÝÝÝÑ 0. ψ MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 29 We prove the claim by showing that ΨpFi`1q Ă Ai and ΨpTi`1q Ă Air1s which suffices by consid- ering the exact sequence (46). First assume that E P Fi`1. Then Ψ´1pEq " 0 by (52). From (53), we get Ψ0pEq P Fi and Ψ1pEq P Ti`1. Since (54) HompΨpEq, T `i`1rpsq " HompE, ΦpT `i`1qrpsq " HompE, T `i`1pKXqrp ´ 1sq " 0, HompΨpEq, T `i`1pKXqrpsq " HompE, ΦpT `i`1pKXqqrpsq " HompE, T `i`1rp ´ 1sq " 0 for p ď 0, Ψ1pEq P Ti. Therefore ΨpEq P Ai. Now assume that E P Ti`1. Then Ψ´1pEq P Fi by (52). We must show that Ψ0pEq P Ti and Ψ1pEq " coker ψ " 0. As a quotient of an object generated by T `i`1 and T `i`1pKXq, Ψ1pEq " coker ψ P Ti`1. Moreover, for the same reason we note that im ψ P Ti`1 as E P Ti`1. As a subobject of an object generated by T `i`1 and T `i`1pKXq, we also have im ψ P Fi. Hence φ`pim ψq " φ`pT `i`1q. Similarly, coker ψ is a σ`-semistable object of φ`pcoker ψq " φ`pT `i`1q. It follows that coker ψ and im ψ are direct sums of T `i`1 and T `i`1pKXq. Now by using (54) we get 0 " HompΨpEq, T `i`1pDqr´1sq " HompΨ1pEq, T `i`1pDqq, for D " 0, KX . Thus we have Ψ1pEq " coker ψ " 0. Writing TipEq and FipEq for the components of E in the torsion pair pTi, Fiq, it follows from im ψ P Fi that TipEq Ă ker ψ, and it is easy to see that then Tipker ψq " TipEq. Applying Homp´, T `i`1pDqq pD " 0, KXq to the short exact sequence we see that hompE, T `i`1pDqq " hompFipEq, T `i`1pDqq for D " 0, KX . As 0 Ñ TipEq Ñ E Ñ FipEq Ñ 0, ext2pT `i`1, Eq " hompE, T `i`1pKXqq, ext2pT `i`1, EpKXqq " hompE, T `i`1q, it follows from the short exact sequence 0 Ñ ker ψ{TipEq Ñ FipEq Ñ Ext2pT `i`1, EpKXqq b T `i`1 ' Ext2pT `i`1, Eq b T `i`1pKXq Ñ 0 (55) that Hompker ψ{TipEq, T `i`1pDqq " 0 by the same reasoning. But this forces ker ψ{TipEq " 0. Indeed, as E P Ti`1, we have FipEq P Fi X Ti`1, so FipEq is a direct sum of copies of T `i`1 and T `i`1pKXq, and thus so is ker ψ{TipEq from (55). Thus we have ker ψ " TipEq P Ti so that Ψ0pEq P Ti`1. Then (54) implies Ψ0pEq P Ti. Therefore ΨpEq P Air1s, as required. As Ai deals in parallel with objects in P0p1q considered with respect to σ´-stability, we have (cid:3) the following result: Proposition 6.11. R´1 T ´ i`1 Proof. We set Φ :" RT ´ (56) i`1 induces an equivalence Ai Ñ Ai`1. and Ψ :" R´1 T ´ i`1 . We only show that ΦpAi`1q Ă Ai . We note that HompT ´i`1, ΦpEqrpsq " HompΦpT ´i`1pKXqqr1s, ΦpEqrpsq " HompT ´i`1pKXq, Erp ´ 1sq " 0, HompT ´i`1pKXq, ΦpEqrpsq " HompΦpT ´i`1qr1s, ΦpEqrpsq " HompT ´i`1, Erp ´ 1sq " 0 for E P P0p1q and p ď 0. Assume that E P Fi`1. For the morphism (57) ker ϕ and im ϕ are generated by T ´i`1, T ´i`1pKXq. By (56), Φ´1pEq " 0 and Φ0pEq P Fi . Since Φ1pEq is generated by T ´i`1, T ´i`1pKXq, Φ1pEq P T i . Therefore ΦpEr1sq P Ai . ϕ : HompT ´i`1, Eq b T ´i`1 ' HompT ´i`1pKXq, Eq b T ´i`1pKXq Ñ E, 30 HOWARD NUER AND K ¯OTA YOSHIOKA Assume that E P T i`1. Then Φ1pEq " 0. Since Φ´1pEq P Fi`1, (56) implies Φ´1pEq P Fi . Since coker ϕ, T ´i`1, T ´i`1pKXq P T i , Φ0pEq P T i . Therefore ΦpEq P Ai . (cid:3) For negative i, we must make the following definition. Definition 6.12. Assume that i ď 0. (1) Let pT i , Fi q be the torsion pair of P0p1q such that is generated by σ`-stable objects E with φ`pEq ě φ`pT `i q. (a) T i :" xT `0 , T `0 pKXq, ..., T `i`1, T `i`1pKXqy. (b) Fi Let Ai " xT i , Fi r1sy be the tilting. (2) Let pTi, Fiq be a torsion pair of P0p1q such that (a) Ti :" xT ´0 , T ´0 pKXq, ..., T ´i`1, T ´i`1pKXqy. (b) Fi is generated by σ´-stable objects E with φ´pEq ď φ´pT ´i q. Let Ai " xTir´1s, Fiy be the tilting. Since F0 " T0 " 0, we have A0 " A0 " P0p1q. We also have a similar result to Remark 6.9. Moreover, we also have the following claims whose proofs are similar to those of Proposition 6.10 and Proposition 6.11. Proposition 6.13. Assume that i ď 0. (1) We have an equivalence R´1 T ` i (2) We have an equivalence RT ´ i : Ai Ñ Ai´1. : Ai Ñ Ai´1. 6.2.2. Preservation of stability. Having defined the abelian categories Ai and Ai , we will relate a certain stability on them to σ-stability. Example 6.14. Recall Definition 6.6 and Proposition 6.7, and let B " P0p1q. (1) We take an orientation preserving injective homomorphism Z : H Ñ C such that Zpw0q, Zpw1q P H Y Ră0, and Zpw1q{Zpw0q P H. The second condition means that φZpw1q ą φZpw0q. Then E P P0p1q is σ`-semistable if and only if E is Z-semistable. In this case, Ai pi ě 0q is an example of B1 for some θ ě 0, and Ai pi ď 0q is an example of B2 for some θ ď 0. (2) We take an orientation reversing injective homomorphism Z : H Ñ C such that Zpw0q, Zpw1q P H Y Ră0, and Zpw0q{Zpw1q P H. Then E P P0p1q is σ´-semistable if and only if E is Z-semistable. In this case, Ai pi ď 0q is an example of B1, and Ai pi ě 0q is an example of B2. Now we can finally prove a sequence of comparison results that allow us to reduce our analysis to the case of minimal Mukai vectors. 1 (2) (1) (a) RT ` Proposition 6.15. where v1 " Rw1pvq. : A0 Ñ A1 induces an isomorphism Mσ´pvq Ñ Mσ`pv1q, : A0 Ñ A1 induces an isomorphism Mσ`pvq Ñ Mσ´pv1q, where v1 " R´1 w1pvq. : A0 Ñ A´1 induces an isomorphism Mσ`pvq Ñ Mσ´pv1q, where v1 " Rw0pvq. : A0 Ñ A´1 induces an isomorphism Mσ´pvq Ñ Mσ`pv1q, where v1 " R´1 w0pvq. Proof. By Example 6.14, we can apply Proposition 6.7 to compare the stabilities on Ai and Ai to σ-stability on A0 " A0. Since the orientation of H is reversed under the reflection, the claims follow from Proposition 6.7 and Propositions 6.10, 6.11, 6.13. (b) R´1 T ´ 1 (a) RT ´ (b) R´1 T ` 0 (cid:3) 0 This result has the following significant corollary: MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 31 Corollary 6.16. If xv, w1y " 0, then RT1 induces an isomorphism Mσ´pvq Ñ Mσ`pvq. xv, w0y " 0, then RT0 induces an isomorphism Mσ`pvq Ñ Mσ´pvq. Proof. We apply Proposition 6.15 and observe that for i " 0, 1 we get Rwipvq " v ` 2xv, wiy " v. If (cid:3) In order to use the sequence of weakly spherical reflections to reduce any v to the minimal Mukai vector in its orbit, as in (44), we consider the next step in the program beyond that in Proposition 6.15. Proposition 6.17. (a) RT ` (1) Assume that i ą 0. i i`1 RT ` i`1 R´1 T ´ i v1 " Rwi`1 Rwipvq. (b) R´1 T ´ : Ai´1 Ñ Ai`1 induces an isomorphism Mσ`pvq Ñ Mσ`pv1q, where : Ai´1 Ñ Ai`1 induces an isomorphism Mσ´pvq Ñ Mσ´pv1q, where wipvq. : Ai`1 Ñ Ai´1 induces an isomorphism Mσ´pvq Ñ Mσ´pv1q, where i`1 (a) RT ´ wi`1 R´1 v1 " R´1 (2) Assume that i ă 0. i RT ´ i R´1 v1 " R´1 Proof. (1) Since RT ` similar argument as in Proposition 6.15. The proof of (2) is similar. v1 " Rwi Rwi`1pvq. wi R´1 i`1 RT ` i`1 R´1 and R´1 T ´ (b) R´1 T ` T ´ i T ` i`1 i : Ai`1 Ñ Ai´1 induces an isomorphism Mσ`pvq Ñ Mσ`pv1q, where wi`1pvq. preserve the orientation, we get the claims by a (cid:3) It turns out that the composition of two consecutive weakly spherical reflections as in the previous proposition is independent of i, which is the content of the following lemma. i ‰ 0, 1, i " 0, 1. i`1 RT ` i RT ` i´1 Lemma 6.18. i pT `i´1q "#T `i`1r1s, T `i`1pKXqr1s, T `i`1, T `i`1pKXq, (1) RT ` i RT ` (2) We have RT ` . In particular, RT ` Proof. (1) Assume that i ě 2. By Proposition 6.17 (1), RT ` isomorphism i´1 " RT ` i`1 RT ` i i " RT1 RT0 for all i. : Ai´2 Ñ Ai induces an i´1pT `i´1pKXqq " RT ` i RT ` i pT `i´1r´1sq " T `i`1, T `i`1pKXq. If i " 0, 1, then Proposition 6.15 implies RT ` Hence RT ` we get RT ` T `i`1, T `i`1pKXq. Assume that i ă 0. Then Proposition 6.17 (2) implies RT ` induces an isomorphism Mσ`pwi´1q Ñ Mσ`pwi`1q. i pT `i´1r´1sq is a σ`-stable object with Mukai vector wi`1. Then i pT `i´1q " : Ai´2 Ñ Ai i RT ` i´1 Mσ`p´wi´1q Ñ Mσ`p´wi`1q. Hence we get RT ` i pT `i´1q " T `i`1r1s, T `i`1pKXqr1s. i RT ` i´1 R´1 i " RR T ` (2) Since RT ` i pT ` T ` In the same way, we also see that RT ´ i´1q i RT ´ by [22, Lemma 8.21], the claim follows from (1). (cid:3) i`1 " RT0 RT1 for all i. This leads us to make the following definition. Definition 6.19. We set R` :" RT1 RT0 and R´ :" RT0 RT1. We have finally studied the action of GH enough to prove our main reduction result. Proposition 6.20. Let vn P Cn be defined as in (44). That is, vn is in the orbit of v0 P C0. 32 (58) (59) (1) If n is even, then R HOWARD NUER AND K ¯OTA YOSHIOKA n 2 n 2 ` R ´ induces a birational map Mσ´pvnq -- Mσ´pv0q 99K Mσ`pv0q -- Mσ`pvnq, which is isomorphic in codimension one. n´1 n´1 (2) If n is odd, then R 2 ` RT1 RT1 R ´ 2 induces a birational map Mσ´pvnq -- Mσ´pv1q -- Mσ`pv0q 99K Mσ´pv0q -- Mσ`pv1q -- Mσ`pvnq, which is isomorphic in codimension one. Proof. (1) Using n{2 applications of Proposition 6.17, we get that R Mσ´pvnq „ÝÑ Mσ´pv0q, and by Lemma 6.2 the open subset of σ0-stable objects M s birational map ´ induces an isomorphism σ0pv0q provides a n 2 Moreover, the complement of M s The proof of (2) follows similarly by using Proposition 6.15 twice in the middle. (cid:3) n 2 σ0pv0q in Mσpv0q has codimension at least 2. Using Proposition ` gives an isomorphism Mσ`pv0q „ÝÑ Mσ`pvnq, which gives the result. 6.17 again, we get that R Mσ´pv0q 99K Mσ`pv0q. The complementary result for v on the boundary of some Cn, which generalizes Corollary 6.16, is the following. Proposition 6.21. Suppose that v P CW satisfies xv, wny " 0 for some n. Then Mσ`pvq -- Mσ´pvq. Proof. First let us suppose that n is odd. Then we take v1 " R 0. isomorphism In this case, by Propositions 6.15 and 6.17, the composition R ` ´ pvq which satisfies xv1, w1y " induces an RT ` 1 R ´ n´1 n´1 n´1 2 2 2 (60) Mσ´pvq -- Mσ´pv1q -- Mσ`pv1q -- Mσ`pvq. n 2 R If instead n is even, then we take v0 " R ` RT ` (61) ´ induces an isomorphism 0 R n 2 Mσ´pvq -- Mσ´pv0q -- Mσ`pv0q -- Mσ`pvq. n 2 ´pvq, which satisfies xv0, w0y " 0. In this case, (cid:3) We can conclude from Propositions 6.20 and 6.21 that, in the case of a non-isotropic wall of type (3)(b), the two moduli spaces Mσ`pvq and Mσ´pvq are birational. As identical statements to these propositions hold for non-isotropic walls of type (3)(a) (see [6, Prop. 6.8 and Lem. 7.5]), the same conclusion holds in that case. Now we move on to the final non-isotropic case we need to deal with. 6.3. Case (3)(c): Exactly one σ0-stable spherical and exceptional object, respectively, up to ´ b OXpKXq. We shall briefly explain the case where H contains effective exceptional and spherical vectors. Denote the Mukai vector of the unique σ0-stable exceptional object (up to ´ b OXpKXq) by w0, and complete it to a basis of H such that H " Zw0 ` Zz, where z satisfies xw0, zy " 0 and D :" z2 ą 0. Furthermore, we know there is an effective w P CW X H with w2 " ´2. Then x2 ´ Dy2 " 2 has an integral solution, which implies ?D is irrational. Let ps, tq be a solution of x2 ´ Dy2 " 2, and set α :" s ` t?D. Then ´α2{2 " p1 ´ s2q ´ st?D, and px, yq " p1 ´ s2,´stq is a solution of x2 ´ y2D " 1. Let px1, y1q be a minimal solution of x2 ´ y2D " 1 such that x1 ă 0 and y1 ą 0, and set β :" x1 ` y1?D. Then ´1 ă β ă 0 (see [2, Thm. 11.3.1]) and α2{2 " βn for some n. 2 satisfies pu ` v?Dq2 " 2. But then We claim that n is odd. If not, then u ` v?D :" α{β , which contradicts the irrationality of ?D. Hence n is odd and α2{2 " ´βn. Set ?D " 2´u2´Dv2 2uv n MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 33 n " 2k ` 1. Then s1 ` t1?D " α{βk satisfies ps1 ` t1?Dq2 " ´2β and s2 1D " 2. So replacing ps, tq by ps1, t1q (and possibly taking a conjugate), we may assume that α :" s1 ` t1?D satisfies Npαq " 2, β " ´α2{2, and s1 ă 0 and t1 ą 0. Moreover, the same argument applied to any other solution γ of the equation x2 ´ Dy2 " 2 shows that γ " α2n`1 2n , where γ2{2 " ´β2n`1. 1 ´ t2 We define psn, tnq P Z Z by (62) s2n ` t2n?D :"p´1qn`1βn " ´α2n{2n, pn ě 1q s0 ` t0?D :"1, s2n´1 ` t2n´1?D :"α2n´1{2n´1, pn ě 1q s´2n ` t´2n?D :" ´ s2n ` t2n?D " p´βq´n, pn ě 1q s´2n`1 ` t´2n`1?D :" ´ s2n´1 ` t2n´1?D " 2np´αq´p2n´1q, pn ě 1q. For n P Z, we set wn :" snw0 ` tnz. Then w2 n "#´1, 2 n ´2, 2 ∤ n It is easy to see that (63) wn`1 " ´ Rwnpwn´1q, pn ě 2q wn´1 " ´ Rwnpwn`1q, pn ď ´1q w2 "Rw1pw0q, w´1 " Rw0pw1q, where Rwn is, by abuse of notation, the action on HalgpX, Zq induced by spherical/weakly spherical reflection through wn. To be concrete, we have Rwnpuq " u ` 2xu, wnywn, 2 n Rwnpuq " u ` xu, wnywn, 2 ∤ n. From this description it is clear that (64) where ∆pXq was defined in (9). By the assumption on H and W, there exists a σ0-stable spherical object T , so we must have vpTq " w2n`1 for some n P Z. In particular, c1pw2n`1q " Z pmod 2q, where Z is a nodal cycle by Theorem 3.8. From (63), we see that ∆pXq X H Ď t wn n P Z u , (65) c1pw2j`1q " c1pw1q pmod 2q, for all j P Z, so the inclusion in (64) is an equality: (66) ∆pXq X H " t wn n P Z u . As in Section 6.2, it is not difficult to see that " u P H Zσ0puq Zσ0pvq ą 0, u2 ě ´2* Ă Qě0w0 ` Qě0w1, and thus CW " Rą0w0 ` Rą0w1. It follows from this and (66) that there are σ0-stable objects, T0 and T1, of classes w0 and w1, respectively. Moreover, as we then have it follows from (65) and Theorem 3.8 that there are σ-stable objects T n with vpT n q " wn. Note that T `i " T ´i " Ti for i " 0, 1. ∆pXq X CW " t wn n P Z u , 34 HOWARD NUER AND K ¯OTA YOSHIOKA Let Cn Ă CW be the region between wKn and wKn`1 as in Section 6.2. Then C0 is again the fundamental domain of the Weyl group GH associated to ∆pXq X H. Up to reordering, we may assume that (67) for any σ`-stable object E with vpEq2 ě 0 and (68) for any σ´-stable object E with vpEq2 ě 0. φ`pT `1 q ą φ`pT `2 q ą ą φ`pEq ą ą φ`pT ` ´1q ą φ`pT `0 q φ´pT ´1 q ă φ´pT ´2 q ă ă φ´pEq ă ă φ´pT ´ ´1q ă φ´pT ´0 q We note that T n pKXq " T n if n is odd and T n pKXq fl T n if n is even. As in Section 6.2, we see that (69) R` :" RT1 RT0 " RT ` i`1 RT ` i is an equivalence which preserves σ`-semistability and (70) R´ :" RT0 RT1 " RT ´ i´1 RT ´ i is an equivalence which preserves σ´-semistability. Here, RT0 and RT1 are the weakly-spherical and spherical reflections associated to the exceptional and spherical objects T0 and T1, respectively. Then we have the following results. Proposition 6.22. Let v P CW X H. (1) Suppose that v P Cn. (a) If n is even, then R n 2 n 2 ` R ´ induces a birational map Mσ´pvq -- Mσ´pv0q 99K Mσ`pv0q -- Mσ`pvq, n 2 isomorphic in codimension one, where v0 " R n´1 n´1 ´pvq. (b) If n is odd, then R induces a birational map 2 ` R2 T1 R ´ 2 Mσ´pvq -- Mσ´pv1q -- Mσ`pv0q 99K Mσ´pv0q -- Mσ`pv1q -- Mσ`pvnq, ´ pvq and v0 " RT1pR isomorphic in codimension one, where v1 " R n´1 2 n´1 2 ´ pvqq. (2) Suppose that xv, wny " 0. (a) If n is even, then R n 2 ` RT ` 0 R ´ induces an isomorphism n 2 (71) (72) (73) (74) Mσ´pvq -- Mσ´pv0q -- Mσ`pv0q -- Mσ`pvq, ´pvq. (b) If n is odd, then R where v0 " R n´1 n´1 n 2 induces an isomorphism 2 2 ` RT ` 1 R ´ n´1 Mσ´pvq -- Mσ´pv1q -- Mσ`pv1q -- Mσ`pvq, ´ pvq. 2 where v1 " R i is spherical. To avoid this word-for-word repetition, we omit these proofs. Note Proposition 6.22 is proven by defining Ai and Ai analogously to Section 6.2 and by showing that induces isomorphisms on moduli with identical proofs except for the minor adjustments when RT i is odd so that T i that the birational map Mσ`pv0q 99K Mσ´pv0q follows again from Lemma 6.2. We have seen in this section that, regardless of its type, a non-isotropic wall W induces a bira- tional map Mσ`pvq 99K Mσ´pvq, and this map is isomorphic outside of a subvariety of codimension at least 2 unless xv, wy " 0 for some w P CW X ∆pXq. We finish off the section by explaining how this birational map relates to the contraction maps π : Mσpvq Ñ M MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 35 induced by ℓσ0,, as in Theorem 3.14. Recall that the curves contracted by π are precisely those curves of σ`-semistable objects that become S-equivalent with respect to σ0. Let E be a σ0-stable object of class v0 P C0. Then it is an irreducible object of A0. By Propositions 6.10, 6.11, and 6.13, and their analogues in the other subcases of Proposition 5.7(3), it follows that for a non-negative integer n, E`n :" RT ` 1 pEq is an irreducible object of An and E´n :" R´1 n R´1 1 pEq is an irreducible object of An. By Remark 6.9 , En are successive extensions of E by T i , T i pKXq. In particular, En is and the definition of RT S-equivalent to n´1 R´1 n´1 RT ` n RT ` T ´ T ´ T ´ i (75) (76) with respect to σ0, where $'&'% , 1 0 ' T 'k1 E ' T 'k0 E ' pT0 ' T0pKXqq'k0 ' pT1 ' T1pKXqq'k1 , E ' pT0 ' T0pKXqq'k0 ' T 'k1 1 , in Case (3)(a) in Case (3)(b) in Case (3)(c) vpEn q "$'&'% vpEq ` k0w0 ` k1w1, vpEq ` 2k0w0 ` 2k1w1, vpEq ` 2k0w0 ` k1w1, in Case (3)(a) in Case (3)(b) in Case (3)(c) . Thus if v " Rwn Rwn´1 Rw1pv0q for v0 P C0, then the contraction map π : Mσ Ñ M is σ0pv0q under the isomorphism Mσpv0q „ÝÑ Mσpvq given by the above injective on the image of M s composition of Fourier-Mukai functors.2 In particular, π is birational. Moreover, the S-equivalence in (75) true even if E is strictly σ0-semistable, so the curves contracted by π : Mσpvq Ñ M are in direct correspondence with the curves of S-equivalent objects contracted by π0 : Mσpv0q Ñ M 0,, where this is the analogous morphism for v0. When v0 P C0 proper, that is, xv, wy ą 0 for all w P ∆pXq X CW , the codimension of the exceptional locus of π0 is at least two, so the same remains true of π. When v P wKn , however, this exceptional locus has codimension one and the two moduli spaces Mσpvq are isomorphic. We explore in the next section when this divisor is contracted upon crossing the wall W. 7. Divisorial contractions in the non-isotropic case In this section we aim to prove the following result. Proposition 7.1. Assume that the potential wall W is non-isotropic. Then W induces a divisorial contraction on Mσ`pv, Lq if and only if either there exists a spherical class w P ∆pXq´2 X CW such that xv, wy " 0 or there exists an exceptional class w P ∆pXq´1 X CW such that xv, wy " 0, v´ 2w is spherical, and L " D` rk v 2 KX pmod 2q for a nodal cycle D. If T is a σ-stable spherical object of class w (or v ´ 2w in the second case), then the contracted divisor can be described as a q ‰ 0 or by Homp , Tq ‰ 0. Brill-Noether divisor for T : it is given either by the condition HompT, By Lemma 6.2, we know that the locus of strictly σ0-semistable objects has codimension one if and only if xv, wy " 0 for some class w P ∆pXq X CW. In particular, this condition must be met if W induces a divisorial contraction. In the next two lemmas, we prove that only when w is spherical does this divisor get contracted. Lemma 7.2. Suppose that H is non-isotropic and W is a potential wall associated to H. If there exists an effective spherical class w with xv, wy " 0, then W induces a divisorial contraction. A generic element E in the contracted divisor D admits a short exact sequence 0 Ñ T Ñ E Ñ F Ñ 0 or 0 Ñ F Ñ E Ñ T Ñ 0, 2As written, this isomorphism exists for n even. For n odd, the composition of Fourier-Mukai functors would give Mσ¯ pv0q „ÝÑ Mσpvq 36 HOWARD NUER AND K ¯OTA YOSHIOKA where T P M s filtration steps in a JH-filtration for E with respect to σ0. σ`pwq and F P M s σ`pv ´ wq, such that the inclusion T ãÑ E or F ãÑ E is one of the Proof. By the discussion at the end of Section 6, we can use a composition of spherical or weakly- spherical reflections, as in [6, Corollary 7.3, Lemma 7.5] (in Case (3)(a) and Proposition 6.22 (in Case (3)(c), to reduce the discussion to the case of a minimal Mukai vector. We assume this to be the case. Then the spherical class must be w0 or w1 (in Case (3)(a)) or w1 (in Case (3)(c)), and we assume it is w1 with the other case being dealt with similarly. As in [6, Lemma 7.4], we first prove that v ´ w1 is also minimal. Let us start by assuming that v2 ě 2. We note that we must in-fact have v2 ě 3, since v2 " 2 gives pv´ w1q2 " 0, contrary to the assumption that H is non-isotropic. Write v " xw0` yw1 with x, y P Q. Then 0 " xw1, vy " 0 gives y " m 2 x, where recall that m " xw0, w1y. As xw1, v´w1y " 2, to show that v ´ w1 is minimal it suffices to check that 0 ` ymq ´ m " my2w2 0 m2 ` 1 ´ m " my1 ` 0 " ´2, so v2 ě 3 implies that 3 m2 ´ 1 . 2 ď y2p1 ´ 4 2w2 0 We now consider the Cases (3)(a) and (3)(c) separately. First suppose we are in Case (3)(a). Then w2 m2q, and as in the proof of Proposition 5.7, we have m ě 3. If m " 3, then it is easy to show that the equations v2 " 3 and xv, w1y " 0 have no rational solutions, so we may assume that v2 ě 4. But then 2 ď y2p1 ´ 4 m2q, so 0 ď xw0, v ´ w1y " pxw2 y21 ´ Taking square-roots, we see that 4 4 2w2 0 9 ě 10 9 ą 1. " y21 ´ m22 y1 ` y21 ´ m2 5 m2 ą 1, m22 0 " ´1, so v2 ě 3 is equivalent to 3 9 8 ą 1, ě 4 and therefore we have xw0, v ´ w1y ą 0. If, instead, m ě 4, then we get p1 ´ 4 it follows that m2q ě 3 4 , from which Now suppose we are in Case (3)(c). Then w2 so indeed xw0, v ´ w1y ą 0. m2q, and now m ě 2. If m " 2, then again one can easily check that the equations v2 " 3 and xv, w1y " 0 have no rational solutions, so v2 ě 4, i.e. y2p1 ´ 2 m2q ě 2. Thus 2 ď y2p1´ 2 2 y21 ´ m22 " y21 ´ m2 1 m2 ě 7 so xw0, v ´ w1y ě 0. If, instead, m ě 3, then`1 ´ 2 y21 ´ ě y21 ´ m22 m2 7 9 ě 2 2 2 2 ě 1, 9 and thus 7 6 ą 1, so we get xw0, v ´ w1y ą 0. As we have shown that v ´ w1 is minimal, Lemma 6.2 guarantees that the generic element F P Mσ`pv ´ w1q is also σ0-stable. But then for the unique σ0-stable spherical object T1 with vpT1q " w1 we have ext2pF, T1q " hompT1pKXq, Fq " hompT1, Fq " 0 " hompF, T1q by stability. Thus ext1pF, T1q " xv ´ w1, w1y " 2, so there is a family of extensions 0 Ñ F Ñ Ep Ñ T1 Ñ 0, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 37 parametrized by p P P1 " PpExt1pT1, Fqq, which are all S-equivalent with respect to σ0. By [6, Lemma 6.9], they are σ`-stable. Thus π` contracts this rational curve. Varying F P M s σ0pv ´ w1q sweeps out a family of σ`-stable objects in Mσ`pvq of dimension 1 ` pv ´ w1q2 ` 1 " v2 " dim Mσ`pvq ´ 1. Thus we get a divisor contracted by π`, which must then have relative Picard rank one, so this is the only component contracted by π`. Finally, suppose that v2 " 1. Then pv ´ w1q2 " ´1, so we must be in Case (3)(c) with v ´ w1 " wn for some n P Z. As wn P Zw0 ` Zw1, it follows that v " xw0 ` yw1 with x, y P Z. But then xv, w1y " 0 is equivalent to x " 2 1 " x2 m2 m y so that v2 " 1 is equivalent to 2 ´ 1 " 2 x2pm2 ´ 2q , whose only solution in the positive integers is x " 1 and m " 2. Thus we have v ´ w1 " w0, so we are, as above, guaranteed that the unique member of Mσ`pv ´ w1q is σ0-stable. The same argument gives a family of extensions 0 Ñ T0 Ñ Ep Ñ T1 Ñ 0, parametrized by p P P1 " PpExt1pT1, T0qq, which are S-equivalent with respect to σ0 but are σ`- stable. This curve is contracted by π` and is a divisor in the two-dimensional moduli space Mσ`pvq, so π` again has relative Picard rank one. These extensions thus give the unique curve contracted by π`, as required. (cid:3) Having confirmed that when w is spherical we do get a divisorial contraction, we prove that this is not the case when w is exceptional with one exception. Lemma 7.3. Suppose that H is non-isotropic and W is a potential wall associated to H. If there exists an effective exceptional class w with xv, wy " 0, then W only induces a divisorial contraction on Mσ`pv, Lq if v2 " 2, L " D` rk v 2 KX pmod 2q for a nodal cycle D, and H falls into Case (3)(c) of Proposition 5.7. In this case, for E in the contracted divisor, there is a short exact sequence 0 Ñ T ' TpKXq Ñ E Ñ S Ñ 0 or 0 Ñ S Ñ E Ñ T ' TpKXq Ñ 0, where T P M s σ`pwq and S P M s σ`pv ´ 2wq, such that the inclusion T ãÑ E or S ãÑ E is one of the filtration steps in a JH-filtration for E with respect to σ0, while the generic E P Mσ`pvq satisfies HompT, Eq " HompE, Tq " 0. In general, there is nevertheless a divisor Dσ`pvq whose generic element E admits a short exact sequence 0 Ñ T Ñ E Ñ F Ñ 0 or 0 Ñ F Ñ E Ñ T Ñ 0, σ`pwq and F P M s where T P M s σ`pv ´ wq, such that the inclusion T ãÑ E or F ãÑ E is one of the filtration steps in a JH-filtration for E with respect to σ0, while the generic E P Mσ`pvq satisfies HompT, Eq " HompE, Tq " 0. Moreover, when v is minimal, this divisor is precisely the locus of strictly σ0-semistable objects. Proof. As before, we may assume that v is minimal. Then in terms of Proposition 5.7, HW must fall into cases (2)(b), (3)(b), or (3)(c). By minimality of v, w must be w0 or w1 in Case (3)(b) or w0 in Case (3)(c), and we assume it is w0 with the other case being dealt with similarly. Furthermore, observe that there cannot exist any n P Z such that v2 " n2 as then pv ´ nw0q2 " 0, contrary to the hypothesis that W is non-isotropic. So, in particular, v2 ě 2, from which it follows that pv ´ w0q2 ą 0. Let us first show that v´ w0 is minimal. As xv´ w0, w0y " 1 ą 0, this is clear in Case (2)(b), so we can restrict ourselves to Cases (3)(b) and (3)(c). Thus it remains to show that 0 ď xv´ w0, w1y. 38 HOWARD NUER AND K ¯OTA YOSHIOKA Writing v " xw0 ` yw1, the conditions xv, w0y " 0, v2 ě 2, and xv ´ w0, w1y ě 0 become respectively. As m ě 2 in either case, we get y " 1 ` w2 1 w2 1 x m , x21 ` m2 "#1 ´ 1 m2 , 1 ´ 2 m2 , x21 ` m22 w2 1 4 , 1 2 , if w2 if w2 1 " ´1, m2 ě 2, and m„x1 ` 1 " ´2 ě# 3 m2 1 m2 ´ 1 ě 0, ě x21 ` x1 ` w2 1 w2 1 w2 1 m2 ´ 1 ě 0, 1 2 , if w2 if w2 1 " ´1, 1 " ´2, ě 2 ě 1. 2 ě 2 1 in either case. Thus Taking square-roots gives that indeed as required. We consider the case when v2 " 2, L " D ` rk v 2 KX pmod 2q, and H falls into Case (3)(c) of Proposition 5.7. As xv ´ w0, v ´ 2w0y " 0 and pv ´ 2w0q2 " ´2, it follows from the minimality of v ´ w0 that v ´ 2w0 " w1 " vpSq, for the unique σ0-stable spherical object S. We denote by T0 the unique σ0-stable object of class w0 (up to ´ b OXpKXq). By stability, we have hompS, T0pDqq " ext2pS, T0pDqq " hompT0pDq, Sq " 0, for D " 0, KX . Thus ext1pS, T0 ' T0pKXqq " xv ´ 2w0, 2w0y " 4, so by [17, Lemma 6.1-6.3] there is a P1 P1 worth of non-isomorphic σ`-stable E fitting into a short exact sequence 0 Ñ T0 ' T0pKXq Ñ E Ñ S Ñ 0. As this gives a contracted divisor, it must be the only contracted divisor, as claimed. We now treat the general case, that is, either v2 ą 2 or v2 " 2 and either L ı D ` rk v 2 KX pmod 2q or H does not fall into Case (3)(c) of Proposition 5.7. As v ´ w0 is minimal, it follows from Lemma 6.2 that there exists a σ0-stable object F of class v ´ w0. By stability hompF, T0q " ext2pF, T0q " hompF, T0pKXqq " 0, so ext1pF, T0q " xv ´ w0, w0y " 1, and there exists a unique non-trivial extension 0 Ñ T0 Ñ E Ñ F Ñ 0, which is σ`-stable by [6, Lemma 6.9]. By a dimension count, upon varying F P M s σ0pv ´ w0q these extensions sweep out a divisor of strictly σ0-semistable objects which does not get contracted by π`. Moreover, from the proof of Lemma 6.2, it follows that this is precisely the locus strictly σ0-semistable objects. (cid:3) Remark 7.4. We will see in Section 9 that, if v2 ě 3, then in the setup of Lemma 7.3, W induces a small contraction, contracting a P1P1. As the weakly-spherical reflection RT induces an isomorphism Mσ`pvq „ÝÑ Mσ´pvq that acts as the identity on Mσ`pvqzDσ`pvq, where HompT, Eq " HompE, TpKXqq " 0, we see that Mσ´pvq cannot be the flop of π`. It is unclear whether examples such as this show that there are minimal models of Mσ`pvq which cannot be obtained by Bridgeland wall-crossing. On the other hand, it may be possible to reach this minimal model by crossing a different wall bounding the chamber containing σ`. Proof of Proposition 7.1. This follows directly from Lemma 6.2, Lemma 7.2, and Lemma 7.3. (cid:3) Remark 7.5. It is important to note that with the exception of Lemma 7.3, everything we have said thus far in the non-isotropic case applies to each component Mσ`pv, Lq, Mσ`pv, L`KXq, where σ0pv´ w, L1 ` KXq in Lemmas c1pvq " rL mod KXs. In particular, by taking F P M s σ0pv´ w, L1q, M s MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 39 7.2 and 7.3, we get divisors with the described properties in each component Mσ`pv, Lq, Mσ`pv, L` KXq. We will see in the next section that we must take great care to treat the determinants differently as the wall-crossing behavior is often radically different in each component, in a similar way to the case v2 " 2, L " D ` rk v 2 KX pmod 2q, and v ´ 2w is spherical in Lemma 7.3. 8. Isotropic walls We finally treat the case of isotropic walls. We divide our discussion in two. We will first discuss the case that H contains a primitive isotropic vector u with ℓpuq " 2, in which case the wall W corresponds, after a Fourier-Mukai transform, to the contraction to the Uhlenbeck compactification, see [30, 33]. We will consider separately the case that H only contains primitive isotropic vectors u with ℓpuq " 1. In both cases, we again use the stack of Harder-Narasimhan filtrations, as in Section 4, to study the wall-crossing behavior. We begin by studying in more detail the isotropic lattice H, its isotropic vectors, and the associated moduli spaces. 8.1. Preliminaries. We state here a result that summarizes the facts we will need for a more detailed study of wall-crossing in the isotropic case. Proposition 8.1. Assume that there exists an isotropic class u P H. Then there are two effective, primitive, isotropic classes u1 and u2 in H, which satisfy PH " Rě0u1 ` Rě0u2 and xv1, uiy ě 0 for i " 1, 2 and any v1 P PH. Moreover, one of the following mutually exclusive conditions holds: (1) CW " PH and ℓpu1q ě ℓpu2q. In this case, M s σ0puiq " Mσ0puiq for each i " 1, 2 and a generic σ0 P W; or (2) There exists an exceptional class w P CW X H. In this case ℓpu1q " ℓpu2q, H " Zw ` Zu1 and u2 " u1 ` 2xu1, wyw. Consequently, xu1, u2y " 2xu1, wy2 and CW " Rě0u1 ` Rě0w. Finally, in this case M s σ0pu1q " Mσ0pu1q for a generic σ0 P W, while W is a totally semistable wall for u2; or (3) There exists a spherical class w P CW X H. In this case, we again have ℓpu1q " ℓpu2q, H " Zw ` Zu1, and u2 " u1 ` xu1, wyw. Consequently, xu1, u2y " xu1, wy2 and H is even in this case.3 Finally, CW " Rě0u1 ` Rě0w and M s σ0pu1q " Mσ0pu1q for a generic σ0 P W, while W is a totally semistable wall for u2. 0 " pxu1 ` yv1q2 " 2xyxu1, v1y ` y2pv1q2 Proof. If u P H is a primitive isotropic class, then up to replacing u by ´u, we may assume that u is effective, so we set u1 " u. Completing u1 to a basis H " Zu1 ` Zv1, we see that (77) has a second integral solution, since we can assume y ‰ 0 and xu1, v1y ‰ 0 from the signature of H. Taking the unique effective primitive class on the corresponding line, we get u2. Clearly PH is as claimed, and the inequality xv1, uiy ě 0 follows accordingly. If CW " PH, then the claim about moduli spaces in Case (1) follows from the fact that u1 and u2 are primitive vectors on extremal rays of CW . Moreover, up to renumbering, we assume that ℓpu1q ě ℓpu2q and, in case of equality, xv, u1y ě xv, u2y. spherical as in Case (2) or (3), respectively. Suppose that there exists a class w P CW that is not in PH. Then either w is exceptional or Let us consider first Case (2), and write w P CW X ∆pXq´1 as w " xu1 ` yv1. Then ´1 " w2 " ypyv12 ` 2xxv1, u1yq implies y " 1. Replacing v1 by w, we see that H " Zu1 ` Zw. Then it is easy to that the other primitive effective isotropic vector must satisfy (78) u2 " u1 ` 2xu1, wyw. 3All of these conclusions continue to hold in Case (1) if H admits a class w R CW with w 2 " ´2. 40 HOWARD NUER AND K ¯OTA YOSHIOKA w u2 PH u1 Figure 2. The red lines are defined by u2 " 0. The dark gray region is the positive cone PH, while the first quadrant is the effective cone CW . Pairing this equality with u1, it is clear that xu1, u2y " 2xu1, wy2. Moreover, we see that c1pu2q " c1pu1q pmod 2q from (78), so we get the last statement that ℓpu1q " ℓpu2q. Now consider Case (3), and write w P CW X ∆pXq´2 as w " xu1 ` yv1. Then ´2 " w2 " ypyv12 ` 2xxu1, v1yq implies that y " 1,2. If y " 2, then ¯1 " 2v12 ` 2xxu1, v1y " 2pv1 ` xxu1, v1yq, which is absurd. Thus y " 1, so that replacing v1 by w we have H " Zu1 ` Zw. In this case, the other primitive effective isotropic vector must satisfy (79) Pairing (79) with u1 gives xu1, u2y " xu1, wy2. To see that ℓpu1q " ℓpu2q, observe that if ℓpuiq " 2 for say i " 1, then 2 c1pu1q implies that 2 xu1, wy as rk w " rk u1 " 0 pmod 2q. Thus u1 " u2 pmod 2q so that ℓpu2q " 2 as well. Otherwise, ℓpu1q " ℓpu2q " 1, and we have equality again. Finally, note that for any v " xu1 ` yw with x, y P Z, we have u2 " u1 ` xu1, wyw. v2 " 2xyxu1, wy ` y2w2 " 2xyxu1, wy ´ 2y2 is even, as claimed. Observe that Cases (2) and (3) are indeed mutually exclusive since H is an odd lattice in the first case and an even lattice in the latter. For the statements about CW and moduli spaces in Cases (2) and (3), observe that CW " PH ` Rě0w and xu2, wy " ´xu1, wy in either case. So, up to reordering, we may suppose that xu1, wy ą 0 and xu2, wy ă 0. In particular, u1 is an extremal ray of CW (see Figure 2), and W is totally semistable for u2 by Lemma 6.1. Thus M s (cid:3) σ0pu1q " Mσ0pu1q, as claimed. Remark 8.2. It is worth noting that if H contains a vector v such that v2 is odd, then ℓpu1q " ℓpu2q. Indeed, as above, we may complete u1 to a basis so that H " Zu1 ` Zv1, and as v2 is odd, we must have v12 is odd as well. From (77), we see immediately that in writing u2 " xu1 ` yv1 with gcdpx, yq " 1, we must have y even and x odd. But then c1pu1q " c1pu2q pmod 2q so that ℓpu1q " ℓpu2q, as claimed. Remark 8.3. We note that ℓpu1q " 2 and 2 rkpu2q force xu1, u2y to be even. Let us be more specific each case. In Case (1), we must have xu1, u2y ě 4 if ℓpu2q " 2. Indeed, if ℓpu2q " 2, then u1´ u2 is divisible by 2 in HalgpX, Zq, and thus in the saturated sublattice H as well, so xu1, u2y " 2 would imply MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 41 that H contains the exceptional class u1´u2 be even by Remark 8.2. , an impossibility. If, instead, ℓpu2q " 1, then H must In Case (3), when there exists an effective spherical class, we also have xu1, u2y ě 4 since then In Case (2), we have xu1, wy is odd so that 4 ∤ xu1, u2y " 2xu1, wy2. Indeed, writing u1 " xu1, wy is even so that xu1, u2y " xu1, wy2 is divisible by 4. p2r1, 2c1, s1q and w " pr, c, s 2q with r, s odd integers, we observe that 2 xu1, wy " 2c1.c ´ rs1 ´ r1s " ´pr1 ` s1q " 1 pmod 2q by Lemma 3.2. Remark 8.4. Up to relabeling σ` and σ´, we may assume that the orientation on HR is as in Figure 2, even in Case (1). That is, we will assume for the remainder of this section that φ`pu1q ă φ`pu2q ă φ`pwq. Our main result about isotropic walls is the following classification: Proposition 8.5. Assume that HW is isotropic and v P HW with v2 ą 0. Set r :" rk v. (1) If W is totally semistable, that is, M s σ0pv, Lq " H, then (a) HW contains an effective exceptional or spherical class w such that xv, wy ă 0; or (b) HW contains a primitive isotropic u such that ℓpuq " 2 and xv, uy " 1; or (c) HW contains a primitive isotropic u and an effective spherical class w such that 2 KX pmod 2q, where D is a nodal cycle; or (d) HW contains a primitive isotropic u and an effective exceptional class w such that xv, uy " ℓpuq " 2, xv, wy " 0, L " D ` r xv, uy " ℓpuq " 2, xv, wy " 0, and L " KX pmod 2q; or xv, uy " 1 " ℓpuq, xv, wy " 0, and L " D ` r cycle. (e) HW contains a primitive isotropic u and an effective spherical class w such that 2 KX pmod 2q, where D is a nodal (2) W induces a divisorial contraction if and only if (a) HW contains a primitive isotropic u and an effective exceptional class w such that (b) HW contains a primitive isotropic u and an effective spherical class w such that Remark 8.6. Let us make a few comment about irreducible components and determinants. σ0pv, Lq " Mσ0pv, Lq. (1) In Case (1)(c), then v " w ` u and v2 " 2. In Case (1)(d), v " 2pw ` uq. 2 KX ` KX pmod 2q, then M s If L " KX pmod 2q, then a connected component of Mσ`pv, Lq has two irreducible components (Proposition 13.10) and W is totally semistable for one of them. If L " 0 pmod 2q, then Mσ`pv, LqzM s σ0pv, Lq is a divisor. In Case (1)(e), we also see that v " w ` 2u. By Proposition 13.13, Mσpvq has two irreducible components and each component becomes totally semistable at walls W (of type (1)(c)) and W1 (of type (1)(e)). If L " D ` r (2) In Case (2)(d) with ℓpuq " 1, assume that u " u1. Then we can show v " v2 2 u1 ` u2 with xu1, u2y " 1. In particular ℓpu1q " ℓpu2q " 1. We will prove Proposition 8.5 in the next two subsections. For clarity of discussion, in the next section we will first tackle the case that v is minimal, that is, xv, wy ě 0 for the unique exceptional/spherical class w P HW. Then we will show in Section 8.3 that if xv, wy ă 0, we may apply the spherical or weakly-spherical twist associated to T of class w to reduce the wall-crossing behavior to that of a minimal v. xv, uy " 1, ℓpuq " 2, and xv, wy ‰ 0; or xv, wy " 0 and xv, uy ą ℓpuq; or w such that xv, uy " ℓpuq and xv, wy ‰ 0; or and v2 ě ℓpuq ` 2. (c) HW contains a primitive isotropic u and an effective exceptional or spherical class (d) HW satisfies CW " PH, contains a primitive isotropic u such that xv, uy " ℓpuq, 42 HOWARD NUER AND K ¯OTA YOSHIOKA 8.2. Minimal Mukai vectors. In this section we assume that v is minimal, i.e. that xv, wy ě 0 for the (unique) spherical or exceptional effective class w, if it exists. When such w exists, let us denote by T the spherical or exceptional σ0-stable object of class w. We divide our analysis into two based on whether ℓpuiq " 2 for some i " 1, 2 or not. We begin with the case where ℓpuiq " 2 for some i. 8.2.1. ℓpuiq " 2 for some i. By Proposition 8.1 above, we may assume that i " 1 so that M s σ0pu1q " Mσ0pu1q and ℓpu1q " 2 implies that Mσ0pu1q -- X by [40, Lemma 9.3]. Using the Fourier-Mukai transform with kernel given by the universal family of Mσ0pu1q, Φ : DbpXq -- DbpXq, we get Φpu1q " p0, 0, 1q. By construction of Stab:pXq, skyscraper sheaves of points on X are Φpσ0q-stable. By Bridgeland's Theorem 3.4, there exist divisor classes ω, β P NSpXqQ, with ω ample, such that up to the GL2pRq-action, Φpσ0q " σω,β. In particular, the category Pω,βp1q is the extension-closure of skyscraper sheaves of points and the shifts, Fr1s, of µω-stable torsion-free sheaves F with slope µωpFq " ω β. Since σ0 by assumption does not lie on any other wall with respect to v, the divisor ω is generic with respect to Φpvq. Under these identifications, we have the following result whose proof is identical to that of [36, Theorem 3.2.7], [6, Proposition 8.2], and [32, Section 5]. Proposition 8.7. An object E of class v is σ`-stable if and only if ΦpEq is the shift Fr1s of a pβ, ωq-Gieseker stable sheaf F on X; therefore r´1sΦ induces the following identification of moduli spaces: Moreover, the contraction morphism π` induced by the wall W is the Li-Gieseker-Uhlenbeck (LGU) morphism to the Uhlenbeck compactification. Mσ`pvq " M β ωp´Φpvqq. Similarly, an object F of class v is σ´-stable if and only if it is the shift F _r1s of the derived dual of a p´β, ωq-Gieseker stable sheaf on X. It follows from the above description that a σ`-stable object E becomes σ0-semistable if and only if F " ΦpEqr´1s is not locally free or if F is not µ-stable, as these are the sheaves contracted by the Uhlenbeck contraction. Proposition 8.8. Assume that HW contains a primitive isotropic vector u with ℓpuq " 2. Suppose that v P HW satisfies v2 ą 0 and xv, wy ě 0 for the (unique) effective spherical or exceptional class w (if it exists). (1) If W is totally semistable for Mσ`pvq, then either (a) xv, uy " 1 for primitive isotropic u P H with ℓpuq " 2; or (b) xv, uy " 2 for primitive isotropic u P H with ℓpuq " 2 and xv, wy " 0 for the unique spherical or exceptional class w. (2) If codimpMσ`pvqzM s σ0pvqq " 1, then either (a) xv, uy " 2 " ℓpuq for primitive isotropic u P H; or (b) xv, wy " 0 for the unique spherical or exceptional class w; or (3) In all other cases, codim Mσ`pvqzM s σ0pvq ě 2. Proof. We assume that u1, u2, and w are labelled and oriented in accordance with Remark 8.4 and the discussion preceeding Proposition 8.7. In particular, we assume that ℓpu1q " 2 and xu1, wy ą 0. For a given E P Mσ`pvq, let the Harder-Narasimhan filtration of E with respect to σ´ correspond to a decomposition v " ři ai. Using Proposition 4.3, we shall estimate the codimension of the sublocus Fpa0, . . . , anqo of destabilized objects which is equal to (80) ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 43 (I) We first assume that one of the ai satisfies a2 i ă 0, say a0 " b0w for an effective spherical or exceptional class w. Then we are Cases (2) or (3) in Proposition 8.1 so that ℓpu1q " ℓpu2q " 2. Assume that a1 and a2 are isotropic. We may set a1 " b1u1 and a2 " b2u2. Then (81) ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy 0 ´ dim Mσ´pa0qq `ÿiě1 ěpa2 " ´ dim Mσ´pa0q ` b0xw, vy ´ b1 ´ b2 ` b1b2xu1, u2y ě ´ dim Mσ´pa0q ´ b1 ´ b2 ` b1b2xu1, u2y, b0xw, aiy ´ b1 ´ b2 ` b1b2xu1, u2y First suppose that w2 " ´2. Then we note that dim Mσ´pa0q " ´b2 where the first inequality follows from Proposition 3.10 and the second inequality follows from the assumption that xv, wy ě 0. 0 and xu1, wy is even because ℓpu1q " 2 and w2 " ´2 is even. From Proposition 3.10 we also have xu1, u2y " xu1, wy2 ě 4. Thus (82) So we must have codim Fpa0, . . . , anqo ě 3 in this case. 0 ` 4b1b2 ´ b1 ´ b2 ě 1 ` 2b1b2 ` b1pb2 ´ 1q ` b2pb1 ´ 1q ě 3. codim Fpa0, . . . , anqo ě b2 If instead w2 " ´1, then dim Mσ´pa0q "Y´ b2 are odd. Thus 2]. Moreover, by Remark 8.3 xu1, wy and xu1,u2y 2 0 (83) codim Fpa0, . . . , anqo ě b2 0 2 ` 2b1b2 ´ b1 ´ b2 ě 1 2 ` b1pb2 ´ 1q ` b2pb1 ´ 1q ě 1 2 . Thus codim Fpa0, . . . , anqo ě 1 with equality only if b0 " b1 " b2 " 1. But then v " w ` u1 ` u2 so that xv, wy " w2 " ´1 ă 0, contrary to the assumption. Thus codim Fpa0, . . . , anqo ě 2 in this case. Now assume that a1 " b1uj and a2 i ą 0 for i ą 1. Then i ´ dim Mσ´paiqq `ÿiăj ÿi pa2 xai, ajy ě a2 (84) 0 ´ dim Mσ´pa0q `ÿiě1 b0xw, aiy ´ b1 `ÿiě2 ě ´ dim Mσ´pa0q ` b0xw, vy ´ b1 ` b1xuj, a2y ě ´ dim Mσ´pa0q ` b1pxuj , a2y ´ 1q. b1xuj , aiy If w2 " ´2, then H is even by Proposition 8.1, so ℓpujq " 2 implies that 2 xuj , a2y. Thus If w2 " ´1, then codim Fpa0, . . . , anqo ě ´ dim Mσ´pa0q ` b1 " b2 codim Fpa0, . . . , anqo ě ´ dim Mσ´pa0q " ´Z´ 0 ` b1 ě 2. b2 0 2^ ě 1, with equality in the last inequality only if b0 " 1, xv, wy " 0 and xuj, a2y " 1. But 1 ď xv, ujy " xw, ujy ` xa2, ujy " xw, ujy ` 1, which forces j " 1. Writing a2 " xu1 ` yw, we see that 1 " xu1, a2y forces y " 1 and xu1, wy " 1. Thus, in addition to xv, wy " 0, we also have xv, u1y " 2. So v falls into both types (2)(a) and (2)(b). Otherwise, we get codim Fpa0, . . . , anqo ě 2. We can now assume that there are no positive classes in the Harder-Narasimhan factors, i.e. v " b0w ` b1uj. But v2 ą 0 forces j " 1 (see Figure 2), so we may assume this outright. Then 0 ď xv, wy " b0w2 ` b1xu1, wy, so our estimate becomes 44 HOWARD NUER AND K ¯OTA YOSHIOKA (85) i ´ dim Mσ´paiqq `ÿiăj codim Fpa0, a1qo "ÿi pa2 "b2 0w2 ´ dim Mσ´pa0q ´ b1 ` b0b1xw, u1y b1 0w2 ´ dim Mσ´pa0q ` ěb2 2 pb0xw, u1y ´ 2q ` b1 ěb2 0w2 ´ dim Mσ´pa0q ` ` 2 pxv, u1y ´ 2q . b0b1xw, u1y b0b1xw, u1y xai, ajy 2 2 If w2 " ´2, then b2 0w2 ´ dim Mσ´pa0q " ´b2 that 2 xv, u1y. The last line of (85) then becomes 0, and H is even so that again ℓpu1q " 2 implies codim Fpa0, a1qo ě ´b2 0 ` b0b1xw, u1y 2 b1 2 pxv, u1y ´ 2q " b0 2 xv, wy ` b1 2 pxv, u1y ´ 2q ě 0, ` with equality only if xv, wy " 0 and xv, u1y " 2, as in Case (1)(b). Moreover, codim Fpa0, a1qo ě 2 unless xw, u1y " 2 and v " w ` 2u1. In this case we have codim Fpa0, a1qo " 1 as in Case (2)(a). 0w2 ´ dim Mσ´pa0q "Q´ b2 2U, so the last line of (85) gives 0 If w2 " ´1, then b2 (86) codim Fpa0, a1qo ě ´ b1 2 pxv, u1y ´ 2q " unless xv, u1y " 1, in which case b0 " 1 " xw, u1y. But then b2 b0b1xw, u1y ` b2 0 2 ` 2 b0 2 xv, wy ` b1 2 pxv, u1y ´ 2q ě 0 0w2 ´ dim Mσ´pa0q " 0, so codim Fpa0, a1qo ě b1 2 ´ b1 2 " 0, and indeed codim Fpa0, a1qo " 0 when xv, u1y " 1 as in Case (1)(a). We get equality in (86) only if xv, wy " 0 and xv, u1y " 2. We can derive from these two equations that xw, u1y " 1 and v " 2w ` 2u1, in which case indeed codim Fpa0, a1qo " 0 as in Case (1)(b). Moreover, codim Fpa0, a1qo ě 2 unless b0 2 pxv, u1y´ 2q " 1; or 2 xv, wy " 1 2 and xv, u1y " 2. In the latter two cases, however, we would have xw, u1y " 2, which is impossible, as noted in Remark 8.3. Thus we are left with the first case, in which v " 2w ` 3u1 and xw, u1y " 1. Notice that xv, u1y " 2 so that we are in Case (2)(a). 2 xv, wy " 1 and xv, u1y " 2; or xv, wy " 0 and b1 b0 Finally, assume that other than a0 " b0w, a2 i ą 0 for all i ą 0. Then the estimate becomes codim Fpa0, . . . , anqo "ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy " ´ dim Mσ´pa0q ` b0xw, vy ` ÿ0ăiăjxai, ajy ě b2 0 2 ą 0. Moreover, codim Fpa0, . . . , anqo ě 2 unless xw, vy " 0 and v " w` a1, in which case Fpa0, a1qo has 1 " pv ´ wq2 " v2 ` w2, codimension one as in Case (2)(b). Note that in this case we require 0 ă a2 so v2 ą 2 or v2 ą 1 if w2 " ´2 or w2 " ´1, respectively. (II) We next assume that a2 Suppose first that a1 " b1u1 and a2 " b2u2. Then we can be in any case of Proposition 8.1, and i ě 0 for all i. our estimate now becomes (87) (88) codim Fpa1, . . . , anqo "ÿi i ´ dim Mσ´paiqq `ÿiăj pa2 ^ ` b1b2xu1, u2y ě ´ b1 ´Z b2ℓpu2q 2 xai, ajy ěb1pb2 ´ 1q ` b2pb1 ´ 1q ě 0, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 45 with equality only if b1 " b2 " 1, ℓpu2q " 2, and xu1, u2y " 2, so that H falls into Case (2) of In particular, we have xv, u1y " 2 and xv, wy " xu1, wy ` xu2, wy " 0, as in Proposition 8.1. Case (1)(b). Furthermore, by Remark 8.3 we have codim Fpa1, . . . , anqo " 1 only if H falls into Case (2) of Proposition 8.1 and v " 2u1`u2 or u1`2u2 with xu1, u2y " 2, so that xv, uy " 2 " ℓpuq as in Case (2)(a), or H falls into Case (1) of Proposition 8.1 and v " u1 ` u2 with ℓpu2q " 1 and xu1, u2y " 2, so that xv, u1y " 2, as in Case (2)(a). Otherwise, codim Fpa1, . . . , anqo ě 2. Now we assume that a1 " b1uj and a2 i ą 0 for i ě 2. In this case, we also see that (89) codim Fpa1, . . . , anqo "ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy b1xuj, aiy ` ÿ1ăiăk " ´Z b1ℓpujq ^ `ÿią1 ěb1pxv, ujy ´ 1q ` ÿ1ăiăk xai, aky 2 ěb1pxv, ujy ´ 1q ě 0. xai, aky Thus codim Fpa1, . . . , anqo " 0 only if xv, ujy " 1, ℓpujq " 2, and v " b1uj ` a2, as in Case (1)(a). Similarly, codim Fpa1, . . . , anqo " 1 only if either v " uj ` a2 with xv, ujy " 2 and ℓpujq " 2 as in Case (2)(a), or v " u2 ` a2, 2u2 ` a2 with xv, u2y " 1 " ℓpu2q. But this latter case is impossible. Indeed, H must be even by Remark 8.2 and ℓpu1q ‰ ℓpu2q. But then we see that we can write u1 " ´ 2 u2 ` a2 so that xu1, u2y " 1, which is impossible as ℓpu1q " 2. a2 2 i ą 0 for all i, then codim Fpa1, . . . , anqo ě 2 by Proposition 5.9. Finally, if a2 (cid:3) Remark 8.9. Proposition 8.8 can be proven by using the Fourier-Mukai transform Φ to translate the problem to the equivalent problem of determining the codimensions of the strictly µ-semistable locus and the non-locally free locus. One could then use the estimates in [53] to get the result. However, there is a small error in Case C there, which misses the spherical case of Case (1)(b), so we use the method above. We explore this example more fully in Proposition 13.13. Remark 8.10. The reader may notice that when xv, wy " 0 and xv, uy " 2 " ℓpuq, we simultane- ously claim that codimpMσ`pvqzM s σ0pvqq is both zero and one. Indeed, we shall prove in Section 13 that for one choice of the determinant L, Mσ`pv, Lq contains a connected component with two ir- reducible components, M0 and M1. For M1, W is a totally semistable wall inducing a P1-fibration over the singular locus, where it meets M0. The strictly σ0 locus on M0 is this singular locus, which is a divisor. When w2 " ´2, this describes all of Mσ`pv, Lq as it is connected. For the other determinant, L ` KX , the strictly σ0-semistable locus is a divisor, if non-empty. Now we demonstrate the converse of 8.8 in the following sequence of lemmas. Furthermore, we determine precisely when the divisor in Case (2) gets contracted. We make free use of the Fourier-Mukai transform Φ to translate the problem to that of moduli of sheaves. Lemma 8.11. Assume that HW contains a primitive isotropic vector u such that ℓpuq " 2 and that xv, uy " 1 for v minimal. Then v2 is odd and HW contains an exceptional class. Moreover, W is totally semistable and, if v2 ą 1, induces a divisorial contraction. Proof. Let us first prove that v2 is odd and H contains an exceptional class. Write v " pr, c, s 2q and u " p2r1, 2c1, s1q, with r1 ` s1 odd, as ℓpuq " 2. Then as r " s pmod 2q, it follows that 1 " xv, uy " 2c.c1 ´ r1s ´ rs1 " ´rpr1 ` s1q " r from which we see that r is odd, or equivalently v2 is odd. Thus v´ v2`1 class. pmod 2q, 2 u P HW is an exceptional 46 HOWARD NUER AND K ¯OTA YOSHIOKA As H falls into Case (2) of Proposition 8.1, we see that ℓpu2q " 2 and (90) We observe from (90) that we may assume that u " u1. Indeed, if xv, u2y " 1, then pairing (90) with v, we get u2 " u1 ` 2xu1, wyw. 1 " xv, u2y " xv, u1y ` 2xu1, wyxv, wy ě xv, u1y ą 0, so that xv, u1y " 1 as well. It follows that Mσ`pvq -- M β ωp´Φpvqq is isomorphic to the Hilbert scheme of points. Finally, W is the Hilbert-Chow wall inducing the Hilbert-Chow morphism HilbnpXq Ñ SymnpXq, which is a divisorial contraction for 1 ă n " v2`1 , and every ideal sheaf is strictly semistable as in [40, Proposition 13.1]. (cid:3) 2 Although the behavior in Lemma 8.11 is analogous with the corresponding case on K3 surfaces, we see some new behavior in the next two lemmas. We begin with the exceptional case of Case (1)(b) in Proposition 8.8. Lemma 8.12. Suppose that HW contains a primitive isotropic u such that xv, uy " 2 " ℓpuq and xv, wy " 0 for an exceptional class w P CW X HW . Then W is totally semistable for, and induces a P1-fibration on, precisely one component of Mσ`pv, 2L1 ` KXq. Proof. As xv, wy " 0, we see from (90) that xv, u2y " xv, u1y, so we may assume that u " u1. After applying r´1sΦ, and possibly tensoring by a line bundle, we may assume that v " p2, 0,´1q, w " p1, 0, 1 2q, and Mσ`pvq is isomorphic to MHp´Φpvqq, the moduli space of H-Gieseker semistable sheaves of Mukai vector ´Φpvq with respect to a generic polarization H. The contraction π` is then the LGU-contraction morphism as in Proposition 8.7. As mentioned in [53, Remark 2.3] and proven in Section 13 (see Proposition 13.10), there is precisely one component of Mσ`pv, KXq consisting of stable non-locally free sheaves E fitting into the short exact sequence 0 Ñ E Ñ F pφ1,φ2q ÝÝÝÝÑ Cp ' Cq Ñ 0, where F :" OX'OXpKXq and φ1 : OX Ñ Cp'Cq and φ2 : OXpKXq Ñ Cp'Cq are both surjective. The polystable object in the same S-equivalence class as E with respect to σ0 is pCp ' Cqqr´1s ' F , and the set of distinct σ` stable objects in the same S-equivalence class is parametrized by P HompF, Cpq P HompF, Cqq{pAutpFq{Cq, which is a curve birational to P1. Thus W is totally semistable for this component and induces a P1-fibration. (cid:3) We get similar behavior in the spherical case: Lemma 8.13. Suppose that HW contains a primitive isotropic u such that xv, uy " 2 " ℓpuq and xv, wy " 0 for a spherical class w P CW X HW. Then W is totally semistable for, and induces a P1-fibration on, precisely one component of Mσ`pv, Lq with L " Z ` rk v 2 KX mod 2, where Z is a nodal cycle. Moreover Mσ`pv, L ` KXq " M s Proof. As in the proof of Lemma 8.12, xv, wy " 0 means that xv, u2y " xv, u1y, so we may assume that u " u1, and applying r´1s Φ, we may assume that v " p2, D, sq with gcdp2, Dq " 1 and s P Z and w " p2, D, D2`2 q, where 4 D2 ` 2 since D " c1pvq " Z pmod 2q for a nodal cycle Z. Then Mσ`pvq is isomorphic to MHp2, D, sq. Let T be the stable spherical bundle with vpTq " w. Then we have a family of non-locally free sheaves E parameterized by a P1-bundle over X: σ0pv, L ` KXq. 4 0 Ñ E Ñ T Ñ Cp Ñ 0. This P1-bundle is one component of Mσ`pv, Lq, and each such E is strictly σ0-semistable, S- equivalent to T ' Cpr´1s. As detpTq " Z ` rkpvq2 KX pmod 2q and detpCpq " 0 so that detpEq " Z ` rkpvq2 KX pmod 2q, we get the first claim. For the other determinant, notice that as gcdp2, Dq " 1, all stable sheaves are µ-stable, so for any non-locally free E P MHpp2, D, sq, L ` KXq, E__ would be a µ-stable locally free sheaf in (91) v2 ě 4, v2 ě 3, v2 ‰ 4, v2 ą 2, $'&'% in Case (1) of Proposition 8.1; in Case (2) of Proposition 8.1; in Case (3) of Proposition 8.1. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 47 MHpp2, D, D2`2 sheaf so that Mσ`pv, L ` KXq " M s 4 q, L ` KXq " H. Thus every E P MHpp2, D, sq, L ` KXq is a µ-stable locally free (cid:3) σ0pv, L ` KXq, as claimed. Having considered the behavior of totally semistable walls, we move on to determining when the codimension one strictly σ0-semistable locus gets contracted. We begin with Case (2)(a): Lemma 8.14. Suppose that xv, u1y " 2 " ℓpu1q for v minimal. Assume further that Then W is not a totally semistable wall and induces a divisorial contraction on Mσ`pv, Lq. Proof. In all cases, we will show that our assumptions imply that W is not totally semistable for σ0pv´u1, L1q Mσ`pv´u1, L1q, Mσ`pv´u1, L1`KXq. Assuming we have done this, then taking F P M s σ0pv ´ u1, L1 ` KXq) and G P Mσ0pu1q, we get a P1 worth of distinct extensions (or M s 0 Ñ G Ñ E Ñ F Ñ 0 of objects in Mσ`pv, Lq (or Mσ`pv, L ` KXq, respectively) that are S-equivalent with respect to σ0. A quick dimension count shows that these sweep out a divisor. Now we show that W is not totally semistable for Mσ`pv´ u1, L1q, Mσ`pv´ u1, L1` KXq. Under the assumptions in (91), we may have pv ´ u1q2 " ´1 in Case (2) of Proposition 8.1. But then v´u1 " w and xu1, wy " 2, which is impossible by Remark 8.3. Otherwise, pv´u1q2 " v2´4 ě 0, with equality only if v " u1 ` ku2. In this case, xv, u1y " 2 implies that k " 1, xu1, u2y " 2, ℓpu2q " 1, and H falls into Case (1) by Remark 8.3 and (91). So v ´ u1 " u2, and M s σ0pu2q " Mσ0pu2q so that W is not totally semistable for Mσ`pu2q (or a wall at all). It remains to show that W is not totally semistable when pv ´ u1q2 ą 0. As both conditions (1)(a) and (1)(b) of Proposition 8.8 for W being totally semistable require the existence of a spher- ical/exceptional class (see Lemma 8.11 for Case (1)(a)), W is automatically not totally semistable in Case (1) of Proposition 8.1. In Cases (2) and (3), if we write v " xu1 ` yw with x, y P Zě0, then by Remark 8.3, xv, u1y " 2 and v2 ą 4 is equivalent to x ą 2, y " 2 and xu1, wy " 1 in Case (2), while in Case (3), it is equivalent to x ą 1, y " 1 and xu1, wy " 2. Thus xv ´ u1, wy ě 0, and in case of equality, we may choose F above to be in the component of Mσ`pv ´ u1, L1q (or Mσ`pv´ u1, L1 ` KXq) that contains σ0-stable objects, as guaranteed by Lemmas 8.12 and 8.13. If xv´u1, wy ą 0, then xv´u1, u2y ą xv´u1, u1y " 2, so W is not totally semistable for Mσ`pv´u1q by Proposition 8.8, as claimed. (cid:3) Indeed, suppose xv, u1y " 2 and 0 ă v2 ă 4. It is worth noting that the possibilities excluded by the condition (91) have either been dealt If v2 " 3, then with already, or are irrelevant. pv ´ u1q2 " ´1, so we must be in Case (2) of Proposition 8.1, which was included in (91). If v2 " 2, then pv ´ u1q2 " ´2, so in Case (3) we have already seen (and will prove in Section 13) that the divisorial component of the strictly σ0-semistable locus is not contracted, while in Case (1) we see that W is not a wall for v. This is also the case if v2 " 1. Finally, if v2 " 4 in Case (2), then again we have already seen (and will prove in Section 13) that the divisorial component of the strictly σ0-semistable locus is not contracted. Now we consider Case (2)(b) of Proposition 8.8: Lemma 8.15. Suppose that HW contains a primitive isotropic vector u with ℓpuq " 2 and v2 ą 2. If xv, wy " 0 for an effective spherical class w, then W induces a divisorial contraction. Proof. Let T be the unique σ0-stable spherical object of class w. Consider a :" v ´ w. Then a2 " v2 ´ 2 ą 0 and xa, wy " 2. By Proposition 8.1 we see that H is even, and from ℓpuq " 2 48 HOWARD NUER AND K ¯OTA YOSHIOKA it follows that ℓpu1q " ℓpu2q " 2. Thus xa, uiy is even and at least 2, by Remark 8.3. By σ0paq, we see that the P1's of S-equivalent Proposition 8.8, M s extensions σ0paq ‰ ∅, so letting A vary in M s sweep out a contracted divisor in Mσ`pvq. 0 Ñ A Ñ E Ñ T Ñ 0 (cid:3) With the exact same proof, one can show that the analogous situation for w exceptional results in a divisor of strictly σ0-semistable objects that does not get contracted. Lemma 8.16. Suppose that HW contains a primitive isotropic vector u with ℓpuq " 2 and v2 ‰ 1, 4. If xv, wy " 0 for an effective exceptional class w, then the locus Mσ`pvqzM s σ0pvq is a divisor that is not contracted. Lemma 8.17. Let W be a potential wall and v minimal, and suppose that HW contains a primitive isotropic class u such that ℓpuq " 2. Assume that either xv, uy " 1, or xv, uy " 2 and (92) v2 ě 4, v2 ě 3, v2 ‰ 4, v2 ą 2, $'&'% in Case (1) of Proposition 8.1; in Case (2) of Proposition 8.1; in Case (3) of Proposition 8.1. Then W induces a divisorial contraction on Mσ`pv, Lq. Proof. The class u is automatically effective. By Lemmas 8.11 and 8.14, the only remaining case is u " u2 and xv, u2y " 2. First, suppose that HW admits an effective exceptional or spherical class w. Then from the minimality of v and the assumptions in (92), we have xv, wy ą 0. Indeed, we may write v " xu2 ` yw with x, y P Z by Proposition 8.1, so the conditions xv, u2y " 2 and xv, wy " 0 imply that v2 " 4 or 2 if w is exceptional or spherical, respectively, contrary to (92). Thus by (90) (and the analogue for spherical w), we must have 0 ă xu1, vy ă xv, u2y " 2, so xv, u1y " 1 and the result follows from Lemma 8.11. Now suppose that HW admits no effective spherical or exceptional classes. Then the proof of (cid:3) Lemma 8.14 applies with u2 instead, giving the result. 8.2.2. ℓpu1q " ℓpu2q " 1. We begin again by determining necessary conditions for a potential wall W to be totally semistable and for codimpMσ`pvqzM s Proposition 8.18. Let W be a potential wall for positive and minimal v such that HW is isotropic with ℓpu1q " ℓpu2q " 1. σ0pvqq " 1. (1) If W is totally semistable, then xv, wy " 0 and xv, u1y " 1, where w is an effective spherical class in HW. (2) If codimpMσ0pvqzM s σ0pvqq " 1, then (a) xv, uy " 1 for a primitive isotropic u P HW, or (b) xv, wy " 0 for a spherical or exceptional class w. Proof. Again we assume that u1, u2, and w are labelled and oriented in accordance with Remark 8.3 and the discussion preceeding Proposition 8.8 so that xu1, wy ą 0. For a given E P Mσ`pvq, let the Harder-Narasimhan filtration of E with respect to σ´ correspond to a decomposition v "ři ai. We shall estimate the codimension of the sublocus Fpa0, . . . , anqo of destabilized objects which is equal to (93) ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy. We will divide the discussion into two cases as we did for Proposition 8.8, depending on whether or not one of the classes has negative square. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 49 (I) We first assume that one of the ai satisfies a2 i ă 0, say a0 " b0w for an effective spherical or exceptional class w. Assume that a1 " b1u1 and a2 " b2u2 are isotropic. Then (94) ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy 0 ´ dim Mσ´pa0qq `ÿiě1 ěpa2 ě ´ dim Mσ´pa0q ` b0xw, vy ´Z b1 2^ ´Z b2 ě ´ dim Mσ´pa0q ´Z b1 b0xw, aiy ´Z b1 2^ ´Z b2 2^ ` b1b2xu1, u2y, 2^ ` b1b2xu1, u2y 2^ ´Z b2 2^ ` b1b2xu1, u2y where b0xv, wy ě 0 from the minimality of v. First suppose that w2 " ´2, so dim Mσ´pa0q " ´b2 b1 2 ´ 0 ` b1b2 ´ (95) codim Fpa0, . . . , anqo ě b2 " 1 ` b2 2 ě 1 ` b1b2 ´ ě 1. 2 b2pb1 ´ 1q ` b1pb2 ´ 1q b1 2 ´ b2 2 0 and thus Thus if codim Fpa0, . . . , anqo " 1 then we must have xv, wy " 0, xu1, u2y " 1, and v " w` u1` u2. But then xv, wy " ´2 ă 0, contrary to assumption. So we must have codim Fpa0, . . . , anq0 ě 2 in this case. If instead w2 " ´1, then dim Mσ´pa0q "Y´ b2 2] and 2 xu1, u2y by part (2) of Proposition 8.1, so 0 (96) codim Fpa0, . . . , anqo ě " so codim Fpa0, . . . , anqo ě 2 in this case. Now assume that a1 " b1uj and a2 (97) ÿi pa2 i ´ dim Mσ´paiqq `ÿiăjxai, ajy ě a2 b1 2 ´ b2 2 ě b2 0 2 ` 2b1b2 ´ 1 2 ` b1p2b2 ´ 1q ` b2p2b1 ´ 1q 2 b1 2 ´ b2 2 1 2 ` 2b1b2 ´ 3 , 2 ě i ą 0 for i ą 1. Then 0 ´ dim Mσ´pa0q `ÿiě1 ě ´ dim Mσ´pa0q ` b0xw, vy ´Z b1 b0xw, aiy ´Z b1 2^ `ÿiě2 2^ ` b1xuj , a2y ě ´ dim Mσ´pa0q ` b1 2 . b1xuj, aiy If w2 " ´2 then while if w2 " ´1, then ´ dim Mσ´pa0q ` b1 2 " b2 0 ` b1 2 ě 3 2 , codim Fpa0, . . . , anqo ě ´ dim Mσ´pa0q ` b1 2 ě b2 0 2 ` b1 2 ě 1, with equality in the last inequality only if b1 " b0 " 1, in which case the first inequality is strict. So we always have codim Fpa0, . . . , anqo ě 2 in this case. 50 HOWARD NUER AND K ¯OTA YOSHIOKA We can now assume that there are no positive classes amongst the Harder-Narasimhan factors, i.e. v " b0w ` b1uj. But v2 ą 0 forces j " 1, so we may assume this outright. Then 0 ď xv, wy " b0w2 ` b1xu1, wy, so our estimate becomes (98) codim Fpa0, a1qo "ÿi i ´ dim Mσ´paiqq `ÿiăj pa2 0w2 ´ dim Mσ´pa0q ´Z b1 "b2 0w2 ´ dim Mσ´pa0q ` ěb2 xai, ajy 2^ ` b0b1xw, u1y b0b1xw, u1y 2 b1 2 pb0xw, u1y ´ 1q . ` If w2 " ´2, then b2 0w2 ´ dim Mσ´pa0q " ´b2 b0b1xw, u1y b1 2 pb0xw, u1y ´ 1q " ` 2 ´b2 0 ` 0, so the last line of (98) becomes b0 2 xv, wy ` b1 2 pxv, u1y ´ 1q ą 0 unless xv, wy " 0 and xv, u1y " 1. But then v " w ` 2u1 and xw, u1y " 1, in which case indeed codim Fpa0, a1qo " 0 as in Case (1). Moreover, codim Fpa0, a1qo ě 2 unless xw, u1y " 2 and v " w ` u1, as in both Cases (2)(a) and (2)(b), or xw, u1y " 1 and v " w ` 3u1 or w ` 4u1, as in Case (2)(a). In each of these cases codim Fpa0, a1qo " 1. 0w2 ´ dim Mσ´pa0q "Q´ b2 2U, so the last line of (98) gives 0 If w2 " ´1, then b2 b2 0 2 ` b0b1xw, u1y codim Fpa0, a1qo ě ´ ` unless again xv, wy " 0 and xv, u1y " 1, In this case, however, codim Fpa0, a1qo " 1 as in both Case (2)(b) and Case (2)(a). Moreover, codim Fpa0, a1qo ě 2 unless xw, u1y " 1 and v " w`2u1, 2pw`u1q, in which case codim Fpa0, a1qo " 1. b1 b1 2 pxv, u1y ´ 1q ą 0 2 pb0xw, u1y ´ 1q " in which case v " w ` u1 and xw, u1y " 1. b0 2 xv, wy ` 2 Finally, assume that other than a0 " b0w, a2 i ą 0 for all i ą 0. Then the estimate becomes (99) i ´ dim Mσ´paiqq `ÿiăj codim Fpa0, . . . , anqo "ÿi pa2 xai, ajy " ´ dim Mσ´pa0q ` b0xw, vy ` ÿ0ăiăj xai, ajy ě b2 0 2 ą 0. Moreover, codim Fpa0, . . . , anqo ě 2 unless xw, vy " 0 and v " w ` a1, as in Case (2)(b). In this case we have codim Fpa0, a1qo " 1, and we require that 0 ă a2 1 " pv ´ wq2 " v2 ` w2, so v2 ą 2 or v2 ą 1 if w2 " ´2 or w2 " ´1, respectively. (II) We next assume that a2 i ě 0 for all i. We assume a1 " b1u1 and a2 " b2u2. Then (100) codim Fpa1, . . . , anqo "ÿi pa2 ě ´Z b1 i ´ dim Mσ´paiqq `ÿiăjxai, ajy 2^ ´Z b2 2^ ` b1b2xu1, u2y b1pb2 ´ 1q ` b2pb1 ´ 1q 2 ě ą 0, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 51 unless v " u1 ` u2 and xu1, u2y " 1, in which case codim Fpa1, a2qo " 1, as in Case (2)(a). If, say, b1 " 1 and b2 ě 2, then we have codim Fpa1, a2qo ě ´Z 1 2^ ´Z b2 2^ ` b2xu1, u2y ě b2 2 ě 1, with equality only if b2 " 2 and xu1, u2y " 1. Thus codim Fpa1, . . . , anqo is always positive in this case. Moreover, Fpa1, . . . , anqo has codimension one only if v " u1 ` u2, 2u1 ` u2, u1 ` 2u2 with xu1, u2y " 1, as in Case (2)(a). Otherwise, codim Fpa1, . . . , anqo ě 2. Now we assume that a1 " b1uj and a2 i ą 0 for i ě 2. In this case, we also see that (101) codim Fpa1, . . . , anqo "ÿi pa2 " ´Z b1 i ´ dim Mσ´paiqq `ÿiăk 2^ `ÿią1 b1xuj , aiy ` ÿ1ăiăk 2q ` ÿ1ăiăk xai, aky 1 xai, aky xai, aky ěb1pxv, ujy ´ ěb1pxv, ujy ´ 1 2q ą 1, unless xv, ujy " 1, and v " b1uj ` a2 with b1 " 1, 2, in which case codim Fpa1, a2qo " 1, as in Case (2)(a). Finally, if a2 i ą 0 for all i, then codim Fpa1, . . . , anqo ě 2 by Proposition 5.9. (cid:3) We prove the converse to Proposition 8.18 in the following lemmas. We begin with the case of a totally semistable wall as in Case (1) of Proposition 8.18. σ0pv, L ` KXq is a divisor which does not get contracted. Lemma 8.19. Suppose that HW contains an effective spherical class w and an isotropic class u such that xv, uy " 1 " ℓpuq and xv, wy " 0. Then W is totally semistable and induces a P1-fibration on Mσ`pv, Lq for L " D ` rk v 2 KX pmod 2q, where D is a nodal cycle. For the other determinant, Mσ`pv, L ` KXqzM s Proof. We first observe that since xv, wy " 0, it follows from the analogue of (90) in the spherical case that xv, u1y " xv, u2y. As ℓpu1q " ℓpu2q, we may assume that u " u1. By Proposition 8.1 we may write v " xw ` yu1 with x, y P Z. Then 1 " xv, u1y " xxw, u1y which implies that σ0p2u1, 2L1q is non-empty x " 1 " xw, u1y. But then xv, wy " 0 forces y " 2. Since ℓpu1q " 1, M s and two-dimensional by Proposition 3.10. Moreover, for the unique σ0-stable spherical object T of σ0p2u1, 2L1q, stability ensures that ext1pT, Aq " x2u1, wy " 2. Then the P1 class w and any A P M s worth of extensions gets contracted by crossing W, and varying A in M s irreducible component of Mσ`pv, Lq, where σ0p2u1, 2L1q generically sweeps out an entire 0 Ñ A Ñ E Ñ T Ñ 0 L " 2L1 ` detpTq " detpTq " D ` rk v 2 KX pmod 2q. For the other determinant, observe that the only decompositions of v into effective classes are v " w ` 2u1 " u1 ` u2, and from the proof of Proposition 8.18 only the former decomposition corresponds to a totally semistable wall. Moreover, in the case of the decomposition v " w ` 2u1 for the determinant L ` KX , the strictly σ0-semistable locus has codimension 1, so W is not totally semistable for Mσ`pv, L ` KXq. Indeed, if E P Mσ`pv, L ` KXq has this decomposition for its Harder-Narasimhan filtration with respect to σ´, then the kernel A of the surjection E ։ T would be in Mσ´p2u1, 2L1 ` KXq. But M s σ´p2u1, 2L1 ` KXq " ∅ by Proposition 3.10, so A would have to be strictly σ´-semistable, and from determinant considerations its Jordan-Holder factors 52 HOWARD NUER AND K ¯OTA YOSHIOKA would have to be A1 P Mσ´pu1, L1q and A2 P Mσ´pu1, L1 ` KXq. But then A1 fl A2, A2pKXq, as detpA1q ‰ detpA2q " detpA2pKXqq, so ext1pA1, A2q " xu1, u1y ` hompA1, A2q ` ext2pA1, A2q " hompA2, A1pKXqq " 0, from which it follows that A " A1' A2. Using [17, Lemmas 6.1-6.3], we thus get a unique σ`-stable extension 0 Ñ A1 ' A2 Ñ E Ñ T Ñ 0, Now consider the other decomposition, v " u1 ` u2, and take A1 P Mσ`pu1, L1q and A2 P which is unique in its S-equivalence class with respect to σ0, and varying the Ai spans a non- contracted divisor of strictly σ0-semistable objects with the prescribed Harder-Narasimhan filtration for σ´. Mσ`pu2q. Then any nontrivial extension (102) is σ`-stable by [6, Lemma 9.3], as the parallelogram spanned by u1 and u2 has no lattices points other than its vertices. In order for detpEq " L ` KX we must have 0 Ñ A1 Ñ E Ñ A2 Ñ 0 detpA2q ` L1 " L ` KX " 2L1 ` detpTq ` KX , so that detpA2q " L1 ` detpTq ` KX . As M s stable factors T and A11 P Mσ`pu1q. It follows that σ0pu2q " ∅, A2 must be strictly σ0-semistable with L1 ` detpTq ` KX " detpA2q " detpTq ` detpA11q, so A11 P Mσ`pu1, L1 ` KXq. In particular, detpA1q ‰ detpA11q, and thus A1 fl A11, A11pKXq. Hence by stability, HompT, A1q " HompA1, Tq " HompA11, A1q " HompA1, A11pKXqq " 0, from which we see that HompA2, A1q " 0 " Ext2pA2, A1q " 0, by applying Homp , A1q to the short exact sequence (102). Thus ext1pA2, A1q " 1, and there exists a unique σ`-stable extension E, which is also unique in its S-equivalence class with respect to σ0. Letting the Ai vary, we again sweep out a divisor that is not contracted by W, as claimed. (cid:3) Now we move on to Case (2)(a). Lemma 8.20. Assume that v is minimal in HW , which contains primitive isotropic classes u1 and u2 such that ℓpu1q " ℓpu2q " 1, and suppose that xv, uy " 1 for a primitive isotropic u P HW with ℓpuq " 1. (1) If v2 ě 3, then W induces a divisorial contraction on Mσ`pv, Lq. (2) If either (a) v2 " 1 or (b) v2 " 2, HW contains a spherical class, and L " D ` p rk v then Mσ`pv, LqzM s σ0pv, Lq is a divisor which is not contracted by W. D is a nodal cycle, 2 ` 1qKX pmod 2q, where Proof. Suppose first that HW contains an effective spherical or exceptional class w and v2 ě 3. Then by minimality of v and (90) (and its analogue in the spherical case), xv, u2y ě xv, u1y with equality only if xv, wy " 0. But if xv, wy " 0, then u could be either u1 or u2, and writing v " xw ` yu with x, y P Z, we see that xv, uy " 1 and xv, wy " 0 imply that v " w ` p´w2qu, where xu, wy " 1. It follows that v2 " ´w2, which were explicitly excluded. Thus xv, wy ą 0 and we see that u " u1. Moreover, we see by the same reasoning that xv, u1y " 1 implies that v " w ` yu1 and xw, u1y " 1 so that v2 " w2 ` 2y " w2 pmod 2q. We will break the proof up into different parts based on the decomposition of v we will use. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 53 Suppose first that v2 ě 3 if w is exceptional and v2 ě 8 if w is spherical. Setting v1 " v ´ 2u1, it follows that v12 " v2 ´ 4 ě ´1 or v12 ě 4 if w is exceptional or spherical, respectively. As xv1, u1y " 1, we have v1 P CW, and from xv1, wy " v2`w2´4 , which is positive if w is spherical, σ0pv1q, we have σ0pv1q ‰ ∅ by Proposition 8.18. Then for E1 P M s σ0p2u1q and E2 P M s we have M s HompE2, E1q " HompE1, E2pKXqq " 0 by stability. It follows that ext1pE2, E1q " xv1, 2u1y " 2, so there is a P1 worth of extensions of the form 2 0 Ñ E1 Ñ E Ñ E2 Ñ 0, If v2 " 6 and L " D ` p rk v which gets contracted by W. Varying E1 and E2 in their moduli sweeps out a divisor in Mσ`pvq. Moreover, as detpE1q " 2L1 by Proposition 3.10, we may choose E2 P M s σ0pv1q to have either determinant L2 or L2 ` KX, where rL2 mod KXs " c1pv1q, to get a divisorial contraction on each Mσ`pv, 2L1 ` L2q and Mσ`pv, 2L1 ` L2 ` KXq. We have proven the first claim of the lemma for w exceptional, and to complete the proof of this claim for w spherical we must consider when v2 " 6, 4. 2 ` 1qKX pmod 2q, then the same argument gives a divisorial contraction on Mσ`pv, Lq, as then v12 " 2 (i.e. σ0pv1, L ´ 2L1q ‰ ∅ by 2 KX pmod 2q, then we will use a different decomposition Lemma 8.19. of v. The conditions xv, u1y " 1 and v2 " 6 force v " w ` 4u1 with xu1, wy " 1, so instead of the above decomposition (v " pv ´ 2u1q ` 2u1), we use a different one, v " w ` v2 where v2 :" 2u1 ` 2u1. Indeed, take the unique σ0-stable object T of class w, and two non-isomorphic objects E1, E2 P M s σ0p2u1, 2L1q. Then by [17, Lemmas 6.1-6.3] the extensions of the form If instead L " D ` rk v xv1, wy " 0) and M s 0 Ñ E1 ' E2 Ñ E Ñ T Ñ 0 σ0p2u1, 2L1qM s are σ`-stable and move in a two-dimensional family contracted to the same point by W. Varying σ0p2u1, 2L1qqz∆, where ∆ is diagonal, sweeps out a contracted divisor pE1, E2q P pM s in Mσ`pv, Lq. If v2 " 4, then the condition xv, u1y " 1 forces v " w ` 3u1. Take the unique σ0-stable object T of class w, let E1 P M s σ0pu1q. We consider extensions of the form σ0p2u1, 2L1q, and E2 P M s 0 Ñ 0 Ñ E1 ' E2 Ñ E Ñ T Ñ 0. These extensions move in a one-dimensional family by [17, Lemma 6.3] and are σ`-stable by [17, Lemma 6.1]. For fixed Ei, this curve of extensions is contracted by W, and varying the Ei sweeps out a divisor. As detpE2q can be either L1 or L1 ` KX , we get a divisorial contraction in each component as before. This concludes the proof of Case (1) for HW that falls into Cases (2) and (3) of Proposition 8.1. For the second claim of the lemma for HW falling into Cases (2) and (3) of Proposition 8.1, we must consider v2 " 1 and v2 " 2, which occur when w is exceptional and spherical, respectively. In the first case, we must have v " w ` u1. Letting F P M s σ0pu1q and T be the unique (up to bOXpKXq) σ0-stable exceptional object of class w, we consider the unique non-trivial extension, 0 Ñ F Ñ E Ñ T Ñ 0. Varying F P M s σ0pu1q, we sweep out a non-contracted divisor in each of Mσ`pv, Lq and Mσ`pv, L` KXq, giving Case (2)(a). The second case has been dealt with in the second statement of Lemma 8.19, giving Case (2)(b). Finally, we suppose that HW contains no effective spherical or exceptional class. We note that since xv, uy " 1, HW must be even. Indeed, it follows from xv, uy " 1 that HW " Zu ` Zv, and if v2 were odd, then v2`1 2 u ´ v would be an exceptional class, so v2 must be even. But then HW must be even as well. We first prove Case (1) in this case. So suppose that v2 ě 3. Then HW being even gives v2 ě 4, and by Remark 8.4 we may write v " v1 ` 2u1 as above to get a divisor swept out by contracted 54 HOWARD NUER AND K ¯OTA YOSHIOKA P1's of extensions 0 Ñ E1 Ñ E Ñ E2 Ñ 0 σ0pv1q. Note that if v2 ą 4, so that v12 ą 0, M s σ0p2u1, 2L1q and E2 P M s σ0pv1q is non- with E1 P M s empty by Proposition 8.18, while if v2 " 4, so that v12 " 0, then v1 " u2 and M s σ0pu2q ‰ ∅ because CW " PH by Proposition 8.1. Either way, we may choose E2 to have the appropriate determinant to give a divisorial contraction in each Mσ`pv, Lq. This completes the proof of Case (1) of the Lemma. The final option to consider is v2 " 2, in which case the only possibility for a destabilizing exact sequence is for Ei P M s σ0puiq, which span a divisor which is not contracted as ext1pE2, E1q " 1. 0 Ñ E1 Ñ E Ñ E2 Ñ 0 Now we prove the converse to Case (2)(b) of Proposition 8.18. (cid:3) Lemma 8.21. Suppose that HW is isotropic with ℓpu1q " ℓpu2q " 1 and xv, wy " 0 for an effective spherical or exceptional class w P HW. (1) If w2 " ´2 and either (2) If either (a) v2 ą 2, or (b) v2 " 2 and xv, u1y ą 1, then W induces a divisorial contraction on Mσ`pv, Lq. (a) w2 " ´2, v2 " 2, xv, u1y " 1, and L " D ` p rk v (b) w2 " ´1, then codimpMσ`pv, LqzM s 2 ` 1qKX pmod 2q, or σ0pv, Lqq " 1 but this divisor is not contracted. Proof. Consider the Mukai vector a :" v ´ w. Then a2 " v2 ` w2 ą w2,xa, vy " v2 ą 0, and xa, wy " ´w2. If a2 ą 0, then since xa, wy ą 0, M s extensions σ0paq ‰ ∅ by Proposition 8.18, and we consider the σ`-stable 0 Ñ F Ñ E Ñ G Ñ 0, σ0paq and G P M s σ0pwq. If w2 " ´2, then G is the unique σ0-stable spherical object where F P M s T of class w, and by stability ext1pT, Fq " xw, ay " 2, so these extensions span a P1 which gets contracted by W. If w2 " ´1, then G is T or TpKXq, and each choice of G gives a unique such non-trivial extension, as ext1pG, Fq " xa, wy " 1 by stability. A dimension count immediately σ0paq spans a divisor which gets contracted if w2 " ´2 but does not if gives that varying F in M s w2 " ´1. Note that we may choose F to have either determinant appropriately to give the claimed behavior for each Mσ`pv, Lq. It remains to consider when a2 " 0, i.e v2 " ´w2. The fact that a is an effective isotropic class such that xa, wy ą 0 implies that a " ku1. If w2 " ´1, then k " xu1, wy " 1, in which case v " w` u1 and the claim follows from Case (2)(a) of Lemma 8.20. On the other hand, if w2 " ´2, then 2 " ´w2 " xa, wy " xku1, wy " kxu1, wy, so either v " w ` 2u1 and xu1, wy " 1 or v " w ` u1 and xu1, wy " 2. In the first case, v2 " 2 and xv, u1y " 1, as in Case (2)(a), so the claim follows from Lemma 8.19. In the second case, we consider the P1 worth of extensions with F P M s M s σ0pu1, L1q (or M s σ0pu1q and T P M s σ0pwq the unique σ0-stable spherical object of class w. Varying F P σ0pu1, L1 ` KXq), these extensions span a contracted divisor in each Mσ`pv, Lq. (cid:3) 0 Ñ F Ñ E Ñ T Ñ 0 MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 55 8.3. Non-minimal case. Finally, we consider the case that xv, wy ă 0. As usual, we follow the orientation described in Remark 8.4 so that φ`pwq ą φ`pvq, and hence φ´pwq ă φ´pvq, and denote by T the unique (up to ´b OXpKXq) σ0-stable object of class w. We set T1 :" xT, TpKXqy if w2 " ´1 (resp. T1 :" xTy if w2 " ´2) and F1 is the full subcategory of Pp1q generated by σ0-stable objects E with φ`pEq ă φ`pTq. We also let T 1 be the full subcategory of Pp1q generated by σ0-stable objects E with φ´pEq ą φ´pTq and F1 :" xT, TpKXqy if w2 " ´1 (resp. F1 :" xTy if w2 " ´2). We set A0 " Pp1q, A1 " xT1r´1s, F1y and A1 :" xT 1 , F1 r1sy. Let Φ : DbpXq Ñ DbpXq be the equivalence defined by T as in Eqs. (2) and (3). Then we have equivalences Φ :A0 „ÝÑ A1 Φ´1 :A0 „ÝÑ A1 . The proof is identical to that for the non-isotropic case. Write n :" ´2xv, wy if w2 " ´1 (resp. n :" ´xv, wy if w2 " ´2), and assume that n ą 0. Then we have isomorphisms (103) (104) Φ : Mσ`pvq Φ : Mσ`pv ´ nwq „ÝÑ „ÝÑ Mσ´pv ´ nwq, Mσ´pvq, where v ´ nw is minimal and xv ´ nw, wy ą 0. By Propositions 8.8 and 8.18 and the fact that xv, wy ‰ 0, we have a birational map (105) unless xv ´ nw, uy " 1 and ℓpuq " 2. In this case, W is totally semistable for v ´ nw as well and induces a divisorial contraction. We note that the S-equivalence class of E P Mσ`pvq is determined by that of ΦpEq P Mσ`pv ´ nwq, as E is an extension of ΦpEq and pT ' TpKXqq'n if w2 " ´1 (resp. T 'n if w2 " ´2), so it follows that W induces a divisorial contraction for Mσ`pvq as well. Otherwise, precomposing and postcomposing the birational map in (105) with the isomorphisms from (104), we get a birational map Mσ´pv ´ nwq 99K Mσ`pv ´ nwq (106) Φ Φ : Mσ`pvq 99K Mσ´pvq. As the birational map in (105) is isomorphic in codimension one unless (1) xv ´ nw, uy " 2 " ℓpuq; or (2) xv ´ nw, uy " 1 " ℓpuq, the same holds true for the birational map in (106). In either of these cases, crossing W induces a divisorial contraction on Mσ`pv ´ nwq by Lemmas 8.17 and 8.20, so the same holds true for Mσ`pvq by the above discussion regarding S-equivalence in the case xv ´ nw, uy " 1, ℓpuq " 2. Remark 8.22. As in the non-isotropic case, if xv, wy " 0, then Φ again induces an isomorphism Φ : Mσ`pvq Ñ Mσ´pvq. Proof of Proposition 8.5. The proposition follows from the above discussion and Lemmas 8.12, 8.13, 8.15-8.17 and 8.19-8.21. The only point that needs mentioning is that the restrictions on v2 in these lemmas can be equivalently rephrased in terms of either the pairing with w or u as in the statement of Proposition 8.5. (cid:3) Example 8.23. An important example arises in the context of Remark 8.22. Suppose that xv, wy " 0 for the effective exceptional class w in the isotropic lattice HW . Then we have seen that if xv, uy ą ℓpuq for all primitive isotropic u P CW X HW, then Mσ`pv, LqzM s σ0pv, Lq is a divisor that does not get contracted. We will see in the next section (specifically Proposition 9.1) that W nevertheless induces a small contraction if v is primitive. This is an interesting situation. On the one hand, we know from Remark 8.22 that Φ induces an isomorphism between Mσ`pvq and Mσ´pvq. Moreover, from its definition, Φ restricts to the 56 HOWARD NUER AND K ¯OTA YOSHIOKA σ0pvq Ă Mσ`pvq X Mσ´pvq. On the other hand, as π` is a identity on the common open subset M s small contraction, we may flop it to get another minimal model M of Mσ`pvq, but M fl Mσ´pvq. This phenomenon, which is present on any Enriques surface, leads one to wonder whether or not the minimal model M can nevertheless be obtained by Bridgeland wall-crossing. It may be possible to reach M by crossing a different wall bounding the chamber containing σ`, or this may be a counter-example to the conjecture that all minimal models of Mσpvq, for σ P Stab:pXq generic, are isomorphic to Mτpvq for some τ P Stab:pXq. 9. Flopping walls In Sections 6-8 we have given necessary and sufficient criteria for the wall W to be totally semistable, to induce a P1 fibration, and to induce a divisorial contraction. In this section, we discuss the remaining possibility for the contraction morphism π`. That is, if W does not induce a fibration or a divisorial contraction, then it must either induce a small contraction, that is, the exceptional locus of π` must have codimension at least two, or W is fake wall so that π` does not contract any curves. In the next result, we give precise criteria for when W is a genuine wall inducing a small contraction, at least for v primitive. It is the only result in our work so far that has assumed that v is primitive. Proposition 9.1. Assume that v is primitive and that W induces neither a divisorial contraction nor a P1-fibration. If either (1) v2 ě 3 and v can be written as a sum v " a1 ` a2 with ai P PH such that L " r i " 0 and ℓpaiq " 2; or (2) there exists an exceptional class w and either 2 KX pmod 2q if for each i, a2 (a) 0 ă xw, vy ď v2 2 , or (b) xv, wy " 0 and v2 ě 3; or (a) 0 ă xw, vy ă v2 (b) xw, vy " v2 2 , or (3) there exists a spherical class w and either 2 and v ´ w is a spherical class, then W induces a small contraction on Mσ`pv, Lq. Proof. Note that it suffices to show that some positive dimensional subvariety of σ`-stable objects becomes S-equivalent with respect to σ0 and thus gets contracted by π`. We consider Case (1) first, so v " a1 ` a2 with ai P PH. Using [6, Lemma 9.2], we may assume that the parallelogram with vertices 0, a1, v, a2 does not contain any lattice point other than its vertices. In particular, the ai are primitive, and without loss of generality, we may assume that φ`pa1q ă φ`pa2q. By Theorem 3.8, there exist σ`-stable objects Ai with vpAiq " ai. If a2 i ą 0 for each i, then the signature of H forces xa1, a2y ě 2 so that ext1pA2, A1q ě 2. If, say, a2 1 " 0, then by part (1) of Theorem 5.8 and the assumptions that v2 ě 3 and that W does not induce a divisorial contraction, we must have either ℓpa1q " 2 and xv, a1y ě 3 or ℓpa1q " 1 and xv, a1y ě 2. So either way xa2, a1y ě 2 and again ext1pA2, A1q ě 2. By [6, Lemma 9.3], any nontrivial extension 0 Ñ A1 Ñ E Ñ A2 Ñ 0 is σ`-stable of class v. All such extensions are non-isomorphic but S-equivalent with respect to σ0, giving a projective space of positive dimension contracted by π`. Moreover, in all of the above cases, we may choose A1 or A2 to have the appropriate determinant so that E can have either of the two possible determinants for v, except when a2 1 " a2 2 " 0 and ℓpa1q " ℓpa2q " 2. When a2 1 " a2 2 " 0 and ℓpa1q " ℓpa2q " 2, then the argument above produces a projective space of positive dimension dimension contracted by π` if L " detpA1q ` detpA2q " rkpA1q 2 KX ` rkpA2q 2 KX " r 2 KX pmod 2q, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 57 as claimed. Observe further that if H contains an exceptional or spherical class w, then a2 " a1 ´ 2xa1,wyw2 w, where we note that if w2 " ´2 then xa1, wy is even by Remark 8.3, so v " a1 ` a2 " 2a1 ´ 2xa1, wy w2 w " 2a1 ´ xa1, wy w2 w is not primitive, contary to our assumptions. Thus this possibility only occurs in Case (1) of Proposition 5.7. We move on to Case (2)(a). Assume first that v is minimal. Then xv, wy ą 0 means w is effective. Since pv ´ wq2 ě ´1 and xv, wy ď v2 2 ă v2, we see that xv, v ´ wy ą 0 so that v ´ w must be effective. From the assumptions, we observe that xw, v ´ wy " xw, vy ` 1 ě 2. As in the proof of [6, Proposition 9.1], we consider the parallelogram P with vertices 0, w, v, v ´ w and the function fpaq " a2 on P, and the same argument as given there shows that fpaq ą ´1 unless a P tw, v ´ wu. It follows that if P contains any lattice point a other than its vertices, then both a2 ě 0 and pv ´ aq2 ě 0. So v is the sum of two positive classes, and we are in Case (1). It is easy to see that indeed v satisfies the extra condition v2 ě 3 if such an a is isotropic. We may therefore assume that no such lattice points exist. Let T be a σ`-stable object of class w and F be any σ`-stable object of class v´ w. Then, assuming φ`pwq ă φ`pv ´ wq without loss of generality, we get ext1pF, Tq ě 2, so there is a positive dimensional projective space worth of σ`-stable extensions that are S-equivalent with respect to σ0 and thus get contracted by π`. Now assume that v is not minimal. Consider the composition of spherical and weakly-spherical twists as in Proposition 6.20 or Proposition 6.22, and denote it by Φ. Then Φpvq is minimal, and as xΦpwq, Φpvqy " xw, vy, we see that Φpwq is an effective exceptional class and satisfies the same inequality in the hypothesis of (2)(a) for Φpvq instead v. Thus Φpv ´ wq is effective and satisfies Φpv´ wq2 ě ´1. Moreover, depending on the parity of the index of the chamber occupied by v, we either get Mσpvq -- MσpΦpvqq or Mσpvq -- Mσ¯pΦpvqq, and the S-equivalence class of E P Mσ`pvq is determined by that of ΦpEq P MσpΦpvqq, respectively. Thus the result follows from the work of the previous paragraph. Now let us consider Case (2)(b). We first assume that v is minimal. As v2 ě 3 by assumption, pv ´ 2wq2 ě ´1, and since xv, wy " 0 and xv, v ´ 2wy " v2 ą 0, we may assume that both w and v ´ 2w are effective. Note also that xw, v ´ 2wy " 2. Let T, TpKXq be the two σ`-stable exceptional objects of class w and let F be a σ`-stable object of class v ´ 2w. As in the previous cases, we may assume that the parallelogram with vertices 0, w, v, v ´ w has no additional lattices points so that therefore the parallelogram with vertices 0, 2w, v, v´ 2w has no lattices points other than 0, w, 2w, v, v ´ w, v ´ 2w. Without loss of generality we may assume that φ`pwq ą φ`pvq. Then for any extension corresponding to non-zero extensions in each of Ext1pT, Fq and Ext1pTpKXq, Fq, E satisfies 0 Ñ F Ñ E Ñ T ' TpKXq Ñ 0 HompT, Eq " HompTpKXq, Eq " 0. It follows that E is σ`-stable of class v. Indeed, if not, then the class of the maximal destabilizing subobject A would satisfy φ`pvpAqq ą φ`pvq and thus must either be w or 2w. But then we would get HompT, Eq ‰ 0 or HompTpKXq, Eq ‰ 0, a contradiction. Thus we get a P1 P1 worth of non-isomorphic σ`-stable objects of class v that gets contracted by π`. If v is not minimal, then as in Case (2)(a), we may apply the composition of spherical and weakly-spherical twists Φ to reduce to the minimal case. Finally, we move on to Case (3) and deal with both subcases at the same time. The proof proceeds exactly as in [6, Lemma 9.1, case (b)] persuant to the following remark: if there is a lattice point a in the parallelogram with vertices 0, w, v, v ´ w that satisfies a2 " ´1, then we are in Case (2), which we have covered already. The argument of [6, Lemma 9.1, case (b)] then carries through without change. (cid:3) 58 HOWARD NUER AND K ¯OTA YOSHIOKA pv ´ w0q2 " ´1 pv ´ w1q2 " ´1 w1 v0 v2 " v2 0 w0 Figure 3. The region between the two hyperbolas when w2 0 " w2 1 " ´1. Now we prove the converse to Proposition 9.1, namely that if HW does not fall into any of the above mentioned cases, then W is not a genuine wall. Proposition 9.2. Assume that v is primitive and that W induces neither a divisorial contraction nor a P1-fibration. Assume further that we are not in Cases (1)-(3) of Proposition 9.1. Then W is either a fake wall, or not a wall at all. Proof. We consider first the case that v is minimal in its GH-orbit. Furthermore, we assume for now that v2 ě 3 and prove that in this case every σ`-stable object E of class v is σ0-stable. If not, then some such E is strictly σ0-semistable, and thus σ´-unstable. Let a1, . . . , an be the Mukai vectors of the HN-filtration factors of E with respect to σ´. By assumption on the failure of condition (1), the ai cannot all be in PH, so E must have a destabilizing spherical or exceptional subobject or quotient T with vpTq " w. If there is only one σ0-stable spherical or exceptional object (in the latter case, uniqueness is only up to ´ b OXpKXq of course), then clearly v ´ w P PH, so v2 ´ 2xw, vy ` w2 ě 0, contradicting the assumption about the failure of conditions (2) and (3) of Proposition 9.1. Now suppose instead that there are two σ0-stable spherical/exceptional objects with Mukai vectors w0, w1. We must have v ´ w P CW , and moreover, by [51, Lemma 4.6] any stable factor of T must also be spherical or exceptional, so v ´ w0 or v ´ w1 must be effective as well. The assumption about the failure of conditions (2) and (3) in addition to the minimality assumption on v force xv, wiy ą v2 2 , and thus that pv ´ wiq2 ă w2 i , for i " 0, 1. But then v must lie above the concave up (region of the) hyperbola pv ´ w1q2 " w2 1 and below the concave down hyperbola 0. In case w2 pv ´ w0q2 " w2 1, these two hyperbolas intersect at 0 and w0 ` w1, as pictured in 1 " ´2, then we must have pv ´ w0q2 ď ´2 and pv ´ w1q2 ă ´2. Fig. 3, while if, say, ´1 " w2 Similarly to the previous case, v must lie above or on the hyperbola pv´ w0q2 " ´2 and below the concave down hyperbola pv ´ w1q2 " ´2. One can easily check that, writing vi " xiw0 ` yiw1, i " 1, 2, these hyperbola intersect at two points v1 and v2 both of which satisfy 0 ă xi, yi ă 1. In either case, it follows that v must be located in the interior of the parallelogram with vertices 0, w0, w1, w0 ` w1. But then neither v ´ w0 nor v ´ w1 can be effective, a contradiction. Thus W is not a wall at all as every E P Mσ`pvq is σ0-stable. Now let us suppose that v is still minimal but v2 " 1, 2. We will show that though there may be some strictly σ0-semistable object E, there are no curves of such objects that σ`-stable. Take a strictly σ0-semistable object E and consider its Jordan-Holder filtration with respect to σ0 with i"1 ai, and the minimality of v σ0-stable factors Ei with vpEiq " ai. Then we may write v "řn forces xv, aiy ě 0 for all i, so we may order the ai such that 0 " w2 0 ‰ w2 0 ď xv, a1y ď xv, a2y ď ď xv, any. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 59 Let us begin with v2 " 1, and we observe that for E to be strictly σ0-semistable we must have xv, a1y " 0, since otherwise nÿi"1xv, aiy ě n. 1 " v2 " 1 ă 0, so that a2 But then the signature of HW forces a2 1 " ´2 or ´1 and E1 is spherical or excep- tional, respectively. We claim that actually E1 must be exceptional. Indeed, if E1 were spherical then HW would be non-isotropic by Proposition 8.1, so W would induce a divisorial contraction for each choice of L by Lemma 7.2, contrary to assumption. Thus E1 must be exceptional. It follows that HW is isotropic as pv ´ a1q2 " 0, and xv ´ a1, a1y " 1 implies that v " a1 ` u1, which is the only decomposition of v into effective classes. If ℓpu1q " 2, then by Lemma 8.11 W is totally semistable but contracts no curves. If ℓpu1q " 1, then by Lemma 8.20 Mσ`pv, LqzM s σ0pv, Lq is a divisor containing no contracted curves for each choice of L, as required. Now we consider v2 " 2. If xv, a1y ą 0, then we must have n " 2 and xv, a1y " xv, a2y " 1. Moreover, 1 " pv ´ a2q2 " v2 ´ 2xv, a2y ` a2 a2 2 " a2 2. 1 ě 0, then a2 i " 0 and xa1, a2y " 1. It follows that ℓpa1q " ℓpa2q " 1, so we get that Mσ`pv, LqzM s 2 ě 0, so from v2 " 2, we see that the only decomposition with ai P PH occurs Thus if a2 when a2 σ0pv, Lq is a divisor containing no contracted curves by the end of the proof of Lemma 8.20. Otherwise, i ă 0, and we have xv, aiy " v2 a2 2 as in Cases (2)(a) and (3)(b) of Proposition 9.1, contrary to our assumption. Thus we must consider xv, a1y " 0, and from the signature of HW we must have a2 1 ă 0 so that a2 1 " ´2 or ´1 and E1 is spherical or exceptional, respectively. If E1 is spherical, then pv ´ a1q2 " 0, so HW is isotropic, and we either have v ´ a1 " u1 with xu1, a1y " 2 or v´a1 " 2u1 with xu1, a1y " 1 (in which case ℓpu1q " 1). If ℓpu1q " 2, then we must have v " a1`u1 with xa1, u1y " 2, and this is the only decomposition of v. But then by Lemma 8.13 W is not a wall at all since by assumption W does not induce a P1-fibration. If ℓpu1q " 1, then from the end of the proof of Lemma 8.21, we cannot have the decomposition v " a1 ` u1 as this would lead to a divisorial contraction for each choice of L, contrary to assumption. Thus we must have v " a1 ` 2u1 with xa1, u1y " 1. But then as W does not induce a P1-fibration, we must have L " D ` p rk v σ0pv, Lq is a divisor that contains no contracted curves by Lemma 8.19, as required. If E1 is exceptional, then HW must be non-isotropic. Indeed, if HW were isotropic, then we could write v " xu1`ya1 with x, y P Z, and the two conditions v2 " 2 and xv, a1y " 0 would force 2 " y2, a contradiction. If L " D ` rk v 2 KX pmod 2q and HW falls into Case (3)(c) of Proposition 5.7, then from Lemma 7.3, W would induce a divisorial contraction, contrary to our assumptions. So either HW falls into Case (3)(c) of Proposition 5.7 and L ı D ` rk v 2 KX pmod 2q or HW does not fall into Case (3)(c) of Proposition 5.7. But then Lemma 7.3 shows that no curves get contracted, as required. 2 ` 1qKX pmod 2q and Mσ`pv, LqzM s If v is not minimal, then v P Cn for some 0 ‰ n P Z and there exists a minimal class v0 in the same orbit. By Propositions 6.20 and 6.22, if n is even then Mσ`pvq -- Mσ`pv0q, induced by a sequence of spherical/exceptional twists, and if n is odd, then Mσ`pvq -- Mσ`pv1q -- Mσ´pv0q. Since the assumptions of the proposition are invariant under GH, the same asumptions apply to v0, so every σ-stable object of class v0 is σ0-stable. If Φ is the sequence of spherical/exceptional twists used in this isomorphism, then from the definition of a spherical/exceptional twist, it is easy to see that the S-equivalence class of ΦpE0q is determined by that of E0. Since this equivalence is trivial on Mσpv0q, it must be trivial on Mσ`pvq as well, implying that π` is an isomorphism. Thus W is a fake wall, as claimed. (cid:3) Finally we can prove Theorem 5.8: 60 HOWARD NUER AND K ¯OTA YOSHIOKA Proof of Theorem 5.8. The theorem follows from Propositions 7.1, 8.5, 9.1 and 9.2 and Lemma 6.1. (cid:3) We will end this section by observing that Theorem 5.8 and Propositions 6.21 and 6.22 prove part (1) of Theorem 1.1. Indeed, connect σ and τ by a path, σptq, with 0 ď t ď 1, σp0q " σ, and σp1q " τ . Observe that as the set of walls is lcoally finite, σptq only crosses finitely many of them, and by perturbing σptq slightly, we may assume that σptq only crosses one wall at a time (that is, if σpt0q P W then σpt0q is a generic point of W). If σptq does not cross any totally semistable walls, then Mσpvq and Mτpvq are clearly birational. Otherwise, it suffices consider that σ " σ` and τ " σ´ are two sufficiently close stability conditions separated by a single totally semistable wall W. If there exists a spherical/exceptional class w P HW such that xv, wy " 0, then Mσ`pvq -- Mσ´pvq by Propositions 6.21 and 6.22. Otherwise, by Theorem 5.8 and Propositions 6.20 and 6.22, we may assume that v is minimal so that we must have xv, uy " 1 for an isotropic u P HW such that ℓpuq " 2. But then we have seen that π` is a divisorial contraction, so we again have Mσ`pvq and Mσ´pvq are birational, as claimed. In Section 11 we will prove part (2) of Theorem 1.1 and the rest of our main results. In particular, we will show that, under generic conditions, the birational map between Mσ`pvq and Mσ´pvq is induced by a Fourier-Mukai transform. This is obvious when W induces a small contraction, but is more subtle in the case of divisorial contractions. But while the behavior of divisorial contractions for Bridgeland moduli on Enriques surfaces is analogous to the K3 case for divisorial contractions of Brill-Noether, Hilbert-Chow, and Li-Gieseker-Uhlenbeck type, there is an additional type of divisorial contraction that requires extra care. We will turn to this subject in the next section. 10. LGU on the covering K3 surface As its name suggests, the induced Li-Gieseker-Uhlenbeck (iLGU) type contraction, which occurs when the hyperbolic lattice HW contains an isotropic u such that xv, uy " 1 " ℓpuq, is induced from the covering K3. The associated HN-filtration factors are objects F P M s σ0p2uq. But while this stable locus has dimension two, it is not proper as Mσ0p2uqzM s σ0p2uq ‰ ∅. So we cannot use the usual machinery as in the regular LGU case where ℓpuq " 2. There we use the universal family associated to the proper two-dimensional moduli space to obtain the requisite Fourier-Mukai transform as in [16, Corollary 2.8]. Instead, we show that we may induce a Fourier-Mukai transform on X by considering the corresponding LGU type contraction on its K3 cover rX. So we turn now to considering LGU contractions more carefully in the K3 case. 10.1. Fourier-Mukai transform associated to the Li-Gieseker-Uhlenbeck contraction. Let X1 be a K3 surface and u1 be a primitive isotropic Mukai vector on X1. For a general stability condition σ P Stab:pX1q, we consider X2 :" Mσpu1q, which is another K3 surface. Let E P DbpX1 X2, 1X1 αq be a universal object for Mσpu1q, considered as a twisted object where α is a 2-cocycle of OX2 . We set (107) Φ :" ΦE _ X1ÑX2 , Ψ :" ΦE X2ÑX1 . For a smooth variety Y , we also set (108) DY pEq :" E_ " RHomOY pE, OY q, E P DbpY q. Let v1 be a Mukai vector on X1 such that xu1, v1y " 2. It is well known that, using the Fourier- Mukai transform Φ, the moduli space Mσpv1q is isomorphic to a moduli space of rank two Gieseker stable sheaves of Mukai vector v2 " vpΦpv1qq on X2. In the next result, we translate a well-known concrete result about crossing the Uhlenbeck wall for v2 on X2 to an instrinsic analogous statement for X1. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 61 Proposition 10.1. There is an object F P DbpX1 X1q which induces a Fourier-Mukai transform ΦF _ X1ÑX1 : DbpX1q Ñ DbpX1q such that preserves Stab:pX1q. X1ÑX1 (1) ΦF _ (2) The induced action on HpX1, Zq is given by the formula (109) DX1 ΦF _ X1ÑX1pvq " ´pv ` v2 1 2 xv, u1yu1 ´ xv, v1yu1 ´ xv, u1yv1q, for any v P HpX1, Zq. Proof. For F P DbpX1q with vpFq " v1, we set (110) N :" det ΦpFq, and v2 :" vpΦpFqq " p2, ξ, aq. As mentioned above, Φ induces an isomorphism between Mσpv1q and the moduli space Mωpv2, αq of ω-Gieseker stable α-twisted sheaves of Mukai vector v2 on X2. It is well known that crossing the LGU wall of the Gieseker chamber for v2 is induced by the Fourier-Mukai transform p q_ b N " pbNq DX2. Pulling this back to X1, we have a sequence of equivalences: (111) DX2ÝÝÝÝÑ DbpX2, αq bNÝÝÝÝÑ DbpX2, α´1q ΨÝÝÝÝÑ DbpX1q, DbpX1q ΦÝÝÝÝÑ DbpX2, α´1q and we denote their composition by (112) Ξ :" Ψ pbNq DX2 Φ, which is a (contravariant)-equivalence DbpX1q Ñ DbpX1q. Thus DX1 Ξ is an autoequivalence of DbpX1q. Since r2s Ξ " DX1 ΦE _ X2ÑX1 pbN_q Φ (by Grothendieck-Serre duality), DX1 Ξ defines a Fourier-Mukai transform ΦF _ X1ÑX1 To see that ΦF _ preserves Stab:pX1q, note first that Φpσq is geometric as Φ identifies the objects of Mσpu1q with the points of X2 considered as Φpσq-stable objects of class p0, 0, 1q. Thus Φ induces an isomorphism Stab:pX1q -- Stab:pX2q. Since bN_ also preserves the space of full numerical stability conditions, ΦF _ : DbpX1q Ñ DbpX1q. X1ÑX1 preserves Stab:pX1q, as claimed in (1). X1ÑX1 Now we prove (2). Since v2 " p2, ξ, aq, it follows that pQv2 ` QX2qK "! eξ{2p0, D, 0q D P H 2pX2, Qq) , where the orthogonal complement is taking inside HpX2, Qq. Moreover, under pbNq DX2, we have (113) Hence (114) v2 X2 eξ{2p0, D, 0q v2 X2 ÞÑ ÞÑ ÞÑ ´eξ{2p0, D, 0q. ΞpQu1`Qv1q " 1pQu1`Qv1q, ΞpQu1`Qv1qK " ´1pQu1`Qv1qK . Then it is easy to see that for any v P HpX1, Zq we have (115) v2 1 Ξpvq " ´pv ` 2 xv, u1yu1 ´ xv, v1yu1 ´ xv, u1yv1q. (cid:3) 62 HOWARD NUER AND K ¯OTA YOSHIOKA 10.2. The induced Li-Gieseker-Uhlenbeck wall and its Fourier-Mukai transform on an Enriques surface. Having delved more deeply into the FM transform associated to a wall of LGU type on K3 surfaces, we explore the corresponding picture on the Enriques quotient. So let X be an Enriques surface with K3 cover rX. Proposition 10.2. Assume that PicprXq " PicpXq. Let u0 and v0 be Mukai vectors such that v1 :" ´pX ` 2v2 Then there is an autoequivalence ΦE _ u0 is primitive and isotropic, xu0, v0y " 1 and ℓpu0q " 1. We set 0xX , u0yu0 ´ 2xX , v0yu0 ´ 2xX , u0yv0q. XÑX : DbpXq Ñ DbpXq such that (1) For any v P HalgpX, Zq, (116) DX ΦE _ (2) vpEXtxuq " v1. (3) DX ΦE _ XÑX " Ξ for Ξ in (112). XÑXpvq " ´pv ` 2v2 0xv, u0yu0 ´ 2xv, v0yu0 ´ 2xv, u0yv0q. (cid:3) (117) Since ΦF _ rXÑrX Proof. We note that v1 is a primitive and isotropic Mukai vector with ℓpv1q " 2. We set v1 :" pv0q and u1 :" pu0q. By Proposition 10.1, we have an object F P DbprX rXq which defines a Fourier-Mukai transform satisfying (1) and (2) of Proposition 10.1. We will show that F descends, in an appropriate sense, to an object E P DbpX Xq that defines the desired autoequivalence of DbpXq. preserves Stab:prXq, which is isomorphic to Stab:pXq from the assumption PicprXq " PicpXq by [35, Theorem 1.2], there is an ι-invariant stability condition σ1 P Stab:prXq such that F is a family of σ1-stable objects. By the ι-invariance of σ1, ιpFrXtrXuq is also σ1-stable, and moreover, 2vpFrXtrXuq " pv1q. Hence we have an isomorphism τ : rX Ñ rX such that pι 1rXqpFq -- p1rX τqpFq b L, where L is a line bundle on rX. Then we have a commutative diagram ĂXÑĂXÝÝÝÝÑ DbprXq DbprXq §§đpbL_qτ ι§§đ DbprXq ÝÝÝÝÑΦF _ DbprXq. ĂXÑĂX rXÑrX ιpvq, v P HprX, Zq. By Proposition 10.1, we see that ι ΦF _ Hence we get (118) ΦF _ rXÑrXpvq " ΦF _ ιpvq " pbL_q τpvq, v P HprX, Zq, (119) (120) which implies L " OrX and ι " τ on HprX, Zq (by substituting v " vpOrXq). By the Torelli theorem, ι " τ . Therefore ΦF _ rXÑrX Theorem 4.5], there is E P DbpX Xq such that (121) p 1rXqpFq -- p1X qpEq. XÑX defines the desired Fourier-Mukai transform. It follows that ΦE _ and ι commute. Then as in the proof of [15, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 63 Remark 10.3. The conditions (1) and (2) are cohomological, and [35, Theorem 1.2 (ii)] shows that there are autoequivalences of DbpXq satisfying (1) and (2) without the assumption PicprXq " PicpXq. Since we need the property (3) for the proof of Proposition 10.4 below, we gave a proof. XÑX to the birational behavior at a Now we can relate the induced Fourier-Mukai transform ΦE _ iLGU wall. XÑX induces an isomorphism Mσ`pvq -- Mσ´pvq. underlying lattice as the torsion free lattice NSpXqtf but with pairing multiplied by 2; and the Proposition 10.4. Assume that PicprXq " PicpXq. Let W be a wall for v defined by an isotropic Mukai vector u such that xu, vy " 1 and ℓpuq " 1. Let σ0 P W be a generic stability condition. Assume that σ are sufficiently close stability conditions in opposite, adjacent chambers separated by W. Then DX ΦE _ Proof. There are two important consequences of our assumption that PicprXq " PicpXq. The first is that HalgprX, Zq " NSpXqtfp2q ' x2y ' x´2y, where we recall that NSpXqtfp2q is the same second is that Stab:prXq " Stab:pXq. In particular, it follows from the first consequence that setting X1 :" rX, v1 :" pvq, and u1 :" puq " u1, there is no w P HalgpX1, Zq with xw, u1y " 1. Thus α as in Section 10.1 defines a non-trivial Brauer class on X2 :" Mpσqpu1q, and there are no α-twisted sheaves on X2 of rank 1. From the second consequence it follows that pσq remains generic with respect to pvq. Considering the autoequivalence Φ constructed in the proof of Proposition 10.1, we see that for F P Mσ`pvq, ΦppFqq is a µ-stable α´1-twisted sheaf of rank two, as there are no twisted sheaves of rank one. Since pσq remains generic, µ-stability is taken with respect to a generic polarization, in which case µ-stability is independent of the choice of the B-field. Thus pσ`q and pσ´q are separated by the single wall associated to the LGU contraction. The construction of Ξ in Proposition 10.1 shows that Ξ switches the chambers containing pσq, so it follows that Ξppσ`qq and pσ´q belong to the same chamber. Therefore Ξ induces an isomorphism Mpσ`qpv1q Ñ Mpσ´qpv1q. By Proposition 10.2 (3), it follows that DX ΦE _ XÑX induces an isomorphism Mσ`pvq Ñ Mσ´pvq, as claimed. 10.3. The case of odd rank. For the construction of the equivalence ΦE _ XÑX, we had to assume Proposition 10.4. In this subsection, we shall construct an isomorphism Mσ`pv, Lq Ñ Mσ´pv, Lq without the construction of ΦE _ is odd. that PicprXq " PicpXq. Then by using the equivalence, we constructed the isomorphism in XÑX, and thus without the assumption PicprXq " PicpXq, if rk v For a stability condition σ P Stab:pXq, we set σ1 :" pσq. Then we have a morphism ψ : Mσpvq Ñ Mσ1pv1q by sending E P Mσpvq to pEq, where v1 " pvq. We need to study when the restriction of ψ to the stable locus gives σ1-stable objects, the content of the next lemma. Lemma 10.5 ([40, Theorem 8.1]). For a σ-stable object E, pEq is properly σ1-semistable if and only if E " pFq, where F is a σ1-stable object with F fl ιpFq. In particular rk E is even. Lemma 10.6. If rk v is odd, then ψ : Ms contained in Ms σpv, Lq Ñ Mσ1pv1q is an embedding and the image is (cid:3) σ1pv1q, where rL mod KXs " c1pvq. Proof. Since rk v is odd, the image is contained in Ms Proposition 7.2], we get the claim. σ1pv1q by Lemma 10.5. Combining with [40, (cid:3) In particular, if v has odd rank and σ is generic, then Lemma 10.6 implies that ψ embeds all of σ1pv1, Lq. Moreover, for any σ2 in a sufficiently small neighborhood of σ1, we also Mσpv, Lq into M s have an embedding (122) M s σpv, Lq ãÑ M s σ1pv1q ãÑ Mσ2pv1q 64 HOWARD NUER AND K ¯OTA YOSHIOKA by openness of stability. For the rest of the section, we assume that rk v is odd. Let W Ă Stab:pXq be a wall defined by a primitive isotropic u with xv, uy " 1 " ℓpuq and σ0 a generic stability condition in W. We denote by σ10 " pσ0q, and we continue to denote by v1 and u1 the pull-backs pvq and puq, respectively. Similarly, let W1 be the wall for v1 in Stab:prXq defined by u1. Take σ2 P W1 in a neighborhood of σ10 such that W1 is the unique wall for v1 in a small neighborhood Uσ2 of σ2 and such that σ2 is generic with respect to u1. We also take σ2 P Uσ2 in opposite and adjacent chambers for v1, separated by the single wall W1. We may assume that pσ`q and pσ´q belong to the pv1q be the embeddings for same chambers as σ2` and σ2´, respectively. Let ψ : Mσpv, Lq Ñ Mσ2 σ in (122). We will show in a moment that there is an isomorphism ϕ : Mσ`pv, Lq Ñ Mσ´pv, Lq despite ϕ (possibly) not being induced by an autoequivalence of DbpXq. But first we start with a cohomo- logical lemma. Lemma 10.7. Let v1 " pvq and u1 " puq where xv, uy " 1 " ℓpuq for the primitive isotropic vector u. Then the sublattice Zu1 ` Zv1 is saturated in HprX, Zq. In particular, if W Ă Stab:pXq is the wall for v defined by u and W1 Ă Stab:prXq is the wall for v1 defined by u1), then we have HW 1 " Zv1 ` Zu1 " HW so that xv1, v2y P 2Z for any v1, v2 P HW 1. Proof. Let L be the saturation of Zu1 ` Zv1 in HprX, Zq. Since u1 is primitive, there is w P L such that L " Zu1 ` Zw. We set v1 " au1 ` bw (a, b P Z). Then 2 " xv1, u1y " bxw, u1y. Hence b " 1,2. If b " 2, then rk v1 " a rk u1 2 rk w P 2Z, which is impossible if rk v1 " rk v is odd. If rk v is even, then let us write v " p2n, D1, sq and u " p2m, D2, tq so that v1 " p2n, D1, 2sq and u1 " p2m, D2, 2tq. Then xv1, u1y " 2 implies that 2pD1, D2q " pD1, D2q " 2 pmod 4q, so pD1, D2q is odd. On the other hand, v1 " 2w ` au1 gives D1 " 2D3 ` aD2, where D3 " c1pwq. But as : PicpXq{xKXy ãÑ PicprXq is a primitive embedding, we must have D3 " D13 for some D3 P PicpXq. It follows that D1 " 2D13 ` aD2 and thus a contradiction. Thus, in either case, b " 1 so that L " Zu1 ` Zv1, as required. The first equality of the last statement follows because HW 1 Ă HalgprX, Zq is a saturated hyper- bolic sublattice of rank two containing Zv1 ` Zu1, and the second equality follows from xv, uy " 1 so that HW " Zv ` Zw. pD1, D2q " 2pD13, D2q ` apD2q2 " 0 pmod 2q, (cid:3) Now we are in position to prove the following result. Proposition 10.8. We have an isomorphism ϕ : Mσ`pv, Lq Ñ Mσ´pv, Lq with a commutative diagram (123) In particular, it is the identity on M s Mσ`pv, Lq ÝÝÝÝÑ Mσ´pv, Lq ϕ ψ`§§đ Mσ2 `pv1q ÝÝÝÝÑΞ σ0pv, Lq. §§đψ´ ´pv1q Mσ2 Proof. For X1 :" rX and X2 :" Mσ2pu1q, we shall apply Proposition 10.1 to get an (anti-) autoequivalence Ξ of DbpX1q. Recall from the construction of Ξ that there is an equivalence `pv1q and the moduli space Φ : DbpX1q Ñ DbpX2, αq inducing an isomorphism between Mσ2 MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 65 `pv1q Ñ Mσ2 Mωp2, D, aq of ω-Gieseker semistable α-twisted sheaves on X2. Moreover, Ξ is the pull-back to ´pv1q. X1 of the autoequivalence E ÞÑ E_ b detpEq, which induces an isomorphism Mσ2 We claim that Mωp2, D, aq consists only of µ-stable sheaves. Granting this for the moment, we see that for locally free E P Mωp2, D, aq, we have E_ b detpEq -- E, so the isomorphism σ2pv1q, which corresponds to the locus `pv1q Ñ Mσ2 ´pv1q induced by Ξ is the identity on M s Mσ2 M lf ω p2, D, aq on X2. Now let us prove our claim that Mωp2, D, aq " M µs ω p2, D, aq. We note first that the claim is obvious if α is non-trivial, as we saw in the proof of Proposition 10.4. So suppose that the Brauer class α is trivial. If there exists a strictly µ-semistable sheaf E P Mωp2, D, aq with subsheaf F Ă E such that µpFq " µpEq, then from the genericity of ω, it would follow that 2 D and vpFq " p1, D 2 , bq so that x´vpFq,p0, 0, 1qy " 1. But 2vpFq " p2, D, aq ` p2b ´ aqp0, 0, 1q, so writing ´vpFq " Φpwq, we get w such that 2w P Zv1 ` Zu1 and xw, u1y " 1. As HW 1 " Zv1 ` Zu1 is saturated by Lemma 10.7, this is impossible, proving the claim in this case as well. By (122), we get an isomorphism ϕ with the commutative diagram (123). (cid:3) We close this subsection by describing the relationship between the Picard groups of Mσpv, Lq. Proposition 10.9. Let R be the cohomological action on KpXq given by RpEq " E ` 2v2xvpEq, uyU ´ 2xvpEq, vyU ´ 2xvpEq, uyV, E P KpXq, where U, V P KpXq satisfy vpUq " u, vpV q " v. Then we have a commutative diagram modulo torsion. (124) KpXqv θv,σ`§§đ RÝÝÝÝÑ KpXqv §§đθv,σ´ PicpMσ`pv, Lqq ÐÝÝÝÝϕ PicpMσ´pv, Lqq Proof. Consider a quasi-universal family E P DbpX Mσpv, Lqq for the moduli space Mσpv, Lq and the map 1 fitting into the commutative diagram: rX Mσpv, Lq (❘❘❘❘❘❘❘❘❘❘❘❘❘ pMσ pv,Lq 1 X Mσpv, Lq pMσ pv,Lq Then by (122), we may consider p 1qE to be a family of objects in Mσ2 pv1q parameterized by Mσpv, Lq. The morphism ψ is the classifying map associated to the family p 1qE, so by [22, Theorem 8.1.5] we get Mσpv, Lq θp1qE " ψ θv1,σ2 , where θp1qE : KprXqv1 Ñ PicpMσpv, Lqq is the Donaldson-Mukai homomorphism associated to the family p1qE. Precomposing with the pull-back from KpXqv gives the following commutative diagram: (125) KpXqv 2θv,σ θp1qE v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ PicpMσpv, Lqq ψ KprXqv1 pv1qq θv1 ,σ2 PicpMσ2 / / ( / / v o o 66 HOWARD NUER AND K ¯OTA YOSHIOKA Indeed, for E P KpXqv, we have θp1qEpEq " detppMσpv,Lq!pp 1qE b p rXpEqqq " detppMσpv,Lq!pp 1qE b p 1qppX Eqqq " detppMσpv,Lq!pp 1qpE b pX Eqqq " detppMσpv,Lq!pp 1qp 1qpE b pX Eqqq " detppMσpv,Lq!ppE b pXpE b pOX ' OXpKXqqqqqq " θv,σpEq ` θv,σpEpKXqq, Using R " ´Ξ and θv1,σ2 so modulo torsion, θp1qEpEq " 2θv,σpEq, as claimed. Ξ to denote both the autoequivalence and the induced isomorphism Ξ : Mσ2 proposition then follows from (123). ´ Ξ, where we have abused notation by using ´pv1q, the ` " ´Ξ θv1,σ2 `pv1q Ñ Mσ2 (cid:3) Remark 10.10. (1) Since the claim is independent of the choice of complex structure, we can reduce to the case where PicprXq " PicpXq. Then the claim is a consequence of (2) Since pξτq " 2ξpτq, it follows that Proposition 10.2. ℓCpτq " θv,σpξτq " ψ pθv1,σ2 pξpτqqq " ψ ℓC2 pτq. This gives another proof of [40, Proposition 10.2]. 10.4. The case of even rank. We show in this subsection that the technique of Section 10.3 also generalizes to Mukai vectors v of even rank. For the rest of the section, we assume that rk v is even. Let W Ă Stab:pXq be a wall defined by a primitive isotropic u with xv, uy " 1 " ℓpuq and σ0 a generic stability condition in W. We denote by σ10 " pσ0q, and we again denote by v1 and u1 the pull-backs pvq and puq, respectively. Similarly, let W1 Ă Stab:prXq be the wall for v1 defined by u1. Take σ2 P W1 in a neighborhood of σ10 such that W1 is the unique wall for v1 in a small neighborhood Uσ2 of σ2 and such that σ2 is generic with respect to u1. We also take σ2 P Uσ2 in opposite and adjacent chambers for v1, separated by the single wall W1. We may assume that pσ`q and pσ´q belong to the same chambers as σ2` and σ2´, respectively. We observe first that we still have ℓpvq " 1 so that v is primitive. Indeed, as in Remark 8.3, if ℓpvq were 2, then 2 rkpuq would imply that xv, uy were even, which is absurd. It therefore pv1q follows from Lemma 10.5 that we have morphisms ψ : Mσpv, Lqsm Ñ M s which are ´etale double covers onto their images. Moreover, by Lemma 8.20 we have a birational map ϕ : Mσ`pv, Lq 99K Mσ´pv, Lq which is the identity on M s σ0pv, Lq. As Mσpv, Lq has only terminal l.c.i. singularities and numerically trivial canonical divisor by [40, Theorem 8.2], it follows that ϕ is an isomorphism in codimension one by [26, Corollary 3.54]. This defines an isomorphism pσqpv1q ãÑ M s ϕ : ClpMσ´pv, Lqq „ÝÑ ClpMσ`pv, Lqq, where Clp´q denotes the group of Weil divisors moduli linear equivalence. If v2 ě 8, then codimpSingpMσpv, Lqq ě 4 so that Mσpv, Lq is locally factorial by [19, Exp. XI, Cor. 3.14] and thus ClpMσpv, Lqq " PicpMσpv, Lqq. We will show in the next result that, even for 2 ă v2 ă 8, ϕ extends to an isomorphism so that we may nevertheless identify PicpMσpv, Lqq. σ2 MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 67 Proposition 10.11. If v2 ą 2, then the birational map ϕ : Mσ`pv, Lq 99K Mσ´pv, Lq extends to an isomorphism fitting into a commutative diagram modulo torsion. (126) KpXqv θv,σ`§§đ RÝÝÝÝÑ KpXqv §§đθv,σ´ PicpMσ`pv, Lqq ÐÝÝÝÝϕ PicpMσ´pv, Lqq `pv1q Ñ Mσ2 Proof. As in the proof of Proposition 10.8, we have an autoequivalence Ξ that induces an isomor- σ2pv1q. Indeed, the same proof goes through phism Mσ2 since we still have HW 1 " Zv1 ` Zu1 by Lemma 10.7. Restricting the birational map ϕ to the smooth locus Mσ`pv, Lqsm, we get a commutative dia- gram ´pv1q which is the identity on M s Mσ`pv, Lqsm ϕ /❴❴❴ Mσ´pv, Lqsm ψ` ψ´ Mσ2 `pv1q Ξ / Mσ2 ´pv1q with ϕ the identity on M s σ0pv, Lqsm. Moreover, we have two homomorphisms ϕ θv,σ´ R : KpXqv Ñ ClpMσ`pv, Lqq θv,σ` : KpXqv Ñ PicpMσ`pv, Lqq, where R is defined as in Proposition 10.9 and we have abused notation by using ϕ for the restriction of ϕ : ClpMσ´pv, Lqq Ñ ClpMσ`pv, Lqq to the subgroup PicpMσ´pv, Lqq Ă ClpMσ´pv, Lqq. For any x P KpXqv, Proposition 10.9, which had nothing to do with rk v being odd, implies that θv,σ`pxq and ϕ θv,σ´ Rpxq agree away from SingpMσ`pv, Lqq, which has codimension at least two since v2 ą 2. It follows that ϕ θv,σ´ R " θv,σ`, which gives the commutative diagram (126). Finally, we prove that ϕ extends to an isomorphism. As ξσ0 P pvK X uKqR, Rpξσ0q " ξσ0 and thus ϕℓC`pσ0q " ℓC´pσ0q. Moreover, letting d " v ´ v2u P HW X vK, it is easy to see that R " Rd is just the reflection of vK in the hyperplane dK defined by Rdpzq " z ´ 2xz,dyd2 d (see Section 12 for more on this). Perturbing σ and σ0 slightly, we may assume that we can write ξσ " xd ` ξσ0 for x P R. In particular, Rpξσ`q " ξσ´. But then ℓC`pσ`q " ϕpθv,σ´pRpξσ`qqq " ϕpθv,σ´pξσ´qq " ϕℓC´pσ´q. As both ℓCpσq are ample, ϕ extends to an isomorphism by [27, Exercise 5.6], as required. (cid:3) 11. Main theorems Proof of Theorem 1.1. We proved part (1) at the end of Section 9, so we focus on part (2). Never- theless, we recall part of the setup form the proof of part (1). We may connect σ and τ by a path which intersects walls in points that lie on no other walls. As the set of walls is locally finite, the path will intersect only finitely many walls, and thus for the purpose of proving the theorem, it suffices to consider one wall W, a generic stability condition σ0 P W, and nearby stability conditions σ. Suppose that xv, wy ă 0 for an effective spherical or exceptional class w P HW . Then suppose that v " vn P Cn with v0 the minimal Mukai vector in the GH-orbit of v. Letting Φ be the sequence of spherical or exceptional twists giving the isomorphism Mσpvq Ñ Mσpv0q if n is even (resp., Mσpvq Ñ Mσ¯pv0q if n is odd), we see that it suffices to prove the theorem in the case that / / 68 HOWARD NUER AND K ¯OTA YOSHIOKA 1 proving the theorem. and we break up the proof into cases. v is minimal. Indeed, if Φ is an autoequivalence inducing a birational map Mσ`pv0q 99K Mσ´pv0q as in the theorem, then pΦ´q´1 Φ Φ` (resp., pΦ´q´1 Φ´1 Φ`) proves the theorem for v if n is even (resp., if n is odd). Thus we may assume that xv, wy ě 0 for all effective spherical and exceptional classes w P HW , Suppose first that xv, wy " 0 for some effective spherical or exceptional class w P HW . But then the minimality of v forces w " w0 " vpT `0 q or w1 " vpT `1 q as in Propositions 6.20 and 6.22 and or the discussions following them. But we showed there that the spherical/exceptional twists RT ` induce an isomorphism Mσ`pvq Ñ Mσ´pvq. So we may take U to be the entire moduli space RT ` Mσ`pvq. Notice that this discussion covers walls inducing P1-fibrations, walls of Brill-Noether type, as well as flopping walls induced by an exceptional class w P HW such that xv, wy " 0 when v2 ě 3. Otherwise xv, wy ą 0 for all effective spherical or exceptional classes w P HW. We first consider the case that W is a flopping wall or a fake wall. Then W being a flopping or fake wall implies that codimpMσ`pvqzM s σ0pvqq ě 2 by Lemma 6.2 and Propositions 8.8 and 8.18. In this case we may take U to be the open subset of σ0-stable objects, so there is nothing to prove (i.e. we just take Φ " Id). Now we consider the case that W induces a divisorial contraction and xv, wy ą 0 for all effective spherical/exceptional classes w P HW , i.e. divisorial contractions coming from Hilbert-Chow, LGU, or iLGU walls. We will show that we may take U to be the entire moduli space Mσ`pvq in the case of Hilbert-Chow and LGU walls. That is, we show that there is an autoequivalence that induces an isomorphism Mσ`pvq „ÝÑ Mσ´pvq. We show that the same is true for an iLGU wall if PicprXq " PicpXq, while in general we can still take U to be an open subset of codimension one, Hilbert-Chow: Here we assume that HW contains an isotropic vector u such that xv, uy " 1 and ℓpuq " 2. Then by Proposition 8.7 and Lemma 8.11 we may, up to the shift by 1, identify Mσ`pvq with the pβ, ωq-Gieseker moduli space M β ωp´vq of stable sheaves of rank one. Up to tensoring with a line bundle, we may assume that Mσ`pvq parametrizes the shifts IZr1s of ideal sheaves of 0-dimensional subschemes Z P HilbnpXq, where n " v2`1 . Moreover, σ´-stable objects are precisely I_Z r1s. But then Φp q :" p q_r2s provides the required autoequivalence. LGU: As in Proposition 8.7, shifting by one identifies Mσ`pvq with the moduli space Mωp2, c, sq of ω-Gieseker stable sheaves F with vpFq " p2, c, sq " ´v. Choosing L P PicpXq with rL mod KXs " c, we get that Φp q :" p q_ b OpLqr2s is the required autoequivalence, as ΦpFr1sq " F _b OpLqr1s is an object of Mσ´pvq, for any F P Mωp2, c, sq. iLGU: If we take U to be the open subset M s σ0pvq Ă Mσ`pvq, then Mσ`pvqzU has codimension one, and we may take Φ " Id. We can say more if PicprXq " PicpXq. Indeed, then we may take U " Mσ`pvq and Φ " DX ΦE _ XÑX as in Proposition 10.4. 0 (cid:3) 2 Now let us prove Theorem 1.2 as an application of Theorem 1.1. Proof of Theorem 1.2. By the proof of [48], there is an (anti)-autoequivalence Φ such that rk Φpvq " 1. Applying Theorem 1.1, we get the first claim. For the second claim, we use [41, sect. 1]. (cid:3) Similarly, Theorem 1.3 follows from Theorem 1.1. Proof of Theorem 1.3. As in [53, Section 4] (for generic X) or the arguments in Lemmas 13.14 and 13.15, there is an autoequivalence Φ such that Φpvq " w " p0, C, χq where C Ñ P1 is a double cover of P1 and χ ‰ 0. Then by Theorem 1.1, we get the first claim. If PicprXq " PicpXq, then by [53], [44, Assumption 2.16] holds for MHpw, L1q. Hence by [44, Thm. 3.1, Thm. 4.4], we get the second claim. (cid:3) MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 69 In this section we prove Corollary 1.4. Before doing so, however, we prove that the maps 12. Picard groups of moduli spaces ℓC : C Ñ NumpMCpv, Lqq fit together to form one coherent map Stab:pXq Ñ NumpMσ`pv, Lqq whose image is contained in MovpMσ`pv, Lqq. 12.1. How the Bayer-Macr`ı maps fit together. Recall from Eqs. (13) and (14) that ℓC can be written as the composition Z C ãÑ Stab:pXq ÝÑ HalgpX, Cq IÝÑ vK (127) where Zpσq " ✵σ, Ip✵σq " ξσ, and θv,σ are cohomological Donaldson-Mukai maps associated to the universal families for Mσpv, Lq as in Eq. (12). To show that these fit together it suffices to study how the ℓC are related across a single wall crossing, and from (127), it therefore suffices to study the relationship between θv,σ` and θv,σ´. θv,σÝÝÝÑ NumpMσ`pv, Lqq, So, for a given Mukai vector v P HalgpX, Zq with v2 ą 0, consider two adjacent chambers C` and C´ separated by a wall W, and, as always, choose a generic σ0 P W and nearby stability conditions σ P C. The signature of HW and its saturatedness force vK X HW " Zd for some integral d. We have already seen in Propositions 6.21 and 6.22 that in some cases d " wn is a spherical or exceptional class, in which case there is an isomorphism Mσ´pv, Lq „ÝÑ Mσ`pv, Lq induced by RT ` , where T `n is the σ`-stable object of class wn. When v is primitive and W is defined by an isotropic vector u with n (1) xv, uy " 1, 2 and ℓpuq " 2, or (2) xv, uy " ℓpuq " 1, u. We certainly have d1 :" v ´ v2 xv,uy u P`vK X HWQ, and when then we claim that d " v ´ v2 xv,uy xv, uy " 1, it is clear that d1 P vK X HW so that d1 " md for some m P Z. But then 1 " xd1, uy forces m " 1. Otherwise, xv, uy " 2 and ℓpuq " 2, forcing rkpvq, and thus v2, to be even, so d1 P vK X HW with d1 " md. If m ‰ 1, then xd1, uy " 2 forces m " 2, and, as d12 " ´v2, we see that 4 v2, so 2 v, contradicting the assumption that v was primitive. Therefore m " 1, and d " v ´ v2 u, as claimed. In any case, we define Rd to be the reflection of vK in the hyperplane xv,uy dK. That is, Rd is defined by the formula Rdpzq " z ´ 2xz, dy d " z ´ 2xz, uy xv, uy d2 d " z ´ 2xz, uy xv, uy xv, uy2 u, z P vK. 2pv ´ v2 xv,uy uq, when We will prove in the next result that in all of the cases mentioned above θv,σ differ by precom- v ` 2v2 xz, uy Observing that Rmdpzq " Rdpzq for any m P Z, we see that even if d " 1 xv, uy " 2 " ℓpuq and v is not primitive, the reflection remains the same. posing with the reflection Rd. Proposition 12.1. Let W be a wall for v P HalgpX, Zq with v2 ą 0, and for generic σ0 P W let σ be sufficiently close stability conditions in the opposite and adjacent chambers separated by W. Then we may identify PicpMσ`pv, Lqq with PicpMσ´pv, Lqq. Moreover, if (1) xv, wy " 0 for a spherical or exceptional class w P CW X HW , or (2) xv, uy " ℓpuq for an isotropic vector u P HW, or (3) xv, uy " 1 for an isotropic vector u P HW with ℓpuq " 2, then θv,σ` " θv,σ´ Rd, where vK X HW " Zd and Rd is the reflection of vK in the hyperplane dK. Otherwise, we have θv,σ` " θv,σ´. 70 HOWARD NUER AND K ¯OTA YOSHIOKA Suppose first that codimpMσ`pv0, L0qzM s Proof. We claim first that we may identify PicpMσ`pv, Lqq with PicpMσ´pv, Lqq. Indeed, by The- orem 5.8 and Propositions 6.20-6.22, Mσ`pv, Lq -- Mσ´pv, Lq if xv, wy " 0 for a spherical or exceptional w P CW X HW; in particular, the two moduli spaces are isomorphic when W is a totally semistable wall of types (TSS3) and (TSS4) and when W is a Brill-Noether wall. The two moduli spaces are also isomorphic when W is a wall of Hilbert-Chow or Li-Gieseker-Uhlenbeck type, as we showed in the proof of part (2) of Theorem 1.1. For walls of iLGU type, we proved that Mσ`pv, Lq -- Mσ´pv, Lq in Propositions 10.8 and 10.11. According to Theorem 5.8, Mσ`pv, Lq and Mσ´pv, Lq are isomorphic away from a locus of codimension at least two in all other cases. Now let us move on to proving the claims about the compatibility of the θ-maps. As usual, let v0 be the minimal Mukai vector in the GH-orbit of v, and if v P Cn, then we consider the composition of spherical or exceptional twists Φ giving the isomorphism Mσpvq Ñ Mσpv0q (resp., Mσpvq Ñ Mσ¯pv0q) if n is even (resp., if n is odd). σ0pv0, L0qq ě 2. Then the two moduli spaces Mσpv0, L0q share a common open subset on which the universal families agree. As the complement of M s σ0pv0, L0q has codimension at least two and the maps θv0,σ are determined by their restriction to curves in this open subset, we get θv0,σ` " θv0,σ´. For v, the autoequivalence giving the birational map Mσ`pv, Lq 99K Mσ´pv, Lq is induced by pΦ´q´1 Φ`. As the classes of the spherical/exceptional objects occuring in Φ` and Φ´ are identical, this autoequivalence does not change the class of the universal family in the K-group. Therefore, we again have θv,σ` " θv,σ´. σ0pv0, L0qq " 1. Then by Propositions 8.8 and 8.18 and Lemma 6.2, we must have either xv0, wy " 0 for a spherical or exceptional class w P CW X HW or xv0, uy " ℓpuq for a primitive isotropic u P HW . induces an isomorphism Mσ`pv0, L0q Ñ Mσ´pv0, L0q by Propositions 6.21 and 6.22. Considering the classes of the uni- versal families in the K-group we see that θv0,σ` " θv0,σ´ Rw. Similarly, we have seen that i Φ` induces an isomorphism Mσ`pv, Lq Ñ Mσ´pv, Lq for i " 0 or 1, so taking the pΦ´q´1 RT ` classes of the universal families in the K-group we see that In the first case, we have that w " w0 or w1 and RT ` Now suppose that codimpMσ`pv0, L0qzM s or RT ` 1 0 θv,σ` " θv,σ´ ppΦ´q´1 Rw pΦ`qq " θv,σ´ R pΦ`q´1 pwq , because pΦ`q " pΦ´q, as we noted above, and Φ RT Φ´1 " RΦpTq for any autoequivalence Φ and spherical/exceptional object T [21, Lemma 8.21]. Setting w1 :" pΦ`q´1 pwq, we get θv,σ` " θv,σ´ Rw1 for the spherical/exceptional w1 P CW X HW such that xv, w1y " 0, as claimed. In the second case, [6] or the arguments in section 10 showed that θv0,σ` " θv0,σ´ Rd. As in the previous cases, we see that giving the claim for v. θv,σ` " θv,σ´ R pΦ`q´1 pdq , Finally, we must consider when M s σ0pv0, Lq " ∅. By Theorem 5.8, we then have v " v0, and either W is a Hilbert-Chow wall or xv, wy " 0 for a spherical or exceptional w P CW X HW and xv, uy " ℓpuq for primitive isotropic u P HW . These latter cases have already been covered above, so it only remains to consider the case of a Hilbert-Chow wall. We saw in the proof of part (2) of Theorem 1.1 that Φp q " p q_r2s induces an isomorphism between Mσ`pv, Lq and Mσ´pv, Lq, where we may assume that v " ´p1, 0, 1 2 ´ nq, u " p0, 0, 1q, and L " OX. Then vK is spanned by ´d " p1, 0, n ´ 1 2q and classes of the form p0, c, 0q for c P NumpXq. As the universal family E´ of Mσ´pv, Lq is given by E_`r2s, where E` is the universal family of Mσ`pv, Lq, we have by Grothendieck duality: In particular, θv,σ`pp1, 0, n ´ 1 θv,σ`pzq " ´θv,σ´pz_q. 2qq and 2qq " ´θv,σ´pp1, 0, n ´ 1 θv,σ`pp0, c, 0qq " ´θv,σ´pp0,´c, 0qq " θv,σ´pp0, c, 0qq. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 71 On the other hand, Rdpp1, 0, n ´ 1 indeed 2qq " ´p1, 0, n ´ 1 2q and Rdpp0, c, 0qq " p0, c, 0q, so we see that as required. θv,σ` " θv,σ´ Rd, (cid:3) Since ξσ0 P pHKWqR, so in particular ξσ0 P pdK X vKqR, we have (128) θv,σ`pξσ0q " θv,σ´pRdpξσ0qq " θv,σ´pξσ0q by Proposition 12.1. It follows that the maps ℓC agree along W and thus fit together to give a piece-wise analytic continuous map ℓ : Stab:pXq Ñ NumpMσ`pv, Lqq, 2 , 1 2s Ñ Stab:pXq with σp´ 1 2q " σ`, σp0q " σ0, and σp 1 as claimed. It is important to note that in any of the three cases enumerated in Proposition 12.1, the image under ℓ of a path σ : r´ 1 2q " σ` is a path contained entirely in the nef cone of Mσ`pv, Lq because of the action of the reflection Rd. Indeed, as σptq approaches σp0q " σ0, ℓpσptqq approaches ℓpσ0q, but then Rd causes ℓpσptqq, t ą 0, to bounce off the wall of NefpMσ`pv, Lqq containing ℓpσ0q and continue back into the interior. Note that while most of these bouncing walls correspond either to divisorial contractions or P1- fibrations, and thus extremal walls of MovpMσpv, Lqq, the Enriques case differs from the K3 case in that there are bouncing walls that induce small contractions. Indeed, in the presence of exceptional classes w P CW X HW X vK, W is a bouncing wall inducing a small contraction if v2 ě 3 as in Example 8.23. Remark 12.2. Recall that the contraction morphisms π : Mσpv, Lq Ñ M are defined by the linear systems nθv,σpξσ0q pn " 0q. From (128), these define morphisms π : Mσpv, Lq Ñ P such that im π` " im π´, and thus M` " M´, as the normalization of im π` " im π´. Let M1σ0pvq be the set of S-equivalence classes of σ0-semistable objects E with vpEq " v. Then im π is a subset of M1σ0pvq, which is a proper subset in general. 12.2. Proof of Corollary 1.4. Now we shall prove Corollary 1.4. For the computation of the Picard groups in odd rank cases, we can use deformation of Enriques surfaces, since θv,σ is well- defined for a relative moduli space over a family of Enriques surfaces. Thus in either case of Corollary 1.4, we may assume that PicpXq " PicprXq, where rX is the covering K3 surface. By Theorems 1.2 and 1.3 and Proposition 12.1, it is sufficient to prove that θv,σ is an isomorphism for a special pair of v and σ. Specifically, we may assume that v " p1, 0, 1 2 ´ nq or v " p2, D, aq with pD, ηq " 1 for a divisor η with pη2q " 0. In the first case, θv,σ is an isomorphism if Mσpv, Lq " HilbnpXq pn ě 2q (see the proof of [53, Corollary A.4]). So we shall treat the second case. Replacing v by v exppkηq, we may assume that D2 " ´2,´4. Since X is unnodal, we can take an ample divisor H with pD, Hq " 0. We set n :" D2 2 ` 1 ´ a. Then v " p2, D, D2 2 ` 1 ´ nq " vpOXpDqq ` vpIZq, IZ P HilbnpXq. We take β0 P PicpXqQ with pβ0, Dq " D2 2 ` n. Then χpOXpD ´ β0qq " χpIZp´β0qq. We take β P PicpXqQ in a neighborhood of β0 such that χpOXpD ´ βqq ă χpIZp´βqq. We take a stability condition σ such that Mσpv, Lq is the moduli space of β-twisted stable sheaves M β Hpv, Lq. For a non-trivial extension 0 Ñ OXpDq Ñ E Ñ IZ Ñ 0, Z P HilbnpXq, E is a β-twisted stable sheaf. Let us denote by Z the universal subscheme Z Ă X HilbnpXq and by pX, pHilbnpXq the projections of X HilbnpXq onto its first and second factors, respectively. Since H 0pX, OXpDqq " H 0pX, OXpD ` KXqq " 0, we see that V :" Ext1 X pIZ , pXOXpDqq is a locally free sheaf on HilbnpXq of rank n ´ 1 ´ D2 2 . For n ą 0, let P " PpV_q and let pHilbn 72 HOWARD NUER AND K ¯OTA YOSHIOKA π : P Ñ HilbnpXq be the projective bundle parameterizing non-trivial extensions of IZ by OXpDq with universal family of extensions 0 Ñ OXpDq b OPpλq Ñ E Ñ πIZ Ñ 0, (129) where OPpλq is the tautological line bundle on P . Hence we have a morphism ψ : P Ñ Mσpv, Lq. Lemma 12.3. θv,σ is an isomorphism if v2 ě 4. Proof. We note that v2 " ´D2 ` 4pn ´ 1q. Hence v2 ě 4 if and only if n ě 2, or n " 1 and D2 " ´4. We fix η P NSpXq with pη, Dq " 1. Then we have a decomposition NSpXq " DK ' Zη. For ξ P NSpXq, let uξ P KpXq be an element of KpXq such that vpuξq " p0, ξ, 0q and c1puξq " ξ. Any u P KpXq can be written u " rOX ` sCx ` uξ, so from xv, vpuηqy " 1 it follows that (130) Hence any u P KpXqv is written as (131) u ´ xv, vpuqyuη " rOX ` sCx ` uξ´pξ,Dqη ` prp2 ` D2 2 ´ nq ` 2squη P KvpXq. u " rOX ` sCx ` uξ ` prp2 ` D2 2 ´ nq ` 2squη, r, s P Z, ξ P DK. For any α P KpXq, we set θ0pαq :" det pHilbnpXq!pIZ b pXpα_qq. Then θ0pCpq " OHilbnpXq. We have an injective homomorphism NSpXq Ñ NSpHilbnpXqqp -- PicpHilbnpXqqq by sending ξ P NSpXq to c1pθ0puξqq. We regard NSpXq as a subgroup of NSpHilbnpXqq by this homomorphism. If n ě 2, then NSpHilbnpXqq " Zδ ' NSpXq with δ :" c1pθ0pOXqq. From the exact sequence (129) and [22, Lemma 8.1.2], we see that writing any u P KpXqv as in (131), we have c1ppP !pE b pXpu_qqq " ´ xvpOXpDqq, vpuqyλ ` πc1pθ0puqq (132) "rpδ ` pn ´ 1qλ ` p2 ` D2 2 ´ nqηq ´ spλ ´ 2ηq ` ξ, where we have surpressed the π in the second equality since π : NSpHilbnpXqq Ñ NSpPq is injective and NSpPq " NSpXq ' Zδ ' Zλ " DK ' Zη ' Zδ ' Zλ. By [22, Theorem 8.1.5], the homomorphism KpXqv Ñ NSpPq in (132) is precisely c1 composed with the pull-back along the classifying morphism ψ of the universal Donaldson-Mukai map θv,σ, it follows from the formula in (132) that c1 ψ θv,σ : KpXqv Ñ PicpM β Hpv, Lqq Ñ PicpPq Ñ NSpPq is injective (up to torsion) and its image is a direct summand of NSpPq. If n " 1 and D2 " ´4, then rk V " n` 1 " 2 and NSpPq " NSpXq' Zλ. Since 2` D2 2 ´ n " ´1, c1ppP !pE b pXpu_qqq " ´rη ´ spλ ´ 2ηq ` ξ. (133) Hence c1 ψ θv,σ is an isomorphism. In either case, it follows that θv,σpKpXqvq is a direct summand of PicpMσpv, Lqq. Since the torsion submodule of PicpMσpv, Lqq is isomorphic to Z{2Z and rk PicpMσpv, Lqq " rk KpXqv ([44, Theorem 5.1]), θv,σ is an isomorphism. (cid:3) 13. Appendix: Some non-normal moduli spaces of sheaves In this final section, we shall describe the two fundamental outlying examples of moduli spaces of sheaves (Bridgeland semistable objects) on Enriques surfaces. Indeed, by Proposition 3.9 for σ P Stab:pXq generic with respect to v satisfying v2 ą 0, Mσpvq is normal unless (1) v " 2v0 with v2 0 " 1, or (2) v2 " 2 and X is nodal. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 73 Moreover, these are precisely the exceptions to M s σpvq having torsion canonical divisor and Goren- stein, terminal, l.c.i. singularities [40, Theorem 8.2]. Similarly, we have seen in Theorem 3.8 when v is primitive Mσpv, Lq is irreducible if X is unnodal and can only be reducible for nodal X if v2 " 2. We will prove here that the failure of normality in these two cases is due to the presence of multiple irreducible components. In the first case, when v " 2v0 with v2 0 " 1, we describe a conneected component of Mσpv, Lq for L " KX pmod 2q consisting of two irreducible components. In the second case, when v2 " 2, we describe all of Mσpv, Lq for L " D ` r 2 KX pmod 2q, which consists In both cases, we also describe the other component of precisely two irreducible components. parametrizing objects with the other determinant. 13.1. Some components of Mσpv, Lq when v " 2v0 with v2 0 " 1. By [48, Theorem 4.6], there exists a Fourier-Mukai transform Φ : DbpXq Ñ DbpXq such that Φpv0q " p1, 0,´ 1 2q. By Theorem 1.1, it follows that Mσpvq is birational to the moduli space Mωp2, 0,´1q of ω-Gieseker semistable sheaves of Mukai vector 2p1, 0,´ 1 2q " p2, 0,´1q for a generic ample divisor ω. In partic- ular, to understand the number of irreducible components, we are reduced to analyzing the moduli space Mωp2, 0,´1q. 13.1.1. A connected component of Mωpp2, 0,´1q, KXq. For the Mukai vector v " p2, 0,´1q, we shall describe a connected component of Mωpv, KXq, which is a refinement of [53, Remark 2.3]. involution ι. We set E0 :" OX ' OXpKXq and begin by concretely studying the singular locus of the stable locus of M s Recall that : rX Ñ X is the quotient map from the covering K3 surface rX with covering ! pIWpDqq IW P HilbnprXq, ιpDq " ´D, n " pD2q{2 ` 2, IWpDq fl ιpIWpDqq) , ωpv, KXq is singular at E, then EpKXq -- E, which implies E " pIWpDqq, where IW is the ideal sheaf of a 0-dimensional subscheme W of rX and D is a divisor on rX (cf. [46, Lemma 2.13]). By the stability of E, ιpIWpDqq fl IWpDq. Since ppIWpDqqq " IWpDq' IιpWqpιpDqq, it follows that D ` ιpDq " 0 and deg W " pD2q{2 ` 2. In particular, as pD, pωqq " pιD, ιpωqq " pιD, pωqq, we get pD, pωqq " 0, so either D " 0 or neither D nor ´D is effective. As rX is a K3 surface, we cannot have pD2q ě ´2, since then D or ´D would be effective, so we must in fact have pD2q ď ´4. From deg W ě 0, however, we get pD2q ě ´4. We conclude therefore that either D " 0 or pD2q " ´4, giving the description in (134). In the open subscheme of locally free sheaves, W " H, so the singular locus there is isolated, consisting of pOrXpDqq for the finitely many divisors D such that ιD " ´D and pD2q " ´4. If D " 0 and deg W " 2, then pIWq is a non-locally free sheaf with pIWq__ -- E0, which gives an irreducible component where D " 0 or pD2q " ´4. In particular, the singular locus consists of isolated singular points in the open subscheme of locally free sheaves, and an irreducible 4-dimensional subscheme. ωpv, KXq of stable sheaves, the singular locus is Proof. If M s Lemma 13.1. In the open subscheme M s ωpv, KXq. (134) ! pIWq P M s ωpv, KXq IW P Hilb2prXq) of the singular locus of dimension 4. (cid:3) From the lemma it follows that the singularities of the open subscheme M s,lf ω pv, KXq parametriz- ωpv, KXq are isolated 5-dimensional hypersurface singularities. There- ing locally free sheaves in M s fore, we get the following corollary: 74 HOWARD NUER AND K ¯OTA YOSHIOKA Corollary 13.2. The open subscheme M s,lf Mukai vector v and determinant KX is normal. ω pv, KXq parametrizing ω-stable locally free sheaves of Now we turn to the 4-dimensional component of the singular locus. We have an injective mor- phism X Ñ Hilb2prXq by sending Iz pz P Xq to pIzq " I´1pzq. The image is the ι-invariant Hilbert scheme Hilb2prXqι. We shall identify X with Hilb2prXqι. We also have a morphism which is a double covering to its image. If IW P Hilb2 sheaf and the fiber over pIWq is tIW , IιpWqu. If IW " I´1pzq, then rX zX, then pIWq is a stable non-locally free Hilb2prXq Ñ Mωpv, KXq IW ÞÑ pIWq, pIWq " ppIzqq " Iz b pOX ' OXpKXqq " Iz b E0, which is a properly semi-stable sheaf. For E " pIWq, we have ´χpE, Eq{2 " 2, and hence by using [46, Fact 2.4] and the paragraph ωpv, KXq is analytic locally defined by a preceeding [46, Lemma 2.13], we see that around E, M s hypersurface Fpt1, t2, ..., t6q " 0 in pC6, 0q such that Fpt1, t2, ..., t6q " nÿi"1 t2 i ` Gpt1, t2, ..., t6q, n ě 2, Gpt1, t2, ..., t6q P pt1, t2, ..., t6q3. Since the singular locus is 4-dimensional at E " pIWq, we must have n " 2. Therefore there are at most two irreducible components intersecting along the 4-dimensional singular locus. We shall prove that there are exactly two such irreducible components M0 and M1 and that their union is connected. Let ψ : Mωpv, KXq Ñ N be the contraction map to the Uhlenbeck compactification. Lemma 13.3. There is an irreducible component M0 of Mωpv, KXq such that M0 contains a µ- stable locally free sheaf and ψpM0q contains Mωpv1, KXq S2X, where v1 " vpE0q. Proof. Consider the stack Mωpv, KXqµss of µ-semistable sheaves E with vpEq " v and detpEq " KX , and let M0 be the irreducible component containing all locally free sheaves fitting in the short exact sequence (135) where IZ P Hilb2pXq. Then by the proof of [53, Lemma 2.8], 0 Ñ OX Ñ E Ñ IZpKXq Ñ 0, M10 :" t E P M0 E is a µ-stable locally free sheaf u is a nonempty open and dense substack of M0. Let M0 be the irreducible component of Mωpv, KXq containing the associated coarse moduli scheme of M10. We shall prove that ψpM0q contains Mωpv1, KXq S2X, where v1 " vpE0q. Let E P M0 be a locally free sheaf fitting in (136) where IZ P Hilb2pXq. Consider a point F P M10 and a generic curve T Ă M0 connecting E and F . This corresponds to a deformation E over the curve T such that Et0 " E and Et1 " F for points t0, t1 P T . In particular, as T is taken generic, we have Et is a µ-stable locally free sheaf for t ‰ t0. Note that E is µ-semistable but not Gieseker semistable, as 0 Ñ OX Ñ E Ñ IZpKXq Ñ 0, µpOXq " 0 " µpIZpKXqq and χpOXq " 1 ą ´1 " χpIZpKXqq, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 75 so E does not induce a morphism T Ñ M0. By Langton's theorem [22, Theorem 2.B.1], however, we obtain another family F over T of semistable sheaves Ft such that Ft " Et for t ‰ t0. This new family does induce a morphism T Ñ M0. Since ψ extends to the family of µ-semistable sheaves Et, (137) ψpFtq " ψpEt0q " pE0,rZsq P Mωpv1, KXq S2X, ψpFt0q " lim tÑt0 where rZs is the 0-cycle defined by Z. Hence our claim holds. (cid:3) For the construction of the other irreducible component M1, we prepare the following lemma. 0 Ñ IZ Ñ E Ñ OXpKXq Ñ 0, Lemma 13.4. For Z P Hilb2pXq and a non-trivial extension (138) E P Mωpv, KXq. Proof. Suppose not, and let F be a saturated, destabilizing, stable subsheaf of E. In particular, rkpFq " 1. As E is µ-semistable, µωpFq ď µωpEq, so we must have µωpFq " µωpEq and χpFq ą χpEq2 " 0. It follows from χpIZq " ´1 that the composition φ : F ãÑ E ։ OXpKXq must be nonzero so that from stability χpFq ď χpOXpKXqq " 1, and thus χpFq " χpOXpKXqq. Then φ is an isomorphism, contradicting the non-triviality of the extension (138). Hence E is semistable, as claimed. (cid:3) Observe that E as in (138) is not locally free and dualizing (138) twice we get another short exact sequence (139) 0 Ñ OX Ñ E__ Ñ OXpKXq Ñ 0. Moreover, we may put Eqs. (138) and (139) together into the following short exact sequence of complexes 0 ÝÝÝÝÑ IZ ÝÝÝÝÑ E ÝÝÝÝÑ OXpKXq ÝÝÝÝÑ 0 §§đ §§đ Id§§đ and apply the Snake lemma to see that E fits into another short exact sequence 0 ÝÝÝÝÑ OX ÝÝÝÝÑ E__ ÝÝÝÝÑ OXpKXq ÝÝÝÝÑ 0 0 Ñ E Ñ E__ Ñ OZ Ñ 0. (140) As Ext1pOXpKXq, OXq " 0, we see that E__ " OX ' OXpKXq, so we may compose the natural embedding OXpKXq ãÑ OX ' OXpKXq with the surjection E__ ։ OZ to obtain a morphism OXpKXq Ñ OZ . We will show in the next lemma that we may determine when E as in (138) is in M s ωpv, KXq based on the surjectivity of the associated morphism OXpKXq Ñ OZ . Lemma 13.5. A sheaf E fitting into the short exact sequence (138) is stable if and only if the associated morphism OXpKXq Ñ OZ is surjective. Proof. The sheaf E is properly semistable if and only if it contains a torsion free subsheaf of Mukai vector p1, 0,´ 1 2q, that is, either the ideal sheaf of a point Iz or its twist IzpKXq. As HompIz, IZq " 0 " HompIz, OXpKXqq, we see that E is properly semistable if and only if E contains Iz1pKXq for some z1 P X. But this is equivalent to E sitting in a short exact sequence (141) for some other point z2 P X. 0 Ñ Iz1pKXq Ñ E Ñ Iz2 Ñ 0 Taking double duals of (141), we again get a short exact sequence of complexes 0 ÝÝÝÝÑ Iz1pKXq ÝÝÝÝÑ E ÝÝÝÝÑ Iz2 ÝÝÝÝÑ 0 §§đ §§đ §§đ 0 ÝÝÝÝÑ OXpKXq ÝÝÝÝÑ E__ ÝÝÝÝÑ OX ÝÝÝÝÑ 0, 76 HOWARD NUER AND K ¯OTA YOSHIOKA so using the Snake lemma we see that, as claimed, the composition OXpKXq ãÑ E__ ։ OZ is not surjective, factoring instead through the proper subsheaf Oz1 Ĺ OZ . Note that we also see that Z " tz1, z2u. Conversely, if the composition OXpKXq ãÑ E__ ։ OZ is not surjective, then it factors through Oz1 for some z1 P Z, and we get a short exact sequence of complexes 0 ÝÝÝÝÑ OXpKXq ÝÝÝÝÑ E__ ÝÝÝÝÑ OX ÝÝÝÝÑ 0 §§đ §§đ §§đ for z2 P Z. It follows that E sits in a short exact sequence as in (141). 0 ÝÝÝÝÑ Oz1 ÝÝÝÝÑ OZ ÝÝÝÝÑ Oz2 ÝÝÝÝÑ 0 (cid:3) We have seen that any E fitting into (138) fits into (140). We show that the converse holds as well if E is stable. Lemma 13.6. Let E be a stable sheaf fitting into a short exact sequence as in (140), 0 Ñ E Ñ OX ' OXpKXq Ñ OZ Ñ 0 for Z P Hilb2pXq. Then E fits into a short exact sequence as in (138). Proof. Indeed, if E is a stable sheaf fitting into (140), then the stability of E and OXpKXq forces p2ÝÑ OXpKXq is surjective, where p2 is the projection onto the composition E ãÑ OX ' OXpKXq the second factor. Then similar arguments as above show that the kernel of this composition is IZ , giving the lemma. (cid:3) Now we are able to define the second irreducible component M1. Since hompOXpKXq, IZq " 0 and ext2pOXpKXq, IZq " hompIZ , OXq " 1, we see that so we have a family of semistable non-locally free sheaves ext1pOXpKXq, IZq " xp1, 0, 1 2q,p1, 0,´ 3 2qy ` 1 " 2, (142) on the projective bundle 0 Ñ IZ b OPp1q Ñ E Ñ OPXpKXq Ñ 0 q : P :" PpExt1 ppOHilb2pXqXpKXq, IZqq Ñ Hilb2pXq, where Z is the universal family on Hilb2pXqX and p : Hilb2pXqX Ñ Hilb2pXq is the projection. By Lemma 13.4, E gives rise to a morphism (143) g : P Ñ Mωpv, KXq Ez. z ÞÑ We set M1 :" gpPq, and prove in the next result that M1 is an irreducible component of Mωpv, KXq. Lemma 13.7. M1 is an irreducible component of Mωpv, KXq and satisfies M1 " ψ´1pMωpv1, KXq S2Xq. Proof. We begin by showing that g is injective on the stable locus. Indeed, for a stable sheaf E in the image of g, Z is determined uniquely by E__{E so that the exact sequence (138) is uniquely determined by E. Now we prove that the set M pss ω pv, KXq " t Iz1 ' Iz2pKXq z1, z2 P X u Ă Mωpv, KXq of properly semistable polystable sheaves is contained in M1. First assume that Z " tz1, z2u pz1 ‰ z2q. If E is properly semistable, then by Lemma 13.5, the associated morphism OXpKXq Ñ OZ is not surjective, say at z1, so E fits into an exact sequence 0 Ñ Iz2pKXq Ñ E Ñ Iz1 Ñ 0, MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 77 giving the point Iz2pKXq'Iz1 of Mωpv, KXq. Note, moreover, that it follows from ext1pIz1 , Iz2pKXqq " 1 that g is injective if z1 ‰ z2. the induced morphism OXpKXq Ñ OZ is not surjective and E fits into an exact sequence If Z " tzu with a non-reduced structure and E is properly semistable, then again by Lemma 13.5, 0 Ñ IzpKXq Ñ E Ñ Iz Ñ 0, giving the point Iz ' IzpKXq. Thus M pss ω pv, KXq Ă M1 as claimed. Note further that the point IzpKXq ' Iz of Mωpv, KXq is independent of the choice of a scheme structure on Z, so we also see that g´1pIz ' IzpKXqq has positive dimension. As g is injective away from q´1p∆Xq, where ∆X Ă Hilb2pXq is the locus of non-reduced sub- schemes, it follows that dimpM1q " 5. As M1 is proper, irreducible, and of maximal dimension, it is an irreducible component of Mωpv, KXq. Since M pss ω pv, KXq Ă M1 Ă ψ´1pMωpv1, KXq S2Xq, to prove the second claim, we must show ωpv, KXq satisfying ψpEq P Mωpv1, KXq S2X is in M1. But such an E is a stable that E P M s sheaf fitting into an exact sequence as in (140). It follows from Lemma 13.6 that E fits into the exact sequence (138). Therefore E P M1, as claimed. (cid:3) We will show that M0 Y M1 is a connected component of Mωpv, KXq, but the first step is to show that M1 does not meet any other components. In the next lemma, we show that away from the locus of properly semistable sheaves S-equivalent to Iz ' IzpKXq, M1 is normal and thus meets no other components. Lemma 13.8. The local deformation space is smooth at E if E is S-equivalent to Iz1 ' Iz2pKXq (z1 ‰ z2). Hence Iz1 ' Iz2pKXq (z1 ‰ z2) is contained in the normal open subscheme Mωpv, KXq :" E P Mωpv, KXq Ext2pE, Eq " 0( of Mωpv, KXq. Proof. We begin by showing that Mωpv, KXq is normal. Let Q be an open subscheme of a suitable quot-scheme QuotOXp´mHq'N{X, which parameterizes quotients OXp´mHq'N Ñ E, such that Mωpv, KXq is a GIT-quotient of Q by P GLpNq. Let Q be the open subscheme of Q such that Ext2pE, Eq " 0. Then Q is smooth. Hence the open subscheme Mωpv, KXq " Q{P GLpNq of Mωpv, KXq is a normal scheme. Now suppose that E is S-equivalent to Iz1 ' Iz2pKXq with z1 ‰ z2. Then Ext2pIz1 , Iz1q " Ext2pIz1 , Iz2pKXqq " Ext2pIz2pKXq, Iz1q " Ext2pIz2pKXq, Iz2pKXqq " 0. Hence Ext2pE, Eq " 0 so that E P Q. It follows that the polystable representative, Iz1 ' Iz2pKXq, is in Mωpv, KXq, as claimed. (cid:3) We can actually prove that Mωpv, KXq is smooth: Lemma 13.9. P Mωpv,KXq Mωpv, KXq Ñ Mωpv, KXq is a closed immersion whose image is a connected component of Mωpv, KXq. In particular Mωpv, KXq is smooth. Proof. We first show that the image of g1 : P Mωpv,KXq Mωpv, KXq Ñ Mωpv, KXq a connected component of Mωpv, KXq. First observe that g1 is proper since g is a proper morphism. We further note that the image of g1 is M1 :" M1 X Mωpv, KXq, which is therefore a connected component of Mωpv, KXq by Lemma 13.8. ω pv, KXq is S-equivalent to Iz'IzpKXq, then we see that Ext2pE, Eq ‰ 0. Hence E is not contained in Mωpv, KXq. Therefore g1 is injective, which implies that g1 is a finite birational map between normal schemes. Therefore, g1 is an isomorphism onto its image, proving the claim. Now let us show that g1 is a closed immersion. If E P M pss 78 HOWARD NUER AND K ¯OTA YOSHIOKA As M pss ω pv, KXq Ă M1, the open subscheme Mωpv, KXqzM1 is smooth by [25] and the definition of Mωpv, KXq. Since M1 is isomorphic to an open subscheme of a P1 bundle over Hilb2pXq, it is smooth, which gives the final statement of lemma. (cid:3) Finally, we put everything together to show that M0YM1 is a connected component of Mωpv, KXq. (1) The singular locus of Mωpv, KXq consists of a 4-dimensional sub- Proposition 13.10. scheme and a finite set of points S1 :" tpIWq IW P Hilb2prXqu S2 :" tpOrXpDqq ιpDq " ´D,pD2q " ´4u. (2) M0 and M1 intersect along S1. (3) M0 Y M1 is a connected component of Mωpv, KXq. Proof. (1) follows from Lemma 13.1 and Lemma 13.9. By Lemma 13.3 and Lemma 13.7, M0 intersects M1 along the 4-dimensional singular locus. By the description of the singular locus, M0 Y M1 is a connected component of Mωpv, KXq. (cid:3) Remark 13.11. The restriction of ψ to S1 " M0 X M1 is a double cover onto Mωpv1, KXq S2X. Indeed, for pIw1,w2q pw1 ‰ w2, ιpw2qq, we push forward the short exact sequence on rX 0 Ñ Iw1,w2 Ñ OrX Ñ Ow1,w2 Ñ 0 to obtain the short exact sequence 0 Ñ pIw1,w2q Ñ OX ' OXpKXq Ñ Opw1q,pw2q Ñ 0. From this it is clear that ψ´1pψppIw1,w2qqq " tpIw1,w2q, pIw1,ιpw2qqu. 13.1.2. A remark on Mωpv, 0q pv " p2, 0,´1qq. We discuss here the structure of the other compo- nent of Mωpvq parametrizing sheaves with determinant OX vis-a-vis the Uhlenbeck contraction. Lemma 13.12. If E P Mωpv, 0q is not a µ-stable locally free sheaf, then E is S-equivalent to Iz1 ' Iz2 or Iz1pKXq ' Iz2pKXq. Proof. If E is not locally free, then we see that χpE__q ą 0. Hence H 0pE__q ‰ 0 or H 0pE_pKXqq ‰ 0. It follows from considerations of µ-stability that E__ contains OXpDq, where D " 0 or KX , respectively, as a saturated subsheaf. Intersecting with E gives an exact sequence 0 Ñ IZpDq Ñ E Ñ IWpDq Ñ 0 with D " 0, KX . By the semi-stability of E, degpZq ě 1, degpWq ď 1, and degpZq ` degpWq " 2. If degpZq " degpWq " 1, then E is S-equivalent to IZpDq ' IWpDq, as claimed. If deg W " 0, then E__ " OXpDq'2 with E__{E -- OZ where now Z P Hilb2pXq. Choosing a torsion free sheaf F with E Ă F Ă E__ and F{E " Oz1, F is µ-semistable and χpFq " 1, so the same argument as for E__ shows that HompOXpDq, Fq ‰ 0. Slope-stability considerations force OXpDq to be a saturated subsheaf, and intersecting with E gives another short exact sequence 0 Ñ IZ 1pDq Ñ E Ñ IW 1pDq Ñ 0 with degpZ1q ě 1, degpW 1q ď 1, and degpZ1q ` degpW 1q " 2. This time, however, we see that 0 ‰ OXpDq{IZ 1pDq " pOXpDq ` Eq{E Ă F{E " Oz1 , so Z1 " z1 and W 1 " z2 for points z1, z2 P X. Therefore E is S-equivalent to Iz1pDq ' Iz2pDq in this case as well. (cid:3) Thus the complement of the open subset of µ-stable locally free sheaves in Mωpv, 0q is parame- terized by two copies of S2X, and this locus is not contracted by the morphism to the Uhlenbeck compactification. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 79 13.2. Irreducible components of Mσpvq with v2 " 2. In this subsection, we shall prove the following claim. Proposition 13.13. Let v :" pr, ξ, aq be a Mukai vector such that v2 " 2. Assume that L " D ` r 2 KX pmod 2q, where D is a nodal cycle. Then Mωpv, Lq has two irreducible components. epY q "řp,q ep,qpY qxpyq, where ep,qpY q :"řkp´1qkhp,qpH k As in the previous subsection, we begin by reducing to studying a certain moduli space of Gieseker semistable sheaves of rank two. To do so, we make use of a motivic invariant called the virtual Hodge polynomial. Recall that for a variety Y , the virtual Hodge polynomial epY q is defined by c pY qq are the virtual Hodge numbers of the natural mixed Hodge structure on the cohomology of Y with compact support (see [52] and the references therein for a more complete explanation). Lemma 13.14. Let v :" pr, ξ, aq be a Mukai vector such that v2 is even. (1) There is v1 " p2, L1, a1q such that epMωpv, Lqq " epMωpv1, L1qq, where L1 ` KX " L` r (2) For a smooth rational curve C, we have an equality pmod 2q. 2 KX (144) epMωpv, Lqq " epMω1pv1, L1qq, where σ1 " Φpσq for the spherical twist Φ " ROCp´1q, v1 " Φpvq " pr, ξ ` pC, ξqC, aq and L1 " L ` pL, CqC. Proof. (1) Since the p´1q-reflection ROX associated to OX preserves Stab:pXq, we have an isomor- phism (145) where σ1 " ROXpσq, v1 " ROXpvq " p´2a, ξ,´r{2q and L1 " L ` r`2a (146) Mσpv, Lqq -- Mσ1pv1, L1q, epMωpv, Lqq " epMω1pv1, L1qq. By similar arguments as in [39] or [53], we get (1). 2 KX. By [39], we have (2) Let Φ : DbpXq Ñ DbpXq be the twist functor associated to the spherical object OCp´1q. The action of Φ preserves Stab:pXq, so we have an isomorphism (147) where σ1 " Φpσq, v1 " pr, ξ ` pC, ξqC, aq and L1 " L ` pL, CqC. By using [39] again, we get (144). Mσpv, Lq -- Mσ1pv1, Lq (cid:3) In the specific case that we are interested in, we can in fact get more explicit: Lemma 13.15. Let v " pr, ξ, aq be a Mukai vector with v2 " 2 and L a divisor such that L " 2 KX pmod 2q, where D is a nodal cycle. Then there is a Mukai vector v1 " p2, ξ1, 0q and an D ` r elliptic fibration π : X Ñ P1 such that (1) epMωpv, Lqq " epMω1pv1, L1qq, (2) pξ1, Cq " 2 for a general fiber C of π and L1 " D1 ` KX pmod 2q, where D1 is a nodal cycle. Proof. For a primitive Mukai vector v such that v2 " 2, we take v1 " p2, ξ1, a1q and L1 satisfying Lemma 13.14. As ℓpvq " 1, we can find an isotropic divisor η such that pξ1 ` 2λ, ηq " 1, where λ P NSpXq. Replacing v1 by v1ekη`λ, for an appropriate choice of k, we may assume that pξ12q " 2 and a1 " 0. Then ξ1 is effective or ´ξ1 is effective. Since v1e´ξ1 " p2,´ξ1, a1q, we may assume that ξ1 is effective. Since L " D ` r 2 KX pmod 2q for a nodal cycle D, by using Lemma 13.16 and (144), we may assume that ξ1 is nef and L1 " D1 ` KX pmod 2q for a nodal cycle D1. As pξ12q " 2, ξ1 is primitive, so we may find an isotropic divisor η with pξ1, ηq " 1. The Riemann- Roch theorem then implies that η is effective. Let η " f `ři Ci be a decomposition of η such that 80 HOWARD NUER AND K ¯OTA YOSHIOKA f is a nef and isotropic divisor, and Ci are smooth rational curves. Then pξ1, fq,pξ1, Ciq ě 0 imply that pf, ξ1q " 1 by using Hodge index theorem. The linear system 2f induces an elliptic fibration π : X Ñ P1 satisfying our requirements. Lemma 13.16. For nodal cycles D and D1, D ` pD, D1qD1 " D2 pmod 2q, where D2 is another nodal cycle. Proof. For nodal cycles D and D1 and an ample divisor H, let E be a H-stable vector bundle with v " vpEq " p2, D ` KX , 0q and E1 a H-stable vector bundle with v1 " vpE1q " p2, D1 ` KX , 0q. Then for EpnHq pn " 0q, (cid:3) F :" kerpHompE1, EpnHqq b E1 Ñ EpnHqq is a stable vector bundle with vpFq " ´enHp2, D ` KX , 0q ´ xenH v, v1yv1 ([49, Thm. 1.7]). Thus vpFq2 " ´2. Since (148) xenH v, v1y " pD, D1q ´ 2n2pH 2q ` 2npH, D1 ´ Dq, we get (149) rk F 2 "1 ` pD, D1q pmod 2q, c1pFq "D ` KX ` pD, D1qpD1 ` KXq pmod 2q "D ` pD, D1qD1 ` KX pmod 2q. 2 rk F As F is thus a stable spherical bundle, it follows that there is a nodal cycle D2 such that D2 ` and the result follows. rk F 2 KX " c1pFq " D ` pD, D1qD1 ` rk F 2 KX pmod 2q, (cid:3) By Lemma 13.15, it is sufficient to describe the irreducible components of Mωpv, Lq for v " p2, ξ, 0q with pξ2q " 2 and L such that L " D ` KX pmod 2q and pL, Cq " 2 for a general fiber C of an elliptic fibration π : X Ñ P1. For this purpose, we first describe a 2-dimensional component of the singular locus of Mωpv, Lq. Lemma 13.17. For a Mukai vector v " p2, ξ, 0q with pξ2q " 2 and a divisor L with L " D ` KX pmod 2q, the singular locus of Mωpv, Lq is (150) ! pIWprLqq IW P HilbnprXq, ιprLq " pLq ´rL, n " prL2q{2 ` 1) , E P Mωpv, Lq Ext2pE, Eq ‰ 0( " t E P Mωpv, Lq E -- EpKXq u . where prL2q " 0,´2. In particular, the 2-dimensional component of the singular locus is irreducible. If EpKXq -- E, then we see that E -- pIWprLqq, where ιpIWprLqq fl IWprLqq and W " ∅ or W " twu. If W " twu, then E__ " pOrXprLqq is a spherical vector bundle with vpE__q " p2, ξ, 1q. Remark 13.18. For x P X, we set ´1pxq " tz, ιpzqu. Then pIzprLqq and pIιpzqprLqq are not locally free at x. Hence the 2-dimensional component of the singular locus is a double covering of X. ωpv, Lq is of expected dimension, so the singular locus is Proof. Since ω is generic, Mωpv, Lq " M s Therefore the claim holds. (cid:3) The existence of a two dimensional component of the singular locus has the following important consequence for the reducibility of Mωpv, Lq. Lemma 13.19. There are at most two irreducible components of Mωpv, Lq. MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 81 Proof. If there are two irreducible components of Mωpv, Lq, then the connectedness of Mωpv, Lq implies they intersect along the 2-dimensional component of the singular locus. By [46], the analytic germ of Mωpv, Lq at a singular point E is described as a hypersurface Fpx1, x2, x3q " 0 in pC3, 0q If Mωpv, Lq is reducible, then each irreducible component is with a non-trivial quadratic term. defined by a factor of Fpx1, x2, x3q. Therefore Mωpv, Lq has at most two irreducible components. (cid:3) Let f be the reduced part of a multiple fiber of π and E0 P Mωpv1, L ´ 2fq be a spherical vector bundle, where v1 " p2, ξ ´ 2f, 0q. We shall prove Proposition 13.13 by constructing two irreducible components M0 and M1 of Mωpv, Lq. We begin by constructing a component M0 containing locally free sheaves using elementary transformations as in [22, Section 5.2]. To this end we must study how E0 restricts to a general fiber of π. Lemma 13.20. Let C be a general fiber of the elliptic fibration. Then E0C -- OCppq ' OCpqq pp ‰ qq. Proof. For the primitive and isotropic Mukai vector u :" p0, C, 1q, ℓpuq " 2 so that Mωpu, Cq is a fine moduli space which is isomorphic to X, where ω is general. As in [11], there is an elliptic fibration Mωpu, Cq Ñ P1 that comes from regarding Mωpu, Cq as a smooth compactification of the relative Picard scheme Pic1pX{P1q Ñ P1 of degree 1. We shall identify Mωpu, Cq Ñ P1 with the elliptic fibration X Ñ P1. XÑX : DbpXq Ñ DbpXq. Then F :" ΦE _ XÑXpE0qr1s is a purely 1-dimensional sheaf whose support is a double cover of P1. Thus for a general fiber C with FC " Cp ' Cq, one can check that E0C -- OCppq ' OCpqq. For a universal family E of Mωpu, Cq, we have a relative Fourier-Mukai transform ΦE _ (cid:3) Let L be a line bundle on a smooth fiber C P 2f with vpLq " p0, 2f, 0q. We set L :" OXpL, OXq. Then L is a line bundle of degree 0 on C so that vpLq " p0, 2f, 0q. Observe that Ext1 xvpE_0 q, vpLqy " xp2, 2f ´ ξ, 0q,p0, 2f, 0qy " ´2 ă 0, so there is necessarily a non-zero homomorphism ψ : E_0 Ñ L. Moreover we may assume that ψ is surjective by Lemma 13.20. We will see in the next two results that these give rise to a component of Mωpv, Lq containing µ-stable locally free sheaves. As E0 is µ-stable with respect to any polarization [24, 43], for any given starting polarization H0, tE0u " MH0pv1, L ´ 2fq. We will have cause to vary the polarization in the next lemma, but it is important to state that the stability of E0 remains unchanged. Lemma 13.21. For a non-zero homomorphism ψ : E_0 Ñ L, we set E :" RHomOXpConepψq, OXqr1s. Then E is a µ-stable torsion free sheaf with respect to H0 ` nf , n " 0. If ψ is surjective, then E is a locally free sheaf. Proof. We first show that E is torsion free, and locally free if ψ is surjective. Dualizing the exact triangle shifting by 1, and taking cohomology sheaves, we immediately see that E is a coherent sheaf fitting into an exact sequence ψ E_0 ÝÑ L Ñ Conepψq Ñ E_0 r1s, (151) Moreover, as RHompE, OXq " Conepψqr´1s, we get that Ext1pE, OXq " H0pConepψqq " cokerpψq, which is supporded in codimension two, and Ext2pE, OXq " 0, so E is torsion free by [22, Propo- sition 1.1.10]. Moreover, if ψ is surjective, so that cokerpψq " 0, then E is reflexive (by [22, ibid.]) and thus locally free. 0 Ñ E0 Ñ E Ñ L Ñ 0. 82 HOWARD NUER AND K ¯OTA YOSHIOKA Now we show that E is µ-stable. Let F be a subsheaf of E with rk F " 1. Then E0 X F is a rank 1 subsheaf of E0. Since E0 is µ-stable for any ample divisor, 2pc1pE0 X Fq, H0 ` nfq ă pc1pE0q, H0 ` nfq pn ě 0q. Hence 2pc1pE0 X Fq, fq ď pc1pE0q, fq " 1 and 2pc1pE0 X Fq, H0q ă pc1pE0q, H0q. In particular pc1pE0XFq, fq ď 0. As c1pFq " c1pE0XFq`2f , pc1pFq, fq ď 0. If n ą p4f, H0q, then 2pc1pFq, H0` nfq ă pc1pEq, H0 ` nfq. Therefore E is µ-stable with respect to H0 ` nf for n " 0, as claimed. (cid:3) Considering the irreducible component containing the sheaves E constructed in Lemma 13.21, we get the following result. Corollary 13.22. For ω " H0 ` nf (n " 0), there is an irreducible component M0 of Mωpv, Lq which contains a µ-stable locally free sheaf. The second irreducible component, M1, parametrizes E fitting into an exact sequence (152) 0 Ñ E Ñ E0pfq Ñ Cx Ñ 0, x P X. In particular, such E are not locally free, so M1 is indeed distinct from M0. Moreover, as in the previous subsection, M1 has the structure of a P1-bundle, this time over X (instead of Hilb2pXq). Moreover, the fibers of this P1 bundle are contracted by the Uhlenbeck morphism. By Lemma 13.19, M0 and M1 are the irreducible components of Mωpv, Lq, which shows Propo- sition 13.13 by Lemma 13.15. References [1] D. Abramovich and A. Polishchuk. Sheaves of t-structures and valuative criteria for stable complexes. J. Reine Angew. Math., 590:89 -- 130, 2006. [2] S¸aban Alaca and Kenneth S. Williams. Introductory algebraic number theory. Cambridge University Press, Cam- bridge, 2004. [3] D. Arcara, A. Bertram, I. Coskun, and J. Huizenga. The minimal model program for the Hilbert scheme of points on P2 and Bridgeland stability. Adv. Math., 235:580 -- 626, 2013. [4] D. Arcara and E. Miles. Projectivity of Bridgeland moduli spaces on del Pezzo surfaces of Picard rank 2. Int. Math. Res. Not. IMRN, 11:3426 -- 3462, 2017. [5] A. Bayer, M. Lahoz, E. Macr`ı, H. Nuer, A. Perry, and P. Stellari. Stability conditions in families. preprint, 2019. [6] A. Bayer and E. Macr`ı. MMP for moduli spaces of sheaves on K3s via wall-crossing: nef and movable cones, lagrangian fibrations. Invent. Math., 198(3):505 -- 590, 2014. [7] A. Bayer and E. Macr`ı. Projectivity and birational geometry of Bridgeland moduli spaces. J. Amer. Math. Soc., 27(3):707 -- 752, 2014. [8] Thorsten Beckmann. Birational geometry of moduli spaces of stable objects on Enriques surfaces. arXiv, 2018. [9] A. Bertram and I. Coskun. The birational geometry of the Hilbert scheme of points on surfaces. In Birational geometry, rational curves, and arithmetic, Simons Symp., pages 15 -- 55. Springer, Cham, 2013. [10] A. Bertram, C. Martinez, and J. Wang. The birational geometry of moduli spaces of sheaves on the projective plane. Geom. Dedicata, 173:37 -- 64, 2014. [11] T. Bridgeland. Fourier-Mukai transforms for elliptic surfaces. J. Reine Angew. Math., 498:115 -- 133, 1998. [12] T. Bridgeland. Stability conditions on triangulated categories. Ann. of Math., 166(2):317 -- 345, 2007. [13] T. Bridgeland. Stability conditions on K3 surfaces. Duke Math. J., 141(2):241 -- 291, 2008. [14] T. Bridgeland. An introduction to motivic Hall algebras. Adv. Math., 229(1):102 -- 138, 2012. [15] T. Bridgeland and A. Maciocia. Fourier-Mukai transforms for quotient varieties. arXiv:math/9811101, 1998. [16] T. Bridgeland and A. Maciocia. Complex surfaces with equivalent derived categories. Math. Z., 236(4):677 -- 697, 2001. [17] I. Coskun and J. Huizenga. The nef cone of the moduli space of sheaves and strong Bogomolov inequalities. arXiv, 12 2015. [18] I. Coskun, J. Huizenga, and M. Woolf. The effective cone of the moduli space of sheaves on the plane. arXiv:math/1401.1613, 2014. [19] A. Grothendieck. Cohomologie locale des faisceaux coh´erents et th´eor`emes de Lefschetz locaux et globaux (SGA 2), volume 4 of Documents Math´ematiques (Paris) [Mathematical Documents (Paris)]. Soci´et´e Math´ematique de France, Paris, 2005. S´eminaire de G´eom´etrie Alg´ebrique du Bois Marie, 1962, Augment´e d'un expos´e de Mich`ele MMP VIA WALL-CROSSING FOR MODULI SHEAVES ON ENRIQUES SURFACES 83 Raynaud. [With an expos´e by Mich`ele Raynaud], With a preface and edited by Yves Laszlo, Revised reprint of the 1968 French original. [20] D. Happel, I. Reiten, and S. Smalo. Tilting in abelian categories and quasitilted algebras. Mem. Amer. Math. Soc., 120(575):viii+ 88, 1996. [21] D. Huybrechts. Fourier-Mukai transforms in algebraic geometry. Oxford Mathematical Monographs. The Claren- don Press, Oxford University Press, Oxford, 2006. [22] D. Huybrechts and M. Lehn. The geometry of moduli spaces of sheaves. Cambridge Mathematical Library. Cambridge University Press, Cambridge, second edition, 2010. [23] M. Inaba. Toward a definition of moduli of complexes of coherent sheaves on a projective scheme. J. Math. Kyoto Univ., 42(2):317 -- 329, 2002. [24] H. Kim. Exceptional bundles on nodal Enriques surfaces. Manuscripta Math., 82(1):1 -- 13, 1994. [25] H. Kim. Moduli spaces of bundles mod Picard groups on some elliptic surfaces. Bull. Korean Math. Soc., 35(1):119 -- 125, 1998. [26] J. Koll´ar and S. Mori. Birational geometry of algebraic varieties, volume 134 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 1998. With the collaboration of C. H. Clemens and A. Corti, Translated from the 1998 Japanese original. [27] J. Koll´ar, K. Smith, and A. Corti. Rational and nearly rational varieties, volume 92 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 2004. [28] W. LeVeque. Fundamentals of number theory. Dover Publications, Inc., Mineola, NY, 1996. Reprint of the 1977 original. [29] C. Li and X. Zhao. The minimal model program for deformations of Hilbert schemes of points on the projective plane. Algebr. Geom., 5(3):328 -- 358, 2018. [30] Jun Li. The first two Betti numbers of the moduli spaces of vector bundles on surfaces. Comm. Anal. Geom., 5(4):625 -- 684, 1997. [31] M. Lieblich. Moduli of complexes on a proper morphism. J. Algebraic Geom., 15(1):175 -- 206, 2006. [32] J. Lo and Z. Qin. Mini-walls for Bridgeland stability conditions on the derived category of sheaves over surfaces. Asian J. Math., 18(2):321 -- 344, 2014. [33] Jason Lo. On some moduli of complexes on k3 surfaces. arXiv, 03 2012. [34] A. Maciocia and C. Meachan. Rank 1 Bridgeland stable moduli spaces on a principally polarized abelian surface. Int. Math. Res. Not. IMRN, 9:2054 -- 2077, 2013. [35] E. Macr`ı, S. Mehrotra, and P. Stellari. Inducing stability conditions. J. Algebraic Geom., 18(4):605 -- 649, 2009. [36] H. Minamide, S.Yanagida, and K. Yoshioka. The wall-crossing behavior for Bridgeland's stability conditions on abelian and K3 surfaces. J. Reine Angew. Math., 735:1 -- 107, 2018. [37] H. Minamide, S. Yanagida, and K. Yoshioka. Some moduli spaces of Bridgeland's stability conditions. Int. Math. Res. Not. IMRN, -(19):5264 -- 5327, 2014. [38] H. Nuer. MMP and Wall-Crossing for Bridgeland moduli spaces on bielliptic surfaces. preprint, 2018. [39] Howard Nuer. A note on the existence of stable vector bundles on Enriques surfaces. Selecta Math. (N.S.), 22(3):1117 -- 1156, 2016. [40] Howard Nuer. Projectivity and birational geometry of Bridgeland moduli spaces on an Enriques surface. Proc. Lond. Math. Soc. (3), 113(3):345 -- 386, 2016. [41] K. Oguiso and S. Schroer. Enriques manifolds. J. Reine Angew. Math., 661:215 -- 235, 2011. [42] Martin Olsson. Algebraic spaces and stacks, volume 62 of American Mathematical Society Colloquium Publica- tions. American Mathematical Society, Providence, RI, 2016. [43] Z. Qin. Chamber structures of algebraic surfaces with Kodaira dimension zero and moduli spaces of stable rank two bundles. Math. Z., 207(1):121 -- 136, 1991. [44] G. Sacca. Relative compactified jacobians of linear systems on enriques surfaces. Trans. AMS, 2018. [45] Y. Toda. Moduli stacks and invariants of semistable objects on K3 surfaces. Adv. Math., 217(6):2736 -- 2781, 2008. [46] K. Yamada. Singularities and Kodaira dimension of moduli scheme of stable sheaves on Enriques surfaces. Kyoto J. Math., 53(1):145 -- 153, 2013. [47] S. Yanagida and K. Yoshioka. Bridgeland's stabilities on abelian surfaces. Math. Z., 276(1-2):571 -- 610, 2014. [48] K. Yoshioka. Twisted stability and Fourier-Mukai transform I. Compositio Math., 138(3):261 -- 288, 2003. [49] K. Yoshioka. Stability and the fourier-mukai transform ii. Compositio Math., 145:112 -- 142, 2009. [50] K. Yoshioka. Bridgeland's stability and the positive cone of the moduli spaces of stable objects on an abelian surface. Adv. Stud. Pure Math., 69:473 -- 537, 2016. [51] K. Yoshioka. Some remarks on Bridgeland stability conditions on K3 and Enriques surfaces. arXiv, 07 2016. [52] K. Yoshioka. A note on stable sheaves on Enriques surfaces. Tohoku Math. J. (2), 69(3):369 -- 382, 2017. [53] K. Yoshioka. Moduli spaces of stable sheaves on Enriques surfaces. Kyoto J. Math., 58:865 -- 914, 2018. 84 HOWARD NUER AND K ¯OTA YOSHIOKA H.N.: Department of Mathematics, Statistics, and Computer Science, University of Illinois at Chicago, 851 S. Morgan Street Chicago, IL 60607 E-mail address: [email protected] K.Y.: Department of Mathematics, Faculty of Science, Kobe University, Kobe, 657, Japan E-mail address: [email protected]
1108.2668	1	1108	2011-08-12T17:28:36	Some metric properties of spaces of stability conditions	[ "math.AG" ]	We show that, under mild conditions, the space of numerical Bridgeland stability conditions Stab(T) on a triangulated category T is complete. In particular the metric on a full component of Stab(T) for which the central charges factor through a finite rank quotient of the Grothendieck group K(T) is complete. As an example, we compute the metric on the space of numerical stability conditions on a smooth complex projective curve of genus greater than one, and show that in this case the quotient Stab(T)/C by the natural action of the complex numbers is isometric to the upper half plane equipped with half the hyperbolic metric. We also make two observations about the way in which the heart changes as we move through the space of stability conditions. Firstly, hearts of stability conditions in the same component of the space of stability conditions are related by finite sequences of tilts. Secondly, if each of a convergent sequence of stability conditions has the same heart then the heart of the limiting stability condition is obtained from this by a right tilt.	math.AG	math	SOME METRIC PROPERTIES OF SPACES OF STABILITY CONDITIONS JON WOOLF 1. Introduction The space of locally-finite stability conditions Stab(C) on a triangulated category C is a (possibly inifinite-dimensional) complex manifold. There is a metric which induces the topology. For numerically-finite C, for instance the coherent bounded derived category of a smooth complex projective variety or the bounded derived category of a finite-dimensional algebra, one can consider the finite-dimensional submanifold of numerical stability conditions. We show that under mild conditions this is complete in the metric and describe limiting stability conditions. The contents are as follows: §2 contains some background material on stability conditions. Section 3 contains the proof that a full component of the space of stability conditions whose central charges factor through some finite rank quotient A of the Grothendieck group is complete. Recall that full means that the space of stability conditions is locally homeomorphic to Hom(A, C), and not to some proper subspace. The key result is Proposition 3.6 which describes the limiting Harder–Narasimhan filtrations. The assumptions are satisified, for instance, for a full component of the space of numerical stability conditions on a numerically-finite triangulated category. In §4 we independently verify that the metric on the space of numerical stability conditions on a smooth complex projective curve of genus ≥ 1 is complete. We compute this metric as follows. There is a natural action of the universal cover G of GL+ 2 R on any space of stability conditions. When the phases of semistable objects are dense for a stability condition σ, the orbit through σ is free and the restriction dG of the metric to it is independent of σ and can be explicitly described, and seen to be complete. In the case of curves of genus g ≥ 1 this density of phases condition is satisfied and the action of G is both free and transitive, so that the space of numerical stability conditions is isometric to (G, dG). The metric dG is closely related to the hyperbolic metric on the upper half plane - the universal cover of the conformal linear maps forms a subgroup of G isomorphic to C and the quotient can be identified with the upper half-plane in such a way that the quotient metric is half the standard hyperbolic metric - see Proposition 4.1. Section 5 contains some observations about relationships between the hearts of stability conditions. Corollary 5.2 states that hearts of stability conditions in the same component of Stab(C) are related by finite sequences of tilts. In the process of proving Theorem 3.7 we obtain a description of the limiting stability condition Date: August 2011. I would like to thank the Newton Institute, Cambridge, for their generous hospitality in April and May 2011, when this paper was completed. I would also like to thank Tom Bridgeland and Arend Bayer for helpful conversations. 1 2 JON WOOLF σ of a convergent sequence σn. As a consequence we obtain Corollary 5.3, which states that if the σn all have the same heart, A say, then the heart of σ must be a right tilt of A. 2. Stability conditions We fix some notation. Let C be an additive category. We write c ∈ C to mean c is an object of C. We will use the term subcategory to mean strict, full subcategory. When C is triangulated with shift functor [1] exact triangles will be denoted either by a → b → c → a[1] or by a diagram _> / b a > > > c where the dotted arrow denotes a morphism c → a[1]. We will always assume that C is essentially small so that isomorphism classes of objects form a set. Given sets Si of objects for i ∈ I let hSi i ∈ Ii denote the ext-closed subcategory generated by objects isomorphic to an element in some Si. We use the same notation when the Si are subcategories of C. Let C be a triangulated category and K(C) be its Grothendieck group. A sta- bility condition σ = (Zσ,Pσ) on C [2, Definition 1.1] consists of an additive group homomorphism Zσ : K(C) → C and full additive subcategories Pσ(ϕ) of C for each ϕ ∈ R satisfying (1) if c ∈ Pσ(ϕ) then Zσ(c) = m(c) exp(iπϕ) where m(c) ∈ R>0; (2) Pσ(ϕ + 1) = Pσ(ϕ)[1] for each ϕ ∈ R; (3) if c ∈ Pσ(ϕ) and c′ ∈ Pσ(ϕ′) with ϕ > ϕ′ then Hom(c, c′) = 0; (4) for each nonzero object c ∈ C there is a Harder–Narasimhan filtration, i.e. a finite collection of triangles 0 = c0 bF F F / c1 F F b1 ··· / cn−1 aD / cn = c xxxxxxxxx D D D bn with bj ∈ Pσ(ϕj ) where ϕ1 > ··· > ϕn. The homomorphism Zσ is known as the central charge and the objects of Pσ(ϕ) are said to be σ-semistable of phase ϕ. The objects bj are known as the σ-semistable factors of c. We define ϕ+ σ (c) = ϕn. The mass of c is defined to σ (c) = ϕ1 and ϕ− i=1 m(bi). be mσ(c) =Pn Lemma 2.1. If a → b → c → a[1] is an exact triangle then σ (b) ≤ max{ϕ+ σ (c)} ≤ ϕ− σ (b) ≤ ϕ+ min{ϕ− σ (a), ϕ− σ (a), ϕ+ σ (c)} for any stability condition σ. σ (b) = sup{t ∃ b′ ∈ Pσ(t) with Hom(b′, b) 6= 0} and similarly Proof. Note that ϕ+ σ (b) = inf{t ∃ b′ ∈ Pσ(t) with Hom(b, b′) 6= 0}. ϕ− Furthermore, both the extreme values are achieved. Suppose ϕ+ There is a semistable b′ with phase ϕ+ σ (c). σ (b) and nonzero morphism b′ → b. Since σ (b) > ϕ+ / _ / / / / / b a SOME METRIC PROPERTIES OF SPACES OF STABILITY CONDITIONS 3 Hom(b′, c) = 0 for phase reasons this morphism factors through a nonzero mor- phism to a. Hence ϕ+ σ (c)}. The other inequality is proved similarly. σ (b) ≤ max{ϕ+ σ (a) ≥ ϕ+ σ (b) and ϕ+ σ (a), ϕ+ (cid:3) By a σ-filtration of 0 6= c ∈ C we mean a filtration as above whose factors are σ- semistable. The Harder–Narasimhan filtration is one example; in general there will be others not satisfying the decreasing phase of factors condition (which uniquely characterises the Harder–Narasimhan filtration). One consequence of the previous lemma is that the Harder–Narasimhan filtration has minimal phase range amongst all σ-filtrations in the sense that if a σ-filtration of c has factors with phases ϕi then min{ϕi} ≤ ϕ− σ (c) ≤ ϕ+ σ (c) ≤ max{ϕi}. For any interval I ⊂ R let Pσ(I) = hPσ(ϕ) ϕ ∈ Ii be the full subcategory gen- erated by σ-semistable objects with phases in I. For any t ∈ R the full subcategory Pσ(t,∞) is a t-structure. In particular the heart Pσ(t,∞) ∩ Pσ(t,∞)⊥[1] = Pσ(t,∞) ∩ Pσ(−∞, t + 1] = Pσ(t, t + 1] is an abelian subcategory of C. Note that Zσ(a) 6= 0 for any 0 6= a ∈ Pσ(t, t + 1] so that the phase ϕσ(a) = 1 π arg Zσ(a) ∈ (t, t + 1] is well-defined. The semistable objects in Pσ(t, t + 1] can be characterised as those b such that a ֌ b implies ϕσ(a) ≤ ϕσ(b), or equivalently those for which b ։ c implies ϕσ(b) ≤ ϕσ(c). See [2, §5] for details. The abelian category Pσ(0, 1] is referred to as the heart of the stability condition σ. When the length I = sup I − inf I of the interval is < 1 the category Pσ(I) is quasi-abelian, see [2, Lemma 4.3]. The strict short exact sequences arise from exact triangles a → b → c → a[1] in C for which a, b and c are in Pσ(I). A stability condition σ is said to be locally-finite if given t ∈ R we can find ǫ > 0 such that the quasi-abelian category Pσ(t − ǫ, t + ǫ) is both artinian and noetherian. (Such a category is called finite length in [2] but we prefer to avoid this terminology to avoid confusion with the a priori stronger notion of length category introduced in the next section.) The set of locally-finite stability conditions can be topologised so that it is a, possibly infinite-dimensional, complex manifold Stab(C) referred to as the space of stability conditions on C. For each component of Stab(C) there is a linear subspace V ⊂ Hom(K(C), C) such that the projection (Z,P) 7→ Z is a local homeomorphism from that component to V [2, Theorem 1.2]. The topology on Stab(C) arises from the generalised metric d(σ, τ ) = sup 06=c∈C max(cid:26)ϕ− σ (c) − ϕ− τ (c),ϕ+ σ (c) − ϕ+ log τ (c),(cid:12)(cid:12)(cid:12)(cid:12) (cid:27) mσ(c) mτ (c)(cid:12)(cid:12)(cid:12)(cid:12) which takes values in [0,∞], see [2, Proposition 8.1]. Note that {τ d(σ, τ ) < ∞} is both open and closed and so is a union of components. Hence d(σ, τ ) = ∞ implies that σ and τ are in distinct components of Stab(C). For the rest of this paper we will loosely refer to d as a metric. The group Aut(C) of automorphisms of C acts on the left of Stab(C) by isometries via (Z,P) 7→(cid:0)Z ◦ α−1, α ◦ P(cid:1) . 4 JON WOOLF There is also a smooth right action of the universal cover G of GL+ 2 R. An element g ∈ G corresponds to a pair (Tg, θg) where Tg is the projection of g to GL+ 2 R under the covering map and θg : R → R is an increasing map with θg(t + 1) = θg(t) + 1 which induces the same map as Tg on the circle R/2Z = R2 − {0}/R>0. In these terms the action is given by (Z,P) 7→(cid:0)T −1 g ◦ Z,P ◦ θg(cid:1) . (Here we think of the central charge as valued in R2.) This action preserves the semistable objects, and also preserves the Harder–Narasimhan filtrations of all ob- jects. The subgroup consisting of pairs for which T is conformal is isomorphic to C with λ ∈ C acting via (Z,P) 7→ (exp(−iπλ)Z,P(ϕ + re λ)) i.e. by rotating the phases and rescaling the masses of semistable objects. This action is free and preserves the metric. The action of 1 ∈ C corresponds to the action of the shift automorphism [1]. The restriction of the metric to an orbit of this C action is given by d(σ, σλ) = max{re λ, πim λ}. No sphere in the metric intersects the orbit in a smooth submanifold. Therefore d is not the path metric of a Riemannian metric. Straight lines in C are geodesics since they have length exactly the distance between their endpoints. By considering geodesic triangles in an orbit we see that Stab(C) can never have strictly negative curvature, although it may have non-positive curvature. Another elementary observation is that the orbits of the C action are closed. Hence there is an induced quotient metric d on Stab(C)/C defined by d(xC, yC) = inf{d(x, yλ) λ ∈ C}. In §4 we will see examples in which this quotient metric arises from a Riemannian metric with constant strictly negative curvature. It would be interesting to see whether the quotient metric has similar good properties in other examples. 3. Completeness of spaces of stability conditions Let C be a triangulated category and K(C) ։ A a finite rank quotient of its Grothendieck group. Consider the subspace of stability conditions on C whose central charge factors through A. Let Stab0(C) be a component of this subspace. Suppose that Stab0(C) is full in the sense of [3, Definition 4.2], i.e. that the pro- jection Stab0(C) → Hom(A, C) is a local homeomorphism. We prove that such a component is complete in the metric. In fact this follows rather quickly from the results of [2, §7,8], which show that Stab0(C) can be covered by closed balls of fixed radius in the metric, each of which is isometric to a complete metric space.1 How- ever, we give a different proof which has the benefit of providing a useful description of the limiting stability condition. By assumption stability conditions in Stab0(C) are locally-finite. We sharpen this slightly, to show that they are locally-length in the sense that given t ∈ R we can find ǫ > 0 such that the quasi-abelian category Pσ(t− ǫ, t + ǫ) is a quasi-abelian 1I would like to thank Arend Bayer for pointing out this quick argument. SOME METRIC PROPERTIES OF SPACES OF STABILITY CONDITIONS 5 length category, or length category for short. By this we mean a quasi-abelian category in which every object a 6= 0 has a length l(a) ∈ N such that any filtration ··· ֌ ai ֌ ai+1 ֌ ··· ֌ a by strict subobjects has length ≤ l(a) with equailty for some filtration. (See [6] for more on quasi-abelian categories.) Clearly a length category is both noetherian and artinian, so that a locally-length stability condition is locally-finite. However, the locally-length condition is a priori stronger in that it requires the existence of an upper bound on the length of any filtration, not merely that any filtration have finite length. Remark 3.1. If A is a noetherian and artinian abelian category the Jordan– Holder theorem asserts that any 0 6= a ∈ A has a finite composition series with simple factors, and that all such composition series have the same length. So this definition of length category agrees with the usual one in case A is abelian, and any noetherian and artinian abelian category is a length category. Furthermore, for an abelian length category the length descends to a homomorphism l : K(A) → Z defined on the Grothendieck group. However, it is not clear that the analogues hold in the quasi-abelian setting. Lemma 3.2. Suppose I ⊂ J are nested intervals of length < 1. Then Pσ(I) is a quasi-abelian length category whenever Pτ (J) is. Moreover the length of a in Pσ(I) is bounded above by its length in Pτ (J). ı−→ a′ −→ Proof. A strict monomorphism ı : a ֌ a′ in Pσ(I) fits into a triangle a a′′ −→ a[1] with a, a′, a′′ ∈ Pσ(I) and therefore also in Pτ (J). Hence ı is a strict monomorphism in Pτ (J) too. The result follows. (cid:3) The next result is a slightly stronger version of [3, Lemma 4.4]; the idea of the proof is the same. Lemma 3.3. Any σ ∈ Stab0(C) is locally-length, indeed Pσ(I) is a quasi-abelian length category whenever I is a closed interval of length I < 1. Proof. Suppose σ ∈ Stab0(C) and that the central charge Zσ is rational, i.e. has image in Q[i]. Then the image is a discrete subgroup of C. Assume, without loss of generality, that I ⊂ (0, 1). Fix 0 6= a ∈ Pσ(I). Then K =(cid:26)z ∈ C (cid:12)(cid:12)(cid:12)(cid:12) 1 π arg z ∈ I and im z ≤ im Zσ(a)(cid:27) is compact, and so contains only finitely many points, say L, in the image of Zσ. Any filtration of a in Pσ(I) is mapped under Zσ to a sequence in K. Since Zσ(b) 6= 0 for 0 6= b ∈ Pσ(I) the images of the subobjects in the filtration are distinct, and hence the filtration cannot have length greater than L. In the general case, since Stab0(C) is full we can approximate σ by a stability condition τ with rational central charge and such that Pσ(I) ⊂ Pτ (J) for some closed interval J of length < 1. The result follows from Lemma 3.2. (cid:3) Suppose that σn is a Cauchy sequence in this component. For ease of reading we write Pn, Zn, ϕn, mn for Pσn , Zσn , ϕσn and mσn . Proposition 3.4. Let 0 6= c ∈ C. The lengths of the Harder–Narasimhan σn- filtrations of c are bounded. 6 JON WOOLF Proof. Fix 0 < ǫ < 1/8 and θ ∈ R. Choose M such that d(σm, σn) < ǫ whenever m, n ≥ M . In particular, for any n ≥ M , Pn(−∞, θ + ǫ) ⊂ PM (−∞, θ + 2ǫ) and Pn(θ − ǫ,∞) ⊂ PM (θ − 2ǫ,∞). n be the truncation functor right adjoint to the inclusion of Pn(−∞, t) in C, Let τ <t and so on. Then τ <θ+ǫ n τ >θ−ǫ n n c = τ <θ+ǫ = τ >θ−ǫ = τ >θ−ǫ n n τ >θ−ǫ n τ <θ+ǫ n τ <θ+ǫ n τ >θ−2ǫ M τ >θ−2ǫ M τ <θ+2ǫ M c c τ >θ−2ǫ M c. Hence the number of σn-semistable factors of c with phase in (θ − ǫ, θ + ǫ) is the same as the number of σn-semistable factors of τ <θ+2ǫ c with phase in (θ − ǫ, θ + ǫ). However, the Harder–Narasimhan σn-filtration of τ <θ+2ǫ c is contained within τ >θ−2ǫ M τ >θ−2ǫ M M M Pn(θ − 3ǫ, θ + 3ǫ) ⊂ PM (θ − 4ǫ, θ + 4ǫ) and the latter is a length category by Lemma 3.3. Therefore there is a uniform (in n ≥ M ) bound on the number of σn-semistable factors of c with phase in (θ − ǫ, θ + ǫ). Note that M depends only on ǫ, and not on θ. Therefore, covering [ϕ− M (c)− ǫ, ϕ+ M (c)+ ǫ] by finitely many intervals of the form (θ− ǫ, θ + ǫ) we obtain a uniform bound on the length of the Harder–Narasimham σn-filtration of c whenever n ≥ M . (cid:3) For θ ∈ R we define the limiting semistable objects of phase θ to be P(θ) = h0 6= c ∈ C ϕ± n (c) → θi. The next lemma shows that limiting semistable objects have non-zero limiting cen- tral charge. Lemma 3.5. Let 0 6= c ∈ C. If ϕ+ Proof. If ϕ+ implies mn(c) → 0, contradicting the assumption that σn is Cauchy. Proposition 3.6. Each 0 6= c ∈ C has a filtration by limiting semistable objects: n (c) → 0 then mn(c) − Zn(c) → 0. Therefore Zn(c) → 0 n (c) → 0 as n → ∞ then Zn(c) 6→ 0. n (c) − ϕ− n (c) − ϕ− (cid:3) 0 = c0 bF / c1 F F F F b1 ··· / cr−1 aC / cr = c yyyyyyyyy C C C br where the factors bi ∈ P(θi) with θ1 > ··· > θr. Proof. By Proposition 3.4 there is some L ∈ N such that the Harder–Narasimhan σn-filtration of c has length ≤ L for all sufficiently large n. We prove the result by induction on L. The case L = 1 is easy, for in this case c is σn-semistable for all sufficiently large n and so ϕ− n (c) → 0 (is in fact 0 for sufficiently large n) and c is itself a limiting semistable. Now assume that the result is true for any object whose Harder–Narasimhan σn-filtrations are of length < L for all sufficiently large n. n (c) − ϕ+ Since σn is a Cauchy sequence both ϕ+ n (c) converge. Suppose that n (c)− ϕ− ϕ+ n (c) → α. If α = 0 then c is a limiting semistable and the result holds. So we may assume α > 0. Choose 0 < ǫ < α/2(L+1). Choose M so that d(σm, σn) < ǫ n (c) and ϕ− / / / / / b a SOME METRIC PROPERTIES OF SPACES OF STABILITY CONDITIONS 7 whenever m, n ≥ M . For any n ≥ M we can find successive semistable factors in the Harder–Narasimhan σn-filtration of c whose phases differ by more than α − 2ǫ L > 2(L + 1)ǫ − 2ǫ L = 2ǫ. Splitting the Harder–Narasimhan σn-filtration of c between the above-mentioned factors one obtains a triangle c′ → c → c′′ → c′[1] with c′, c′′ 6= 0 and such that n (c′) > ϕ+ ϕ− n (c′′) whenever n ≥ M . Thus the Harder–Narasimhan σn-filtration of c for any n ≥ M is obtained by concatenating those of c′ and c′′. It follows that the Harder–Narasimhan σn-filtrations of c′ and c′′ must have length < L for any n ≥ M . Hence, by the inductive hypothesis, both c′ and c′′ have filtrations by limiting semistable objects of decreasing phase. Moreover, the minimal phase of a factor of c′ is at least as large as the maximal phase of a factor of c′′. The result follows by concatenating these limiting Harder–Narasimhan filtrations of c′ and c′′. (cid:3) n (c) > ϕ+ Theorem 3.7. Let Stab0(C) be a full component of the space of locally-finite stabil- ity conditions on C whose central charges factor through a given finite rank quotient of the Grothendieck group. Then Stab0(C) is complete in the natural metric. Proof. Let σn be a Cauchy sequence. Then the central charges Zn converge, say to Z : K(C) → C. We claim that σ = (Z,P) is a locally-length stability condition with σn → σ. It is immediate from the definition of limiting semistable objects that P(θ + 1) = P(θ)[1]. By Lemma 3.5 if 0 6= c ∈ P(θ) then Z(c) = m exp(iπθ) for some m > 0. If c ∈ P(θ) and c′ ∈ P(θ′) with θ > θ′ then, for sufficiently large n, we have ϕ− n (c′) so that Hom(c, c′) = 0. The existence of Harder–Narasimhan filtrations is guaranteed by Proposition 3.6. Hence σ is a stability condition. If the kernel of K(C) ։ A is annihilated by each of the Zn then it is also annihilated by Z, i.e. the central charge of σ factors through A. Given ǫ > 0 we can choose M ∈ N so that d(σm, σn) < ǫ for m, n ≥ M . Then P(t − ǫ, t + ǫ) ⊂ Pm(t − 2ǫ, t + 2ǫ) whenever m ≥ M . Hence σ is locally-length by Lemma 3.2, in particular σ is locally-finite. By construction σn → σ, so that σ ∈ Stab0(C) and the component is complete. Remark 3.8. The condition that Stab0(C) is a full component of the space of locally-finite stability conditions on C whose central charges factor through a given finite rank quotient of the Grothendieck group enters only in the proof that the lengths of the Harder–Narasimhan σn-filtrations of a fixed object are bounded as n → ∞. Hence Theorem 3.7 shows that any Cauchy sequence in the space of locally-finite stability conditions for which this is true converges. (cid:3) 4. Examples In general the metric d, and the induced quotient metric on Stab(C)/C, are hard to compute. In this section, under the assumption that the phases of σ-semistables are dense in R, we compute the restricted metric on the orbit σG, where G is the universal cover of GL+ 2 R, and show that it is independent of σ. In this case the induced metric on σG/C ∼= H is half the standard hyperbolic metric on the upper half-plane. This allows us to compute the metric on Stab(X) whenever X is a smooth complex projective curve of genus ≥ 1, and verify directly that the metrics are complete in these cases. 8 JON WOOLF Proposition 4.1. Suppose σ ∈ Stab(C) is a stability condition for which the phases of semistable objects are dense in R. Then the G orbit through σ is free, the restric- tion of the metric d to it is independent of σ and thus gives a well-defined metric dG on G. This metric is invariant under the action of G by left multiplication and is complete. The induced metric on σG/C ∼= H is half the hyperbolic metric dhyp on the upper half-plane. Proof. Let g ∈ G correspond to the pair (Tg, θg) where Tg ∈ GL+ 2 R and θg : R → R is increasing with θg(t + 1) = θg(t) + 1 and induces the same map as Tg on the circle R/2Z = R2 −{0}/R>0. Since σ and σg have the same semistable objects and Harder–Narasimhan filtrations d(σg, σ) = sup semistable c = = sup semistable c sup semistable c log max(cid:26)ϕσg(c) − ϕσ(c) ,(cid:12)(cid:12)(cid:12)(cid:12) max(cid:26)θg(ϕσ(c)) − ϕσ(c) ,(cid:12)(cid:12)(cid:12)(cid:12) mσg(c) mσ(c) (cid:12)(cid:12)(cid:12)(cid:12) (cid:27) Zσ(c) (cid:12)(cid:12)(cid:12)(cid:12) log TgZσ(c) max{θg(ϕσ(c)) − ϕσ(c) ,log (Tgv)} (cid:27) where v = Zσ(c)/Zσ(c) is the unit vector in the direction of Zσ(c) ∈ R2. Under the assumption that phases of semistables are dense we thus have d(σg, σ) = max(sup t θg(t) − t, sup v=1log (Tgv)) (cid:9) which depends only on g, and not on σ. Set ∆(g) = d(σg, σ). It is easy to check that ∆(g) = 0 if and only if g = 1, so that = max(cid:8)θg − id, log Tg, log T −1 g dG(g, h) = d(σg, σh) = d((σh)h−1g, σh) = ∆(h−1g) is a metric on G, and the orbit σG is free. Note that dG(f g, f h) = ∆((f h)−1f g) = ∆(h−1g) = dG(g, h) so that dG is invariant under left multiplication. The explicit description shows that dG is complete. isomorphism is given by We have identifications G/C ∼= GL+ (cid:18) a b 2 R/C∗ ∼= SL2R/SO(2) ∼= H where the last d (cid:19) 7−→ ai + b ci + d Under this identification the left multiplication action of SL2R corresponds to its action by Mobius transformations on H. Hence the induced quotient metric on σG/C ∼= H is invariant under the Mobius action of SL2R. To pin down the metric it therefore suffices to compute the distance between i and ai for real a ≥ 1. This is given by inf λ∈C ∆(gλ) where c . Tg =(cid:18) √a 0 0 1/√a (cid:19) and θg(t) = 1 π a (cid:19) . tan−1(cid:18) tan(πt) By direct computation we find that inf λ∈C ∆(gλ) = 1 2 log a = 1 2 dhyp(i, ai). (cid:3) If X is a smooth complex projective curve of genus ≥ 1 then Stab(X) ∼= G is a single orbit [5]. Furthermore, the phases of semistables are dense since, for the standard stability conditon, there are semistable sheaves of any rational slope. SOME METRIC PROPERTIES OF SPACES OF STABILITY CONDITIONS 9 (Both of these facts are false for the genus 0 case.) Therefore the spaces of stability conditions are isometric to (G, dG) for any such X, and in particular are complete. Furthermore, Stab(X)/C is isometric to the upper half-plane with metric 1 2 dhyp. 5. Tilting and hearts of stability conditions A torsion theory in an abelian category A is a pair (T ,F ) of subcategories such that F ⊂ T ⊥ and every a ∈ A sits in a short exact sequence 0 → t → a → f → 0 with t ∈ T and f ∈ F , see for example [1, Definition 1.1]. In fact F = T ⊥ and T = ⊥F so that it is only necessary to specify either T or F . Let A be the heart of a t-structure D. A torsion theory T in A determines a new t-structure hD,T [−1]i, see [4, Proposition 2.1]: we say this new t-structure is obtained from the original by left tilting at T and denote its heart by LT A. Explicitly LT A = hF ,T [−1]i. A torsion theory also determines another t-structure hD⊥,F [1]i⊥ by a 'double dual' construction. This is the shift by [1] of the other: we say it is obtained by right tilting at T and denote the new heart hF [1],T i by RT A. Left and right tilting are inverse to one another: F is a torsion theory in LT A and right tilting with respect to this we recover the original heart A. In the opposite direction, given a t-structure E with D ⊂ E ⊂ D[−1] Beligiannis and Reiten [1, Theorem 3.1] show that T = (E ∩ D⊥)[1] = ha ∈ A a = H 1e for some e ∈ Ei determines a torsion theory in A such that LT A is the heart of E. Lemma 5.1. If d(σ, τ ) < 1/2 then Aτ can be obtained from Aσ by performing a left and then a right tilt. (Of course one or both of these tilts may be trivial.) Proof. Recall that Aσ = Pσ(0, 1] is the heart of the t-structure Pσ(0,∞), and analogously for Aτ . Since d(σ, τ ) < 1/2 there are inclusions of t-structures Pτ (0,∞) Pσ(0,∞) / Pσ(−1,∞) / Pτ(cid:0)− 1 2 ,∞(cid:1) Pτ (−1,∞) So Pτ (−1/2,∞) determines torsion theories T in Aσ and T ′ in Aτ with LT Aσ = Pτ (−1/2, 1/2] = LT ′Aτ . Hence Aτ = RF ′LT Aσ. Corollary 5.2. If σ and τ are in the same component of Stab(C) then the hearts Aσ and Aτ are related by a finite sequence of left and right tilts. Proof. The connected components of Stab(C) are the path components. Choose a path from σ to τ , cover it with finitely many balls of diameter < 1/2 and apply Lemma 5.1. (cid:3) (cid:3) For a heart A let U (A) = {σ ∈ Stab(C) Aσ = A}. Note that U (A) may be empty and need not be either open or closed. The next result describes the hearts of stability conditions in the closure of U (A). / / 10 JON WOOLF Corollary 5.3. Suppose σn is a sequence in U (A) with σn → σ. Then Aσ = RT A for the torsion theory with T = h0 6= a ∈ A ϕ− n (a) 6→ 0i n (a) → 0i. and F = h0 6= a ∈ A ϕ+ Proof. It is easy to check that T is a torsion theory; the required short exact sequences arise from the description of the Harder–Narasimhan σ-filtration as a limiting filtration. The description of the limiting semistable objects shows that RT A ⊂ Aσ. Hence they are equal as hearts of distinct non-degenerate t-structures cannot be nested. (cid:3) It is not always the case that a torsion theory has the form h0 6= a ∈ A ϕ− n (a) 6→ 0i for some Cauchy sequence σn of stability conditions, and it is not true that B = RT A implies U (A) ∩ U (B) 6= ∅. For example, let A be the category of representations of the Kronecker quiver. This is well-known to be derived equivalent to the coherent derived category of P1 and (after fixing an appropriate equivalence) A = LT Coh(P1) where T = ⊥hO(d) d < 0i. However there is no sequence σn in U (A) with σn → σ where Aσ = Coh(P1); see [7, §3.2] for details. References [1] A. Beligiannis and I. Reiten. Homological and homotopical aspects of torsion theories. Mem. Amer. Math. Soc., 188(883):viii+207, 2007. [2] T. Bridgeland. Stability conditions on triangulated categories. Ann. of Math. (2), 166(2):317– 345, 2007. [3] T. Bridgeland. Stability conditions on K3 surfaces. Duke Math. J., 141(2):241–291, 2008. [4] D. Happel, I. Reiten, and S. Smalø. Tilting in abelian categories and quasitilted algebras. Mem. Amer. Math. Soc., 120(575), 1996. [5] E. Macr`ı. Stability conditions on curves. Math. Res. Lett., 14(4):657–672, 2007. [6] J.-P. Schneiders. Quasi-abelian categories and sheaves. M´em. Soc. Math. Fr. (N.S.), (76):vi+134, 1999. [7] J. Woolf. Stability conditions, torsion theories and tilting. J. London Math. Soc., 2010.
0911.1773	3	0911	2010-04-16T00:39:53	Perverse coherent sheaves on blow-up. III. Blow-up formula from wall-crossing	[ "math.AG", "math-ph", "math-ph" ]	In earlier papers arXiv:0802.3120, arXiv:0806.0463 of this series we constructed a sequence of intermediate moduli spaces $\bM^m(c)$ connecting a moduli space $M(c)$ of stable torsion free sheaves on a nonsingular complex projective surface and $\bM(c)$ on its one point blow-up. They are moduli spaces of perverse coherent sheaves on the blow-up. In this paper we study how Donaldson-type invariants (integrals of cohomology classes given by universal sheaves) change from $\bM^m(c)$ to $\bM^{m+1}(c)$, and then from $M(c)$ to $\bM(c)$. As an application we prove that Nekrasov-type partition functions satisfy certain equations which determine invariants recursively in second Chern classes. They are generalization of the blow-up equation for the original Nekrasov deformed partition function for the pure N=2 SUSY gauge theory, found and used to derive the Seiberg-Witten curves in arXiv:math/0306198.	math.AG	math	PERVERSE COHERENT SHEAVES ON BLOW-UP. III. BLOW-UP FORMULA FROM WALL-CROSSING HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA To the memory of the late Professor Masaki Maruyama Abstract. In earlier papers [21, 22] of this series we constructed a sequence of interme- In this pa- per we study how Donaldson-type invariants (integrals of cohomology classes given by diate moduli spaces {cM m(c)}m=0,1,2,... connecting a moduli space M (c) of stable torsion free sheaves on a nonsingular complex projective surface X and cM (c) on its one point blow-up bX. They are moduli spaces of perverse coherent sheaves on bX. universal sheaves) change from cM m(c) to cM m+1(c), and then from M (c) to cM (c). As an application we prove that Nekrasov-type partition functions satisfy certain equations which determine invariants recursively in second Chern classes. They are generalization of the blow-up equation for the original Nekrasov deformed partition function for the pure N = 2 SUSY gauge theory, found and used to derive the Seiberg-Witten curves in [18]. Introduction Let X be a nonsingular complex projective surface and p : bX → X the blow-up at a point 0. Let C = p−1(0) be the exceptional divisor. Let c = (r, c1, ch2) ∈ H ev(bX) be a cohomological data. Let cM(c) be the moduli space of stable torsion free sheaves E on bX with ch(E) = c and M(p∗(c)) the corresponding moduli space on X. In [21, 22] we constructed a sequence of intermediate moduli spaces cM m(c) connected by birational morphisms as cM m(c) ξm )SSSSSS (∗) ukkkkkkk · · · such that cM m,m+1(c) ξ+ m ujjjjjj cM m+1(c) ξm+1 UUUUUU vmmmmmmmm · · · cM m+1,m+2(c) (1) cM m(c) ∼= cM(c) if m is sufficiently large, and (2) cM 0(c) ∼= M(p∗(c)) if (c1, [C]) = 0 under the natural homomorphism given by E 7→ p∗(E). (See Proposition 1.2 for the statement when 0 < (c1, [C]) < r.) The diagram (∗) is an example of those often appearing in variations of GIT quotients [27], and similar to ones for moduli spaces of sheaves (by Thaddeus, Ellingsrud-Gottsche, Friedman-Qin and others) when we move polarizations. See [21, 22] for more references on earlier works. 2000 Mathematics Subject Classification. Primary 14D21; Secondary 16G20. 1 ) u u v 2 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA In this paper, we study how Donaldson type invariants (certain integrals of cohomology of framed sheaves for which quiver description was given in [21]. We conjecture that the results are universal, i.e., independent of the choice of a surface. Moreover we have classes given by universal sheaves) change from cM m(c) to cM m+1(c). By a technical reason we restrict ourselves to the case when X is P2 and cM m(c) is replaced by the moduli space a natural (r + 1)-dimensional torus eT = (C∗)2 × T r−1 action on cM m(c) from the (C∗)2- action onbP2 and the change of framing. Thus we can consider equivariant Donaldson type invariants to which we can apply Atiyah-Bott-Lefschetz fixed point formula to perform a further computation. In this sense, we think our situation is most basic. Our first main result says that the difference of invariants is given by a variant of Mochizuki's weak wall-crossing formula [14], i.e., it is expressed as a sum of an integral once (∗) is understood as a variation of GIT quotients. Summing up the weak wall-crossing formula from 0 to m, we get the formula for the over cM m(c′) with smaller c′ (Theorem 1.5). Our argument closely follows Mochizuki's, difference of cM (c) and M(p∗(c)) by integrals over various cM m′(c′), as a result. We nor- to apply M(p∗(c′)) ∼= cM 0(c′) for (c1(c′), [C]) = 0 and its modification Proposition 1.2. over even smaller cM m′′(c′′). We apply the same argument for cM m′′(c′′). We thus do this argument recursively to give an algorithm to express cM(c) by a linear combination of malize the first Chern class of c′ in the interval [0, r− 1] twisting by a line bundle in order Then those integrals themselves can be expressed by integrals over M(p∗(c′)) and ones integrals over M(c♭) for various c♭. Since this algorithm is complicated (see Figure 1 for the flowchart), we do not try to write down an explicit formula in general. We instead focus on vanishing theorems for special cases when integrands are not twisted too much along C. This is our second main result. See §2. Our motivation of study in this series is an application to the Nekrasov partition func- tion [24]. Let us explain it briefly. The Nekrasov partition function is the generating function of an equivariant integral over M(c). One of the main conjecture on it states that the leading part F0 of its logarithm is given by the Seiberg-Witten prepotential, a certain period integral on the Seiberg-Witten curves. The three consecutive coefficients (denoted by H, A, B) are also important for the application to the wall-crossing formula for usual or K-theoretic Donaldson invariants for projective surfaces with pg = 0 [7, 8]. When the integrand is (1) 1, (2) slant products of Chern classes of universal sheaves with the fundamental classes of C2, or (3) the Todd class of M(c), the authors proved that the partition function satisfies functional equations, called the blow-up equations, which determine coefficients recursively in second Chern numbers of c [18, 19, 20]. The functional equations induce a nonlinear partial differential equation for F0, which has been known as the contact term equation in the physics literature [12, 6]. In particular, the Seiberg-Witten prepotential satisfies the same equation and hence is equal to F0. This was our proof of the above mentioned conjecture. There are other completely independent proofs by [25, 2]. But so far H, A, B can be determined only from the blow-up equation. Nekrasov's partition functions have more variants by replacing the integrand. Let us give three examples: PERVERSE COHERENT SHEAVES ON BLOW-UP. III 3 (a) We integrate Euler classes of vector bundles given by pushforward of universal sheaves. They are called the theories with fundamental matters in the physics literature. (b) When we integrate the Todd classes, we can cap with powers of the first Chern classes of the same bundles. They are called 5-dimensional Chern-Simons terms. (c) We can also incorporate universal sheaves to which Adams operators are applied. They are called (higher) Casimir operators. They give coefficients appearing in the defin- ing equation of the Seiberg-Witten curve. The blow-up equation was derived by analyzing relation between integrals over M(c) andcM (c). Our vanishing results in §2 enable us to generalize our proof for those variants1. In this paper, we explain it for theories with 5-dimensional Chern-Simons terms and Casimir operators. The case for the theories with matters will be given elsewhere [23]. The paper is organized as follows. In §1 we state our results after preparing the nec- essary notations. In §2 we prove several versions of vanishing theorems as applications In §3 we study the Nekrasov partition function for theories with of the results in §1. 5-dimensional Chern-Simons terms. The blow-up equation is derived. This section is ex- pository since the derivation of Nekrasov's conjecture was already given in [8] assuming the vanishing theorems. The actual proof starts from §4. We review the quiver description of the framed moduli spaces obtained in [21] and the analysis of the wall-crossing in [22], and add a few things. In §5 we define enhanced The quiver description is necessary to define master spaces. master spaces. We follow Mochizuki's method [14], but give the construction in detail for the sake of a reader. In §6 we prove Theorem 1.5, the variant of Mochizuki's weak wall-crossing formula. Again the proof is the same as Mochizuki's. Acknowledgments. The first named author is supported by the Grant-in-Aid for Scien- tific Research (B) (No. 19340006), Japan Society for the Promotion of Science. The second named author is supported by the Grant-in-aid for Scientific Research (B) (No. 18340010), (S) (No. 19104002), JSPS and Max Planck Institute for Mathematics. Both authors thank Takuro Mochizuki for explanations of his results, and the referee for many valuable com- ments. 1. Main result Notations. Let [z0 : z1 : z2] be the homogeneous coordinates on P2 and ℓ∞ = {z0 = 0} closed subvariety of P2 × P1 defined by the line at infinity. Let p : bP2 → P2 be the blow-up of P2 at [1 : 0 : 0]. Then bP2 is the (cid:8)([z0 : z1 : z2], [z : w]) ∈ P2 × P1 z1w = z2z(cid:9) , where the map p is the projection to the first factor. We denote p−1(ℓ∞) also by ℓ∞ for brevity. Let C denote the exceptional divisor given by z1 = z2 = 0. Let O denote the structure sheaf ofbP2, O(C) the line bundle associated with the divisor C, and O(mC) its mth tensor product O(C)⊗m when m > 0, (O(C)⊗−m)∨ if m < 0, and O if m = 0. And 1The proof in [25] can be generalized to those variants. See [26] for the theory with 5-dimensional Chern-Simons terms, and [13] for higher Casimir operators, but not with Todd genus. 4 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA we use the similar notion O(mC + nℓ∞) for tensor products of O(mC) and tensor powers of the line bundle corresponding to ℓ∞ or its dual. The structure sheaf of the exceptional divisor C is denoted by OC. If we twist it by the line bundle OP1(n) over C ∼= P1, we denote the resulted sheaf by OC(n). Since C has the self-intersection number −1, we have OC ⊗ O(C) = OC(−1). For c ∈ H∗(bP2), its degree 0, 2, 4-parts are denoted by r, c1, ch2 respectively. If we want to specify c, we denote by r(c), c1(c), ch2(c). For brevity, we twist the push-forward homomorphism p∗ by Todd genera of P2 andbP2 as in [22, §3.1] so that it is compatible with the Riemann-Roch formula. We also use the following notations frequently: 2r(c) c1(c)2. • ∨ is the involution on the K-group given by taking the dual of a vector bundle. • ∆(E) := c2(E) − r−1 • Cm denotes OC(−m − 1). • em := ch(OC(−m − 1)). 2r c1(E)2, ∆(c) := − ch2(c) + 1 • For an integer N let N = {1, 2, . . . , N}. • pt is a single point in X, bX or sometimes an abstract point. Its Poincar´e dual in H 4(X) or H 4(bX) is also denoted by the same notation. For a sheaf E onbP2, we denote H 1(E(−ℓ∞)), H 1(E(C − ℓ∞)) by V0(E), V1(E) respec- tively (and simply by V0, V1 if there is no fear of confusion). In this paper we mainly treat sheaves E with H i(E(−ℓ∞)) = 0 = H i(E(C − ℓ∞)) for i 6= 1. This is clear after we will recall the quiver description of framed moduli spaces in §4: Vα(E) appears as a vector space on the vertex α, and any sheaf in this paper corresponds to a representation of the quiver. Under this assumption we have dim V0 = dim H 1(E(−ℓ∞)) = −(ch2(E), [bP2]) + dim V1 = dim H 1(E(C − ℓ∞)) = −(ch2(E), [bP2]) − 1 2 (c1(E), [C]), 1 2 (c1(E), [C]) by Riemann-Roch. • V0(E) := R1q2∗(E ⊗ q∗1O(−ℓ∞)), • V1(E) := R1q2∗(E ⊗ q∗1O(C − ℓ∞)). Let cM be a moduli scheme (or stack) and q1, q2 be projections to the first and second factors ofbP2 ×cM . For a sheaf E (e.g., the universal sheaf) onbP2 ×cM , let Let Ext•q2 denotes the derived functor of the composite functor q2∗ ◦ Hom. We often consider Ext•q2(E, Cm), where Cm is considered as a sheaf onbP2 ×cM via the pull-back by q1. 1.1. Framed moduli spaces. A framed sheaf (E, Φ) on P2 is a pair of • a coherent sheaf E, which is locally free in a neighborhood of ℓ∞, and • an isomorphism Φ : Eℓ∞ → O⊕r ℓ∞ , where r is the rank of E. An isomorphism of framed sheaves (E, Φ), (E′, Φ′) is an isomorphism ξ : E → E′ such that Φ′ ◦ ξℓ∞ = Φ. When r = 0, we understand that a framed sheaf is an ordinary sheaf PERVERSE COHERENT SHEAVES ON BLOW-UP. III 5 of rank 0 whose support does not intersect with ℓ∞. We have the corresponding definition of a framed sheaf on the blow-upbP2. Definition 1.1. Let m ∈ Z≥0. A framed sheaf (E, Φ) onbP2 is called m-stable if (1) Hom(E,OC(−m − 1)) = 0, (2) Hom(OC(−m), E) = 0, and (3) E is torsion free outside C. Though it is not obvious from the definition, an m-stable sheaf must have r > 0. (See [21, §2.2].) framing unlike the case of ordinary moduli spaces. As special cases with m = 0 and m sufficiently large, we have fine moduli schemes M(c) We have a smooth fine moduli scheme cM m(c) of m-stable framed sheaves (E, Φ) with ch(E) = c ∈ H∗(bP2) such that (c, [ℓ∞]) = 0. It is of dimension 2r(c)∆(c). (See Theo- rem 4.2.) Let E be the universal sheaf on bP2 × cM m(c), which is unique thanks to the and cM (c) of framed torsion free sheaves (E, Φ) on P2 andbP2 respectively. For M(c), we take c ∈ H∗(bP2) with (c1, [C]) = 0. (See [21, §7] or [22, §3.1, §3.9].) They are connected by a sequence of birational morphisms as explained in the introduction. See §4.4. In fact, M(c) was studied earlier in [17, Chap. 2,3] (denoted there by M(r, n)). We need to recall one important property: We have a projective morphism π : M(c) → M0(c), where M0(c) is the Uhlenbeck (partial) compactification of the moduli space M reg 0 (c) of framed locally free sheaves (E, Φ). In [loc. cit.] M0(c) was constructed via the quiver description, and bijective to Gc′ 0 (c′) ⊗ S∆(c)−∆(c′)(C2) M reg set-theoretically. Here Sn(C2) denotes the nth symmetric product of C2. from the quiver description (Theorem 4.2) or [22, §3.2]. It is compatible with the diagram 1.2. Grassmann bundle structure. As we mentioned above, we have M 0(c) ∼= M(p∗(c)) when (c1, [C]) = 0. For 0 < (c1, [C]) < r, we have a similar relation as follows. We need For any m, we still have a projective morphism bπ : cM m(c) → M0(p∗(c)). This follows (∗) and induced from a projective morphism cM m,m+1(c) → M0(p∗(c)). to consider cM 1(c) with 0 > (c1, [C]) > −r instead after twisting by the line bundle O(C). bN (c, n) relating cM 1(c) and cM 1(c − ne0) through a diagram cM 1(c − ne0) Proposition 1.2 ([22, §3.10]). Suppose 0 < n := −(c1, [C]) < r. There is a variety bN (c, n) satisfying the followings: cM 1(c) f1 vmmmmmm f2 TTTTTT q2(OC(−1),E′)) of n-planes in the vector bun- (1) f1 is surjective and birational. (2) f2 is the Grassmann bundle Gr(n, Ext1 dle Ext1 q2(OC(−1),E′) of rank r over cM 1(c − ne0). v 6 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA (3) We have a short exact sequence 0 → (idbP2 ×f2)∗E′ → (idbP2 ×f1)∗E → OC(−1) ⊠ S → 0. Hence Ext1 Remark that Exti Here E, E′ are the universal sheaves for cM 1(c) and cM 1(c− ne0) respectively, and S is the q2(OC(−1),E′) over Gr(n, Ext1 universal rank n subbundle of Ext1 q2(OC(−1),E′)). q2(OC(−1),E′) = 0 for i = 0, 2 by the remark after Lemma 4.11 below. q2(OC(−1),E′) is a vector bundle, and its rank is r by Riemann-Roch. after crossing the wall between 0-stability and 1-stability. 1.3. Torus action and equivariant homology groups. Let T be the maximal torus We have (c1(c− ne0), [C]) = (c1, [C]) + n = 0. Therefore cM 1(c− ne0) becomes M(p∗(c)) of SLr(C) consisting of diagonal matrices and let eT = C∗ × C∗ × T . We have a eT -action on cM m(c) induced from the C∗ × C∗-action onbP2 given by ([z0 : z1 : z2], [z : w]) 7→ ([z0 : t1z1 : t2z2], [t1z : t2w]) Let H eT (pt) of a point, isomorphic to the symmetric and the change of the framing Φ. See [22, §5]. It was defined exactly as in the case of framed moduli spaces of torsion free sheaves, given in [18, §3]. The action is compatible over the equivariant cohomology group H∗ eT universal sheaf E. rational coefficients. Let H∗ eT coefficients. They are defined for X satisfying a reasonable condition, say an algebraic with one on M0(c), i.e.,bπ is eT -equivariant. All the constructions, which we have explained so far, are canonically eT -equivariant. For example, we have the canonical eT -action on the ∗ (X) be the eT -equivariant Borel-Moore homology group of a eT -space X with (X) be the eT -equivariant cohomology group with rational variety with an algebraic eT -action. See, for example, [19, App. C]. They are modules product of the dual of the Lie algebra, which we denote by S(eT ). The projective morphismbπ : cM m(c) → M0(p∗(c)) induce a homomorphism We denote this homomorphism bπ∗ byRcM m(c), since we also use similar push-forward ho- ∆(p∗(c)). ThenRcM m(c) takes values in H eT (M0(c′)). We can also makeRcM m(c) with values in S(eT ), the quotient field of S(eT ) as follows: Recall that eT has the unique fixed point 0 in M0(p∗(c)) [18, Prop.2.9(3)]. We composeRcM m(c) with the inverse ι−1 (pt) = S(eT ). This is compatible with the above ing tensor products with S(eT ) over H∗ 0∗ of the push-forward homomorphism ι0∗ for the inclusion {0} → M0(p∗(c)) by using the localization theorem for the equivariant homology group, which says ι0∗ becomes an isomorphism after tak- On the other hand, the target space M0(p∗(c)) is not at all important. We can com- pose the push-forward homomorphism for the inclusion M0(p∗(c)) ⊂ M0(c′) for ∆(c′) ≥ momorphisms from homology groups of various moduli schemes or stacks and want to emphasize the domain. inclusion. See [18, §4] for more detail. ∗ (cM m(c)) → H eT bπ∗ : H eT ∗ (M0(p∗(c))). ∗ eT PERVERSE COHERENT SHEAVES ON BLOW-UP. III 7 1.4. Weak wall-crossing formula. We now state our first main result in this subsection. Let Φ(E) ∈ H∗ eT a cohomology group, for example, (cM m(c)) be an equivariant cohomology class on cM m(c) defined from a sheaf E onbP2 ×cM m(c) by taking a slant product by a cohomology class onbP2, or taking (1.3) Φ(E) := exp" ∞Xp=1ntp chp+1(E)/[C] + τp chp+1(E)/[bC2]o# , NfYf =1 e(Va(E) ⊗ emf ) a = 0 or 1, where tp, τp are variables and the exponential defines formal power series in tp, τp in the first case, and m1, . . . , mNf are variables for the equivariant cohomology H∗ (C∗)Nf (pt) of the Nf -dimensional torus of a point, and emf is the corresponding equivariant line bundle. For E we typically take the universal sheaf, or its variant. For the latter Φ(E) = QNf f =1 e(Va(E) ⊗ emf ), we need to enlarge eT to eT × (C∗)Nf but keep the notation eT for brevity. And e( ) denotes the equivariant Euler class. Remark 1.4. The notation Nf is taken from physics literature. It is the number of flavors. But we denote the rank by r, though it is denoted by Nc (number of colors) in physics literature. or Φ(E) := The above examples of Φ are multiplicative, i.e., Φ(E ⊕E′) = Φ(E)Φ(E′). This condition is useful when we will study the vanishing theorem in §2. But we do not assume it in general. For j ∈ Z>0 we consider the j-dimensional torus (C∗)j acting trivially on moduli schemes. We denote the 1-dimensional weight n representation of the ith factor by eni. The equivariant cohomology H∗(C∗)j (pt) of the point is identified with C[1, . . . , j]. In the following formula we invert variables 1,. . . , j. See §6.1 for the precise definition. Also we identify Φ(E) with the homology class Φ(E) ∩ [cM m+1(c)] and apply the push-forward homomorphismRcM m+1(c). ZcM m+1(c) Φ(E) −ZcM m(c) Theorem 1.5. Φ(E) = ∞Xj=1ZcM m(c−jem) j =0· · · Res Res 1=0"Φ(E♭ ⊕ Q1≤i16=i2≤j(−i1 + i2) where E♭ is the universal sheaf for cM m(c − jem) and Ψj(E♭) := i=1 e(N(E♭, Cm) ⊗ e−i) e(N(Cm,E♭) ⊗ ei) , Cm ⊠ e−i)Ψj(E♭)# , jMi=1 1 j! Qj (−1)a Exta N(E♭, Cm) := − 2Xa=0 q2(E♭, Cm), N(Cm,E♭) := − (−1)a Exta q2(Cm,E♭). 2Xa=0 8 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA (Note that Ψj(E♭) depends on j, but not on c − jem if we consider E♭ as a variable.) The proof will be given in §6.4. 1.5. Blow-up formula. Recall that cM m(c) is isomorphic to the framed moduli space cM (c) of torsion free sheaves onbP2 if m is sufficiently large. Using Proposition 1.2, Theo- rem 1.5 and twist by the line bundle O(C), we can expressRcM(c) Φ(E) as a sum of various RM (c′) Φ′(E)'s for some c′, Φ′. Unfortunately the procedure, which we will explain below 1.5.1. For a sequence ~j = (j0, j1, . . . , jm−1) ∈ Zm Ψ~j m = 1 by in detail, is recursive in nature and rather cumbersome. See Figure 1 for the flowchart. In particular, we do not solve the recursion and do not give the explicit formula. n recursively starting from ~j 1 jn! i ) × n+1(• ⊕ Cn ⊠ e−n ~j jn are variables. Then we set Ψ~j = Ψ 0. By Theorem 1.5 we get ≥0, we define Ψ~j Q1≤i16=i2≤jn(−n Qjn i=1 e(N(•, Cn) ⊗ e−n Φ(E) =X~j ZcM 0(c−Pn jnen) i )Ψ~j(E♭), Cn ⊗ en m−1Mn=0 jnMi=1 Φ(E♭ ⊕ i1 + n i2) −→Res ~=0 i ) e(N(Cn,•) ⊗ en i ) , Ψ~j n(•) := Ψ where n 1 , . . . , n (1.6) ZcM m(c) where −→Res~=0 is the iterated residues =0· · · Res := Res −→Res ~=0 0 1=0 0 j0 =0· · · Res Res 1 j1 1=0 · · · Res m−1 jm−1 1 =0· · · Res m−1 1 =0 . (1.7) jnen) if (j1, j2, . . . ) 6= 0. Note that if (c1, [C]) ≥ 0, we have Since cM (c) is isomorphic to cM m(c) for a sufficiently large m, an integral over cM (c) can be written in terms of integrals over cM 0(c′) with various c′ thanks to this formula. dimcM 0(c −Xn = dimcM m(c) −X r(2n + 1)jn −X jn(cid:16)X jn + 2(c1, [C])(cid:17) < dimcM m(c) 1.5.2. We usually consider the moduli space cM (c) with 0 ≤ (c1, [C]) < r. This is always RcM (c) Φ(E) to RM (p∗(c)) Φ(E) thanks to Proposition 1.2. But look at (1.6). The right hand side of (1.6) contains integrals over cM 0(c′) (c′ = c −Pn jnen) for which we have a term in the right hand side of (1.6) asRcM 0(c) Φ(E), and start from it. We can assume 0 ≤ (c1(c′), [C]), but not necessarily < r. Thus we need to tensor a line bundle again. achieved by tensoring a power of the line bundle O(C). And then we can hope to relate Since keeping track the precise form of the formula is a rather tiresome work, we redefine (c1, [C]) ≥ 0, as we explained. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 9 M(p∗(c)) by (E, Φ) 7→ (p∗(E), Φ), where the higher direct image sheaves R>0p∗(E) van- ishes. Moreover its inverse is given by (F, Φ) 7→ (p∗F, Φ), and L<0p∗F = 0. (See [22, Prop. 3.3 and §1].) If we denote the universal sheaf for M(p∗(c)) by F , the universal sheaf 1.5.3. First consider the case (c1, [C]) = 0. We have an isomorphism Π : cM 0(c) ∼= E for cM 0(c) is equal to (p × Π)∗(F ). Therefore we have Φ((p × Π)∗(F )). ZcM 0(c) Φ(E) =ZcM 0(c) Since L<0(p × Π)∗(F ) vanishes, this holds in the level of K-group. pression Φ(E), but we can use the projection formula to rewrite the right hand side as We may have expressions O(C) or [C], which do not come from M(p∗(c)) in the ex- ZM (p∗(c)) Φ′(F ). ZcM 0(c) Φ(E) =ZcM 1(ce[C]) Φ(E(−C)) for a possibly different cohomology class Φ′(•). 1.5.4. Now we may assume (c1, [C]) > 0. We have (1.8) We have −r < (c1(ce[C]), [C]) = (c1, [C]) − r. from the former under this isomorphism, and it is the reason why we have Φ(E(−C)). If this is negative, in other words, if we have (c1, [C]) < r, we go to the step which will be explained in §1.5.5. So we assume back to §1.5.1 and apply (1.6) with m = 1. by the isomorphism cM 0(c) ∋ (E, Φ) 7→ (E(C), Φ) ∈ cM 1(ce[C]). Two universal sheaves for cM 0(c), cM 1(ce[C]) are denoted by the same notation, but the latter is twisted by O(C) (c1(ce[C]), [C]) ≥ 0. We now redefine the right hand side of (1.8) asRcM 1(c) Φ(E) and return We repeat this procedure until all terms are integrals overcM 1(c′) with −r < (c1(c′), [C]) < 0, orcM 0(c′′) with c1(c′′) = 0. From the dimension estimate (1.7), the procedure ends after For Φ(E) = QNf f =1 e(Va(E) ⊗ emf ), this process requires a care, since Va(E(−C)) = R1q2∗(E(−C)⊗ q∗1(O(aC − ℓ∞))) may not be a vector bundle on M 0(c), so e(Va(E(−C))⊗ emf ) does not make sense, and we cannot apply (1.6) with m = 1. We overcome this difficulty by replacing e(Va(E(−C)) ⊗ emf ) by a product of e(Va(E) ⊗ emf ) and a certain class, which is well-defined on M 0(c). See the proof of Theorem 2.1 for detail. finite steps. 1.5.5. Now we redefine the right hand side of (1.8) asRcM 1(c) Φ(E) and consider it under the assumption −r < (c1, [C]) < 0. Let n := −(c1, [C]). By Proposition 1.2 we have ZcM 1(c) = Z bN (c,n) Φ(E) =Z bN (c,n) Φ((idbP2 ×f1)∗E) Φ(cid:0)(idbP2 ×f2)∗(E) ⊕ C0 ⊠ S(cid:1) , 10 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA where we denote the universal bundle overcM 1(c−ne0) by E for brevity. Since f2 : bN (c, n) → cM 1(c − ne0) is the Grassmann bundle of n-planes in Ext1 cM 1(c − ne0) to get q2(C0,E), we can pushforward to Z bN (c,n) ′Φ(•) := ZGr(n,r) Φ(cid:0)(idbP2 ×f2)∗(E) ⊕ C0 ⊠ S(cid:1) =ZcM 1(c−ne0) Φ (• ⊕ (C0 ⊠ S))(cid:12)(cid:12)(cid:12)(cid:12)c(Cr)=c(Ext1 q2 ′Φ(E), . (C0,E)) where q2 We need to explain the notation. We consider the Grassmannian Gr(n, r) of n-planes (C0,E)) means that we substitute the Chern classes of Ext1 a moduli space and • ⊕ C0 ⊠ S is a universal sheaf, and apply the function Φ. Finally q2(C0,E)) to the ( )c(Cr)=c(Ext1 equivariant Chern classes of Cr in H∗GL(r)(pt). in Cr, and RGr(r−n,r) is the pushforward H∗GL(r)(Gr(r − n, r)) → H∗GL(r)(pt). The • is a variable living in the K-group K(bP2 × pt). The universal subbundle of the trivial bundle Cr is denoted by S. And C0 ⊠ S is a sheaf on bP2 × Gr(r, n). We consider Gr(r, n) as We now redefineRcM 1(c) Φ(E) asRcM 1(c−ne0) ′Φ(E), and return to §1.5.1. Since dimcM 1(c− ne0) < dimcM 1(c), this procedure eventually stop. 1.6. Example. Consider the case c ∈ H∗(bP2) with r(c) = r, c1(c) = 0, (∆(c), [bP2]) = 1. In Theorem 1.5 the wall-crossing term appears only in the case m = 0, j = 0. Therefore (1.9) ZcM (c) Φ(E) −ZcM 0(c) Φ(E) =ZcM 0(c−e0) Res 1=0 Φ(E♭ ⊕ C0 ⊠ e−1) e(N(E♭, C0) ⊗ e−1) e(N(C0,E♭) ⊗ e1) In the quiver description Theorem 4.2 for cM 0(c − e0) we have V0 = C, V1 = 0 and hence see that E♭ ∼= Ker(cid:2)O⊕r cM 0(c − e0) ∼= Pr−1. P → OP(1) ⊠ OC(cid:3) . Then we have N(E♭, C0) ∼= OP(−1), N(C0,E♭) ∼= OP(1)⊕2 ⊕ S, This also follows from Proposition 1.2. In fact, f1 is an isomorphism in this case. We also where OP(1) is the hyperplane bundle of P = Pr−1 and S is the universal subbundle, i.e., kernel of O⊕r P → OP(1). This also follows from Lemmas 4.9,4.11. Therefore e(N(E♭, C0) ⊗ e−1) = −c1(OP(1)) − 1, e(N(C0,E♭) ⊗ e1) = (c1(OP(1)) + 1)2e(S ⊗ e1) = r 1(c1(OP(1)) + 1). For Φ, we consider a simplest nontrivial case. Let µ(C) be the cohomology class on cM m(c) given by (1.10) µ(C) := ∆(E)/[C], 11 PERVERSE COHERENT SHEAVES ON BLOW-UP. III ✩ ✬ ✪RcM m(c) Φ(E) ✫ ❄ Use Theorem 1.5 (m − 1) times. ✲ ✲ ❄ ✩ ❄✬ ✪RcM 1(c) Φ(E) ✫ ✩ ❄✬ ✪RcM 0(c) Φ(E) ✫ Theorem 1.5 ❄ ❜ ✧ ✧ ✧ ❜ ❜ ✧ ❜ ✧ ❜ ❜ ✧ (c1, [C]) = 0 ? ✧ ❜ ❜ ✧ Yes ❜ ✧ ❜ ❜ ✧ ✧ No ❜ ✧ ✲ §1.5.3 ✏ ✲✓ ✒ ✑STOP ❄ Take • ⊗ O(C) No ✛ §1.5.5 ✧ ❜ ✧ ❜ ❄ ❜ ✧ ✧ ❜ ❜ ❜ ✧ ✧ ❜ ✧ ❜ ✧ (c1, [C]) ≥ 0 ? ❜ ❜ ✧ ✧ ✧ Yes ❜ ❜ ✧ ❄ §1.5.4 Figure 1. flowchart Hence NfYf =1 e(V1(E♭ ⊕ C0 ⊠ e−1) ⊗ emf ) = V1(E♭ ⊕ C0 ⊠ e−1) ∼= e−1. NfYf =1 We assume 2r − Nf ≥ 1 and take Φ(E) = µ(C)2r−NfQNf NfYf =1 (−c1(OP(1)) − 1 + ε1 + ε2)2r−Nf 1(c1(OP(1)) + 1)2 ZPr−1 −r Res 1=0 right hand side of (1.9) becomes (mf − 1). (mf − 1). f =1 e(V1(E) ⊗ emf ). Then the 12 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA where / denotes the slant product / : H d eT is the µ-map appearing in the usual Donaldson invariants. We have i (bP2)) → H d−i ∆(E♭ ⊕ C0 ⊠ e−1)/[C] = −c1(OP(1)) − 1 + ε1 + ε2, (bP2 ×cM m(c)) ⊗ H eT eT where ε1, ε2 are generators of Lie(C∗ × C∗) corresponding to t1, t2. We also have (cM m(c)). This By the degree reason, this must be a constant in ε1, ε2, mf . Therefore we may set all 0. Then this is equal to −ZPr−1 Nf−r 1 Res 1=0 (c1(OP(1)) + 1)2r−Nf−2 = −(cid:18)2r − Nf − 2 r − 1 (cid:19). If r = 2, Nf = 0, the answer is −2. This is a simplest case of the blow-up formula, which was used to define Donaldson invariants for c2 in the unstable range. 2. Applications -- Vanishing theorems As we mentioned above, the wall-crossing formula only gives us a recursive procedure to give the blow-up formula. In this section, we concentrate on a rather special Φ(E) and derive certain vanishing theorems. They turn out to be enough for applications to the instanton counting. 2.1. Theory with matters. Let µ(C) be as in (1.10). We consider Φ(E) = e(V0(E) ⊗ emf ) × exp (tµ(C)) , NfYf =1 and study the coefficient Φd(E) of td with small d. We assume Nf ≤ 2r hereafter. Theorem 2.1. Suppose (c1, [C]) = 0. Then ZcM m(c) Φ(E) =ZM (p∗(c)) NfYf =1 e(V(E) ⊗ emf ) + O(tk), where k = max(r + 1, 2r − Nf ). Here V(E) = R1q2∗(E ⊗ q∗1(O(−ℓ∞))) is defined from the universal sheaf E on bP2 × M(p∗(c)) as in the case of V0(E), V1(E). PERVERSE COHERENT SHEAVES ON BLOW-UP. III 13 This, in particular, means that Φd(E) = 0 for d = 1, . . . , k − 1. When Nf = 0, this vanishing was shown in [18, §6] by the dimension counting argument. The key point was that dim M(c) = 2r∆, and hence the smaller moduli spaces have codimension greater than or equal to 2r. Once the wall-crossing formula is established as in the previous section, the remaining argument below is similar, and the bound 2r − Nf comes from the f =1 e(V(E) ⊗ emf ) ∩ [M(p∗(c))] has dimension (2r − Nf )∆. Proof. Let us compute the cohomological degrees of the both sides of the equality in 2k(M0(p∗(c))). We fact that the 'virtual fundamental class'QNf Theorem 1.5, where we sayRcM m(c) ♥ has degree k if it is contained in H eT (2.2) degZcM m+1(c) have Φd(E) Φd(E) = degZcM m(c) = dimcM m(c) − Nf dim V0(c) − d = −(2r − Nf )(ch2(c), [bP2]) − d. Cm ⊠ e−i)Ψj(E♭) =ZcM m(c−jem) NfYf =1 jMi=1 e(V0(E♭) ⊗ emf ) ∪ ♥ dimcM m(c − jem) − Nf dim V0(c − jem) = − (2r − Nf )(ch2(c), [bP2]) − j(m(2r − Nf ) + r + j). Φ(E♭ ⊕ for some cohomology class ♥. Therefore its degree is at most ZcM m(c−jem) On the other hand, we can write i ), Φ(E♭ ⊕ jnMi=1 m−1Yn=0 Since j(m(2r − Nf ) + r + j) ≥ r + 1, it is zero if d ≤ r. In order to prove the vanishing for d ≤ 2r− Nf − 1, we need a refinement of the general machinery in §1.5. We need to look at each step in the flowchart (Figure 1) more closely. The first step §1.5.1 has no problem. In (1.6) we have Cn ⊗ en,1 jnYi=1 Φ(Cn ⊗ en,1 n=0 Ljn as we have a decomposition of a vector bundle V0(E ⊕Lm−1 n=0 Ljn Lm−1 m−1Mn=0 i=1 V0(Cn ⊗ en,1 i ) = Φ(E♭) Whitney sum. i ) = V0(E) ⊕ i ), and the Euler class has a multiplicative with respect to the i=1 Cn ⊗ en,1 In §1.5.4, 1.5.5 we consider the tensor product E(−C), where E is the universal family on the moduli space of 1-stable sheaves. This causes a trouble because R1q2∗(E(−C−ℓ∞)) is not a vector bundle, as mentioned ealier. We need a closer look. By [21, Lemma 7.3] the natural homomorphism H 1(E(−ℓ∞)) → H 1(E(C − ℓ∞)) is surjective for a 0-stable framed sheaf (E, Φ). Therefore H 1(E(−C − ℓ∞)) → H 1(E(−ℓ∞)) is surjective for a 1-stable framed sheaf (E, Φ). Let us give a direct proof since we need to understand the kernel. Suppose E is 1-stable. Since Hom(E,OC(−2)) = 0, we have an exact sequence 0 = H 2(Tor1(E,OC)) → H 1(E ⊗L OC) → H 1(E ⊗ OC), it implies H 1(E ⊗L OC) = 0, and hence H 1(E(−C − ℓ∞)) → H 1(E(−ℓ∞)) is surjective. E ⊗ OC/torsion =L OC(ai) with ai ≥ −1. Therefore H 1(E ⊗ OC) = 0. Since we have 14 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Since the kernel of this surjective homomorphism is H 0(E ⊗L OC), we have e(V0(E(−C)) ⊗ emf ) = e(V0(E) ⊗ emf ) e(q2∗(E ⊗L OC) ⊗ emf ) = e(V0(E) ⊗ emf ) ca(q2∗([E] ⊗ [OC]) ⊗ emf ) replace E, OC, q2∗ by their K-theory classes and the K-theory pushforward in the last expression. on cM 1(c), where a = dim H 1(E(−C − ℓ∞)) − dim H 1(E(−ℓ∞)) = (c1(E), [C]) + r and we Now V0(E) is a vector bundle over cM 0(c), cM 1(c) and master spaces from the quiver description in §4. Therefore e(V0(E) ⊗ emf ) (and also ca(q2∗([E] ⊗ [OC]) ⊗ emf )) are well- defined so we replace e(V0(E(−C)) ⊗ emf ) by the right hand side and continue the flow in Figure 1. We may still need to treat ⊗O(C) in a subsequent process in the flowchart. Then we again get e(V0(E(−C))⊗ emf ), so use the same procedure to replace by the right hand side. As a result we can write (2.3) ZcM m(c) Φd(E) =XJ ZcM 0(c−cJ) NfYf =1 e(V0(E♭) ⊗ emf )ΩJ d(E♭) for various cJ with c1(c − cJ) = 0 and cohomology classes ΩJ degree as in (2.2). On the other hand, the degree of the right hand side is at most d(E♭). The left hand side has dim M(p∗(c − cJ)) − Nf rankV0(E♭) = (2r − Nf )(∆(c − cJ), [bP2]). If cJ is nonzero, then it is at most (2r − Nf ){(∆(c), [bP2])− 1}, since (∆(c − cJ), [bP2]) is an if d < 2r−Nf . For cJ = 0, we getRcM 0(c) Φd(E), but it is equal to δd0RM (p∗(c))QNf emf ) as cM 0(c) → M(p∗(c)) is an isomorphism and µ(C) = 0 on cM 0(c). integer and we have (1.7). Therefore there is no contribution to the wall-crossing formula f =1 e(V(E)⊗ For a slightly modified version (cid:3) Φ′(E) = NfYf =1 e(V1(E) ⊗ emf ) × exp (tµ(C)) , the second part of the argument works, we get Theorem 2.4. Suppose (c1, [C]) = 0. Then ZcM m(c) Φ′(E) =ZM (p∗(c)) NfYf =1 e(V(E) ⊗ emf ) + O(t2r−Nf ). Moreover the coefficient of t2r−Nf is −(cid:18)2r − Nf − 2 r − 1 (cid:19)ZM (p∗(c)+pt) NfYf =1 e(V(E) ⊗ emf ), if Nf < 2r. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 15 For the last assertion, it is enough to calculate the case (∆(c), [bP2]) = 1 by the same argument. (See the proof of Theorem 2.6 for more detail.) Hence §1.6 gives us the answer. Next consider the c1 6= 0 case. Theorem 2.5. Suppose 0 < n := (c1, [C]) < r. Then ZcM m(c) Φ′(E) = O(tn(r−n)). In fact, we have degZcM m(c) Φ′d(E) = (2r − Nf ) dim V1(c) + n(r − n) − d, and all terms in the right hand side of (2.3) has degrees at most (2r − Nf ) dim V1(c) as dim V1(c − cJ) ≤ dim V1(c). So the same argument works. Let us state what we observed in the above proof as a general structure theorem. Let Φ(E) be a multiplicative class in the universal family E. Then Theorem 2.6. Let us fix c1 with 0 ≤ −(c1, [C]) < r. There exists a class Ωj(E, t), which is a polynomial in ci(E)/[0] (i = 2, . . . , r) with coefficients in H∗C∗×C∗(pt)[[t]] = C[ε1, ε2][[t]], and independent of ∆(c) such that (2.7) ZcM (c) Φ(E) exp(tµ(C)) =Xj≥0ZM (p∗(c)+j pt) Φ(E)Ωj(E, t). Moreover Ωj(E, t) is unique if Φ(E) 6= 0 for H∗ eT (M(r, 0, 0)) = H∗ eT For the theory with matters, the coefficients of Ωj(E, t) are in H∗ (pt) = S(eT ). C∗×C∗×(C∗)Nf (pt)[[t]] = C[ε1, ε2, m1, . . . , mNf ][[t]]. Proof. As in the derivation of (2.3), we obtain a formula as above, where M(p∗(c)+j pt) is [OC(m)]), q2∗([E]∨ ⊗ [OC(m)]) of various m. This Ωj(E, t) is independent of ∆(c), as replaced by cM 0(p∗(p∗(c) + j pt)) and Ωj(E, t) is a polynomial in Chern classes of q2∗([E]⊗ • Ψj(E♭) in Theorem 1.5 depends only on j, • the choice, whether we perform twist by O(C) or not, is determined by c1, and • the Grassmannian in Proposition 1.2 is determined by r and c1. Thus it only remains to show that we can further replace Ωj so that it is a polynomial in ci(E)/[0] (i = 2, . . . , r). by the remark after Lemma 4.11 below. Therefore Ext1 By the Grothendieck-Riemann-Roch theorem, these classes can be expressed by ci(E)/[0] = ci(q2∗([E]⊗ [OC])). Note that q2∗([E]⊗ [OC]) = −P2 a=0(−1)a Exta(OC(−1),E). If we cross q2(OC(−1),E) on cM 1 back the wall from cM 0 to cM 1, we have Ext0 rank r, and ci(E)/[0] vanishes for i > r. Since the difference of the integrals over cM 1 and cM 0 are expressed by integrals over small moduli spaces, we can eventually express Ωj as a polynomial in ci(E)/[0] (i = 2, . . . , r) by a recursion. q2(OC(−1),E) is a vector bundle of q2(OC(−1),E) = 0 = Ext2 16 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA possible ∆(c) with cM (c) 6= ∅, i.e., the case when c + (c1, [C])e0 = (r, 0, 0) (cf. Proposi- Let us show the uniqueness of Ωj(E, t) by a recursion on j. Let us take the smallest tion 1.2). Then we only have the term with j = 0 in the right hand side of (2.7). In this case, p∗(c) = (r, 0, 0) and the moduli space M(r, 0, 0) is a single point. Therefore Ω0(E, t) is determined by (2.7). Now suppose that Ωj(E, t) with j < n are determined. Then we take c whose ∆(c) is n larger than the previous smallest ∆(c). Then j in the summation in (2.7) runs from 0 to n. Moreover M(p∗(c) + n pt) = M(r, 0, 0) is a single point again. Therefore Ωn(E, t) is determined from (2.7) and Ωj(E, t) with j < n. Suppose that the degree of RcM (c) Φ(E) is of a form γ∆(c) + a(c1, [C]) + b for some constants γ, a and b, depending only on r. We further assume that γ > 0, as in the case γ = 2r − Nf > 0 for the theory with matters. Then (cid:3) degZcM (c) degZM (p∗(c)+j pt) Φ(E)µ(C)d = γ∆(c) + a(c1, [C]) + b − d, Φ(E)Ωj(E, t) ≤ γ(∆(p∗(c)) − j) + b, Since we are fixing (c1, [C]) as in Theorem 2.6, the two degree cannot match if j is too large. This means that we only need to calculate finitely many Ωj(E, t) to determine RcM (c) Φ(E)µ(C)d for a fixed d. (In practice, the maximal j can be calculated explicitly.) 2.2. K-theory version. We derive the wall-crossing formula for a K-theoretic integra- tion via the Grothendieck-Riemann-Roch formula. Let td(α) be the Todd class of a K-theory class α on various moduli spaces. The Todd class of the tangent bundle TM of a variety M is denoted by td M. We have For integers d, l and a = 0 or 1, we consider tdcM m(c) = td(N(E,E)). Φ(E) = td(N(E,E)) exp(lc1(Va(E))) exp(−d ch2(E)/[C]). (2.8) By a discussion in [20, several paragraphs preceding Def. 2.1], − ch2(E)/[C] is the first Chern class of an equivariant line bundle up to a (rational) cohomology class in H 2 (pt), eT which is zero if (c1, [C]) = 0. Since this difference is immaterial in the following discussion (in particular in Theorem 2.11), we identify − ch2(E)/[C] with the equivariant line bundle, which we denote by µ(C). (This is the same as µ(C) in the previous subsection up to a class in H 2 eT (pt).) Then the equivariant Riemann-Roch theorem (see e.g., [9]) we have ZcM m(c) Φ(E) = τ(cid:0)bπ∗(µ(C)⊗d ⊗ detVa(E)⊗l)(cid:1) , where τ is the equivariant Todd homomorphism τ : K eT (M0(p∗(c))) → H eT ∗ the Uhlenbeck partial compactification M0(p∗(c)), andbπ∗ is the push-forward homomor- phism in the equivariant K-theory. (M0(p∗(c))) for We have td(N(E,E)) = td(N(E1,E1)) td(N(E1,E2)) td(N(E2,E1)) td(N(E2,E2)) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 17 if E = E1 ⊕ E2. Then from Theorem 1.5 we get ZcM m+1(c) j!" jYi=1 1 Φ(E) = ∞Xj=1ZcM m(c−jem) Φ(E) −ZcM m(c) exp(cid:0)lc1(Va(Cm ⊗ e−i)) − d ch2(Cm ⊗ e−i)/[C](cid:1) Φ(E♭) ∪ Res eK(N(•, Cm) ⊗ e−i) eK(N(Cm,•) ⊗ ei) Here eK is the (Chern character of) K-theoretic Euler class: where Ψj(•) := j =0· · · Res 1=0 Y1≤i16=i2≤j Ψj(E♭), (1 − ei1−i2 )# . eK(α) = e(α) td(α)−1 = where e(α) is the usual Euler class as before. ∞Xp=0 (−1)p ch(Vpα∨), Strictly speaking, we need to consider the completion C[−1 , i]] for the coefficient rings of the localized equivariant homology groups of moduli spaces, as for example, e−i is not allowed. Here C[−1 {j < 0 aj 6= 0} is finite. The modification appears only at §6.1 and the beginning of §6.3, and the rest of the proof remains unchanged. Observe that i appears always as a function in xi := e−i − 1 in the above formula. We change the coefficient ring from C[−1 We have , i]] is the algebra of formal power series P ajj such that , i]] to C[x−1 , xi]]. i i i i Res i=0 f (e−i − 1) = − Res xi=0 f (xi) xi + 1 as dxi = −e−idi = −(xi + 1)di. Therefore Theorem 2.9. where ZcM m+1(c) Ψj(•) := Φ(E) −ZcM m(c) jYi=1 1 j! Φ(E) = ∞Xj=1ZcM m(c−jem) Φ(E♭) ∪ Res xj =0· · · Res x1=0 Ψj(E♭), exp (lc1(Va(Cm ⊗ (1 + xi))) − d ch2(Cm ⊗ (1 + xi))/[C]) eK(N(•, Cm) ⊗ (1 + xi)) eK(N(Cm,•) ⊗ 1 )(−(1 + xi)) 1+xi × Y1≤i16=i2≤j xi1 − xi2 1 + xi1 . In view of Proposition 1.2 the following is useful to replace the K-theoretic integration on cM 1(c) by one on bN(c, n). Lemma 2.10. Consider the diagram in Proposition 1.2. We have Rf1∗(O bN (c,n)) = OcM 1(c). Proof. Since f1 is a proper birational morphism between smooth varieties, this is a well- known result (see e.g., [11, §5.1]). (cid:3) 18 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA In the remaining of this section we study the vanishing theorem for small d. Theorem 2.11. Assume 0 ≤ l ≤ r. (1) Suppose (c1, [C]) = 0. If 0 ≤ al + d ≤ r, ZcM(c) Φ(E) =ZM (p∗(c)) td(N(E,E)) exp(lc1(V(E))), where V(E) is the vector bundle defined as in Theorem 2.1. (2) Suppose a = 1 and 0 < (c1, [C]) < r. If 0 < d ≤ min(r + (c1, [C]) − l, r − 1), Φ(E) = 0. This result was conjectured in [8, (1.37),(1.43)]. ZcM (c) Proof. We study factors of Ψj(E♭) more closely. Note that E♭ is the universal sheaf for cM m(c − jem), hence rank N(Cm,E♭) = (m + 1)r + (c1, [C]) + j, rank N(E♭, Cm) = mr + (c1, [C]) + j. If {αa}, {βb} are Chern roots of N(Cm,E♭), N(E♭, Cm) respectively, we have 1 (2.12) eK(N(Cm,E♭) ⊗ 1 1+xi ) 1 eK(N(E♭, Cm) ⊗ (1 + xi)) On the other hand, we have = exp(Pa αa) (−xi)(m+1)r+(c1,[C])+jYa xi(cid:19)mr+(c1,[C])+jYb =(cid:18)1 + 1 1 , 1 + 1−eαa xi 1 1 + 1−e−βb xi . exp(lc1(Va(Cm ⊗ (1 + xi)))) = exp(lc1(Va(Cm)))(1 + xi)l(m+a) as rankVa(Cm) = m + a. Also exp(−d ch2(Cm ⊗ (1 + xi))/[C]) = exp(−d ch2(Cm)/[C])(1 + xi)d, as ch1(Cm)/[C] = −1. remaining factor Let us expand Ψj(E♭) into formal Laurent power series in x1. Note that we have the = Qi26=1(x1 − xi2) (1 + x1)j 1 x1 − xi2 1 + x1 (1 + x1)Yi26=1 x1(cid:19)mr+(c1,[C])+j (cid:18)1 + 1 Qi26=1(x1 − xi2) (1 + x1)j from Ψj(•) in Theorem 2.9. This term can be absorbed into the second equality of (2.12) as Note that mr + (c1, [C]) ≥ 0, and hence (1 + 1 write x1 )mr+(c1,[C]) is a polynomial in x−1 1 . We also (1 + x1)l(m+a)+d = xl(m+a)+d 1 (1 − xi2 x1 ). = 1 x1(cid:18)1 + 1 x1(cid:19)mr+(c1,[C])Yi26=1 x1(cid:19)l(m+a)+d 1 (cid:18)1 + PERVERSE COHERENT SHEAVES ON BLOW-UP. III 19 as a Laurent polynomial in x−1 assumption. Therefore we have 1 . Note that we have l(m + a) + d ≥ la + d ≥ 0 by our for some formal power series f (x−1 1 xN 1 f (x−1 1 ) Ψj(E♭) = 1 ) in x−1 1 with N = (m + 1)r + (c1, [C]) + j + 1 − l(m + a) − d = m(r − l) + (r − d) + (c1, [C]) − la + j + 1. Since 0 ≤ r − l and 0 ≤ m, the first term m(r − l) is nonnegative. We also have j ≥ 1. Therefore we have N ≥ 2 if d+la ≤ r+(c1, [C]). This shows that there are no wall-crossing term, i.e., ZcM(c) Φ(E) =ZcM 0(c) Φ(E). (p∗(E), Φ), Π−1(F, Φ) = (p∗(F ), Φ) (see §1.5.3 for a precise statement). Therefore the If (c1, [C]) = 0, we have an isomorphism Π : cM 0(c) → M(p∗(c)) given by Π(E, Φ) = tangent bundles N(E,E) for cM(c) and M(p∗(c)) are isomorphic to each other. Vector bundles V0(E) for cM (c) and M(p∗(c)) are isomorphic to each other from the description To show (2), we consider cM 0(c) ∼= cM 1(ce[C]) given by (E, Φ) 7→ (E(C), Φ). Then ZcM 0(c) of Π in §1.5.3. We also have ch2(E)/[C] = 0. These show (1). Proposition 1.2 is applicable. We then have td(N(E,E)) exp(lc1(V0(E))) exp(−d ch2(E(−C))/[C]) Φ(E) =ZcM 1(ce[C]) where E in the right hand side is the universal sheaf for cM 1(ce[C]). As we explained in the beginning of this subsection, we may replace this integral by τ (bπ∗(µ(C)⊗d ⊗ det V0(E)⊗l)). The difference between ch2(E(−C))/[C] and ch2(E)/[C] is immaterial, as it is a class pulled back from M0(p∗(c)). By Lemma 2.10 and the projection formula, we can replace the above by the push-forward from N(ce[C], n) with n = r − (c1, [C]), where µ(C), V0(E) are replaced by their pull-backs by f1. From the exact sequence in Proposition 1.2(3) we have f∗1V0(E) = f∗2V0(E′) where E′ is the universal sheaf for cM 0(ce[C] − ne0). We also have up to the pull-back of a line bundle from cM 0(ce[C] − ne0) by f2. Therefore it is enough to f∗1 µ(C) = det S show that f2∗(detS⊗d) = 0 for 0 < d < r. Since f2 is a Grassmann bundle of n-planes in a rank r vector bundle, this vanishing is a special case of Bott vanishing theorem [1] or a direct consequence of Kodaira vanishing theorem. (cid:3) 20 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA 2.3. Casimir operators. We generalize the vanishing result in the previous subsection to the case when we integrate certain K-theoretic classes given by universal sheaves on moduli spaces. × exp Xp∈Z\{0} Let ψp be the pth Adams operation (see e.g., [5, I.§6]). We will use it for p < 0 defined by ψp(x) = ψ−p(x∨) . For indeterminates ~τ = (· · · , τ−2, τ−1, τ1, τ2,· · · ) and ~t = (· · · , t−2, t−1, t1, t2,· · · ) we consider a generalization of (2.8) with a = 0: Φ(E) = td(N(E,E)) exp(lc1(V0(E)) − d ch2(E)/[C]) τp ch(cid:16)ψp(E)/[bC2](cid:17) + tp ch(cid:16)ψp(E) ⊗ OC(−1)/[bC2](cid:17) , where the K-theoretic slant product •/[bC2] is defined by •/[bC2] = q2∗(•⊗q∗1p∗(K 1/2 p : bC2 → C2. We have two remarks for the definition. First we define the push-forward q2∗ by the localization formula, i.e., the sum of the fixed point contributions, since q2 is not proper. This is a standard technique in instanton counting and its meaning was explained in detail in [18, §4]. Second K 1/2 is the trivial line bundle together twisted by the square root of the character of (C∗)2, which we consider as a character of its double cover. Having the above two remarks in mind, we see that the above integral is essentially defined by the equivariant K-theory push-forward as before. C2 )) with C2 We expandRcM m(c) Φ(E) in tp, τp and consider coefficients p ZcM m(c) ZcM m(c) Φ~n, ~m(E), where ~n = (· · · , n−1, n1,· · · ), ~m = (· · · , m−1, m1,· · · ). If we set , τ np p tmp Φ(E) =X~n, ~mYp6=0 (cid:18) ∂ :=Yp6=0(cid:18) ∂ ∂~τ(cid:19)~n ∂τp(cid:19)np ~n ! ~m !(cid:18) ∂ ∂~τ(cid:19)~n(cid:18) ∂ ∂~t(cid:19) ~mZcM m(c) × ch Op (cid:16)ψp(E)/[bC2](cid:17)⊗np 1 np!, ~n ! =Yp6=0 Φ(E)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)~τ =~t=0 ⊗(cid:16)ψp(E) ⊗ OC(−1)/[bC2](cid:17)⊗mp! . we have Φ~n, ~m(E) = ZcM m(c) ~n ! ~m !ZcM m(c) 1 = td(N(E,E)) exp(lc1(V0(E)) − d ch2(E)/[C]) Proposition 2.13. Suppose (c1, [C]) = 0. (1) Assume the followings: (a) 0 ≤ d +Pp<0 pnp + pmp, and (b) d +Pp>0 pnp + pmp ≤ r. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 21 Then the wall-crossing term is zero, i.e., (2) We further assume mp 6= 0 for some p. Then Φ~n, ~m(E). ZcM m(c) Φ~n, ~m(E) =ZcM 0(c) ZcM m(c) Φ~n, ~m(E) = 0. Proof. (1) We note that ψp(E♭ ⊕ Cm ⊗ (1 + xi)) = ψp(E♭) + ψp(Cm) ⊗ (1 + xi)p, as the Adams operation is a homomorphism. Then the proof exactly goes as before. (2) Recall thatcM 0(c) ∼= M(p∗(c)) under (E, Φ) 7→ (p∗(E), Φ). Then ψk(E)⊗OC(−1)/[bC2] vanishes, since p∗(OC(−1)) = 0. (cid:3) 3. Partition function and Seiberg-Witten curves In this section we explain an application of the vanishing theorems to Nekrasov partition functions. Here a reader is supposed to be familiar with [18, 19, 20, 8]. (ε1, ε2, ~a; Λ, ~τ ) :=Xc ×ZM (c) 3.1. Partition function. Let us fix l ∈ Z. We define a partition function as the gener- ating function of integrals considered in §§2.2, 2.3: (3.1) Z inst (Λ2re−(r+l)(ε1+ε2)/2)(∆(c),[P2]) l generators ai of H∗T (pt) and ε1, ε2 ones of H∗(C∗)2(pt), and the integrals, more coefficients (pt) as explained in §1.3. td M(c) exp(lc1(V(E))) exp Xp∈Z\{0} τp ch(cid:0)ψp(E)/[C2](cid:1) , where the rank r = r(c) is fixed. Here ~a = (a1, . . . , ar) (P aα = 0) is the vector given by of monomials in τp's, take values in the quotient field S(eT ) of H∗ (More precisely the quotient field of the representation ring R(eT ) = Z[e±ε1, e±ε2, e±aα] as in §2.2 since this is a K-theoretic partition function.) And the K-theoretic slant product •/[C2] is defined by q2∗(• ⊗ q∗1(K 1/2 (3.2) (cid:18) ∂ ∂~τ(cid:19)~n (ε1, ε2, ~a; Λ, ~τ = 0) =Xc (Λ2re−(r+l)(ε1+ε2)/2)(∆(c),[P2]) C2 )) as in §2.3. We have Z inst eT l ×ZM (c) ch Op (cid:0)ψp(E)/[C2](cid:1)⊗np! td M(c) exp(lc1(V(E))). Therefore Z inst operators applied to the universal sheaves. l (ε1, ε2, ~a; Λ, τ ) gives us integrals of any tensor products of various Adams 22 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA We consider fixed points of the eT -action on M(c) as in [18, §2]: they are parametrized by r-tuples of Young diagrams ~Y = (Y1, . . . , Yr) with ~Y =PYα = (∆(c), [P2]) corre- sponding to direct sums of monomial ideals in C[x, y]. The character of the fiber of V(E) at the fixed point ~Y is given by ch(V(E)~Y )(ε1, ε2, ~a) = rXα=1 where a′(s), l′(s) are as in [18, §2]. We also have exp(lc1(V(E)~Y )) = exp"l rXα=1Xs∈Yα Therefore we have e−l′(s)ε1−a′(s)ε2, eaαXs∈Yα (aα − l′(s)ε1 − a′(s)ε2)# . Z inst l (3.3) (ε1, ε2, ~a; Λ, ~τ) =X~Y × exp"l rXα=1Xs∈Yα × exp"Xp,α τp (Λ2re−r(ε1+ε2)/2)~Y n~Y α,β(ε1, ε2, ~a) Yα,β (aα − l′(s)ε1 − a′(s)ε2 − ε1 + ε2 2 )# epaα(cid:8)1 − (1 − e−pε1)(1 − e−pε2)Ps∈Yα e−pl′(s)ε1−pa′(s)ε2(cid:9) (eε1/2 − e−ε1/2)(eε2/2 − e−ε2/2) # , where n~Y α,β(ε1, ε2, ~a) is the alternating sum of characters of exterior powers of the cotangent space of M(r, n) at the fixed point ~Y . Its explicit formula was given in [20, §1.2], where it was denoted by n~Y α,β(ε1, ε2, ~a; β), and we put β = 1. We have Z inst (3.4) where ′~τ is given by ′τp = τ−p. This symmetry is a simple consequence of [8, the displayed formula one below (1.33)], or the Serre duality. −l (−ε1,−ε2,−~a; Λ, ′~τ ), (ε1, ε2, ~a; Λ, ~τ ) = Z inst l Let d ∈ Z≥0. We consider a similar partition function on the blow-up: × ZcM (c) (Λ2re−(r+l)(ε1+ε2)/2)(∆(c),[P2]) l,k,d(ε1, ε2, ~a; Λ, ~τ, ~t) :=Xc bZ inst tdcM (c) exp(lc1(V0(E)) − d ch2(E)/[C]) × exp Xp∈Z\{0} where we also fix (c1(c), [C]) = −k in this case. τp ch(cid:16)ψp(E)/[bC2](cid:17) + tp ch(cid:16)ψp(E) ⊗ OC(−1)/[bC2](cid:17) , PERVERSE COHERENT SHEAVES ON BLOW-UP. III 23 This is related to the partition function (3.1) by l,k,d(ε1, ε2, ~a, Λ, ~τ , ~t) (e(ε1+ε2)(d−(r+l)/2)Λ2r)(~k2)/2e(d−l/2)(~k,~a) (3.5) P kα=k bZ inst = X~k=(kα)∈Zr × exp"l 1 × Z inst l × Z inst l ~α(ε1, ε2, ~a) Q~α∈∆ l~k 2Xα (ε1 + ε2)Xα 6 (ε1, ε2 − ε1, ~a + ε1~k; Λe (ε1 − ε2, ε2, ~a + ε2~k; Λe k3 α + 1 k2 αaα!# ε1, e−ε1/2(~τ + (eε1 − 1)~t)) d−(r+l)/2 2r d−(r+l)/2 2r ε2, e−ε2/2(~τ + (eε2 − 1)~t)), where l~k ~α(ε1, ε2, ~a) is a function given in [20, (2.3)]. Let us briefly explain how this formula is proved. It is a consequence of the Atiyah- Bott-Lefschetz fixed point formula applied to the eT -action on cM (c). The fixed points are parametrized by (~k, ~Y 1, ~Y 2), where ~k ∈ Zr corresponds to a line bundle O(kαC), and ~Y 1, ~Y 2 are Young diagrams corresponding to monomial ideals in the toric coordinates at the C∗ × C∗-fixed point p1 = ([1 : 0 : 0], [1 : 0]) and p2 = ([1 : 0 : 0], [0 : 1]) in bP2. (See [18, §3] for more detail.) The structure of the above formula, i.e., the sum over ~k of the product of two partition functions comes from this description of the fixed point set. The shift of variables in the partition functions come from study of tangent bundles, universal sheaves at fixed points. All these are done in [18, §3], [8, §1.7], except the expression e−εa/2(~τ + (eεa − 1)~t) (a = 1, 2) appears for variables for the Adams operators. If we replace the Adams operator ψp by the degree p part of the Chern character, the expression was given in [19, §4], where we just need to change variables as ~τ + εa~t. In our situation, ~t is multiplied by ch(OC(−1))pa = eεa − 1 instead of εa. The factor e−εa/2 appears as the 'square root' of KC2 ⊗ K−1 bC2 at the fixed point pa, since the K-theoretic slant product •/[bC2] was defined as q2∗(• ⊗ q∗1p∗(K 1/2 bC2 ))'. (We C2 )), not as 'q2∗(• ⊗ q∗1(K 1/2 bC2 , which cannot be defined.) avoid K 1/2 We have bZ inst l,k,d(ε1, ε2, ~a; Λ, ~τ = 0, ~t) = bZ inst −l,k,r−d(−ε1,−ε2,−~a; Λ, ~τ = 0,−′~t). This is proved exactly as in [8, the last displayed formula in §1.7.1] and (3.4), or the Serre duality. 24 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA We define the perturbation part, see [20, §4.2] and [8, §1.7.2] for more details. We set + x2 log Λ! , 1 1 1 2 1 n e−nx γε1,ε2(x; Λ) := (ε1 + ε2)(cid:19)3 (enε1 − 1)(enε2 − 1) 1 2ε1ε2 − 6(cid:18)x + +Xn≥1 ε1ε2(cid:18)π2x 6 − ζ(3)(cid:19) eγε1,ε2(x; Λ) := γε1,ε2(x; Λ) + 2ε1ε2 (cid:18)x log Λ + 6(cid:19) + for (x, Λ) in a neighborhood of √−1R>0 × R>0. Zl(ε1, ε2, ~a; Λ, ~τ) := exp"−Xα6=βeγε1,ε2(aα − aβ; Λ) − l rXα=1 bZl,k,d(ε1, ε2, ~a; Λ, ~τ , ~t) 6ε1ε2#bZ inst := exp"−Xα6=βeγε1,ε2(aα − aβ; Λ) − l We define the full partition function by rXα=1 ε1 + ε2 a3 α π2 + ε2 1 + ε2 2 + 3ε1ε2 12ε1ε2 log Λ a3 α 6ε1ε2# Z inst l (ε1, ε2, ~a; Λ, ~τ ), l,k,d(ε1, ε2, ~a; Λ, ~τ, ~t). Using the difference equation satisfied by the perturbation terms (see [20, §4.2] and [8, §1.7.2]), we can absorb the factors in (3.5) coming from line bundles O(kαC) into the partition function to get (3.6) bZl,k,d(ε1, ε2, ~a; Λ, ~τ , ~t) = exp(cid:20)(cid:26)− × X{~l}=− k r (4 (d + l (−1/2 + k/r)) − r) (r − 1) + 48 k3l 6r2(cid:27) (ε1 + ε2)(cid:21) ε1 ε2 kε1 kε2 r ⋆ e− ε1 2}Λ, e r )− r 2 + k 2}Λ, e r )− r 2r{d+l(− 1 ~l; e 2 + k 2r{d+l(− 1 2 (~τ + (eε1 − 1)~t)) Zl(ε1, ε2 − ε1, ~a + ε1~l; e × Zl(ε1 − ε2, ε2, ~a + ε2 where ~l runs over the set {~l = (lα)r is defined as 2 (~τ + (eε2 − 1)~t)), α=1 ∈ Qr P lα = 0, lα ≡ −k/r mod Z}, and ekεa/r ⋆ ~τ We shift from the previous ~k to ~l by lα = kα − k/r. The effect of this shift was calculated in [8, (1.36)] when ~τ = ~t = 0, and we have used it here. And ekεa/r⋆ comes from O(kαC) = O((lα + k/r)C). The part lα is absorbed into the shift ~a + εa~l, but we need the remaining contribution from k/r. ekεa/r ⋆ ~τ = (· · · , e−2kεa/rτ−2, e−kεa/rτ−1, ekεa/rτ1, e2kεa/rτ2,· · · ). r ⋆ e− ε2 Remark 3.7. We do not make precise to which ring the full partition functions belong, as a function in Λ. We just use them formally to make a formula shorter as above. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 25 This applies all formulas below until they (more precisely their leading coefficients) are identified with one defined via Seiberg-Witten curves, which are really functions defined over an appropriate open set in Λ. 3.2. Regularity at ε1 = ε2 = 0. We assume 0 ≤ l ≤ r hereafter. Theorem 2.11(1) means bZl,0,d(ε1, ε2, ~a; Λ, ~τ = 0, ~t = 0) = Zl(ε1, ε2, ~a; Λ, ~τ = 0) (3.8) for 0 ≤ d ≤ r. This was conjectured in [8, (1.37)]. Combined with (3.5), we see that coefficients of Z inst(ε1, ε2, ~a; Λ, ~τ = 0) in Λn are determined recursively if the above holds two different values of d, as explained in [19, §5.2]. The equation (3.8), the left hand side replaced by (3.6), is called the blow-up equation. It gives a strong constraint on the partition function Z(ε1, ε2, ~a; Λ, ~τ = 0). As an application, in [loc. cit., Prop. 1.38] we proved the followings under [loc. cit., (1.37)]: (3.9) (3.10) l (ε1,−2ε1, ~a; Λ, ~τ = 0) = Z inst Z inst (2ε1,−ε1, ~a; Λ, ~τ = 0) ε1ε2 log Zl(ε1, ε2, ~a; Λ, ~τ = 0) is regular at ε1 = ε2 = 0. l if l 6= r, (More precisely only the proof of (3.9) was given in [loc. cit.,(1.37)]. The proof of (3.10) was omitted since it is the same as [20, Th. 4.4].) Remark 3.11. Though it was not stated explicitly in [loc. cit.], the second assertion holds even if l = r. On the other hand, the first one follows from (3.8) for d and r + l− d, which must be different. Therefore l 6= r is required. Let us apply the same argument to Proposition 2.13. We expand Z inst as before: l (3.12) Z inst l τ np p Z inst ~n (ε1, ε2, ~a; Λ) (ε1, ε2, ~a; Λ, ~τ) =X~n Yp6=0 ~n !(cid:18) ∂ ∂~τ(cid:19)~n 1 with ~n = (· · · , n−1, n1,· · · ). Thus Z inst ~n (ε1, ε2, ~a; Λ) = Z inst l (ε1, ε2, ~a; Λ, ~τ = 0). By Proposition 2.13 we have (3.13) (3.14) if ~n is nonzero and satisfies (cid:18) ∂ ∂~t(cid:19)~n bZl,0,d(ε1, ε2, ~a; Λ, ~τ = 0, ~t = 0) = 0 −Xp<0 pnp ≤ d ≤ r −Xp>0 pnp. After substituting (3.6) to the left hand side, we also call this as the blow-up equation. For simplicity we assume ~n is supported on either Z>0 or Z<0, i.e., np = 0 for any p < 0 or np = 0 for any p > 0. We say ~n is positive in the first case, and negative in the second case. 26 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA We define 'log Z(ε1, ε2, ~a; Λ, ~τ)' as follows. Since the perturbative part is already the exponential of something, we only need to define 'log Z inst'. Then observe that Z inst(ε1, ε2, ~a; Λ, ~τ) = exp Xp,α (eε1/2 − e−ε1/2)(eε2/2 − e−ε2/2)! × (1 + O(Λ)). τpepaα formal power series in Λ. τpepaα/(eε1/2−e−ε1/2)(eε2 /2−e−ε2/2) and a Proposition 3.15. (1) Suppose that ~n satisfies the followings: Therefore log Z inst can be defined as the sum ofPp,α (a) If P np is odd, −(r + l)/2 ≤Pp<0 pnp (negative case) or Pp>0 pnp ≤ (r − l)/2 (b) IfP np is even, the strict inequality holds. (2) If −r <Pp<0 pnp (negative case) orPp>0 pnp < r (positive case), Z~n(ε1,−2ε1, ~a; Λ) = Z~n(2ε1,−ε1, ~a; Λ). (positive case). Then log Zl(ε1, ε2, ~a; Λ, ~τ = 0) ∂~τ(cid:19)~n ε1ε2(cid:18) ∂ is regular at ε1 = ε2 = 0. Proof. Since the proof of (2) is the same as that of (3.10), we only prove (1). We expand Z inst l as Z inst l (ε1, ε2, ~a; Λ, ~τ ) = Note that Z0(ε1, ε2, ~a, ~τ ) = exp"Xp,α τp ZN (ε1, ε2, ~a, ~τ )Λ2rN . ∞XN =0 (eε1/2 − e−ε1/2)(eε2/2 − e−ε2/2)# . epaα Therefore the assertion is true for Z0. We prove the assertion by the induction on N. We further expand as ZN (ε1, ε2, ~a, ~τ ) =X~n Yp6=0 τ np p ZN,~n(ε1, ε2, ~a) as in (3.12). Fix ~n and N and consider the coefficient of Λ2rNQ τ np p have 0 = X(~k,~k)/2+N1+N2=N ~n1+~n2=~n (3.16) (−1)P n2,p e(d−l/2)(~k,~a)e(N1−N2)(d−(r+l)/2)ε1 αaα(cid:19) 2X k2 × ZN1,~n1(ε1,−2ε1, ~a + ε1~k)ZN2,~n2(2ε1,−ε1, ~a − ε1~k) Qα∈∆ l~k exp(cid:18) l α(ε1,−ε1, ~a) in (3.5). Setting ε2 = −ε1, we PERVERSE COHERENT SHEAVES ON BLOW-UP. III 27 pnp. if ~n 6= 0 by the blow-up equation ((3.13) and (3.5)). Here the summation is over ~k, ~n1, ~n2, N1, N2 and we write ~n2 = (· · · , n2,−1, n2,1,· · · ). We assume We suppose that the same equality holds for r + l − d. So we assume By our assumption, there exists d satisfying both inequalities, e.g., we take d = r − pnp ≤ d ≤ r −Xp>0 pnp ≤ d ≤ r + l +Xp<0 −Xp<0 l +Xp>0 Pp>0 pnp in the positive case, r + l +Pp<0 pnp in the negative case. Moreover, we may assume d 6= (r + l)/2 ifP np is even. Substituting r + l − d into d in (3.16), and replacing ~k by −~k, (N1, N2) by (N2, N1) and 0 = X(~k,~k)/2+N1+N2=N (−1)P n1,p e(d−l/2−r)(~k,~a)e(N1−N2)(d−(r+l)/2)ε1 αaα(cid:19) exp(cid:18) l 2X k2 × ZN1,~n1(2ε1,−ε1, ~a + ε1~k)ZN2,~n2(ε1,−2ε1, ~a − ε1~k). Qα∈∆ l−~k (~n1, ~n2) by (~n2, ~n1), we get α (ε1,−ε1, ~a) ~n1+~n2=~n pnp. Note er(~k,~a)/2 ~α(ε1,−ε1, ~a) = Q~α∈∆ l~k e−r(~k,~a)/2 ~α (ε1,−ε1, ~a) Q~α∈∆ l−~k by [20, Lem. 4.1 and (4.2)]. Note also that P n1,p = P np −P n2,p and hence we can replace (−1)P n1,p by (−1)P n2,p in the above formula. Then the only difference of the above two equations are variables for ZN1,~n1 and ZN2,~n2. By the induction hypothesis those are also equal if N1, N2 < N. Therefore we have 0 = X~n1+~n2=~n (3.17) = X~n1+~n2=~n (−1)P n2,p(cid:16)eN (d−(r+l)/2)ε1 ZN,~n1(ε1,−2ε1, ~a)Z0,~n2(2ε1,−ε1, ~a) + e−N (d−(r+l)/2)ε1 Z0,~n1(ε1,−2ε1, ~a)ZN,~n2(2ε1,−ε1, ~a) − eN (d−(r+l)/2)ε1 ZN,~n1(2ε1,−ε1, ~a)Z0,~n2(ε1,−2ε1, ~a) − e−N (d−(r+l)/2)ε1 Z0,~n1(2ε1,−ε1, ~a)ZN,~n2(ε1,−2ε1, ~a)(cid:17), (−1)P n2,pheN (d−(r+l)/2)ε1 Z0,~n2(2ε1,−ε1, ~a) + e−N (d−(r+l)/2)ε1 Z0,~n1(ε1,−2ε1, ~a) × {ZN,~n1(ε1,−2ε1, ~a) − ZN,~n1(2ε1,−ε1, ~a)} × {ZN,~n2(2ε1,−ε1, ~a) − ZN,~n2(ε1,−2ε1, ~a)}i, if ~n 6= 0. (3.9). We now prove the assertion by induction on ~n. The case ~n = 0 is treated already in 28 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Now we assume that the assertion holds for smaller ~n. Note that the assumption on ~n implies that on smaller ones. Then the only remaining terms in (3.17) are either ~n1 = 0 or ~n2 = 0. Therefore we have so the above is a nontrivial equality. 0 =(cid:0)eN (d−(r+l)/2)ε1 − (−1)P npe−N (d−(r+l)/2)ε1(cid:1) {ZN,~n(ε1,−2ε1, ~a) − ZN,~n(2ε1,−ε1, ~a)} . Hence we have the assertion for ZN,~n. Note that we take d 6= (r + l)/2 whenP np is even, Remark 3.18. We need the vanishing (3.16) for the case whenP np is odd, but it is enough to suppose that the right hand side of (3.16) is the same for d and r + l − d whenP np is even. In particular, if we use (3.8) instead of (3.13), the above argument works even for ~n = 0. This is nothing but the proof of (3.9) in [8]. (cid:3) 3.3. Contact term equations. We expand as ε1ε2 log Zl(ε1, ε2, ~a; Λ, ~τ ) = F0(~a; Λ, ~τ) + (ε1 + ε2)H(~a; Λ, ~τ ) + ε1ε2A(~a; Λ, ~τ ) + Proposition 3.15(2) so that singular terms do not appear. where we only consider (∂/∂~τ )~n(cid:12)(cid:12)~τ =0 applied to this function, with ~τ in the range in 0) = −π√−1h~a, ρi, where ρ is the half sum of the positive roots. More generally by Propo- By (3.9) H(~a; Λ, ~τ = 0) comes only from the perturbative part. Thus we have H(~a; Λ, ~τ = sition 3.15(1) we have ε2 1 + ε2 2 B(~a, Λ, ~τ) + · · · , 3 (cid:18) ∂ ∂~τ(cid:19)~n H(~a; Λ, ~τ = 0) = 0 if ~n satisfies the condition there. We introduce a new coordinate system for ~a by ai = a1 + a2 +· · · + ai (i = 1, . . . , r − 1) Let as in [18, §1]. We also define ki for ~k in the same way. 1 ∂2F0(~a; Λ, ~τ = 0) . (3.19) τij = − ΘE(~ξτ ) = X~k∈Zr−1 ∂ai∂aj Let ΘE(~ξτ ) be the Riemann theta function defined by 2π√−1 exp π√−1Xi,j We substitute (3.6) into the left hand side of (3.8), and take the limit of ε1, ε2 → 0. Using H(~a; Λ, ~τ = 0) = −π√−1h~a, ρi, we obtain exp(B − A) = exp"− τijkikj + 2π√−1Xi )! . ki(ξi + 1 8r2(cid:18)d − 2 (cid:19)2 r + l 1 2 ∂2F0 (∂ log Λ)2#× ΘE(cid:18)− 2π√−1 1 1 2r(cid:18)d − r + l ∂ log Λ∂~a(cid:12)(cid:12)(cid:12)(cid:12) τ(cid:19) 2 (cid:19) ∂2F0 for 0 ≤ d ≤ r as in [20, §4]. Here F0, A, B are evaluated at (~a; Λ, ~τ = 0). PERVERSE COHERENT SHEAVES ON BLOW-UP. III 29 In particular, the right hand side is independent of d. Dividing by the expression for d = (r + l)/2 (if r + l is even), d = (r + l − 1)/2, (if r + l is odd), we have (3.20) ∂2F0 ΘE(0τ ) ΘE(cid:16)− 1 2 (cid:1) 4π√−1r(cid:0)d − r+l ΘE(cid:16)− 1 2 (cid:1) 4π√−1r(cid:0)d − r+l ∂ log Λ∂~a(cid:12)(cid:12)(cid:12) τ(cid:17) ΘE(cid:16) ∂ log Λ∂~a(cid:12)(cid:12)(cid:12) τ(cid:17) ∂ log Λ∂~a(cid:12)(cid:12)(cid:12) τ(cid:17) 8π√−1r ∂2F0 ∂2F0 1 r + l = exp" 1 = exp" 1 8r2(cid:18)d − 8r2((cid:18)d − 2 (cid:19)2 2 (cid:19)2 r + l ∂2F0 (∂ log Λ)2# , 4) ∂2F0 (∂ log Λ)2# − 1 according to either r + l is even or odd. This equation is called the contact term equation. This equation determines F0(~a; Λ, ~τ = 0) recursively in the expansion with respect to Λ, starting from the perturbation part. Here we remind again that our full partition function, and hence τij and ΘE(~ξτ ), etc, do not have the rigorous meaning. (See Remark 3.7.) The rigorous form of the contact term equation is given by rewriting it as an equation for the instanton part. This was given in [18, (7.5)], for the homological version of Nekrasov partition function, but we do not give here since it is not enlightening. Similarly as the limit of (3.13), we obtain (3.21) 0 = − 1 r + l 2r(cid:18)d − 2π√−1Xi − 1 2 (cid:19) ∂ ∂ log ΘE ∂ log Λ(cid:18) ∂ ∂~τ(cid:19)~n (cid:12)(cid:12)(cid:12)(cid:12)~ξ=− 1 ∂ξi F0(~a; Λ, ~τ = 0) 4π√−1r (d− r+l 2 ) ∂2F0 ∂ log Λ∂~a ∂ ∂~τ(cid:19)~n ∂ai(cid:18) ∂ F0(~a; Λ, ~τ = 0) when ~n is nonzero, either positive or negative and satisfies (3.14) and (3.15)(1)(a),(b). We call this as the contact term equation for (∂/∂~τ )~nF0. This derivation of the contact term equation from the blow-up equation can be done in the same way as in [19, §5.3]. the expansion with respect to Λ starting from the constant term: Since ∂ log ΘE/∂ξi is divisible by Λ, this equation determines (∂/∂~τ )~nF0 recursively in F0(~a; Λ = 0, ~τ = 0) = (cid:18) ∂ ∂~τ(cid:19)~nXp,α if np = 1 and nq = 0 for q 6= p, otherwise. τpepaα(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)~τ ,Λ=0 (3.22) In particular, we have ∂~τ(cid:19)~n (cid:18) ∂ = (Pα epaαΛ=0 ∂~τ(cid:19)~n (cid:18) ∂ 0 F0(~a; Λ, ~τ = 0) = 0 unless np = 1 and nq = 0 for q 6= p. The remaining ∂F0/∂τp will be determined in the next subsection. 30 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Since higher derivatives vanish, we have (cid:0)(eε1/2 − e−ε1/2)(eε2 − e−ε2/2)(cid:1)Pp np(cid:0) ∂ ∂~τ(cid:1)~n Zl(ε1, ε2, ~a; Λ, ~τ = 0) By (3.2) this is equal to Zl(ε1, ε2, ~a; Λ, ~τ = 0) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ε1=ε2=0 =Yp (cid:18) ∂F0 (~a; Λ, ~τ = 0)(cid:19)np ∂τp (3.23) Z inst l 1 (ε1, ε2, ~a; Λ, ~τ = 0)Xc ch Op ×ZM (c) (Λ2re−(r+l)(ε1+ε2)/2)(∆(c),[P2]) (ψp(E)/[0])⊗np! td M(c) exp(lc1(V(E)))(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)ε1=ε2=0 where •/[0] is the slant product q2∗(• ⊗ q∗1(C0)) = (eε1/2 − e−ε1/2)(eε2/2 − e−ε2/2) • /[C2] with the skyscraper sheaf C0 at the origin, which is nothing but the restriction to the origin. 3.4. Seiberg-Witten prepotential. We give a quick review of the Seiberg-Witten pre- potential for the theory with 5-dimensional Chern-Simons term in this subsection. See [8, App. A] for more detail. (We change the notation slightly: m in [loc. cit.] is our l. ai is our aα. β is set to be 1.) We consider a family of hyperelliptic curves parametrized by ~U = (U1, . . . , Ur−1): C ~U ,l : Y 2 = P (X)2 − 4(−X)r+lΛ2r P (X) = X r + U1X r−1 + U2X r−2 + · · · + Ur−1X + (−1)r for −r < l < r, l ∈ Z. Note that we set β = 1 from [loc. cit.] for brevity. We call them Seiberg-Witten curves. We define the Seiberg-Witten differential by dS = 1 2π√−1 log X 2XP ′(X) − (r + l)P (X) 2XY dX. We choose zα (α = 1, . . . , r) so that Xα = e−√−1zα are zeroes of P (X) = 0. Then we take the cycles Aα, Bα (α = 2, . . . , r) in a way explained in [loc. cit.]. We define aα, aD α by α =ZBα in aα. Here we use aα = −√−1zα + O(Λ) [loc. cit., (A.1)]. Then one can show that there exists a locally defined function F0 = F0(~a; Λ) such that We then invert the role of aα and Up, so we consider aα as variables and Up are functions aα =ZAα dS. dS, aD . , aD α = − 1 2π√−1 ∂F0 ∂aα . This defines F0 up to a function (in Λ) independent of ~a. This ambiguity is fixed by specifying ∂F0/∂ log Λ and the perturbation part of F0. See [loc. cit.] for detail. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 31 It was proved in [loc. cit., (A.27), (A.33)] that F0 satisfies the contact term equations (3.20) where the period matrix (τ ) is defined by the same formula as (3.19) by replacing F0 by F0. Moreover, it was also proved that F0 has the same perturbation part as F0 [loc. cit., Prop. A.6]. Therefore the recursive structure of (3.20) implies that F0 = F0, i.e., the leading part of the Nekrasov partition function is equal to the Seiberg-Witten prepotential. In [loc. cit., (A.25)], it was proved (3.24) 0 = 1 2r ∂Up ∂ log Λ + 1 2π√−1Xi ∂ log ΘE ∂ξi (cid:12)(cid:12)(cid:12)(cid:12)~ξ=− 1 4π√−1r ∂Up ∂ai ∂2F0 ∂ log Λ∂~a under the assumption r + l is even. This is the same as the equation (3.21) with d − (r + l)/2 = ±1. When r + l is odd, we assume that Up satisfies (3.21) with d = (r + l + 1)/2 for a moment, and gives a proof later. The initial condition for Up is Up = (−1)pep(X1, . . . , Xr) = (−1)pep(ea1, . . . , ear ) at Λ = 0, where ep is the pth elementary symmetric polynomial. Noticing that (3.21) holds for polynomials in ∂F0/∂τp, we see that (−1)pUp and ∂F0/∂τp are related exactly in the same way as an elementary symmetric polynomial and a power sum by (3.22). Theorem 3.25. ∂F0 ∂τp (~a; Λ, ~τ = 0) = X p 1 + · · · + X p r holds for −(r + l)/2 ≤ p ≤ (r − l)/2. Note that all Up's are written by polynomials in ∂F0 ∂τp (~a; Λ, ~τ = 0) in the above range, as X1 · · · Xr = 1. Thanks to (3.23) the polynomials can be replaced by those in Adams operators. We thus have Up = Z inst l (−1)p (ε1, ε2, ~a; Λ, ~τ = 0)Xc (Λ2re−(r+l)(ε1+ε2)/2)(∆(c),[P2]) ×ZM (c) ch (Vp(E/[0])) td M(c) exp(lc1(V(E)))(cid:12)(cid:12)(cid:12)(cid:12)ε1=ε2=0 for 0 < p ≤ (r − l)/2. For (r − l)/2 ≤ p < r, the equation holds if we replaceVp(E/[0]) byVr−p(E/[0])∨. This equation gives a moduli theoretic description of the coefficients Up It remains to show that Up satisfies (3.21) with d = (r + l + 1)/2 in the case r + l is odd. Let us give a sketch of the argument. We use the notation in [loc. cit.], e.g., E(X1, X2) is the prime form, ω∞+−0− is the meromorphic differential with the vanishing A-periods having poles 0− and ∞+ of residue −1, +1 respectively, etc. in the Seiberg-Witten curves. 32 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA By [4, Cor. 2.11] we have Θ2 0−+∞+ E( 1 E( 1 Θ2 2R 2X 2R ∞+ 0− ~ω) ~ω) E(0−,∞+) E(X, 0−)E(X,∞+) = ω∞+−0− By [loc. cit., (A.18)], we have (P (X) − Y )dX 2XY 1 2r − Therefore it is enough to show (3.26) (P (X) − Y )dX 2XY = (X) + 2 × 1 2π√−1Xi ∂ log ΘE ∂ξi (cid:12)(cid:12)(cid:12)(cid:12)~ξ= 1 2 R ∞+ 0− ωi(X). ~ω ∂Up ∂ log Λ r−1Xp=1 2R 2X 2R ∞+ E( 1 E( 1 Θ2 0− 0−+∞+ Θ2 X r−p−1dX Y = ω∞+−0− (X). ~ω) ~ω) E(0−,∞+) E(X, 0−)E(X,∞+) . W 7→ X = W 2. It is given by (3.27) As in [loc. cit., §A.7] we take the branched double cover p : bC ~U,l → C ~U ,l given by = (cid:0)P (W 2) − 2(√−1W )r+lΛr(cid:1)(cid:0)P (W 2) + 2(√−1W )r+lΛr(cid:1) . Y 2 = P (W 2)2 − 4(−W 2)r+lΛ2r We consider the Szego kernel for bC ~U,l given by where bΘ bE is the Riemann theta function for the curve bC ~U ,l. As in [loc. cit., the first two Ψ bE(W1, W2) = W1 , displayed formulas in §A.6] we have Ψ bE(W,∞+)2 = − ~ω) bΘ bE(R W2 bΘ bE(0)E(W1, W2) dW d(cid:18) 1 From the defining equation (3.27) of the double cover, this has zero of order 2(r + l) at 0+ and of order 2(r − l − 1) at ∞−. Therefore 2Y Y − P (W 2) W2(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)W2=∞+ divbΘ bE(W − ∞+) = (r + l) · 0+ + (r − l − 1) · ∞−. . (See [4, pp.16, 17] for basic properties of E(W1, W2).) (3.28) If we identify the half-integer characteristic bE with a vector in C2r−1 so that bΘ bE(ξ) = bΘ(ξ − bE), we have where b∆ is the Riemann's divisor class ([4, Th. 1.1]). bE = p∗E − [0, c∗, 0], bE = (r + l) · 0+ + (r − l − 1) · ∞− − ∞+ −b∆, By [loc. cit., Lem. A.32] we have (3.29) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 33 where p∗ : J0(C ~U ,l) → J0(bC ~U ,l). On the other hand, we have (3.30) by [4, Prop. 5.3]. b∆ − p∗∆ = 0− + ∞− + p∗(cid:18) 1 2Z ∞+ 0− ~ω(cid:19) + [0, c∗, 0] From (3.28,3.29,3.30) we get r + l − 1 E = · 0+ + r − l − 1 2 · ∞− − 0− − 2 ~ω − ∆, 1 0− 2Z ∞+ 2R 2X where we have used 0+ − ∞− = ∞+ − 0−. Therefore ΘE( 1 (r + l − 1)/2 at 0+, and of order (r − l − 1)/2 at ∞− again by [4, Th. 1.1]. On the other hand, the left hand side of (3.26) has zero of order r + l − 1 at 0+, and of order r − l − 1 at ∞−. Both sides of (3.26) have poles at 0−, ∞+ with residues −1, 1 respectively. (See [4, Cor. 2.11].) Therefore we have (3.26). Although it is not necessary, let us also sketch how to prove (3.21) for more general d. We first assume r + l is even and generalize (3.24) as ~ω) has zero of order 0−+∞+ (3.31) 0 = d 2r ∂Up ∂ log Λ + 1 2π√−1Xi ∂ log ΘE ∂ξi (cid:12)(cid:12)(cid:12)(cid:12)~ξ=− d 4π√−1r ∂Up ∂ai ∂2F0 ∂ log Λ∂~a for d ≤ (r − l)/2. This equation is nothing but (3.21) with d − (r + l)/2 replaced by d. In terms of the d in (3.21), the condition is l ≤ d ≤ r. This is exactly one under which we have proved (3.21) for all p from the vanishing theorem. To show (3.31) we use [loc. cit., the first displayed formula in §A.6.2], which is [4, Cor. 2.19 (43)]. We replace d by d + 1 and take x0 = X, y0 = X′, x1 = · · · = xd = 0−, y1 = · · · = yd = ∞+. We then get d0−+X ~ω(cid:17) ΘE(cid:16)R d∞++X′ ΘE(0)E(X, X′) (cid:18) E(X, 0−)E(∞+, X′) E(X,∞+)E(0−, X′)(cid:19)d E(0−,∞+)pdX1(cid:12)(cid:12)(cid:12)X1=0 √dX2 X2 ΨE(0−, X′)ΨE(X,∞+) 1 − (X/X′)d 1 − (X/X′) (cid:12)(cid:12)(cid:12)(cid:12)X2=∞!−d2 ΨE(0−,∞+) = ΨE(X, X′) − . This is proved exactly as in [loc. cit., §A.6.2], so the detail is omitted. We multiply both sides E(X, X′), differentiate with respect to X′ and set X′ = X. We get (3.32) ∂ log ΘE ∂ξα (dZ ∞+ 0− rXα=2 ~ω)ωα(X) + d × ω∞+−0− (X) = d ΨE(X, 0−)ΨE(X,∞+) ΨE(0−,∞+) as in [15, IIIb §3, p.226]. When d = 1, this is nothing but [loc. cit., (A.24)]. From the argument in [loc. cit., §A.6.1] we get (3.31). We next consider the case r+l is odd. We take the branched double cover p : bC ~U ,l → C ~U,l given by W 7→ X = W 2 as before. Then r, l become 2r, 2l for bC ~U,l, and hence we have 34 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA (3.32) for bC ~U ,l, with d replaced by 2d: rXα=2 rXα=2 ~ω)bωα(W ) + 0− (2dZ ∞+ ∂ logbΘ bE ∂bξα ∂ logbΘ bE ∂bξ′∗ ∂ logbΘ bE ∂bξ′α (2dZ ∞+ ~ω)bω∗(W ) + 2d ×bω∞+−0− 0− + (W ) = 2dbΨ bE(W, 0−)bΨ bE(W,∞+) coordinates bξα, bξ∗, bξ′α on J0(bC ~U ,l). This holds if d ≤ (r − l)/2 as above. The right hand where we take cycles Aα, Bα, A∗, B∗, A′α, B′α as in [loc. cit., §A.7] and the corresponding side is bΨ bE(0−,∞+) , ~ω)bω′α(W ) (2dZ ∞+ 0− (P (W 2) − Y )dW 2d = dp∗(cid:18) (P (X) − Y )dX 2XY (cid:19) as [loc. cit., the second displayed equation in §A.6.1]. We also have 2W Y bω∞+−0− (W ) = p∗ω∞+−0− (X) 1 2 by definition. and the second displayed formula in p.1105]. We then get On the other hand, we rewrite the theta function bΘ bE by ΘE by using [loc. cit., (A.29) rXα=2(cid:26) ∂ log ΘE ~ω)(cid:27) ωα(X) )Z ∞+ )Z ∞+ ((d − ∂ log ΘE ((d + ~ω) + ∂ξα ∂ξα 1 2 1 2 1 2 0− 0− + d × ω∞+−0− (X) = d (P (X) − Y )dX . 2XY From this we get 0 = d 2r ∂Up ∂ log Λ + 1 2π√−1Xi 1 2 ∂ log ΘE ∂ξi (cid:12)(cid:12)(cid:12)(cid:12)~ξ=− d+1/2 4π√−1r + ∂ log ΘE ∂ξi ∂2F0 ∂ log Λ∂~a ∂ log Λ∂~a! ∂Up ∂ai ∂2F0 (cid:12)(cid:12)(cid:12)(cid:12)~ξ=− d−1/2 4π√−1r as in [loc. cit., §A.6.1]. This is nothing but the sum of (3.21) for d replaced by d + (r + l + 1)/2 and d + (r + l − 1)/2. Since we have already proved (3.21) for (r + l − 1)/2, we have (3.21) for l ≤ d ≤ r. 4. Quiver description In this section we review the result of [21], rephrase that of [22] in the quiver description, and add a few things on Ext-groups. 4.1. Moduli spaces of m-stable sheaves. We take vector spaces V0, V1, W with r = dim W, (c1, [C]) = dim V0 − dim V1, We consider following datum X = (B1, B2, d, i, j) (ch2, [bP2]) = − 1 2 (dim V0 + dim V1). PERVERSE COHERENT SHEAVES ON BLOW-UP. III 35 • B1, B2 ∈ Hom(V1, V0), d ∈ Hom(V0, V1), i ∈ Hom(W, V0), j ∈ Hom(V1, W ), B1,B2 V0 / V1 ~}}}}}}}} j d `AAAAAAAA i W • µ(B1, B2, d, i, j) = B1dB2 − B2dB1 + ij = 0. Let Q := µ−1(0) be the subscheme of the vector space Hom(V1, V0)⊕2 ⊕ Hom(V0, V1) ⊕ Hom(W, V0) ⊕ Hom(V1, W ) defined by the equation µ = 0. It is acted by G := GL(V0) × GL(V1) g · (B1, B2, d, i, j) = (g0B1g−1 1 , g0B2g−1 1 , g1dg−1 0 , g0i, jg−1 1 ). Let ζ = (ζ0, ζ1) ∈ Q2. Definition 4.1. We say X = (B1, B2, d, i, j) is ζ-semistable if (1) for subspaces S0 ⊂ V0, S1 ⊂ V1 such that Bα(S1) ⊂ S0 (α = 1, 2), d(S0) ⊂ S1, (2) for subspaces T0 ⊂ V0, T1 ⊂ V1 such that Bα(T1) ⊂ T0 (α = 1, 2), d(T0) ⊂ T1, Ker j ⊃ S1, we have ζ0 dim S0 + ζ1 dim S1 ≤ 0. Im i ⊂ T0, we have ζ0 codim T0 + ζ1 codim T1 ≥ 0. We say X is ζ-stable if the inequalities are strict unless (S0, S1) = (0, 0) and (T0, T1) = (V0, V1) respectively. We say X1, X2 are S-equivalent when the closures of orbits intersect in the ζ-semistable locus of Q. By a standard argument, we can see that these come from quotients of Q by G in the geometric invariant theory. We only explain the result, see [10] for detail. Let χ : G → C∗ be the character given by χ(g) = det g−ζ0 , where we assume (ζ0, ζ1) ∈ Z2 by multiplying a positive integer if necessary. We have a lift of G action on the trivial line bundle Q × C given by (B1, B2, d, i, j, z) = (g · (B1, B2, d, i, j), χ(g)z). Let Lζ denote the corresponding G-equivariant line bundle. Then we can consider the GIT quotient det g−ζ1 1 0 cMζ = Proj Mn≥0 A(Q)χ,n! , quotient of ζ-stable locus modulo the action of G. where A(Q)χ,n is the relative invariants in the coordinate ring A(Q) of Q: {f ∈ A(Q) . Then cMζ is f (g · X) = χ(g)nf (X)}, which is the space of invariant sections of L⊗n the quotient of ζ-semistable locus modulo S-equivalence relation. It contains cM s We have a natural projective morphism bπ : cMζ → µ−1(0)//G, where µ−1(0)//G is the affine geometric invariant theory quotient of µ−1(0) by G, i.e., cM0. By [21, §1.3] µ−1(0)//G is isomorphic to M0, the Uhlenbeck partial compactification on P2. Now the main result of [21] says ζ of the ζ / o o o o ~ ` 36 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA We use this result as the definition of the moduli scheme of m-stable framed sheaves in Theorem 4.2. Let m ∈ Z≥0. Suppose that ζ0 < 0, 0 > mζ0 + (m + 1)ζ1 ≫ −1. Then ζ and it is bijective to the set of isomorphism classes of m-stable framed we have cMζ = cM s sheaves on bP2. this paper. Therefore cMζ will be denoted by cM m (or cM m(c) when we want to write the and (b) the action of G is free on the ζ-stable locus. Therefore cMζ is a smooth fine moduli Chern character of sheaves) hereafter. It was proved in [21, 2.4] that (a) dµ is surjective The construction is given as follows: We consider the complex scheme. V0 ⊗ O(C − ℓ∞) V1 ⊗ O(−ℓ∞) ⊕ α−−−→ C2 ⊗ V0 ⊗ O ⊕ C2 ⊗ V1 ⊗ O ⊕W ⊗ O β −−−→ V0 ⊗ O(ℓ∞) ⊕ V1 ⊗ O(−C + ℓ∞) , (4.3) with α = z0B1 z0B2 z w 0 z1 − z0dB1 0 z2 − z0dB2 0 z0j , β =(cid:20) z2 −z1 B2z0 −B1z0 dw −dz −z w iz0 0(cid:21) .   The equation µ(B1, B2, d, i, j) = B1dB2 − B2dB1 + ij = 0 is equivalent to βα = 0. The stability condition ensures the injectivity of α and the surjectivity of β. Then the sheaf corresponding to (B1, B2, d, i, j) is defined by E = Ker β/ Im α. By the definition it is endowed with the framing Eℓ∞ → W ⊗ Oℓ∞ . The ζ-stability is identified with the m- stability. The inverse construction is given by V0 := H 1(E(−ℓ∞)), V1 := H 1(E(C − ℓ∞)), and B1, B2, d, i, j are homomorphisms between cohomology groups induced from certain natural sections. From this construction V0, V1 naturally define vector bundles over cM m, which are denoted by V0, V1 respectively. The above α, β in (4.3) are interpreted as homomorphisms between vector bundles and the universal family E is given by Ker β/ Im α. When we prove that the ζ-stability corresponds to the m-stability in Definition 1.1, it is crucial to observe that the sheaf OC(−m − 1) corresponds to the datum V0 = Cm, V1 = Cm+1, W = 0, d = 0 and B1 = [ 1m 0 ] , B2 = [ 0 1m ] , where 1m is the identity matrix of size m. We denote this datum by Cm as above. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 37 4.2. Tangent complex. From the construction the tangent space is the middle coho- mology group of the complex Hom(V0, V1) Hom(V0, V0) ⊕Hom(V1, V1) ι−−−→ ⊕ C2 ⊗ Hom(V1, V0) ⊕Hom(W, V0) ⊕Hom(V1, W ) dµ−−−→ Hom(V1, V0), (4.4) with ι(cid:20)ξ0 ξ1(cid:21) = dξ0 − ξ1d B1ξ1 − ξ0B1 B2ξ1 − ξ0B2 ξ0i −jξ1  , (dµ) =   ed eB1 eB2 ei ej B1deB2 + B1edB2 + eB1dB2 − B2deB1 − B2edB1 − eB2dB1 +eij + iej, where dµ is the differential of µ, and ι is the differential of the group action. Remark that dµ is surjective and ι is injective by the above remark if X is ζ-stable. 4.3. A modified quiver. We fix the vector space W with dim W = r 6= 0. We define a new quiver with three vertexes 0, 1, ∞. We write two arrows from 1 to 0 corresponding to the data B1, B2, and one arrow from 0 to 1 corresponding to the data d. Instead of writing one arrow from ∞ to 0, we write r-arrows. Similarly we write r-arrows from 1 to ∞. And instead of putting W at ∞, we replace it the one dimensional space C on ∞. We denote it by V∞. It means that instead of considering the homomorphism i from W to V0, we take r-homomorphisms i1,i2,. . . , ir from V∞ to V0 by taking a base of W . (See Figure 2.) 0 / 1 . . . . . . ∞ Figure 2. modified quiver We consider the full subcategory of the abelian category of representations of the new quiver with the relation, consisting of representations such that dim V∞ = 0 or 1. An object can be considered as a representation of the original quiver with dim W = 0, or dim W = r, according to dim V∞ = 0 or 1. Note that we do not allow a representation of the original quiver with dim W 6= r, 0. It is also suitable to modify the stability condition for representations for the new quiver. Let (ζ0, ζ1, ζ∞) ∈ Q3. For a representation X of the modified quiver, let us denote the / o o o o u u o o V V i i 38 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA underlying vector spaces by X0, X1, X∞. We define the rank , degree and slope by (4.5) rank X := dim X0 + dim X1 + dim X∞, ζ · dim X := ζ0 dim X0 + ζ1 dim X1 + ζ∞ dim X∞, θ(X) := , ζ · dim X rank X where θ(X) is defined only when X 6= 0. We only consider the case dim X∞ = 0 or 1 as before. We say X is θ-semistable if we have θ(S) ≤ θ(X) for any subrepresentation 0 6= S of X. We say X is θ-stable if the inequality is strict unless S = X. If θ(X) = 0, θ-(semi)stability is equivalent to ζ-(semi)stability. In fact, if a subrepresentation S has S∞ = 0, then θ(S) ≤ 0 is equivalent to ζ0 dim S0 + ζ1 dim S1 ≤ 0. If a subrepresentation T has T∞ = C, then θ(T ) ≤ 0 is equivalent to ζ0 codim T0 + ζ1 codim T1 ≥ 0. The θ-stability is unchanged even if we add c(1, 1, 1) (c ∈ R) to (ζ0, ζ1, ζ∞). Therefore once we fix dim X0, dim X1, dim X∞, we can always achieve the condition θ(X) = 0 without the changing stable objects. 4.4. Wall-crossing. Let us fix a wall {ζ mζ0 + (m + 1)ζ1 = 0, ζ0 < 0} and a parameter ζ 0 = (ζ 0 1 ) from the wall. We take ζ +, ζ− sufficiently close to ζ 0 with 0 , ζ 0 1 ≫ mζ + 0 + (m + 1)ζ + 1 > 0, −1 ≪ mζ−0 + (m + 1)ζ−1 < 0. (See Figure 3.) Then ζ− is nothing but the parameter ζ appeared in Theorem 4.2 corre- sponding to the m-stability. On the other hand, we also know that ζ + corresponds to the (m + 1)-stability by the determination of the chamber structure in [21, §2]. ζ1 ζ0 + ζ1 = 0 mζ0 + (m + 1)ζ1 = 0 ζ + ζ 0 ζ− ζ0 Figure 3. wall-crossing semistability. sheaves set-theoretically, where Let us define the scheme cM m,m+1 as the GIT quotient cMζ 0 with respect to the ζ 0- From [21] it follows that cM m,m+1 is the S-equivalence classes of (m, m + 1)-semistable Definition 4.6. (a) A framed sheaf (E, Φ) onbP2 is called (m, m + 1)-semistable if PERVERSE COHERENT SHEAVES ON BLOW-UP. III 39 (1) Hom(E,OC(−m − 2)) = 0, (2) Hom(OC(−m), E) = 0, and (3) E is torsion free outside C. (b) A framed sheaf (E, Φ) onbP2 is called (m, m + 1)-stable if it is either OC(−m − 1) (with the trivial framing) or both m-stable and (m+1)-stable, i.e., we have Hom(E,OC(−m)) = 0 and Hom(OC(−m), E) = 0 instead of (1),(2). (c) A (m, m + 1)-semistable framed sheaf (E, Φ) has a filtration (0 = E0 ⊂ E1 ⊂ · · · ⊂ EN = E) such that Ei/Ei−1 is (m, m + 1)-stable with the induced framing from Φ. We say (m, m + 1)-semistable framed sheaves (E, Φ) and (E′, Φ′) are S-equivalent if there exists an isomorphism from Li Ei/Ei−1 to Lj E′j/E′j−1 respecting the framing in the (m, m + 1)-stable factors. The main result [21, Th. 1.5] was stated for m-stable framed sheaves, but it can be generalized to the case of (m, m + 1)-stable framed sheaves. It follows from Proposi- tions 4.1,4.2 there that X is ζ 0-semistable if and only if it satisfies the condition (S2) in [loc. cit., Th. 1.1] and the condition corresponding to (a1-2) above. Then the remaining arguments are the same. Since ζ±-stability implies ζ 0-semistability, we have natural morphisms ξm : cM m → cM m,m+1, ξ+ m : cM m+1 → cM m,m+1. Thus cM m, cM m+1 and cM m,m+1 form the diagram (∗). This definition of cM m,m+1 is different from what are given in [22] for ordinary moduli spaces without framing, but they are the same at least set-theoretically thanks to the construction in [loc. cit., §§3.6,3.7]. It is also possible to show directly that ξm, ξ+ m have structures of stratified Grassmann bundles described there. From the definition it is clear that ξm, ξ+ m are compatible with projective morphisms cM m → M0, cM m,m+1 → M0, cM m+1 → M0, (all denoted bybπ before). The change of the moduli spaces under the wall-crossing is described as follows: Proposition 4.7 ([22, 3.15]). (1) If (E−, Φ) is m-stable, we have an exact sequence with V = Hom(Cm, E−) such that 0 → V ⊗ Cm → E− → E′ → 0 (a) E′ is (m, m + 1)-stable, and (b) the induced homomorphism V → Ext1(E′, Cm) is injective. Conversely if (E′, Φ) is (m, m + 1)-stable and a subspace V ⊂ Ext1(E′, Cm) is given, (E−, Φ), defined by the above exact sequence, is m-stable. (2) If (E+, Φ) is (m + 1)-stable, we have an exact sequence 0 → E′ → E+ → U∨ ⊗ Cm → 0 with U = Hom(E+, Cm) such that (a) E′ is (m, m + 1)-stable, and (b) the induced homomorphism U∨ → Ext1(Cm, E′) is injective. Conversely if (E′, Φ) is (m, m + 1)-stable and a subspace U∨ ⊂ Ext1(Cm, E′) is given, (E+, Φ), defined by the above exact sequence, is (m + 1)-stable. 40 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA 4.5. Computation of Ext-groups. In this subsection, we continue to fix a wall mζ0 + (m + 1)ζ1 = 0, ζ0 < 0. Take X = (B1, B2, d, i, j) defined on V0, V1, W such that µ(B1, B2, d, i, j) = 0. We consider the complex Hom(V0, Cm) ⊕ Hom(V1, Cm+1) σ−−−→ (4.8) with Hom(V0, Cm+1) C2 ⊗ Hom(V1, Cm) τ−−−→ Hom(V1, Cm), ⊕ ⊕ Hom(W, Cm) ξ0i ξ1d  , ξ0B1 − [ 1m 0 ] ξ1 ξ0B2 − [ 0 1m ] ξ1 ξ1(cid:21) = σ(cid:20)ξ0  τ ed eB1 = [ 1m 0 ]edB2 − [ 0 1m ]edB1 +eij + eB1dB2 − eB2dB1. eB2 ei [ z w ] This complex is constructed as follows. We take the dual of (4.3), and replace the part −−−→ V ∗0 ⊗ O(−C + ℓ∞) by V ∗0 ⊗ O(C − ℓ∞). We then take the tensor C2 ⊗ V ∗0 ⊗ O (E, Φ), the cohomology groups of the complex are Ext•(E, Cm). product with Cm and apply H∗(bP2,•). Therefore when X corresponds to a framed sheaf Lemma 4.9. Suppose that X corresponds to a framed sheaf (E, Φ). (1) Hom(E, Cm) ∼= Ker σ, Ext1(E, Cm) ∼= Ker τ / Im σ, and Ext2(E, Cm) ∼= Coker τ . (2) Suppose further that X is (m, m + 1)-semistable. Then τ is surjective. Proof. (1) These are already explained. (2) By the Serre duality we have Ext2(E, Cm) = Hom(Cm−1, E)∨. But the right hand side vanishes if E is (m, m + 1)-semistable. (cid:3) If X is ζ−-stable (i.e., (E, Φ) is m-stable), we also have Ker σ = 0. But this does not hold in general if X is only (m, m + 1)-semistable. Next we consider the complex (4.10) Hom(Cm, V0) ⊕ Hom(Cm+1, V1) σ−−−→ Hom(Cm, V1) C2 ⊗ Hom(Cm+1, V0) τ−−−→ Hom(Cm+1, V0), ⊕ ⊕ Hom(Cm+1, W ) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 41 with dξ0 jξ1  , B1ξ1 − ξ0 [ 1m 0 ] B2ξ1 − ξ0 [ 0 1m ] ξ1(cid:21) = σ(cid:20)ξ0  τ ed eB1 = B1ed [ 1m 0 ] − B2ed [ 0 1m ] + iej + B1deB2 − B2deB1. eB2 ej Lemma 4.11. Suppose that X corresponds to a framed sheaf (E, Φ). (1) Hom(Cm, E) ∼= Ker σ, Ext1(Cm, E) ∼= Ker τ / Im σ, and Ext2(Cm, E) ∼= Coker τ . (2) Suppose further that X is (m, m + 1)-semistable. Then τ is surjective. Proof. The proof of (1) is the same as in Lemma 4.9, so is omitted. (2) We have Ext2(Cm, E) ∼= Hom(E, Cm+1)∨ by the Serre duality. But the right hand side vanishes if E is (m, m + 1)-semistable. (cid:3) If X is ζ +-stable (i.e., (E, Φ) is (m + 1)-stable), we also have Ker σ = 0. But this does not hold in general if X is only (m, m + 1)-semistable. Corollary 4.12. Let Qss(ζ 0) be the open subscheme of Q consisting of ζ 0-semistable, i.e., (m, m + 1)-semistable objects. The differential dµ is surjective on Qss(ζ 0). Hence Qss(ζ 0) is smooth. Proof. Since the surjectivity of dµ is an open condition, it is enough to check the assertion when X is a direct sum of ζ 0-stable objects. As explained in Definition 4.6, we have m , where X 0 is ζ 0-stable with X∞ 6= 0 and p ≥ 0. Then the tangent complex X = X 0⊕ C⊕p (4.4) decomposes into four parts, the tangent complex for X 0, the sum of p-copies of the complex (4.8) for X 0, the sum of p-copies of the complex (4.10) for X 0, and the sum of p2-copies of the tangent complex for Cm. The differential of dµ for X 0 is surjective since X 0 is ζ 0-stable by [21, 2.4]. The sur- jectivity of τ for (4.8) and (4.10) was proved in Lemma 4.9 and Lemma 4.11 respectively. The surjectivity of dµ for Cm follows from either [21, 2.4], Lemma 4.9 or Lemma 4.11 since Cm is (−1/m, 1/(m + 1))-stable by [21, §2.2]. (cid:3) 5. Enhanced master spaces A naive idea to show the weak wall-crossing formula (=Theorem 1.5) is to compare the intersection products on cM m(c) and cM m+1(c) through the diagram (∗). Such an idea works when cM m(c), cM m+1(c) are replaced by moduli spaces of stable rank 2 torsion free sheaves over a surface with pg = 0 with respect to two polarizations separated by a wall which is good [3]. (See also [7].) However the idea does not work since the morphisms ξm, ξ+ m are much more complicated in our current situation, basically because dimensions of vector spaces V , U in Proposi- tion 4.7 can be arbitrary. We use a refinement of the idea, due to Mochizuki [14], which 42 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA was used to study wall-crossing formula for moduli spaces of higher ranks stable sheaves. It consists of two parts: (Ma): Consider a pair of a sheaf and a full flag in the sheaf cohomology group. (Mb): Use the fixed point formula for the C∗-equivariant cohomology on Thaddeus' master space. In this section we describe the part (Ma). The results are straightforward modifications of those in [14, §4], possibly except one in §5.4. We remark that Yamada also used moduli spaces of pairs of a sheaf and a full flag in the sheaf cohomology group to study wall-crossing of moduli spaces of higher rank sheaves [29]. We continue to fix m and choose parameters ζ 0, ζ± as in the previous section. We ∞ so that the normalization condition fix a dimension vector v = (v0, v1, 1) and define ζ 0 ζ 0 0 v0 + ζ 0 = 0 is satisfied. 1 v1 + ζ 0 ∞ 5.1. Framed sheaves with flags in cohomology groups. Let ((E, Φ), F •) be a pair of a framed sheaf and a full flag F • = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1(E)) (N = dim V1(E)). Let ℓ be an integer between 0 and N. Definition 5.1. An object ((E, Φ), F •) is called (m, ℓ)-stable if the following three con- ditions are satisfied (1) (E, Φ) is (m, m + 1)-semistable. (2) For a subsheaf S ⊂ E isomorphic to C⊕p (3) For a subsheaf S ⊂ E such that the quotient E/S is isomorphic to C⊕p m with p ∈ Z>0, we have V1(S) ∩ F ℓ = 0. m with p ∈ Z>0, we have F ℓ 6⊂ V1(S). This notion makes a bridge between the m-stability and the (m + 1)-stability by the fol- lowing observation: If ℓ = 0 (resp. = N), then (m, ℓ)-stability of ((E, Φ), F •) is equivalent to m (resp. (m + 1))-stability of (E, Φ). (See Proposition 4.7.) M0(p∗(c)). The proof will be given in the next subsection. Proposition 5.2 ([14, 4.2.5]). We have a smooth fine moduli scheme fM m,ℓ(c) of (m, ℓ)- stable objects ((E, Φ), F •) with ch(E) = c. There is a projective morphism fM m,ℓ(c) → Let E be the universal sheaf overbP2 ×fM m,ℓ(c) and q1, q2 be the projection to the first and second factors of bP2 ×fM m,ℓ(c) respectively as before. As well as the vector bundle F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1) of flags of vector bundles over fM m,ℓ(c). For ℓ = 0 or N, the preceding remark implies that fM m,0(c), fM m,N (c) are the full flag bundles Flag(V1, N) associated with vector bundles V1 over cM m(c), cM m+1(c) respectively. Here N = {1, . . . , N} and the notation Flag(V1, N ) means that the flag is indexed by the set N . This notation will become useful when we consider the fixed points in the enhanced master space. V1 ≡ V1(E) := R1q2∗(E ⊗ q∗1O(C − ℓ∞)), we also have the universal family F• = (0 = PERVERSE COHERENT SHEAVES ON BLOW-UP. III 43 5.2. Proof of Proposition 5.2. Let us rephrase the (m, ℓ)-stability in the quiver de- scription. We consider a pair (X, F •) = ((B1, B2, d, i, j), F •) of X ∈ µ−1(0) and a flag F • = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1) of V1 with dim(F i/F i−1) = 0 or 1. In Definition 5.1 we have assumed that F • is a full flag, i.e., dim(F i/F i−1) = 1, but we slightly generalize it for a notational simplicity. In terms of a quiver representation, Definition 5.1 is expressed as follows: Definition 5.3. For 0 ≤ ℓ ≤ N, we say (X, F •) is (m, ℓ)-stable if the following conditions are satisfied: (1) X is ζ 0-semistable, (2) for a nonzero submodule 0 6= S ⊂ X with ζ 0 · dim S/ rank S = 0 and S∞ = 0, we (3) for a proper submodule S ( X with ζ 0 · dim S/ rank S = 0 and S∞ = C, we have have F ℓ ∩ S1 = 0, and F ℓ 6⊂ S1. The equivalence of (2),(3) and 5.1(2),(3) is an immediate consequence of [21, Th. 2.13, Prop. 5.3]. If ℓ = 0 (resp. = N), then (m, ℓ)-stability is equivalent to the ζ− (resp. ζ +)-stability of X. (See Proposition 4.7.) The idea of the proof of Proposition 5.2 is to relate the above condition to an usual stability condition for a linearization on the product of µ−1(0) and the flag variety with It is the tensor product of linearizations on µ−1(0) respect to the group action of G. and the flag variety. For µ−1(0), we take one as in §4.1. On the flag variety, we take >0, where F i is the ith universal bundle. (See [14, §4.2].) Then we have a natural projective morphism to µ−1(0)//G = M0(p∗(c)) as before. The corresponding stability condition is expressed as before, but an extra term for flags N(detF i)⊗(−ni) for n = (n1, . . . , nN ) ∈ QN is added to θ (cf. [16, Ch. 4, §4]): For a nonzero graded submodule S ⊂ X we define ζ · dim S +Pi ni dim(S ∩ F i) rank S . µζ,n(S) := We say (X, F •) is (ζ, n)-(semi )stable if µζ,n(S)(≤)µζ,n(X) holds for any nonzero proper submodule 0 6= S ( X. Here (≤) means that ≤ for the semistable case, and < for the stable case. We say (X, F •) is strictly (ζ, n)-semistable if it is (ζ, n)-semistable and not (ζ, n)-stable. A standard argument shows the following: Lemma 5.4. If (X, F •), (Y, G•) are (ζ, n)-stable and have the same µζ,n-value, a nonzero homomorphism ξ : X → Y with ξ(F i) ⊂ Gi must be an isomorphism. We take N = dim V1 so that F • is a full flag of V1. Consider the following conditions when ℓ 6= 0: (5.5a) ζ0v0 + ζ1v1 + ζ∞ = 0, 44 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA (5.5b) nk > rank X ini > 0 for k = 1, . . . , N − 1, NXi=k+1 0 < max S⊂V ζ 0·dim S6=0(cid:12)(cid:12)(cid:12)(cid:12) ℓXi=1 NXi=l+1 rank X (ζ 0 − ζ) · dim S rank S (cid:12)(cid:12)(cid:12)(cid:12) + ini < min ℓXi=1 rank X 2m + 1 ini − (5.5c) (5.5d) (5.5e) NXi=1 ini < min S⊂V ζ 0·dim S6=0(cid:12)(cid:12)(cid:12)(cid:12) ζ 0 · dim S rank S (cid:12)(cid:12)(cid:12)(cid:12) , (mζ0 + (m + 1)ζ1) < nℓ, ini − rank X 2m + 1 (mζ0 + (m + 1)ζ1), ini + rank X 2m + 1 (mζ0 + (m + 1)ζ1)!, nℓ − ℓXi=1 where S ⊂ V runs over the set of all graded subspaces (not necessarily submodules, as X is not fixed yet) with ζ 0 · dim S 6= 0 in (c). For example, we can take ζ, n in the following way: First note that the right hand side of (c) is a fixed constant independent of ζ, n. Therefore it holds if ni and ζ 0 − ζ are sufficiently small. Next note that (d) says that i=1 ini. Therefore we can first take ζ and n1, . . . , nℓ so that (b) (for k = 1, . . . , ℓ),(d) and (c) hold if we set nℓ+1 = · · · = nN = 0. Then we take sufficiently small ni (i = ℓ + 1, . . . , N) so that (e) holds. We can choose them very small so that (b) and (c) are not violated. ζ 0− ζ, up to a constant multiplication, is between (Pl i=1 ini)− nl andPl Lemma 5.6 ([14, 4.2.4]). Assume (5.5). Then the followings hold: (1) The (m, ℓ)-stability is equivalent to the (ζ, n)-stability. (2) The (ζ, n)-semistability automatically implies the (ζ, n)-stability. Lemma 5.4 also implies that (ζ, n)-stable objects have the trivial stabilizer. Therefore we have Proposition 5.2 from this lemma. From the construction the universal family F• = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1) is the descent of the universal flag over Flag(V1, N). Using descents of V0, V1 (which will be denoted by V0, V1) in the complex (4.3), we also get the universal family E over Proof of Lemma 5.6. Suppose that S ⊂ X is a nonzero submodule with ζ 0 · dim S = 0, S∞ = 0. Then dim S = p(m, m + 1, 0) for some p ∈ Z>0. Then µζ,n(S) ≤ µζ,n(X) means mζ0 + (m + 1)ζ1 2m + 1 i=1 ni dim(S1 ∩ F i) p(2m + 1) +PN ≤ PN i=1 ini rank X . By (5.5d,e) this holds if and only if S1 ∩ F ℓ = 0. Moreover the equality never holds. dim X/S = p(m, m + 1, 0) for some p ∈ Z>0. Then µζ,n(S) ≤ µζ,n(X) means Next suppose that S ⊂ X is a proper submodule with ζ 0 · dim S = 0, S∞ = C. Then mζ0 + (m + 1)ζ1 2m + 1 i=1 ni dim(F i/S1 ∩ F i) p(2m + 1) +PN ≥ PN i=1 ini rank X . bP2 ×cM m(c). PERVERSE COHERENT SHEAVES ON BLOW-UP. III 45 By (5.5d,e) this holds if and only if F i/S1 ∩ F i 6= 0 for some i = 1, . . . , ℓ, i.e., F ℓ 6⊂ S1. Moreover the equality never holds. Now let us start the proof. Suppose that (X, F •) is (ζ, n)-semistable. Then by (5.5c) µζ,n(S) ≤ µζ,n(X) implies ζ 0 · dim S ≤ 0, i.e., X is ζ 0-semistable. Then the above consideration shows that (X, F •) is (m, ℓ)-stable and (ζ, n)-stable. Conversely suppose (X, F •) is (m, ℓ)-stable. We want to show that µζ,n(S) < µζ,n(X) for any nonzero proper submodule S. Thanks to (5.5c), it is enough to check this inequality for S with ζ 0· dim S = 0. We have either S∞ = 0 or S∞ = C, and the above consideration shows that the inequality holds. (cid:3) Remark 5.7. A closer look of the argument gives that it is enough to assume (5.5c) for S ⊂ V satisfying either of (1) ζ 0 · dim S > 0 and µζ,n(S) ≤ µζ,n(X), (2) ζ 0 · dim S < 0 and µζ,n(S) ≥ µζ,n(X). 5.3. Oriented sheaves with flags in cohomology groups. The following variant of objects in the previous subsection will show up during our analysis for the enhanced master space. Definition 5.8. (1) Let (E, F •) be a pair of a sheaf and a full flag F • = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1(E)) (N = dim V1(E)). We say (E, F •) is (m, +)-stable if the following two conditions are satisfied: (a) E ∼= C⊕p (b) For a proper subsheaf S ( E isomorphic to C⊕q F 1 = 0. m for p ∈ Z>0. m with q ∈ Z>0, we have V1(S) ∩ (2) For (m0, m1) ∈ Z2\{(0, 0)}, an orientation of (E, F •) is an isomorphism ρ : det H 1(E)⊗m0⊗ det H 1(E(C))⊗m1 ∼=−→ C. We set D := mm0 + (m + 1)m1. (3) An oriented (m, +)-stable object means an (m, +)-stable object (E, F •) together with an orientation. We will choose (m0, m1) later when we define the enhanced master space. At this stage we only need that we will have D > 0. Since the orientation will be used frequently, we use the notation L(E) := det H 1(E(−ℓ∞))⊗m0 ⊗ det H 1(E(C − ℓ∞))⊗m1 for a sheaf E on If E is a universal family for some moduli stack, we denote the corresponding line bundle by L(E). Note that we deal only with those sheaves given by quiver representations, we have vanishing of H 0 and H 2. bP2. (In the above case, E is supported on C and the twisting by O(−ℓ∞) is unnecessary.) We will show that we have a moduli stack fM m,+(pem) of oriented (m, +)-stable objects F N−1 ⊂ F N = V1) is a flag of vector bundles over fM m,+(pem). In the following proposition we identify fM m,+(pem) with a quotient stack related to the with ch(E) = pem with the universal family (E,F•) where F• = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ Grassmann variety Gr(m + 1, p) of p-dimensional quotients of V1(Cm)∗ = H 1(Cm(C))∗ = Cm+1. Let us fix the notation. Let Q denote the universal quotient bundle over Gr(m + 1, p), detQ its determinant line bundle, and (detQ)⊗D its Dth tensor power. Let πG : ((detQ)⊗D)× → Gr(m + 1, p) be the associated C∗-bundle. Let OGr(m+1,p) → V1(Cm) ⊗ Q be the homo- morphism obtained from the universal homomorphism V1(Cm)∗ → Q. 46 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Proposition 5.9. (1) Forgetting F i for i 6= 1, we identify fM m,+(pem) with the total space of the flag bundle Flag(V1/F 1, N \{1}), where the base is isomorphic to the quotient stack ((detQ)⊗D)×/C∗, where C∗ acts by the fiber-wise multiplication with weight −pD. (2) Let C∗s = Spec C[s, s−1] be a copy of C∗ and let C∗ → C∗s be the homomorphism given by t 7→ t−pD = s. It induces an ´etale and finite morphism h : ((detQ)⊗D)×/C∗ → ((detQ)⊗D)×/C∗s = Gr(m + 1, p) of degree = 1/pD. The vector bundles V1, F 1 and the universal sheaf E are related to objects on Gr(m + 1, p) by V1 ⊗ (F 1)∗ = h∗(V1(Cm) ⊗ Q), (F 1)⊗−pD = h∗((detQ)⊗D), E ⊗ (F 1)∗ = (id×h)∗(Cm ⊠ Q), and the inclusion (F 1 → V1) is h∗(OGr(m+1,p) → V1(Cm) ⊗ Q) ⊗ idF 1. The proof will be given in the next subsection. 5.4. Proof of Proposition 5.9. Since F i with i > 1 does not appear in the stability condition, the moduli stack has a structure of the flag bundle Flag(V1/F 1, N \ {1}) over the moduli stack parametrizing (X, F 1). Here V1 is a vector bundle over the moduli stack and F 1 is its line subbundle, coming from V1 and F 1 respectively. Next we determine the moduli stack parametrizing (X, F 1). Since we already know X ∼= C⊕p m , the remaining parameter is only a choice of F 1, which is a 1-dimensional subspace in V1(X) ∼= V1(Cm) ⊗ Cp. We have an action of the stabilizer GLp(C) of X. The above stability condition means that F 1, viewed as a nonzero homomorphism V1(Cm)∗ → Cp, is surjective. Therefore the moduli stack is (5.10) where the action of GLp(C) on C∗ is given by g·u = (det g)D u with D = mm0 +(m+1)m1. ({ξ ∈ P(Hom(V1(Cm)∗, Cp)) ξ is surjective} × C∗) / GLp(C), We consider (5.10) as (tξ, u) = t idCp ·(ξ, t−pDu), ({ξ ∈ Hom(V1(Cm)∗, Cp) ξ is surjective} × C∗) / GLp(C) × C∗, (5.11) where C∗ ∋ t acts by (ξ, u) 7→ (tξ, u). If we take the quotient by GLp(C) first, we get L× = L \ (0-section), where L = (detQ)⊗D, the Dth tensor power of the determinant line bundle detQ of the universal quotient over the Gr(m + 1, p). Since C∗ ∋ t acts by (5.12) it is the fiber-wise multiplication with weight −pD on the quotient L×. Thus the moduli stack parametrizing (X, F 1) is isomorphic to the quotient stack [L×/C∗]. This action factors through ρ : C∗ → C∗ given by t 7→ s = t−pD, where the latter action is free on L× and the quotient is Gr(m + 1, p). Let us denote the latter C∗ by C∗s. Then the stack L×/C∗s is represented by Gr(m + 1, p), as L× is a principal C∗s-bundle. Since a C∗-bundle induces a C∗s-bundle by taking the quotient by Ker ρ ∼= Z/pD, we have a morphism H : L×/C∗ → L×/C∗s = Gr(m + 1, p) between stacks. It is ´etale and finite of degree 1/pD. Let us identify the pair (F 1 ⊂ V1) of the vector bundle V1 and its line subbundle F 1 In the description (5.10), it is the over the moduli stack in the description [L×/C∗]. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 47 descent of the restriction of (OP(−1) ⊠ OC∗ ⊂ V1(Cm) ⊗ Cp ⊗ OP ⊠ OC∗) with respect to the natural GLp(C)-action, where P = P(Hom(V1(Cm)∗, Cp)). Then in the description (5.11), it becomes the descent of the pair (OV ⊂ V1(Cm) ⊗ Cp ⊗ OV ) , where GLp(C) acts naturally and the C∗-action is twisted by the weight −1 action on the first factor OV . Here V = {ξ ∈ Hom(V1(Cm)∗, Cp) ξ is surjective} × C∗. Finally in the description [L×/C∗], it is the descent of (OL× ⊂ π∗G(V1(Cm) ⊗ Q)) , where the C∗-action is twisted by weight −1 on both OL× and π∗G(V1(Cm) ⊗ Q. Here πG : L× → Gr(m + 1, p) is the projection. The twist on the second factor π∗G(V1(Cm)⊗Q) comes from the term t idCp in (5.12). From the above description of V1, we have V1 ⊗ (F 1)∗ = h∗(V1(Cm) ⊗ Q). On the other hand, (F 1)⊗−pD is the descent of OL× with the C∗-action twisted by weight pD. The action factors through the C∗s-action, and it is twisted by weight −1. Therefore it descends to L on Gr(m + 1, p). 5.5. 2-stability condition. This subsection is devoted to preliminaries for a study of enhanced master spaces. We consider the following condition on n: (5.13) kini 6= 0 for any (k1, . . . , kN ) ∈ ZN \ {0} with ki ≤ 2N 2. NXi=1 Our flag F • of V1 again may have repetition, but assume dim(F i/F i−1) = 0 or 1 as before. Lemma 5.14. Assume that ζ satisfies (m+1)ζ0 +(m+2)ζ1 < 0, (m−1)ζ0 +mζ1 > 0, and n satisfies (5.13). If (X, F •) is strictly (ζ, n)-semistable, then there exists a submodule 0 6= S ( X such that (1) µζ,n(S, S1 ∩ F •) = µζ,n(X, F •), (2) (S, S1 ∩ F •) and (X/S, F •/S1 ∩ F •) are (ζ, n)-stable. Moreover the submodule S is unique except when (X, F •) is the direct sum (S, S1 ∩ F •) ⊕ (X/S, F •/S1 ∩ F •). In this case the another choice of the submodule is X/S. Proof. Take a submodule S violating the (ζ, n)-stability of X. Then we have (1). Moreover (S, S1 ∩ F •) and (X/S, F •/S1 ∩ F •) are (ζ, n)-semistable. We have either S∞ = 0 or (X/S)∞ = 0. Assume either (S, S1 ∩ F •) or (X/S, F •/S1 ∩ F •) is strictly (ζ, n)-semistable. Then we have a filtration 0 = X 0 ( X 1 ( X 2 ( X 3 = X with µζ,n(X a/X a−1, F •a ) = µζ,n(X, F •) for a = 1, 2, 3, where F •a denote the induced filtration on X a/X a−1 from F •. Among X a/X a−1 (a = 1, 2, 3), one of them has C and two of them have 0 at the ∞- component. Assume X 1 has C at the ∞-component for brevity, as the following argument can be applied to the remaining cases. We have dim X 2/X 1 = p2(m, m + 1, 0), dim X 3/X 2 = p3(m, m + 1, 0) for some p2, p3 ∈ Z>0. Then µζ,n(X 2/X 1, F •2 ) = µζ,n(X 3/X 2, F •3 ) implies X ni(cid:18)dim(F i p2 2) dim(F i 3) p3 (cid:19) = 0. − 48 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA By the assumption (5.13) we have p3 dim(F i 2) = p2 dim(F i 3) and hence (5.15) p3 dim(F i 2/F i−1 2 ) = p2 dim(F i 3/F i−1 3 ) for any i. On the other hand, we have 2/F i−1 ) + dim(F i 1/F i−1 dim(F i 2 1 ) + dim(F i 3/F i−1 3 ) = dim(F i/F i−1) = 0 or 1 for any i. Therefore at most one of three terms in the left hand side can be 1 and the other terms are 0. Combined with (5.15) this implies dim(F i ) = 0 for any i. Thus we get a contradiction X 1 = X 2 = X 3. ) = dim(F i 3/F i−1 2/F i−1 2 3 If we have another submodule S′ of the same property, the (ζ, n)-stability implies (cid:3) S ∩ S′ = 0 or S ∩ S′ = S = S′. In the former case we have S′ = X/S. Lemma 5.16 ([14, 4.3.9]). Let (ζ, n) as in Lemma 5.14. If (X, F •) is (ζ, n)-semistable, its stabilizer is either trivial or C∗. In the latter case, (X, F •) has the unique decomposition (X♭, F •♭ ) ⊕ (X♯, F •♯ ) such that both (X♭, F •♭ ), (X♯, F •♯ ) are (ζ, n)-stable, and µ(ζ,n)(X♭) = µ(ζ,n)(X♯). The stabilizer comes from that of the factor (X♯, F •♯ ) with (X♯)∞ = 0. Proof. Suppose g stabilizes (X, F •). If g has an eigenvalue λ 6= 1, then we have the generalized eigenspace decomposition (X, F •) = (X♭, F •♭ ) ⊕ (X♯, F •♯ ) with (X♭)∞ = C, (X♯)∞ = 0. By Lemma 5.14 (X♭, F •♭ ), (X♯, F •♯ ) are (ζ, n)-stable. Since they have the same µζ,n and are not isomorphic, there are no nonzero homomorphisms between them. Therefore the stabilizer is C∗ in this case. The uniqueness follows from that in Lemma 5.14. Next suppose g is unipotent and let n = g−1. Assume n 6= 0 and let j such that nj 6= 0, nj+1 = 0. We consider the submodule 0 6= Ker nj ( X. From the (ζ, n)-semistability of (X, F •) we have µζ,n(Ker nj, F • ∩ (Ker nj)1) = µζ,n(X, F •). Therefore (Ker nj, F • ∩ (Ker nj)1) and (X/ Ker nj, F •/F • ∩ (Ker nj)1) are (ζ, n)-stable by Lemma 5.14. They are not isomorphic since they have different ∞-components. However nj : X/ Ker nj → Ker nj is a nonzero homomorphism, and we have a contradiction by Lemma 5.4. 5.6. Enhanced master space. We continue to fix c, m ∈ Z≥0, ℓ ∈ N . As we mentioned respectively. Their descents will be denoted by the same notations. but with different lifts of the action. Here V1 is a vector space of dimension N, on which G acts naturally. And Flag(V1, N) be the variety of full flags in V1. Let us denote by above, cM m is constructed as a GIT quotient of a common space Q independent of m. Then the moduli schemes fM m,0 and fM m,ℓ will be also constructed as GIT quotients of eQ = Q× Flag(V1, N) by the action of the group G with respect to a common polarization, L− and L+ the corresponding equivariant line bundles over eQ to define fM m,0 and fM m,ℓ We consider the projective bundle P(L− ⊕ L+) → eQ with the canonical polarization OP(1). Here P(L− ⊕ L+) is the space of 1-dimensional quotients of L− ⊕ L+. We have the natural lifts of the G-action to P(L− ⊕ L+) and OP(1). Let (5.17) be the quotient stack of OP(1)-semistable objects of P(L− ⊕ L+) divided by G, where P(L− ⊕ L+)ss denotes the semistable locus. This is called the enhanced master space. This space was introduced in [27] to study the change of GIT quotients under the change of linearizations. We have an inclusion Ma := P(La)ss/G → M for a = ±. M ≡ M(c) ≡ Mm,ℓ(c) := P(L− ⊕ L+)ss/G (cid:3) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 49 overbP2 × M. The tautological flag of vector bundles over Flag(V1, N) descends to M. We denote it by F• = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1). We also have the universal sheaf E We have a natural C∗-action on P(L− ⊕ L+) given by t · [z− : z+] = [tz− : z+] where [z− : z+] is the homogeneous coordinates system of P(L− ⊕ L+) along fibers. It descends to a C∗-action on M. We have a natural C∗-equivariant structure on the universal family E, F•. The following summarizes properties of M. Theorem 5.18. (1) M is a smooth Deligne-Mumford stack. There is a projective mor- phism M → M0(p∗(c)). (2) The fixed point set of the C∗-action decomposes as MC∗ = M+ ⊔ M− ⊔ GI∈Dm,ℓ(c) MC∗(I), and we have isomorphisms M+ ∼= fM m,ℓ(c), M− ∼= fM m,0(c). The universal family (E,F•) on M is restricted to ones on M+ ∼= fM m,ℓ(c) and M− ∼= fM m,0(c) (which were denoted by the same notation (E,F•)). And the restriction of the C∗-equivariant structure are trivial. (3) There is a diagram (5.19) S(I) )RRRRRR G F vllllll MC∗(I) fM m,min(I♯)−1(c♭) ×fM m,+(c♯), where S(I) is a smooth Deligne-Mumford stack and both F , G are ´etale and finite of degree 1/pD. There is a line bundle LS over S(I) with L⊗pD = G∗(L(E♭)∗) and the restriction of the universal family (E,F•) over M and the universal families (E♭,F•♭ ), (E♯,F•♯ ) over fM m,min(I♯)−1(c♭), fM m,+(c♯) are related by F ∗E ∼= G∗(E♭) ⊕ G∗(E♯) ⊗ LS, F ∗F• ∼= G∗(F•♭ ) ⊕ G∗(F•♯ ) ⊗ LS. Moreover the restriction of the C∗-equivariant structure on the universal family (E,F ) is trivial on the factor (E♭,F♭) and of weight 1/pD on (E♯,F♯) under the above identification. S We need to explain some notations: • Dm,ℓ(c) is the set of decomposition types: (5.20) Dm,ℓ(c) :=(cid:26)I = (I♭, I♯)(cid:12)(cid:12)(cid:12)(cid:12) For I ∈ Dm,ℓ(c), we set c♯ = pem, c♭ = c − c♯ ∈ H∗(bP2). N = I♭ ⊔ I♯, I♭, I♯ 6= ∅, I♯ = p(m + 1) for p ∈ Z>0, min(I♯) ≤ ℓ(cid:27) . object (see Definition 5.8) will be determined by the choice of L±. • integers (m0, m1) appeared in the definition of an orientation of a (m, +)-stable The isomorphism F ∗F• ∼= G∗(F•♭ ) ⊕ G∗(F•♯ ) ⊗ LS of universal flags in (3) means as follows: From the first statement F ∗E ∼= G∗(E♭) ⊕ G∗(E♯) ⊗ LS we have a decomposition F ∗(V1(E)) = G∗(V1(E♭))⊕G∗(V1(E♯))⊗LS. Then we have F ∗(F i) = G∗(F i ♯ )⊗LS ♭ = F i−1 where F•♭ , F•♯ are flags indexed by N. If we forget irrelevant factors F i ♭ )⊕G∗(F i ♭ with F i ♭ ) v 50 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA ♯ ♯ ♯ ♯ with F j ♯ = F j−1 ⊂ F max(I♯) , we get the universal flags over fM m,min(I♯)−1(c♭), fM m,+(c♯). The and F j above sets I♭, I♯ consist of indexes of relevant factors. In particular, (F 1 ⊂ V1) appearing in Proposition 5.9 is identified with (F min(I♯) ♯ by V ♯ hereafter. The fixed point substack MC∗ is defined as the zero locus of the fundamental vector field generated by the C∗-action. Note that this does not imply that the action of C∗ is trivial on MC∗, but becomes trivial on the finite cover C∗s = Spec C[s, s−1] → C∗; s 7→ spD = t. Therefore the restriction of a C∗-equivariant sheaf to the fixed point locus is a sheaf tensored by a C∗s-module. In the statements (2),(3), we wrote the weights of C∗s-modules divided by pD, considered formally as weights of rational C∗-modules. The proofs of (1),(2),(3) will be given in §5.7, §5.8, §5.9 respectively. ♯ ). Let us denote F N = F N 1 Remark 5.21. We can define the master space connecting cM m(c) and cM m+1(c) in the same way. However it will be not necessarily a Deligne-Mumford stack as a semistable point possibly has a stabilizer of large dimension. This is the reason why we, following Mochizuki, consider pairs of framed sheaves and flags in cohomology groups. 5.7. Smoothness of the Enhanced Master Space. Let us write the enhanced master space in the quiver description. We first take ζ− sufficiently close to ζ 0. For l = 1, . . . , N, we choose (ζ, n) satisfying (5.5,5.13). We take ζ so that ζ−ζ 0, n are sufficiently smaller than ζ − ζ−. We take a large number k so that k(ζ−, n) and k(ζ, n) are integral. Let L− (resp. L+) be the G-equivariant line bundle over µ−1(0) × Flag(V1, N) corresponding to the stability condition k(ζ−, n) (resp. k(ζ, n)). We consider the projective bundle P(L− ⊕ L+) with the canonical polarization OP(1). We have the natural lifts of the G-action to P(L− ⊕ L+) and OP(1). Let M ≡ M(c) := P(L− ⊕ L+)ss/G be the quotient stack of the semistable locus P(L− ⊕ L+)ss divided by G. The following was shown in e.g., [27, §§3,4]. − ⊗ L⊗t Lemma 5.22. A point x of P(L− ⊕ L+)\ (P(L−) ⊔ P(L+)) is semistable if and only if the corresponding (X, F •) is semistable with respect to a Q-line bundle Lt = L⊗(1−t) + for some t ∈ [0, 1] ∩ Q. Proposition 5.23. M is a smooth Deligne-Mumford stack. Proof. Let x be a semistable point in P(L− ⊕ L+). Then the corresponding point (X, F •) in µ−1(0) × Flag(V1, N) is (ζ′, n)-stable for some ζ′ on the segment connecting ζ and ζ− (Lemma 5.22). We can apply Lemma 5.16 as ζ′ satisfies (m + 1)ζ0 + (m + 2)ζ1 < 0, (m − 1)ζ0 + mζ1 > 0, and n satisfies (5.13). Therefore either the stabilizer of (X, F •) is trivial or (X, F •) decomposes as (X♭, F •♭ ) ⊕ (X♯, F •♯ ). Since (X♯)∞ = 0, X♯ ∼= C⊕p m for some p ∈ Z>0 as explained in Definition 4.6. In this case the stabilizer is C∗, coming from the automorphisms of (X♯, F •♯ ). Its action on the fiber is given by t· u = tpk(m,m+1)·(ζ−ζ−)u for t ∈ C∗. Therefore x only has a finite stabilizer. It is also reduced as the base field is of characteristic 0. Therefore M is Deligne-Mumford. Since P(L− ⊕ L+) is smooth, M is also smooth. (cid:3) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 51 We set (m0, m1) := k(ζ − ζ−) (and hence D = k(m, m + 1) · (ζ − ζ−)) which was used in Theorem 5.18. From our choices of ζ, ζ−, we have D > 0. 5.8. C∗-action. We have a natural C∗-action on P(L− ⊕ L+) given by t· [z− : z+] = [tz− : z+] where [z− : z+] is the homogeneous coordinates system of P(L− ⊕ L+) along fibers. It descends to a C∗-action on M, as it commutes with the G-action. Letting C∗ act trivially on V0, V1 and the universal flag over Flag(V1, N), we have the C∗-equivariant structure on the universal family E, F• = (0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F N−1 ⊂ F N = V1). The fixed point substack MC∗ is defined as the zero locus of the fundamental vector field generated by the C∗-action. Note that this does not imply that the action of C∗ is trivial on MC∗. The action becomes trivial after a finite cover C∗ → C∗. We have an inclusion Ma := P(La)ss/G → M for a = ±. Then Ma is a component of the fixed point set MC∗. From the construction Ma is the moduli stack of objects (X, F •), which are stable with respect to La. From our choice of (ζ, n), we have M+ ∼= fM m,ℓ by Lemma 5.6. Since n is sufficiently smaller than ζ − ζ−, (X, F •) is stable with respect to L− if and only if X is ζ−-stable. Thus we have M− ∼= fM m,0. Next consider a fixed point in MC∗ other than M+ ⊔ M−. Suppose that a point x = ((X, F •), [z− : z+]) in P(L−⊕L+)ss\(P(L−)⊔P(L+)) is mapped to a fixed point in the quotient M. It means that the tangent vector generated by the C∗-action at x is contained in the subspace generated by the G-action. In view of Lemma 5.16 this is possible only if (X, F ∗) has a nontrivial stabilizer, and hence decompose as (X♭, F •♭ ) ⊕ (X♯, F •♯ ) to the direct sum of two (ζ′, n)-stable objects with the equal µζ′,n for some ζ′ on the segment connecting ζ and ζ−. (See Lemma 5.22 and Lemma 5.16.) We number the summand so that (X♭)∞ = C. Therefore (X♯)∞ = 0 and hence X♯ ∼= C⊕p m for some p ∈ Z>0. The data u = z+/z− corresponds to an isomorphism L(X) ∼= C. Conversely suppose we have such a decomposition (X, F •) = (X♭, F •♭ ) ⊕ (X♯, F •♯ ). Let V = V ♭ ⊕ V ♯ be the corresponding decomposition of V . We lift the C∗-action on M to P(L− ⊕ L+)ss by (5.24) which is well-defined on the covering C∗ → C∗; s 7→ spD = t, and fixes ((X, F •), [z− : z+]) = ((X♭, F •♭ )⊕ (X♯, F •♯ ), [z− : z+]). Since this C∗-action is equal to the original one up to the G-action, it is the same on the quotient M. Therefore the point x = ((X, F •), [z− : z+]) is mapped to a fixed point in M. ((X, F •), [z− : z+]) 7→ (idV ♭ ⊕t1/pD idV ♯) · ((X, F •), [tz− : z+]), Let α/F i−1 α ) = 1} Iα := {i ∈ N dim(F i for α = ♭, ♯. Then we have the decomposition N = I♭ ⊔ I♯. The datum (I♭, I♯) is called the decomposition type of the fixed point. Since dim(X♯)1 = p(m+1), we have I♯ = p(m+1). Lemma 5.25 ([14, 4.4.3]). min(I♯) ≤ ℓ. Proof. Suppose min(I♯) > ℓ. Then (5.5) implies µζ,n(X♯) < µζ,n(X). (See the proof of Lemma 5.6.) On the other hand we have µζ−,n(X♯) < µζ−,n(X) since n is sufficiently smaller than ζ − ζ−. Therefore we cannot have µζ′,n(X♯) = µζ′,n(X) for any ζ′ on the segment connecting ζ and ζ−. This contradicts with the assumption. (cid:3) 52 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Conversely suppose an object (X, F •) = (X♭, F •♭ ) ⊕ (X♯, F •♯ ) with the decomposition type (I♭, I♯) with min(I♯) ≤ ℓ is given. We also suppose X♯ ∼= C⊕p m . We take a point x of P(L− ⊕ L+) \ (P(L−) ⊔ P(L+)) from the fiber over (X, F •). Since we have µζ,n(X♯) > µζ,n(X) and µζ−,n(X♯) < µζ−,n(X) by the same argument as above, we can find ζ′ with µζ′,n(X♯) = µζ′,n(X). Then x is semistable if and only if both (X♭, F♭) and (X♯, F♯) are (ζ′, n)-stable. Lemma 5.26 ([14, 4.4.4]). (1) (X♭, F♭) is (ζ′, n)-stable if and only if (m, min(I♯) − 1)- stable. m , and we have (2) (X♯, F♯) is (ζ′, n)-stable if and only if (m, +)-stable, i.e., X♯ ∼= C⊕p S1 ∩ F min(I♯) Note that dim F min(I♯) = 0 for any proper submodule S ( X♯ of a form S ∼= C⊕q m . = 1, dim F min(I♯)−1 = min(I♯) − 1. So the definitions in §5.1 apply though F♭, F♯ are flags which possibly have repetitions. Proof. (1) Let S ⊂ X♭ be a submodule. We need to study the stability inequalities when ζ 0 · dim S = 0. We first suppose S∞ = 0. Then the inequality µ(ζ′,n)(S) < µ(ζ′,n)(X♭) = µ(ζ′,n)(X♯) is equivalent to ♯ ♭ ♯ Pi ni dim(S1 ∩ F i rank S ♭ ) < Pi ni dim F i rank X♯ ♯ since ζ′ · dim S/ rank S = ζ′ · dim X♯/ rank X♯ = (2m + 1)−1 (mζ′0 + (m + 1)ζ′1). Since ni (i ≥ min(I♯)) is much smaller than nmin(I♯)−1 by (5.5b), we must have S1 ∩ F min(I♯)−1 = 0 = 0. Then S1 ∩ F min(I♯) if the inequality holds. Conversely suppose S1 ∩ F min(I♯)−1 Next suppose S∞ = C. Then the inequality µ(ζ′,n)(S) < µ(ζ′,n)(X♭) = µ(ζ′,n)(X♯) is Thus the inequality holds again by (5.5b). = 0. ♭ ♭ ♭ equivalent to Pi ni dim(F i ♭ /S1 ∩ F i ♭ ) > Pi ni dim F i rank X♯ ♯ . rank(X♭/S) 6⊂ S1 by the same argument as above. Thus (X♭, F♭) is (2) First note that X♯ must be ζ 0-semistable as ζ′ is close to ζ 0 and n is small. Then m , as explained in Definition 4.6. To prove the remaining part, the same argument (cid:3) This is equivalent to F min(I♯)−1 (ζ′, n)-stable if and only if (m, min(I♯) − 1)-stable. X♯ ∼= C⊕p as above works. ♭ If we rephrase what we have observed in terms of sheaves, we get Proposition 5.27 ([14, 4.5.2]). MC∗(I) is the moduli stack of objects (((E♭, Φ), F •♭ ), (E♯, F •♯ ), ρ) where • ((E♭, Φ), F •♭ ) is (m, min(I♯) − 1)-stable, • (E♯, F •♯ ) is (m, +)-stable, • ρ is an isomorphism L(E♭ ⊕ E♯) ∼=−→ C. Moreover the restriction of the universal family (E,F•) on M decomposes as E = ME♭ ⊕ ME♯, F• = MF•♭ ⊕ MF•♯ , PERVERSE COHERENT SHEAVES ON BLOW-UP. III 53 where MF•♭ , MF•♯ are flags (0 = MF 0 ♯ ⊂ · · · ⊂ MF N ♯ = V1(ME♯)). The restriction of the C∗-equivariant structure on the universal family (E,F ) is trivial on the factor (ME♭, MF♭) and of weight 1/pD on (ME♯, MF♯) under the above identifica- tion. ♭ = V1(ME♭)), (0 = MF 0 ♭ ⊂ · · · ⊂ MF N The last assertion follows from the description of the C∗-action at (5.24). 5.9. Decomposition into product of two moduli stacks. Let ¯M be the moduli stack of objects ((E♭, Φ), F •♭ , ρ♭) where • ((E♭, Φ), F •♭ ) is (m, min(I♯) − 1)-stable, • ρ♭ is an isomorphism L(E♭) ∼=−→ C. We have a natural projection ¯M → fM m,min(I♯)−1 forgetting ρ♭. It is a principal C∗-bundle. On the other hand, let fM m,+ be the moduli stack of oriented (m, +)-stable sheaves with flags as in §5.3. In order to distinguish from ρ, we denote the orientation by ρ♯. Then we have MC∗(I) ∼= ( ¯M ×fM m,+)/C∗, where C∗ acts by ρ♭ 7→ tρ♭, ρ♯ 7→ t−1ρ♯. Let us take a covering C∗s → C∗ s 7→ spD = t. 1/pD. Then we have an ´etale and finite morphism F : ( ¯M × fM m,+)/C∗s → MC∗(I) of degree The action of C∗s on the second factor fM m,+ is trivial, since it can be absorbed in the isomorphism s−1 id : E♯ ∼=−→ E♯ as (5.28) E♯ s−1 id E♯ Therefore we have L(E♯) = det H 1(E♯)⊗m0 ⊗ det H 1((E♯)(C))⊗m1 s−pD id s−pDρ♯ C id L(E♯) = det H 1(E♯)⊗m0 ⊗ det H 1((E♯)(C))⊗m1 / C. ρ♯ ( ¯M ×fM m,+)/C∗s = ¯M /C∗s ×fM m,+. associated with the representation of C∗ with weight 1. It is a line bundle LS such that L⊗pD S Furthermore we have an ´etale and finite morphism G : ¯M /C∗s → ¯M /C∗ of degree 1/pD. But the latter is nothing but fM m,min(I♯)−1. Hence we have the diagram in Theorem 5.18 with S(I) = ( ¯M ×fM m,+)/C∗s. From (5.28) the universal sheaf E♯ over fM m,+ is twisted by the line bundle over ¯M /C∗ = G∗L(E♭)∗. Therefore we have F ∗(cid:0)ME♯(cid:1) = G∗(E♯) ⊗ LS. On the other hand, we have F ∗(cid:0)ME♭(cid:1) = G∗(E♭). 5.10. Normal bundle. Let us describe the normal bundles N(M±) of M± and N(MC∗(I)) of MC∗(I) in M in this subsection. We need to prepare several notations. Recall first that the covering C∗s → C∗; s 7→ spD = t acts trivially on MC∗(I) (while the original C∗ does not). Hence the tangent space at a fixed point has a natural C∗s-module structure. We formally consider it as a module structure of the original C∗ dividing weights by pD. / / / 54 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Recall also that the restriction of the universal sheaf E decomposes as ME♭ ⊕ ME♯ over MC∗(I) (see Proposition 5.27). Let Ext•q2 denotes the higher derived functor of the composite functor q2∗ ◦ Hom. Let (5.29) N(ME♭, ME♯) := − (−1)a Exta q2(ME♭, ME♯), where this is a class in the equivariant K-group of MC∗(I). We use similar notation N(ME♯, ME♭) exchanging the first and second factors. Later we will also use N(•,•) replacing ME♯, ME♭ by similar universal sheaves. We have already used this notation in Theorem 1.5. 2Xa=0 Let Flag(V α 1 , indexed by N, F i 1 , Iα) := {a flag F •α of V α α = 0 if and only if i /∈ Iα} for α = ♭, ♯. We have an embedding Flag(V ♭ 1 , I♯) → Flag(V1, N) given by (F •♭ , F •♯ ) 7→ F •♭ ⊕ F •♯ . Let N0 denote its normal bundle. It has a natural C∗-equivariant 1 , I♯) is a component of C∗-fixed points in Flag(V1, N) structure as Flag(V ♭ with respect to the C∗-action induced by C∗ ∋ t 7→ idV ♭ 1 ∈ GL(V1). More precisely, when we write α/F i−1 1 , I♭) × Flag(V ♯ 1 , I♭) × Flag(V ♯ 1 ⊕t1/pD idV ♯ N0 =Mi>j ) ⊕Mi>j Hom(F i ♭ /F i−1 ♭ , F j ♯ /F j−1 ♯ Hom(F i ♯ /F i−1 ♯ , F j ♭ /F j−1 ♭ ), the first term has weight 1/pD and the second term has weight −1/pD. We have an associated vector bundle, denoted also by N0, over MC∗(I), induced from the flag bundle structure eQ/G → Q/G between quotient stacks. ∓ ⊗ L± with the C∗-action (2) The normal bundle N(MC∗(I)) of MC∗(I) is equivariant K-theoretically given by Theorem 5.30. (1) The normal bundle N(M±) of M± is L∗ of weight ±1. N0 + N(ME♭, ME♯) ⊗ I1/pD + N(ME♯, ME♭) ⊗ I−1/pD, where In denotes the trivial line bundle over MC∗(I) with the C∗-action of weight n. Proof. First consider the case of M± = P(L±)ss/G. The normal bundle is the descent of ∓ ⊗ L± with the C∗-action of weight the normal bundle P(L±) ⊂ P(L− ⊕ L+). Then it is L∗ ±1. In the remaining of the proof we consider the case of MC∗(I). The normal bundle is the sum of nonzero weight subspaces in the restriction of the tangent bundle of M to MC∗(I). Take a point ((X, F •), [z− : z+]) ∈ P(L− ⊕ L+)ss which descends to a fixed point in MC∗(I). We have a decomposition (X, F •) = (X♭, F •♭ ) ⊕ (X♯, F •♯ ) as above. Let V = V ♭ ⊕ V ♯ be the corresponding decomposition of V . We lift the C∗-action on M to P(L− ⊕ L+)ss as in (5.24). The tangent space at the point corresponding to ((X, F •), [z− : z+]) is the quotient vector space Ker dµ ⊕ TF • Flag(V1, N) ⊕ T[z−:z+]P1/ Hom(V0, V0) ⊕ Hom(V1, V1), PERVERSE COHERENT SHEAVES ON BLOW-UP. III 55 Hom(V1, V1)) + TF • Flag(V1, N) + T[z−:z+]P1 where dµ and Hom(V0, V0) ⊕ Hom(V1, V1) → Ker dµ are as in (4.4), and Hom(V0, V0) ⊕ Hom(V1, V1) → TF • Flag(V1, N) ⊕ T[z−:z+]P1 is the differential of the G-action. This quo- tient space has the C∗-module structure induced from the above lift of the C∗-action. In the equivariant K-group, we can replace this space by Ker dµ − (Hom(V0, V0) ⊕ The expression Ker dµ− (Hom(V0, V0)⊕ Hom(V1, V1)) is equal to the alternating sum of cohomology groups of the tangent complex (4.4) in the K-group. Moreover, the complex (4.4) decomposes into four parts, the tangent complex for X♭, the sum of p-copies of the complex (4.8) for X♭, the sum of p-copies of the complex (4.10) for X♭, and the sum of p2-copies of the tangent complex for Cm. Since C∗ acts on V ♭ with weight 0 and V ♯ with weight 1/pD, the first and fourth parts do not contribute to the normal bundle. The second and third terms give p(ME♭, ME♯) + Ext1 (cid:0)− Ext0 p(ME♭, ME♯)(cid:1) ⊗ I1/pD +(cid:0)− Ext0 = N(ME♭, ME♯) ⊗ I1/pD + N(ME♯, ME♭) ⊗ I−1/pD. p(ME♯, ME♭)(cid:1) ⊗ I−1/pD p(ME♯, ME♭) + Ext1 The contribution from TF • Flag(V1, N) is given by N0. Thus we have Theorem 5.30. (cid:3) We have no contribution from T[z−:z+]P1. 5.11. Relative tangent bundles of flag bundles. Let Θrel be the bundles over various moduli stacks induced from the relative tangent bundle of the flag bundle eQ/G → Q/G rel over fM m,min(I♯)−1(c♭) ×fM m,+(c♯) as in §6.2 coming from the tangent bundles of Flag(V ♭ bundle N0 of Flag(V ♭ 1 , I♯). Recall the normal 1 , I♯) in Flag(V1, N) is considered as a vector bundle On the other hand, we have vector bundles Θ♭ 1 , I♭) and Flag(V ♯ rel, Θ♯ 1 , I♭) × Flag(V ♯ over fM m,min(I♯)−1(c♭) ×fM m,+(c♯) (see §5.10), so we have an exact sequence rel) → F ∗ΘrelMC∗ (I) → G∗(N0) → 0 0 → G∗(Θ♭ rel ⊕ Θ♯ (5.31) of C∗-equivariant vector bundles, where Θ♭ equivariant structure as describe in §5.10. , I♯ \ {min(I♯)}) over the quotient stack (det Q⊗D)×/C∗ (see Proposition 5.9). Let Θ′rel be the relative tangent bundle of the fiber. Then we have an exact sequence of vector bundles Recall also that the second factor fM m,+(c♯) is the flag bundle Flag(V ♯ rel has weight 0 and N0 has the C∗- 1/F min(I♯) rel ⊕ Θ♯ ♯ 0 → Θ′rel → Θ♯ (5.32) coming from the fibration Flag(V ♯ rel → Hom(V ♯ 1/F min(I♯) 1/F min(I♯) ) → Flag(V ♯ ,F min(I♯) ) → 0 1) → P(V ♯ 1). ♯ ♯ ♯ 6. Wall-crossing formula We now turn to Mochizuki method (Mb). (See [14, §7.2].) We will apply the fixed point formula to the equivariant homology groups H eT×C∗ (M(c)) of the enhanced moduli space M(c), which is a module over H∗C∗(pt) ∼= C[]. For the definition of the homology group of a Deligne-Mumford stack, see [28]. ∗ 56 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA 6.1. Equivariant Euler class. In the fixed point formula we put an equivariant Euler class in the denominator and we later consider an equivariant Euler class of a class in the K-group. Let us explain how we treat them in this subsection. Let Y be a variety (or a Deligne-Mumford stack) with a trivial C∗-action. The Grothen- dieck group of C∗-equivariant vector bundles decomposes as KC∗(Y ) = K(Y ) ⊗Z R(C∗) where R(C∗) is the representation ring of C∗. Let In denote the 1-dimensional represen- tation of C∗ with weight n. For a class α ∈ K(Y ), we set ci(α)(n)r(α)−i ∈ H∗(Y )[−1, ] := H∗(Y ) ⊗C C[−1, ], e(α ⊗ In) :=Xi≥0 where ci(α) is the ith Chern class of α and r(α) = ch0(α) is the (virtual) rank of α. If n 6= 0, this element is invertible. In general, if α ∈ KC∗(Y ) is a sum of αn⊗ In with n 6= 0, its equivariant Euler class e(α) is defined in H∗(Y )[−1, ]. ∗ possibly infinite i's. (M(c)) → H eT×C∗ (M0(p∗(c))). Since C∗ acts trivially ∗ ∗ (M0(p∗(c))) ∼= H eT on M0(p∗(c)), we have H eT×C∗ ∗ (M0(p∗(c))) ⊗ C[]. For a cohomology use similar push-forward homomorphisms from homology groups of various moduli stacks We also consider the case when another group eT acts on Y . Then e(α ⊗ In) can be H∗(Y ×eT En)[−1, ], where En → En/eT is a finite still defined as an element in lim←−n dimensional approximation of the classifying space EeT → BeT for eT . Note ci(α) 6= 0 for 6.2. Wall-crossing formula (I). The projective morphism bπ : M(c) → M0(p∗(c)) in- duce a homomorphism bπ∗ : H eT×C∗ class • on M(c), we denote the pushforwardbπ∗(• ∩ [M(c)]) byRM(c) • as in §1. We also and denote them in similar ways, e.g.,RM±(c),RMC∗ (I), etc. Let Φ(E) be as in §1.4. We denote also by Φ(E) the class on fM m,ℓ(c) given by the same Let Θrel be the relative tangent bundle of the flag bundle eQ/G → Q/G. Then we have the induced bundle over fM m,ℓ(c) by the restriction. We denote it also by Θrel for brevity. Θrel. It has a natural C∗-equivariant structure. We introduce eΦ(E) := 1 Let e(F ) denote the equivariant Euler class of the equivariant vector bundle F , that is the top Chern class ctop(F ). See §6.1 for its generalization to a K-theory class F in our situation. We also have the pullback to the enhanced master space M(c), which is again denoted by v1(c)! Φ(E) ∪ e(Θrel) formula. On the enhanced master space M(c), we can consider the class defined by the same formula as Φ(E), which is regarded as a C∗-equivariant class. so that we have where 'Res=0' means taking the coefficient of −1. Here v1(c) = dim V1(E) for a sheaf E with ch(E) = c. Then the fixed point formula in the equivariant homology group gives us Φ(E). ZfM m,0(c)eΦ(E) =ZcM m(c) Φ(E) −ZfM m,ℓ(c)eΦ(E) = XI∈Dm,ℓ(c) ZcM m(c) =0ZMC∗ (I) Res eΦ(E) e(N(MC∗(I))) , PERVERSE COHERENT SHEAVES ON BLOW-UP. III 57 By Proposition 5.9 and Theorem 5.18 together with computations of normal bundles §5.10, we can rewrite the right hand side to get Theorem 6.1. Res Φ(E) = XI∈Dm,ℓ(c)ZfM m,min(I♯)−1(c♭) ZfM m,ℓ(c)eΦ(E) −ZcM m(c) =0heΦ(E♭ ⊕ (Cm ⊠ Q ⊗ I−1))Φ′(E♭)i , whereeΦ(E♭⊕(Cm⊠Q⊗I−1)) is defined exactly aseΦ(E) by replacing E by E♭⊕(Cm ⊠Q⊗I−1) everywhere, and Φ′(E♭) is another cohomology class given by (6.2) Φ′(E♭) = Here In denotes the trivial line bundle with the C∗-action of weight n, and N(•,•) is the equivariant K-theory class given by the negative of the alternating sum of Ext-groups (see (5.29)). (Note that Φ′ depends on I.) e(N(E♭, Cm) ⊗ Q ⊗ I−1) e(N(Cm,E♭) ⊗ Q∗ ⊗ I1) e ((V1(Cm) ⊗ Q/O)∗) ZGr(m+1,p) (v1(c♯) − 1)!v1(c♭)! v1(c)! . The proof will be given in the next subsection. 6.3. Fixed point formula on the enhanced master space. Let ι±, ιI be the in- clusions of M±(c), MC∗(I) into M(c). Let ι∗ , ι∗I be the pullback homomorphisms, ± which are defined as M(c) is smooth. They will be omitted from formulas eventually. Let e(N(M+(c))), e(N(MC∗(I))) be the equivariant Euler class of the normal bundles N(M+(c)), N(MC∗(I)). The localization theorem in the equivariant cohomology groups says the following diagram is commutative: (M(c) ×eT En) ⊗C[] C[−1, ] H∗(M(c)C∗ ×eT En)[−1, ] H C∗ ∗ ∼=−−−→ lim←−n lim←−n R M(c)y lim←−n H∗(M0(p∗(c) ×eT En))[−1, ] where the upper horizontal arrow is given by M+(c) +R M−(c) +PI R MC∗ (I) H∗(M0(p∗(c) ×eT En))[−1, ], yR lim←−n +XI ι∗+ e(N(M+(c))) + ι∗ − e(N(M−(c))) ι∗I e(N(MC∗(I))) , Let T (1) be the trivial line bundle with the C∗-action of weight 1. We have and En → En/eT is a finite dimensional approximation of EeT → BeT as above. + XI∈Dm,ℓ(c)ZMC∗ (I) eΦ(E)c1(T (1)) ZM(c)eΦ(E)c1(T (1)) =Xa=±ZMa(c) eΦ(E)c1(T (1)) the C∗-action, the left hand side is restricted to 0 at = 0: RM(c)eΦc1(T (1))(cid:12)(cid:12)(cid:12)=0 holds in lim←− H∗(M0(p∗(c) ×eT En))[−1, ]. Since T (1) is a trivial line bundle if we forget e(N(MC∗(I))) e(N(Ma)) = 0. On 58 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA the other hand, c1(T (1))Ma(c) = and c1(T (1))MC∗ (I) = . Moreover we have 1 e(N(Ma)) = a( − ω)−1 = a ∞Xi=0(cid:16) ω (cid:17)i =0ZMC∗ (I) We now use the diagram (5.19) to rewrite the integral in the right hand side of (6.3): for a = ±, where ω = c1(L∗+ ⊗ L−). Combining with Theorem 5.18(2) we get (6.3) Res . eΦ(E) e(N(MC∗(I))) ZfM m,ℓ(c)eΦ(E) −ZcM m(c) eΦ(E) Φ(E) = − XI∈Dm,ℓ(c) = pDZS(I) e(N(MC∗(I))) ZMC∗ (I) F ∗eΦ(E) F ∗(Φ(E)) = Φ(G∗(E♭) ⊕ G∗(E♯) ⊗ LS ⊗ I1/pD). F ∗e(N(MC∗(I))) . From Theorem 5.18(3) we have Since L⊗pD S = G∗(L(E♭)∗), we have c1(LS) = − 1 pD G∗c1(L(E♭)). Since LS appears as c1(LS) in Φ(G∗(E♯) ⊗ LS ⊗ I1/pD), we can formally write Φ(G∗(E♭) ⊕ G∗(E♯) ⊗ LS ⊗ I1/pD) = G∗Φ(E♭ ⊕ E♯ ⊗ L(E♭)−1/pD ⊗ I1/pD), meaning that we replace c1(L(E♭)−1/pD) by −c1(L(E♭))/pD. Similarly from Theorem 5.30 and Theorem 5.18(3) we have F ∗e(N(MC∗(I))) = F ∗(cid:0)e(N(ME♭, ME♯) ⊗ I1/pD)e(N(ME♯, ME♭) ⊗ I−1/pD)e(N0)(cid:1) = G∗(cid:0)e(N(E♭,E♯) ⊗ L(E♭)−1/pD ⊗ I1/pD)e(N(E♯,E♭) ⊗ L(E♭)1/pD ⊗ I−1/pD)e(N0)(cid:1) . From (5.31,5.32) we have rel)e(N0)(cid:17) 1/F min(I♯) ♯ ,F min(I♯) ♯ )(cid:17) e(N0)(cid:17) . Therefore we get rel)e(Θ♯ F ∗(e(Θrel)) = G∗(cid:16)e(Θ♭ = G∗(cid:16)e(Θ♭ rel)e(Θ′rel)e(cid:16)Hom(V ♯ F ∗eΦ pDZS(I) v1(c)!ZfM m,min(I♯)−1(c♭)×fM m,+(c♯) F ∗e(N(MC∗(I))) 1 = Φ(E ♭ ⊕ E♯ ⊗ L(E♭)−1/pD ⊗ I1/pD) e(N(E♭,E♯) ⊗ L(E♭)−1/pD ⊗ I1/pD) e(Θ♭ rel)e(Θ′rel)e(cid:16)Hom(V ♯ × 1/F min(I♯) ♯ ,F min(I♯) ♯ )(cid:17) . e(N(E♯,E♭) ⊗ L(E♭)1/pD ⊗ I−1/pD) PERVERSE COHERENT SHEAVES ON BLOW-UP. III 59 We use the claim that fM m,+(c♯) is a flag bundle over (detQ⊗D)×/C∗ (Proposition 5.9(1)) to rewrite this further as (v1(c♯) − 1)! v1(c)! ZfM m,min(I♯)−1(c♭)×(det Q⊗D)×/C∗ Φ(E ♭ ⊕ E♯ ⊗ L(E♭)−1/pD ⊗ I1/pD) e(Θ♭ rel) e(N(E♭,E♯) ⊗ L(E♭)−1/pD ⊗ I1/pD) ,F min(I♯) 1/F min(I♯) e(cid:16)Hom(V ♯ e(N(E♯,E♭) ⊗ L(E♭)1/pD ⊗ I−1/pD) )(cid:17) × ♯ ♯ . Gr(m + 1, p), and then plug into (6.3) to get v1(c♭)! Φ(•)e(Θ♭ rel), use Proposition 5.9(2) to replace (detQ⊗D)×/C∗ by Φ(E) = XI∈Dm,ℓ(c)ZfM m,min(I♯)−1(c♭) ZGr(m+1,p) ♥, (v1(c♯) − 1)!v1(c♭)! v1(c)! pD 1 Res =0 eΦ(E♭⊕E♯⊗L(E♭)−1/pD⊗I1/pD)Φ′(E♭), We set eΦ(•) := 1 ZfM m,ℓ(c)eΦ(E)−ZcM m(c) where Φ′(E♭) = − ♥ = × e(N(E♭, Cm) ⊗ Q ⊗ detQ−1/p ⊗ L(E♭)−1/pD ⊗ I1/pD) e ((V1(Cm) ⊗ Q/O)∗) . e(N(Cm,E♭) ⊗ Q∗ ⊗ detQ1/p ⊗ L(E♭)1/pD ⊗ I−1/pD) Here • RGr(m+1,p) means the pushforward with respect to the projection fM m,min(I♯)−1(c♭)× Gr(m + 1, p) → fM m,min(I♯)−1(c♭). • detQ−1/p is understood as before: we replace c1(detQ−1/p) by −c1(detQ)/p. Let us slightly simplify the formula. First note that I±1/pD appears with detQ∓1/p ⊗ L(E♭)∓1/pD. Let ω = −(c1(L(E♭)) + c1(Q))/pD. Thus Φ′(E♭) is written as By a direct calculation we have Therefore we have Res =0 Aj( − ω)j. ∞Xj=−∞ ( − ω)j =(1 ∞Xj=−∞ Res =0 0 if j = −1, otherwise. ∞Xj=−∞ Aj( − ω)j = Res =0 Ajj. This means that we can erase det Q∓1/p ⊗L(E♭)∓1/pD from Φ′(E♭). (The last simplification appeared in [14, Proof of Theorem 7.2.4]) Next note that nontrivial contributions of the C∗-action appear as I±1/pD. If we take the covering C∗s → C∗; s 7→ t = s−pD, I±1/pD is of weight ∓1 as a C∗s-module by our 60 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA (pt) = C[s] ∼= H∗C∗(pt) = C[] convention. Since that we have a natural isomorphism H∗C∗s by s = −/pD, we can replace by s noticing Res=0 f () = −pD × Ress=0 f (spD). We use this replacement and then replace back s by again. Therefore we get the formula (6.2). We have completed the proof of Theorem 6.1. 6.4. Proof of Theorem 1.5. The right hand side of the formula in Theorem 6.1 can For j ∈ Z>0 let We then get Theorem 1.5. The proof has combinatorial nature and is the same as one in [14, §7.6]. We reproduce it here for reader's convenience. In fact, it is enough to consider the case m = 0, as a general case follows from m = 0 formula in Theorem 6.1 is simplified as we can assume p = 1 as Gr(m + 1, p) = Gr(1, p) is empty otherwise. Theorem 1.5 is obtained in this way, but we give a proof for general m. be expressed by an integral over cM m(c♭) by using the formula again. We continue this procedure recursively, we will get a wall-crossing formula comparingRcM m(c) andRcM m+1(c). since we have an isomorphism cM m(c) ∼=−→ cM 0(ce−m[C]); (E, Φ) 7→ (E(−mC), Φ). Then the >0(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) piem = c) . j (c) :=(~c = (c♭, p1, . . . , pj) ∈ H∗(bP2) × Zj We denote the universal family for cM m(c♭) by E♭ as above. Let Sm(c) =F∞j=1 Sm >0 we consider the product of Grassmannian varietiesQj For ~p = (p1, . . . , pj) ∈ Zj torus (C∗)j acting trivially on Qj j (c). i=1 Gr(m+ 1, pi). Let Q(i) be the universal quotient of the ith factor. We consider the j-dimensional i=1 Gr(m + 1, pi). We denote the 1-dimensional weight n representation of the ith factor by eni. (Denoted by In previously.) The equivariant cohomology H∗(C∗)j (pt) of the point is identified with C[1, . . . , j]. Theorem 6.4. (c♭, [ℓ∞]) = 0, c♭ + jXi=1 Sm ZcM m+1(c) Φ(E)−ZcM m(c) Φ(E) = X~c∈Sm(c)ZcM m(c♭) j=0· · · Res Res 1=0 Φ(E♭⊕ jMi=1 Cm⊠Q(i)⊗e−i)∪Ψ~p(E♭), where Ψ~p(•) is another cohomology class given by (6.5) Ψ~p(•) := (m + 1)j 1 ♥ = jYi=1 × Y1≤i16=i2≤j i=1 Gr(m+1,pi) ♥, 1 P1≤k≤i pkZQj jYi=1 e(cid:0)(V1(Cm) ⊗ Q(i)/O)∗(cid:1) e(Q(i1) ⊗ Q(i2)∗ ⊗ e−i1 +i2 ). e(N(•, Cm) ⊗ Q(i) ⊗ e−i) e(N(Cm,•) ⊗ Q(i)∗ ⊗ ei) (Note that Ψ~p depends on ~p = (p1, . . . , pj), but not on c♭.) Let us prepare notation before starting the proof. ♯ )}, ♭ with i < min(I (j) ♯ ). We also (j) , ~I♯ ) given by PERVERSE COHERENT SHEAVES ON BLOW-UP. III 61 We defined a cohomology class Ψ~p(E♭) by the formula (6.5), and it is an element in i=1 Gr(m+1, p(i)))[±1 , . . . , ±j ] for some moduli stack M with the universal H∗ eT family E. (bP2×M ×Qj Let Dec(j)(c) be the set of pairs I(j) = (I (j) ) as follows: (j) , ~I♯ ♭ ♯ ) of subsets of v1(c) such that I (i) ♯ = pi(m + 1) i=1 I (i) ♯ . (j) ♯ ♯ ♭ , I (j−1) , . . . , I (1) is a tuple (I (j) • ~I♯ for some pi ∈ Z>0 (1 ≤ i ≤ j). • min(I (1) • I (j) For I(j) ∈ Dec(j)(c) set ♯ ) > min(I (2) is also a subset of v1(c) and we have v1(c) = I (j) ♯ ) > · · · > min(I (j) ♯ ). k(I(j)) := max{i ∈ I (j) ♭ ⊔Fj i < min(I (j) where we understand this to be 0 if there exists no i ∈ I (j) put jXi=1 ) 7→ (I (j) ♯ ). We have a map πj : Dec(j+1)(c) → Dec(j)(c); (I (j+1) := piem (1 ≤ i ≤ j), := c − , . . . , I (1) := (I (j) piem. c(j) ♭ c(i) ♯ , ~I♯ (j+1) ~I♯ (j) , ♯ ♭ ♭ ♭ ♯ ♭ I (j) ♭ := I (j+1) ⊔ I (j+1) fM(I(j)) := fM m,ℓ(c(j) ♭ ), Let where in the first equality we take the unique order preserving bijection I (j) #I (j) ♭ ) corresponding to k(I(j)). and take ℓ ∈ v1(c(j) ♭ ♭ ∼= v1(c(j) ♭ ) = cM (I(j)) := cM m(c(j) ♭ ), For the universal family E (j) (E (j) ♭ ) := Ψ~p(E (j) ♭ ) ΨI(j) ♭ for fM (I(j)) or cM (I(j)) let v1(c(j) ♯ ) − 1)! i=1(v1(c(i) v1(c)! ♭ )!Qj where ~p = (p1, . . . , pj). Lemma 6.6 ([14, 7.6.5]). For each j, we have the formula (m + 1)j jYi=1 iXk=1 pk, Φ(E) (6.7) ZcM m+1(c) Φ(E) −ZcM m(c) ZcM(I(i)) = X1≤i<j + XI(j)∈Dec(j)(c)ZfM (I(j)) I(i)∈Dec(i)(c) i=0· · · Res Res 1=0 Φ(E (i) ♭ ⊕ iMk=1 1=0eΦ(E (j) ♭ ⊕ j =0· · · Res Res (E (i) ♭ ) Cm ⊠ Q(k) ⊗ e−k)ΨI(i) jMk=1 Cm ⊠ Q(k) ⊗ e−k )ΨI(j) (E (j) ♭ ). 62 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA Proof. We prove the assertion by an induction on j. If j = 1, this is nothing but Theo- rem 6.1 applied to ℓ = v1(c). Suppose that the formula is true for j. We apply Theorem 6.1 to get j=0· · · Res Res ♭ ⊕ Cm ⊠ Q(k) ⊗ e−k)ΨI(j) (E (j) ♭ ) jMk=1 1=0eΦ(E (j) −ZcM (I(j)) ZfM (I(j+1)) Φ(E (j) ♭ ⊕ jMk=1 1=0"eΦ(E (j+1) ♭ j+1=0· · · Res Res j =0· · · Res Res 1=0 Cm ⊠ Q(k) ⊗ e−k )ΨI(j) (E (j) ♭ ) Cm ⊠ Q(k) ⊗ e−k)Φ′(E (j+1) ♭ )# , ⊕ j+1Mk=1 ZfM (I(j)) = XI(j+1)∈Dec(j+1)(c) πj(I(j+1))=I(j) where Φ′(E (j+1) ♭ ) = v1(c(j+1) ♭ ) − 1)! )!(v1(c(j+1) ♯ v1(c(j) ♭ )! ZGr(m+1,pj+1) ♥ , Cm) ⊗ Q(j+1) ⊗ e−j+1)e(N(Cm,E (j+1) ⊕ Cm ⊠ Q(j+1) ⊗ e−j+1)e(cid:0)(V1(Cm) ⊗ Q(j+1)/O)∗(cid:1) ) ⊗ Q(j+1)∗ ⊗ ej+1) ♭ . ΨI(j)(E (j+1) ♭ e(N(E (j+1) ♭ ♥ = We have Φ′(E (j+1) ) thanks to the multiplicative property of the Euler class and N(Cm, Cm) = − Hom(Cm, Cm) = −C idCm . Hence the formula holds for j + 1. ) = ΨI(j+1)(E (j+1) (cid:3) ♭ ♭ If j is sufficiently large, Dec(j)(c) = ∅. Hence we have ZcM m+1(c) Φ(E) Φ(E) −ZcM m(c) ∞Xj=1 XI(j)∈Dec(j)(c)ZcM(I(j)) = jMi=1 j =0· · · Res Res 1=0 Φ(E (j) ♭ ⊕ Cm ⊠ Q(i) ⊗ e−i)ΨI(j) (E (j) ♭ ). We have a map ρj : Dec(j)(c) → Sj(c) given by ρj(I(j)) = (c♭, p1, . . . , pj) = (c(j) ♭ Therefore the right hand side is equal to , I (1) m + 1 ♯ , . . . , I (j) m + 1 ♯ ). ∞Xj=1 X~c∈Sj(c) #ρ−1 j (~c) v1(c♭)!Qj i=1(pi(m + 1) − 1)! (m + 1)j v1(c)! ×ZcM m(c♭) j=0· · · Res Res 1=0 pk jYi=1 iXk=1 jMi=1 Φ(E♭ ⊕ Cm ⊠ Q(i) ⊗ e−i)Ψ~p(E♭). Thus Theorem 6.4 follows from the following lemma. PERVERSE COHERENT SHEAVES ON BLOW-UP. III 63 Lemma 6.8 ([14, 7.6.7]). #ρ−1 j (~c) v1(c♭)!Qj i=1(pi(m + 1) − 1)! v1(c)! = 1 (m + 1)j jYi=1 . 1 P1≤k≤i pk Proof. The set ρ−1 j (~c) is (I (j) ♭ , I (j) ♯ , . . . , I (1) v1(c) = I (j) ♭ ⊔ jGi=1 I (i) ♯ ♭ ,I (j) = v1(c♭),I (i) ♯ ) > · · · > min(I (j) ♯ ) ♯ min(I (1) ♯ ) > min(I (2) = pi(m + 1),  ♯ )(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) .  ♭ ♯ ⊂ N \ I (j) . ♭ ). Let x be this number. Nj−1 (cid:1) possibilities. (cid:1) possibilities. We continue ♯ ♯ Nj−1−1 We first choose I (j) Put N := v1(c), N0 := v1(c♭), Ni := pi(m + 1) (1 ≤ i ≤ j). N0(cid:1) possibilities. Next we choose I (j) ♭ ⊂ N. We have (cid:0) N ♯ ) = min(N \I (j) From the second condition, we must have min(I (j) ♭ )\ {x}. We have(cid:0)N−N0−1 ♯ \ {x} ⊂ (N \ I (j) Then the remaining choice is I (j) ♯ ). We have(cid:0)N−N0−Nj−1 Next we choose I (j−1) ⊂ N \ (I (j) ♭ ∪ I (j) until we choose I (1) . Therefore we have N0(cid:19) jYi=1(cid:18)N − N0 −Pk>i Nk − 1 (cid:19) j (~c) =(cid:18) N P1≤k≤i Nk i=1(Ni − 1)! × N0!Qj jYi=1 P1≤k≤i pk(m + 1) P1≤k≤i Nk jYi=1 jYi=1 Ni − 1 Moreover #ρ−1 N! = = 1 1 1 . . References (cid:3) [1] R. Bott, Homogeneous vector bundles, Ann. of Math. (2) 66 (1957), 203 -- 248. [2] A. Braverman and P. Etingof, Instanton counting via affine Lie algebras II: from Whittaker vectors to the Seiberg-Witten prepotential, Studies in Lie theory, 61 -- 78, Progr. Math., 243, Birkhauser, Boston, 2006; arXiv:math/0409441. [3] G. Ellingsrud and L. Gottsche, Variation of moduli spaces and Donaldson invariants under change of polarization, J. Reine Angew. Math. 467 (1995), 1 -- 49. [4] J.D. Fay, Theta functions on Riemann surfaces, Lecture Notes in Math., 352, Springer, Berlin, 1973. [5] W. Fulton and S. Lang, Riemann-Roch algebra, Grundlehren der Mathematischen Wissenschaften, 277, Springer-Verlag, New York, 1985. x+203 pp. [6] A. Gorsky, A. Marshakov, A. Mironov and A. Morozov, RG equations from Whitham hierarchy, Nucl. Phys. B 527 (1998), 690 -- 716; arXiv:hep-th/9802007. [7] L. Gottsche, H. Nakajima and K. Yoshioka, Instanton counting and Donaldson invariants, J. of Differential Geometry, 80 (2008), 343 -- 390; arXiv:math/0606180. [8] , K-theoretic Donaldson invariants via instanton counting, Pure and Applied Math. Quar- terly, 5 (2009), 1029 -- 1110; arXiv:math/0611945. 64 HIRAKU NAKAJIMA AND K ¯OTA YOSHIOKA [9] R. Joshua, Equivariant Riemann-Roch for G-quasi-projective varieties. I., K-Theory 17 (1999), no. 1, 1 -- 35. [10] A. King, Moduli of representations of finite dimensional algebras, Quarterly J. of Math. 45 (1994), 515 -- 530. [11] J. Koll´ar and S. Mori, Birational geometry of algebraic varieties, Cambridge Tracts in Mathematics, 134. Cambridge University Press, Cambridge, 1998. viii+254 pp. [12] A. Losev, N. Nekrasov and S. Shatashvili, Issues in topological gauge theory, Nuclear Phys. B 534 (1998), 549 -- 611; arXiv:hep-th/9711108. [13] A. Marshakov and N. Nekrasov, Extended Seiberg-Witten Theory and Integrable Hierarchy, J. High Energy Phys. 2007, no. 1, 104, 39 pp; arXiv:hep-th/0612019. [14] T. Mochizuki, Donaldson type invariants for algebraic surfaces. Transition of moduli stacks, Lecture Notes in Mathematics, 1972. Springer-Verlag, Berlin, 2009. xxiv+383 pp. [15] D. Mumford, Tata lectures on theta. II, Birkhauser Boston, Boston, MA, 1984. [16] D. Mumford, J. Fogarty and F. Kirwan, Geometric invariant theory, Third Edition, Ergebnisse der Mathematik und ihrer Grenzgebiete (2), 34, Springer-Verlag, Berlin, 1994, xiv+292. [17] H. Nakajima, Lectures on Hilbert schemes of points on surfaces, Univ. Lect. Ser. 18, AMS, 1999. [18] H. Nakajima and K. Yoshioka, Instanton counting on blowup. I. 4-dimensional pure gauge theory, Invent. Math 162 (2005), no. 2, 313 -- 355; arXiv:math/0306198. , Lectures on instanton counting, Algebraic structures and moduli spaces, 31 -- 101, CRM Proc. Lecture Notes, 38, Amer. Math. Soc., Providence, RI, 2004; arXiv:math/0311058. , Instanton counting on blowup. II. K-theoretic partition function, Transform. Groups 10 (2005), 489 -- 519; arXiv:math/0505553. , Perverse coherent sheaves on blow-up. I. A quiver description, preprint, arXiv:0802.3120. , Perverse coherent sheaves on blow-up. II. Wall-crossing and Betti numbers formula, preprint, [19] [20] [21] [22] arXiv:0806.0463, J. of Algebraic Geometry, to appear. [23] [24] N. Nekrasov, Seiberg-Witten prepotential from instanton counting, Adv. Theor. Math. Phys. 7 (2003), , Instanton counting on blowup. III. Theories with matters, in preparation. no. 5, 831 -- 864; arXiv:hep-th/0206161. [25] N. Nekrasov and A. Okounkov, Seiberg-Witten prepotential and random partitions, The unity of mathematics, 525 -- 596, Progr. Math., 244, Birkhauser Boston, Boston, MA, 2006; arXiv:hep-th/0306238. [26] Y. Tachikawa, Five-dimensional Chern-Simons terms and Nekrasov's instanton counting, J. High Energy Phys. 2004, no. 2, 050, 13 pp; arXiv:hep-th/0401184. [27] M. Thaddeus, Geometric invariant theory and flips, J. Amer. Math. Soc. 9 (1996), no. 3, 691 -- 723. [28] A. Vistoli, Intersection theory on algebraic stacks and on their moduli spaces, Invent. Math. 97 (1989), no. 3, 613 -- 670. [29] K. Yamada, Blowing-ups describing the polarization change of moduli schemes of semistable sheaves of general rank, preprint, arXiv:0704.2870. Research Institute for Mathematical Sciences, Kyoto University, Kyoto 606-8502, Japan E-mail address: [email protected] Department of Mathematics, Faculty of Science, Kobe University, Kobe 657-8501, Japan Max-Planck-Institut fur Mathematik, Vivatsgasse 7 53111 Bonn, Germany E-mail address: [email protected]
1103.3930	3	1103	2013-09-09T19:28:33	Complements and higher resonance varieties of hyperplane arrangements	[ "math.AG" ]	Hyperplane arrangements form the geometric counterpart of combinatorial objects such as matroids. The shape of the sequence of Betti numbers of the complement of a hyperplane arrangement is of particular interest in combinatorics, where they are known, up to a sign, as Whitney numbers of the first kind, and appear as the coefficients of chromatic, or characteristic, polynomials. We show that certain combinations, some nonlinear, of these Betti numbers satisfy Schur positivity. At the same time, we study the higher degree resonance varieties of the arrangement. We draw some consequences, using homological algebra results and vector bundles techniques, of the fact that all resonance varieties are determinantal.	math.AG	math	COMPLEMENTS AND HIGHER RESONANCE VARIETIES OF HYPERPLANE ARRANGEMENTS NERO BUDUR Abstract. Hyperplane arrangements form the geometric coun- terpart of combinatorial objects such as matroids. The shape of the sequence of Betti numbers of the complement of a hyperplane arrangement is of particular interest in combinatorics, where they are known, up to a sign, as Whitney numbers of the first kind, and appear as the coefficients of chromatic, or characteristic, poly- nomials. We show that certain combinations, some nonlinear, of these Betti numbers satisfy Schur positivity. At the same time, we study the higher degree resonance varieties of the arrangement. We draw some consequences, using homological algebra results and vector bundles techniques, of the fact that all resonance varieties are determinantal. 1. Introduction Hyperplane arrangements form the geometric counterpart of combi- natorial objects such as matroids. The shape of the sequence of Betti numbers of the complement of a hyperplane arrangement is of partic- ular interest in combinatorics, where they are known, up to a sign, as Whitney numbers of the first kind, and appear as the coefficients of chromatic, or characteristic, polynomials. Using this equivalent termi- nology, lower bounds have been determined by Dowling-Wilson [9] and improved for connected matroids by Brylawski [1]. Recently, using sin- gularity theory, Huh [19] proved that the sequence of Betti numbers is log concave. A more general question is what other polynomials in the Betti numbers satisfy positivity. We show that certain combinations, some nonlinear, of Betti numbers satisfy Schur positivity, see Theorem 1.6. At the same time, we study the higher degree resonance varieties R i j of the arrangement. These combinatorial invariants were first intro- duced by Falk [14] and are related to the cohomology of local systems 2010 Mathematics Subject Classification. 32S22; 52C35. Key words and phrases. hyperplane arrangements; connected matroids; Betti numbers; chromatic and characteristic polynomials; resonance varieties. This work was partially supported by NSA grant H98230-11-1-0169. 1 on the complement [22] and, conjecturally, to the lower central series and Chen ranks of the fundamental group of the complement [30]. Res- onance varieties are also connected with the critical points of master functions [3, 4], with the Bethe ansatz equations for Gaudin models of complex simple Lie algebras [32], and are key objects in the conjectured combinatorial invariance of characteristic varieties and of Milnor fiber [23]. The varieties R1 cohomology of hyperplane arrangements, e.g. j have been studied in detail by many people, e.g. Libgober-Yuzvinsky [24], Falk-Yuzvinsky [15], etc. In contrast, our knowledge of the higher degree R i j are deter- minantal some results complementing the existing ones on resonance varieties. Thus the main contribution of this note is to point out the usefulness to the, already diverse, hyperplane arrangement theory of some homological algebra and vector bundles results. That R i 1 admit equations in terms of minors of matrices is also contained in Denham- Schenck [7], an unpublished preprint that was brought to our attention by the authors. See also loc. cit. for an interpretation of R i 1 in terms of Ext modules and for the role played by the double Ext spectral sequence. j is very limited. We derive from the fact that all R i To present the results, let D be a hyperplane arrangement of degree d in Cn. We assume that D is central, essential, and indecomposable; see 2.1 for definitions. Let U = Pn−1 − P(D). We denote by H i(U) the complex cohomology group H i(U, C). The cohomology ring of U is a combinatorial invariant of the hyperplane arrangement D [28]. Let bi : = bi(U) = dimC H i(U) βi : = bi − bi−1 + . . . + (−1)ib0. Thus βn−1 = χ(U) is the Crapo invariant of the matroid of D, and βi is the Crapo invariant of a truncation this matroid. Corollary 2.3 gives lower bounds on bi and βi following [9, 1]. In Proposition 3.2, although we show bounds in general weaker than the ones of Corollary 2.3 as long as d is not too small compared with n, we derive them from general algebraic results. Let P = P(H 1(U)). By [12], we have a linear locally free resolution (1) 0 −→ OP(−n + 1) ⊗ H 0(U) φ0−→OP(−n + 2) ⊗ H 1(U) φ1−→ . . . . . . φn−2−→ OP ⊗ H n−1(U) −→ F −→ 0 . 2 Here F is the sheaf version of the singular module of the arrangement as defined by Eisenbud-Popescu-Yuzvinksy using the BGG-correspondence [12]. An element v ∈ H 1(U) defines a complex (H •(U), v ∪ .) via the cup product. Define the resonance varieties of U to be j (U) := {v ∈ H 1(U) dim H i(H •(U), v ∪ .) ≥ j}. R i The resonance varieties of U are combinatorial invariants of the hyper- plane arrangement D. We will use the notation Ri(U) for R i 1(U). We have Ri(U) = R i 3(U) ⊃ . . .. It is known that Ri(U) are union of vector subspaces of H 1(U), see [6, 5, 12]. 2(U) ⊃ R i 1(U) ⊃ R i Define the Fitting ideal Ik(φi) to be the ideal generated by the k- minors of the matrix of linear forms representing φi in (1). Matei-Suciu [26] have shown that R1 j (U) admit equations in terms of minors of the linearized Alexander matrix, which is the same as φ1. This can be generalized, see also Denham-Schenck [7]-Proposition 2.9 for R i j (U)) is the support of the ideal Iβi+1−j(φi). j (U) ≤ min{d − 1, (βi−1 + j)(βi+1 + j)}. Corollary 1.2. codim R i Theorem 1.1. P(R i 1(U). When the above inequality on the codimension of R i j (U) is useful, we can say something stronger about R i j (U). Note that in contrast with the conclusion of the next result, it is known that the irreducible components of P(R 1 j (U)) are mutually disjoint [24], see also [8]. Corollary 1.3. If (βi−1 + j)(βi+1 + j) < d − 2, then P(R i connected. j (U)) is It is known that resonance propagates, that is, Ri(U) ⊂ Ri+1(U), [12]. We show that a deeper propagation holds: Corollary 1.4. We have: • R i(U) ⊂ R i+1 • R i(U) ⊂ R i+1 2 j (U) (i ≤ n − 2), (U) (i < n − 2 and j ≤ 1 + n−3 i+1 ). Corollary 1.2 can be improved when j = 1. Define from now qi := codim Ri(U) = d − 1 − dim Ri(U) for 0 ≤ i ≤ n − 1. Theorem 1.5. n − 1 − i ≤ qi ≤ min{d − 1, (βi−1 + 1)(βi+1 + 1), βi+1 + i + 1}. 3 The computational complexity of the resonance varieties is discussed briefly in section 5.1, see for example Proposition 5.2. Regarding the shape of the Betti numbers of the complement, using a vector bundle method of Popa-Lazarsfeld [20, 25] we obtain Schur positivity of certain combinations of the Betti numbers of U. Define for j > 0 j+1 c(j) t := (1 − kt)(−1)k bj+1−k . Yk=1 Let c(j) i denote the coefficient of ti in c(j) t . Theorem 1.6. Let 0 < j < n − 1. If j = n− 2, assume that qn−2 > 1. Then: qj−1 is non- (a) Any Schur polynomial of weight < qj negative. In particular, c(j) 1 , . . . , c(j) in c(j) i ≥ 0 for 1 ≤ i < qj, and Xk=1 (−1)k+1 · k · bj+1−k ≥ 0. j+1 (2) c(j) 1 = i = 0 for i > min{βj+1, qj − 1}. (b) c(j) (c) qj > max{i c(j) (d) The coefficients c(j) quence. i i 6= 0}. of the polynomial c(j) t form a log concave se- Recall that the first few Schur polynomials are: c1; c2, c2 c1c2−c3, c3 can determined from Theorem 1.6, see 4.3. 1 − c2; c3, 1−2c1c2+c3. Other lower bounds on bi, in terms of b1, . . . , bi−1 The outline of the article is the following. In the second section we recall the basic definitions and lower bounds on Betti numbers of U from [9, 1]. The third section is the core of the article, where we prove the statements from this Introduction. Next section is a brief discussion of Theorem 1.6. We end the article with a section containing some remarks about resonance varieties. I would like to thank A. Dimca, M. Falk, A. Suciu, and S. Yuzvin- sky for comments and suggestions, to G. Denham and H. Schenck for sharing their preprint [7], and to the Max Planck Institute in Bonn and the Johns Hopkins University for their hospitality during the writing of the article. 4 2. Preliminaries 2.1. Hyperplane arrangements. Let us recall the basic terminology. An affine (resp. projective) hyperplane arrangement in Cn (resp. Pn−1) is a finite set of hyperplanes. We will abuse notation and identify the hyperplane arrangement with the corresponding reduced divisor. An arrangement is essential if the intersection of all hyperplanes has di- mension at most zero. An arrangement is central if the intersection of all hyperplanes is nonempty. Non-essential implies central. An ar- rangement is indecomposable if it is not the product of two distinct hyperplane arrangements. In other words, there is no choice of coor- dinates for which the equation of the arrangement is a product of two non-constant polynomials in two disjoint sets of variables. An affine central arrangement D will tacitly be assumed to contain the origin in any of its hyperplanes. For a hyperplane arrangement D and a linear subspace S of the ambient space, we will denote by DS the hyperplane arrangement D ∩ S in S. For every affine hyperplane arrangement D in Cn we will consider, initially, the following sets of numbers: hi, bi, βi. These are as follows: hi = dim H i(Cn − D, C), bi = hi − hi−1 + . . . + (−1)ih0, βi = bi − bi−1 + . . . + (−1)ib0 = hi − 2hi−1 + 3hi−2 − . . . + (−1)ih0. When D is central, the numbers hi are also known as the absolute values of the Whitney numbers of the first kind, and the number βn−1 is commonly called the Crapo invariant. 2.2. Central affine arrangements. Define and P B(d, n, i) := (cid:18)n P DW (d, n, i) := (cid:18)n i − 1(cid:19), i(cid:19) + (d − n)(cid:18)n − 1 i − 1(cid:19) − δi,n−1. i(cid:19) + (d − n)(cid:18) n Theorem 2.1. (Dowling-Wilson [9]) Let D be a central essential hyperplane arrangement in Cn of degree d. Then hi ≥ P DW (d, n, i). In the indecomposable case, we have the following improvement. Let Ω := {(x, y) ∈ N2 x − y ≥ 2, (x, y) 6= (7, 4)}. 5 Theorem 2.2. (Brylawski [1]) Let D be an indecomposable central essential hyperplane arrangement in Cn of degree d. If (d, n) ∈ Ω then if i < n, and hi ≥ P B(d, n, i) βn−1 ≥ max{1, d + 2 − 2n}. For examples when the bounds are achieved see loc. cit. Let D be a central hyperplane arrangement in Cn. Then Indeed, bi = H i(Pn−1 − P(D), C). π(D, t) = (1 + t)π(P(D), t), where we denote by π(Z, t) the Poincar´e polynomial of the complement of Z, [27]. Note: hi = bi + bi−1, βi = bi − bi−1 + . . . + (−1)ib0, bi = hi − hi−1 + . . . + (−1)ih0 = βi + βi−1. Let S be a generic subvector space of Cn of dimension s + 1. Then bi(D) = bi(DS) for 0 ≤ i ≤ s, by the combinatorial invariance of bi, [28]. Hence also hi(D) = hi(DS), βi(D) = βi(DS), for 0 ≤ i ≤ s. If D is indecomposable, then so is DS by Corollary 3.1. If D is essential, so is DS. Applying Theorem 2.2, we obtain: Corollary 2.3. Let D be a central essential indecomposable hyperplane arrangement in Cn of degree d. For 0 < i < n, hi ≥ max{P B(d, s + 1, i) i ≤ s ≤ n − 1, (d, s + 1) ∈ Ω}; if (d, i + 1) ∈ Ω, then βi ≥ max{1, d − 2i}; and if in addition (d, i) ∈ Ω, then bi ≥ 2(d − 2i − 1). 6 Remark 2.4. Lower bounds on hi = bi + bi−1 do not automatically translate into lower bounds for bi, unless the following is true: the Betti numbers of the complement of an affine arrangement of degree d− 1 in Cn−1 are the first n − 1 Betti numbers of the complement of a central affine arrangement of degree d−1 in some Cn′ with n′ ≥ n. Here n′ = n iff the hyperplane at infinity is in general position. 3. Proofs 3.1. Lower bounds. Let us point out how homological algebra results about linear free resolutions imply lower bounds on hi, bi, βi. These bounds are in general weaker than the lower bounds of Corollary 2.3, as long as d is not too small compared with n. We start with a different proof of the following result. Note that it is well-known that the positivity of the Crapo invariant βn−1 is equivalent to the indecomposability of the central arrangement. Proposition 3.1. Let D be a central essential indecomposable hyper- plane arrangement in Cn. Then βi > 0 for 0 ≤ i < n. Proof. We can assume 0 < i < n. Let F and φi be as in (1). Then βi = rank φi. So βi ≥ 0 since the rank is a nonnegative number. If βi = 0 then φi = 0. This is a contradiction. Indeed, by the definition of φi in [12], the condition that φi = 0 implies that H k(U) = 0 for k > i. In particular, βn−1 = 0 which contradicts the indecomposability of the arrangement. Note that since the entries of the matrix representing φi are linear forms and not all vanishing, a Nakayama Lemma argument implies that (1) is the minimal locally free resolution of F , see [11]- Lemma 19.4. (cid:3) Proposition 3.2. (a) bi ≥ (cid:0)n−1 i (cid:1), hence hi ≥ (cid:0)n i(cid:1). (b) βi ≥ n − 1 − i for i > 0. Proof. By the proof of Proposition 3.1, the projective dimension of F is pd(F ) = n − 1. A conjecture of Buchsbaum-Eisenbud and Hor- rocks about Betti numbers of minimal free resolutions implies that bi ≥ (cid:0)pd(F ) i (cid:1). This conjecture is proved for graded modules with a lin- ear minimal free resolution, which is our case, by Herzog-Kuhl [18]. This is part (a). Part (b) is the lower bound for syzygy modules of a module with projective dimension n− 1 due to Evans-Griffith [13]. (cid:3) 3.2. Resonance, singular modules, and truncations. Let D be a central indecomposable hyperplane arrangement in Cn. Consider the 7 locally free resolution (1) of the singular module F of D. Denote by Ki the cokernel of φi: φi−→OP(−n + 2 + i) ⊗ H i+1(U) −→ Ki −→ 0. OP(−n + 1 + i) ⊗ H i(U) Proposition 3.3. The complex (1), without the last term F , is the complex of sheaves on P obtained from the complexes of vector spaces (H •(U), v ∪ .) when v varies along P. Proof. Let S = C[x1, . . . , xb1]. The complex of sheaves on P obtained by varying v ∈ P comes from the complex of free S-modules with maps S ⊗ H i(U) −→ S ⊗ H i+1(U) given by 1⊗ w 7→ P xi⊗ (ei∪ w), where ei is a basis of H 1(U) and xi is the dual basis of H 1(U)∨. These are exactly the maps of the complex of S-modules corresponding to (1) by [12]-3. (cid:3) Let S be a generic subvector space of Cn of dimension s + 1. By Proposition 3.1, the hyperplane arrangement DS is also indecompos- able. Since H i(U) = H i(U∩P(S)) for 0 ≤ i ≤ s, the resonance varieties also agree up to degree s: Ri(U) = Ri(U ∩ P(S)), for 0 ≤ i ≤ s. The singular modules of D and DS are related as follows. To denote the dependance on D, we will briefly use the notation FD for F . Lemma 3.4. The singular module of DS is FDS = OP(n−1−s)⊗Ks−1. Proof. The locally free resolution (1) of FD is minimal, as in the proof of Proposition 3.1. Thus the complex 0 → OP(−s) ⊗ H 0(U) → . . . → OP ⊗ H s(U), formed from the truncation of the complex (1) after twisting by OP(n− 1 − s), is the minimal locally free resolution of OP(n − 1 − s) ⊗ Ks−1. However, by Proposition 3.3, the minimal locally free resolution of FDS has the same shape due to the invariance of the cup product maps up to degree s. Hence FDS = OP(n − 1 − s) ⊗ Ks−1. (cid:3) 3.3. The resonance varieties Ri(U). Before discussing the refined resonance varieties R i j (U), we focus on the case j = 1. Let I(φ) denote the ideal Iβi(φi) given by the βi-minors of φi. We prove first a particular case of Theorem 1.1. Proposition 3.5. P(Ri(U)) is the support of the Fitting ideal I(φi). 8 Proof. Let Ki := coker φi. We need to show that P(Ri(U)) is the locus of points where Ki fails to be locally free. Indeed, by [11]-20.6, the non-locally free locus of Ki is the support of the Fitting ideal I(φi). Let v ∈ P = P(H 1(U)) and let κ(v) be the residue field of v. By [12]-Theorem 4.1 (a), H n−1−i(H •(U), v ∪ .) = TorOP,v i (κ(v),Fv) In particular, we have H n−2(H •(U), v ∪ .) = TorOP,v (κ(v),Fv). By [11]-Ex. 6.2 (a), the space on the right is zero iff Fv is free. Hence v ∈ P(Rn−2(U)) iff Kn−2 = Fv is not free. To prove the claim for i < n − 2, we reduce to the above case by truncating and using Lemma 3.4. In this case we have (κ(v), (Ki)v). (3) Hence, as above, P(Ri(U)) is the locus of points where Ki fails to be locally free. H i(H •(U), v ∪ .) = TorOP,v (cid:3) 1 1 Remark 3.6. Note that propagation of resonance, i.e. the fact that Ri(U) ⊂ Ri+1(U), follows from [11]-20.12 where it is shown that the support of I(φi) is included the support of I(φi+1). Proof of Theorem 1.5. Let us prove the first inequality. By [11]-20.9, we have that depth I(φi) ≥ n − 1 − i. Since S = C[x1, . . . , xb1] is Cohen-Macaulay, depth equals codimension. Now the claim follows by Proposition 3.5. It is known that the depth of Fitting ideals is bounded above by that of generic determinantal varieties. More precisely, for a map f : P → Q of projective modules, we have by [10] that (4) depth Ik(f ) ≤ (rank(P ) − k + 1)(rank(Q) − k + 1), where Ik(f ) is the ideal generated by the k-minors of the matrix rep- resenting f . In our case, qi = depthI(φi) ≤ (bi − βi + 1)(bi+1 − βi + 1) = = (βi−1 + 1)(βi+1 + 1). The remaining inequality qi ≤ βi+1 + i − 1 is part (d) of Theorem 3.7 below. (cid:3) Theorem 1.6 consists of the parts (a)-(c) and (e) in: Theorem 3.7. Let 0 < j < n − 1. If j = n− 2, assume that qn−2 > 1. Then: qj−1 is non- (a) Any Schur polynomial of weight < qj negative. 1 , . . . , c(j) in c(j) 9 i = 0 if βj+1 < i < qj. 6= 0}. (b) c(j) (c) qj > max{i c(j) (d) βj+1 ≥ qj − j − 1 . (e) The coefficients c(j) quence. i i of the polynomial c(j) t form a log concave se- Let W be a vector subspace of H 1(U) that is transversal to Rn−2 Proof. The proof is a combination of [12] and [20]. Let us prove first the case when j = n − 2. Denote qn−2 by q, and βn−1 by β. (U). Then P′ := P(W ) has dimension q − 1 ≥ 1. Restrict the linear locally free resolution of F to P′. Then we have a linear locally free resolution (5) 0 → OP′(−n + 1) ⊗ H 0(U) → OP′(−n + 2) ⊗ H 1(U) → . . . . . . → OP′ ⊗ H n−1(U) → F ′ → 0 , 1 It follows that c(n−2) where F ′ is a vector bundle on P′. is the Chern polynomial of F ′. Since F ′ is globally generated, the Chern classes ci(F ′), which equal c(n−2) , and the Schur polynomials in these, are nonnegative, [16]-12.1.7-(a). This proves (a). t i 6= 0. As in [20], the parts (b) and (c) follow from the fact that rank(F ′) = β, that ci(F ′) = 0 for i > max{rank(F ′), q − 1}, and that there exist an i such that ci(F ′) 6= 0. The proof of part (d) is essentially the same as the one in [20]. Since β > 0, we can assume q > n − 1. If q = n then we need to show If F ′ = 0 then (5) cannot be an exact sequence, as that F ′ the alternating product of Chern polynomials cannot be 1. So we can assume that q > n. Chasing through (5) we have that H j(P′,F ′(k)) = 0 for all k and 0 < j < q−n+1. The splitting criterion of Evans-Griffith, see [21]-3.2.12, implies, if rank(F ′) ≤ q−n+1, that F ′ splits as a direct sum of line bundles. This cannot happen for the same reason as before. Hence rank(F ′) > q − n + 1. from (1) by truncation: 0 −→ OP(−n + 1)⊕b0 −→ OP(−n + 2)⊕b1 −→ . . . . . . −→ OP(−n + 1 + j)⊕bj −→ OP(−n + j + 2)⊕bj+1 −→ Kj −→ 0 . We tensor this complex with OP(n − j − 2) to obtain a linear locally free resolution Now we prove the case j < n − 2. Consider the complex obtained 0 −→ OP(−1 − j)⊕b0 −→ OP(−j)⊕b1 −→ . . . . . . −→ OP(−1)⊕bj −→ O⊕bj+1 P −→ Kj(n − j − 2) −→ 0 . 10 By part (c), c(j) Now Kj(n− j − 2) is globally generated, and the rest of the proof goes as for the case j = n − 2. is a polynomial which has all coefficients and all roots real. An old theorem of Newton, see [29]-Theorem 2, implies that the coefficients must form a log concave sequence. This shows part (e). (cid:3) t Question 3.8. Is qn−2 > 1 always true ? We note that this is true for n = 3: the non-local components of R1(U) have small dimension [33], and the local components have codi- mension ≥ 2 by the indecomposability of D, cf. [15]. We give an algebraic reformulation of this question and a partial answer in section 5.2 . 3.4. The refined resonance varieties R i remaining statements from Introduction. j (U). We prove now the Proof of Theorem 1.1. By (3) and Lemma 3.4, it is enough to prove the case i = n − 2. Hence, we need to show that the locus of points v such that dim TorOP,v (κ(v),Fv) ≥ j is the support of Iβn−2+1−j(φn−2). This ideal is actually an invariant of F , by [11]-Corollary/Definition 20.4. In the notation of loc. cit., 1 Iβn−2+1−j(φn−2) = Fittβn−1−1+j(F ), since rankF = βn−1. By [11]-Proposition 20.6, the support of the Fitting ideal Fittβn−1−1+j(F ) is the locus of points v where Fv cannot be generated by βn−1 − 1 + j elements. To summarize, it is enough to show: if (R, P ) is a local noetherian domain, and M is an R-module minimally generated by k elements, then k − rank(M) ≤ dim TorR 1 (R/P, M). Here rank(M) is the dimension of the K-vector space M ⊗R K, where K is the quotient field of R. To prove this claim, consider a minimal set of generators of M and the short exact sequence attached to them: (6) Since the operation . ⊗R K is exact, we have 0 → N → Rk → M → 0. k − rank(M) = rank(N). By a similar reasoning, rank(N) ≤ k′, where k′ is the minimal number of generators of N. Tensoring (6) with R/P we have an exact sequence 11 of R/P -vector spaces 0 → TorR 1 (R/P, M) → N ⊗R R/P → (R/P )k → M ⊗R R/P → 0. By Nakayama Lemma and the minimality of k, the two vector spaces on the right are isomorphic. Hence, also the two vector spaces on the left are isomorphic, and k′ = dim TorR 1 (R/P, M). (cid:3) Proof of Corollary 1.2. It follows from Theorem 1.1 and (4). (cid:3) Proof of Corollary 1.3. By a theorem of Fulton-Lazarsfeld [21] -7.2, the support of the ideal sheaf Ik+1(φ) is connected if φ : E → E ′ is a vector bundle map on a variety X such that E ∨ ⊗ E ′ is ample and dim(X) > (rank(E) − k)(rank(E ′) − k). We apply this to Iβi+1−j(φi) and X = Pd−2. Note that (OP(−n + 1 + i)⊕bi)∨ ⊗ O(−n + 2 + i)⊕bi+1 = OP(1)⊕bibi+1 is an ample vector bundle since it is the direct sum of ample line bun- dles. (cid:3) Proof of Corollary 1.4. Because of Theorem 1.1, we can apply to the refined resonance varieties facts about minors of matrices in a finite free resolution. For example, the second claim follows directly from Buchsbaum-Eisenbud [2]- Corollary 6.2 together with [2]-Theorem 3.1, part (b). The first claim follows from [2]-(10.5) if the Conjecture 10.1 in [2] holds. This conjecture is proven by Tchernev-Weyman [31]. (cid:3) 4. Dissection of Theorem 1.6 4.1. Linear combinations. We note that c(i) 1 = βi − βi−1 + βi−2 − . . . = bi − 2bi−1 + 3bi−2 − . . . = hi − (1 + 2)hi−1 + (1 + 2 + 3)hi−2 − (1 + 2 + 3 + 4)hi−3 + . . . . 1 give lower , lower bounds on the numbers c(i) 1 + c(i−1) Since βi = c(i) bounds on βi. 1 4.2. Nonlinear combinations. The numbers c(i) for i ≥ 2, and the higher degree Schur polynomials, are nonlinear combinations of hi, bi, βi, c(i) 1 . Denote for simplicity j ai = c(i) 1 . 12 We have for example: c(2) 2 ≥ 0 ⇐⇒ a2 [c(2) 1 ]2 − c(2) 2 ≥ 0 ⇐⇒ a2 c(3) 2 ≥ 0 ⇐⇒ a2 c(3) 3 ≥ 0 ⇐⇒ a2 + 3a2 3 ≥ 0 ⇐⇒ a3 3 ≥ 0 ⇐⇒ a3 c(3) 1 c(3) [c(3) 1 ]3 − 2c(3) 1 c(3) 2 − c(3) 2 + c(3) 2 + 2a0 − 2a1 + a2 ≥ 0 2 − 2a0 + 2a1 − a2 ≥ 0 3 − 2a0 + 2a1 − 2a2 + a3 ≥ 0 3 + 6(a1a3 − a0a3 + a2a3)+ 3 − 36a0 + 24a1 − 12a2 + 2a3 ≥ 0 3 + 18a0 − 12a1 + 6a2 − a3 ≥ 0 3 + 6(a0a3 − a1a3 + a2a3) − 3a2 3− − 36a0 + 24a1 − 12a2 + 2a3 ≥ 0. 4.3. Lower bounds on bi. Although we regard the positivity prop- erties from Theorem 1.6 on various combinations of the numbers bi as mainly telling us something about the shape of the sequence bi, we can also use these inequalities to derive lower bounds for bi. Let us do so for the first few ones. For n = 4 we note that if q2 > 1 then c(2) 1 ≥ 0 gives b2 ≥ 2d − 5. If q2 > 2, then solving the quadratic equation c(2) we deduce the stronger inequality 2 ≥ 0, as in [20]-3.4, (7) b2 ≥ 2d + √8d − 31 2 11 2 . − We note that expression under the square root is always positive, since d ≥ 5 by the irreducibility assumption on D. We have an equality in (7) when c(2) 2 = 0, which by Theorem 1.6-(b) is guaranteed when β3 = 1. This is the case of a generic arrangement with d = 5. For n = 5 we note that c(2) 1 ≥ 0 gives b2 ≥ 2d − 5. If q2 > 2, then solving the quadratic equation c(2) stronger inequality 2 ≥ 0 we deduce the b2 ≥ 2d + √8d − 31 2 11 2 . − Note that these are the same inequalities as in n = 4 case. The in- equality c(3) 1 ≥ 0 gives, if q3 > 1, b3 > 2b2 − 3d + 7. 13 Solving for the quadratic equation c(3) the stronger inequality 2 ≥ 0 gives as in [20], if q3 > 2, √73 − 24d + 8b2 . (8) b3 ≥ 13 2 − 3d + 2b2 + 2 This is available as long as the expression under the square root is nonnegative, that is if b2 ≥ 3d − 9. We note that we get an equality in (8) if c(3) 2 = 0, which by Theorem 1.6-(b) is guaranteed when β4 = 1. This is the case of the generic arrangement with d = 6. Remark 4.1. From the above computations we note the following trend: the inequality c(j) i+1 ≥ 0 is better than c(j) i ≥ 0. Example 4.2. Let D be a generic central hyperplane arrangement of degree d in Cn, with d > n. Then i (cid:19), bi = (cid:18)d − 1 i (cid:19), βi = (cid:18)d − 2 i (cid:19), 1 = (cid:18)d − 3 c(i) if 0 ≤ i ≤ n − 1, otherwise all three numbers are zero. Also, c(i) when d = j + 3 and n = j + 2. 2 = 0 5. Remarks on resonance varieties 5.1. Explicit equations for R1. There are well-known explicit bases for the vector spaces H i(U) such as the "no broken circuits" sets, [27]. Hence the matrices representing φi, and thus by Theorem 1.1, the equa- tions for the resonance varieties R i j (U) are very explicit. Example 5.1. Let us spell out how to obtain R1 for the case when the hyperplane arrangement D is the cone over a planar line arrangement {H1, . . . , Hd} with at most triple points. Each line Hj with j 6= d defines a basis element of H 1(U), and of its dual, i.e. a generator xj of the symmetric algebra S = C[x1, . . . , xd−1]. For each intersection point P of two lines from {H1, . . . , Hd−1}, define iP := max{i < d P ∈ Hi}. Consider the pairs (i, iP ) with i < iP and P ∈ Hi. These pairs index a basis of H 2(U). The b2 × b1 matrix M = (M(i,iP ),j) representing the map φ1 has entries M(i,iP ),j =   0 −xiP xi 14 if j 6∈ {i, iP}, if j = i, if j = iP , if P is a multiplicity-two point of {H1, . . . , Hd−1}, and if j 6∈ {i, k, iP}, if j = iP , if j = i, if j = k, M(i,iP ),j = 0 xi −xiP − xk −xiP + xi   if P is a multiplicity-three point of {H1, . . . , Hd−1} and {Hi, Hk, HiP} are the three lines passing through P . The equations of R1 in S are the (d − 2)-minors of the matrix M. Coming back to the general case, note that R1(U) is the common β1(cid:1) ·(cid:0)b2 zero locus of (cid:0)b1 sition states that only (cid:0)b2 β1(cid:1) polynomials, the β1-minors of φ1. Next propo- β1(cid:1) of these polynomials are necessary. To our knowledge this is currently the best reduction of the computational complexity of R1(U). Proposition 5.2. Let 1 ≤ i ≤ b1. Then R1(U) is the common zero locus of x−1 i m, where m ranges over the maximal minors of the b2× β1- matrix obtained from φ1 by removing the i-th column. β1(cid:1)-matrix such that [x1 . . . xb1]t · A represents the map Vβ1 φ∗ Proof. By [2]-Theorem 3.1, the map Vβ1 φ∗ the 1×(cid:0)b2 1 factors through the map φ0. The map φ0 is represented by the matrix [x1 . . . xb1]t. Let A be 1. Then the entries of A are the elements x−1 i m, with m as above. By loc. cit., part (b), the common vanishing locus of the entries of A is the same as the support of the Fitting ideal I(φ1). By Corollary 3.5, the support of I(φ1) is R1(U). (cid:3) A similar reduction of the number of necessary polynomials to define the higher resonance varieties Ri(U) is available following [2]. 5.2. Reformulation of Question 3.8. Let I = I(φn−2) ⊂ S be the homogeneous ideal of Rn−2 from Corollary 3.5. That is I is the ideal generated by the βn−2-minors of φn−2. Let q = qn−2 be the codimension of I. The following gives an algebraic reformulation of Question 3.8. Proposition 5.3. q > 1 iff for every linear form f and homogeneous element m ∈ S, f m ∈ I implies m ∈ I. Proof. We have that q > 1 iff every associated prime ideal P of I has codimension > 1. Since I is homogeneous, every associated prime ideals of I is also homogeneous. By definition, P is an associated prime ideal of I if it annihilates a nonzero element m + I of S/I. This is equivalent to annihilating a homogeneous element, [11]-3.12. Since 15 S is a polynomial ring, a homogeneous prime ideal P has codimension 1 iff P is generated by a linear form f . (cid:3) We note that Question 3.8 has a positive answer if D is itself the generic central hyperplane section of an indecomposable central essen- tial hyperplane arrangement in a higher number of variables, by Lemma 3.4 and the first inequality of Theorem 1.5, j (U) are supported on finitely many linear subspaces: 5.3. Structure of R i j . It is known that the refined resonance varieties R i [5, 6] using Hodge theory, [22] using deformations, [24] using linear algebra. Let us mention a few facts that follow from Theorem 1.1. The refined resonance varieties R i j are finite intersections of varieties swept by vector spaces of particular type. More precisely, for a p × q matrix of linear forms with p ≤ q, the support of the locus defined by the vanishing of the p-minors is swept by the linear spaces determined by the simultaneous vanishing of the coefficients of a linear combination of rows (or of columns), [11]-Exercise A2.19. On a different note, let us see next what it means to prove, using only linear algebra, that R1 j are supported on linear subspaces. Fix a basis e1, . . . , ep of Cp. Let M(x) be a matrix of size q × p, p ≤ q, with entries linear in C[x1, . . . , xp] = C[x], such that M(a)b = −M(a)b for any complex vectors a and b of size p, and rank(M(a)) = p − 1 for generic vectors a. Define R1 j to be the common zero locus of the (p− j)-minors of M(x). For example, if M(x) is the matrix representing φ1 in (1), then R1 Lemma 5.4. Let a ∈ R1 j+1. Let N be a submatrix of M(a) of size (p− j− 1)× p that contains a (p− j − 1)× (p− j − 1) submatrix N0 with nonzero determinant. Let b1, . . . , bj+1 be the vectors obtained as the (p − j)-minors, containing N0, of the (p − j) × p matrix obtained from concatenating vertically the matrix (e1 . . . ep) with N. Then ker M(a) is a dimension j + 1 linear subspace with basis b1, . . . , bj+1. j is the refined resonance variety by Theorem 1.1. j − R1 Proof. We prove first that b1, . . . , bj+1 are nonzero and linearly inde- pendent. Let k ∈ {1, . . . j + 1}. Since the minor corresponding to N0 is nonzero, there exists a coordinate eik with nonzero entry in bk. The set {ik′ k′ ∈ {1, . . . j + 1}} corresponds to the columns of N not contributing to the minor N0. Hence the ik′-th coordinate of bk, with k′ 6= k, is zero. This shows that b1, . . . , bj+1 are linearly independent. j+1, the codimension of ker M(a) is p− j − 1. Thus, we only need to show now that M(a)bk = 0 for all k ∈ {1, . . . j + 1}. Consider the i-th entry (M(a)bk)i of M(a)bk. This entry is, by the Since a ∈ R1 j − R1 16 definition of bk, either a (p − j)-minor of M(a), or the determinant of the vertical concatenation of a (p − j − 1) × (p − j) submatrix of N containing N0 with one of its rows. While in the second case (M(a)bk)i vanishes because of the repeated row, in the first case it vanishes be- cause the rank of M(a) is p − j − 1. (cid:3) Given the explicit description of generators of the spaces ker M(a), one would like then to prove, via linear algebra, that there are only finitely many such ker M(a) for a ∈ R1 j+1, and they form the support of R1 j+1. This is done for hyperplane arrangements in [24]-Corollary 3.7. Outside this case, it is not clear how to characterize the class of matrices M(x) having these properties. j − R1 j − R1 5.4. Other possible bounds. As in [20], one can try to obtain bounds on Betti numbers of U by displaying certain subspaces of H i(U) and counting their dimension. Note that ΛiH 1(U) → H i(U) is not injec- tive on decomposable forms, as it is known that there are monomials that vanish in the Orlik-Solomon algebra. However, we can ask the following. For each 0 ≤ i < n− 1 let Wi be a vector subspace of H 1(U) transversal to Ri(U), such that H 1(U) = W0 ⊃ . . . ⊃ Wn−2. Define for 0 < i < n Σi := {v1 ∧ . . . ∧ vi ∈ ΛiH 1(U) dim Span(v1, . . . , vi) = i and vj ∈ Wj−1}. Question 5.5. Is the natural map Σi → H i(U) an injection ? We note that this is true for i = 2 and that a converse holds, in a certain sense, see [14] -2.13 and 3.1. The numerical counterpart of this is: Proposition 5.6. If Question 5.5 is true then, for 0 < i < n, q0 + . . . + qi−1 < bi + i(i + 1) 2 . Proof. Consider the Grassmanian G(i, b1) of dimension i subspaces of H 1(U), with the Plucker embedding in P. Then P(Σi) is the subvariety of G(i, b1) consisting of subspaces L such that dim(L ∩ Wj) ≥ i − j for 0 ≤ j ≤ i − 1. Noting that dim Wj = qj, the codimension of this Schubert variety in G(i, b1) can be computed by a standard formula, see for example [17], and equals codim Σi = (b1 − i)i + i(i + 1) 2 17 − (q0 + . . . + qi−1). Since the dimension of G(i, b1) is i(b1 − 1), we have i(i + 1) dim Σi = (q0 + . . . + qi−1) − . 2 The claim now follows from a positive answer to Question 5.5. (cid:3) References [1] T. Brylawski, Connected matroids with the smallest Whitney numbers. Dis- crete Math. 18 (1977), no. 3, 243 -- 252. 1, 2, 4, 6 [2] D. Buchsbaum and D. Eisenbud, Some structure theorems for finite free res- olutions. Advances in Math. 12 (1974), 84 -- 139. 12, 15 [3] D. Cohen, G. Denham, M. Falk, and A. Varchenko, Critical points and resonance of hyperplane arrangements. To appear in Canad. J. Math. arXiv:0907.0896. 2 [4] D. Cohen, G. Denham, M. Falk, and A. Varchenko, Vanishing products of one-forms and critical points of master functions. arXiv:1010.3743. 2 [5] D. Cohen and P. Orlik, Arrangements and local systems, Math. Res. Lett. 7 (2000), 299 -- 316. 3, 16 [6] D. Cohen and A. Suciu, Characteristic varieties of arrangements, Math. Proc. Cambridge Phil. Soc. 127 (1999), 33 -- 53. 3, 16 [7] G. Denham and H. Schenck, The double Ext spectral sequence and rank va- rieties. Preprint. 2, 3, 4 [8] A. Dimca, S¸. Papadima, and A. Suciu, Topology and geometry of cohomology jump loci, Duke Math. J. 148 (2009), no. 3, 405 -- 457. 3 [9] T. Dowling and R.M. Wilson, The slimmest geometric lattices. Trans. Amer. Math. Soc. 196 (1974), 203 -- 215. 1, 2, 4, 5 [10] J.A. Eagon and D.G. Northcott, Ideals defined by matrices and a certain complex associated with them. Proc. Roy. Soc. Ser. A 269 (1962), 188 -- 204. 9 [11] D. Eisenbud, Commutative algebra. With a view toward algebraic geome- try. Graduate Texts in Mathematics, 150. Springer-Verlag, New York, 1995. xvi+785 pp. 7, 9, 11, 15, 16 [12] D. Eisenbud, S. Popescu, and S. Yuzvinsky, Hyperplane arrangement coho- mology and monomials in the exterior algebra. Trans. Amer. Math. Soc. 355 (2003), no. 11, 4365 -- 4383. 2, 3, 7, 8, 9, 10 [13] E.G. Evans and P. Griffith, The syzygy problem. Ann. of Math. (2) 114 (1981), no. 2, 323 -- 333. 7 [14] M. Falk, Arrangements and cohomology, Ann. Comb. 1 (1997), 135 -- 157. 1, 17 [15] M. Falk and S. Yuzvinsky, Multinets, resonance varieties, and pencils of plane curves, Compositio Math. 143 (2007), no. 4, 1069 -- 1088. 2, 11 [16] W. Fulton, Intersection theory, 2nd edition, Springer-Verlag, Berlin, (1998), xiv+470 pp. 10 [17] P. Griffiths and J. Harris, Principles of algebraic geometry. Wiley Classics Library. John Wiley & Sons, Inc., New York, 1994. xiv+813 pp. 17 [18] J. Herzog and M. Kuhl, On the Betti numbers of finite pure and linear reso- lutions. Comm. Algebra 12 (1984), no. 13-14, 1627 -- 1646. 7 [19] J. Huh, Milnor numbers of projective hypersurfaces and the chromatic poly- nomial of graphs. arXiv:1008.4749. 1 18 [20] R. Lazarsfeld and M. Popa, Derivative complex, BGG correspondence, and numerical inequalities for compact Kahler manifolds. arXiv:0907.0651. 4, 10, 13, 14, 17 [21] R. Lazarsfeld, Positivity in algebraic geometry I , Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 48, Springer-Verlag, Berlin, 2004. 10, 12 [22] A. Libgober, First order deformations for rank one local systems with a non- vanishing cohomology. Arrangements in Boston: a Conference on Hyperplane Arrangements (1999). Topology Appl. 118 (2002), no. 1-2, 159 -- 168. 2, 16 [23] A. Libgober, On combinatorial invariance of the cohomology of Milnor fiber of arrangements and Catalan equation over function field. arXiv:1011.0191. 2 [24] A. Libgober and S. Yuzvinsky, Cohomology of the Orlik -- Solomon algebras and local systems, Compositio Math. 121 (2000), 337 -- 361. 2, 3, 16, 17 [25] L. Lombardi, Inequalities for the Hodge numbers of irregular compact Kahler manifolds. arXiv:1103.1704v1. 4 [26] D. Matei and A. Suciu, Cohomology rings and nilpotent quotients of real and complex arrangements, in: Arrangements -- Tokyo 1998, pp. 185 -- 215, Adv. Stud. Pure Math., vol. 27, Math. Soc. Japan, Tokyo, 2000. 3 [27] P. Orlik and H. Terao, Arrangements of hyperplanes. Springer-Verlag, Berlin, 1992. xviii+325 pp. 6, 14 [28] P. Orlik and L. Solomon, Combinatorics and topology of complements of hy- perplanes. Invent. Math. 56 (1980), no. 2, 167 -- 189. 2, 6 [29] R. P. Stanley, Log-concave and unimodal sequences in algebra, combinatorics, and geometry, in Graph theory and Its Applications: East and West (Jinan 1986), 500 -- 535, Ann. New York Acd. Sci., 576, 1989. 11 [30] A. Suciu, Fundamental groups of line arrangements: enumerative aspects. Advances in algebraic geometry motivated by physics (Lowell, MA, 2000), 43 -- 79, Contemp. Math., 276, Amer. Math. Soc., Providence, RI, 2001. 2 [31] A. Tchernev and J. Weyman, Free resolutions for polynomial functors. J. Algebra 271 (2004), no. 1, 22 -- 64. 12 [32] A. Varchenko, Bethe ansatz for arrangements of hyperplanes and the Gaudin model. Mosc. Math. J. 6 (2006), no. 1, 195 -- 210, 223 -- 224. 2 [33] S. Yuzvinsky, A new bound on the number of special fibers in a pencil of curves. Proc. Amer. Math. Soc. 137 (2009), no. 5, 1641 -- 1648. 11 19 ERRATUM The proof of [1, Theorem 1.1, p. 8671] is incomplete: for the claimed equality of two sets, only one of the inclusions is proved. It was assumed wrongly that the other inclusion is easy. In fact, the other inclusion holds only under an additional hypothesis. We correct here Theorem 1.1 and Corollary 1.2 of [1]. The rest of the results of [1] stay the same, including the main results of the paper on deeper propagation and dimensional bounds for resonance varieties, and the positivity results on Betti numbers. Theorem 1.1. Let i ≤ n − 2. (a) P(Ri Iβi(φi). (b) P(Ri equals it away from P(Ri−1 (U)). (c) P(R1 Iβ1+1−j(φ1). (d) P(Ri P(Ri (f ) P(Ri 1(U)) is the support of the ideal j(U)) contains the support of the ideal Iβi+1−j(φi), and j (U)) is the support of the ideal 2(U)) is the support of the ideal Iβ2+1−j(φ2). (e) j(U)) is the support of the ideal Iβi+1−j(φi) if j ≤ 1 + (n − 3)/i. j(U)) is the support of the ideal Ibi+1−j(φi−1 ⊕ φi). 1 Proof. Part (a) is [1, Proposition 3.4]. The first claim of (b) is what is actually proven on p. 867 of [1]. Indeed, it follows from: if (R, P ) is a local noetherian domain, and M is an R-module minimally generated by k or more elements, then k − rank(M) ≤ dim TorR 1 (R/P, M). The proof of this statement involves the obvious fact that rank(N) ≤ k′ where N are the first syzygies of M and k′ is the minimal number of generators of N (note that there is a typo in loc. cit.). The converse of the statement holds if and only if N is free. Since N and M are the stalks of coker(φi−1) and coker(φi) in our case, by part (a) this gives the second claim in (b). Since P(R0 1(U)) = ∅, (c) follows. [1, Corollary 1.3] holds as stated due to the right inclusion in (b) and it follows from the same statements about the corresponding determinantal ideals. Thus P(Ri−1 (U)) is a subset of the support of Iβ2+1−j(φ2), and (d) follows by (b). Similarly, (e) follows from the second part of [1, Corollary 1.3] for determinantal ideals. Part (f) is well-known. (cid:3) 1 We do not know if the original statement of Theorem 1.1, that P(Ri j(U)) is the support of Iβi+1−j(φi), holds in general. Due to the right inclusion of part (b) in Theorem 1.1 above, this change affects in [1] only the statement of Corollary 1.2: Corollary 1.2. Let i ≤ n − 2. If (βi−1 + j)(βi+1 + j) < d − 2, then P(Ri j(U)) is connected whenever it is the support of Iβi+1−j(φi), or more generally, P(Ri j(U)) is connected away from the components of P(Ri−1 (U)) which are disconnected from the support of Iβi+1−j(φi). 1 1corresponding to p.11 in this arxiv version. 20 I would like to thank Botong Wang for very useful discussions. References [1] N. Budur, Complements and higher resonance varieties of hyperplane arrange- ments, Math. Res. Lett. 18 (2011), no. 5, 859 -- 873. 20 Department of Mathematics, University of Notre Dame, 255 Hurley Hall, IN 46556, USA E-mail address: [email protected] 21
1212.0521	1	1212	2012-12-03T20:22:22	Log-uniruled affine varieties without cylinder-like open subsets	[ "math.AG" ]	A classical result of Miyanishi-Sugie and Keel-McKernan asserts that for smooth affine surfaces, affine-uniruledness is equivalent to affine-ruledness, both properties being in fact equivalent to the negativity of the logarithmic Kodaira dimension. Here we show in contrast that starting from dimension three, there exists smooth affine varieties which are affine-uniruled but not affine-ruled.	math.AG	math	LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS ADRIEN DUBOULOZ AND TAKASHI KISHIMOTO Abstract. A classical result of Miyanishi-Sugie and Keel-Mckernan asserts that for smooth affine surfaces, A1-uniruledness is equivalent to A1-ruledness, both properties being in fact equivalent to the negativity of the logarithmic Kodaira dimension. Here we show in contrast that starting from dimension three, there exists smooth affine varieties which are A1-uniruled but not A1-ruled. Introduction 2 1 0 2 c e D 3 ] . G A h t a m [ 1 v 1 2 5 0 . 2 1 2 1 : v i X r a Complex uniruled projective varieties are nowadays considered through the Mori Minimal Model Program (MMP) as the natural generalization to higher dimensions of birationaly ruled surfaces (see e.g. [12]). In particular, as in the case of ruled surfaces, these are the varieties for which the program does not yield a minimal model, but a Mori fiber space. These varieties are also conjectured to be the natural generalization to higher dimensions of surfaces of negative Kodaira dimension, the conjecture being in fact established as long as the abundance conjecture holds true [11], hence in particular for smooth threefolds. During the past decades, the systematic study of the geometry of rational curves on these varieties has been the source of many progress in the structure theory for higher dimensional, possibly singular, projective varieties to which the MMP can be applied. The situation is much less clear for non complete varieties, in particular affine ones. The natural analogue of ruledness in this context is the notion of A1-ruledness, a variety X being called A1-ruled if it contains a Zariski dense open subset U of the form U ≃ A1 × Y for a suitable quasi-projective variety Y . A landmark result of Miyanishi-Sugie [14] asserts that a smooth affine surface is A1-ruled if and only if it has negative logarithmic Kodaira dimension [6]. For such surfaces, the projection prY : U ≃ A1 × Y → Y always extends to a fibration p : X → Z with general fibers isomorphic to the affine line A1 over an open subset of the smooth projective model of Y , providing the affine counterpart of the fact that a smooth birationaly ruled projective surface has the structure of a fibration with general fibers isomorphic to P1 over a smooth projective curve. This result, together with the description of the geometry of degenerate fibers of these fibrations, has been one of the cornerstones of the structure theory of smooth affine surfaces developed during the past decades. But in contrast, the foundations for a systematic study of A1-ruled affine threefolds have been only laid recently in [5]. On the other hand, from the point of view of logarithmic Kodaira dimension, the appropriate counterpart of the notion of uniruledness for a non necessarily complete variety X is to require that X is generically covered by images of the affine line A1, in the sense that the set of points x ∈ X with the property that there exists a non constant morphism f = fx : A1 → X such that x ∈ fx(A1) is dense in X with respect to the Zariski topology. Such varieties are called A1- uniruled, or log-uniruled after Keel and McKernan [8], and can be equivalently characterized by the property that they admit an open embedding into a complete variety X which is covered by proper rational curves meeting the boundary X \ X in at most one point. In particular, a smooth A1-uniruled quasi-projective variety X has negative logarithmic Kodaira dimension. It is conjectured that the converse holds true in any dimension, but so far, the conjecture has been only established in the case of surfaces by Keel and McKernan [8]. It follows in particular from these results that for smooth affine surfaces the notions of A1-ruledness and A1-uniruledness coincide. Pursuing further the analogy with the classical projective notions, it seems then natural to expect that these do no longer coincide for higher dimensional affine varieties. Our main result confirms that this is indeed the case. Namely, we establish the following: Theorem. For every n ≥ 3, the complement of a smooth hypersurface Qn of degree n in Pn is A1-uniruled but not A1-ruled. The anti-ampleness of the divisor KPn + Qn enables to easily deduce the A1-uniruledness of affine varieties of the form Pn \ Qn from the general log-deformation theory for rational curves developed by Keel and McKernan [8]. The failure of A1-ruledness is then obtained in a more indirect fashion. Indeed, it turns out that for the varieties under consideration, A1-ruledness is equivalent to the stronger property that they admit a non trivial action of the additive group Ga. We then exploit two deep results of projective geometry, namely the non rationality of the smooth cubic threefold in P4 in the case n = 3 and the birational super-rigidity of smooth hypersurfaces Qn ⊂ Pn if n ≥ 4, to exclude the existence of such non trivial actions. In the last section, we consider more closely the case of complements of smooth cubic surfaces in P3 which provides a good illustration of the subtle but crucial difference between the two notions of A1-ruledness and A1-uniruledness in higher dimension. We show in particular that even though such complements are not A1-ruled they admit natural fibrations by A1-ruled affine surfaces. We study automorphisms of such fibrations in relation with the problem of deciding whether every automorphism of the complement of a smooth cubic surface is induced by a linear transformation of the ambient space P3. The first author was partialy supported by ANR Grant "BirPol" ANR-11-JS01-004-01. The second author was supported by a Grant-in-Aid for Scientific Research of JSPS No. 24740003. This work was done during a stay of the first author at Saitama University and a stay of the second author at the Institut de Mathématiques de Bourgogne (Dijon). The authors thank these institutions for the hospitality. 1 LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 2 1. Recollection on affine ruled and affine uniruled varieties 1.1. Affine ruledness and algebraic Ga-actions. Here we review general properties of affine ruled varieties, with a particular focus on the interplay between A1-ruledness of an affine variety and the existence of non trivial algebraic actions of the additive group Ga on it. We refer the reader to [4] for basic properties of the correspondence between such actions on affine varieties and their algebraic counterpart, the so-called locally nilpotent derivations of their coordinate rings. Definition 1. A quasi-projective variety X is called affine ruled or A1-ruled if it contains a Zariski dense open subset U of the form U ≃ A1 × Y for a suitable quasi-projective variety Y . An open subset of this type is called an A1-cylindrical open subset of X. 1.1.1. An A1-ruled variety X is in particular ruled in the usual sense, the inclusion A1 × Y ֒→ X of an A1-cylindrical open subset of X inducing a dominant birational map P1 × Y 99K X. The converse is not true since for instance the product of the punctured affine line A1 ∗ = A1 \ {0} with a smooth projective curve of positive genus is ruled but does not contain any A1-cylinder. In particular, in contrast with ruledness, the property of being A1-ruled is not invariant under birational equivalence. Note similarly that the existence of an A1-cylindrical open subset of X is a stronger requirement than that of a dominant birational morphism A1 × Y → X for a suitable variety Y : for instance, the affine cone X ⊆ A3 over a projective plane curve C ⊆ P2 of positive genus does not contain A1-cylinders but on the other hand, blowing-up the vertex of the cone X yields a smooth quasi-projective variety σ : X → X with the structure of a locally trivial A1-bundle ρ : X → C hence, restricting to a subset Y of C over which ρ is trivial, a dominant birational morphism σ ρ−1(Y ): ρ−1(Y ) ≃ A1 × Y → X. 1.1.2. Typical examples of A1-ruled varieties are normal affine varieties X = Spec(A) admitting a nontrivial algebraic action of the additive group Ga. Indeed, if X is equipped with such an action induced by a locally nilpotent C-derivation ∂ of A, then for every local slice f ∈ Ker∂2 \ Ker∂, the principal open subset X∂f = Spec(A∂f ), where A∂f denotes the localization A[(∂f )−1] of A, is Ga-invariant and the morphism Ga × (V (f ) ∩ X∂f ) → X∂f induced by the Ga-action on X is an isomorphism. In particular, X∂f is a principal A1-cylindrical open subset of X. Note on the contrary that the existence of an A1-cylindrical open subset U ≃ A1 × Y of an affine variety X = Spec(A) is in general not enough to guarantee that X can be equipped with a Ga-action whose general orbits coincide with the general fibers of the projection prY : U ≃ A1 × Y → Y . For instance, the general fibers of the projection pr1 : X = P1 × P1 \ ∆ → P1, where ∆ ⊂ P1 × P1 denotes the diagonal, cannot coincides with general orbits of an algebraic Ga-action on X. Indeed, otherwise every invariant function on X would descend to a regular function on P1 and hence would be constant, in contradiction with the fact that for an affine variety X, the field Frac(Γ(X, OX )Ga ) has transcendence degree dim X − 1 over C. In fact, it is classically known that the general fibers of an A1-cylindrical affine open subset U ≃ A1 × Y of a normal affine variety X = Spec(A) coincide with the general orbits of an algebraic Ga-action on X if and only if U is a principal affine open subset of X (see e.g. [9, Proposition 3.1.5]). Let us briefly recall the argument for the convenience of the reader: the existence of a principal A1-cylinder in X is equivalent to the existence of an element a ∈ A \ {0} and an isomorphism of ∂t of Aa is then locally nilpotent C-algebra ϕ : Aa and since A is a finitely generated algebra, there exists n ≥ 0 such that an∂0 induces a C-derivation ∂ of A. Noting that being invertible in Aa, a is necessarily contained in the kernel of ∂0, we conclude that ∂ is a locally nilpotent derivation of A, defining a Ga-action on X whose restriction to the principal invariant open subset U = Spec(Aa) ≃ Spec(B) × A1 is by construction equivariantly isomorphic to that by translations on the second factor. → B[t], where t is transcendental over Frac(B). The derivation ∂0 = ϕ∗ ∂ ∼ It follows in particular that for a normal affine variety X whose divisor class group Cl(X) is torsion, the existence of A1-cylindrical open subsets is essentially equivalent to that of nontrivial Ga-actions. More precisely, we have the following criterion: Proposition 2. Let X = Spec(A) be a normal affine variety such that Cl(X) ⊗Z Q = 0. Then for every A1-cylindrical open subset U ≃ A1 × Y of X there exists an action of Ga on X whose general orbits coincide with the general fibers of the projection prY : U → Y . In particular, X admits an A1-cylinder open subset if and only if it admits a nontrivial Ga-action. Proof. By replacing Y by an affine open subset of it, we may assume that U is affine whence that its complement D = X \ U has pure codimension one in X as X itself is affine. The hypothesis guarantees precisely that D is the support of a principal divisor and hence the assertions follow immediately from the discussion above. (cid:3) 1.2. Basic facts on log uniruled quasi-projective varieties. Definition 3. A quasi-projective variety X is called A1-uniruled, or (properly) log-uniruled after [8] if it contains a Zariski dense open subset U with the property that for every x ∈ U , there exists a non constant morphism fx : A1 → X such that x ∈ fx(A1). 1.2.1. Equivalently, X is A1-uniruled if through a general point there exists a maximally affine rational curve, that is, an affine rational curve whose normalization is isomorphic to the affine line A1. An A1-uniruled variety is in particular uniruled in the usual sense, i.e. there exists a dominant, generically finite rational map P1 × Y 99K X for a suitable quasi-projective variety Y . More precisely, letting (V, D) be a pair consisting of a projective model V of X and a, possibly empty, boundary divisor D such that V is smooth along D, it follows from [8, 5.1] that there exists a closed sub-scheme Y of Mor(P1, V ) consisting of morphisms f : P1 → V with the property that f −1(f (P1) ∩ D) consists of at most one point and on which the restriction of the canonical evaluation morphism ev : P1 × Mor(P1, V ) → V induces a dominant morphism P1 × Y → V . LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 3 1.2.2. It is not difficult to check that a smooth A1-uniruled quasi-projective variety X has logarithmic Kodaira dimension κ(X) = −∞ (see e.g. [8, 5.11]). The converse holds true in dimension ≤ 2 thanks to the successive works of Miyanishi-Sugie [14] and Keel-McKernan [8] and in fact, for a smooth affine surface X, the three conditions κ(X) = −∞, X is A1-uniruled and X is A1-ruled turn out to be equivalent to each other. Furthermore, the surface X has then the stronger property that every of its point belongs to a maximally affine rational curve. This follows from the fact that the projection prY : U ≃ A1 ×Y → Y from an A1-cylindrical open subset U extends to a fibration p : X → Z over an open subset of a smooth projective model of Y , with general fibers isomorphic to A1 and whose degenerate fibers consist of disjoint unions of affine lines (see e.g. [13]). Meanwhile, much less is known in higher dimension where one lacks in particular a good logarithmic analogue of Mori's Bend-and-Break techniques. Nevertheless, the following simple criterion due to Keel and McKernan [8, Corollary 5.4] enables to easily confirm A1-ruledness of certain smooth affine varieties and will be enough for our purpose: Theorem 4. Let X be a smooth affine variety and let X ֒→ (V, D) be a projective completion where V is a smooth projective variety and D = V \ X a reduced divisor on V . If −(KV + D) is ample then X is A1-uniruled. Example 5. The above criterion guarantees for instance that for every n ≥ 1, the complement X of a hypersurface D ⊆ Pn of degree d ≤ n is A1-uniruled. In dimension 2, we recover in particular the fact that the complement of smooth conic Q2 in P2 is A1-uniruled, actually A1-ruled say for instance by the restriction to P2 \ Q2 of the rational pencil generated by Q2 and two times its tangent line at a given point. On the other hand, even though the criterion does not require D to be an SNC divisor, it does not enable to deduce the fact that the complement of a cuspidal cubic D is A1-uniruled, in fact again A1-ruled, namely via the restriction of the rational pencil on P2 generated by D and three times its tangent line at the singular point. 2. Complement of smooth hypersurfaces of degree n in Pn, n ≥ 3 According to Theorem 4, the complement X of a hypersurface Qn of degree n in Pn, n ≥ 2, is a smooth A1-uniruled affine variety. If n = 2, then Q2 is a smooth conic whose complement is in fact A1-ruled (see Example 5). Here we show in contrast that for every n ≥ 3, the complement of a smooth hypersurface Qn ⊆ Pn is not A1-ruled. It is worthwhile to note that the strategy to establish this fact in case of n = 3 is quite different from that of the case of n ≥ 4. 2.1. The case n ≥ 4. Here we will exploit the birational super-rigidity of smooth hypersurfaces Qn ⊂ Pn of degree n ≥ 4 to deduce that the corresponding affine varieties X = Pn \ Qn are not A1-ruled. Let us first briefly recall the notion of birational super-rigidity for a class of variety which is enough for our needs (see e.g. [17] for the general definition). Definition 6. A smooth Fano variety V with Pic(V ) ≃ Z is called birationaly super-rigid if the following two conditions hold: a) The variety V is not birational to a fibration V ′ → S onto a variety of dimS > 0 whose general fibers are smooth varieties with Kodaira dimension −∞, b) Any birational map from V to another smooth Fano variety V ′ with Pic(V ′) ≃ Z is a biregular isomorphism. The essential ingredient we will use is the following result, first established by Pukhlikov [17] under suitable generality assumptions and recently extended to arbitrary smooth hypersurfaces by De Fernex [3]: Theorem 7. Let n ≥ 4 and let Qn ⊆ Pn be a smooth hypersurface of degree n. Then Qn is birationaly super-rigid. n ⊂ Pn between smooth Remark 8. A noteworthy consequence of the above result is that every birational map Qn 99K Q′ hypersurfaces of degree n is in fact a biregular isomorphism which has the additional property to be induced by the restriction of a linear transformation of the ambient space Pn. This follows from the fact that anti-canonical divisors on Qn coincide with hyperplane sections of Qn. In particular, the group Bir(Qn) of birational automorphisms of Qn coincides with that of projective automorphisms Lin(Qn) and the latter is a finite group. Now we are ready to prove the following: Proposition 9. The complement of every smooth hypersurface Qn ⊆ Pn of degree n ≥ 4 is A1-uniruled but not A1-ruled. Proof. The A1-uniruledness of X = Pn \ Qn is an immediate consequence of Theorem 4. Since Cl(X) = Z/nZ, it follows from Proposition 2 that the A1-ruledness of X is equivalent to the existence of a non trivial algebraic Ga-action on it. Every given Ga-action on X induces a one parameter family of automorphisms {ϕt}t∈Ga of X that we interpret through the open inclusion X = Pn \ Qn ֒→ Pn as birational automorphisms Φt : Pn 99K Pn, t ∈ Ga. Since Qn is birationally super-rigid by virtue of Theorem 7 above, it follows that every Φt is in fact a biregular automorphism. Indeed, otherwise, noting that Φt cannot be an isomorphism in codimension one (see e.g. [2]) and letting Pn p q → Pn be a resolution of it, where p consists of successive blow-ups of smooth centers, it would follow that the proper transform q−1 ∗ Qn of Qn in V is an exceptional divisor of p whence is birationally ruled in contradiction with condition a) in Definition 6. Therefore the Ga-action on X extends to a linear one on Pn which leaves X whence Qn invariant. Since the automorphism group Aut(Qn) is finite (see Remark 8 above), the induced Ga-action on Qn is the trivial one and so, being linear, the Ga-action on Pn extending that on X is trivial on the linear span of Qn. But Qn is obviously not contained in any hyperplane of Pn and hence the initial Ga-action on X is necessarily trivial, which completes the proof. (cid:3) ← V LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 4 2.2. The case n = 3: complements of smooth cubic surfaces in P3. Since a smooth cubic surface Q3 in P3 is rational, the same argument as in the previous section depending on birational super-rigidity is no longer applicable to deduce the non A1-ruledness of its complement X = P3 \ Q3. Instead, we will derive it from the unirationality but non rationality of smooth cubic threefolds in P4 together with a suitable finite étale covering trick. The divisor class group of the complement X of a smooth cubic surface Q3 ⊆ P3 being isomorphic to Z/3Z generated by the class of a hyperplane section, it follows again from Proposition 2 that X is A1-ruled if and only if it admits a non trivial algebraic Ga-action. On the other hand, since algebraic Ga-actions lift under finite étale covers (see e.g. [18]), to establish the non A1-ruledness of X it is enough to exhibit a finite étale cover π : X → X of X with the property that X does not admit any non trivial algebraic Ga-action. Proposition 10. Let X = P3 \ Q3 be the complement of a smooth cubic surface and let π : X → X be the canonical étale Galois cover of degree three associated with the canonical sheaf ωX ≃ OP1(−1) X of X. Then X is a smooth affine threefold which does not admit any non trivial algebraic Ga-action. Consequently, X is not A1-ruled. Proof. Letting Q3 ⊆ P3 = Proj(C[x, y, z, u]) be defined as the zero locus of a homogeneous polynomial F (x, y, z, u) ∈ C[x, y, z, u] of degree three, it is straightforward to check that the canonical triple étale covering X of X is isomorphic to the affine variety A4 = Spec(C[x, y, z, u]) defined by the equation F (x, y, z, u) = 1, and that the morphism π : X → X coincides with the restriction of the natural morphism A4 \ {0} → P3, (x, y, z, u) 7→ [x : y : z : u]. Note that by construction, X is an open subset of the smooth cubic threefold V ⊆ Proj(C[x, y, z, u, v]) with equation F (x, y, z, u) − v3 = 0, hence is unirational but not rational by virtue of a famous result of Clemens-Griffiths [1]. Now suppose that there exists a non trivial algebraic Ga-action on X. Then by virtue of [19], the algebra of invariants Γ( X, O X )Ga is a finitely generated integrally closed domain of transcendence degree two over C. Letting q : X → Z = Spec(Γ( X, O X )Ga) be the corresponding quotient morphism, the unirationality of X implies that of Z whence its rationality since these two notions coincide in dimension two. But then the existence of a principal affine open subset Za of Z such that q−1( Za) ≃ A1 × Za (see § 1.1.2 above) would imply in turn that X itself is rational, a contradiction. (cid:3) Remark 11. Complements of singular cubic surfaces Q3 ⊆ P3 may turn out to be A1-ruled. For instance, it is shown in [10] that if Q3 has Du Val singularities worse than A2 then P3 \ Q3 admits a non trivial algebraic Ga-action. It may even happen that P3 \ Q3 contains an A2-cylinder, i.e., an open subset of the form A2 × Y for a smooth curve Y . This holds for instance for the complement of the cubic surface Q ⊆ P3 = Proj(C[x, y, z, u]) defined as the vanishing locus of the homogeneous polynomial F (x, y, z, u) = yu2 + z(xz + y2) which has a unique isolated singularity [1 : 0 : 0 : 0] of type D5. Indeed, noting that f (x, y, z) = F (x, y, z, 1) is a component of the Nagata automorphism [15], it follows that the general fibers of the morphism ρ : P3 \ Q → A1 induced by the restriction of the rational pencil on P3 generated by Q and three times the hyperplane {u = 0} are isomorphic to A2 hence, by virtue of [7], that there exists an open subset Y ⊆ A1 such that ρ−1(Y ) ≃ A2 × Y . A similar construction holds for all normal cubic surfaces listed in [16] which arise as closures in P3 of zero sets of components of automorphisms of A3. 3. Rigid rational fibrations on complements of smooth cubic surfaces in P3 In contrast with the higher dimensional case where a similar argument as the one used in the proof of Proposition 9 shows that for every smooth hypersurface Qn of the degree n in Pn, n ≥ 4, the group Aut(Pn \ Qn) is embeded as a subgroup of Aut(Qn) hence consists of a finite group of linear transformations, it is not known whether every automorphism of the complement of a smooth cubic surface Q = Q3 in P3 is induced by a linear transformation of P3. Even though P3\Q is not A1-ruled (cf. Proposition 10), it turns out that it is fibered in a natural way by A1-ruled affine surfaces whose general closures in P3 are smooth cubic surfaces. Since A1-ruled affine surfaces are usually good candidates for having non trivial automorphisms, one can expect that these fibrations have interesting automorphisms, in particular automorphisms induced by strictly birational transformations of the ambient space P3. In this section, we first review the construction of these fibrations ρ : P3 \ Q → A1 by A1-ruled affine surfaces. We describe the automorphism groups of their general fibers and we check in particular that for a general smooth cubic hypersurface Q ⊆ P3, these fibers do indeed carry interesting automorphisms induced by strictly birational Geiser involutions of their projective closures. We show in contrast that every automorphism of the full fibration comes as the restriction of a linear transformation of P3. 3.1. Special rational pencils on the complement of a smooth cubic surface. Given a smooth cubic surface Q ⊆ P3 and a line L on it, the restriction to Q of the rational pencil HL = OP3(1) ⊗ IL on P3 generated by planes containing L can be decomposed as HL Q= L + L where L is a base point free pencil defining a conic bundle ΦL : Q → P1 with five degenerate fibers, each consisting of the union of two (−1)-curves intersecting transversally. The restriction ΦL L: L → P1 is a double cover and for every branch point x ∈ P1 of ΦL L, the intersection of Q with the corresponding hyperplane Hx ∈ HL consists either of a smooth conic tangent to L or of two distinct lines intersecting L in a same point, which is called an Eckardt point of Q. Letting say H0(L) and H∞(L) be the planes in HL such that ΦL L is ramified over the points ΦL(Ht(L) Q −L), t = 0, ∞, the divisors Q and 3Ht(L) generate a pencil of cubic surfaces ρt(L) : P3 99K P1 with a unique multiple member 3Ht(L) and whose general members are smooth cubic surfaces. 3.1.1. From now on we consider a pencil ρ = ρt(L) as above associated to a fixed line L ⊆ Q and a fixed distinguished plane H ∈ HL intersecting Q either along the union L ∪ C of a line and conic intersecting each other in a single point p or along the union L ∪ L1 ∪ L2 of three lines meeting at an Eckardt point p of Q. We denote by ρ : P3 \ Q → A1 the morphism induced LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 5 by the restriction of ρ to the complement of Q. By construction, the closure in P3 of a general fiber S of ρ is a smooth cubic surface V such that the reduced divisor D = H ∩ V has either the form D = L + C or D = L + L1 + L2. In each case we denote by α : W → V the blow-up of p, with exceptional divisor E. If p is an Eckardt point, then we let V = W and we denote by D ⊆ V the reduced divisor α−1(D)red = L + L1 + L2 + E. Otherwise, if p is not an Eckardt point, then we let V be the variety obtained from W by blowing-up further the intersection point of E and of the proper transform of C, say with exceptional divisor E2, and we let D ⊆ V be the reduced divisor D = L + C + E + E2. By construction, D is an SNC divisor and the induced birational morphism σ : ( V , D) → (V, D) provides a minimal log-resolution of the pair (V, D). The following lemma summarizes basic properties of the general fibers of fibrations of the form ρt(L) : P3 \ Q → A1, t = 0, ∞. Lemma 12. For a general fiber S = V \ D of ρt(L) : P3 \ Q → A1, the following holds: a) S is a smooth affine surface with a trivial canonical sheaf and logarithmic Kodaira dimension κ(S) = −∞, b) If D = L + C (resp. D = L + L1 + L2) then the Picard group of S is isomorphic to Z5 (resp. Z4). Proof. The affineness of S is clear as D is a hyperplane section of V . Since V is a smooth cubic surface, it follows from adjunction formula that ωV ≃ ωP3 ⊗ OP3 (3) V ≃ OP3(−1) V ≃ OV (−D). This implies in turn the triviality of the canonical sheaf of S as ωS ≃ ωV S≃ OV (−D) S≃ OS. We may identify S with V \ D via the birational morphism σ : ( V , D) → (V, D) constructed above. Since D is an SNC divisor by construction, we have κ(S) = κ( V , K V + D). Using the fact that −D is a canonical divisor on V , we deduce from the logarithmic ramification formula for σ that (K V + D = −E if D = L + L1 + L2 K V + D = −E2 if D = L + C. Since E and E2 are exceptional divisors, one has H 0( V , m(K V + D)) = 0 for every m > 0 and hence κ(S) = −∞. To determine the Picard group of S, we will exploit the conic bundle structure ΦL : V → P1 described in the beginning of 3.1. By contracting a suitable irreducible component of each of the five degenerate fibers of ΦL : V → P1, we obtain a birational morphism τ : V → P1 × P1 fitting into a commutative diagram V τ→ P1 × P1 ΦL ↓ ↓ pr1 P1 = P1 and such that the proper transform of L is a smooth 2-section of pr1 with self-intersection 4, whence a section of the second projection pr2 : P1 × P1 → P1. We can then identify the divisor class group Cl(V ) ≃ Z7 of V with the group generated by the proper transforms of a fiber of pr1, a fiber of pr2, and the five exceptional divisors of τ . The proper transform τ∗(D) of D in P1 × P1 is the union of the proper transform of L and of a fiber of pr1. Thus if D = L + C then L and C together with the five exceptional divisors of τ generate Cl(V ) and hence Cl(S) ≃ Pic(S) is isomorphic to Z5 generated by the classes of the intersections of these exceptional divisors with S. Similarly, if D = L + L1 + L2 then L1 or L2, say L2, is an exceptional divisor of τ and so Cl(V ) is generated by L, L1, L2 and the four other exceptional divisors of τ . This implies in turn that Cl(S) ≃ Pic(S) is isomorphic to Z4 generated by the classes of the intersections of these four exceptional divisors with S. (cid:3) 3.1.2. Since a general fiber S = V \ D of ρt(L) : P3 \ Q → A1 has logarithmic Kodaira dimension −∞ by Lemma 12, it is A1-ruled and hence, by virtue of [13], it admits an A1-fibration q : S → C over a smooth curve. Let us briefly explain how to construct such fibrations using the birational morphism τ : V → P1 × P1 which contracts an irreducible component of each of the five degenerate fibers of ΦL : V → P1 as described in the proof of the previous Lemma. According to the configuration of D, we have the following: a) If D = L + C where C is a smooth conic, then each of the points blown-up by τ is the intersection point of the proper transform of L with a fiber of pr2 : P1 × P1 → P1. The proper transforms F1, . . . , F5 ⊆ V of these fibers are disjoint (−1)-curves which do not intersect L. Let µ : V → P2 be the contraction of L and F1, . . . , F5. Since each Fi, i = 1, . . . , 5 intersects C transversally and L is tangent to C, the image of C in P2 is a cuspidal cubic. The rational pencil on P2 generated by the image of C and three times its tangent line T at its unique singular point lifts to a rational pencil q : V 99K P1 having i=1 Fi and 3T + L as singular members. Its restriction to S = V \ D is an A1-fibration q : S → P1 with two degenerate fibers: one is irreducible of multiplicity three consisting of the intersection of the proper transform of T with S and the other is reduced, consisting of the disjoint union of the curves Fi ∩ S ≃ A1, i = 1, . . . , 5. the divisors C +P5 b) The construction in the case where D = L + L1 + L2 is very similar: we may suppose that L1 is contracted by τ so that contracting the same irreducible components F1, . . . , F4 of the degenerate fibers of ΦL : V → P1 but L2 instead of L1, we obtain a birational morphism τ1 : V → F1. The image of L in F1 is a smooth 2-section of the P1-bundle structure π1 : F1 → P1 with self-intersection 4. Letting C0 be the exceptional section of π1 with self-intersection (−1), one has necessarily L ∼ 2C0 + 2F where F is a fiber of π1. So L does not intersect C0 and its image in P2 by the morphism q : F1 → P2 contracting C0 is a smooth conic. Since L1 is tangent to L in F1, its image in P2 is the tangent to the image of C at a point distinct from the one blown-up by q. The rational pencil on P2 generated by the image of L and two times that i=1 Fi as degenerate members. Its restriction to S = V \ D yields an A1-fibration ρ : S → P1 with two degenerate fibers : one is irreducible of multiplicity two consisting of the intersection of the proper transform of C0 with S and the other is reduced, consisting of the disjoint union of the curves Fi ∩ S ≃ A1, i = 1, . . . , 4. of L1 lifts to a rational pencil ρ : V 99K P1 having 2L1 + L2 + 2C0 and L +P4 LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 6 Remark 13. It follows from Corollary 15 below that the automorphism group of a smooth affine surface S = V \ D as above In particular, such surfaces are A1-ruled but do not admit non trivial Ga-actions and so the complement of a is finite. smooth cubic surface Q ⊆ P3 is a smooth affine A1-uniruled threefold without non trivial Ga-action which has the structure of a family ρt(L) : P3 \ Q → A1 of A1-ruled affine surfaces without non trivial Ga-actions. In contrast, the total space of a family of affine varieties with non trivial Ga-actions often admits itself a non trivial Ga-action. For instance, suppose that ϕ : X = Spec(B) → S = Spec(A) is a dominant morphism between complex affine varieties and that ∂ : B → B is an A-derivation of B which is locally nilpotent in restriction to general closed fibers of ϕ, defining non trivial Ga-actions on these fibers. Then ∂ is a locally nilpotent A-derivation of B and hence X carries a non trivial Ga-action. Indeed, up to shrinking S if necessary, we may assume that for every maximal ideal m ∈ Specmax(A), the A/m-derivation ∂m : B/mB → B/mB induced by ∂ is locally nilpotent. Given an element f ∈ B and an integer n ≥ 0, denote by K n(f ) the closed sub-variety of S whose points are the maximal ideals m ∈ Specmax(A) for which the residue class of f in B/mB belongs to Ker∂n m. The hypothesis implies that S is equal to the increasing union of its closed sub-variety K n(f ), n ≥ 0, and so this sequence must stabilizes as the base field C is uncountable. So there exists n0 = n0(f ) such that the restriction of ∂n0 f to every closed fiber of ϕ : X → S is the zero function, and hence ∂n0f = 0 since these fibers form a dense subset of X. 3.2. Automorphisms of general fibers of special rational pencils. This sub-section is devoted to the study of auto- morphisms of general fibers of the rational fibrations ρ : P3 \ Q → A1 constructed as above. We show in particular that such general fibers admit biregular involutions induced by Geiser involutions of their projective closures. 3.2.1. To simplify the discussion, let us call a pair (V, D) as in 3.1.1 special if V is a smooth cubic hypersurface in P3 and D is composed of three concurrent lines L + L1 + L2 meeting in an Eckardt point p of V , or of the union of L and a smooth conic C intersecting L with multiplicity two in a single point p. Similarly as in § 3.1.1, we denote by α : W → V the blow-up of p and we denote by E its exceptional divisor. By construction, W is a weak del Pezzo surface (i.e., −KW is nef and big) of degree 2 on which the anticanonical linear system − KW defines a morphism θ : W → P2. The latter factors into a birational morphism W → Y contracting L (resp. L, L1 and L2) if D = L + C (resp. if D = L + L1 + L2) followed by a Galois double covering Y → P2 ramified over a quartic curve ∆ with a unique double point (resp. three double points) located at the image of L (resp. at the images of L, L1 and L2). The non trivial involution of the double covering Y → P2 induces an involution GW : W → W fixing L and exchanging the proper transform of D and the exceptional divisor E (resp. fixing L, L1 and L2). The latter descends either to birational involution GV,p : V 99K V if D = L + C, or to a biregular involution GV,p : V → V having p as an isolated fixed point if D = L + L1 + L2. In each case, we say that GV,p is the Geiser involution of V with center at p. By construction, GV,p restricts further to a biregular involution jGV,p of the affine surface S = V \ D which we call the affine Geiser involution of S with center at p. Proposition 14. Let ϕ : S = V \ D (V, D) and (V ′, D′) respectively. Then the following assertions hold: ∼ → S ′ = V ′ \ D′ be an isomorphism between affine surfaces associated to special pairs a) If either D or D′ consists of three concurrent lines then ϕ extends to an isomorphism of pairs ϕ : (V, D) b) Otherwise, if both D = L+C and D′ = L′ +C ′ consist of the union of a line and a conic, then the induced birational map ϕ : V 99K V ′ is either an isomorphism of pairs, or it can be decomposed as ϕ = Ψ ◦ GV,p or ϕ = GV ′,p′ ◦ Ψ′ where Ψ, Ψ′ are isomorphisms of pairs and where GV,p and GV ′,p′ denote the Geiser involutions of V and V ′ with centers are p = L ∩ C and p′ = L′ ∩C ′ respectively. Furthermore, every birational map V 99K V ′ of the form Ψ ◦ GV,p (resp. GV ′,p′ ◦ Ψ′) can be uniquely re-written in the form GV ′,p′ ◦ Ψ′ (resp. Ψ ◦ GV,p). In particular, if there exists an isomorphism ϕ : S = V \ D → S ′ = V ′ \ D′ then V and V ′ are isomorphic smooth cubic surfaces. → (V ′, D′). ∼ ∼ Proof. Let σ : ( V , D) → (V, D) and σ′ : ( V ′, D′) → (V ′, D′) be the minimal log-resolutions of the pairs (V, D) and (V ′, D′) respectively described in § 3.1.1. Since V \ D ≃ V \ D and V ′ \ D′ ≃ V ′ \ D′ by construction, every isomorphism ϕ : S → S ′ extends in a natural way to a birational map ϕ : V 99K V ′ restricting to ϕ on S. We claim that ϕ is in fact a biregular β ′ isomorphism of pairs. Indeed, suppose on the contrary that ϕ is strictly birational and let V → V ′ be its minimal resolution. Recall that the minimality of the resolution implies in particular that there is no (−1)-curve in X which is exceptional for β and β′ simultaneously. Since V is smooth and D is an SNC divisor, β′ decomposes into a finite sequence of blow-downs of successive (−1)-curves supported on the boundary B = β−1( D)red = (β′)−1( D)red. In view of the structure of D, the only possible (−1)-curve in B which is not exceptional for β is the proper transform of E if D = L + L1 + L2 or the proper transform of E2 if D = L + C. But after the contraction of these curves, the boundary would no longer be an SNC divisor, a contradiction. This implies that ϕ : V → V ′ is a morphism and the same argument shows that it does not contract any curve in the boundary D. So ϕ : V → V ′ is an isomorphism restricting to an isomorphism between V \ D and V ′ \ D′ whence an isomorphism between the pairs ( V , D) and ( V ′, D′). It follows in particular that the intersection matrices of the divisors D and D′ must be the same up to a permutation. In view of the description given in § 3.1.1, we conclude that either D and D′ simultaneously consist of three concurrent lines or of the union of a line and a smooth conic. β ← X In the first case, since the exceptional divisors E of the blow-up of V at p and E ′ of the blow-up of V ′ at p′ are the only (−1)-curves in D and D′ respectively, the biregular isomorphism ϕ : V → V ′ extending ϕ necessarily maps E isomorphically onto E ′. So ϕ descends to an isomorphism of pairs ϕ : (V, D) → (V ′, D′) which gives a). In the second case, letting D = L + C and D′ = L′ + C ′, a similar argument implies that the isomorphism ϕ : V → V ′ maps E2 onto E ′ 2 and the proper transform of L onto that of L′. This implies in turn that ϕ descends to an isomorphism ϕ1 : W → W ′ between the surfaces obtained from V and V ′ by blowing-up the points p = L ∩ C and p′ = L′ ∩ C ′ with LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 7 respective exceptional divisors E and E ′. Now we have the following alternative: either ϕ1 maps E and C onto E ′ and C ′ respectively, and then it descends to a biregular isomorphism of pairs ϕ : (V, D) → (V ′, D′), or ϕ1 maps E and C onto C ′ and E ′ respectively. In the second case, by composing ϕ : V 99K V ′ either by GV,p on the left or by GV ′,p′ on the right, we get a new birational maps Ψ, Ψ′ : V 99K V ′ which lift to a birational map Ψ1 : W 99K W ′ mapping L, E and C isomorphically onto L′, E ′ and C ′ respectively. The previous discussion implies that Ψ and Ψ′ are isomorphism of pairs and so we get the desired decompositions ϕ = Ψ ◦ G−1 V ′,p′ ◦ Ψ′ = GV ′,p′ ◦ Ψ′. The last assertion about the uniqueness of the re-writing is clear by construction. (cid:3) V,p = Ψ ◦ GV,p and ϕ = G−1 Corollary 15. Let (V, D) be a special pair and let S = V \ D be the corresponding affine surface. Then the following holds: a) If D consists of three concurrent lines then Aut(S) is a nontrivial finite subgroup of PGL(4, C) which coincides with the automorphism group Aut(V, D) of the pair (V, D). b) Otherwise, if D consists of a line and a smooth conic, then we have an exact sequence 0 → Aut(V, D) → Aut(S) → Z2 · jGV,p → 0 and an isomorphism Aut(S) = Aut(V, D)⋊Z2 ·jGV,p , where the affine Geiser involution jGV,p acts on Aut(V, D) by jGV,p ·Ψ = Ψ′ where for every Ψ ∈ Aut(V, D), Ψ′ is the unique automorphism of the pair (V, D) such that Ψ ◦ GV,p = GV,p ◦ Ψ′. Proof. Since V ⊆ P3 is embedded by its anti-canonical linear system − KV , every automorphism of V is induced by a linear transformation of P3. Furthermore, since D is a hyperplane section of V , Aut(V, D) is an algebraic sub-group of the group of linear transformation of P3 preserving globally the corresponding hyperplane. The automorphism group of a smooth cubic surface in P3 being finite, it follows that Aut(V, D) is a finite algebraic group. If D consists of three concurrent lines, then their common point p is an Eckardt point of V which is always the isolated fixed point of a biregular involution of V preserving D. This shows that Aut(V, D) is never trivial in this case. In case b), where D = L + C is the union of a line and smooth conic, the assertion is an immediate consequence of the previous Proposition 14. (cid:3) Remark 16. If D = L+C then every automorphism Ψ of the pair (V, D) preserves the rational pencil q : V 99K P1 constructed in § 3.1.2. Indeed such an automorphism certainly preserves the conic C and it maps the five (−1)-curves F1, . . . , F5 to five other (−1)-curves each intersecting C transversally. But it follows from the construction of V from P2 that F1, . . . , F5 are the only five (−1)-curves in V with this property. Similarly, since L and p = L ∩ C are Ψ-invariant, the image of T by Ψ must be a smooth rational curve intersecting L and C transversally at p but the construction of V again implies that T is the only i=1 Fi and 3T + L which generate the pencil q : V 99K P1 are Ψ-invariant whence that q is globally preserved by Ψ as claimed. This implies in turn that every automorphism ϕ of S which extends to a biregular automorphism of the pair (V, D) fits into a commutative diagram curve in V with this property. It follows that the divisors C +P5 S q ↓ P1 ϕ → S ↓ q ξ → P1, where ξ ∈ PGL(2, C) fixes the two points of P1 corresponding to the non isomorphic degenerate fibers 3T ∩S andP5 i=1 Fi ∩S of q. In contrast, it is straightforward to check from the construction that q : S → P1 is not globally preserved by the affine Geiser involution jGV,p of S with center at p = D ∩ L and hence that S carries a second A1-fibration q ◦ jGV,p : S → P1 whose general fibers are distinct from that of q. In particular, if V is chosen generally so that Aut(V, D) is trivial, then Aut(S) is isomorphic to Z2, generated by the affine Geiser involution jGV,p (cf. Corollary 15, (b)), and S carries two conjugated A1-fibrations q : S → P1 and q ◦ jGV,p : S → P1, each of these having no non trivial automorphism. 3.3. Automorphisms of special rational pencils. This subsection is devoted to the proof of the following result: Proposition 17. Let Q ⊆ P3 be a smooth cubic surface and let ρ : P3 99K P1 be the special pencil associated to a line L ⊆ Q and a distinguished hyperplane H ∈ OP3(1) ⊗ IL as in § 3.1.1. Then every automorphism of the induced rational fibration ρ : P3 \ Q → A1 is the restriction of an automorphism of P3. Proof. Let us denote by Aut(P3 \ Q, ρ) the subgroup of Aut(P3 \ Q) consisting of automorphisms preserving the fibration ρ : P3 \ Q → A1 = Spec(C[λ]) fiberwise. Taking restriction over the generic point η of A1 induces an injective homomorphism from Aut(P3 \ Q, ρ) to the group Aut(Sη) of automorphisms of the generic fiber Sη of ρ, and we identify from now on Aut(P3 \ Q, ρ) with its image in Aut(Sη). By definition, Sη is a nonsingular affine surface defined over the field C(λ) whose closure in P3 C(λ) is a nonsingular cubic surface Vη such that Dη = Vη \Sη consists of either three lines Lη, Lη,1 and Lη,2 meeting a unique C(λ)-rational point or the union of a line Lη and nonsingular conic Cη intersecting Lη at a unique C(λ)-rational point p. The same argument as in the proofs of Proposition 14 and its Corollary 15 implies that Aut(Sη) = Aut(Vη, Dη) in the first case whereas Aut(Sη) = Aut(Vη, Dη) ⋊ Z2 · jGVη ,p , where jGVη ,p denotes the affine Geiser involution with the center at p, in the second case. In the first case, we conclude that every automorphism of Aut(P3 \ Q, ρ) is generically of degree one when considered as a birational self-map of P3 and hence is the restriction of an automorphism of P3. So it remains to show that in the second case, one has Aut(P3 \ Q, ρ) ⊆ Aut(Vη, Dη) necessarily. Suppose on the contrary that Aut(P3 \ Q, ρ) 6⊆ Aut(Vη, Dη). Then since we have an extension 0 → Aut(Vη, Dη) → Aut(Sη) → Z2 · jGVη,p → 0, LOG-UNIRULED AFFINE VARIETIES WITHOUT CYLINDER-LIKE OPEN SUBSETS 8 it would follow that jGVη,p ∈ Aut(P3 \ Q, ρ). On the other hand, letting πp : P3 99K P2 be the projection from the point p, the rational map (ρ, πp) : P3 99K P1 × P2 is generically of degree 2, contracting the plane H to a line πp(H) ∈ ρ(H) × P2 ≃ P2 and inducing a Galois double covering P3 \ (Q ∪ H) → P1 \ {ρ(Q) ∪ ρ(H)} × P2 \ (πp(Q) ∪ πp(H)) ≃ A1 ∗ × A2 \ (πp(Q)). By definition, the affine Geiser involution jGVη ,p of Sη is simply the restriction to the generic fiber of ρ of the nontrivial involution J of this covering. So jGVη,p ∈ Aut(P3 \ Q, ρ) considered as a birational self-map of P3 would coincide with J which is absurd since the latter contracts H whence the fiber ρ−1(ρ(H)) of ρ. Thus Aut(P3 \ Q, ρ) ⊆ Aut(Vη, Dη) as desired. (cid:3) References 1. C. Clemens, and P. Griffiths, The intermediate Jacobian of the cubic threefold,The Annals of Mathematics, Vol. 95 (1972), No. 2 281-356. 2. A. Corti, Factoring birational maps of threefolds after Sarkisov, J. Algebraic Geom. 4 (1995), no. 2, 223? 223-254. 3. T. de Fernex, Birationally rigid hypersurfaces, Invent. Math., to appear, preprint arXiv:math/0604213. 4. G. Freudenburg, Algebraic Theory of Locally Nilpotent Derivations, Encyclopaedia of Mathematical Sciences Invariant Theory and Algebraic Transformation Groups, Volume VII R. V. Gamkrelidze, V. L. Popov, Subseries Editors Springer Verlag (Berlin, Heidelberg, New York) 2006. 5. R. Gurjar, M. Masuda and M. Miyanishi, A1-fibrations on affine threefolds, Journal of Pure and Applied Algebra, 216 (2012), 296-313. 6. S. Iitaka, On logarithmic Kodaira dimension of algebraic varieties, Complex Analysis and Algebraic Geometry (W.L. Baily, Jr. and T. Shioda, eds.), Iwanami Shoten, Tokyo, 1977, 175-189. 7. S. Kaliman and M. Zaidenberg, Families of affine planes: The existence of a cylinder, Michigan Math. J. 49 (2001), 353-367. 8. S. Keel and J. McKernan, Rational curves on quasi-projective surfaces, Memoirs of AMS, No. 669, American Mathematical Society, 1999. 9. T. Kishimoto, Yu. Prokhorov, M. Zaidenberg, Group actions on affine cones, Affine Algebraic Geometry, The Russell Festschrift (Eds. D. Daigle, R. Ganong and M. Koras), CRM Proceedings & Lecture Note Vol. 54, American Mathematical Society, 2011, pp. 123-163. 10. T. Kishimoto, Yu. Prokhorov and M. Zaidenberg, Affine cones over Fano threefolds and additive group actions, preprint arXiv:1106.1313v1 [math.AG]. 11. J. Kollár et al, Flips and abundance for algebraic threefolds, Astérisque 211, Société Mathématique de France, 1992. 12. J. Kollár and S. Mori, Birational Geometry of Algebraic Varieties, Cambridge Tracts in Math. 134, Cambridge University Press. 13. M. Miyanishi, Singularities on normal affine surfaces containing cylinderlike open sets, J. Algebra, 68 (1981), 268-275. 14. M. Miyanishi and T. Sugie, Affine surfaces containing cylinderlike open sets, J. Math. Kyoto Univ. 20 (1980), no. 1, 11-42. 15. M. Nagata, Polynomial rings and affine spaces, Regional Conference Series in Math. Vol.37, AMS, Providence, RI, 1978. 16. T. Ohta, The structure of algebraic embeddings of C2 into C3 (the normal quartic hypersurface case. I), Osaka J. Math. 38 (2001), 507-532. 17. A.V. Pukhlikov, Birational automorphisms of Fano hypersurfaces, Invent. Math. 134 (1998), 401-426. 18. A. Seidenberg, Derivations and integral closure, Pacific Journal, 16(1966), 167-173. 19. O. Zariski, Interpretations algébrico-géometriques du quatorzième probleme de Hilbert, Bulletin des Sciences Mathéematiques 78 (1954), pp. 155-168. Institut de Mathématiques de Bourgogne, 9 avenue Alain Savary, 21 078 Dijon, France E-mail address: [email protected] Department of Mathematics, Faculty of Science, Saitama University, Saitama 338-8570, Japan E-mail address: [email protected]

Subsets and Splits